Automation

AUTOMATION
EditedbyFlorianKongoli

Automation
Edited by Florian Kongoli

Published by InTech
Janeza Trdine 9, 51000 Rijeka, Croatia

Copyright 2012 InTech
All chapters are Open Access distributed under the Creative Commons Attribution 3.0
license, which allows users to download, copy and build upon published articles even for
commercial purposes, as long as the author and publisher are properly credited, which
ensures maximum dissemination and a wider impact of our publications. After this work
has been published by InTech, authors have the right to republish it, in whole or part, in
any publication of which they are the author, and to make other personal use of the
work. Any republication, referencing or personal use of the work must explicitly identify
the original source.

As for readers, this license allows users to download, copy and build upon published
chapters even for commercial purposes, as long as the author and publisher are properly
credited, which ensures maximum dissemination and a wider impact of our publications.

Notice
Statements and opinions expressed in the chapters are these of the individual contributors
and not necessarily those of the editors or publisher. No responsibility is accepted for the
accuracy of information contained in the published chapters. The publisher assumes no
responsibility for any damage or injury to persons or property arising out of the use of any
materials, instructions, methods or ideas contained in the book.

Publishing Process Manager Romina Skomersic
Technical Editor Teodora Smiljanic
Cover Designer InTech Design Team

First published July, 2012
Printed in Croatia

A free online edition of this book is available at www.intechopen.com
Additional hard copies can be obtained from [email protected]

Automation, Edited by Florian Kongoli
p. cm.
ISBN 978-953-51-0685-2

Contents

Preface IX
Chapter 1 Optimization of IPV
6
over 802.16e WiMAX
Network Using Policy Based Routing Protocol 1
David Oluwashola Adeniji
Chapter 2 Towards Semantic Interoperability in
Information Technology: On the Advances in Automation 17
Gleison Baico, Anilton Salles Garcia and Giancarlo Guizzardi
Chapter 3 Automatic Restoration of Power Supply in
Distribution Systems by Computer-Aided Technologies 45
Daniel Bernardon, Mauricio Sperandio,
Vincius Garcia, Luciano Pfitscher and Wagner Reck
Chapter 4 Automation of Subjective Measurements
of Logatom Intelligibility in Classrooms 61
Stefan Brachmanski
Chapter 5 Automation in Aviation 79
Antonio Chialastri
Chapter 6 Power System and Substation Automation 103
Edward Chikuni
Chapter 7 Virtual Commissioning of Automated Systems 131
Zheng Liu, Nico Suchold and Christian Diedrich
Chapter 8 Introduction to the Computer
Modeling of the Plague Epizootic Process 149
Vladimir Dubyanskiy, Leonid Burdelov and J. L. Barkley
Chapter 9 Learning Automation to Teach Mathematics 171
Josep Ferrer, Marta Pea and Carmen Ortiz-Caraballo
VI Contents

Chapter 10 Rapid Start-Up of the Steam Boiler,
Considering the Allowable Rate of Temperature Changes 199
Jan Taler and Piotr Harchut
Chapter 11 A Computer-Aided Control and
Diagnosis of Distributed Drive Systems 215
Jerzy wider and Mariusz Hetmaczyk
Chapter 12 Optical Interference Signal Processing
in Precision Dimension Measurement 241
Haijiang Hu, Fengdeng Zhang, Juju Hu and Yinghua Ji
Chapter 13 Land Administration and Automation in Uganda 259
Nkote N. Isaac
Chapter 14 A Graphics Generator for Physics Research 269
Eliza Consuela Isboiu
Chapter 15 Genetic Algorithm Based Automation
Methods for Route Optimization Problems 293
G. Andal Jayalakshmi
Chapter 16 Automatic Control of the Software
Development Process with Regard to Risk Analysis 311
Marian Jureczko
Chapter 17 Automation in the IT Field 339
Alexander Khrushchev
Chapter 18 VHDL Design Automation
Using Evolutionary Computation 357
Kazuyuki Kojima
Chapter 19 Applicability of GMDH-Based
Abductive Network for Predicting Pile Bearing Capacity 377
Isah A. Lawal and Tijjani A. Auta
Chapter 20 An End-to-End Framework
for Designing Networked Control Systems 391
Alie El-Din Mady and Gregory Provan
Chapter 21 The Role of Automation
in the Identification of New Bioactive Compounds 417
Pasqualina Liana Scognamiglio,
Giuseppe Perretta and Daniela Marasco
Contents VII

Chapter 22 Automatic Stabilization of Infrared
Images Using Frequency Domain Methods 435
J. R. Martnez de Dios and A. Ollero
Chapter 23 SITAF: Simulation-Based Interface Testing
Automation Framework for Robot Software Component 453
Hong Seong Park and Jeong Seok Kang
Chapter 24 Advanced Bit Stuffing Mechanism
for Reducing CAN Message Response Time 471
Kiejin Park and Minkoo Kang
Chapter 25 A Systematized Approach to Obtain
Dependable Controller Specifications for Hybrid Plants 487
Eurico Seabra and Jos Machado
Chapter 26 Answering Causal Questions
and Developing Tool Support 507
Sodel Vazquez-Reyes and Perla Velasco-Elizondo
Chapter 27 An Intelligent System for
Improving Energy Efficiency in Building
Using Ontology and Building Automation Systems 531
Hendro Wicaksono, Kiril Aleksandrov and Sven Rogalski

Preface

Automation is closely related to the modern need for sustainable development in the 21
st

century. One of the principles of sustainability is Doing More with Less which in other
words, is also one of the goals of automation. By replacing the routine part of human
labor with the use of machines, automation not only increases productivity and the
quality of products beyond what can be achieved by humans but also frees space, time
and energy for humans to deal with the new, non-routine challenge of developing
innovative and more advanced technologies. This magnificent cycle in which established
developments are automated and the free resources achieved by this automation are
used to develop newer technologies that are subsequently automated is one of the most
successful recipes for the human race towards the goal of sustainable development.
Automation as a new discipline emerged in the second half of the 20
th
century and
quickly attracted some of the brightest scientists in the world. Although in the beginning
its theory and industrial applications were mostly in the electrical, mechanical, hydraulic
and pneumatic fields, the computer revolution has made possible the invasion of
automation in all fields of life without exception. This phenomenon has perpetuated
itself by such a degree that today computerization and automation can hardly be clearly
distinguished from one another. This is also reflected in the contents of this book where
interesting contributions cover various fields of technical as well as social sciences. At
first sight it seems that the chapters of the book have nothing in common since they
cover different fields and are directed only to specialists in their respective domains. In
fact, they are bound together by the need for automation in the various routine aspects of
our lives and it is here where the overlapping interest of the reader is expected. In order
to facilitate and encourage broad interest from scientists of various fields, each chapter
starts similarly with a plain and simple description of the related scientific field where
automation is applied, a short review of previous achievements, and a capsule detailing
what the new developments are.

Dr. Florian Kongoli, BSc (Honors), MASc, PhD,
MTMS, MGDMB, MCIM, MSME, MAIST,MISIJ,
MSigmaXi, MIFAC, MACS, MASM, MMRS, MACerS, MECS
Executive President (CEO), FLOGEN Technologies Inc.
Chairman, FLOGEN Star OUTREACH

1
Optimization of IPV
6
over 802.16e WiMAX
Network Using Policy Based Routing Protocol
David Oluwashola Adeniji
University of Ibadan
Nigeria
1. Introduction
Internet application needs to know the IP address and port number of the remote entity
with which it is communicating during mobility. Route optimization requires traffic to be
tunneled between the correspondent node(CN) and the mobile node (MN). ). Mobile IPv6
avoids so-called triangular routing of packets from a correspondent node to the mobile node
via the Home Agent. Correspondent nodes now can communicate with the mobile node
without using tunnel at the Home Agent..The fundamental focal point for Optimization of
IPv6 over WiMAX Network using Policy Based Routing Protocol centered on the special
features that describe the goal of optimization mechanism during mobility management. To
reiterate these features there is need to address the basic concept of optimization as related
to mobility in mobile IPv6 Network. WiMAX which stands for Worldwide Interoperability
for Microwave Access, is an open, worldwide broadband telecommunications standard for
both fixed and mobile deployments. The primary purpose of WiMAX is to ensure the
delivery of wireless data at multi-megabit rates over long distances in multiple ways.
Although WiMAX allows connecting to internet without using physical elements such as
router, hub, or switch. It operates at higher speeds, over greater distances, and for a greater
number of people compared to the services of 802.11(WiFi).A WiMAX system has two units.
They are WiMAX Transmitter Tower and WiMAX Receiver. A Base Station with WiMAX
transmitter responsible for communicating on a point to multi-point basis with subscriber
stations is mounted on a building. Its tower can cover up to 3,000 Sq. miles and connect to
internet. A second Tower or Backhaul can also be connected using a line of sight, microwave
link. The Receiver and antenna can be built into Laptop for wireless access.This statement
brings to the fact that if receiver and antenna are built into the laptop, optimization can take
place using a routing protocol that can interface mobile IPv6 network over WiMAX
802.16e.However the generic overview of optimization possibilities most especially for a
managed system can be considered.
What then is Optimization? Basically optimization is the route update signaling of
information in the IP headers of data packets which enable packets to follow the optimal
path and reach their destination intact. The generic consideration in designing route
optimization scheme is to use minimumsignaling information in the packet header.
Furthermore the delivery of managed system for optimization describes the route
optimization operation and the mechanism used for the optimization. In order for

Automation

2
optimization to take place, a protocol called route optimization protocol must be introduced.
Route optimization protocol is used basically to improve performance. Also route
optimization is a technique that enables mobile node and a correspondent node to
communicate directly, bypassing the home agent completely; this is based on IPv6 concept.
The use of domains enables a consistent state of deployment to be maintained. The main
benefits of using policy are to improve scalability and flexibility for the management system.
Scalability is improved by uniformly applying the same policy to large sets of devices ,
while flexibility is achieved by separating the policy from the implementation of the
managed system. Policy can be changed dynamically, thus changing the behavior and
strategy of a system, without modifying its implementation or interrupting its operation.
Policy-based management is largely supported by standards organizations such as the
Internet Engineering Task Force (IETF) and the Distributed Management Task Force
(DMTF) and most network equipment vendors.However the Architectures for enforcing
policies are moving towards strongly distributed paradigms, using technologies such as
mobile code, distributed objects, intelligent agents or programmable networks.
Mobile IP is the standard for mobility management in IP networks. New applications and
protocols will be created and Mobile IP is important for this development. Mobile IP
support is needed to allow mobile hosts to move between networks with maintained
connectivity. However internet service driven network is a new approach to the provision of
network computing that concentrates on the services you want to provide. These services
range from the low-level services that manage relationships between networked devices to
the value-added services provided to the end-users. The complexity of the managed systems
results in high administrative costs and long deployment cycles for business initiatives, and
imposes basic requirements on their management systems. Although these requirements
have long been recognized, their importance is now becoming increasingly critical. The
requirements for management systems have been identified and can be facilitated with
policy-based management approach where the support for distribution, automation and
dynamic adaptation of the behaviour of the managed system is achieved by using policies.
IPV6 is one of the useful delivery protocols for future fixed and wireless/mobile network
environment while multihoming is the tools for delivering such protocol to the end users.
Optimization of Network must be able to address specific market requirements, deployment
geographical, end-user demands, and planned service offerings both for today as well as for
tomorrow.
2. IP mobility
The common mechanism that can manage the mobility of all mobile nodes in all types of
wireless networks is the main essential requirement for realizing the future ubiquitous
computing systems. Mobile IP protocol V
4
or V
6
considered to be universal solutions for
mobility management because they can hide the heterogeneity in the link-specific
technology used in different network. Internet application needs to know the IP address and
port number of the remote entity with which it is communicating during mobility. From a
network layer perspective, a user is not mobile if the same link is used, regardless of
location. If a mobile node can maintain its IP address while moving, it makes the movement
transparent to the application, and then mobility becomes invisible. From this problem the
basic requirement for a mobile host is the

Optimization of IPV
6
over 802.16e WiMAX Network Using Policy Based Routing Protocol

3
Mobile IP works in the global internet when the mobile node (MN) which belong to the
home agent (HA) moves to a new segment, which is called a foreign network (FN). The MN
registers with the foreign agent (FA) in FN to obtain a temporary address i.e a care of
address(COA).The MN updates the COA with the HA in its home network by sending BU
update message. Any packets from the corresponding node (CN) to MN home address are
intercepted by HA. HA then use the BU directly to the CN by looking at the source address
of the packet header.

Fig. 1. Mobile IP Network
The most critical challenges of providing mobility at the IP layer is to route packets
efficiently and securely. In the mobile IP protocol all packets are sent to a mobile node while
away from home are intercepted by its home agent and tunneled to the mobile node using
IP encapsulation within IP.
2.1 Limitation of mobile IP
Mobile IP can only provide continuous Internet connectivity and optimal routing to a
mobile host, and are not suitable to support a mobile network. The reasons is that, not all
devices in a mobile network is sophisticated enough to run these complicated
protocols.Secondly, once a device has joined a mobile network, it may not see any link-level
handoffs even as the network moves.
2.2 Detailed description of NEMO
Network Mobility describes the situation of a router connecting an entire network to the
Internet dynamically changes its point of attachment. The connections of the nodes inside
the network to the Internet are also influenced by this movement. A mobile network can be

Automation

4
connected to the Internet through one or more MR, (the gateway of the mobile network),
there are a number of nodes (Mobile Network Nodes, MNN) attached behind the MR(s). A
mobile network can be local fixed node, visiting or nested. In the case of local fixed node:
nodes which belong to the mobile network and cannot move with respect to MR. This node
are not able to achieve global connectivity without the support of MR. Visiting node belong
to the mobile network and can move with respect to the MR(s).Nested mobile network
allow another MR to attached to its mobile network. However the operation of each MR
remains the same whether the MR attaches to MR or fixed to an access router on the
internet. Furthermore in the case of nested mobile network the level mobility is unlimited,
management might become very complicated. In NEMO basic support it is important to
note that there are some mechanism that allow to allow mobile network nodes to remain
connected to the Internet and continuously reachable at all times while the mobile network
they are attached to changes its point of attachment.Meanwhile, it would also be meaningful
to investigate the effects of Network Mobility on various aspects of internet communication
such as routing protocol changes, implications of real-time traffic, fast handover and
optimization. When a MR and its mobile network move to a foreign domain, the MR would
register its care-of-address (CoA) with its HA for both MNNs and itself. An IP-in-IP tunnel
is then set up between the MR and its HA. All the nodes behind the MR would not see the
movement, thus they would not have any CoA, removing the need for them to register
anything at the HA. All the traffic would pass the tunnel connecting the MR and the HA.
Figure 2.3 describes how NEMO works.

Fig. 2. Network Mobility
2.3 Micro mobility and Macro mobility
This discussion on Micromobility and Macromobility is centered on wireless
communication architecture that focuses on the designed of IP Micromobility protocol that
compliment an IETF standard for Macromobility management which is usually called
Mobile IP. From this point of view, Macro-mobility concerns with the management of users
movements at a large scale, between different wide wirelesses accesses networks connected
to the Internet. Macro-mobility is often assumed to be managed through Mobile IP. On the

Optimization of IPV
6

5
other hand, Micro-mobility covers the management of users movements at a local level,
inside a particular wireless network.The standard Internet Protocol assumes that an IP
address always identifies the node's location in the Internet. This means that if a node moves
to another location in the Internet, it has to change its IP address or otherwise the IP packets
cannot be routed to its new location anymore. Because of this the upper layer protocol
connections have to be reopened in the mobile node's new location.
The Technologies which can be Hierarchical Mobile IP,Cellular IP,HAWAII at different
micro mobility solutions could coexist simultaneously in different parts of the Internet. Even
at that the Message exchanges are asymmetric on Mobile IP. In cellular networks, message
exchanges are symmetric in that the routes to send and receive messages are the same.
Considering the mobile Equipment the appropriate location update and registration
separate the global mobility from the local mobility. Hence Location information is
maintained by routing cache. During Routing most especially in macro mobility scenario a
node uses a gateway discovery protocol to find neighboring gateways Based on this
information a node decides which gateway to use for relaying packets to the Internet. Then,
packets are sent to the chosen gateway by means of unicast. With anycast routing, a node
leaves the choice of gateway to the routing protocol.
The routing protocol then routes the packets in an anycast manner to one of the gateways.In
the first case, a node knows which gateway it relays its packets to and thus is aware of its
macro mobility.
The comparative investigation of different requirement between Micro mobility and Macro
mobility are based on following below:
- Handoff Mobility Management Parameter:The interactions with the radio layer, initiator
of the handover management mechanism, use of traffic bicasting were necessary. The
handoff latency is the parameter time needed to complete the handoff inside the network.
Also potential packet losses were the amount of lost packets during the handoff must be
deduced. Furthermore the involved stations: the number of MAs that must update the
respective routing data or process messages during the handover are required.
- Passive Connectivity with respect to paging required an architecture that can support
via paging order to control traffic against network burden.This architecture is used to
support only incoming data packet.Therefore the ratio of incoming and outcoming
communication or number of handover experienced by the mobiles are considered for
efficiency support purposes.To evaluate this architecture an algorithm is used to
perform the paging with respect to efficiency and network load.
- Intra Network Traffic basic in micro mobility scenario are the exchanged of packet
between MNs connected to the same wireless network. This kind of communication is a
large part of todays GSM communications and we can expect that it will remain an
important class of traffic in future wireless networks.
- Scalability and Robustness: The expectation is that future large wireless access
networks will have the same constrains in terms of users load. These facts are to be
related to the increasing load of todays Internet routers: routing tables containing a few
hundreds of thousandsentries have become a performance and optimization problem.
The review of micromobility via macromobility of key management in 5G technology must
addressed the following features:

Automation

6
5G technology offer transporter class gateway with unparalleled consistency.
The 5G technology also support virtual private network
Remote management offered by 5G technology a user can get better and fast solution.
The high quality services of 5G technology based on Policy to avoid error.
5G technology offer high resolution for crazy cell phone user and bi-directional large
bandwidth shaping.
2.4 Concept of Multihoming
Mobile networks can have multiple points of attachment to the internet, in this case they are
said to be multihomed. Multihoming arises when the MR has multiple addresses, multiple
egress interfaces on the same link, or multiple egress interfaces on different links. Basically
the classification of configuration can be divided into :Configuration-Oriented Approach,
Ownership-oriented Approach and Problem-Oriented Approach. The multihoming analysis
classifies all these configuration of multihomed mobile networks using (x, y, z) notation.
Variables x, y, and z respectively mean the number of MRs connected to the Internet (so
called root MRs), the number of HAs, and the number of Mobile Network Prefix (MNP) s. In
case of 1, each variable implies that there exists a single node or prefix. If the variable is N,
then it means that one or more agents or prefixes exist in a single mobile network. From
different combinations of the 3-tuple (x, y, z), various types of multihoming scenarios are
possible. For example the (N, 1, 1) scenario means there is multiple MRs at the mobile
network, but all of MRs are managed by single HA and use same MNP.

Fig. 3. A Multihoming of Nested Mobile Network
The Figure above shows how a train provide a Wifi network to the passengers with MR3, the
passenger could connect to MR3 with MR1 (for example his Laptop). The passenger could also
connect directly to Internet with MR2 (his Phone with its GPRS connectivity).The train is
connected to Internet with Wimax connectivity. The MNNs can be a PDA and some sensors.
Multihoming from the above nested mobile network provides the advantages of session
preservation and load sharing. During optimization the key data communication that must
be taking into consideration are:
Session preservation by redundancy.-the session must be preserved based on the
available stable mobile environment either via wireless or wired.

Optimization of IPV
6

7
Load balancing by selecting the best available interface or enabling multiple interfaces
simultanousely.Traffic load balancing at the MR is critical since in mobile networks, all
traffic goes through the MR.
The apprehension from the above can be justified by specifying the mobile message notification
mobile node as well as the procedure for node joining. A mobility notification message contains
two important information:(i) the notification interval for multihoming; and (ii)the prefix of the
access network that the sending gateway belongs. The optimal choice of the notification interval
depends on the mobility of the nodes as well on the amount of traffic sent.
Processing of packets from multihomed nodes is more complex and requires the gateway to
perform two tasks. First,the gateway has to verify if a node has recently been informed that its
packets are relayed through this access network. If this does not take place, the gateway sends a
mobility notification message to the mobile node to inform it about the actual access network.
For reducing the amount of mobility notification messages, the gateway records the node
address combined with a time stamp in a lookup table. After a notification interval, the gateway
deletes the entry and if it is still relaying packets for this node, notifies the mobile node
again.Secondly the gateway substitutes the link-local address prefix of the IP source address of
the packet with the prefix of the access network it belongs to and forwards the packet to the
Internet. When a multihoming node receives a mobility notification message, it adjusts its
address prefix to topologically fit the new access network. Subsequently,it informs about its
address change using its IP mobility management protocols. In the case where packets of a node
are continuously forwarded over different access networks,multihoming support is an
advantage to prevent continuous address changes. When a multihoming node receives a
mobility notification message, it checks if it already is aware of X or Y access network.
2.5 Requirement of Multihoming configuration
The requirement for Multihomed configurations can be classified depending on how many
MRs are present, how many egress interfaces, Care-of Address (CoA), and Home Addresses
(HoA) the MRs have, how many prefixes (MNPs) are available to the mobile network nodes,
etc. The reader of this chapter should note that there are eight cases of configuration of
multihomed mobile network. The 3 key parameter associated to differentiate the
configuration are referred to 3-tuple X, Y, Z. To describe any of this requirement
configurationin respect to macro mobility, a detection mechanism and notification protocol
is required. The below table present the most significant features of the eight classification
approach for NEMO. Although there are several configuration but NEMO does not specify
any particular mechanism to manage multihoming.

Configuration Class Requirement
Prefix
Advertisement
1 Configuration 1, 1, 1 Class 1,1,1 MR, HA, MNP 1 MNP
2 Configuration 1, 1, 1 Class 1,1,n 1 MR, 1 HA, More MNP 2 MNP
3 Configuration 1, 1, 1 Class 1,n,1 1 MR, More HA, 1 MNP 1 MNP
4 Configuration 1, 1, 1 Class 1,n,n 1 MR, More HA, More MNP Multiple MNP
5 Configuration 1, 1, 1 Class n,1,1 More MR, 1 HA, 1 MNP MNP
6 Configuration 1, 1, 1 Class n,1,n More MR, 1 HA, More MNP Multiple MNP
7 Configuration 1, 1, 1 Class n,n,1 More MR, More HA, MNP 1 MNP
8 Configuration 1, 1, 1 Class n,n,n More MR, More HA, More MNP Multiple MNP
Table 1. Analysis of Eight cases of multihoming configuration

Automation

8
What then about the reliability of these configurations during NEMO? Internet connection
through another interface must be reliable. The levels of redundancy cases can be divided
into two: if the mobile nodes IP address is not valid any more, and the solution is to use
another available IP address; the order is that the connection through one interface is
broken, and the solution is to use another. If one of the interfaces is broken then the
solution is to use another interface using a Path Exploration. However in the case of
multiple HAs, the redundancy of the HA is provided, if one HA is broken, another one
could be used. The important note here is the broken of one interface can also lead to
failure. Failure Detection in all the cases in which the number of MNP is larger than 1,
because the MNN could choose its own source address, if the tunnel to one MNP is
broken, related MNNs have to use another source address which is created from another
MNP. In order to keep sessions alive, both failure detection andredirection of
communication mechanisms are needed. If those mechanisms could not perform very
well, the transparent redundancy can not be provided as well as in the cases where only
one MNP is advertised.
The only difference between using one MR with multiple egress interfaces and using
multiple MRs each of which only one egress interface is Load sharing. Multiple MRs could
share the processing task comparing with only one MR, and of course it provides the
redundancy of the disrupting of the MR. Therefore the mechanisms for managing and
cooperating between each MRs are needed.Also the common problem related to all the
configuration is where the number of MNP are larger than 1 and at the same time the
number of MR or the number of HA or both is larger than one. So a mechanism for solving
the ingress filtering problem should be used. In most cases the solution is to use second
binding on the ingress interface by sending a Prefix-BU through the other MRs and then the
HA(s) get(s) all other CoAs.
How do we then distinguish between CoAs.? We use Preference Settings One solution is to
use an extra identifier for different CoAs and include the identifier information in the
update message. This kind of situation exists a lot, except for the cases in which one MNP is
only allowed to be controlled by one CoA.
2.6 Policy Based Routing Protocol
Policy is changing the behavior and strategy of a system, without modifying its
implementation or interrupting its operation. Policy-based management is largely
supported by Standards organizations such as the Internet Engineering Task Force (IETF)
and the Distributed Management Task Force (DMTF) and most network equipment
vendors. The focal point in the area of policy-based management is the notion of policy as a
means of driving management procedures. Although the technologies for building
management building management systems are available, work on the specification and
deployment of policies is still scarce.Routing decisions and interface selection are based
entirely on IP/network layer information.In order to provide adequate information the level
of hierarchy must be considered.The specific information during optimization and
deployment is based on the following approach:
Link Layer Information: Interface selection algorithm should take into account all available
information and at the same time minimize resource consumption and make decisions with

Optimization of IPV
6

9
as light computation as possible .However , link quality must be constantlymonitored and
the information must be made available for the network layer and user applications in a
form that suits them best.
IP layer Information: Several attributes can be retrieved from the IPv6 header without
looking into the data, e.g., source address and destination address etc.Some attributes can
also be retrieved from IPv6 extension headers (e.g. HOA) only transport protocols like TCP
and UDP can be identified directly from the IP header.
Network Originated Information: A service provider may disseminate information about
cost, bandwidth and availability of the Internet access in an area using WiMax,WLAN,
GPRS and Bluetooth. To advertise such information the default gateway or the access router
can send information on cost and bandwidth. The mobile users could then have preferences
for connections, like maximize bandwidth or minimize price and the host would select the
appropriate interface satisfying these preferences.
This section of the chapter considered the below Algorithm for message notification of
mobile node.

Fig. 4. Algorithm for mobility notification messages at a mobile node during optimization.

Automation

10
2.7 Mechanism for interface selection
The separation of policy and mechanism makes it possible to implement a dynamic
interface selection system. The mechanism evaluates connection association and transport
information against the actions in policies, using principles.The interface selection system
is based on four basic components, entities, action, policy and mechanism. Entities define
actions. An entity may be a user, peer node or 3rd party, e.g., operator. Action is an
operation that is defined by an entity and is controlled by the system. Actions specify the
interfaces to be used for connections on account of entitys requirements. Actions can be
presented as conditional statements. Policy governs the actions of an entity. Only one
action can take place at a time in a policy. A policy set contains several policies possible
defined by different entities. Mechanism evaluates actions against connection related
information and decides which interface is to be used with a specific connection.

Fig. 5. Algorithm to optimizing the interface selection mechanism.
2.7.1 Network model and optimization
Network model topology to be optimized must contain futures that addressed parameter
such as assurance of service delivery and security. Since Wimax is a Flexible Access Point
System that delivers on the promise of personal broadband and rich service delivery. Paired
with a converged IP core and communicating with feature-rich, multimodal devices
combining one network, one service delivery platform and seamless experience that is

Optimization of IPV
6

11
transparent to the end users. The below fig 6 consider the network model topology for
optimization of IPv6 over wimax.

Fig. 6. Network Model Topology For Optimization of IPv6 over WIMAX
From the Network model the wimax BS1, wimax BS2,AR was coined from the architectural
specification that depict the concept of Wimax deployment. The access Service Network
(ASN) mainly was used for regrouping of BS and AR.The connectivity service Network
(CSN) offers connectivity to the internet. To optimize using policy based routing protocol.
The link layer information,IP layer information,Network originated information are
initialized.
2.8 Standard for WiMAX architecture
WiMAX is a term coined to describe standard, interoperable implementations of IEEE 802.16
wireless networks, similar to the way the term Wi-Fi is used for interoperable
implementations of the IEEE 802.11 Wireless LAN standard. However, WiMAX is very
different from Wi-Fi in the way it works. The architecture defines how a WiMAX network
connects with other networks, and a variety of other aspects of operating such a network,
including address allocation, authentication. An overview of this specification for different
architectures in order to deploy IPv6 over WiMAX is depicted below in Fig 7 by WiMAX
forum .

Automation

12

Fig. 7. Architectural Specification for Deployment of IPv6 over WiMAX.
In the proposed network model and optimization the reminder should note that :
Regrouping of BS and AR into one entity is named the Access Service Network (ASN) for
WiMAX. It has a complete set of functions such as AAA (Authentication, Authorization,
Accounting), Mobile IP Foreign agent, Paging controller, and Location Register to provide
radio access to a WiMAX Subscriber. The Connectivity Service Network (CSN) offers
connectivity to the internet. In the ASN, the BS and AR (or ASN-Gateway) are connected by
using either a Switch or Router. The ASN has to support Bridging between all its R1
interfaces and the interfaces towards the network side; forward all packets received from
any R1 to a network side port and flood any packet received from a network side port
destined for a MAC broadcast or multicast address to all its R1 interfaces. The SS are now
considered as mobile (MS), the support for dormant mode is now critical and a necessary
feature. Paging capability and optimizations are possible for paging an MS are neither
enhanced nor handicapped by the link model itself. However, the multicast capability
within a link may cause for an MS to wake up for an unwanted packet.
The solution can consist of filtering the multicast packets and delivering the packets to MS
that are listening for particular multicast packets. To deploy IPv6 over IEEE 802.16, SS enters
the networks and auto-configure its IPv6 address. In IEEE 802.16, when a SS enters the
networks it gets three connection identifier (CID) connections to set-up its global
configuration. The first CID is usually used for transferring short, sensitive MAC and radio
link control messages, like those relating to the choice of the physical modulations. The
second CID is more tolerant connection, it is considered as the primary management
connection. With this connection, authentication and connection set-up messages are
exchanged between SS and BS. Finally, the third CID is dedicated to the secondary
management connection.

Optimization of IPV
6

13
2.8.1 WiMAX security
WiMAX Security is a broad and complex subject most especially in wireless communication
networks. The subject mechanism of Wimax Technology must meet the requirement design
for security architecture in Wimax. Each layer handles different aspects of security, though
in some cases, there may be redundant mechanisms. As a general principle of security, it is
considered good to have more than one mechanism providing protection so that security is
not compromised in case one of the mechanisms is broken. Security goals for wireless
networks can be summarized as follows. Privacy or confidentiality is fundamental for
secure communication, which provides resistance to interception and eavesdropping.
Message authentication provides integrity of the message and sender authentication,
corresponding to the security attacks of message modification and impersonation. Anti-
replay detects and disregards any message that is a replay of a previous message. Non-
repudiation is against denial and fabrication. Access control prevents unauthorized access.
Availability ensures that the resources or communications are not prevented from access by
DoS attack. The 802.16 standard specifies a security sub layer at the bottom of the MAC
layer. This security sub layer provides SS with privacy and protects BS from service
hijacking. There are two component protocols in the security sub layer: an encapsulation
protocol for encrypting packet data across the fixed BWA, and a Privacy and Key
Management Protocol (PKM) providing the secure distribution of keying data from BS to SS
as well as enabling BS to enforce conditional access to network services. The model below
was adapted based on security in wimax. This chapter is still investigating the protocol in
the sub layer that can mitigate encapsulation of packet data.

Fig. 8. Security model

Automation

14
2.8.2 Authentication in WiMAX
Basically in WIMAX/802.16 there are three options for authentication: device list based,
X.509 based or EAP-based. If device list-based authentication is used only, then the
likelihood of a BS or MA masquerading attack is likely. Impact can be high. The risk is
therefore high and there is a need for countermeasures. If X.509-based authentication is
used, the likelihood for a user (a MS) to be the victim of BS masquerading is possible
because of the asymmetry of the mechanism.
There are specific techniques that identify theft and BS attack. Identity theft consists of
reprogramming a device with the hardware address of another device. This is a well know
problem in unlicensed services such as WiFi/802.11, but in cellular networks because it had
been made illegal and more difficult to execute with subscriber ID module (SIM) cards. The
exact method of attack depends on the type of networks.
The proposed policy based routing protocol for optimization evaluates connection
association and transport information against the actions in policies, using the following
principles:
- The mechanism must allow dynamic management of policies such as add, update and
remove operations.
- The evaluation of policies should always result in exactly one interface for any traffic
flow or connection. This is reached by having a priority order for actions.
- All attribute-value pairs in an action must match for a traffic flow or connection for the
action to take place.
- The mechanism selects an interface based on the priority order of interfaces in an
action.
- The mechanism uses default actions which match to all flows and connections if no
other matching action is found. The mechanism should support distributed policy
management and allow explicit definition of priorities. The below table consider our
model for optimization during Authentication of WIMAX/802.16.

Threat DoS on BS or MS
Kind Mechanism
Device Device List : RSA / X.509 Certificate
User Level EAP + EAP TLS (X.509) or EAP SIM (subscriber ID module)
Data Traffic AES-CCM CBC-MAC
Physical Layer
Header
None
MAC Layer Header None
Management
messages
SHA 1 Based MAC
AES Based MAC
Table 2. Authentication in WIMAX 802.16
The intended proposed concept is to mitigate and prevent Dos on the BS or MS by
introducing Policy Repository. In a WiMax/802.16 network, it is more difficult to do these
because of the time division multiple access model. The attacker must transmit while the

Optimization of IPV
6

15
legitimate BS is transmitting. The signal of the attacker, however, must arrive at the targeted
receiver MS(s) with more strength and must put the signal of the legitimate BS in the
background, relatively speaking. Again, the attacker has to capture the identity of a
legitimate BS and to build a message using that identity. The attacker has to wait until a
time slot allocated to the legitimate BS starts. The attacker must transmit while achieving a
receive signal strength. The receiver MSs reduce their gain and decode the signal of the
attacker instead of the one from the legitimate BS.
2.8.3 Key management in Wimax for 5G technology
5G technology has a bright future because it can handle best technologies. The primary
concern that should be focused on in 5G is the automated and optimization capability to
support software. Although the issue of handover is being address since the Router and
switches in this Network has high connectivity capability. Security is under studied in this
regard. The knowledge base for the key management for 5G technology centered on
physical layer, privacy sub layer threat, mutual Authentication, Threat of identity theft,
water Torture and Black hat threat in wimax technology. The protocol used is not rolled out
because some flexible framework created by the IETF (RFC 3748), allows arbitrary and
complicated authentication protocols to be exchanged between the supplicant and the
authentication server. Extensible Authentication Protocol (EAP) is a simple encapsulation
that can run over not only PPP but also any link, including the WiMAX link. A number of
Extensible Authentication Protocol (EAP) methods have already been defined to support
authentication, using a variety of credentials, such as passwords, certificates, tokens, and
smart cards. For example, Protected Extensible Authentication Protocol (PEAP) defines a
password- based EAP method, EAP-transport-layer security (EAP-TLS) defines a certificate-
based Extensible Authentication Protocol (EAP) method, and EAP-SIM (subscriber identity
module) defines a SIM cardbased EAP method. EAP-TLS provides strong mutual
authentication, since it relies on certificates on both the network and the subscriber
terminal.(Chong li).
3. Conclusion
Considering the complex issues and areas that have been addressed in this book chapter.
The main focus of the chapter is how to provide techniques on automation and optimization
using Algorithm based on policy based routing protocol. However, the various issues on
this subject matter have been addressed. Analysis of micro mobility via Macro mobility
based on comparative investigation and requirement was advanced. Furthermore the key
optimization and data communication of IPv6 over wimax deployment must consider:
session preservation and interface selection mechanism. The account of policy based routing
protocol must provide: link layer information, IP layer information and network originated
information. Our network model topology for optimization evaluates connection association
and transport information against the actions in the policies using the aforementioned
Algorithm. The remainder of this report should note that there are limitations in wimax
deployment such as: low bit rate, speed of connectivity and sharing of bandwidth.
Finally, the chapter provides the basic Algorithm for optimization of IPv6 over wimax
deployment using policy based routing protocol.

Automation

16
4. References
[1] Mobile IPv6 Fast Handovers over IEEE 802.16e Networks H. Jang, J. Jee, Y. Han, S. Park,
J. Cha, June 2008
[2] IPv6 Deployment Scenarios in 802.16 Networks M-K. Shin, Ed., Y-H. Han, S-E. Kim, D.
Premec, May 2008.
[3] Transmission of IPv6 via the IPv6 Convergence Sublayer over IEEE 802.16 Networks B.
Patil, F. Xia, B. Sarikaya, JH. Choi, S. Madanapalli, February 2008.
[4] Mobility Support in IPv6 D. Johnson, C. Perkins, J. Arkko,RFC 3775. June 2004.
[5] Analysis of IPv6 Link Models for 802.16 Based Networks S. Madanapalli, Ed. ,August
2007.
[6] Threats Relating to IPv6 Multihoming Solutions E. Nordmark, T. Li, October 2005.
[7] Koodli, Fast Handovers for Mobile IPv6, IETF RFC- 4068, July 2005.
2
Towards Semantic
Interoperability in Information Technology:
On the Advances in Automation
Gleison Baico, Anilton Salles Garcia and Giancarlo Guizzardi
Federal University of Esprito Santo
Brazil
1. Introduction
Automation has, over the years, assumed a key role in various segments of business in
particular and, consequently, society in general. Derived from the use of technology,
automation can reduce the effort spent on manual work and the realization of activities that
are beyond human capabilities, such as speed, strength and precision. From traditional
computing systems to modern advances in information technology (IT), automation has
evolved significantly. At every moment a new technology creates different perspectives,
enabling organizations to offer innovative, low cost or custom-made services. For example,
advents such as artificial intelligence have enabled the design of intelligent systems capable
of performing not only predetermined activities, but also ones involving knowledge
acquisition. On the other hand, customer demand has also evolved, requiring higher quality,
lower cost or ease of use. In this scenario, advances in automation can provide innovative
automated services as well as supporting market competition in an effective and efficient
way. Considering the growing dependence of automation on information technology, it is
observed that advances in automation require advances in IT.
As an attempt to allow that IT delivers value to business and operates aligned with the
achievement of organizational goals, IT management has evolved to include IT service
management and governance, as can be observed by the widespread adoption of innovative
best practices libraries such as ITIL (ITIL, 2007) and standards such as ISO/IEC 20000
(ISO/IEC, 2005). Nonetheless, as pointed out by Pavlou & Pras (2008), the challenges arising
from the efforts of integration between business and IT remain topic of various studies. IT
management, discipline responsible for establishing the methods and practices in order to
support the IT operation, encompasses a set of interrelated processes to achieve this goal.
Among them, configuration management plays a key role by providing accurate IT
information to all those involved in management. As a consequence, semantic
interoperability in the domain of configuration management has been considered to be one
of the main research challenges in IT service and network management (Pras et al., 2007).
Besides this, Moura et al. (2007) highlight the contributions that computer systems can play
in terms of process automation, especially when they come to providing intelligent
solutions, fomenting self-management. However, as they emphasize, as an emerging
paradigm, this initiative is still a research challenge.

Automation 18
According to Pras et al. (2007), the use of ontologies has been indicated as state of the art for
addressing semantic interoperability, since they express the meaning of domain concepts
and relations in a clear and explicit way. Moreover, they can be implemented, thereby
enabling process automation. In particular, ontologies allow the development of intelligent
systems (Guizzardi, 2005). As a result, they foment such initiatives as self-management.
Besides this, it is important to note that ontologies can promote the alignment between
business and IT, since they maximize the comprehension regarding the domain
conceptualization for humans and computer systems. However, although there are many
works advocating their use, there is not one on IT service configuration management that
can be considered as a de facto standard by the international community (Pras et al., 2007).
As discussed in Falbo (1998), the development of ontologies is a complex activity and, as a
result, to build high quality ontologies it is necessary to adopt an engineering approach
which implies the use of appropriate methods and tools. According to Guizzardi (2005,
2007), ontology engineering should include phases of conceptual modeling, design and
implementation. In a conceptual modeling phase, an ontology should strive for
expressiveness, clarity and truthfulness in representing the domain conceptualization. These
characteristics are fundamental quality attributes of a conceptual model responsible for its
effectiveness as a reference framework for semantic interoperability. The same conceptual
model can give rise to different ontology implementations in different languages, such as
OWL and RDF, in order to satisfy different computational requirements. Thus, each phase
shall produce different artifacts with different objectives and, as a consequence, requires the
use of languages which are appropriate to the development of artifacts that adequately meet
their goals. As demonstrated by Guizzardi (2006), languages like OWL and RDF are focused
on computer-oriented concerns and, for this reason, improper for the conceptual modeling
phase. Philosophically well-founded languages are, conversely, committed to expressivity,
conceptual clarity as well as domain appropriateness and so suitable for this phase.
Considering these factors, Baico et al. (2009) present a conceptual model of the IT service
configuration management domain based on foundational ontology. Subsequently, Baico
& Garcia (2010) present an implementation of this ontology, describing how a conceptual
model can give rise to various implementation models in order to satisfy different
computational requirements. The objective of this chapter is to provide further details about
this IT service configuration management ontology, describing the main ontological
distinctions provided by the use of a foundational ontology and how these distinctions are
important to the design of models aligned with the universe of discourse, maximizing the
expressiveness, clarity and truthfulness of the model and consequently the semantic
interoperability between the involved entities. Moreover, this chapter demonstrates how to
apply the entire adopted approach, including how to generate different implementations
when compared with previous ones. This attests the employed approach, makes it more
tangible and enables to validate the developed models as well as demonstrating their
contributions in terms of activity automation.
It is important to note that the approach used in this work is not limited to the domain of IT
service configuration management. In contrast, it has been successfully employed in many
fields, such as oil and gas (Guizzardi et al., 2009) as well as medicine (Gonalves et al., 2011).
In fact, the development of a computer system involves the use of languages able to
adequately represent the universe of discourse. According to Guizzardi (2005), an imprecise
representation of state of affairs can lead to a false impression of interoperability, i.e.

Towards Semantic Interoperability in Information Technology: On the Advances in Automation 19
although two or more systems seem to have a shared view of reality, the portions of reality
that each of them aims to represent are not compatible. As an alternative, ontologies have
been suggested as the best way to address semantic interoperability. Therefore, in
particular, the ontological evaluation realized in this work contributes to the IT service
configuration management domain, subsidizing solutions in order to address key research
challenges in IT management. In general, this chapter contributes to promote the benefits of
the employed approach towards semantic interoperability in IT in various areas of interest,
maximizing the advances in automation. Such a contribution is motivated in considering
that although recent research initiatives such as that of Guizzardi (2006) have elaborated on
why domain ontologies must be represented with the support of a foundational theory and,
even though there are many initiatives in which this approach has been successfully
applied, it has not yet been broadly adopted. As reported by Jones et al. (1998), most existing
methodologies do not emphasize this aspect or simply ignore it completely, mainly because
it is a novel approach.
In this sense, this chapter is structured as follows: Section 2 briefly introduces the IT service
configuration management domain. Section 3 discusses the approach to ontology
development used in this work. Section 4 presents the conceptual model of the IT service
configuration management domain. Section 5 shows an implementation model of the
conceptual model presented in Section 4 and finally Section 6 relates some conclusions and
future works.
2. IT service configuration management
The business of an organization requires quality IT services economically provided.
According to ITIL, to be efficient and effective, organizations need to manage their IT
infrastructure and services. Configuration management provides a logical model of an
infrastructure or service by identifying, controlling, maintaining and verifying the versions
of configuration items in existence. The logical model of IT service configuration
management is a single common representation used by all parts of IT service management
and also by other parties, such as human resources, finance, suppliers and customers. A
configuration item, in turn, is an infrastructure component or an item that is or will be under
the control of configuration management (ITIL, 2007; ISO/IEC, 2005). For innovative IT
management approaches such as ITIL and ISO/IEC 20000, configuration items are viewed
not only as individual resources but as a chain of related and interconnected resources
compounding services. Thus, just as important as controlling each item is managing how
they relate to each other. These relationships form the basis for activities such as impact
assessment.
According to ITIL and ISO/IEC 20000, a configuration item and the related configuration
information may contain different levels of detail. Examples include an overview of all
services or a detailed view of each component of a service. Thus, a configuration item may
differ in complexity, size and type, ranging from a service, including all hardware, software
and associated documentation, to a single software module or hardware component.
Configuration items may be grouped and managed together, e.g. a set of components may
be grouped into a release. Furthermore, configuration items should be selected using
established selection criteria, grouped, classified and identified in such a way that they are
manageable throughout the service lifecycle.

Automation 20
As with any process, IT service configuration management is associated with goals that in
its case include: (i) supporting, effectively and efficiently, all other IT service management
processes by providing configuration information in a clear, precise and unambiguous way;
(ii) supporting the business goals and control requirements; (iii) optimizing IT infrastructure
settings, capabilities and resources; (iv) subsidizing the dynamism imposed on IT by
promoting rapid responses to necessary changes and by minimizing the impact of changes
in the operational environment. To achieve these objectives, configuration management
should, in summary, define and control the IT components and maintain the configuration
information accurately. Based on best practices libraries such as ITIL and standards such as
ISO/IEC 20000 for IT service management, the activities of an IT service configuration
management process may be summarized as: (i) planning, in order to plan and define the
purpose, scope, objectives, policies and procedures as well as the organizational and
technical context for configuration management; (ii) identification, aiming to select and
identify the configuration structures for all the items (including their owner,
interrelationships and configuration documentation), allocate identifiers and version
numbers for them and finally label each item and enter it on the configuration management
database (CMDB); (iii) control, in order to ensure that only authorized and identified items
are accepted and recorded, from receipt to disposal, ensuring that no item is added,
modified, replaced or removed without appropriate controlling documentation; (iv) status
accounting and reporting, which reports all current and historical data concerned with each
item throughout its life cycle; (v) verification and audit, which comprises a series of reviews
and audits that verify the physical existence of items and check that they are correctly
recorded in the CMDB.
As an attempt to promote efficiency and effectiveness, IT management has evolved to
include IT service management and governance, which aims to ensure that IT delivers value
to business and is aligned with the achievement of organizational goals. As emphasized by
Sall (2004), in this context, IT processes are fully integrated into business processes. Thus,
one of the main aspects to be considered is the impact of IT on business processes and vice
versa (Moura et al., 2008). As a consequence, IT management processes should be able to
manage the entire chain, i.e. from IT to business. For this reason, the search for the
effectiveness of such paradigms towards business-driven IT management has been the topic
of several studies in network and service management (Pavlou & Pras, 2008). According to
Moura et al. (2007), one of the main challenges is to achieve the integration between these
two domains. Configuration management, in this case, should be able to respond in a clear,
precise and unambiguous manner to the following question: what are the business
processes and how are they related to IT services and components (ITIL, 2007)?
Furthermore, as cited by ITIL, due to the scope and complexity of configuration
management, keeping its information is a strenuous activity. In this sense, research
initiatives consider automation to be a good potential alternative. In fact, the automation of
management processes has been recognized as one of the success factors to achieve a
business-driven IT management, especially when considering intelligent solutions
promoting self-management (Moura et al., 2007). Besides its scope and complexity,
configuration management is also closely related to all other management processes. In IT
service management and governance, this close relationship includes the interaction among
the main entities involved in this context, such as: (i) business, (ii) people, (iii) processes, (iv)
tools and (v) technologies (ITIL, 2007). Thus, semantic interoperability among such entities
has been characterized as one of the main research challenges, not only in terms of the

configuration management process, but in the whole chain of processes that comprise the
discipline of network and service management (Pras et al., 2007).
According to Pras et al. (2007), the use of semantic models, in particular, the use of
ontologies, has been regarded as the best way with respect to initiatives for addressing
issues related to semantic interoperability problems in network and service management.
According to these authors, ontologies make the meaning of the domain concepts such as
IT management, as well as the relationships between them, explicit. Additionally, this
meaning can be defined in a machine-readable format, making the knowledge shared
between humans and computer systems, enabling process automation, as outlined by these
authors. From this point of view, it is worth mentioning that ontologies are considered as
potential tools for the construction of knowledge in intelligent systems (Guizzardi, 2005).
Thus, they allow the design of intelligent and above all interoperable solutions, fomenting
initiatives as self-management. Finally, it is important to note that ontologies can promote
the alignment between business and IT when applied in the context of IT service
management and governance since they maximize the expressiveness, clarity and
truthfulness of the domain conceptualization for humans and computer systems. However,
Pras et al. (2007) point out that despite the efforts of research initiatives, there are still many
gaps to be addressed.
Several studies claim that the use of ontologies is a promising means of achieving
interoperability among different management domains. However, an ontology-based model
and formalization of IT service configuration management remains a research challenge.
Regarding limitations, it should be mentioned that ontologies are still under development in
the management domain. In fact, the technology is not yet mature and there is not an
ontology that can be considered as a de facto standard by the international community (Pras
et al., 2007). In general, the research initiatives have not employed a systematic approach in
the development of ontologies. According to Falbo (1998), the absence of a systematic
approach, with a lack of attention to appropriate methods, techniques and tools, makes the
development of ontologies more of an art rather than an engineering activity. According to
Guizzardi (2005, 2007), to meet the different uses and purposes intended for the ontologies,
ontology engineering should include phases of conceptual modeling, design and
implementation. Each phase should have its specific objectives and thus would require the
use of appropriate languages in order to achieve these goals. However, in most cases, such
research initiatives are engaged with the use of technologies and tools such as Protg and
OWL. Sometimes these technologies and tools are used in the conceptual modeling phase,
which can result in various problems relating to semantic interoperability, as shown in
Guizzardi (2006). At other times, however, they are employed in the implementation phase,
ignoring previous phases such as conceptual modeling and design. As a result, such
initiatives are obliged to rely on models of low expressivity. Moreover, in most cases, such
initiatives propose the use of these technologies and tools for the formalization of network
management data models, such as MIB, PIB and the CIM schema. It is noteworthy that data
models are closely related to the underlying protocols used to transport the management
information and the particular implementation in use. In contrast, information models work
at a conceptual level and they are intended to be independent of any particular
implementation or management protocol. Working at a higher level, information models
usually provide more expressiveness (Pras et al., 2007). Following this approach, Lopez de
Vergara et al. (2004) propose an integration of the concepts that currently belong to different

Automation 22
network management data models (e.g. MIB, PIB and the CIM schema) in a single model,
formalized by ontology languages such as OWL. In an even more specific scenario, i.e. with
no intention to unify the various models but only to formalize a particular model, Majewski
et al. (2007) suggest the formalization of the CIM schema through ontology languages such
as OWL. Similarly, while differentiating the type of data model, Santos (2007) presents an
ontology-based network configuration management system. In his work, the proposed
ontology was developed according to the MIB data model concepts. As MIB is limited in
describing a single system, a view of the entire infrastructure, including the relationships
between its components, is not supported by the model. In practice, this gap is often filled
by functionalities provided by SNMP-based network management tools which, for example,
support the visualization of network topologies (Brenner et al., 2006). Aside from the fact
that, in general, the research initiatives are committed to the use of technologies and tools, it
is also observed that they are characterized by specific purposes in relation to peculiar
applications in information systems that restrict their conceptualizations. In Xu and Xiao
(2006), an ontology-based configuration management model for IP network devices is
presented, aiming at the use of ontology for the automation of this process. In Calvi (2007),
the author presents a modeling of the IT service configuration management described by the
ITIL library based on a foundational ontology. The concepts presented and modeled in his
work cover a specific need regarding the demonstration of the use of ITIL processes for a
context-aware service platform. Finally, there are approaches that seek to establish semantic
interoperability among existing ontologies by means of ontological mapping techniques, as
evidenced in Wong et al. (2005). However, it is not within the scope of such approaches to
develop an ontology but rather to integrate existing ones.
Therefore, in considering the main challenges as well as the solutions which are considered to
be state of the art and in analyzing the surveyed works, it is observed that there are gaps to
be filled, as highlighted by Pras et al. (2007). In summary, factors such as the adoption of ad
hoc approaches, the use of inappropriate references about the domain, the intention of specific
purposes and, naturally, the integration of existing ontologies, all result in gaps. As a
consequence, such factors do not promote the conception of an ontology able to serve as a
reference framework for semantic interoperability concerning the configuration management
domain in the context of IT service management and governance. This scenario demonstrates
the necessity of a modeling which considers the gaps and, therefore, promotes solutions in
line with those suggestions regarded as state of the art for the research challenges discussed
earlier in this chapter. In particular, it demonstrates the necessity of an appropriate approach
for the construction of ontologies as a subsidy for such modeling. In this sense, the next
section of this chapter presents an approach for ontology development.
3. Ontology engineering
In philosophy, ontology is a mature discipline that has been systematically developed at
least since Aristotle. As a function of the important role played by them as a conceptual tool,
their application to computing has become increasingly well-known (Guizzardi et al., 2008).
According to Smith and Welty (2001), historically there are three main areas responsible for
creating the demand for the use of ontologies in computer science, namely: (i) database and
information systems; (ii) software engineering (in particular, domain engineering); (iii)
artificial intelligence. Additionally, Guizzardi (2005) includes the semantic web, due to the
important role played by this area in the current popularization of the term.

According Guizzardi et al. (2008), an important point to be emphasized is the difference
between the senses of the term ontology when used in computer science. In conceptual
modeling, the term has been used as its definition in philosophy, i.e. as a philosophically
well-founded domain-independent system of formal categories that can be used to articulate
domain-specific models of reality. On the other hand, in most other areas of computer
science, such as artificial intelligence and semantic web, the term ontology is generally used
as: (i) an engineering artifact designed for a specific purpose without giving much
importance to foundational issues; (ii) a representation of a particular domain (e.g. law,
medicine) expressed in some language for knowledge representation (e.g. RDF, OWL).
From this point of view, the development of ontologies should consider the various uses
and, consequently, the different purposes attributed to ontologies as well as any existing
interrelationship in order to enable the construction of models that satisfactorily meet their
respective goals. However, despite the growing use of ontologies and their importance in
computing, the employed development approaches have generally not considered these
factors, resulting in inadequate models for the intended purpose. In considering such
distinctions Guizzardi (2005, 2007) elaborates and discusses a number of questions in order
to elucidate such divergences and thus provide a structured way with respect to the use of
ontologies. In addition, Guizzardi and Halpin (2008) describe that the interest in proposals
for foundations in the construction of ontologies has been the topic of several studies and
they report some innovative and high quality research contributions. It is based on such
questions that are elaborated the further discussions contained in this section and thus the
approach used for the construction of the ontological models proposed in this work.
As discussed in Falbo (1998), the development of ontologies is a complex activity and,
hence, in order to build high quality ontologies, able to adequately meet their various uses
and purposes, it is necessary to adopt an engineering approach. Thus, unlike the various ad
hoc approaches, the construction of ontologies must use appropriate methods and tools.
Falbo (2004) proposes a method for building ontologies called SABiO (Systematic Approach
for Building Ontologies). This method proposes an life cycle by prescribing an iterative
process that comprises the following activities: (i) purpose identification and requirements
specification, which aims to clearly identify the ontologys purpose and its intended use by
means of competence questions; (ii) ontology capture, viewing to capture relevant concepts
existing within the universe of discourse as well as their relationships, properties and
constraints, based on the competence questions; (iii) ontology formalization, which is
responsible for explicitly representing the captured conceptualization by means of a formal
language, such as the definition of formal axioms using first-order logic; (iv) integration
with existing ontologies, in order to search for other ones with the purpose of reuse and
integration; (v) ontology evaluation, which aims to identify inconsistencies as well as
verifying truthfulness in line with the ontologys purpose and requirements; (vi) ontology
documentation. Noticeably, the competence questions form an important concept within
SABiO, i.e. the questions the ontology should be able to answer. They provide a mechanism
for defining the scope and purpose of the ontology, guiding its capture, formalization and
evaluation - regarding this last aspect, especially with respect to the completeness of the
ontology.
The elements that constitute the relevant concepts of a given domain, understood as domain
conceptualization, are used to articulate abstractions of certain states of affairs in reality,
denominated as domain abstraction. As an example, consider the domain of product sales.

Automation 24
A conceptualization of this domain can be constructed by considering concepts such as, inter
alia: (i) customer, (ii) provider, (iii) product, (iv) is produced by, (v) is sold to. By means of
these concepts, it is possible to articulate domain abstractions of certain facts extant in
reality, such as: (i) a product is produced by the provider and sold to the customer. It is
important to highlight that conceptualizations and abstractions are abstract entities which
only exist within the mind of a user or a community of users of a language. Therefore, in
order to be documented, communicated and analyzed, they must be captured, i.e.
represented in terms of some concrete artifact. This implies that a language is necessary for
representing them in a concise, complete and unambiguous way (Guizzardi, 2005). Figure 1-
a presents Ullmanns triangle (Ullmann, 1972), which illustrates the relation between a
language, a conceptualization and the part of reality that this conceptualization abstracts.
The relation represents concerns the definition of language semantics. In other words, this
relation implies that the concepts are represented by the symbols of language. The relation
abstracts, in turn, denotes the abstraction of certain states of affairs within the reality that
a given conceptualization articulates. The dotted line between language and reality
highlights the fact that the relation between language and reality is always intermediated by
a certain conceptualization. This relation is elaborated in Figure 1-b, which depicts the
distinction between an abstraction and its representation, as well as their relationships with
the conceptualization and representation language. The representation of a domain
abstraction in terms of a representation language is called model specification (or simply
model, specification or representation) and the language used for its creation is called
modeling language (or specification language).

Fig. 1. Ullmanns triangle and relations between conceptualization, abstraction, modelling
language and model, according to Guizzardi (2005).
Thus, in addition to the adoption of appropriate methods, able to systematically lead the
development process, ontology engineering as an engineering process aims at the use of
tools, which should be employed in accordance with the purpose of the product that is
being designed. In terms of an ontology development process such tools include modeling
languages or even ontology representation languages. According to Guizzardi (2005), one of
the main success factors regarding the use of a modeling language is its ability to provide its
users with a set of modeling primitives that can directly express the domain
conceptualization. According to the author, a modeling language is used to represent a

conceptualization by compounding a model that represents an abstraction, which is an
instance of this conceptualization. Therefore, in order for the model to faithfully represent
an abstraction, the modeling primitives of the language used to produce the model must
accurately represent the domain conceptualization used to articulate the abstraction
represented by the model. According to Guizzardi (2005), if a conceptual modeling language
is imprecise and coarse in the description of a given domain, then there can be specifications
of the language which, although grammatically valid, do not represent admissible state of
affairs. Figure 2-a illustrates this situation. The author also points out that a precise
representation of a given conceptualization becomes even more critical when it is necessary
to integrate different independently developed models (or systems based on these models).
As an example, he mentions a situation in which it is necessary to have the interaction
between two independently developed systems which commit to two different
conceptualizations. Accordingly, in order for these two systems to function properly
together, it is necessary to ensure that they ascribe compatible meanings to the real world
entities of their shared subject domain. In particular, it is desirable to reinforce that they
have compatible sets of admissible situations whose union (in the ideal case) equals the
admissible states of affairs delimited by the conceptualization of their shared subject
domain. The ability of entities (in this case, systems) to interoperate (operate together) while
having compatible real-world semantics is known as semantic interoperability (Vermeer,
1997). Figure 2-b illustrates this scenario.

Fig. 2. Consequences of an imprecise and coarse modelling language (Guizzardi, 2005).
In Figure 2-b, C
A
and C
B
represent the conceptualizations of the domains of systems A and
B, respectively. As illustrated in this figure, these conceptualizations are not compatible.
However, because these systems are based on poor representations of these
conceptualizations, their sets of considered possible situations overlap. As a result, systems
A and B agree exactly on situations that are neither admitted by C
A
nor by C
B
. In summary,
although these systems appear to have a shared view of reality, the portions of reality that
each of them aims to represent are not compatible. Therefore, the more it is known about a
given domain and the more precisely it is represented, the bigger the chance of obtaining
interpretations that are consistent with the reality of that domain and, therefore, of
achieving semantic interoperability between the entities involved in these interpretations.
Thus, Guizzardi (2005) concludes that, on the one hand, a modeling language should be
sufficiently expressive to adequately characterize the conceptualization of the domain and,
on the other hand, the semantics of the produced specifications should be clear, allowing
users to recognize what language constructs mean in terms of domain concepts. Moreover,
the specification produced by means of the language should facilitate the user in
understanding and reasoning about the represented state of affairs.

Automation 26
In view of the different purposes, Guizzardi (2005, 2007) highlights that ontology
engineering, analogous to software engineering and information systems, must include
phases of conceptual modeling, design and implementation. Each phase has its specific
objectives and thus requires different types of methods and tools to meet its particular
characteristics. As mentioned, during a conceptual modeling phase, an ontology must strive
for expressivity, clarity and truthfulness in representing the domain conceptualization.
Therefore, the conceptual modeling phase requires specialized languages so as to create
ontologies that approximate as closely as possible to the ideal representation of the domain.
The same conceptual model can give rise to different implementation models in different
languages, such as OWL and RDF, in order to satisfy different non-functional requirements,
such as decidability and completeness. The section delimited as Level in Figure 3 illustrates
this approach based on relations between conceptualization, abstraction, modeling language
and model, shown in Figure 1-b. According to Guizzardi (2006), semantic web languages
such as OWL and RDF are focused on computation-oriented concerns and are therefore
inadequate for the conceptual modeling phase. Philosophically well-founded languages, on
the other hand, are engaged in expressivity, conceptual clarity and domain appropriateness
and are therefore suitable for this phase. To support his assertion, Guizzardi (2006) presents
several problems of semantic interoperability from the use of semantic web languages in the
representation of the domain and demonstrates how philosophically well-founded
languages are able to address these problems.
As shown in Guizzardi (2005), while domain conceptualizations and, consequently,
domain ontologies are established by the consensus of a community of users with respect
to a material domain, a conceptual modeling language (which can be used to express these
domain ontologies) must be rooted in a domain independent system of real-world
categories, philosophically and cognitively well-founded, i.e. a foundational ontology.
Foundational ontologies aggregate contributions from areas such as descriptive
metaphysics, philosophical logic, cognitive science and linguistics. The theories inherent to
these areas are called (meta-) conceptualizations and describe knowledge about reality in a
way which is independent of language and particular states of affairs. A foundational
ontology, in turn, is the representation of these theories in a concrete artifact. Thus,
foundational ontologies, in the philosophical sense, can be used to provide real-world
semantics for modeling languages as well as to constrain the possible interpretations of
their modeling primitives, increasing the clarity of interpretation and, consequently,
reducing ambiguities (which are key success factors in achieving semantic
interoperability). Accordingly, it is possible to build domain ontologies by means of
conceptual modeling languages based on foundational ontologies. In this sense, the Meta-
level section in Figure 3, in addition to the Level section, represents the approach proposed
by Guizzardi (2005, 2007).
An example of a foundational ontology is UFO (Unified Foundational Ontology). UFO was
initially proposed in Guizzardi and Wagner (2004) and its most recent version is presented
by Guizzardi et al. (2008). It is organized in three incrementally layered compliance sets: (i)
UFO-A, which is essentially the UFOs core, defining terms related to endurants (objects,
their properties etc); (ii) UFO-B, which defines as an increment to UFO-A terms related to
perdurants (events etc); (iii) UFO-C, which defines as an increment to UFO-B terms
explicitly related to the spheres of social entities.


Fig. 3. Ontology engineering approach proposed by Guizzardi (2005, 2007).
As a conclusion, the ontology engineering approach described in this section considers the
distinctions of the term ontology as well as their interrelationships, thereby establishing the
phases, their respective objectives, as well as the methods and tools appropriate for the
characteristic of each phase, allowing thus the construction of models capable of meeting the
various purposes intended for them. Therefore, based on the study about IT service
configuration management, as well as the study about ontology development, it is possible
to construct an ontology of this domain as the objective of this chapter.
4. Conceptual model of the IT service configuration management domain
In considering the main research challenges, as well as the initiatives for solutions which are
regarded state of the art, which properly lead the identification of gaps, as much as the
appropriated approach to fulfil them, this section presents the conceptual models proposed
in this work. On the basis of Figure 3, which shows the ontology development approach
adopted in this work, the conceptual model proposed in this section concerns the domain
ontology, whose conceptualization in discussion is the IT service configuration
management. Regarding the foundational ontology, this work uses UFO, which represents
the meta-conceptualization responsible for promoting the philosophical base of the work.
As discussed in Section 2, configuration management is responsible for maintaining
information about configuration items and providing them to all the other management
processes. In the context of IT service management and governance, configuration
management must be able to answer questions such as: what are the business processes and
how do they relate to the IT services and components? Based on this question and given that
the main goal of this ontology is to describe a theory of the domain of IT service
configuration management independent of specific applications, the defined competency
questions reflect this intention. In this case, they lead to a mapping between IT and business
concepts, as follows:
CQ1: How do the IT services and the business processes of an organization relate?
CQ2: How do the IT services and the IT components such as hardware and software relate?

Automation 28
To answer these questions, it is necessary to address others, such as: (i) what is an IT
service? (ii) what is a business process? For this reason, Baico et al. (2009) propose an IT
service configuration management ontology, which addresses such questions in order to
provide a basis for the ontology presented in this chapter. As described through the SABiO
method, if the domain of interest is too complex, a decomposing mechanism should be used
in order to better distribute this complexity. In this case, a potentially interesting approach is
to consider sub-ontologies. Therefore, to answer the competency questions, the following
sub-ontologies were developed: (i) business process; (ii) IT service; (iii) IT component and
lastly (iv) configuration item. These sub-ontologies complement each other in constituting
the IT service configuration management ontology discussed in this work.
In terms of reusing existing ontologies, it is important to mention that besides the adopted
literature, the conceptual modeling of this section also takes into consideration the
discussions inherent to processes in general done in Falbo (1998) and Guizzardi et al.
(2008), as well as the discussions inherent to IT services done in Calvi (2007) and Costa
(2008). Still in line with the SABiO method, during the capture of the ontology the use of a
graphic representation is essential to facilitate the communication between ontology
engineers and domain experts. However, a graphical model is not enough to completely
capture an ontology. This way, axioms should be provided to reinforce the semantics of the
terms and establish the domain restrictions. Thus, the sub-ontologies developed in this
work are connected by relations between their concepts and by formal axioms. Due to
limitations of space, will be shown only those axioms also used in the next section. To
distinguish the subject domain concepts and the UFO concepts, these last ones are
presented in blank in the conceptual model that follows. Figure 4 presents part of the
proposed ontology.

Fig. 4. Part of the IT service configuration management ontology.

According to the ITIL library, an IT service is a service which is provided by an IT
organization to one or more clients. Therefore, in accordance with the UFO terminology, an
IT service is characterized as a type of plan, i.e. an intentional event. As a result, the
properties inherent to events are applied to IT services. As established in UFO, events are
possible changes from one portion of reality to another, which means they can transform
reality by altering the state of affairs from a pre-state to a post-state. Consequently they can
produce, direct or indirectly, situations that satisfy the necessary conditions for other events
to happen. On that account, events can cause other events, including then the service
executions as represented in the model by the relation causes. An IT service execution
(ITSE), in turn, denotes one or more particular actions that occur at specific time intervals,
aiming to satisfy the propositional content of a commitment. On this note, an IT service
execution is an action that instantiates a type of plan, in this case an IT service. This
distinction, derived from the foundational ontology, provides greater adequacy to the
domain, making it possible to distinguish services from their executions and allowing the
comparison, for example, between achieved and planned results.
Because it denotes one or more actions, an IT service execution can be atomic or complex. As
a complex execution, it is decomposed into other smaller service executions, termed
subservices. In this way, a subservice is a service execution that is part of a bigger service
execution, its super-service. As the properties inherent to events are applied to the IT
services, the decomposition of service executions, as any other event decomposition, is
characterized as a transitive, asymmetric and irreflexive relation. This inheritance of
properties from the foundational ontology facilitates modeling decisions and minimizes the
possibilities of incoherent descriptions of the domain.
1

According to the ITIL library, an IT service execution aims to produce resources in order to
satisfy the needs of its customers. On the conceptual model proposed in this section, the
produced resources are said to be outputs. On the other hand, an IT service execution can
consume resources, seen as raw material, to produce results. On the model, these resources
are said to be inputs. From the point of view of UFO, an artifact of a service execution is a
type of resource (UFO::Resource) which in turn is mapped to the notion of an object. As
such, the subartifact and superartifact relations are then governed by the axioms defined for
the (different types of) parthood relations between substantials, as described in Guizzardi
(2005).
According to UFO, a resource (UFO::Resource) is a role that an object plays in an event.
Thus, the artifacts of a service execution are roles played by objects in the scope of this
service instance. This being said, it is important to highlight a contribution from UFO
attributed to the model. As the notions of objects and roles are defined, it becomes possible
to represent real situations of the domain, as with those where the same object plays the role
of an output to a service execution and input to another, distinguishing only the type of
participation performed by the object and keeping its identity throughout its existence. This
is because, according to UFO, an object is a type of endurant which, in contrast to a

1
UFO makes explicit distinctions often ignored by many languages. For example, while it is possible to
consider that an event x is a part of an event z because x is a part of an event y that is a part of z, it is not
the case that the musicians hand (and so a part thereof) is a part of the band within which the musician
is a part. In the first case, there is transitivity, but in the latter this does not exist. In this sense, parthood
relations denote distinctions that should be considered.

Automation 30
perdurant (e.g. an event), is an individual that keeps its identity throughout its existence. On
the model, the participation of the output is represented through the relation produces
while the participation of the input is represented through the relation consumes.
As objects, the participation of artifacts in a service execution should correspond to the types
of participation of objects in an action, defined in UFO. In fact, being produced or consumed
does not express the exact notion on the participation of an object in a service execution.
Namely, outputs can be created or modified. Inputs, on the other hand, can be used,
modified into new products or else terminated. As such, the relation produces, as well as
the relation consumes, designates distinct notions which should be considered. Therefore,
in light of the UFO terminology and the subject domain, the relation produces denotes the
participation of creation or change, while the relation consumes indicates the participation
of usage, change or termination as defined in UFO. This demonstrates the foundation of the
domain concepts and relations in terms of philosophically well-founded concepts and
relations, making the representation of the universe of discourse even more clear, expressive
and coherent with reality.
As described by the ITIL library, an IT service is based on the use of the information
technology, which includes, among other things, IT components such as hardware and
software. Therefore, an IT service execution, as any other activity, presumes the use of
resources, in particular IT components, in order to achieve results. Essentially, under the
UFOs perspective, the resources of a service execution are types of resources
(UFO::Resource), i.e. objects participating in an action. Therefore, not only artifacts but also
resources are roles played by objects within the scope of a service execution. Again, the
distinction between objects and roles contributes to the representation of real situations in
the universe of discourse, including those where the same object is produced by a service
execution but is required by another, though remaining as the same individual. In the
model, the participation of a resource is represented by the relation requires. Once these
resources are used as support tools in a service execution, the foundation associated with
this relation leads to only one of the types of resource participation defined in UFO, i.e. the
usage participation.
In considering the definitions of artifacts and resources, especially regarding the foundation
provided by UFO, it is noted that the inputs and the resources are objects which can play the
same type of participation in a service execution, i.e. the usage participation. However,
taking a service execution as a transformation primitive, inputs indicate raw materials
which are incorporated into the product. Resources, on the other hand, refer to components
that support a service execution, but are not intended as products of this execution. Thus,
the resources employed in a service execution cannot be considered as products within the
scope of this execution. Therefore, by clearly representing the participation of inputs and
resources in a service execution, this foundation promoted the identification with regard to
the similar type of participation of these roles. As a consequence, it required the use of
domain definitions able to characterize such roles, since fundamentally they are similar.
Hence, as pointed out, this observation becomes evident, in this case, especially due to the
foundation of the domain concepts in terms of UFO. This demonstrates the contribution of a
foundational ontology, in general, and of UFO, in particular, in supporting the construction
of appropriate models regarding domain and comprehension.

As described in Costa (2008), an IT service execution is an intended and orderly execution of
one or more actions in order to satisfy the propositional content (goal) of a commitment
agreed with an agent. This ordering, in some cases, is governed according to certain
situations that culminate into the execution of the service. In this sense, the causality relation
between events leads to a dependence relation between them. This happens because, by
changing the state of affairs of the reality from one state (pre-state) to another (post-state),
events can generate (directly or indirectly) a situation which satisfies a necessary condition
for other events to occur. In summary, an event depends on another if the first is caused,
directly or indirectly, by the second. Hence, service executions can depend on events (or on
other service executions) in order to occur. It is relevant to mention that artifacts and
resources can imply a dependency relation between service executions. In fact, as described
in UFO, situations are complex entities that agglutinate other entities (including objects) and
denote the pre- and post-state of an event. So, as a type of object, artifacts and resources are
present in situations that symbolize the pre- and post-states of service executions. As an
example, a service may require resources or consume inputs that are products of other
services, characterizing thus a dependence relation between them because of their respective
objects. Besides this, it is important to highlight that if an event depends on the other then
there is a temporal relation between them that should be taken into account. Concerning
dependence between service executions, this temporal relation is represented by the pre-
service and post-service relations. As the causality relation between events, the dependence
relation between them (including the IT service executions) is transitive, asymmetric and
irreflexive. These properties are also valid for the relations pre- and post-service. Temporal
relations between events make possible to define the temporal ordering in which the events
(including the service executions) are submitted and, therefore, establish the order in which
they occur, even when there is no dependence relation defined between them. This is
because, according to UFO, events are framed into temporal intervals, from which originate
the temporal relations. Thus, the model proposed in this section is based on a framework of
concepts and relations defined in UFO which makes possible to specify the flow in which
the events are associated and, consequently, the ordering associated with the IT service
executions. In this way, this structure supports the modeling decisions at the same time as it
contributes to the creation of a more expressive, clear and truthful model with regard to the
universe of discourse.
An IT service is described by a normative description, termed IT service description. The
description of an IT service, for instance, describes the roles played by each agent in a
service execution. Agents, as well as objects, are substantial from the UFO point of view.
However, agents differ from objects because of the fact that they can possess beliefs,
desires and intentions. Intentions are characterized as desired states of affairs for which the
agent commits itself to pursuing, i.e. an internal commitment. For this reason, intentions
cause the agent to perform actions. In this sense, the participation of an agent in an action
is characterized as an action contribution. Consequently, service executions are performed
by agents. Indeed, the action contribution of an agent in a service execution is caused by a
social commitment of the agent in performing this service execution (or part thereof, the
subservice) with its consequent permissions and obligations. Therefore, the role modeling
pattern described by UFO applies to the IT service domain. In fact, as advanced in
Guizzardi (2006), as a domain independent knowledge representation language,
foundational ontologies in general and UFO in particular aim to support the construction

Automation 32
of domain-dependent models by acting as a reference framework and thus guiding
modeling decisions and allowing the creation of models that clearly and accurately
represent, as much as possible, the real situations in a universe of discourse. This pattern
has resolved various role modeling problems in the literature. Thus, the execution of a
service instance is characterized by an agent playing a role in an IT service, in this case,
acting as a service provider. It is worth noting that not all types of agents are responsible
for the service executions. According to ITIL, an IT service is one provided to one or more
customers by an IT organization. On the other hand, not all the instances of service
providers are IT organizations, but possibly another type of agent. Therefore, the execution
of an IT service occurs by means of an IT organization acting as a service provider.
Nonetheless, given that IT organizations can play other roles, such as the role of a
customer, it is not the case that all instances of IT organizations act as service providers in
every situation, but only when they perform IT services. So, the relation between service
provider and IT organization cannot be direct in any sense. On the contrary, it should be
intermediated by a role (in this case, IT Service Provider - ITSP) that aggregates the criteria
of identity of the species (in this case, IT Organization) and performs the mixed role (in this
case, Service Provider). In this way, it is defined that not all service provider is
characterized as an IT organization, which may be another type of agent. Furthermore, it is
defined that there are IT organizations which in certain circumstances are service providers
but not in others, possibly playing other roles. The responsibility of each played role is
described by the IT service description.
The execution of an IT service occurs by means of a request, which is motivated according to
the requestors needs. Therefore, a customer is a type of requestor, i.e. an agent who
requests an IT service. The process of requesting an IT service, as well as other concepts
inherent to the domain of service-level management, is discussed and modeled in Costa
(2008). In summary, the author describes that an IT service is appropriate for a certain need
when the product from its execution satisfies the requestors requirements, i.e. its needs. As
such, an IT service can achieve a need if, and only if, the post-state of an occurrence of this
service is a situation that satisfies the propositional content of the referred need, as
formalized by the axiom A1.
(A1) x,y (IT-Service(x) Need(y) can-achieve(x,y) a,b,c (ITSE(a) Situation(b)
Proposition(c) instance-of(a,x) postState(b,a) satisfies(b,c) propositional-content-
of(c,y)))
An IT service is requested by means of a document that describes the need of the requestor,
termed service-level requirement (SLR). Thus, given a SLR that describes a certain need
which can be achieved by an IT service, then this SLR is used to request this service
(COSTA, 2008). This definition is formalized by the axiom A2.
(A2) x,y,z (IT-Service(x) Need(y) SLR(z) can-achieve(x,y) describes(z,y) used-
to-ask-for(z,x))
So, through a SLR, a requestor can search for an IT service that satisfies its needs. This
allows the requestor to find out whether or not there are services that can achieve its needs
and request, in case there is, the most adequate, according to its demand. This definition is

(A3) x,y,z,w (IT-Service(x) Need(y) SLR(z) Requestor(w) can-achieve(x,y)
describes(z,y) used-to-ask-for(z,x) characterizes(y,w) asks-for(w,x))
Once the requester asks for services that can achieve its necessities and finds such services
offered by the providers, this requestor is able to request such services. According to Costa
(2008), when an IT service is requested, the requestor hires the provider responsible for this
service, as formalized by the axiom A4.
(A4) x,y,z (Requestor(x) ITSP(y) IT-Service(z) asks-for(x,z) requests(x,z)
provides(y,z) hires(x,y))
According to the ITIL library, the provision of an IT service to a requestor is mediated by an
agreement. Thus, once the requesting process has been established, there will be an
agreement that will mediate all the IT service provision, establishing the claims and
obligations related to both parts, the requester and the provider. This relation is formalized
by the axiom A5.
(A5) x,y (Requestor(x) ITSP(y) hires(x,y) z (Agreement(z) mediates(z,x)
mediates(z,y)))
With regard to UFO, an agreement is a type of social relation. A social relation is composed
of social moments, called claims and commitments. A social commitment is a type of
commitment and thus the motivating cause of an action performed by an agent. Thus, the
agreement established between the requestor and the provider will cause the execution of
the service requested by the requestor, for it is composed of social commitments inherent to
the provider.
The requesting of a service is motivated by a need of a requestor and it is fulfilled through a
service execution performed by a provider. In this context, there is a relation of dependence
between the requestor and the provider. From the UFO point of view, a dependence relation
between agents leads to a delegation relation. According to UFO, an agent a depends on an
agent b regarding a goal g if g is a goal of agent a, but a cannot achieve g and agent b can
achieve g. This matter may be the reason why agent a decides to delegate such goal
achievement to agent b. A delegation is thus associated with a dependency, but it is more
than that. As a material relation, it is founded on more that its connected elements. In this
case, the connected elements are two agents, namely, the requestor (delegator) and the
provider (delegatee), as well as a goal (delegatum), which is the delegation object that
represents the needs of the requestor. The foundation of this material relation is the social
relator (a pair of commitments and claims), i.e. the agreement established between the two
agents involved in this delegation, entitled requestor and provider. In other words, when a
requestor delegates a goal to a provider, besides the fact that the requestor depends on the
provider with regard to the goal, the provider commits itself to achieving the goal on behalf
of the requester. This commitment is established by means of an agreement. As described in
Calvi (2007), an IT service delegation (ITSD) is a delegation committed to achieve a goal
according to a specific plan, the IT service. Therefore, according to UFO, an IT service
delegation is a closed delegation, since it describes a specific plan which the provider should
adopt to achieve the goal delegated by the requestor. In summary, if there is an agreement
between the requestor and the provider which causes the execution of a service, then there
will be a service delegation associated with the agreement. In addition, this delegation will

Automation 34
be committed with the service and consequently with its execution. In this context, the
requestor assumes the role of delegator while the provider assumes the role of delegatee.
This relation is formalized by the axiom A6.
(A6) x,y,z (Requestor(x) ITSP(y) Agreement(z) mediates(z,x) mediates(z,y) w
(ITSD(w) associated-to(w,z) delegator-of(x,w) delegatee-of(y,w)))
As defined by UFO, when an agent is said to be able to achieve a certain goal it means that
such agent can achieve this goal by itself or else delegate it to another agent that would be
able to achieve it on its behalf. Thus, when a provider receives a delegation by the means of
a service-level agreement, this provider analyses the delegated service and, if needed,
delegates that service to other service providers, termed internal providers (e.g. an IT
infrastructure department) and external providers (e.g. suppliers). In this manner, each
subservice execution contributes to the service delegated by the customer, in this case, the
super-service.
In view of this discussion, it is defined the relationship between business process and IT
service inherent to the competency question CQ1. In summary, information technology is
frequently used to support the business process activities through IT services. Given that a
business process activity is an activity that is owned and performed by the business
(commonly by a business unit) and an IT service is a service provided by an IT organization,
the relation between business process and IT service occurs as a result of the dependence
relation between the respective agents, which means business unit and IT organization. In
other words, the fact of a certain business process activity occurrence being performed by
the business denotes a social commitment of this agent in performing this occurrence. This
social commitment contains a propositional content, i.e. a goal. Thereby this agent has a
commitment to perform a certain action which satisfies this goal. In case this agent, i.e. the
business, needs an IT service to achieve such a goal, this agent can delegate the goal (or part
thereof) to the agent responsible for providing the IT service. In this context, the agent
responsible for the business process activity assumes the role of requestor (more specifically,
the role of customer) and the agent responsible for providing the service assumes the role of
IT service provider. Therefore, the relation between business process and IT service is
derived from the dependency relation between its respective agents, leading to a delegation
relation. In this sense, it is worth noting the UFO contribution as it fundaments the entire
service delegation process. Regarding the relationship between IT services and IT
components, which is inherent to the competence question CQ2, the discussions during this
section have defined that such components are seen as resources under the UFO point of
view, assuming the usage participation. Moreover, in considering the definition of hardware
and software presented by IEEE 610.10 (IEEE, 1994) as well as by ISO/IEC 2382-1 (ISO/IEC,
1993) it is possible to conclude that a hardware is a physical component that processes the
instructions described by a software. In terms of UFO, a software is a type of normative
description which describes a computer process, i.e. a type of event (an event universal). An
instance of this process denotes an occurrence of such a process, termed computer process
occurrence, i.e. a predetermined course of events whose execution includes the participation
of a hardware. In this sense, the concept of hardware, as well as of software, is mapped to
the notion of substantial. As such, if a service execution requires a software resource which
is processed by a hardware resource, then this service execution also requires this hardware
resource. In other words, if a service execution x requires a software resource y and this last

resource describes a computer process z whose occurrence w includes the participation of a
hardware q, then the service execution x also requires the hardware resource q, as
(A7) x,y,z,w,q (ITSE(x) Software(y) Computer-Process(z) Computer-Process-
Occurrence(w) Hardware(q) requires(x,y) describes(y,z) instance-of(w,z)
participation-of(w,q) requires(x,q))
In this sense, it is useful to note the UFO contribution, as it allows the distinction between
the notions pertinent to software, its processes and its occurrences, as well as the proper
participations of hardware components, granting more expressiveness, clarity and veracity
to the model. Indeed, the use of appropriate tools such as UFO and methods such as SABiO
promoted o development of important ontological distinctions, as discussed in this section.
Considering the ontology engineering approach adopted in this chapter, the next section
explores the benefits from the implementation of the conceptual model.
5. An implementation and application of the proposed ontology
In view of the contributions that ontologies provide towards automated solutions,
including important features such as artificial intelligence and, above all, interoperability,
this section aims to present a case study in order to: (i) perform a proof of concept of the
models developed in the previous section and also (ii) demonstrate how the concepts
modeled by using ontologies can be implemented and applied in a computational system,
in order to enable automation, including important features such as cited in this paragraph.
Taking into accounting Figure 3, which illustrates the ontology development approach
adopted in this work, the ontology proposed in this section concerns to the implementation
model, which represents an implementation of the conceptual model proposed in the
previous section.
As discussed in Guizzardi (2005, 2007), each ontology engineering phase requires the use of
appropriate languages in relation to the context within which the model is being designed.
In the implementation phase, the choice of a language must be conducted by the end-
application requirements. This refers to languages focused on computational requirements,
such as decidability and efficient automated reasoning. In terms of the configuration
management process, factors such as decidability, completeness and expressiveness are
considered to be key requirements because of its role in relation to all other processes in
service management. Thus, for the implementation of the conceptual models, this work used
the OWL DL sublanguage, since it allows a greater degree of expressiveness, as compared
with OWL Lite, while maintaining computational guarantees such as completeness and
decidability, features not guaranteed by OWL Full (Bechhofer et al., 2004). In addition to
OWL DL, the implementation of the models also used SWRL (Horrocks et al., 2003). The
SWRL language allows the representation of the axioms defined in the conceptual models
presented in Section 4 in an integrated way with the concepts and relations implemented by
means of OWL. Finally, for the implementation of the models in OWL and the definition of
the axioms in SWRL, this work used the Protg tool (Protg, 2011), an ontology editor that
enables the integration of different languages such as OWL and SWRL inside the same
implementation environment.

Automation 36
Once defined the development environment, the implementation models were developed
according to the modular structure of the conceptual models, as follows: (i) UFO.owl; (ii)
BusinessProcess.owl; (iii) ITService.owl; (iv) ITComponents.owl and finally (v)
ConfigurationItem.owl. Due to the expressivity restrictions inherent in the implementation
languages, the main issue concerning the mapping from conceptual models into
implementation models is related to the treatment of the reduction in semantic precision. In
order to maintain this reduction at an acceptable level, the most relevant losses that were
found were related to the transformation of all ontologically well-founded concepts and
relations into OWL classes and properties, respectively. Regardless of the application
scenario, this mapping must consider the information contained in the notation used for the
development of the conceptual models, such as cardinality, transitivity, domain and range.
With respect to cardinality and transitivity, in OWL it is not possible to represent them
simultaneously (Bechhofer et al., 2004). As a result, this work considers that the
representation of cardinality restrictions is more relevant to the implementation models
developed in this section. In addition, to represent the cardinality restrictions in both
directions inverse relations were used. For instance, the relation requests is represented by
the pair of relations requests and is requested by. However, according to Rector and
Welty (2001), the use of inverse relations significantly increases the complexity of automated
reasoning. Thus, they should be used only when necessary. With respect to domain and
range, an issue that should be considered is how to organize and represent many generic
relations. For example, if a generic relation describes is created, it is not possible to restrict
the domain and the range. In this case, the design choice was to use specific relations like
describes_Software_ComputerProcess, which is represented as a sub-relation of a generic
relation describes. Finally, with respect to SWRL restrictions, this language has neither
negation operators nor existential quantifiers (Horrocks et al., 2003). In addition, the SWRL
language might lead to undecidable implementation models. Nevertheless, this issue may
be worked around by restricting the use of rules and manipulating only those that are DL-
safe (Motik et al., 2005). As an attempt to make this tangible, consider an implementation of
the axiom A7, which concerns the competence question QC2, discussed in Section 4. This
implementation is represented by the rule R7a.
(R7a) IT_Service_Execution(?IT-SERVICE-EXECUTION) Software(?SOFTWARE)
ComputerProcess(?COMPUTER-PROCESS)
ComputerProcessOccurrence(?COMPUTER-PROCESS-OCCURRENCE)
Hardware(?HARDWARE) requires_ITServiceExecution_Resource(?IT-SERVICE-
EXECUTION,?SOFTWARE)
describes_Software_ComputerProcess(?SOFTWARE,?COMPUTER-PROCESS)
isInstanceOf_ComputerProcessOccurrence_ComputerProcess(?COMPUTER-PROCESS-
OCCURRENCE,?COMPUTER-PROCESS)
hasParticipationOf_ComputerProcessOccurrence_Hardware(?COMPUTER-PROCESS-
OCCURRENCE,?HARDWARE) requires_ITServiceExecution_Resource(?IT-SERVICE-
EXECUTION,?HARDWARE)
The axiom A7 constitutes the set of axioms that establishes the relationship between the
computational resources that are required by an IT service execution as a response to the
competence question QC2. Thus, this axiom involves concepts such as IT service execution,
hardware and software, as well as the interrelationship between these concepts, such as the

relations requires and describes. As discussed earlier, concepts are implemented as
classes, while relations are implemented as properties, according to OWL. According to
described throughout this work, the same conceptual model can give rise to a variety of
implementation models in order to meet different requirements, in accordance with the
purpose of the application scenario. The rule R7a is intended to meet configuration
management activities, especially the activities of identification and inference of managerial
information. In Baico and Garcia (2010), the axiom A7 is implemented mainly in order to
meet the latter activity. In this case, the axiom is implemented as formalized by the rule R7b.
(R7b) IT_Service_Execution(?IT-SERVICE-EXECUTION) Software(?SOFTWARE)
ComputerProcess(?COMPUTER-PROCESS)
ComputerProcessOccurrence(?COMPUTER-PROCESS-OCCURRENCE)
Hardware(?HARDWARE) requires_ITServiceExecution_Resource(?IT-SERVICE-
EXECUTION,?SOFTWARE)
describes_Software_ComputerProcess(?SOFTWARE,?COMPUTER-PROCESS)
isInstanceOf_ComputerProcessOccurrence_ComputerProcess(?COMPUTER-PROCESS-
OCCURRENCE,?COMPUTER-PROCESS)
hasParticipationOf_ComputerProcessOccurrence_Hardware(?COMPUTER-PROCESS-
OCCURRENCE,?HARDWARE) query:select(?IT-SERVICE-EXECUTION,?SOFTWARE,
?HARDWARE) query:orderByDescending(?IT-SERVICE-EXECUTION)
There are numerous contributions offered by ontology engineering for the construction of
autonomous, intelligent and above all interoperable computational applications. Although
done in a different area, Gonalves et al. (2008) presents an application for the interpretation
of electrocardiogram results where the use of an ontology model provides a graphical
simulation of the heart behavior of an individual and the correlation of the heart behavior
with the known pathologies. Regarding configuration management, this process identifies,
controls, maintains and checks the versions of the existing configuration items and reports
the information of the IT infrastructure to all those involved in the management. Thus, this
section aims to demonstrate how implementation models can be applied in a computational
environment in order to support management activities in an automated manner. In
addition, the results will provide a proof of concept of the developed ontology.
The first part comprises the mapping between business processes activities and business
units responsible for these activities. In addition, it includes the needs that characterize these
business units. Figure 5-a shows, for example, that the business process activity BPAO_Sales
is composed of the activity BPAO_Ordering, which is owned by the business unit BU_Sales,
as shown in Figure 5-b. The business unit BU_Sales, in turn, has need inherent to this activity,
as presented in Figure 5-c. It is worth mentioning that such information, as well as any
assertion that appears highlighted in blue, concerns information previously inserted into the
implementation models. On the other hand, assertions that appear highlighted in yellow are
information automatically inferred by the implementation models, which denote knowledge
acquisition. Such inferences are performed by the Pellet reasoner (Sirin et al., 2007).
As discussed in Section 4, IT services can achieve business needs by supporting its activities.
Thus, Figure 5-d illustrates the IT services that can achieve the needs of the business. This
information is inferred by executing the axiom A1, implemented in this section. Figure 5-d
illustrates, for example, that the service IT_Service_Ordering can achieve the need

Automation 38
Need_Ordering. Figure 5-e, in turn, presents the requirements used for service requests.
This inference is performed by executing the implemented axiom A2. This scenario
motivates the requestor, in this case the business unit, to ask for services which can achieve
its needs, as shown in Figure 5-f. This inference is performed by means of the implemented
axiom A3. It should be mentioned that information of this nature is fundamental to
processes such as service level management, which interacts with configuration
management in requesting information in order to find services able to meet the needs of the
requestors.
According to the axiom A4, once a service capable meeting the need is found, the requestor
can then initiate the delegation process by contracting the service. Thus, Figure 5-g shows
the provider hired by the business unit. As shown in Figure 5-g, the provider hired by the
business unit BU_Sales is the IT_Department, because it is the provider responsible for
providing the requested service. As discussed in Section 4, the hiring process is mediated by
an agreement, as illustrated in Figure 5-h. This inference is performed by means of the
implemented axiom A5. Figure 5-h illustrates, for example, that the agreement
SLA_Ordering mediates the requestor BU_Sales and the provider IT_Department. This
agreement, in turn, characterizes the delegation process, which has a delegator (in this case
the requestor) and a delegatee (in this case the provider), as shown in Figure 5-i. This
inference is performed by means of the implemented axiom A6. Figure 5-i illustrates, for
example, that the delegation ITSD_Ordering is associated to the agreement SLA_Ordering
and has as the delegator the BU_Sales and as the delegatee the IT_Department.
As described in Section 4, the hired provider receives the service delegation from the
requestor and provisions the necessary resources for the service execution. In this sense, the
received service execution is characterized as a complex action which is delegated to the
support groups by means of subservices. In this context, the hired provider, in a manner
similar to that of the business unit, plays the role of requestor and the support groups, in
turn, play the role of service provider. Thus, Figure 5-j shows, for instance, that the
execution ITSE_Ordering, which represents an instance of the service IT_Service_Ordering,
is composed of the sub-executions ITSE_Ordering_Processing and ITSE_Ordering_Printing.
The delegation performed by the provider to the support groups is mediated by agreements.
Figure 5-k shows, for example, that the agreement OLA_Ordering_Processing mediates the
delegation process of the sub-execution ITSE_Ordering_Processing between the
IT_Department and the IT_Department_System. In this case, IT_Department plays the role
of delegator while the support groups play the role of delegatee, as shown in Figure 5-l.
IT services are based on the use of information technology. Thus, Figure 5-m relates the
software required by each service execution. Figure 5-m shows, for example, that the
execution ITSE_Ordering_Processing requires the software Software_Ordering_Processing.
As discussed in Section 4, software is processed by hardware. Thus, Figure 5-n presents the
hardware associated with the processing of the concerned software. In addition, Section 4
states that if a service execution requires a software and this software is processed by a
hardware, then this service execution also requires this hardware, as illustrated in Figure 5-
o. This inference is performed by means of the implemented axiom A7. Figure 5-o
illustrates, for example, that the execution ITSE_Ordering_Processing requires the hardware
Hardware_Sales, since such hardware processes the software
Software_Ordering_Processing (as shown in Figure 5-n) required by such an execution.

Thus, this case study concludes the mapping between business and IT. This mapping is
fundamental to other management processes. As an illustration, the configuration
management process of this case study is able to correlate and determine that a particular
event of unavailability on the hardware Hardware_Sales affects the software
Software_Ordering_Processing, which is used by the service execution
ITSE_Ordering_Processing. This execution is part of the execution ITSE_Ordering which is
instance of the service IT_Service_Ordering and, in turn, supports related activity of the
business process activity BPAO_Sales. This correlation, provided by the implementation
model, is the basis for activities such as: (i) event correlation in event management; (ii)
workaround identification in incident management; (iii) root cause analysis in problem
management and (iv) impact analysis in change management.

Fig. 5. Application of the implementation model.
To complete this case study, consider intelligent software agents playing the roles of
requestor and provider and, consequently, negotiating the provision of services that meet the
needs of the environment. Regarding the role of provider, such activities refer to the various
management disciplines, as presented throughout this section. In particular, this scenario
denotes implementation models subsidizing paradigms known as autonomous networks. In
general, it denotes implementation models promoting automation in various areas of interest.
6. Conclusion
As discussed in this chapter, automation enables organizations to explore opportunities as
well as supporting challenges in an effective and efficient way. The important role played by

Automation 40
IT as an instrument for automation has made automation increasingly dependent on IT,
consequently rising the demands for advances, as observed by the growing challenges
arising from the conception of systems continuously more complex, intelligent and, above
all, interoperable. Moreover, the need for an efficient and effective IT management has
grown substantially, as evidenced by widespread adoption of innovative best practices
libraries and standards. For this reason, this work presented an ontology of IT service
configuration management. The objective was not only to adopt the state of the art in order
to address key research challenges in IT management, but also to foment novel approaches
which can be applied in IT in various areas of interest.
The diverse uses attributed to the ontologies in computer science and the interrelation
between their purposes promote the search for approaches capable of providing the
construction of ontological models able to achieve the various objectives assigned to them.
Based on innovative and high quality research initiatives, this chapter discussed about a
systematic approach for building ontologies known as Ontology Engineering. In considering
the various uses and purposes, as well as their interrelationships, these initiatives attempt to
establish a structured means of development as an alternative to the various ad hoc
approaches that characterize the current developments and imply in models unable to
achieve their goals. In summary, this approach allowed the development of conceptual
models which are application-independent artifacts and, as a result, it enabled their use as a
reference ontology for the subsequent development phases, deriving implementation
models in order to address the different purposes of end applications.
According to Guizzardi and Halpin (2008), the practice of conceptual modeling is permeated
by philosophical questions. This demonstrates the need for an appropriate theoretical
foundation for conceptual modeling languages so as to ensure that the quality requirements
of domain and comprehensibility appropriateness can be fulfilled by the produced
conceptual models. In this sense, they advance that philosophically well-founded ontologies
play a key role in this initiative. They complement this line of reasoning by citing Guarino
and Guizzardi (2006) and emphasizing that although typical conceptual modeling
languages provide facilities for structuring domain elements, such as taxonomies and data
value structures, the justification for the validity of many structuring choices, as much as the
justification for the grammar of many natural language sentences, can only be made on
ontological grounds, in this sense, on a philosophical basis. As a final consideration,
Guizzardi and Halpin (2008) point out that philosophical foundations are vital components
with respect to conceptual modeling, in general, and domain ontology engineering, in
particular, as mature disciplines with sound principles and practices. Thus, in quoting the
physicist and philosopher of science Mario Bunge, every science presupposes some
metaphysics, they conclude that a scientific field can either choose to develop and make
explicit its philosophical foundations or to remain oblivious to its inevitable and often ad hoc
ontological and epistemological commitments.
Accordingly, in addition to the appropriate methods and techniques, such as the SABiO
method, this approach used a philosophically well-founded ontology, termed UFO. The
SABiO method provided a systematic approach that led the development of ontology
proposed in this chapter, describing an iterative process, closely related to evaluation. Thus,
with emphasis on the concept of competence questions, the SABiO method provided a
means for defining the scope and purpose of the ontology, leading its capture and

formalization, the reuse of existing ontologies, as well as its evaluation and documentation,
enabling the ontology proposed in this chapter to adequately meet the requirements for
which it was designed, as confirmed by proof of concept. The UFO ontology, in turn, was
useful in building a conceptual model committed to maximizing the expressivity, clarity and
truthfulness of the modeled domain. These characteristics are key quality attributes of a
conceptual model, responsible for its effectiveness as a reference framework for the tasks of
semantic interoperability and reuse. In fact, as a knowledge representation language
philosophically well-founded, the UFO ontology aims to provide a sound basis for the
representation of a conceptualization and therefore inhibit arbitrary descriptions of concepts
and relationships of a universe of discourse. As discussed during the development of the
conceptual models proposed in this chapter, the UFO ontology guided diverse modeling
decisions, contributing to the derivation of new knowledge or the identification and
elucidation of ambiguous and inconsistent representations of the domain, often represented
in various literatures.
As discussed in this chapter, the more it is known about a universe of discourse and the
more precisely it is represented, the bigger the chance of producing models that reflect, as
much as possible, the appropriate conceptualization of the domain. In this sense, besides the
use of appropriate languages, such as ontologies philosophically well-founded, as a
common and shared specification, it is important that the specification of the domain
considers appropriate literatures and, especially, that it considers its main concepts and
relationships as well as application independence. Considering the universe of study of this
chapter, these aspects associated with the use of appropriate methods and tools enabled the
development of models able to maximize the alignment between IT and business for
humans and computers. Moreover, these aspects allude to a point of view which should be
mentioned. By maximizing the capacity of a model in acting as a common and shared source
about a universe of discourse, conceptual models, representing norms and standards, can be
potentially used as an addendum to such literatures. This is because, in general, these
libraries are described in natural languages, which are susceptible to ambiguities and
inconsistencies, as opposed to conceptual models, which are formally described.
Considering the importance of automation, as well as the contributions that an
implementation provides in terms of ontology evaluation, it was developed an
implementation model, derived from the conceptual model proposed in this chapter. In
addition, by applying the entire approach discussed in this chapter, it is possible to attest it,
as well as making it more tangible, promoting its benefits. It should be mentioned that the
development of conceptual models followed by the development of implementation models
became evident the distinction between ontology representation languages, as discussed
throughout this chapter. This demonstrates that the approach adopted in this chapter shows
itself appropriate by considering the various uses and purposes assigned to ontologies. In
this way, despite the expressivity restrictions inherent in implementation languages, it was
possible to perform a proof of concept of the ontology developed in this chapter as well as
demonstrating how such models can be derived and implemented in computing
environments with a view to the different computational requirements. In particular, it was
possible to show how implementation models can support automation, including special
characteristics such as knowledge acquisition and interoperability. From this point of view it
is important to note that the concepts inherited from UFO are important in promoting
paradigms such as artificial intelligence by making explicit, for computational agents,

Automation 42
concepts that express the daily lives of human agents, such as intentions, goals and actions.
These factors are especially important in face of the increasing need for integration of
complex as well as heterogeneous systems and also when considering autonomous systems.
Therefore, the contributions of this chapter are not restricted to the domain of
configuration management. Instead, they promote semantic interoperability in IT in
diverse areas of interest, maximizing the advances in automation. Additionally, this
chapter makes possible other researches. In this sense, future works include: (i) the
extension of the ontology for covering other business-driven IT management concepts,
such as IT services metrics and business measures as well as their relationships,
improving the alignment between these two domains; (ii) the extension of the ontology
in order to cover other configuration management concepts, such as baseline, version
and variant; (iii) the extension of the ontology to cover other management process, such
as change management and release management; (iv) the application of conceptual
models as an addendum to norms and standards; (v) the application of the
implementation models, especially with techniques like artificial intelligence,
promoting paradigms such as autonomous networks.
7. References
Baico, G., Costa, A.C.M., Calvi, C.Z. & Garcia, A.S. (2009). IT Service Management and
Governance Modeling an ITSM Configuration Process: a Foundational
Ontology Approach, In 4th IFIP/IEEE International Workshop on Business-driven
IT Management, 11th IFIP/IEEE International Symposium on Integrated Network
Management, 2009
Baico, G. & Garcia, A.S. (2010). Implementation and Application of a Well-Founded
Configuration Management Ontology, In 5th IFIP/IEEE International Workshop on
Business-driven IT Management (BDIM), 12th IFIP/IEEE Network Operations and
Management Symposium (NOMS), Osaka, 2010
Bechhofer, S. et al. (2004). OWL Web Ontology Language Reference, In W3C
Recommendation, Oct 2011, Available from http://www.w3.org/TR/owl-ref/
Brenner, M., Sailer, M., Schaaf, T. & Garschhammer, M. (2006). CMDB - Yet Another
MIB? On Reusing Management Model Concepts in ITIL Configuration
Management, In 17th IFIP/IEEE Distributed Systems Operations and Management
(DSOM), 2006
Calvi, C. Z. (2007). IT Service Management and ITIL Configuration Process Modeling in
a Context-Aware Service Platform (in Portuguese), Master Dissertation, UFES,
2007
Costa, A.C.M. (2008) ITIL Service Level Management Process Modeling: An Approach
Using Foundational Ontologies and its Application in Infraware Platform (in
Portuguese), Master Dissertation, UFES, 2008
Falbo, R.A. (1998). Knowledge Integration in a Software Development Environment (in
Portuguese), Doctoral Thesis, COPPE/UFRJ, 1998
Falbo, R.A. (2004). Experiences in Using a Method for Building Domain Ontologies, In
16th Conference on Software Engineering and Knowledge Engineering (SEKE),
Canada, 2004

Gonalves, B. et al. (2009). An Ontology-based Application in Heart Electrophysiology:
Representation, Reasoning and Visualization on the Web, In ACM Symposium on
Applied Computing, 2009
Gonalves, B.N., Guizzardi, G. & Pereira Filho, J.G. (2011). Using an ECG reference
ontology for semantic interoperability of ECG data, In Journal of Biomedical
Informatics, Special Issue on Ontologies for Clinical and Translational Research,
Elsevier, 2011
Guarino, N., Guizzardi, G. (2006). In the Defense of Ontological Foundations for
Conceptual Modeling, In Scandinavian Journal of Information Systems, ISSN 0905-
0167, 2006
Guizzardi, G. & Wagner, G. (2004). A Unified Foundational Ontology and some
Applications of it in Business Modeling, In Open INTEROP Workshop on Enterprise
Modelling and Ontologies for Interoperability, 16th CAiSE, Latvia, 2004
Guizzardi, G. (2005). Ontological Foundations for Structural Conceptual Models, Ph.D. Thesis,
University of Twente, ISBN 90-75176-81-3, The Netherlands, 2005
Guizzardi, G. (2006). The Role of Foundational Ontology for Conceptual Modeling and
Domain Ontology Representation, In Companion Paper for the Invited Keynote
Speech, 7th International Baltic Conference on Databases and Information Systems,
2006
Guizzardi, G. (2007). On Ontology, ontologies, Conceptualizations, Modeling
Languages, and (Meta)Models, In Frontiers in Artificial Intelligence and
Applications, Databases and Information Systems IV, IOS Press, ISBN 978-1-58603-
640-8, Amsterdam, 2007
Guizzardi, G., Falbo, R.A. & Guizzardi, R.S.S. (2008). The Importance of Foundational
Ontologies for Domain Engineering: The case of the Software Process Domain (in
Portuguese), IEEE Latin America Transactions, Vol. 6, No. 3, July 2008
Guizzardi, G. & Halpin, T. (2008). Ontological foundations for conceptual modeling, In
Journal of Applied Ontology, v.3, p.1-12, 2008
Guizzardi, G., Lopes, M., Baio, F. & Falbo, R. (2009). On the importance of Truly
Ontological Distinctions for Ontology Representation Languages: An Industrial
Case Study in the Domain of Oil and Gas, In Lecture Notes in Business Information
Processing, 2009
Horrocks, I. et al. (2003). SWRL: A Semantic Web Rule Language Combining OWL and
RuleML, In DAML, Oct 2011, From http://www.daml.org/2003/11/swrl/
IEEE (1994). IEEE Standard Glossary of Computer Hardware Terminology, IEEE Std 610.10-
1994
ISO/IEC (1993). IT Vocabulary Part 1: Fundamental terms, ISO/IEC 2382-1:1993
ISO/IEC (2005). Information technology Service management, ISO/IEC 20000, 2005
ITIL (2007). ITIL Core Books, Office of Government Commerce (OGC), TSO, UK, 2007
Jones, D.M., Bench-Capon, T.J.M. & Visser, P.R.S. (1998). Methodologies For Ontology
Development, In 15th IFIP World Computer Congress, Chapman-Hall, 1998
Lopez de Vergara, J.E., Villagra, V.A. & Berrocal, J. (2004). Applying the Web Ontology
Language to management information definitions, IEEE Communications
Magazine
Majewska, M., Kryza, B. & Kitowski, J. (2007). Translation of Common Information
Model to Web Ontology Language, In International Conference on Computational
Science, 2007

Automation 44
Moura, A., Sauve, J. & Bartolini, C. (2007). Research Challenges of Business Driven IT
Management, In 2nd IEEE/IFIP Business-driven IT Management, 10th IFIP/IEEE
IM
Moura, A., Sauve, J. & Bartolini, C. (2008). Business-Driven IT Management - Upping
the Ante of IT: Exploring the Linkage between IT and Business to Improve Both
IT and Business Results, In IEEE Communications Magazine, vol. 46, issue 10,
October 2008
Motik, B., Sattler, U. & Studer, R. (2005). Query Answering for OWL-DL with Rules, In
Journal of Web Semantics, Vol. 3, No. 1, pp. 41-60, 2005
Pavlou, G. & Pras, A. (2008). Topics in Network and Service Management, In IEEE
Communications Magazine, 2008
Pras, A., Schnwlder, J., Burgess, M., Festor, O., Prez, G.M., Stadler, R. & Stiller, B.
(2007). Key Research Challenges in Network Management, IEEE Commun
Magazine, 2007
Protg (2011). The Protg Ontology Editor and Knowledge Aquisition System, October
2011, Available from http://protege.stanford.edu/
Rector, A. & Welty C. (2001). Simple part-whole relations in OWL Ontologies, In W3C
Recommendation, Oct 2011, Available from
http://www.w3.org/2001/sw/BestPractices/OEP/SimplePartWhole
Sall, M. (2004). IT Service Management and IT Governance: Review, Comparative Analysis
and their Impact on Utility Computing, In HPL-2004-98, 2004
Santos, B.C.L. (2007). O-bCNMS: An Ontology-based Network Configuration Management
System (in Portuguese), Master Dissertation, UFES, 2007
Sirin E. et al. (2007). Pellet: a Practical OWL-DL Reasoner, In Journal of Web Semantics:
Science, Services and Agents on the World Wide Web, Vol. 5, No. 2, pp 51-53, 2007
Smith, B. & Welty, C. (Eds.). (2001). Ontology: Towards a new synthesis, Chris Welty and
Barry Smith, Formal Ontology in Information Systems, ACM Press, 2001
Ullmann, S. (1972). Semantics: An Introduction to the Science of Meaning, Basil Blackwell,
Oxford, 1972
Vermeer, M.W.W. (1997). Semantic interoperability for legacy databases, PhD Thesis,
University of Twente, The Netherlands, 1997
Wong, A.K.Y., Ray, P., Parameswaran, N. & Strassner, J. (2005). Ontology Mapping for the
Interoperability Problem in Network Management, IEEE Journal on Selected Areas in
Communications (JSAC), Vol. 23, No. 10, October 2005
Xu, H. & Xiao, D. (2006). A Common Ontology-Based Intelligent Configuration
Management Model for IP Network Devices, In Proceedings of the First International
Conference on Innovative Computing, Information and Control (ICICIC), 2006
3
Automatic Restoration of
Power Supply in Distribution
Systems by Computer-Aided Technologies
Daniel Bernardon
1
, Mauricio Sperandio
2
,
Vincius Garcia
1
, Luciano Pfitscher
1
and Wagner Reck
2

1
Federal University of Santa Maria
2
Federal University of Pampa
Brazil
1. Introduction
The need to improve quality and reliability of Power Systems has contributed to the
advancement of research on Smart Grids and related topics. Some challenges that motivate
the deployment of such researches include the increasing in energy consumption,
environmental aspects, the integration of distributed and renewable generation, and new
advances in computer aided technologies and automation. Smart Grids are characterized by
a series of integrated technologies, methodologies and procedures for planning and
operation of electrical networks. A survey of the main projects and researches related to
Smart Grid is presented in (Brown, 2008).
Some of the main desired features in a Smart Grid are the low operating and maintenance
costs and the ability to self-healing. In this context, utilities have concentrated significant
efforts in order to improve the continuity of the supplied electrical energy, especially
because of regulatory policies, besides the customer satisfaction and the improvement of the
amount of energy available to commercial and industrial activities. However, supply
interruptions are inevitable due to implementation of the expansion of the system,
preventive maintenance on network equipment, or even by the action of protective devices
due to defects.
A distribution network can have its topology changed by opening or closing switches,
allowing to isolate faults and to restore the supply in contingency situations, and also in case
of scheduled shutdowns. In addition, the change of topology allows a better load balancing
between feeders, transferring loads of heavily loaded feeders to other feeders, thus
improving the voltage levels and reducing losses and increasing levels of reliability.
The reconfiguration of a distribution network is considered an optimization problem in
which the objective is to seek for one configuration, among several possible solutions, that
leads to better performance, considering the ultimate goal of the reconfiguration and
observing the network constraints. One factor that increases the complexity of the problem

Automation

46
is the large number of existing switches in a real distribution network, which leads to a lot of
different possible configurations to be analyzed.
The stages of planning the reconfiguration of distribution networks usually involve the
collection of information of the network and load, the application of methodologies for
estimating loads and calculation of power flows, and the application of the methodology for
network optimization.
The behavior of the load is essential in the study of reconfiguration of distribution networks,
because load variations cause demand peaks at different periods, and thus the
reconfiguration at a particular period may not meet the objectives and constraints at later
periods.
Several researches are related to the reconfiguration of distribution networks, based on
mathematical methods, heuristics and artificial intelligence techniques, as shown in (Takur
& Jaswanti, 2006). Generally, the reconfiguration aims to meet the maximal number of
consumers as much as possible, or the maximum energy demand, with loss minimization
and load balancing. The reduction of losses is often taken as the primary objective of the
reconfiguration. When more than one objective function is defined, it is necessary to apply
multicriteria methods for decision making. Among the restrictions, there is the need to
maintain a radial network, and the operation within the limits of voltage and current
capacity of equipment and conductors of the line. The coordination of the protection system
of the new network configuration should be taken into consideration.
In recent years, new methodologies of reconfiguration of distribution networks have been
presented, exploring the greater capacity and speed of computer systems, the increased
availability of information and the advancement of automation, in particular, the SCADA
(Supervisory Control and Data Acquisition). With the increased use of SCADA and
distribution automation - through the use of switches and remote controlled equipment - the
reconfiguration of distribution networks become more viable as a tool for planning and
control in real time.
Act as soon as possible in a contingency situation may result in a minimum cost to the utility
and consumers. When a fault is identified at any point of the network, the following
procedures must be performed: identify the location of the problem; isolate the minimum
portion of the distribution system by opening normally closed switches; restore the power
supply to consumers outside the isolated block reclosing the feeder breaker and/or
normally open switches; correct the problem; re-operate the switches to get back to the
normal network status. To do that, generally a maintenance crew has to travel long
distances, and may be affected by traffic jams and unfavorable accesses.
The automation of distribution systems, with the installation of remote controlled switches,
plays an important role on reducing the time to implement a service restoration plan
(Sperandio et al., 2007). These devices have shown to be economically viable due to the
growth of a large number of automation equipment suppliers and new communication
technologies.
The commitment of an efficient system to operate these devices is quite important for the
utilities, aiming to guarantee the technical feasibility of the network reconfiguration with
minimal time necessary to restore the energy supply of the affected consumers.
Automatic Restoration of Power Supply in
Distribution Systems by Computer-Aided Technologies

47
In this chapter, a methodology for the automatic restoration of power supply in distribution
systems by means of remote controlled switches is presented. It includes a validation of the
technical feasibility for the reconfiguration of the network in real time using computer
simulations. Since there may be many configuration options with different gains, an
algorithm based on a multiple criteria decision making method, the Analytic Hierarchical
Process AHP (Saaty, 1980), is employed to choose the best option for load transfers after
contingencies. The AHP method has proven to be effective in solving multi-criteria
problems, involving many kinds of concerns including planning, setting priorities, selecting
the best choice among a number of alternatives and allocating resources. It was developed
to assist in making decisions where competing or conflicting evaluation criteria difficult to
make a judgment (Saaty, 1994). Additionally, the algorithms for load modeling and load
flow are also presented, since they are essential for analysis of the maneuvers.
As a result, load transfers are carried out automatically, being preceded by computer
simulations that indicate the switches to be operated and that ensure the technical feasibility
of the maneuver, with the characteristics of agility and safety for the power restoration and
in agreement with the Smart Grids concepts. The developed tool has been applied in a pilot
area of a power utility in Brazil. The reduction in the displacement of maintenance teams
and improving indices of continuity characterize the greatest benefits for the company,
making a difference in the market and, consequently, generating economic and productivity
gains.
2. Modeling of power load profiles
The most common information on power loads profile comes from utilities charges, based
on measurements of monthly energy consumption. Unfortunately, these data are
insufficient for the analysis of the distribution systems, since they do not reflect the daily
power behavior. Therefore, a methodology for power load modeling is need, which usually
employs typical load curves for their representation. A technique for building the typical
load curves that has advantages in relation to the traditional statistical methods was
presented in (Bernardon et al., 2008). It reduces the influence of random values and also the
amount of measurements required to form a representative sample of the load types.
Instead of using the simple average to determine the active and reactive power values for a
typical load curve ordinate, the following equation is used:
{ } { } { }
t t t t
1
X 2M X 2Me X Mo X
5
( = + +

(1)
where:
X
t
active (P
t
) or reactive (Q
t
) power value for the time t of the typical load curve;
M{X
t
} sample average;
Me{X
t
} sample median (central number of a sample);
Mo{X
t
} sample mode (most frequently occurring value of a sample).
According to the monthly power consumption data and economic activity developed, each
customer is associated to a typical load curve. Based on the load factor (LF) and monthly
energy consumption values (W), the maximum power demand (P
max
) for a k group of
consumers is calculated:

Automation

48

k
k
max
k
W
P
T LF
=
(2)
The typical curves utilized are normalized in relation to the maximum active demand, and
the load curve for a k group of consumers is built by multiplying each ordinate by this
value:

k
*
kUt max kUt
P P P =

k
*
kSt max kSt
P P P = (3)

k
*
kDt max kDt
P P P =

k
*
kUt max kUt
Q P Q =

k
*
kSt max kSt
Q P Q = (4)

k
*
kDt max kDt
Q P Q =
where:
*
kt
P and
*
kt
Q - active and reactive power values, normalized for ordinate t of typical curve k;
U, S and D Working days, Saturdays, Sundays/holidays, respectively.
The integral load curves for Working days, Saturdays and Sundays for the distribution
transformer j is done through the sum of the load curves of the different groups of N
k

consumers connected to it:

k
N
jUt iUt
i 1
P P
=
=

;
k
N
jSt iSt
i 1
P P
=
=

and
k
N
jDt iDt
i 1
P P
=
=

(5)

k
N
jUt iUt
i 1
Q Q
=
=
;
k
N
jSt iSt
i 1
Q Q
=
=

and
k
N
jDt iDt
i 1
Q Q
=
=

(6)
Thus, it is possible to consider the load levels corresponding to the period in which the
failure occurred in the distribution network. Usually, the load transfers are analyzed
considering the faults time of the occurrence and the next four consecutive hours, thus
ensuring the technical feasibility of the load transfers until the network returns to its original
configuration.
3. Load flow method
A version of the classical backward/forward sweep method by (Kersting & Mendive, 1976)
was performed to calculate the load flow in radial distribution networks. Since the electrical
loads are defined by a constant power according to the applied voltage, the circulating

49
current varies with the voltage drops. So, the solution is found only iteratively. The resulting
procedure is described as follows:
Step 1. It is considered that the voltage in all points of the feeder is the same as the voltage
measured in the substation bar. This information can be automatically received by
the remote measurement systems installed at the substations. Voltage drops in the
branches are not taken into account in this step.
Step 2. Active and reactive components of the primary currents absorbed and/or injected
in the system by the electrical elements are calculated.
Step 3. The procedure to obtain the current in all network branches consists of two stages:
(a) a search in the node set is performed adding the current values in the set of
branches and (b) currents from the final sections up to the substation are
accumulated.
Step 4. Voltage drops in primary conductors are determined.
Step 5. From the substation bus it is possible to obtain the voltage drops accumulated at
any other part of the primary network, and, consequently, the voltage values at any
point.
Step 6. The difference between the new voltage values for all nodes and the previous
values is checked. If this difference is small enough comparing to a previously
defined threshold, the solution for the load flow calculation was found and the
system is said to be convergent. Otherwise, steps 2 to 6 are repeated, using the
calculated voltages to obtain the current values. The threshold of 1% was chosen,
because it leads to accurate values for the status variables without requiring too
much processing time.
At the end of the process, the active and reactive powers and the technical losses in the
primary conductors are defined for all branches of the feeder.
This load flow method was implemented in the proposed methodology for the automatic
power supply restoration. It was used for analyzing the technical feasibility of the load
transfers and the results were considered as constraints in the optimization procedure. That
is, such transfers may neither cause an overload on the electrical elements (conductors and
transformers), nor reach the pickup threshold of the protective devices, nor exceed the limits
of voltage range of the primary network. The checking of the constraints is performed by
considering the load profile compatible with the period of the failure.
4. Methodology for automatic operation of remote controlled switches to restore
power supply
The logic for power restoration is presented considering the hypothetical example of the
simplified distribution network illustrated in Figure 1. Normally close (NC) switch and
normally open (NO) switches in Figure 1 are remotely controlled. Assuming that an outage
has occurred in feeder FD-1, the procedure for electric power restoration is:
- Fault downstream of NC-1 switch: in the event of this fault, the current values of short-
circuit will be flagged online in the SCADA (Supervisory Control and Data Acquisition)
system. So, it is assumed that the failure occurred downstream of NC-1 switch; then,
this switch is operated automatically to isolate the defect.

Automation

50
- Fault upstream of NC-1 switch: in the event of this fault, the current values of short-
circuit will not be flagged in the SCADA system. So, it is assumed that the failure
occurred upstream of NC-1 switch, automatically operating the remote controlled
switches to open NC-1 and to close NO-1 or NO-2 in order to transfer consumers
downstream of NC-1 to another feeder.

Fig. 1. Example of switches in a distribution network (FD feeder, NC normally closed
switch, NO normally open switch).
The technical and operational feasibility of load transfers using the remote controlled
switches is verified by computer simulations. If there is more than one option of load
transfer (e.g. to FD-2 or FD-3), the best option will be chosen considering the defined
objective functions and constraints by a multiple criteria decision making algorithm. After
this analysis, the developed tool automatically sends the necessary commands to maneuver
the equipment.
Moreover, the automatic operation of remote controlled switches are carried out only after
all attempts to restart the protection devices have been tried, i.e., they are done only in case
of permanent fault, after a maximum of 3 minutes needed to complete the computer
simulations and maneuvers of the switches since the instant of the fault identification.
Figure 2 illustrates the architecture of the proposed system.

Fig. 2. Architecture of the developed system.

51
The flowchart of the proposed methodology is shown in Figure 3.

Fig. 3. Flowchart of the proposed methodology.
5. Methodology to determine the optimized reconfiguration based on the AHP method
After a contingency, the challenge is to decide which is the best load transfer scheme using
the remote controlled switches among all possibilities, depending on previously defined
objective functions and constraints. This is a multiple criteria decision making problem,
since various types of objective functions can be considered.

Automation

52
The most common objectives are the maximization of restored consumers and of restored
energy; however, generally it is not possible to optimize the grid for both objectives
simultaneously. Furthermore, it is also important to ensure the reliability of distribution
systems, through continuity indicators. The basic parameters are the SAIDI (System
Average Interruption Duration Index) and the SAIFI (System Average Interruption
Frequency Index), according to (Brown, 2009). In this approach, we adopt expected values
based on the systems failure probability.
The constraints considered are the maximum loading of electrical elements, the protection
settings and the allowable voltage drop in the primary network. Typically, the last two
restrictions are the hard ones. On the other hand, a percentage of overloading of the
network elements is acceptable in a temporary situation, assuming that the fault can be fixed
in a couple of hours.
In our approach the following objective functions and constraints were defined to be used in
the analysis of load transfers in case of contingencies:
Objective functions:
- Maximization of the number of restored consumers;
- Maximization of the amount of restored energy;
- Minimization of the expected SAIFI:

Expected Total Number of Customer Interruptions
ESAIFI
Total Number of Customers Served
= (7)
Constraints:
- Current magnitude of each element must lie within its permissible limits:

i i max
I I s (8)
- Current magnitude of each protection equipment must lie within its permissible limits:

i jprot
I I s (9)
- Voltage magnitude of each node must lie within its permissible ranges:

j min j j max
V V V s s (10)
where:
ESAIFI - expected value of system average interruption frequency (failures/year);
i
I - current at branch i;
i max
I - maximum current accepted through branch i;
jprot
I - pickup current threshold of the protection device j;
j
V - voltage magnitude at node j;
j min
V - minimum voltage magnitude accepted at node j;
j max
V - maximum voltage magnitude accepted at node j.

53
The verification of the objective functions and constraints is made by calculation of the load
flow for the various alternatives in real time. The ESAIFI is obtained by applying the
classical equations of reliability during the process of calculating the load flow (Tsai, 1993).
Identifying the best option for load transfers is not simple since three objective functions are
employed. For example, one particular option may have the largest number of consumers to
be transferred, the other the largest amount of energy to be transferred, and the other the
least expected value of consumers interrupted per year.
To solve this, the Analytic Hierarchical Process - AHP method was chosen, because of its
efficiency in handling quantitative and qualitative criteria for the problem resolution. The
first step of the AHP is to clearly state the goal and recognize the alternatives that could lead
to it. Since there are often many criteria considered important in making a decision, the next
step in AHP is to develop a hierarchy of the criteria with the more general criteria at the top
of it. Each top level criteria is then examined to check if it can be decomposed into
subcriteria.
The next step is to determine the relative importance of each criterion against all the other
criteria it is associated with, i.e., establish weights for each criterion. The final step is to
compare each alternative against all others on each criterion on the bottom of the hierarchy.
The result will be a ranking of the alternatives complying with the staged goal according to
the defined hierarchy of the criteria and their weights (Baricevic et al., 2009).
In the proposed approach, the main criterion is to choose the best option for load transfers
and the subcriteria are the proposed objective functions. The alternatives are the options for
load transfers.
An example of the AHP algorithm is defined in (Saaty, 1980):
1. The setup of the hierarchy model.
2. Construction of a judgment matrix. The value of elements in the judgment matrix
reflects the users knowledge about the relative importance between every pair of
factors. As shown in Table 1, the AHP creates an intensity scale of importance to
transform these linguistic terms into numerical intensity values.

Intensity of
Importance
Definition
1 Equal importance
3 Weak importance of one over another
5 Essential or strong importance
7 Very strong or demonstrated importance
9 Absolute importance
2, 4, 6, 8 Intermediate values between adjacent scale values
Table 1. Intensity scale of importance (Yang & Chen, 1989).
Assuming C
1
, C
2
,, C
n
to be the set of objective functions, the quantified judgments on pairs
of objectives are then represented by an n-by-n matrix:

Automation

54

1 12 1n
2 12 2n
n 1n 2n
1 2 n
C 1 a a
C 1 a 1 a
C 1 a 1 a 1
C C C
(
(
(
=
(
(

M
(11)
Where n is the number of objective functions and the entries a
i,j
(i, j = 1, 2, , n) are defined
by the following rules:
- if a
i,j
= o, then a
j,i
= 1/o, where o is an intensity value determined by the operators, as
shown in Table 1;
- if C
i
is judged to be of equal relative importance as C
j
, then a
i,j
= 1, and a
j,i
= 1; in
particular, a
i,i
= 1 for all i.
3. Calculate the maximal eigenvalue and the corresponding eigenvector of the judgment
matrix M. The weighting vector containing weight values for all objectives is then
determined by normalizing this eigenvector. The form of the weighting vector is as
follows:

1
2
:
n
w
w
W
w
(
(
(
=
(
(

(12)
4. Perform a hierarchy ranking and consistency checking of the results. To check the
effectiveness of the corresponding judgment matrix an index of consistency ratio (CR) is
calculated as follow (Saaty & Tran, 2007):

max
n
n 1
CR
RI
| |
|
\ .
= (13)
where:
max
= the largest eigenvalue of matrix M;
RI = random index.
A table with the order of the matrix and the RI value can be found in Saaty, 1980. In general,
a consistency ratio of 0.10 or less is considered acceptable.
The AHP method was implemented in the proposed methodology. Its application is
presented considering the example of distribution network illustrated in Figure 4. Normally
close (NC) switch and normally open (NO) switches are remotely controlled.
The main goal is to define the best option for load transfer, considering that an outage has
occurred in the feeder F1 (fault upstream of NC-1). In this case, there are two transfer
options:
a. open the NC-1 switch and close the NO-1 switch; or
b. open the NC-1 switch and close the NO-2 switch.

55

Fig. 4. Distribution network with three feeders.
First, the judgment matrix was obtained:

(
(
=
(
(

1 2 3
1
2
3
C C C
C 1 3 5
C 1 /3 1 3
C 1 /5 1 /3 1
M
(14)
where:
M = judgment matrix;
C
1
= number of restored consumers;
C
2
= amount of restored energy;
C
3
= expected number of interrupted consumers per year.
Thus, the weight values for the three objective functions were determined:

0.64
W 0.26
0.10
(
(
=
(
(

(15)
where:
max
= 3.07.
The consistency ratio (CR) was calculated by Equation 13:

3.07 3
3 1
CR 0.0673
0.52
| |
|
\ .
= = (16)

Automation

56
The consistency ratio is lower than 0.10 and it is considered acceptable.
Tables 2 and 3 show the results of the analysis for each load transfer. In this example, there
is no violation on the constraints.

Options
Number of
Restored
Consumers
Amount of
Restored
Energy (kW)
Expected Number of
Interrupted Consumers
per Year
1
(open NC-1 and
close NO-1)
14,000 1,930.00 1,800
2
(open NC-1 and
close NO-2)
14,000 1,930.00 2,300
Base Selected 14,000 1,930.00 1,800
Table 2. Results of the analysis for each load transfer.

Options
Number of
Restored
Consumers
Amount of
Restored
Energy
Expected Number of
Interrupted Consumers
per Year
1 1.00 1.00 1.00
2 1.00 1.00 0.78
Table 3. Normalized values of Table 2.
The results using AHP method were obtained by Equation 17:

0.64
Op.1 1.00 1.00 1.00 1.00
0.26
Op.2 1.00 1.00 0.78 0.98
0.10
(
( ( (
(
= =
( ( (
(

(

(17)
According to the proposed method the option 1 is considered the best solution. Thus, the
system performs the commands to make the load transfer, open NC-1 and close NO-1,
without violating the set of constraints.
6. Experimental analysis
To verify the performance of the proposed methodology several case studies were carried
out in the concession area of a power utility in Brazil. The developed methodology was
applied on the distribution system shown in Figure 5, which has 20 distribution substations,
125 feeders, 214 remote controlled switches, and 523,619 consumers.

57

Fig. 5. Distribution network of a power utility in Brazil.
Figures 6 and 7 show the results for the calculation of load flow and the typical load curves
used, respectively:

Fig. 6. Results for the calculation of load flow with the indication of values of current in
branches and voltage in nodes.

Automation

58

Fig. 7. Typical load curves used.
Table 4 shows the results obtained by the application of this methodology in case of outage
of a feeder when considering the power restoration time.

Description
Mean Time to Restore Energy
Faults upstream the NC switch Faults downstream the NC switch
Clients upstream
the switch
Clients downstream
the switch
Clients upstream
the switch
Clients downstream
the switch
Before installing the
remote controlled switches
1h54min 58min 43min 1h34min
After installing the remote
controlled switches
1h54min 0min 0min 1h34min
Reduction - 58min 43min -
Table 4. Results obtained with the use of remote controlled switches.
It should be noted that a reduction of approximately 30 % on the annual SAIDI index of this
system is expected, assuming the number of faults in the main feeder.
Figures 8 and 9 show the picture of a pole top remote controlled switch been installed and
the screen of the SCADA system, respectively:

59

Fig. 8. Installation of a remote controlled switch.

Fig. 9. Screen of the SCADA system.
7. Conclusion
This work presented a methodology developed for automatic power restoration system,
which operates remote controlled switches in the distribution network. It was show how to
assess the technical feasibility of the load transfers in real time by means of computer
simulations, and how the best maneuver option to execute after a contingency is defined
based on the AHP multicriterial method. The system will automatically handle the load
transfers in accordance with the defined objective functions, without violating the
established constraints. Case studies with real data from utilities were conducted to evaluate
the performance of software developed presenting satisfactory results.

Automation

60
8. Acknowledgment
The authors would like to thank the technical and financial support of AES Sul
Distribuidora Gacha de Energia SA, Conselho Nacional de Desenvolvimento Cientfico e
Tecnolgico (CNPq), and Coordenao de Aperfeioamento de Pessoal de Nvel Superior
(CAPES).
9. References
Baricevic, T., Mihalek, E., Tunjic, A. & Ugarkovic, K. (2009). AHP Method in Prioritizing
Investments in Transition of MV Network to 20kV, CIRED 2009 - 20th International
Conference on Electricity Distribution, pp. 1-4, Jun. 2009.
Bernardon, D.P., Comassetto, L. & Canha, L.N. (2008). Studies of parallelism in distribution
networks served by different-source substations, Electric Power Systems Research,
Elsevier, v. 78, p. 450-457, 2008.
Brown, R.E. (2008). Impact of Smart Grid on distribution system design, IEEE Power and
Energy Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st
Century, pp.1-4, 2008.
Brown, R.E. (2009). Electric Power Distribution Reliability, CRC Press, Second Edition, ISBN
978-0-8493-7567-5, New York, 2009.
Kersting, W.H. & Mendive, D.L. (1976). An application of ladder network theory to the
solution of three-phase radial load-flow problems, IEEE Power Engineering Society
Winter Meeting, vol. A76 044-8, pp. 1-6, 1976.
Saaty, T.L. & Tran, L.T. (2007). On the invalidity of fuzzifying numerical judgments in the
Analytic Hierarchy Process, Mathematical and Computer Modelling, vol. 46, pp. 962
975, 2007.
Saaty, T.L. (1980). The Analytic Hierarchy Process: Planning, Priority Setting, Resource Allocation,
McGraw-Hill, ISBN 0-07-054371-2, New York, 1980.
Saaty, T.L. (1994). Highlights and Critical Points in the Theory and Application of the
Analytic Hierarchy Process, European Journal of Operational Research, vol. 52, pp.
426-447, 1994.
Sperandio, M., Coelho, J., Carmargo, C.C.B., et al. (2007). Automation Planning of Loop
Controlled Distribution Feeders, 2nd International Conference on Electrical
Engineering (CEE'07), Coimbra, 2007.
Thakur, T. & Jaswanti (2006). Study and Characterization of Power Distribution Network
Reconfiguration, Proc. 2006 IEEE Transmission & Distribution Conference and
Exposition: Latin America, pp. 1-6.
Tsai, L. (1993). Network reconfiguration to enhance reliability of electric distribution
systems, Electric Power Systems Research, Elsevier, no. 27, pp. 135-140, 1993.
Yang, H.T. & Chen, S.L. (1989). Incorporating a multi-criteria decision procedure into the
combined dynamic programming/production simulation algorithm for generation
expansion planning, IEEE Transaction Power System, vol. 4, pp. 165175, Feb. 1989.
4
Automation of Subjective Measurements of
Logatom Intelligibility in Classrooms
Stefan Brachmanski
Wroclaw University of Technology
Poland
1. Introduction
A great number of rooms are dedicated to a voice communication between a singular
speaker and a group of listeners. Those rooms could be as small as meeting room and
classrooms or larger like auditoriums and theatres. There is a major demand for them to
assure the highest possible speech intelligibility for all listeners in the room. Classrooms are
an example of rooms where a very good speech intelligibility is required (a teacher talks to a
group of students who want to understand the teachers utterance). To determine the
intelligibility degree (its maps) of rooms, it is necessary to take measurements in many
points of those rooms. The number of measurement points depends on rooms size and
precision of created intelligibility rooms maps. Despite of the crucial progress in the
instrumental measurement techniques, the only reliable method subjective speech
intelligibility measurement is still very time consuming, expensive, demanding high skills
and specially trained group of listeners. The first part of this chapter presents the idea of
speech intelligibility subjective measurements. The measurements with properly trained
team are taken in described standards, conditions have to be controlled and repeated. The
subsequent sections of the chapter are focusing on one of the classic subjective speech
intelligibility measurement method in rooms (classrooms) and its automated version which
is named as the modified intelligibility test with forced choice (MIT-FC). Finally, in the last
section of the chapter, there compared results of subjective speech intelligibility
measurements in rooms taken with classic and automated methods are and the relation
between intelligibility taken with the forced choice method is given as well. The presented
relation let us compare results taken with both methods and use relations known from
earlier research carried in domain of speech intelligibility. However, the biggest advantage
of the speech intelligibility measurement automation is the shortening of measurement time
and the possibility of taking simultaneous measurements in several points of the room. The
result of speech intelligibility is obtained just after the end of measurements, it is then
possible to obtain the intelligibility map in few minutes. Of course the precision is growing
with number of listeners in particular points.
Speech quality is a multi-dimensional term and a complex psycho-acoustic phenomenon
within the process of human perception. Every person interprets speech quality in a

Automation

62
different way. The pioneering work on speech intelligibility was carried out by Fletcher at
Bell Labs in the early 1940s. Fletcher and his team not only established the effects of
bandwidth on intelligibility but also the degree to which each octave and -octave band
contributed.
One of the fundamental parameters for quality assessment of speech signal transmission
in analogue and digital telecommunication chains as well as in rooms, sound
reinforcement systems and at selection of aural devices is speech intelligibility (ANSI,
2009; Basciuk & Brachmanski, 1997; Brachmanski, 2002, 2004; Davies, 1989; International
Organization for Standardization [ISO], 1991; Majewski, 1988, 1998, 2000; Polish
Standard, 1991, 1999; Sotschek 1976). Satisfactory speech intelligibility should be provided
by telecommunication channels, rooms and hearing aids. It is obvious, that speech
intelligibility concerns only the linguistic information (i.e. what was said) and does not
take into account such features of speech, like its naturalness or the speaker voice
individuality. Nevertheless, intelligibility should be and - up till now - is viewed as a
basic and most important aspect of the quality of all systems which transmit, code,
enhance and process the speech signals. Satisfactory speech intelligibility requires
adequate audibility and clarity.
In general the evaluation of the speech quality may be done by subjective (intelligibility,
quality rating) (Farina, 2001; Howard & Angus, 2009; Mller, 2010) and objective methods
e.g. Articulation Index - AI (American National Standards Institute [ANSI], 1997; French &
Steinberg, 1947; Kryter, 1962), Speech Transmission Index - STI (Brachmanski, 1982, 2004,
2006, Houtgast & Steeneken, 1973; International Electrotechnical Commission [IEC], 1991;
Lam & Hongisto, 2006; Steeneken & Houtgast, 1980, 2002), Perceptual Evaluation of Speech
Quality PESQ (International Telecommunication Union - Telecommunication
Standardization Sector [ITU-T], 2003), Perceptual Objective Listening Quality Assessment
POLQA, (ITU-T, 2011) . Designers of devices and systems intended for speech transmission
incline to usage of objective measurement methods, not always taking into consideration the
application limitations and exactitude which depends on type of examined object and
measure conditions. However the final verifier of quality of speech transmission devices is
their user, that is to say man.
The aim of aural evaluation is quantitative evaluating and qualitative differentiation of
acoustical signals reaching a listener. The process of aural evaluation can be presented as
follows
B S R (1)
where: B stimulus reaching the listener, S listener, R listeners reaction
The reaction R is dependent on the signal S reaching the listeners receptors and on
conditions in which the listener is based. Generally it can be assumed that the reaction R
depends on the sum of external stimuli having the effect on the listener and internal factors
having the effect on his organism. That assumption, however, doesnt take into
consideration the listeners characteristic features such as cognitive abilities, rate of
information processing, memory etc. The reaction can be then presented as a function

Automation of Subjective Measurements of Logatom Intelligibility in Classrooms

63
R = f(B,S) (2)
Physiological and psychological process connected with a reaction to sound signal (audio)
consists of sensational reaction and emotional reaction. The total listeners reaction is a sum
of both types of reaction.
The sensational reaction is the effect of a physiological process which occurs during the
listening. It arises when a certain stimulus overdraws sensitivity levels or aural sensation
category levels. The emotional reaction is more complex and difficult in analysis because it
isnt a direct result of received signal features but the listeners habits and individuality. As
a conclusion it can be stated that the sensational reaction is the reflection of an acoustical
picture created in a persons (listeners) mind, whereas the emotional reaction is the
reflection of a persons attitude to that picture. Psychological and psychoacoustical research
have proven that when provided stable in time conditions of evaluation, the differences in
sensational reactions of particular listeners are substantially smaller than the differences in
their emotional reactions. Therefore, one aim of objectivisation of aural evaluation is limiting
the influence of emotional reaction on the final assessment result. That aim is achieved by
introducing appropriately numerous assessments statistics, a proper choice and training of a
listeners team and proper choice of testing material and rules of carrying listeners tests. The
results analysis also has the big role in minimization the influence of the emotional reaction
on the assessment result.
Among the different subjective methods that have been proposed for assessment of
speech quality in rooms, the preferred are methods based on intelligibility tests or
listening-only tests (ITU-T, 1996a, 1996b; ITU-R, 1997). The subjective measure results
should be mostly dependent on physical parameters of the tested room and not on the
structure of the tested language material. The elimination of semantic information is done
by means of logatom
1
(i.e. pseudo-words) lists on the basis of which the logatom
intelligibilities are obtained. The problem ist hat speech (logatom) intelligibility is not a
simply parameter to measure.
2. The traditional method of logatom intelligibility
The measurement of logatom intelligibility consists in the transmission of logatom lists,
read out by a speaker, through the tested channel (room), which are then written down by
listeners and the correctness of the record is checked by a group of experts who calculate
the average logatom intelligibility. It is recommended to use lists of 50 or 100 logatoms
(Fig.1).
Logatom lists are based on short nonsense word of the CVC type (consonant-vowel-
consonant). Sometimes only CV, but also CVVC, CVCVC-words are used. The logatoms are
presented in isolation or in carrier phrase e.g. Now, please write down the logatom you
hear. Each list should be phonetically and structurally balanced (Fig.2).

1
Logatom (logos (gr) - spoken phrase, atom (gr) indivisible) vocal sound, generally insignificant,
usually made by the sound of a consonant or the first consonant, then by an intermediate vowel, finally
by a consonant or a final consonant sound.

Automation

64

Fig. 1. Illustrative 100-logatom list from set 3 for the Polish language.
The measurement should be carried out in rooms in which level of internal noise together
with external noise (not introduced on purpose) does not exceed 40 dBA. If no requirements
as to background noise are specified for a tested chain, articulation should be measured at a
noise level of 60 dBA in the receiving room and for the Hoth spectrum (Fig. 3).
The listeners should be selected from persons who have normal, good hearing and normal
experience in pronunciation in the language used in the test. A person is considered to have
normal hearing if her/his threshold does not exceed 10 dB for any frequency in a band of
125 Hz 4000 Hz and 15 dB in a band of 4000 Hz 6000 Hz. Hearing threshold should be
tested by means of a diagnostic audiometer.


65

Fig. 2. Example of phonetic balance of two sets, three lists of 100 logatom each for the Polish
language.

Fig. 3. The room noise power density spectrum (Hoth noise) (Polish Standard 1991, 1999).

Automation

66
The size of the listening group should be such that the obtained averaged test results do not
change as the group size is further increased (minimum 5 persons). The group of listeners
who are to take part in logatom intelligibility measurements should be trained (2-3 training
sessions are recommended). Logatoms should be spoken clearly and equally loudly without
accenting their beginnings or ends. The time interval between individual logatoms should
allow the listener to record the received logatom at leisure. It is recommended that logatoms
should be spoken with 3-5 sec. pauses in between. The time interval between sessions
should not be shorter than 24 h and not longer than 3 days. The total duration of a session
should not exceed 3 hours (including 10 minute breaks after each 20 minute listening
period).
Listeners write the received logatoms on a special form on which also the date of the test,
the test list number, the speakers name or symbol (no.), the listeners name and
additional information which the measurement manager may need from the listener is
noted. The recording should be legible to prevent a wrong interpretation of the logatom.
The received logatoms may be written in phonetic transcription (a group of specially
trained listeners is needed for this) or in an orthographic form specific for a given
language. In the next step, the group of experts checks the correctness of received
logatoms and the average logatom intelligibility is calculated in accordance to the
equation (3) and (4)

= =
=
N
n
K
k
k n L
W
K N
W
1 1
,
[%]
1
(3)
N - number of listeners, K - number of test lists, W
n,k
- logatom intelligibility for n-th listener
and k-th logatom list,

[%] 100
,
,
=
k
k n
k n
T
P
W
(4)
P
n,k
- number of correctly received logatoms from k-th logatom list by n-th listener,
T
k
- number of logatoms in k-th logatom list.
Standard deviation s, calculated in accordance to Eq.(3), expresses the distribution of
logatom intelligibility values W
L
over listeners.

2
1
N
1 n
2
K
1 k
L k , n
) W W (
1 K N
1
s
(
(

=

= =
(5)
If W
n,k
W
L
>3s, the result of measurement is not taken into account, when an average
intelligibility is calculated and calculation of W
L
and s must be done in accordance to Eq. (1)
and (2) for reduced number of measurements.
The obtained average logatom intelligibility value can be used to determine quality classes
according to Table 1 (Polish Standard 1991).


67

Quality
class
Description of quality class
Logatom
intelligibility [%]
I
Understanding transmitted speech without slightest
concentration of attention and without subjectively
detectable distortions of speech signal
above 75
II
Understanding transmitted speech without difficulty
but with subjectively detectable distortions of speech
60 75
III
Understanding transmitted speech with concentration
of attention but without repetitions and return queries
48 60
IV
Understanding transmitted speech with great
concentration of attention and with repetitions and
return queries
25 48
V
It is impossible to fully understand transmitted
speech (breakdown of communication)
to 25
For each quality class lowest logatom articulation values are lowest admissible values
Table 1. Speech intelligibility quality classes for analog channels in the traditional logatom
intelligibility method.
3. Modified intelligibility test with forced choice (MIT-FC)
In the traditional intelligibility tests the listeners write down (in ortographic form) received
utterences on a sheet of paper and next a professional team revises the results and calculates
the average intelligibility. This is the most time-consuming and difficult operation. To
eliminate hand-made revision of the tests a method, called modified intelligibility test with
forced choice (MIT-FC) has been designed and investigated in the Institute of
Telecommunications, Teleinformatics and Acoustics at the Wroclaw University of
Technology .
In the MIT-FC method all experiments are controlled by a computer. The automation of the
subjective measurement is connected with the basic change in generation of logatoms and in
making decision by a listener. The computer generates logatoms and presents the utterances
(for logatom test the list consists of 100 phonetically balanced nonsense words), via a D/A
converter and loudspeaker to the listeners subsequently and for each spoken utterance
several logatoms that have been previously selected as perceptually similar are visually
presented.

Automation

68
It has been found (Brachmanski, 1995) that the optimal number of logatoms presented
visually to the listeners is seven (six alternative logatoms and one transmitted logatom to be
recognized). The listener chooses one logatom from the list visually presented on the
computer monitor. The computer counts the correct answers and calculates the average
logatom intelligibility and standard deviation. The measurement time for one logatom set (3
lists) consisted of 300 logatoms is 20 minutes.
All measurement procedures are fully automatized and an operator has a flexible
possibilities to set the measurement parameters and options. It is also possible to upgrade
the application which realizes the MIT-FC method with more sophisticated scores
processing. Block diagram system for the subjective measurements of logatom intelligibility
by MIT-FC method in the rooms is presented in Fig.4.

Fig. 4. The measuring system for the assessment of logatom intelligibility in room.
The obtained with MIT-FC method average logatom intelligibility value can be used to
determine quality classes according to Table 2.


69

Quality
class
Description of quality class
Logatom
intelligibility [%]
I
Understanding transmitted speech without slightest
concentration of attention and without subjectively
detectable distortions of speech signal
above 70
II
Understanding transmitted speech without difficulty
but with subjectively detectable distortions of speech
60 70
III
Understanding transmitted speech with concentration
of attention but without repetitions and return queries
54 60
IV
Understanding transmitted speech with great
concentration of attention and with repetitions and
return queries
40 54
V
It is impossible to fully understand transmitted
speech (breakdown of communication)
to 40
For each quality class lowest logatom articulation values are lowest admissible values

Table 2. Classes of speech intelligibility quality for analog channels for the MIT-FC method.
4. MIT-FC measurement system
The program for the subjective assessment of speech transmission in rooms with logatom
intelligibility method with forced choice (MIT-FC) is based on TCP Client/Server
technology i.e. the communication is done by local network. Requirements of the program
are following: PC computer with Windows 9x, a network card and hub for communication
between the Server and Clients (members of the team of listeners).
The work with the program starts with the installation on the Server computer a Server
program. The Server is supervised by the person leading the subjective assessment. The next
step is the installation of the Client program on the Client (listener) computers. The Client
computers are used by members of the team of listeners. Before starting the assessment the
Server and Client programs should be configured. During the configuration of the Server
program it is necessary to give the path to the directory with signal files (testing signals -
logatoms), number of logatoms per a session, intervals between reproduced logatoms,
number of sessions and the port for the communication with the Client (usually 3000)
(Fig.5).
The configuration of the Client program is done by giving the name of the measure point
(e.g. a room, location of listener in the room etc.), listeners login, IP address and the port
number (usually: 3000) (Fig.6).

Automation

70

Fig. 5. The example of the window of the Server program

Fig. 6. The example of the window of the Client program
The next step after the configuration is the connection process. The number of people being
logged on can be seen on the right-hand side of the main window of the program. During a
multi-session assessment (the number of session should be chosen during the configuration
of the Server), after finishing of the first session the program waits for reconfiguration of
measurement positions. For example, changing the listener on certain measurement
position, we should first disconnect the Client and change the user name. After this and
choosing the Continue option, the next session can be started. The computer generates
logatoms and for each spoken utterance visually shows six alternative logatoms and one
transmitted logatom to be recognized (Fig.7a). The listener chooses one logatom from the list
visually presented on the computer monitor (Fig. 7b). During the tests, the listener is
confirming his response by using key from 1 to 7. Other keys are non-active during all
testing session.
After finishing all measurement sessions, the dialog window presenting results in two
options shows up:
1. Result of session nr - in this option the number of session for which the results will be
shown should be given.
2. Summary - in this option the summary of all sessions with detailed list of listeners and
measurement points shows up.


71

Fig. 7. The example of the listeners window.
5. Experiments
The goals of experiment
- decision if the results of traditional and modified with forced choice methods let finding
the relation which would allow to convert results from one method to the other and the
classification of rooms tested with both methods,
- measurement of experimental relations between traditional and modified logatom
intelligibility methods.
Taking into account the comparative character of the experiment it was planned to be done
only in a function of used logatom intelligibility method. With this end in view each
measurement was done with both traditional and modified methods, not changing:
- listeners,
- surroundings.
- measurement system (only the logatom lists),
The subjective tests were done according to Polish Standard PN-90/T-05100 with the team of
listeners made up of 12 listeners in age from 18 to 25 years. The listening team was selected
from students at Wroclaw University of Technology with normal hearing. The qualification
was based on audiometric tests of hearing threshold. The measurements of logatom
intelligibility were done using the traditional method and the MIT-FC method. The
measurements were taken in two unoccupied rooms (Fig. 8). In each room, four measure point
(Mp) were selected. These positions were chosen in the expectation of yielding a wide range of
logatom intelligibility. Sound sources (voice and white or rose noise) were positioned in the
part of the room normally used for speaking. One loudspeaker was the voice source and the
second the noise source. The various conditions were obtained by combination 14 level of
white and rose noise and four measure point. The 14 signal-to-noise ratios (SNR) were used: 39
(without noise, only background noise), 36, 33, 30, 27, 24, 21, 18, 15, 12, 9, 6, 3, 0dB. As a result
b)
a)

Automation

72
112 different transmission conditions were obtained (14 SNR*4Mp*2rooms). The speech and
noise signal levels were controlled by means of a 2606 Bruel & Kjaer instrument by measuring
them on a logarithmic scale according to correction curve A.

Fig. 8. Plan view of the rooms 1 and 2 showing source position and four receiver positions
(I loudspeaker source of the logatoms, III - loudspeaker source of the noise, 1, 2, 3, 4
measurement points).
The testing material consisted of phonetically and structurally balanced logatoms and
sentences lists uttered by professional male speaker, whose native language was Polish.
Logatom lists reproduced through the loudspeaker were recorded on the digital tape
recorder in an anechoic chamber using a linear omnidirectional microphone. The
microphone was positioned 200mm from the speakers lips. The active speech level was
controlled during recording with a meter conforming to Recommendation P.56 (ITU-T,
1993). At the beginning of each logatom set recording, 20 seconds 1000Hz calibration tone
and 30 seconds rose noise are inserted at a level equal to the mean active speech level.
For each measure point (Mp) (the place where the measure position was situated) a list of
300 logatoms has been prepared. The same logatom lists were never played for any
condition day after day. The logatom lists at the four listener locations were recorded on the
digital tape recorder. These recordings were played back over headphones to the subject
afterward. This way of subjective measurements realization provides the same listening
conditions for both traditional and with choice methods. In each room for each position of
listener (Mp) and for each signal-to-noise ratio (SNR) the logatom intelligibility was
obtained by averaging out the group of listeners results.


73
The listening tests were done in the studio of Institute of Telecommunications,
Teleinformatics and Acoustics of Wroclaw University of Technology (Fig.9). Background
noise level didnt exceed 40 dBA. Subjective measurements of logatom intelligibility were
taken in conditions of binaural listening using headphones for the optimum speech signal
level of 80 dBA.
Prior to the proper measurements the listening team was subjected to a 6 - hour training
(two 3 - h sessions). The measurement sessions duration did not exceed 3h (together with 10
min breaks after every 20 min of listening).

Fig. 9. Speech quality subjective measurement stand.
The subjective logatom intelligibility measurement results, obtained for different speech to
noise ratio, are presented graphically in fig. 10. The curves, representing the relationship
between logatom intelligibility and the signal-to-noise ratio in a room were approximated
by a four-degree polynomial calculated on the basis of the least-squarees method. The
obtained relations are presented in fig. 10. which also includes values of correlation
coefficient R
2
a measure of the conformity between the polynomial and the results
obtained from the subjective tests. As one can see there is very good agreement between the
theoretical curve and the empirical results; the value of correlation coefficient R
2
exceeds
0.99 in each case.

Automation

74

Fig. 10. Relationship between logatom intelligibility and signal-to-noise ratio (SNR) for
rooms (Logatom intelligibility measured with traditional and MIT-FC method).
For a few randomly selected measuring points the distribution of W
n,k
values was compared
with the normal distribution. The agreement between the W
n,k
distribution and the normal
distribution were tested by applying the Kolmogorov-Smirnov test. It has been found that at
significance level o = 0.05 there are no grounds to reject the hypothesis about the goodness
of fit of the distributions. Thus, it is reasonable to use average logatom hypothesis W
L
as an
estimator of the logatom intelligibility for a given measuring point.
The main goal of presented research was assessment if there exists a relation between
measurement results of traditional method and modified intelligibility test with forced
choice (MIT-FC) in rooms, and if such a relation exists, its finding out. The obtained
results revealed that there exists monotonic relation between results of traditional method
and MIT-FC. The results were presented on the surface: logatom intelligibility MIT-FC
traditional, and approximated by a fourth order polynomial. The obtained relations
are presented in fig. 11. which also includes values of correlation coefficient R
2
a
measure of the conformity between the polynomial and the results obtained from the
subjective tests.


75

Fig. 11. Relationship between logatom intelligibility measured with traditional and MIT-FC
method for rooms.
6. Conclusion
The presented MIT-FC method offers a simple, easy to use, stable, and fully automatized
speech system to assessment of speech quality in rooms. The results of the experiments
have shown that the MIT-FC method is very useful in the evaluation of speech quality in
rooms. The time needed to carry out the measurement with MIT-FC method is the same
as in traditional one but we obtain the results right after finishing the measurement
process.
The experiments carried out in finding the relations between logatom intelligibility
measured with traditional and semi-automatic with forced choice methods for the rooms
have shown that there exist the multi-value and repetitive relation between them. It allows
using both methods interchangeably and converting results between them.
The obtained relations are applied in the Institute of Telecommunications, Teleinformatics
and Acoustics of Wrocaw University of Technology to the design of subjective tests for the
verification of results yielded by a new objective method based on automatic speech
recognition techniques.

Automation

76
7. References
ANSI S 3.2, (2009), Methods for Measuring the Intelligibility of Speech over Communication
Systems, American National Standards Institute.
ANSI S 3.5, (1997), Methods for the Calculation of the Speech Intelligibility index (SII). American
National Standards Institute
Basciuk K., Brachmanski S., (1997), The Automation of the Subjective Measurements of
Logatom Intelligibility, Prep. 4407, 102-nd Convention AES, Munich, Germany,
March 22-25, 1997.
Brachmanski S., (1982), Modulation Transfer Function (MTF) as a Measure of Polish Speech
Transmission Quality (in Polish), Ph. D. Thesis, Wrocaw.
Brachmanski S., (1995), Choosing Optimum Number of Test Items in Subjective Logatom
Measurements (in Polish). Proc. XLII Open Seminar on Acoustics, pp. 423-428, ISBN
83-902146-3-6, Warszawa-Bialowieza, September 12-15, 1995
Brachmanski S., (2002) The Automation of Subjective Measurements of Speech Intelligibility
in Rooms, 112
th
Conv. AES, Munich, Germany, May 10-13, 2002.
Brachmanski S., (2004) Estimation of Logatom Intelligibility with the STI Method for Polish
Speech Transmitted via Communication Channels, Archives of Acoustics, Vol. 29,
No. 4, (2004), pp. 555-562, , ISSN 0137-5075.
Brachmanski S., (2004), The Subjective Measurements of Speech Quality in Rooms, Proc. of
Subjective and Objective Assessment of Sound, Poznan, September 1-3, 2004.
Brachmanski S., (2006), Experimental Comparison between Speech Transmission Index (STI)
and Mean Opinion Scores (MOS) in Rooms,. Archives of Acoustics, Vol .31, No.4,
(2006), pp. 171-176, ISSN 0137-5075.
Davies D.D., Davies C., (1989), Application of Speech Intelligibility to Sound Reinforcement.
J. Audio Eng. Soc., Vol. 37, No.12, (December 1989), pp. 1002-1018.
Farina A., (2001), Acoustic Quality of Theatres: Correlations between Experimental
Measures and Subjective Evaluations, Applied Acoustic, Vol. 62, No 8, (August 2001),
pp. 889-916, ISSN: 0003682X.
French, N.R., Steinberg, J.C., (1977), Factors Governing the Intelligibility of Speech Sounds, ,
In: Speech Intelligibility and Speaker Recognition, Hawley M. E., pp. 128 152,
Dowden, Hutchinson & Ross Inc., ISBN 0-470-99303-0, Stroudsburg, Pennsylvania.
Houtgast T., Steeneken H.J.M., (1973), The Modulation Transfer Function in Room Acoustics
as a Predictor of Speech Intelligibility, J. Acoust. Soc. Am. Vol. 54, No. 2, (August
1973), pp. 557-557, pp. 66-73, ISSN 0001-4966
Howard D.M, Angus J.A.S, Acoustics and Psychoacoustics, Elsevier-Focal Press, 2009, ISBN
978-0-240-52175-6, Oxford, UK.
IEC 60268-16, (2003), Sound system equipment Part16: Objective Rating of Speech Intelligibility
by Speech Transmission Index, International Electrotechnical Commission, Geneva,
Switzerland.
ISO/TR 4870, (1991), Acoustics The Construction and Calibration of Speech Intelligibility Tests.
International Organization for Standardization.
ITU-T Recommendation P.56, (1993), Objective Measurement of Active Speech Level,
International Telecommunication Union, Geneva, Switzerland.


77
ITU-T Recommendation P.800, (1996), Method for Subjective Determination of Transmission
Quality, International Telecommunication Union, Geneva, Switzerland.
ITU-T Recommendation P.830, (1996), Method for Objective and Subjective Assessment of
Quality, International Telecommunication Union, Geneva, Switzerland.
ITU-T Recommendation P.862, (2003), Perceptual Evaluation of Speech Quality (PESQ), an
Objective Method for End-To-End Speech Quality Assessment of Narrowband Telephone
Networks and Speech Codecs, International Telecommunication Union, Geneva,
Switzerland.
ITU-T Recommendation P.863, (2011), Methods for Objective and Subjective Assessment of
Speech Quality. Perceptual Objective Listening Quality Assessment, International
Telecommunication Union, Geneva, Switzerland.
ITU-R Recom. BS.1116-1, (1997), Method for the Subjective Assessment of Small Impairments in
Audio Systems Including Multichannel Sound Systems, International
Telecommunication Union, Geneva, Switzerland.
Kryter, K.D., (1962), Methods for the Calculation and Use of the Articulation Index, J. Acoust.
Soc. Am., Vol. 34, No. 11, (November 1962), pp. 1689-1697, ISSN 0001-4966.
Lam P., Hongisto V., (2006), Experimental Comparison between Speech Transmission Index,
Rapid Speech Transmission Index, and Speech Intelligibility Index, J. Acoust. Soc.
Am., Vol. 119, No. 2, (February 2006), pp.1106-1117, ISSN 0001-4966.
Majewski W., Basztura Cz., Mylecki W., (1988), Relation between Speech Intelligibility and
Subjective Scale ff Speech Transmission Quality., Proc. 7
th
FASE Symposium,
Proceedings SPEECH88, 719-726, Edinburgh, August 22-26, 1988.
Majewski W., Mylecki W., Baciuk K., Brachmaski S., (1998), Application of Modified
Logatom Intelligibility Test in Telecommunications, Audiometry and Room
Acoustics, Proceedings 9
th
Mediterranean Electrotechnical Conference Melecon98, pp.
25-28, Tel-Aviv, Israel, May 18-20, 1998.
Majewski W., Mylecki W., Brachmaski S., (2000), Methods of Assessing the Quality of
Speech Transmission (in Polish), Proceedings of 47
th
Open Seminar on Acoustics, pp.
66-75, ISBN 83-914391-0-0, Rzeszow-Jawor, September 19-22, 2000.
Mller S. Assessment and Prediction of Speech Quality in Telecommunications, Kluwer Academic
Publisher, ISBN 978-1-4419-4989-9, Dordrecht, Netherlands.
Polish Standard PN-90/T-05100, (1991), Analog Telephone Chains. Requirements and Methods of
Measuring Logatom Articulation (in Polish), Polski Komitet Normalizacyjny (Polish
Committee for Standardization), 1990, Warszawa, Poland
Polish Standard PN-V-90001, (1999), Digital Communication Systems. Requirements and
Methods for Measurement of Logatom Articulation (in Polish), Polski Komitet
Normalizacyjny (Polish Committee for Standardization), 1999, Warszawa, Poland
Sotschek J., (1976), Methoden zur Messung der Sprachgte I: Verfahren zur Bestimmung der
Satz- und der Wortverstndlichkeit, Der Fernmelde Ingenieur, Vol. 30, No. 10, pp. 1-
31.
Srinivasan S.H., (2004), Speech Quality Measure Based on Auditory Scene Analysis, Proc.
IEEE 6th Workshop on Multimedia Signal Processing, pp. 371 374, ISBN: 0-7803-8578-
0 September 29- October 1, 2004, Siena, Italy.

Automation

78
Steeneken H.J.M., Houtgast T., (1980), A Physical Method for Measuring Speech-
Transmission Quality, J. Acoust. Soc. Am., Vol. 67, No. 1, (January 1980), pp. 318
326, ISSN 0001-4966.

Steeneken H.J.M., Houtgast T., (2002), Validation of Revised STI Method, Speech
Communication, Vol. 38, No. 3-4, (November 2002), pp.413-425, ISSN:0167-6393.
5
Automation in Aviation
Antonio Chialastri
Medicair, Rome,
Italy
1. Introduction
An aircraft landed safely is the result of a huge organizational effort required to cope with a
complex system made up of humans, technology and the environment. The aviation safety
record has improved dramatically over the years to reach an unprecedented low in terms of
accidents per million take-offs, without ever achieving the zero accident target. The
introduction of automation on board airplanes must be acknowledged as one of the driving
forces behind the decline in the accident rate down to the current level.
Nevertheless, automation has solved old problems but ultimately caused new and different
types of accidents. This stems from the way in which we view safety, systems, human
contribution to accidents and, consequently, corrective actions. When it comes to aviation,
technology is not an aim in itself, but should adapt to a pre-existing environment shared by
a professional community.
The aim of this paper is to show why, when and how automation has been introduced, what
problems arise from different ways of operating, and the possible countermeasures to limit
faulty interaction between humans and machines.
This chapter is divided into four main parts:
1. Definition of automation, its advantages in ensuring safety in complex systems such as
aviation;
2. Reasons for the introduction of onboard automation, with a quick glance at the history
of accidents in aviation and the related safety paradigms;
3. Ergonomics: displays, tools, human-machine interaction emphasizing the cognitive
demands in fast-paced and complex flight situations;
4. Illustration of some case studies linked to faulty human-machine interaction.
2. What is automation
According to a shared definition of automation, the latter may be defined in the following
way: Automation is the use of control systems and information technologies to reduce the
need for human work in the production of goods and services. Another plausible
definition, well-suited the aviation domain, could be: The technique of controlling an
apparatus, a process or a system by means of electronic and/or mechanical devices that

Automation 80
replaces the human organism in the sensing, decision-making and deliberate output
(Webster, 1981).
The Oxford English Dictionary (1989) defines automation as:
1. Automatic control of the manufacture of a product through a number of successive
stages;
2. The application of automatic control to any branch of industry or science;
3. By extension, the use of electronic or mechanical devices to replace human labour.
According to Parasumaran and Sheridan, automation can be applied to four classes of
functions:
1. Information acquisition;
2. Information analysis;
3. Decision and action selection;
4. Action implementation.
Information acquisition is related to the sensing and registration of input data. These
operations are equivalent to the first human information processing stage, supporting
human sensory processes. If we adopt a decision-making model based on perception,
identification, mental process, decision, action, follow-up and feedback, information
acquisition could be likened to the first step: perception. Lets imagine a video camera
and the aid it offers in monitoring activity. It helps to replace continuous, boring,
monotonous human observation with reliable, objective and detailed data on the
environment.
Automation may handle these functions, as it is more efficient in detecting compared to
humans, while at the same time it offers the possibility of positioning and orienting the
sensory receptors, sensory processing, initial data pre-processing prior to full perception,
and selective attention (e.g.: the focus function in a camera).
Information analysis is related to cognitive functions such as working memory and
inferential processes. It involves conscious perception and manipulation of processed items.
It allows for quick retrieval of information in the working memory. In aviation, this kind of
system is broadly used to provide pilots with predictive information, such as how much fuel
will be available at destination, where the top of climb or top of descent will be in order to
optimize the flight path, and so forth.
With regard to decision and action selection, automation is useful because it involves
varying levels of augmentation or replacement of human decision-making with machine
decision-making. It is generally acknowledged that human decision-making processes are
subject to several flaws, among them a tendency to avoid algorithmic thought, a biased
development of pros and cons based on the laws of logic, a partial view of the overall
system and, often, the heavy influence of emotions.
The fourth stage involves the implementation of a response or action consistent with the
decision taken. Generally, it this stage automation replaces the human hands or voice.
Certain features in the cockpit allow automation to act as a substitute for pilots. For instance,
this occurs when following an alert and warning for windshear conditions the
automation system detects an imminent danger from a power setting beyond a pre-set

Automation in Aviation 81
threshold. In this case, the autopilot automatically performs a go-around procedure, which
avoids a further decline in the aircrafts performance.
Besides being applicable to these functions, automation has different levels corresponding to
different uses and interactions with technology, enabling the operator to choose the
optimum level to be implemented based on the operational context (Parasumaran, Sheridan,
2000). These levels are:
1. The computer offers no assistance; the human operator must perform all the tasks;
2. The computer suggests alternative ways of performing the task;
3. The computer selects one way to perform the task and
4. Executes that suggestion if the human operator approves, or
5. Allows the human operator a limited time to veto before automatic execution, or
6. Executes the suggestion automatically then necessarily informs the human operator, or
7. Executes the suggestion automatically then informs the human operator only if asked.
8. The computer selects the method, executes the task and ignores the human operator.
3. History of accidents
Automation in the aviation world plays a pivotal role nowadays. Its presence on board
airplanes is pervasive and highly useful in improving the pilots performance and
enhancing safety. Nevertheless, certain issues have emerged in the recent past that
evidence automation misuse by pilots. This could depend on a series of factors, among
them human performance, capabilities and limitations on one side, and poor ergonomics
on the other.
We should first investigate the reasons leading to the introduction of onboard automation.
During the Fifties and Sixties, the main causes of aviation accidents were believed to be
related to the human factor. The immediate cause of an accident was often to be found in
active failures, e.g. loss of control of the aircraft in which pilots failed to keep the aircraft
under control, reaching over-speed limits, stalling, excessive bank angles, etc.
In these cases the root cause was a flawed performance that eventually caused the loss of
control (the effect). Factors related to human performance, e.g. the impact of fatigue,
attention, high workload sustainability, stress mismanagement, etc. were consequently
addressed. Technological solutions were sought to help pilots manage these factors.
Innovation at that time eventually led to the introduction of the auto-pilot, auto-throttle,
flight director, etc. After the mid-Fifties, as a result of these innovations, the accident curve
dropped sharply.
Looking at the graph below, we can clearly notice the impact of such innovations on flight
safety. The vertical axis corresponds to the number of accidents per million take-offs, while
the horizontal axis corresponds to the relative decades.
As we can observe, after a dramatic improvement the accident curve rose again during the
mid-Seventies. Aviation safety experts were faced with accidents involving a perfectly
functioning aircraft, with no evidence whatsoever of malfunctions. In these cases (known as
Controlled Flight Into Terrain - CFIT), the aircrafts were hitting obstacles with the pilots
in full control. The accidents were caused by loss of situational awareness, either on the

Automation 82
horizontal or on the vertical path. The evidence showed that an improper interaction
between pilots was the main cause behind the accident, so this time the solution came from
psychology. Human factor experts developed techniques and procedures to enhance
cooperation between pilots, and specific non-technical training aspects became mandatory
for pilots. For instance, a Cockpit Resource Management (CRM) course is nowadays
mandatory in a pilots curriculum: its topics may include leadership, cross-checking and
criticizing fellow colleagues, assertiveness, resolution of conflicts, communication skills, etc.

Source: ICAO doc. 9683/950. Accident rate over the years
In the last decade, the pendulum has swung back to loss of control as a major cause of
aviation accidents, however, compared to the accidents occurring during the Fifties, the
factors leading to loss of control appear to be different. Whereas in the beginnings of
aviation, human performance was impaired by under-redundancy, that is, insufficient
aids available to pilots for avoiding the effects of factors like fatigue, distraction, workload
and stress which reduced the pilots performance, nowadays many domain experts are
pointing at possible cases of over-redundancy. This means that increasing automation
might be putting the pilot out-of-the-loop, thus causing reduced situational awareness,
automation complacency or over-confidence and loss of skills, due to lack of practice in
manually flying the aircraft. As a result, pilots may not be able to regain control once
automation has failed, or may be incapable of effectively monitoring the performance of
automated systems (and questioning it when required).
The safety philosophy behind the adoption of increasing onboard automation is based on
the assumption that human error is the main cause of accidents. Therefore, since the
human (liveware) component of the system is the flawed link in the accident chain, we
ought to look for a substitute capable of handling the tasks once performed by pilots. This
is partially true, as well see later on. It is first necessary to understand what are the pros
and cons of human contribution to safety, at what levels of operation does automation offer
undoubted advantages and where the latter should end to leave room for pilots decisions.
Pilots and machines are not alternatives, but complementary factors in ensuring flight
safety. Achieving the correct balance between these components of the aviation domain
0.5 accidents per million take-offs

benefits safety, since the role of technology appears to have reached a standstill. Until the
mid-1990s, it appears that pilots tried to adjust their behaviour to a given challenge (new
automation), which was conceived regardless of their actual need and deeply rooted
habits. In fact, one of the main drivers of the recent cockpit design philosophy was the
reduction of costs related to better performances, lower fuel consumption, cheaper
maintenance and flexible pilot training. Concern for the adaptability of pilots to these new
solutions only came at a later stage and following some severe mishaps. To achieve this
balance, well briefly analyse what levels of operation are involved in flight and where
automation primarily conceived to replace certain human operator tasks should give
way to the pilots intervention.
4. Skill, rules and knowledge
According to a paradigm proposed by Rasmussen and also developed by James Reason,
human activity can be grouped into three main fields: skill-based action, rule-based action
and knowledge-based action.
The first field is based on the human capability to accomplish physical tasks, such as
providing correct input to flight control (so-called stick and rudder training), responding
to external stimuli in a quick and consistent way, and coordinating the body in order to
obtain a desired result. It is mainly an area in which the psycho-physiological aspect is
paramount. Moreover, we could also include monitoring tasks in this field, such as
detection, identification and response to external signals stemming from habits (body
automatism or conditional reaction). Automation has played an important role at this level
by replacing human performance rather well. As a result, autopilots, auto-throttle (and later
on, auto-thrust computers) have come to gradually replace pilots in hand-flying.
Generally speaking, an autopilot can tolerate workloads that are hardly sustainable by a
pilot. Lets imagine an oceanic flight during the night; an autopilot is able to maintain (with
no effort at all) altitude, speed, track and so forth, whereas pilots are subject to tiredness,
attention lapses, distraction, etc. On the other hand, the systematic replacement of basic
flying skills has led to the erosion of competence, because, as Germans put it: Die bung
macht den meister (Use makes master). In the U.S. the FAA (Federal Aviation Authority)
has suggested adopting back to basics training, in which pilots are taught how to fly
without the help of automatisms and how to retrieve the elementary notions of
aerodynamics, in order to avoid grossly misreading altitude, speed and power.
The second level the rule base action is the conceptual layer. It indicates compliance with
the rules, norms, laws, and everything laid down in the official documentation. It is
unproblematic to apply a given rule whenever conditions warrant it. This is the case of a
limit set for a device, e.g. the maximum temperature for operating an engine (EGT). When
the upper limit is exceeded, something happens: red indications on instruments, alarms,
flashing light on to attract the pilots attention, automatic exclusion of the failed system and
so forth. A machine can easily detect whether the operating conditions are normal or
abnormal, by matching the real values with an operating envelope. Since the pilot may
forget some rules, apply them incorrectly, or fail to apply valid norms, certain functions
(especially those relating to the monitoring activity that induces boredom and complacency)
are assigned to automation. It is a consequence of automation, therefore, that the flight
engineer is no longer required in the cockpit. Some problems were initially detected in the

Automation 84
normal flying activity of newly designed cockpits, since two pilots were required to manage
a three-pilot cockpit, with automation playing the role of a silent crew-member.
The third level the knowledge-base includes the sound judgement of pilots in deciding if
and when the given rules are applicable. This implies the notion of a complex system.
Complexity is evident at every level of reality, from physics to biology, from
thermodynamics to meteorology (Morin, 2001). Different conceptions of complexity emerge
in the current scientific debate, but generally speaking, we may highlight some
commonalities between the different theories: refusal of reductionism, different level of
system description (be it physical or biological or even man-made) according to the level of
observation, emergent proprieties, etc.
Since aviation is a complex system made up of complex subsystems such as humans,
technology and the environment, it is almost impossible to govern everything in advance
through rules and norms. There will always be a mismatch between the required task and
the final outcome (Hollnagel, 2006). The resilience engineering approach to safety is aware
of this complexity and focuses on the ambivalent role in such a system of man, who
simultaneously constitutes a threat and resource in coping with unexpected events,
unforeseeable situations and flawed procedures. Much of this activity, which
continuously and strategically adapts the means to the goal, is undetected either by the
top management, or by the front-line operators themselves (pilots). These micro-
corrections are so pervasive that the person involved in accomplishing a task fails to even
realize how much he/she deviates from a given rule. The front-end operator should
always seek to compromise between efficiency and thoroughness. Hollnagel calls this
compromise the ETTO (Efficiency, Thoroughness Trade Off) principle. The paradox
emerging from the blind application of rules the so-called white strike or working by
the rule is that it leads to a paralysis of the entire production activity (Hollnagel, 2009).
The effects of such shortcuts during normal operations are another area of concern
affecting flight safety, due to the systems opacity, the operators superficial knowledge,
uncertainties and ambiguities of the operational scenario. If we turn to the initial
introduction of onboard automation, we may detect some commonalities regarding the
way pilots have coped with the innovations.
5. Evolution of automation
Automation seems to follow an evolutionary path rather than a revolutionary approach. Its
adoption on board aircrafts does not respond to the planned purpose of enhancing safety
from scratch in a consistent way, but rather resembles a biological organism trying to
continuously adapt to the challenges posed by its environment (fly-fix-fly).
This trial-and-response approach can be observed regardless of the fact that innovation
introduced on board generally lags a step behind the overall level achieved by the industry.
In fact, one of the requirements for a certain technology to be implementable in the aviation
domain is its reliability; it is preferable to have a slower yet reliable system rather than a
high-speed one that is not completely tested or tried in an operational environment.
We may identify three main generations of onboard automation systems: mechanical,
electrical and electronic.

In the beginnings of commercial flight there were no instrumental aids to help pilots to fly.
A piece of string was attached to the wing to indicate whether the airflow over it was
sufficient to sustain flight. Later on, the first anemometers and altimeters were introduced to
indicate pilots the airspeed and altitude, respectively. These were the first steps toward the
virtualization of the environment (Ralli, 1993). The invention of the pneumatic gyroscope
(replaced, shortly after, by the electric gyroscope), used to stabilize an artificial horizon,
helped pilots to understand their situation even in meteorological conditions characterised
by extremely poor visibility, while at the same time preventing dangerous vestibular
illusion (a false sense of equilibrium stemming from the inner ear). These simple
instruments were merely capable of providing basic indications. Early signs of automation
were introduced on board aircrafts during the decade from 1920 to 1930, in the form of an
autopilot based on a mechanical engineering concept that was designed to keep the aircraft
flying straight: a very basic input to control the flight at a skill level. Moreover, as
airplanes became bigger and bigger, it became necessary to apply some form of
amplification of the pilots physical force, because of the airflow over large aerodynamic
surfaces. Servo-mechanisms were introduced on board, alongside certain devices aimed at
facilitating perception of the force acting on such surfaces (artificial feel load, mach trim
compensator) and absorbing the effects of the so-called Dutch-roll, an abnormal behaviour
whereby the airplane yaws and oscillates in an uncoordinated manner (yaw damper). This
(mechanical) innovation was the first of multiple steps that began widening the gap between
the pilots input (action on the yoke) and the final outcome (aerodynamic movement).
Instead of direct control, with the yoke mechanically attached to the ailerons, airplanes
began to be constructed with a series of mechanisms intervening between the pilots input
and the expected output. In this case, the virtualization of flight controls accompanied the
parallel virtualization of flight instruments introduced by the artificial horizon (Attitude
Display Indicator). At this stage, automation aided pilots mainly in their skill-based activity.
The second generation of automation included electric devices replacing the old
mechanisms. Electric gyroscopes instead of pneumatic ones, new instruments such as the
VOR (Very High Frequency Omni-directional Range) to follow a track based on ground
aids, the ILS (Instrumental Landing System) to follow a horizontal and a vertical path till the
runway threshold, and so forth. The 1960s saw plenty of innovations introduced on board
aircrafts that enhanced safety: electric autopilots, auto-throttle (to manage the power setting
in order to maintain a selected speed, or a vertical speed), flight directors (used to show
pilots how to manoeuvre to achieve a pre-selected target such as speed, path-tracking and
so forth), airborne weather radars, navigation instruments, inertial platforms, but also
improved alarming and warning systems capable of detecting several parameters of engines
and other equipment. Whereas the first generation of automation (mechanical) managed the
pilots skill-based level, the second generation managed the skill-based and rule-based
levels previously assigned to pilots. The airplane systems were monitored through a
growing number of parameters and this gave rise to a new concern: the inflation of
information with hundreds of additional gauges and indicators inside the cockpit, reaching
almost 600 pieces (Boy, 2011). At this stage, pilots used the technology in a tactical manner.
In other words, their inputs to automation were immediately accessible, controllable and
monitored in the space of a few seconds. For example, if the pilot wanted to follow a new
heading, he would use a function provided by the autopilot. The desired heading value was
selected in the glare-shield placed in front of the pilots eye and the intended outcome

Automation 86
would be visible within a few seconds: the airplane banked to the left or right to follow the
new heading. End of the task. At this stage, automation helped pilots also at a rule-based
level, since monitoring of thousands of parameters required efficient alarming and warning
systems, as well as recovery tools.
The third generation of innovation involved electronics, and was mainly driven by the
availability of cheap, accessible, reliable and usable technology that invaded the market,
bringing the personal computer into almost every home. The electronic revolution occurring
from the mid-80s also helped to shape the new generation of pilots, who were accustomed
to dealing with the pervasive presence of technology since the early years of their life.
Electronics significantly helped to diminish the clutter of instruments on board and allowed
for replacing old indicators - gauges in the form of round-dial, black and white mechanical
indicators for every monitored parameter - with integrated coloured displays (e.g: CRT:
Cathode Ray Tube, LCD: Liquid Crystal Display) capable of providing a synthetic and
analytic view of multiple parameters in a limited area of the cockpit.
It is worthy mentioning that the type of operations implied by the Flight Management
System shifted from tactical to strategic. In fact, whilst in the previous electrical automation
stage, pilots were accustomed to receiving immediate feedback visible shortly after the
entered input, in the new version, a series of data entered by the operator would show their
effects at a distance of hours. The data was no longer immediately accessible and visible,
therefore this new way of operating placed greater emphasis on crew co-ordination, mutual
cross-checking, operational discipline, not only in the flying tasks but also in the monitoring
activity. The Flight Management System database contains an impressive amount of data,
from navigational routes, to performance capabilities and plenty of useful information that
can be retrieved from pilots. Further on we analyze the traps hidden behind this kind of
automation.
However, it is important to point out that the actual discontinuity introduced by this
generation of automation: was the notion of electronic echo-system. Compared to the past,
when pilots were acquainted with the inner logic of the systems they used, their basic
components, and normal or abnormal procedures for coping with operational events, in the
new cockpit pilots are sometimes out of the loop. This occurrence forces them to change
their attitude towards the job. On an airplane such as the A-320 there are almost 190
computers located in almost every area of the fuselage. They interact with each other
without the pilot being aware of this interaction. Every time the pilot enters an input to
obtain a desired goal (e.g. activating the hydraulic system), he/she starts a sequence in
which not only the selected system is activated, but also a number (unknown to the pilot) of
interactions between systems depending on the flight phase, operational demands, the
airplanes conditions, etc.
The unmanageable complexity of the electronic echo-system is a genuine epistemological
barrier for the pilot. Whereas before, the pilot had thorough knowledge of the entire
airplane and could strategically operate in a new and creative manner whenever
circumstances required, the evolution of the cockpit design and architecture has brought
about a new approach to flight management that is procedural and sequential. Only actions
performed in accordance with computer logic and with the given sequence are accepted by
the system. In acting as a programmer who cannot perform tasks that are not pre-planned

by the computer configuration, the pilot has lost most of his expertise concerning the
hardware part of the system (the aircraft). He too is constrained by the inner logic which
dictates the timing of operations, even in high-tempo situations. Consequently, in the
international debate on automation and the role of pilots, the latter are often referred to as
system operators. This holds true up to a certain point but, generally speaking, it is an
incorrect assessment. Recalling what has been said about the levels of operation, we may say
that at the skill-based level, pilot have become system operators because flying skills are
now oriented to flight management system programming. As a flight instructor once said:
we now fly with our fingers, rather than with a hands-on method. What he meant was
that the pilot now appears to push-the-button rather than govern the yoke and throttle.
At the skill-based level, automatisms may be in charge for the entire flight, relegating the
pilot to a monitoring role. However, at a rule-based level, computers manage several tasks
once accomplished by pilots, including monitoring the pressurization system, air
conditioning system, pneumatic system and so forth. This statement is no longer valid when
referred to the knowledge-based level. Here, the pilot cannot be replaced by any computer,
no matter how sophisticated the latter is. Sound judgment of an expert pilot is the result of a
series of experiences in which he/she fills the gap between procedures and reality. Since this
paper discusses automation rather than the human factor in aviation, anyone wanting to
investigate concepts such as flexibility, dealing with the unexpected, robustness, etc. may
refer to authors adopting a Resilience Engineering approach (Hollnagel, 2008) (Woods,
Dekker, 2010).
The fact that pilots are no longer so acquainted with the airplane as in the past, has led them
to only adopt the procedural way to interact with the airplane. This is time-consuming,
cognitively demanding, and above all, in some cases it may lead to miss the big picture, or
situation awareness. This new situation introduces two major consequences: automation
intimidation (ICAO, 1998) and a restriction of the available tools for coping with unexpected
events. In this sense, we can compare the natural echo-system made up of a complex
network of interactions, integrations, retro-actions that make it mostly unforeseeable and
unpredictable with the electronic system. On a modern aircraft, the electronic echo-system
is mostly not known by its user. Pilots only know which button they have to press, what the
probable outcome is, and ignore what lies is in between. It is a new way of operating that
has pros and cons. To assess the real impact on safety, we ought to analyze the relationship
between man and technology, by shifting from HMI (Human Machine Interaction) to HCI
(Human Computer Interaction). A further step will lead to Human Machine Engineering
and/or Human Machine Design (Boy, 2011). The core of this approach is that the focus of
the research should be on human-centred design; in other words, the final user with
his/her mental patterns activated in real scenarios should be regarded as the core of the
entire project for a new form of automation.
From an engineering perspective, it is very strange that concepts routinely used to describe
the role of onboard automation miss a basic focal point: the final user. In fact, when we talk
about technology, we refer to bolt-on versus built-in systems. These expressions indicate the
different pattern of integration of onboard technology. Bolt-on indicates the introduction of
a new technology on board an airplane conceived without automation. It is a reactive mode
that strives to combine the old engineering philosophy with new devices. It is a kind of
patchwork which requires several local adjustments to combine new requirements with

Automation 88
old capabilities. The expression built-in indicates the development of a new technology
incorporated in the original project. Every function is integrated with the airplanes systems,
making every action consistent with the original philosophy of use. The paradox is that the
final user is left out of the original project. Nowadays, after several avoidable accidents,
pilots are involved in the early stage of design in order to produce a user-friendly airplane.
6. Why automation?
Two main reasons led to the decision to adopt onboard automation: the elimination of
human error and economic aspects. The first element stems from the general view whereby
human performance is regarded as a threat to safety. As such a topic would require a paper
in itself, it is more appropriate to briefly mention some references for students eager to
investigate the topic thoroughly. The second element is easier to tackle since we can even
quantify the real savings related to, say, lower fuel consumption. According to IATA
estimates, a one percent reduction in fuel consumption translates into annual savings
amounting to 100,000,000 dollars a year for IATA carriers of a particular State. (ICAO,
1998). Aside from fuel, the evolution of onboard technology over the years has led to a
dramatic improvement in safety, operational costs, workload reduction, job satisfaction, and
so forth. The introduction of the glass cockpit concept allows airlines to reduce maintenance
and overhauling costs, improve operational capabilities and ensure higher flexibility in pilot
training.
a. Fuel consumption
A crucial item in an airlines balance sheet is the fuel cost. Saving on fuel is vital to remain
competitive on the market. The introduction of the fly-by-wire concept helps to reduce
fuel consumption in at least three areas: weight, balance and data predictions.
1. The fly-by-wire concept has brought a tangible innovation. Inputs coming from the
pilots control stick are no longer conveyed via cables and rods directly to the
aerodynamic surfaces. In fact, the side-stick (or other devices designed to meet
pilots demands regarding a conventional yoke) provides input to a computer
which via optic fibres sends a message to another computer placed near the
ailerons or stabilizer. This computer provides input to a servo-mechanism to move
the surfaces. Therefore, there is no longer any need for steel cables running through
the fuselage and other weighty devices such as rods, wheels, etc. This also
significantly reduces the aircrafts weight and improves fuel consumption, since
less power is required to generate the required lift.
2. The second area that contributes to saving fuel is aircraft balance. The aircraft must
be balanced to maintain longitudinal stability (pitch axis in equilibrium). This
equilibrium may be stable, unstable or neutral. In a stable aircraft, the weight is
concentrated in front of the mean aerodynamic chord. Basically, this means that the
stabilizer (the tail) should push-down (or, technically, induce de-lifting) to
compensate for the wing movements. In an unstable equilibrium, the balance of
weight is shifted sensibly backwards compared to a stable aircraft. In other words,
the stabilizer should generate lift to compensate for the wing movements. Where
does the problem lie with unstable aircrafts? A stable aircraft tends to return to its
original state of equilibrium after it deviates from the latter, but is less

manoeuvrable since the excursion of the stabilizer is narrower. On the contrary,
unstable equilibrium causes an increasingly greater magnitude of oscillations as it
deviates from the initial point. It makes the aircraft more manoeuvrable but
unstable. In practical terms, in these kinds of airplanes pilots are required to make
continuous corrections in order to keep the aircraft steady. This is why the
computer was introduced to stabilize the airplane with continuous micro-
corrections. This significantly reduces the pilots workload for flying smoothly.
Moreover, due to the distribution of weight concentrated on the mean aerodynamic
chord, an unstable aircraft consumes less fuel.
3. The third factor helping pilots is a database capable of computing in real time any
variation to the flight plan either on the horizontal path (alternative routes, short-
cut, mileage calculations, etc.) or on the vertical profile (optimum altitude, top of
descent to manage a low drag approach, best consumption speed, and so forth).
This enhances the crews decision-making task in choosing the best option in order
to save fuel.
b. Maintenance costs
The glass cockpit concept enables airlines to reduce maintenance and overhauling costs. In
conventional airplanes, every instrument had its box and spare part in the hangar.
Whenever a malfunction was reported by the crew, maintenance personnel on the ground
fixed it by replacing the apparatus or swapping the devices. All these actions required a new
component for every instrument. If we consider that an airliner has roughly one million
spare parts, we can easily understand the economic breakthrough offered by the glass
cockpit concept.
In these airplanes, a single computer gives inputs to several displays or instruments. The
maintenance approach is to change a single computer rather than every component or
actuator. Based to this operating method, few spare parts are required in the hangar: no
more altimeters, no more speed indicators, no more navigation displays (often supplied
different manufacturers). Moreover, training of maintenance personnel is simplified as it
focuses on a few items only which, in turn, allows for increased personnel specialisation.
c. Selection and Training costs
The fast growth pace of the airline industry over the last decades has generated concern
about the replacement of older pilots, since training centres cannot provide the necessary
output for airline requirements. Hiring pilots from a limited base of skilled workers creates
a bottleneck in the industrial supply of such an essential organizational factor as are pilots
for an airline. Automation has facilitated the hiring of new pilots, since the basic skills are no
longer a crucial item to be verified in the initial phase of a pilots career. If, barely thirty
years ago, it would not have been sufficient to have walked on the moon to be hired by a
major airline, as an expert pilot ironically put it, nowadays the number of would-be pilots
has increased exponentially. In the current industrial philosophy, almost everyone would be
able to fly a large airliner safely with a short amount of training. This phenomenon gives
rise to new and urgent problems, as well now see.
Besides the selection advantages, broadening the potential pilot base enables airlines to save
money for the recurrent training of pilots or to reduce transition costs. Indeed, once the
manufacturer sets up a standardized cockpit display, the latter is then applied to a series of

Automation 90
aircraft. If we look at the Airbus series comprising A-319, A-320, A-321, A-330, A-340, etc.,
we realize how easy is to switch from one airplane to another. In this case, the transition
course costs are significantly lower since pilots require fewer lessons; indeed, aside from
certain specific details, the only difference involves the performance (take-off weight, cruise
speed, landing distances, etc.).
Since pilot training costs make up a considerable portion of an airlines budget, it is
financially convenient to purchase a uniform fleet made up of same family of airplanes.
Often, the regulations enable airlines to use a pilot on more than one airplane belonging to a
family. Consequently, this leads to shorter transition courses, operational flexibility as
pilots get to fly several aircrafts at a time, and in the long-run, better standardization among
pilots.
Some authors have pointed out how automation has redefined the need for different
training processes and crew interaction (Dekker, 2000).
d. Operational flexibility
A pilot flying with no aids at all, be they mechanical, electrical or electronic, is limited in
many ways.
He/she must fly at low altitude, because of his/her physiological limits (hypoxia), he/she
cannot fly too fast since the effort on the yoke exceeds his/her physical power, he/she
cannot even fly in bad weather (clouds or poor visibility) since he/she must maintain visual
contact with the ground. Automation allows for overcoming such limitations. Higher flight
levels also mean lower fuel consumption and the possibility of flying out of clouds. Faster
speeds allow for reaching the destination earlier and completing multiple flights a day.
Onboard instruments enable pilots to achieve better performance. Lets imagine an
approach in low visibility conditions. In the beginnings, pilots would rely on a Non-
Directional Beacon (NDB) as an aid to find the final track to land on the runway. Safety
measures were implemented such as the operating minima. These implied that during a
final landing approach, the pilot had to identify the runway before reaching a certain
altitude (landing minima). Obviously, as instruments became more reliable, the landing
minima were lowered. Subsequently, the introduction of the VOR (Very High Frequency
Omnidirectional Range), which provided a more accurate signal, allowed for performing an
approach closer to the runway and at a lower decision altitude, as the safety margins were
assured. When the ILS (Instrumental Landing System) was introduced on board, pilots
could also rely on vertical profile indications. This implied higher safety margins that led
the regulatory bodies to once again lower the landing minima. As the ILS became
increasingly precise and accurate, the landing minima were lowered till they reached the
value of zero. This means that an airplane can touch down with such low visibility that
pilots must rely on autopilot to perform such an approach. This is due to human limitations
such as visual and vestibular illusions (white-out, wall of fog, duck under, etc.) that may
impair the pilots performance. Lets imagine an approach during the 50s: with a visibility
of 1,000 metres at the destination airport, the airplane should have diverted to another
airport because the pilots would not have been able to attain the airport visual reference at
the decision altitude (landing minima). Nowadays, the same airport could operate with 100
metres visibility due to improved transmitting apparatus on the ground and flight
automation that enhances pilot performance.

A problem evidenced by several authors concerns the shift in responsibility of the people
managing the automation system. With the performances brought about by automation that
exceed human capabilities, a failure in the automatic system places pilots in a situation in
which they have no resources available for coping with an unexpected event. In one case,
where the pilots were performing a low visibility approach, the automatic system went out
of control at very low altitude, causing the airplane to pitch down and hit hard on the
runway. Experts acknowledge that the pilots recovery in that case was beyond reasonable
intervention, that is to say impossible (Dismukes, Berman, Loukopoulos, 2008). Indeed,
the time available for detecting, understanding and intervening was so short that it was
virtually impossible to solve the airplanes faulty behaviour. Who is responsible in such a
case? Is it correct to refer to the pilots mismanagement, poor skill, or untimely reaction
when they are operating outside their safety boundaries? If the operations are conducted
beyond human capabilities we should also review the concept of responsibility when coping
with automation. As Boy and Grote put it: Human control over technical systems, including
transparency, predictability, and sufficient means of influencing the systems, is considered to be the
main prerequisite for responsibility and accountability (Boy, 2011).

7. Different cockpit presentation
In recent years, cockpit design has undergone somewhat of a revolution. Old aircraft were
designed to satisfy pilots needs and relied upon slight modifications in the previous
onboard scanning pattern. A common standardized cockpit display emerged in the early
50s: the T-model. It included the basic instruments such as artificial horizon (in the top-
centre), anemometer (speed indicator) on the left, the altimeter on the right and compass
(indicating the heading and track) in the bottom-centre position. Two further instruments,
the side-slip indicator and vertical speed indicator were added later on. The adjacent
picture shows the classic T-model.
ClassicTmodel

Automation 92

With the introduction of the glass cockpit, the traditional flight instrument display was
replaced by a different presentation encompassing more information, greater flexibility,
colour coding and marking, but at the same time, it could lead to information overload, as
several parameters are displayed in a compact area. In a single display, known as the PFD
(Primary Flight Display), multiple information is included not only for basic parameters, but
also for navigation functions, approach facilities, automatic flight feedback, flight mode
awareness, and so forth.
Since the pilot needs to cross-check a great number of instruments at a glance, the design of
traditional instruments was developed according to a pattern of attention, from the more
important information to the more marginal information. Each instrument had its own case
and its functional dynamics was perfectly known by the pilot; it was easy to detect, easy to
use during normal operations and easy to handle in case of failure. The switch positions and
shapes were positioned on board according to a pattern of use and were instantly
recognisable at the touch. It was common for pilots to perform the checklist according to the
touch and feel principle, by detecting whether a switch was in the required configuration
by confirming its on/off position. Almost every switch had a peculiar shape: i.e., rough and
big for operative ones, smooth and small for less important switches. A form of training
implemented in most flying schools was the blind panel exercise, which consisted in
covering the trainees eye with a bandage and asking him/her to activate the switch
prompted by the instructor. Being familiar with the physical ergonomics of the cockpit
proved very helpful. The pilot knew how much strength was required to activate a
command, how much the arm had to be stretched to reach a knob and so forth. This exercise
enhanced the pilots skill, or the skill level of operation (according to Rasmussens Skill-
Primary flight Display

Rule-Knowledge paradigms). Regarding the input to the flight controls, it was assured via
flight control wheels and sticks, cables and rods, in order to enable the pilot to intervene
directly on the aerodynamic surfaces such as the ailerons, rudder and stabilizer.
With regard to displays, the traditional instrument configuration made up of drum
pointers, single instrument boxes and touch and feel overhead panel was replaced by a
new approach towards onboard information available to the pilot. The touch and feel
philosophy was replaced by the dark cockpit approach. A dark, flat, overhead panel was
adopted to show the pilot that everything was OK. Failures or anomalies were detected by
an illuminated button, recalled by a master caution/warning light just in front of the pilots
eyes. In the unlikely (but not impossible) situation of smoke in the cockpit severely
impairing the pilots sight, (due to the concentration of a thick layer of dense, white fog),
pilots of conventional airplanes could retrieve the intended systems configuration by
heart, by detecting the switch position with a fingertip.
Vice-versa, lets imagine a dark panel with smoke on board. It is truly challenging to spot
where to place your hand and, above all, what the systems feedback is, since the system
configuration is not given by the switch position but by an ON/OFF light, which is often
undetectable.
8. Ergonomics
The problem with innovation, especially in a critical context for safety as is aviation, lies in
the interaction between a community of practice and a new concept, conceived and
implemented by engineers. This implies a relationship between automation and ergonomics.
Ergonomics is a word deriving from the Greek words ergon (work) and nomos (law), and it is
a field of study aimed at improving work conditions so as to guarantee optimal adaptation
of the worker to his/her environment. We may identify three main types of ergonomics:
physical, cognitive and social (or organizational).
Initially, with the onset of the first generation of automation (mechanical), ergonomists
studied physical ergonomics, namely: how to reach a control, how much force is required to
operate a lever, visibility of displayed information , seat position design, and so forth. For
example, several accidents occurred due to misuse of the flap lever and landing gear lever.
Indeed, these were positioned near one another and had similar shapes, leading the pilots to
mistake them (Koonce, 2002). The ergonomists found a straightforward and brilliant
solution: they attached a little round-shaped rubber wheel to the lever tip to indicate the
landing gear, and a wing-shaped plastic cover indicating the flaps. Moreover, the two levers
were separated in order to avoid any misuse.
In the next generation of automation, namely electrical automation, the ergonomists aimed
to improve the cockpit design standardization. A pilot flying on an airplane had difficulty
in forming a mental image of the various levers, knobs and displays, as every manufacturer
arranged the cockpit according to different criteria. Dekker has clearly illustrated a case of
airplane standardization: the position of the propeller, engine thrust and carburettor in a
cockpit (Dekker, 2006). Pilots transitioning to other aircrafts struggled to remember the
position of every lever since they were adjacent to each other, and the possibility of
mistaking them was very high. Moreover, during the training process there was a risk of

Automation 94
the so-called negative transfer, that is, the incorrect application of a procedure no longer
suited to the new context. For example, the switches on Boeing aircrafts have top-down
activation, while Airbus has chosen the swept-on system (bottom-up). A pilot transitioning
from a Boeing 737 to an Airbus A-320 may quickly learn how to switch on the systems, but
in conditions of stress, fast-paced rhythm and high workload, he/she may return to old
habits and implement them incorrectly. Social ergonomics tries to eliminate such
difficulties.
Cognitive ergonomics studies the adaptability of the technology to the mental patterns of
the operator. The third generation of automation introduced on board aircrafts gave rise to
many concerns about whether the instrument logic was suitable for being correctly used by
pilots. From a cognitive perspective, the new system should be designed according to the
users need, while bearing in mind that pilot-friendly is not equivalent to user-friendly
(Chialastri, 2010). A generic user is not supposed to fly, as some engineers erroneously tend
to believe. Pilots belong to a professional community that share a mental pattern when
coping with critical operational situations. The modus operandi is the result of a life-long in-
flight experience. A professional community such as the pilot community is target-oriented,
and aims at the final goal rather than observing all the required procedural tasks step-by-
step. This is why we talk about shortcuts, heuristics, and so forth.
For example purposes, lets discuss the introduction of a new airplane with variable wings:
the F-111. It was conceived by the designers with a lever in the cockpit to modulate the
wings from straight (useful at low speed) to swept (to fly at high speed). In the mind of the
designer (who is not a pilot), it seemed perfectly sensible to associate the forward movement
of the lever in the cockpit with the forward movement of the wings: lever forward straight
wings, lever back swept wings. When the airplane was introduced in flight operations,
something strange occurred. Pilots associated the forward movement with the concept of
speed. So, when they wanted high speed, they pushed the thrust levers forward, the yoke
down to dive and, consequently, the wing-lever forward (incorrectly so). For a thorough
analysis of this case, see Dekker (2006).
All this occurred due to the confusion between generic users and specialized users, as are
pilots. As Parasumaran puts it: Automation does not simply supplant human activity but
rather changes it, often in ways unintended and unanticipated by the designers of
automation and, as a result, poses new coordination demands on the human operator
(Parasumaran, Sheridan, 2000).
9. Automation surprise
The interaction between pilots and technologies on board the aircraft raises some concerns
regarding an acknowledged problem: the automation surprise. This occurs when pilots no
longer know what the system is doing, why it is doing what it does and what it will do next.
It is an awful sensation for a pilot to feel to be lagging behind the airplane, as ever since
the early stages of flight school every pilot is taught that he/she should be five minutes
ahead of the airplane in order to manage the rapid changes occurring during the final part
of the flight. When approaching the runway, the airplanes configuration demands a higher
workload, the communication flow with air traffic controllers increases and the proximity to
the ground absorbs much of the pilots attention.

Once all the fast-pace activities are handled with the aid of automation, the pilot may shed
some of the workload. When automation fails or behaves in a strange manner, the
workload increases exponentially. This is only a matter of the effort required to cope with
automation, and could be solved with additional training. A more complicated issue
arising from investigations on some accidents is the difference between two concepts,
namely: situational surprise and fundamental surprise (Wears, 2011). To summarise our
explanation, we may refer to the example mentioned by Wears regarding an everyday
situation: a wife returns home unexpectedly and finds her husband in bed with the waitress.
She is shocked and the only thing she can say is: Darling, I am surprised. Her husband
calmly replies, no, darling, I am surprised; you are astonished. In this example, the
husband knows he is doing something regrettable, but accepts the risk even if he doesnt
expect his wife to return home. He is surprised, but has calculated even the worst case (wife
back home sooner than expected). The wife comes home and finds a situation she is not
prepared for. She is surprised and doesnt know what to do at first, because she lacks a plan.
No doubt she will find one
Therefore, situation surprise arises when something occurs in a pre-defined context, in
which the pilots know how to manage the situation. Such a situation is unexpected, being an
abnormal condition, but nonetheless remains within the realm of the pilots knowledge.
Moreover, it returns to being a manageable condition once the pilot acknowledges the
difference between a normal and an abnormal condition, by applying the relevant abnormal
procedure. After the initial surprise due to the unexpected event, the pilot returns to a
familiar pattern of operation acquired with training.
A so-called fundamental surprise is a very different circumstance: it implies a thorough re-
evaluation of the situation starting from the basic assumptions. It is an entirely new
situation for which there is no quick-fix recipe and is difficult to assess, since it is
unexampled. The pilots literally do not know what to do because they lack any cognitive
map of the given situation. This concept can be well summarised by the findings following
an accident involving a new generation airplane. When analysing the black box after an
accident, it is not rare to hear comments such as: I dont know whats happening this is a
something of a pilots nightmare.
It should be noted that some failures occurring on board highly automated aircrafts are
transient and are very difficult to reproduce or even diagnose on the ground. Even the
manufacturers statements regarding electrical failures are self-explaining, as a failure may
produce different effects in each case or appear in different configurations depending on the
flight phase, airplane configuration, speed and so forth. Sometimes, an electrical transient
(also influenced by Portable Electrical Devices activated by unaware passengers) could
trigger certain airplane reactions that leave no memory of the root cause.
10. Automation issues
Although automation has improved the overall safety level in aviation, some problems tend
to emerge from a new way of operating. A special commission was set up in 1985 by the
Society of Automotive Engineers to determine the pros and cons arising from flight deck
automation. Nine categories were identified: situation awareness, automation complacency,
automation intimidation, captains command authority, crew interface design, pilot

Automation 96
selection, training and procedures, the role of pilots in automated aircrafts (ICAO, 1998).
Sometimes, pilots loose situation awareness because they lag behind the automation logic.
A pilot should always know what the automation system is going to do in the next five
minutes, to either detect any anomalies or take control of the airplane. In fast-paced
situations such as crowded skies, proximity with the terrain and instrumental procedures
carried out in poor visibility a high number of parameters must be monitored. As the
workload increases, the pilot tends to delegate to the automation system a series of
functions in order to minimise the workload. It is important to specify that the physical
workload (the number of actions performed within a given time frame) differs from the
cognitive workload in that the latter implies a thorough monitoring, understanding and
evaluation of the data coming from the automation system. The paradox involving airplane
automation is that it works as an amplifier: with low workloads, it could lead to
complacency (let the automation system do it) that reduces alertness and awareness,
while the latter increase with high workloads, due to the high number of interactions and
data involved in fast-paced situations. Plainly said: When good, better; when bad, worse.
Poor user interface design is another issue. Norman, Billings et al. have studied the way
humans interact with automation and have developed cognitive ergonomics. There are
several studies concerning the optimal presentation (displays, alarms, indicators, etc.) for
pilot use. For example, old displays were designed in such a way that each single
instrument provided data for a specific domain only: the anemometer for speed, the
altimeter for altitude and flight levels, and so forth. With the integrated instruments, all the
required data is available at a glance in a single instrument, which not only provides colour-
coded information but also on-demand information by means of pop-up features, so that it
can be concealed during normal operations and recalled when needed.

Several issues are related to displays, but only two are addressed here: cognitive mapping of
instruments and human limitations in applying the colour-code to perception.
We have to consider that the replacement of round-dial black and white instruments, such
as the old anemometer, with coloured multi-function CRT (Cathodic Ray Tube) instruments,
has determined a new approach by pilots in mentally analysing speed data. Indeed, with
round-dial indicators, pilots knew at a glance in which speed area they were flying: danger
area, safe operations or limit speed. Although it is not important to know the exact speed,
according to a Gestalt theory concept we need to see the forest rather than every single tree.

Furthermore, from a cognitive perspective, we tend to perceive the configuration as a
multiple instrument: for example, during the approach phase, the anemometer needle will
point to 6 oclock, the vertical speed indicator to 8 oclock and so forth.
In the new speed indicator, the values appear on a vertical speed tape rolling up and down,
with the actual speed magnified for greater clarity. Moreover, additional indicators are
displayed on the speed tape, such as minimum speed, flap retraction speed and maximum
speed for every configuration. In many ways, this system facilitates the pilots task.
Problems arise whenever failures occur (electrical failure, unreliable indication etc.), as the
markings disappear.

The situation worsens compared to the old indicators, as pilots needs to create a mental
image of the speed field in which they are flying. For example, it is unimportant for us to
know whether we are flying at 245 or 247 knots, but it becomes important to know that we
are flying at 145 knots rather than 245 knots. This means that the mental mapping of the
speed indication involves a higher workload, is time-consuming and also energy-
consuming. Lets take a look at these two different speed indicators to grasp the concept.
The second issue is related to human performance and limitations. The human eye may
detect colours, shape, light and movement with foveal vision. There are two kinds of
receptors in the fovea: the cones and the rods.
The cones are responsible for detecting shapes and colours, in a roughly 1 cent large area,
and account for focused vision. Instead, rods are responsible for peripheral vision and are
able to detect movement and light. This means that colour-coded information should appear
in an area just in front of the pilots head, as only the cones may such information. In fact,
since an input appearing to the side can be detected by the rods, which are insensitive to
colour and shape, the pilot will not notice it unless he stares at it directly.
Other relevant indicators for pilots are symmetry and context. The needle in a round-dial
instrument satisfies both these requirements, since we may easily detect which field of
operation we are currently in (near the upper limit, in the middle, in the lower margin), in
addition to the trend (fast-moving, erratic, gradual) and symmetry with nearby parameters.
A digital indicator shows this information in an alternative way, by arranging numbers on a
digital display, as can be clearly seen in the above picture. The configuration on the left can
be grasped at first glance, while the second set of information requires a considerable effort
to detect differences.

Automation 98

Disrupted symmetry Digital information
Another area worthy of attention is the greater magnitude of the errors made by pilots when
flying highly automated airplanes. Pilots flying older aircrafts normally relied on their
ability to obtain the required data using a heuristic and rule-of-thumb approach. This
method was not extremely accurate but roughly precise. Nowadays, with the all-round
presence of computers on board, flight data can be processed very precisely but with the
risk of gross errors due to inaccurate entries made by pilots. An example taken from
everyday life may help to explain the situation: when setting older-version alarm clocks, the
users error might have been limited to within the range of a few minutes, whereas the new
digital alarm clocks are very accurate but subject to gross errors for example 8 PM could be
mistaken for 8AM.
Lower situation awareness means that pilots no longer have the big picture of the available
data. The aviation truism, trash in-trash out, is applicable to this new technology as well.
Having lost the habit of looking for the frame, namely the context in which the
automation systems computed data should appear, pilots risk losing the big picture and,
eventually, situation awareness. Unsurprisingly, there have been cases of accidents caused
by inconsistent data provided by the computer and uncritically accepted by pilots. This is
neither a strange nor uncommon occurrence. Fatigue, distractions, and heavy workload may
cause pilots to lower their attention threshold. In one case, a pilot on a long haul flight
entered the available parameters into the computer to obtain the take-off performance
(maximum weight allowed by the runway tables, speed and flap setting). He mistook the
take-off weight with other data and entered the parameters into the computer. The end
result was a wrong take-off weight, wrong speed and wrong flap setting. The aircraft
eventually overrun the runway and crashed a few hundred meters beyond the airport
fences.
Another issue linked the introduction of onboard automation is a weakening of the
hierarchy implied by the different way of using the automation system. The captain is
indeed the person with the greatest responsibility for the flight and this implies a
hierarchical order to establish who has the final word on board. This hierarchical order is
also accompanied by functional task sharing, whereby on every flight there is a pilot flying
and a pilot not flying (or monitoring pilot). Airplanes with a high level of automation
require task sharing, in which the pilot flying has a great degree of autonomy in
programming the Flight Management System, in deciding the intended flight path and type

of approach. Above all, the pilot flying also determines the timing of the collaboration
offered by the pilot not flying, even in an emergency situation. In fact, since the crew must
act in a procedural way in fulfilling the demands of the automation system, both pilots must
cooperate in a more horizontal way compared to the past. The hierarchical relationship
between captain and co-pilot is known as the trans-authority gradient (Hawkins, 1987)
and, from a human factor perspective, should not be too flat nor too steep.
11. Case studies
Currently, several case studies could be cited to illustrate the relationship between pilots
and automation. In the recent past, accidents have occurred due to lack of mode awareness
(A-320 on Mount St. Odile), automation misuse (Delhi, 1999), loss of braking leading to loss
of control (S. Paulo Garulhos, 2006), loss of control during approach (B-737 in Amsterdam,
2008), and several other cases.
From these accidents, we chose a couple of peculiar cases. The first case occurred in 1991
and involved an Airbus flying at cruise altitude. While the captain was in the passengers
cabin, the co-pilot tried to play with the FMS in order to learn hands-on. He deselected
some radio-aids (VOR: Very High Frequency Omni-directional Range) from the planned
route. After the tenth VOR was deselected, the airplane de-pressurized, forcing the pilot to
perform an emergency descent. From an engineering perspective there was a bug in the
system but, moreover, it was inconceivable for the pilot to link the VOR deselection (a
navigation function) to the pressurization system.
The second case is related to a B-727 performing a low-visibility approach to Denver. The
flight was uneventful till the final phase. The sky was clear but the city was covered by a
layer of fog that reduced the horizontal visibility to 350 ft. and the vertical visibility to 500 ft.
The captain was the pilot flying, in accordance with the company regulations, and used the
autopilot as specified in the operating manual. The chosen type of approach was an ILS Cat
II. The ILS is a ground based navigation aid that emits a horizontal signal to guide the
airplane to the centre of the runway together with a vertical signal to guide the aircraft
along a certain slope (usually 3) in order to cross the runway threshold at a height of 50 ft.
The regulations state that once the aircraft is at 100 ft., the crew must acquire the visual
references to positively identify the runway lights (or markings). If not, a go-around is
mandatory. Once the captain has identified the runway lights he must proceed to land
manually, by switching off the autopilot at very low altitude. There are certain risks
associated with this manoeuvre, since with impaired visibility the pilot loses the horizon,
depth perception and the vertical speed sensation, and could be subject to spatial
disorientation due to the consequences of the so-called white-out effect. In this case,
however, nothing similar occurred. The autopilot duly followed the ILS signals, but due to a
random signal emitted in the last 200 ft., the aircraft pitched down abruptly. The captain
tried to take over the controls, though unsuccessfully. He later reported during the
investigation that the window was suddenly full of lights, meaning that the aircraft
assumed a very nose-down pitch attitude. The high rate of descent, coupled with the
surprise factor and low visibility, did not allow the captain to recover such a degraded
situation: the airplane touched down so hard that it veered off the runway, irreversibly
damaging the fuselage.

Automation 100
This case emphasizes the relationship between pilots and liability. Operating beyond human
capabilities, the pilot finds himself/herself in a no-mans land where he/she is held
responsible even when he/she cannot sensibly regain control of the airplane. As stated by
Dismukes and Berman, it is hardly possible to recover from such a situation.
12. Conclusion
This brief paper highlights certain aspects concerning the introduction of automation on
board aircrafts. Automation has undeniably led to an improvement in flight safety.
Nevertheless, to enhance its ability to assure due and consistent help to pilots, automation
itself should be investigated more thoroughly to determine whether it is suitable in terms of
human capabilities and limitations, ergonomics, cognitive suitability and instrument
standardization, in order to gradually improve performance.
13. References
Air Pilot Manual, (2011), Human Factors and pilot performance, Pooleys, England
Alderson D.L., Doyle J.C., Contrasting view of Complexity and Their Implications For
Network-Centric Infrastructures, in IEEE Transaction on Systems, Man and
Cybernetics part A: Systems and Humans, vol 40 No. 4, July 2010
Amalberti R., The paradox of the ultra-safe systems, in Flight Safety Australia, September-
October 2000
Bagnara S., Pozzi S., (2008). Fondamenti, Storia e Tendenze dell'HCI. In A. Soro (Ed.),
Human Computer Interaction. Fondamenti e Prospettive. Monza, Italy: Polimetrica
International Scientific Publisher.
Lisanne Bainbridge, (1987), The Ironies of Automation, in New technology and human
error, J. Rasmussen, K. Duncan and J. Leplat, Eds. London, UK: Wiley,.
Barnes C., Elliott L.R., Coovert M.D., Harville D., (2004), Effects of Fatigue on Simulation-
based Team Decision Making Performance, Ergometrika volume 4, Brooks City-
Base, San Antonio TX
Boy G., a cura di, (2011), The Handbook of human Machine Interface A Human-centered design
approach, Ashgate, Surrey, England
Chialastri Antonio (2011), Human-centred design in aviation, in Proceedings of the Fourth
Workshop on Human Centered Processes, Genova, February 10-11
Chialastri Antonio (2011), Resilience and Ergonomics in Aviation, in Proceedings of the
fourth Resilience Engineering Symposium June 8-10, 2011, Mines ParisTech, Paris
Chialastri Antonio (2010), Virtual Room: a case study in the training of pilots, HCI aero-
conference, Cape Canaveral
Cooper, G.E., White, M.D., & Lauber, J.K. (Eds.) (1980) "Resource management on the
flightdeck," Proceedings of a NASA/Industry Workshop (NASA CP-2120)
Dekker S., Sharing the Burden of Flight Deck Automation Training, in The International
Journal of Aviation Psychology, 10(4), 317326 Copyright 2000, Lawrence
Erlbaum Associates, Inc.
Dekker S. (2003), Human Factor in aviation: a natural history, Technical Report 02 - Lund
University School of Aviation
Dekker S., Why We need new accident models, Lund University School of
Aviation,Technical Report 2005-02

Dismukes, Berman, Loukopoulos (2008), The limits of expertise, Ashgate, Aldershot,
Hampshire
Ferlazzo F. (2005), Metodi in ergonomia cognitiva, Carocci, Roma
Flight Safety Foundation (2003), The Human Factors Implications for Flight Safety of
Recent Development In the Airline Industry, in Flight Safety Digest, March-April
Hawkins Frank (1987), Human factor in Flight, Ashgate, Aldershot Hampshire
Hollnagel E., Woods D., Leveson N. (2006) (a cura di), Resilience Engineering Concepts and
Precepts, Ashgate, Aldershot Hampshire
Hollnagel Erik, (2008) Critical Information Infrastructures : should models represent
structures or functions ?, in Computer Safety, Reliability and Security, Springer,
Heidelberg
Hollnagel Erik, (2009), The ETTO Principle Efficiency-Thoroughness Trade-Off, Ashgate,
Surrey, England
Hutchins Edwin (1995), How a cockpit remembers its speeds, Cognitive Science, n. 19, pp.
265-288.
IATA (1994), Aircraft Automation Report, Safety Advisory Sub-Committee and Maintenance
Advisory Sub-committee.
ICAO - Human Factors Digest No. 5, Operational Implications of Automation in Advanced
Technology Flight Decks (Circular 234)
ICAO (1998) Doc. 9683-AN/950, Montreal, Canada
Khler Wolfgang (1967), Gestalt psychology, Psychological Research, Vol. 31, n. 1, pp. 18-
30
Koonce J.M., (2002), Human Factors in the Training of Pilots, Taylor & Francis, London
Maurino D., Salas E. (2010), Human Factor in aviation, Academic Press, Elsevier, MA, USA
Morin E. (2008), Il Metodo Le idee: habitat, vita organizzazione, usi e costumi, Raffaello Cortina
editore, Milano
Morin E. (2004), Il Metodo La vita della vita, Raffaello Cortina editore, Milano
Morin E. (2001), Il Metodo La natura della natura, Raffaello Cortina editore, Milano
Morin E. (1989), Il Metodo La conoscenza della conoscenza, Feltrinelli editore, Milano
David Navon (1977), Forest before trees: the precedence of global features in visual
perception, Cognitive Psychology, n. 9, pp. 353-383
Norman D. (1988), The psychology of everyday things, New York, NY: Basic Books, 1988.
Parasumaran, Sheridan , A Model for Types and Levels of Human Interaction with
Automation IEEE Transactions on Systems, Man, and Cyberneticspart A:
Systems and Humans, Vol. 30, No. 3, MAY 2000
Parasumaran R., Wickens C., Humans: Still Vital After All These Years of Automation, in
Human Factors, Vol. 50, No. 3, June 2008, pp. 511520.
Ralli M. (1993), Fattore umano ed operazioni di volo, Libreria dellorologio, Roma
Rasmussen J., Skills, Rules, Knowledge: Signals, Sign and Symbol and Other Distinctions in
Human Performance Models, In "IEEE Transactions Systems, Man & Cybernetics",
SMC-13, 1983
Reason James, (1990) Human error, Cambridge University Press, Cambridge
Reason James (2008), The human contribution, Ashgate, Farnham, England
Salas E., Maurino D., (2010), Human factors in aviation, Elsevier, Burlington, MA, USA

Automation 102
Wears Robert, Kendall Webb L. (2011), Fundamental on Situational Surprise: a case study
with Implications for Resilience in Proceedings of the fourth Symposium on Resilience
Engineering, Mines-Tech, Paris.
Woods D., Dekker S., Cook R., Johannesen L., Sarter N., (2010), Behind human error, Ashgate
publishing, Aldershot, England
Woods D. D., Modeling and predicting human error, in J.Elkind S. Card J. Hochberg and B.
Huey (Eds.), Human performance models for computer aided engineering (248-274),
Academic Press 1990
Wright Peter, Pocock Steven and Fields Bob, (2002) The Prescription and Practice of Work
on the Flight DeckDepartment of Computer ScienceUniversity of YorkYork YO10 5DD
6
Power System and Substation Automation
Edward Chikuni
Cape Peninsula University of Technology
South Africa
1. Introduction
Automation is the application of machines to tasks once performed by human beings, or
increasingly, to tasks that would otherwise be impossible, Encyclopaedia Britannica [1].
The term automation itself was coined in the 1940s at the Ford Motor Company. The idea of
automating processes and systems started many years earlier than this as part of the
agricultural and industrial revolutions of the late 18
th
and early 19
th
centuries. There is little
disputing that England was a major contributor to the Industrial Revolution and indeed was
the birth place of some prominent inventors, for example. in the area of textiles:
James Hargreaves: Spinning Jenny
Sir Richard Arkwright: Mechanical Spinning Machine
Edmund Cartwright: Power Loom
Cartwrights power loom was powered by a steam engine. At these early stages we see the
symbiotic relationships between automation, energy and power. The early forms of
automation can only largely be described as mechanisation, but the emergence of electrical
power systems in the late 19
th
century and the entry of electronic valves in the early 20
th

century heralded the humble beginnings of modern automation. With electronic valves
came computers. One of the earliest computers was the ENIAC (Electronic Numerical
Integrator and Automatic Computer) built over two years between 1943 and 1946. It
occupied an area of 1000 square feet (about 93 square metres), had 18000 valves and
consumed 230 kW [2].
Before the deployment of computers in industrial automation, relays and RELAY LOGIC,
the wiring of circuits with relays to achieve automation tasks, was in common use. Today,
however, relay logic is far less used than computer-based, PROGRAMMABLE LOGIC,
which has followed the invention of the transistor, integrated circuits and microprocessors.
2. Automation in the automobile (car / truck) industry
The motor assembly line pioneered by Ransom Olds and Henry Ford, the maturity of
computer technology and the structured nature of the car assembly process led early
entrepreneurs in the motor industry to view automation as key to business success. Indeed
there was all to be gained in automation and today automation is viewed to have, among
many other attributes, the following [3]:

Automation 104
a. relieves humans from monotony and drudgery
b. relieves humans from arduous and dangerous tasks
c. increases productivity and speeds up work rates
d. improves product quality
e. reduces costs and prices
f. increases energy and material savings
g. improves safety
h. provides better data capture and product tracking.
The automotive industry was very competitive right from the early days and automation
soon came to be seen as key to commercial success. General Motors embraced it and so did
many other auto manufacturing corporations in the US, Europe and Japan. The computer
system used was designed for the usually harsh industrial environments. Programmable
Logic Controllers (PLC) as such computers are known typically to have many inputs and
outputs; the inputs receiving sensor signals, the outputs being for displaying information or
driving actuators (Figure 1).

Fig. 1.
The need for standardization was realized early, especially when many PLCs are required to
achieve the automation task. The standards generating body IEEE has been a driving force
in computer communication standards over the years. One of its standards, the token ring
802.4 was implemented in modified form by General Motors in its Manufacturing
Automation Protocol (MAP).
3. Power system automation
The early power plants had a modest number of sensor and action variables, of the order of
several hundreds. Modern large power stations have in excess of hundreds of thousands,
even tens of million variables [3]. It is therefore easy to see that automation took root in

Power System and Substation Automation 105
power and generating stations earlier than in transmission and distribution. One of the
useful applications of automation is in the railway industry where remote control of power
is often vital. This manifests itself in systems to control and manage power supplies to
electric locomotives. The deprivation of power to any locomotives in a section could
seriously disrupt schedules, inconvenience customers and in the end have serious adverse
financial implications. Consider Figure 2

Fig. 2.
In earlier systems, the mini computer would have been sourced from specialised
computer companies such as IBM, Perkin Elmer, Digital Equipment Corporation. The
operating systems would have belonged to the same equipment provider (e.g., VMS VAX,
in the case of Digital Equipment Corporation). The microwave link would have been part of
the infrastructure costs for the project. The synoptic panel, allowed visual representation of
the system states to the operator (and also some capabilities for remote switching). For
Figure 1, the operator at the control centre is able switch a circuit breaker ON or OFF
SYNOPTIC
PANEL
PROGRAMMING
TERMINAL
IC
RTU
TRANSFORMER
Actuator and
Circuit Breaker
M
ic
r
o
w
a
v
e

L
in
k
Mini Computer

Automation 106
through the microwave link. Interface cards (IC) provide the necessary I/O capabilities. The
remote terminal unit (RTU) is able to send a switching signal to the circuit breaker through a
suitably designed actuator. The circuit breaker is also able to send its position (ON or OFF)
through the RTU and the same microwave link to the operators panel. The older type of
hardwired panel is largely being replaced by electronic displays and the controller is able to
receive and view system status through touch panel capabilities or through networked
computers in the control room.
The proprietary nature of the old legacy systems meant that embarking on the path to
automation was not a trivial matter. The cost of ownership, (infrastructure, hardware and
software costs) was very high. Also very high were the costs of maintaining and upgrading
this hardware and software. The cost of software licences was often prohibitive. On the
other hand cyber threats and virus attacks were unheard of. The systems themselves were
broadly secure, but not necessarily reliable. Communication link (microwave failure) meant
that there was at the time, as there is even in this generation, a need for manual back-ups.
This usually meant sending a technician to do manual switching operations.
4. Modern grid and substation automation
Power system automation happens in segments of the power system (Northcote-Green,
Wilson) [4] which can serve different functions. One segment is bulk transmission of power
which traditionally was handled by the power producer, but increasingly (in de-regulated
environments), is handled by an independent transmission system operator (TSO). Bulk
transmission is usually associated with outdoor switchyards and high voltage operating
voltage levels (in excess of 132 kV). Bulk transmission substations play a critical role in
energy trading and power exchanges. Wholesale electricity is sold through the transmission
system to distributors. Figure 3 shows a portion of the 400 kV outdoor substation at AUAS
near Windhoek, the capital of Namibia. Namibia is a net importer of electrical power most
of it from neighbouring South Africa and this substation is of vital importance. The
substation with which it connects in South Africa is over 800 km away at Aries near
Kenhardt. Figure 4 shows the Namibian electrical power transmission network. The
complex interconnections between equipment, such as transformers, reactors, lines and bus-
bars, is such that manual operation is not a practical proposition. In the case of the AUAS
substation, effective control is in the hands of Namibia Power Corporations (Nampower)
headquarter-based National Control Centre in Windhoek.
The other segment of automation is at distribution level. Large distributors are typically
municipal undertakings or in countries in which electrical power is de-regulated, the so
called DISCOS. Automation has existed at the distribution level for many years, but has
been restricted to situations involving either large numbers of customers or critical loads. As
a result of this the quality of service given has been very good, while the rural consumers
have been at a disadvantage. For power distributors, automating rural supplies was not
cost-effective due to the dispersed nature of the lines and loads (Alsthom Network
Protection and Application Guide) [5]. Technology changes in recent years, national power
quality directives as well as increased consciousness by consumers themselves has led to
radical changes in our Power System Control infrastructure and ways of operation with
serious implications for those organisations that lag behind [NPAG].


Fig. 3. A group of Polytechnic of Namibia students visit the AUAS 400 kV substation
4.1 Distribution systems automation
From experience, faults at transmission levels are less frequent than at distribution levels
[6]. At the same time distribution networks are not only complex, but the consequences of
failure are quite severe. For this reason investment in distribution automation will increase.
The elements that characterise distribution automation systems are given the definition by
the IEEE. According to the IEEE, a Distribution Automation System (DAS) is a system that
enables an electric utility to remotely monitor, coordinate and operate distribution
components, in a real-time mode from remote locations [7]. In this chapter we shall deal in
more detail with the components involved and how they are coordinated within a DAS. In
countries or situations where there are large networks, the network (primary distribution)
itself is subdivided into more segments, namely, one for large consumers (no
transformation provided) and for the rest at a lower HV voltage (secondary distribution)
(Figure 6).

Automation 108

Fig. 4. (courtesy AREVA, NPAG)


Fig. 5. (courtesy, Namibian Power Corporation)

Fig. 6. (courtesy AREVA, NPAG)

Automation 110
5. Power system automation components
Power system automation components may be classified according to their function:
Sensors
Interface Equipment
Controllers
Actuators
Thus we see that Figure 1 is still a good representation of what is needed to effect
automation, whether it is for EHV transmission, sub-transmission or distribution. Figure 7
depicts the control philosophy of a power system automation scheme.

Fig. 7.
5.1 Overview of power system components
5.1.1 Sensors
5.1.1.1 Current and voltage transformers
Individually or in combination current and voltage transformers (also called instrument
transformers) are used in protective schemes such as overcurrent, distance and carrier
protection. Also in combination current and voltage transformers are also used for power
measurements. In general custom specified voltage and current transformers are used for
power metering, because of the increased accuracy requirements. Figure 8 shows instrument
transformers in one of the substation areas (called bays).
Action and
Feedback
MONITORING
AND
SUPERVISION
CONTROL
ACQUISITION


Fig. 8.
5.1.1.2 Other sensors
For reliable electrical power system performance the states, stress conditions and the
environmental conditions associated with the components have to be monitored. A very
costly component in a substation is a transformer. For a transformer, monitoring is done, for
example, for pressure inside the tank, winding temperature and oil level. For circuit
breakers, sensing signals may need to be obtained from it such as gas pressure and number
of operations.
5.1.2 Switches, isolators, circuit breakers
A most important function of a substation is the enabling of circuit configuration changes
occasioned by, for example, planned maintenance, faults feeders or other electrical
equipment. This function is of course in addition to the other important function of circuit
protection which may also necessitate configuration changes. Modern switches and circuit
breakers will have contacts or sensors to indicate their state or position. Figure 9 shows the
ABB HH circuit breaker mechanism. The plant required to achieve the desired operation is
usually quite elaborate and includes controls and protection to ensure that it operates
reliably.

Automation 112

Fig. 9. Portion of HH ABB circuit breaker mechanism
6. IEC 61850 substation automation: Origin and philosophy
The International Electrotechnical Commission is one of the most recognisable standard
generating bodies for the electrical power industry. Its standard the IEC 61850
Communication Networks and Systems in Substations is a global standard governing
communications in substations. The scope of the standards is very broad and its
ramifications very profound. So profound in fact that it is hard to imagine any new modern
substation that would not at least incorporate parts of this standard. In addition, the
standard is almost sure to be adopted albeit in customised / modified form in Generation,
Distributed Energy Resources (DER) and in manufacturing. The standard has its origins in
the Utility Communications Architecture (UCA), a 1988 initiative by the Electrical Power
Research Institute (EPRI) and IEEE with the initial aim of achieving inter-operability
between control centres and between substations and control centres. In the end it was
found to be more prudent to join efforts with similar work being done by the Working
Group 10 of Number 57 (TC57). The emerged document IEC 61850 used work already done
by the UCA as a basis for further development.
6.1 IEC 61850 substation architecture
6.1.1 Substation bays
In an IEC 61850 compliant substation, equipment is organized into areas or zones called
bays. In these areas we find switching devices (e.g., isolators and circuit breakers) that
connect, for example, lines or transformers to bus-bars.

Examples of the bays would be:
Incomer bay
Bus-coupler bay
Transformer bay
Figure 8 above, for example, could depict a transformer bay.
6.1.2 Merging units
Merging units are signal conditioners and processors. For example, they accept, merge and
synchronise sampled current and voltage signals (all three phases quantities of the CT/VT)
from current and voltage transformers (conventional and non-conventional) and then
transfer them to intelligent electronic devices (see IEDs, in the next section). So called
electronic VTs and CTs are being manufactured by some companies which use new ways of
sensing with the overall size being reduced. With electronic sensing, the sensing and
merging are combined. Figure 10 gives an overview of the functions and associated inputs
of a merging unit.

Fig. 10. (Jansen & Apostolov)

Automation 114
As technology progresses it is believed that there will be a move away from copper
connections from field devices to the substation control room in favour of fibre. Figure 11
shows a merging unit (Brick) by vendor GE.

6.1.3 Intelligent Electronic Devices (IEDs)
An IED is any substation device which has a communications port to electronically transfer
analog, status or control data via a proprietary or standard transmission format (BPL Global
IEC 61850 Guide) [8]. Examples of IEDs are:
Modern IEC 61850 protection relays (distance, over-current, etc.)
Equipment-specific IED (e.g., for transformer bay protection and control, with tripping
logic, disturbance monitoring, voltage, current, real and reactive power, energy,
frequency, etc.).
Bay controllers
Figure 11 shows some IEDs from various vendors with multiple functionality.


Fig. 11.
In reality todays IEDs have mutated to the form of programmable logical controllers
(PLCs) of another kind with multiple capabilities.
6.1.4 Device/system integration: Substation functional hierarchy
An IEC 61850-designed substation has the following hierarchical zones:
Process
Bay
Station
Diagrammatically this is illustrated in Figure 12 (Jansen & Apostolov) [9]. A complete
representation that includes aspects, such as links to remote control centres and GIS, is given
in Figure 13.

Automation 116

Fig. 12.

Fig. 13. (courtesy SISCO & GE)

[O1]
Fig. 14. Fibre-based

Fig. 15.

Automation 118

7. Substation communications and protocols
With the IEC 61850 technology and with all the components and systems described in
previous sections functioning normally, we have in fact a virtual substation. The remote
terminal units (RTU) increasingly with IED functionality, pass on analog and digital data
through either copper or fibre to IEDs in the substation control room in the form of relays or
bay controllers. The process of transferring data and communicating it to various devices
has been greatly simplified with the aid of the standard. The data arriving at the IEDs comes
already formatted / standardized. The situation is similar to the plug and play
philosophy applied to computer peripherals of today.
7.1 Virtualisation
With the IEC 61850 a real substation is transformed into a virtual substation, i.e., real devices
transformed into objects with unique standardized codes. In Figure 16, a real device, a
transformer bay is transformed into a virtual, logical device with descriptive name, e.g.,
Relay1. Inside the device are logical nodes (LN) named strictly in accordance with the IEC
standard. For example, a circuit breaker inside this logical device is given XCBR1 [10]. In turn
the breaker has other objects associated with it, e.g., status (open / closed) and health. The

services associated with this data model are defined in the Abstract Communications System
Interface (ACSI). The following ACSI functions are listed by Karlheinz Schwartz [11]:
Logical Nodes are used as containers of any information (data objects) to be monitored
Data objects are used to designate useful information to be monitored
Retrieval (polling) of the values of data objects (GetDataObjectValues)
Send events from a server device to a client (spontaneous reporting)
Store historical values of data objects (logging)
Exchange sampled values (current, voltages and vibration values)
Exchange simple status information (GOOSE)
Recording functions with COMTRADE files as output
7.2 Mapping
IEC 61850 is a communications standard, a main aim of which is interoperability. A good
definition is:
Interoperability is the ability of two or more IEDs (Intelligent Electronic Devices) from
the same vendor, or different vendors to exchange information and uses that information
for correct co-operation [12]. Although ACSI models enable all IEDs to behave
identically from a general network behaviour perspective, they still need to be made to
work with practical networks in the power industry, (Baigent, Adamiak and Mackiewicz)
[10]. This universal compatibility is achieved through mapping of the abstract services to
universal, industry-recognised protocols. Presently the protocol most supported is the
Manufacturing Message Specification (MMS). MMS was chosen because it has an
established track record in industrial automation and can support the complex and
service models of IEC 61850.
Table 1 gives an idea of the naming process:

Automation 120

Fig. 16. Karlheinz Schwartz

8. Communication of events in an IEC 61850 substation
In his IEC 61850 Primer, Herrera states that IEC 61850 provides a standardized framework
for substation integration that specifies the communications requirements, the functional
characteristics, the structure of data in devices, the naming conventions for the data, how
applications interact and control the devices, and how conformity to the standard should be
tested. In simpler terms, IEC 61850 it is an open standard protocol created to facilitate
communications in electric substations.
8.1 The communication structure of the substation
The IEC 61850 architecture there are two busses:
Process bus
Station bus
IEC 61850 station bus interconnects all bays with the station supervisory level and carries
control information such as measurement, interlocking and operations [13].
IEC 61850 process bus interconnects the IEDs within a bay that carries real-time
measurements for protection called sampled values or sampled measured values [13].
Figure 17 shows the basic architecture.

Fig. 17.

Automation 122
The process bus is designed to be fast since it must carry crucial I/O between IEDs and
sensors/actuators.
The requirements for the process bus cited in various literature sources are as follows:
High environmental requirements for the terminal equipment (electromagnetic
compatibility, temperature, shock, where applicable) in the area of the primary system
Adequate bandwidth for several SV data streams
Highly prioritized trip signals for transmitting from the protection device to the
CBC
Permeability of data to the station bus/data filtering at the coupling point
Simultaneous TCP/IP traffic for normal control and status signal traffic as well as
reports on the process bus
Download/upload channel for setting or parameterizing functions
Highly precise time synchronization
Redundancy
For reasons of speed, the process bus is based on optical fibre with high data
throughput of about 10Gbits/s. Because of its enhanced data capacity it is capable of
carrying both GOOSE (Generic Object Oriented Substation Event) and SMV (Sampled
Measured Values). The station bus is used for inter-IED communications. Only GOOSE
messaging occurs in the station bus.
9. Substation control and configuration
Although the strengths of the IEC 61850 in the capturing, virtualisation, mapping and
communication of substation information are undoubted, it will still be necessary to link
everything together and to design a control strategy. This strategy must utilize the
experience and expertise of the asset owner. The substation must also respond in accordance
with the operational and safety criteria set by the organization.
9.1 Substation configuration
Automation of the substation will require in the first instance the capture of its
configuration. This requires the capture of the information on all the IEDs in the substation.
In some cases the IEDs could be from different vendors. The information has to be in a
standardized IED Capability Description (ICD). Then, using a system configuration tool, a
substation description file is created (Figure 18). The SCD (Substation Configuration
Description) is then used by relay vendors to configure individual relays [14].
10. Wider implications of the IEC 61850: The Smart Grid
Smart Grid is a term used to describe the information driven power systems of the future.
This will involve introducing new electronic, information and computer technology into the
whole value chain of electrical energy systems from generation, transmission and
distribution down to the consumer level. Figure 19 shows the linkages between the
technology of electricity production and commerce.


Fig. 18.

Fig. 19. Source NIST Smart Grid Framework
We have seen that automation started on the factory floor and some of the IEC 61850
functions use manufacturing protocols such as MMS. We already start to see the trend
I CD FI LE
I CD FI LE I CD FI LE
SYSTEM CONFI GURATI ON TOOL
SCD FI LE
I EC 61 8 50

Automation 124
towards extending the IEC 61850 to generating stations. It is therefore not hard to
imagine IEC 61850 like protocols encompassing every facet of engineering, including
manufacture.
10.1 Smart Grid benefits
Among the benefits of Smart Grid are:
Increased grid efficiency - the use of control systems to achieve optimum power flow
through, for example, centrally controlled FACTS devices which can increase efficiency
of the transmission system
Better demand control - a Smart Grid would incorporate an energy management system
to manage demand (e.g., managing the peaks and valleys)
Asset optimization - the IEC 61850 information model already has the capability to store
not only the status of a logical device / node, but also condition / health
Management of renewable energy sources - renewable energy sources, such as wind
and solar, tend to be unpredictable, therefore, the Smart Grid system can enable
predictions on the availability of these resources at any moment and ensure proper
energy scheduling decisions are taken
Management of plug in electric vehicles - the Smart Grid can inform electric vehicle
motorists of the nearest charging stations
Smart metering - with smart metering, power usage and tariffs can be administered
remotely to the advantage of both the supplier and the consumer
11. Security threats in automated power systems
In this chapter we have seen the central role computer hardware and software in the control
and management of the power system bring tremendous benefits. However, investing in
these high technology, information technology reliant assets also brings threats. The threats
are quite serious especially when it is realized that every critical component of the
substation becomes a virtual computer. The IED mentioned numerous times in this chapter
is itself a computer. What are the threats?
11.1 SCADA vulnerabilities Chikuni, Dondo [15]
Computing vulnerabilities
Hardware: RTUs, IEDs and SCADA Masters belong to the class of computer hardware and
suffer from the same vulnerabilities of regular computer systems such as interruption
(denial of services [DoS]) and eavesdropping
Communication links: the vulnerabilities are also similar to those in regular computer
networks - if messages are not encrypted, data or passwords can be intercepted. Radiation
emissions from equipment can be read by unauthorized people
SCADA software: the most common attacks come in the form of interruption,
interception and modification. Software bugs, if not fixed in time, can attract hobbyist
hackers to attack unpatched SCADA [15]

Data: SCADA data has more value to the attacker than hardware and software. Data may be
stolen by competitors of saboteurs. To safe guard the data, encryption needs to be included
11.2 Other vulnerabilities
Equipment location: we have seen that some IEDs and RTUs are located in usually
unmanned remote locations; where this applies these must be housed or mounted securely
Remote access: access to relays, controllers, IEDs and RTUs should be password
protected; encryption modems are available for secure dial-up communications
Human element: the employee could be the most vulnerable part of the automated
power system, therefore, no unauthorized persons should have access to the control
terminals. Strong authentication and smart card access are recommended
Integrity and confidentiality: software and hardware should have at least the US
National Computer Security Centre (NCSC) class 2 rating. In Europe criteria similar to
that of NCSC is managed through the European Information Technology Security
Evaluation Criteria (ITSEC)
11.3 Attack examples
Nadel et al. [16] list what they describe as generic attack categories to which all network-
based threats to substation automation systems can be reduced, namely:
Message modification
Message injection
Message suppression
They demonstrate the various paths that an attacker can take in a given attack. An example
of circuit breaker attack scenario is shown in Figure 20. In this case an attacker may take one
of the paths in the graph to prevent a circuit breaker from opening when it is supposed to.
11.4 Countermeasures
Nadel et al. [16] list some of the important assumptions / precautions necessary before any
meaningful countermeasures can be instituted. These include static configuration of the SAS
and the number and types of devices in the bay level are known; configuration changes only
to occur during major maintenance or modification work; also that the SAS is not used for
billing and no general purpose PCs are allowed at bay level.
Message modification
In this attack parts of a valid, existing message are modified in transit. Detection is
facilitated through encryption and digital signatures, with the receiver having a record of all
authorized senders.
Message injection and replay
The attacker sends messages which are not intended to be sent by any authorized sender.
These may be entirely new messages from the attacker or original untampered with
replayed messages. Digital signatures are a way of combating message injection. To mitigate
against replay, a digital signature and message sequence number are required.

Automation 126

Fig. 20. Example of attack graph for a circuit breaker
Message suppression
In this attack certain messages between SAS devices are prevented from reaching the
receiver, e.g., circuit breaker control devices are isolated from protection devices. Message
suppression can involve several other types of attack, e.g., re-configuration of routers or
switches, cutting wires or congesting the network so that genuine messages cannot get
through (denial of service attack).
Security protocols
The multiplicity and varied nature of SAS attacks makes it imperative to institute robust
security protocols capable of handling all eventualities. Such protocols include the use of
private keys (only known to the sender), encryption and sequence numbers (initial number
known between sender and receiver at the start).
Messages
Suppressed
Fake Sensor
Messages
Inj ected
Message
Content
Changed
Communication
Sensor->PDIS
Disturbed
Communication
PDIS->XCBR
Disturbed
Wrong Setti ngs
Messages
Suppressed
PDIS
Trigger
Conditions
XCBR
Switching
Conditions
XCBR energy
Control loop
Message
Content
Changed
Message
Content
Changed
Fake Messages
Injected
Wrong Settings
in PDIS
reclosing
wrongly
activated
CB doesn't
Switch
short circuit in li ne
and li ne not
disconnected
unselective
switch-off
primary
equipment
damage


Automation 128

12. Effects on educational curricula
To give an idea of the profound changes the power system will have on our education
systems and also to give some suggestions on how to mitigate some of the challenges, we
reproduce the following excerpts from a paper by Chikuni, Engelbrecht and Dongo [17] at
the PowerCon 2010 conference:
When we analyse a modern substation incorporating the new substation technology based
on IEC 61850, it would seem that the role of an electrical engineer is notable by its absence.
Certainly some of the responsibilities of both engineers and technicians have shifted and
there needs to be new breed of electrical engineers altogether. Some questions need to be
answered:
Should we retrain the power engineer in networks or
Should we train network engineers so that they acquire power engineering knowledge
or
Should we work with a completely new curriculum which merges power systems,
electronics and networks into one programme?
We first need to acknowledge that there are already a lot of good power engineers out there
trained in the traditional manner, i.e., starting off with physics, circuits and systems,
electrical machines and power systems (including electrical protection). For these engineers,
one needs to identify those who can benefit both themselves and their organizations by

going through this additional training. The process of training engineers has been quite
formal, especially if they wish to attain professional status. One typically needs four to five
years of formal training and a further two years of guided industrial training before
attaining the status of chartered (CEng) or professional (PrEng) engineer. A great debate
will ensue, therefore, when a computer network engineer is designated the responsible
person in an electrical substation, notwithstanding the obviously immense power this
individual will have in making sure that the substation operates correctly, safely and
efficiently.
The other route is to include networking as part of any electrical engineering curriculum. A
few programmes today include industrial automation and a few even include computer
networking. In the University of Zimbabwe model all electrical engineering students have a
chance to complete at least the first semesters of a CISCO network academy programme.
Indeed some complete the CCNA (four semesters). Whatever solution is arrived at, it is clear
that electrical engineering training curricula inevitably have to include more and more
electronics, sensors, automation and networking, not as peripheral subjects, but as part of
the core.
13. Summary and conclusions
In this chapter we have seen the extremely rapid development of automation, starting from
the years of mechanisation, production lines and the taking root of computer-based
automation in the car manufacturing industry. Then we noticed rapid increases in computer
power in both hardware and software forms. There has also been tremendous moves in
standardization in North America and Europe. We have seen too IEC 61850 international
cooperation in standards development and the benefits that are already being reaped from
this. Interoperability brings some relief to customers, giving them the ability to choose
hardware from an increasing variety of vendors. Quite striking is the increasing dominance
of ICT in power system control and massive changes in power system operation and
practice. The power systems have become more complex - more interlinked. The complexity
presents new challenges. The traditionally trained power systems engineer lacks the know
how to understand or tackle faults that could arise in these systems. On the other hand the
network engineer may lack the underlying principles of power and energy systems. A new
type of multi-discipline power systems engineer has to be trained. The Smart Grid will soon
be a reality. Generation, transmission, distribution consumption and commerce will be
information driven. Finally, when automation is combined with mechatronics and robotics,
our lives are poised to be drastically changed.
14. References
[1] Mikel P. Groover, Britannica Online Encyclopedia
[2] Benjamin F. Shearer,. Home front heroes: a biographical dictionary of Americans during
wartime, Volume 1, Greenwood Publishing Group 2007
[3] Edward Chikuni, Concise Higher Electrical Engineering, Juta Academic Publishers, 2008
[4] James Northcote-Green, Robert Wilson, Control and Automation of Electrical Power
Distribution Systems, Taylor& Francis, 2006
[5] AREVA, ALSTOM, Network Protection and Application Guide, 2011 Edition

Automation 130
[6] Su Sheng; Duan Xianzhong; W.L. Chan, Probability Distribution of Fault in Distribution
System , Power Systems, IEEE Transactions on, Aug. 2008
[7] D. Bassett, K. Clinard, J. Grainger, S. Purucker, and D.Ward, Tutorial course: distribution
automation, IEEE Tutorial Publ. 88EH0280-8-PWR, 1988
[8] IEC 61850 Guide Serveron TM8TM and TM3TM On-line Transformer Monitors 810-
1885-00 Rev A August 2011
[9] M.C. Janssen,, A. Apostolov IEC 61850 Impact on Substation Design
[10] Drew Baigent, Mark Adamiak, Ralph Mackiewicz, GE and SISCO Communication
Networks and Systems In Substations, An Overview for Users,.
[11] Karlheinz Schwarz, SCC, Monitoring and Control of Power Systems and Communication
Infrastructures based on IEC 61850 and IEC 61400.
[12] Dipl.-Ing. H. Dawidczak, Dr.-Ing. H. Englert Siemens AG, Energy Automation
Nuremberg, Germany IEC 61850 interoperability and use of flexible object modeling
and naming
[13] R. Moore, IEEE Member, R. Midence, IEEE, M. Goraj, Practical Experience with IEEE
1588 High Precision Time Synchronization in Electrical Substation based on IEC 61850
Process Bus
[14] J. Holbach, J. Rodriguez, C. Wester, D. Baigent, L. Frisk, S. Kunsman, L. Hossenlop,
First IEC 61850 Multivendor Project in the USA, Protection, Automation and Control
World, August 2007
[15] Edward Chikuni, Maxwell Dondo, Investigating the Security of Electrical Power Systems
SCADA, IEEE Africon 2007, Windhoek
[16] Martin Naedele , Dacfey Dzung , Michael Stanimirov, Network Security for Substation
Automation Systems, Proceedings of the 20th International Conference on
Computer Safety, Reliability and Security, p.25-34,September 26-28, 2001
[17] E Chikuni, F Engelbrecht, and M Dondo, The emergence of substation automation in
Southern Africa, opportunities, challenges and threats, IEEE Africon Conference,
Windhoek, September, 2007
7
Virtual Commissioning of Automated Systems
Zheng Liu
1
, Nico Suchold
2
and Christian Diedrich
2
1
Otto-von-Guericke University, Magdeburg,
2
Institut fr Automation und Kommunikation (ifak), Magdeburg,
Germany
1. Introduction
Concepts for the digitalization of products and all the production related tasks in the
manufacturing and process industry have been developed for several decades. The
development began with the introduction of computer-based 2D design (Figure 1).

Fig. 1. Development stages for the Digital Factory (Br, 2004)
Because of the technological developments in the information technologies field, more
complex tasks in the product development process can be processed digitally (Brac, 2002).
This belongs to the scope of the digital factory which includes the digitalization of models of
the products and their integration into the manufacturing process chain. In (VDI 4499, 2008)
the digital factory was defined as a generic term for a comprehensive network of digital
models and methods, including simulation and 3D visualization. Its purpose is the

Automation 132
integrated planning, implementation, control and on-going improvement of all important
factory processes and resources related to the product.
One of the topics in the digital factory is the Virtual Commissioning. As we know, the
commissioning of the automated system is an important phase in the engineering, which
makes visible whether the systems and components are planned, designed, produced and
installed correctly according to the user requirements. However, this phase has been known
as time-consuming and cost-consuming in practice (section 2.1) and can be improved by
Virtual Commissioning.
State of the art is that the simulation of the system components is frequently used and the
simulation is used without integration of automation devices and components, for example
without PLCs (Programmable Logic Controller). HiL (Hardware in the loop) technologies
are used only for individual components. The concept of digital factory combines both
technologies.
During Virtual Commissioning the real plant is replaced by a virtual model according to the
concept of digital factory. But the questions are: what digital data is necessary for the virtual
plant model, how is the virtual plant implemented and configured, and how can these
processes be integrated into the engineering lifecycle of manufacturing systems? This
chapter gives comprehensive answers to these questions.
2. The Virtual Commissioning
The basic idea of Virtual Commissioning is to connect a digital plant model with a real plant
controller (e.g. PLC- or HMI-Human Machine Interface) so that engineers from different
fields such as design process and control have a common model to work with together.
Thus, for instance, the PLC program can be tested virtually before the physical
implementation is finished. Furthermore, the general functionalities of the plant can be
validated and finally tested in an earlier phase. In this section, the general workflow of a
traditional engineering process is introduced and its disadvantage discussed. Thereafter the
Virtual Commissioning, as a possibility to overcome these problems, is introduced and
described in detail in the subsequent subsections.
2.1 General workflow in the engineering process
The engineering of manufacturing plants begins with the construction and the rough
mechanic planning. In this phase the general structure (layout) of the manufacturing system
as well as the assignment of the machines in different production phases are determined,
which is based on the result of product development.
The detailed planning is done in the next step. The kinematic simulation for the offline
programming (OLP) of the robots is carried out, therefore the robot programs can already be
created without the real robots, e.g. for the generation of a trajectory free of collision with
other devices. In the phase of mechanic detail planning, a detailed CAD-based 3D cell layout
(M-CAD) is created. The production workflow of a single machine is to be defined and
every mechanical component in the cell can also be selected and documented in the list of
materials. The results of the mechanical detail planning are used as input for the subsequent
electrical detail planning. At this step the electrical energy supply as well as the control

Virtual Commissioning of Automated Systems 133
signals for the machines and their interconnection with each other is defined. This can be
called E-CAD and is documented in the circuit diagram. The last phase of the detail
planning is the PLC programming.
The last step in the classic engineering process is the commissioning, which can be
understood as a phase of preproduction. At this phase the functionality of every single
component in a function group is examined. After that it follows the interconnection check
in turn from the level of function groups to the level of stations until the whole plant is
complete - the bottom-up principle. The result of the commissioning is a functional plant
which is ready for the production.
2.2 Todays problem in the commissioning
The above mentioned approach of the engineering process can be seen as the water fall
model. A single step in this model can only be executed one after the other. There is no
iteration step back to the previous steps. As a result, the debugging takes place only at the
phase of commissioning. The following two main problems are identified here.
Bugs in the software
(Zh & Wnsch, 2005) describes the time consuming for commissioning and PLC-
programming in a whole project. It can be seen that the commissioning takes up 15-20% of
the overall project duration, of which up to 90% is for commissioning the industrial electrics
and control systems (e.g. PLC). Within the commissioning, up to 70% of the time is incurred
by software errors (Figure 2).
The debugging of the PLC-Software is usually time consuming and may cause hardware
damage. Since the commissioning is usually carried out under extreme time pressure, then a
delay of the project must be calculated with financial penalties. Therefore it is desirable that
the software debugging be carried out in an earlier phase.

Fig. 2. Contribution of control software to project delay (Wnsch, 2008)
The error is detected too late
Many errors that are detected during the commissioning arise in the earlier phases. This is a
big problem since the late detection of errors and the costs of their elimination can be very
high (Figure 3). When an error occurs in the early phase, an overwork of the rest phases is
not avoidable. For example, if a robot is wrongly placed in the 3D cell layout and this error

Automation 134
is not detected until the commissioning phase, then it is a great effort with extra costs to
eliminate this error because all the planning data related to this robot has already been
modified.

Fig. 3. Costs incurred by the correction of errors, depending on the time of error detection
(Kiefer, 2007)
2.3 State of the technology
One approach to overcome the disadvantages mentioned in subsection 2.2 is the full digital
planning and simulation of the production line. Concepts, concrete process sequences and
the necessary resources can be defined and approved already in the early planning phase.
This activity belongs to the concept of digital factory.

Fig. 4. The engineering chain in the production with Virtual Commissioning
The focus of digital factory is the digital processing of product development and production
planning based on the existing CAx data with a seamless workflow in PLM (product
lifecycle management). Digital simulation methods and technologies are used to secure the
planning results, to optimize the process and to respond more quickly to the changes than
before. In this way the product qualities are improved. Another benefit is the reduction of
the planning time as well as the overall project duration. The Time-to-Market is thus
shortened which means another advantage with respect to economy.
To test the PLC-program before the real commissioning, the Virtual Commissioning can be
applied (Figure 4). It serves as a smooth transition between digital and real factory. In the
case of Virtual Commissioning, the physical plant, which consists of mechatronic
components, is simulated with the virtual model. This simulated system is connected to the
real controller (PLC) via simple connection or real industrial communication systems. The
goal is to approximate the behavior of the simulated system to that of the real physical plant
by connecting the commissioned PLC to the real plant without changes (Figure 5). Therefore
the development and test of automation systems can be done parallel to the electric and
mechatronic development. In the case of the real commissioning, the connection can be
switched to the real system again.


Fig. 5. From Virtual Commissioning to real commissioning
2.4 Component of the Virtual Commissioning
The virtualized plant, a mechatronic plant model, simulates the behavior of the real physical
plant. It should respond correctly to the PLC control signals just as a real system. Besides,
the whole process should be visualized for a better human observation. To fulfill these
requests, a mechatronic model (Mhlhause et al., 2010) should be divided into a control-
oriented behavior model and a kinematic 3D-model.
The behavior model simulates the uncontrolled behavior of the system. The behavioral
states of the production resource are modeled and calculated by means of the logic and
temporal components, which operate based on the control signals. In one PLC scan period,
the behavior model should react to the control signals (output signals of PLC) according to
the physical feature of the plant and give the feedback signal (input signals of PLC) back to
the PLC just as the behavior of a real system.
The kinematic 3D-model can be understood as a geometric model, which is based on the 3D-
CAD model and enriched with additional information e.g. the grouping of the components
which should be moved together and the definition of the DOF (degree of freedom) for
every moving part. In contrast to todays mechanical CAD models, the mechatronic
components do not just solely consist of the pure 3D CAD model, but also kinematic
(including end positions) and electrical information like the electrical name of the respective
device. For the purpose of synchronization, a signal coupling between the behavior model
and kinematic 3D-model is needed.

Automation 136
The kinematic 3D-model is optional for the Virtual Commissioning, which helps for a better
understanding by visualizing the movement behavior in 3D.
The communication between the mechatronic cell model and the real PLC can be realized
via TCP/IP, field bus, Ethernet etc. There should be an interface to emulate the behavior of
the communication. In the case of PROFIBUS, for example, a SIMBA-Card is needed.
2.5 Configuration of a Virtual Commissioning workstation
For the planning of a new production system, a layout plan for the necessary components
(roles for the production resource to be used) and a rough workflow of the systems will be
worked out based on a pattern (Figure 6). Based on this, a simulation model of the plant,
including the geometric and kinematic design, will be built. This simulated layout is the
starting point for the Virtual Commissioning, which is to be extended by the offline robot
programming and planning process. At the step of detailed planning, the roles of resource
are replaced by concrete units (function groups) under some fixed rules. For example, a
certain combination of frequency inverter, the motor for a required electric moment of a
torque, the type of clamp and valve cluster with the proper nominal pressure, or the robot
and the weight of products to be operated, must be adhered to. These rules are to be
observed by planners in the detailed planning. In a robot cell, for example, about 100 of such
function groups must be matched and built. This function group can be stored in a
kinematic library for the reusability. To build a simulation model, these previously created
library elements can be added and connected with each other to describe the behavior of the
corresponding component. Such a behavior simulation model has in average 30 inputs and
outputs, which must be connected to the kinematics or to the PLC.
Workplace of the virtual
commissioning
simulation system PLC HMI
Simulation model
control programm HMI configuration configuration of
the signal interface
layout
process definition
robot simulation
Use of engineering data
plant
model
d
a
t
a
b
a
s
e
Libraries
mechanic
rough
planning
kinematic
simulation
mechanic /
electric detail
planning
virtual / hybrid
commissionin
g
A
P
I
s
E
x
c
h
a
n
g
e

f
o
r
m
a
t
s

Fig. 6. Concept for integration into the Virtual Commissioning

After or in parallel to the layout planning and kinematic simulation, a suitable control
program and operator visualization will also be developed, which run on a PLC or a PCS
(Process Control System) system. The PLC program is developed in the following way: for
each resource a corresponding control function block (Proxy FBs) is used, which is included
in the control library. These Proxy FBs will be invoked in one or more SCF (Sequential
Function Chart language IEC 61131-3). This control program will then be supplemented by
plant-specific locks, a switch-off and parameterizations.
Afterwards, the PLC and the simulation systems are interconnected via the communication
system or direct wiring. The I/O lists of the simulation system and the I/O lists of the
control program must be properly connected with each other. Furthermore, each I/O has a
number of features such as data type, range and unit, which need to be customized. The
connection between the I/O of the Proxy FBs (or the communication configuration of the
PLC) and the I/O lists of the simulation model must therefore be checked accordingly to the
signal features.
A workstation for the Virtual Commissioning (Figure 6) consists of a simulation system on
which the simulation model runs, a PLC which is usually connected with the simulation
model via a bus system (e.g. PROFINET) and optionally a PCS.
In the phase of Virtual Commissioning, the plant which is to be tested is completely
simulated in a digital model instead of a real one. This includes the automation equipment
such as the clamps, robots and electric drives. When the PLC is connected to the real and
digital systems at the same time, then it can be called the hybrid commissioning. For
example, a robot with its controller can be integrated into a single system with other digital
components, and the system should not detect any differences between the real and the
digital components.
The current technology state assumes that the geometry and kinematics of the resources are
provided by the manufacturer and as available libraries. In addition, the features of reused
equipment may be stored in proprietary formats. These features and libraries are available
in different storage systems (e.g. databases, repositories of tools and file systems).
Typical tools for Virtual Commissioning are Process Simulate/Process Simulate
Commissioning and Delmia Robotics/Delmia Automation. In additionthe authors use the
software WinMOD of the German company Mewes & Partner GmbHwhich includes the
component for 3D visualization. A circuit plan for electric planning can be edited for
example with COMOS. There are also templates such as in Process Simulate, WinMOD,
COMOS and Simatic S7, in which the corresponding features already exist in the library
elements for converters, motors, robots etc. Today, the behavior models are not supplied by
the resource producers. Instead, they are created by the users themselves for specific
applications.
Besides the integration of a mechatronic plant model in the line-balancing scenario, a
concept for the integration into the Virtual Commissioning was developed. The process of
creating the workstation for the Virtual Commissioning has been explained above. The use
of a mechatronic plant model should contribute to the construction of the workstation for
Virtual Commissioning. The information which is extracted from existing data sources by
using the mechatronic plant is able to represent the structure of the workstation (kinematic
and behavioral simulation, PLC program, link-up between PLC and simulation model).

Automation 138
3. The mechatronic plant model with semantic techniques
The domain specific models in the individual life cycle phases usually have their own
concepts depending on their physical phenomenon and tasks. This is essential because the
models reflect only the necessary characteristics of the domains. A universal valid model is
not thinkable. However, for seamless information flow it is necessary for the information
consumers to have knowledge about the context of the transmitted information so that they
can act in a proper way. (Rauprich et al., 2002) and (Wollschlaeger et al., 2009) point out that
the historical separation of engineering disciplines, their workflows and thus the
information integration is still an unsolved problem. One of the most common model
structures in the manufacturing domain is the Product, Process Resource approach. It is the
basis of digital planning of major tools like the Delmia tool set of Dassault Systemes (Kiefer,
2007).
Therefore, each piece of information has to be unambiguously identified and validated
before it can be used in one of the engineering tasks. Additionally, its semantic including, its
context information has to be specified and must be available. Mechatronic components, i.e.
plant or machine parts consisting of mechanical, electrical and logical aspects, build up the
modular blocks of the model, which are used in the engineering process. Thus these
mechatronic components have to be represented by its semantic enriched information.
Consequently, the mechatronic model must be based on semantic methods.
3.1 Mechatronic plant model
Today the information management in companies is characterized by a variety of available
information, which is mostly stored distributed in a monolithic form and in different data
sources. Howeverthe size of each data source increases continuously over time, so that a
connection between the information can only be realized through significant human work.
As a result, much of the available information is very difficult to find by the user. Hereby a
situation relevant lack of information arises within the information stream in the company.
The time required for the complex information search and retrieval rises exponentially with
the increasing amount of informationwhich results in significant costs. The mechatronic
plant model is a logical layer above the design data and can be divided into the basic aspects
of product, process and resource (Figure 7). Planning and other tasks in the plant life cycle
can be executed by tool applications. These tools add new data to the existing data pool.
These applications are modeled as so called facets (roles for the user data) in the
mechatronic plant model. The resources could be further refined (see Figure 7). The basic
idea of the model is to disperse all aspects (products, process and resource) in smaller
features. For example, a construction group has the feature of the diameter of the pressure
pipe connection or movement duration of a valve. The typical process features are process
step number and number of the predecessor or successor step. If the mechatronic model is
used for the acquisition of the data storage, the addressing data (e.g. data source type, URI
of the data source) is also to be modeled. The mechatronic model is thus a hierarchical
structure of sub-elements based on the three aspects product, process and resource. These
three aspects are then to be further dissected into even smaller features, until no more
division can be undertaken. In addition, these features are categorized so that semantic
information about the individual elements of an aspect and its relationship with other
elements of other aspects in the model can be enriched. (Figure 8)


Fig. 7. Basic structure of the mechatronic plant model

Fig. 8. The intern structure of the mechatronic plant model

Automation 140
3.2 Potential of the feature-based modeling approach
The basic idea of focusing on the features of engineering is to make engineering decisions
and the underlying rules simpler to relationships between features. A unified set of features,
for example coming from the standardization of technical product data, enables an
automated execution of rules. This makes it possible to resolve the relationship with defined
rules between feature carriers (e.g. valve or clamp or behavioral description and proxy FB).
This means the following assistance functions are available for the planner:
Consistency check
(semi-) automatic assignment
Restrict the solution
An important element of the mechatronic plant model is the coupling between kinematics
and behavior of the mechatronic components (aggregation relationship between the module
as well as the kinematics and behavior model in Figure 7) and the integration of the PPR
model (product, process and resource). Figure 7 shows the connection between resource and
process (executeProcess). For the purpose of semantically unambiguous modeling, the
description language OWL was used, which can clearly formulize the relations between the
classes.
The extensive relationship between kinematics and behavior model are described in Figure
9.

Fig. 9. Kinematic and behavioral model

In summary, the mechatronic plant model has the following essential basic elements:
Basic structure is the grouping of product, process and resource
Relationships are defined between the basic elements
The PPR elements are then further specialized
The properties of the PPR elements are described with features
The features can be structured in a specialized hierarchy
The features can build relationships with each other
4. Application of the model for Virtual Commissioning
4.1 Assistance by means of mechatronic cell model in the Virtual Commissioning
As described above, the starting point to establish a workplace for Virtual Commissioning is
the layout and the kinematic model of the plant. In addition, it can be assumed that the
information about the resources and process is located in component related files
(kinematics, geometry, behavioral descriptions), in databases or tools.
The goal is to build a workplace for Virtual Commissioning. A tool should be implemented
to assist the configuration of the workspace. This assistance tool is called VIB configurator or
VIBCon (Figure 10).
workplace
for the virtual
commissioning
Simulation system PLC
simulation model
control program configuration of the
signal interface
layout
process definition
Robot simulation
Use and model based integration of planning results
databases
engineering tools
exhange formats
Database access
AutomationML / PLMXML
APIs
mechanic rough
planning
kinematic simulation
mechanic electric
detail planning
virtual / hybrid
commissioning
VIB Configurator
starting point:
goal:

Fig. 10. The demonstrator for the Virtual Commissioning

Automation 142
The methodological foundations of the assistance tool allow the rule-based processing of
raw data, which is related to a specific working step. The raw data can be extracted from
various data sources. The result is the necessary configuration data or files for the
workspace of Virtual Commissioning (Figure 11). In this way a semantic-based model is
established in the mechatronic plant model. The model contains relationships such as the
structure-based relationship "consistOf" or "isCharacterizedBy" (Figure 7). In addition, there
are naming conventions that make, by using prefixes or suffixes, uniqueness semantic out of
the mechatronic model. In this way, a check list based on the previously finished work can
be generated. Based on this list, the library elements are instantiated. All these rules are
defined in the VIB configurator. The following functions were implemented by the
mechatronic plant model:
Specification of the relationships between model elements and the workflow and the
corresponding rules
Generation of the needed configuration information by semantic inference or support of
their derivation through a selection of possible alternatives

Fig. 11. Methodological approach of the rules
An example can be found at the beginning of the configuration of a Virtual Commissioning
workstation (Figure 11). The needed document types (e.g. RobCAD file, VRML file, PLC
program...) should be selected according to the context. These files will be earmarked and

stored in the VIB configurator. When a new project is started, which uses the pattern of an
old project, only the references to the types of these files and information are needed. These
references are derived from the structural knowledge of the relationships between the types
of resource and the types of their description.

Fig. 12. Sample rule from the knowledge of structure creating a project
A second example is the acquisition of information for all modules or components
(implementation of the resources) that are included in the layout of a plant (Figure 13).
The proper types of the files and information are needed for each resource in the project.
The relationships between the model elements are integrated in the mechatronic plant
model. Following the chain of these relationships between the elements leads to the
references of the data sources. Naming conventions are applied here, such as extensions
of the file name.
The mechatronic plant model serves as a knowledge model between the heterogeneous data
sources and the data which is to be generated (Figure 13). The semantic model contains the
know-how of the plant. The proper types of the files and information are needed for each
resource in the project. The whole model is embedded in a software environment, which is
described in subsection 4.2.

Automation 144

is referenced in
(naming convention)
WinMOD Macro
SIMATIC function blocks
data reference
real graphic
list of IOs
Component description
AVILUS-Web-Service
AutomationML-Service
database reference
component database
VRML kinematics
Layout

Fig. 13. Sample rule from naming convention acquisition of information resource
4.2 Practical implementation of mechatronic cell model based on the semantic model
The main focus of the mechatronic model is the semantic modeling of the mechatronic
components and the processes of the plant. This includes not only the description of the
structural characteristics, e.g. the mechanical or electrical features, but also details like
resource behaviors and process steps. Furthermore, a unified structure for the formal
description of system properties is set (see Subsection 3.1). A framework (Figure 14) is
implemented in order to access the information contained in the model like rules and
relationships. Therefore, the needed information is found from different data resources
(e.g. web services, exchange files of different formats). This process can be called
reasoning.
The components and tasks of this framework include:
Design and modeling (Figure 14, green)
Standard technologies of Semantic Web (Figure 14, gray)
Integration and evaluation (Figure14, orange)

model
plant
mechatronic
I
n
t
e
g
r
a
t
e
d

M
o
d
e
l
i
n
g

E
n
v
i
r
o
n
m
e
n
t
T
o
p
b
r
a
i
d
C
o
m
p
o
s
e
r
OWL-Reasoner, SPARQL-Engine
(e.g. Pellet OWL2 Reasoner, Jena SPARQL query engine)
data integration
import and convert
different data sources
rule- and
query design
K
B
S

K
n
o
w
l
e
d
g
e
B
a
s
e

S
e
r
v
e
r

Import
(e.g. product DB, PLMXML)
SPARQL-API
Export
(e.g. Excel, AutomatonML)
Feature-API
Semantic Web Framework Semantic Web Solutions Platform
data modelling
edit the
mechatronic model

Fig. 14. Framework of the mechatronic plant model

workplace
for the virtual
commissioning
simulation system PLC
simulation model control program
Konfiguration der
Signalkopplung
layout
process definition
robot simulation
Use and model based integration of planning results
Databases
Engineering tools
exchange formats
Database access
AutomationML / PLMXML
APIs
VIB Configurator
model
plant
mechatronic
Anwendungsintegration
mechanic rough
planning
kinematic simulation
mechanic / electric
detail planning
virtual / hybrid
commissioning
document management,
workflow assistance
(rule based)

Fig. 15. Role of the mechatronic plant model in the preparation of a workstation for Virtual
Commissioning

Automation 146
The most important technologies of the methods are data modeling and design of the rules
and inquiries. For data modeling, as mentioned above, a semantic-based approach was
deployed. The design of the rules and inquiries was based on this approach, which forms
the basis of the knowledge processing known as automated rules-based reasoning. The
instance data, which is based on the mechanisms of import and conversion, was integrated
into the data model.
The integration of existing standard technologies became the focus in the development of
the mechatronic plant model. The necessary technologies for knowledge processing have
been implemented by applying the well-established tools and frameworks (e.g. OWL
reasoned and SPARQL engines) for the Semantic Web. In this way, a semantic-based model
with the functions, such as consistency check, reasoning and graph-based query, was built.
Different interfaces, which are based on a synchronous communication, are implemented as
prototypes for the integration of the model into the engineering process. These interfaces
enable the feature-based access (Feature API) and the query-based access (SPAQL API).
The concept, based on standardized exchange formats such as PLMXML and
AutomationML (Draht et al., 2008), enables the flexible import and export of engineering
data. The Knowledge Base Server (ifak-KBS) is a middleware which facilitates the
integration of the mechatronic model using different interfaces into the engineering tools.
A conclusion of the important features of the VIB configurator is listed as follows:
The development of a Virtual Commissioning workstation is a combination of tasks
such as selection, classification and design, which are specified in a defined workflow.
Assistance functions can be offered based on the knowledge of workflow and structure
as well as naming conventions.
The fundament of the assistance functions is the mechatronic plant model that was
implemented in a software framework, based on semantic technologies.
The introduced functions serve only as examples here, which can be supplemented by
other detailed analysis of the process.
Checklists and possible tests can be derived out of the available digital data.
5. Acknowledgment
This work is funded by the German Federal Ministry of Education and Research (BMBF)
within the project AVILUS.
6. References
Br, T. (2004). Durchgngige Prozesskette vom digitalen Produkt bis zur realen Fabrik, Congress
Digitale Fabrik Ludwigsburg, Germany (2004)
Bergert, M.; Diedrich, C. (2008). Durchgngige Verhaltensmodellierung von Betriebsmitteln
zur Erzeugung digitaler Simulationsmodelle von Fertigungssystemen, atp
Automatisierungstechnische Praxis, Vol.7 (2008), pp. 61-66, Oldenbourg
Industrieverlag, ISSN 2190-4111
Bracht, U. (2002). Anstze und Methoden der Digitalen Fabrik, Tagungsband Simulation und
Visualisierung, Magdeburg, pp. 1-11

Draht, R.; Lder, A.; Peschke, Jrn. & Hundt, L. (2008). AutomationML - The glue for
seamless automation engineering, Proceedings of the 13th international conference on
Emerging Technologies and Factory Automation, pp.616-623, ISBN 978-1-4244-1505-2,
Patras, Hamburg, September 15-18, 2008
Heeg, M. (2005). Ein Beitrag zur Modellierung von Merkmalen im Umfeld der Prozessleittechnik,
VDI-Verlag, ISBN 978-3-18-506008-3, Dsseldorf, Germany
Jain, Atul; Vera, D. A.; Harrison, R. (2010). Virtual commissioning of modular automation
systems, Proceedings of the 10th IFAC Workshop on Intelligent Manufacturing Systems,
pp. 79-84, ISBN: 978-3-902661-77-7, Lissabon, Portugal, July 1-2, 2010
Kiefer, J. (2007). Mechatronikorientierte Planung automatisierter Fertigungszellen im Bereich
Karosserierohbau, PhD thesis, Universitt des Saarlandes, Schriftenreihe
Produktionstechnik, ISBN 978-3-930429-72-1, Saarbrcken, Germany
Kiefer, J.; Bergert M. & Rossdeutscher M. (2010). Mechatronic Objects in Production
Engineering, atp Automatisierungstechnische Praxis, Vol.12 (2010), pp. 36-45,
Oldenbourg Industrieverlag, ISSN 2190-4111
McBride, B. (2010). An Introduction to RDF and the Jena RDF API, available from
http://jena.sourceforge.net/tutorial/RDF_API/index.html, Retrived on
22/02/2012
McGuinness, D.; van Harmelen, F. (2011). OWL web ontology language overview, available
from http://www.w3.org/TR/owl-features/
Mhlhause, M.; Suchold, N. & Diedrich, C. (2010). Application of semantic technologies in
engineering processes for manufacturing systems, Proceedings of the 10th IFAC
Workshop on Intelligent Manufacturing Systems, pp. 61-66, ISBN: 978-3-902661-77-7,
Lissabon, Portugal, July 1-2, 2010
Prud'hommeaux, E.; Seaborne, A. (2008). SPARQL Query Language for RDF, W3C
Recommendation
Rauprich G.; Haus C. & Ahrens W. (2002). PLT-CAE-Integration in gewerkebergreifendes
Engineering und Plant-Maintenance, atp - Automatisierungstechnische Praxis, Vol.2
(2002), pp. 50-62, Oldenbourg Industrieverlag, ISSN 2190-4111
Schreiber, S.; Schmidberger, T.; Fay, A.; May, J.; Drewes, J. & Schnieder, E. (2007). UML-
based safety analysis of distributed automation systems, Proceedings of the 12th
international conference on Emerging Technologies and Factory Automation, pp.1069-
1075, ISBN 1-4244-0826-1, Patras, Greece, September 25-28, 2007
TopQuadrant (2011). Getting Started with SPARQL Rules (SPIN), Version 1.2, Available from
http://www.topquadrant.com/spin/tutorial/SPARQLRulesTutorial.pdf, Retrived
on 22/02/2012
VDI 4499 Blatt 1 (2008). Digitale Fabrik Grundlagen. VDI-Handbuch Materialfluss und
Frdertechnik, Bd. 8, Grndruck. Berlin: Beuth
Volz, R. (2008). Web Ontology Reasoning with Logic Databases, PhD thesis, 17 Februar 2008,
Universitt Karlsruhe, Germany
Wollschlger M.; Braune A.; Runde S.; Topp U.; Mhlhause M.; Drumm O.; Thomalla C.;
Saboc A. & Lindemann L. (2009), Semantische Integration im Lebenszyklus der
Automation, Tagungsband zum Automation 2009 Kongress Baden-Baden, VDI-
Berichte 2067, pp. 167-170, VDI-Verlag, Dsseldorf, Germany

Automation 148
Wnsch, G. (2008). Methoden fr die Virtuelle Inbetriebnahme automatisierter Produktionssysteme,
PhD thesis, Technische Universitt Mnchen, Herbert Utz Verlag, ISBN 97838316
07952, Mnchen, Germany
Zh, M. F.; Wnsch, G. (2005). Schnelle Inbetriebnahme von Produktionssystemen, wt
Werkstattstechnik 95, Vol. 9 (2005), pp. 699-704
8
Introduction to the Computer
Modeling of the Plague Epizootic Process
Vladimir Dubyanskiy
1
, Leonid Burdelov
2
and J. L. Barkley
3
1
Stavropol Plague Control Research Institute,
2
M. Aikimbaevs Kazakh Science Center of Quarantine and Zoonotic Diseases,
3
Virginia Polytechnic Institute and State University
1
Russian Federation
2
Republic of Kazakhstan
3
USA
1. Introduction
Bubonic plague, caused by the bacteria, Yersinia pestis, persists as a public health problem
in many parts of the world, including Central Asia and Kazakhstan. Plague is a vector-borne
disease, i.e. the disease is spread by arthropod vectors that live and feed on hosts (Gage and
Kosoy, 2005). Great gerbil (Rhombomys opimus Liht., 1823, Rodentia, Cricetidae) is a
significant plague host in Central Asia and Kazakhstan region. Transmission happens
through bites of infected eas that earlier fed on infected hosts. This plague epizootic
process study can help future plague control work.
Modeling the epizootic process of plague in great gerbil settlements allows for quantitative
analyses of epizootic characteristics, which we are unable to control for in nature (Soldatkin
et al, 1973). Only one such attempt of this type of modeling is known (Soldatkin et al, 1966;
Soldatkin and Rudenchik, 1967; Soldatkin et al, 1973). In this study, the authors
demonstrated the high potential for this model, yet it possessed many limitations and had a
small work space (4,096 holes). The development of high-performance workstations, and
integration of geographic information systems (GIS) and remote sensing now makes it
possible to model complex, epizootic processes of plague.
We created a probabilistic, cellular automaton model of the epizootic process of plague by
using all of the above-named technologies. A probabilistic, cellular automaton with Monte-
Carlo, as the chosen statistical method, is the basic mechanism of the model (Etkins, 1987;
Grabovskiy, 1995). It is the SEIR (susceptible, exposed, infected and removed) model with
specific additions.
2. Description of the model
The model was created using Microsoft Excel 2003 with Visual Basic as the code language.
Microsoft Excel has 255 columns and 65,000 rows. A user-defined, fixed size of the modeling
area depends on the density of the gerbil colonies. A probabilistic, cellular automaton with
Monte-Carlo, as the chosen statistical method, is the basic mechanism of the model

Automation

150
(Dubyanskiy V.M., Burdelov L.A. , 2010, Dubyanskiy V.M, 2010). The choice for this model
type was rationalized by the following:
Commonly-used mathematical models that rely on differential equations require input
from many data sources. While, at the same time, the availability of plague surveillance
source data is rather insufficient.
Temporal models, (based on differential equations or other methods), are being used
for modeling without consideration of spatial distribution. In these models, it is
assumed that all events are happening at an abstract point rather than considering
plague focus, landscape region, or primary square.
If plague simulates nature, it analyzes only a static structure of plague focus or its parts.
The probabilistic, cellular automaton simulation allows the user to combine all of these
rationales. The epizootic process of plague in Great gerbil settlements was a good subject for
this experiment. The gerbils burrow system represents a time-spatial, discrete unit of an
epizootic process (Sedin, 1985). The data on burrow systems distribution within a plague
focus was determined by GIS and remote sensing. Satellite images of the burrow systems
provide a coherent picture of the colony structure (Burdelov et al., 2007; Addink et al., 2010)
at many plague foci (Fig 1.).
The cell/row structure of the models workspace allowed us to observe the proportion
between burrow systems and settlements and density of colonies. The density of colonies
was the main factor approximating the number of rodents. Both the density of colonies and
the number of rodents are significant factors for plague prediction (Klassovskiy et al., 1978;
Davis et al., 2004, 2008).
Several quantitative and qualitative relationships and characteristics between components
of the plagues parasitic system were identified early on (Burdelov et al., 1984; Begon et al.,
2006). Analyzing these numerous, character-combinations required that we use Monte-Carlo
simulation, rather than differential equation models, to describe each specific time stamp in
question. The analysis of burrow systems distribution on satellite images demonstrated that
the number of nearest neighbors in the burrow system often exceeded four (von Neumann
neighborhood). Hence, our model used the square (eight-cell) Moore neighborhood (Fig 2.).
The state of colonies can be categorized as infected, infecting, immune, or readied
for reinfecting. An infected colony is one that is infected, (in host, in vector or in
substratum), but has not spread the plague microbe (latent period). An infecting colony is
one that is infected and is spreading the plague microbe among other colonies. A colony that
has been cleared of the plague microbe and is not infectious is called an immune colony. A
readied colony is one which can be infected again after the immune period.
Data on the vectors and hosts infection were used to the extent that it was necessary for
calculating the models parameters. An infecting agent can be transmitted to other colonies
independent of the main mechanism of host-vector-host transmission. These processes
define whether a colony is infected or not and constitutes the black box of the model.
Nuances of an epizootic process which might be connected to specific mechanisms of plague
transmission have not been reviewed.
The cycle of infection is a duration when an infected colony is being converted to an
infecting colony and can spread the plague microbe among other colonies. The obligatory

Introduction to the Computer Modeling of the Plague Epizootic Process

151
scan of the models workspace occurs at the end of each cycle of infection. The probability of
transfer success is defined as one attempt of the plague microbe to transfer from an infected
colony to a non-infected colony. The number of attempts required to infect neighboring
colonies from one infecting colony during one cycle of infection is termed the coefficient
of colony infection. This term was proposed by Litvin et al., (1980). However, this term, in
our case, unites the coefficient of infection of a colony by Litvin and the coefficient of
transmission by Soldatkin et al., (1973). The rate (or the value) of epizootic contact is the
product of the successive transfer probability and the coefficient of a colony infection.

Fig. 1. Colonies of Great Gerbil in a satellite image. Each bright disc represents colonies 10-
40m in diameter. The image was captured using the publicly available software Google
Earth (http://earth.google.com/). Copyright 2008 DigitalGlobe; Europa Technologies.
1

Fig. 2. Moore's environment allows using all cells in model for its greater efficiency.
*
i t
L C C (1)
where L is a rate (or the value) of epizootic contact,
i
C the coefficient of colony infection,
t
C
the probability of transfer success.

Automation

152
The relative remote drift is the possibility of more or less remote colonies infecting without
nearer colonies infecting. This parameter can be regulated by the change of plague drift
distance as well as by the change of plague drift realization probability. It also allows
changing the speed of epizootic spread.
The structure of great gerbil settlements was reflected by altering colony-cells and non
colony-cells. This allowed working with different configurations of settlements and with
different densities of burrow systems. The scheme of burrow systems spatial location could
be integrated into the work space of the model from a map or from a satellite image if
necessary (Fig. 3). If the epizootic process was imitated, then the spatial location of burrow
systems would be defined by a random-number generator. Initial results were published
earlier (Dubyanskiy, Burdelov, 2008).

A B C
Fig. 3. Carry of a real configuration of settlement of Great Gerbil from the satellite image
adhered in GIS in working space of model. A - Allocated by means of GIS colonies of Great
Gerbil. B - Layer of GIS ready for carry to working space of model. C - Working space of
model with developing of plague epizooty.
There are two ways of spreading the plague microbe. The infected colony could attempt
to infect the other colony-cell as well as non colony-cell. In the latter case, the attempt of
infecting is useless. However, the tendency of gerbils to migrate among colonies will spread
the plague microbe to occupied colonies. The probability of transfer success is a probability
that a sick flea (or sick gerbil) will be transferred from an infected colony to a non-infected
colony and a microbiocenose of colony will be involved in the epizootic process. The
coefficient of a colony infection demonstrates how often transfers occurred (i.e. a gerbil can
visit non-infected colony two times or 10 times and so on).
The basic quantitative and qualitative characteristics were entered into the parameter sheet.
The value of parameters can be changed before modeling, although the dynamic changing
of parameters is possible. The parameters could be, for example, quantities of the cycle of
infection for one season (or session of modeling), the duration of the cycle of infection, the
density of colonies, and the probability of infecting neighbor colonies. The technical
conditions of the model work were also defined. The built-in menu allows comfortably
managing the modeling options.
Cellular flattened models have the edge effect. Naturally, this effect took place in our model
too. It was decided that all transmission of plague microbes outside the model workspace
was not consequential. In this case the model was made simpler. The model settlement
looked like an isolated island of the gerbils settlement in nature.


153
2.1 Basic parameters of the model
Even with the seeming simplicity of the model, the real quantity of fulfilled interrelations is
rather great. For example, infection of a colony depends on: the duration of cycle of
infection; the duration of the colonys status as infected; infectious quotient and immunity;
the probability of success of transfer; the coefficient of infection of a colony; the density and
distribution of colonies in the workspace of the model; and the number and arrangement of
infected colonies. Considering that further behavior of a cell depends on probability and
status of surrounding cells at PCA, to give the optimum and full analysis of influence of
such quantities of the interconnected parameters is very difficult. Therefore, initial selection
of the basic working parameters of the model were carried out in a literature search of data,
including results of direct or physical modeling of epizootic process using radioactive
isotopes in real settlements of the great gerbils. It is necessary to make a stipulation that the
analysis of literature has shown an absence of any unequivocal quantitative characteristic of
epizootic process in a required aspect.
After infection by the plague microbe, a colony becomes infected. This period in the
model lasts for 10 days where three days are the time of bacteremia and seven days are the
time of fleas prohibition (block of the proventriculus) if the flea fed on a sick animal. The
colony has the possibility to infect neighbors from day 11 to 50 from the beginning of its
own infection. It is not infectious from day 51 to 90, and it is immune (Novikova et al., 1971;
Naumov et al., 1972). The colony can be infected again after 90 days (Samsonovich et al.,
1971; Rothschild et al., 1975; Rothschild, 1978).
0
0,05
0,1
0,15
0,2
0,25
0,3
0,35
0,4
0,45
0,5
1 2 3 4 5 6 7 8 9 10
Distance in cells of working space of model
P
r
o
b
a
b
i
l
i
t
y

o
f

d
r
i
f
t

o
f

a

p
l
a
g
u
e

m
i
c
r
o
b
e

Fig. 4. Probability of drift of a plague microbe from colony to colony in working space of
model during one cycle of infection.

Automation

154
The probabilities distribution of plague microbe drift approximately corresponds to data of
direct modeling of drift of the fleas in the neighboring holes (Soldatkin, Rudenchik, 1971),
and to data on movement distances of gerbils to the neighbor colonies (Alekseev, 1974;
Davis et al., 2008) (Fig. 4). The probability of transfer success could be modified by the user
within the limits from zero to one. The coefficient of a colony infection could be varied from
one up to experimentally received or any other logically admissible value. The density,
distribution, occupation, and number of a plague microbe hosts expressed through
occupation were given as cell-colonies percentage, from quantity of cells in all working
spaces of the model and could be changed from zero to 100. It would be desirable to
emphasize that it was possible to change any parameters in the model.
The maximum drift of plague microbe in model can mark as l . Every infecting colony can
infect the neighbours in a square with diagonal 2l .The infecting colony is situated in the
center of this square. The probability of contact with any colonies inside this square defined as

1 2
( )
l
l
N i i R
Pz
N

(2)
where Pz is the probability of contact with the neighbour colony,
l
N is the general number
of colonies in the square with 2l diagonal,
1
i is the infected colony,
2
i is the infecting
colony, R is the immune colony.
The probability of colony infect is defined as
* Pi Pz Ct Ci (3)
where Pi is the probability of colony infect, Ct is the probability of transfer success (input
from experimentalist), Ci is the coefficient of colony infection (input from experimentalist).
Taking into consideration the formula (3)
*
I Z
P P L (4)
The intensity of model's epizooty is defined as

1 2
( ) / * 100 Y i i B (5)
where Y is the intensity of epizooty, B is the general number of colonies.
The quantity of infected colonies are increasing in the beginning as Fibonacci sequence

2 1
,
n n n
F F F n N

(6)
However when immune colonies appear, the complexity of the epizootic system increases.
The description of epizooty by differential equations are not possibe.
The basic parameters described above would be for a case studying the influence of spatial
parameters for epizootic processes. It is possible to establish a reverse problem to study the
influence of temporary parameter changes at a static, spatial configuration of the settlement.
In an ideal case, data can be accumulated on epizootic processes in nature and in
experiments on models.


155
For this study, the plague spread in various settlement types of great gerbils used the
following items: 1. The probability of transfer success was equal to one, 2. the coefficient of a
colony infection could be varied from one up to 10, and 3. The coefficient of a colony
infection was always one.
A two dimensional array of cells was used for the modeling world. The modeling square
was equal to 900 ha. The experiment was successful if the epizootic process continued
during 18 epochs (models half year) 11 times without a break. It was a 0.9999 probability
that the experiment was nonrandom (Rayfa, 1977). The start value of the coefficient of a
colony infection was equal to one. The start density of colonies were 1.33 per ha. The density
of colonies declined for 0.11 colonies per ha after each successful experiment for continuous
settlements and it was increasing the distance among colonies per one cell on vertical and
horizontal lines for narrow-band settlements. As a result, the density of the colonies
declined exponentially. If the experiment was not successful the coefficient of a colony
infection increase up per unit and recurred again.
3. Results
We gave an example of developmental epizooty in the feigned settlement of the great gerbil
(Fig. 5). The goal of the experiment was an examination of qualitative likeness between
modeling and natural epizootic processes. The size of the models working field was 2x2
km. It corresponded to an area of 400 hectares with density of colonies at 2.4 per hectare. A
conditional start date of epizooty was March 1
st
. The epizootic process was developed as a
result of the distribution of a plague microbe from one infected and random established
colony. The probability of transfer success and the coefficient of a colony infection were
constant and equal, 0.8 and 1 respectively.
The epizooty developed slowly at the initial stage. The analysis of the modeled epizootic
process showed that the critical time was the initial transfer of a plague microbe between
colonies. This was also noted by Soldatkin et al. (1971). The epizootic process could proceed
indefinably long if the infected colony transferred the plague microbe even to one not
infected colony during its infecting ability. The random hit of a primary infected
colony on a site of settlement where transfer of a plague microbe would occur raises the
probability of epizootic fade-out. The epizooty will end if the transfers of plague microbe
do not provide even short-term constant increase of number in the infected colonies (1.1
2 times during a minimum of 3 cycles of infection). If the start of epizooty was successful,
after three months, small foci, consisting of 4 8 infected colonies would have been formed
(Fig. 5a). The number of the infected colonies increased exponentially. On reaching a certain
level of infected colonies, the intensity of epizootic process started varying cyclically (Fig.
6). The exponential increase of the infected colonies number was terminated by virtue of two
causes. The first cause is the probabilistic character of the plague microbe transfer success
from colony to colony. If two or more colonies are infected by an infected colony, the
epizootic process intensity will exponentially increase. At the same time, an infected colony
can be a source of infection for all neighboring colonies or infect nothing. The second cause
is the relatively remote drift of the plague microbe that forms many small foci, consisting of
the infected colonies (Fig. 5b). These foci are dissemination centers of epizooty. The spatial
structure of the epizooty site is a complicating factor. It begins as a saturated epizooty site of
infected and immune colonies, whereby the plague microbe transfers to other

Automation

156
infected and immune colonies in greater numbers, decreasing the probability of
transferring to non infected colonies.

A B

C
Fig. 5. The visual representation of epizootic process had been developed at model. A) The
initial stage of the epizooty development (90 days). B) Remote drift and formation of small
foci (210 days). Quantity of the "infected" colonies had started an exponential increase. C) A
stage of "saturation" of the epizooty site and transition to cyclic fluctuations
Typically, two peaks of epizooty activity are registered during the year, in the spring and in
the fall (21). The epizootic contact rate increases during the rut in early spring and at the end
of the spring, as young gerbils explore their territory. Epizootic contact rate and process
intensity decreases during the high summer temperatures as the gerbils spend less time on
the surface. This also coincides with a generational replacement of the Xenopsilla fleas.
Contact increases again during the fall when the gerbils become hyperactive, resulting in a
second peak of epizooty activity. In autumn, the gerbils cache food and redistribute it
among the colonies before winter. The younger generations of fleas are actively feeding,
possibly transferring the plague microbe among the animals. This complex mechanism is
approximated through the rate of epizooty contact within the model. The exact data about


157
probability of infecting colonies in different seasons of the year are absent in literature.
Thus, for a qualitative look at the model and nature, the following conclusions were
garnered from the model: the probability of a colony infection in the summer is half of that
in the spring; a colony infection in the autumn is 0.15% less than in the spring; and a colony
infection in the winter is three times below that in the autumn.
Numerical values of the probability of transfer success and the coefficient of colony
infections were calculated earlier (Soldatkin et al, 1966, 1971). The probability of transfer
success was equal to 0.8 and the coefficient of a colony infection was equal to 1.0, for spring.
After entering these parameters, the picture of epizootic process in the model was
appreciably changed and looked much like a natural occurrence (Fig. 7).

0,00
5,00
10,00
15,00
20,00
25,00
30,00
1 19 37 55 73 91 109 127 145 163 181 199 217 235 253 271 289 307 325 343
Cycle of infection
"
I
n
f
e
c
t
e
d
"

c
o
l
o
n
i
e
s

(
%
)

Fig. 6. Intensity of epizootic process in model
There were rare, separate infected colonies present in the models space after the winter
period. Those colonies were in the initial stages of the annual cycle of epizooty (Fig. 7a).
There were many immune colonies that remained after the last epizooty in autumn. The
epizooty became sharper at the end of spring when the number of plagued colonies
increased, but the number of immune colonies decreased (Fig. 7b). In summer, the
epizooty recessed and the number of immune colonies increased again (Fig. 7c). In
autumn, the epizooty raised again as the number of plagued colonies increased and the
number of immune colonies decreased (Fig. 7d). Thus, the annual epizootic process
intensity had qualitative correspondence to a classic, twice-peak epizooty (Fig. 8) as in some
natural, plague foci.

Automation

158

A B

C D
Fig. 7. The epizootic process was developing with a changeable rate of epizootic contact.
Intensity of epizooty in the end: A) winter; B) spring; C) summer; D) autumn
0,00
5,00
10,00
15,00
20,00
25,00
W
i
n
t
e
r
S
p
r
i
n
g
S
u
m
m
e
r
F
a
ll
Seasons
"
I
n
f
e
c
t
e
d
"

c
o
l
o
n
i
e
s

(
%
)

Fig. 8. The intensity of epizootic process modeling within a year


159
It is interesting, that when a changeable rate of epizootic contact was entered into the models
parameters, the average number of the infected and immune colonies decreased from
18.92 to 9.31 (p <0.001, Mann-Whitney U test). It was caused by the increase in colonies which
were ready for a new infection. As a result, much sharper epizooty could develop. We
modeled the sharper epizooty. The coefficient of colony infection was increased to two for this.
The picture of epizooty is shown on Fig. 9 and Fig. 10. The epizooty intensity increased almost
double in spring and in the fall. However, the number of infected colonies was on par with
the pre-epizootic level in the winter-summer recession of epizootic activity. The speed of the
epizootic spreading fluctuated from 640 to 920 meters per month in these experiments.
It is known that the colonies of the great gerbil have a non-uniform distribution on the
surface. According to the distribution pattern, the groups of colonies - the settlement, may
be visible as a narrow-band, a wide-band, a large patch, a small patch and continuous
settlement (fig 11, 12). Some authors assumed that spatial distribution of the great gerbil
settlements played a big role in the epizootic process functioning (Pole, Bibicov, 1991;
Dubyanskiy et al., 1992). However that supposition was described as a quality parameter
without any quantity values and was quantitatively checked up by modeling. The narrow-
band settlements were presented as single colonies, arranged on an identical distance on the
vertical and horizontal lines (fig 13).

A B

C D
Fig. 9. The epizooty was developing with increasing of the rate of epizootic contact. Intensity
in the end A) autumn; B) winter; C) spring; D) summer

Automation

160
First experimental results demonstrated a significant difference of plague spread into
continuous and narrow-band settlements (table 1). When the coefficient of a colony infection
is one (it is the minimum value of model) the plague can be spread from a density of 1.33
colonies per ha into the continuous settlement. At the same time a density of 0.44 colonies
per ha is enough for plague epizootic into the narrow-band settlement. The difference of
densities equals three. As the density declines the coefficient of a colony infection
proportionally increases into the continuous settlement. A critical threshold is achieved
from a density of 0.78 colonies per ha. At the following step of density (0.67 colonies per ha)
the epizootic process is recommencing only from value of the coefficient of a colony
infection equal 10.
0,00
5,00
10,00
15,00
20,00
25,00
30,00
35,00
40,00
45,00
50,00
1 12 23 34 45 56 67 78 89 100 111 122 133 144 155 166 177 188 199 210
Cycle of infection
"
I
n
f
e
c
t
e
d
"

c
o
l
o
n
i
e
s

(
%
)

Fig. 10. The intensity of epizootic process which developed at the model with increasing of
the rate of epizootic contact (the right part of curve after cycle of infection 171).
In the narrow-band settlement, the minimum limit of density was achieved at 0.11 colonies
per ha, and the value of the coefficient of the colony infection was equal to eight. It is not
possible to reduce the density of colonies more because of the model limits in the plague
microbe drift.
Previous explanation of this phenomenon follows. As monitoring of the model shows, the
colonies in the continuous settlement were aggregated to a few groups. The plague epizootic
could develop under certain conditions:
1. if the number of colonies in separate groups are enough for plague circulating only in
this group;


161
2. if the distance among groups of colonies is not more than a limit of the relatively remote
drift;
3. if the movement activity of animals is enough for connection among the groups of the
colonies;
4. if the rate of epizootic contact is not the highest at the given density of colonies.
These conditions were not often established in the continuous settlement. The plague
epizootic was faded by the Allee principle (Odum, 1975). It is possible to check up these
items by the statistical data from table 1. The critical threshold of the colonies density
appeared when the average distance among the colonies had exceeded the maximum
possibility of the plague drift. At the same time the mode of spacing demonstrated that the
circulation plague microbe inside the groups of colonies was not limited by the distance.
The rare connections among the colonies and groups of colonies which were situated in the
plague microbe drift distance were more stable on the highest level of epizootic contact. It
could be about 25% of all colonies at 0.67 colonies per ha density. It demonstrated the first
quartile of the distance among colonies. Thus the plague microbe could circulate among 150
colonies in the model's space.

Fig. 11. The continuous settlement of the great gerbil. Colonies of Great Gerbil in a satellite
image. Each bright disc represents colonies 10-40m in diameter. The image was captured
using the publicly available software Google Earth (http://earth.google.com/). Copyright
2008 DigitalGlobe; Europa Technologies.

Automation

162

Fig. 12. The narrow-band settlement of the great gerbil. Colonies of Great Gerbil in a
satellite image. Each bright disc represents colonies 10-40m in diameter. The image was
captured using the publicly available software Google Earth (http://earth.google.com/).
Copyright 2008 DigitalGlobe; Europa Technologies.

A B
- plague colony, - immune colony, - susceptible colony
Fig. 13. The development of plague epizooty into narrow-band settlement (A) and into
continuous settlement (B).


163
Density per ha
The statistical characteristics of distance among colonies
(in cells)
The rate of
epizootic
contact
Continuous
settlement
Narrow-
band
settlement
Average 1 quartile Mode Median
1 1.33 0.44 7,33 2 1 5
2 1.11 0.23 8,09 2 0 0
3 1.00 0,17 8,99 2 0 6
4 0.89 0.14 10,11 2 0 7
5 0.78 0,11 10,11 2 0 7
6 0,78 0,11 10,11 2 0 7
7 0,78 0,11 10,11 2 0 7
8 0,78 0.11 10,11 2 0 7
10 0.67 0,11 15,61 4 2 11
Table 1. Dependence of plague spread from colonies density
In the narrow-band settlement the groups of colonies were absent. The colonies were
arranged in the form of a lattice and the colonies stood in the vertexes of this lattice. The
distance among the colonies was not more than the limit of the relatively remote drift. Each
surrounding colony contained enough numbers of colonies for supporting plague epizootic.
The colonies were situated in the better places for digging holes in nature (Dubyanskiy,
1970). The migration routes of rodents were not voluntary, and arrangement depended on
the settlement structure (Naumov et. al., 1972). The rodents moved along migration routes
despite low density, for searching mates, dissemination of young animals, etc., increasing
the epizootic contact rate (the dynamic density (Rall, 1965)). The bird-view (fig. 12)
demonstrated the limited routes of migration of the animals. These routes are lying along
verges of lattice. The similar direction of migration plays a big role in supporting of the
plague epizootic.
There is a great difference in the intensity of epizootic processes between the narrow-band
and the continuous settlements (fig. 14). The epizooty in the narrow-band settlement is more
intensive as in the continuous settlement. The significant difference is represented according
to Mann-Whitney U test p<=0.01.
If the epizootic contact rate was variable, the intensity of the epizootic process changed
according to the density of the colonies. As the epizootic process was more stable in the
narrow-band settlement, this type was used for experiments in the model. When the rate of
epizootic contact was to equal one the epizootic process was registered by the minimum
experimental density of colonies 0.11 per ha. However the probability of long-term
epizootic process was fewer less - 0.9-0.99.

Automation

164
Is persistence of plague possible when the number of great gerbil is in depression? It is an
important question for a problem of an interepizootic plague save solution.
It is known that great gerbils depression, inhabited colonies are often saved as groups which
were situated into better ecological niches (Marin et al., 1970; Burdelov L.A. & Burdelov V.A.,
1981; Dubrovskiy et. al., 1989). It was interesting that the number of colonies in separate
groups were necessary for persistent plague epizooty. The colonies were arranged as a united
group in the model space. The spacing among the colonies and the level (rate) of epizootic
contact changed during experiments. Results demonstrated in the table 2.
The minimum number of colonies necessary for plague persistent during the modeling year
was equal to 50 with maximum value of the probability of transfer success, the coefficient of
the colony infection and the spacing among the colonies. After decreasing the spacing among
the colonies to eight cells, the epizootic process was developed inside a group consisting of 40
colonies, but after decreasing the spacing to seven cells, 60 colonies were required. Each
decrease in the spacing among the colonies (and increasing of plague contact) the number of
colonies necessary for supporting plague epizooty increased. The spatial structure of
epizootic process was reminiscent of a "layer cake" (fig. 15) or wave of infected colonies.
The external layer is a layer of infected coloniesfollowed by a layer of immune colonies.
The "readied" colonies were arranged around the first infected colony inside the "Layer
cake". The drift of plague microbe from infected to "readied" colonies was difficult. The
irregularity of epizootic process, which is one of the main conditions of long existing
epizooty, disappeared. Thus the epizootic process degenerated. If the rate of contact
decreases, the effect of "wave is neutralized. In total the plague can persist in a group of 25
colonies when epizootic contact rate is equal to one. The biology base of this effect was
confirmed in nature. When great gerbils' density was in deep depression, the movement of
rodents decreased and the level of epizootic contacts decreased too (Naumov et al., 1972).
0,00
5,00
10,00
15,00
20,00
25,00
30,00
35,00
40,00
1 18 35 52 69 86 103 120 137 154 171 188 205 222 239 256 273 290 307
Cycle of infection
I
n
f
e
c
t
e
d

c
o
l
o
n
i
e
s

(
%
)
continuous settlement
the narrow-band settlement

Fig. 14. The intensity of epizootic process in the different type of great gerbils settlements in
model. The density of narrow-band settlement is 0.44 colonies per ha, the density of the
continuous settlement is 1.33 colonies per ha.


165

- immune colonies, - infected colonies

Fig. 15. The spatial development of epizooty at the high density of colonies and the highest
epizootic contact rate.
The modelling showed that 25 colonies were enough for the plague to persist for a long
time. The epizootic contact level had non-line (threshold) dependence from density and
number of colonies.

The level of epizootic
contact
The colonies quantity
The distance among
colonies (cells)
10 50 10
10 40 8
10 60 7
10 50 6
10 70 5
10 400 4
6,48 60 4
1,2 40 2
1,14 30 2
1,0 25 0
Table 2. The level of epizootic contact, the distance among colonies and minimum of
colonies quantity, when the plague persistence during 1 year period

Automation

166
4. Discussion
It is important to remember the following for a quality assessment of the modeling and real
epizootic process. The investigation of the plagues spatial structure had shown that the
small foci of plagued colonies represented pools of the neighbor colonies where there had
been living sick or immune gerbils and their infested ectoparasites (Dubyanskiy, 1963;
Klassovskiy et al., 1978; Rothschild, 1978; Rivkus et al., 1985). Each pool was irregularly
located on the surface and separated by significant distance from other identical pools. It
allowed for observing all phases of the epizootic process in a small focus at one time. Sick
rodents were observed in some parts of colonies, while an increased fraction of immune
rodents were observed in other parts of the colonies. Some colonies were without the plague
and these were reserved for epizooty prolongation. All authors agreed on the irregularity of
epizootic process in the small focus and permanent moving of epizooty in the territory of
the plague focus.
The model-based process was analogous to the described-above epizooty at the small focus
or at the pool of the small foci (so-called epizooty spot). The analogous types of colonies
were present in the model. The well-defined pools of colonies (small foci) were separated
from each other and non-uniformity distributed in the working space of the model. The
speed of the model-based epizooty spreading coordinated well with field experiments
through the marking of animals with radioisotopes and the direct observation of epizooty in
nature. The cyclicity arising in the model simulated fluctuations of natural epizootic process
intensities (Dubyanskiy and Burdelov, 2010) that confirmed an adequacy of the model-
based process to real epizooty.
We have shown the possibility of modeling an annual epizootic cycle by changing the
seasons epizootic contact rate. This demonstrated the adequacy of the model in
transmission theory framework. At the same time, the qualitative picture of epizootic cycle
phases was simulated in the model, shown by an increase in the immune colonies fraction
after autumn or spring epizooty peaks. Even a minor increase (up to two) in the single
parameter, coefficient of colony infection, was an important factor, as it led to epizooty
outbreak. A doubling in the coefficient of colony infection could indicate a gerbil familys
contact rate with other families as two times in 10 days instead of only one time. Our
thinking, based on literature data, was that both values of the coefficient of colony infection
(1 or 2) are usual for the density of a colony, which we used in the model. It demonstrates
how minor fluctuations of the epizootic contact rate can lead to the strong changes in the
epizootic situation. It is necessary to note that the endless epizootic process can develop
even in a small model area. It can be considered as indirect acknowledgement of high
survivability of a similar probabilistic process. It was correct for the real epizootic process at
natural foci and it was often proved in practice.
It was quantitatively demonstrated in the big role of spatial distribution of great gerbils for
plague epizootic. The various kinds of burrow systems constituting settlements, was
marked by M. A. Dubyanskiy (1963, 1963a, 1970). He showed that lines of separate kinds of
burrow systems played a different role in the plague spread. The modeling demonstrated
spatial arrangement is very significant for plague epizootic too. It was known some plague
foci had band settlements. These are the foci from North desert subzone. The
settlements from these foci are arranged along dry riverbeds, ravines, gullies. The plague in
these foci was more intense and had more rare interepizootic periods as in the foci with the


167
continuous settlements (Aikimbaev et al., 1987). We believe the spatial kind of great
gerbil settlements was a significant influence to this phenomenon.
The background epizooty of low density in the occupancy colonies have proved the
opinion that a negative result of surveillance can result not only from a real absence of
plague in the focus, but also from insufficient intensity and extent of the surveillance
(Dubyanskiy et al., 2007; Dubyanskiy, 2008). Culling surveillance area of plague foci is
possible provided 0.1 0.6 colonies per ha.
When the great gerbils density was set in the deep depression, the plague microbe was not
detected. However, the plague epizooties were restarted after the gerbils density
recovered. Such disappearing of the plague microbe, resulted in a long discussion about
routes of plague microbe conservation during interepizootic period. It was believed that the
plague microbe requires specific mechanisms for conservation as a soil plague and the
transmission mechanism by fleas was not effective (Akiev, 1970; Burdelov, 2001; Popov,
2002). Special works were made in Mangyshlak (Marin et al., 1970) in the South Balhash area
(Burdelov L.A. & Burdelov V.A., 1981) and in Kizilkum (Dubrovskiy et al., 1989) it was
shown that the occupancy colonies were arranged individually as a group, from five
colonies to several. It is enough for plague microbes circulating for a long time. At the same
time the most difficult to detect plague by unspecific surveillance because the groups of
colonies were rare arranged on the biggest (more over 1000 square kilometers) area.
Moreover the plague microbe was not conserved for every colony group. It is a cause,
possibly, unregistering of epizooty and existing interepizootic periods. The modeling
experiments demonstrated that the classical transmission theory can explain a complexity of
plague detected during depression of gerbils and a beginnings of plague at a normal density
of rodents.
The modeling allows searching numeric parameters of the epizootic process of plague.
However, the simulation may appear a perspective method of forecasting plague epizootics,
if applied along with permanent comparing and correcting of modeling results with the real
surveillance situation.
The number of great gerbil burrow systems was nearest to natural data about quantity and
distribution colonies during deep depressions of great gerbils number. Thus, the permanent
plague epizootics could occur even during deep depressions of great gerbils number.
5. Conclusion
The model can be useful for studying the general appropriateness of the plagues epizootic
process and for obtaining formerly inaccessible quantitative characteristics. It is the result of
model simplicity, availability of high-performance computing workstations, model
parameter input by anyone, and a large display area in the model. The model allows
displaying a real and modeled structure of the rodents settlement. Thus, it was possible to
study the spatial regulation of the epizootic process depending on the density of colonies,
distribution of colonies, intrasettlement migration, and so forth. Remote sensing methods
and GIS can be used as a base for modeling. The basic characteristics of simulated epizootic
processes qualitatively also are quantitatively close to natural analogues. It is important to
emphasize the model is not constrained by species-specificity. It can be used for modeling
epizootic processes among different species and for different vector-borne diseases.

Automation

168
6. References
Addink E.A., S.M. de Jong, S.A. Davis, V. Dubyanskiy, L.A. Burdelov & H. Leirs. (2010). The
use of high-resolution remote sensing for plague surveillance in Kazakhstan.
Remote Sensing of Environment No. 114, pp. 674-681
Akiev A.K. (1970). Condition of a question on studying the mechanism of preservation of the
plague in interepizootic years. (Review). Problems of especially dangerous infections.
Vol 4 (14), pp. 13-33
Aikimbaev A.M., Aubakirov S.A., Burdelov A.S., Klassovskiy L.N., Serjanov O.S. The Nature
desert plague focus of Middle Asia. Alma-Ata. Nauka. 1987. 207 p.
Alekseev A. F. (1974). The long of live and peculiarity of moving of Great Gerbils in North-
West Kizilkum. The material VIII scientist conference plagues control establishments of
Central Asia and Kazakhstan pp. 220-222, Alma-Ata
Begon M., Klassovskiy N., Ageyev V., Suleimenov B., Atshabar B. et al.. (2006).
Epizootiological parameters for plague (Yersinia pestis infection) in a natural
reservoir in Kazakhstan. Emerging Infectious Diseases. Vol. 12, Issue 2, pp. 268-273
Burdelov L. A. (2001). Sources, the reasons and consequences of crisis in plague epizootology.
Quarantine and zoonotic infections in Kazakhstan, Vol. 3, pp. 20-25.
Burdelov L. A., Burdelov A. S., Bondar E. P., Zubov V. V., Maslennikova Z. P., Rudenchik N.
F. (1984). Hole using of great gerbil (Rhombomys opimus, Rodentia, Cricetidae)
and epizootically significance empty colonies in the Central Asia foci. Zoology
journal, No. LXIII, Vol. 12, pp. 1848-1858.
Burdelov L. A., Burdelov V. A. (1981). The optimum ecotope and ways of great gerbils survive in
unfavorable conditions. The ecology and medical significance gerbils in the USSR fauna,
Proceedings of national meeting by ecology and medical significance gerbils as importance
rodents of arid zone, pp 129-130, Moscow, Russia.
Burdelov L. A., Davis S., Dubyanskiy V.M., Ageyev V. S., Pole D. S., Meka-Mechenko V. G.,
Heier L., Sapojnikov V. I. (2007) The prospect use remote sensing in monitoring of
epidemiology of plague. Quarantine and zoonotic infections in Kazakhstan, Vol. 1-2 (15-
16), pp. 41-48
Davis S., Trapman P., Leirs H., Begon M., Heesterbeek J.A.P. (2008) The abundance
threshold for plague as a critical perculation phenomenon. Nature. Vol. 454, pp.
634-637
Davis S, Begon M, De Bruyn L, Ageyev VS, Klassovskiy NL, Pole SB, Viljugrein H, Stenseth
NC, Leirs H. (2004). Predictive thresholds for plague in Kazakhstan. Science, Vol. 304,
pp. 736-738. (doi:10.1126/ science.10.95854).
Rivkus J.Z.,Mitropolskiy O.V.,Urmanov R.A.,Belyaeva S.I. (1985).
Features of epizootic plague among rodents Kyzylkum, In: Fauna and ecology of the
rodents, ol. 16. pp. 5-106. Publishing house of Moscow University
Dubrovskiy J. A., Mitropolskiy O.V., Urmanov R.A., Tretjakov G.P., Zakirov R.H., Sarjinskiy
B.A., Fotteler E.R. (1989). The natural laws of great gerbil arrangement during
perennial depression of its number in Central Kizilkum. The gerbils - importance
rodents of USSR arid zone. Fan, Tashkent, pp. 42-44.
Dubyanskiy M. A. (1963). About exterior of plague epizooty on a different phases in
settlements of great gerbil. Materials of scientist conference, pp. 76-78. Alma-Ata
Dubyanskiy M. A. (1970). The ecology structure of great gerbils setllements in Priaral Karakum.
Abstract of Ph. D. thesis. Alma-Ata


169
Dubyanskiy M. A. (1963). The settlement types of great gerbil and theirs epizotology
significant in Priaral Karakum Zoology journal, Vol. 42, No. 1, pp. 103-113
Dubyanskiy M. A., Kenjebaev A., Stepanov V. V., Asenov G. A., Dubyanskaja L. D. (1992).
Forecasting epizootic activity of a plague in Near- Aral region and Kizilkum. Nukus:
Karakalpakstan.
Dubyanskiy V.M., Burdelov L.A. (2008). A possibility of mathematic modeling of plague
epizootic process in GIS-space. Modern technologies in the implementation of the global
strategy of combating infectious diseases over the territories of the CIS member states.
Release IX international. science.-pract. conf. CIS member states, pp. 195-197. Volgograd
Dubyanskiy V.M., Burdelov L.A. (2010). The computer model of plagues epizootic process
in Rhombomys opimus settlements and some results of its researching. Zoology
journal, Vol 89, No 1, pp. 7987
Dubyanskiy, V. M. (2008). Interval evaluation of intensity of plague epizootics. Quarantinable
and zoonotic infections in Kazakhstan. Almaty, Vol 1-2 (17-18), pp. 52-57
Dubyanskiy, V. M., Pole S. B., Burdelov L.A., Klassovskaja E.. V., Sapojnikov V. I. (2007).
Depending on plague detection in Bakanas plain on completeness of
epizootologival survey. Quarantinable and zoonotic infections in Kazakhstan. Vol. 1-2
(15-16), pp.62-66.
Dubyanskiy V. M. (2010). Modelling plague spread in the different kind settlements of Great
Gerbil (Rhombomys opimus Liht., 1823, Rodentia, Cricetidae), In: Current issuaes of
zoonotic diseases. International conference. 29 March 2010. Ulaanbaatar, pp. 92-99
Etkins P. (1987). Order and disorder in Nature. M.: Mir.
Grabovskiy V. I. (1995). Cellular automata as a simple model of complication system.
Progress of modern biology, No. 115, pp. 412 418.
Gage, K. L., & Kosoy, M. Y. (2005). Natural history of plague: Perspectives from more than a
century of research. Annual Review of Entomology, No. 50, pp. 505528
Klassovskij L. N, Kunitskiy V. N., Gauzshtejn D. M, Burdelov A. S., Aikimbaev M. A.,
Dubovizkiy N. M., Novikov G. S., Rasin B. V., Savelov U.V., (1978). To a question
on short-term forecasting epizootic situations in Ili-Raratal country between two
rivers. The condition and prospects prophilaxis of plagues: Thesis on USSRs conference
plague control institute "Microbe", pp. 69-70. Saratov
Litvin V. U., Karulin B. E., Vodomorin N. A., Ohotskij J. V. (1980). Radioisotope modeling
epizootic situations and stochastic model of rabbit-fever epizooty in stack, In: Fauna
and ecolgy of rodents. Vol. 14, pp. 63-84. Publishing house of Moscow University
Marin S.N., Kamnev P.I., Korinfskiy A.N., Nikitenko G.I. (1970). The depression of great
gerbils number on Mangishlak. Problems of particularly dangerous infections, Vol. 4
(14), pp. 149-157.
Naumov N. P., Lobachev V. S., Dmitriev P. P., Smirin V. M. (1972). The nature plague focus in
Priaral Karakum. Moscow. MGU.
Novikova T. N., Pershin I. B., Gubaydullina G. S., Reshetnikova P. I., Sviridov G. G.,
Kunitskiy V. N. (1971). Results of serology study of Great Gerbils after its infected
by plague fleas. The material V scientist. conference plague control establishments of
Central Asia and Kazakhstan, pp. 148-151. Alma-Ata
Odum E. (1975). Fundamentals of ecology. oscow. Mir.

Automation

170
Pole S. B., Bibicov D. I.. (1991). The dynamic of structure and the mechanisms supporting the
optimal density of grey marmot population. The structure of marmots population,
pp. 148-171. Moscow
Popov N.V. (2002). Discrete character the basic spatial-temporal peculiarity of plague
manifestations in natural foci of souslic type. Saratov University Press.
Rall J.M. (1965). Natural nidality and epizootology of plague. Moscow. Medicina
Rayfa G. (1977). Decision analysis. oscow. Nayka
Rothschild E.V. (1978). Spatial structure of the natural foci of a plague and methods of its studying.
Moscow. The Moscow State University.
Rothschild .., Postnikov G.B., Dankov S.S, Kosarev V.P. (1975). Development of groups of
plague holes at epizooty among Great Gerbils. Problems of particularly dangerous
infections, Vol. 3-4 (43-44), pp. 70-77.
Samsonovich L. G., Peysahis L. A., Gordienko O. J. (1971). Susceptibility, sensitiveness and
intensity immunity of Great Gerbils after infected it by leucine-depended strain of
plague microbe from different foci. The material V scientist. conference plague control
establishments of Central Asia and Kazakhstan, pp. 169-172. Alma-Ata
Sedin V.I. (1985). Hole's of Great Gerbils as elements of structure plague epizooty in the
Central and the Western Kara-Kum. XII interrepublican scientist conference plague
control establishments of Central Asia and Kazakhstan on prophylaxis of a plague, pp. 141-
143. Alma-Ata
Soldatkin I. S, Rodnikovskiy V. B., Rudenchik J. V. (1973). Experience of statistical modeling
of plague epizootic process. Zoology journal, Vol. 52, pp. 751-756.
Soldatkin I. S., Rudenchik J. V. (1967). About correlation between model and nature plague
epizooty process. The material V scientist. conference plague control establishments of
Central Asia and Kazakhstan, to the 50 anniversary of the Great October Socialist
revolution, pp 70-71. Alma-Ata
Soldatkin I. S., Rudenchik J. V., 1971. Some questions a plague enzooty as a self-regulating
system rodent-flea-causative agent, In: Fauna and ecology of the rodents, Vol. 10. pp.
5 29. Publishing house of Moscow University
Soldatkin I. S., Rudenchik J. V., Ostrovsky I. B., Levoshina A. I. (1966). Quantitative
characteristics of condition development of plague epizooty in Great Gerbils
settlements. Zoology journal, Vol. 14, pp. 481-485
0
Learning Automation to Teach Mathematics
Josep Ferrer
1
, Marta Pea
1
and Carmen Ortiz-Caraballo
2
1
Universitat Politecnica de Catalunya
2
Universidad de Extremadura
Spain
1. Introduction
One of the recurrent controversies at our Engineering Faculty concerns the orientation of
rst year basic courses, particularly the subject area of mathematics, considering its role
as an essential tool in technological disciplines. In order to provide the basic courses with
technological applications, a mathematical engineering seminar was held at the Engineering
Faculty of Barcelona. Sessions were each devoted to one technological discipline and aimed
at identifying the most frequently used mathematical tools with the collaboration of guest
speakers from mathematics and technology departments.
In parallel, the European Space of Higher Education process is presented as an excellent
opportunity to substitute the traditional teaching-learning model with another where students
play a more active role. In this case, we can use the Problem-Based Learning (PBL)
method. This environment is a really useful tool to increase student involvement as well as
multidisciplinarity. With PBL, before students increase their knowledge of the topic, they
are given a real situation-based problem which will drive the learning process. Students
will discover what they need to learn in order to solve the problem, either individually or
in groups, using tools provided by the teacher or facilitator, or found by themselves.
Therefore, a collection of exercises and problems has been designed to be used in the PBL
session of the rst course. These exercises include the applications identied in the seminar
sessions and would be considered as the real situation-based problems to introduce the
different mathematic topics. Two conditions are imposed: availability for rst year students
and emphasis on the use of mathematical tools in technical subjects in later academic years.
As additional material, guidelines for each technological area addressed to faculties without
engineering backgrounds are dened.
Some material on Electrical Engineering has been already published ((Ferrer et al., 2010)).
Here we focus on control and automation. The guidelines and some exercises will be
presented in detail later on. As general references on linear algebra see for example (Puerta,
1976) and on system theory (Kalman et al., 1974) and (Chen, 1984). For other applications, see
(Lay, 2007).
Here we describe some of the items regarding control and automation that are presented in
the guideline (Section 2).
9
2 Will-be-set-by-IN-TECH
(1) The input-output description: black box, input-output signals, impulse response, linearity,
causality, relaxedness, time invariance, transfer-function matrix, time domain, frequency
domain, Laplace transform, Fourier transform, gain, phase, poles, zeros, Bode diagram,
lters, resonances.
(2) The state-variable description: the concept of state, state equation, output equation,
transfer-function matrix, linear changes, feedbacks, realizations, stability, reachable
states, controllability, control canonical form, pole assignment, observability, Kalman
decomposition.
The exercises (Section 3) related to this area cover the following topics:
Matrices. Determinant. Rank: Composition of Systems (ex. 1), Controllability Matrix,
Controllable Systems and Controllability Indices (ex. 2), Realizations (ex. 3).
Vector Spaces. Bases. Coordinates: States in Discrete Systems (ex. 4), Control Functions in
Discrete Systems (ex. 4).
Vector Subspaces: Reachable States for one or several Controls and Sum and Intersection
of these Subspaces (ex. 5).
Linear Maps: Changes of Bases in the System Equations and Invariance of the
Transfer-Function Matrix (ex. 6), Controllability Subspace and Unobservable Subspace
(ex. 7), Kalman Decomposition (ex. 8).
Diagonalization. Eigenvectors, Eigenvalues: Invariant Subspaces and Restriction to an
Invariant Subspace (ex. 7), Controllable Subsystem (ex. 10), Poles and Pole assignment
(ex. 9).
Non-Diagonalizable Matrices: Control Canonical Form (ex. 11)
2. Guideline for teachers
(1) External system description
Systems are considered as "black boxes" in which each input u(t) (input, control, cause,...)
causes an output y(t) (output, effect,...), both multidimensional vectors, in general. We
consider only known inputs, ignoring other ones like, for example, disturbance, noise...
u
y
The most usual inputs will be piecewise continuous functions, built from the elementary
functions or standard "signals" (impulse or delta, step, ramp, sawtooth, periodics...). Simple
systems are adders, gain blocks, integrators, pure delays, lters...
In general, the aimis to analyse their behaviour without looking inside. Indeed, if we consider
the "impulse response" g(t; ) (that is, the output when the input is an "impulse" in ,
) and
the system is linear, it results
y(t) =
g(t; )u()d.
172 Automation
Learning Automation to Teach Mathematics 3
The upper integration limit will be t if we assume that the system is "causal" (that is, if the
current response does not depend on the future inputs, as in all physical systems) and the
lower will be t
0
(or simply 0) if it is "relaxed" at t
0
(that is, y(t) = 0 for t t
0
, when u(t) = 0
for t t
0
). Finally, if it is "time-invariant", instead of g(t; ) we can write g(t ), resulting in
the following expression of y(t) as a convolution product of g(t) and u(t)
y(t) =
g(t )u()d = g(t) u(t),

which is the general system representation in the "time domain".
Applying Laplace transform we get the general representation in the "s-domain"
y(s) = G(s) u(s),
where G(s) = g(s) is called "transfer function matrix". Indeed, it is the focus of study in this
external representation.
If we do the change of variable s = j (imaginary axis of the "complex plane") we obtain the
representation in the "frequency domain", (more usual in engineering)
y() = G()u(),
where G() is called "isochronous transfer function matrix". It can be obtained directly as a
"Fourier" transform when appropriate hypotheses hold. This allows the use of tools such as
Fourier transform, Parseval theorem... if we have some basic knowledge of functional analysis
and complex variable: function spaces, norms, Hilbert spaces, integral transforms...
Its denomination shows that G() indicates the system behaviour for each frequency. So,
if u(t) = Asin(t), then y(t) = B sin(t + ), being the "gain" B/A the module of G(),
and the "phase" is its argument. A widely used tool in engineering is the "Bode diagrams"
which represent these magnitudes on the ordinate (usually, the gain is in logarithmic scale, in
decibels: dB = 20log|G|) as a function of the frequency as abscissa (also in log scale).
In generic conditions (see next section) the coefcients of G are "proper rational fractions", so
that the systems behaviour is largely determined by its "degrees", "zeros" and "poles". So:
The relative degree (denominator degree minus numerator degree) gives the "order of
differentiability" at the origin of the response to the input step signal.
As already mentioned, this difference must be strictly positive (if zero, the Parseval
theorem would give innite energy in the output signal; if negative, it would contradict
the causality), so that the gain tends to 0 for high frequencies.
Roughly speaking, the zeros indicate ltered frequencies (of gain 0, or in practice far
below), so that in the Bode diagram place the "inverted comb pas" as "lters". On the
contrary, the poles indicate dangerous frequencies because of resonance (innite gain).
(2) Internal system description
In addition, state variables x(t) (not univocally dened) are considered. They characterize the
state in the sense that they accumulate all the information fromthe past, that is, future outputs
are determined by the current state and the future inputs. Typically, the derivative x(t)
173 Learning Automation to Teach Mathematics
functionally depends on x(t), u(t) and t (state equation), and y(t) as well (output equation),
although in this case we may obviate the dependence on u(t). In the linear case:
x(t) = A(t)x(t) + B(t)u(t); y(t) = C(t)x(t).
From elementary theory of ordinary differential equations, it holds that for every continuous
(or piecewise continuous) control u(t), there exists a unique "solution" x(t) for every "initial
condition" x
0
= x(0).
In particular, it can be applied a "feedback control" by means of a matrix F
u(t) = Fx(t).
One of the rst historical (non linear) examples is the Watt regulator which controlled the
velocity of a steam engine acting on the admission valve in the function of the centrifugal
force created in the regulator balls by this velocity. Nowadays this "automatic regulation" can
be found in simple situations such as thermostats.
When this kind of feedback is applied, we obtain an autonomous dynamical system
x(t) = Ax(t) + BFx(t) = (A + BF)x(t)
called a "closed loop" system. It is natural to consider if we can adequately choose F such that
this system has suitable dynamic properties. For example, for being "stable", that is, that the
real parts of the eigenvalues of A+BF are negative. Or more in general, that these eigenvalues
have some prexed values. As we will see later, one of the main results is that this feedback
"pole assignment" is possible if the initial system is "controllable".
If it is time-invariant (that is, A, B, and C are constant) and we assume x(0) = 0, the Laplace
transform gives
y(t) = G(s) u(s); G(s) = C(sI A)
1
B
recovering the previous transfer matrix. Reciprocally, the "realization" theory constructs
triples (A, B, C) giving a prexed G, formed by proper rational fractions: it can be seen that it
is always possible; the uniqueness conditions will be seen later.
It is a simple exercise to check that when introducing a "linear change" S in the state variables,
the system matrices become S
1
AS, S
1
B and CS, respectively, and G(s) do not change.
In this description it is clear that the coefcients of G(s) are proper rational fractions and in
particular its poles are the eigenvalues of A.
One of the main results is that the set of "reachable states" (the possible x(t) starting from
the origin, when varying the controls u(t)) is the image subspace ImK(A, B) of the so-called
"controllability matrix"
ImK(A, B); K(A, B) =
B AB . . . A
n1
B
,
that is, the subspace spanned by the columns of B and successive images for A, called
"controllability subspace". It is an interesting exercise to justify that we can truncate in n 1,
since the columns of higher powers are linearly dependent with the previous ones. In fact we
can consider each control individually (each column of B separately), in such a way that the
174 Automation
"sum and intersection of subspaces" give the reachable states when using different controls at
the same time or for each one of them.
The systemis "controllable" if all the states are reachable, that is, if and only if K(A, B) has full
"rank". Hence, it is a generic condition (that is, the subset of controllable pairs (A, B) is open
and dense).
In the single input case (A, b), it can be seen that if the system is controllable it can be
transformed, by means of a suitable basis change, in the so-called "control canonical form"
A =
0 1 0 . . . 0 0
0 0 1 . . . 0 0
. . . . . .
0 0 0 . . . 0 1
. . .
,

b =
0
0
.
.
.
0
1
.
Observe that

A is a "companion" (or Sylvester) matrix. For these kinds of matrices, it can be
easily seen that the coefcients of the characteristic polynomial are the ones of the last row
(with opposite sign and in reverse order), and they are "non derogatory matrices", that is, for
each eigenvalue there is a unique linearly independent eigenvector, and therefore only one
Jordan block. Hence, if some eigenvalue is multiple, the matrix is non diagonalizable.
In the multi input case, another reduced form is used (Brunovsky, or Kronecker form), which
is determined by the so-called "controllability indices" that can be computed as a conjugated
partition from the one of the ranks of: B, (B, AB),...,K(A, B).
From these reduced forms it is easy to prove that the pole assignment is feasible, as well as to
compute the suitable feedbacks.
If the system is not controllable, it is easy to see that the subspace ImK(A, B) is "A-invariant",
and that in any "adapted basis" (of the state space) the matrices of the system are of the form:
A
c

0
B
c
0
,
being (A
c
, B
c
) controllable. It is also easy to deduce that if x(0) belongs to ImK(A, B), x(t)
also belongs, for all time t and all control u(t), and that its trajectory is determined for the pair
(A
c
, B
c
), which enables considering the "restriction" of the system (A, B) to ImK(A, B), called
the "controllable subsystem" of the initial one.
In the same direction, the "Kalman decomposition" is obtained by considering "(Grassman)
adapted bases to the pair of subspaces" ImK(A, B) and Ker L(C, A), being
L(C, A) =
C
CA
. . .
CA
n1
that is, the "transposed matrix" of K(A

t
, C
t
). In fact there are interesting properties of "duality"
between the systems (A, B, C) and (A
t
, C
t
, B
t
).
We remark the equivalence between the controllability of (A, B, C) and the "observability"
of (A
t
, C
t
, B
t
), in the sense that the initial conditions are computable if the outputs for
determined inputs are known.
On the other hand it is interesting to note that the transfer matrix of the initial system is the
same as that of the controllable subsystem, as well as that of the "complete subsystem" (that
is, controllable and observable) obtained by means of the Kalman decomposition.
This means that, given any transfer matrix (formed by proper rational fractions), not only it
is possible to nd realizations, but also "controllable realizations" and even complete ones. In
fact it can be seen that all the complete realizations are equivalent. In particular, they have
the same number of state variables. Moreover, a realization is complete if and only if it is a
"minimal realization", in the sense that there are not realizations with a smaller number of
state variables. This minimal number of state variables of the realizations is called "McMillan
degree", which coincides with the dimension of the complete realizations.
3. Exercises for students
We will present here the guideline of proposed exercises for the students and their solutions:
3.1 Proposed exercises
1. Composition of systems
A control system
u
y
is dened by the equations
x(t) = Ax(t) + Bu(t); y(t) = Cx(t),
or simply by the "triple of matrices" (A, B, C). This triple determines the "transfer matrix"
G(s) = C(sI A)
1
B
which relates the Laplace transforms of u, y:
y(s) = G(s) u(s).
Given two systems:
1
: x
1
= A
1
x
1
+ B
1
u
1
; y
1
= C
1
x
1
2
: x
2
= A
2
x
2
+ B
2
u
2
; y
2
= C
2
x
2
they can be composed of different ways to obtain a new system. For example in:
(i) Series
1

2
u
y
176 Automation
(ii) Parallel
2
u
y
(iii) Feedback
2
u
y
In all these cases the new state variables are:
x =
x
1
x
2
(1) Deduce the following relations between the different input and output variables:
(1.i) u = u
1
; y
1
= u
2
; y
2
= y
(1.ii) u = u
1
= u
2
; y = y
1
+ y
2
(1.iii) u + y
2
= u
1
; y
1
= y = u
2
(2) Deduce the expression of the triple of matrices of each composed system, in terms of
A
1
, B
1
, C
1
, A
2
, B
2
, C
2
.
(3) Deduce the expression of the transfer matrix of each composed system, in terms of G
1
(s)
and G
2
(s).
2. Controllable systems; controllability indices
A control system
x(t) = Ax(t) + Bu(t), A M
n
, B M
n,m
is "controllable" (that is, any change in the values of x is possible by means of a suitable control
u(t)), if and only if the rank of the so-called "controllability matrix" is full
K(A, B) = (B, AB, A
2
B, . . . , A
n1
B).
(a) Discuss for which values of , R the system dened by
A =
0 1 0
2 1 0
0 1
, B =
1 0
0
0 1
is controllable.
(b) Discuss for which values of , R the system is controllable with only the second
control, that is, when instead of the initial matrix B we consider only the column matrix
b
2
=
.
(c) In general, the "controllability indices" are determined by the rank of the matrices
B, (B, AB), (B, AB, A
2
B), . . . , K(A, B).
Compute these ranks, in terms of parameters , R, for the system dened by
A =
0 1
0 1
0 1
0 1 0
0
0 0 1
0
, B =
0 0
0 0
0
0
1 0
0 0
0 1
.
3. Realizations
Given a matrix G(s) of proper rational fractions in the variable s, it is called a "realization" any
linear control system
x(t) = Ax(t) + Bu(t); y(t) = Cx(t)
which has G(s) as "transfer matrix", that is,
G(s) = C(sI A)
1
B.
In particular, the so-called "standard controllable realization" is obtained in the following way,
assuming that G(s) has p rows and m columns.
(i) We determine the least common multiple polynomial of the denominators
P(s) = p
0
+ p
1
s + . . . + p
r
s
r
.
(ii) Then, G(s) can be written
G(s) =
1
P(s)
G
11
(s) G
1m
(s)

G
p1
(s) G
pm
(s)
,
where G
ij
(s) are polynomials with degree strictly lower than r(= deg P(s)).
(iii) Grouping the terms of the same degree we can write G(s) in the form
G(s) =
1
P(s)
(R
0
+ R
1
s + . . . + R
r1
s
r1
),
where R
0
, . . . , R
r1
M
p,m
(R).
178 Automation
(iv) Then, the standard controllable realization is given by the triple of matrices (where 0
m
is
the null matrix of M
m
(R)):
A =
0
m
I
m
0
m
. . . 0
m
0
m
0
m
I
m
. . . 0
m
. . . . . . . . .
0
m
0
m
0
m
. . . I
m
p
0
I
m
p
1
I
m
. . . . . . p
r1
I
m
M
mr
(R), B =
0
m
.
.
.
0
m
I
m
M
mr,m
(R),
C =
R
0
. . . R
r1
M
p,mr
(R).
We consider, for example,
G(s) =
1/s
1/(s 1)
.
(1) Following the above paragraphs, compute the triple of matrices (A, B, C) which give the
standard controllable realization.
(2) Check that it is controllable, that is, that the matrix
K(A, B) = (B, AB, . . . , A
mr1
B)
has full rank.
(3) Check that it is a realization of G(s), that is,
1/s
1/(s 1)
= C(sI A)
1
B.
4. Reachable states; control functions
Given a linear control system
x(k +1) = Ax(k) + Bu(k); A M
n
(R), B M
n,m
(R),
the h-step "reachable states", from x(0), are:
x(h) = A
h
x(0) + A
h1
Bu(0) + . . . + ABu(h 2) + Bu(h 1)
where all possible "control functions" u(0), u(1), . . . are considered. More explicitly, if we write
B = (b
1
. . . b
m
) and u(k) = (u
1
(k), . . . , u
m
(k)), we have that:
x(h) = A
h
x(0)
+ (A
h1
b
1
u
1
(0) + . . . + A
h1
b
m
u
m
(0)) + . . .
+ (Ab
1
u
1
(h 2) + . . . + Ab
m
u
m
(h 2)) + (b
1
u
1
(h 1) + . . . + b
m
u
m
(h 1))
Let
A =
2 1 1
2 1 1
2 1 3
, B =
1 1
1 0
.
(1) Assume that only the rst control acts, that is: u
2
(0) = u
2
(1) = . . . = 0. Show that in this
case, the state x = (1, 2, 1) is not 3-step reachable from the origin.
(2) Assume now that only the second control acts.
(2.1) Prove that x = (1, 2, 1) is 2-step reachable from the origin.
(2.2) Compute the corresponding control function.
(3) Assume that both controls act.
(3.1) Determine the control functions set to reach x = (1, 2, 1) from the origin, at second
step.
(3.2) In particular, check if it is possible to choose positive controls.
(3.3) Idem for u
1
(0) = u
2
(0).
5. Subspaces of reachable states
x(k +1) = Ax(k) + Bu(k); A M
n
(R), B M
n,m
(R),
the h-step "reachable states", from x(0), are
x(h) = A
h
x(0) + A
h1
Bu(0) + . . . + ABu(h 2) + Bu(h 1),
where all possible control functions u(0), u(1), . . . are considered.
(1) We write K(h) the set of these states when x(0) = 0. Show that:
(1.1) K(h) = [B, AB, . . . , A
h1
B] R
n
(1.2) K(1) K(2) . . . K(j) . . .
(1.3) K(h) = K(h +1) K(h +1) = K(h +2) = . . .
(1.4) K(n) = K(n +1) = . . .
This maximal subspace of the chain is called "subspace of reachable states":
K = [B, AB, . . . , A
n1
B] R
n
(2) Analogous results hold when only the control u
i
(k) acts. In particular, the subspace of
reachable states, from the origin, with only this control is:
K
i
= [b
i
, Ab
i
, . . . , A
n1
b
i
] R
n
; 1 i m
where B = (b
1
, . . . , b
m
).
(2.1) Reason that the reachable states by acting the controls u
i
(k) and u
j
(k) are K
i
+ K
j
.
(2.2) Reason that K
i
K
j
is the subspace of reachable states by acting any of the controls
u
i
(k) or u
j
(k).
(3) Let us consider the linear control system dened by the matrices
A =
2 1 1
2 1 1
2 1 3
, B =
0 0
1 1
1 0
.
180 Automation
(3.1) Determine the subspaces K, K
1
and K
2
, and construct a basis of each one.
(3.2) Idem for K
1
+ K
2
, K
1
K
2
.
6. Change of state variables in control systems
In the linear control system
x(t) = Ax(t) + Bu(t); y(t) = Cx(t)
A M
n
, B M
n,m
, C M
p,n
,
we consider a linear change in the state variables given by:
x = S
1
x.
(1) Prove that in the new variables the equations of the system are:
x(t) =

A x(t) +

B u(t); y(t) =

C x(t)
A = S
1
AS,

B = S
1
B,

C = CS.
(2) The "controllability indices" of the system are computed from the ranks:
rank(B, AB, . . . , A
h
B), h = 1, 2, 3, . . .
Deduce from (1) that they are invariant under linear changes in the state variables.
(3) The "transfer matrix" of the system is:
G(s) = C(sI A)
1
B.
Deduce from (1) that it is invariant under linear changes in the state variables.
7. Controllability subspaces and unobservability subspaces
x(t) = Ax(t) + Bu(t); y(t) = Cx(t)
A M
n
(R), B M
n,m
(R), C M
p,n
(R),
the following subspaces are called "controllability subspace" and "unobservability subspace",
respectively:
K = Im
B AB . . . A
n1
B
L = Ker
C
CA
. . .
CA
n1
.
(1) Show that they are A-invariant subspaces.
(2) Let us consider
A =
1 1 0 0
0 1 0 0
0 0 1 1
0 0 0 1
, B =
1
0
0
1
, C =
0 1 0 1
.
(2.1) Compute the dimensions of K and L.
(2.2) Construct a basis of each one.
(2.3) Obtain the matrices, in these bases, of the restrictions A|
K
and A|
L
.
(2.4) Idem for the subspace K L.
8. Kalman decomposition
x(t) = Ax(t) + Bu(t); y(t) = Cx(t)
A M
n
(R), B M
n,m
(R), C M
p,n
(R),
we consider the subspaces
K = ImK(A, B), L = Ker L(C, A),
where
K(A, B) =
B AB . . . A
n1
B
,
L(C, A) =
C
CA
. . .
CA
n1
.
A (Grassman) adapted basis to both subspaces is called a "Kalman basis". More specically,
the basis change matrix is of the form
S =
S
1
S
2
S
3
S
4
S
2
: basis of K L
(S
1
S
2
) : basis of K
(S
2
S
4
) : basis of L
where some of the submatrices S
1
, . . . , S
4
can be empty.
(1) Prove that with a basis change of this form, the matrices of the systembecome of the form
A = S
1
AS =
A
11
0 A
13
0
A
21
A
22
A
23
A
24
0 0 A
33
0
0 0 A
43
A
44
182 Automation
B = S
1
B =
B
1
B
2
0
0
C = CS =
C
1
0 C
3
0
.
It is called "Kalman decomposition" of the given system.
(2) Consider the system given by
A =
1 1 0 0
0 1 0 0
0 0 1 1
0 0 0 1
, B =
1
0
0
1
, C =
0 1 0 1
.
(2.1) Determine a Kalman basis.
(2.2) Determine a Kalman decomposition.
(2.3) Check that
(A
11
, B
1
, C
1
) is controllable and observable,
A
11
0
A
21
A
22
B
1
B
2
C
1
0
is controllable,
A
11
A
13
0 A
33
B
1
0
C
1
C
3
is observable,
that is, that the following matrices have full rank:
K(A
11
, B
1
) =
B
1
A
11
B
1
. . . (A
11
)
n1
B
1
,
(L(C
1
, A
11
))
t
=
C
t
1
A
t
11
C
t
1
. . . (A
t
11
)
n1
C
t
1
,
K
A
11
0
A
21
A
22
B
1
B
2
B
1
B
2
. . .
A
11
0
A
21
A
22
n1
B
1
B
2
C
1
C
3
A
11
A
13
0 A
33
t
=
C
t
1
C
t
3
. . .
A
t
11
0
A
t
13
A
t
33
n1
C
t
1
C
t
3
.
9. Pole assignment by state feedback
B
A
u x x
x(t) = Ax(t) + Bu(t); A M
n
, B M
n,m
,
the eigenvalues of the matrix A are called the "poles" of the system and play an important
role in its dynamic behaviour. For example, they are the resonance frequencies; the system is
"BIBO stable" if and only if the real part of its poles is negative.
If we apply an automatic control by means of a "feedback" F (for example, a thermostat, a
Watt regulator,...)
B
A
F
u x x
we obtain an autonomous dynamic system (called a "closed loop" system):
x(t) = Ax(t) + BFx(t) = (A + BF)x(t)
with matrix A+ BF instead of the initial matrix A. We ask if it is possible to choose adequately
F such that the new poles (that is, the eigenvalues of A + BF) have some prexed desired
values. For example, with negative real part, so that the automatic control is stable.
A main result of control theory ensures that this pole assignment by feedback is possible if the
initial system is controllable, that is, if rank(B, AB, A
2
B, . . . , A
n1
B) = n.
We will prove and apply that in the one-parameter case (m = 1). We will do it by transforming
the initial systemin the so-called "control canonical form" by means of a suitable basis change.
In this form, the feedback computation is trivial, and nally it will only be necessary to undo
the transformation.
(1) Consider a one-parameter control system
x(t) = Ax(t) + bu(t); A M
n
, b M
n,1
,
and let
1
, . . . ,
n
be the desired eigenvalues for the feedback system (each one repeated
so many times as its algebraic multiplicity). The main hypothesis is that the system is
controllable, that is, rank(b, Ab, A
2
b, . . . , A
n1
b) = n.
Then it is possible to nd a linear change S in the state variables such that the new matrices
of the system have the so-called "control canonical form":
A
c
= S
1
AS =
0 1 0 . . . 0 0
0 0 1 . . . 0 0
. . . . . .
0 0 1 . . . 0 1
a
n
a
n1
a
n2
. . . a
2
a
1
, b
c
= S
1
b =
0
.
.
.
0
1
for certain coefcients a

1
, . . . , a
n
.
184 Automation
(1.1) Check that these coefcients are the same (with opposite sign) than the ones of the
characteristic polynomial:
Q(t) = (1)
n
(t
n
a
1
t
n1
. . . a
n1
t a
n
).
(1.2) Deduce that it is straightforward to nd F
c
such that:
eigenvalues(A
c
+ bcF
c
) = {
1
, . . . ,
n
}.
(1.3) Prove that F = F
c
S
1
is the sought feedback, that is, that:
eigenvalues(A + bF) = {
1
, . . . ,
n
}.
(2) Consider the particular case:
A =
0 0 1
0 1 1
1 0 1
, b =
1
1
1
.
(2.1) Check that the basis change
S =
2 1 1
2 0 1
1 0 1
transform the initial matrices into the control canonical form.

(2.2) Compute a feedback F such that:
eigenvalues(A + bF) = {
1
,
2
,
3
}.
10. Controllable subsystem
x(t) = Ax(t) + Bu(t); y(t) = Cx(t) (1)
A M
n
, B M
n,m
, C M
p,n
,
the following matrix is called its "controllability matrix"
K(A, B) =
B AB A
2
B . . . A
n1
B
.
It can be seen that the subspace spanned by its columns K = ImK(A, B) is the set of reachable
states from the origin, and it is called "controllability subspace". We denote d = dimK.
(1) Prove that K is an A-invariant subspace.
We consider a basis change S in the state space adapted to K, and we denote by x
c
the rst
d new coordinates and by x
uc
the remaining ones:
x = S
1
x =
x
c
x
uc
, x
c
M
d,1
.
(2) Check that:
x K x
uc
= 0
that is:
x K x = S
1
x =
x
c
0
.
(3) Prove that the equations of the system in the new variables are:
x
c
(t)
x
uc
(t)
=

A
x
c
(t)
x
uc
(t)
+

Bu(t)
A = S
1
AS =
A
c

0
, A
c
M
d
B = S
1
B =
B
c
0
, B
c
M
d,m
C = CS =
C
c
C
uc
, C
c
M
p,d
.
(4) Deduce that if the initial state belongs to K, it also belongs all the trajectory, for any applied
control u(t):
x(0) K x(t) K, t, u(t).
Therefore, it makes sense to consider the "restriction" to K of the initial system:
x
c
(t) = A
c
x
c
(t) + B
c
u(t); y
c
(t) = C
c
x
c
(t). (2)
The following paragraphs justify the interest of this subsystem and its denomination as
"controllable subsystem" of the initial one.
(5) Justify that the trajectories of system (1) in K can be computed by means of equations (2)
of the subsystem and relation (2).
(6) Prove that this subsystem is "controllable", that is, that the following matrix has full rank:
B
c
A
c
B
c
. . . A
d1
c
B
c
.
(7) Prove that the controllable subsystem has the same "transfer matrix" (which, we recall,
reects the input/output behaviour) than the initial system, that is, that:
C(sI A)
1
B = C
c
(sI A
c
)
1
B
c
.
11. Control canonical form
For controllable systems, the so-called "control canonical form" simplies the computations,
for example, for the pole assignment by feedback. We are going to obtain it for the
one-parameter case.
(1) Consider the system
x(t) = Ax(t) + bu(t)
186 Automation
A =
0 0 1
0 1 1
1 0 1
, b =
1
1
1
.
(1.1) Check that it is controllable, that is, that:
rank
b Ab A
2
b
= 3.
(1.2) Check that doing the basis change
S
1
=
A
2
b Ab b
we get
A = S
1
1
AS
1
=
2 1 0
0 0 1
1 0 0
,

b = S
1
b =
0
0
1
.
(1.3) Check that with the additional change
S
2
=
1 0 0
2 1 0
0 2 1
we get the control canonical form

A
c
= S
1
2

AS
2
=
0 1 0
0 0 1
1 2 0
, b
c
= S
1
2

b =
0
0
1
.
(2) Consider the system
x(t) = Ax(t) + bu(t); A M
n
, b M
n,1
which is assumed to be controllable:
rank
b Ab . . . A
n1
b
= n
and let Q(t) be the characteristic polynomial of A
Q(t) = (1)
n
(t
n
a
1
t
n1
. . . a
n1
t a
n
).
(2.1) Check that doing the basis change
S
1
=
A
n1
b . . . Ab b
we get
A = S
1
1
AS
1
=
a
1
1 0 . . . 0
a
2
0 1 . . . 0
. . .
a
n1
0 0 . . . 1
a
n
0 0 . . . 0
,

b = S
1
1
b =
0
.
.
.
0
1
.
(2.2) Check that with the additional change
S
2
=
1 0 . . . 0 0
a
1
1 . . . 0 0
a
2
a
1
. . . 0 0
. . .
a
n1
a
n2
. . . a
1
1
we get the control canonical form

A
c
= S
1
2

AS
2
=
0 1 0 . . . 0 0
0 0 1 . . . 0 0
. . .
0 0 0 . . . 0 1
a
n
a
n1
a
n2
. . . a
2
a
1
, b
c
= S
1
2

b =
0
.
.
.
0
1
.
(3) As an application, we use the control canonical form to check that A is non-derogatory:
(3.1) Prove that if
i
is an eigenvalue of A, then:
rank(A
C
i
I) = n 1.
(3.2) Deduce that A is non-derogatory, and so it does not diagonalize if some of its
eigenvalues are multiple.
3.2 Solutions
1. Solution
(1) It follows immediately from the observation of the diagrams.
(2.i)
x
1
= A
1
x
1
+ B
1
u
1
= A
1
x
1
+ B
1
u
x
2
= A
2
x
2
+ B
2
u
2
= A
2
x
2
+ B
2
y
1
= A
2
x
2
+ B
2
C
1
x
1
y = y
2
= C
2
x
2
Hence:
x =
x
1
x
2
A
1
0
B
2
C
1
A
2
x
1
x
2
B
1
0
u
y =
0 C
2
x
1
x
2
(2.ii) Reasoning in the same way, it results:
A
1
0
0 A
2
B
1
B
2
,

C
1
C
2
188 Automation
(2.iii) Analogously:
A
1
B
1
C
2
B
2
C
1
A
2
B
1
0
,

C
1
0
(3.i) y = y
2
= G
2
u
2
= G
2
y
1
= G
2
G
1
u
1
= G
2
G
1
u. Hence, the transfer matrix is:
G
2
G
1
(3.ii) Analogously:
G
1
+ G
2
(3.iii) y = y
1
= G
1
u
1
= G
1
( u + y
2
) = G
1
u + G
1
G
2
u
2
= G
1
u + G
1
G
2
y
(I G
1
G
2
) y = G
1
u
y = (I G
1
G
2
)
1
G
1
u
2. Solution
(a) The controllability matrix is
K(A, B) =
1 1 0 2
0 2 2
0 1 1 +1
,
which clearly has full rank for all , . Hence, the given system is controllable for all , .
(b) When only the second control acts, that is to say, when u
1
(t) = 0 for all t, the system can
be written
x(t) = Ax(t) + b
2
u
2
(t).
Then, the controllability matrix is reduced to
K(A, b
2
) =
1

1 1 +1
,
which has full rank for = 0, =
2
.
(c) Now one has
K(A, B) =
0 0 0 0 0 0 1 0
0 0 0 0 1 0 0 0
0 0 1 0 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0
1 0 0 0 0 0 0 0 0 0
0 0 0 1 0 0 0 0 0 0
0 1 0 0 0 0 0 0 0 0
.
The "controllability indices" can be computed fromthe ranks of the 2, 4, 6, . . . rst columns:
rank(B) = 2, for all ,
rank(B, AB) = 4, for all ,
rank(B, AB, A
2
B) = 5 if = = 0; rank(B, AB, A
2
B) = 6 otherwise
rank(B, . . . , A
3
B) = 6 if = = 0; rank(B, . . . , A
3
B) = 7 otherwise
rank(B, . . . , A
4
B) = 7
3. Solution
(1) Let
G(s) =
1/s
1/(s 1)
.
Then: p = 2, m = 1. Following the given pattern:
(i) P(s) = s(s 1) = s + s
2
r = 2, p
0
= 0, p
1
= 1, p
2
= 1
(ii) G(s) =
1
s+s
2
s 1
s
(iii) G(s) =
1
s+s
2
1
0
1
1
R
0
=
1
0
, R
1
=
1
1
(iv) A =
0 1
0 1
, B =
0
1
, C =
1 1
0 1
(2) rank K(A, B) = rank
0 1
1 1
= 2
(3) (sI A)
1
=
1 0
0 1
0 1
0 1
1
=
s 1
0 s 1
1
=
1
s
2
s
s 1 1
0 s
C(sI A)
1
B =
1 1
0 1
1
s
2
s
s 1 1
0 s
0
1
=
1
s
2
s
1 1
0 1
1
s
=
1
s
2
s
1 + s
s
= G(s)
4. Solution
(1) x = (1, 2, 1) is 3-step reachable, from the origin, if there are u
2
(0), u
2
(1), u
2
(2) such that:
x = A
2
b
1
u
1
(0) + Ab
1
u
1
(1) + b
1
u
1
(2)
=
10
8
8
u
1
(0) +
2
2
2
u
1
(1) +
0
1
1
u
1
(2)
190 Automation
or equivalently::
1
2
1
10 2 0
8 2 1
8 2 1
u
1
(0)
u
1
(1)
u
1
(2)
No solutions exist because

rank
10 2 0
8 2 1
8 2 1
= 2 rank
10 2 0 1
8 2 1 2
8 2 1 1
= 3
(2) Analogously:
1
2
1
1 0
1 1
1 0
u
2
(0)
u
2
(1)
whose solution is: u

2
(0) = 1, u
2
(1) = 1.
(3.1) Analogously:
1
2
1
2 1 0 0
2 1 1 1
2 1 1 0
u
2
(0)
u
2
(0)
u
1
(1)
u
2
(1)
The solutions can be parameterized by u

1
(0) as follows:
u
2
(0) = 1 + u
1
(0)
u
1
(1) = 1 2u
1
(0) u
2
(0) = 4u
1
(0)
u
2
(1) = . . . = 1 4u
1
(0)
(3.2) It is not possible, because u
1
(0) and u
1
(1) have opposite signs.
(3.3) It is possible u
1
(0) = u
2
(0) = 1. Then u
1
(1) = 4, u
2
(1) = 5.
5. Solution
(1.1) When x(0) = 0, one has:
x(k) = (A
k1
b
1
u
1
(0) + . . . + A
k1
b
m
u
m
(0))
.
.
.
+ (Ab
1
u
1
(k 2) + . . . + Ab
m
u
m
(k 2))
+ (b
1
u
1
(k 1) + . . . + b
m
u
m
(k 1))
where B = (b
1
, . . . , b
m
) and u(0), . . . , u(k 1) run over all possible control functions.
Therefore
K(h) = [A
k1
b
1
, . . . , A
k1
b
m
, . . . , Ab
1
, . . . , Ab
m
, b
1
, . . . , b
m
] = [B, AB, . . . , A
k1
B].
(1.2) It is obvious from (1.1).
(1.3) K(h) = K(h +1) if and only if A
h
b
1
, . . . , A
h
b
m
K(h).
Then A
h+1
b
1
, . . . , A
h+1
b
m
K(h +1), so that K(h +2) = K(h +1).
(1.4) As dimK(h) n, the length of the increasing chain is n at most.
(2)
K
i
+ K
j
= [b
i
, Ab
i
, . . . , A
n1
b
i
] + [b
j
, Ab
j
, . . . , A
n1
b
j
]
= [(b
i
, b
j
), A(b
i
, b
j
), . . . , A
n1
(b
i
, b
j
)]
(3.1) K = Im
0 0 2 1 10 2
1 1 2 1 8 2
1 0 2 1 8 2
= R
3
K
1
= Im
0 2 10
1 2 8
1 2 8
= Im
0 2
1 2
1 2
A basis of K
1
: (0, 1, 1), (1, 0, 0)
K
2
= Im
0 1 2
1 1 2
0 1 2
= Im
0 1
1 1
0 1
A basis of K
2
: (0, 1, 0), (1, 0, 1)
(3.2) K
1
+ K
2
= Im
0 1 0 1
1 0 1 0
1 0 0 1
= R
3
K
1
= {y + z = 0}
K
2
= {x + z = 0}
K
1
K
2
= {y + z = x + z = 0}
A basis of K
1
K
2
: (1, 1, 1)
6. Solution
(1) If x = S
1
x, then:
x = S
1
x = S
1
(Ax + Bu) = S
1
AS x + S
1
Bu
y = Cx = CS x
(2) rank(

B,

A

B, . . . ,

A
h

B) = rank(S
1
B, S
1
ASS
1
B, . . . , (S
1
AS)
h
S
1
B)
= rank S
1
(B, AB, . . . , A
h
B) = rank(B, AB, . . . , A
h
B)
(3)

G(s) =

C(sI

A)
1

B = CS(sI S
1
AS)
1
S
1
B = CS(S
1
(sI A)S)
1
S
1
B
= CSS
1
(sI A)
1
SS
1
B = G(s)
7. Solution
(1) From Cayley-Hamilton theorem:
A
n
= a
0
I + a
1
A + . . . + a
n1
A
n1
192 Automation
Therefore, if x Im
B AB . . . A
n1
B
, then
Ax Im
AB A
2
B . . . A
n
B
Im
B AB . . . A
n
B
,
and
Im
B AB . . . A
n
B
= Im
B AB . . . (a
0
B + a
1
AB + . . . + a
n1
A
n1
B)
= Im
B AB . . . A
n1
B
= K.
If x L, then:
Cx = CAx = . . . = CA
n1
x = 0
Clearly:
C(Ax) = . . . = CA
n2
(Ax) = 0
It is sufcient to prove that CA
n1
(Ax) = 0, but
CA
n1
(Ax) = CA
n
x
= C(a
0
I + a
1
A + . . . + a
n1
A
n1
)x
= a
0
Cx + a
1
CAx + . . . + a
n1
CA
n1
x = 0
(2.1) K = Im
1 1 1 1
0 0 0 0
0 1 2 3
1 1 1 1
; dimK = 3
L = Ker
0 1 0 1
0 1 0 1
0 1 0 1
0 1 0 1
; dimL = 4 2 = 2
(2.2) Basis of K: (u
1
, u
2
, u
3
)
u
1
= (1, 0, 0, 1), u
2
= (1, 0, 1, 1), u
3
= (1, 0, 2, 1)
Basis of L: (v
1
, v
2
)
v
1
= (1, 0, 0, 0), v
2
= (0, 0, 1, 0)
(2.3) Au
1
= u
2
; Au
2
= u
3
; Au
3
=
1
0
3
1
= u
1
+ u
2
u
3
Mat A|
K
=
0 0 1
1 0 1
0 1 1
Av
1
= v
1
; Av
2
= v
2
Mat A|
L
=
1 0
0 1

(2.4) K = {(x
1
, x
2
, x
3
, x
4
) : x
2
= 0}
L = {(x
1
, x
2
, x
3
, x
4
) : x
2
= x
4
= 0}
K L = L
8. Solution
(1) We recall (see ex. 7) that K and L are A-invariant subspaces. Therefore:
A(S
2
) [S
2
]
A(S
1
) [S
1
, S
2
]
A(S
4
) [S
2
, S
4
]
so that

A = S
1
AS has the stated form.
Moreover ImB K. Hence,

B = S
1
B [S
1
, S
2
].
Finally, as L Ker C, we have C(S
2
) = C(S
4
) = 0.
(2.1) According to the solution of ex. 7, we can take:
S
2
=
1 0
0 0
0 1
0 0
S
1
=
0
0
0
1
S
3
=
0
1
0
0
and S
4
is empty. Hence:
S =
0 1 0 0
0 0 0 1
0 0 1 0
1 0 0 0
(2.2)
A = S
1
AS =
1 0 0 0
0 1 0 1
1 0 1 0
0 0 0 1
B = S
1
B =
1
1
0
0

C = CS =
1 0 0 1
That is to say:
A
11
=
A
13
=
0 0
A
21
=
0
1
A
22
=
1 0
0 1
A
23
=
1
0
A
33
=
B
1
=
B
2
=
1
0
C
1
=
C
3
=
(2.3) rank K(A

11
, B
1
) = rank
1 1
= 1
rank(L(C
1
, A
11
))
t
= rank
1 1
= 1
194 Automation
rank K
A
11
0
A
21
A
22
B
1
B
2
= rank
1 1 1
1 1 1
0 1 2
= 3
rank
C
1
C
3
A
11
0
A
21
A
33
t
= rank
1 1
1 1
= 2
9. Solution
(1.1) It is easy to check that there is only one main minor of each sign. Then:
a
1
= tr A
c
, a
2
= det
0 1
a
2
a
1
, a
3
= det
0 1 0
0 0 1
a
3
a
2
a
1
, . . . , a
n
= det A
c
.
(1.2) If F
c
=
f
1
. . . f
n
, then
A
c
+ b
c
F
c
=
0 1 . . . 0 0
0 0 . . . 0 0
. . .
0 0 . . . 0 1
a
n
+ f
1
a
n1
+ f
2
. . . a
2
+ f
n1
a
1
+ f
n
Therefore, f
1
, . . . , f
n
are the solutions of:
t
n
(a
1
+ f
n
)t
n1
. . . (a
n1
+ f
2
)t (a
n
+ f
1
) = (t
1
)(t
2
) (t
n
).
(1.3)
eigenvalues(A + bF) = eigenvalues(SA
c
S
1
+ Sb
c
F
c
S
1
)
= eigenvalues(S(A
c
+ b
c
F
c
)S
1
)
= eigenvalues(A
c
+ b
c
F
c
) = {
1
, . . . ,
n
}.
(2.1) It is straightforward to check that:
S
1
AS =
0 1 0
0 0 1
1 2 0
, S
1
b =
0
0
1
.
(2.2) First, we look for F
c
=
f
1
f
2
f
3
such that:
t f
3
t
2
(2 + f
2
) (1 + f
1
) = (t
1
)(t
2
)(t
3
)
Therefore:
f
3
=
1
+
2
+
3
f
2
=
1
2
+
2
3
+
3
1
2
f
1
=
1
3
+1
Finally: F = F
c
S
1
.
10. Solution
(1) From Cayley-Hamilton theorem, if the characteristic polynomial of A is
Q(t) = (1)
n
(t
n
a
1
t
n1
. . . a
n1
t a
n
),
then
A
n
= a
n
I + a
n1
A + . . . + a
1
A
n1
.
Therefore, if x K = Im
B AB . . . A
n1
B
, then
Ax Im
AB A
2
B . . . A
n
B
Im
B AB . . . A
n1
B (a
0
B + . . . + a
n1
A
n1
B)
= K
(2) If (u
1
, . . . , u
d
, . . . , u
m
) is a basis of the state space adapted to K, then a state x belongs to K
if and only if the last n d coordinates are 0.
(3) In the conditions of (2), the d rst columns of

A are Au
1
, . . . , Au
d
, which belong to K (see
(1)). Hence, again from (2), their last n d coordinates are 0. The same argument works
for B, because its columns belong to K.
(4) From (3), it is clear that x
uc
(0) = 0 implies x
uc
(t) = 0 for any control u(t).
(5) Again from (3), if x
uc
(t) = 0, the remainder coordinates x
c
(t) are determined by system
(2).
(6) By hypothesis
rank
B AB . . . A
n1
B
= d.
The rank is preserved under changes of bases. Hence:
d = rank

B

A

B . . .

A
n1

B
= rank
B
c
0
A
c
B
c
0
. . .
A
n1
c
B
c
0
= rank
B
c
A
c
B
c
. . . A
n1
c
B
c
= rank
B
c
A
c
B
c
. . . A
d1
c
B
c
where the last equality follows from Cayley-Hamilton theorem.

(7) Recall that the transfer matrix is preserved under changes of bases:
C(sI A)
1
B =

CS
1
(sI S

AS
1
)
1
S

B =

CS
1
(S(sI

A)S
1
)
1
S

B =
=

CS
1
(S
1
)
1
(sI

A)S
1
S

B =

C(sI

A)

B
From (2):
(sI

A)
1
=
sI
A
c

1
=
sI
d
A
c

0
1
=
(sI
d
A
c
)
1
196 Automation
where it does not matter the form of the blocks *:
C(sI

A)
1

B =
C
c
C
uc
(sI
d
A
c
)
1
B
c
0
C
c
C
uc
(sI
d
A
c
)
1
B
c
0
= C
c
(sI
d
A
c
)
1
B
c
11. Solution
(1)
A =
0 0 1
0 1 1
1 0 1
; b =
1
1
1
(1.1)
rank
b Ab A
2
b
= rank
1 1 2
1 0 2
1 2 3
= 3
(1.2)
S
1
=
2 1 1
2 0 1
3 2 1
; S
1
1
=
1
3
2 1 1
5 1 4
4 1 2
A =
1
3
6 3 0
0 0 3
3 0 0

b =
1
3
2 1 1
5 1 4
4 1 2
1
1
1
=
1
3
0
0
3
(1.3)
S
2
=
1 0 0
2 1 0
0 2 1
; S
1
2
=
1 0 0
2 1 0
4 2 1
A
c
=
0 1 0
0 0 1
1 0 2
; b
c
=
1 0 0
2 1 0
4 2 1
0
0
1
0
0
1
(2) The computations are analogous to those in (1).

(3.1) If
i
is an eigenvalue of A
c
, then:
rank(A
c
i
I) < n.
On the other hand,
rank(A
c
i
I) = rank
i
1 0 . . . 0 0
0
i
1 . . . 0 0
. . . . . .
0 0 0 . . .
i
1
. . .
n 1.
(3.2) If
i
is an eigenvalue of A, then
i
is also en eigenvalue of A
c
. Then, from (3.1):
dimKer(A
i
I) = n rank(A
i
I) = n rank(A
c
i
I) = n (n 1) = 1.
Therefore, A is non-diagonalizable if the algebraic multiplicity of
i
is greater than 1.
4. References
Chen, Chi-Tsong (1984). Linear System Theory and Design, Holt-Saunders International
Editions, Japan. ISBN: 4-8337-0191-X.
Ferrer, J.; Ortiz-Caraballo, C. & Pea, M. (2010). Learning engineering to teach mathematics,
40th ASEE/IEEE Frontiers in Education Conference
Kalman, R. E.; Falb, P. L. & Arbib, M. A. (1974). Topics in mathematical system theory, PTMH,
India, McGraw-Hill
Lay, D.C (2007). Algebra lineal y sus aplicaciones, Ed. Pearson, Mexico. ISBN: 978-970-26-0906-3
Puerta Sales, F. (1976). Algebra Lineal, Universidad Politecnica de Barcelona. ETS Ingenieros
Industriales de Barcelona in collaboration with Marcombo Boixareu Ed. ISBN:
84-600-6834-X
198 Automation

10
Rapid Start-Up of the Steam Boiler,
Considering the Allowable Rate
of Temperature Changes
Jan Taler and Piotr Harchut
Department of Power Installations, Cracow University of Technology
Poland
1. Introduction
The development of wind turbine engineering, which can be characterized by large
inconsistencies in the amount of delivered electricity over time, generated new problems
regarding the regulation of power engineering systems. At high and low wind speeds, the
energy production falls rapidly, meaning that the conventional heat or steam-gas blocks
have to be activated very quickly. Steam boilers in blocks have to be designed in a way that
allows for a start-up of the block within a few dozen minutes. The main elements limiting
the quick activation of steam boilers are thick-walled structural elements, in which heat
induced stresses can occur during the start-up cycle.
The start-up and the shut-down processes shall be conducted in a way that ensures that the
stresses at the places of stress concentration do not exceed the allowable values.
To be able to ensure the appropriate durability and safety of the blocks adapted for a rapid
start up, a precise analysis of the flow-heat processes, together with the strength analysis, is
required. Such analysis is covered in this paper.
German boiler regulations, TRD 301 and the European Standard, EN-12952-3 allow
determining two allowable rates for heating the pressure element:
-
1 T
v the pressure p
1
at the beginning of the start-up process,
-
2 T
v at the pressure p
2
at the end of the start-up process.
Heating rates for intermediate pressures shall be determined using the method of linear
interpolation (Fig. 1).
To be able to heat the boiler evaporator at the maximum allowable rate, heat flow
rate
k
Q
shall be delivered to the evaporator in the boiler furnace chamber (Fig. 2). The value
of
k
Q
is a function of many parameters, but for the boiler of a specific construction, it

depends mainly on the rate of changes of the saturation pressure /
n
dp dt and the mass of the
saturated steam mass flow rate.

Automation

200

Fig. 1. Heating rate of the pressure boiler element as a pressure function, determined in
accordance with TRD-301 regulations.
2. Pressure and temperature changes in the evaporator during the boiler heating
process
In accordance with TRD 301 regulations (Fig. 1), the changes of the heating rate
T n
v dT dt = in
the pressure function are interpolated using a straight line.

2 1 1 2 2 1
2 1 2 1
( )
n T T T T
n n
dT p p
p T
dt p p p p
v v v v
= +

(1)
The water saturation pressure ( )
n n
p T from Eq. (1) can be expressed in the following way in
the temperature function:
exp( 19, 710662 4, 2367548 )
n n
p T = + (2)
Where pressure p
n
is expressed in bars and temperature T
n
is expressed in
o
C.
Eq. (1) has been integrated using the Runge-Kutty method of the 4-th rank at the following
initial condition:

0 1
( )
n t n
T T p
=
= (3)
After determining, by solving the initial condition (1-3), the flow of the saturation
temperature in a function of time, that is T
n
(t), the pressure changes p
n
(t) can be determined,
e.g., from function (2). The pressure changes rate /
n
dp dt and the saturation temperature
changes rate /
n
dT dt in a function of time shall be determined on the basis of the assumed
run v
T
(p
n
) (Fig. 1).
In this paper, the dynamics of the boiler evaporator of the OP-210M boiler will be analysed.
OP-210M is a fine coal fuelled boiler with natural circulation. The T
n
(t) and p
n
(t) flows,
determined for p
1
=0 bar, v
T1
=2 K/min and p
2
=108,7 bar, v
T2
=5 K/min are presented in Fig. 2.

Rapid Start-Up of the Steam Boiler, Considering the Allowable Rate of Temperature Changes

201

Fig. 2. Temperature and saturation pressure changes in the boiler evaporator during the
heating process.
After determining the T
n
(t) and dT
n
/dt time run, it is possible to determine the rate of
pressure changes dp
n
/dt using the following formula:

n n n
n
dp dp dT
dt dT dt
= (4)
Where functions dp
n
/dT
n
for saturated steam can be determined by the integration of
formula (2).

Fig. 3. Rates of the pressure dp
n
/dt and temperature dT
n
/dt changes, during the evaporator
heating process.

Automation

202
The pressure dp
n
/dT
n
changes rate can also be calculated using an approximate formula:

( ) ( )
2
n n n
p t t p t t dp
dt t
+ A A
=
A
(5)
Where: At is the time step.
For the calculations, it can be assumed, for example, that At = 1 s. Temperature dT
n
/dt and
pressure dp
n
/dt changing rates are presented in Fig. 3.
3. Mass and energy balance for the boiler evaporator
Modelling of the transient state effects occurring in the boiler evaporator is usually done
while assuming that it is an object having a concentrated mass and heat capacity
4
. The start
point for the determination of the heat flow rate
k
Q
, which assures that the evaporator is

heated at the desired rate vT(t), are the mass and energy balance equations for the
evaporator (Fig. 4):

Fig. 4. Diagram of the boiler evaporator.

( ) ' ' " "
w p od
d V V
m m m
d

t
+
= (6)

( ) ' ' " "
w p od
d V V
m m m
d

t
+
= (7)
After transforming formulas (6) and (7) we obtain:


203

'( " ') " ' ' " "
( " ') ( )
' " ' " ' "
"( " ') ( " ') ' "
' '
[ '( ' 1) "( " 1) ]
' " ' "
h h h h
Q m h h m h m
k od w w p
dT dp
h h h h d d
dh dh n n
V V m c
dp dp dp dp m m dp dt
n

= + +

+ + + +

(8)
Formula (8) allows determining the heat flow rate
k
Q
, which shall be delivered to the

evaporator from the furnace chamber in order to ensure the assumed rate of the pressure
change dp
n
/dt . Changes of the d/dp and d/dp functions and the dh/dp and dh/dp
functions, appearing in formula (8) are presented in Fig. 5 and Fig. 6.

Fig. 5. Changes of the d/dp and d/dp derivatives as a function of pressure.

Fig. 6. Changes of the dh/dp i dh/dp derivatives as a function of pressure.

Automation

204
The precision in determining those functions is very important. The best way is to determine
those functions through analytical differentiation of the (p) and (p) functions, obtained
earlier using the least square methods on the basis of the pair tables.
4. Heat flow rate absorbed in combustion chamber
From the conducted calculations, we can see, that the heat flow rate
k
Q
delivered to the
evaporator can be lowered significantly by reducing the mass of the steam mass flow rate
p
m delivered from the boiler drum. The minimum steam mass flow rate
p
m , which shall be
produced in the boiler during the start-up process can be determined from the condition,
stating that the maximum allowable temperatures for the particular stages of the heater shall
not be exceeded. The furnace oil mass flow rate
pal
m , delivered to the burners during the
boiler start up process, which is necessary to assure the heat flow rate
k
Q
shall be
determined on the basis of the calculations of the boiler furnace chamber.
The heat flow rate
, k kom
Q
absorbed by the walls of the boiler furnace chamber, expressed in

W, can be calculated from the following formula:

, , ,
0 0
T T
k kom pal d pow p pow pow sp p sp sp
pow sp
Q m W m c T m c T
''
'' = +
(9)

pal
m
fuel mass flow rate, kg/s,
W
d
fuel calorific value, J/kg,
pow
m
air mass flow rate, kg/s,
,
0
T
p pow
pow
c
average specific heat of air, at constant pressure and at temperature from
0 to T
pow
[
o
C], J/(kgK)
T
pow
temperature of air delivered to the furnace chamber,
o
C
sp
m
flue gas mass flow rate, kg/s
,
0
T
p sp
sp
c
''

average specific heat of flue gas, at constant pressure and at temperature
from 0 to
sp
T
''
[
o
C], J/(kgK)
sp
T'' flue gas temperature at the exit of the furnace chamber,
o
C
The temperature of flue gases at the exit of the furnace chamber,
sp
T
''
is calculated from the
equation:

0,6
273, 15
273, 15
1
ad
sp
p
T
T
a
M
Bo
+
'' =
| |
+
|
\ .
(10)
Where Bo is the Boltzmann number, determined by the formula:

,
3
sp p sp
k ad
m c
Bo
A T o
=

(11)


205
, p sp
c is the average specific heat of flue gases in J/(kgK), at temperature ranging from
sp
T
''
to
T
ad

273,15
, ,
273,15
T
p sp p sp
T
ad
sp
c c

''
= (12)

a
p
emissivity of the furnace chamber,
M parameter characterizing the area, in which the flame temperature in the
chamber is highest,
A
k
area of the walls of the boiler furnace chamber, m
2

T
ad
theoretical (adiabatic) combustion temperature,
o
C
o
Stefan-Boltzmann constant, o = 5,6710
-8
W/(m
2
K
4
)

thermal efficiency coefficient of the screens, being the quotient of the heat flux
absorbed by the screen and the incident flux.
The calculation of the heat flow rate
, k kom
Q
absorbed by the walls of the furnace chamber

of the OP-210M boiler has been calculated using (9). In the function of the fuel mass flow
rate
pal
m and the air excess coefficient , the calorific value of the oil was assumed to be:
W
d
= 41060 kJ/kg.
The results of the calculations are presented in Fig. 7.

Fig. 7. Heat flow rate
, k kom
Q
, in kW, delivered to the evaporator of the OP-210M boiler, as a

function of the fuel mass flow rate
pal
m in kg/s and the air excess coefficient .

Automation

206
It was proven that the increase of the air excess coefficient at the assumed fuel mass flow
rate
pal
m leads to the reduction of the heat flux delivered from flue gases in the furnace
chamber to the evaporator. This is due to the fact that the temperature of the flue gases in
the furnace chamber T
pl
was lowered significantly, which in turn led to a significant
decrease of the heat flow rate
, k kom
Q
, as the heat flux is proportional to temperature

difference between the flame temperature and the temperature of the chamber walls raised
to the fourth power, that is
( )
4 4
pl sc
T T when increases
, k kom
Q
is reduced, and the heat flux

absorbed by the steam over heaters increases. Increasing the fuel mass flow rate
pal
m at the
assumed air excess coefficient leads to the increase of the heat flow rate
, k kom
Q
absorbed by
the walls of the furnace chamber.
Next, the fuel mass flow rate
pal
m delivered to the boiler during the start-up (heating)
process was determined. After determining from (8) the heat flow rate ( )
k
Q t
, the non-linear
algebraic equation is solved.
( )
,
( ) ,
k k kom pal
Q t Q m =

(13)
p
m
pal
m The fuel mass flow rate, at the assumed value of the air excess coefficient , is
determined from (13). Changes of the mass flow rate in a time function, during the start-up
process of the OP-210M boiler, determined from (13), for = 1,1 is presented in Fig. 9a and
9b. The calculations were conducted for different steam mass flows produced in the boiler
evaporator. The mass of the fuel oil used for the boiler start up process from time t=0 to t=t
k

can be calculated using the formula:

0
t
pal pal
k
m m dt =
}

(14)
The integral (14) was calculated numerically, using the trapezium rule for integration.
5. Calculation results
The calculation has been completed for the evaporator of the OP-210M boiler, for the
pressure change rate dp
n
/dt presented in Fig. 3. The temperature of the incoming water T
w

during the evaporator heating process is at a given moment lower by 10K than the
saturation temperature Tn. The temperature run of the incoming water T
w
, the saturation
temperature T
n
and the pressure in the evaporator p
n
are presented in Fig. 8.
The calculations have been conducted for the following data:
0, 51
ods
m = kg/s, 17, 1
w
m = kg/s, 16, 57
p
m = kg/s, m
m
=171900 kg, c
m
= 511 J/(kgK),
V = 43,6 m
3
, V = 15,9 m
3
.
The results of the
k
Q
calculations, determined for various saturated steam mass flow rates,

are presented in Fig. 9a and Fig. 9b.


207

Fig. 8. Temperature run of the incoming water temperature T
w
, saturation temperature T
n

and pressure in the evaporator p
n
during the boiler start up cycle.
After analysing the results shown in Fig. 9a we can see that if the mass capacity of the
evaporator
p
m equals zero, then the flow rate of the heat flow rate
k
Q
delivered to the
evaporator shall increase. In this case, all heat delivered from the furnace chamber is used
mainly for heating water contained in the boiler evaporator. This is due to the increase of the
specific heat of water c
w
, which increases in line with the pressure increase. For example, at
pressure p
n
=1 MPa water specific heat equals: c
w
= 4405 J/(kgK), and at the pressure p
n
=10
MPa water specific heat equals: c
w
= 6127 J/(kgK). Moreover, at the end of the start-up
process a large increase of the rate of temperature changes in the evaporator was noticed
(Fig. 3), which required increased heat flow rate
k
Q
to be delivered to the evaporator.

In larger mass capacities
p
m the heat flow rate
k
Q
, which shall be delivered from the

furnace chamber to the evaporator, increases significantly, because there is a large power
demand on water evaporation (Fig. 9). For the evaporator efficiencies greater than 90 t/h
(Fig. 9b), however, it has been noted that for the assumed flux
p
m , the reduction of the heat
flow rate
k
Q
occurred during the end phase of the heating process, despite the increase of
the rate of evaporator heating (Fig. 3). This is due to the fact that the water evaporation heat,
r = h h decreases in line with the pressure increase in the evaporator. For example, at the
pressure p
n
=1 MPa, water evaporation heat equals: r = 2014,4 J/(kgK), and at the pressure
p
n
=10 MPa, the water evaporation heat is reduced to: r = 1317,7 J/(kgK).

Automation

208

a)

b)
Fig. 9. Heat flow rate
k
Q
delivered to the evaporator of the boiler in the furnace chamber for

various steam fluxes
p
m .
Fuel mass flow rate
pal
m during the start-up of the OP-210M boiler obtained from Eq. (13)
for the excess air number =1,1 are shown in Figs. 10a and 10b.


209

a)

b)
Fig. 10. Changes of the fuel mass flow rate
pal
m in kg/s, during the start-up of the OP-210M
boiler, depending on the steam mass flow rate
p
m in t/h, produced in the boiler evaporator.
The calculations are carried out for various steam flow rates
p
m generated in the evaporator.

Automation

210

a)

b)
Fig. 11. Fuel consumption m
pal
, in kg, during the start-up process of the OP-210M boiler,
depending on the steam mass flow rate
p
m , in t/h, produced in the boiler evaporator.
The results of the calculations of m
pal
corresponding to the runs of the fuel mass flow rates
pal
m from Fig. 10 are presented in Fig. 11.


211
From the analysis of the results shown in Fig. 11 we can see that the fuel consumption
depends mainly on the steam mass flow rate
p
m removed from the boiler drum. During the
start-up process, lasting 8000s, the fuel consumption varies from m
pal
= 3308 kg, at the steam
mass flow rate
p
m = 10 t/h, to m
pal
= 27207 kg, at the steam mass flow
p
m = 120 t/h. Due to
the high price of the fuel oil, the boiler start up process shall be conducted at the minimum
steam mass flow rate
p
m , ensuring the appropriate cooling of the pipes of the over heaters.
Strength analyses show that temperature can change at the beginning of warm-up by leaps
and bounds, without exceeding the allowable stress.

Fig. 12. The allowable process of the temperature due to the thermal stresses on the inner
surface of the boiler drum.
The allowable temperature of the process due to the thermal stresses on the inner surface of
the drum can be approximated by using:
( )
2 n
c
T t a bt
t
= + + (15)
where the constants a, b and c are determined from the condition:
( ) ,
in i a
r t
o o ~ ( ) 1
i
t i t = A 1, 2, 3,..., i n = (16)
Condition (16) means that for optimal temperature of the process defined by formula (15)
constants a, b and c should be chosen so that the thermal stress on the inner circumferential
surface of the drum was, for n time points, equal to the allowable stress. In the case of the
drum analysed in this paper, for allowable stress 162, 7
a
o = MPa, the following constants
were obtained: a=79,95
o
C, b=0,1699
o
C/s, c=153,86 s. Starting with an optimal start-up boiler
water temperature change in the drum is very difficult, because the optimal temperature is
very high at the beginning of start-up and then very quickly reduced to the minimum value.

Automation

212
Therefore, during start-up an approximation of optimal temperature change will be
implemented by the step up temperature increase at the beginning and then heating at a
constant speed. When the pressure in the drum is higher than the atmospheric pressure, after
the jump of temperature, the speed of temperature change is equal v
T1
(at the beginning) and
v
T2
(at the end of start-up). Determination of the heating or cooling rate of the drum for any
pressure p is the same as in TRD 301 regulations or European Standard BS EN 12952-3.
Fig. 13a and fig. 13b show the changes in temperature and pressure of water in the drum at
different initial temperature jumps.

a)

b)
Fig. 13. Temperature and pressure of the water in the drum during boiler start-up.


213

Fig. 14. Fuel consumption and boiler start-up time at different initial jumping changes in
water temperature in the evaporator boiler.
An analysis of the results in Fig. 14 shows that the release of the initial jump in the
temperature of the boiler start-up reduces the time and reduces fuel consumption.
6. Conclusions
The model of the dynamics of the boiler evaporator, presented in this paper, can be used for
the determination of the heat flux, flowing from flue gases in the furnace chamber to the
boiler evaporator, necessary to warm up the boiler at the assumed rate of the medium
temperature change in the boiler evaporator. The fuel mass flow rate
pal
m , for which the
heat flux delivered to the evaporator equals
k
Q
, is determined on the basis of the

calculations of the boiler furnace chamber. Fuel consumption during the boiler start up
process, necessary for heating the boiler evaporator to the assumed temperature, from the
assumed nominal start temperature, assuming that the heating of the boiler drum is
conducted at the maximum allowable speed, due to the limitations caused by heat stresses
at the boiler drum/down comers intersection, has been determined. In order to limit the
consumption of the expensive fuel oil during the boiler start up process, the process shall be
conducted at the minimum steam mass flow rate, ensuring the appropriate cooling of the
over heaters and at the minimum allowable air coefficient. The dependencies derived in this
paper can be used for the preparation of the optimal technology of the boiler start up
process.

Automation

214
It is shown that allowing the rapid increase of water temperature at the start of
commissioning of the boiler reduces the boiler start-up time and helps to reduce fuel oil
consumption.
7. References
TRD 301 (2001), Zylinderschalen unter innerem berdruck. Technische Regeln fr Dampfkessel
(TRD), pp. 143-185, Heymanns Beuth Kln Berlin, Germany.
Staff Report, (2007). Dealing with cycling: TRD 301 and the Euro Norm compared, Modern
Power Systems, Vol 27, No 5, p 33-38.
European Standard, EN 12952-3, Water-tube boilers and auxiliary installations Part 3: Design
and calculation for pressure parts , CEN European Committee for Standardization,
rue de Stassart 36, B-1050 Brussels, 25. July 2001.
Profos, P., (1962). Die Regelung von Dampfanlagen, Springer, Berlin, Germany.
Styrykowicz, M. A., Katkowskaja, K. J., Sierow, E. P. (1959). Kotielnyje agregaty.
Gosenergoizdat, Moscow, Russia.
Sierow, E.P., Korolkow, B. P. (1981). Dinamika parogienieratorow. Energija, Moscow, Russia.
Taler, J., Dzierwa, P. (2010): A new method for optimum heating of steam boiler pressure
components. International Journal of Energy Research, Volume 35, Issue 10, pages
897908, Wiley, August 2011.
11
A Computer-Aided Control and
Diagnosis of Distributed Drive Systems
Jerzy wider and Mariusz Hetmaczyk
The Silesian University of Technology,
Poland
1. Introduction
Modern systems of the industrial processes automation most often are composed of several
basic components, among others: controllers (PLCs, regulators, industrial computers, etc.),
actuators (electric, pneumatic, hydraulic and their combinations), control and measuring
equipment (sensors and detectors), industrial networks and SCADA/HMI systems (wider
& Hetmaczyk, 2009c). Functional subassemblies have been divided into several groups
(mechanical elements, electrical and electronic parts, control equipment and software
packets). All of the enumerated components are susceptible to failures, therefore, the main
problem is an estimation of the trouble-free service period. In the case of individual
components the described task is easy, but an analysis of distributed applications defined as
a collection of dependent subsystems the reliability function is dependent on all reliability
factors assigned to all components (Dhillon, 2002).
Sequential distributed systems include most often: automated machining centres, assembly
lines, conveyors etc. Specified examples (for safety reasons) involve extended monitoring
procedures strictly connected with requirements of the employed maintenance model.
The preventive maintenance model includes several meaningful assumptions, among other
things: risk reduction, failure avoidance, the continuous improving of the reliability of the
considered system, defects elimination (with attention focused on minimization of the time
needed for the detection of error sources and their elimination), maximization of
a productivity combined with decreasing of the total production costs, with continuous
improvement of the quality level. An elaboration of the method that fulfils all the specified
requirements with enumerated factors will be the most important task for engineers derived
from different domains. A utilization of more advanced devices allows for a combination of
diagnostic and control functions at various levels. Furthermore, in the most of contemporary
applications, an integration was omitted for many reasons (economic premises, time and
work efforts, the complexity of the algorithm etc.).
2. Problems connected with the maintenance of distributed systems
The maintenance of distributed systems is a very important task in many engineering
applications (Ryabinin, 1976). Industrial control systems are described by several qualities:

Automation

216
control system organization (centralized or distributed), application size, type of measuring
equipment (sensors and detectors), technical orientation, etc.
Industrial productive systems run at a risk of impact by many external factors. Among
meaningful impacts occur two main groups in the form of: command signals (purposefully
dispatched by users) and disturbances. Progress at a complexity of manufacturing
departments, productive systems, as well as machines, causes a continuous increase in the
number of failure sources (De Silva, 2005). Nowadays the diagnosis of automated industrial
productive systems is a significant issue.
Design, implementation and operation of machines without taking into consideration
aspects related to the monitoring of operating conditions and planning of the duty are not
possible in applications taking advantage of mechatronics devices.
The described problem is especially apparent in the case of the monitoring and control of
the states of electric drives equipped with AC asynchronous motors that are subjected to
extensive investigation in many research institutes. However, tests on a single drive treated
as an autonomous machine constitute a predominating tendency nowadays.
Such an approach does not meet the needs of operational and maintenance engineers, as
industrial conditions compel the requirements of complying with several rules, including:
fast detection and removal of failures, minimization of the number of configuration
procedures run on the diagnostic equipment (a universality securing the operation of
numerous communication protocols and methods of access to the diagnosed object) as well
as a wide range of available tools (Bloch & Geitner, 2005).
Up to this moment, devising a concept of diagnosing distributed drive systems operated by
frequency converters coupled with the ProfiBus industrial network (Mackay & Wright,
2004) and a centralized PLC control has not been developed.
2.1 A structural analysis of the electric drive unit
Electric motors dominate in industrial drive systems of modern machines. The reasons for
such wide spread popularity are their functional features such as: high efficiency of the
individual drive unit, wide ranges of torque and speed adjustments, small dimensions, ease
of a power supply, good prices regarding functional features.
The availability of a wide selection of the components belonging to the group of the
automatic control equipment has made it possible to construct new technical devices that
have not previously corresponded to the traditional definition of a machine (Fig. 1a). The
determination of the state of the machine on the grounds of control, energy, disturbances
and mass fluxes does not consider many factors to ensure a proper course of work of
distributed units. The problems of control and identification of failures of units that have
a different degree of technological advancement are still important tasks to be addressed.
Automation involves the necessity of providing man-free operation of the duty, for which
a two-directional information flow is required (enabling both control of the parameters and
change in the states).
The specification of the components of the machine block (Fig. 1b) enables a complex
identification of the internal information fluxes (signals exchanged at the level of individual

A Computer-Aided Control and Diagnosis of Distributed Drive Systems

217
control of drives by means of frequency inverters) and external ones (a composite flux,
corresponding to the operational state of the machine as a whole).
Each drive subsystem has features that predispose it to a group of autonomous machines. In
each subsystem were extracted an electric drive, extended by a reduction gear and a control
subsystem (in the form of a frequency inverter or an I/O module). Detailed relationships
between functional subsystems of the drive system are shown in Figure 1c.

Fig. 1. Definitions of the machine: a) a traditional approach (Korbicz et al., 2002), b) the
approach adjusted to the requirements of the analysed systems, c) simplified view of the
drive subsystem in the form of coupled components collection
The proposed structure refers to, among others, conveyors, production lines treated as a one
cell, in which the proper course of the work cycle depends on all the components.
Distributed systems with the appropriately formed information flux may be subjected to an
analysis in view of the system parameters at different levels (from the values recorded by
sensors, through the state of the control elements, to production data).

Automation

218
2.2 An overview of the most often used approaches for the monitoring of distributed
systems
The supervision (Boyer, 2004) over industrial processes and the change of the parameters of
the states of distributed systems are most frequently performed by three means (Fig. 2).
Portable analysers (Fig. 2a) make it possible to make measurements of the operational
parameters of machines at a work place, yet, in view of the nature of their construction
(specialized units enabling the determination of parameters, mostly for the one type of
phenomenon), the measurements are conducted only at planned time intervals or in the case
of visible symptoms of damage. An additional disadvantage of such an approach is the need
to provide the service using experienced diagnostics staff.

Fig. 2. Methods of controlling the states of technical systems: a) analysis of the machine
under operation made by the analyser, b) use of the input/output card, c) implementation
of SCADA system and data collector, d) a concept of the customized computer environment
Computer input/output cards (Fig. 2b) are used in systems with a small number of
measuring or executive devices. SCADA systems (Fig. 2c) do not facilitate access to special
functions of the supervised equipment and are, at the same time, relatively expensive
(Boyer, 2004). Furthermore, their satisfactory recording resolution requires the configuration
of an optional collector. The remaining two means have many limitations and are most
frequently used as elements supporting exact diagnostic activities.
The last of the concepts (Fig. 3d) is based on the application of customized software, which
combines benefits of presented supervisory and control systems, expanding simultaneously
opportunities for additional features such as: dedicated diagnostic and control blocks,
utilization of the ProfiBus main line as a carrier of the diagnostic information, advanced
features of the ProfiBus protocol. The main advantage of the proposed concept is the ability
of the implementation in any process (or machine) operated by the PLC controller.


219
3. Integration of diagnostic and control functions
The developed method is characterized by some degree of universality (all applications of
control and diagnostic activities of electric drives supported by PLC controllers), but at the
design stage the authors have adopted significant limitations related to the organization of
hardware and software structures (Kwaniewski, 1999).
Usage of programmable logic controllers and distributed drives equipped in industrial
network interfaces is a common case in industrial applications. PLCs popularity results
from a fact that exchanged data most often have a status of control commands (Legierski,
1998).
In the majority of cases diagnostic and control functions are separated. The traditional
structure of an organization of control algorithms is oriented to the failure-free execution of
the course of work, but the diagnostic information is most often omitted. The main cause of
this situation arises from a substantial complexity of the control algorithm extended by
diagnostic functions. Integrated diagnostic and control functions compound specified
functional qualities, which combine presented domains.
A set of diagnostic and control functions creates a program for realizing self-diagnostic
functions of the controlled system. A fault detection function should be prepared for every
unit (wider & Hetmaczyk, 2010). A superior level function is performed by the computer
system, which consists of dedicated screens that allow interpretation of current and
historical data. To primary parameters belonged, among other things current, minimum and
maximum bus cycles, diagnostic information, a current work status, error codes, reactive
and active powers, current status of I/Os, etc.
3.1 The model of the information flow
The proposed solution is intended for the real time analysis in the case of industrial
applications that use automatically controlled equipment and are mainly operated by means
of customized controllers with properly selected modules and communication standards.
A preventive maintenance of programmable logic controllers contains several simple
actions. On the basis of this statement the PLC unit could be assigned to the group of the
most reliable elements of automated productive systems (Prasad, 2004).
The reliability of constituent elements and a whole system depends on a fast assessment of
operating conditions and prediction of failures on the grounds of the current state.
So that execution of detailed results could be undertaken, research objects were restricted to
Mitsubishi PLC controllers. A functional extension can be obtained on most PLCs available
in engineering applications.
In the diagram of a distributed system shown in Fig. 3a, the system has been divided into
particular functional subsystems. The PLC holds a superior function as it enables the
execution of the control parameters for all drives (wider & Hetmaczyk, 2009d).
The basic assumption of the decentralized system, in the form of distributed actuators and
centrally located PLC (or separated control systems connected by the communication main
line, facilitating access to the data in the entire system) leads to the conclusion that the

Automation

220
system components are connected hierarchically. An elaborated method in the form of
Computer Control and Diagnostic System of Distributed Drives (CCaDSoDD) enables the
control of the states of the PLCs, actuators and industrial sensors without interferences into
the operation of the distributed system, and a simultaneous running of other software
connected with the actuators or control units (wider & Hetmaczyk, 2009a). The structure
of the described system, based on connection relationships, conforms to an automation
hierarchy (Fig. 3):
electric drives and frequency inverters are placed in the lower level (a primary
technology level),
a Programmable Logic Controller performs the role of a supervised unit (within the
algorithm defined in the PLC memory) and is placed at the group control level (wider
& Hetmaczyk, 2009b),
the elaborated expert system is located on the highest level (the PLC and the ProfiBus
DP interface mediate at the stage of data exchange between frequency inverters and the
expert system),
work of the CCaDSoDD does not interfere with an operation of the PLC controller, but it
is possible to enforce certain conditions of drives (performance of active diagnostic tests).

Fig. 3. Schematic diagrams of: a) the considered system, b) a data flow
Despite a precise division, a problem with an inspection of distributed mechatronic systems
still exists. The maintenance of big systems requires utilization of advanced tools.


221
The proposed approach is based on making direct or indirect connections by means of
a coupled network. The topology of the connection depends on the type of the
communication network.
The choice of Mitsubishi controllers has been made in view of the ease of connection with
selected logical units at each level of the defined hierarchy of the system (Fig. 4).

Fig. 4. A schematic diagram of dependences of the considered system taking into account
the automation hierarchy
The acquisition of data from remote units is possible by means of the access via gate
function (for example, the HMI panel - Fig. 5a).

Fig. 5. Manners of access to the discussed distributed system: a) 1:1 connection (direct or by
means of the operators panel), b) and c) by means of 1:n connection
The software provided by producers must be connected directly (Fig. 6) to the drive section
(frequency converter) by means of a network coupling the entire manufacturing company
(the Industrial Ethernet is very popular). It is also possible to use the communication socket
and the USB port of a PC, however this is still 1:1 type connection. Moreover, the contents of
the data file are limited to control signals of the electric motor.

Automation

222
The classification of industrial networks in view of their possible applications and the
configuration of the system in accordance with the assumed hierarchy of a control enables
access from any level (Fig. 6). Such a solution offers many advantages, as it facilitates the
configuration of the connection in a coupled or in a direct mode using any unit extended by
a selected network module.

Fig. 6. The pyramid of access methods to the resources of control units
Knowing the address or the relations of the connections among the system elements, it is
possible to utilize a direct access to the resource (a coupled network), both for the ascending
graph trend (from lower to higher level networks) and for the descending one.
3.2 Integrated control and diagnostic blocks
In the discussed case, the communication and control of the operational parameters of each
drive are performed by means of a control algorithm, saved in a PLC memory (the sequence
of operating conditions). Correct course of work results from a faultless control algorithm
definition. Elimination of errors leads to acquiring a faultless control algorithm, but is rather
time consuming. A better approach is elimination of roots of problems and that method has
been used in elaboration of integrated control and diagnostic blocks. An essential drawback
of industrial networks is difficulty in direct access to transmitted data.
Integrated control and diagnostic blocks have been divided, at an angle of possibilities of
applications, into several groups (Fig. 7):
primary control blocks (for: the Movimot frequency inverter, the Movifit frequency
inverter, distributed I/O modules),
universal diagnostic blocks (of: the SLAVE unit, an interpretation of diagnostic
information, the interpretation of extended diagnostic information, the interpretation of
alarm information, a distribution of control data).
Basic blocks are necessary for correct system operation and are required (in case of the
elementary configuration) to proper realization of the control process, however,
maintenance of diagnostic functions is unfeasible. Achievement of the complete
functionality requires utilization of optional blocks. Error elimination, based on internal
algorithms, permits a programmer to abandon of the PLC memory allocation and algorithm
testing stages (wider & Hetmaczyk, 2010, 2009c).


223

Fig. 7. A schematic representation of relationships between types of blocks, degrees of
universality and the signal flow dependency
Moreover the algorithm structure for each SLAVE unit is unequivocally defined, without
possibilities of mistakes. Each presented block has different tasks in the control algorithm.
An elaboration of integrated control and diagnostic blocks has been based on characteristic
features of Programmable Logic Controllers. Principal assumptions of the presented method
can be characterized in several points:
each of the SLAVE unit is identified by an unique ordinal number (dependent on the
ProfiBus network address),
for proper system operation, the configuration of minimum three primary control
blocks is necessary,
additional functions (extended diagnostic functions, alarm control, etc.) demand an
extension at the level of diagnostic blocks,
exists the possibility of programming without the knowledge about a network
configuration and a system structure,
presented universal blocks have been characterized by general-purpose internal
structure (implementation to control functions of all SLAVE devices with ProfiBus
interfaces),
control blocks have been elaborated at an angle of individual units.
The presented scheme of the signal flow and blocks arrangement is universal for all units
(not only for the presented devices). Accommodation for an arbitrary configuration requires
an implementation of the control block dedicated to the concrete unit.
Up to now, the method of traditional programming with algorithms acquiring the faultless
data exchange between PLCs networks modules and SLAVE units requires:
a continuous inspection of addresses defined in options of a network module and
addresses assigned by a programmer,
multiple testing connected with the task of a proper memory allocation,
a definition of the amount of input and output words of each SLAVE unit (most often
on the basis of manuals),
realization of time consuming additional operations (in the case of requirements
concerning diagnostic information).

Automation

224

Fig. 8. A scheme of the signal flow and the blocks configuration for the single Slave unit
The proposed solution is based on the memory stack (Fig. 9) which is the primary principle
of operation of all devices that convert digital data. The main problem arises from
difficulties with an identification of registers which store corresponding data.
The first step is checking of statistical data of a concerned system, at an angle of number of
devices, number of input (control) words and output words (diagnostic words). On the
grounds of those information a programmer is able to identify the type of the SLAVE unit,
without a knowledge of the functional structure of a distributed system. All described
functions are performed by the universal block (Fig. 8).
The main difference between a traditional programming method and the proposed solution
might be defined as a direct data exchange between the PLC processor and a special
network module.
An internal algorithm of the universal block is presented in the Figure 11. The described
program block belongs to the group of main elements of the control algorithm. Basic
features of this block are presented in Figure 8, but the most important task is
the dispatching of reply words from the SLAVE unit. A contrary task characterizes
a distribution block of the control data. The block located at the centre (Fig. 8) is dependent on
two described blocks and must be matched to the individual features of each SLAVE unit.


225

Fig. 9. A graphical interpretation of a PLC memory (the memory stack)
An internal algorithm formulates contents of control words dispatched to a block of
distribution of the control data. Due to required organization of each control and diagnostic
blocks, the set must be arranged in accordance with the presented scheme.

Fig. 10. A block diagram of the block of a distribution of the control data

Automation

226

Fig. 11. Block diagrams of: a) the main universal block, b) the data interpretation block of the
diagnostic information


227
The principle of an operation of other blocks is similar to the data interpretation block of
diagnostic information and its description has been neglected.
4. Embedded functions of identification and interpretation of errors
4.1 The binary states vector method a manual definition of failure states
A common malfunction of systems with implemented industrial networks is a loss of the
communication. In consequence, often parameter reading of the inverters (or I/O modules)
becomes impossible, resulting mostly from improper configuration, electromagnetic
interferences, interruptions or damages in bus structures, etc.
The CCaDSoDD system allows a definition of the binary diagnostic vector (wider &
Hetmaczyk, 2011). The vector is defined on the basis of damage or failure symptoms
indicated by external elements of control devices (status diodes placed on front plates).
In the case of integrated frequency inverter of the MOVIFIT unit, contents of the states
vector are defined according to the following formula:
W
i
=[ S
t
T
r
K
zi
K
zl
K
c
P
r
S
rp
P
dk
P
zp
K
r
2x K
r
3x K
r
4x K
r
5x K
r
6x ] (1)
where:
i the unique ordinal number of the diagnosed unit,
S
t
the current state of a status diode (0 - passive, 1 active),
T
r
a work mode (0 a continuous signal, 1 the pulse signal),
K
zi
, K
zl
, K
c
colours (accordingly: green, yellow or red; marker equals logical true
corresponds to the current colour),
P
r
, S
rp
, P
dk
, P
zp
accordingly: a uniform pulsation, the fast uniform pulsation, the pulsation
in two colours, the pulsation with a pause,
K
r
times of pulsation (3, 4, 5, 6-times)
Each failure has been assigned to the unequivocally defined diagnostic vector, which is the
pattern used for comparison of functions that define errors.
In the case of determination of bus work parameters, the described vector has been divided
into components, with the aim of easier identification of damaged states:
W
i
=[ [W
RUN
] [W
BUS
] [W
SYS
] ] (2)
where:
W
RUN
a six-elements vector of the hardware components diagnosis,
W
BUS
the three-elements vector of the communication status diagnosis,
W
SYS
the five-elements vector of the configuration diagnosis, a data exchange management
program, etc.
Table 1 shows forms of vectors constituting the main vector. The result of the submission is
the fourteen-elements diagnostic vector, whose components were determined as:
S
tA
, S
tP
, S
tpls
current states, accordingly: active, passive or pulse,
K
zi
, K
zl
, K
c
as previously.

Automation

228

Vector denotation Contents
W
RUN
W
RUN
=[ S
tA
K
zi
K
zl
K
c
S
tp
S
tpls
]
W
BUS
W
BUS
=[ S
tA
S
tp
S
tpls
]
W
SYS
W
RUN
=[ S
tA
S
tp
S
tpls
K
c
K
zl
]

Table 1. Vectors corresponded to isolated components
The encoded states vector is unreadable for the end user. The transformation of the vector
content W
i
into a type of damage U
n
is realized through the use of rules of an inference in
the form of a simple conjunction, for example, the first vector:
If (S
t
=s
j
) (T
r
=s
j
) (K
zi
=s
j
) (K
r
6x=s
j
) Then U
n
) (3)
where:
s
j
a current state (0 or 1, j=2),
U
n
- n-th damage (defined as a unique tag of a data record).
An error description is identified on the basis of the unique damage marker Un (syntax
consistent with the equation 3). The selection function of corrective actions is defined in the
following form:
If ZU=U
n
Then [Z
AKR
= k[z]] (4)
where:
ZU a damage marker,
Z
AKR
the z-th element of the corrective actions vector.
The correct definition of the vector can lead to two situations: determined as:
the corrective actions means the vector is strictly assigned to one type of damage,
a given state vector corresponds to n failures.
In the second case the selection function of corrective actions takes the form of a logical
alternative and its form of notation can be presented in the following form:
If ZU=U
n
Then [Z
AKR1
= k[z] or Z
AKR2
= k[s] or or Z
AKRm
= k[p] ] (5)
A size of corrective action vectors is variable and depends on the type of error, impact
degree at the diagnosed system and number of steps required in order to restore the correct
status.
It is worth mentioning that the presented system is complete (contains all possible
combinations of fault states of frequency inverters). An implementation of the method was
performed using a high order language. Figure 12 shows a screen of Computer Control and
Diagnostic System of Distributed Drives used to determine the status of the integrated
frequency inverter of a MOVIFIT unit. Representation of the diagnostic vector is realized by
a states definition in subsequent dialog windows (Fig. 13).


229

Fig. 12. A view of the diagnosis window of an integrated frequency converter of the
MOVIFIT unit

Fig. 13. A view of the MOVIFIT communication bus diagnosis window

Automation

230
In the case of diagnosis of the MOVIFIT bus, the exemplary diagnostic vector is presented in
Figure 13. The division into three sub-vectors were introduced by defining the weight
marker of diagnosed subsystems. The weight is introduced to define the subsystem, among
which exist an irregularity. Selecting a tag "RUN LED" will search for solutions among the
hardware components of the application. The other markers work similarly. The whole
advisory system has been elaborated in fourteen windows.
4.2 An automatic mode of the definition of the failure states
In the case of the automatic failure detection, a phase of diagnostic vector definition is
omitted. Content of the failure vector U
n
is directly defined by using internal functions of
frequency inverters and I/O modules. The error is identified on the basis of the individual
8-bit code contained in the status word of SLAVE units.
The primary disadvantage of this approach is dependence on the correct condition of the
PLC controller and the ProfiBus network. The communication disturbance or incorrect PLC
operational status is indicated by the diagnosing system of the PLC controller. The structure
of the CCaDSoDD is hierarchical; that allows a reciprocal supplementation of defined
subsystems.
It is necessary to present the main assumption, which confirms the presented stand of the
authors, in the following form: entrusting to the advanced mechatronics devices of the
control functions (which require a high level of reliability) allows functionality extension at
additional functions and data connected with diagnostic functions without loss of the
accepted quality factor.

Fig. 14. A general view of the MOVIMOT functional block, dedicated to the fault
identification of frequency inverter, where: 1 a fault marker, 2 a number of the identified
error
The group of frequency inverters coupled by the ProfiBus DP network are controlled and
monitored via integrated control and diagnostic blocks (placed in PLC memory) and
elaborated expert system, at an angle of functional features of integrated self-diagnostic
algorithms of frequency inverters (acquiring error codes) which are utilized in the
CCaDSoDD system (a set of implication rules that converts numerical codes to text strings
with user directions).


231
4.3 Self-diagnostic functions of frequency inverters
The presented example of frequency inverters allows the identification of a domain which
could be used for computer-aided failures detection and their elimination. It is worth
stressing that modern mechatronic devices offer very reliable and expanded functions
connected with monitoring of internal subsystems.
Utilization of inner self-diagnostic algorithms coverage by specialized electronic circuits,
integrated inside mechatronic devices, brings many benefits, among others:
a separation of system functions (discharging of resources),
shortening of time assigned to detection and elimination of identified failures,
increasing application reliability.
A result of self-diagnostic functions is given in the form of individual error codes. Error
codes contain 8-bits and are easy to gather by the PLC connected with the frequency
inverters by the ProfiBus DP network. Additionally, the PLC performs several functions at
the same time:
processing of the control algorithm determined by dependences defined in accordance
with the cyclogram,
gathering the diagnostic data from frequency inverters circuits,
checking current parameters of additional elements (like sensors, electric drives,
frequency inverters, etc.) within the defined dependences (relations between states of
constituent elements) determining the current state.

Ordinal
number
Assigned error set Description of sources
1. AE-PTE
incorrect configuration of hardware resources,
causes generated in the stage of loading of
programs, settings or parameters,
an overflow of a memory stack,
disturbances of the beginning or an execution of
program actions,
abnormalities of data conversion, settings or
an active mode change.
2. EE-HTE
the ProfiBus DP bus failures,
incorrect matching of the hardware layer,
an abnormal work of hardware devices (reading
errors, incorrect hardware configurations, etc.),
diagnostic information,
failures of hardware components like electronic
circuits, feeding interfaces,
abnormal conditions of supervised units (i.e. decay
of feeding voltage, phases, etc.).
3. CE-OTE
complex relationships between identified failures
belonged to the AE-PTE and EE-HTE collections
Table 2. Types of identified failure states with description of their sources

Automation

232
All identified failure conditions were grouped by the authors into three main domains:
AE-PTE application errors (connected with programming activities, configuration
parameters, etc.),
EE-HTE hardware errors,
CE-OTE complex errors (coupled by a conjunction of previously specified groups).
The end user has not got the possibility of any influence over the form and size of the error
set notified by the internal resources of frequency inverters, because access to the embedded
circuits and functions in many cases is blocked by several methods (lack of sufficiently
documentation, resource interlocking and simplification of accessible functions).
A complete exploitation of advantages of AE-PTE and EE-THE collections enables
continuous monitoring of values stored in variables included in the control algorithm.
A designer of the control algorithm decides how to use codes and how to define mutual
dependences which define complex errors CE-OTE. Error codes are collected and gathered
in the PLC memory. On the basis of their current values and conditions of other variables,
a user realizes the diagnostic inference. Each of the frequency inverter error codes is treated
like an independent diagnostic premise.
An identification of complex errors is performed by a comparison of two (or more)
simultaneously identified errors belonging to different groups within the defined relations
placed in the statements of the expert system of the CCaDSoDD application. Data gathered
in the CCaDSoDD expert system are transmitted in on a package in every cycle of the PLC
controller.
The identified error code is sent to a PLC controller in the form of one byte, transformed up
to the hexadecimal number. For the considered groups of frequency inverters produced by
the SEW Eurodrive group, the authors isolated more than 70 failures identified only by
embedded self-diagnostic functions.
Taking into consideration functional constraints of considered devices, the final results are
satisfactory, because a defined number describes attributes referred to the hardware
subsystems (a group of diagnostic premises identified directly). The identification of
residual conditions could be realized as a result of an elaboration of decision rules matched
to the concrete application of distributed drives. A diagnostic inference is possible in two
independent modes:
directly on the basis of premises delivered in the form of frequency inverters error codes,
indirectly by connection of frequency inverters error codes and additional statements
implemented in the expert system.
Isolated errors (used in the direct diagnostic inference) expanded by relations connected
with monitored values of the control algorithm allow defining a group of corrective actions
with a quantity exceeding 270. The accepted inference method is based on dynamic
checking of a diagnosed subsystem and a failures localization model at an assumption
supported by the serial inference algorithm.
The method of identification and definition of reciprocal dependences of complex errors
were defined in accordance with the following assumptions:


233
a complex error is implicated by two (or more) errors belonging to the set of AE-PTE or
EE-HTE errors coupled with a conjunction condition (definition of used implications
shows Figure 15),
in the case of the occurrence of two or more errors from the same set (AE-PTE or
EE-HTE), errors are treated as individual diagnostic premises (adequately u
AE-PTE
and
uEE-HTE
); occurrence of many errors in the domain of one collection do not implicate
a complex error (identified diagnostic premises take part in the inference process in
accordance with defined rules in the CCaDSoDD system),
each of the errors (AE-PTE or EE-HTE) can belong to many rules implicating complex
errors.
The mediate form of the inference has a lot of advantages, among others:
a possibility of an unbounded forming of statements, fitted to the concrete applications,
shortening of time connected with an elaboration of expert systems,
an open structure and feasibility of development.

Fig. 15. A definition of the complex error implication (CE-OTE type), where:
AE-PTE
n
, EE-HTE
n
u, u diagnostic premises (AE-PTE
n
or EE-HTE
n
, nN), ERR
n
the
logical implication of a conjunction of u premises
Defined relational databases are part of the expert system. On the basis of rules, the
CCADSoDD system chooses descriptions of the identified error and corrective actions (steps
enabling a restoration of correct operating conditions). All relational databases based on the
SQL language (Structured Query Language) were divided into tables (the main key is
equivalent to the error code).
At the stage of an error identification, the authors accepted a few meaningful assumptions:
the monitoring and collection of all data are realized within PLC tasks,
maximal time of a symptom generation is equal to the PLC cycle, increased at the time
connected with ProfiBus DP bus refreshing, within the equation (Korbicz et al., 2002):

{ }
2
jk
) F(s f : k
j
max =
j k
(6)

Automation

234
where:
F(s
j
) - a failures set detected by the diagnostic signal s
j
,
2
jk
- maximal time from the moment of occurrence of the k failure to generation of the j
symptom.
a set of diagnostic tests does not include elements corresponding to symptoms with
time occurrence exceeding the time predicted to a diagnosis period,
protective functions (safety actions) of the analysed case are executed in the time equal
to the value of one PLC cycle,
dynamic properties of the diagnostic symptoms formation is dependent on the PLC
cycle,
in the elaborated expert system, the authors do not improve a states detection on the
basis of the time and detection sequence of diagnostic symptoms (a packet block
transfer).
Implication rules are referred to all groups of diagnosed units and databases of the
CCaDSoDD system. Recording of error values are stored within a defined standard (Fig. 16).

Fig. 16. A view of entry error patterns of the frequency inverter: a) basic, b) extended, where:
Nuz a user name, Padm a authorization level, Db date, Cb current time, Nord the
ordinal number of the SLAVE unit, EprfAddr network address error, Endor ordinal
number error of the SLAVE unit, Ealkd alarm code, Ealdl length of the alarm
information, Ealtp alarm type, Ealsn socket number with an active error state, Enext
extended diagnostic information code, Eextdl data length of the extended diagnostic
information, Eiden identification number of SLAVE unit
The pattern of error entries has been divided into two main subgroups:
basic - main errors of the SLAVE (frequency inverter) unit,
extended - includes all data of the basic pattern widen of diagnostic information of the
ProfiBus DP network and hardware devices.
5. An investigation of the impact of volume and syntax of the algorithm at the
performance of the PLC controller
From the standpoint of the operation of the control unit, a main parameter is refresh time
values stored in registers of the control device. The assumption of the entrusting of control
and diagnostic tasks to one device requires a high reliability factor of the data processing
element (in this case the PLC controller).
During the study, the influence of the complexity of the integrated control and diagnostic
algorithm at processing parameters of the PLC was specified (wider & Hetmaczyk,
2009d). Table 3 summarizes the configuration parameters and their numerical values,
adapted to the verification needs of the PLC algorithm.


235
Networks configuration
parameters
Configuration Physical unit
Data transmission type
Digital differential signal compatible with
RS485 standard, NRZ coding type
-
Data rate variable kbits
Data protection type
HD=4, parity bit, start/end transmission
separator
-
Data carrier Two-wire twisted shielded pair (A type) -
Power supply type With additional cable -
Network topology Main line -
Bus length 12 [m]
Number of stations 9 (MASTER + 8 x SLAVE) -
Table 3. Basic configuration parameters of the tested network
Refresh time values of the Profibus DP bus were recorded on the basis of parameters
recorded by the MASTER QJ71PB92V_ContDiagBLK dedicated block. A set of variable
parameters of realized verification contains the data transfer rate of the ProfiBus DP bus and
organization of the syntax of the control and diagnostic algorithm.
Time values of maximum and minimum cycles were subjected to an exceptional reading,
while the current time is calculated as the arithmetic mean of samples, obtained within 10-
minutes cycles of observation (recorded with a sampling resolution equal to the PLC cycle,
labelled as TC_PLC).
The presented graphs also show values of the PLC cycle time. It should be noted that the
numerical value of the first cycle of the PLC controller is not known. The only value
registered during the first cycle is the maximum refresh time value of the bus network,
given in the second cycle of the controller.
The first series of measurements were performed using the following assumptions:
program blocks were placed in the main program,
control blocks of drives are supported from the calling subroutine placed in the main
program,
additional blocks of a control algorithm (determining a course of work of controlled
objects) were positioned inside individually called subprograms.
Figure 17 shows measured values of PLC refresh times, with a characteristic of work
specified in the form of assumptions defined with respect to the first measurement cycle and
configuration of slave units shown in Table 3.
The second of the considered cases assumed placement of all blocks in the main program
window (Fig. 18). This configuration of the algorithm affects the growth of the recorded
maximal value of the refresh time of the ProfiBus network bus. The identified phenomenon
is related to the need for processing a larger number of variables, their support and
assignment of parameters contained in the main program.

Automation

236
SLAVE/MASTER
unit type
Category Network address Unit configuration
QJ71PB92V M 0 -
- NU 12 -
MFP/MOVIMOT S 3 [000] 3PD+DI/DO (MFP2x)
MFP S 4 [000] 0PD+DI/DO (MFP2x/3x)
MFP/MOVIMOT S 5 [000] 3PD+DI (MFP2x/3x)
MFP S 6 [000] 0PD+DI (MFP2x/3x)
MFP/MOVIMOT S 7 [000] 3PD+DI/DO (MFP2x)
MFP S 8 [000] 0PD+DI/DO (MFP2x/3x)
MFP/MOVIMOT S 9 [000] 3PD+DI (MFP2x/3x)
MOVIFIT S 10
[000] Slot unused
[001] MOVIFIT Status
[002] FC 3PD
[003] Slot unused
[004] FC/SC: 6/8 DI
[005] 2/4 D0
[006008] Slot unused
Table 4. A matching of units and configuration parameters of the considered application

Fig. 17. The dependence of characteristic time periods of the ProfiBus DP network (case of
the complete configuration including all diagnostic and control blocks called from
subroutines), where: TC_MIN minimum value [ms], TC_B current value [ms], TC_MAX
maximum value, TC_OBL calculated value (predicted value), TC_PLC PLC cycle time
(onefold time of an algorithm processing)


237
In the first configuration case a deficiency of variable that initiate subroutine execution
causes an omission of the algorithm execution contained in its structure, which significantly
limits the data amount subjected to processing in the first cycle of the driver.

the complete configuration including all diagnostic and control blocks placed in the main
program)

the configuration based on main blocks)

Automation

238
Because of an application of the rapid CPU unit (Q25HCPU) there were no significant
differences in the recorded values of current bus cycle times and refresh times of the PLC
resources.
Estimated (computed) values of the refresh time of the bus network, obtained from the
MELSOFT GX CONFIGURATOR DP, are on average two times smaller than the actual
recorded values. The presented fact must be taken into account at the design stage of the
industrial network structure (step of matching network module and a data transfer rate
choice).
A third case of measurements (Fig. 19) was carried out using the configuration shown in
Table 4 and the PLC algorithm, without loading of the control code (only control blocks of
the data exchange, i.e. QJ71PB92V_ContDiagBLK and PROFIBUS_NET).
In the described test cycle a slight decrease in values of maximum refresh time of bus was
noticed, both in relation to the previously described configuration. However, current bus
refresh times oscillate around previously identified values.
It can be concluded that the current cycle time of a ProfiBus DP bus does not depend on the
configuration architecture of a control and diagnostic algorithm, which confirms the
usefulness of the application in complex appliances extended at Fieldbus network buses.
Increasing data transmission speed of the network bus does not cause significant problems
in the recorded times, despite changes in the structure of the configuration of the control
and diagnostic algorithm.
The proposed form of the structure organization of the developed control and diagnostic
system is configuration organized in subroutines called from the main program.
The configuration is beneficial for reasons arising from the need to fulfil the requirements of
smooth operation of the PLC processor and exchange of data with remote units.
Small time values of the current bus cycle (TC_B) confirm the validity of the correct
implementation of the described diagnostic functions.
6. Conclusions
The presented expert system of the CCaDSoDD software is one of the modules of the
computer diagnostic system of distributed drives and it is equipped with the following
embedded functions:
on-line diagnosis,
off-line diagnosis,
preliminary diagnosis.
The computer environment is convenient to use, but an elaboration of the expert system is
rather complicated. Significant difficulties can arise for the following reasons:
the process of data gathering involves participation of high-class engineers,
a database designer must identify all possible failures,
the expert system requires performing of the additional database with stored corrective
actions,


239
the readable user interface is necessary,
work stability depends on the specified protocol.
Self-diagnostic functions are frequently used in removal of states of failures by qualified
staff, but an intention of the authors was to turn the attention to the utilization of embedded
diagnostic functions at the stage of design of the final form of applications included in
modern mechatronic devices.
Advantages of the presented approach are meaningful and can be summarized by several
points:
increasing reliability of the considered application,
a possibility of preparation of fault detection functions for every mechatronic unit
(taking into account its functional constraints and embedded functions),
utilization of more advanced mechatronic units allowing a widening of the amount of
possible identified operating conditions by embedded subsystems,
to primary parameters belong, among other things: current, minimal and maximal bus
cycles (detected by a industrial network module of the PLC unit), the diagnostic
information, a current work status, error codes, reactive and active powers, etc.
Beyond the specified advantages a group of functional constraints also exist, resulting in an
enlargement of expenditure connected with:
a preparation of the control algorithm widened by memory cells which include values
of diagnostic premises,
an elaboration of the expert system and inference rules (diversified at an angle of the
concrete application).
The application of developed control and diagnostic blocks with the proposed structure of
the algorithm allows for:
reduction of the time needed for an elaboration of the complex control and diagnostic
system,
resignation of the phase of copying the contents of global memory addresses to
memory buffers (a direct manipulation on memory buffer registers),
the implementation of different configuration types:
with an implementation of the function corresponding to the control algorithm,
using diagnostic functions,
combining these functions into the one system extended with additional
capabilities of an expert system.
7. References
Ryabinin, I. (1976). Reliability of engineering systems, MIR Publishers, ISBN: 0714709492,
Moscow
Legierski, T. (1998). PLC Controllers Programming, Jacek Skalmierski Publishing, ISBN:
838664416-8, Gliwice
Kwaniewski, J. (1999). Industrial programmable controllers in control systems, AGH Publishers,
ISBN: 83-86320-45-1, Krakw

Automation

240
Dhillon, B.S. (2002). Engineering Maintenance - A Modern Approach, CRC Press, ISBN 978-1-
58716-142-1, New York
Korbicz, J. & Kocielny, J. & Kowalczuk, Z. & Cholewa, W. (2002) Processes diagnostic: models,
methods of the artificial intelligence, applications, WNT, ISBN 83-204-2734-7, Warsaw
Boyer, S. (2004). SCADA: Supervisory Control and Data Acquisition, ISA, ISBN: 1556178778,
2004, New York Fraden J. (2004) Handbook of modern sensors physics, design,
applications, Springer, ISBN 1441964657, New York
Prasad, K.V. (2004). Principles of Digital Communication Systems and Computer Networks,
Charles River Media, ISBN 158450-3297, New York
Mackay, St. & Wright, E. (2004). Practical Industrial Data Networks: Design, Installation and
Troubleshooting, Elsevier Linacre House, ISBN 07506-5807X , Oxford
Bloch, H.P. & Geitner, F. K. (2005). Practical Machinery Management for Process Plants
Machinery Component Maintenance and Repair, Elsevier Linacre House, ISBN
0750677260, Oxford
De Silva, C.W. (2005). Mechatronics: an integrated approach, CRC Press, ISBN 0849312744, New
York
wider, J. & Hetmaczyk, M. (2009). Hardware and software integration of mechatronic
systems for an example measurement path for temperature sensors, Solid State
Phenomena, Vol. 147-149, 2009, pp. 676-681, ISSN 1012-0394
wider, J. & Hetmaczyk, M. (2009). Method of indirect states monitoring of dispersed
electric drives, [in:] Proceedings of the The Sixth International Conference on
Condition Monitoring and Machinery Failure Prevention Technologies, Dublin,
(June 2009), pp. 1171-1179
wider, J. & Hetmaczyk, M. (2009). The visualization of discrete sequential systems,
Journal of Achievements in Materials and Manufacturing Engineering, Vol. 34, No. 2,
(April 2009), pp. 196-203, ISSN 1734-8412
wider J. & Hetmaczyk, M. (2009) Computer-aided diagnosis of frequency inverter states,
Problems of Working Machines, pp. 57-66, Vol. 33, ISSN 1232-9304
wider, J. & Hetmaczyk, M. (2010). Adaptation of the expert system in diagnosis of the
connection of the PLC user interface system and field level, Solid State Phenomena,
Vol. 164, 2010, ISSN 1012-0394
wider, J. & Hetmaczyk, M. (2011). The computer integrated system of control and diagnosis of
distributed drives, Silesian Technical University Publishing, ISBN: 978-83-7335-844-7,
Poland
12
Optical Interference Signal Processing
in Precision Dimension Measurement
Haijiang Hu
1,*
, Fengdeng Zhang
2
, Juju Hu
1
and Yinghua Ji
1

1
Jiangxi Normal University, Nanchang,
2
University of Shanghai for Science and Technology, Shanghai,
China
1. Introduction
The optical techniques are widely used in the automatic measurement and test. In the
precision dimension measurement, there are some optical instruments that are widely
applied for the microscale or nanoscale dimension measurement, such as the grating, the
homodyne interferometer, and the heterodyne interferometer. An optical instrument system
is often divided into two main subsystems: the optical subsystem and the electrical
subsystem. Moreover, some optical instrument systems also have the mechanical subsystem
and the computer subsystems. The optical subsystem includes optical source and other
optical elements, such as the polarizing beam splitter, the quarter-wave plate, and so on. The
electronic subsystem turns the received optical signals into the electrical signals by the
photoelectric receiver, and uses the analog signal system and the digital signal system to
process the received signals. A typical structure of the optical instruments is shown as Fig.1.

Fig. 1. A typical structure of optical instruments: ADC, analog-digital converter.
In the performance evaluation of the optical measurement instrument, the resolution and
the precision (JCGM, 2008) are two important parameters. The resolution describes the
ability of a measurement system to resolve detail in the object that is measured. The
precision

of a measurement system, also called reproducibility or repeatability, is the degree
to show the same results when the measurement process is repeated under the unchanged
conditions. This chapter mainly discusses two topics: the signal processing method for the
improvement of measurement resolution and the elimination errors for the improvement of
measurement precision. Generally speaking, the optical source, the design of optical path,
and the signal processing are the main factors to improve the measurement resolution. For

*
Corresponding Author

Automation

242
example, in a heterodyne Michelson interferometer with double path optical difference that
is shown as Fig.2, the traveling distance of the micro-displacement platform L is as follows:

0 2
= A
}
,
T
L fdt (1)
where is the wavelength of laser, and f is the Doppler frequency shift. Obviously, is
smaller, the measurement resolution is higher.

Fig. 2. A heterodyne Michelson interferometer with double path optical difference: BS, beam
splitter; PBS, polarizing beam splitter; P, polarization analyzer.

Fig. 3. A heterodyne Michelson interferometer with fourfold path optical difference: BS,
beam splitter; PBS, polarizing beam splitter; P, polarization analyzer; QWP, quarter-wave
plate.
The multiple reflection of optical path, also called the optical subdivision, is the second
factor to influence the measurement resolution. A heterodyne Michelson interferometer

Optical Interference Signal Processing in Precision Dimension Measurement

243
with fourfold path optical difference is shown as Fig.3. When the micro-displacement
platform M
2
moves the same distance L in Fig.2 and Fig.3, the Doppler frequency shift f in
Fig.3 is twice as large as f in Fig.2. In the same configuration of laser source and electrical
signal processing system, the measurement resolution in Fig.3 is twice as large as the
measurement resolution in Fig.2. Therefore, the optical subdivision relies on the optical path
design of optical system and can be realized by the multiple reflections in the optical path
(Cheng et al., 2006).
The third method is the electrical signal processing. Because of the light attenuation in
optical parts, the optical subdivision is hard to realize the high subdivision number.
Moreover, the measurement error from the nonideal characters and arrangement of optical
parts is gradually increased when the number of optical subdivision is added. So the
electrical signal processing becomes the most useful method to improve the measurement
resolution.
Moreover, the detection and elimination of errors is the important research targets for the
improvement of measurement precision. These errors have very serious influence on the
measurement precision. In this chapter we discuss electrical signal processing of orthogonal
signals (fringes), which is widely used in the optical gratings, the homodyne
interferometers, and other optical instruments.
2. Signal preprocessing
In the optical gratings and the homodyne interferometers, the photoelectric converter turns
the intensity of light into the electrical signal. Because of the requirement of the direction
recognition when the measured object moves, the orthogonal signals whose phase
difference is /2 are used in these instruments. In order to obtain the orthogonal signals, 4-
channel signal receiving systems are designed in the optical systems. Fig.4 shows a typical 4-
channel signal receiving systems (Keem et al., 2004).

Fig. 4. 4-channel signal receiving systems: BS, beam splitter; PBS, polarizing beam splitter;
QWP: quarter-wave plate; HWP: half-wave plate;
When the measured object moves forward, the 4-channel electrical signals are collected from
the photosensors and shown as follows:

1 1 1
2 2 2
3 3 3
4 4 4
= +
= +
sin
cos
sin
cos
I A B
I A B
I A B
I A B
, (2)

Automation

244
where A
n
is the DC component of channel n, B
n
is the amplitude of AC component of
channel n, and is the change of phase that is related to the displacement L when the
measured object moves, respectively. In the homodyne interferometer with two path optical
differences, the relationship between and L is shown as follows:

4
= A
, (3)
where is the wavelength of laser.
Two signals I
x
and I
y
can be obtained by subtracting:

1 3
4 2
=
y
x
I I I
I I I
. (4)
From Eqs.2 and Eqs. 4,

1 3 1 3
4 2 2 4
= + +
= + +
( )sin
( )cos
y
x
I A A B B
I A A B B
. (5)
Under the ideal condition, A
2
=A
4
, A
1
=A
3
and B
2
+B
4
= B
1
+B
3
. So, two orthogonal signals are
obtained as follows when the measured object moves forward:

=
sin
cos
y
x
I B
I B
, (6)
where B=B
1
+B
3
=B
2
+B
4
.
Similarly, the 4-channel signals are shown as follows when the measured object moves
backward:

1 1 1
2 2 2
3 3 3
4 4 4
= +
= +
sin
cos
sin
cos
I A B
I A B
I A B
I A B
, (7)
where the parameters of Eqs.7 are the same as the parameters of Eqs.2. Similarly, from Eqs.4
and Eqs.7, two orthogonal signals are obtained as follows when the measured object moves
backward:

=
sin
cos
y
x
I B
I B
, (8)
where B=B
1
+B
3
=B
2
+B
4
.
So, L is obtained by using the half periodic counting of orthogonal signals and the
calculation of as follows:


245

2
A = +
L N , (9)
with
4
A
=

,
where is the wavelength of laser, N is the counting value of half period, and is the
additional displacement value that is shorter than /2 and can be calculated from the
orthogonal signals, respectively. For the improvement of measurement resolution, some
bidirectional subdivision methods are studied to raise the resolution of N and .
3. Bidirectional subdivision
3.1 /4 bidirectional subdivision
The simplest bidirectional subdivision method (Yoshizawa, 2005) is shown as Fig.5. This
method is applied on both the analog orthogonal signals and the digital orthogonal signals.
The principle of this method is shown in the right part in Fig.5. During the measured object
moves forward, when signal u
2
crosses zero from positive to negative, signal u
1
is positive;
when signal u
1
crosses zero from positive to negative, signal u
2
is negative; when signal u
2
crosses zero from negative to positive, signal u
1
1
crosses zero from
negative to positive, signal u
2f
is positive. Therefore, the counting of four times in a period
for forward movement is realized.
Similarly, during the measured object moves backward, when signal u
2
crosses zero from
negative to positive, signal u
1
is positive; when signal u
1
crosses zero from positive to
negative, signal u
2
is positive; when signal u
2
crosses zero from positive to negative, signal
u
1
1
crosses zero from negative to positive, signal u
2
is negative.
From the above detection, the counting of four times in a period for bidirectional movement
is realized.

Fig. 5. /4 bidirectional subdivision

Automation

246
In order to realize the counting of more than four times in a period, some methods are
studied. The first method uses two thresholds to realize /8 bidirectional subdivision by
(Chu et al., 2003). The method defines two thresholds T
h
and T
l
. T
h
is the amplitude of
normalized signal in the /4 phase while T
l
is the negative of T
h
. When u
1
cross across T
h

from the bottom to the top, the movement direction is positive if u
2
is positive at the same
time or negative if u
2
is negative at the same time. Similarly, when u
1
cross across T
l
from the
bottom to the top, the movement direction is positive if u
2
is positive at the same time or
negative if u
2
is negative at the same time. By combination between /4 bidirectional
subdivision and two thresholds detection, the /8 bidirectional subdivision is realized.
(Cui et al., 2000) proposes a method to realize /8 bidirectional subdivision by the
constructed function. The constructed function is as follows:

1 2
= f u u . (10)
Fig.6 shows the function images of u
1
, u
2
and f. According to the change of sign of u
1
, u
2
and
f, /8 bidirectional subdivision is achieved. The principle of this method is as shown in Fig.7.

Fig. 6. Function images of u
1
, u
2
and f

Fig. 7. Principle of /8 bidirectional subdivision by Eq.10


247
Similar to (Cui et al., 2000), (Chen et al., 2005) proposes another constructed function and
status code method to achieve /16 bidirectional subdivision. The difference from (Cui et al.,
2000), (Chen et al., 2005) uses 4 bits code as status parameter.
3.4 /2
n+2
bidirectional subdivision
(Hu et al., 2009) and (Hu & Zhang, 2012) proposes a method of /2
n+2
bidirectional
subdivision that uses both the constructed function and the cross-zero detection. The /8
and /16 bidirectional subdivision methods are simply introduced at first. Commonly, two
orthogonal signals are defined as follows when the measured object moves forward:

1 0
2 0
=
sin
cos
f
f
u U
u U
, (11)
where U
0
is the amplitude of the orthogonal signals, and is the change of phase that is
related to the displacement when the measured object moves.
The following two relative functions are constructed to be the reference signals in order to
achieve the bidirectional subdivision:

11 0 0
12 0 0
= +
sin cos
sin cos
f
f
r U U
r U U
. (12)
Reference signals r
11f
and r
12f
can be acquired by the digital adder. Therefore, the /8
subdivision of the two orthogonal signals can be realized easily by the zero-cross detection
which consists of the orthogonal signals and the reference signals.
Based on the /4 bidirectional subdivision When the positive voltage drives the measured
object to move forward, reference signal r
12f
crosses zero from negative to positive, signals
u
1f
and u
2f
are positive; when reference signal r
11f
crosses zero from positive to negative,
signal u
1f
is positive and signal u
2f
is negative; when reference signal r
12f
crosses zero from
positive to negative, signals u
1f
and u
2f
are negative; when reference signal r
11f
crosses zero
from negative to positive; signal u
1f
is negative and signal u
2f
is positive; Therefore, the
counting of eight times in a period for the forward movement is realized.
The same to the forward counting, its easy to realize the counting of eight times in a period
for the backward movement. When the measured object moves backward, the two
orthogonal signals are defined as follows:

1 0
2 0
=
sin
cos
b
b
u U
u U
. (13)
The same to Eqs.12, the two relative functions are constructed to be the reference signals:

11 0 0
12 0 0
=
= +
sin cos
sin cos
b
b
r U U
r U U
. (14)

Automation

248
Similarly, the counting of eight times in a period for the backward movement is also
achieved. So, a set of reference signals are only built as follows:

11 1 2
12 1 2
= +
r u u
r u u
. (15)
The bidirectional subdivision of /8 can be realized by the zero-cross detection of the
orthogonal signals and the reference signals.
On the basis of the /8 subdivision, a group of the reference signals is built which is
constituted by the two sets of reference signals. During the time of forward counting, the
reference signal sets are defined as follows:

11 0 0
12 0 0
= +
sin cos
sin cos
f
f
r U U
r U U
, (16)

21 0 0
22 0 0
2 2
2 2
= +
sin cos
sin cos
f
f
r U U
r U U
. (17)
In order to calculate the reference signals r
21f
and r
22f
, the two orthogonal signals are used to
express them. r
21f
is obtained as follows:

| |
21 0 0 0
2 2 2 1 = + = + sin cos cos (sin cos )
f
r U U U . (18)
Similarly, r
22f
is obtained as follows:

| |
22 0
2 1 = + sin (sin cos )
f
r U . (19)
U
0
=1 is still assumed. The function images of the two orthogonal signals and reference
signals for the /16 subdivision are shown in Fig. 8 and Fig. 9 during the time of the
forward counting and the backward counting. The same as the /8 subdivision, the cross-
zero detection is also used to realize the bidirectional subdivision of /16. Therefore, on the
basis of the /8 subdivision, the bidirectional subdivision of /16 for the orthogonal signals
is realized by adding a set of reference signals:

( )
( )
2
21 1 2 0
0
1
22 1 2 0
0
2
2
= +
= +
u
r u u U
U
u
r u u U
U
. (20)
Based on the /8 subdivision and the /16 subdivision, n sets of reference signals can be
built to realize the /2
n+2
bidirectional subdivision as follows:

1 1
1 0 0
1 1
2 0 0
2 2
2 2

= +
sin cos
sin cos
n n
n
n n
n
r U U
r U U
, (21)


249
where n is integer and n>0.

Fig. 8. Forward orthogonal signals and reference signals for /16 subdivision

Fig. 9. Backward orthogonal signals and reference signals for /16 subdivision
From Euler's formula:
= + cos sin
j
e j , (22)
the following equation is obtained:
= + = + cos sin (cos sin )
jn n
e n j n j . (23)
Then Eq. 23 is expanded as follows:

Automation

250
1 2 2 2
0 1 2

| | | | | | | |
+ = + + + +
| | | |
\ . \ . \ . \ .
cos sin cos cos sin cos sin sin
n n n n n
n n n n
n j n j j j
n

1
1
1
1
4
0 1
4 1
0 1
4 2
0 1
0 1
| | | | | |
+ + =
| | |
\ . \ . \ .
| | | | | |
+ + = +
| | |
\ . \ . \ .
=
| | | | | |
+ = +
| | |
\ . \ . \ .
| | | |
+
| |
\ . \ .
cos cos sin sin ,

cos cos sin sin ,
cos cos sin sin ,
cos cos
n n n
n n n
n n n
n n
n n n
j n N
n
n n n
j j n N
n
n n n
j n N
n
n n
j 4 3
| |
= +
|
\ .
sin sin ,
n
n
j n N
n
, (24)
where N is integer and N0.
From Eqs.24, the following equations are obtained:

2 2
2 2
2
=
sin sin cos

cos cos sin

(25)

1 3 3 1
2 2
1 3 1
4
2

| | | | | |
= +
| | |
\ . \ . \ .
=
| |
= + +
|
\ .
sin cos sin cos sin cos sin

,
cos cos cos sin sin
n n n
n n n
n n n
n
n
n N
n
n
, (26)
where N is integer and N1.
So, from Eqs.16, Eqs.25 and Eqs.26, the following equations are obtained:

11 0 0
12 0 0
1
= +
=
sin cos
,
sin cos
r U U
n
r U U
, (27)

( )
( )
21 0
22 0
2 1
2
2 1
= + (
= + (

cos sin cos
,
sin sin cos
r U
n
r U
, (28)

1 1 1 1
1 1 1 1
1 1
2 2 1 2 2 2 2
1 0
1 1
2 2 1 2 2 2 2
2 0
2 2
1 2
2
2 2
1 2

| | | | | |
| = + + | |
| | |
\ . \ . \ .
>
| | | | | |
| = + + | |
| | |
\ . \ . \ .
cos cos sin cos sin sin

,
cos cos sin cos sin sin
n n n n
n n n n
n n
n
n n
n
r U
n
r U
(29)
where n is integer.


251
The same as the /8 subdivision and the /16 subdivision, u
1
and u
2
are used to express r
n1

and r
n2
:

11 1 2
12 1 2
1
= +
=
,
r u u
n
r u u
, (30)

( )
( )
2
21 1 2 0
0
1
22 1 2 0
0
2
2
2
= +
= +
,
u
r u u U
U
n
u
r u u U
U
, (31)

1 1 1 1
1
1 1 1 1
1
1 1
2 2 1 2 2 2 2
1 2 2 1 2 1 1
2 1
0
1 1
2 2 1 2 2 2 2
2 2 2 1 2 1 1
2 1
0
1 2 2
1 2
2
1 2 2
1 2

| | | | | |
| = + + | |
| | |
\ . \ . \ .
>
| | | | | |
| = + + | |
| | |
\ . \ . \ .
,
n n n n
n
n n n n
n
n n
n
n n
n
r u u u u u u
U
n
r u u u u u u
U
, (32)
where n is integer.
For n=1, the number of subdivision times is 8.
For n=2, the number of subdivision times is 16. This is just shown in Fig.9 during the time of
the forward counting and Fig.10 during the time of the backward counting.
Similarly, the /32, /64, /128, , /2
n+2
subdivision can be obtained where n is integer
and n>2.
Consequently, a conclusion is obtained as follows:
On the basis of the /2
n+1
bidirectional subdivision, the /2
n+2
bidirectional subdivision of
the orthogonal signals can be realized by adding the two reference signals r
n1
and r
n2
and
their zero-cross detection. The zero-cross detection uses the original orthogonal signals and
the reference signals r
11
, r
12
, r
21
, r
22
, , r
(n-1)1
and r
(n-1)2
as the positive-negative judgment
items for the forward counting and the backward counting.
4. Error detection and elimination
In the ideal conditions, the phase difference of the orthogonal signals is /2. However,
because of the imperfect design and manufacture of measurement system, the
environmental disturbance and the system noise, the signals often have some errors, such as
the nonorthogonality, the non-equality of amplitude, the drift of DC signals, and so on
(Heydemann, 1981), which have very serious influence on the precision of the fringe
subdivision. The typical orthogonal signals with errors are shown as follows:

1 1
2 2 1
1
= +
= +
( cos sin )
d
d
u u p
u u u q
r
, (33)

Automation

252
where p,q are drift of DC, r is non-equality of amplitude coefficient, is nonorthogonal error,
respectively.
(Heydemann, 1981) proposes a universal method to compensate the errors. From Eqs.33,
another equation is obtained as follows:

2
2 2
1 2 1 0
1 1 | |
+ + + =
|
\ .
( ) cos sin u p u u q U
r r
, (34)
and

2
2 2 2 1
1 0
(
+
+ = (
(

( ) ( )sin
( )
cos
d d
d
u q r u p
u p U
. (35)
From Eq.35, the error factors are obtained as follows:

2 2
1 2 1 2 1 2
1 + + + + =
d d d d d d
Au Bu Cu u Du Eu , (36)
where

2 2 2 2 2 1
2
2
2
2
2
=
=
=
= +
= +
( cos sin )
sin
( sin )
( sin )
A R p r q rpq
B Ar
C Ar
D A p rq
E Ar rq p
,
respectively. Obviously this is an elliptic function express. The parameter A, B, C, D, and E
are called the error factors. Therefore, if the error factors A, B, C, D, and E are obtained, p, q,
r, are calculated as follows:

1
2
1
2
2
2
4
2
4
2
4

arcsin ( )
( )
C AB
B
r
A
BD EC
p
C AB
AE DC
q
C AB
. (37)
From Eqs.37, u
1
and u
2
are obtained as follows:

1 1
2 1 2
1
(
= +
( )sin ( )
cos
d
d d
u u p
u u p r u q
. (38)


253
(Heydemann, 1981) sets up a complete method for the error compensation. This method
needs at least a period of signals for the calculation of the error factors. Therefore it has the
poor real-time performance. (Zumberge et al., 2004, Eom et al., 2001) also calculate error
factors for the compensation by least-squares fitting. (Song et al., 2000) uses phase-
modulated grating to compensate for nonorthogonal errors. (Keem et al., 2004) uses Jones
matrix calculation to investigate the remaining error of a homodyne interferometer with a
quadrature detector system. (Cheng et al., 2009) uses the normalized waveforms to eliminate
DC offsets and amplitude variation. (Hu & Zhang, 2011) proposes a method for detection
and elimination of the nonorthogonal errors based on digital sampled signals as follows
when (Hu et al., 2009, Hu & Zhang, 2012) are applied in the fringe subdivision.
In (Birch, 1990), when the measured object moves forward, the two orthogonal signals
which only contain nonorthogonality error are shown as follows:

1 0
2 0
2
=
= + + A

sin
sin( )
f
u U
u U
, (39)
where is the nonorthogonality error. Similarly, when the measured object moves backward,
the original signals which only contain the nonorthogonality error are shown as follows:

1 0
2 0
2
=
= A

sin
sin( )
b
u U
u U
. (40)
The detection method of is as follows:
At first, a counter is defined as C that records the sample number of the original signal u
2
.
Then, in the forward movement of the measured object, when the original signal u
1
crosses
zero from negative to positive, the counter C resets itself and begins to count. When the
original signal u
2
crosses zero, the value of the counter C
e
is saved, and the counter C goes on
counting. When the original signal u
1
crosses zero from positive to negative, the value of the
counter C
hp
is saved. Consequently, the nonorthogonality error of two original signals can be
calculated as follows:

1
2

A =
hp e
hp
C C
C
. (41)
Similarly, when the measured object moves backward, the counter C resets itself and begins
to count. When the original signal u
2
crosses zero, the value of the counter C
eb
is saved and
the counter C goes on counting. When the original signal u
1
crosses zero from positive to
negative, the value of the counter C
hp
is saved. the nonorthogonality error of two original
signals is calculated as follows:

1
2
A =
eb hp
hp
C C
C
. (42)

Automation

254
The precision of is related with the sample number in the half period obviously. When
the sample number in the half period is higher, the precision of is higher. Moreover, the
multiple-averaging technique can be used based on sampling many times to improve the
computational accuracy.
If d is defined as the displacement value when the original signal moves a period, the
relationship between the displacement L and the phase is as follows:

0 2
2
= + e , [ , )
L Nd d
, (43)
where N is the number of periods when the measured object moves, which can be obtained
by the cross-zero detection, and can be acquired by the method of subdivision and the
counting. If the nonorthogonality error does not exist, Eq. 43 can be represented by the /8
bidirectional subdivision from (Hu et al., 2009) as follows:

8
4
0
8 4
= + e , [ , )
N d
L d
, (44)
where N
8
is the number of 1/8 periods when the counting system uses /8 subdivision
algorithm, and

8
8
=
N
N
. (45)
When the nonorthogonality error exists, the algorithm can not divide a signal period into
eight subintervals evenly. Therefore Eq.44 is useless if the nonorthogonality error exists. The
relationship of the displacement L and has to be redefined.
C
p
is defined as the sampling value of the signal in a period and can be obtained by means of
the cross-zero detection and the counting for the original signals. When the
nonorthogonality error does not exist, Eqs.43 and 44 can be shown as follows:
( )
= + e , ,
p p
p
n
L Nd d n C C
C
, (46)
and

8
8 8 8
| |
= + e
|
|
\ .
, ,
p p
p
C C
N d n
L d n
C
, (47)
where n is positive when the measured object moves forward or negative when the
measured object moves backward. In the Eq. 46, n is the count value of the original signal
when the original signal does not pass through a complete period. Similarly, in the Eq.47, n
is the count value of the original signal when the original signal does not pass through one
eighth periods.


255

Fig. 10. Subintervals of nonorthogonal signals in /8 subdivision
When the nonorthogonality error exists, eight subintervals can be arranged in two groups. It
is shown as Fig. 12. The first group includes the subinterval I, II, V, and VI, while the second
group includes the subinterval III, IV, VII, and VIII. Every subinterval in a same group has
the same interval length. So the interval length of eight subintervals can be obtained as
Table 1:

Number of
subinterval
Range of in [0,2) Interval length
I
0
4 2
A
[ , )

4 2
A

II
4 2 2
A
A [ , )

4 2
A

III 3
2 4 2
A
A [ , )

4 2
A
+

IV 3
4 2
A
[ , )

4 2
A
+

V 5
4 2
A
[ , )

4 2
A

VI 5 3
4 2 2
A
A [ , )

4 2
A

VII 3 7
2 4 2
A
A [ , )

4 2
A
+

VIII 7
2
4 2
A
[ , )

4 2
A
+

Table 1. The interval length of eight subintervals

Automation

256
So the displacement L can be calculated when the nonorthogonality error exists:

8 8
2 2
8 8
A + A
= + + ( )
a b
p
n
L N N d d
C
, (48)
where N
8a
is the count value of the subinterval when the original signal passes by the
subinterval I, II, V, and VI, N
8b
is the count value of the subinterval when the original signal
passes by the subinterval III, IV, VII, and VIII, and n is the count value of the original signal
when the original signal does not pass through a subinterval. N
8a
, N
8b
, and n are positive
when the measurement platform moves forward or negative when the measurement
platform moves backward.
Therefore, without the extra compensation for the nonorthogonality error, the displacement
L can be also calculated and the influence of the nonorthogonality error can be eliminated by
means of Eq.48.
5. Implementation platform
With the development of electronic and computer technologies, many platforms are widely
used in the signal processing for orthogonal signals, such as the hardware (Birch, 1990,
Downs & Birch, 1983), FPGA (Hu et al., 2009), the software programming (Zhang et al.,
1994, Su et al., 2000), LabVIEW (Yacoot et al., 2001), DSP(Zumberge et al., 2004), and so on.
Besides the analog signal processing and the digital signal processing, the image processing
is also an important method to realize the fringe analysis (Takeda et al., 1982, Qian, 2004).
The image processing system often get data from the CMOS or CCD cameras, and use
digital image processing methods to calculate the phase shift. The image processing
technique offers many new solutions for the fringe analysis (Patorski & Styk, 2006,
Marengo-Rodriguez et al., 2007, Bernini et al., 2009, Larkin et al., 2001, Pokorski & Patorski,
2010, Zhang, 2011) and extends the application fields of fringe analysis (Geng, 2011,
Riphagen et al., 2008).
6. Summary
In this chapter the signal processing methods for orthogonal signals are introduced. The
orthogonal signals are widely used in the optical grating and the homodyne interferometers.
In order to improve the resolution of these optical instruments, some bidirectional
subdivision methods are introduced. Besides the subdivision methods, the algorithms for
error detection and elimination are also stated for improving the measurement precision.
With the development of the hardware technology and the software technology, the
precision dimension measurement technology will be further developed and improved on
the resolution, the precision, the real-time characteristics, the applicability, and the
maneuverability.
7. References
JCGM. (2008). International vocabulary of metrology. Available from:
<http://www.bipm.org/utils/common/documents/jcgm/JCGM_200_2008.pdf>.


257
Cheng, Z. Gao, H. Zhang, Z. Huang, H. & Zhu, J. (2006). Study of a dual-frequency laser
interferometer with unique optical subdivision techniques. Appl. Opt., Vol. 45, pp.
2246-2250.
Keem, T. Gonda, S. Misumi, I. Huang, Q. & Kurosawa T. (2004). Removing Nonlinearity of a
Homodyne Interferometer by Adjusting the Gains of its Quadrature Detector
Systems. Appl. Opt., vol. 43, pp. 2443-2448.
Yoshizawa, T. (2008). Handbook of Optical Metrology - Principles and Applications, CRC Press.
Chu, X. L, H. & Cao, J. (2003). Research on direction recognizing and subdividing method
for Moir(interference) fringes. Chin. Opt. Lett., vol. 1, pp. 692-694.
Cui, J. Li, H. & Chen, Q. (2000). New digital subdividing snd rester-sensing technique for
moir fringe of grating. Opt. Tech., vol. 26, pp. 591-593.
Chen, B. Luo, J. & Li, D. (2005). Code counting of optical fringes: methodology and
realization. Appl. Opt., vol. 44, pp. 217-222.
Hu, H. Qiu, X. Wang, J. Ju, A. & Zhang, Y. (2009). Subdivision and direction recognition of
/16 of orthogonal fringes for nanometric measurement. Appl. Opt., vol. 48, pp.
6479-6484.
Hu, H. & Zhang, F. (2012). Advanced bidirectional subdivision algorithm for orthogonal
fringes in precision optical measurement instruments. Optik, in press. Available
from <http://www.sciencedirect.com/science/article/pii/S0030402611005973>.
Heydemann, P. L. M. (1981). Determination and correction of quadrature fringe
measurement errors in interferometers. Appl. Opt., vol. 20, pp. 3382-3384.
Zumberge, M. A. Berger, J. Dzieciuch, M. A. & Parker R. L. (2004). Resolving quadrature
dringes in real time. Appl. Opt., vol. 43, pp. 771-775.
Eom, T. B. Kim, J. Y. & Jeong, K. (2001). The dynamic compensation of nonlinearity in a
homodyne laser interferometer. Meas. Sci. and Technol., vol. 12, pp. 1734-1738.
Song, J. H. Kim, K. C. & Kim, S. H. (2000). Reducing tilt errors in moir linear encoders using
phase-modulated grating. Rev. Sci. Instrum., vol. 71, pp. 2296-2300.
Cheng, F. Fei, Y. T. & Fan, G. Z. (2009). New method on real-time signal correction and
subdivision for grating-based nanometrology. Proc. SPIE., vol. 7284, 728403.
Birch, K. P. (1990). Optical fringe subdivision with nanometric accuracy. Precis. Eng., vol. 12,
pp.195-198.
Downs, M. J. & Birch, K. P. (1983). Bi-directional fringe counting interference refractometer.
Precis. Eng., vol. 5, pp. 105-110.
Zhang, G. X. Wang, C. H. & Li, Z. (1994). Improving the accuracy of angle measurement
system with optical grating. CIRP Ann. Manuf. Technol., vol. 43, pp. 457-460.
Su, S. L, H. Zhou, W. & Wang, G. (2000). A software solution to counting and subdivision
of Moir fringe with wide dynamic range. Proc. SPIE, vol.4222, pp. 308-312.
Yacoot, A. Kuetgens, U. Koenders, L. & Weimann, T. (2001). A combined scanning
tunnelling microscope and x-ray interferometer. Meas. Sci. Technol., vol. 12, pp.
1660.
Takeda, M. Ina, H. & Kobayashi, S. (1982). Fourier-transform method of fringe-pattern
analysis for computer-based topography and interferometry. J. Opt. Soc. Am., vol.
72, pp. 156-160.
Qian, K. (2004).Windowed Fourier Transform for Fringe Pattern Analysis. Appl. Opt., vol. 43,
pp. 2695-2702.

Automation

258
Patorski, K & Styk, A. (2006). Interferogram intensity modulation calculations using
temporal phase shifting: error analysis. Opt. Eng., vol. 45, 085602.
Marengo-Rodriguez, F. A. Federico, A. & Kaufmann, G. H. (2007). Phase measurement
improvement in temporal speckle pattern interferometry using empirical mode
decomposition. Opt. Commun., vol. 275, pp. 38-41.
Bernini, M. B. Federico, A. & Kaufmann, G. H. (2009). Normalization of fringe patterns
using the bidimensional empirical mode decomposition and the Hilbert transform.
Appl. Opt., vol. 48, pp. 6862-6869.
Larkin, K. G. Bone, D. J. & Oldfield, M. A. (2001). Natural demodulation of two-dimensional
fringe patterns. I. General background of the spiral phase quadrature transform. J.
Opt. Soc. Am., vol. 18, pp.18621870.
Pokorski, K. & Patorski, K. (2010). Visualization of additive-type moir and time-average
fringe patterns using the continuous wavelet transform. Appl. Opt., vol. 49, pp.
36403651.
Zhang, S. (2011). High-resolution 3-D profilometry with binary phase-shifting methods.
Appl. Opt., 50(12), 1753-1757.
Geng, J. (2011). Structured-light 3D surface imaging: a tutorial. Adv. Opt. Photon., vol. 3, pp.
128-160.
Riphagen, J. M. van Neck, J. W. van Adrichem, & L. N. A. (2008). 3D surface imaging in
medicine: a review of working principles and implications for imaging the
unsedated child. J. Craniofac. Surg., pp. 19, vol. 517-524.
Hu, H. & Zhang, F. (2011). Real-time detection and elimination of nonorthogonality error in
interference fringe processing. Appl. Opt., vol. 50, pp. 2140-2144.

13
Land Administration
and Automation in Uganda
Nkote N. Isaac
Makerere University Business School, Kampala
Uganda
1. Introduction
This chapter examines automation and land administration in Uganda with the goal of
assessing the introduction of automation and whether it has contributed to more efficiency
and reliability in the operation of land administration in Uganda. From the findings we
make recommendations about the appropriate level of automation that can fully solve the
challenges of land administration in Uganda.
Land administration refers to management of land issues that involve keeping custody of
the land title records, documentation and ensuring land transactions are secure to promote
investments. The land administration system in Uganda prior to automation was paper-
based and premised on the Torrens system that guarantees land title by registration. The
land administration system was regionally based, providing separate registry offices
countrywide and one in Kampala, the capital city of Uganda. The managerial and technical
functions were vested in the Land Registrars.
Since land is an important factor in production and the governments desire for more
investments to meet the needs of the growing population, this relatively stable paper-based
system was deemed incompatible with the complex and increasing number of the land
transactions. The solution to these cumbersome and manual procedures was embracing
automation of the processes. Automation in this context called for the use of computers,
information technology and adopting online service channels that have been a dominant
theme in land administration circles worldwide. It is argued that countries that have
adopted automation of their land administration have had a complete transformation of the
way in which land registries operate. This led Uganda to embrace the initial automation by
computerizing some transactions in the 1990s in order to meet modern service level
expectations. This computerized land registration system saw the conversion of existing
paper-based registers, to being maintained electronically and providing online delivery of
title registration functions.
These developments have improved the efficiency of land administration operations and
made records more accessible to all stakeholders. However, the system does not fully allow
for online transaction links to the key players in land transaction such as lawyers, banks,
investors, sellers and buyers. Therefore, future plans will require that land administration in

Automation

260
Uganda is fully automated, with the possibility of online title registration system that
enables subscribers to submit land title transfers, and discharge of mortgage for registration
in electronic form over the Internet, thereby enhancing/adopting e-dealings access by the
licensed law firms, banks, other transacting parties and administrators. It is anticipated that
obtaining digital certificates and existing office procedures and document management
systems must be aligned to the electronic environment.
2. Automation
Automation has gained widespread usage in recent years in various processes of both
public and private organizations. Indeed there is agreement among scholars that automation
application is usually pursued judiciously in organizations (Qazi, 2006). Sheridan (2002)
offers varied conceptualizations of automation, such as data processing and decision
making by computers, while Moray, Inagaki and Itoh (2000) perceive it to be any sensing,
information processing, decision making and control action that could be performed by
humans, but is actually performed by machines. As Sheridan and Parasuraman (2006), and
earlier Kaber and Endsley (2000), contended, human-automation interaction explains the
complex and large scale use of automation in various fields. Further, it explains the ability of
humans to interact with adaptive automation in information processing, hence enhancing
the achievement of optimal performance within an organization.
The literature argues that automated assistance is usually adaptively applied to information
acquisition, information analysis, decision making and action implementation aspects, (Lee
and Moray, 1994). However, the choice of a framework for analysing and designing
automation systems is grounded in the theoretical framework and addresses aspects such as
the role of trust, system acceptability and awareness measures (Lee and Moray, 1992). While
automation is deemed to perform higher level problemsolving tasks, often the human
capacity limitations lead to errors thereby intensifying the challenges of adopting of
automation Frank (1998). This constraint subsequently determines the level of automation in
allocating the functions to be automated, in particular, the level of the desired autonomy
that represents the scale of delegation of tasks to automation and the associated implications
for reliability, use and trust (Lee and See, 2004; Lewandowsky et al., 2000).
One driver of adopting automation is its ability to enhance operational efficiency of processes
and the associated positive changes in the productivity of the organization. Notwithstanding
this increased output rate, automation is often problematic, especially if people fail to rely
upon it appropriately (Adam et al., 2003). The support for this contention is that technology is
shaped by the social setting on the one side and trust that guides the assurance and reliance
of the stakeholders on the other. This is in line with the fact that automation characteristics
and cognitive processes affect the appropriateness of trust (Itoh, 2011).
Endsley and Kaber (1999) look at another dimension of automation by arguing that
automation is applicable to different aspects of organizations, but to a varying degree, hence
creating different levels of task autonomy. Therefore, the debate on automation can only be
conclusive if it addresses the issue of whether partial or full automation is the desired goal
of the organization. The relevance of full automation achieving the desired performance of
that organization is usually at contention, though achieving a reasonable level of

Land Administration and Automation in Uganda

261
automation, especially for those tasks that are performed by human beings, is certainly the
aspiration of every organization.
The automation of individual intellectual capabilities would allow accomplishing a higher
level of automation in organizations, since within the information systems domain,
managerial decision making is of special importance (Kaber and Endsley 2004). The variety
of tasks that are performed by managers is immense, hence justifying automation of some,
especially those that are simple, complex and repetitive with a distinct application domain
while leaving those that seem not to permit automation. The argument for limiting
automation is that managers act in an environment that is characterized by ambiguity and
risk, hence the creation of room for discretion or judgment.
The above position finds support in the management automation scenarios advanced by
Koenig et al. (2009), however, they castigate the failure to understand the effects of
automation. Muir (1992) states that this is due to the problem associated with either under- or
over-trusting of the automation process. The answer to the problem lies in embracing
progressive automation as a way to introduce automation in a manner that quickly leverages
the positive effects of automation, while reducing the potential negative effects through a
gradual increase in automation. This will help align the various management automation
systems to a simple common model, where users will have a better knowledge and more
experience when pursuing a high level of automation in the future. Further, one contribution
of progressive automation is its ability to have people build up the appropriate amount of
trust in the automation systems structure and behaviour, its components and data.
According to Lee and Moray (1992), the decision to rely on automation by the users depends
on both trust and self-confidence. Where there is overriding self-confidence in ones ability
to perform a task over ones trust in automation, one is most likely to perform the task
manually. The reverse is true in incidences where trust in automation is greater, then
reliance on automation will dominate. However, Riley (1996) introduces mediating domains
of automation reliability and the level of risk associated with the particular situation. The
decision to use automation can depend upon different system management strategies (Lee,
1992) and user attitudes (Singh, Molloy & Parasuraman, 1993).
Another distinction in how to use automation is reliance (Meyer, 2001). Reliance refers to the
assumption that the system is in a safe state and operates within a normal range, (Dzindolet
2003).Over-reliance is attributed to factors of workload, automation reliability and
consistency, and the saliency of automation state indicators (Parasuraman & Riley, 1997).
Inappropriate reliance on automation relative to the automations capabilities may reflect
poorly on calibrated trust, automation bias and complacency, and may also reflect failure
rate behaviours (Moray, 2003).
Trust as an attitude is a response to knowledge, but other factors do intervene to influence
automation usage or non-usage decisions (Muir, 1997). While trust is an important element
in those decisions, it is far from the only one. According to Lee and See (2004) humans use
alternate routes by which they develop their trust, namely the analytic methods that assume
rational decision making on the basis of what is known about the motivations, interests,
capabilities and behaviours of the other party. Cialdini (1993) argues that we tend to trust
those people and devices that please us more than those that do not. Further, there is
realization of the temporal element to trust building that takes time to acquire, whether
through experience, training or the experiences of others.

Automation

262
3. Extent of automation in land administration
In setting the stage for automation there is acknowledgement that automation is applied to
various organizations, however, the feasibility of any applicability necessitates
understanding the nature of the transactions involved and the managers that perform the
tasks (McLaughlin, 2001).
Introduced in 1908, the land administration management in Uganda is based on the Torrens
system developed in 18
th
century Britain. The tenets of the Torrens system are that the
government office is the issuer and the custodian of all original land titles and all original
documents registered against them. Further, the government employees in their
management tasks examine documents and then guarantee them in terms of accuracy
(Barata, 2001).
The Torrens system has three principles: the mirror, curtain and insurance principles. While
the mirror principle refers to certificates of title, which accurately and completely reflect the
current facts about a persons title, the curtain principle ensures that the current certificate of
title contains all the relevant information about the title that creates certainty and offers
assurance to the potential purchaser about the dealings on any prior title. The whole trust
and confidence in the transactions covered by the insurance principle will guarantee the
compensation mechanism for loss of the correct status of the land. These combined
principles contribute to secure land transactions and the development of the land market in
any given country.
Practitioners in land transactions argue that the Torrens system ensures that the rights in
land are transferred cheaply, quickly and with certainty. For manually managed systems,
this is conceivable in cases of low volumes of transactions; otherwise, large volumes of
transactions necessitate automation. It is argued that the benefits of an automated system
will lead to efficiency, accuracy, integrity and cost containment. This will overcome the
challenges of retrieval of documents and the inability to manage and store large amounts of
data efficiently (Ahene, 2006).
The management tasks in land offices include capturing the precise parcel of land, the
owner, limitations of the right of ownership and any right or interest which has been
granted or otherwise obtained. Another management task is the cancellation and creation of
certificates of title, land notifications and transfers, subdivisions, showing all outstanding
registered interest in the land, such as mortgages, caveats and easements
In the management of land transactions there is legal examination of all the associated data
entry on documents to make sure that the documents are correct and in compliance with the
law affecting land transactions. The objective of the examination is to ensure that the
document complies with all applicable law and therefore, this process involves making
judgments upon the relevancy of the law and ensuring certainty of the transaction.
Another management task within land transactions are those roles performed by the survey
staff which involve reviewing and making the associated data entries, and comparing and
interpreting the existing land survey evidence with the new ones to ensure that the land
surveyed on the new plan does not encroach upon adjacent lands (Barata, 2001).
In reference to the existing documents, the management task involves processing of searches
of the records in the land office that are classified as public records which can be searched
by anyone wishing to transact in land matters. Request for searches are usually received


263
from agents, particularly lawyers and bank staff, that seek to verify the documents to
conclude the transactions of their clients. To obtain a title search you must know the legal
description, land identification numbers or the title number for the property you want to
search. The primary purpose of the name search facility is to enable creditors and other
parties with statutory rights to determine what interests in land are owned by the person
affected by instruments.
While the above management tasks are simple, with increasing volumes of transactions, the
land office in Uganda was characterized as inefficient and riddled with unethical behaviour.
This outcry by stakeholders led to the adoption of the automation of the processes of the
land office in two phases. The first step of automation was initiated by the government
through the Ministry of Land and Urban Development. The approach involved capturing of
the records in a digital format to create a reliable computer database to allow prompt
searches and retrieval of information. The second phase saw the introduction of automation-
based decision making. These two phases meant that automation included data entry,
indexing and scanning of Mailo land records. These fully automated interventions in the
land registry helped generate computer-related information on land ownership, information
on Mailo land transactions such as changes in property ownership and encumbrances
thereon. The automation of the land office meant all live paper titles were converted from
their paper format into an electronic medium. In addition to each certificate of title being
assigned a unique title number at the time of its creation, each parcel of land contained
within each title was also assigned a unique code number.
A critical review of the documents produced during the process of automation reveal that
the conceptual plan included a progressive computerization of the legal and administrative
records, and cadastral maps. The aim was to address the shortcomings of the manual
system, restore the integrity of the land registry and ensure modernization of land registry
operations to meet the needs of a growing economy. It was quite logical to start the process
of the actual computerization of the land registry after getting the filing right and re-
organizing all the registry records, reconstructing the torn and damaged records and vetting
of all records in the manual system. This helped identify and get rid of any forgeries or
problem land titles in circulation.
As a result of automation, most titles, all registered documents with a registration number
and all plans can now be searched and retrieved electronically, in comparison to the
previous practice of only being able to search through a registry agent.
There are many advantages in the automation of the land office over the manual system,
meaning that documents can be searched electronically and in a fast and expeditious manner.
The documents can also be easily replicated, averting potential loss from natural disasters
and lastly the documents are linked with other documents. This chimes well with Maggs
(1970) assertion that automating land systems is only commercially practical where highly
formalized rules are applied to highly standardized data. Where rules and data are not of a
highly formal nature and so require a large degree of individual judgment, progress can only
be achieved through the substitution of new formal rules and data structures.
4. Methodology
The aim of the research for this book chapter was to examine the automation and efficiency
of the land transactions management office in Uganda. In analysing the automation of the

Automation

264
land transactions in the Ugandan land office, we adopt a cross sectional, qualitative and
analytical study design in order to examine automation operations in the three districts of
Kampala, Mukono and Mpigi. Primary information was acquired via semi-structured in-
depth interviews and focus group discussions (FGDs). The main instrument of data
collection was a structured questionnaire for the Key Informant Interviews (KIIs)
comprising of lawyers, bankers and staff of the land office.
The questionnaire contained open-ended questions and was developed based on the
literature review, as well as drawing from other studies on automation. The key questions
that were used related to trust in automation, reliance, usability and confidence. The
information obtained formed the basis for the analysis of the findings. The instrument was
pilot tested prior to beginning fieldwork to assess the clarity, flow and appropriate context of
the questions. Following the pilot test, the questionnaire was refined based on the initial
feedback.
While carrying out the study, there should be a population from which a sampling frame
can be drawn. The population of the study was drawn from the three district land offices of
Kampala, Wakiso and Mpigi that have fully automated their operations. Further, focus was
on the institutions that were identified as being key stakeholders to the land transactions,
namely the banks, law firms and the administrators of the land offices. Therefore, interviews
and focus group discussions comprising lawyers, land office administrators and bankers
were conducted to evaluate how the three principle users gauged the automation efforts by
the land office.
Using a purposive sampling technique, 11 lawyers were purposively selected from law firms
that frequently transact land issues, seven staff from the legal departments of seven banks
and 15 staff of the land administration. The purposive sampling was based on the frequency
of land transaction by the law firms and the banks in the last six months preceding the study.
To capture different perspectives and experiences, the staff of the land offices had varying
roles and levels of participation in activities and decision-making processes.
The data collection fieldwork in Kampala commenced in July 2011 and lasted through to
August 2011 in the other two districts. To deal with the ethical considerations of the study,
the firms and institutions were briefed about the study and appointments were made in
order to have meaningful and well-planned discussions without disrupting their normal
work. The participants were briefed and verbal consent was sought from all participants.
The exercise started with giving the respondents background information about the study.
5. Findings on automation and land management
The proponents of automation argue that when organizations undertake automation it
impacts positively on employee productivity and improves efficiency and decision making,
besides allowing better management control of operations. At the same time, the opponents
of automation are critical of it because it lessens the responsiveness of the bureaucrats to
citizens, puts technological elite in charge of processes and shifts power to managers, among
others. These diametrically opposed positions are therefore considered when analysing the
impact of automation in the organizations that have adopted or intend to adopt it. The
findings follow the objective of this study which was to examine automation and the
efficiency of land administration in Uganda.


265
The results of the study show that there has been successful automation in land administration
in the three districts following the adoption of a comprehensive computerization, with online
access to the system minimizing the need for the paper-based system.
The study sought an assessment of the experiences of the employees with automation of
land management. The response was varied as the consequences to the clerical and
managerial level staff were differentiated. Most of the clerical staff thought that their own
interest in the job had been eroded. However, when responding to how automation has
influenced the productivity of the employees, they revealed a registered increase in output.
This is quite conceivable since automation released the employees from the clumsy, tedious
and dusty manual tasks, thereby allowing them to learn new skills that are ideal for
enhancing the production of better quality work at a fast rate. Automation has changed our
perception of our jobs which has resulted in improved quality, quantity and accessibility of
data and hence, improved services from the land office. This was against the findings that
indicated an initial high sensitivity and fear of learning the new systems. The automated
institutional infrastructure so far has reduced the internal resistance to operate the
computerized system to its full potential. While this has been achieved, there remains the
fact that the transactions of the stakeholders are not automated and in this the non-linkage
to key offices, such as law and banking firms, is a limitation on the availability of the
information. This means they have to come to the land office to undertake transactions
instead of doing it online through networked systems.
Secondly automation has, besides improving productivity, impacted on efficiency. As
indicated by the responses of participants it has been positive because there has been a
substantial reduction in the transaction time from years in some instances to just a month or
two from the time of lodging the transaction. This can be attributed to far easier searches
and retrieval of land records in digital format through the automated system. The quick,
easy information retrieval and search procedure means that the partial automation of the
tasks has facilitated the public sector managers to continue making decisions in line with the
set procedures and rules. This implies that while full automation has eased the retrieval of
information, the decision making processes still require judgement based on the procedures
and hierarchy, and this impacts upon the level of adoption of automation and the efficiency
of public managers action. Indeed the respondents reported cases of a similar approach and
therefore, no indication of a drastic reduction in red tape. This argument points to what is
discussed in the literature about the level of automation in that certain decisions have to be
made by the managers hence, limiting the much anticipated benefits of automation.
Thirdly the sceptics of automation doubt the rationale for automation, a fact reiterated by
many stakeholders who contributed their insight to the study. They argue that even under
the manual system, retrieval personnel were always quick if one complied with their rent
seeking motives. By proxing that the manual process was slow, one negates the fact that the
motivation of rent seeking created the delays, rather than the inability to get the
information. The clerical personnel usually made standard statements to the effect that
information was not available, could not be traced or that documents were missing with the
stakeholders having no avenue by which to prove otherwise. As one respondent noted:
the problem hasnt been really the manual system, rather the attitudinal behaviour of the clerical
staff that perform their duties upon extra facilitation or inducement by those seeking the services
(comment by one legal respondent).

Automation

266
The findings point to the fact that notwithstanding the positive effects of automation, the
services of the land administration have been depersonalized. While this is good because it
is likely to limit the interface between the service providers and the stakeholders and
therefore, reduce the rent seeking motives of the managers, on the other hand it erodes the
trust, confidence and networks that have been built by the frequent and regular users of the
system. Stakeholders continue to express misgivings about automation, including mistrust
about the extent to which automation will free them from forgeries and malpractice as
experienced in the past. Automation has removed the avenue by which to obtain
preferential treatment by the managers, particularly in matters of urgency. The question is,
has this reduced the avenue by which to pursue personal or institutional vested interests.
While some would answer in the affirmative, others think that even with automation, the
respondents revealed that still a number of officials intervene in a single transaction. Hence,
quick access and transparency of the land records is not fostering investments in land as it
can facilitate the speeding up of processes of transactions, credits, transfers and mortgages.
One of the findings from the public managers indicates their pleasure with the ability to
offer easier access to information and services through the use of the web and integrated
database. With automation we cant reach citizens in different environments and we are now
handling public requests for information and service in a timely manner (remarks by public
manager). However, there is also a fear in the minds of the people about the potential
manipulation of these records.
6. Conclusion
The qualitative findings have been presented and the evidence from the respondents point
to increased efficiency, productivity and job satisfaction in land administration. The
automation of the land administration has been an achievement given the fact that it had
been crippled for years and was unable to cope with the increasing volume of transactions.
Secondly, automation of the land transactions offers the society a great opportunity in
public access to information that had hitherto been under the sole custody of state
functionaries, thereby leading to rent seeking practices on a massive scale. The automation
of land administration has facilitated citizens access to it and, by implication, increased
tenure security.
While the public managers have gradually gained trust in automation, the daunting task
that remains is to build trust among the stakeholders that the information generated
through automation will not be subject to fraud and the risk of loss of property. Many
people still insist on the old paper documents as a reassurance to the different market actors
and state public managers. Associated with this is the fact that while automation benefits
have eased decision making, the findings revealed that in order to restore citizens
confidence in land administration, induced by the questionable ethical behaviour of the
managers, internal change is required.
What remains to be seen is whether the new system steers clear of corruption or
alternatively aims at grappling with the existing system to change its dynamics by bringing
about structural and technological change.
The project in the three districts has served as a testing ground for working towards
achieving a more efficient nationally managed system. It will enhance efforts to increase


267
easy public access to land records and the idea of running an automated system parallel to
the existing manual system raises the expectation that the automated-based system will
demonstrate its superiority over the manual system.
Lastly automation and creating an electronic system containing the possibility for electronic
searches, analysis, retrieval and manipulation with an electronic version of the document in
a non-revocable electronic form would give the required safeguards.
Like any study, we experienced the shortcoming of unwillingness by respondents to release
information due to a fear that it would be revealed. This was overcome by asking the
respondents to remain anonymous. Further limitations of the study were the fact that the
system is still new and therefore, there is still a lot of old practice, which make assessment of
the gains rather difficult. Despite all this the findings reveal there is a general positive
attitude to automation.
7. References
Adams B.D, Bruyn L.E, Houde S and Angelopoulos P (2003) Trust in Automated Systems
Literature Review Human system incorporated, Toronto
Ahene, R. (2006) Moving from Analysis to Action: Land in the Private Sector
Competitiveness Project :World Bank / Private Sector Foundation Uganda.
Barata, K. (2001) Rehabilitating the Records in the Land Registry, Final Report;
USAID/SPEED (www.decorg/partners/dex_public/index)
Cialdini, R.B. (1993). Influence: Science and Practice. 3rd Ed. New York; Harper Collins.
Dzindolet,M. (2003) The role of trust in automation reliance, International Journal of Human-
Computer Studies,Vol. 58, Issue 6, Publisher: Academic Press, Inc., pp.697-718
Endsley, M.R.(1996) Automation and systems awareness: In Parasuraman & M Mouloua (Eds)
Automation and Human Performance: Theory and Applications (pp. 163-181)
Endsley, M.R. & Kaber, D.B. (1999) Level of automation effects on performance, situation
awareness and workload in a dynamic control task Ergonomics, 42(3), 462-492
Frank U (1998) Increasing the level of automation in organization, remarks on formalization,
contingency and social construction of reality. In: The Systemist, Vol.2091998) p, 98-
113
Itoh, M. (2011) A model of trust in automation: Why humans over-trust? SICE Annual
Conference (SICE), 2011 Proceedings:
Kaber, D B. and Endsley, M (2000) Situation Awareness & Levels of Automation, Hampton,
VA: NASA Langley Research Center; SA Technologies -
Kaber, D B. and Endsley, M (2004) The effect of level of automation and adaptive
automation on human performance, situation awareness and workload in dynamic
control tasks: Theoretical issues in Ergonomics science 5, 113-153
Koenig, R., Wong, E.,& Danoy, G (2009) Progressive automation to gain appropriate trust in
management automation systems. Degstuhl Seminar proceedings, Combanatorial
scientific computing, http://drops.degstuhl.de/opus/volliexte/2009/2093
Lee, J. (1992). Trust, self-confidence, and operators adaptation to automation. Unpublished
doctoral thesis, University of Illinois, Champaign.
Lee, J. & See, K. (2004). Trust in Automation: Designing for Appropriate Reliance. Human
Factors, 46 (1), 50-80.

Automation

268
Lee J &Moray N (1994) Trust, self-confidence and operations adaptation to automation.
International Journal of Human Computer Studies 40,153-184
Lee, J.& Moray,N. (1992) Trust, control strategies and allocation of functions in human-
machine systems. Ergonomics 35, 1243-1270
Lewandowsky,S., Mundy, M.&Tan, G. (2000) The dynamics of trust: comparing humans to
automation.Journal of experimental Psychology: Applied, 6(2) 104-123
McLaughlin, J. (2001) Land Information Management: From Records to Citizenship Online
Meyer (2001) Effects of warning validity and proximity on response warnings.Human Factors
43(4) 563-572
Maggs, P. (1970) Automation of the land title system
Moray, N. (2003) Monitoring, complacency, scepticism and eutectic behavior : International
Journal of Industrial Ergonomics, Volume 31, Number 3, pp. 175-178(4) Publisher:
Elsevier
Moray, N., Inagaki, T.&Itoh,M. (2000) Adaptive automation, trust and self-confidence in
fault management of time critical tasks, Journal of Experimental Psychology Applied
6(2) 44 -58
Muir, B.M. (1997) Trust between humans and machines: the design of decision aids,
International Journal of Man-Machines studies 27, 527- 539
Muir B.M (1992 Trust in automation: Part I. Theoretical issues in the study of trust and
human intervention in automated systems Ergonomics Volume 37, Issue 11, 1994
(1905-1922)
Parasuraman, R. & Riley, V. (1997). Humans and automation: Use, misuse, disuse, abuse.
Human Factors, 39, 230-253.
Qazi, M. U. (2006) Computerization of Land Records in Pakistan, LEAD International,
Islamabad.
Riley, V. (1996). Operator reliance on automation: theory and data. In R. Parasuraman & M.
Mouloua (Eds.), Automation and human performance: theory and applications (pp. 19-
35). Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
Sheridan, T.M. (2000) Humans and automation: systems design and research issues, HFES
Issues in Human Factors and Ergonomic Series volume 3 Santa Monica CA; Whiley
and Sons Publication Inc.
Sheridan and Parasuraman, 2006. Human-automation interaction. In: Nickerson, R.S. (Ed.),
Reviews of Human Factors and Ergonomics, vol. 1.
Singh, I., Molloy, & Parasuraman, (1993). Automation induced complacency: development
of complacency potential rating scale: International Journal of Aviation Psychology
14
A Graphics Generator for Physics Research
Eliza Consuela Isboiu
Department Informatics, Faculty of Accounting and Finance,
Spiru Haret University,
Romania
1. Introduction
The great quantity of information which emerges from the necessity of an informational
society can be mastered in a proper way, due to the perfection of the informatics systems
used in all the fields. The pieces of information, due to the fact that they are numerous as
concerns their content, become hard to work with, on the whole; thus, only a part of the
content is represented as compound and flexible objects, by assuring both their direct usage
and re-usage within other conditions and situations. An important factor in the trend of
spreading and adapting the application on the web zone is represented by distributed
systems and web services.
Generally speaking, the administration of resources and fluxes in distributed systems of
great dimensions are confronted with many difficulties. These are provoked by the
complexity of the hardware and software platforms and, especially, by the dynamism which
is characteristic of these systems compounds.
These drawbacks can be met through two necessary conditions, based on the understanding
of the performing characteristics of the resources of administration systems and of the
delivery solutions for a better dynamic adaptation.
This chapter proposes to demonstrate the utility of distributed systems in the design of
educational and research software in physics. Educational systems are generally rich in
structured knowledge, but not so complex, which does not imply behavioural relations
among users.
However, all of the existing architectures being their construction with the experts in the
experience of the learning domain, but not from the users perception, as well as from the
study of the materials of the way one constructs the schemes of mental concepts (Hansen,
2007). One has to underline that there is an intrinsic connection between the results obtained
by the users and the method of teaching. Due to the restriction imposed by time and
resources, the great part of the implementation is reduced to simple models of knowledge.
Thus, the present educational systems do not propose to determine the real state of
knowledge, and so the strategies used by the users are omitted as are the specific strategies
for each and every domain.

Automation

270
2. Theoretical background
Because of problem which one follows wants to develop consists in realizing web services
with numerical calculating methods, I provide a presentation of some points of view
connected to the subject. The basic standards, originally proposed by IBM, Microsoft, HP
and others, are: SOAP (Simple Object Access Protocol), which offers a standard modality of
connecting to the web, UDDI (Universal Description, Discovery and Integration), which
represents a standard modality of issuing ones own web services, and WSDL (Web Service
Description Language), which provides further information for understanding and using
the offered services (Landau 2008). WSIF (Web Services Invocation Framework) represents
the invoking of the web services, defined by the standards WSDL and UDDI (Universal
Resource Locator). This kind of invoking does not suppose the creation of SOAP XML
messages. There are four kinds of transmitting an invoking: one-way - the final point
receives an answer; request-answer the final point receives a message and transmits a
correlated message; solicitation-response - the final point transmits a message and receives a
correlated message; notification the final point transmits a message. The most used are
one-way and request-answer. The location transparency is realized through link patterns, as
a UDDI type. The web services revolute the way in which one application communicates
with other applications, thus offering a universal pattern of data which can easily be
adapted or transformed. Based on XML, the web services can communicate with platforms
and different operating systems, irrespective of the programming language used for their
writing. For easy communication among the business parameters: the Java language allows
the applications realization regardless of the platform used. The apparition of the Web
services represents a means of communication on a large scale. This can be done by
applying some standardized protocols, defined by standardized public organizations. The
basic protocols of the Web service are not complete as regards the description and
implementer terms, their security, reliability and complexity. It is known that by the fact of
using Web technology, the information uses many sources. The portal is the mediator for
the access to the information. The portal servers are, in their turn, elementary components in
the interchange of information. The importance of these technologies is due to the fact that
they emerged as a response to the present tendencies manifested in the software industry:
distributed remaking, the development of applications based on the components, the
development of the services for companies and the modifying of the Web paradigm, in the
sense of the development of applications.
3. Services for numerical calculation
The purpose of the present study is the analysis of the surplus introduced by the Web
service interface and its exposure in as simple as possible a variant of the facilities
introduced by it. The study uses the book Numerical Library in Java for Scientists and
Engineers, and so the present work does not re-implement the codes in that book, but only
uses those already implemented. The passing from an interface with the user in a command
line to an interface based on Web services represents a big advantage because it allows the
combining and re-using of some procedures with a prior definition. Many discussions about
the algorithms in the speciality literature on the Web omit important details, which can only
be uncovered by encoding or be made suitable by reading the code. We also need the real
code for object-oriented programming, found in the Numerical Library in Java for Scientists


271
and Engineers (Lau, 2004). An orientation of these solutions is for scientific explanations
which are based on numerical calculation, beginning with elementary structures,
progressing to complicated ones, as the integration of ordinary differential equations (Lau,
2004). The digital technologies introduced to the market include Web technology; this
completely fulfils all of the requests connected with the information cost, stocking and
spreading (Petcu, 2008). From these first steps where the sites were using simple visiting
cards, iterative processes have been created and developed. Once, with the movement
towards an informational society, a novelty is represented by the distributed systems of
large dimensions, by which facility is enabled access to a great variety of resources. The
above mentioned book offers a general discussion for each subject, a certain amount of
mathematical analysis (IBM, 2009), a certain discussion of the algorithm and, most
importantly, the implementation of these ideas in a real mode as routines. There is a proper
equilibrium among these ingredients for each subject. A starting point is concerned with the
construction of the services and the inclusion of one or more methods in implementation is
constituted by the analysis of the data structures from the elementary to the complex
(Isbasoiu, 2009). The majority of the books on numerical analysis use a specific subject
"standard", such as: linear equations, interpolation, extrapolation and ordinary differential
equations. Other discussed points are: the functions evaluation and especially mathematical
functions, aleatory numbers, the Monte Carlo method, sorting, optimization,
multidimensional methods, the Fourier transformation, the spectral methods and other
methods concerning statistical descriptions and data modelling and relaxing methods. The
book (Lau, 2004) says that there are seven groups of themes that should be focused on:
vectors and matrices, algebraic evaluations, linear algebra, analytical evaluations, analytical
problems, special functions, proximity and interpolation. Each group has the name Basic
followed by the appropriate number for each theme. A schematic representation of these
groupings is presented in Figure 1 (Isbasoiu, 2009):

Fig. 1. The basic grouping

Automation

272
Every group contains an impressive number of functions. Through an attentive analysis, we
noticed the following: Basic 1 is the base for the construction of the other basics, and some
functions of Basic 1 are applied even with the construction of the other functions of this
basic; every basic interacts with the other basics; there are functions which do not interact or
else do not represent the basis for the construction of other functions. In trying to conclude
all of these representations in one, it appears in the following figure:

Fig. 2. All dependencies representations
In the following table is presented the number of times a function is used by other functions:

Table 1. The dependencies representation


273
A graphic representation of these dependencies, depending on the number of functions that
appear, is realized as follows:

Fig. 3. The graphical representation of the dependencies according to the number of
functions
3.1 The functions for numeric calculation Time comparison
The services were realized using the Eclipse platform. For each function, the response time
is specified. We can notice that it varies a lot. We repeat the measurements for more complex
input data, both for double type and integer type and vectors and matrices because these
units of time depend on this input data.

Function
Name
Time of response with Web
Service (val/ms) -simple values
Time of response with Web
Service (val/ms)-complex values
Function
Name
Rnk1min 16 143 Rnk1min
Praxis 11 580 Praxis
Marquardt 13 541 Marquardt
Gssnewton 14 642 Gssnewton
Multistep 10 533 Multistep
Ark 15 570 Ark
Efrk 13 13 Efrk
Efsirk 561 616 Efsirk
Liniger1vs 456 603 Liniger1vs
Liniger 2 473 583 Liniger 2
Gms 605 470 Gms
Impex 562 524 Impex
Peide 628 587 Peide
Minmaxpol 611 552 Minmaxpol
Table 2. Time of response for every function with the Web Service
An elegant graphical representation of the response variations can be seen in the following
graphic:

Automation

274

Fig. 4. Graphical representation of the response variations for every function
We repeat the measurements, without the Web Service, for more complex input data, both
for double type and integer type and vectors and matrices, because these units of time
depend on this input data.

Function Name
Time of response without
Web Service (val/ms) -
simple values
Time of response without Web
Service (val/ms) - complex
values
Function Name
Rnk1min 10 36 Rnk1min
Praxis 8 368 Praxis
Marquardt 9 336 Marquardt
Gssnewton 7 398 Gssnewton
Multistep 8 327 Multistep
Ark 11 352 Ark
Efrk 10 7 Efrk
Efsirk 314 326 Efsirk
Liniger1vs 289 318 Liniger1vs
Liniger 2 293 359 Liniger 2
Gms 348 236 Gms
Impex 311 312 Impex
Peide 362 362 Peide
Minmaxpol 353 342 Minmaxpol
Table 3. Time of response for every function without the Web Service
A more elegant representation of time variation can be seen in the following graphic:

Fig. 5. The graphical representation of time response of each function


275
In analysing with a focus on the two graphics (Fig. 3 and Fig. 4), we can notice the following:
a. generally, the response times are longer for complex parameters, the main cause stands
for the value itself of those parameters
b. the longest time belongs to those functions which resolve determinants, matrices and
vectors
c. there are extremely close response times for some functions
d. we can see two groups of functions, corresponding to two types of functions
e. it is clear that the only function which does not modify its time is Efrk, having the value
of 13 ms - with the Web Service
f. irrespective of the type of data introduced - or of their complexity - the response time is
bigger, but still close
g. by comparing the bits of response time with the Web Service there are significant
differences, so with this case it is important to consider the complexity of the input data
3.2 Scenarios
The evaluation of a function and the discovery of its minimum (using the function Rnk1min)

Fig. 6. The graphical representation of a scenario for the evaluation of a function and the
discovery of its minimum
Procedures used: vectors product, the determination of a superior triangle matrix.
The determination of the minimum of a function with a minimum number of searches
(using the function Praxis):

Fig. 7. The graphical representation of the scenario for the determination of the minimum of
a function with a minimum number of searches
The resolution
The value of the function for a minimum
A limited number of searching
Searching in a line mood
The resolution
The determination of a vector for a
function
The settling of the minimum, the determination of the
smallest integer number
The determination of a
matrix rank
The gradient of a function
calculated for the
minimum

Automation

276
Procedures used: initialization of the constants, initialization of the sub matrices, matrices
product, matrices rank.
The resolution[L1] of a non-linear system (using the function Marquardt)

Fig. 8. The graphical representation of the resolution of a non-linear system
Procedures used: the product of two vectors, the product of a vector and a matrix.
The determination of the solutions of a non-linear system (using the function Gssnewton)

Fig. 9. The graphical representation for the determination of the solutions of a non-linear
system
Procedures used: pivot method, the resolution of the systems through successive appeals,
the interchanging of lines and columns for the inversion of a matrix, the product of two
vectors, adding a multiple constant, systems with the same matrix coefficient, resolved
through successive appeals.
The resolution of the systems with differential equations (using the function Multistep)

Fig. 10. The graphical representation of the scenario for the resolution of systems with
differential equations
The resolution
The back differentiation method
The rapid method for systems of differential
equations with more variables
Data stocking
The resolution
Systems of derivative equations
The inverse of a matrix
The obtaining of a vector with unknown values
The resolution
The function applied
The choice of the correct answer
The inverse of a matrix


277
Procedures used: the product of two vectors, the product of a vector and a matrix, matrices
decomposition, particular cases: a smaller number of variables as against the number of
binary numbers in the number of representation, the resolution of the linear systems whose
matrix was distorted triangularly.
The resolution of the parabolic and hyperbolic equations (using the function Ark)

Fig. 11. The graphic representation of the scenario for the resolution of parabolic and
hyperbolic equations
Procedures used: the initialization of a vector after certain constants, vectors with multiple
setting forms, factors for multiplication, duplicate elements in a vector, the product of an
element belonging to a vector and other elements belonging to another vector, adding a
multiple constant from a vector to other elements from another vector being in a certain
state, the resolution of the systems of linear equations
The resolution of the systems with ordinary differential equations (using the function Efrk)

Fig. 12. The graphical representation of the scenario for the resolution of the systems with
ordinary differential equations
Procedures used: adding a multiple constant from a vector to an element belonging to
another vector which is in a certain state, matrices decomposition, permuting, resolution
linear systems.
Resolution the autonomous systems of differential equations (using the function Efsirk)
The resolution
Number of evaluations
The precision order of the method
The absolute and relative tolerance
The resolution
The evaluation of the changing of the procedure of derivates
The determination of the Boolean values
The calculus reduction through constraints

Automation

278

Fig. 13. The graphical representation of the scenario for the resolution of the autonomous
systems of differential equations
Procedures used: the product of an element of a vector and another element of another
vector, the product of a vector and a matrix, the product of a vector represented by the index
of a row and a matrix represented by the index of a column, matrices decomposition,
permuting, the use of the partial pivot, factors for multiplication, decomposition of the
linear system as the matrices type.
The resolution of the autonomous systems for differential equations (using the function
Liniger1vs)

systems for differential equations
Procedures used: the initialization of the matrices and constants, factor of multiplication,
replacement the elements of a matrix with a succession of elements in a rectangular matrix,
duplication of an element in a vector, the product of a vector and a matrix, adding a
multiple constant, the product of two vectors, matrix decomposition, permutation of
matrices, resolution the linear systems.
The resolution of the autonomous systems for differential equations (using the function
Liniger2)
The resolution
The determination of the system functioning matrix
The variables can be autonomic; they must not be used in
derivation
The reduction of the calculating time for the linear systems
The reduction of the calculating time for the non-linear systems
The resolution
Use the Newtonian scheme
Repetitions used in the Newtonian scheme for implicit
equations
The great number of integrations in Newtons procedure


279

systems for differential equations
Procedures used: the product of an element of a vector and another element of another
vector, the product of a vector and a matrix, the product of a matrix line and the column of
another matrix, matrices decomposition, permuting, the resolution of the linear systems.
The determination of the number of differential equations (using the function Gms)

Fig. 16. The graphical representation of the scenario for the determination of the number of
Procedures used: the product of a vector and another vector, the product of a line in a
matrix and a column in another matrix, adding a multiple constant, the resolution of the
linear systems, the replacement of a matrix column with constant complex elements and
factors of multiplication.
The determination of the number of differential equations (using the function Impex)

Fig. 17. The graphical representation of the scenario for the determination of the number of
The resolution
The elimination of some parameters
The resolution of the implicit equations using Newtons
procedure
The resolution
The determination of the asymptote
The number of integration steps
Solving nonlinear systems
The resolution Extrapolation
Stability Great precision
Uniformity
The global elimination of
errors
The elimination of
contrarydictions

Automation

280
Procedures used: the initializing of the matrices and constants, the initializing of a sub
matrix, the replacement of an element in a vector, the replacement of a column in a matrix
with complex constant elements, the product of an element of a vector and another element
of other vector, the product of a vector and a matrix, the product of a line in a matrix and the
column of another matrix, matrices decomposition, permuting, resolution the linear
systems.
Estimations on the unknown variables (using the function Peide)

Fig. 18. The graphical representation of the scenario for the estimations on the unknown
variables
Procedures used: initializing of the matrices and constants, initializing of a sub-matrix,
factor of multiplication, replacement of an element in a vector, replacement of a column in a
matrix with complex constant elements, the product of an element of a vector and another
element of another vector, the product of a vector and a matrix, the product of a line in a
matrix and the column of another matrix, matrix decomposition, permuting, resolution the
linear systems, the maximum number of repetitions to be made.
Calculating the polynomial coefficients (using the function Minmaxpol)

Fig. 19. The graphical representation of the scenario for the calculating polynomial
coefficients
Procedures used: adding a multiple constant from a vector into another vector, the duplicate
settling of an element of a vector, calculating the Newtonian polynomial coefficients
obtained through the interpolation of the corresponding coefficients, calculating the
polynomial sum, determination of the polynomial degree.
3.3 Tests realized on the basis of the scenario
I realized a series of tests on the basis of these scenarios. I calculated the sum, the
subtraction, the product, the division, the determinant, the inverse, the medium matrices
and the covariant, for matrices beginning with 2X2 up to 9X9, both with the Web Service
and without the Web Service.
The settling of the
unknown variables The resolution
The settling of the
Newtonian method of
determining polynomials
Polynomials determined
through interpolation
The determination of the maximum
number of repetitions
The resolution


281

*) Time expressed in micro seconds
*) WS Web Service
*) Con Executed in console
Table 4. The time obtained with testing
Due to the calculation of the inverse the execution time, beginning with 9x9, is bigger than
30 seconds.

Fig. 20. The graphical representation of the response time for tests, with and without WS
The advantages of the web service use with these scenarios:
a. the possibility of resolution problems of great dimensions which do not go in the client
computers memory
b. the identification of the common costs
c. the reduction of the costs
d. the reduction of the response time
e. the client can be involved with the problem description

Automation

282
4. A graphics generator for physics research
The proper architecture for the web service used for physics simulations is divided into a
working area. At first, we realized a series of Web services which grounds the numerical
calculation. These facts are oriented towards scientific applications, beginning with
elementary structures, up to complex ones, such as the integration of ordinary differential
equations. The passing from an interface of the type line command with the user, to an
interface based on Web services, represents a great advantage because it allows the
combination and use of certain procedures routines - defined earlier on.
The methods for numerical calculation are practical, efficient and elegant, and the
realization of a proper architecture of a Web service will significantly contribute to this
process [2].
Beginning here, I would like to develop this idea, applying it to both physics and physics
research. By selecting this, I chose the most representative phenomena in physics which
allow for applications; next, I followed the interpretation of them from the point of view of
numerical calculation.
The chosen phenomena are:
a. A motor with a continuous current;
b. The forces of inertia on the earths surface;
c. The dynamics of a variable mass point;
d. Admissible resistance;
e. Floating bodies.
In order to set out every phenomenon, they are described physically and are mentioned in
terms of the manner in which they are to be found within the created service. The platform
is divided into two parts: the former is designated to numerical calculation; the latter is for
the realization of graphics. Also, within this service, there is a static zone where I present the
selected physical phenomena, as can be seen in the following image (retrieved on
http://www.eliza-isbasoiu.muscel.ro 05/10/2009).

Fig. 21. WS Static content (link-uri)


283
The user is first welcomed by the image presented beneath: (retrieved on http://www.eliza-
isbasoiu.muscel.ro 05/10/2009)

Fig. 22. The starting image of the platform
In order to know the variables of the constants and equations at the basis of every
simulation, the user receives pieces of information about the physical phenomena and the
possible discussions after the simulation.
A motor with continuous current
One should notice that if we apply a tension to the motor terminals - given the same polarity
as where it works as a generator - the rotor spins in the opposite direction as that of the
generator (Maxwell, 1892).

Fig. 23. The graphical representation of the generator and the motor
The parameters that vary are: external voltage, ranging from a minimum value and a
maximum, R and r, and contrary[L2] electromotor tension. According to these values, the

Automation

284
graph appears. Such findings may be established for efficiency (retrieved on

Fig. 24. The graphical representation of the contrary[L3] electromotor tension
Forces of inertia on the earths surface
The Earth makes a revolutionary motion around the Sun, under the influence of the
gravitational forces of different cosmic bodies in our solar system.
We consider a material mass point m, being at rest against the Earth in an A point. This can
be seen by the following picture:

Fig. 25. The graphical representation of the material point as against the Earth


285
In the inertial system Si against the material points, thus act the gravitational forces of the
Sun, Moon and Earth.
The determination of inertial forces on the earths surface has been, for a long time, a very
attractive notion in physics.
Mass: the point of mass is variable. The angle is also variable. (retrieved on

Fig. 26. The graphical representation of the inertial forces on the earths surface
The dynamics of the variable mass points applications in the case of detachment.
We consider a material point M, with the mass m, whereby the variability of its mass is
given by the joining or detaching of some particles from M. These two situations are, in fact,
two cases: the joining case and the detaching case.
The variables in this situation are greater. The time ranges between a maximum and a
minimum. The matter the value it receives is v0. The angle, the reaction speed and the
gravitational acceleration influence the graph (retrieved on http://www.eliza-
isbasoiu.muscel.ro 05/10/2009).

Fig. 27. The graphical representation of a variable mass point speed, after an angle of
= 30
0

Automation

286
The floating bodies
Knowledge of the conditions which determine the stability of floating bodies is very
important in naval construction techniques. We assume a floating body which has a centre
in the same place with the O point, as with the figure below, and that over the superior part
of the body outside the liquid on which the F horizontal force acts, by tending to turn the
body towards the left (Balibar, 2007).

Fig. 28. The graphical representation of the centre of mass an inclined floating body
The factor which influences this phenomenon the most is represented by the surface marked
by S. After we introduced the minimum and maximum of the value, we obtained the
following graph (retrieved on http://www.eliza-isbasoiu.muscel.ro 05/10/2009).

Fig. 29. The graphical representation of the limits between a body can or cannot float for a
specific surface S


287
The graphical interface which the user employs is a simple one but extremely efficient,
because it is generalized for every area of study (retrieved on http://www.eliza-
isbasoiu.muscel.ro 05/10/2009).

Fig. 30. The image of the graphical interface of the Web service graphics generator
When we press the button generate graphs, the user is asked which essential parameter
they want as a graphical representation. For example, for control electromotor tension, there
are three variants: the exterior variation of U tension, the variable E contro electromotor
tension, the R variable.
The following image presents this fact (retrieved on http://www.eliza-isbasoiu.muscel.ro
05/10/2009).

Fig. 31. Select the variant for the graphic
The platform is divided into two working areas: the former is for numerical calculation; the
latter is for the graphs realization.

Automation

288
We can represent the sounds intensity; we introduce two values, the minimum and the
maximum, the representation being made after a desired pattern (retrieved on

Fig. 32. The graphical representation of the sounds intensity
If we modify the weight, then by increasing it simultaneously with the increasing of the
angle, the graph is as follows: (retrieved on http://www.eliza-isbasoiu.muscel.ro
05/10/2009)

Fig. 33. The graphical representation of the inertial force at the earth surface, for the angle
30

The above representations suggest the following conclusions:
a. the necessity of knowing of all of the equations, but also the particular cases which
represent the basis of the physical phenomena;


289
b. the possibility of interpretation, depending upon the input values;
c. simulations for finding other possible particular cases;
d. the understanding of the physical phenomena, at a practical level.
All these solutions are oriented towards scientific applications, which are grounded on
numerical calculations. They are followed by more complicated constructions.
There are many authors who deeply value such software architectures. Some of them
consider it to be the separate profession of the software engineer [5]. Others consider it be
independent, separate for the role of the software engineer and, thus, in need separate
learning approaches.
The activity can be considered from: psychological points of view; systematic points of view;
organizational points of view.
From the psychological point of view, the application is a creative process, which implies
knowledge within close fields, such as: software engineering, computer science, logic,
cognitive science and programming etc. From systematic point of view, the project is seen as
an activity which implies the discovery of optimal solutions for a set of problems, taking
into account the balance between obstacles and forces.
The organizational perspective offers the possibility that some software elements are re-used
for other products.
The main ideas which result from the above examples are as follows:
a. the necessity of knowing all of the equations and the particular cases which represent
the bases of the physical phenomena;
b. the possibility of their interpretation depending upon the input data;
c. simulations for finding other particular cases;
d. the practical understanding of the phenomena of physics;
All of these comparisons between values and interpretations, based upon the graphs,
depend upon users objectives. It is very important that a possible user of the application
can use the interface despite different levels of knowledge. The created Web service presents
a few physics applications based upon numerical calculations. The selected phenomena can
assure the construction of more complicated applications. This method displays ideal cases
which are not present in nature, but also specific cases. If one interprets them correctly, one
can find optimal solutions to a set of problems; besides this, there is the possibility of the
further use of the results obtained. The main advantages of using the Web service are: the
possibility of solving problems of great dimensions which exceed the computers memory;
the identification of common costs and their reduction; the reduction of the time response
value; the fact that client can be involved in the problems description.
The graphic must be defined within the text editor under the form equation
(formula,min,max,step,precision).
a. the formula must be x variable, framed by double inverted commas this x. The x
variable must have a space before it and another space after it in an equation, whenever
it appears. For example: "sin( x ) + cos( x + 4) + 2 * x "
b. min, max represent the interval margins where the calculation is made;

Automation

290
c. the step represents the incrementation value of the values for which the calculation was
made;
d. the precision represents the ranging interval of the graphical representation. It is
recommended to use 0.01; one can also use other values in order to see the results.
A few examples of graphs generated by this method are presented below: (retrieved on

Fig. 34. The movement of a projectile
Figure 32 presents the movement of a projectile according to the equations of the theory
depicted in the area the description of physical phenomenon (retrieved on

Fig. 35. Newtons Second Law
Figure 33 presents the Second Law of Newton, in conformity with the equations from the
theory depicted in the area related to the description of physical phenomena.


291
5. Conclusion
We presented in this article concepts connected to the architecture of distributed systems,
demonstrating the criteria for a proper choice in order to use adequate technology. The
problem that we developed was represented by the realization of certain Web Services using
methods of numeric calculation. The applications were created on the Eclipse platform;
these were realized with the solutions offered by the book Numerical Library in Java for
Scientists and Engineers. These solutions are oriented towards scientific applications based
upon numeric calculation. They begin with elementary structures move on to more complex
structures. By grouping and creating services for each and every structure, we realized that
many of the functions used for their construction appeal to other functions; we created a
table where those functions appear only once. We created a service for every function and
followed the response time. We saw that, depending upon the functions complexity, this
time differs. By way of comparison, we constructed a graphical representation with the time
for both individual functions, with simple input data, and parameters with complex values.
The next step would be to realize Web services for education in physics, as well as services
for scientific research in physics, using these services.
So long as informatics and communications technology evolve, the implications that they
will have on the educational system are difficult to be foresee. There will be always new
opportunities and new difficulties, but there will also be results and benefits.
Computer simulations seem to offer the most efficient methods for using computers in
physics. There, the processes used in research physics are encouraged: for determining the
cause of an effect, in prognosis, and also for the interpretation of research data. As a rule,
such simulations develop an inductive and deductive way of thinking, by assuring the
capacity of problem solving, the formulation of new hypotheses and the realization of tests.
We feel that users should be allowed to realize other experiments as a preliminary to the
physics laboratory.
Beginning with these elements and using the Web services for numerical calculation
subsequently realized, I selected a series of mathematical functions, namely those useful for
the equations used for continuous electricity, the calculation of inertial forces on the earths
surface, the calculation of the dynamics of variable weight points for finding admissible
resistance, and in the production and propagation of sounds.
6. References
Balibar F., (2007), Einstein, bucuria gndirii, pp. 95 -100, Publisher Univers, ISBN 978-1-
60257-025-2, Bucureti
Hang T.L., (2004), Numerical Library in Java for Scientists and Engineers, pp. 38 - 751,
ISBN-13: 978-1584884309, Publisher Chapman & Hall/CRC, USA
Isbasoiu E.C. (2009), Services for Numerical Calculation, pp. 245 - 252, CSSim - Conference
on Computational Intelligence, Modelling and Simulation, ISBN: 978-1-4244-5200-2,
Brno, Czech, 2009
Landau R.H. & Bordeianu C.C., (2008), Computational Physics. A Better Model for Physics
Education?, pp. 22 - 30, Computing in Science & Engineering, ISSN 1521-9615,
Oregon State University, USA, 2006

Automation

292
Maxwell J.C., (1892), A Treatise on Electricity and Magnetism, pp. 98 115, 3rd ed., vol. 2.
Publisher Oxford: Clarendon
Petcu D., (2008), Distributed Systems- Lecture notes, pp. 35 87, Publisher UVT, Retrieved
from http://web.info.uvt.ro/~petcu/distrib.htm, Timisoara
http://www.eliza-isbasoiu.muscel.ro retrieved on 05/10/2009
http://www.eliza-isbasoiu.muscel.ro retrieved on 21/10/2010
15
Genetic Algorithm Based Automation
Methods for Route Optimization Problems
G. Andal Jayalakshmi
Intel,
Malaysia
1. Introduction
Genetic Algorithms (GA) are robust search techniques that have emerged to be effective for
a variety of search and optimization problems. The primary goal of this chapter is to explore
various Genetic Algorithm (GA) based automation methods for solving route optimization
problems. Three real world problems: Traveling Salesman, Mobile Robot Path-Planning and
VLSI global routing are considered here for discussion. All the three problems are Non-
deterministic Polynomial (NP)-complete problems and require a heuristic algorithm to produce
acceptable solutions in a reasonable time.
The basic principles of GAs were first laid down by Holland. GAs work with a population of
individuals each representing a solution to the problem. The individuals are assigned a
fitness value based on the solutions quality and the highly fit individuals are given more
opportunities in the reproduction. The reproduction process generates the next generation
of individuals by two distinct processes named crossover and mutation. The new
individuals generated by crossover share some features from their parents and resemble the
current generation whereas the individuals generated by mutation produces new characters,
which are different from their parents. The probability of crossover operation is usually very
high compared to the probability of mutation operation due to the nature of their
operations. The reproduction process is carried out until the population is converged which
usually takes hundreds of iterations for complex real world problems. The time taken for
convergence is dependent on the problem and it is the progression towards increasing
uniformity among the individuals of a population.
A standard GA described by Goldberg uses binary encoding for representing the individuals,
one-point crossover for reproduction which exchanges two consecutive sets of genes and
random mutation which randomly alters the selected gene. The probability for applying
crossover typically ranges from 0.6 to 1.0 and the probability for mutation operation is
typically 0.001. Generally crossover enables the rapid exploration of the search space and
mutation provides a small amount of random search to ensure that no point in the search
space is given zero probability of being examined.
Traditional GAs are generally not the successful optimization algorithms for a particular
domain as they blindly try to optimize without applying the domain knowledge. L.Davis
states in the Handbook of Genetic Algorithms, that the Traditional genetic algorithms

Automation 294
although robust are generally not the most successful optimization algorithms on any
particular domain. Davis argues that the hybridization will result in superior methods.
Hybridizing the genetic algorithm with the optimization method for a particular problem
will result in a method which is better than the traditional GA and the particular
optimization method. In fact this will produce a more superior method than any of the
individual methods. The standard GA can be improved by introducing variations at every
level of the GA component including the encoding techniques, the reproduction
mechanisms, population initialization techniques, adaptation of genetic parameters and the
evolution of the individuals. These thoughts have resulted in a class of genetic algorithms
named Hybrid Genetic Algorithms (HGA). These are the customized genetic algorithms to fit
the traditional or simple GA to the problem rather than to fit the problem to the
requirements of a genetic algorithm. The HGAs use real valued encodings as opposed to
binary encodings and employs recombination operators that may be domain specific.
We will explore further on the use of HGA by discussing a solution to the Traveling Sales
Person (TSP) problem. The TSP problem has been a typical target for many approaches to
combinatorial optimization, including classical local optimization techniques as well as
many of the more recent variants on local optimization, such as Simulated Annealing, Tabu
Search, Neural Networks, and Genetic Algorithms. This problem is a classic example of Non
deterministic Polynomial hard problem and is therefore impossible to search for an optimal
solution for realistic sizes of N. The HGA that is described here is as proposed by Jayalakshmi
et. al. which combines a variant of an already existing crossover operator with a set of new
heuristics. One of the heuristics is for generating the initial population and the other two are
applied to the offspring either obtained by crossover or by shuffling. The heuristics applied
to the offspring are greedy in nature and hence the method includes proper amount of
randomness to prevent getting stuck up at local optimum.
While the hybrid GAs exploit the domain knowledge, in many realistic situations, a priori
knowledge of the problem may not be available. In such cases, it is fortunately possible to
dynamically adapt aspects of the genetic algorithms processing to anticipate the
environment and improve the solution quality. These are the Adaptive GAs which are
distinguished by their dynamic manipulation of selected parameters or operators during the
course of evolving a problem solution. In this chapter we will see an adaptive GA solution
to the mobile robot path planning problem which generates collision free paths for mobile
robots. The problem of generating collision-free paths has attracted considerable interest
during the past years. Recently a great deal of research has been done in the area of motion
planning for mobile robots as discussed by Choset et al. Traditional planners often assume that
the environment is perfectly known and search for the optimal path. On the other hand on-line
planners are often purely reactive and do not try to optimize a path. There are also approaches
combining offline planers with incremental map building to deal with a partially known
environment such that global planning is repeated whenever a new object is sensed and added
to the map. The developments in the field of Evolutionary Computation (EC) have inspired the
emergence of EC-based path planners. However traditional EC-based planners have not
incorporated the domain knowledge and were not adaptive and reactive to the changing
environments. Recent research has offered EC-based planners for dynamic environments.
The solution to the mobile robot path planning problem discussed here is as proposed by
Jayalakshmi et. al, which incorporates domain knowledge through domain specific operators

Genetic Algorithm Based Automation Methods for Route Optimization Problems 295
and uses an initialization heuristics to convert infeasible paths into feasible ones. The fitness
of the solution is measured based on the number of fragments, acute edges and the angle
between the turns in the path. The algorithm plans the path for the current environment and
the robot travels in that direction. If an obstacle is found in its path, the robot senses the
presence of the obstacle before the critical time to avoid collision and calls the path planner
algorithm again to find the new path from that point onwards.
The CHC genetic algorithm proposed by Eshelman has emerged as an alternative to resolve the
perennial problem with simple GAs which is the premature convergence. The simple GA
allows a sub-optimal individual to take over a population resulting in every individual being
extremely alike and thus causing premature convergence. The consequence of premature
convergence is a population which does not contain sufficient genetic diversity to evolve
further. The CHC genetic algorithm uses crossover using generational elitist selection,
heterogeneous recombination by incest prevention and cataclysmic mutation to restart the
search when the population starts to converge. The CHC GA has a very aggressive search by
using monotonic search through survival of the best and offset the aggressiveness of the search
by using highly disruptive operators such as uniform crossover.
In this chapter we will also explore a solution to the VLSI global routing problem using CHC
GA. One of the most important VLSI Design Approaches is the Macro Cell design. Macro
cells are large, irregularly sized parameterized circuit modules that are generated by a
silicon compiler as per a designers selected parameters. Usually the physical design process
for macro cells is divided into Floor Planning/Placement, Global Routing and Detailed
Routing. Floor Planning/Placement constructs a layout indicating the position of the macro
cells. The placement is then followed by routing, which is the process of determining the
connection pattern for each net to minimize the overall routing area. Before the global
routing process begins, a routing graph is extracted from the given placement and routing is
done based on this graph. Computing a global route for a net corresponds to finding a
corresponding path in the routing graph. Each edge represents a routing channel and the
vertex is the intersection of the two channels. First the vertices that represent the terminal of
the net are added to the routing graph and then the shortest route for the net is found. Both
the placement and routing problems are known to be NP-complete. Thus it is impossible to
find optimal solutions in practice and various heuristics are used to obtain a near optimal
solution. There has been a lot of work on optimization for routing, including Simulated
Annealing algorithms and Genetic Algorithms.
The details of the genetic algorithm solutions to each of these problems are described in the
following sections.
2. Design of a hybrid GA for TSP
A heuristic approach employs some domain knowledge in providing a solution to the
problem. A good heuristics can be devised provided one has the knowledge of the problem
being solved. In cases, where there is no knowledge of the problem, it is best to use a more
general heuristic, often called a meta-heuristic. Meta-heuristics are sometimes also called
black-box optimization algorithms or simply, general-purpose optimization algorithms.
Coding complex data structures by simple lists of bits or real values leads to the problem
that has no one-to-one correspondence between these lists and the problem instances. Hence

Automation 296
problem knowledge is necessary either to repair operators to deal with invalid solutions or
to design special operators tailored to the problem.
Of the present evolutionary algorithms, hybrid genetic algorithms have received increasing
attention and investigation in recent years. This is because of the reason that the hybrid GAs
combine the global explorative power of conventional GAs with the local exploitation
behaviours of deterministic optimization methods. The hybrid GAs usually outperform the
conventional GAs or deterministic methods in practice. To hybridize the genetic algorithm
technique and the current algorithm, the following three principles are suggested by Davis:
Use the current algorithms encoding technique in the hybrid algorithm. This
guarantees that the domain expertise embodied in the encoding used by the current
algorithm will be preserved.
Hybridize where possible by incorporating the positive features of the current method
in the hybrid algorithm.
Adapt the genetic operators by creating new crossover and mutation operators for the
new type of encoding by analogy with bit string crossover and mutation operators.
Incorporate domain based heuristics on operators as well.
Theoretical work as well as practical experience demonstrates the importance to progress
from fixed, rigid schemes of genetic algorithms towards a problem specific processing of
optimization problems.
This section explores how a HGA is used to solve the TSP problem. The TSP is probably the
most studied optimization problems of all times. In the Travelling Sales Person problem,
given a set {c
1
, c
2
, . . . , c
n
} of cities, the goal is to find an ordering of the cities that
minimizes the quantity
d(c
(i)
,c
(i+1)
)+d(c
(n)
,c
(1)
) 1 i n-1
Where d (c
i
,c
j
) is the distance associated with each pair of distinct cities <c
i
,c
j
>. This quantity
is referred to as the tour_length, since it is the length of the tour a salesman would make
when visiting the cities in the order specified by the permutation, returning at the end, to
the initial city. The Euclidean Travelling Sales Person problem involves finding the shortest
Hamiltonian Path or Cycle in a graph of N cities. The distance between the two cities is just
the Euclidean distance between them.
In a symmetric TSP, the distances satisfy d(c
i
,c
j
) = d(c
j
,c
i
) for 1i, jN. The symmetric traveling
salesman problem has many applications, including VLSI chip fabrication X-ray
crystallography and many other. It is NP-hard and so any algorithm for finding optimal
tours must have a worst-case running time that grows faster than any polynomial. This
leaves researchers with two alternatives: either look for heuristics that merely find near-
optimal tours, but do so quickly, or attempt to develop optimization algorithms that work
well on real-world rather than worst-case instances. Because of its simplicity and
applicability the TSP has for decades served as an initial proving ground for new ideas
related to both these alternatives.
2.1 The hybrid GA solution
The HGA proposed by Jayalakshmi et al. to solve the TSP problem use heuristics for
initialization of population and improvement of offspring produced by crossover. The

Initialization Heuristics algorithm is used to initialize a part of the population and the
remaining part of the population is initialized randomly. The offspring is obtained by
crossover between two parents selected randomly. The tour improvement heuristics:
RemoveSharp and LocalOpt are used to bring the offspring to a local minimum. If cost of the
tour of the offspring thus obtained is less than the cost of the tour of any one of the parents
then the parent with higher cost is removed from the population and the offspring is added
to the population. If the cost of the tour of the offspring is greater than that of both of its
parents then it is discarded. For shuffling, a random number is generated within one and if
it is less than the specified probability of the shuffling operator, a tour is randomly selected
and is removed from the population. Its sequence is randomized and then added to the
population.
2.1.1 The crossover operator
The initial city is chosen from one of the two parent tours. This is the current city and all the
occurrences of this city are removed from the edge map. If the current city has entries in its
edgelist then the city with the shortest edge is included in the tour, and this becomes the
current city. Any ties are broken randomly. This is repeated until there are no remaining
cities. An example is given below:
Let the distance matrix be

0 10 4 15 5 20
10 0 5 25 5 10
4 5 0 13 6 2
15 25 13 0 6 10
5 5 6 6 0 20
20 10 2 10 20 0
Let the genotype p
1
be equal to (2,3,4,5,0,1) which encodes the TSP tour (2,3,4,5,0,1,2) and p
2

be equal to (2,3,1,4,0,5) which encodes the TSP tour(2,3,1,4,0,5,2). The combined edge map
M
12
contains the combined edge relationships from both the parents. The first gene value in
p1 i.e. 2 is added to the child c1. Then the gene value 2 is removed from the edge map. The
combined edge map before and after are given below:

Gene
value
Edge
map
(p
1
)
Edge
map
(p
2
)
Combined
Edge map
M
12
(before)
Combined
Edge map
M
12
(after)

0 5,1 4,5 1,4,5 1,4,5
1 0,2 3,4 0,2,3,4 0,3,4
2 3,1 3,5 1,3,5 1,3,5
3 2,4 2,1 1,2,4 1,4
4 3,5 1,0 0,1,3,5 0,1,3,5
5 4,0 0,2 0,2,4 0,4
Table 1. Combined edge map

Automation 298
Now |E(2)| = 3 therefore an edge j is chosen such that j E(2) and |<2,j,>| is minimum and
is added to the child c
1
. In this example j is 5. Now j is removed from the edge map, and the
same procedure is followed until the child c
1
is filled with all the genes. For the example the
child c
1
will become (2,5,0,4,1,3).
2.1.2 The Initialization Heuristics
The Initialization Heuristics (IH) initializes the population based on a greedy algorithm which
arranges the cities depending on their x and y coordinates. The tours are represented in linked-
lists. First an initial list is obtained in the input order which is the Input List. The linked-list
that is obtained after applying the Initialization Heuristics is the Output List. During the
process of applying the Initialization Heuristics all the cities in the Input List will be moved
one by one to the Output List. Four cities are selected, first one with largest x-coordinate
value, second one with least x-coordinate value, third one with largest y-coordinate and fourth
one with least y-coordinate value. These are moved from the Input List to the Output List.
The sequence of the four cities in the Output List is changed based on minimum cost. The
elements in the Input List are randomized and the head element is inserted into the Output
List at a position where the increase in the cost of the tour is minimal. This process is repeated
until all the elements in the Input List are moved to the Output List.
Figure 1(a) shows a 8-city problem. Figure 1(b) shows the Boundary Tour formed from four
extreme cities. Figure 2 (a), (b), (c) & (d) shows the four possible tours that can be formed
when city 'E' is moved to the Output List. It is obvious from the figures that the Tour in
Figure 2(a) will result in minimum increase in the cost of the tour in the Output List.
Similarly other cities will be moved one by one to the Output List.

Fig. 1. (a) Input cities and (b) boundary tour formed by four extreme cities

(a) (b) (c) (d)
Fig. 2. Various possible tours, which can be formed by moving city E to the output list

2.1.3 The Removesharp heuristics
The RemoveSharp algorithm removes sharp increase in the tour cost due to a city, which is
badly positioned. It rearranges the sequence of a tour by considering the nearest cities of a
badly positioned city such that the tour_cost is reduced. A list containing the nearest m cities
to a selected city is created. The selected city from the tour is removed and a tour with N-1
cities is formed. Now the selected city is reinserted in the tour either before or after any one
of the cities in the list previously constructed with m nearest cities and the cost of the new
tour length is calculated for each case. The sequence, which produces the least cost, is
selected. This is repeated for each city in the tour.
An example is given below:

(a) (b)
Fig. 3. (a) A tour with a badly positioned city and (b) The tour after RemoveSharp is applied
In Figure 3(a) the city E is in between the cities A and D, while it is obvious that the nearest
cities to it are city C and B. RemoveSharp will move city E between the cities C and D,
resulting in a decrease in the tour cost as shown in Figure 3(b).
2.1.4 The local heuristics
The heuristics finds a locally optimal tour for a set of cities, by rearranging them in all
possible orders. The LocalOpt algorithm will select q consecutive cities (S
p+0
, S
p+1
, . . . . . , S
p+q-
1
) from the tour and it arranges cities S
p+1
, S
p+2
, . . . . , S
p+q-2
in such a way that the distance is
minimum between the cities S
p+0
and S
p+q-1
by searching all possible arrangements. The
value of p varies from 0 to n-q, where n is the number of cities. In Figure 4(a) it is quite clear
that the distance between the cities A and G can be reduced if some rearrangements are
made in the sequence of the cities between them. LocalOpt will make all possible
rearrangements and replace them to the sequence as shown in Figure 4(b).

(a) (b)
Fig. 4. (a) A bad tour and (b) The tour after LocalOpt is applied

Automation 300
2.2 Results
The results for the HGA solution for 3 standard TSP problems are compared with the results
for GA and SA solutions. The best integer tour_length and the best real tour_length (in
parenthesis) are tabulated below in Table 2. NA represents "Not Available".
The heuristics with which Hybrid GA is compared here are GA and SA as reported by
Whitley. D et al. and TSPLIB. The difference between integer and real tour length is that in
the first case distances are measured by integer numbers, while in the second case by
floating point approximations of real numbers. In TSPLIB website only Eil51 and Eil76 are
available which have an additional city to Eil50 and Eil75 respectively.

Problem name HGA GA SA
Eil50
50-city problem
426
(428.871)
{for Eil51}
428
(NA)

443
(NA)

Eil75
75-city problem
538
(544.36)
{for Eil76}
545
(NA)

580
(NA)

KroA100
100-city problem
21282
(21285.44)
21761
(NA)
NA
(NA)
Table 2. Comparison of HGA with other heuristics on geometric instances of the symmetric
TSP.
3. Design of an adaptive GA for mobile robots
Adaptive Genetic algorithms dynamically manipulate selected parameters or operators
during the course of evolving a problem solution. Adaptive GAs are advantageous over
SGAs in that they are more reactive to unanticipated characteristics of the problem and can
dynamically acquire information about the problem characteristics and exploit them. As
described by Davis, adaptive GAs can be categorized based on the level at which the
adaptive parameters operate. Population-level techniques dynamically adjust parameters that
are global to the entire population. Individual-level adaptive methods modify a particular
individual within the population depending on how it is affected by the mutation operators.
Component-level adaptive GAs dynamically alter the individual components depending on
how each individual will be manipulated independently from each other. The operator
probabilities play a major role in determining the solution quality and the convergence rate.
Since the range of potential applications of genetic algorithms is infinite, it is difficult to
measure the goodness of the parameter values. This suggested the idea of adapting the
operator probabilities during the evolution of the GA.
The path planner proposed by Jayalakshmi et al. incorporates domain knowledge in the
algorithm through domain specific operators. An initialization heuristics is used to convert
infeasible paths into feasible ones. The fitness is measured based on the number of
fragments, acute edges and the angle between the turns in the path. The algorithm plans the
path for the current environment and the robot travels in that direction. If an obstacle is
found in its path, the robot senses the presence of the obstacle before the critical time to
avoid collision and calls the path planner algorithm again to find the new path from that
point onwards. The following sections describe the solution.

3.1 The adaptive GA solution
The path planner algorithm design has four major phases. The first phase is the design of
the Initialization Heuristics (IH), which includes the Backtrack and Change_Y operators. The
algorithm initializes the population randomly and then repairs the population by applying
these two operators. This leaves the initial population free of any infeasible solutions and
reduces the search region considerably. The heuristics is discussed in detail in a later
section.
The second phase is the design of domain specific genetic operators End_New, Mid_New,
Shake and Adjacent to change the characteristics of the path. The operators tune the path
generated by removing the sharp edges, inserting adjacent segments in the path and
introducing new vertices. The operators are discussed in detail in section 3.1.4.
The third phase is the design of the objective function which is designed to include the
smoothness factors of the path in calculating the fitness value. This ensures that the path is
not only optimal in length but also smooth without any sharp turns. The objective function
is discussed in detail in section 3.1.5. The fourth phase of the algorithm is the design of the
adaptive rules to evolve the operator probabilities. The operator probabilities are adapted
based on the smoothness factors. The adaptive rules are discussed in section 3.1.6. The
binary tournament selection scheme is used to select a parent chromosome and the
reproduction is carried out by refining the paths in the previous generation using the
domain specific genetic operators described in section 3.1.4. The complete algorithm is given
below:
Adaptive_Path_Planner()
Begin
Initialize the population using the heuristics
Calculate the objective function and evaluate the population
While not convergence do
Begin
Repeat
Select the parent using binary tournament selection
Apply the domain specific genetic operators with the optimal
probability and produce offspring
Replace the parent with the offspring
Until the next generation is filled
Evaluate the population based on the objective function
Tune the operator probability
End
End

Automation 302
3.1.1 The chromosome structure
A chromosome in a population represents a feasible/infeasible path for the robot to reach
the goal location. It consists of the starting point followed by the intersection points of the
line segments of the path, and finally the goal position or the ending point. A path can have
a varied number of intermediate nodes. And hence the length of the chromosomes will be a
variable. An initial population of chromosomes is randomly generated such that it has a
random number of intermediate edges with random coordinates. A sample path and its
representation are given below:

Fig. 5. Example path and its equivalent chromosome
3.1.2 The selection scheme
The binary tournament selection scheme is used to select a parent chromosome. In binary
tournament selection, pairs of individuals are picked at random from the population;
whichever has the higher fitness is copied into a mating pool. This is repeated until the
mating pool is full. In this, the better individual wins the tournament with probability p,
where 0.5 < p < 1. By adjusting tournament size or win probability, the selection pressure
can be made arbitrarily large or small.
3.1.3 The initialization heuristics
Each chromosome shall represent a feasible or an infeasible path. A random initialization
generally leads to a large number of infeasible paths and hence a heuristics is used to
convert infeasible paths to feasible ones. The heuristics involve two operators: Change_Y and
Backtrack. The Change_y operator is used to change the value of Y coordinate of the vertex
which when included leads to an infeasible path. This makes the robot go up a step to take
a different path so that the collision with the obstacle is avoided. The Backtrack operator is
used to take a different path when the path already taken by the robot leads to an obstacle. It
allows the robot to go back by two steps in the original path and take up a new path. It is
designed to go back two steps because when the robot realizes an obstacle, it will be very
nearer to the obstacle and going back one step may not lead to a feasible path.
3.1.4 The genetic operators
The traditional genetic operators Mutation and Crossover cannot be used as such here,
hence they are given new forms to accommodate the requirements of the problem to be
solved. The operators are designed having in mind the nature inspired actions, a person
shall take to avoid collisions with obstacles. The operators are described below:

Mid_New: A vertex is chosen randomly and the edges connecting it to the previous and the
next vertices are altered so that any steep increase in the path can be eliminated. The mid
points of the edges are considered recursively until a feasible path is found. This helps to
take up a closer but safer path around the obstacle
End_New: A vertex is chosen randomly and is removed from the path provided the resultant
path is feasible. This operator helps to reduce the number of fragments in a path.
Shake: This operator chooses a vertex randomly and changes its Y coordinate by either
adding or subtracting a constant value. This works like a mutation operator.
Adjacent: This operator changes the original path by interchanging a segment of a path with
a parallel segment by either adding or subtracting a constant value with the Y coordinate of
each vertex in that segment.
3.1.5 The evaluation function
The fitness of a chromosome is calculated based on the following factors:
Length
Feasibility
Number of acute edges
Number of bends
Number of fragments
Two objective functions are designed; one based on length and the other on smoothness.
The objective function Simple_Obj calculates the fitness of a path on the basis of the length of
the path and its feasibility. The objective function Smooth_Obj calculates the fitness based on
the smoothness of the path. The objective functions are described in the following sections.
3.1.5.1 The objective function based on length and feasibility
The fitness is calculated based on the length and feasibility. The feasibility is measured by
checking whether the next node on the path generated so far leads to a collision with any of
the obstacles. The length_constant is a large integer value which when divides the path
length will give high fitness value for the path with minimal length.
Fitncss F(potb) = (
Kl
totol lcngtb (potb)
) + K Fcosibility(potb)
Where K
l
is the length_constant (a large integer value) and K
f
is the feasibility_constant
(an integer value).
total_length(path) = length( i , i + 1) , 1 i n, n = No. of noues
cosibility(potb) = _
1, or o cosiblc potb
-1 , or on incosiblc potb

3.1.5.2 The objective function based on smoothness
The objective function Simple_Obj discussed above takes into account only the path length
and the feasibility factor whereas the Smooth_Obj takes into account the other factors such as

Automation 304
the number of acute increases/decreases, turns with angle 90 and the number of fragments
in the path. The new objective function Smooth_Obj is given below:
Fitncss F(potb) = _
Kl
totol_lcngtb(potb)
] + Ks smootbncss(potb)
Smootbncss(potb) = _
Su SA
1uu
] + _
4u SB
1uu
] + _
1u SF
1uu
]
where
SA = 1 -
Acutc_Count 1uu
NF

SB = 1 -
BcnJ_Count 1uu
NF

SF =
IJcol Frogcmcnts 1uu
NF

And NF

is the number of Fragments
The Acute_Count is the number of sharp increases / decreases in the path, the Bend_Count is
the number of rectangular turns and the Ideal_Fragments is the minimum number of
fragments of all the paths that occur in a generation. Since the number of acute edges affects
the smoothness of the path largely, its contribution is 50% in the calculation of the fitness
value. The number of bends is given 40% share and the fragments 10% share in the
calculation of the fitness value.
3.2 The dynamic environment
The robot travels in an environment where the obstacles may get introduced suddenly in the
path planned by the robot for travelling. In such situations the robot has to decide at every
step whether to take up the already planned path or a new path. The robot is assumed to
travel at a fixed speed and has a sensor in it to detect the existence of any obstacle. When an
obstacle is recognized by the robot it calls the dynamic path planner algorithm and plans its
path from that position onwards. The sensor is assumed to sense the existence of any
obstacle before the robot reaches the region and the path planner returns the path within the
critical time. Now the new path is taken up from that position and the same procedure is
repeated until the robot has reached the destination.
3.3 Results
A sample output for the adaptive GA solution is given below. The Robot is assumed to
travel in a dynamic environment of dimension (5,5) to (400,400) with different kinds of
obstacles placed randomly. A dynamic environment is created by adding new obstacles on
the path planned by the robot.
The path planning algorithms for dynamic environments are computationally intensive and
hence will take longer time to converge for an environment with non-rectilinear obstacles.

Faster genetic operators and multi-threading methods might help to speed up the path
planning process in these environments.

(a) (b)
Fig. 6. (a) A dynamic obstacle in the planned path and (b) the final generation of paths from
the point of intervention
4. Design of a Hybrid CHC GA for VLSI routing
The Minimum Rectilinear Steiner Tree (MRST) problem arises in VLSI global routing and
wiring estimation where we seek low-cost topologies to connect the pins of signal nets. The
Steiner tree algorithm is the essential part of a global routing algorithm. It has been an active
field of research in the recent past. This section presents a Hybrid CHC (HCHC) genetic
algorithm for global routing. The Minimum Rectilinear Steiner Tree problem is to construct a
tree that connects all the n points given in the Euclidean plane. If the edges in this tree are to
be selected from all possible edges that are from the complete graph on the points, it is the
familiar problem of finding a spanning tree in an undirected graph. If the edges of the tree
must be horizontal and vertical, the additional points where the edges meet are called the
Steiner points, and the resulting tree is a Rectilinear Steiner Tree [RST]. A shortest such tree
on a set of given points is a minimum rectilinear Steiner tree.
4.1 Construction of Minimum Spanning Tree
The Spanning tree algorithm presented here is based on the shortest path heuristic as
described by Ellis Horowitz et al. A simple genetic algorithm is used for the construction of a
Minimum Spanning Tree (MST), which is then used in the generation of the Steiner minimal
tree. The spanning tree is generated by initializing the population with random solutions.
The random solutions are then repaired using a repair heuristics. The offspring are
generated by applying one point crossover and exchange mutation which exchanges edges
in an individual. The exchange of edges may lead to a totally different tree, thus justifying
the purpose of mutation. The new population is evolved and the same procedure is repeated
until convergence.
The algorithm for the construction of the minimum spanning tree is given below.

Automation 306
Minimal_Spanning_Tree()
Begin
Initialize parameters: generation count, crossover and mutation probabilities
Initialize parent population randomly
Apply repair heuristics to the parent population
While termination condition not reached
Begin
Select parents based on the total length of the Spanning tree
Apply crossover and mutation
Evolve new population
Replace previous population by new population
End
End
The repair heuristics removes cycles and repeated edges from the population and makes it a
set of feasible solutions. The vertices of the graph are stored in separate sets, so that it can be
later combined whenever an edge is included in the final spanning tree. The union algorithm
unions two sets containing V
i
and V
j
respectively when an edge E <V
i
, V
j
> is added to the
final spanning tree. The find algorithm verifies whether a particular vertex belongs to a set.
The union algorithm combines all vertices that are connected, in to a single set. When a
particular edge is selected for addition into a partially constructed spanning tree, it is
checked whether the vertices of that edge are already present in the same set using the find
algorithm. If they are in the same set, then the inclusion of this edge will lead to a cycle. The
repeated edges can be checked easily with the adjacency matrix.
4.2 Construction of minimum steiner tree
The Steiner tree problem can be defined as the subset of minimum spanning tree problem.
In minimum spanning tree construction, a tree is constructed with vertices V
1
,V
2
,V
n

connected without loops at the lowest cost. In the Steiner tree problem, extra vertices are
added besides the existing vertices V
1
,V
2
,V
n
, to construct a lower cost tree connecting
V
1
,V
2
,.V
n
. The extra vertices are called the Steiner points. There are various heuristics
available to construct a MRST, and most of them use MST as a starting point. The I-Steiner
algorithm as discussed by Kahng A.B et al., constructs the MRST by evaluating all possible
Steiner points for their impact on MST cost. The algorithm operates on a series of passes, in
each pass the single Steiner point which provides the greatest improvement in spanning tree
cost is selected and added to the set of demand points. Points are added until no further
improvement can be obtained.
The heuristics used here for the construction of MRST is the BOI or edge-removal
technique of Borah, Owen and Irwin. The algorithm constructs the Steiner tree through
repeated modification of an initial spanning tree as discussed by Jayalakshmi et al. An edge

and a vertex pair that are close to each other in an MST are determined. For each vertex V
i
,
edge E
i
pairing of the spanning tree, an optimal Steiner point is found to merge the
endpoints of the edge E
j
with vertex V
i
. This will create a cycle, so the longest edge on this
cycle is found and a decision is made about removing this edge from the cycle based on the
cost. Among all possible eliminations, whichever leads to the lowest cost is removed and
the tree is modified. The edges are removed and new connections are inserted until no
improvements can be obtained. The resulting tree is the minimum Steiner tree. The
approach has low complexity with performance comparable to that of I-Steiner. The
algorithm for the construction of the minimum Steiner tree is given below:
Steiner_Tree()
Begin
Build the routing graph G
For each net do
Begin
Initialize weights for edges
Find the minimum cost spanning tree T
For each <vertex,edge> pair of the spanning tree
Begin
Find the optimum Steiner point to connect this edge to the vertex
at a suitable point
Find the longest edge on the generated cycle
Compute the cost of the modified tree and store the pair in a list
if the cost is less than the MST
End
While the list is not empty do
Begin
Remove the pair from the list which results in lowest cost
Re-compute the longest edge on the cycle and the cost of the tree
If the edges to be replaced are in the tree and the cost is less then
modify the tree
End
End
End

Automation 308
4.3 Results
The HCHC solution for four standard test problems B1, B3 B6 and B9 from the problem sets
of J.E.Beasley are given below in Table 3. A simple GA with one point crossover and
exchange mutation is compared with the HCHC solution. In HCHC, Uniform crossover and
External mutation are used for reproduction.

Test
problem
Optimum
Solution Error Generations
SGA HCHC SGA HCHC SGA HCHC
B1 82 187 95 105 13 150 200
B3 138 145 140 7 2 150 200
B6 122 128 125 6 3 400 250
B9 220 241 224 21 4 150 200
Table 3. The solutions obtained by SGA and HCHC for Beasleys test problems B1, B3,B6
and B9
For HCHC algorithm the maximum error is for the test problem B1 and for the rest of the
problems the error is less than 6. And SGA has performed very poorly for B1 and B9. For the
other problems, SGA has performed moderately well with error less than 10.
5. Summary
With Genetic Algorithms emerging as strong alternative to traditional optimization
algorithms, in a wide variety of application areas, it is important to find the factors that
influence the efficiency of the genetic algorithms. The simple GAs are found to be ineffective
for most of the real world problems. Hence there arises the need for the customization of the
traditional GAs. This chapter explored the variants of the simple genetic algorithm and their
application to solve real world problems. TSP is a problem of a specific domain and
required hybridization for quicker convergence. In particular the local search algorithm
chosen has a determining influence on the final performance. The heuristics used were
simple and easy to implement when compared to other algorithms. The solution to the
mobile robot path planning problem explored the design of different operators and showed
that the adaptation of operators has a significant impact in improving the solution quality. A
hybrid CHC algorithm was used to solve the VLSI global routing problem. And this
example showed that the simple GA could only find a sub optimal solution and could not
go beyond certain values due to the lack of techniques that avoid premature convergence.
6. References
Beasley.J.E. (1989). An SST-Based Algorithm for the Steiner Problem in Graphs, Networks,
Vol.19, pp.1-16, l989.
Borah.M, R. M. Owens & M. J. Irwin. (1994). An edge-based heuristic for Steiner routing,
IEEE Trans. Computer-Aided Design, vol. 13, pp. 15631568, Dec. 1994.

Choset, H., Lynch, K. M., Hutchinson, S., Kantor, G., Burgard, W., & Kavraki, L. E. (2005).
Principles of robot motion: theory, algorithms, and implementations. Boston, MIT
Press.
Davis.L, (Editor). (1991). Handbook of Genetic Algorithms , Van Nostrand Reinhold.
Ellis Horowitz, Sartaj Sahni & Sanguthevar Rajasekaran. (2007). Computer Algorithms, Silicon
Pr.
Eshelman,L.J. (1991). The CHC Adaptive Search Algorithm: How to have safe search when
engaging in nontraditional genetic recombination, Rawlins G.(Editor), Foundations of
Genetic Algorithms, Morgan Kaufmann, pp.265-283.
Goldberg, David E. (1989). Genetic Algorithms in Search Optimization and Machine Learning,
Addison Wesley, ISBN 0201157675.
Goldberg, David E. (2002). The Design of Innovation: Lessons from and for Competent Genetic
Algorithms, Addison-Wesley, Reading, MA.
Holland, John H. (1975). Adaptation in Natural and Artificial Systems, University of Michigan
Press, Ann Arbor.
Hu.J & Sapatnekar.S.S. (2001). Performance driven global routing through gradual
refinement, IEEE Int. Conf.on Comp Design, 2001, pp.481-483
Jayalakshmi.G.A, S. Sathiamoorthy & R. Rajaram. (2001). A Hybrid Genetic Algorithm A
New Approach to Solve Traveling Salesman Problem, International Journal of
Computational Engineering Science, Volume: 2 Issue: 2 p.339 355.
Jayalakshmi, G.A, Prabhu. H, & Rajaram. R. (2003). An Adaptive Mobile Robot Path Planner
For Dynamic Environments With Arbitrary-Shaped Obstacles, International Journal
of Computational Engineering Science(2003),pp. 67-84
Jayalakshmi, G.A., Sowmyalakshmi.S & Rajaram.R (2003). A Hybrid CHC Genetic
Algorithm For Macro Cell Global Routing, Advances in Soft Computing Engineering
Design and Manufacturing Benitez, J.M.; Cordon, O.; Hoffmann, F.; Roy, R. (Eds.)
pp. 343 350, Springer-Verlag, Aug 2003
Kahng.A.B and G.Robins. (1992). A new class of iterative Steiner tree heuristics with good
performance, IEEE Trans on Computer Aided design of Integrated circuits and
systems, 11(7), pp.893-902.
Li-Ying Wang, Jie Zhang & Hua Li. (2007). An improved Genetic Algorithm for TSP,
Proceedings of the Sixth International Conference on Machine Learning and Cybernetics,
Hong Kong, 19-22 August 2007.
TSPLIB: http://elib.zib.de/pub/mp-testdata/tsp/tsplib/tsplib.html retrievable as of 3
rd

Feb 2012
Whitley.D, Lunacek.M & Sokolov.A. (2006). Comparing the Niches of CMA-ES, CHC and
Pattern Search Using Diverse Benchmarks, Parallel Problem Solving from Nature
Conference (PPSN 2006), Springer.
Whitley.D, T. Starkweather & D. Shaner .(1991). The Traveling Salesman and Sequence
Scheduling: Quality Solutions Using Genetic Edge Recombination. The
Handbook of Genetic Algorithms. L. Davis, ed., pp: 350-372. Van Nostrand
Reinhold.

Automation 310
Zhou.H. (2004). Efficient Steiner Tree Construction Based on Spanning Graphs, IEEE
Transactions on Computer-Aided Design of Integrated Circuits And Systems, Vol.
23, No. 5, May 2004, pp:704-710.
0
Automatic Control of the Software Development
Process with Regard to Risk Analysis
Marian Jureczko
Wrocaw University of Technology
Poland
1. Introduction
In this paper, we present a method for analysing risk in software development. Before
describing the details, let us dene the basic terms that are important for the further
discussion. Among them, the most import is risk. In the context of software engineering
it represents any potential situation or event that can negatively affect software project in
the future and may arise from some present action. Software project risks usually affect
scheduling, cost and product quality. Further details are given in Section 2.2. The risk
will be discussed in the context of project requirements and releases. The requirements (or
requirement model) represent requests from the customer that should be transformed into
features and implemented in the software product. The releases refer to a phase in the
software development life cycle. It is a term used to denote that the software product is
ready for or has been delivered to the customer. This work takes into consideration agile
development (i.e. processes with short iterations and many releases) therefore, it should be
emphasized that some of the releases may be internal. These releases are not scheduled to be
delivered to the customer; nevertheless they are potentially shippable.
Considerable research has been done on controlling the software development process. The
mainstream regards process improvement approaches (e.g. CMMI SEI (2002)) where time
consuming activities are performed in order to monitor and control the work progress. In
effect the maturity is assessed. Furthermore, mature processes indicate lower project failure
risk.
The agile software development shifts the focus toward product Beck & Andres (2005).
Moreover, it is recommended to ensure high level of test coverage. The tests are automated
Jureczko & Mynarski (2010), and therefore it is possible to execute them frequently in a
continuous integration system. Despite of the fact that the agile approach is more automated,
risk assessment may be challenging. There are few risk related metrics right out of the
box. Fortunately, due to the high level of automation considerable amount of data regarding
quality and risk might be collected without laborious activities Stamelos & Sfetsos (2007). It
is possible to build simple metrics tools (e.g. BugInfo Jureczko (Retrived on 01/11/2011a),
Ckjm Jureczko & Spinellis (Retrived on 06/05/2011)) as well as a complex solution for
identifying architectural smells (Sotograph Hello2morrow Inc. (Retrived on 01/11/2011)
into the continuous integration system. Unfortunately, it is not evident which metrics should
16
be collected and what analysis should be performed to accomplish the risk related goals. This
chapter discusses which data should be collected and howit should be done in order to obtain
accurate risk estimations with smallest effort. It will be considered how to mitigate the effort
by using automation. We rst suggest a risk analysis method that reects the common trends
in this area (however the focus is slightly moved toward agile software development) and then
we decompose the method to wellknown in software engineering problems. The potential
solutions to the aforementioned problems are reviewed with regard to automation in order
to identify improvement possibilities. In result, a direction for further research is dened.
Moreover, the main value of this chapter is the review of the status and recent development
of risk related issues with regard to automation.
The remainder of the chapter is organized as follows: in the next section an approach
to risk analysis is suggested and decomposed into three problems, i.e. effort estimation,
defect prediction and maintainability measurement. The next three sections survey the three
aforementioned issues. The sixth section discusses the automation possibilities. Finally, the
seventh section presents concluding remarks and future research direction.
2. Problem formulation
There are tools that support the risk analysis. Unfortunately, such tools usually do not
offer automation, but require expert assistance, e.g. RiskGuide Grski & Miler (Retrived on
01/11/2011) or are proprietary black box approaches, e.g. SoftwareCockpit Capgemini CSD
Research Inc. (Retrived on 01/11/2011). The users do not know what exactly is going on
inside and the software projects may differ from each other Jureczko & Madeyski (2010);
Zimmermann et al. (2009). In order to make good risk related decisions, the inference
mechanism should be transparent and understandable.
2.1 Requirement model
The requirement model that incorporates both functional and non-functional features might
lay foundations for a method of risk assessment. Each and every requirement is characterized
by its specication and so-called time to delivery. Since most of the real world projects
undergo not only new feature development, but also maintenance, defect repairs are
considered requisite and hence also belong to the requirement model.
In case of any functional requirement, it may or may not be met in the nal product, there
are only two satisfaction levels: satised or not satised. In the case of nonfunctional
requirements the satisfaction level might be specied on the ordinal or interval scale
depending upon the requirement kind.
Failing to deliver some requirement in due time usually causes severe nancial losses. When
a functional requirement is missing, a penalty is a function of delivery time, which in turn is
a constant dened in the customer software supplier agreement. On the other hand, when a
nonfunctional requirement is not satised, a penalty is a function of both: delivery time and
satisfaction level. These values should be dened in the very same agreement or evaluated
using empirical data from similar historical projects.
312 Automation
Automatic Control of the Software Development Process with Regard to Risk Analysis 3
Among the nonfunctional requirements two are especially interesting: quality and
maintainability. Both are very difcult to measure Heitlager et al. (2007); Kan (1994) and
both may cause severe loss in future. Let us consider a release of a software system with
low level of quality and maintainability. There are many defects since the quality is low.
The development of the next release will generate extra cost because the defects should
be removed. Furthermore, the removal process might be very time consuming since the
maintainability is low. There is also strong possibility that developing new features will
be extremely expensive due to maintainability problems. Therefore, quality as well as
maintainability should be considered during risk analysis.
2.2 The risk
The risk Rptq specied over a set of possible threats (t represents a threat) is dened as follows:
Rptq Pptq Sptq (1)
where Pptq stands for a probability that t actually happens and Sptq represents the quantitative
ramications expressed most often in nancial terms. The following threat sets a good
example: a product complying with some requirement is not delivered in due time. When
the requirement is delivered with delay and the ramications depends on the value of the
delay, the Equation 1 should be drafted into the following form:
Rptq
8
0
Ppt, q Spt, qd (2)
where denotes the value of delay.
When risk is estimated for a threat on the grounds of delivery time distribution, the mean loss
value becomes the actual value of risk. Then, summing up risk values for all requirements
belonging to the requirement model of a particular release, the mean cumulative risk for the
release r is obtained:
RRprq
tPT
r
Rptq (3)
where T
r
is the set of threats connected with release r. If all the releases are considered, the
mean threat cost is obtained for the whole project p:
RPppq
rPp
RRprq (4)
The risk can be used to evaluate release plan as well as the whole project. When the RPppq
calculated for the whole project surpasses the potential benets, there is no point in launching
the project. Furthermore, different release plans may be considered, and the RRprq value may
be used to choose the best one.
One may argue that summing risks that were estimated for single threats (Rptq) in order
to evaluate the risk regarding the whole project can result in cumulating estimation errors.
Therefore, the provided estimates of RPppq can be signicantly off. Such scenario is likely
when the requirements come late in the software development life cycle and are not known
313 Automatic Control of the Software Development Process with Regard to Risk Analysis
up front. Nevertheless, the suggested approach also has advantages. As a matter of fact,
considering each threat separately can be benecial for agile methods where plans are made
for short periods of time, e.g. the Planning Game in XP Beck & Andres (2005). The agile
methods usually employ high level of automation Beck & Andres (2005), Stamelos & Sfetsos
(2007), and hence create a good environment for automated risk analysis. Furthermore, the
mean cumulative risk for the release RRprq is designed to be the main area of application of the
suggested approach and the release is usually the main concern of planning in agile methods
Beck & Andres (2005), Schwaber & Beedle (2001).
The suggested risk analysis method partly corresponds with the ISO/IEC standard for risk
management ISO(2004). The standard is focused on the whole software life cycle whereas our
method can cover the whole life cycle but is focused on shorter terms, i.e. releases. Similar, but
slightly more complex method was suggested by Penny Penny (2002), who emphasized the
relevance of resources quality and environmental conditions (specically interruptions and
distractions in work). We do not question the value of those parameters. Nevertheless, they
may cause difculties in collecting the empirical data that is necessary to estimate the threat
occurrence probabilities. Therefore, data regarding those parameters is not required in our
risk analysis method, but when available, denitely recommended.
2.3 Threat occurrence probability
According to Equation 1 the risk consist of two elements : probability of occurrence of a threat
Pptq and expected loss Sptq. Let us assume that the threat t
i
regards failing to deliver the
requirement req
i
in the due time. Therefore, in order to calculate Ppt
i
q the distribution of
expected effort, which is necessary to fulll requirement req
i
, must be evaluated. The methods
of effort estimation are discussed in Section 3.
There is also a second factor that is relevant to the probability of occurrence. It is the
availability of resources for the planned release, which means that the effort estimated for
the requirements and the available manpower indicates the probability of failing to deliver a
requirement in due time. Furthermore, not every requirement could be implemented with
equal effort by any participant of the developers team Helming et al. (2009). Therefore,
the effort estimations should be combined with a workow model. The workow model
is indicated by the employed software development methodology, and hence should be
available in every non chaotic process. Nevertheless, what derives from the methodology
denition can be on a very high level of abstraction or exibility, especially in the case of
agile methods Schwaber & Beedle (2001), and detailing the release plan is not a trivial task.
According to Bagnall et al. Bagnall et al. (2001) the problem is NP hard even without any
dependencies between requirements. Fortunately, there are methods for semi-automated
planning of release Helming et al. (2009).
Estimation of probability of occurrence of a threat Pptq must take into consideration two
factors: effort estimation for different requirements on the one hand and the allocation of
resources on the other. Resources allocation can be done semi-automatically Helming et al.
(2009) or may come from a manager decision. In both cases the resources allocation will
be available and can be used for further analysis. However, in some cases (i.e. in the
agile methods) it may be done on a very low level of granularity. Effort estimation is more
314 Automation
challenging. The difculties that may arise during automating those estimations are discussed
in the next section.
2.4 Loss function
As it has been stated in Section 2.1 the value of loss function might be explicitly dened in
the customer software supplier agreement. Unfortunately, such simple cases do not cover all
possibilities. There are quality related aspects that might not be mentioned in the agreement,
but still may cause loss. According to the ISO/IEC 9126 ISO (2001) there are six characteristics
to consider: functionality, reliability, usability, efciency, maintainability, and portability.
Some of these characteristics may be covered by the customer software supplier agreement
and when are not delivered on the contracted level there is a penalty. However, there are
two aspects that should be explicitly considered, i.e. software defects and maintainability (the
defects might regard not satisfying one of the ve other ISO quality characteristics).
It is a typical case that software contains defects Kan (1994). Therefore, the customers, in order
to protect their interests, put defect xing policies in the agreement. As a result, the defects
create loss. First of all they must be repaired, and sometimes there is also a nancial penalty.
There are many different grounds for defects. Nevertheless, there is a general rule regarding
the relation between cost, time and quality called The Iron Triangle Atkinson (1999), Oisen
(1971), see Fig. 1. According to The Iron Triangle cost, time and quality depend on each other.
In other words, implementation of a set of requirements generates costs and requires time in
order to achieve the desired quality. When one of those two factors is not well supported,
the quality goes down, e.g. skipping tests of a component may result in a greater number
of defects in this component. However, it should be taken into consideration that in the case
of skipping tests we have savings in current release. The issues regarding lower quality and
greater number of defects will arise in the next release. Therefore, it is important to estimate
the risk regarding lower quality since it sometimes could be reasonable to make savings in
current release and pay the bill in the next one. Risk related to low quality regards the loss
that comes from defect xing activities in the next release. Hence, the number of defects as
well as the effort that is necessary to x them should be predicted and used to construct the
loss function. The defect prediction is discussed in Section 4.
Fig. 1. The Iron Triangle
Low level of maintainability, which reects problems with analysing, changing and testing
the software increases the efforts of developing new features. Since the efforts are greater,
the threat occurrence probability for forthcoming requirements grows as well. If the current
release is not the last, the lowmaintainability related loss should be estimated and considered.
Methods of estimating maintainability are discussed in Section 5.
2.5 Automation possibilities
The goal of this chapter is to present how risk can be estimated in an automated way. Risk has
been decomposed into ve factors: effort estimation, release planning, missing functionality
penalty, defect prediction, and maintainability measuring (Fig. 2. The release planning is a
managers decision. Nevertheless, the decision may be supported by a tool. Specically, the
suggested risk evaluation approach can be used to assess a release plan. The penalty is usually
dened in the customer software supplier agreement and is a function of both: delivery time
and satisfaction level of the requirement implementation.
Fig. 2. Problem decomposition; elements denoted with * are detailed in forthcoming
sections.
The three other aspects (effort, hidden defects and maintainability) are evaluated using
empirical data, therefore there is room for automation. When there is no estimation
method available, historical data must be collected and used to train a prediction model. If
there is no historical data, cross-project prediction methods should be considered. All the
aforementioned activities may be automated to some extent. We can employ tools to collect
data, we can use tools to make the estimations and nally to generate the risk related reports.
Unfortunately, not all aspects of discussed above risk analysis were investigated closely
enough to give conclusive remarks and not all of them are supported by tools. The following
sections will discuss what has been already done and where there are gaps that must be lled
by practitioners who would like to automate the risk analysis.
3. Effort estimation
Considerable research has been done on effort estimation. It regards different techniques and
at least some of them may be used in risk analysis. As it was discussed in Section 2, threats
like a product complying with some requirement is not delivered in due time should be
considered. Therefore, efforts for each requirement must be estimated to assess the risk.
In order to choose the most appropriate effort estimation approach let us review the available
ones. Boehm et al. Boehm et al. (2000) summarized the leading techniques (the content of this
316 Automation
section is mostly based on their work). Later, the estimation methods were also reviewed by
Jorgensen and Shepperd Jorgensen & Shepperd (2007).
3.1 Model-based techniques
Most of the models have functional form. Metrics which describe features of the object of
estimation or features of the organization, which is going to develop the object, are used
as model input. After some calculations the model produces value of the expected effort.
Unfortunately, many models are proprietary and hence cannot be analysed in terms of model
structure. On the other hand, the proprietary solutions are usually ready to use tools with
high level of automation.
3.1.1 Constructive Cost Model (COCOMO 81 and COCOMO II)
The COCOMOmodel was rstly suggested in Boehm(1981). It is an algorithmic software cost
estimation model that uses regression formula. The intuition behind COCOMO model is that
the development grows exponentially as the developed system grows in size:
Ef f ort
PM
a pKLOC
b
q p
j
EM
j
q (5)
Where, Ef f ort
PM
is the whole effort in person months, KLOC is thousands of lines of source
code, EM is an effort multiplier generated via regression on historical data, a and b denote
software type coefcients.
The COCOMO 81 model was based on a study of 63 projects that were created using the
waterfall software development model, which reected the practices of the 80. Nevertheless,
software development techniques changed dramatically, namely highly iterative processes
were employed. Therefore, applying the COCOMO 81 model becomes problematic. The
solution to the problem was to redesign the model to accommodate different approaches to
software development.
The revised cost estimation model was called COCOMO II and was initially described by
Boehm et al. Boehm et al. (1995). The model is focused on estimating the effort that regards
the whole system. Moreover, it is based on empirical data; it is widely used and hence well
validated. Web based tools that support the COCOMO II model are available on the models
web page Center for Systems and Software Engineering (Retrived on 01/11/2011). There are
commercial implementations: COSTAR Softstar Systems (Retrived on 01/11/2011) and Cost
Xpert Cost Xpert AG (Retrived on 01/11/2011) as well, so the automation is relatively easy.
There are also derived models, e.g. COSYSMO Valerdi et al. (2003) or AFCAA REVIC Air
Force Cost Analysis Agency (Retrived on 01/11/2011).
3.1.2 Putnams Software Life-cycle Model (SLIM)
The Putnams Software Life-cycle Model Putnam (1978) is based on analysis of software
development lifecycle in terms of the wellknown Rayleigh distribution of resources (project
personnel) usage versus time. Quantitative Software Management offers a tools suite based
on metrics collected from over 10 000 completed software projects that employ the Putnams
SLIM model Quantitative Software Management Inc. (Retrived on 01/11/2011).
The tools assure high level of automation. SLIM was originally designed to make estimations
for the whole project. However, the project can be interpreted as a set of requirements, what
partly corresponds with the described in Section 2 risk analysis.
3.1.3 SPR KnowledgePLAN
SPR KnowledgePLAN (following in the footsteps of previous SPR products: SPQR/20
and Checkpoint) is a knowledge-based software estimation solution Software Productivity
Research (Retrived on 01/11/2009). It is based on a proprietary database of 14 531 projects
(version 4.3). The estimation is calculated according to detailed project description.
The tool predicts effort at four levels of granularity: project, phase, activity and task, therefore
it is suitable for risk analysis. Furthermore, since there is good support in tools the automation
is relatively simple.
3.1.4 Functionalitybased estimations
3.1.4.1 Function Points
The function points were suggested by Albrecht Albrecht (1979) and are a standard unit
of measure for representing the functional size of a software application. The functional
requirements are categorized into ve types: outputs, inputs, inquiries, internal les, and
external interfaces. Subsequently, the requirements are assessed for complexity and a
number of function points are assigned. Since 1984 this method has been promoted by The
International Function Point Users Group (IFPUG). This group produced a set of standards,
known as the ISO/IEC 14143 series. Function Point Analysis was also developed by
NESMA Netherlands Software Metrics Association (Retrived on 01/11/2011). It resulted
in method called NESMA Functional Size Measurement that is compliant with ISO/IEC
24570. Moreover, there are good automation possibilities since the Function Point Analysis
is supported by a tool called Metric Studio TSA Quality (Retrived on 01/11/2011).
3.1.4.2 MKII FPA
Mk II Function Point Analysis Symons (1991) is a software sizing method that belongs to
the function point group of techniques. The most important feature of this method is the
simple measurement model. There are only three components to consider: inputs (processes
that enter data into software), outputs (processes that take data from software), and entity
references (storage, retrieval and deletion of date from permanent storage).
3.1.4.3 Weighted Micro Function Points (WMFP)
The Weighted Micro Function Points (WMFP) is a software size estimating method developed
by Logical Solutions Logical Solutions (Retrived on 01/11/2011). The method measures
several different software metrics obtained from the source code that represents effort and
are translated into time: ow complexity, object vocabulary, object conjuration, arithmetic
intricacy, data transfer, code structure, inline data, and comments.
318 Automation
3.1.4.4 Common Software Measurement International Consortium (COSMIC)
The COSMIC project was launched in 1998, it is a voluntary, world-wide grouping of software
metrics experts Common Software Measurement International Consortium (Retrived on
01/11/2009). The focus is on developing a method of measuring functional size of software.
The method is entirely open, and all the documentation is available for free of charge
download. Unfortunately, the consortium provides only the method description and hence
automation may be time consuming. However, there are third party vendors that support the
method
1
.
3.1.4.5 Object Points
Object Points is an object oriented software size estimating method developed by Banker et al.
Banker et al. (1991). The method is based on the following object types (type denitions after
Banker et al. Banker et al. (1991)): rule set (a collection of instructions and routines written in a
high level language of a CASE tool), 3GL module (a precompiled procedure, originally written
using third-generation language), screen denition (logical representation of an on-screen
image), and user report. The object points are obtained by summing the instances of objects of
the aforementioned types and weighting each type by the development effort associated with
it.
3.1.4.6 Use Case Points (UCP)
UCP is a measure of software size that takes into consideration use cases. It is calculated by
counting the number of actors and transactions that were included in the ow of use case
scenarios. Transaction is an event that occurs between the actor and the system and can be
made entirely or not at all.
There are tools that support the UCP, e.g. Kusumoto et al. (2004). Nevertheless, identifying
transactions in the use cases is not a trivial task. Natural language processing methods are
usually employed Ochodek &Nawrocki (2007) and there is a considerable mistake possibility.
3.1.5 PRICE-S
The PRICE-S is a parametric cost model that was originally developed for internal use in
NASA software projects. It is implemented in the TruePlanning PRICE Systems (Retrived on
01/11/2011), which is a tool that encapsulates the experience with 3 212 projects from over 30
years.
According to the authors, risk refers to the fact that because a cost estimate is a forecast, there
is always a possibility of the actual cost being different from the estimate (such situation very
well corresponds with our goal). Furthermore, using TruePlanning tool allows risk analysis
that quanties probability.
The tool makes estimations on work package level hence it is appropriate for the described in
Section 2 risk analysis. Unfortunately, according to Boehm et al. (2000) the model equations
were not released in public domain, still, it is a welldesigned commercial solution and
therefore the automation is supported.
1
http://www.cosmicon.com/vendorsV3.asp
3.1.6 SEER for Software (SEER-SEM)
SEER-SEM is an algorithmic, parametric project management system that supports
estimating, planning and tracking effort and resource consumption for software development
and maintenance Galorath Inc. (Retrived on 01/11/2011). The estimations are calculated
using effective size (S
e
). It is a generic metric that covers development and integration of
new, reused, and commercial off-the-shelf software modules. Several sizing metrics can be
translated into S
e
, this includes lines of code (LOC), function points, IFPUG function based
sizing method, and use cases. Although the tool is focused on estimation for the whole project,
there are great automation possibilities due to the support for integration with MicroSoft and
IBM products.
3.1.7 PROxy-Based Estimating (PROBE)
Proxy based estimating uses previous developments in similar application domains. The
method is a part of the Personal Software Process Humphrey (1995). A proxy is a unit of
software size that can be identied early in a project (e.g. screen, le, object, logical entity,
function point). Proxies can be translated into lines of source code according to historical sizes
of similar proxies in earlier developed projects.
The concept of proxies corresponds well with the risk analysis. Nevertheless, according to
authors knowledge there is no tool that supports the method. Therefore, the automation may
be challenging.
3.2 Expertbased techniques
The expertbased techniques capture the knowledge of domain experts and use it to obtain
reliable estimations. Those techniques are especially useful when there is not enough
empirical data to construct a model. Jorgensen and Shepperd Jorgensen & Shepperd (2007)
pointed out two important facts regarding those methods of estimation. Firstly, they
concluded that the formal estimation techniques have no documented accuracy higher than
the expertbased approach. Secondly, the expertbased technique is the most commonly used
method in the industry. Unfortunately, using expert knowledge is very inconvenient with
regard to automation.
Two expertbased techniques have been developed. These are the Work Breakdown Structure
and the Delphi method.
3.2.1 Work Breakdown Structure (WBS)
The WBS is a method of organizing project into a hierarchical structure of components or
activities (as a matter of fact, two structures can be created; one for the software product itself,
and one for the activities that must be performed in order to build the software) in a way
that helps describe the total work scope of the analysed project. The hierarchical structure
is developed by starting with the end objective and dividing it into gradually smaller ones.
A tree, which contains all actions that must be executed in order to achieve the mentioned
objective is obtained as the result. The estimation is done by summing the efforts or resources
assigned to single elements.
320 Automation
3.2.2 Delphi
The Delphi method is a structured, interactive estimation method that is based on experts
judgements. A group of experts is guided to a consensus of opinion on an issue.
Participants (experts) make assessments (estimations) regarding the issue without consulting
each other. The assessments are collected and presented to each of the experts individually.
Subsequently, the participants make second assessment using the knowledge regarding
estimations provided by other experts. The second round should result in a narrowing of
the range of assessments and indication of a reasonable middle value regarding the analysed
issue.
The Delphi technique was developed at The Rand Corporation in the late 1940s as a way of
making predictions regarding future events Boehm et al. (2000); currently a similar method,
i.e. the planning poker is very popular.
3.3 Lerning-oriented techniques
Learning-oriented techniques include a wide range of different methods. There are
traditional, manual approaches like case studies as well as highly automated techniques,
which build models that learn from historical data, e.g. neural networks, classication trees.
Unfortunately, even the most highly automated approaches require some manual work since
the historical data must be collected and the estimation models must be integrated with
the software development environment. The learning-oriented techniques are commonly
employed in other approaches.
3.3.1 Case studies
Case studies represent an inductive process, whereby estimators and planners try to learn
useful general lessons and estimation heuristics by extrapolation from specic examples
Boehm et al. (2000). Case studies take into consideration the environmental conditions and
constraints, the decisions that inuenced software development process, and the nal result.
The goal is focused on identifying relations between cause and effect that can be applied in
other projects by analogy. Case studies that result in a wellunderstood cause effect relations
constitute good foundation for designing automated solutions that can be applied to similar
project. Nevertheless, in order to dene the boundaries of similarity further analysis is usually
required.
According to Jorgensen and Shepperd Jorgensen & Shepperd (2007) there are few case studies
focused on the effort estimation. In consequence, there is a possibility that we still can learn a
lot from well described reallife cases.
3.3.2 Regression based techniques
Regression based techniques are the most popular method in the effort estimation studies
Jorgensen & Shepperd (2007). Moreover, they were used in many of the model based
approaches, e.g. COCOMO, SLIM.
The regression based methods refer to the statistical approach of general linear regression with
the least square technique. Historical data is employed to train the model and then the model
is used for making estimations for newprojects. The regression based methods are well suited
for automation since they are supported by almost all statistic tools (including spreadsheets).
Unfortunately, the data which is necessary to train the model is not always available and its
collection may cause automation related issues.
3.3.3 Neural networks
There are several articial intelligence approaches to effort estimation and the neural networks
are the most popular among them. Nevertheless, according to Jorgensen and Shepperd
Jorgensen & Shepperd (2007) less than 5% of research papers oriented to effort estimation
considered the neural networks. There is no no good reason for such low popularity since the
neural networks have proven to have adequate expression power. Idri et al. Idri et al. (2002)
used backpropagation threelayer perceptron on the COCOMO 81 dataset and according to
the authors, the obtained accuracy was acceptable. Furthermore, the authors were able to
interpret the obtained network by extracting if-then fuzzy rules from it, hence it alleviates to
some extent the main neural networks drawback, namely low readability.
Other example of neural networks usefulness was provided by Park and Baek Park & Baek
(2008), who investigated 148 software projects completed between 1999 and 2003 by one of
the Korean IT service vendors. The authors compared accuracy of the neural networks model
with human experts judgments and two classical regression models. The neural network
outperformed the other approaches by producing results with MRE (magnitude of relative
error) equal to 59.4 when expert judgements and the regression models produced results with
MRE equal to 76.6, 150.4 and 417.3 respectively.
3.3.4 Analogy
The analogybased estimation methods are based on projects characteristic in terms of
features, e.g. size of the requirements, size and experience of the team, software development
method. The features are collected from historical projects. In order to forecast effort for a
new project the most similar historical projects must be found. Similarity can be dened as
Euclidean distance in n-dimensional space where n is the number of project features. The
dimensions are standardized since all of them should have the same weight. The most similar
projects, i.e. the nearest neighbours, are used as the basis for effort prediction for the new
project. According to Jorgensen and Shepperd Jorgensen & Shepperd (2007) the popularity of
research on analogybased estimation models is increasing.
Good prediction abilities of the analogybased methods were shown by Shepperd and
Schoeld in a study conducted on 9 datasets, a total of 275 software projects Shepperd &
Schoeld (1997). The analogy (case based reasoning) was compared upon stepwise regression
analysis and produced superior predictive performance for all datasets when measured in
MMRE (mean magnitude of relative error) and for 7 datasets when measured in Pred(25)
(percentage of predictions that are within 25% of the actual value).
Gupta et al. Gupta et al. (2011) surveyed the recent analogy based techniques for effort
estimation and identied following approaches (some of them overlap estimation methods
mentioned in other subsections):
322 Automation
Machine learning based approaches (Multi Layer Perceptrons, Radial Basis Functions,
Support Vector Regression, Decision Trees)
Fuzzy logic based approaches (Fuzzy analogy, Generalized Fuzzy Number Software
Estimation Model)
Grey Theory based approaches (Grey Relational Analysis based software project effort
(GRACE), Improved GRA, GLASE, Weighted GRA, GRACE
`
)
Hybrid And Other Approaches (Genetic Algorithms with GRA, Fuzzy Grey Relational
Analysis, AQUA (case-based reasoning & collaborative ltering))
Gupta et al. also investigated the efciency of the above listed methods. Moreover, the authors
collected and presented information regarding the prediction accuracy of those methods. The
best result was obtained in the case of Decision Trees. Nevertheless, it is not conclusive since
the methods were evaluated on different data sets.
3.3.5 Analysis effort method
The analysis effort method is suited to producing the initial estimates for a whole project.
The method is based on doing some preliminary detailed design and measuring the time
consumed to accomplish these tasks. Then, the results are used to estimate effort for the
rest of the analysis, design, coding and testing activities. For the purpose of this method,
a simplied, waterfall based software development lifecycle is assumed. This method uses
three key factors that apply to the estimated project in terms of its size, familiarity and
complexity. Unfortunately, the Analysis Effort Method requires some manual work in the
initial phase, and therefore impedes the automation. A detailed description of the Analysis
Effort Method was given by Cadle and Yeates Cadle & Yeates (2007).
3.4 Dynamicsbased techniques
The dynamics based techniques take into consideration the dynamics of effort over the
duration of software development process. The environment conditions and constraints (e.g.
available resources, deadlines) may change over time. Some of the changes are scheduled,
some not, but all of them may have signicant inuence on productivity and, in consequence,
the likelihood of project success.
3.4.1 System dynamics approach
System dynamics is a continuous simulation modeling methodology whereby model results
and behavior are displayed as graphs of information that change over time. Models are
represented as networks modied with positive and negative feedback loops. Elements
within the models are expressed as dynamically changing levels or accumulations (the nodes),
rates or ows between the levels (the lines connecting the nodes), and information relative to
the system that changes over time and dynamically affects the ow rates between the levels
(the feedback loops). Description after Boehm et al. Boehm et al. (2000).
There are examples of application of the systemdynamics approach e.g. Madachy (1996), Ruiz
et al. (2001). Detailed description of the concept was provided by Abdel-Hamid and Madnick
Abdel-Hamid & Madnick (1991).
3.5 Composite techniques
Composite techniques employ two or more different effort estimation approaches (including
those mentioned above) to produce the most appropriate model.
3.5.1 Bayesian approach
Bayesian approach is a method of inductive reasoning that minimizes the posterior expected
value of a loss function (e.g. estimation error). Bayesian Belief Nets are an approach that
incorporates the case effect reasoning. The causebased reasoning is strongly recommended
by Fenton and Neil Fenton & Neil (2000). The authors gave a very interesting example of
providing wrong conclusion when ignoring the causeeffect relations. The example regards
analysing car accidents. The relation between month of year and the number of car accidents
shows that the safest to drive months are January and February. The conclusion, given the
data involved, is sensible, but intuitively not acceptable. During January and February the
weather is bad and causes treacherous road conditions and fewer people drive their cars and
when they do they tend to drive more slowly. According to the authors, we experience similar
misunderstanding in the effort estimation since most models use equation with following
form: e f f ort f psizeq. The equation ignores the fact that the size of solution cannot cause
effort to be expanded.
Fenton and Neil employed Bayesian Belief Nets to represent casual relationships between
variables investigated in the model Fenton & Neil (1999a). In consequence, they obtained
model that has the following advantages:
explicit modeling of ignorance and uncertainty in estimations,
combination of diverse type of information,
increasing of visibility and auditability in the decision process,
readability,
support for predicting with missing data,
support for whatif analysis and predicting effects of process change,
rigorous, mathematical semantic.
3.5.2 COSEEKMO
COSEEKMO is an effort-modeling workbench suggested by Menzies et al. Menzies et al.
(2006) that applies a set of heuristic rules to compare results from alternative models:
COCOMO, Local Calibration Boehm (1981), Least Squares Regression and Quinlans M5P
algorithm Quinlan (1992). The model was trained and evaluated using data from 63 projects
originally used to create the COCOMO 81 model, 161 projects originally used to create the
COCOMO II model and 93 projects from six NASA centers.
The model is focused on estimating the effort that regards the whole application. According to
the authors the COSEEKMOanalysis is fully automated; all model procedures were published
as programs written in pseudocode or sets of heuristics rules.
324 Automation
3.5.3 ProjectCodeMeter
The ProjectCodeMeter Logical Solutions (Retrived on 01/11/2011) is a commercial tool that
implements four different cost models: Weighted Micro Function Points (WMFP), COCOMO
81, COCOMO II, and Revic 9.2. When consider which models are supported, it is not
surprising that the tool is focused on making estimations for the whole project. There is a
great automation possibility since the tool integrates with many IDEs, i.e. MicroSoft Visual
Studio, CodeBlock, Eclipse, Aptana Studio, Oracle JDeveloper, and JBuilder.
3.5.4 Construx Estimate
Estimate Construx (Retrived on 01/11/2011) predicts effort, budget, and schedule for
software projects based on size estimates. It is calibrated with industry data. However, it
is possible (and recommended) to calibrated Estimate with own data. Estimate is based on
Monte Carlo simulation and two mentioned above estimation models: Putnams SLIM and
COCOMO. It is focused on estimations for the whole project.
3.5.5 CaliberRM ESTIMATE Professional
CaliberRM ESTIMATE Professional Borland Software Corporation (Retrived on 01/11/2011)
uses two models in order to make initial estimate of total project size, effort, project schedule
(duration) as well as staff size and cost: COCOMOII and Putnams SLIM. Later, a Monte Carlo
Simulation may be used to generate a range of estimates with different probabilities later the
most appropriate one can be selected and executed.
4. Defect prediction
The described in Section 2 risk analysis considers defect repairs as a vital part of the risk
assessment process. Delivering lowquality software will result in a number of hidden defects,
which is why the development of the next release will generate extra cost (these defects must
be removed). In consequence, it is important to estimate the number of defects and the effort
that is needed to x them. Considerable research has been conducted on defect prediction.
The current state of the art has been recently reported in systematic reviews Catal (2011);
Kitchenham (2010).
4.1 Predicting the number of defects
Traditionally, the defect prediction studies were focused on predicting the number or density
of defects Basili & Perricone (1984); Gaffney (1984); Lipow (1982). However, these approaches
were extensively criticized Fenton & Neil (2000; 1999b). It was concluded that the models
offer no coherent explanation of how defect introduction and detection variables affect defect
count and hence there is no causeeffect relation for defect introduction. Furthermore, the
employed variables are correlated with gross number of defects, but are poor in predicting. In
consequence, these approaches are not adequate for risk analysis.
Recent defect prediction studies are focused on supporting resource management decisions
in the testing process. The studies investigate which parts of a software system are especially
defect prone and hence should be tested with special care and which could be defect free and
hence may be not tested. The most popular approaches are focused on organizing modules
in an order according to the expected number of defects Jureczko & Spinellis (2010); Weyuker
et al. (2008) or on making boolean decision regarding whether a selected module is defect
free or not Menzies et al. (2007); Zimmermann & Nagappan (2009). Unfortunately, such
approaches are not effective in estimating the overall number of defects. Probably the most
appropriate approach to defect prediction with regard to risk analysis is Bayesian belief nets
Fenton & Neil (1999a),Fenton et al. (2007). This approach can be employed to predict the
number of defects Fenton et al. (2008) and can provide the causeeffect relations. The Bayesian
belief nets approach might be a source of automation related issues. Unfortunately, designing
and conguring such a net without expert assistance is challenging. However, when the net is
dened, it is readable and understandable Fenton & Neil (2000) and hence no further expert
assistance in necessary.
There is also an alternative approach that is much easier to automate, but unfortunately
does not support the causeeffect analysis, namely, reliability growth models. The reliability
growth models are statistical interpolation of defect detection in time by mathematical
functions, usually Weibull distribution. The functions are used to predict the number of
defects in the source code. Nevertheless, this approach has also drawbacks. The reliability
growth models require detailed data regarding defects occurrence and are designed for the
waterfall software development process. It is not clear whether they are suitable for highly
iterative processes. Detailed description of this approach was presented by Kan Kan (1994).
4.2 Predicting the defect xing effort
The number of residual defects does not satisfy the requirements of risk analysis, when the
xing effort remains unknown. There are two types of studies focused on xing time. First
type estimates the time from bug report till bug x Giger et al. (2010); Gokhale & Mullen
(2010); Kamei et al. (2010). However, this type of estimation is not helpful in risk analysis.
Fortunately, the second type of studies is more appropriate since it is focused on the xing
effort.
Weiss et al. Weiss et al. (2007) conducted an empirical study regarding predicting the
defect-xing effort in the JBoss project. The authors used text similarity techniques to execute
analogy (nearest neighbor approach) based prediction. All types of issues were considered
(defects, feature requests, and tasks). However, the best results were obtained in the case of
defects; the average error was below4 hours. Nevertheless, the study has lowexternal validity
and according to the authors should be considered as a proof of concept. Further research in
similar direction was conducted by Hassouna and Tahvildari Hassouna & Tahvildari (2010).
The authors conducted an empirical study regarding predicting the effort of defectxing
in JBoss (the same data set as in Weiss et al. (2007)) and Codehaus projects. The suggested
approach (Composite Effort Prediction Framework) enhanced the technique suggested by
Weiss et al. (2007) by data enrichment and employing majority voting, adaptive threshold,
and binary clustering in the similarity scoring process. From the automation point of view,
this study provides us with an important nding. The authors developed PHP scripts to crawl
and collect issue information from a project tracking system called JIRA. Hence, the authors
showed a method of gathering empirical data with regard to defect xing effort.
326 Automation
Zeng and Rine Zeng & Rine (2004) used dissimilarity matrix and selforganizing neural
networks for software defects clustering and effort prediction and obtained good accuracy
when applied to similar projects (average MRE from 7% to 23%). However, only 106 samples
corresponding to 15 different software defect x efforts were used to train the model, which
signicantly affects the external validity.
Datasets from NASA were also used by Song et al. Song et al. (2006). The study investigated
200 software projects that were developed over more than 15 years and hence the external
validity was signicantly better. The authors used association rule mining based method
to predict defect correction effort. This approach was compared to C4.5, Naive Bayes and
Program Assessment Rating Tool (PART) and showed the greatest prediction power. The
accuracy was above 90%.
The studies regarding predicting defect xing effort present various approaches.
Nevertheless, the external validity is limited and they do not conclude which approaches
is optimal or how it can be automated.
4.3 Crossproject prediction
In order to employ a defect prediction model, it must be trained. When analysing the rst
release of a project, or when no historical data is collected, the model must be trained using
data from other projects. Unfortunately, the cross-project defect prediction does not give
good results. Some studies have investigated this issue recently He et al. (2011); Jureczko
& Madeyski (2010); Zimmermann et al. (2009). However, all of them reported low success
rate.
5. Measuring maintainability
Asoftware systemthat delivers all functionalities correctly may still cause problems in future.
Specically, the effort connected with implementing new features may differ depending on
the character of already written source code. There are phenomena that affect the easiness
of modifying and improving an application, e.g. the code smells (symptom in the source
code that possibly indicates a deeper problem). Pietrzak and Walter Pietrzak & Walter (2006)
investigated the automatic methods of detecting code smells. The work was later extended
in Martenka & Walter (2010) by introducing a hierarchical model for software design quality.
In consequence, the model employs the interpretation of software metrics as well as historical
data, results of dynamic behavior, and abstract syntax trees to provide traceable information
regarding detecting anomalies, especially code smells. Thanks to the traceability feature it is
possible to precisely locate the anomaly and eliminate it. The model is focused on evaluating
highlevel quality factors that can be a selection of quality characteristic.
Unfortunately, code smells and readability do not cover all aspects of software that are related
to the easiness of modifying, improving and other maintenance activities. These aspects
are covered by a term called software maintainability, which is dened in the IEEE Std
610.12-1990 as: the ease with which a software system or component can be modied to
correct faults, improve performance or other attributes, or adapt to a changed environment
IEEE (1990).
Considerable research has been conducted on measuring the maintainability; therefore there
is guidance regarding which approach can be used in the mentioned in Section 2 risk analysis
with regard to the automation.
5.1 ISO Quality model
The ISO/IEC 9126 Quality Model ISO (2001) identies several quality characteristics,
among them maintainability. The model also provides us with sub-characteristics. In
the case of maintainability they are: analyzability (the easiness of diagnosing faults),
changeability (the easiness of making modications), stability (the easiness of keeping
software in consistent state during modication), testability (the easiness of testing the
system), maintainabilitycompliance (the compliance with standards and conventions
regarding maintainability). Such multidimensional characteristic of maintainability seems to
be reasonable. Software that is easy to analyse, change, test, and is stable and complies with
standards and conventions should be easy to maintain.
The ISO Quality Model recommends sets of internal and external metrics designed for
measuring the aforementioned subcharacteristics. For instance, for the changeability there
is the change implementation elapse time suggested as external metric and the change
impact (it is based on the number of modications and the number of problems caused by
these modications) as internal metric. Nevertheless, the ISO Quality Model was criticized
by Heitlager et al. Heitlager et al. (2007). The authors noted that the internal as well as the
external metrics are based on the observations of the interaction between the product and its
environment, maintainers, testers and administrators; or on comparison of the product with
its specication (unfortunately, the specication is not always complete and correct). Heitlager
et al. Heitlager et al. (2007) argued that the metrics should be focused on direct observation of
the software product.
5.2 Maintainability Index (MI)
The Maintainability Index was proposed in order to assess the software system
maintainability according to the state of its source code Coleman et al. (1994). It is a composite
number obtained from an equation that employs a number of different software metrics:
MI 171 5.2 lnp
Vol 0.23
Vpg
1
q 16.2 lnp
LOC `50 sinp

a
2.46 perCM (6)
Where
Vol,
Vpg
1
q, and
LOC are the average values of Halstead Volume, McCabes Cyclomatic

Complexity, and Lines Of Code respectively, and perCMis the percentage of commented lines
in module.
The MI metric is easy to use, easy to calculate and hence to automate. Nevertheless, it is not
suitable to the risk analysis since it does not support causeeffect relations. When analysing
risk, it is not enough to say that there is a maintainability problem. An action that can mitigate
the problem should always be considered. Unfortunately, in the case of the MI we can only
try to decrease the size related characteristics (Vol, V(g), and LOC) or increase the percentage
of comments, what not necessarily will improve the maintainability. The MI is correlated
with the maintainability, but there are no proofs for cause-effect relation and hence we cannot
expect that changing the MI will change the maintainability.
328 Automation
5.3 Practical model for measuring maintainability
Heitlager et al. Heitlager et al. (2007) not only criticized other approaches, but also suggested
a new one, called Practical model for measuring maintainability. The authors argued
that the model was successfully used in their industrial practice and to some extend is
compatible with the widely accepted ISOQuality Model ISO(2001). The model consists of ve
characteristics (source code properties that can be mapped onto the ISO 9126 maintainability
subcharacteristics: Volume, Complexity per unit, Duplication, Unit size, and Unit testing.
The authors dened metrics for each of the aforementioned properties that are measured with
an ordinal scale: ++, +, o, , ; where ++ denotes very good and very bad maintainability.
5.3.1 Volume
Volume represents the total size of a software system. The main metric of volume is
Lines Of Code. However, the authors noted that this metric is good only within a single
programming language since different set of functionalities can be covered by the same
number of lines of code written. Therefore, the Programming Languages Table Jones (Retrived
on 01/11/2011) should be considered in order to express the productivity of programming
languages. Heitlager et al. gave an ready to use example for three programming languages
(Tab. 1).
rank Java Cobol PL/SQL
++ 066 0131 046
+ 66246 131491 46173
o 246665 4911,310 173461
6651,310 1,3102,621 461992
>1,310 >2,621 >992
Table 1. Values of the volume metric in correspondence with source code size measured in
KLOC according to Heitlager et al. (2007).
5.3.2 Complexity per unit
Complex source code is difcult to understand and hence to maintain. The complexity per
unit (method) can be measured using the McCabes Cyclomatic Complexity (CC). Heitlager
et al. pointed out that summing or calculating the average value of CC does not reect the
maintainability well, since the maintenance problems are usually caused by few units, which
has outlying values of the complexity metric. To mitigate the problem, the authors suggested
to assign risks to every unit according to the value of CC: CC P r1, 10s low risk; CC P
r11, 20s moderate risk; CC P r21, 50s high risk; CC P r51, 8s very high risk. Using the
risk evaluations, the authors determine the complexity rate according to Tab. 2.
5.3.3 Duplication
Heitlager et al. suggested calculating duplication as the percentage of all code that occurs
more than once in equal code blocks of at least 6 lines. Such measure is easy to automate
since there are tools that support identifying code duplication (e.g. PMD). Details regarding
mapping the value of duplication onto maintainability are presented in Tab. 3.
rank Moderate risk High risk Very high risk
++ 25% 0% 0%
+ 30% 5% 0%
o 40% 10% 0%
50% 15% 5%

Table 2. Values of complexity per unit according to the percentage of units that do not exceed
the specied upper limit for percentage of classes with given risk according to Heitlager et al.
(2007).
rank Duplication rank Unit test coverage
++ 03% ++ 95100%
+ 35% + 8095%
o 510% o 6080%
1020% 2060%
20100% 020%
Table 3. Rating scheme for duplication and test code coverage according to Heitlager et al.
(2007).
5.3.4 Unit size
The size per unit should be measured using LOC in a similar way to cyclomatic complexity
per unit. Unfortunately, the authors do not provide us with the threshold values that are
necessary to map the measured values to risk categories and rank scores.
5.3.5 Unit testing
Heitlager et al. suggested that the code coverage should be used as the main measure of unit
tests. Details are presented in Tab. 3.
Measuring test code coverage is easy to automate due to good tool support (e.g Clover
2
,
Emma
3
. Unfortunately, this measure does not reect the quality of tests. It is quite easy to
obtain high code coverage with extremely poor unit tests, e.g. by writing tests that invoke
methods, but do not check their behavior. Therefore, the authors recommend supplementing
the code coverage with counting the number of assert statements. Other measure of unit tests
quality is becoming very popular nowadays, namely mutation score, see Madeyski & Radyk
(2010) for details.
5.4 Other approaches
The presented above approaches do not cover all maintainability measuring techniques
that are used or investigated. The researchers also study the usefulness of regression
based models, Bayesian Networks, fuzzy logic, articial intelligence methods and software
reliability models. A systematic review of software maintainability prediction and metrics
was conducted by Riaz et al. Riaz et al. (2009).
2
http://www.atlassian.com/software/clover/
3
http://emma.sourceforge.net/
330 Automation
6. Automation possibilities and limitations
Three areas of risk analysis were discussed: effort estimation, defect prediction and
maintainability measurement. The most common techniques and tools were described.
Unfortunately, the automation of these analyses is not as trivial as installing offtheshelf
products. There are still gaps that should be lled. This section points out what is missing in
this area and what is already available.
6.1 Effort estimation
The research conducted on effort estimation as well as the set of available tools is very
impressive. Unfortunately, the focus is oriented towards estimating the whole project what is
not suitable for the dened in Section 2 risk analysis. There is support for estimating effort for
single requirement (e.g. SLIM, SPR KnowledgePLAN, PRICE-S). However, the methods are
not optimised for such task and not as well validated as in the case of larger scope estimation.
For the risk analysis it is crucial to incorporate the uncertainty parameters of the prediction.
Lets assume that there are 5 requirements and each of them is estimated to 5 days. Together
it makes 25 days. And on the other hand, there are 30 mandays to implement these
requirements. If the standard deviation of the estimations is equal to 5 minutes, the release will
presumably be ready on time. However, when the standard deviation is equal to 5 days, there
will be great risk of delay. Fortunately, some of the effort estimation techniques incorporate
uncertainty, e.g. the COCOMO II method was enhanced with the Bayesian approach Chulani
et al. (1999); Putnam considered mapping the input uncertainty in the context of his model
Putnam et al. (1999).
6.2 Defect prediction
Defect prediction is the most problematic part in risk analysis. There are methods for
predicting a number of defects and some pioneer works regarding estimating the effort of
the removal process. However, there is no technique of combining these two methods.
Furthermore, there are issues related to model training. Crossproject defect prediction
has low success rate and intraproject prediction always requires extra amount of work for
collecting data that is necessary to train the model.
Full automation of defect prediction requires further research. Currently, there are methods
for evaluating the software quality using defect related measures, e.g. the Bayesian belief
nets approach or the reliability growth models (both described in Section 4). However, it is
not clear how such measures are correlated with the defect xing effort and hence further
empirical investigation regarding incorporating these measures into risk analysis is necessary.
6.3 Measuring maintainability
There are a number of maintainability measuring methods. We recommend the Practical
model for measuring maintainability suggested by Heitlager et al. Heitlager et al. (2007).
This approach not only measures the maintainability, but also shows what should be done in
order to improve it. The authors put the model in use in a number of projects and obtained
satisfactory results. Nevertheless, no formal validation was conducted and hence it is hard
to make judgments about the external validity. There is also a possibility that automation
related issues will arise, since the authors omitted detailed description of some of the model
dimensions. It is also not clear howto map the model output to project risk. Therefore, further
research regarding incorporating the model into the risk analysis is necessary.
6.4 Data sources
Further research requires empirical data collected from real software projects. The collection
process requires tools that are well adjusted to the software development environment. Such
tools can be also benecial during the risk analysis, since usually they can be used to monitor
the projects and hence to assess its current state. A number of tools were listed in the
Section 3. We are also working on our own solutions (Ckjm Jureczko & Spinellis (Retrived
on 06/05/2011), and BugInfo Jureczko (Retrived on 01/11/2011a)). It is possible to collect
data from opensource projects with such tools. Limited number of projects track the effort
related data, but fortunately, there are exceptions e.g. JBoss and some of the Codehaus
projects Codehaus (Retrived on 01/11/2011). There are also data repositories that can be
easily employed in research. That includes public domain datasets: Promise Data Repository
Boetticher et al. (Retrived on 01/11/2011), Helix Vasa et al. (Retrived on 01/11/2011), and
our own Metrics Repository Jureczko (Retrived on 01/11/2011b); as well as commercial:
ISBSG ISBSG (Retrived on 01/11/2011), NASAs Metrics Data Program NASA (Retrived on
01/05/2006), SLIM, SPR KnowledgePLAN, or PRICE-S.
7. Summary and future works
Method of software project risk analysis was suggested and decomposed into well known
in software engineering problems: effort estimation, defect prediction and maintainability
measurement. Each of these problems was investigated by describing the most common
approaches, discussing the automation possibilities and recommending suitable techniques.
The potential problems regarding automation were identied. Many of themcan be addressed
using the available tools (see Section 3). Unfortunately, some of them are not covered and
require further research or new software solutions.
Validation in the scope of single requirement is necessary in the case of effort estimation. The
tools are adjusted to the scope of the whole project and hence the quality of estimations for
single requirement should be proven. Such validation requires great amount of empirical data.
Potential data sources were given in Section 6.4. We are developing a tool (extension of our
earlier solutions: BugInfo and Ckjm; working name: Quality Spy) that will be able to collect
the data and once we have nished we are going to conduct an empirical study regarding the
validation.
The discussion regarding defect prediction was not conclusive due to a number of major issues
related to incorporating the prediction methods into the risk analysis. It is not clear how to
combine the predicted number of defects with the expected xing effort. Furthermore, the
studies regarding prediction of xing effort have low external validity and the cross-project
defect prediction has low success rate. We are going to conduct further studies regarding
estimating effort of defect removal and formalizing the method of using results of defect
prediction in the risk analysis by providing a mapping to the loss function.
332 Automation
A recommendation regarding measuring maintainability was made (the model suggested in
Heitlager et al. (2007)). Unfortunately, we have no empirical data to support it and therefore,
we are going to implement the model in a tool and validate it against opensource projects.
We are expecting that the validation will help to solve the second maintainability related issue,
namely, mapping the Heitlagers model output to values of the loss function.
8. Acknowledgement
Thanks to Z. Huzar, M. Kowalski, L. Madeyski and J. Magott for their great contribution to
the design of the risk analysis method presented in this paper.
9. References
Abdel-Hamid, T. & Madnick, S. E. (1991). Software project dynamics: an integrated approach,
Prentice-Hall, Inc., Upper Saddle River, NJ, USA.
Air Force Cost Analysis Agency (Retrived on 01/11/2011). AFCAA REVIC.
URL: https://sites.google.com/site/revic92
Albrecht, A. J. (1979). Measuring Application Development Productivity, Proceedings of
the Joint SHARE, GUIDE, and IBM Application Development Symposium, Monterey,
California, USA, pp. 8392.
Atkinson, R. (1999). Project management: cost, time and quality, two best guesses and a
phenomenon, its time to accept other success criteria, International Journal of Project
Management 17(6): 337 342.
Bagnall, A. J., Rayward-Smith, V. J. & Whittley, I. M. (2001). The next release problem,
Information and Software Technology 43(14): 883 890.
Banker, R. D., Kauffman, R. J. & Kumar, R. (1991). An empirical test of object-based output
measurement metrics in a computer aided software engineering (case) environment,
J. Manage. Inf. Syst. 8: 127150.
Basili, V. R. & Perricone, B. T. (1984). Software errors and complexity: an empirical
investigation0, Commun. ACM 27: 4252.
Beck, K. & Andres, C. (2005). Extreme Programming Explained: Embrace Change; 2nd ed.,
Addison-Wesley, Boston, MA.
Boehm, B., Abts, C. & Chulani, S. (2000). Software development cost estimation approaches
a survey, Ann. Softw. Eng. 10: 177205.
Boehm, B., Clark, B., Horowitz, E., Westland, C., Madachy, R. & Selby, R. (1995). Cost models
for future software life cycle processes: Cocomo 2.0, Annals of Software Engineering,
pp. 5794.
Boehm, B. W. (1981). Software Engineering Economics, Prentice Hall.
Boetticher, G., Menzies, T. & Ostrand, T. (Retrived on 01/11/2011). Promise repository of
empirical software engineering data.
URL: http://promisedata.org
Borland Software Corporation (Retrived on 01/11/2011). CaliberRMESTIMATE Professional.
URL: http://www.ktgcorp.com/borland/products/caliber/index.html#estimate_pro
Cadle, J. & Yeates, D. (2007). Project Management for Information Systems, 5th edn, Prentice Hall
Press, Upper Saddle River, NJ, USA.
Capgemini CSD Research Inc. (Retrived on 01/11/2011). Software Cockpit.
URL: http://www.de.capgemini.com/capgemini/forschung/research/software/
Catal, C. (2011). Review: Software fault prediction: A literature review and current trends,
Expert Syst. Appl. 38: 46264636.
Center for Systems and Software Engineering (Retrived on 01/11/2011). COCOMO II.
URL: http://csse.usc.edu/csse/research/COCOMOII/cocomo_main.html
Chulani, S., Boehm, B. &Steece, B. (1999). Bayesian analysis of empirical software engineering
cost models, Software Engineering, IEEE Transactions on 25(4): 573 583.
Codehaus (Retrived on 01/11/2011). Codehaus.
URL: http://jira.codehaus.org
Coleman, D., Ash, D., Lowther, B. & Oman, P. (1994). Using metrics to evaluate software
system maintainability, Computer 27: 4449.
Common Software Measurement International Consortium (Retrived on 01/11/2009).
Cosmic.
URL: http://www.cosmicon.com
Construx (Retrived on 01/11/2011). Estimate.
URL: http://www.construx.com
Cost Xpert AG (Retrived on 01/11/2011). Cost xpert.
URL: http://www.costxpert.eu/en/research/COCOMOII/cocomo_comsoftware.htm
Fenton, N. E. & Neil, M. (1999a). Software metrics and risk, Proceedings of the 2nd European
Software Measurement Conference, FESMA 99, pp. 3955.
Fenton, N. E. & Neil, M. (2000). Software metrics: roadmap, Proceedings of the Conference on
The Future of Software Engineering, ICSE 00, ACM, New York, NY, USA, pp. 357370.
Fenton, N. & Neil, M. (1999b). A critique of software defect prediction models, Software
Engineering, IEEE Transactions on 25(5): 675 689.
Fenton, N., Neil, M., Marsh, W., Hearty, P., Radlinski, L. & Krause, P. (2007). Project data
incorporating qualitative factors for improved software defect prediction, ICSEW 07:
Proceedings of the 29th International Conference on Software Engineering Workshops, IEEE
Computer Society, Washington, DC, USA.
Fenton, N., Neil, M., Marsh, W., Hearty, P., Radli nski, L. & Krause, P. (2008). On the
effectiveness of early life cycle defect prediction with bayesian nets, Emp. Softw. Engg.
13: 499537.
Gaffney, J. E. (1984). Estimating the number of faults in code, Software Engineering, IEEE
Transactions on SE-10(4): 459 464.
Galorath Inc. (Retrived on 01/11/2011). SEER for Software.
URL: http://www.galorath.com/
Giger, E., Pinzger, M. & Gall, H. (2010). Predicting the x time of bugs, Proceedings of the 2nd
International Workshop on Recommendation Systems for Software Engineering, RSSE 10,
ACM, New York, NY, USA, pp. 5256.
Gokhale, S. S. & Mullen, R. E. (2010). A multiplicative model of software defect repair times,
Empirical Softw. Engg. 15: 296319.
Grski, J. & Miler, J. (Retrived on 01/11/2011). RiskGuide.
URL: http://iag.pg.gda.pl/RiskGuide/
Gupta, S., Sikka, G. & Verma, H. (2011). Recent methods for software effort estimation by
analogy, SIGSOFT Softw. Eng. Notes 36: 15.
334 Automation
Hassouna, A. & Tahvildari, L. (2010). An effort prediction framework for software defect
correction, Information and Software Technology 52(2): 197 209.
He, Z., Shu, F., Yang, Y., Li, M. & Wang, Q. (2011). An investigation on the feasibility of
cross-project defect prediction, Automated Software Engineering pp. 133.
Heitlager, I., Kuipers, T. & Visser, J. (2007). A practical model for measuring
maintainability, Proceedings of the 6th International Conference on Quality of Information
and Communications Technology, IEEE Computer Society, Washington, DC, USA,
pp. 3039.
Hello2morrow Inc. (Retrived on 01/11/2011). Sotograph.
URL: http://www.hello2morrow.com/products/sotograph
Helming, J., Koegel, M. & Hodaie, Z. (2009). Towards automation of iteration planning, Proc.
of the 24th ACM SIGPLAN conference companion on Object oriented programming systems
languages and applications, OOPSLA 09, ACM, New York, NY, USA, pp. 965972.
Humphrey, W. S. (1995). A Discipline for Software Engineering, 1st edn, Addison-Wesley
Longman Publishing Co., Inc., Boston, MA, USA.
Idri, A., Khoshgoftaar, T. & Abran, A. (2002). Can neural networks be easily interpreted in
software cost estimation?, Fuzzy Systems, 2002. FUZZ-IEEE02. Proceedings of the 2002
IEEE International Conference on, Vol. 2, pp. 1162 1167.
IEEE (1990). IEEE Standard Glossary of Software Engineering Terminology, Technical report,
IEEE.
ISBSG (Retrived on 01/11/2011). The global and independent source of data and analysis for
the it industry.
URL: http://www.isbsg.org
ISO (2001). ISO/IEC 9126-1:2001, Software engineering Product quality Part 1: Quality
model, Technical report, International Organization for Standardization.
ISO(2004). ISO/IEC 16085:2001, systems and software engineering life cycle processes risk
management, Technical report, International Organization for Standardization.
Jones, C. (Retrived on 01/11/2011). Programming languages table.
URL: http://www.cs.bsu.edu/homepages/dmz/cs697/langtbl.htm
Jorgensen, M. & Shepperd, M. (2007). A systematic review of software development cost
estimation studies, IEEE Trans. Softw. Eng. 33: 3353.
Jureczko, M. (Retrived on 01/11/2011a). BugInfo.
URL: http://kenai.com/projects/buginfo
Jureczko, M. (Retrived on 01/11/2011b). Metrics repository.
URL: http://purl.org/MarianJureczko/MetricsRepo
Jureczko, M. & Madeyski, L. (2010). Towards identifying software project clusters with regard
to defect prediction, PROMISE2010: Proceedings of the 6th International Conference on
Predictor Models in Software Engineering, ACM.
Jureczko, M. &Mynarski, M. (2010). Automated acceptance testing tools for web applications
using test-driven development, Electrotechnical Review 86(09): 198202.
Jureczko, M. & Spinellis, D. (2010). Using Object-Oriented Design Metrics to Predict Software
Defects, Vol. Models and Methodology of System Dependability of Monographs of
System Dependability, Ocyna Wyd. Politechniki Wroclawskiej, Wroclaw, Poland,
pp. 6981.
Jureczko, M. & Spinellis, D. (Retrived on 06/05/2011). Ckjm extended.
URL: http://gromit.iiar.pwr.wroc.pl/p_inf/ckjm/
Kamei, Y., Matsumoto, S., Monden, A., Matsumoto, K.-i., Adams, B. & Hassan, A. (2010).
Revisiting common bug prediction ndings using effort-aware models, Software
Maintenance (ICSM), 2010 IEEE International Conference on, pp. 1 10.
Kan, S. H. (1994). Metrics and Models in Software Quality Engineering, 1st edn, Addison-Wesley
Kitchenham, B. (2010). Whats up with software metrics? - a preliminary mapping study, J.
Syst. Softw. 83: 3751.
Kusumoto, S., Matukawa, F., Inoue, K., Hanabusa, S. & Maegawa, Y. (2004). Estimating
effort by use case points: Method, tool and case study, Proceedings of the Software
Metrics, 10th International Symposium, IEEE Computer Society, Washington, DC, USA,
pp. 292299.
Lipow, M. (1982). Number of faults per line of code, IEEE Trans. Softw. Eng. 8: 437439.
Logical Solutions (Retrived on 01/11/2011). Project Code Meter.
URL: http://www.projectcodemeter.com
Madachy, R. J. (1996). System dynamics modeling of an inspection-based process, Proceedings
of the 18th international conference on Software engineering, ICSE 96, IEEE Computer
Society, Washington, DC, USA, pp. 376386.
Madeyski, L. & Radyk, N. (2010). Judy a mutation testing tool for java, Software, IET 4(1): 32
42.
Martenka, P. & Walter, B. (2010). Hierarchical model for evaluating software design quality,
e-Informatica 4: 2130.
Menzies, T., Chen, Z., Hihn, J. & Lum, K. (2006). Selecting best practices for effort estimation,
IEEE Trans. Softw. Eng. 32: 883895.
Menzies, T., Greenwald, J. &Frank, A. (2007). Data mining static code attributes to learn defect
predictors, IEEE Transactions on Software Engineering 33: 213.
NASA (Retrived on 01/05/2006). Nasas metrics data program.
URL: http://mdp.ivv.nasa.gov/repository.html
Netherlands Software Metrics Association (Retrived on 01/11/2011). NESMA Function Point
Analysis.
URL: http://www.nesma.nl/section/home/
Ochodek, M. & Nawrocki, J. (2007). Automatic transactions identication in use cases,
Proceedings of the Second IFIP TC 2 Central and East European Conference on Software
Engineering Techniques, Springer, pp. 3346.
Oisen, R. P. (1971). Can project management be dened?, Project Management Quarterly 2(1): 12
14.
Park, H. & Baek, S. (2008). An empirical validation of a neural network model for software
effort estimation, Expert Syst. Appl. 35: 929937.
Penny, D. A. (2002). An estimation-based management framework for enhancive
maintenance in commercial software products, Software Maintenance, 2002.
Proceedings. International Conference on, pp. 122 130.
Pietrzak, B. & Walter, B. (2006). Leveraging code smell detection with inter-smell relations,
Vol. 4044 of Lecture Notes in Computer Science, pp. 7584.
PRICE Systems (Retrived on 01/11/2011). TruePlanning.
URL: http://www.pricesystems.com/products/true_h_price_h.asp
Putnam, L. H. (1978). Ageneral empirical solution to the macro software sizing and estimating
problem, IEEE Trans. Softw. Eng. 4: 345361.
336 Automation
Putnam, L. H., Mah, M. & Myers, W. (1999). First, get the front end right, Cutter IT Journal
12(4): 2028.
Quantitative Software Management Inc. (Retrived on 01/11/2011). SLIM Suite.
URL: http://www.qsm.com/tools
Quinlan, J. R. (1992). Learning With Continuous Classes, Proceedings of the 5th Australian Joint
Conf. Articial Inteligence, pp. 343348.
Riaz, M., Mendes, E. & Tempero, E. (2009). A systematic review of software maintainability
prediction and metrics, Proceedings of the 2009 3rd International Symposium on
Empirical Software Engineering and Measurement, ESEM 09, IEEE Computer Society,
Washington, DC, USA, pp. 367377.
Ruiz, M., Ramos, I. &Toro, M. (2001). Asimplied model of software project dynamics, Journal
of Systems and Software 59(3): 299 309.
Schneider, G. & Winters, J. P. (2001). Applying use cases, Second Edition, Addison-Wesley
Schwaber, K. & Beedle, M. (2001). Agile Software Development with Scrum, Prentice Hall PTR,
Upper Saddle River, NJ, USA.
SEI (2002). Capability Maturity Model Integration (CMMI), Version 1.1 - Continuous
Representation, Technical Report CMU/SEI-2002-TR-011, Software Engineering
Institute.
Shepperd, M. & Schoeld, C. (1997). Estimating software project effort using analogies,
Software Engineering, IEEE Transactions on 23(11): 736 743.
Softstar Systems (Retrived on 01/11/2011). Costar.
URL: http://www.softstarsystems.com/
Software Productivity Research (Retrived on 01/11/2009). SPR KnowledgePLAN.
URL: http://www.spr.com/project-estimation.html
Song, Q., Shepperd, M., Cartwright, M. & Mair, C. (2006). Software defect association mining
and defect correction effort prediction, IEEE Trans. Softw. Eng. 32: 6982.
Stamelos, I. & Sfetsos, P. (2007). Agile Software Development Quality Assurance, IGI Publishing,
Hershey, PA, USA.
Symons, C. R. (1991). Software sizing and estimating: Mk II FPA (Function Point Analysis), John
Wiley & Sons, Inc., New York, NY, USA.
Thaw, T., Aung, M. P., Wah, N. L., Nyein, S. S., Phyo, Z. L. & Htun, K. Z. (2010). Comparison
for the accuracy of defect x effort estimation, Computer Engineering and Technology
(ICCET), 2010 2nd International Conference on, Vol. 3, pp. 550 554.
TSA Quality (Retrived on 01/11/2011). Metric Studio 2011.
URL: http://www.tsaquality.com/pages/english/nav/produtos.html
Valerdi, R., Boehm, B. & Reifer, D. (2003). Cosysmo: A constructive systems engineering cost
model coming age, Proceedings of the 13th Annual International INCOSE Symposium,
pp. 7082.
Vasa, R., Lumpe, M. & Jones, A. (Retrived on 01/11/2011). Helix - Software Evolution Data
Set.
URL: http://www.ict.swin.edu.au/research/projects/helix
Weiss, C., Premraj, R., Zimmermann, T. & Zeller, A. (2007). How long will it take to x this
bug?, Proceedings of the Fourth International Workshop on Mining Software Repositories,
MSR 07, IEEE Computer Society, Washington, DC, USA, pp. 1.
Weyuker, E. J., Ostrand, T. J. & Bell, R. M. (2008). Do too many cooks spoil the broth? using
the number of developers to enhance defect prediction models, Empirical Softw. Engg.
13: 539559.
Zeng, H. & Rine, D. (2004). Estimation of software defects x effort using neural networks,
Computer Software and Applications Conference, 2004. COMPSAC 2004. Proceedings of
the 28th Annual International, Vol. 2, pp. 20 21.
Zimmermann, T. & Nagappan, N. (2009). Predicting defects with program dependencies,
Proceedings of the 2009 3rd International Symposium on Empirical Software Engineering
and Measurement, ESEM 09, IEEE Computer Society, Washington, DC, USA,
pp. 435438.
Zimmermann, T., Nagappan, N., Gall, H., Giger, E. & Murphy, B. (2009). Cross-project defect
prediction: a large scale experiment on data vs. domain vs. process, Proceedings of
the the 7th joint meeting of the European software engineering conference and the ACM
SIGSOFT symposium on The foundations of software engineering, ESEC/FSE 09, ACM,
New York, NY, USA, pp. 91100.
338 Automation
17
Automation in the IT Field
Alexander Khrushchev
Instream Ltd.
Russia
1. Introduction
To begin with, what is Information Technology (IT)? If you ask any person not related to IT
you would most likely receive an answer that the term IT means any activity related to
computers, software and communications. According to the definition adopted by
UNESCO, the IT is a set of interrelated scientific, technological and engineering disciplines
are studying methods of effective labor organization of people employed information
processing and storage, computing, and methods of organization and interaction with
people and manufacturing facilities, their practical applications, as well as associated with
all these social, economic and cultural issues..
This formal definition does not give a complete picture of the IT industry as a whole, since it
does not enumerate specific areas of activity usually included in the concept of IT: design
and development of computer technology, software development, testing, technical support,
customer care, communication services etc.
Over the last three decades IT has made a qualitative and quantitative leap. Everything started
with the first attempts of informational support of production, bulky scientific computers and
the first simple personal computers. Over time the IT industry has made a huge step toward
with modern technologies: supercomputers, mobile computers, virtualization of calculations,
the Internet and many others. Computing performance has increased by hundreds of
thousands of times, while the cost of computers has decreased significantly, making
information technologies available to billions of people all over the planet.
The progress of IT is impressive. Things that until recently seemed fantastic, such as a
device transferring mental signals to a computer, systems of three-dimensional printing and
android robots hardly distinguishable from a human, have already been created and are
operating.
But still automation of daily human activities remains one of the main objectives of IT.
Computer systems and robotized lines perform routine, complex and dangerous operations
instead of human staff. Activity fields where it was formerly impossible to manage without
human control nowadays are operated by robots and unmanned machines operating with
jeweller accuracy.
Having freed millions of people from the necessity to work manually, at the same time IT
has opened up a huge market for intellectual labour for millions of IT specialists. Moreover,

Automation 340
the IT industry constantly acquires new areas and technologies, generating deeper division
of labour between experts.
Year on year the IT industry faces an increased array of problems, starting with
technological restrictions and finishing with solutions with moral and ethical implications.
Some problems are solved, while others constantly appear. Among them there is a big and
important issue: who can automate the work of the one who automates the work of others?
There is no universal solution. Ironic as it may be, many IT processes are extremely difficult
to be automated. As an example, the process of software development, testing and support
is quite creative work, despite the continuing tendency to automate the processes. This is
what makes programmers, testers and support staff much like street artists. The initial
requirements are a little like a client who wants a portrait painted, and the result, i.e., a
finished portrait, depends on the clients preferences and mood, as well as on the artists
style of painting.
The issue of automation of IT processes is too complex and it is extremely difficult to cover
this area completely, however, in this chapter we will try to examine the possibilities and
specific features of automation in the main areas IT.
2. Features of the IT business
Lets consider a significant difference between the IT business and a classic industry, such as
heavy industry. This is well illustrated by two giants: GM and IBM.

Company Turnover Revenue Total number of employees
IBM $ 99.9 billion $ 14.8 billion ~400,000
GM $ 137 billion $ 6.5 billion ~202,000
Table 1. Activity indicators of IBM and GM in 2010.
Both of them have comparable turnovers ($137 billion and $100 billion in 2010, respectively),
but the number of IBM employees is twice that of GM. At the same, the revenue of IBM is
twice that of GM ($14.8 billion vs. $6.5 billion in 2010, respectively).
Most of IBMs revenue is provided by Global Services. This is the largest division of IBM
(with over 190,000 employees) and is the worlds largest business and technology services
provider. Second place in IBMs revenue structure is Software. Therefore, most of the
services and products offered by IBM are intellectual products created directly by the
employees. There are no automated assembly lines, no conveyors and no additional costs
for any resources (except for electricity maybe).
At the same time, GMs main income is still provided by car sales and sales of car parts. The
production of car parts and the assembly of complete vehicles have a high degree of
automation. In addition, there are high costs for raw materials and resources (such as steel,
electricity, etc.).

Automation in the IT Field 341

Fig. 1. Structure of IBMs income in 2010.
The main difference of the IT from industrial production is that industrial production
requires complex organization of the product copies production (technological processes,
components supply and quality control at the output). For software product copies
production you only need to spend on the protection of intellectual property rights. In
addition, there is a huge market for special products (tailor-made), where products can be
used only by one or several clients without reaching the mass market. If we draw an
analogy with car manufacturing, even big IT companies are more like workshops that
produce hand-made automobiles rather than giant manufacturers.
At the same time the IT field is the driving force behind automation processes in any other
industries. The work of IT specialists can be related to automated production processes, data
collection systems, billing systems, etc. They produce products and solutions that reduce
manual labour and calculations.
All this suggests a very different approach to automation in the IT business. In the IT
business automation is an auxiliary tool. Many IT companies have a minimal set of
automated processes (or have no automated processes), but their business is still successful.
Of course IT professionals try to automate as many processes as possible.
3. Areas of the IT business
The IT business can be clearly divided into three areas:
Software
Hardware
Services

Automation 342
The software production process includes many parts which starts from programming and
finishes by software sales. Each process in the production of software looks like a brick in a
wall. If one of the bricks is destroyed or removed then the entire building (production) will
collapse. Each of these processes has its own characteristics and a degree of possible
automation.

Fig. 2. Structure of software production
Hardware production has the same types of processes as software production (with minimal
difference). Therefore, the approach to automate the development of hardware is not very
different from the automation of software production. But it also includes direct production
of components and computers (servers, network equipment, etc.). Since this process is a
classic industrial production, it will not be considered in this chapter.
Providing services is an extensive part of the IT business. Some services may be provided in
fully automatic mode, while others can only be provided directly by employees of the
service division. Examples of services will be considered more below.
4. Software production
4.1 Software testing
Usually when people involved in IT business talk about automation in software production,
they mean software testing. It is true that in software testing automation technology is most
widely used. Historically, this was due to the rapid development of IT technology from the
1990s to the present.
The constant growth in the complexity of information systems on the one hand and
appearance of new automation technologies on the other hand has pushed the development
of testing automation over the last 20 years.
Software testing includes a set of different operations, some of which are recurrent while
others are unique and product/line specific. It is quite obvious that routine recurrent
processes are much easier to be automated. Testing is the easiest part of the software

development for automation, since it consists of many such processes. Lets consider the
most typical parts of the testing process.
4.1.1 Functional testing
Functional testing can be partly automated. It especially refers to recurrent basic
functionality check-ups from one version to another, which is known as regression
testing. Such automation is only justified when the product development (several versions
production) is guaranteed and the product life cycle lasts at least one year. Besides, there are
tools that permit conducting automated user interface testing. Functional testing consists of
three phases: unit- testing, complex testing and regression testing.
4.1.1.1 Smoke-testing and unit-testing
As a rule, these tests are made by the developer. They consist of a set of calls of concrete
modules with a defined set of parameters. Usually, developers do not have enough time to
generate the necessary data sets and to make complex tests. In addition, tools which the
developer has do not allow proper control of the quality of an output. There is another
problem: while performing a module check, the developer is determined to prove the
functionality of this module (a psychological feature of most developers). The testers have
different motivation their goal is to prove that the module does not meet the accepted
requirements. The tester also looks at the working capacity of the unit from the users point
of view. Therefore, the most reasonable solution to this dilemma is to leave developers to
generate and to make smoke-testing independent, assigning the making of unit-tests to
testers. Meanwhile, in order to save testers time, developers can carry out automatic unit-
tests during a development cycle and correct errors at once, without activating the bug
tracking process and thus keeping testers free of the tens of iterations of checks.
Meanwhile, it is very important that developed unit-tests could to be used, in the same or
modified form, on complex or regression testing in the sequel.
4.1.1.2 Complex (system) testing
Complex testing tests the complete system (bundled software) to check its working capacity.
If autotests were not used at earlier stages of testing, development of autotests just for
complex testing does not always justify itself. A good case in point is the application of the
modified autotests used on unit-testing.
4.1.1.3 Regression testing
At this stage the autotests developed earlier for other stages of testing (for current or earlier
versions of the product) should be used. If the scale regression testing of system is
significant, but special regression autotests are not available, their creation makes sense only
if the full regress would be done again several times (in the case of enduring product life
cycle).
4.1.2 Load testing
Load testing is a check of the system capability in the real world. Usually, the real world is
understood like productive volume and composition of the processed data. It helps to define
the limit of system performance or compliance of performance requirements.

Automation 344
Load testing is the easiest and the most effective process to automate. Load testing uses
systems that imitate simultaneous work of thousands of users. Also to simulate the real
content of the database, testers use data generators. This is quite natural and 100% justified,
since it is pretty difficult to module the functioning of a popular Internet service or
hundreds of user calls coming into a call centre in any other way.
Stress testing. Stress testing is a kind of load testing. It is usually applied to software and
hardware complexes, and critical information systems (such as a hardware and software
complexes of aircraft). This testing includes verification of complex systems as a whole
under extreme conditions far from the operational environment. As part of this test,
resistance to extreme environmental conditions, component redundancy, alarm systems and
disaster recovery capabilities are checked. Sometimes, the conditions for a stress test can be
simulated only by automation.
Integration testing. This is type of testing verifies the interaction of the object under test with
adjacent systems or complexes. In practice, manual integration testing with external systems
is much more effective than automatic testing. Firstly, this is because the majority of man-
hours (up to 50%) within the limits of integration testing are spent on various options and
interaction adjustments between adjacent systems. Practically, this work cannot be
automated.
4.1.3 Advantages of automated testing
4.1.3.1 Cost
Decrease of man-hours spent on testing;
Reduce to a minimum time for each test-iteration after bug fixing;
Automatic creation of test-reporting, tester does not waste time on reports;
Automatic registration of issues in bug tracking systems;
After hours performance of autotests (overnight or over weekend).
4.1.3.2 Quality
Reduced human factor;
More tests per unit of time make testing more qualitative.
4.1.3.3 Productivity
More iterations in time unit;
Higher productivity than in manual testing;
Limitation on the number of tests by system performance only, not on the available
float time of a tester.
4.1.3.4 Employee motivation
In many IT companies testers have a less profitable position with respect of both prestige
and wages than developers. Naturally, some of them are not satisfied and view the position
of tester as an intermediate step in their career. Instead of the development of testing skills,
they prefer to improve their software development skills in order to switch to more
profitable positions in the current company or find a job as a developer in another company.
As a result, the quality of work of both testers and developers may be affected.

Participation in the automation of the testing process is an excellent chance for career
growth for these workers. A good specialist in the field of automated tests is more valuable
to a company than an ordinary developer or tester. In addition, they can develop their skills
and build a career without leaving the testing process or without changing their place of
work. Anyway, the making of automated tests is a more creative process than manual
testing. It brings more satisfaction to workers, increasing their motivation.
4.1.3.5 Other
Efficiency of autotesting is expressed not only in the saving of man-hours on testing. One of
the main advantages of the autotest is earlier error detection and lower total cost for their
correction.
4.1.4 Automated testing is not advantageous in the following cases
The scale of an autotests covering of the system is incorrectly estimated. In such cases,
the cost of producing autotests is many times greater than the benefit of its use;
Employees dont have appropriate qualifications. In this case, autotest quality is low
and the cost of their creation exceeds that planned;
Autotests cannot be used in a sequel. Usually, if autotests cannot be re-used as is or
with minimal modification, the cost of their creation will never be paid back (except in
special cases, e.g., load testing);
The cost of licensing the autotesting tools exceeds the economic profit engendered from
using autotesting in projects in the first place. It is necessary to remember that
autotesting tools often require separate servers that have to be purchased.
4.2 Software development
At first sight you might think that the term software development automation belies
common sense. It sounds pretty much like novel-writing automation or picture-painting
automation. But if we look closer at the process of software development over the last two
or three decades, we will see that it makes total sense. Yes, the artistic part of programming
cannot be automated by the existing means of automation, but they can provide the
developers with handy tools that multiply the speed and efficiency of their work. There is a
reason why the development tools market is worth billions of dollars and still continues to
grow ($6.9 billion in 2007, $7.3 billion in 2008 according to Gartner).
The automation of software development is generally intended for reducing the required
qualification level of the workers and for speeding up the software development process.
This is clearly shown by the programming languages, technologies and platforms evolution.
Many environments with possibilities in visual development permit minimizing the
expenses on user interface routine operations. Frameworks serve as components and
protocols connectors. Automated syntax checks, code formatting and debugging make the
programmers job significantly easier. To help developers there are tools to ensure the
normal functioning of the development team, such as version control systems, tools and
team coding. Modern means of communication and techniques of project management
allow constructing the distributed team, staffed by specialists from different countries and
areas of programming. Thus, the developers have more time for the artistic part of
programming and experience exchange.

Automation 346
The same process of natural evolution led to the creation of business integration systems
and technologies. Also these systems can hardly be called automated, but the aim of their
creation was to reduce man-hours during corporate applications development and
integration. Practically these systems made it possible to delegate part of the corporate
applications development and integration job to analysts and integrators who do not need
any special knowledge of programming languages or technologies to perform their jobs. In
addition, this helps to reduce time spent on worker communication with each other in the
course of software product development.
4.2.1 An approach to the development of a program code
The possibility of applying autotests depends very strongly on the methods of development
of a program code. Therefore, the use of test automation can strongly influence the
development. Implementation of some principles of development makes autotest using
easier and reduces time for autotest development. The features of adaptation of a program
code for autotests depend on technology and testing tools used. The following basic
manoeuvres can be used:
Option of quick rollback of changes for great volumes of data (this is especially
important for databases);
Functionality switches. Flags or the parameters that limit a set of operations of standard
modules;
Hidden parameters of a test mode, including additional logging or emulation of any
operations;
System messages (logs, warnings etc.) that should be easy for parse in the software
tools;
Option to fix a units state at the intermediate stages of work.
Testers can put up additional requirements to a program code, based on the selection of the
available tools of testing and features of the testing process. Usually, compliance to these
requirements does not consume lots of developer man-hours, but it makes autotest making
easier and decreases the number of total man-hours on development and testing. The
majority of testers requirements should probably be included as a part of the process of
program code development.
4.3 Automation of project management
It is often said about project management that it is The Art and Science of Getting Things
Done. In fact, the process of project management cannot be automated, as with any of the
arts. There are only auxiliary tools which can help the project manager, like an easel or a
palette for the artist.
Among the tools the project manager can identify the main ones are: a planning tool and
means of gathering information.
At the moment there are lots of tools for planning projects. The criterion for selecting a
planning tool is ease of use.
In general, useful information provided by the ERP system makes the work of the project
manager more automated. The keyword is useful, since just the multiplicity of
information in the ERP system is not the key to success.

At present, there are comfortable online tools of project management that combine the
following features:
Ability to manage requirements;
Possibility of automatic notification to interested parties on changes to the project;
Ease of use;
Assigning and tracking tasks;
Storing, categorization and versioning of files;
Forums to discuss tasks and projects;
Project planning;
Tracking the progress of the project by all participants;
Tracking time spent;
Report about the project state at any time;
Implementation of communication on the project with reference to the aims and
objectives (comments).
Currently, among such solutions the best known are Basecamp and TeamLab.
4.4 An adhesive force of the processes within an IT company
It makes no sense to consider further the automation of processes involved in software
development without mention of the force capable to link these processes together.
Lack of integration can negate the benefits of automation of individual processes. Therefore,
the task of information exchange between processes is one of the most important in the
construction of software production.

Fig. 3. Areas of IT business. Integrated processes of software production.
4.5 Enterprise resource planning
ERP it is an information system used to manage information flows in the organization and
to automate the various areas of the company. In 1990 Gartner Group first employed the

Automation 348
acronym ERP as an extension of material requirements planning (MRP), later manufacturing
resource planning and computer-integrated manufacturing. Not all ERP packages were
developed from a manufacturing core. Vendors of ERP solutions variously began with
accounting, maintenance and human resources. By the mid1990s ERP systems were
adjusted to all core functions of enterprises. Naturally, IT companies widely use ERP
systems to automate their processes. Moreover, many IT companies are trying to automate
all internal and external processes in order to reduce the influence of human factors on the
quality and outcome of their work.
Implementation of ERP systems allows automating the many workflows in the IT company.
Conventionally, these processes can be divided into two groups: major (productive activities
of the company) and minor (non-productive activities of the company).
Major workflows are:
Manufacturing (engineering, work orders, manufacturing process, manufacturing
projects, manufacturing flow, activity-based costs);
Project management (human resource planning, costing, billing, time and expense,
performance units, activity management, risk management);
Product management (capacity planning, workflow management, quality control, cost
management, product lifecycle management);
Customer relationship management (sales and marketing, commissions, service,
customer contact, call centre support, demand management, customer care);
Data services (self-care, knowledge base).
Minor workflows are aimed at the support of normal functioning of the company and not
directly aimed at the production or provision of services:
Finance/Accounting (general ledger, payables, cash management, budgeting);
Human resources (payroll, training, recruiting, time tracking);
Communications and business structure (business structure, integrated channels of
communication, document repository, access control, planning meetings and events);
Supply chain management (order to cash, inventory, order entry, purchasing, supply
chain planning, supplier scheduling, inspection of goods, claim processing,
commissions).
Different ERP systems can include support of different sets of workflows or components,
but the characteristic feature of all of these systems is the tight integration between the
components. These solutions also support the common protocols, interfaces and integration
tools. Thus, companies can integrate multiple solutions to meet the requirements for
automation.
There are some difficulties in the implementation of generic ERP solutions around the world
that relate to the peculiarities of doing business, the laws and taxation in each country. This
allows local markets of similar solutions to exist. Due to this, specific solutions of local
vendors can compete with solutions offered by industry giants such as SAP, Oracle or
Microsoft. A good example is the company 1C (Russia), which in 2008 took second place
(after SAP) in the ERP market in Russia, increasing its stake to 18.7% in monetary terms.
Among them, many Russian IT companies prefer to use some components of the solutions
1C for running their businesses.

4.6 Automating the minor enterprise workflows
4.6.1 Finance
Accounting for financial operations, budgeting, accounting, financial planning and other
financial operations are all critical to the business of any company. The complexity of their
implementation within the information system in part relates to complex laws (in many
countries the law requires a special licence for similar systems) and increased requirements
for accuracy and reliability; also finance is a complex subject area and without the help of
financial advisors a similar system cannot be properly developed.
For these reasons, the range of existing solutions automating finances is not so wide. The
implementation and development of such systems produces qualified professionals. Even in IT,
there are not many companies that have their own solutions of full automation for financial
operations.
4.6.2 Human resources, communications, business structure and supply chain
management
These workflows of any company (including IT) are very easily automated. To do this, there
are many solutions in complex ERP systems and in single solutions (including free
solutions). Here are just some processes of these workflows that can be automated by
existing ERP solutions:
Maintaining a register of positions and instructions;
Management of the structure of the company;
Maintaining personnel files;
Time management (tracking and registration);
Calculation of the efficiency of staff;
Training of employees;
Registration of skills, experiences and work results of staff;
Planning of meetings;
Control of room reservation;
Accounting of inventory;
Document circulation;
Maintenance of purchasing equipment and supplies;
Maintaining a register of vendors;
Recruitment of staff;
Control of authority staff (up to integration with access control system).
The main criteria for successful application of such solutions are flexibility and ease of use.
These two requirements usually mean a flexible adjustment of the instrument according the
companys business processes, user-friendly interface and easy integration with other solutions.
4.7 Integration of minor processes with major processes
Information obtained by an ERP system with the support of secondary processes is often
used to adjust the main processes and vice versa. The relationship between internal and
external processes within the companys ERP system is just as important as it is in real life.
As example, information of results and quality of work of employees are aggregated and

Automation 350
can be used for calculation of wages or benefits. Another example, the statistics obtained
from the accounting systems of work are used in project planning and information about the
skills and productivity of employees helps planning of the project.
4.8 Customer care (self-care systems)
Customer care organization is one of the most important processes. The customer care
service quality reflects a lot on the reputation of the product and the company itself. At the
same time it is quite difficult to forecast the quantity and content of the user calls which
makes it necessary for IT companies to reserve a lot of support staff or resort to the services
of external call centres. In addition, the higher the level of the customer care service, the
more expensive it gets, which in the end reflects on the price of the software product.
Customer care automated systems can handle a great part of user calls, transferring them to
operator only if necessary. The processing cost of such a self-care system, licence and
technical support inclusive, is ten times lower than having it processed by operators.
Besides, modern automated systems can integrate many channels of call processing like
email, telephone, IP telephony, SMS, Internet portals, etc., which makes it possible to create
customer care solutions with higher capacity than a standard help desk.
At the same time it is necessary to take into account that most users (45% according to
Forrester) still prefer to be attended by a real person on the telephone, rather than by an
automated system. Modern systems offer compromises in the form of IVR
1
services and
animated digital characters integrated into customer care portals. However, these systems
cannot process even half of the calls entering.
Therefore, despite the successful experience of implementation of self-care solutions in
many large companies, the developers of such systems must still solve many difficult
problems in the future, such as:
Increasing intellectualization of these systems;
Implementation of new technologies (speech recognition, the formation of visual
images, etc.);
Increase the capacity of self-care systems;
Support for customers with disabilities, etc.
The successful solution to these problems can raise the level of user confidence in these
systems, expand their range of applications and improve the quality of the automated
service. That in turn may reduce costs and increase CSI
2
of customers.
4.9 Customer relationship management
Despite the fact that the term customer relationship management (CRM) has only arisen
recently, it is based on systematic studies in marketing relationships that have been carried
out systematically for more than 20 years.
Today it is very difficult to sustain an IT business with many thousands of customers without
the help of automated customer interaction. Automated information systems CRM is now

1
IVR (Interactive Voice Response) is system of pre-recorded voice messages, performing the function of
routing calls within the call-center, using the information entered by the client using tone dialing.
2
Customer Satisfaction Index

firmly established in the business processes of many companies. IT companies are also very
widely using similar solutions for their businesses. Basically, these systems are used by
companies operating in the business to customer (B2C) sector. These companies have many
customers, a wide range of products and many partners around the world. Additionally it is
these companies who obtain greater benefit from an automation of processes of interaction with
customers. It is especially effective if such systems are integrated with the companys business
processes and existing information systems, such as self-care and bug tracking.
The implementation and development of CRM systems in companies may vary. CRM
systems are very complex and critical to the business, therefore, most non-IT companies
prefer to use third-party solutions. Only a few companies are willing to create their own
departments to develop and implement CRM solutions. However, many IT companies, with
large resources of specialists and experience in developing complex solutions, decide to
create their own solutions. Some companies start to develop their own CRM solutions based
on experience in the use of automated CRM third-party solutions.
An example of the successful implementation of a self-made solution is the experience of
Microsoft. This company has millions of customers and hundreds of thousands of partners. For
such a big company automating routine tasks is critical to ensure the efficient operation of sales.
For a long time Microsoft used a third-party CRM solution, but this solution was too complex
and inconvenient for business users. In addition, this solution has sufficiently high TCO
3
,
because it requires high maintenance costs, extension of warranty, licensing and acquisition of
equipment. The procedure of adding new features to the old platform was too long.
This had a negative effect on the companys flexibility, which is necessary in order for the
company to be at the forefront of the software industry.
To remove these problems and improve the efficiency of sales, in 2009, Microsoft
implemented Global Sales Experience (GSX) solution based on Microsoft Dynamics CRM,
Microsoft SQL Server and other Microsoft products. The effect of the implementation of the
solution has been very positive:
Time spent on training sales managers was reduced;
Total time savings amounted to about of 3 hours per week;
More user friendly interface made users CSI higher (CSI has grown by 25% as of 2009);
Total cost of ownership of the CRM system was decreased.
In 2011 Microsoft was recognized by Gartner
4
as a leader and visionary in sales force
automation.
4.10 Automation of product management
Product management is an extremely important component of any IT business. It includes
management of the image of products, pricing and management of user satisfaction.
Product management is often viewed as an integral part of CRM, but in the IT business this
process is particularly important, so it should be considered separately.

3
TCO (total cost of ownership)
4
According Gartners Magic Quadrant. Gartner rates vendors upon two criteria: completeness of vision
and ability to execute. Leaders score higher on both criteria. Visionaries score higher on the
completeness of vision only.

Automation 352
The range of tasks in product management also includes forecasts for capacity, markets and
user expectations. A proper product management strategy helps to increase user satisfaction
of the products or services of the company and helps to extend the life cycle of products. It
also avoids negative effects such as requirements death spiral
5
. In its turn it increases the
range of proposed solutions, comprehensive income and enhances the companys image.
So which tools exist for product management in IT companies? First of all, the market can
offer a lot of solutions for product lifecycle, gathering of users expectations, products for
market analysis and long-term forecasting. Typically, such solutions are already integrated
into the existing complex ERP systems or can be added to them by a simple customization.
Some single solutions also exist that can solve a separate task of product management or a
wide range of those tasks.
However, small IT companies often rely on their abilities in the automation of product
management. Automation of only a part of the process of product management allows those
companies to make proper product policy and constantly improve products. Automated tools
are also widely used to monitor usage and diagnose problems (such as those made by a third-
party and those made by the company itself). In the case of corporate clients, these tools can be
integrated with customer information systems (CRM, ERP). This can be beneficial for both the
vendor and customer (the vendors reaction time is reduced and its awareness is increased).
4.11 Integration of Information systems
The effectiveness of implementing new information systems in the IT infrastructure of the
company will drop sharply if they are not integrated with existing processes or solutions. IT
companies as owners of a wide range of information systems must solve the problem of
integration. IT companies are suitable to the tasks of integration in different ways.
4.11.1 Fragmented integration environment
The easiest way to integrate information systems is the creation of several integration
applications that can convert data streams from different systems. Each such application has a
simple logic. Its easy and cheap to develop and maintain, but has some significant drawbacks:
The complexity of the configuration. If the data format of the one of information
systems has been changed then integration application must also be changed;
The lack of uniformity in the technologies and techniques of development. Development
of information technologies and changes in staff skills occur in parallel with the
development of the IT infrastructure of a company. It often happens that in an
environment of integration there are two types of applications running simultaneously:
applications that use the old technologies and the latest solutions. This phenomenon has
been called the zoo. This phenomenon is usually harmless, except in two situations:
The company has no experts in the old technology (specialists were fired or
switched to using other technologies);
The solution based on the old technology can no longer cope with the increasing
volumes of data.

5
Customer expectations permanently warm up by advertising and promises of management. At the
same time production of goods and services cannot keep up with customer expectations.

Normally, the second problem entails the first one, since until there are no problems with
the old solution, usually no one thinks about the lack of the necessary specialists. Therefore,
this method of integrating information systems into a single space is practiced in small IT
companies (because of the economy), or giving way to a common integration environment.
4.11.2 Common integration environment
The alternative is a single integration environment that is capable of taking on the task of
integrating all of the companys information systems. This environment is the complex
solution and has following abilities:
Own internal data format;
Ability to convert the output data formats of external systems into internal data format;
Ability to convert the internal data format to output data formats of external systems;
Ability to route internal data streams;
Ability to build the simplest logical processes.

Fig. 4. Usage of enterprise service bus
This principle of interaction of information systems is called ESB (enterprise service bus).
This concept has been developed in analogy to the bus concept found in computer hardware
architecture combined with the modular and concurrent design of high-performance
computer operating systems. Motivation was to find a standard, structured and general
purpose concept for describing implementation of loosely coupled software components
(called services) that are expected to be independently deployed, running, heterogeneous
and disparate within a network.
Companies that have the ability to create these custom solutions typically use their own
unified integration environment, but many IT companies prefer to use third-party solutions
or do not use a single integration environment. The integration solutions market is

Automation 354
constantly growing (up to the 2008 crisis, the growth was about 10% per year, according to
the Forrester report, 2008), existing solutions are also continuing to develop.
Integration solutions offer a variety of flexible solutions that are suitable for large and
medium-sized businesses, but existing solutions that implement the concept of ESB have the
same drawbacks (in comparison with fragmented environment integration):
Low performance. Multiple transformations of data formats and the universalization of
the components lead to significant reduction in overall system performance;
High hardware requirements. Due to the fact that the system becomes a central part of
the entire IT infrastructure of the company, its low performance is unacceptable,
because this reduces the productiveness of all involved processes. In addition, such
systems have special requirements for equipment reliability and redundancy.
In some companies, ESB solutions coexist with specialized simple applications of integration.
The companies take these measures if eliminating the disadvantages of ESB (listed above) is
impossible. This situation occurs also during the gradual implementation of a common
integration environment. Such a principle of building the integration environment combines
the drawbacks of both approaches with all of the possible negative effects.
5. Services
Services provided by IT companies can be divided into two groups: consulting and
resources providing.
It is clear that consulting services principally cannot be provided automatically and the
production of such services is always based on human activity. The exception may be a
variety of specialized knowledge bases or guided help systems. But such services are closer
to the second group - the provision of resources.
Providing of resources is a large group, which includes the following specific services:
Hosting;
Rental computing resources;
Provision of communications services;
Internet service providing;
Data storage;
Creation and storage of special accounts;
Virtual cash;
Social networks, etc.
All of these services have one major feature - they are provided by automated systems and
solutions. The basis of any of these services is the product manufactured by the IT business.
Therefore, the chain of automation of these services starts from the development of relevant
IT products. A review of solutions and products to automate the providing of these services
is beyond the scope of this chapter.
6. Conclusion
As we can see, automation of labour is widely used in the IT field, however, due to the
inherent features of this area, there are a number of typical problems faced by the majority
of IT companies during automation of its operations.

6.1 Common problems of automation in IT field
The capabilities of automation tools are considerably inferior to the flexibility of the human
brain, so the main drawback is a necessity to add the ability of using automation tools to
specific processes. The wider use of automation in the processes of developing and
supporting software makes these processes less flexible.
Application of automation technologies reduces the influence of the human factor on the
results, however, the total cost of a bug or failure in the automation process can be much
higher than the price of an error in manual labour. This happens for several reasons: first, the
result of an automatic process is more trusted than the result of human activity; in any case,
failure of an automated process can stop all related processes, either manual or automatic.
Finally, the most important factor: automation is very expensive. Automation tools, hiring
experts in automation, the necessity to adjust the companys processes for automation all
of these requires a significant investment. For this reason, management often looks
sceptically at the benefits of automation.
IT companies often want quick profits, i.e., the investment paying off as soon as possible.
But investment in automation looks like an investment in the basic research:
It is too expensive;
It gives no obvious benefit;
It has a very long payback period.
6.2 How to minimize the impact of these problems?
Automation is not for the sake of automation, and not a passing fad. Automation offers
convenience and time saving. The implementation of automation into a companys
processes must be based on a principle that can be called RGB:
Relevant, that need to be automated only those processes that need it;
Gradual - the simultaneous and widespread introduction of automated technologies
adversely affects the stability of the workflow in the company. Automation should be
implemented from simple to complex;
Balanced - an initiative to implement various processes must come from a place where
it is required. Centralized implementation of automation processes in the company
does not bring real benefits.
Building automated processes must be done by the most experienced staff that well
understand the subject of automation. It is desirable for complex processes or calculations at
the first introduction of automation to be duplicated manually. Also, you must have a
backup scheme in case of failure.
6.3 A few words about automation tools
There is no standard tool for all possible problems in the market. Many IT areas deal with
highly tailored tasks, but even if such a tool did exist, sometimes a company, especially a small
one, could not afford it. Licensing costs for some automation products can be compared to the
monthly turnover of a small company. In such cases a company has some possible alternatives:
Continue using manual labour;

Automation 356
Buy the expensive product;
Use freeware tools or technologies;
Develop the necessary product themselves.
The choice of a free tool seems the easiest, however, it is very difficult to find a free tool to
automate a particular process, because most of them have a limited range of options.
Developing a good tool for automating is very expensive, therefore, among freeware
automation tools there are few good or universal ones available.
Development of automation tools by the company itself seems quite reasonable, but in many
cases it turns out to be a trap. The sum invested into the development of the tool can exceed
many times the cost of the ones that already exist. This is quite logical, as it is impossible to
create in just one week what has been developed by another company over many years.
However, sometimes the risk pays off.
But what can we do if there is no tool on the market that could serve for a specific problem?
Should the process be automated anyway? In any case, automation always pays off over the
long-term. So the company management has to decide on whether the company is ready to
start implementing automation right now to realize the profits in about a years time or
whether it is better to leave everything as it is. With the constantly changing market
conditions, in a year a small company could change its specialization or reconstruct the
whole development process, and therefore all the efforts would be wasted.
7. References
Definition of Information Technology. 2011, Available from:
http://definitionofinformationtechnology.com. Retrieved on 02/10/2011.
Vinnichenko, I. (2005) Automation of process of testing, Piter, ISBN, St.-Petersburg, Russian
Federation.
Enterprise Resource Planning. (October 16, 2011), In: Wikipedia The Free Encyclopedia. 2011,
Available from: http://en.wikipedia.org/wiki/Enterprise_resource_planning.
Retrieved on 12/10/2011.
Genina, N. Enterprise Management: 1C surpassed Oracle in Russia. (September 18, 2009),
In: C-News, Russian IT-Review. 2009, Available from:
http://www.cnews.ru/news/top/index.shtml?2009/09/18/362481.
Retrieved on 20/10/2011.
Smulders, C. Magic Quadrants and Market Scopes: How Gartner Evaluates Vendors Within
a Market (April 26, 2011), In: Gartner, 2011, Available from:
http://www.gartner.com/resources/154700/154752/magic_quadrants_and_mark
etsc_154752.pdf . Retrieved on 20/10/2011.
Payne, I. (June 6, 2005) Handbook of CRM: Achieving Excellence in Customer Management.
ButterworthHeinemann, 0750664370
Lawley, B.; Cohen, G. & Lowell L. (October 14, 2010) 42 Rules of Product Management: Learn
the Rules of Product Management from Leading Experts from Around the World, Super
Star Press, 1607730863
0
VHDL Design Automation
Using Evolutionary Computation
Kazuyuki Kojima
Saitama University
Japan
1. Introduction
This chapter describes the automatic generation method of VHDL which lays out the control
logic of a control system. This framework releases a designer from the work of describing the
VHDL directly. Instead, the designer inputs the equation of motion of a system and target
operation.
In this chapter, rst, FPGA, CPLD, VHDL and evolutionary computation are outlined. This
is basic knowledge required for an understanding of this chapter. Next, the framework
of automatic generation of FPGA using evolutionary computation is described. VHDL
description is expressed by several kinds of data structures called a chromosome. VHDL
expressed in a chromosome is changed using evolutionary computation and changes to a
more suitable code for controller purposes. Finally two example applications are shown.
The rst one is the controller for a simple inverted pendulum. After that, the framework
is applied to a more complicated system, an air-conditioning system. The simulation results
show that the controller automatically generated using this framework can control the system
appropriately.
2. Computer-aided controller design using evolutionary computation
2.1 FPGA/CPLD/ASIC and VHDL
CPLDs and FPGAs are both sorts of programmable LSIs. The internal logic of both can be
designed using HDL. The ASIC is one example of a device that can be designed using HDL
in the same way as programmable LSIs. CPLDs and FPGAs can be immediately evaluated on
the system for the designed logical circuit. In addition, they are exible for the rearrangement
of a specication. These merits make them suitable for the intended use in the case of a
rapid prototyping. For this reason, a CPLD is used as a controller. However, the proposed
framework is applicable to all devices that can be designed by HDL. VHDL is one of the most
popular HDLs and is therefore used in this paper.
The logic described by VHDL is veried and synthesized using a simulator or a logic synthesis
tool so that it can be written into a device. When CPLD or FPGA serves as target devices,
the programming code which determines the function of the target device can be, through a
download cable, written into it in order to obtain the target LSI easily. The VHDL for a simple
logical circuit is shown in Fig. 1.
18
library IEEE;
use IEEE.std_logic_1164.all;
entity HALF_ADDER is
port(
A,B : in std_logic;
S,CO : out std_logic);
end HALF_ADDER;
architecture DATAFLOW of HALF_ADDER is
signal C, D : std_logic;
begin
C <= A or B;
D <= A nand B;
CO <= not D;
S <= C and D;
end DATAFLOW;
library IEEE;
use IEEE.std_logic_1164.all;
entity HALF_ADDER is
port(
A,B : in std_logic;
S,CO : out std_logic);
end HALF_ADDER;
architecture DATAFLOW of HALF_ADDER is
signal C, D : std_logic;
begin
C <= A or B;
D <= A nand B;
CO <= not D;
S <= C and D;
end DATAFLOW;
A
B
S
CO
A
B
S
CO
Fig. 1. VHDL for simple logical circuit
2.2 Genetic algorithm
The genetic algorithm used as a basis of this framework is outlined below (Fig. 2). The
decision-variable vector x of an optimization problem is expressed with the sequence of N
notations s
j
(j = 1, , N) as follows:
x : s = s
1
s
2
s
3
s
N
(1)
It is assumed that symbol string is a chromosome consisting of N loci. s
j
is a gene in the
j-th locus and value s
j
is an allelomorph. The value is assumed to be a real number, a mere
notation, and so on of the group of a certain integer or a certain range of observations as an
allelomorph. The population consists of K individuals expressed with Eq. (1). Individual
population p(n) in generation n changes to individual population p(n +1) in next generation
n + 1 through the reproduction of a gene. If reproduction in a generation is repeated and
if the individual who expresses solution x nearer to an optimum value is chosen with high
probability, then the value increases and an optimum solution is obtained (Goldberg, 1989;
Koza, 1994).
2.3 Evolvable hardware
Higuchi et al. proposed evolvable hardware that regards the architecture bit of CPLD as a
chromosome of a genetic algorithm to nd a better hardware structure by genetic algorithm
(Higuchi et al., 1992). They applied this to myoelectric controllers for electrical prosthetic
hands, image data compression and so on (Kajitani et al., 2005; Sakanasi et al., 2004). This
approach features coding of an architecture bit directly for a chromosome. The designer
must determine which CPLD is used beforehand, implicitly determining the meaning of the
architecture bit. If hardware is changed for the problem of circuit scale or structure, it is
necessary to recalculate evolutionary computation.
Henmi et al. evolved a hardware description language (HDL) called structured function
description language (SFL) and applied it to robot walking acquisition (Hemmi et al., 1997).
358 Automation
VHDL Design Automation Using Evolutionary Computation 3
Crossover
Mutation
Selection
n th generation
n+1 th generation
0 1 0 0 0 1 1 1 0 1 0 0 1 0 1 0
Chromosome
Gene
x: s = s
1
s
2
s
3
s
N
p(n) p(n+1)
Crossover
Mutation
Selection
n th generation
n+1 th generation
0 1 0 0 0 1 1 1 0 1 0 0 1 0 1 0 0 1 0 0 0 1 1 1 0 1 0 0 1 0 1 0
Chromosome
Gene
x: s = s
1
s
2
s
3
s
N
p(n) p(n+1)
Fig. 2. Genetic algorithm overview
The basic motion, called an "action primitive", must be designed in a binary string. In
our approach, the designer needs only to dene I/O pins for CPLD. VHDL rather than an
architecture bit is coded directly onto the chromosome, so the chromosome structure does not
depend on CPLD scale or type, and after VHDL is generated automatically, CPLD is selected
appropriately to the VHDL scale. Pin assignment is set after VHDL generation.
2.4 Controller design framework using evolutionary computation
The study of optimizing rewritable logical-circuit ICs, such as CPLD, using a genetic
algorithmhas increased. The framework that changes an internal logic circuit ICconguration
to attain evolution is called evolvable hardware (EHW). With this framework, the designer
needs only to dene the criteria used to evaluate a controller. We explain the controller design
framework using evolutionary computation with XC95144 (Xilinx, 1998) as the test device.
Internal blocks of XC95144 are shown in Fig. 3. XC95144 is a small CPLD that has 144 pins
(117 user input-outputs), 144 macro cells and 3200 usable gates. A designer chooses input and
output signals from 117 user I/Os and assigns these pins. Each signal is dened for each I/O.
If CPLD is used in control, sensors and actuators can be associated with CPLD I/O pins (Fig.
4). Here, I/O pins are associated with two sensors and one actuator. The sensor values are
inputs to CPLD as two 16-bit digital signals and a 10-bit digital signal is output as a reference
signal to the actuator.
VHDL, which describes internal CPLD logic, is encoded on a chromosome. An example of
the VHDL generated is shown in Fig. 5, corresponding to Fig. 4. This VHDL consists of
three declarations (a) entity declaration, (b) signal declaration, and (c) architecture body.
CPLD I/O signals are dened in (a) and internal CPLD signals are in (b). Signals mainly used
in VHDL are a std_logic type and a std_logic_vector type. The std_logic type
is used when dealing with a signal alone and the std_logic_vector type is used when
dealing with signals collectively. The std_logic and the std_logic_vector types should
be combined to optimize maintenance and readability. A description of VHDL can be restored
if all of the input output signals and internal signals are used as the same std_logic type
and only the number is encoded on a chromosome. The number of input signals, output
359 VHDL Design Automation Using Evolutionary Computation
JTAG Port
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O/GCK
I/O/GSR
I/O/GTS
JTAG
Controller
In-System Programming Controller
I
/
O

B
l
o
c
k
s
F
a
s
t
C
O
N
N
E
C
T

S
w
i
t
c
h

M
a
t
r
i
x
3
1
18
54
18
54
18
54
18
54
3
1
4
Function
Block 1
Function
Block 2
Function
Block 3
Function
Block 4
Macrocells
1 to 18
Macrocells
1 to 18
Macrocells
1 to 18
Macrocells
1 to 18
117
Fig. 3. XC95144 architecture
I
/
O

B
l
o
c
k
s
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
Sensor1 ( 16 bit input )
Actuator1 ( 10 bit output )
Sensor2 ( 16 bit input )
Fig. 4. Example of CPLD application
signals and internal signals are encoded on the head of a chromosome (Fig. 6). In Fig. 4, two
input signals are set up with 16 bits and an output signal with 10 bits.
A chromosome that represents a VHDL assignment statement is shown in Fig. 7. A
chromosome structure corresponding to a process statement is shown in Fig. 8. The value
currently described is equivalent to the process statement in which "DI009" and "DI014" are
enumerated in the sensitivity list.
360 Automation
000: -------------------------------------------------------------------------------
001: -- VHDL for I/P controller generated by _hiGA_AC with _hiGA.dll
002: -- (c) 2011, Saitama Univ., Human Interface Lab., programmed by K.Kojima
003: -- Chromo#:E/50 Length:13 Inputs:32 Outputs:10 Signals:2
004: -- Simulation: Initial condition: 180.0, dt=0.0100
005: -------------------------------------------------------------------------------
006: library IEEE;
007: use IEEE.std_logic_1164.all;
008: use IEEE.std_logic_arith.all;
009: use IEEE.std_logic_unsigned.all;
010:
011: entity GA_VHDL is
012: port(
013: DI000 : in std_logic;
045: DO000 : out std_logic;
054: DO009 : out std_logic
055: );
056: end GA_VHDL;
057:
(a) LnuLy declarauon
Fig. 5. (a) Automatically generated VHDL (List-1)
The VHDL description has an if-statement inside of a process statement and the description
has two nesting levels. The hierarchy of the list structure is deep compared to the substitution
statement. When the gene of such a multilist structure is prepared, it is possible to represent
various VHDL expressions.
2.5 Variable length chromosome and genetic operations
The structure of a chromosome changes with the control design specication. The number
of internal signals is set arbitrarily and different descriptions in VHDL are expressed with
different locus lengths. The length of a chromosome is determined by the VHDL line count.
The length determines the number of internal signals enumerated on the sensitivity list or
058: architecture RTL of GA_VHDL is
059:
060: signal S000 : std_logic;
061: signal S001 : std_logic;
062:
063: begin
064:
065:
066: process(DI013) begin
067: S000 <= not DI013;
068: end process;
069:
070:
071: process(DI002, DI009) begin
072: S001 <= DI009;
073: end process;
074:
075: DO000 <= (((DI002 or not DI011) nor DI002) nand DI003);
076: DO001 <= (((((((((((((((((((((((((((((DI028 nor not DI011) or DI001) or
077: DI008) nor DI017) and not DI008) nor DI016) or not DI030) or DI025) or not
078: DI015) or not DI010) and DI010) nand not DI028) nand not DI010) and not DI012)
079: and DI008) nand DI002) or not DI009) nor not DI031) nor DI007) nor not DI009)
080: nand DI023) or DI014) nor DI019) and not DI025) nand DI024) nand not DI010) or
081: not DI031) nor not DI019) nand DI000);
082:
083: process(DI006, DI029, DI014) begin
084: DO002 <= ((not DI014 or DI029) or DI006);
085: end process;
086:
087:
088: process(DI029) begin
089: DO003 <= DI029;
090: end process;
091:
092:
094: if(DI009'event and DI009='0')then
095: DO004<=DI009 nand not DI007;
096: end if;
097: end process;
098:
099: DO005 <= ((((((((((((((((((((((((((((DI014 and DI016) nand not DI007) and
100: not DI022) nor not DI005) nand DI018) nor DI021) and not DI000) nand not DI013) or
101: not DI022) or DI030) or not DI027) and DI031) and DI020) or DI023) nor not DI025)
102: or not DI000) nor DI002) nand DI002) and not DI027) nand not DI017) nand not
103: DI025) nand not DI007) nor DI031) and DI012) nand not DI012) or not DI030) nand
104: not DI026) and not DI011);
105: DO006 <= ((((((((((((((((((((((((DI010 and not DI028) nand DI009) nor DI011) nand
106: not DI016) or DI029) nand not DI001) and DI011) or DI017) or DI011) nand not
107: DI030) nor DI022) or not DI005) or DI001) and DI019) nor DI027) nand DI010) or not
108: DI025) or not DI023) and DI019) or not DI019) and DI023) nor not DI008) or not
109: DI005) nand not DI021);
110: DO007 <= DI009;
111:
112: process(DI025, DI013, DI023) begin
113: DO008 <= not DI023;
114: end process;
115:
116:
118: DO009 <= DI014;
119: end process;
120:
121:
122: end RTL;
(b) Slgnal declarauon
(c) ArchlLecLure body
(d) AsslgnmenL
sLaLemenL
(e) rocess sLaLemenL
Fig. 5. (b) Automatically generated VHDL (List-2)
the length of the right-hand side of an assignment statement. When dealing with such a
variable length chromosome, the problem is that genetic operations will generate conict on
a chromosome. To avoid this problem, we prepared the following restrictions:
1. With a top layer, the length of a chromosome is equal to the number that added one to the
summary of the number of internal signals and the number of output signals.
2. All signals are encoded on a chromosome using a reference number.
362 Automation
Number of inputs
Number of outputs
Number of signals
Chromosome
1st locus
32
10
2
Fig. 6. Signal denition on the rst locus
Chromosome
n th locus
Architecture body
Reference No.
Command ID
ID
ID_CMD_SUBSTITUTION
Left term
35
1st Right term
Reference No. not
Right terms
Reference No. not Operator
3
0
1 12 ID_OP_OR
Fig. 7. Gene structure for assignment statement
3. The signal with a large reference number is described by only the signal whose reference
number is smaller than the signal.
4. The top layer of a chromosome describes the entity declaration using all internal signals
and output signals in order with a low reference number. Each signal can be used only
once.
5. Crossover operates on the top layer of a chromosome.
These restrictions avoid the conict caused by genetic operations.
Figure 9 shows an example of crossover operation. The back of the 6th gene is chosen in this
example. Chromosome (A) and chromosome (B) cross and change to chromosome (A) and
chromosome (B). Only the gene before and behind the crossover point of each chromosome
Chromosome
n th locus
Architecture body
Command ID
ID
ID_CMD_PROCESS_SENSITIVITY
Sensitivity list
10
Process body
ID_CMD_PS_IFTHENELSE
[0] [1]
15
ID
Command ID
Reference No.
Event
Edge
Substitution
Reference No.
Left term
38
1st Right term
Reference No. not
Right terms
Reference No. not Operator
10
0
1 8 ID_OP_NAND
10
1
0
Fig. 8. Gene structure for if-then-else statement in process body
show the gene of a lower layer. In the gure, chromosome (A) has two sensitivity lists and
chromosome (B) has two assignment statements. The structure of a chromosome changes by
replacing the gene from the back of a top gene to before a crossover point. Both chromosome
(A) and chromosome (B) came to have an assignment statement and a sensitivity list.
3. Application to HDL-based controller of inverted pendulum
In this section an application to an HDL-based controller for the inverted pendulum is
described (Kojima, 2011).
364 Automation
Chromosome (A)
Crossover point
Crossover point
Crossover point
Chromosome (B)
Chromosome (A)
Chromosome (B)
Crossover point
ID
5 16
ID ID
5 16
2 16 2 16
5 16 5 16
2 16 2 16
Chromosome (A)
Crossover point
Crossover point
Crossover point
Chromosome (B)
Chromosome (A)
Chromosome (B)
Crossover point
ID
5 16
ID ID
5 16
2 16 2 16
5 16 5 16
2 16 2 16
Fig. 9. Crossover operation
l
z
c
1

c
2
z
f
d
m, J
M
Fig. 10. Inverted pendulum
3.1 Equations of motion
Figure 10 shows the model of the inverted pendulum. The equations of motion are given by
(ml
2
+ J)
= mgl sin ml z cos c

1

+ d (2)
(M + m) z = f c
2
z (3)
CPLD
A/D
A/D
Motor Driver
Inverted
Pendulum
10
Duty
PWM
16
16
Position
Angle
z
Fig. 11. Control system block diagram
I
/
O

B
l
o
c
k
s
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
Angle ( 16 bit input )
Duty ( 10 bit output )
Position ( 16 bit input )
Fig. 12. CPLD pin assignment
f = Kv (4)
where [rad] is the angle of the pendulum, m [kg] is the mass of the pendulum, M [kg] is the
mass of the cart, J [kgm
2
] is the inertia of the rod, l [m] is the length between the gravity point
and fulcrum, z [m] is the position of the cart, d [Nm] is the disturbance torque, c
1
[Ns/rad] is
the viscous resistance of the pendulum, c
2
[Ns/m] is the viscous resistance of the cart, f [N]
is the controlling force, v is the parameter for the control input and K is the gain.

J, a and b are
dened as:
J = J + ml
2
(5)
a =
c
2
M + m
(6)
b =
K
M + m
(7)
366 Automation
then, equations (2) and (3) become:
=
mgl
J
sin
ml z
J
cos
c
1
+
d
J
(8)
z = a z + bv. (9)
Each parameter can be determined by measurement and experiment as l = 0.15, m = 46.53
10
3
,

J = 1.58 10
3
, c
1
= 2.05 10
2
, a = 4.44 and b = 2.46 10
1
. Using these parameters,
evolutionary simulations are conducted.
3.2 Control system and CPLD pin assignment
Figure 11 shows the control system block diagram. The CPLD is used as the controller of the
inverted pendulum. The position of the cart and the angle of the pendulum are converted
16 bit digital signals respectively and input to the 32 CPLD pins. The control logic in the
CPLD, which is formed using the framework previously mentioned, determines the 10 bit
control signal driving the motor of the inverted pendulum. Figure 12 shows the CPLD pin
assignment for this application. In this case, 32 bit parallel inputs and 10 bit parallel outputs
are adopted. Instead of using them, we can use serial I/Os connected with A/D converters
and D/A converters. In this case, serial to parallel logics should be formed in the CPLD and
even though serial inputs are used, automatically generated VHDL can be used as a VHDL
component.
3.3 Fitness
The tness value is calculated as a penalty to the differences in the rod angle and the cart
position.
tness =
1
30
0
||dt
30
0
|z|dt (10)
Disturbances are given as a random torque during the control simulation. When calculating
tness value, disturbance torque is always initialized. Therefore, all individuals are given
different disturbance at each evaluation. This kind of disturbance makes the controller robust
to various disturbances.
3.4 Simulation
Simulations are conducted under two conditions (1)
0
= 1
, z
0
= 0, (2)
0
= 180
, z
0
=
0. Population size is 50, mutation rate is 0.5, crossover rate is 1.0, tournament strategy,
tournament size is 10 and the elite strategy is adopted.
Figures 13 and 14 show the simulation results. Figure 13 shows the result of condition (1),
Figure 14 shows the result of condition (2). (a) the result at 0 generation and (b) the result
at 1000 generation are represented respectively. The angle of the rod, the position of the
cart, disturbance and the control signal are shown at each generation. At zero generation,
in both conditions (1) and (2), the obtained controller cannot control the inverted pendulum
adequately. The rod moves in a vibrating manner at around 180
. At 1000 generation,
the controller controls the inverted pendulum successfully in both conditions. Further, in
condition (2), swing up motion can be observed (Fig.14(b)). The control signal at 1000
generation has more various patterns than the signal at 0 generation.
(a) 0th generation
(b) 1000th generation
Fig. 13. Simulation results (
0
= 1
, z
0
= 0 )
368 Automation
(a) 0th generation
Fig. 14. Simulation results (
0
= 180
, z
0
= 0 )
4. Application to an air-conditioning controller
Next, the framework is applied to an air-conditioning controller. Simulation model, task
denition and tness function are described in this section. Simulation results will be shown
at the end of the section (Kojima et al., 2007; Kojima, 2009).
4.1 Simulation model
In the targeted air-conditioning system(Fig. 15), the outside air imported is cooled or warmed
and sent to the console to adjust temperature. Air entering from the inlet (a) is cooled by the
refrigerator (c). Part of the cooled air is warmed with a heater (e), then mixed with cool air to
adjust the temperature. The angle of the mixture door (d) controls the mixing ratio of warm
and cold air. Mixed air is sent to console (f), changing the indoor temperature. The system
controller controls blower motor rotation and the angle of the air mixture door.
Outdoor
(b) Blower (d) Air mixture door
(e) Heater
(f) Console
(c) Refrigerator
5 80
Full cool 0%
Full hot 100%
Air temperature 30
Humidity 50%
500 m
3
/h max.
(0)
(2)
(1)
(3)
(4)
(5)
(6)
(8) (7)
(a) Inlet
Outdoor
(b) Blower (d) Air mixture door
(e) Heater
(f) Console
(c) Refrigerator
5 80
Full cool 0%
Full hot 100%
Air temperature 30
Humidity 50%
500 m
3
/h max.
(0)
(2)
(1)
(3)
(4)
(5)
(6)
(8) (7)
(a) Inlet
Fig. 15. Air-conditioning system
Fig. 16. Control volumes
To evaluate the controller performance, an air-conditioning simulation model is required in
evolutionary computation. We consider an air-conditioning model combining the console and
duct using nine control volumes (Fig. 16). Temperature and humidity, but not compressibility,
are considered in this model. To calculate the predicted mean vote (PMV), which indicates
thermal comfort, we require globe temperature, metabolization, ow rate, insulation of
clothes and external work.
Figure 16 shows three control volumes in this system. We assume that air ows toward
the control volume indexed with (i + 1) from the control volume indexed with (i). Air
temperature T
i+1
[K] and relative humidity
i+1
vary with temperature T
i
[K], relative
370 Automation
humidity
i
, mass ow G
i
[kg/s] and heat transfer Q[kJ/s]. In air-conditioning simulation,
mass ow is proportional to the opening of blower motor .
G
0
= G
i
= G
max
(0 1) (11)
where G
max
is maximum ow at full blower opening. Flow rate G
0
is the sum of air ow G
a0
and steam ow G
w0
.
G
0
= G
a0
+ G
w0
(12)
Water vapour pressure P
w0
is as follows:
P
w0
=
0
P
s0
(13)
where P
s0
saturation water vapour pressure from Tetens formula (Tetens, 1930):
P
s0
= 610.78 e
17.2694(T
0
273.15)
(T
0
273.15)+238.3
(14)
Specic humidity x
0
is given by water vapour pressure P
w0
.
x
0
= 0.622
P
w0
P
0
P
w0
(15)
Air mass ow G
a0
is calculated from the following gas equation:
G
a0
= P
a0

V
R
a
T
0
(16)
where R
a
is a gas constant. Water vapour mass ow G
w0
is given by:
G
w0
= x
0
G
a0
(17)
Air ow rate G
a
is constant. Steam ow G
w
is constant except during dehumidication.
Considering the air ow into control volume (i) from control volume (i 1) in unit time
dt[s], temperature T
i
, humidity x
i
and mass of control volume M
i
after dt is given as follows:
T
i
=
G
i1
C
i1
T
i1
dt + (M
i
G
i
dt)C
i
T
i
G
i1
C
i1
dt + (M
i
G
i
dt)C
i
+
Q
i
dt
M
i
C
i
(18)
x
i
=
M
i
x
i
+ (1 + x
i
)(G
wi1
G
wi
)dt
M
i
+ (1 + x
i
)(G
ai1
G
ai
)dt
(19)
M
i
= M
i
+ (G
i1
G
i
)dt (20)
where specic heat C
i
[kJ/kg K] is given by
C
i
=
1.005 + x
i
{(2501.6/T
i
) + 1.859}
1 + x
i
(21)
At the mixture door, air is divided into two ows. The ratio of the mass of divided air depends
on the mixture door angle. Here G
1
is the mass owat location (1) in Fig. 15, G
2
that at location
(2) and G
3
that at location (3),
G
2
= G
1
(22)
G
3
= (1 )G
1
(23)
where (0 1) is the opening ratio of the mixture door. Adding two mass ow rates
enables us to calculate the downstream mass at a juncture. Here G
4
is the mass owat location
(4) in Fig.15, G
5
that at location (5) and G
6
that at location (6),
G
6
= G
4
+ G
5
(24)
Temperature and humidity at a juncture are given in the same way as for when three control
volumes are considered.
4.2 Predicted mean vote (PMV)
PMV is the predicted mean vote of a large population of people exposed to a certain
environment. PMV represents the thermal comfort condition on a scale from -3 to 3, derived
from the physics of heat transfer combined with an empirical t to sensation. Thermal
sensation is matched as follows: "+3" is "hot." "+2" is "warm." "+1" is slightly warm." "0" is
"neutral." "-1" is slightly cool." "-2" is "cool." "-3" is "cold." Fanger derived his comfort equation
from an extensive survey of the literature on experiments on thermal comfort (Fanger,
1970). This equation contains terms that relate to clothing insulation I
cl
[clo], metabolic
heat production M[W/m
2
], external work W[W/m
2
], air temperature T
a
[
C], mean radiant

temperature T
r
[
C], relative air speed v[m/s] and vapour pressure of water vapour P[hPa].
PMV = {0.33 exp(0.036M) + 0.028}
(M W)
3.05{5.73 0.007(M W) P}
0.42{(M W) 58.1}
0.0173M(5.87 P)
3.96 10
8
f
cl
{(T
cl
+ 273.15)
4
(T
mrt
+ 273.15)
4
}
f
cl
h
c
(T
cl
T
a
)
(25)
f
cl
is the ratio of clothed and nude surface areas given by:
f
cl
= 1.0 + 0.2I
cl
(I
cl
0.5)
f
cl
= 1.05 + 0.1I
cl
(I
cl
> 0.5) (26)
where T
cl
is the clothing surface temperature given by repeated calculation of:
T
cl
= 35.7 0.028(M W)
0.155I
cl
3.96 10
8
f
cl
{(T
cl
+ 273.15)
4
(T
mrt
+ 273.15)
4
} + f
cl
h
c
(T
cl
T
a
)
(27)
372 Automation
where h
c
is the heat transfer coefcient,
h
c
= max{2.38(T
cl
T
a
)
0.25
, 0.0121
v} (28)
and T
mrt
is mean radiant temperature. PMV is detailed in (Fanger, 1970).
4.3 Task denition
The task is to adjust PMV in the console despite heat transfer from outside changes. The
air-conditioning controller controls the opening ratio of blower n and the opening ration of
mixture door m appropriately.
4.4 Control system and CPLD pin assignment
Figure 17 shows the control system block diagram. PMV is input to the CPLD as a 8 bit signal.
The control logic in the CPLD determines the 8 bit blower opening control signal and the 8 bit
mixture door control signal. Figure 18 shows the CPLD pin assignment for this application.
PMV
8
8
8
Blower opening
Mixture door opening
CPLD
I
/
O

B
l
o
c
k
s
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
PMV ( 8 bit input )
Mixture door ( 8 bit output )
Blower motor ( 8 bit output )
I
/
O

B
l
o
c
k
s
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
I/O
PMV ( 8 bit input )
Mixture door ( 8 bit output )
Blower motor ( 8 bit output )
4.5 Fitness
The tness function is as follows:
tness =
t
end
0
|PMV
ctrl
PMV
target
|dt (29)
where t
end
is the end of simulation time. The difference between target and controlled PMV is
integrated as a penalty in the controller simulation.
4.6 Simulation results
Blower
Mix-door
(a) 0th generation
Blower
Mix-door
Blower
Mix-door
(c) 10000th generation
Fig. 19. Simulation results
Figure 19 shows the simulation results. PMV in a console is fed back to the controller. In the
graphs, trends change every 60 seconds. Variations are based on the load effect change each
60 seconds. At zero generation (Fig. 19 (a)), temperature rises or falls with the change in heat
load. PMV also changes simultaneously. This means that simply optimizing a controller is
not enough. After 100 generations of calculation, the difference between the target and the
374 Automation
estimated value decreases (Fig. 19 (b)). In the 10,000th generation (Fig. 19 (c)), tolerance
decreases further. These results show that hardware corresponding to the purpose is obtained
automatically using this framework.
Figure 20 shows simulation results under other conditions. Nine graphs result for three
thermal loads. Three graphs temperature, PMV and control are shown for each thermal
condition. The controller used under three conditions was obtained after 10,000 generations
of calculation. Although thermal load and load change timing were random, the blower and
mix door were controlled so that PMV is set to zero.
Blower
Mix-door
Blower
Mix-door
Blower
Mix-door
Fig. 20. Simulation results
5. Conclusion
In this chapter, to automate controller design, CPLD was used for controller data-processing
and VHDL to describe the logical circuit was optimized using the genetic algorithm. Two
example cases (an inverted pendulum and an air-conditioner) were shown and we conrmed
this framework was able to be applied to both systems. Since this framework is a generalized
framework, so in these kinds of systems which process data from some sensors and drive
some actuators, this framework will work functionally.
6. References
Balderdash, G., Nol, S., & Parisi, D. (2003). Evolution of Collective Behavior in a Team of
Physically Linked Robots, Proceedings of the Applications of Evolutionary Computing,
EvoWorkshops 2003, pp.581-592, Tbingen, April 2009.
Barate, R. & Manzanera, A. (2008). Evolving Vision Controllers with a Two-Phase Genetic
Programming System Using Imitation, Proceedings of the 10th international conference
on Simulation of Adaptive Behavior: From Animals to Animats, pp.73-82, ISBN:
978-3-540-69133-4, Osaka, July 2008.
Fanger, P. O. (1970). Thermal Comfort, McGraw-Hill.
Goldberg, D. E. (1989). Genetic Algorithm in Search, Optimization and Machine Learning,
Addison-Wesley.
Hemmi, H., et al., (1997). AdAM: A Hardware Evolutionary System, Proc. 1997 IEEE Conf.
Evolutionary Computat.(ICEC97), pp.193-196.
Higuchi, T. et al. (1992). Evolvable Hardware with Genetic Learning: A First Step Towards
Building a Darwin Machine, Proceedings of the 2nd International Conference on the
Simulation of Adaptive Behavior, MIT Press, pp.417-424.
Kajitani,I. & Higuchi, T. (2005). Developments of Myoelectric Controllers for Hand Prostheses,
Proceedings of the Myoelectric Controls Symposium 2005, pp.107-111, Fredericton,
August 2005.
Kojima, K. et al. (2007). Automatic Generation of VHDL for Control Logic of Air-Conditioning
Using Evolutionary Computation, Journal of Advanced Computational Intelligence and
Intelligent Informatics, Vol.11, No.7, pp.1-8.
Kojima, K. (2009). VHDL Design Automation Using Evolutionary Computation, Proceedings
of 2009 International Symposium on Industrial Electronics (IEEE ISIE 2009), pp.353-358,
Seoul, July 2009.
Kojima, K. (2011). Emergent Functions of HDL-based Controller of Inverted Pendulum in
Consideration for Disturbance, Proceedings of 2011 IEEE/SICE International Symposium
on System Integration, Kyoto, December 2011.
Koza, J. (1994). Genetic Programming, A Bradford Book, ISBN0262111705.
Sakanashi, H. et al., (2004). Evolvable Hardware for Lossless Compression of Very High
Resolution Bi-level Images, IEEE Proceedings-Computers and Digital Techniques,
Vol.151, No.4, pp.277-286, Ibaraki, August 2004.
Tetens, O. (1930). Uber einige meteorologische Begriffe, Zeitschrift fur Geophysik, Vol.6, pp.297-309.
(in German)
wyns10] Wyns, B., et al., (2010). Evolving Robust Controllers for Corridor Following
Using Genetic Programming, Proceedings of the International Conference on Agents and
Articial Intelligence, volume 1 : articial intelligence; pp.443-446, Valencia, January
2010.
Xilinx. (1998). DS056(v.2.0) XC95144 High Performance CPLD Product Specication, Xilinx; 2.
376 Automation
19
Applicability of GMDH-Based Abductive
Network for Predicting Pile Bearing Capacity
Isah A. Lawal
1
and Tijjani A. Auta
2

1
University of Genoa
2
ATBU, Bauchi
1
Italy
2
Nigeria
1. Introduction
A pile is a type of foundation commonly used in civil construction. They are made using
reinforced concrete and pre tensioned concrete to provide a firmer base where the earth
around a structure is not strong enough to support a conventional foundation [Pile, 2011].
Accurate prediction of the ultimate bearing capacity of a structural foundation is very
important in civil and construction engineering. Conventional method of estimating the pile
bearing capacity has been through pile load test and other in situ test such as standard
penetration test and cone penetration test [Bustamante & Gianeselli, 1982]. Though these
tests may give useful information about ground conditions, however the soil strength
parameters which can be inferred are approximate.
In recent time, advances in geotechnical and soil engineering research have presented more
factors that can affect pile ultimate bearing capacity. However, due to nonlinearity of these
factors, the use of statistical model analysis and design has proved difficult and impractical [Lee
& Lee, 1996]. So there is need to provide civil and structural engineers with intelligent assistance
in the decision making process. Soft computing and intelligent data analysis techniques offer a
new approach to handle these data overload. They automatically discover patterns in data to
provide support for the decision-making process. Tools used for performing such functions
include: Artificial Neural Networks (ANN) [Abu- keifa, 1998; Chow et al, 1995; Teh et al, 1997],
Support Vector Machine (SVM) [Samui, 2011],Genetic programming [Adarsh, in-press] and
Gaussian process regression [Pal & Deswal, 2008, 2010]. The results from using these tools
suggest improved performances for various datasets. However, neural networks like other tool
suffer from a number of limitations, e.g. long training times, difficulties in determining
optimum network topology, and the black box nature with poor explanation facilities which do
not appeal to structural engineers [Shahin et al, 2001]. This research work proposes the use of
self-organising Group Method of Data Handling (GMDH) [Mehra, 1977] based abductive
networks machine learning techniques that has proved effective in a number of similar
applications for performing modelling of pile bearing capacity.
Recently, abductive networks have emerged as a powerful tool in pattern recognition,
decision support [ElSayed & AbdelAal, 2008], classification and forecasting in many areas
[Abdel-Aal, 2005, 2004]. Inspired by promising results obtained in other fields, we explore
the use of this approach for the prediction of pile bearing capacity.

Automation 378
2. Related work
In recent years some researchers have developed computational intelligence models for the
accurate prediction of pile bearing capacity. In [Lee and Lee, 1996], the authors used back-
propagation neural networks to predict the ultimate bearing capacity of piles. A maximum
error of prediction not exceeding 20% was obtained with the neural network model
developed by using data set generated from calibration chamber. Also, in [Pal,2008], the
author investigates the potential of support vector machines based regression approach to
model the static pile capacity from dynamic stress-wave data set. The experiments shows
excellent correlation coefficient between the predicted and measured values of the static pile
capacity investigated. Similarly, in [Samui, 2011], the author studied the potential of Support
Vector Machine (SVM) in prediction of bearing capacity of pile from pile load data set. In
the study the author introduces -insensitive loss function and the sensitivity analysis of the
model developed shows that the penetration depth ratio has much effect on the bearing
capacity of the pile.
However, in [Pal, 2010], the author took a different approach and investigated the potential
of a Gaussian process (GP) regression techniques to predict the load-bearing capacity of
piles. The results from the study indicated improved performance by GP regression in
comparison to SVM and empirical relations. However, the author reported that despite the
encouraging performance of the GP regression approach with the datasets used, it will be
difficult to conclude if the method can be used as a sole alternative to the design methods
proposed in the literature. The reason been that soft computing based modelling techniques
are data-dependent. Their results may change depending on the dataset, the scale at which
the experiments are conducted or the number of data available for training.
The potential for GMDH-based abductive network in pile bearing capacity prediction has
not been explored before in the literature. However, compared to neural networks and other
learning tools, the method offers the advantages of faster model development requiring
little user intervention, faster convergence during model synthesis without the problems of
getting stuck in local minima, automatic selection of relevant input variables, and automatic
configuration of the model structure.
3. GMDH and AIM abductive networks
Abductory Inductive Mechanism (AIM) is a powerful supervised inductive learning tool for
automatically synthesizing network models from a database of input and output values
[AbTech, 1990]. The model emerging from the AIM synthesis process is a robust and
compact transformation implemented as a layered abductive network of feed-forward
functional elements as shown in Figure 1. An abductive network model numerical input
output relationships through abductive reasoning. As a result, the abductive network can be
used effectively as a predictor for estimating the outputs of complex systems [Lee, 1999], as
a classifier for handling difficult pattern recognition problems [Lawal et al, 2010] or as a
system identifier for determining which inputs are important to the modelling
system[Agarwal. 1999]. With the model represented as a hierarchy of polynomial
expressions, resulting analytical model relationships can provide insight into the modelled
phenomena, highlight contributions of various inputs, and allow comparison with
previously used empirical or statistical models.

Applicability of GMDH-Based Abductive Network for Predicting Pile Bearing Capacity 379
3.1 Abductive machine learning
The abductive machine learning approach is based on the self-organizing group method of
data handling (GMDH) [Fallow, 1984]. The GMDH approach is a proven concept for
iterated polynomial regression that can generate polynomial models in effective predictors.
The iterative process involves using initially simple regression relationships to derive more
accurate representations in the next iteration in an evolutionary manner.

Fig. 1. Abductive network showing various types of functional elements
The algorithm selects the polynomial relationships and the input combinations that
minimize the prediction error in each phase. This prevents exponential growth in the
polynomial model generated. Iteration is stopped automatically at a point in time that
strikes a balance between model complexity for accurate fitting of the training data and
model simplicity that enables it to generalise well with new data. In the classical GMDH-
based approach abductive network models are constructed by the following 6 steps [Fallow,
1984]:
i. Separating the original data into training data and testing data.
The available dataset are split into training dataset and testing dataset. The training dataset
is used for estimating the optimum network model and the testing dataset is used for
evaluating the network model obtained on the new data. Usually a 70-30 splitting rule is
employed on the original data, but in this work a pre-determined split used by earlier
published work was adopted to allow direct comparison of results.

Automation 380
ii. Generating the combinations of the input variables in each layer.
Many combinations of r input variables are generated in each layer. The number of
combinations is p!((p -r)! r!). Here, p is the number of input variables and the value of r is
usually taking as 2.
iii. Calculating the partial descriptors
For each input combination, a partial descriptor which describes the partial characteristics of
the model is calculated by applying regression analysis on the training data. The following
second order polynomial regression relationship is usually used
y
k
= b
0
+b
1
x
+ b
2
x
]
+b
3
x
]
+ b
4
x
2
(1)

The output variables y
k
in Eq. (l) are called intermediate variables.
iv. Selecting optimum descriptors.
The classical GMDH algorithm employs an additional and independent selection data for
selection purposes. To prevent exponential growth and limit model complexity, the
algorithm selects only relationships having good predicting powers within each phase. The
selection criterion is based on root mean squared (RMS) error over the selection data. The
intermediate variables which give the smallest root mean squared errors among the
generated intermediate variables (y
k
) are selected.
v. Iteration
Steps III and IV are iterated where optimum predictors from a model layer are used as
inputs to the next layer. At every iteration, the root mean squared error obtained is
compared with that of the previous value and the process is continued until the error starts
to increase or a prescribed complexity is achieved. An increasing root mean squared error is
an indication of the model becoming overly complex, thus over-fitting the training data and
will more likely perform poorly in predicting the selection data.
vi. Stopping the multi-layered iterative computation
Iteration is stopped when the new generation regression equations start to have poorer
prediction performance than those of the previous generation, at which point the model
starts to become overspecialized and, therefore, unlikely to perform well with new data.
Computationally, the resulting GMDH model can be seen as a layered network of partial
descriptor polynomials, each layer representing the results of iteration. Therefore, the
algorithm has three main elements: representation, selection, and stopping. Figure 2 shows
the flow chart of the classical GMDH-based training.
Abductory Inductive Mechanism (AIM) is a later development of the classical GMDH that
uses a better stopping criterion that discourages model complexity without requiring a
separate subset of selection data. AIM adopts a well-defined automatic stopping criterion
that minimizes the predicted square error (PSE) and penalises model complexity to keep the
model as simple as possible for best generalization. Thus, the most accurate model that does
not overfit the training data is selected and hence a balance is reached between accuracy of
the model in representing the training data and its generality which allows it to fit yet
unseen new evaluation data.


Fig. 2. The flow chart of the classical GMDH-based training.
The PSE consists of two terms [AbTech, 1990]:
PSE = FSE + KP (2)
Where FSE is the average fitting squared error of the network for fitting the training data
and KP is the complexity penalty for the network, expressed as [AbTech, 1990].

Automation 382
KP = CPH (2o
2
N) K (3)
Where CPM is the complexity penalty multiplier, K is the number of coefficients in the
network, and
2
is a prior estimate of the model error variance. Usually, a complex network
has a high fitting accuracy but may not generalize well on new evaluation data unseen
previously during training. Training is automatically stopped to ensure a minimum value of
the PSE for the CPM parameter used, which has a default value of 1. The user can also
control the trade-off between accuracy and generality using the CPM parameter. CPM
values greater than 1 will result in less complex models that are more likely to generalise
well with unseen data while values less than the default value will result in a more complex
models that are likely to over fit training data and produce poor prediction performance.
Figure 3 shows the relationship between PSE, FSE and KP, [AbTech, 1990].

Fig. 3. The predicted square error
3.2 AIM functional elements
The used version of AIM supports several functional elements [AbTech, 1990], see Figure 1,
including:
Normaliser: Transforms the original input into a normalized variable having a mean of
zero and a variance of unity.
y = z
0
+z
1
x (4)
Where x is the original input, y is the normalized input, z
0
and z
1
are the coefficients of the
normaliser
Unitizer: Converts the range of the network outputs to a range with the mean and variance
of the output values used to train the network.

Single Node: The single node only has one input and the polynomial equation is limited to
the third degree, i.e.
y = z
0
+ z
1
+z
2
x
2
+z
3
x
3
(5)
Where x is the input to the node, y is the output of the node and z
0
, z
1
, z
2
and z
3
are the node
coefficients.
Double Node: The double node takes two inputs and the third-degree polynomial equation
includes cross term so as to consider the interaction between the two inputs, i.e.
y = z
0
+z
1
x
+ z
2
x
]
+z
3
x
2
+ z
4
x
]
2
+z
5
x
x
]
+ z
6
x
3
+z
7
x
]
3
(6)
Where x
i
, x
j
are the inputs to the node, y is the output of the node and z
0
, z
1
, z
2
and z
7
are
the node coefficients
Triple Node: Similar to the single and double nodes, the triple node with three inputs has a
more complicated polynomial equation allowing the interaction among these inputs.

Fig. 4. Non-representative training data
However, not all terms of an elements equation will necessarily appear in a node since AIM
will throw away or carve terms that do not contribute significantly to the solution. The eligible
inputs for each layer and the network synthesis strategy are defined as a set of rules and
heuristics that form an integral part of the model synthesis algorithm as described earlier.
On a final note, any abductive network model is only as good as the training data used to
construct it. To build a good model it is important that the training database be representative
of the problem space. Figure 4 [AbTech, 1990], illustrates a scenario where the training
database used to create the AIM model does not cover an important portion of the problem.
Training AIM using only the data to the left of the dotted line will result in a model that
generalizes well within the training data range but will be inaccurate in the other region.

Automation 384
4. The dataset and feature discussion
To evaluate and compare the performance of the proposed approach we used the
experimental dataset developed in [Lee & Lee, 1996]. The dataset were generated from a
calibration chamber, in which field stress conditions were simulated with poorly graded,
clean, fine and uniformly layered sand that was dried in air below 2% of water content. The
Sand was deposited in the calibration chamber using a method that allows a uniform sand
deposit of known relative density to be obtained. The setup was allowed to settle for 24
hours before the model pile was driven into it using a guided steel rod (hammer). The
ultimate bearing capacities of the model pile were assumed to be affected by the penetration
depth ratio of the model pile, the mean normal stress and the number of blows. So the
dataset consist of the following features: penetration depth ratio (i.e. penetration depth of
pile/pile diameter), the mean normal stress of the calibration chamber and the number of
blows as input and the ultimate capacity (kN) as the output. More detailed description of
the dataset generation experiment can be found in [Lee & Lee, 1996].
5. Experiments and results
This section describes the development of abductive networks model for predicting the
ultimate bearing capacity using the experimental dataset described in section 3. To allow
direct comparison of results, the same splitting used by earlier published work using the
dataset was adopted. Two experiments were conducted. In the first experiment, the 28
instances in the dataset were split into a training set of 21 instances and an evaluation of 7
instances. While in the second experiment, 14 instances were selected for training purposes
and 14 instances for evaluation. The full training set was used to synthesize an abductive
network model with all the 3 features present in the dataset enabled as network inputs. The
best model was obtained by adjusting the CPM value. The effect of the CPM values on the
models performance is shown in Table 1. The numbers (e.g. Var_4) indicated at the model
input represent the feature selected as input to the model during training, while Var_6
represent the network output corresponding to ultimate pile bearing capacity. It is noted
that the model uses 2 inputs out of the 3 inputs which indicates that almost all the features
are relevant, with only little redundancy in the feature set
To measure the performance of the trained model after evaluation, two statistical measures
namely Root Mean Squared Error (RMSE) and Correlation Coefficient (R
2
) were used. A
brief description and mathematical formulae are shown below:
5.1 Root Mean-Squared Error
The root mean square error (RMSE)) is a measure of the differences between values
predicted by a model and the values actually observed from the phenomenon being
modeled or estimated. Since the RMSE is a good measure of accuracy, it is ideal if it is small.
This value is computed by taking the square root of the average of the squared differences
between each predicted value and its corresponding actual value.
The formula is:
RHSE = _
| (x
1
-
1
)
2
+ (x
2
-
2
)
2
++ (x
i
-
i
)
2
]
n
(7)

Where x
and y
are the predicted and actual values respectively while n is the size of the
data used.

Model CPM Values
Number of
input features
selected
during model
synthesis
Model
Performance
with training
dataset in
experiment 1

0.5
2 out of the 3
features
RMSE = 34.11 kN
Correlation = 0.98

1
2 out of the 3
features
RMSE = 70.22 kN
Correlation = 0.91

2.5
2 out of the 3
features
RMSE = 70.22 kN
Correlation = 0.91
Table 1. The effect of the CPM values on the synthesized models performance
5.2 Correlation coefficient
A correlation coefficient is measure that determines the degree to which two variable's
movements are associated. It gives statistical correlation between predicted and actual
values. This coefficient is unique in model evaluations. A higher number means a better
model, with a value of one (1) indicating a perfect statistical correlation and a value of zero
(0) indicating there is no correlation.
Corrclotion Cocicicnt =
(
c
-
c
1
)(
p
-
p
1
)
_|(
c
-
c
1
)
2
] (
p
-
p
1
)
2
(8)

Automation 386
Where y
u
and y
p
are the actual and predicted values while y
u
1
and y
p
1
are the mean of the
actual and predicted values.
In experiment 1, the best model with CPM = 0.5 was evaluated using the evaluation sets. A
satisfactory agreement between the predicted and measured values of the ultimate pile
bearing capacity was obtained, which is shown by the cross plots in Figure 5. The maximum
error of prediction was 18%. A Root Mean Squared Error (RMSE) and a correlation
coefficient (R
2
) of 59.22kN and 0.82 were obtained respectively in the first experiment as
shown in Table 2.

Fig. 5. The cross plot of predicted and measured values of the pile bearing capacity for
experiment 1
In the second experiment, after training the abductive network with just 14 instances, the
best model was evaluated with the evaluation set of 14 instances. The cross plots in Figure 6
shows the trained and the predicted values of the ultimate bearing capacity. The result
showed widely scattered plots, with a RMSE of 92.5kN and Correlation Coefficient of 0.83.
The reason for the poor result in this case can be attributed to the small number of the
training data which was not enough for the model to learn the entire pattern in the data set.
Therefore, it could be concluded that a certain number of training data sets was needed to
obtain reasonable predictions.


Fig. 6. The cross plot of predicted and measured values of the pile bearing capacity for
experiment 2
Finally, the maximum prediction error of other modeling tool used previously on this
dataset was compared with that of abductive model synthesized in this work as shown in
Figure 5. The comparison indicates that the abductive network model performed much
better in terms of prediction error, with almost 9% improvement compared to that obtained
with Neural Network reported in [Lee & Lee 1996].

Experiment 1 Experiment 2

Training set
(21 instances)
Evaluation set
(7 instances)
Training set
(14 instances)
Evaluation set
(14 instances)
RMSE (kN) 34.11 59.22 70.2 92.5
Correlation(R
2
) 0.98 0.82 0.91 0.83
Table 2. Performance measure of the two experiments on the training and test data set
0
100
200
300
400
500
600
700
800
900
0 200 400 600 800 1000
P
r
e
d
i
c
t
e
d

U
l
t
i
m
a
t
e

P
i
l
e

B
e
a
r
i
n
g

C
a
p
a
c
i
t
y

(
k
N
)
Measured Ultimate Pile Bearing Capacity (kN)
Evaluated Trained

Automation 388

Fig. 7. Comparison of GMDH-based abductive network model with other previous model
used on this dataset in term of Maximum Prediction Error (%)
6. Conclusion
This work demonstrates the use of abductive machine learning techniques for the prediction of
pile bearing capacity. A RMSE value of 59.22kN and 92.5kN and a correlation coefficient of 0.82
and 0.83 were obtained with respect to the pile bearing capacity values predicted in two
separate experiment conducted respectively. An improvement of almost 9% in terms of
prediction error was recorded. This result indicated that the proposed abductive network
approach yields a better performance compared to the other already implemented technique
using the same dataset mentioned in the introduction section. However, the experiments
conducted revealed that for a good prediction, a large number of training set is required to train
the model before evaluation. So, to validate the performance of the abductive network approach
it is recommended that data set obtained from the fields are use in further studies. This will help
realize the full potential of abductive network approach in pile bearing capacity prediction.
Meanwhile, the work has outlined the advantages of abductive networks and has placed it
in the perspective of geotechnical engineering problem computing point of view. Thus,
researchers are encouraged to consider them as valuable alternative modeling tool.
Hopefully, future work will consider the possibility of extending the approach to modelling
of soil behaviour and site characterization.
7. Acknowledgement
The authors will like to acknowledge Dr. R. E. Abdel-Aal of the Department of Computer
Engineering, King Fahd University of Petroleum & Minerals for providing the tool used in
this study.
17
18
19
20
Abductive
Network Model
with 21 training
data
Neural Network
Model with 21
training data
Neural Network
Model with 14
training data
M
a
x
i
m
u
m

P
r
e
d
i
c
t
i
o
n

E
r
r
o
r

(
%
)

8. References
AbdelAal, R.E. (2004). ShortTerm Hourly Load Forecasting Using Abductive Networks,
IEEE Transaction on Power Systems, vol. 19, No.1, pp. 164 -173
Abdel-Aal, R.E. (2005). Improved classification of medical data using Abductive network
committees trained on different feature subsets, Computer Methods and Programs in
Biomedicine, vol. 80, no.2, pp. 141-153
AbTech, Corporation (1990). AIM users manual, Charlottesville, VA
Abu-Kiefa, M. A. (1998). General regression neural networks for driven piles in cohesionless
soils, Journal of Geotechnical & Geoenviromental Engineering, Vol. 124, No.12,
Adarsh, S.; Dhanya,R. Krishna,G. Merlin,R. & Tina,J. (in press). Prediction of Ultimate
Bearing Capacity of Cohesionless soils using Soft Computing Techniques, Journal of
ISRN Artificial Intelligence.
Agarwal, A. (1999). Abductive networks for two-group classification: A comparison with
neural networks. The Journal of Applied Business Research, Vol. 15, No. 2, pp. 112.
Bustamante, M. & Gianeselli, L. (1982). Pile bearing capacity prediction by means of static
penetrometer CPT. Proceeding of the second European symposium on penetration testing,
Amsterdam, May
Chow, Y.K; Chan, W.T. Liu, I.F. & Lee, S.L. (1995). Predication of pile capacity from stress
wave measurements: a neural network approach, International Journal of Numerical
& Analytical Methods in Geomechanics, Vol. 19, pp. 107126
ElSayed, E.M., AbdelAal, R.E. (2008). Spam Filtering with Abductive Networks,
International Joint Conference on Neural Networks, Hong Kong
Farlow, S. J. (1984). The GMDH algorithm. In Self-organizing methods in modeling: GMDH type
algorithms, S. J. Farlow (Ed.), pp. 124, New York: Marcel-Dekker
Lawal, I. A.; Abdel-Aal, R. E. & Mahmoud, S. A. (2010). Recognition of Handwritten Arabic
(Indian) Numerals Using Freeman's Chain Codes and Abductive Network
Classifiers, IEEE 20th International Conference on Pattern Recognition, Istanbul,
pp.1884-1887
Lee, B. Y; Liu, H. S. & Tarng, Y. S. (1999). An Abductive Network for Predicting Tool Life in
Drilling, IEEE Transactions on Industry Applications, Vol. 35, No. 1, January
Lee, I. & Lee, J. (1996). Prediction of Pile Bearing Capacity Using Artificial Neural Networks,
Journal of Computers and Geotechnics, Vol. 18, No. 3, pp. 189-200,
Mehra, R.K. (1977). Group Method for Data handling (GMDH): Review and Experience,
Proceedings of the IEEE Conference on Decision and Control, New Orleans, pp. 2934
Pal, M. & Deswal, S. (2008). Modeling Pile Capacity Using Support Vector Machines and
Generalized Regression Neural Network, Journal of Geotechnical and Geoenviromental
Engineering, Vol.134, No. 1021 pp. 134-7
Pal, M. & Deswal, S. (2010). Modeling pile capacity using Gaussian process regression,
Journal of Computers and Geotechnics, Vol. 37, pp. 942947,
Pile, (2011).In Encyclopeadia Britannica, Retrieved from
http://www.britannica.com/EBchecked/topic/460377/pile pp. 1177-1185
Samui, P. (2011). Prediction of pile bearing capacity using support vector machine,
International Journal of Geotechnical Engineering, pp. 95-102,
Shahin, M. A.; Jaksa, M. B. & Maier, H. R. (2001) Artificial Neural Network Applications In
Geotechnical Engineering, Australian Geomechanics, Vol. 36, No.1, pp. 49-62

Automation 390
Teh, C.I; Wong, K.S. Goh, A.T.C. & Jaritngam, S. (1997). Prediction of pile capacity using
neural networks, Journal of Computing Civil Engineering Vol. 11, No.2, pp. 12938.
0
An End-to-End Framework for Designing
Networked Control Systems
Alie El-Din Mady and Gregory Provan
Cork Complex Systems Lab (CCSL), Computer Science Department
University College Cork (UCC), Cork
Ireland
1. Introduction
Designing a control system over Wireless Sensor/Actuator Network (WSAN) devices
increases the coupling of many aspects, and the need for a sound discipline for
writing/designing embedded software becomes more apparent. Such a WSAN-based control
architecture is called a Networked Control System (NCS). At present, many frameworks
support some steps of the NCS design ow, however there is no end-to-end solution that
considers the tight integration of hardware, software and physical environment. This chapter
aims to develop a fully integrated end-to-end framework for designing an NCS, from
system modelling to embedded control-code generation. This framework aims to generate
embedded control code that preserves the modelled system properties, and observes the
hardware/software constraints of the targeted platform.
Existing approaches for control code design typically ignore the embedded system
constraints, leading to a number of potential problems. For example, network delays and
packet losses can compromise the quality of control that is achievable Mady & Provan
(2011). Designing embedded control that accounts for embedded system constraints requires
dealing with heterogeneous components that contain hardware, software and physical
environments. These components are so tightly integrated that it is impossible to identify
whether behavioural attributes are the result of computations, physical laws, or both working
together.
Contemporary embedded control systems are modelled using hybrid system Henzinger
(1996) that captures continuous aspects (e.g. physical environment) and discrete-event
behaviour (e.g. control decision). Even though many tools support model-based code
generation (e.g. Simulink), the emphasis has been performance-related optimizations, and
many issues relevant to correctness are not satisfactorily addressed, including: (a) the precise
relationship between the model and the generated code is rarely specied or formalized;
(b) the generated code targets a specic embedded platform and cannot be generalized to
multi-targeted platforms, moreover there is no generator considers the embedded WSAN; (c)
the generated code does not respect the targeted platform hardware/software constraints; (d)
the continuous blocks are either ignored, or discredited before code generation. Therefore, the
correspondence between the model and the embedded code is lost.
20
This chapter proposes a framework for designing embedded control that explicitly accounts
for embedded system constraints. We develop an end-to-end framework for designing an
NCS where we model the system using a hybrid systems language. We focus on adopting
a distributed control strategy that explicitly considers hardware/software constraints. Our
approach enables us to generate code for multiple embedded platforms.
Fig. 1 shows an overview of the framework. The framework consists of four design
stages, outlining the role for each stage as follows: (i) the reference model captures the
control/diagnosis
1
strategies and the physical plant using a hybrid system; (ii) the control
model is a projection of the source model, in which the physical/continuous aspects are
abstracted/transformed to a generic control specication; (iii) the embedded control model
combines the control model and the hardware/software constraints to dene an embedded
control model satisfying the platform constraints; (iv) the target embedded model, that
captures the embedded platform code, is generated from the control model considering the
hardware/software constraints.
Consequently, the reference and control models abstract the hardware/software constrains
effect on the control algorithm. These constrains can be classied to: (a) processing resources
constrains; and (b) memory space constrains. The processing resources constrains consider the
hardware/software factors that affect the algorithm execution time (e.g., CPU speed), which
can lead to incompatibility between the processing capacity and the algorithm execution.
Whereas, the memory space constrains check if there is enough memory space for executing
the control algorithm.
Fig. 1. Framework Overview
An NCS is widely used in many applications, such as habitat monitoring, object tracking, re
detection and modern building control systems. In particular, Building Automation Systems
(BAS) often uses a large wireless/wired sensor network. BAS is selected as our application
domain as it is considered as a cornerstone application for decreasing energy consumption;
around 40%of total energy use in the West is consumed in the industrial building sector, which
accounts for nearly one-third of greenhouse gas emissions. Of this gure, approximately
1
In this chapter, we consider only the control aspects in the reference model.
392 Automation
An End-to-End Framework for Designing Networked Control Systems 3
one-third can be attributed to Heating, Ventilation and Air-Conditioning (HVAC) systems
present in buildings. In this chapter, we consider an Air Handling Unit (AHU) control system
as a case study for the framework development. We apply the framework design stages on
this case starting from the reference model to the embedded control code.
We assume that we are given as input a set R of top-level system requirements, for example
user comfort requirements that any set-point will differ from operating value by less or equal
to 10%. We provide empirical guarantees that the requirements R are met by the generated
models and the corresponding transformation rules, by empirically evaluating the system
property for each generated model (i.e. reference, control, target model). If the system
property is respected for each design stage model, then the models are transformed correctly.
Our contributions in this chapter will be as follows:
we formulate an end-to-end framework for designing a network control system.
This framework preserves the modelled systems properties under hardware/software
constraints;
we identify the transformation rules between the framework design stages;
we formulate a typical AHU control system as a case-study of the framework;
we empirically check that our framework preserves the systems requirements R by
applying it to an Air Handling Unit (AHU) model as a case-study.
The remainder of the chapter is organized as follows: Section 2 provides a survey covering the
related work and discusses our contribution comparing to the state-of-the-art. The framework
architecture and the description for each design stage modelling are discussed in Section 3.
The model transformation rules between the design stages are explained in Section 4. The
application domain for the case-study are highlighted in Section 5, and its experiments design
is shown in Section 6. We end in Section 7 by giving a discussion of our work and outlining
future perspectives.
2. Related work
Modelling Frameworks: Modeling languages dene a representation method for expressing
system design. Given the heterogeneity of engineering design tasks, modelling languages
consequently cover a wide range of approaches, from informal graphical notations (e.g., the
object modelling technique (OMT) Rumbaugh et al. (1991)), to formal textual languages (e.g.,
Alloy for software modelling Jackson (2002)).
Semantic meta-modelling is a way to uniformly abstract away model specicities while
consolidating model commonalities in the semantics meta-model. This meta-modelling
results in a mechanism to analyze and design complex systems without renouncing the
properties of the system components. Meta-modelling enables the comparison of different
models, provides the mathematical machinery to prove design properties, and supports
platform-based design.
An abstract semantics provides an abstraction of the systemmodel that can be rened into any
model of interest Lee & Sangiovanni-Vincentelli (1998). One important semantic meta-model
393 An End-to-End Framework for Designing Networked Control Systems
framework is the tagged signal model (TSM) Lee & Sangiovanni-Vincentelli (1998), which can
compare system models and derive new ones.
There are several prior studies on the translational semantics approach. For example, Chen
et al. (2005) use the approach to dene the semantic anchoring to well-established formal
models (such as nite state machines, data ow, and discrete event systems) built upon AsmL
Gurevich et al. (2005). Further, they use the transformation language GME/GReAT (Graph
Rewriting And Transformation language) Balasubramanian et al. (2006). This work, through
its well-dened sets of semantic units, provides a basis for similar work in semantic anchoring
that enables for future (conventional) anchoring efforts.
Tools Several embedded systems design tools that use a component-based approach have
been developed, e.g., MetaH Vestal (1996), ModelHx Hardebolle & Boulanger (2008),
Model-Integrated Computing (MIC) Sztipanovits & Karsai (1997), Ptolemy Lee et al. (2003),
and Metropolis Balarin et al. (2003). These tools provide functionality analogous to the
well-known engineering tool MATLAB/Simulink. In particular, Metropolis and Ptolemy
II are based on semantic metamodelling, and hence obtain the entailed abstract semantics
(and related abstract metamodels) of the approach. In these tools, all models conforming to
the operational versions of the TSMs abstract semantics also conform to the TSMs abstract
semantics. One drawback of these tools is that components" can be assembled only in the
supporting tool. As a consequence, different systems and components must all be developed
in the same environment (tool) to stay compatible. However, the most recent version of
Metropolis, Metropolis II, can integrate foreign tools and heterogeneous descriptions.
One tool that is closely related to our approach is the Behaviour Interaction Priority (BIP)
tool Basu et al. (2006). BIP can combine model components displaying heterogeneous
interactions for generating code for robotics embedded applications. BIP components are
described using three layers, denoting behaviour, component connections and interaction
priorities. Our approach focuses more on the higher-level aspects, in that it uses two levels of
meta-model (i.e., meta-model and meta-meta-model) to dene all underlying specications.
Moreover, our approach considers the consistency check between the system model and the
hardware/software constraints for the embedded platform.
To our knowledge, our approach is unique in its use of two levels of meta-models, a
single centralized reference model with a hybrid systems semantics, and its generation of
embeddable code directly from the centralized meta-model considering hardware/software
constrains.
3. Modelling framework architecture
In this section, we provide a global description for modelling framework. In addition, each
design stage of our framework is formulated.
3.1 Modeling objectives
Our objective is to abstract the essential properties of the control generation process so that we
can automate the process. It is clear that the process has two quite different types of inputs:
394 Automation
Control constraints The control-theoretic aspects concern sequences of actions and their
effects on the plant. These constraints cover order of execution, times for actions to be
executed, and notions of forbidden states, etc. Note that these constraints assume innite
computational power to actually compute and effect to stated control actions.
Embedded System constraints These aspects concern the capabilities of the hardware and
software platforms, and are independent of the applications being executed on the
platform.
It is clear that both property types are needed. Because of the signicant different between the
two types, we must use different constraint representations for each type. As a consequence,
we then must use a two-step process to enforce each constraint type. This is reected in the
fact that we have a two-step model-generation process. Step 1 maps from a reference model
R
to a control model
C
using mapping rules !
C
; the second step maps
C
to
E
using
mapping rules !
E
. If we represent this transformation process such that
C
= f (
R
, !
C
), and
E
= f (
C
, !
E
), then we must have the full process represented by
E
= f ( f (
R
, !
C
), !
E
).
3.2 Meta-meta-model formal denition
The meta-meta-model is formulated using a typical hierarchal component-based modelling
Denckla & Mosterman (2005), as shown in Fig. 2. We can dene a meta-meta-model as
= C, . C represents a set model components, where control components C
C
, plant
components C
P
and building-use components C
B
are C instances, i.e., C C
C
, C
P
, C
B
. The
connection between the components C are described by .
Each component c C is represented as c = P
io
, , C, where P
io
2
is a set of component
input/output ports and describes the relation between the input and output ports.
Moreover, c can contain a set of components C to represent hierarchical component levels.
Relation can be expressed based on the use (application domain) of the framework. In
this article, we consider four meta-model instances of :
R
,
C
,
E
,
T
, where
R
is
the reference meta-model,
C
is the control meta-model,
E
is the embedded meta-model
and
T
is the target embedded meta-model. Each of these meta-models captures the DSML
abstraction for the framework design stage. In this case, for each design stage we can
use any Domain-Specic Modelling Language (DSML) that can be represented using the
corresponding meta-model. In this article we have selected the DMSLs that support designing
NCS for BAS system. However, this design methodology can be adapted to match any other
application domain.
3.3 Reference meta-model formal denition
In the context of BAS modelling, Hybrid Systems (HS) are used to create our reference model
R
. We have used HS to present BAS models as it captures both discrete (e.g. presence
detection) and continuous (e.g. heat dissipation) dynamics, where the continuous dynamic
is represented using algebraic/differential equation. Hence, Linear Hybrid Automata (LHA)
become a suitable HS candidate for this model.
2
In our framework, we assume that the used ports are unidirectional ports.
System
Component Connection
I/O Port
Relation
0..* 1..*
0..*
1
0..*
0..*
0..*
read write
Reference
Model
Control Model
0..1 0..1
Embedded
Model
Target Model
0..1 0..1
N
e
t
w
o
r
k

T
o
p
o
l
o
g
y
0..* 1..*
Composite Primitive
Fig. 2. Framework Meta-Meta-Model
The reference meta-model is considered as one instance of the component relation =
R
. As
shown in Fig. 3, we represent
R
as a standard LHA (
R
= H) representation, as described in
Def. 3.1.
Denition 3.1. [Linear Hybrid Automata] A linear hybrid automaton H is a 3-tuple, i.e. H =
V, M
s
, E, with the following structural extensions:
V is a nite set of component binding variables. It is used to bind/connect the component ports
P
io
with . V is represented as following:

V = v
1
, ..., v
n
for real-valued variables, where n is
the dimension of H.

V = v
1
, ..., v
n
represents the rst derivatives during continuous change.
V = v
1
, ..., v
n
represents values at the conclusion of discrete change.
M
s
is a set of hierarchical system-level modes that describe the system statuses. m
s
M
s
captures
the system-level status using either H or single mode m, i.e., m
s
H, m. m can be diagnosis
mode m
d
or control mode m
c
, i.e. m m
d
, m
c
. We assume that m
c
controls nominal system
behaviour and does not consider fault modes (e.g., fault actuation). Two vertex functions assigned
to each mode m M or m
s
M
s
. Invariant (inv(m)) condition is a predicate whose free variables
396 Automation
0..1
Reference
Model
Binding
Variable
Switch Mode
Linear Hybrid
Automata
1
1..*
0..*
0..*
Invariant
Algebraic
Equation
Differential
Equation
Event Guard Flow Dynamic
Probabilistic
Guard
0..1 0..*
0..* 0..*
0..1
0..1
0..1
Fig. 3. Meta-Model for the Framework Reference Model
are from V and ow dynamic f low(m) is described using algebraic equation f low
alg
(m) and/or
differential equation f low
di f
(m) for

V

V binding variables.
E is set of switches (edges). An edge labelling function j that assigns to each switch e E a
predicate. E is also assigned to a set of events, where event is executed if the corresponding
predicate j is true, where j can be presented using probabilistic or deterministic predicate. Each
jump/guard condition j J is a predicate whose free variables are from

V

V. For example,
e(m
i
, m
l
) is a switch that moves from mode m
i
to m
l
under a guard j then executes , i.e.,
e(m
i
, m
l
) : m
i
j/
m
l
.
3.4 Control meta-model formal denition
The control meta-model
C
aims to capture the control components C
C
in a discrete behaviour.
Therefore, Fig. 4 shows a Finite-State Machine (FSM) F that used to describe the components
relation =
C
. In Def. 3.2, we formally dene the control meta-model using FSM, i.e.,
C
= F.
Denition 3.2. [Finite-State Machine] A nite-state machine F is a 3-tuple, i.e. H = V, S, T,
with the following structural extensions:
V is a nite set of component binding variables. It is used to bind/connect the component ports P
io
with . V is represented as discrete change only.
S is a set of states that used to identify the execution position.
T is set of transitions to move from one state s to another. Similar to E in LHA, a set of events
(actions) and jump/guard condition J are assigned to T. For example, a transition t(s
i
, s
l
) T is
Control Model
Binding
Variable
Transition State
Finite State
Machine
1
1..*
0..*
0..*
Action Guard
Probabilistic
Guard
0..1
0..1
0..1
Fig. 4. Meta-Model for the Framework Control Model
used to reect the move from state s
i
to s
l
under a guard j (i.e., deterministic or probabilistic) then
executing , i.e., t(s
i
, s
l
) : s
i
j/
s
l
.
3.5 Embedded meta-model formal denition
In order to check the consistency between our control model and the hardware/software
constrains, we identify the embedded model following its corresponding meta-model
E
shown in Fig. 5. One standard modelling language that can present our embedded model
is Analysis and Design Language (AADL)
3
(i.e., SAE Standard). However, any other
language/tool that can capture
E
elements can be used in the hardware/software consistency
check.
Denition 3.3. [Embedded Model] An embedded model
E
is a 5-tuple, i.e.
E
=
Mem, Pro, Bus, Thr, Dev, with the following structural extensions:
Mem is a RAM memory identication used in the embedded platform during the control algorithm
execution. This memory contains 3-tuple used to identify the memory specication, i.e., Mem =
Spc, Wrd, Prt, where Spc identies the memory space size, Wrd identies the memory word size,
and Prt identies the memory communication protocol (i.e., read, write, read/write).
Pro is a processor identication used in the embedded platform to execute the control algorithm. In
E
, the Pro specication is identied using the processor clock period Clk, i.e., Pro = Clk.
Bus is a bus identication used to communicate between different hardware components, such as
Mem and Pro. This Bus contains 2-tuple to identify its specication such as the bus latency Lcy
and massage size Msg, i.e., Bus = Lcy, Msg.
3
http://www.aadl.info/
398 Automation
Memory
Embedded
Model
0..*
Processor Bus Device Thread
Space Size Word Size Protocol Clock Period Latency Message Size
Dispatch
Protocol
Period
Execution Time
1
1
1
1
1
1
1
0..*
0..*
0..*
1..*
Read Write Read/Write
1
Periodic Sporadic
1
Communication
Device
Sensing Device
Dispatch
Protocol
Period
Periodic Sporadic
1
0..1
0..1
0..1
1 1
0..*
0..*
Fig. 5. Meta-Model for the Framework Embedded Model
Thr is a thread identication used to run/execute the control algorithm over Pro. This Thr contains
a 2-tuple to identify its specication, such as the dispatch protocol Dprt and execution time Tex,
i.e., Thr = Dprt, Tex. The Dprt can be either sporadic or periodic. In case of periodic dispatch
protocol, we need to identify the period for the thread switching.
Dev is a device identication used to sense/actuate the physical plant, or activate (sending or
receiving) the wireless communication devices. In Dev, we identify dispatch protocol Dprt to
trigger Dev, i.e., Dev = Dprt. The Dprt can be either sporadic or periodic. In case of periodic
dispatch protocol, we need to identify the Dev activating period.
We can deduce that the hardware/software constraints needed for the consistency check are:
1. Memory space size.
2. Memory word size.
3. Memory communication protocol.
4. Processor clock period (CPU).
5. Bus latency.
6. Bus massage size.
7. Each thread execution time.
8. Each thread switching protocol.
9. Device activation/re protocol.
The embedded model evaluates:
1. The binding consistency between the hardware and software description, considering the
hardware/software constrains.
2. The processing capacity, i.e., how much the described processor is loaded with the
execution for the described software. This evaluation factor must be less than 100%,
otherwise the designer has to increase the processing resources (e.g., number of the
processors).
3.6 Target embedded meta-model formal denition
The target embedded model presents the embedded code that will be deployed on the
embedded platform. Fig. 6 depicts the meta-model
T
for the target embedded model.
This
T
captures the primitive building elements that needed to describe an embedded code.
Consequently, we can use any embedded language by identifying the corresponding semantic
for each
T
element. In this article, we focus on embedded Java for Sun-SPOT
4
sensors and
base-station.
Sensor/Actuator
Node
Target
Embedded Model
0..1
Switch
Condition Action State
Send
Assignment Sensing Actuating
Wired
Command
Receive
Wireless
Command
Send Receive Sensor Type Actuator Type
Thread
Property
Base Station
Switch
Condition Action State
Send
Assignment
Receive
Wireless
Command
Thread
Property
0..1
1..*
1..*
0..*
0..*
0..*
1..*
1..*
0..* 0..*
1..*
0..*
0..*
1..* 1..* 1..* 1..*
1..* 1..*
0..* 1..*
0..*
1..*
1..*
Binding
Variable
1..*
Binding
Variable
1..*
Fig. 6. Meta-Model for the Framework Targeted Model
The target embedded meta-model can formally be described as in Def. 3.4.
Denition 3.4. [Target Embedded Model] An embedded model
T
is a 2-tuple, i.e.
T
= N
S
, N
B
,
where N
S
= T, Thr, V and N
B
= T, Thr, V are the semantic denitions needed in case of sensor
node and base-station deployment, respectively, with the following structural extensions:
T is a set of switch semantic denition. Considering a switch t T, t = j, s, , where j is the
predicate that allows executing the action , and s the a Boolean variable to identify the execution
position. State s is included in the semantic of j and , as following:
if (state && condition){
action;
assignment(s);
}
The action semantic can contain wireless/wired commands, typical assignment, and/or
sensing/actuation commands. However, at this stage some hardware/software constrains are needed
4
http://www.sunspotworld.com/
400 Automation
in order to identify the sensor/actuator type and the sensors ID, which identied from the network
topology. For example, in case of sending a wireless data from sensor node, the semantic is as
follows:
try {
SendDg.reset();
SendDg.writeInt(NodeID);
SendDg.writeDouble(DataToBeSent);
SendConn.send(SendDg);
}
catch (Exception e) {
throw e;
}
The thread Thr is used to identify the semantics of a thread as follows.
5
public class Thread_name extends Thread{
public void run(){
while(true/SynchCond){
Switches semantics;
}
}
}
Thread_name.sleep(period);
V is a nite set of binding variables that can be identied as Integer, Double, Boolean.
Regarding the base-station node N
B
, it has the same aforementioned description with less
actions. Typically, N
B
is used to communicate wirelessly between the sensors and the actuator,
therefore its action contains the wireless command and assignment semantics only.
4. Model transformation
In this section, we provide the transformation rules R used to transform an instance of
the source relation to the corresponding instance of the target representation. As mentioned
before, the developed framework aims to generate control code, therefore R considers the
control components C
C
(i.e., sensor, actuator, controller, base-station). The connections and
Ports P
io
that identied in the meta-meta-model remain the same. However, each model
has a corresponding representation for and P
io
. For example, in the reference model the
components can be connected through a variable, whereas in the target model the connection
can be a wireless channel.
5
If there is only one thread under execution, the thread doesnt have a critical role in this case. Moreover,
Thr has a period property to be identied.
4.1 Reference model to control model transformation
The transformation rules R
C
transforms a
R
model to a
C
model. For simplifying the
transformation rules R
C
, we have used the same symbols for the unchanged terms between
R
and
C
. Using R
C
, the control model can be generated from the reference model as
C
= f (
R
, R
C
), where R
C
abstracts the ow dynamics, invariants, and hierarchy in
R
. In
this section, we have highlighted the transformation rules as follows:
Binding variables V in H are transformed to an equivalent format in F, i.e., V
H
R
C
V
F
. In
this case, the continuous change variables in H will be updated discretely in F.
The control modes M
c
in H are transformed to states S in F, i.e., M
H
c
R
C
S
F
. In this
case, the states S capture the positions of M
c
without its operations (e.g. invariant, ow
dynamics). This transformation assumes that there is no fault control modes, therefore the
diagnosis modes M
d
(presenting faulty modes) is not transformed.
The switches E in H (E
H
) are transformed to transitions T in F, i.e., E
H
R
C
T
F
.
The invariant inv(m) and ow dynamics equations f low(m) for each mode in H are
transformed in F to a transition from/to the same state, i.e., t(s, s) : s
j/
s. The transition
guard j is triggered if the mode invariant is true. In order to avoid activating multiple
transitions at the same time, we add (logical and) to j the set
`
J of the inverse guards
for the transitions moving out from state s, i.e., j = inv(m)

`
J, where
`
J is the guards
of the transitions leaving s, i.e., t(s, ). This multiple transitions activation can happen if
the invariant and any other transition guard are true, which is acceptable in H but not
in F. The ow dynamics equations f low(m) are transformed to the transition action ,
i.e., f low(m)
R
C
. However, differential equations f low
di f
(m) in the form of

x(t) = f
are approximated to discrete behaviuor using Eulers method, i.e., x(k + h) = x(k) + f h,
where k is the step number and h is the model resolution that reects the time step (as
much h is small, as the approximation is accurate).
These rules can be summarised as: f low(m), inv(m)
R
C
t(s, s) : s
inv(m)
`
J/f low(m)
s.
The aforementioned rules do not consider the hierarchical feature in the system-level modes
M
s
. Given a system-level mode m
s
M
s
that contains a set of inherited control modes M
c
,
i.e., m
s
_ M
c
. An initial control mode m
init
is identied in M
c
, m
init
is the rst mode under
execution whenever m
s
is triggered. The transformation of these hierarchical modes is as
follows:
The set of control modes M
c
in m
s
is transformed to a set of states S
c
. Consequently, m
init
is transformed to s
init
.
The union of: (a) the set of switches E
H
c
that enters m
init
, i.e., E
H
c
(, m
init
); and (b) the set of
the switches E
H
s
that enters m
s
, i.e., E
H
s
(, m
s
) is transformed to a set of transitions T
c
in F
that enters the initial state s
init
, called T
F
c
(, s
init
). Therefore, E
H
c
(, m
init
) E
H
s
(, m
s
)
R
C
T
F
c
(, s
init
).
402 Automation
The set of switches E
H
c
that enters m
c
M
c
, i.e., E
H
c
(, m
c
) is transformed to a set of
transitions T
c
in F that enters s
c
S
c
s
init
6
, i.e., T
F
c
(, s
c
). Therefore, E
H
c
(, m
c
)
R
C
T
F
c
(, s
c
).
The set of switches E
H
c
that exits m
c
, i.e., E
H
c
(m
c
, ), is transformed to a set of transitions
T
c
in F that exits s
c
S
c
, called T
F
c
(s
c
, ). Therefore, E
H
c
(m
c
, )
R
C
T
F
c
(s
c
, ). However,
in order to give a higher priority to the system transitions in case of activating control
and system transitions at the same time, we introduce the inverse of the condition for the
system transition to J of T
F
c
(s
c
, ), i.e., J(T
F
c
(s
c
, )) = J(E
H
c
(m
c
, )) J(E
H
s
(m
c
, )).
The set of the switches E
H
s
that exits m
s
, i.e., E
H
s
(m
s
, ) is transformed to be added to the
set of transitions T
c
in F that exits s
c
S
c
. Therefore, E
H
s
(m
s
, )
R
C
T
F
c
(s
c
, ).
The invariants inv(m
c
) and inv(m
s
), and ow dynamics equations f low(m
c
) and f low(m
s
)
for m
c
and m
s
, respectively, are transformed in F to a transition from/to the same state,
i.e., t(s
c
, s
c
) : s
c
j/
s
c
. The transition guard j is triggered if the mode invariant for
m
c
or m
s
, and

`
J (explained earlier) are true, i.e., j = (inv(m
s
) inv(m
c
))

`
J. The
ow dynamics equations f low(m
c
) and f low(m
s
) are assigned to t(s
c
, s
c
) action , i.e.,
f low(m
c
) f low(m
s
)
R
C
.
4.2 Control model to embedded model transformation
The embedded model is performed from the control model as
E
=
C
R
E
and can be
performed from the reference model as
E
=
R
(R
C
R
E
), where the transformation rules
R
E
considers the hardware/software constrains and the hardware/software architecture.
As mentioned before, the control model
C
focuses on the control components C
C
. In a BAS
system, C
C
can be classied to a set of sensor component instances C
S
C
, actuator component
instances C
A
C
and processing/controller component instances C
P
C
(which execute the control
algorithm), i.e., C
C
C
S
C
, C
A
C
, C
P
C
. Based on the type of the C
C
instance,
E
creates the
corresponding hardware architecture, e.g., Fig. 7, Fig. 8, and Fig. 9 show a sensor component
architecture, an actuator component architecture, and a processing/controller component
architecture, where each architecture contains the appropriate Pro, Mem, Bus, Dev. As shown
in Fig. 5, each of these hardware components has its own hardware/software constrains.
In order to add the software impact to this architecture, we consider a transformation rule
R
E
from control model
C
. Typically, each control model
C
contains only one F, therefore F
transformed to a thread in
E
, i.e., F
R
E
Thr
E
, where Thr
E
captures the algorithm execution
constrains.
The multi-threading appears in case if a control component C
C
contains several inherited
components. In this case, each component is transformed to a thread and all the transformed
threads will be executed under the same processor.
6
This rule does not consider s
init
Processor Memory
Sensing
Device
Transmitting
Device
Buss
Fig. 7. Sensor Hardware Architecture
Processor Memory
Actuating
Device
Receiving
Device
Buss
Fig. 8. Actuator Hardware Architecture
Processor Memory
Transmitting
Device
Receiving
Device
Buss
Fig. 9. Controller Hardware Architecture
4.3 Control model to target embedded model transformation
The target embedded model is performed from the control model as
T
=
C
R
T
and
can be performed from the reference model as
T
=
R
(R
C
R
T
). The transformation
rules R
T
considers only the hardware/software constrain, whereas R
E
considers both
hardware/software constrains and architecture. The R
T
considers only constrains as
T
directly uses the hardware architecture on the targeted platform.
In this section, we show the transformation rules R
T
between the control model
C
and the
target embedded model
T
, as follows:
The set of binding variables V of
C
(called V
C
) is transformed to an equivalent set of
variables of corresponding type (e.g., Double, Integer, Boolean) in the binding variables
V
T
of
T
, i.e., V
C
R
T
V
T
.
404 Automation
The set of transitions T in
C
(called T
C
) is transformed to a set of switches T
T
of
T
, i.e.,
T
C
R
T
T
T
.
The set of guards J, including probabilistic and deterministic, of T
C
is transformed to a set
of conditions J of T
T
, i.e., J(T
C
)
R
T
J(T
T
).
The set of actions of
C
is transformed to a set of actions of
T
, i.e., (T
C
)
R
T
(T
T
).
However, additional P
io
constrains of N
B
and N
S
are needed to R
T
for identifying the P
io
specication. For example, for a sensor node N
B
one port should read from a sensor, this
port name should be identied and the corresponding sensor type. Therefore, whenever
an action reads from this port, a sensor reading action is added. The same is done for
the ports read/write from/to a wireless network, and these ports are identied with the
connected ID node, where this process identies the network topology.
The set of states S of
C
is transformed to a set of states (i.e., Boolean variables) in
T
,
i.e., S
C
R
T
S
T
. In addition, the set of transitions T
T
(s
T
, ) that exiting s
T
S
T
includes
s
T
in its conditions and s
T
= f alse in its actions. The set of transitions T
T
(, s
T
) that
enter s
T
includes s
T
= true in its actions. This transformation rule should be followed in
the specied order, which means starting from the transitions exiting the states and then
applying the ones entering the states.
5. Application domain
Heating, Ventilating, and Air-Conditioning (HVAC) systems provide a specied ambient
environment for occupants with comfortable temperature, humidity, etc. One way to regulate
the temperature in a room is Air Handling Unit (AHU), it is a set of devices used to condition
and circulate air as part of an HVAC system. Several control strategies have been introduced
to control the temperature regulation, where an operating scheduling is pre-dened. In
this context, standard PI control algorithms are adequate for the control of HVAC processes
Dounis & Caraiscos (2009). Therefore, we consider an AHU model used to regulate the
temperature for a single room, as out framework case-study. The AHU heating/cooling coils
are controlled using a PI algorithm, where the switch between cooling and heating coil is
performed using a system level decision.
We describe in this section the system specication for the used model and evaluation metrics
for the system property. Moreover, we apply the end-to-end transformation process for the
system reference model.
5.1 System evaluation metric
In this case-study, we regulate the temperature (for maintaining thermal comfort) with respect
to user discomfort. The expected discomfort metric penalizes the difference between the
measured indoor air temperature y(k) at time-step k (where k varies from k
s
to k
f
), and the
reference temperature r(k). We use the Root Mean Square Error (RMSE) to reect the indoor
temperature variation around the reference temperature in
o
C, during only the scheduling
period p(k) = 1:
DI =
k
f
k=k
s
(y(k) r(k))
2
k
f
p(k) (1)
In order to validate the transformation rules between the models, each generated model
7
has
to respect the system property DI 2
o
C.
5.2 System specication
As shown in Fig. 10, an AHU uses coils to heat or cool the indoor temperature. The
heated/cooled air is pumped to the room using a supply-air fan. In order to use the
heated/cooled indoor air more efciently, the AHU recycles some of the return air via an
air loop, which has a return-air fan to mix the indoor air with the outdoor air, as controlled by
three dampers. In our case-study, we assume xed settings for the fan speeds and the damper
settings, and we control only the valve settings for the coils.
Outside Air
Exhaust Air
Exhaust-air Damper
Outside-air Damper
Return Air
Return-air
Damper
Supply Air
Supply-air
Fan
Return-air
Fan
Single-Zone Room
Heating
Coil
Cooling
Coil
Fig. 10. AHU Structure
Fig. 11 shows the reference model that used as our framework case-study. This model contains
two main groups of components: (a) environment/plant components reect the plant physics
(e.g., walls, coils); (b) control components show the control/sensing algorithm that used to
modify/monitor the environment.
5.2.1 Environment components
All models described below are lumped-parameter models. Two variables are identied
to evaluate the model behaviour: external temperature T
ext
and set-point temperature
T
sp
. Moreover, ve environment/plant components have been used: Wall, Window,
Heating/Cooling Coil, Return Air and Indoor Air models, as follows:
7
This rule will not be applied to embedded model
E
, as this model used to validate the
hardware/software consistency
406 Automation
External
Temperature
Wall Model Window Model
Indoor Air Model
Temperature
Sensor
Controller
Set-Point
Temperature
walltemp.
indoor temp.
window temp.
indoor
temp. indoor
temp.
sensed
temp.
valve
actuation
Environment
Components
Control
Components
Heating Coil
Model
Cooling Coil
Model
Supply-air Fan
Speed
Return-air Fan
Speed
Return-air Damper
Occlusion
Return Air Model
return-air
temp.
coil
temp.
Valve Actuator
Fig. 11. AHU Component-Based Model
1. Wall Model: One of the room walls is facing the building faade, which implies heat
exchanges between the outdoor and indoor environments. In general, a wall can be
modelled using several layers, where greater delity is obtained with increased layers in
the wall model. In our case, four layers have been considered to reect sufcient delity
Yu & van Paassen (2004), using the following differential equation; Eq. 2:
wall
V
wall
c
wall
dT
wall
dt
=
wall
A
wall
(T
ext
T
wall
) (2)
wall
is the wall density [kg/m
3
], V
wall
is the wall geometric volume [m
3
], c
wall
is the wall
specic heat capacity [J/kg.K], T
wall
is the wall temperature [
o
C],
wall
is the wall thermal
conductance [W/(m
2
.K)], A
wall
is the wall geometric area [m
2
] and T
ext
is the outdoor
temperature [
o
C].
2. Heating/Cooling Coil Model: The coil model uses the temperature difference between the
water-in and water-out in order to heat/cool the room. In this case, the temperature is
controlled through the radiator water ow using the valve occlusion (called actuation
variable u). Moreover the coil exchanges temperature with its environment, such as the
indoor air temperature and returned air temperature (considering the supply fan speed)
as shown in equations: Eq. 3 and Eq. 4.
M
wtr
c
wtr
dT
coil
dt
= m
.
wtr
c
wtr
(T
wtrin
T
wtrout
) Q (3)
Q =
air
A
coil
(T
coil
T
air
) + m
.
sup f an
c
air
(T
coil
T
retair
) (4)
Where, M
wtr
is water mass [kg], c
wtr
is water specic heat capacity [J/kg.K], T
coil
is coil
temperature [
o
C], m
.
wtr
is water mass ow rate throw the coil valve [kg/s], T
wtrin
is water
temperature going to the coil [
o
C], T
wtrout
is water temperature leaving from the coil [
o
C],
air
is air thermal conductance [W/(m
2
.K)], A
coil
is coil geometric area [m
2
], T
air
is indoor
air temperature [
o
C], m
.
sup f an
is air mass ow rate throw the supply-air fan [kg/s], c
air
is
air specic heat capacity [J/kg.K], and T
reair
is returned air temperature [
o
C].
3. Return Air Model: This model acts as an air mixer between the indoor air ow and the
external air based on return-air damper occlusion , as shown in equations: Eq. 5.
T
retair
= T
air
+ (
m
.
ret f an
m
.
sup f an
)(T
air
T
ext
) (5)
Where, m
.
ret f an
is air mass ow rate throw the return-air fan [kg/s].
4. Indoor Air Model: In order to model the indoor temperature T
air
propagation, all HVAC
components have to be considered as they exchange heat with the air inside the controlled
room following equation 6 .
air
V
air
c
air
dT
air
dt
= Q
air
+ Q
wall
+ Q
window
(6)
Q
wall
=
air
A
wall
(T
wall
T
air
) (7)
Q
air
= m
.
sup f an
c
air
(T
coil
T
air
) (8)
Q
window
=
air
A
window
(T
window
T
air
) (9)
Where,
air
is air density [kg/m
3
], V
air
is air geometric volume [m
3
], A
window
is window
geometric area [m
2
], and T
window
is window temperature [
o
C].
5. Window Model: A window has been modelled to calculate the effects of glass on the indoor
environment. Since the glass capacity is very small, the window has been modelled as an
algebraic equation, Eq. 10, that calculates the heat transfer at the window node.
air
(T
ext
T
window
) +
air
(T
air
T
window
) = 0 (10)
5.2.2 Control components
A temperature sensor, PI-controller, and valve actuator are used to monitor, control and
actuate the environment model, respectively. The temperature sensor samples the indoor
temperature (T
s
=1 min) and sends the sampled value to the controller. In case the
sampled value within the operating time slot (identied by the operating schedule), then the
PI-controller calculates the next actuation value and sends it to the valve actuator in order to
adjust the coil valve occlusion.
The reference model for the control algorithm has two levels of hierarchy, as shown in Fig. 12.
The high level represents the system modes of the system decisions, where the heating mode
is activated if T
ext
is greater than T
sp
, and the cooling mode is activated if T
sp
is greater than
T
ext
. In these system modes, a PI-control algorithm is used to adjust the valve occlusion for
the corresponding coil and deactivate the other coil. Consequently, the controller updates the
actuation value each sampling period implemented using a continuous variable to present the
controller timer.
408 Automation
In this document, we consider the controller component as an example to apply the
end-to-end transformation rules. Fig. 12 shows the reference model for the controller
component with its transitions, actions, invariants and ow dynamics. However, we have
used the same design approach to design the reference model for the sensor and actuator
components.
j
1
: ControlTime > SamplingPeriod Schedule
1
: ControlTime = 0
: u
heating
= f
PI
(T
sp
, T
air
)
: u
cooling
= 0
j
2
: ControlTime > SamplingPeriod Schedule
2
: ControlTime = 0
: u
heating
= 0
: u
cooling
= 0
f low
di f 1
:
dControlTime
dt
= 1
inv
1
: ControlTime SamplingPeriod
Fig. 12. Reference model for the controller component, where the high level automata
describes the system level modes M
C
and the inherited level represents the control modes
M
C
.
Where, ControlTime is a continuous variable used to reect the controller time,
SamplingPeriod is the sampling period needed to receive the sensor sample and then update
the actuator, Schedule is a Boolean variable to indentify if the operation schedule is triggered or
not, u
heating
is the valve occlusion value for the heating-coil, f
PI
is the heating PI optimization
function, and u
cooling
is the valve occlusion value for the cooling-coil.
The cooling system mode uses similar LHA as in heating. However, it actuates on u
cooling
using
`
f
PI
instead of u
heating
, where
`
f
PI
is the cooling PI optimization function.
5.3 Case-study control model
We have followed the transformation rules R
C
in Sec. 4.1 to transform the controller reference
model in Fig. 12 to the controller control model in Fig. 13 with the corresponding transition
conditions J and actions .
We can deduce that system modes hierarchy has been removed by adding the complement
of its transition conditions to the conditions in the control model transitions. Moreover, the
mode dynamic ow and invariant are transformed to t
3
(j
3
,
3
).
j
1
: ControlTime > SamplingPeriod Schedule T
ext
> T
sp
1
: ControlTime = 0
: u
heating
= f
PI
(T
sp
, T
air
)
: u
cooling
= 0
j
2
: ControlTime > SamplingPeriod Schedule T
ext
> T
sp
2
: ControlTime = 0
: u
heating
= 0
: u
cooling
= 0
j
3
: ControlTime SamplingPeriod T
ext
> T
sp
3
: ControlTime = ControlTime + h
j
7
: T
ext
> T
sp
7
: ControlTime = 0
Fig. 13. Control model for the controller component, where the system mode hierarchy,
continuous variables are abstracted.
The rest of the control model transitions follow the same conditions and actions as the
aforementioned transitions, but for the cooling-coil.
5.4 Case-study embedded model
In our case-study, we create the controller component hardware architecture for the embedded
model as in Fig. 9 following R
E
in Sec. 4.2. We have emulated the embedded model using
Sun-SPOTs with the following hardware specications: (a) CPU=180MHz, 32 bit ARM 920T
Processor, and (b) RAM=512 Kbyte. Therefore, hardware/software constraints captured by
the embedded model as follows:
1. Memory space size: Spc(Mem) = 512kbyte.
2. Memory word size: Spc(Wrd) = 32bit.
410 Automation
3. Memory communication protocol: Spc(Prt) = read/write.
4. Processor clock period (CPU): Clk(Pro) = 1/180M .
5. Bus latency: Lcy(Bus) = 11ns.
6. Bus massage size: Msg(Bus) = 32bit.
7. Thread execution time: Tex(Thr) = 5 10ms. In this example, we have considered only
one thread as the controller component is a primitive component. However, in case of a
composite component, it is translated to multi-thread.
8. Thread switching protocol: Dprt(Thr) = periodic with switch period 20ms. This constrain
is not critical for our case-study as the processor execute only one thread.
9. Sending/Receiving Communication Device activation/re protocol: Dprt(Dev) =
periodic with a period equals to the network sampling rate T
s
.
5.5 Case-study target embedded model
In the target embedded model, we follow R
T
in Sec. 4.3 to generate the embedded language
semantic. In our case-study we have used embedded Java for Sun-SPOT devices. In the
controller component, u
heating
, u
cooling
and T
air
are identied as the component wireless P
io
,
which are corresponding to u_heating, u_cooling, and IndoorTemp variables in Fig. 14,
respectively. These P
io
identications are transformed to wireless communication commands
in case of reading/writing from/to P
io
. Moreover, the sending/receiving IDs are identied for
each port in order to structure the network topology, and consequently these IDs are included
in the wireless commands. For example, Fig. 14 shows the rst transition transformation
from the control model, where sensorID is the sensor ID that sends T
air
(IndoorTemp ) to the
controller and the controller ID is identied as hostID.
6. Experimental design
In order to check if the developed framework preserves the systems properties, we run a set
of experiments for the AHU system as a case-study at each model (i.e., reference, control, and
target). Then, we evaluate if each model respects the identied system property R = DI
2
o
C.
In our case-study, we vary external temperature and the set point temperature (T
ext
, T
sp
) and
then evaluate the corresponding DI. For example, if we have an input 2-tuple (15, 23) to
present (T
ext
, T
sp
), we will measure outputs DI = 0.73
o
C.
6.1 Dependent/independent variables
In our case-study, we evaluate each model over the domains of the independent variables as
follows:
1. External Temperature T
ext
5, 10, 15, 20;
2. Set-point Temperature T
sp
18, 20, 23;
where some constraints are identied for the variables search space to eliminate the
physically unrealizable solutions as [ T
ext
T
sp
[ T
max
, T
max
= 13
o
C is the maximum
allowable temperature difference and T
ext
,= T
sp
.
i f ( Heati ng_Control && ControlTime > SamplingPeriod && Schedule
&& ! ( ExternalTemp > OptimalTemp ) ) {
ControlTime = 0 . 0 ;
/ / Re c e i vi ng i ndo o r t e mpe r a t ur e f rom s e ns o r wi t h ID : s e ns or I D .
t r y {
RecConn . r ecei ve ( RecDg ) ;
recAddr = RecDg . getAddress ( ) ;
sensorID=RecDg . r eadI nt ( ) ;
i f ( ID=sensorID ) IndoorTemp = RecDg . readDouble ( ) ;
} cat ch ( Except i on e ) {
throw e ;
}
/ / Updat i ng t he he a t i ng a c t ua t i o n va l ue us i ng PI a l go r i t hm .
u_heati ng=heat i ngPI Opt i mi zat i on ( OptimalTemp , IndoorTemp ) ;
/ / Sendi ng t he he a t i ng a c t ua t i o n va l ue p a c k e t t o t he a c t ua t o r
/ / wi t h t he b a s e s t a t i o n ID ( hos t I D ) .
t r y {
SendDg . r e s e t ( ) ;
SendDg . wr i t eI nt ( hostID ) ;
SendDg . writeDouble ( u_heat i ng ) ;
SendConn . send ( SendDg ) ;
throw e ;
}
/ / De a c t i v a t i ng t he c o o l i ng c o i l .
u_cool i ng =0. 0;
/ / Sendi ng t he a c t ua t i o n va l ue p a c k e t t o t he a c t ua t o r
/ / wi t h t he b a s e s t a t i o n ID ( hos t I D ) .
t r y {
SendDg . r e s e t ( ) ;
SendDg . wr i t eI nt ( hostID ) ;
SendDg . writeDouble ( u_cool i ng ) ;
SendConn . send ( SendDg ) ;
throw e ;
}
/ / Update t he e x e c ut i o n s t a t e .
Heati ng_Control =t r ue ;
Cool i ng_Cont rol =f a l s e ;
}
Fig. 14. Target embedded model of the rst transition t
1
of the controller component.
412 Automation
The dependent variable for each model is the average of DI over the independent variable
cross-product space given by T
ext
T
sp
.
6.2 Empirical results
Fig. 15, 16 and 17 show the DI experiments for
R
,
C
, and
T
. The reference and control
models are empirically evaluated using the CHARON tool
8
, whereas the target embedded
model is emulated using Sun-SPOT nodes, with the following hardware specications: (a)
CPU=180MHz, 32 bit ARM 920T Processor, and (b) RAM=512 Kbyte. The embedded model is
captured using the AADL language and then evaluated using the OSATE tool
9
.
Figs. 15, 16 and 17 show that the DI decreases when the difference between T
ext
and T
sp
decreases, as the initial overshoot the settling time decreases. Moreover, we see that control
model gives the worst DI values, the reference model improves the DI and the target model
gives the best DI values. Because of the model resolution h for updating the dynamic ow
( f low
di f
, f low
alg
) decreases in case of the target model (in range of msec) in comparison to the
reference mode (in range of sec), and it increases for the control model (in range of 3 sec) over
the reference model. However, the average of DI for all models respects the system property
DI 2
o
C, where DI = 0.95, 2, 0.44
o
C for
R
,
C
, and
T
, respectively.
We have evaluated the binding consistency and the processing capacity for the embedded
model
E
to show a consistency binding and 50% processing capacity.
0
0.5
1
1.5
2
2.5
3
3.5
4
4 6 8 10 12 14 16 18 20
U
s
e
r

D
i
s
c
o
m
f
o
r
t

I
n
d
i
c
a
t
o
r

(
D
I
)

[
C
]
External Tempreture [C]
Reference Model
Control Model
Target Embedded Model
Fig. 15. User Discomfort Evaluation of Framework Models for T
sp
= 18
8
http://rtg.cis.upenn.edu/mobies/charon/index.html
9
http://www.aadl.info/aadl/currentsite/tool/index.html
0
0.5
1
1.5
2
2.5
10 11 12 13 14 15
U
s
e
r

D
i
s
c
o
m
f
o
r
t

I
n
d
i
c
a
t
o
r

(
D
I
)

[
C
]
Reference Model
Control Model
sp
= 20
0
0.5
1
1.5
2
2.5
3
3.5
10 12 14 16 18 20
U
s
e
r

D
i
s
c
o
m
f
o
r
t

I
n
d
i
c
a
t
o
r

(
D
I
)

[
C
]
Reference Model
Control Model
sp
= 23
414 Automation
7. Conclusion
In this article, we have proposed a novel end-to-end framework for designing an NCS for BAS
systemembedded over a distributed WSAN. This framework considers the design owstages
along from modelling to embedded control-code generation. The developed framework is
empirically validated using AHU system as a case-study.
A future trend of this framework is integrating a de facto standard tool, such as Simulink, to
the reference model. Moreover, we need to consider the timing and delay for the generated
embedded code.
8. References
Balarin, F., Watanabe, Y., Hsieh, H., Lavagno, L., Passerone, C. & Sangiovanni-Vincentelli, A.
(2003). Metropolis: An integrated electronic system design environment, Computer
pp. 4552.
Balasubramanian, D., Narayanan, A., van Buskirk, C. & Karsai, G. (2006). The graph rewriting
and transformation language: Great, Electronic Communications of the EASST 1.
Basu, A., Bozga, M. & Sifakis, J. (2006). Modeling heterogeneous real-time components in BIP,
SEFM, Vol. 6, Citeseer, pp. 312.
Chen, K., Sztipanovits, J. & Neema, S. (2005). Toward a semantic anchoring infrastructure
for domain-specic modeling languages, Proceedings of the 5th ACM international
conference on Embedded software, ACM New York, NY, USA, pp. 3543.
Denckla, B. & Mosterman, P. (2005). Formalizing causal block diagrams for modeling a class
of hybrid dynamic systems, Proc. 44th IEEE Conference on Decision and Control,and
European Control Conference (CDC-ECC05) pp. 41934198.
Dounis, A. & Caraiscos, C. (2009). Advanced control systems engineering for energy and
comfort management in a building environment-a review, Proc. of Renewable and
Sustainable Energy Reviews pp. 12461261.
Gurevich, Y., Rossman, B. & Schulte, W. (2005). Semantic essence of AsmL, Theoretical
Computer Science 343(3): 370412.
Hardebolle, C. & Boulanger, F. (2008). ModHelX: A component-oriented approach to
multi-formalism modeling, Lecture Notes In Computer Science 5002: 247258.
Henzinger, T. (1996). The theory of hybrid automata, Proc. 11th Annual IEEE Symposium on
Logic in Computer Science (LICS 96) pp. 278292.
Jackson, D. (2002). Alloy: a lightweight object modelling notation, ACM Transactions on
Software Engineering and Methodology (TOSEM) 11(2): 256290.
Lee, E., Neuendorffer, S. & Wirthlin, M. (2003). Actor-oriented design of embedded hardware
and software systems, Journal Of Circuits Systems And Computers 12(3): 231260.
Lee, E. & Sangiovanni-Vincentelli, A. (1998). A framework for comparing models of
computation, IEEE Transactions on computer-aided design of integrated circuits and
systems 17(12): 12171229.
Mady, A. & Provan, G. (2011). Co-design of wireless sensor-actuator networks for building
controls, the 50th IEEE Conference on Decision and Control and European Control
Conference (IEEE CDC-ECC).
Rumbaugh, J., Blaha, M., Premerlani, W., Eddy, F. & Lorensen, W. (1991). Object oriented
modeling and design, Prentice Hall, Book Distribution Center, 110 Brookhill Drive,
West Nyack, NY 10995-9901(USA).
Sztipanovits, J. & Karsai, G. (1997). Model-integrated computing, IEEE computer
30(4): 110111.
Vestal, S. (1996). MetaH programmers manual, Technical report, Version 1.09. Technical
Report, Honeywell Technology Center.
Yu, B. & van Paassen, A. (2004). Simulink and bond graph modeling of an air-conditioned
room, Simulation Modelling Practice and Theory, Elsevier .
416 Automation
21
The Role of Automation in the Identification
of New Bioactive Compounds
Pasqualina Liana Scognamiglio
1
, Giuseppe Perretta
2
and Daniela Marasco
1

1
Department of Biological Sciences, University Federico II of Naples, Naples,
2
Institute of Motors, CNR, Naples,
Italy
1. Introduction
Automation is nowadays implemented in many areas of the drug discovery process, from
sample preparation through process development. High-Throughput Screening (HTS) is a
well-established process for lead discovery that includes the synthesis and the activity
screening of large chemical libraries against biological targets via the use of automated and
miniaturized assays and large-scale data analysis. In recent years, high-throughput
technologies for combinatorial and multiparallel chemical synthesis and automation
technologies for isolation of natural products have tremendously increased the size and
diversity of compound collections. The HTS process consists of multiple automated steps
involving compound handling, liquid transfers and assay signal capture. Library screening
has become an important source of hits for drug discovery programmes. Three main
complementary methodologies are actually used: 1) in silico virtual screening of libraries to
select small sets of compounds for biochemical assays, 2) fragment-based screening using
high-throughput X-ray crystallography or NMR methods to discover relatively small related
compounds able to bind the target with high efficiency and 3) HTS of either diverse
chemical libraries or focused libraries tailored for specific gene families. Furthermore a
variety of assay technologies continues to be developed for high-throughput screening;
these include cell-based assays, surrogate systems using microbial cells and systems to
measure nucleic acid-protein and receptor-ligand interactions. Modifications have been
developed for in vitro homogeneous assays, such as time-resolved fluorescence,
fluorescence polarization and the scintillation proximity. Innovations in engineering and
chemistry have led to delivery systems and sensitive biosensors for Ultrahigh-Throughout
Screening working in nanoliter and picoliter volumes. Spectroscopic methods are now
sensitive to single molecule fluorescence. Technologies are being developed to identify new
targets from genomic information in order to design the next generation of screenings. As
HTS assay technologies, screening systems, and analytical instrumentation the interfacing of
large compound libraries with sophisticated assay and detection platforms will greatly
expand the capability to identify chemical probes for the vast untapped biology encoded by
genomes.
As a consequence the growing demand for new, highly effective drugs is driven by the
identification of novel targets derived from the human genome project and from the

Automation

418
understanding of complex protein-protein interactions that contribute to the onset and
maintaining of pathological conditions. To illustrate the dynamics of quantitative and
qualitative process approaches to accelerated drug development, a model pipeline is
depicted in Fig. 1.

Fig. 1. Drug Discovery pipelines: IND: Investigational New Drug, CTA: Clinical Trial
Application, NDA: New Drug Application, MAA: Marketing Authorization.
At the basis of HTS is the simultaneous employment of different sets of compounds that are
rapidly screened for the identification of active components. This approach is nowadays
regarded as a powerful tool for the discovery of new drug candidates, catalysts and
materials. It is also largely utilized to improve the potency and/or the selectivity of existing
active leads by producing analogues derived by systematic substitution or introduction of
functional groups or by merging active scaffolds. In Combinatorial Chemistry (CC) the
higher is the number of building blocks and the number of transforming synthetic steps, the
more attainable molecules can be had. However, the higher is the number of attainable
molecules, the higher should be the handling capacity of so many reagents and products;
thus automation in CC plays a very critical role. Automation in chemical synthesis has been
essentially developed around the solid phase method introduced by R. B. Merrifield for the
synthesis of peptides, which, after its introduction, has been continuously improved in
terms of solid supports, linkers, coupling chemistry, protecting groups and automation
procedures, making it nowadays one of the most robust and well-established synthetic
methods (Shin et al., 2005). For these reasons, the basic concepts of CC, such as compound
libraries, molecular repertoires, chemical diversity and library complexity have been
developed using peptides and later transferred to the preparation of libraries of small
molecules and other oligomeric biomolecules (Houghten et al., 2000). The easiness of
preparation, characterization and the robustness of the available chemistry provide high
purity levels and the built-in code represented by their own sequences have promoted the
employment of large but rationally encoded mixtures instead of single compounds, leading
to the generation and manipulation of libraries composed of hundreds of thousands and
even millions of different sequences. The broad complexity of mixture libraries has also led
to the development of several screening procedures based on iterative or positional
scanning deconvolution approaches for soluble libraries.
2. HTS
High-throughput screening (HTS) has achieved a dominant role in drug discovery over the
past two decades. Its aim is to identify active compounds (hits) by screening large numbers
of diverse chemical compounds against selected targets and/or cellular phenotypes. The
HTS process consists of multiple automated steps involving compound handling, liquid
transfers, and assay signal capture, all of which unavoidably contribute to systematic
variation in the screening data. It represents the process of testing a large number of diverse
chemical structures against disease targets to identify 'hits'.

The Role of Automation in the Identification of New Bioactive Compounds

419
Compared to traditional drug screening methods, HTS is characterized by its simplicity,
rapidness, low cost, and high efficiency, taking the ligand-target interactions as the
principle, as well as leading to a higher information harvest.
Independent of the precise nature of the applied screening technology, lead discovery
efforts can always be analyzed and optimized along the same fundamental principles of
performance management: time, costs, and quality of the process (Fig. 2).
time/well
wells/day
screens/year
project time
Ti me
few false positives
few false negatives
S/N, H/L, Z-factor
validated binders
Qual i t y
reagents
consumables
instrumentation
personnel
Cost s

Fig. 2. HTS magic triangle.
As a multidisciplinary field, HTS involves an automated operation-platform, highly
sensitive testing system, specific screening model (in vitro), abundant component libraries,
and a data acquisition and processing system. Several technologies such as fluorescence,
nuclear-magnetic resonance, affinity chromatography, surface plasmon resonance and DNA
microarray, are actually available, and the screening of more than 100 000 samples per day
is already possible.
The data analysis challenge is to detect biologically active compounds from assay variability.
Traditional plates, controls-based and non-controls-based statistical methods have been
widely used for HTS data processing and active identification by both the pharmaceutical
industry and academic sectors. Recently, the introduction of improved robust statistical
methods has reduced the impact of systematic row/column effects in HTS data. In practice,
no single method is the best hit detection method for every HTS data set. Nevertheless, to
help in the selection of the most appropriate HTS data-processing and active identification
methods a 3-step statistical decision methodology has been developed: Step 1) to determine
the most appropriate HTS data-processing method and establish criteria for quality control
review and active identification from the 3-day assay signal window and validation tests.
Step 2) to perform a multilevel statistical and graphical review of the screening data to
exclude data that fall outside the quality control criteria. Step 3) to apply the established
active criterion to the quality-assured data to identify the active compounds.

Automation

420
The principles and methods of HTS find their application for screening of combinatorial
chemistry, genomics, protein, and peptide libraries. For the success of any HTS assay or
screening several steps -like target identification, reagent preparation, compound
management, assay development and high-throughput library screening- should be carried
out with utmost care and precision. Historically, the majority of all targets in HTS-based
lead discovery fall into a rather small set of just a few target families (Table 1). Enzymes
such as kinases, proteases, phosphatases, oxidoreductases, phosphodiesterases, and
transferases comprise the majority of biochemical targets in todays lead discovery efforts.
Among cell-based targets, many GPCRs (G-protein-coupled receptors, 7-transmembrane
receptors), nuclear hormone receptors, and some types of voltage- and ligand-gated ion
channels (e.g., Ca2+-channels) are very well suited for screening large compound
collections. Despite the large number of human genes (>25,000) and the even larger number
of gene variants and proteins (>100,000), the number of molecular targets with drugs
approved against the target is still fairly limited (~350 targets). The explanations for this
discrepancy can be due to some targets might not be feasible at all for modulation via low
molecular weight compounds. Others, however, might simply not be approachable by
current technologies and therefore constitute not only a great challenge but also a
tremendous potential for future lead discovery. Among those, a large number of ion
channels, transporters, and transmembrane receptors but also protein-protein, protein-
DNA, and protein-RNA interactions, even RNA/DNA itself, might form innovative targets
for modulation via low molecular weight compounds. With better tools with sufficient
predictivity, one might be able to expand the current HTS portfolio into novel classes of
pharmaceutical targets. Besides the obvious classes with targets on the cell surface, such as
ion channels, transporters and receptors, we predict that modulation of intracellular
pathways, particularly via protein-protein interactions, might have a great potential for
pharmaceutical intervention. In this regard, some modern technologies such as subcellular
imaging (High Content Screening [HCS]) will certainly enable novel and powerful
approaches for innovative lead discovery.

Established Target Classes Novel Target Classes
Kinases/Phosphatases
Oxidoreductases
Transferases
Proteases
Nuclear Hormone Receptors
G-protein-coupled receptors, 7-
transmembrane receptors (GPCRs)
Ion Channels (some)
Signaling Pathways (some)
Ion Channels
Transporters
Transmembrane Receptors
Signaling Pathways
Protein-Protein Interactions
Protein-DNA Interactions
Protein-RNA Interactions
DNA/RNA
Table 1. Established and potential future target classes for HTS


421
HTS not only helps in drug discovery but it is also important in improving present drug
moieties to optimize their activity. In past years many advances in science and technology
and economic pressures have kept every researcher to develop speedy and precise drug
discovery and screening technologies to tackle the ever increasing diseases and the many
pathogens acquiring resistance to currently available drugs. This also applies to screening
the ever increasing compound libraries waiting to be screened due to increase in the parallel
and combinatorial chemical synthesis. Research is also carried out so to cut the drug
development costs, so that industries keep abreast with ever increasing competition.
2.1 Screenings of libraries
The greatest potential of combinatorial chemistry is represented by the number and variety
of screenable compounds. Major efforts of researchers in the last decades have been focused
on the development of methodologies to further increase molecular diversity. An interesting
approach is based on general reversible reactions that produce dynamic mixtures: in
these, reactants and products are present in thermodynamic equilibrium. The assayed
biological system selects the best binding structures among the different mixture
components, thus producing a shifts of the reaction equilibrium by subtracting the product.
Importantly, dynamic libraries do not need a deconvolution step and ligands are selected
directly in the reaction mixture. Although innovative and promising, dynamic
combinatorial libraries have been so far limited only to a small number of reversible
reactions and libraries of moderate size. Similarly, the methodology named libraries from
libraries has represented an innovation compared to the traditional concepts of
combinatorial chemistry. By this approach, combinatorial libraries of peptides are built on a
solid phase and are subsequently modified in order to maximize the chemical diversity.
Oxidations, reductions, alkylations and acylations can be performed, exponentially
increasing the number of new compounds. Once synthesized, libraries are employed in
screening processes to determine active components for a given target. The choice of the
assay is of utmost importance to succeed with a screening program. Indeed, different assays
can be chosen depending on whether binders for an unspecified site are searched for, or
ligands with predetermined properties are needed. Competition assays ensure the selection
of active molecules with a specificity for the interacting interface of the proteins employed
as targets in the screening. Binding assays can also be performed on bead, in this case the
libraries have been prepared with the Mix and Split method, the use of bead bound
molecules has the advantage of having a local high concentration of ligands (several
picomoles on a very limited surface area). The assay principle is simple and is based on the
interaction between molecules and a labelled target-any natural or artificial receptor,
enzymes, antibodies, nucleic acids or even small molecules in solution. When the labelled
target binds the bound molecule, the bead will also be labelled and therefore can be
visualized/detected by chromogenic methods and using micromanipulators. The labeled
bead can be isolated and the molecule/peptide microsequenced. Though the on-bead assay
is faster and easier to perform homogeneous phase assays in solution it can provide more
specific functional evaluations. However highly charged or very hydrophobic molecules,
when locally highly concentrated, can result in a high number of false-positive hits. Partially
releasable libraries, based on light sensitive chemical linkers have also been described. These
linkers, being very resistant to cleavage under acidic conditions, allow the complete removal
of amino acid side chain protections, while they can be easily cleaved by irradiation at a

Automation

422
defined wavelength. By tuning light intensity and duration, small amounts of peptides can
be released in solutions, while the molecules bound to the bead can be microsequenced to
reveal the molecule identity. The deconvolution of libraries bound to solid supports can be
performed following different approaches. Peptide arrays can be distinguished on the basis
of their preparations: some ones involve the immobilization of pre-synthesized peptide
derivatives and others the in situ synthesis directly on the array surface. The pre-synthesis
approach requires a chemoselective immobilization on solid supports that can provide a
useful method for controlling the orientation and the density of the immobilized peptides.
A critical factor for the screening of compound arrays is the accessibility to the immobilized
molecules by the target proteins. Several spacers between peptides and chip surface have
been proposed, such as 11 mercaptoundecanoic acid, hydrophilic polyethylene glycol chain,
dextran, bovine serum albumin or human leptin and water-compatible supramolecular
hydrogels.
In situ parallel synthesis methods can provide cheaper and faster miniaturized spatially
addressed peptide arrays. Two general approaches have been introduced for the
preparation of peptide arrays on solid supports: the photolithographic and the SPOT
synthesis. The photolithographic approach consisted in the translation of the solid phase
method for peptide synthesis into the preparation of supported peptide arrays by mean of
photolabile protecting groups that allow the synthesis to proceed only on defined surface
spots that are illuminated by light at a given wavelength. In the SPOT synthesis, the
peptides arrays are synthesized in a stepwise manner on a flat solid support, such as
functionalized cellulose membrane, polypropylene and glass, following the standard Fmoc-
based peptide chemistry. Each spot is thus considered as an independent microreactor:
thereby the selection of the solid support is a key factor. The support has to meet the
chemical and the biological requirements of the target and it determines the synthesis and
screening methods. It also dictates the functionalization type and the insertion of spacers
and/or linkers. Described supports include ester-derivatized planar supports, CAPE
[celluloseamino-hydroxypropyl ether] membranes, amino-functionalized polypropylene
membranes and glass surface. Libraries can thus be seen as a source of bioactive molecules
that are selected on the basis of the biochemical properties pre-defined by appropriate assay
settings. The more diverse will be the library, the higher will be the probability to select
good hits. The concept of diversity is therefore of utmost importance in choosing a library
for a particular screening and good libraries must first fulfill the requirement of highest
diversity instead of the highest complexity. Molecules with overlapping structures will
contribute little or nothing at all to the overall probability to find out a positive hit. The
diversity of a library is generally associated to its complexity, that in turn depends on the
number of different components. However this is not always true: small but smart libraries
can display a higher diversity than libraries with a huge number of components. Indeed the
synthesis of random combinatorial libraries of peptides generates a large number of quasi-
duplicates deriving from the strong similarity between several side chains. Using common
amino acids, in L- or D- configuration, sequences where Glu is replaced by Asp, Leu by Ile
or Val, Gln by Asn and so on, can display very similar properties. Such residues, although
being different in their propensity to adopt secondary structures, can be considered almost
equivalent in terms of intrinsic physico-chemical properties, as for example the capacity to
establish external interactions or to fit in a given recognition site. In large libraries, the need
to manage large arrays of tubes and codes can puzzle the way to the identification of lead


423
compounds and slows down the synthetic and deconvolution steps. To simplify the
synthesis and deconvolution procedures without significantly affecting the probability to
find out active peptides, we have reasoned on the general properties of L and D-amino
acids, reaching a compromise between the need to maintain the highest possible diversity
and that of reducing the number of building blocks. We have called these new libraries
Simplified Libraries, intending with this any new ensemble of possible sequences
achievable with a reduced and non redundant set of amino acids. The set of non
redundant different residues (12 instead of 20) is chosen following several simple rules.
Firstly quasi-identical groups of amino acids are selected: Asp and Glu; Asn and Gln; Ile
and Val; Met and Leu; Arg and Lys; Gly and Ala; Tyr and Trp; Ser and Thr. Then, from each
group, one is selected trying to keep a good distribution of properties, including
hydropathic properties, charges, pKa, aromaticity. The cysteine is converted to the stable
acetamidomethyl-derivative to prevent polymerization and to increase the number of polar
residues within the final set. The distribution of residue molecular weights is also an
important parameter in cases of deconvolution of the library by mass spectrometry
approaches. The molecular weights do not overlap, allowing unambiguous assignment of
sequences by tandem mass spectrometry. Using this set of amino acids in the L- or D-
configuration, several libraries have been designed and synthesized (Marasco et al, 2008).
2.2 Automated libraries characterization
A key element for a successful screening of peptide libraries is the preparation of good
quality libraries. Unclear screening results are often ascribed to the random generation of
large populations of impurities, therefore even for very large mixture libraries, gross
analytical characterizations must be performed to assess the relative amounts and the
distribution of experimental molecular weight of library components. While classical
analytical methods, as for example LC-MS, can be in most cases enough for the
characterization of single compounds, more sophisticated techniques are required for the
analysis of complex mixtures. The recent progresses in parallel separation methods, such as
orthogonal chromatography, associated to synergic combinations of different detectors of
complementary selectivities, have allowed the development of high-throughput analytical
technologies based on traditional methods as HPLC, capillary electrophoresis (CE), NMR,
FTIR, LC-MS, evaporative light scattering (ELSD), chemiluminescent nitrogen (CLND). A
list of useful techniques is reported in Table 2.
Mass spectrometry (MS) is the best method to assess compound identity and purity due to its
sensitivity, speed and specificity. Electrospray ionization (ESI) coupled to quadrupole
analyzers or Matrix Assisted Laser Desorption Ionization Time-Of-Flight (MALDI-TOF) are
useful for peptide library analysis. LC-MS systems using ESI and different analyzers (single
or triple quadrupoles, ion traps, TOF), offer the advantage of chromatographic separation
prior to on-line mass analysis, thus enormously simplifying data interpretation. Single
quadrupole instruments typically provide unit mass resolution and therefore do not resolve
isobaric ions (i.e. two compounds with the same nominal mass but different exact mass),
which are often encountered in combinatorial libraries. LC-MS systems equipped with ESI
and TOF analyzers (ESI-TOF) or with combined quadrupole-TOF (Q-TOF) analyzers provide
increased mass accuracies up to 5 ppm and great versatility for tandem mass analysis.
Limitations of MS include the inability to distinguish isomers and, with the increase
of potential molecular formulae, differences in masses are too close to measurement error.

Automation

424

Detection mode Advantages Disadvantages
UV-absorbance
spectroscopy
High sensibility and
response linearity
Easy on-line integration
Relative purity
measurements
Compounds have to bear
chromophores
Responses depende on
extinction coefficients
Evaporative light
scattering
Response factors of diverse
compounds
Relative purity
measurements
Limited for compounds
with MW>300 amu
Right response depends on
geometry molecule
Chemiluminescent
nitrogen
Universal response to
number of nitrogen
structure
Quantitative purity (if
used with an internal
standards)
Compounds must contain
nitrogen
MS
Excellent sensibility
Exact mass identification
Response factors depend on
compound ionizability and
matrix effects
Limited resolution with
compounds with a high
MW
NMR
Universal response
indipendent from
structure
Quantitative purity (if
used with an internal
standards)
Detailed structure
information
Complex data interpretation
UV-absorbance Peak
assignement difficult in
presence of impurities
Difficult on-line integration
Table 2. Rivelator Method for HTS Characterization Of Compound Libraries
In addition to standard HPLC columns systems for interfacing to MS, several alternative
separation techniques or platforms have been investigated for high-throughput purity
analysis. An interesting application regards imaging time-of-flight secondary ion mass
spectrometry (TOF-SIMS) to perform high-throughput analysis of solid phase synthesized
combinatorial libraries. Also NMR techniques have been extensively utilized for the


425
identification and quantification of combinatorial compounds. NMR has the capability of
elucidating compound structure and it is a quantitative detector whose response factor is
directly proportional to the number of nuclei for a given signal, independent of molecular
structure. Recently advances in NMR probe design have enabled sampling from smaller
sample volumes and have significantly improved sample throughput overcoming some
limitations of traditional NMR for combinatorial analysis included its low sensitivity, low
sample throughput and complexity of data interpretation. The characterization of large
mixture libraries has also been performed by pool amino acid analysis and by pool N-
terminal sequencing techniques. Such techniques, though not providing single compounds
purity and identity, can offer a rough but useful indication of equimolarity and
representativeness of components.
2.3 Assays preparation
Automation is an important element in HTS's usefulness. Typically, an integrated robot
system consisting of one or more robots transports assay-microplates from one station to
another for sample and reagent addition, mixing, incubation and finally readout or
detection. A HTS system can usually prepare, incubate, and analyze many plates
simultaneously, further speeding the data-collection process. Currently HTS robots are able
to test up to 100,000 compounds per day. The key labware or testing vessel of HTS is the
microtiter plate: a small container, usually disposable and made of plastic, that features a
grid of small, open divots called wells. Modern microplates for HTS generally have either
384, 1536, or 3456 wells. These are all multiples of 96, reflecting the original 96 well
microplate with 8 x 12 9mm spaced wells. Most of the wells contain experimentally useful
matter, often an aqueous solution of dimethyl sulfoxide (DMSO) and some other chemical
compounds, the latter of which is different for each well across the plate. The other wells
may be empty, intended for use as optional experimental controls. A screening laboratory
typically holds a library of stock plates, whose contents are carefully catalogued, and each of
which may have been created by the same lab or obtained from a commercial source. These
stock plates themselves are not directly used in experiments; instead, separate assay plates
are created as needed. An assay plate is simply a copy of a stock plate, created by pipetteing
a small amount of liquid (often measured in nanoliters) from the wells of a stock plate to the
corresponding wells of a completely empty plate. To prepare an assay each well of the plate
is filled with some quantity of a protein, or an animal embryo. After some incubation time
has passed to allow the biological matter to absorb, bind to, or otherwise react (or fail to
react) with the compounds in the wells, measurements are taken across all the plate's wells,
either manually or by a machine. Manual measurements are often necessary when the
researcher is using microscopy to (for example) seek changes or defects in embryonic
development caused by the wells compounds, looking for effects that a computer could not
easily determine by itself. Otherwise, a specialized automated analysis machine can run a
number of experiments on the wells. In this case, the machine outputs the result of each
experiment as a grid of numeric values, with each number mapping to the value obtained
from a single well. A high-capacity analysis machine can measure dozens of plates in the
space of a few minutes like this, generating thousands of experimental datapoints very
quickly. Depending on the results of this first assay, the researcher can perform follow up
assays within the same screen by "cherrypicking" liquid from the source wells that gave
interesting results into new assay plates, and then re-running the experiment to collect

Automation

426
further data on this narrowed set, confirming and refining observations. Problems
associated with screening robotics have included long design and implementation time,
long manual to automated method transfer time, non-stable robotic operation and limited
error recovery abilities. These problems can be attributed to robot integration architectures,
poor software design and robotworkstation compatibility issues (e.g., microplate readers
and liquid handlers). Traditionally, these integrated robot architectures have involved
multiple layered computers, different operating systems, a single central robot servicing all
peripheral devices, and the necessity of complex scheduling software to coordinate all of the
above. Usually robot-centric HTS systems have a central robot with a gripper that can pick
and place microplates around a platform. They typically process between 40 and 100
microplates in a single run (the duration of the run depends on the assay type). The screener
loads the robotic platform with microplates and reagents at the beginning of the experiment
and the assay is then processed unattended. Robotic HTS systems often have humidified
CO
2
incubators and are enclosed for tissue culture work. Like in assembly-line
manufacturing, microplates are passed down a line in serial fashion to consecutive
processing modules. Each module has its own simple pick and place robotic arm (to pass
plates to the next module) and microplate processing device. The trend towards assay
miniaturization arose simultaneously with move towards automation from the direct need
to reduce development cost. Although at present most HTS is still carried out in 96-well
plate format, the move towards 384-well and higher density plate formats is well taking
place. Instrumentation for accurate, low-volume dispensing into 384-well plates is
commercially available, so they are sensitive plate-readers that accommodate this format.
The combination of liquid handling automation and information management processes
supports the entire compound management cycle from compound submission to delivery of
assay ready compound plates. The compound management often consists in the
organization of hundreds of thousands of compounds (small molecules, natural products
and peptide libraries in different formats). Compounds are stored offline in an automated
compound store in 96-well tube format and online in 384-well format for hitpicking
operations, 1536-well for HTS operations. Often there are dedicated LC-MS systems with
multimode mass spectrometry (ESI, APCI, ELSD) for quality control of incoming HTS
compound libraries and hit confirmation efforts. This platform is designed for automated
analysis and generates database-ready reports.
2.4 Screening assays
Assays are either heterogeneous or homogeneous. Heterogeneous assays are a bit complex,
requiring additional steps like filtration, centrifugation etc. beyond the usual steps like fluid
addition, incubation and reading. Homogeneous assays are simpler, consisting of the latter
three usual steps - this may also be called a true homogeneous assay. However at times
homogeneous assays could be complex due to the need for multiple addition and different
incubation times. Though homogeneous assays are advantageous, many companies prefer
to continue to use heterogeneous assay, eyeing their better precision over its counterpart,
though it is true only in few number of cases. The driving force for use of homogeneous
assays is the lower number of steps, which will help reduce assay cost. This simplicity may
also reduce the robotic complexity requirement for automation. When performing HTS of
free compounds in solution, automation, miniaturization and very sensitive detection
methodologies are required. Different approaches such as ELISA, cell-based cytotoxic,


427
antimicrobial, radiometric and fluorescence-based assays, affinity chromatography
methodologies can be used. Fluorescence-based techniques are likely to be among the most
important detection approaches used for HTS due to their high sensitivity and amenability
to automation, given the industry-wide drive to simplify, miniaturize, and speed up assays.
Fluorescence resonance energy transfer (FRET), fluorescence polarization (FP), and
fluorescence correlation spectroscopy (FCS) are indeed already broadly utilized for
screenings of large collections of compounds and many optical readers capable of handling
multi-well plates are commercially available. FRET is based on energy transfer between
appropriate energy donor and acceptor molecules. It is typically used in protein-protein
interaction studies where one protein partner is labelled with a fluorescence donor and the
other one is labelled with an appropriate fluorescence acceptor molecule. The donor has the
specific property of being excitable, emitting a fluorescence photon at a wavelength that is
able to in turn excite the acceptor. When the two proteins interact and the system is excited
at the donor excitation wavelength, a specific fluorescence emission at the acceptor emission
wavelength is recorded. The energy transfer occurs only when the donor-acceptor pair is
within a minimum distance, the Frster distance (the distance at which energy transfer
efficiency is half-maximal), which is around 50 . When performing a screening assay, if a
library component is able to disrupt the protein-protein interaction, the effect can be
quantitatively measured by a reduction of the FRET effect. Similar assays can be performed
using the Time Resolved FRET (TR-FRET) whereby the fluorescence of the acceptor
molecule has a duration on the milliseconds time scale, allowing fluorescence measurement
after a short time delay to remove interference by the excitation energy or by inhibitors. TR-
FRET can be integrated on large time intervals to increase sensitivity. Typical donor-
acceptor molecules are the Allophycocyanin-Europium chelates or the fluorescein - Terbium
chelates. FRET can also be used to determine enzyme activity using internally quenched
probes. In these systems short peptides corresponding to the sequence for a natural cleavage
site of the enzyme is synthesized and labelled at opposite ends with appropriate donor and
quencher pairs. Before cleavage, donor and quencher are very close and the effective
fluorescence emitted is low; once the two parts drift apart, the fluorescent signal increases.
This approach has been successfully utilized to screen the substrate specificity of an alkaline
serine proteinase. FP experiments allow measurements of changes in the emitted light
intensity of small labelled probes on binding to larger molecules. The sample is excited with
polarized light and, when a binding equilibrium is established, the observed polarization of
the emitted light increases. In FCS the main detected parameter is the spontaneous intensity
fluctuations caused by the minute deviations of the small system from the thermal
equilibrium. FCS is an emerging technique for HTS. In this case, measurements are carried
out using confocal optics to provide the highly focused excitation light. It is used to monitor
binding interactions as well as other molecular events. At times it was reported that assays
for biological targets cannot be conveniently designed to fit with standard cellular or
biochemical assay formats. For example, in the search for new antibacterial agents, genomic
experiments have indicated a large number of proteins that are essential for the survival of
the bacterium but whose function in the cell is unknown. In this case there is no known
biological function that will allow the design a biochemical or cellular screen. To screen
these types of target, an alternative to conventional chemical or cellular screenings may be
used. One alternative screening approach that does not require knowledge or analysis of the
biological function of the target of choice is direct measurement of compound interaction
with protein. A range of techniques are available to measure the direct binding events such

Automation

428
as NMR, SPR and calorimetry. One advantage afforded by NMR is that it can provide direct
information on the affinity of the screening compounds and the binding location of protein.
The structure-activity relationship acquired from NMR analysis can sharpen the library
design, which will be very important for the design of HTS experiments with well-defined
drug candidates. Affinity chromatography used for library screening will provide
information on the fundamental processes of drug action, such as absorption, distribution,
excretion, and receptor activation; also the eluting curve can give directly the possibility of
candidate drug. SPR can measure the quantity of a complex formed between two molecules
in real-time without the need for fluorescent or radioisotopic labels. SPR is capable of
characterizing unmodified biopharmaceuticals, studying the interaction of drug candidates
with macromolecular targets, and identifying binding partners during ligand fishing
experiments. They indeed allow real-time detection of soluble peptide binding to different
biomolecules, like proteins, nucleic acids or sugars and also to cell membranes or whole
cells. In addition, they allow the one-step measurement of kinetic rates and affinity of
binding, thus providing an affinity ranking of molecules peptides.
2.5 Data analysis
In validating a typical HTS assay, unknown samples are assayed with reference controls.
The sample signal refers to the measured signal for a given test compound. The negative
control (usually referred to as background) refers to set of individual assays from control
wells that give minimum signals. The positive control refers to the set of individual assay
from control wells that give maximum signals. In validating the assay, it is critical to run
several assay plates containing positive and negative control in order to assess
reproducibility and signal variation at two extremes of the activity range. The positive and
negative control data can then be used to calculate their means and standard deviations
(SD). The difference between the mean of the positive controls and the mean of the negative
controls defines the dynamic range of the assay signal. The variation in signal measurement
for samples, positive control, and negative controls (i.e., SDs) may be different. The mean
and SD of all the test samples are largely governed by the assay method and also by intrinsic
properties of the compound library. Because the vast majority of compounds from an
unbiased library have very low or no biological activity, the mean and SD of all the sample
signals should be close to those of the positive controls for inhibition/antagonist type assays
and near to those of the negative controls for activation/agonist types assays.
The Z-factor is a measure of statistical effect. It was proposed for use in HTS to judge
whether the response in a particular assay is large enough to warrant further attention. The
Z-factor is defined by four parameters: the means () and standard deviations () of both the
positive (p) and negative (n) controls (p, p, and n, n). Given these values, the Z-factor is
defined as:

3
Z-factor=1-
p n
p n

Generally HTS suffers from two types of errors false positives and negatives. A poor
candidate or an artifact gives an anomalously high signal, exceeding an established
threshold. While a perfectly good candidate compound is not flagged as a hit, because it


429
gives an anomalously low signal. Moreover, a low degree of relevance of the test may
induce a high failure rate of type. Much more attention is given to false-positive results than
to false-negative results. Some of the false positives are promiscuous compounds that act
non competitively and show little relationship between structure and function.
In HTS each biochemical experiment in a single well is analyzed by an automated device,
typically a plate reader or other kind of detectors. The output of these instruments comes in
different formats depending on the type of reader. Sometimes multiple readings are
necessary, and the instrument itself may perform some initial calculation. These
heterogeneous types of raw data are automatically fed into the data management software.
In the next step raw data are translated in contextual information by calculating results.
Data on percentage inhibition or percentage of control are normalized with values obtained
from the high and low controls present in each plate. The values obtained depend on the
method used for the normalization step (e.g. fitting algorithms used for dose-response
curve) and have to be standardized for screens. All the plates that fail against one or more
quality criteria are flagged and discarded. A final step in the process requires the
experimenter to monitor visually the data that have been flagged, as a final quality check.
This is a fundamental step to ensure the system has performed correctly. In addition to
registering the test data, all relevant information about the assay has to be logged, e.g. the
supplier of reagents, storage conditions, a detailed protocol, plate layout, and algorithms for
the calculation of results. Each assay run is registered and its performance documented.
HTS will initially deliver hits in targeted assays. Retrieval of these data has to be simple and
the data must be exchangeable between different project teams to generate knowledge from
the total mass of data.
3. Virtual screening
Even with HTS, the discovery of new lead compounds largely remains a matter of trial and
error. Although the number of compounds that can be evaluated by HTS methods is
apparently large, these numbers are small in comparison to the astronomical number of
possible molecular structures that might represent potential drug-like molecules. Often, far
more compounds exist or can be synthesized by combinatorial methods than can be
reasonably and affordably evaluated by HTS. As the costs of computing decreases and as
computational speeds increase, many researchers have directed efforts to develop
computational methods to perform virtual screens of compounds. Thus since performing
screens in silico can be faster and less expensive than HTS methods, virtual screening
methods may provide the key to limit the number of compounds to be evaluated by HTS to
a subset of molecules that are more likely to yield hits when screened. For the practical
advantages of virtual screening to be realized, computational methods must excel in speed,
cheapness, and accuracy. Striking the right balance of these criteria with existing tools
presents a formidable challenge. An inspiring example study

is related to the structure-
based virtual screening applied to the selection of new thyroid hormone receptor
antagonists when only a related receptor structure is available.

Receptor-based virtual
screening uses knowledge of the target protein structure to select candidate compounds
with which it is likely to favorably interact (Schapira et al., 2003). Even when the structure of
the target molecule is known, the ability to design a molecule to bind, inhibit, or activate a
biomolecular target remains a daunting challenge. Although the fundamental goals of

Automation

430
screening methods are to identify those molecules with the proper complement of shape,
hydrogen bonding, electrostatic and hydrophobic interactions for the target receptor, the
complexity of the problem is in reality far greater. For example, the ligand and the receptor
may exist in a different set of conformations when in free solution than when bound. The
entropy of the unassociated ligand and receptor is generally higher than that of the
complexes, and favorable interactions with water are lost upon binding. These energetic
costs of association must be offset by the gain of favorable intermolecular proteinligand
interactions. The magnitude of the energetic costs and gains is typically much larger than
their difference, and, therefore, potency is extremely difficult to predict even when relative
errors are small. While several methods have been developed to more accurately predict the
strength of molecular association events by accounting for entropic and solvation effects,
these methods are costly in terms of computational time and are inappropriate for the
virtual screening of large compound databases. The challenge in developing practical
virtual screening methods is to develop an algorithm that is fast enough to rapidly evaluate
potentially millions of compounds while maintaining sufficient accuracy to successfully
identify a subset of compounds that is significantly enriched in hits. Accordingly, structure-
based screening methods typically use a minimalist grid representation of the receptor
properties and an empirical or semiempirically derived scoring function to estimate the
potency of the bound complex (Schulz-Gasch & Stahl, 2003). Several programs now employ
a range of scoring functions, but it is often difficult to assess their effectiveness on difficult
real-world problems.

Virtual screening based on receptor structure therefore has the
distinct advantage of aiding the discovery of new antagonist structural classes or
pharmacophores.

First, and most importantly, from the many case studies published over
the past decade, it has become evident that the applicability and thus the usefulness of a
particular virtual screening method for a given drug discovery project depends on the
macromolecular target being investigated. Thus it seems more appropriate to consider
virtual screening from a problem-centric rather than a method centric perspective.
Depending on what is already known about a target and its ligands, different approaches to
virtual screening and consequently different sets of methods are preferred. In addition,
some virtual screening methods that have been reported might be premature or simply not
sufficiently accurate.

Second, the perceived success or failure of virtual screening in a
particular organization depends on the depth and mode of integration of virtual screening
in the organizations hit identification process, and whether expectations are realistic. For
example, an important factor for the chances of success of virtual screening could be
whether hits with interesting characteristics, such as structural novelty and/or patentability,
can be sufficiently nurtured by medicinal chemists to produce leads that can compete with
those arising from HTS. Indeed, this is typically the scenario for hits arising from alternative
lead discovery approaches such as fragment-based screening.

Third, a major achievement of
virtual screening so far has been to help eliminate the bulk of inactive compounds (negative
design), rather than to actually select bioactive molecules for a given target (positive design).
Although this statement is a simplification it highlights the extent of the challenges for
developing improved virtual screening methods.

It will be important to systematically determine which ligandreceptor interactions are
amenable to such an approach and which require other or additional features to be
considered indeed, dynamic descriptions of molecules will have to replace our


431
predominantly static view of both targets and ligands. Molecular dynamics simulations can
sample conformational ensembles of targets and ligands. However, some of the popular
force-field approaches used to describe the energetics of molecular systems might be
inadequate for drug design. Furthermore, although in general it might be more valuable to
identify ligand chemotypes for which receptorligand complex formation is dominated by
enthalpy changes rather than entropy changes, improvements are required to allow for a
more accurate estimation of both enthalpic and entropic contributions (Freire, 2008). The
thermodynamics of ligandreceptor interactions are commonly treated in a similar way as
molecular reactions, and this may not always be appropriate. How can we reliably and
efficiently predict that some proteinligand interactions become stronger with increasing
temperature, or identify the role of buried water on ligand binding? Questions like these
must be answered by computational chemistry as the forces that govern ligandreceptor
interactions are only understood at a rudimentary level: flexible fit phenomena, the role of
water molecules, protonation states in proteinaceous environments, and the entropic and
enthalpic contributions and compensations upon complex formation are not satisfactorily
addressed by the existing virtual screening methods. In this respect, advanced and
specialized computer hardware might enable extended (>100 ns) dynamics simulations of
macromolecular targets and receptorligand complexes on a routine basis, which might help
our understanding of allosteric effects and flexible fit phenomena of druglike ligands and
effector molecules. As a consequence, modelling of dynamic molecular features (in contrast
to static properties such as molecular mass, logP and other time-invariant molecular
properties), which cannot be accurately achieved at present, could improve the accuracy of
future predictions of novel bioactive compounds.
4. Fragment screening
Fragment-based drug discovery has proved too to be a very useful approach particularly in
the hit-to-lead process, acting as a complementary tool to traditional HTS. Over the last ten
years, fragment-based drug discovery has provided in excess of 50 examples of small
molecule hits that have been successfully advanced to leads and therefore resulted in useful
substrate for drug discovery programs. The unique feature of fragment-based drug
discovery is the low molecular weight of the hit. It has the potential to supersede traditional
HTS based drug discovery for molecular targets amenable to structure determination. This
is because the chemical diversity coverage is better accomplished by a fragment collection of
reasonable size than by larger HTS collections. Fragments represent smaller, less complex,
molecules than either drug compounds or typical lead series compounds. It is now widely
acknowledged within the pharmaceutical and biotech industries that weakly active
fragment hit molecules can be efficiently optimised into lead compound series if structural
insight is obtained at the outset for the binding interaction between each fragment hit and
the target protein of interest. This is supported by recent reports of the progression into
human clinical trials of drug molecules developed from weakly active fragment starting
points (Jhoti et al., 2007). Fragment-based drug discovery can explore the drug-like chemical
diversity space in an efficient and effective manner. Two key factors govern this approach,
firstly the coverage of fragment chemical diversity space during the screening stage and
then, that drug chemical diversity space is explored in an efficient iterative fashion during
the optimisation stage as fewer combinations need to be evaluated than through a purely

Automation

432
random screening and undirected optimisation approach. For example, a fragment
collection of 10 000 molecules may virtually represent the diversity of one billion molecules
if one considers the combinatorial power of fragment merging or linking (e.g. by assuming
two adjacent binding sites to which fragments bind and 10 different possibilities of fragment
linking) but only a small part of the larger chemical space defined by fragment merging and
linking needs to be explored in the structure- directed elaboration of fragments into leads.
Employing fragment-based drug discovery a relatively small number of low molecular
weight fragment molecules can provide a higher degree of sampling of the chemical
diversity space for fragments than a very large number of higher molecular weight
compounds is able to sample the respective chemical diversity space for drug-like
compounds. Furthermore lower molecular weight molecules exhibit reduced complexity
than the larger molecules in drug-like collections and it can be hypothesized a model to
rationalise ligandreceptor interactions in the molecular recognition process. Accordingly
the theoretical probability of a useful interaction falls dramatically with increasing
molecular complexity of the ligand. The selection of the fragment screening method is of key
importance since there are two general factors that have an impact for all methods. The first
is sensitivity of the screening method and the second is throughput. Sensitive screening
methods enable weakly active fragment molecules of lower molecular weights to be
identified as hit compounds and so fragment libraries with a lower molecular weight range
can be used. On the contrary the use of a low throughput screening method necessitates the
use of a smaller fragment library with a concomitant sparser coverage of fragment chemical
diversity space. Perhaps the most elegant method of fragment screening is by X-ray
crystallography, in that it provides directly structural information on the interaction
between fragment ligands and the protein target. However, owing to the methods low
throughput, even when fully automated, the technique can only be effectively applied to
targets for which a robust crystallographic system is available that allows soaking of
preformed crystals with fragment cocktail mixtures of up to 10 compounds at high
concentrations. This requirement imposes two key limitations: in the number of evaluable
fragment compounds, typically limited to no more than 1000 fragments, and for this, there
is a significant possibility of missing active fragments owing to the protein being locked into
a conformation, in the crystals used for the soaking studies, that does not allow the
interaction of fragments that require induced fit to bind. Although, no data are available on
the false negative rate for fragment screening by X-ray crystallography it may be significant
for certain targets. Each fragment screening technique has its advantages; X-ray
crystallography provides immediate structural information, NMR provides binding site and
affinity information of a very high quality while bioassays provide functionally relevant
activity data for larger collections of fragments. The best approach is to combine the
methods in order to maximize their value to fragment-based drug discovery. NMR and
biochemical screening of fragments are complementary orthogonal methods that can be
used individually or in concert to provide the most effective way of addressing each new
biological target of interest. The strength of biochemical screening is that its throughput
allows large fragment collections to be screened in a short length of time. This ensures that
the most ligand efficient diverse starting points are available for medicinal chemists to select
for subsequent optimisation. A further advantage is that screening related targets using
generic biochemical assay formats enables insights into target selectivity from the outset.
The large number of fragment hits that are obtained through use of biochemical screening of


433
large fragment libraries can be effectively triaged ahead of crystallography by the use of
protein NMR. Thus, the most effective way to perform fragment screening is not to rely on a
single method but to use orthogonal methods in concert.
5. Conclusion
HTS is the most widely applicable technology delivering chemistry entry points for drug
discovery programs, however it is well recognized that even when compounds are
identified from HTS they are not always suitable for further medicinal chemistry
exploration. It is evident that in the future the overwhelming number of emerging target
will dramatically increase the demand put on HTS and that this will call for new hit and
lead generation strategies to curb costs and enhance efficiency. The collections of large
pharmaceutical companies are approaching approximately one million entities, which
represents historical collections (intermediates and precursors from earlier medicinal or
agrochemical research programs), natural products and combinatorial chemistry libraries.
This about one order of magnitude higher than ten years ago when HTS and combinatorial
chemistry first emerged. However today purchasing efforts in many pharmaceutical
companies are directed towards constantly improving and diversifying the compounds
collections and making them globally available for random HTS campaigns. The
combinatorial explosion- meaning the virtually number of compounds that are synthetically
tractable-has fascinated and challenged chemists ever since the inception of the concept.
Independent of the library designs, the question of which compounds should be made from
the huge pool of possibilities always emerges immediately, once the chemistry is established
and the relevant building blocks are identified. The original concept of synthesize and
test, without considering the targets being screened, was frequently questioned by the
medicinal chemistry community and is nowadays considered of lower interest due to the
unsatisfactory hit rates obtained so far. As a consequence there is now a clear trend to move
away from huge and diverse random combinatorial libraries towards smaller and focused
drug-like subsets. Hit and lead generation are key processes involved in the creation of
successful new medicinal entities and it is the quality of information content imparted
through their exploration and refinement that largely determines their fate in the later stages
of clinical development. The combination of virtual screening and parallel and medicinal
chemistry, in conjuction with multi-dimensional compound-property optimization, will
generate a much-improved basis for proper and timely decisions about which lead series to
pursue further.
6. References
Freire, E. (2008). Do enthalpy and entropy distinguish first in class from best in class?. Drug
discovery today 13(19-20): 869-874.
Houghten, R.A., Wilson, D. B, & Pinilla, C. (2000). Drug discovery and vaccine development
using mixture-based synthetic combinatorial libraries. Drug discovery today 5(7):
276-285.
Jhoti, H., Cleasby, A., Verdonk. M., & Williams, G. (2007). Fragment-based screening using
X-ray crystallography and NMR spectroscopy. Current opinion in chemical biology
11(5): 485-493.

Automation

434
Marasco, D., Perretta, G., Sabatella, M., & Ruvo, M. (2008). Past and future perspectives of
synthetic peptide libraries. Curr Protein Pept Sci. 9(5):447-67.
Schapira, M., Raaka, B.M., Das, S., Fan, L., Totrov, M., Zhou, Z., Wilson, S.R., Abagyan, R.,
& Samuels H.H. (2003). Discovery of diverse thyroid hormone receptor antagonists
by high-throughput docking. Proceedings of the National Academy of Sciences of
the United States of America 100(12): 7354-7359.
Schulz-Gasch, T. & Stahl, M.. (2003). Binding site characteristics in structure-based virtual
screening: evaluation of current docking tools. Journal of molecular modeling 9(1):
47-57.
Shin, D.S., Kim, D.H., Chung, W.J., & Lee, Y.S. (2005). Combinatorial solid phase peptide
synthesis and bioassays. Journal of biochemistry and molecular biology 38(5): 517-
525.
22
Automatic Stabilization of Infrared Images
Using Frequency Domain Methods
J. R. Martnez de Dios and A. Ollero
Robotics, Vision and Control Research Group, University of Seville
Spain
1. Introduction
In the last decade the decrease in the cost of infrared camera technology has boosted the use
of infrared images in a growing number of applications. Traditional uses of infrared images,
such as thermal analyses (Carvajal et al., 2011), nondestructive testing and predictive
maintenance (Maldague, 2001), have been extended to new fields in which infrared cameras
are mounted on vehicles and mobile robots and are used in applications such as border
surveillance, building inspection, infrastructure maintenance, wildlife monitoring, search and
rescue and surveillance, among many others. Image vibrations are harmful disturbances that
perturb the performance of image-processing algorithms. Many of the aforementioned
applications require having stabilized sequences of images before applying automatic image-
processing techniques. Also, in cases where humans visualize the images, image vibrations
induce significant stress and decrease the attention capacity of the operator.
Two main approaches for image stabilization have been developed. The first one aims at
stabilizing the camera vibrations using different devices ranging from simple mechanical
systems for handheld camcorders to high-performance inertial gyro-stabilized platforms.
The first ones usually have low accuracy and perform vibrations reduction instead of
vibrations cancellation. The high size, weight and cost of gyro-stabilized platforms
constrain their use in a good number of applications.
The second main approach aims at correcting the images by applying image-processing
techniques. Classical image-processing methods for image stabilization are based on
detecting a set of local features e.g. corners, lines and high-gradient points- in one image
and tracking them along the images of the sequence. The relative motion of features from
one image to another is used to model the motion between both images. Once the motion
between images has been estimated, the second image is compensated such that no
vibration can be perceived between both images. These methods have demonstrated very
good performance with images with high contrast and low noise level, where a number of
local features can be robustly detected. However, they do not perform well in images where
these features cannot be robustly identified. This is the case of infrared images, which often
have low contrast and resolution and high noise levels.
This chapter describes an automatic image-processing technique for the stabilization of
sequences of images using frequency domain image representations obtained by means of

Automation

436
the Fourier transform. In particular, the method described in this chapter uses Fourier-
Mellin transforms (FMT) and the Symmetric Phase Only Matched Filtering (SPOMF). This
frequency domain representation provides advantages when stabilizing images with low
contrast and high noise levels: the described method does not rely on local features but on
the global structure of the image. The proposed method can correct translations, rotations
and scalings between images, which is sufficient for a high number of stabilization
problems. This chapter presents the main principles of the method, gives implementation
details and describes its adaptation to different image stabilization applications.
The method was implemented efficiently for real-time execution with low computing
resources. Several versions were tested and validated in different applications. The
proposed method has also been validated with visual images but it is with infrared images
where the advantages with respect to feature-based methods are more evident.
The main strengths of the proposed technique are the following:
1. it relies on the global structure of the image and thus, it is suitable for images with low
contrast and resolution and high noise levels;
2. the method can be tuned balancing the stabilization accuracy and the computational
burden, allowing adaptation to specific image stabilization needs.
This chapter is structured as follows:
- Section 2 introduces the problem of stabilization of sequences of infrared images.
- Section 3 presents the principles of image matching using Fourier-Mellin transform and
Symmetric Phase Only Matched Filtering.
- Section 4 shows examples that illustrate the operation of the proposed method.
- Section 5 proposes some practical aspects to increase the robustness, accuracy and
efficiency of the method.
- Section 6 presents various implementations in different applications.
Finally, Section 7 is devoted to the final discussions and conclusions.
2. Stabilization of infrared images
2.1 Infrared images
Infrared cameras generate images of the scene that contain the radiation intensity field
within the infrared band. Infrared cameras can "see" in pitch black conditions and through
smoke. They can transform radiation measurements in temperature estimations and can also
generate thermograms containing temperature of the objects in the scene (Hudson, 1969). In
the last decade infrared camera technology has evolved significantly. While in the decade of
the 90s most infrared cameras weighted several kilograms, consumed hundreds of watts
and their cryogenic cooling systems required frequent maintenance, now it is possible to
find radiometric infrared cameras that weight less than 150 gr, consume less than 1,5 W and
require no maintenance since they have no cooling system. These advances together with a
remarkable cost decrease have motivated their use in a growing number of applications.
Two infrared cameras are shown in Fig. 1, one Mitsubishi IR-M300 (left) and one Indigo
Omega micro-camera (right).

Automatic Stabilization of Infrared Images Using Frequency Domain Methods

437

Fig. 1. Left) Mitsubishi IR-M300 infrared camera. Right) Indigo Omega infrared micro-camera
However, despite these advances the quality of the images from infrared cameras is
significantly lower than that of visual images due to several physical and technological
reasons. Infrared camera detectors are highly affected by different types of noise including
thermal noise, shot noise and flicker noise (Hudson, 1969). In fact, many cameras,
particularly those operating in the mid-infrared spectral window [3-5] m, use cooling
systems -often cryogenic- to ensure that the detector operates at low and constant
temperatures. Noise is crucial in infrared technology. In fact, infrared cameras usually use
several characteristics to measure the influence of noise in the images, such as the Noise
Equivalent Temperature (NET) or the Noise Equivalent Temperature Difference (NETD), among
others (Maldague, 2001).
Also, infrared detectors usually have significantly lower sensitivity than CMOS and CCD
detectors commonly used in visual cameras. This lack of sensitivity originates images with
low contrast levels. Many infrared cameras compensate this lack of sensitivity by increasing
camera exposure times, which originate image blurring if the camera is under vibrations.
Also, infrared detectors usually have lower resolution than visual cameras. Thus, infrared
images usually have lower resolutions and details- than visual images.
Two images of the same scene are shown in Fig. 2: one taken with a visual camera (left) and
one taken with a FLIR ThermaCam P20 infrared camera (right). The differences in contrast,
noise levels and detail levels are noticeable.
2.2 Brief description of image stabilization methods
Assume that we have a sequence of images taken by one camera under vibrations. Each
image is represented by Im
t
(x,y), where t is the time when the image was captured. The
camera vibrations cause relative motions between the images in the sequence. Assume
image Im
0
(x,y) is considered the reference image. Stabilization of image Im
t
(x,y) consists of
detecting and cancelling the motion between Im
t
(x,y) and Im
0
(x,y), such that there is no
apparent motion between Im
0
(x,y) and the stabilized version of Im
t
(x,y). The process of
detecting the relative motion between two images is called image matching.
There are two main groups of image-matching methods: spatial domain and frequency
domain methods. Other techniques such as invariant moments (Abu-Mostafa & Psaltis,
1984) have poorer matching performance. A detailed survey can be found in (Zitov &
Flusser, 2003).

Automation

438

Fig. 2. Two images of the same scene but taken with different cameras one visual camera
(left) and one FLIR ThermaCam P20 infrared camera (right)
Spatial domain matching methods are based on local features that can be detected in both
images Im
0
(x,y) and Im
t
(x,y). Assume that both images contain enough and sufficiently
distributed features perceptible in both images. These feature-based methods usually have
four steps. The first step is to detect local features in both images. These features are
typically selected as corners (Tomasi, 1991), high-contrast points or local patterns with
invariant properties, (Bay et al., 2008). The feature detector is applied to both images. The
second step is to associate features detected in Im
0
(x,y) with features detected in Im
t
(x,y).
Maximum likelihood criteria are often used for feature association. This step provides
associations between features in both images, which can be used to estimate the relative
motion between them.
The third step is to determine the motion between Im
0
(x,y) and Im
t
(x,y) using the
aforementioned associations. The homography matrix is often used to model the motion
between two images since it allows describing the transformations originated by changes on
the location and orientation of the camera when the scene can be approximated by a plane
(Hartley & Zisserman, 2004). Methods such as Least Median of Squares and RANSAC
(Fischler & Bolles, 1981) are used to increase the robustness of the motion estimation. Once
m
0i
, the motion from Im
0
(x,y) to Im
t
(x,y), has been estimated, the fourth step is to apply
image-processing methods to induce in Im
t
(x,y) a motion inverse to m
0i
.
Spatial domain techniques have good performance if both images contain clear and robust
local features but are not suitable for infrared images since a low number of local features
can be robustly detected in infrared images.
On the other hand, frequency domain image-matching methods exploit the properties of
images in the frequency domain. Normalized cross-correlation, see e.g. (Barnea & Silverman,
1972) and (Segeman, 1992), has been a common approach to match images with relative
translations. In this case, the location of the maximum peak of the cross-correlation function
between Im
0
(x,y) and Im
t
(x,y) corresponds to the relative translation between Im
0
(x,y) and
Im
t
(x,y). Cross-correlation is commonly applied using the Fourier transform due to significant
computational savings. This method does not have good performance if Im
0
(x,y) and Im
t
(x,y)
have high noise levels: it produces broad peaks, which originate inaccuracies when


439
determining the location of the maximum peak. Some alternatives, such as the Symmetric
Phase-Only Matched Filtering (SPOMF) (Ersoy & Zeng, 1989) produce significantly sharper
peaks and are more resilient to noise, partial occlusions in the images and other defects.
Another disadvantage of Fourier transform methods is that they can only match translated
images. The Fourier-Mellin transform (FMT) improves also this disadvantage since FMT
methods can match translated, rotated and scaled images, (Chen et al., 1994).
2.3 Proposed frequency domain image stabilization method
The method proposed in this chapter uses image-matching methods based on frequency
domain transformations. The method relies on the global structure of the image and not on
local features. This fact provides significant robustness to noise and to the lack of contrast
and resolution. Moreover, the proposed method is capable of stabilizing sequences which
images are related through translations, rotations and scalings.
The proposed stabilization method adopts an incremental approach: the first image of the
sequence, Im
0
(x,y), is considered reference and each of the remaining images in the
sequence, Im
t
(x,y), is matched and corrected with respect to the stabilized version of the
previous image, Im
S
t-1
(x,y). Section 3 briefly describes the principles of the proposed method.
Section 4 illustrates the method with some examples.
Image matching can originate small -sub-pixel- errors. The accumulation of these errors in
an incremental stabilization approach can originate drifts. To avoid drifts we developed
another absolute stabilization approach in which all the images of the sequence are matched
and corrected with respect to the same reference image Im
0
(x,y). This and other practical
issues are discussed in Section 5.
The proposed method was customized to several problems with different stabilization
requirements. In all cases the method was implemented and validated. This is described in
Section 6.
3. Image matching using the Fourier-Mellin transform and Symmetric Phase-Only
Matched Filtering
Assume that Im
S
t-1
(x,y) is the stabilized version of the image gathered at time t-1. In the
incremental stabilization approach adopted Im
S
t-1
(x,y) is considered the reference image for
stabilization of Im
t
(x,y) at time t. Assume that Im
t
(x,y) is a rotated, scaled and translated
replica of Im
S
t-1
(x,y). The proposed stabilization method consists of two steps. The first one
detects and corrects the rotation and scaling between Im
S
t-1
(x,y) and Im
t
(x,y). Im
R
t
(x,y) is the
rotation and scaling corrected version of Im
t
(x,y). The second step detects and corrects the
translation in axes x and y between Im
S
t-1
(x,y) and Im
R
t
(x,y). The stabilized version of Im
t
(x,y),
i.e. the translation-corrected version of Im
R
t
(x,y), is denoted Im
S
t
(x,y).
Assume that s(x,y) and r(x,y) are the central rectangular regions of Im
t
(x,y) and Im
S
t-1
(x,y),
respectively. Thus, s(x,y) is a replica of r(x,y) rotated with an angle , scaled with a factor
and translated with translational offsets (x
0
, y
0
):
( ) - - -
0 0
( , ) ( cos sin ) , ( sin cos ) s x y r x y x x y y o o o o o o = + + (1)

Automation

440
The proposed method uses Symmetric Phase-Only Matched Filtering to match translated
images. Below the Symmetric Phase-Only Matched Filtering and the two aforementioned
steps in the algorithm are summarized.
3.1 Symmetric Phase-Only Matched Filtering
Assume that b(x,y) is replica of c(x,y) translated (x
0
,y
0
) in a noisy scene,
b(x,y)=c(x-x
0
,y-y
0
)+n(x,y), where n(x,y) represents a white zero mean random noise. A
traditional matched filter for b(x,y) has the following transfer function:

2
1
( , ) * ( , )
MF
w
H u v C u v
n
=
, (2)
where C*(u,v) is the complex conjugate of the Fourier transform of c(x,y) and |n
w
| is the
noise intensity. The output of the matched filter in (2) has a maximum peak at (x
0
,y
0
). The
location of this peak determines the translational offset (x
0
,y
0
).
One of the main limitations of traditional matched filters is that the output is affected by the
energy of the image while translation between images only influences the spectral phase. To
minimize this dependence the alternative adopted in our method is to use Symmetric Phase-
Only Matched Filtering (SPOMF) to determine translations between images. The SPOMF
between b(x,y) and c(x,y) is as follows:

| |
( , ) * ( , )
( , ) exp ( , )- ( , )
( , ) ( , )
SPOMF B C
B u v C u v
Q u v j u v j u v
B u v C u v
= = , (3)
where
B
(u,v) and
C
(u,v) are the spectral phases of b(x,y) and c(x,y), respectively. In the
absence of noise Q
SPOMF
(u,v)=exp[-j2(ux
0
+vy
0
)] and the inverse Fourier transform of
Q
SPOMF
(u,v) is a Dirac delta located at (x
0
,y
0
). SPOMF yields to a sharp peak, significantly
sharper than in case the matched filters in (2). The location of the maximum is easier to
estimate and more tolerant to noise, which, as previously mentioned, is a significant
advantage when dealing with infrared images.
3.2 Rotation and scaling correction step
This step aims to detect and correct the rotation and scaling between s(x,y) and r(x,y). The
corrected version will be called s
R
(x,y). From (1) the Fourier transform of s(x,y),
S(u,v)=Fourier{s(x,y)}, can be expressed by the following expression:

( )
| |
- - -
- -
2 1 1
( , ) ( cos sin ), ( sin cos ) exp ( , )
S
S u v R u v u v j u v o o o o o o o = + + , (4)
where R(u,v)=Fourier{r(x,y)} and
S
(u,v) is the phase of S(u,v). From (4) it is easy to notice
that |S(u,v)| is affected by rotations and scalings but is invariant to translations:

( )
- - -
-
2 1 1
( , ) ( cos sin ), ( sin cos ) S u v R u v u v o o o o o o o = + + (5)
The rotation and the scaling can be decoupled by converting |R(u,v)| and |S(u,v)| to polar
coordinates. Let r
p
(,) and s
p
(,) be |R(u,v)| and |S(u,v)| in polar co-ordinates (,):


441

( , ) ( cos , sin )
p
r R u u u =
(6)
( , ) ( cos , sin )
p
s S u u u = (7)
Thus, it is easy to check that:

( ) ( )
-
-
1
cos sin cos u v

o o o u o
o
+ =
(8)

( ) ( )
-
- -
1
sin cos sin u v

o o o u o
o
+ =
(9)
Hence, from (5), s
p
(,) can be expressed as:

-
-
2
( , ) ( , )
p p
s r u o u o o = (10)
Thus, the rotation of an angle has been transformed into a translation in the axis. Scaling
can be also transformed into a translation by applying a logarithimc scale on axis, i.e.
=log(). In log-polar coordinates (5) can be expressed as:

-
- -
2
( , ) ( , log( )) ( , log( ))
pl pl pl
s s r u u o u o o = = , (11)
where s
pl
(,) and r
pl
(,) are the log-polar versions of |S(u,v)| and |R(u,v)|. Thus, in (11)
the rotation and scaling have been transformed into translations. Let S
pl
(,) and R
pl
(,) be
the Fourier transforms of s
pl
(,) and r
pl
(,), respectively. Using shift properties of the
Fourier transform, (11) can be rewritten as:

{ }
( ) ( , ) exp - ( , ) ( , ) exp - ( , )-2 log( )
pl Spl pl Spl
S j R j u e u e u e u e t u o eo
(
= +

(12)
Thus, |S
pl
(,)|=|R
pl
(,)|. In (12) the rotation and the scaling appear as shifts between
Spl
(,) and
Rpl
(,), the phases of S
pl
(,) and R
pl
(,), respectively. The phase shift is:
( ) ( , )- ( , ) 2 log( )
Spl Rpl
o u e u e t u o eo = = + (13)
Then, if we apply a SPOMF, the location of the peak of q
r
(,), (
max
,
max
), represents the
rotation angle =
max
and the scaling factor
max
=log()

between s(x,y) and r(x,y):
| | { }
-
-1
( , ) exp
r
q Fourier j u o = , (14)
where Fourier
-1
stands for the inverse Fourier transform. The corrected image, Im
R
t
(x,y), is
obtained by rotating Im
t
(x,y) an angle -
max
and scaling by factor exp(
max
). The central
rectangular part of Im
R
t
(x,y), s
R
(x,y), is used for the translation correction step of the
algorithm.
3.3 Translation correction step
Once the rotation and scaling have been corrected, s
R
(x,y) and r(x,y) are related only by
translations in axes x and y. This step detects and corrects the translational offsets between
s
R
(x,y) and r(x,y). The SPOMF can be used to determine the translations between them:

Automation

442

| | - ( ) - ( ) ( , ) exp ( , , )
t SR R
Q u v j u v u v = , (15)
where
SR
(u,v) and
R
(u,v) are respectively the phases of S
R
(u,v) and R(u,v), the Fourier
transforms of s
R
(x,y) and r(x,y). The translational offset between s
R
(x,y) and r(x,y) can be
obtained by computing:
{ } ( )
-1
( , ) ,
t t
q x y Fourier Q u v = (16)
The peak of q
t
(x,y) is located at x=x
max
and y=y
max
. Thus, the translations between s
R
(x,y) and
r(x,y) in axes x and y are x
0
=x
max
and y
0
=y
max
. The stabilized image, Im
S
t
(x,y), is computed by
shifting Im
R
t
(x,y) by -x
0
in axis x and by -y
0
in axis y, respectively.
4. Experiments
This section presents some experiments that illustrate the proposed method. Two infrared
images gathered from an infrared camera under vibrations are shown in Fig.3. The image at
the left will be consider stabilized, Im
S
t-1
(x,y), and will be used as reference to stabilize the
image at the right, Im
t
(x,y). The method determined and corrected the relative rotation
angle, scaling factor and translations of Im
t
(x,y) with respect to Im
S
t-1
(x,y).

Fig. 3. Two infrared images from a sequence of images under vibrations. The image at the
left is taken as reference
First, the central parts of both images, r(x,y) and s(x,y), are selected, see Fig. 4. Their size
is chosen to be power of two in order to optimize the computational burden of the Fourier
Transform FFT algorithm. The original images are 640x480 and the central parts are
256x256.
The first step of the algorithm described in Section 3 starts. First, R(u,v) and S(u,v), the
Fourier transforms of r(x,y) and s(x,y) are computed. |R(u,v)| and |S(u,v)| are shown in
Fig. 5. It can be noticed that rotation between r(x,y) and s(x,y) results in a rotation between
|R(u,v)| and |S(u,v)| as predicted by (5).


443

Fig. 4. r(x,y) and s(x,y) for images in Fig. 3

Fig. 5. Left) 3D view of |R(u,v)|; Center) and Right) 2D views |R(u,v)| and |S(u,v)|,
respectively
Then, r
p
(,) and s
p
(,), the versions of |R(u,v)| and |S(u,v)| in log-polar co-ordinates are
computed. r
p
(,) and s
p
(,) are shown in Fig. 6. The colour represents the magnitude. As
predicted by (11) the rotation between |R(u,v)| and |S(u,v)| has been transformed into a
shift in the axis between r
p
(,) and s
p
(,). The value of the shift corresponds to the
rotation angle between Im
S
t-1
(x,y) and Im
t
(x,y).
Next, we apply the SPOMF between r
p
(,) and s
p
(,) using (13) and (14). The resulting
q
r
(,) is shown in Fig. 7left. The maximum peak in q
r
(,) is sharp. It is located at
max
=-
9.14,
max
=0. The second peak is 50% lower in magnitude and is located at high distance
from the maximum peak, which facilitates the identification of the maximum peak. Notice
that
max
=log(
max
)=0 involves
max
=1, i.e. there is no scaling between both images. The
rotation between both images is
max
=-9.14. Next, Im
t
(x,y) is corrected: Im
t
(x,y) is rotated an
angle - =9.14. Im
R
t
(x,y), the rotated-corrected version of Im
t
(x,y), is shown in Fig. 7right.
The central rectangular part of Im
R
t
(x,y), s
R
(x,y), is selected for the second part of the
method.

Automation

444

Fig. 6. Resulting r
p
(,) (left) and s
p
(,) (right)

Fig. 7. Left) Resulting q
r
(,). Right) Im
R
t
(x,y), i.e. Im
t
(x,y) after rotation correction
Once the rotation and scaling between Im
S
t-1
(x,y) and Im
t
(x,y) have been cancelled, the
following step is to compute the translations between s
R
(x,y) and r(x,y). The SPOMF is
computed using (15) and (16). The resulting q
t
(x,y) is in Fig. 8left. The peak was very sharp.
The maximum peak was located at x
max
=-4 and y
max
=7. Thus, s
R
(x,y) and r(x,y) are translated
x
0
=-4 in axis x and y
0
=7 in axis y. Next, Im
R
t
(x,y) is shifted by -x
0
and -y
0
in axes x and y
respectively in order to compensate the vibration. Fig. 8right shows Im
S
t
(x,y), the stabilized
version of Im
t
(x,y), i.e. the translation-corrected version of Im
R
t
(x,y).
5. Practical issues
This section presents practical aspects that have been developed to increase the robustness,
accuracy and efficiency of the proposed method.
5.1 Image enhance
Preliminary experiments revealed that applying preprocessing image enhance methods
improved significantly the performance of the proposed method. An efficient histogram
stretching method was used. The objective is to transform the image levels such that the
enhanced image has desired mean intensity -MI
ref
- and contrast -C
ref
- values. The histogram
stretching implements the following linear transformation function:


445
( ) ( , ) ( , )-
ref
E ref
C
Im x y Im x y MI MI
C
= + , (17)
where C and MI are the contrast and mean values before the histogram stretching. Im(x,y)
and Im
E
(x,y) are respectively the images before and after applying image enhance methods.
Image enhance increases the level of noise in the images. Although higher noise levels have
harmful effects in feature-based matching methods, the high robustness to noise of SPOMF
avoids the degradation in our case. In images with very low contrast levels it is interesting
to apply edge detectors. Fig. 9 shows the resulting images after applying Sobel edge detector
to r(x,y) and s(x,y) in Fig. 4.

Fig. 8. Left) Resulting q
t
(x,y). Right) Im
S
t
(x,y), stabilized version of Im
t
(x,y)

Fig. 9. Resulting images after applying Sobel edge detector to r(x,y) and s(x,y) in Fig. 4
5.2 Drift correction
Constrained resolutions of digital images generate small sub-pixel errors in image matching.
These errors have no perceptible consequences unless they accumulate originating drifts in
incremental stabilization schemes. The simplest way to avoid drifts is to adopt an absolute
stabilization approach, in which images are stabilized with respect to the same reference
image Im
0
(x,y). Different schemes were analysed to update the reference images depending

Automation

446
on the application. The reference image is considered valid if the overlapping between
Im
t
(x,y) and Im
0
(x,y) is above a certain threshold. Otherwise, a new reference image is
needed.
5.3 Operating modes
The resolution in the computation of , , x
0
and y
0
is highly dependent on the size of the
images. Although it is not easy to establish mathematically this dependence in real noisy
images, it is possible to analytically obtain some qualitative conclusions.
Assume that the size of r(x,y) and s(x,y) is MxM and that the size of s
p
(,) and r
p
(,) is
WxW. The resolution of the rotation angle depends on the size of S
p
(,) and R
p
(,), i.e.
WxW. The minimum detectable rotation angle is
min
~2/W. The value of W also has
influence on the errors in the computation of the rotation angle. The higher W, the higher
accuracy in the computation of the rotation angle. SPOMF produces broader peaks in q
r
(,)
in case of using smaller W. The value of W depends on the number of different radius values
considered in the polar conversion, which is constrained by the size of S(u,v) and R(,),
which size is MxM.
The size of r(x,y) and s(x,y), i.e. M, has influence on the resolution and errors in the
computation of the translations. Low values of M involve broader peaks in q
t
(x,y), which
involves poorer peak detection.
To deal with different stabilization accuracies and computer burden, two operation modes
have been selected: Mode1 and Mode2. Mode1 uses low values of M and W. Mode1 has
moderate stabilization capability. Mode2 uses high values of M and W. Mode2 stabilizes more
accurately at the expense of higher computer burden. Mode2 is applied when high accuracy
is required or in case of vibrations of high magnitude.
5.4 Increase accuracy through sub-pixel resolution
As described in the above sub-section the resolutions in the computation of , , x
0
and y
0

are limited by sizes of q
r
(,) and q
t
(x,y) and thus, by the sizes of the images. Noise broadens
the peaks of the SPOMF. A sub-pixel estimation method is used to determine with accuracy
and robustness the location of the peak in q
r
(,) or q
t
(x,y). This sub-pixel method detects the
maximum of q
r
(,) or q
t
(x,y) and defines a neighborhood around it. The centroid in the
neighborhood around the maximum peak is considered as the location of the peak. This
simple method allows incrementing the accuracy in peak localization. It also increases the
robustness of the maximum peak detection in case q
r
(,) and q
t
(x,y) have secondary peaks
near the maximum peak.
6. Implementation
The proposed method has been implemented in various problems illustrating its flexibility.
6.1 Stabilization of images from a hovering UAV
The proposed technique has been tested with the HERO3 helicopter, see Fig. 10 developed
by the Group of Robotics, Vision and Control (GRVC) at the University of Seville (Spain).


447
The application is to perform building inspection and particularly the detection of thermal
heat losses on the building envelope. The main perception sensors on board HERO3 are an
Indigo Omega infrared micro-camera and a visual camera. The Indigo Omega use infrared un-
cooled detectors, with low sensitivity and high noise levels. The helicopter is equipped with
an onboard PC-104 computer for stabilizing the infrared images and applying heat losses
detection.

Fig. 10. HERO3 during building inspection experiments carried out at the School of
Engineering of Seville (Spain)
The vibrations in the images have diverse origins. When the UAV is hovering, the images
can be affected by low-frequency oscillations originated by the compensations of the UAV
control systems. Also, images are affected by high-frequency vibrations induced by the UAV
engine. When the helicopter is hovering at certain position and orientation, the low-
magnitude distortions between consecutive images can be approximated by translations and
rotations.
The method was customized for the application. It was assumed that there is no change in the
scaling factor between images: it only corrected rotations and translations. The onboard
infrared micro-camera weights 120 gr. and has a resolution of 160x120 pixels. r(x,y) and s(x,y)
were selected of size 64x64. During the inspection the objective is to stabilize images when
regions of interest are present in the images. An automatic function was developed to detect
when an object of interest -windows in this problem- is present. At that moment the image
stabilization method is triggered and this image is used to initialize the stabilization method.
Fig. 11 shows consecutive infrared images taken from an infrared camera onboard HERO3
in an experiment carried out in December 2005. The image noise level is rather high. The
translations between the images can be observed. The stabilized images are in Fig. 12. The
image stabilization time was 12,6 ms in a PC-104 with computing power similar to a
Pentium III 800 MHz.
6.2 Stabilization of infrared images from a ship
The proposed method has been used for stabilization of images from an infrared camera on
a ship. The objective was to detect other ships and obstacles using automatic computer
vision methods. The stabilization method should keep the images stable despite the sea
condition and ship motion. This application required a high-performance infrared camera

Automation

448
and optical system with high resolution and low noise level. Infrared images have some
difficulties in sea scenes. Water absorbs infrared radiation: water appears in images with
low radiation levels. Sky also appears in infrared images with low radiation levels. The
horizon line can be well perceived in infrared images.

Fig. 11. Sequence of consecutive images from an Indigo Omega onboard HERO3

Fig. 12. Stabilized sequence of images


449
In images with no ships or objects, the stabilization method should greatly rely on the
horizon line, what is useful to compensate rotations and translations in the vertical axis but is
not enough to correct translations in the horizontal axis. If the image contains objects the
method can correct also translations in the horizontal axis. Fortunately, in this application the
method had to operate only if some threats for the ship navigation were found. An automatic
tool was used to detect the presence of objects and to trigger the stabilization method.
The implemented version included the image enhance methods to deal with the lack of
contrast. The resolution of the cameras used was 512x512. Hence, r(x,y) and s(x,y), input of
the algorithm, were selected of size 256x256 in Mode1 and 512x512 in Mode2. A short video
showing the performance of the method can be found in (web1). In this video the ship was
in the port and the camera was pointing at nearby forested area. Only very low magnitude
vibrations can be observed. The method corrects the vibrations except the sub-pixel errors
described in Section 5. The translational offsets x
0
(left) and y
0
(right) computed along one
sequence of images are shown in Fig. 13. The time to stabilize one image was 102,1 ms in
Mode1 and 346,6 ms in Mode2 in a computer with processing power similar to a Pentium III t
800 MHz.

Fig. 13. Values of x
0
(left) and y
0
(right) obtained in the stabilization of a sequence of 1000
images from an infrared camera mounted on a ship
6.3 Stabilization of images from a ground vehicle in motion
The infrared camera is mounted on a ground vehicle. The camera is located on a pole at
certain height and pointing such that it has a general view of an area. The vehicle can be
static or moving on the ground. If the vehicle is static the method employed is similar to that
used in Section 6.2. If the vehicle is moving, the stabilization method compensates rotations
and vertical translations but do not cancel horizontal translations. When the vehicle is
moving on the horizontal plane, even in the absence of vibrations, the images have
horizontal translations. Without using any other sensors but the camera itself, it is not
possible to differentiate if a horizontal translation is originated by the vehicle motion or by
vibrations.
Fig. 14 shows 6 consecutive images taken from one camera under vibrations. The camera
moves on the horizontal plane along the image sequence. The images resulting from the
complete stabilization version of the method -taking the first image as reference- are in Fig.
15. The objects appear static along the sequence. The reduction in the overlapping between
the images generates black patches of increasing size on the corrected images. The resulting

Automation

450
images after partial stabilization, i.e. without correcting the horizontal translations, are
shown in Fig. 16. The scene appears static except for the horizontal axis.

Fig. 14. Consecutive images taken from a camera under vibrations and moving on the
horizontal plane
.

Fig. 15. Images resulting from the complete stabilization version of the method
Both versions, complete and partial stabilization, were developed, each of them with two
modes: Mode1 -image size is 256x256- and Mode2 -image size is 512x512. An operator can
change the option (complete or partial stabilization) and mode during operation. In both
cases the stabilization times were similar to those in Section 6.2.


451

Fig. 16. Images resulting from partial stabilization. The scene appears static except for the
horizontal axis
7. Conclusions
This chapter presents a technique for stabilization of sequences of infrared images using
frequency domain methods. The work is motivated by the lack of local features and high
noise levels commonly present in infrared images, which hampers stabilization methods
based on feature matching. Instead, the proposed method relies on the global structure of
the image. The described method makes extensive use of Fourier-Mellin transforms and
Symmetric Phase Only Matched Filtering. The Fourier-Mellin transform allows determining
the rotation, scaling and rotation between two images by converting the images to log-polar
coordinates. Symmetric Phase Only Matched Filtering enhances the performance of image
matching in the presence of noise.
The main strength of the proposed method is the capability to deal with images in which it
is difficult to find clear and repeatable features. Its flexibility and capability for balancing
between stabilization accuracy and computational cost is also remarkable and allows its
customization to applications with different requirements.
The chapter also concerns several practical aspects that have been considered to increase its
robustness and accuracy including image contrast enhance and the definition of modes with
different performance and computer burden. The method was implemented for real-time
execution with low computing resources. Different versions were implemented and
validated in several applications. Three of them are briefly summarized in the chapter.
Software implementation and real-time execution have been two main requirements in the
design of the methods. They originated the development of stabilization modes to balance
between accuracy and computer cost. The implementation of the stabilization method in
FPGA is interesting to reduce the computer burden of the main processor. In this case,
Mode2, with better stabilization performance, could be used in all conditions.

Automation

452
The combination of visual and infrared images provides interesting synergies in a growing
number of problems. The differences between images from cameras in different spectral
bands would also hamper the application of feature-based methods. The use of the
proposed method for matching images in different spectral bands is object of current work.
8. References
Abu-Mostafa, Y.S. & Psaltis, D. (1984). Recognition aspects of moment invariants. IEEE
Trans. Pattern Anal. Mach. Intel., Vol. 16, No. 12, pp. 1156-1168, ISSN 0162-8828
Barnea, D. I. & Silverman, H. F. (1972). A class of algorithms for fast image registration. IEEE
Trans. Computers C-21, Vol. C-21, No. 6, pp. 179-186, ISSN 0018-9340
Bay, H; Ess, A.; Tuytelaars, T; Van Gool, L. (2008). SURF: Speeded up Robust Features.
Computer Vision and Image Understanding, Vol. 110, No. 3, pp. 346-359, ISSN 1077-
3142
Carvajal, E; Jimenez-Espadafor, F; Becerra, J.A.; Torres, M. (2011). Methodology for the
Estimation of Cylinder Inner Surface Temperature in an Air-Cooled Engine. Applied
Thermal Engineering, Vol. 31, No. 8-9, pp. 1474-1481, ISSN 1359-4311
Chen, Q.-S.; Defrise, M.; Deconinck, F. (1994). Symmetric phase-only matched filtering of
Fourier-Mellin transforms for image registration and recognition. IEEE Trans. on
Pattern Analysis and Machine Intel., Vol. 16, No. 12, pp. 1156-1168, ISSN 0162-8828
Ersoy, O.K. & Zeng, M. (1989). Nonlinear matched filtering. J. Opt. Soc. Am. A., Vol. 6, No. 5,
pp. 636-648, ISSN 1084-7529
Fischler, M.A. & Bolles, R.C. (1981). Random Sample Consensus: A Paradigm for Model
Fitting with Applications to Image Analysis and Automated Cartography.
Communications of the ACM, Vol. 24, No. 6, pp. 381395, ISSN 0001-0782
Hartley, R. & Zisserman, A. (2004). Multiple View Geometry in Computer Vision. Cambridge
University Press, ISBN 0521540518, Cambridge (United Kingdom)
Hudson, R.D. (1969). Infrared System Engineering. John Wiley & Sons, ISBN 0471418501,
New York (USA)
Maldague, X.P.V. (2001). Theory and Practice of Infrared Technology for Nondestructive
Testing. John Wiley & Sons, ISBN 0471181900, New York (USA)
Segeman, J. (1992). Fourier cross correlation and invariant transformations for an optimal
recogntion of functions deformed by affine groups. J. Opt. Soc. Am. A, Vol. 9, No. 6,
pp. 895-902, ISSN 1084-7529
Tomasi, C. (1991). Shape and motion from image streams: A factorization method, PhD
Thesis Carnegie Mellon University.
Zitov, B. & Flusser, J. (2003). Image registration methods: a survey. Image and Vision
Computing, Vol. 21, No. 11, pp. 977-1000, ISSN 0262-8856
http://www.youtube.com/watch?v=inPUcDIHC4s, Retrieved on 14/03/2012

23
SITAF: Simulation-Based
Interface Testing Automation
Framework for Robot Software Component
Hong Seong Park and Jeong Seok Kang
Kangwon National University
South Korea
1. Introduction
Many researchers in robotics have proposed a Component-based Software Engineering
(CBSE) approach to tackle problems in robot software development (Jawawi et al., 2007).
Especially in the component-based robot system, the system quality depends on the quality
of each component because any defective components will have bad effects on the system
built with them. Thus, component interface test is critical for checking the correctness of the
components functionality. It is especially difficult to test robot software components
because of the following two main problems.
First, the preparation of all hardware modules related to robot software and the
configuration of a test environment is labor-intensive. Second, it is difficult to define or
generate test cases for testing robot software components.
The simulation plays an important role in the process of robotic software development.
The simulation allows testing of robot software components and experimentation with
different configurations before they are deployed in real robots. Traditional simulation-
based approaches (Hu, 2005, Martin & Emami, 2006, Michel, 2004) focus on architectures
or methods (e.g., computer-based simulation, hardware-in-the-loop-simulation, and
robot-in-the-loop- simulation), rather than testing. Many software engineering researchers
(Buy et al., 1999, Bundell et al., 2000, Zamli et al., 2007, Momotko & Zalewska, 2004,
Edwards, 2001) have investigated software component testing, but they have not
considered simulation environments. Simulations can be used within a specification-
based testing regime, which helps robot software developers define and apply effective
test case. Note that the generation of test case is an important approach in the field of
automated testing.
In this paper, we propose a Simulation-based Interface Testing Automation Framework
(SITAF) for robot software components. SITAF automatically generates test cases by applying
specification-based test techniques and considering simulation-dependent parameters. SITAF
also performs the interface testing in distributed test environments by interacting with a
simulation. SITAF controls test parameters during testing, which affect the behavior of a
component under testing (CUT); examples of such parameters are simulation-dependent

Automation

454
parameters, input/output parameters of provided/required CUT interface. The main
advantage of this technique is that it identifies errors caused by interactions between the CUT
and the external environment.
The primary contribution of this paper is the integration of specification-based test into the
simulation for automatic interface testing of robot software components.
The rest of this paper is organized as follows. Section 2 presents the SITAF architecture. The
two main functions of SITAF are presented in Section 3. Section 4 discusses the evaluation of
SITAF. Finally, we have some conclusions in Section 5.
2. SITAF architecture
The main aim of SITAF is to automate as much of test process for robot software component
as is possible. To achieve this aim, the architecture of proposed framework consists of a
Web-based Interface Testing Automation Engine Server (ITAES), a Test Build Agent (TBA),
and a robot simulator and is shown in Fig. 1.

Fig. 1. Simulation-based Interface Testing Automation Framework Architecture
ITAES is the core of the framework to which the user is accessible via a Web service, and
generates test cases for interface test of robot software components. And it also generates a
test driver component, a test stub component, and a simulation control component, which
are required for the testing, and links the generated components to each other. Furthermore
ITAES manages test resources such as test cases, test applications, test results, and test logs
in a unit of test activity which is a workspace for execution of test operations. The ITAES
consists of three main modules of Interface Test Case Generator (ITCG), Interface Test
Application Generator (ITAG), and Automatic Test Build Manager (ATBM).
The ITCG module automatically generates test cases by extended test schemes based on
specification-based testing techniques such as equivalence partition (Ostrand & Balcer,

SITAF: Simulation-Based Interface Testing Automation Framework for Robot Software Component

455
1998), boundary value analysis (Hoffman et al., 1999), random test (Ntafos, 1997, Zhu,
1997), and pairwise test (Williams, 2000). It receives the interface representation
information in the form of Interface Definition Language (IDL) or eXtensible Markup
Language (XML) and verifies the specification information for a CUT before test cases are
automatically generated. The test cases are stored as XML files in a database. The user
accesses a Web interface to modify test cases in the database and inputs the expected
result values for each test case. The ITAG module generates the source code for the test
application. The test application is composed of a test driver component, a test stub
component, and a simulation control component. All source codes are shared with TBAs.
The ATBM module connected with the distributed TBAs manages a test build which
means compilation and execution of a test application. And it provides three types of the
test time for the test build: immediate, reserved, and periodic.
An individual TBA can exist in different test environments, and communicates with the
ATBM in ITAES. TBAs are in charge of automatic building of test application. The TBA
contains three modules of Test Build Agent Manager (TBAM), Test Application Compiler
(TAC), and Test Application Executor (TAE). The TBAM module manages a TBA and
receives a test build request from the ATBM in ITAES, and then downloads the test
application and test case files. The TAC module and the TAE module automatically compile
and execute a test application. These modules upload the logs and the test results to ITAES
after the completion of compilation and execution.
The robot simulator for interface test provides a simulation control API and a virtual robot
hardware API. The simulation control API is used to control the virtual test environment in
the simulation. The simulation control component in the test application dynamically
modifies or controls a virtual test environment for each test case by using the simulation
control API. The virtual robot hardware API is used to control virtual robot hardware or to
receive data. If the test component is a hardware-related component, the component
controls the virtual robot hardware or receives data using the API for simulation. The
simulation control data is used to generate effective test cases.
3. Automatic interface test operations for robot software components
In this section, we describe two main functions of SITAF, which are the automatic
generation of interface test case and the automatic execution of interface test by simulation.
3.1 Automatic generation of interface test cases
Specification-based test techniques are applied to the generation of test cases for robot
software components by simulation. A test case for a robot software component consists of
an input vector that requires Test Data of Input (TDI), Test Data of Simulation Dependency
(TDSD), and Test Data of Test Stub (TDTS) because the behavior of CUT is affected by these
test data. TDI is an input parameter of the interface under testing. Because TDSD refers the
simulation control data for interface testing, it is the data affecting the CUT through the
simulation. TDTS is an output data from a required interface of the CUT.
The process for the automatic interface test case generation has two steps, which is shown in
Fig. 2.

Automation

456

Fig. 2. Process of the automatic interface test case generation
3.1.1 Definition of test specifications
The Interface Parser module in the Interface Test Case Generator parses the interface
representation of a CUT and the information on input test parameters of CUT is extracted.
Fig. 3 shows a simplified XML schema of the interface representation. The main elements of
the interface representation are type_name, method, and param. The type_name represents type
of the interface and the method and param describe prototype of methods in the interface.
The test specification includes valid range values, specific candidate values, pre/post-
conditions, and other values, for each test parameter. And a test model is a set of the test
specifications, as shown in Fig. 4. The essential elements of the test model are name,
input_spec_list, and order_of_combinations. The name is a name of the test model, as identifier.
The input_spec_list includes test specifications of TDI, TDSD, and TDTS. The
order_of_combinations is a number of interaction strength for pairwising the each test
parameters.
In Fig. 4, the plus sign (+) indicates that the element further consists of test_element which is
used to generate the test data values. The test_element is a key element in the test
specification, which contains information on a test parameter such as type of the parameter,
specific candidate value, and the method for test data generation. The type in test_element
presents the type of a test parameter such as TDI, TDSD, and TDTS. The test_spec describes a
method and additional information for generating test data, and has different XML schema
for each type of test_spec, as shown in Fig. 5, Fig. 7, and Fig. 8.


457
<xs:schema>
<xs:element name="service_port_type">
<xs:complexType>
<xs:sequence>
<xs:element name="type_name" type=NCName/>
<xs:element ref="method" maxOccurs="unbounded" />
</xs:sequence>
</xs:complexType>
</xs:element>
<xs:element name="method">
<xs:complexType>
<xs:attribute name="name" type="xs:NCName" use="required"/>
<xs:attribute name="return_type" type="xs:NCName" use="required"/>
<xs:attribute name="call_type" type="xs:NCName" default="blocking"/>
<xs:sequence>
<xs:element maxOccurs="unbounded" ref="param"/>
</xs:sequence>
</xs:complexType>
</xs:element>
<xs:element name="param">
<xs:complexType>
<xs:sequence>
<xs:element name="type" type="xs:NMTOKEN"/>
<xs:element name="name" type="xs:NCName"/>
</xs:sequence>
<xs:attribute name="index" type="xs:integer" use="required"/>
</xs:complexType>
</xs:element>
</xs:schema>
Fig. 3. Simplified XML Schema of an interface representation

<xs:schema>
<xs:element name=test_model>
<xs:complexType>
<xs:sequence>
<xs:element name=name type=xs:string use=required />
<xs:element name=description type=xs:string/>
+ <xs:element ref=pre_condition_spec_list />
+ <xs:element ref=input_spec_list use=required/>
+ <xs:element ref=output_spec_list />
+ <xs:element ref=post_condition_spec_list />
<xs:element name=order_of_combinations type=xs:integer
use=required/>
</xs:sequence>
</xs:complexType>
</xs:element>
<xs:element name=test_element>
<xs:complexType>
<xs:sequence name=name type=xs:string use=required/>
<xs:sequence name=type type=xs:string use=required/>
<xs:sequence name=description type=xs:string/>
<xs:sequence ref=test_spec use=required />
<xs:sequence ref=user_value_list />
<xs:attribute name=abstract_type type=xs:string use=required />
<xs:attribute name=real_type type=xs:string use=required/>
</xs:complexType>
</xs:element>
</xs:schema>
Fig. 4. Simplified XML Schema of a test model

Automation

458
3.1.2 Generation of test cases
The Test Data Generator (TDG) module in the ITCG generates the test data satisfying the
test specification for each test parameter. The TDG automatically generates the numeric test
data by applying an equivalence partitioning scheme (ECP), a boundary value analysis
scheme (BVA), and a random testing scheme. Furthermore this paper generates the test data
of string type using BVA and random testing scheme.
The ECP scheme (Ostrand & Balcer, 1998) is a software testing technique that divides input
data for a software unit into partitions of data from which test cases cane be derived. In
principle, test cases are designed to cover each partition at least once. This technique aims to
define test cases and uncover classes of errors, thereby reducing the total number of test
cases that must be developed. Additionally this paper defines types of equivalence class,
listed in Table 1. The TDG automatically generates test data by each type of equivalence
class. Fig. 5 shows simplified XML schema of test_spec element for ECP.

Type of equivalence class Description
NEC_NUMERIC_ONE_BOUDARY
The type includes just one boundary.
If x < 10, there are two equivalence classes,
x<10 and x>10.
NEC_NUMERIC_TWO_BOUDARY
The type includes two boundaries.
If -1 < x < 10, there are three equivalence
classes, x<-1, -1<x<10, x>10.
NEC_BOOLEAN
If x=true, there are two equivalence classes,
x=true, x=false.
NEC_NUMERIC_CONSTANT
If x=3, there are two equivalence classes,
x=3, x!=3
NEC_NUMERIC_SET
If x={-1,0,1}, there are two equivalence classes,
x={-1,0,1}, x!= {-1,0,1}.
Table 1. Types of equivalence class for ECP scheme

<test_spec type="ECP" >
<xs:complexType>
<xs:element ref= equiv_class />
</xs:complexType>
<xs:element name=equiv_class>
<xs:attribute name=type use=required>
<xs:simpleType>
<xs:restriction base=xs:string>
<xs:enumeration value= NEC_NUMERIC_ONE_BOUDARY/>
<xs:enumeration value= NEC_NUMERIC_TWO_BOUDARY/>
<xs:enumeration value= NEC_BOOLEAN/>
<xs:enumeration value= NEC_NUMERIC_CONSTANT/>
<xs:enumeration value= NEC_NUMERIC_SET/>
</xs:restriction>
</xs:simpleType>
</xs:attribute>
.
</xs:element>
</test_spec>
Fig. 5. Simplified XML schema of test_spec element for ECP scheme


459
The BVA scheme (Hoffman et al., 1999) is a software testing technique that designs tests
including representatives of boundary values. Values on the minimum and maximum edges
of an equivalence partition are tested. The values could be input or output ranges of a
software component. Boundaries are common locations for errors that result in
software faults, so they are frequently explored in test cases. Furthermore this paper defines
the offset value of boundary for generation of elaborate test data, as shown in Fig. 6. This
paper automatically generates the test data by the BVA such as the values of
min_low_off_set, minimum boundary, min_high_off_set, max_low_off_set, maximum
boundary, max_high_off_set, and additionally a middle value.

Fig. 6. Offset and test data of BVA scheme
In the BVA, there are just two types of equivalence class because the scheme needs range
values, as shown in Fig. 7.

<test_spec type="BVA" offset= >
<xs:complexType>
<xs:element ref= equiv_class />
</xs:complexType>
<xs:element name=equiv_class>
<xs:attribute name=type use=required>
<xs:simpleType>
<xs:restriction base=xs:string>
<xs:enumeration value= NEC_NUMERIC_ONE_BOUDARY/>
<xs:enumeration value= NEC_NUMERIC_TWO_BOUDARY/>
</xs:restriction>
</xs:simpleType>
</xs:attribute>
.
</xs:element>
</test_spec>
Fig. 7. Simplified XML schema of test_spec element for BVA scheme
A random testing scheme (Ntafos, 1997, Zhu, 1997) is a strategy that requires the random
selection of test cases from the entire input domain. For random testing, values of each test
case are generated randomly, but very often the overall distribution of the test cases has to
conform to the distribution of the input domain, or an operational profile. In this paper, the

Automation

460
scheme is used to generate test data of numeric and string types. In particular, we combine
the random testing scheme and the BVA scheme for generation of test data of string type.
This paper analyses the boundary value of minimum and maximum length of string, and
then randomly generates test data, which is satisfied with the options such as alphabet,
number, special character, space, and negative character, as shown in Fig. 8.

<test_spec type="RANDOM_STRING">
<xs:complexType>
<xs:sequence>
<xs:element name=min_length type=xs:integer use=required />
<xs:element name= max_length type=xs:integer use=required />
<xs:element name= alphabet type=xs:boolean use=required />
<xs:element name= number type=xs: boolean use=required />
<xs:element name= special_char type= xs:boolean use=required />
<xs:element name= space type=xs:boolean use=required />
<xs:element name= negative_char_list type=xs:string />
</xs:sequence>
</xs:complexType>
</test_spec>
Fig. 8. Simplified XML schema of test_spec element for random string scheme
The Test Data Combinator (TDC) module in ITCG combines the test data using a pairwise
scheme (Williams, 2000) for reducing the number of test cases. The pairwise scheme is an
effective test case generation technique, which is based on the observation that most faults
are caused by interactions among input vectors. The TDC enables two-way combination,
three-way combination, and all possible combinations of the test data, which allows the user
to remove overlapping test cases from pairs of parameter combinations. The combined test
cases are stored in an XML file, as shown in Fig. 9. The case_param_info describes name and
type of a test parameter and the case_list consists of values of the test case.

<xs:schema>
<xs:element name=test_suite_data>
<xs:complexType>
<xs:sequence>
<xs:element name=name type=xs:string use=required />
<xs:element name=description type=xs:string/>
<xs:element name=test_ model_name type=xs:string />
+ <xs:element ref=case_param_info />
+ <xs:element ref=case_list />
</xs:sequence>
</xs:complexType>
</xs:schema>
Fig. 9. Simplified XML schema of the test cases
3.2 Automatic test execution by simulation
The test application performs testing by interacting with the robot simulator, as shown in
Fig. 10. This paper automatically generates skeleton source codes for test applications and
links the components to each other for simplifying testing.


461

Fig. 10. Structure and operation sequence of a test application
The test application consists of a Test Driver (TD) component, a Simulation Control (SC)
component, a Test Stub (TS) component, and a CUT, as shown in Fig. 10. The TD component
controls the overall operation of test. During the testing, the TD component reads test cases,
and sets the simulation environment and the required interface of the CUT. After the end of
testing, the component stores the test results in a file. The SC component sets the simulation
environment through the simulation control API in the robot simulator. TS component
provides virtual interfaces of the same type as the required interface of the CUT. The TS
component simulates the behavior of CUT-dependent software components. Thus, the
component is used instead of an actual software component which is needed for execution
of the CUT.
The test application and the robot simulator are connected to each other and the following
operations shown in Fig. 10 are performed to test the CUT: 1) Read test case file, 2) Call the
interface of SC component for control to the simulation environment, 3) Set up the TS
component using the TDTS values, 4) Call the interface of the CUT, 5) Save the test results in
a file.
The TD component reads a test case file and divides it into the TDI, the TDSD, and the
TDTS. The TD component calls the interface of SC component using the TDSD values for
setting the simulation environment. The SC component changes the virtual test

Automation

462
environment through the simulation control API using TDSD values. After the virtual test
environment setup is completed, the TD component set up the output of the required
interface of the CUT via the TS component interface using the TDTS. After the
configuration of the simulation environment and the TS component are completed, the TD
component calls interface of the CUT using the TDI as the input parameters. The CUT
calls the interface of the TS component and requests or receives data via the robot
hardware API during the simulation. When the operation is completed, the value
resulting from the operation may be returned to the TD component. The TD component
compares the actual resulting value with the expected resulting value and saves the test
result in a file. After all of the testing, the TBA uploads the test result file and the test log
file to ITAES. The test log file contains log information on compilation and execution of
the test application.
Fig. 11 shows simplified XML schema of the test result descriptor. The summary consists of
the number of pass and fail, and information on processing times of the interface of the CUT
such as a minimum time, a maximum time, an average time, and a standard deviation time.
The test_result_list contains detailed information on the test result, such as expected and
actual test results and a processing time of the interface of the CUT, for each test case. The
ITAES reads the test result file and log files, and then shows the information through web
interfaces which are table-based view and graphic-based view, for easily analysing the test
result of the CUT.

<xs:schema>
<xs:element name=test_result_descriptor>
<xs:complexType>
<xs:sequence>
<xs:element name=name type=xs:string />
+ <xs:element ref=summary />
+ <xs:element ref=test_result_list />
</xs:sequence>
</xs:complexType>
</xs:element>
</xs:schema>

Fig. 11. Simplified XML schema of a test result descriptor
4. Evaluation
In this section, the proposed framework is evaluated using an example of the test of the
Infrared Ray (IR) sensor component interface.
This paper implements the Interface Testing Automation Engine Server (ITAES) and Test
Build Agent (TBA) in Java and Flex. And the robot simulator used in this paper is OPRoS
simulator (http://www.opros.or.kr/). This paper develops the test simulation environment
for testing shown in Fig. 12. The environment consists of an IR sensor robot which has some
virtual IR sensors linked to the IR sensor simulation API and an obstacle which can move by
the obstacle distance control API.


463

Fig. 12. Application for testing the IR component interface
We tested an Open Platform for Robotic Service (http://www.opros.or.kr/) with the
Infrared Ray (IR) sensor component interface, named GetInfraredData interface, which had
two input parameters of IndexOfSensor and NumOfSensor. The function of the interface is to
get a distance value using IR sensors. This paper defines a new test parameter for TDSD,
named Distance, to control the virtual obstacle in the test simulation environment. Thus,
there are two input test parameters and one simulation-dependent parameter, which are
shown in Table 2.

Automation

464

Name Type Test Specification Description
IndexOfSensor TDI 1 <= IndexOfSensor <= 7,
The offset value is 1.
It is index of IR sensor.
NumOfSensor TDI 1 <= NumOfSensor <= 7,
The offset value is 1.
It is the number of IR sensor.
Distance TDSD 0.0 <= #Distance <= 10.0,
The offset value is 0.5.
It is a distance between an IR
sensor and an obstacle in the
test simulation environment.
The obstacle is moved using
the value of the distance test
parameter.
Expected
Return Value
If all test parameter are valid
values, the value is same of
the distance value.
If the test parameters are
not valid values, the value
is -1.
-1 means the operation is
failed.
Table 2. Test Specifications and an expected test result of GetInfraredData Interface
The test application contains the Test Driver (TD) component, OPRoS IR sensor component,
and the virtual obstacle control component. The test application does not have a test stub
component because the OPRoS IR sensor component does not have a required interface.
First the TD component moves the virtual obstacle in the simulation via the virtual obstacle
control component using the value of Distance test parameter. Then the TD component calls
the interface of the IR sensor component using values of the IndexOfSensor and the
NumOfSensor test parameter. If the return value of the interface of the IR sensor component
is same of Distance value, the test case is a success.
We validate three functions of proposed framework, which are the creation of test activity,
the automated test case generation, and the automatic test execution by simulation.
The process of the creation of test activity has four steps, which are shown in Fig. 13(a) Fig.
13(d). In the creation of the test activity, the information on the IR sensor component such as
component profile, dll file, and interface profile, external library (optional), and the type of
the skeleton test code are used.
Fig. 13(e) shows the generated directory of the test activity after completion of the process.
The directory includes the test profile, the test driver component, the test stub component,
and the concrete test driver.


465

Fig. 13. Process of creating the test activity for IR sensor component
The process of automatic test case generation has four steps, which are shown in Fig. 14(a)
Fig. 14(d). We input the test specifications through the web user interface, which are the
IndexOfSensor, the NumOfSensor, and the Distance. We analyse valid range value of each test
parameter using BVA scheme. And then we select 2-way, as order of combination for
pairwise. After completion of the process, 60 test cases are generated as shown in Fig. 15.
Then we input the expected test result into each test case.

Automation

466

Fig. 14. Process of automatic test case generation of GetInfraredData interface


467

Fig. 15. Test Case list of GetInfraredData interface
We perform and evaluate the example of test application for the IR sensor component in the
test simulation environment, as shown in Fig. 12. Finally we compare the return value of the
interface of the IR sensor component with the value of expected test result. Fig. 16 shows
C++ code of test driver component used in this evaluation.

Fig. 16. Test driver code for testing GetInfraredData Interface

Automation

468
Two types of test result view are provided and shown in Fig. 17. The table-based view
shows the test result of each test case. And the graphic-based view shows the summary test
result, including the percentage of test success and the processing time of the IR sensor
component. The example used shows that the proposed framework is working well.

Fig. 17. Test result views of GetInfraredData interface


469
5. Conclusions and future works
This paper proposes and develops the interface testing framework, SITAF, based on
simulation and specification-based test for robot software components and develops the
automatic test case generation technique for interface testing. SITAF uses three types of the
test parameters, which are the input parameter, the test stub parameter, and the simulation
dependent parameter and applies specification-based test techniques. SITAF also performs
the automatic interface testing to identify errors caused by CUT interactions with an
external environment.
SITAF is evaluated via the example of the test of the IR sensor component used in the
distance measurement. The example shows that the SITAF generates test cases and performs
the automatic interface testing by interactive simulation.
As future works, are considering an automatic regression test by applying software
configuration management, and the mixed test environment of a simulation-based
environment and a real environment for testing robot software component.
6. References
A. Martin, A. & Emami, M. R. (2006). An Architecture for Robotic Hardware-in-the-Loop
Simulation, Proceedings of the International Conference on Mechatronics and
Automation, pp.2162-2167, June 2006
Bundell, G. A.; Lee, G., Morris, J. & Parker, K. (2000). A Software Component Verification
Tool, Proceedings of the Conf. Software Methods and Tools, pp. 137-146, 2000
Buy, U.; Ghezzi , C., Orso , A., Pezze, M. & Valsasna, M. (1999). A Framework for Testing
Object-Oriented Components, Proceedings of the 1st International Workshop on Testing
Distributed Component-Based Systems, 1999
Edwards, S. H. (2001). A framework for practical, automated black-box testing of
component-based software, International Journal of Software Testing, Verification and
Reliability, Vol. 11, No. 2, pp. 97-111, June, 2001
Hoffman, D.; Strooper, P. & White, L. (1999). Boundary Values and Automated Component
Testing, Journal of Software Testing, Verification, and Reliability, Vol. 9, No. 1, 1999, pp.
326,
http://www.opros.or.kr/
Hu, X. (2005). Applying Robot-in-the-Loop-Simulation to Mobile Robot Systems, Proceedings
of the 12th International Conference on Advanced Robotics, pp. 506-513, July 2005
Jawawi, D.N.A.; Mamat, R. & Deris, S. (2007). A Component-Oriented Programming for
Embedded Mobile Robot Software, International Journal of Advanced Robotics
Systems, Vol.4, No.2, 2007, pp.371-380, ISSN 1729-8806
Michel, O. (2004). Webots: Professional Mobile Robot Simulation, International Journal of
Advanced Robotic Systems, Vol. 1, No. 1, 2004, pp. 39-42, ISSN 1729-8806
Momotko, M. & Zalewska, L. (2004). Component+ Built-in Testing: A Technology for
Testing Software Components, Foundations of Computing and Decision Sciences, pp.
133-148, 2004
Ntafos, S. (1998). On Random and Partition Testing, Proceedings of International Symposium on
Software Testing and Analysis (ISSTA), 1998, pp. 4248

Automation

470
Ostrand, T.J. & Balcer, M.J. (1998). The Category-Partition Method for Specifying and
Generating Functional Tests, Communications of the ACM, Vol. 31, No. 6, 1988, pp.
676686
Williams, A.W. (2000). Determination of test configurations for pair-wise interaction
coverage, Proceedings of the 13th Conf. Testing of Communicating Systems, pp. 59-74,
August 2000
Zamli, K. Z.; Isa, N. A. M., Klaib, M. F. J. & Azizan, S. N. (2007). A Tool for Automated Test
Data Generation(and Execution) Based on Combinatorial Approach, International
Journal of Software Engineering and Its Applications, Vol. 1, No. 1, pp. 19-35, July, 2007
Zhu, H.; Hall, P.& May, J. (1997). Software Unit Testing Coverage and Adequacy, ACM
Computing Surveys, Vol. 29, No. 4, December 1997, pp. 366427

24
Advanced Bit Stuffing Mechanism
for Reducing CAN Message Response Time
Kiejin Park and Minkoo Kang
Ajou University
Republic of Korea
1. Introduction
As customer requirements for safety and convenience in automobiles increases, so does the
quantity of electronics and software installed in them. The amount of signal data from the
electronics systems needs to be managed, making the design of communication protocols for
in-vehicle networks (IVN) more important (Navet et al., 2005). The IVN protocols can be
classified into two paradigms: event-triggered and time-triggered (Obermaisser, 2004). The
event-triggered protocols are efficient in terms of network utilization, because the messages
transmitted within event-triggered protocols are only transmitted when specific events
occur. This differs from time-triggered communication in that the response time of message
transmission is not predictable (Fabian & Wolfgang, 2006).
The controller area network (CAN) is a well-known event-triggered protocol originally
developed in the mid-1980s for multiplexing communication between electronic control
units (ECUs) in automobiles (ISO 11898, 1993). In recent years, CAN has been used in
embedded control systems that require high safety and reliability because of its appealing
features and low implementation costs (Navet et al., 2005; Johansson et al.; 2005). Appealing
features of CAN protocols are that the error detection mechanisms can identify multiple
types of error (e.g. bits error, bit stuffing error, cyclic redundancy checksum error, frame
error, and acknowledgement error). Moreover, error counters in a CAN controller can be
used to represent which states of the controller are associated with specific errors, which
include an error-active state, an error-passive state, and a bus-off state (Gaujal & Navet,
2005).
In spite of low implementation costs and wide acceptance of the CAN protocol in
automotive control systems and industrial factory automation, limited bandwidth and
nondeterministic response time have restricted the wider use of CAN in safety-critical real-
time embedded control systems such as x-by-wire applications (Rushby, 2003; Wilwert et al.,
2004). To mitigate the effects of these problems, the worst-case response time of a CAN
message should be reduced as much as possible. Calculating the worst-case response time of
CAN messages has been studied in order to guarantee its schedulability (Tindell et al., 1994,
1995), and this approach has been cited in over 200 subsequent papers. More recently, the
schedulability analysis of CAN has been studied as the revised version of the original
approach (Davis & Burns, 2007).

Automation

472
To reduce the length of CAN messages, the pre-processing mechanism using bitwise
manipulation before bit stuffing has been suggested (Nolte et al., 2002, 2003). According to
this mechanism, the worst-case response time can be reduced by minimizing stuffing bits in
CAN messages. However, this mechanism cannot be applied to CAN network systems
because the problem of message priority inversion has not been addressed. In our previous
work, to resolve the problem of message priority inversion, a mechanism with a new bit
mask for reducing the length of CAN message as well as preserving message priorities has
been proposed (Park et al., 2007). Subsequently, we found that the mechanism has a
problem which causes the frame shortening error and proposed the advanced bit stuffing
(ABS) mechanism for resolving the problems with the previous approach (Park & Kang,
2009). In this paper, we describe the ABS mechanism in detail and extend the generation
procedure of the bit mask of the ABS mechanism for the extended 2.0B frame format.
The outline of this paper is as follows. Section 2 presents a summary of the CAN protocol,
and describes the impossibility of the worst-case bit stuffing scenario proposed by Nolte et
al. Then calculating response time of CAN messages is presented. In Section 3, the ABS
mechanism for reducing CAN message response time using generation of a new mask is
described in detail. Also, we describe the examples of problems with priority inversion and
frame shortening error. In Section 4, we evaluate the performance of the ABS mechanism
with various CAN message sets. Finally, Section 5 concludes the paper.
2. Background
2.1 CAN message frame format
Controller area network (CAN) is the ISO standard for communication in automotive
applications. It is designed to operate at network speeds of up to 1 Mbps for message
transmission. Each CAN message contains up to 8 bytes of data (Farci et al., 1999). The
frame format of a CAN message is classified into two categories that include the standard
2.0A with 11-bit identifier and the extended 2.0B with 29-bit identifier. Furthermore,
message transmission over a CAN is controlled by four different types of frame: data frame,
remote transmit request (RTR) frame, overload frame, and error frame (Etschberger, 2001).
Fig. 1 shows the format of a CAN standard 2.0A data frame.

Fig. 1. Standard 2.0A data frame format of a CAN message.
As shown in fig. 1, a data frame consists of start-of-frame (SOF), arbitration field, control
field, data field, acknowledgement (ACK) field, and end-of-frame (EOF). An SOF bit marks

Advanced Bit Stuffing Mechanism for Reducing CAN Message Response Time

473
the beginning of a data frame, and is represented by one dominant bit (value=0). The
arbitration field consists of an 11-bit identifier and a dominant delimiter bit. The identifier
indicates the priority of the message. A message with identifier 00000000000 has highest
priority, and a message with identifier 11111111111 has lowest priority. The last four bits of
the control field are called the data length code (DLC). Its value represents the length of data
field. Data field contains up to 8 bytes of data to be transmitted. The CRC field consists of a
15-bit CRC code and a recessively transmitted delimiter bit. The ACK field has two delimiter
bits. The EOF consists of a sequence of 7 recessive bits (Etschberger, 2001).
In recent years, a luxury car may incorporate as many as 2500 signals exchanged by up to 70
ECUs (Albert, 2004). In the standard 2.0A frame format of a CAN message, the length of the
identifier is 11 bits. This means that 2048 different CAN messages are distinguishable in the
CAN communication system, so, the 11-bit identifier is insufficient to distinguish all signals.
For this reason, the extended 2.0B frame format with 29-bit identifier has been defined. Fig.
2 shows the format of a CAN extended 2.0B data frame (Pfeiffer et al., 2003)

Fig. 2. Extended 2.0B data frame format of a CAN message.
2.2 Worst-case bit stuffing scenario
When a CAN node detects an error in a transmitted message, it transmits an error flag
which consists of six bits of the same polarity. The bit stuffing mechanism prevents six
consecutive bits from having the same polarity by inserting a bit of opposite polarity after
the fifth bit. Moreover, the main purpose of the bit stuffing mechanism is used to
synchronize transmitter and receiver when the same values are to be transmitted
consecutively (Nolte et al., 2001). Bits exposed to bit stuffing are from an SOF bit to a 15-bit
CRC code without a CRC delimiter (see Fig. 1 and Fig. 2). The stuffing bits of the received
frame are removed at the receiving node before the message is processed (Wolfhard, 1997).
The worst-case scenario of the bit stuffing has been presented as shown in Fig. 3 (Nolte et
al., 2007).

Fig. 3. The worst-case scenario of the bit stuffing.

Automation

474
According to the worst-case scenario, the number of bits of a CAN message is given by:
g + 8s + 1S + j
g+8s-1
4
[ (1)
where g is 34 for the standard format or 54 for the extended format, s is the number of data
bytes of a CAN message (Nolte et al., 2007). However, it is impossible that stuffing bits be
inserted in the worst-case scenario. The causes are:
1. Several bit values are fixed by the CAN frame format (e.g., the SOF bit and delimiter
bits in the arbitration field and the control field).
2. The DLC in the control field depends on its number of data bytes.
3. The CRC field of a CAN message depends on the bit sequence from an SOF bit to the
data field.
Accordingly, the worst-case number of stuffing bits in a standard 2.0A data frame is
reduced by 21~40% from previous values.

Fig. 4. The CAN arbitration process.
2.3 Bitwise bus arbitration
Each ECU of a CAN network can initiate the transmission of a message as soon as the bus is
free. Because it may happen that more than one ECU begins a message transmission at the
same time, an arbitration process is necessary. To prevent the ECUs from destroying each
others transmitted data, the message with the highest priority of all simultaneously
arbitrating messages is determined in an arbitration phase. As mentioned in Section 2.1, the
message having the lowest value message identifier is assigned highest priority
(Etschberger, 2001). Fig. 4 shows the CAN arbitration process. Each ECU monitors the signal
level on the bus during the arbitration phase. The arbitation phase consists of the
transmission of the message identifier and of the RTR bit. If an ECU detects a dominant bus


475
level, although it has switched a recessive level itself, it aborts the transmission process
immediately, as in this case a message with higher priority is obviously being transmitted at
the same time, and goes into the receive state (Etschberger, 2001).
2.4 Response time model of CAN messages
The worst-case response time of a CAN message m can be calculated as the sum of the
queuing delay t
m
and the transmission delay C
m
as follows:
R
m
= t
m
+C
m
(2)
The queueing delay t
m
is composed of the blockng time B and the interference and is given
by:
t
m
= B + _
t
m
+]
]
+:
bit
1
]
_ C
] ]hp(m)
(3)
where the set bp(m) consists of all the messages in the system of higher priority than
message m, [
]
is the jitter on the queueing of the message ], I
]
is the transmission period of
the message ], and
bt
is the transmission time for a single bit.
The blocking time B can be calculated by the transmission time of the longest CAN message
within the system. The transmission delay C
m
can be calculated by multiplying the number
of bits of the message m as in (1) and
bt
(Tindell et al., 1995).
3. Advanced Bit Stuffing (ABS) mechanism
As mentioned in Section 1, the mechanism proposed by Nolte et al. cannot be applied to
reduce the length of CAN messages because the message transmission priorities can be
shuffled as shown in Table 1.

High priority CAN
ID
Low priority CAN
ID
Description
Original CAN
Message
00001010111 01111110101
Low bit value has higher
priority
XORing
(by Nolte et al.)
01011111101 00101011111 Mask: 01010101010
Bit Stuffing Message 010111110101 001010111110 Priority inversion occurs.
Table 1. Counter example of priority inversion problem.
To solve the problem of priority inversion, a mechanism for minimizing the length of CAN
messages in bit stuffing, and for preserving message priorities, has been proposed (Park et
al., 2007). However, the previous mechanism contains a flaw which causes frame shortening
errors. The frame shortening error means that the receiver anticipates a frame of different
length than the original (Charzinski, 1994; Tran, 1999). It can occur when the first bit and the
last four bits of the control field are changed by the XOR operations with the bit mask. An
example of the frame shortening error is shown in Fig. 5.

Automation

476

Fig. 5. Example of the frame shortening error.
As shown in Fig. 5, two bits are changed by bit errors. The first error is the bit in the DLC
and the second one is the ACK slot. Thus the receiver can expect a message with a smaller
than the original message.
In this section, we propose an advanced bit stuffing (ABS) mechanism which adopts XOR
operations and prevents priority inversion and frame shortening errors at the same time. In
order to develop the ABS mechanism, an assignment scheme for CAN message identifiers
and generation rules of a new XOR bit mask are presented.
3.1 Message identifier assignment
To better understand the number of bits used for message identifiers in CAN-based control
systems, we assumed that there are two assignment schemes of message identifiers. The first
scheme is to assign to messages consecutive identifiers starting from 1. In this scheme, when
the number of message identifiers is n, the number of used bits is |log
2
n]. For instance, if the
system requires 256 messages, than the number of used bits for message identifiers are 8. On
the other hand, the second scheme is based on the grouping of message identifiers in
accordance with their level of importance. In this scheme, the number of used bits can be
calculated by:
n
uscd
= |log
2
m] +|log
2
n] (4)
where m and n represent the number of groups and the maximum number of identifiers in a
group, respectively. For example, the system requires 4 message groups and each group
consists of up to 32 message identifiers. In this case, only 7 bits are used for message
identifiers. 2 bits out of 7 bits are required for representing message groups and 5 bits out of
7 bits are required for representing message identifiers of each group. Because the first
scheme is a special case (i.e., m = 1) of the second scheme, in this paper, we have applied the
second scheme in order to assign message identifiers.


477
3.2 XOR mask generation for standard 2.0A frame format
In order to generate a new mask for standard 2.0A frame format, we assumed that the bits
for group and those for group identifiers are assigned to the most significant bit (MSB) and
the least significant bit (LSB) field in the arbitration field, respectively. When the number of
data bytes (s), the number of groups (m) and the maximum number of identifiers in a group
(n) are determined, the following mask generation procedure is constructed for the standard
2.0A frame format.
1. The length of a mask is the length of the bits exposed to bit stuffing, 8s +S4. The mask
is initially set to 010101
2. A value of 0 is assigned to |log
2
m] bits of the MSB and |log
2
n] bits of the LSB of the 11-
bit identifier in the mask.
3. A value of 0 is assigned to the RTR bit of the arbitration field in the mask.
4. The string 010000 is assigned to 6 bits of the control field in the mask. Then, the mask
can be generated as depicted in Fig. 6.

Fig. 6. XOR mask generation for standard 2.0A frame format.
3.3 XOR mask generation for extended 2.0B frame format
We extended the mask generation procedure in Section 3.2 for extended 2.0B frame format.
In the same manner, we assumed that the bits for a group and those for group identifiers are
assigned to the MSB and the LSB field in the arbitration field of the extended 2.0B frame
format, respectively. When the number of data bytes (s), the number of groups (m) and the
maximum number of identifiers in a group (n) are determined, the following mask
generation procedure is constructed for the extended 2.0B frame format.
1. The length of a mask is the length of the bits exposed to bit stuffing, 8s +S4. The mask
is initially set to 010101
2. A value of 0 is assigned to |log
2
m] bits of the MSB and |log
2
n] bits of the LSB of the 29-
bit identifier in the mask.
3. A value of 00 is assigned to 2 medial bits of the arbitration field, and a value of 0 is
assigned to the RTR bit of the arbitration field in the mask.
4. The string 010000 is assigned to 6 bits of the control field in the mask. Then, the mask
can be generated as depicted in Fig. 7.

Automation

478

Fig. 7. XOR mask generation for extended 2.0B frame format.
In both standard and extended frame format, r0 bit should be a dominant bit. But the bit of
r0 bit location in the generated mask is assigned to a recessive bit in the generation
procedure for both standard and extended frame format. This assignment is for separating
XOR masked CAN messages from original CAN messages. If an ECU receives a CAN
message with dominant bit of r0 bit location, the received message is unmasked, and
otherwise (i.e., a CAN message with recessive bit level of r0 bit location), a received message
is masked.

Fig. 8. Implementing the ABS mechanism
3.4 Guidance for implementing the ABS mechanism
In Section 3.2 and Section 3.3, the XOR masks are generated for standard and extended
frame format. The ABS mechanism reduces the number of stuffing bits in the CAN message
by a bitwise manipulation using the XOR masks (Fig. 8).
4. Performance evaluation
As in our previous work, the example of the case of |log
2
m] = 2 illustrates the efficiency of
the proposed mechanism (Park et al., 2007). For the SOF bit, the arbitration field, and the
control field, the expected number of stuffing bits has been calculated with a variable


479
number of |log
2
n]. The expected number of stuffing bits in the original messages is shown
in Table 2. For comparison, the expected number of stuffing bits in the XOR masked
messages is shown in Table 3, where P(X = i) represents the probability that the message
has i stuffing bits, and E(X) is the expected number of the stuffing bits. In order to calculate
the probability, we assume that the probability of a bit having value 1 or 0 is equal.

|log
2
n] P(X = u) P(X = 1) P(X = 2) P(X = S) E(X)
1 0 0.234 0.594 0.172 1.938
2 0 0.305 0.609 0.086 1.781
3 0 0.324 0.633 0.043 1.719
4 0.086 0.402 0.477 0.035 1.461
5 0.157 0.477 0.341 0.025 1.234
6 0.199 0.509 0.268 0.024 1.117
7 0.251 0.561 0.176 0.012 0.949
8 0.260 0.572 0.159 0.009 0.917
9 0.260 0.572 0.159 0.009 0.917
Table 2. The expected number of the stuffing bits in the original messages.
In order to evaluate the performance of the ABS mechanism, the number of stuffing bits of
masked message frames is compared with the number of stuffing bits of the original
message frames. The original and masked message frames are generated by a simulation
program. The flowchart of the simulation program is shown in Fig. 9. Simulation
procedures are as follows: 1) Initially, the simulation program obtains input parameters. In
this simulation, the number of runs is 1000 and each CAN message has four data bytes; 2)
Then, the original message frame is generated and the number of stuffing bits is calculated;
3) Next, the bit mask is generated by using mask generation as shown in Section 3.2 and
Section 3.3, and masked message frame generated by XOR operation in both the bit mask
and the original message frame; 4) Finally, the number of stuffing bits of the masked
message frame is calculated. When a simulation is complete, the average and maximum
number of stuffing bits in the original and masked message frames are calculated,
respectively.
In this experiment, the number of runs is 1000 and the assumptions that are made in these
experiments are as follows: 1) the probability that a bit which does not have a specific value
can be set to 0 or 1 is the same, and 2) it is independent of other bit values, and finally, 3)
there is no transmission error during experiment.

Automation

480
|log
2
n] P(X = u) P(X = 1) P(X = 2) P(X = S) E(X)
1 1 0 0 0 0
2 1 0 0 0 0
3 0.875 0.125 0 0 0.125
4 0.875 0.125 0 0 0.125
5 0.875 0.125 0 0 0.125
6 0.844 0.156 0 0 0.156
7 0.754 0.234 0.012 0 0.258
8 0.728 0.254 0.018 0 0.291
9 0.728 0.254 0.018 0 0.291
Table 3. The expected number of the stuffing bits in the XOR masked messages.
Fig. 10 and Fig. 11 show the average and maximum number of stuffing bits in the standard
data frame as the number of used bits in the arbitration field changes, in both the original
frame and the masked frame. In the same manner, the number of stuffing bits in the
extended data frame is shown in Fig. 12 and Fig. 13. Here we can see that the more the
number of used bits decreases, the more does the number of stuffing bits of the original
message frame increase. However, the number of stuffing bits of the masked message frame
showed little variation regardless of the change in the number of used bits.
From the experimental results, it can be found that the expected number of stuffing bits is
reduced by the maximum of 58.3% with an average of 32.4%. This may effect the response
time of CAN messages because message response time is in proportion to the length of
message frame.
The average and maximum number of stuffing bits in the standard data frame with
|log
2
m] = 2, |log
2
n] = 4 as the number of data bytes changes, in both the masked frame and
the original frame, are shown in Fig. 14. In the same manner, the number of stuffing bits in
the extended data frame is shown in Fig. 15. As shown in Fig. 14 and Fig. 15, the number of
stuffing bits of the masked message frame is smaller than the number of stuffing bits of the
original message frame.
As mentioned in Section 2.4, the response time of CAN messages is determined by the
queuing delay and the transmission delay. If the number of stuffing bits decreases, both the
queuing delay and the transmission delay can be reduced. Furthermore, this may have
effects on the reduced network utilization of the embedded systems using CAN network
protocols.


481

Fig. 9. Flowchart of the simulation program.

Automation

482
The number of used bits for message idenfifiers
0 1 2 3 4 5 6 7 8 9 10 11
T
h
e

n
u
m
b
e
r

o
f

s
t
u
f
f
i
n
g

b
i
t
s
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Original, Maximum
Masked, Maximum
Original, Average
Masked, Average

Fig. 10. The number of stuffing bits in the standard frame with 4 data bytes.
The number of used bits for message identifiers
0 1 2 3 4 5 6 7 8 9 10 11
T
h
e

n
u
m
b
e
r

o
f

s
t
u
f
f
i
n
g

b
i
t
s
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Original, Maximum
Masked, Maximum
Original, Average
Masked, Average

Fig. 11. The number of stuffing bits in the standard frame with 8 data bytes.


483
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
T
h
e

n
u
m
b
e
r

o
f

s
t
u
f
f
i
n
g

b
i
t
s
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Original, Maximum
Masked, Maximum
Original, Average
Masked, Average

Fig. 12. The number of stuffing bits in the extended frame with 4 data bytes.
0 2 4 6 8 10 12 14 16 18 20 22 24 26 28
T
h
e

n
u
m
b
e
r

o
f

s
t
u
f
f
i
n
g

b
i
t
s
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Origianl, Maximum
Masked, Maximum
Original, Average
Masked, Average

Fig. 13. The number of stuffing bits in the extended frame with 8 data bytes.

Automation

484
The number of data bytes
1 2 3 4 5 6 7 8
T
h
e

n
u
m
b
e
r

o
f

s
t
u
f
f
i
n
g

b
i
t
s
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
Original, Maximum
Masked, Maximum
Original, Average
Masked, Average

Fig. 14. The number of stuffing bits in the standard frame with |log
2
m] = 2, |log
2
n] = 4
The number of data bytes
1 2 3 4 5 6 7 8
T
h
e

n
u
m
b
e
r

o
f

s
t
u
f
f
i
n
g

b
i
t
s
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
Original, Maximum
Masked, Maximum
Original, Average
Masked, Average

Fig. 15. The number of stuffing bits in the extend frame with |log
2
m] = 6, |log
2
n] = 9


485
5. Conclusions
Minimizing the response time of the CAN message is necessary to guarantee real-time
performance improvement. In this paper, an effective mechanism called advanced bit
stuffing (ABS) mechanism is presented. The ABS mechanism develops an assignment
scheme of CAN message identifiers and generation rules of a new XOR bit mask, to prevent
the problems with message priority inversion and frame shortening error. From the
experimental result, the number of stuffing bits of the masked message frame showed little
variation regardless of the change of the number of used bits. It has been also found that it is
more effective in embedded systems in which the number of CAN messages is less than that
of bits that are used for message priority.
6. References
Albert, A. (2004). Comparison of Event-triggered and Time-triggered Concepts with
Regards to Distributed Control Systems, Embedded World Conference, Nrnberg,
Germany.
Charzinski, J. (1994). Performance of the Error Detection Mechanisms in CAN, Proceedings of
the 1st International CAN Conference, Mainz, Germany.
Davis, R. & Burns, A. (2007). Controller Area Network (CAN) Schedulability Analysis:
Refuted, Revisited and Revised, Real-Time Systems, Vol. 35, No. 3, pp. 239-272.
Etschberger, K. (2001). Controller Area Network (CAN): Basics, Protocols, Chips, and
Applications, IXXAT Automation GmbH, ISBN 978-3000073762, Weingarten,
Germany.
Fabian, S. & Wolfgang, S. (2006). Time-Triggered vs. Event-Triggered: A Matter of
Configuration, Proceedings of the GI/ITG Workshop on Non-Functional Properties on
Embedded Systems, pp. 1-6.
Farsi, M.; Ratcliff, K.; & Barbosa, M. (1999). An Overview of Controller Area Network,
Computing & Control Engineering Journal, Vol. 10, pp. 113-120.
Gaujal, B. & Navet, N. (2005) Fault Confinement Mechanisms on CAN: Analysis and
Improvements, IEEE Transactions on Vehicular Technology, Vol. 54, pp. 1103-1113.
International Standards Organization. (1993). Road Vehicles - Interchange of Digital
Information Controller Area Network (CAN) for High-Speed Communication.
ISO 11898.
Johansson, K.H.; Torngren, M.; & Nielsen, L. (2005). Vehicle Applications of Controller Area
Network, Handbook of Networked and Embedded Control Systems, ISBN 0-8176-3239-5,
pp. 741-766.
Navet, N.; Song, Y.; Simonot-Lion, F.; & Wilwert, C. (2005). Trends in Automotive
Communication Systems, In Proceeding of the IEEE, Vol. 93, No.6, pp. 1204-1223.
Nolte, T.; Hansson, H.; Norstrom, C.; & Punnekkat, S. (2001). Using bit-stuffing distributions
in CAN analysis, IEEE/IEE Real-Time Embedded Systems Workshop (RTES01).
Nolte, T.; Hansson, H.; & Norstrom, C. (2002). Minimizing CAN Response-Time Jitter by
Message Manipulation, Eighth IEEE Real-Time and Embedded Technology and
Applications Symposium (RTAS'02), pp. 197-206.
Nolte, T.; Hansson, H.; & Norstrom, C. (2003). Probabilistic Worst-Case Response Time
Analysis for the Controller Area Network, Ninth IEEE Real-Time and Embedded
Technology and Applications Symposium (RTAS'03), pp. 200-207.

Automation

486
Obermaisser, R. (2004). Event-Triggered and Time-Triggered Control Paradigms, Sptinger-
Verlag, ISBN 978-1441935694.
Park, K.; Kang, M.; & Shin, D. (2007). Mechanism for Minimizing Stuffing-bits in CAN
Messages, The 33rd Annual Conference on the Industrial Electronics Society (IECON
07), pp. 735-737.
Park, K. & Kang, M. (2009). Advanced Bit Stuffing Mechanism for Reducing the Length of
CAN Messages, International Conference on Convergence Technologies and Information
Convergence (CTIC09).
Pfeiffer, O.; Ayre, A.; & Keydel, C. (2003). Embedded Networking with CAN and CANopen, RTC
Books, ISBN 978-0929392783.
Rushby, J. (2003). A Comparison of Bus Architecture for Safety-Critical Embedded Systems,
NASA/CR, Technical Report, NASA/CR-2003-212161.
Tindell, K.W. & Burns, A.K. (1994). Guaranteed Message Latencies for Distributed Safety-
critical Hard Real-time Networks, Technical Report YCS 229, Department of
Computer Science, University of York.
Tindell, K.; Burns, A.; & Wellings, A. J. (1995). Calculating Controller Area Network (CAN)
Message Response Times, Control Engineering Practice, Vol. 3, No. 8, pp. 1163 1169.
Tran, E. (1999). Multi-bit Error Vulnerabilities in the Controller Area Network Protocol, Carnegie
Mellon University, Thesis.
Wilwert, C.; Navet, N.; Song, Y.-Q.; & Simonot-Lion, F. (2004). Design of automotive X-by-
Wire systems, The Industrial Communication Technology Handbook, R. Zurawski, Ed.
Boca Raton, FL: CRC.
Wolfhard, L. (1997). CAN System Engineering: From Theory to Practical Applications, Springer,
ISBN 978-0387949390.
25
A Systematized Approach to Obtain Dependable
Controller Specifications for Hybrid Plants
Eurico Seabra and Jos Machado
Mechanical Engineering Department, CT2M Research Centre, University of Minho
Portugal
1. Introduction
This chapter focuses on the problem that a designer of an automation system controller
must solve related with the correct synchronization between different parts of the controller
specification when this specification obeys a previously defined structure. If this
synchronization is not done according to some rules, and taking some aspects into
consideration, some dependability aspects concerning the desired behaviour for the system
may not be accomplished. More specifically, this chapter will demonstrate a systematized
approach that consists of using the GEMMA (Guide dÈtude des Modes de Marches et
dÀrrts) (Agence Nationale pour le Developpement de la Production Automatise)
[ADEPA], 1992) and the SFC (Sequential Function Chart) (International Electrotechnical
Commission [IEC], 2002) formalisms for the structure and specification of all the system
behaviour, considering all the stop states and functioning modes of the system. The
synchronization of the models, corresponding to the controller functioning modes and the
controller stop states, is shown in detail and a systematized approach for this
synchronization is presented. For this the advantages and disadvantages of the vertical
coordination and horizontal coordination proposed by the GEMMA formalism are
discussed and a case study is presented to explain the proposed systematic approach. A
complete safe controller specification is developed to control a hybrid plant. Also this
chapter presents and discusses a case study that applies a global approach for considering
all the automation systems emergency stop requirements. The definition of all the
functioning modes and all the stop states of the automation system is also presented
according the EN 418 (European Standard [EN], 1992) and EN 60204-1 (EN, 1997) standards.
All the aspects related to the emergency stop are focused in a particular way. The proposed
approach defines and guarantees the safety aspects of an automation system controller
related to the emergency stop. For the controller structure the GEMMA methodology is
used; for the controller entire specification the SFC is used and for the controller behaviour
simulation the Automation Studio software (FAMIC, 2003) is used.
In order to achieve the goals presented above, the chapter is organized as follows: section 1
presents the challenge addressed to this chapter; in section 2 the main formalisms and
methodologies used to define the controller behaviour specification are presented, namely,
showing how to deal with complex specifications before the implementation into a physical
controller device to help the designer to improve the specifications performance; section 3

Automation 488
discusses different possible approaches for using the coordination of a controller
specification when this specification is previously structured by the use of the GEMMA
method. This is often applied because the complexity of the specification behaviour
demands a separate modular specification, named task, corresponding to the functioning
or failure modes or stop states of the automation system; section 4 presents a case study and
shows in detail defining and structuring a controller specification; section 5 presents and
illustrates how to coordinate a complex specification applied to a case study with possible
extrapolation for similar cases; section 6 is exclusively devoted to the discussion of different
possibilities for emergency stop application; further section 7 presents and discusses an
example of emergency stop application and, finally, section 8 presents some conclusions.
2. Formalisms used to develop the controller behaviour specification
From the desired behaviour specifications, until the implementation of a controller program
for an automation system, the controller designer needs to use some different and
complementary methods, formalisms and tools that help him in all the necessary steps.
Taking into account aspects related to the systems dependability, the designer must be able
to use together these formalisms and tools in order to achieve the desired behaviour for the
system. There are many methods, formalisms and tools for helping the designer during all
necessary steps. For the structure of the controller it is possible to use the GEMMA method
(ADEPA, 1992), Multi-Agent formalism (Sohier, 1996) can be used: for the specification Petri
Nets (Murata, 1989), SFC (IEC, 2002), Statecharts (Harel, 1987), UML (Booch et al., 2000) can
be used; for the implementation, the PLCs (Programmable logic controllers) (Moon, 1994),
Industrial computers (Koornneef and Meulen, 2002), Microprocessors (Brusamolino et al.,
1984) and others can be used.
From the analysis of needs, passing by the conception, realization into the implementation
and exploitation of an automation system there are several steps that must be realized
(Fig. 1). During each step of the controller design a corresponding step of the development
of the plant (physical part of the system: motors, cylinders, sensors etc.) exists. For instance,
step 3 corresponds to the specification of the controller and step 3 corresponds to the
specification of the plant.
The main objective of this chapter is to show how to deal with complex specifications before
the implementation into a physical controller device. Usually there are some methods,
formalisms and tools that help the designer to improve the specifications performance, but if
the coordination of all the parts of the specification is not well done, some aspects related to
the dependability of the system may not be accomplished.
This paper applied a case study and then extrapolated to systems of the same kind, in this it
is more detailed and more related to steps 3 and 4 presented in Figure 1, than related to the
design of a controller.
Currently, some suitable methods and formalisms for the development and creation of the
structure and specification of an automated production system controller exist. Among
them there are GEMMA (Guide dtude des Modes de Marche et dArrt) and SFC
(Sequential Function Chart), both developed in France. GEMMA is well adapted to defining
the controller structure and SFC is well adapted to complete controller specification.

A Systematized Approach to Obtain Dependable Controller Specifications for Hybrid Plants 489

Fig. 1. Steps considered on the design of an automation system
According to SFC rules, the implementation of the automation system requires, in
particular, a description relating to cause and effect. To do this, the logical aspect of the
desired behaviour of the system will be described. The sequential part of the system, which
is accessed via Boolean input and output variables, is the logical aspect of this physical
system. The behaviour indicates the way in which the output variables depend on the input.
The object of the SFC is to specify the behaviour of the sequential part of the systems. This
formalism is characterized, mainly, by its graphic elements, which, associated with an
alphanumerical expression of variables, provides a synthetic representation of the behaviour
based on a description of the situation of the system.
GEMMA was developed in France by ADEPA (Agence Nationale pour le Developpement
de la Production Automatise) and is a method that on the basis of a very precise
vocabulary proposes a simple structured guide for the designer based on a graphical chart
that contains all the functioning modes and stop states that a machine or an automated
system can assume. It is a tool for helping with system analysis, being used for its
supervision, maintenance and evolution definition.
The GEMMA method is based in three basic concepts:
- The operating modes are seen, from the point of view of the command module, as
always available. All the systems are composed by a command module and an
operative module. In the application of GEMMA, it is assumed that the command
module is always on power.
- The production criteria. Two states are considered for the production systems: ON
production and OUT of production. Those states are shown on the graphical chart of
the method.
- The three groups of functioning modes and stop states of the plant, namely:
- States A: Stop states
- States D: Failure modes
- States F: Running modes

Automation 490
The graphical chart of GEMMA is composed of three parts, each one corresponding to each
group of functioning modes and stop states described in Figure 2.

Fig. 2. Functioning modes and stop states considered on the design of an automation system
3. Coordination of a complex specification
Very often modes other than F1 (Normal Production Mode) demand a specification
behaviour with complex cycles and this complex specification demands also a specific
treatment for each of the functioning or failure modes or stop states of the automation
system.
So, it seems useful to separate each modular specification corresponding to each of the
functioning or failure modes or stop states of the automation system. Each specification
module is named task: a task is associated with the F1 mode, other tasks to the F2 mode
and so on for all the functioning and failure modes and stop states of the automation
system.
From a practical point of view, and for implementation of SFC, the division of tasks is
particularly adapted to these needs and it is possible to make the correspondence between a
mode/state, the respective SFC and the respective task, based on the specification SFC for
the corresponding mode (Fig. 3).


Fig. 3. Structure of a task
3.1 Horizontal coordination
This is a very interesting way of coordinating tasks because any task can be dominant over
the others and also each task may launch other tasks (Fig. 4).

Fig. 4. Horizontal coordination
Lets consider a generic task F1 (normal production mode) and a generic task F3 (closing
mode). As F3 appears after task F1, F1 must launch F3. When F1 ends (step 16) the Boolean
variable X16 makes possible the evolution of the task F3, from the step 100 to the step 101.
At the end of task F3 (step 120) the next task is launched by the Boolean variable X120 and
so on with similar behaviour, task by task. The variables Xi are Boolean variables associated
to step i defined by (IEC, 2002).

Automation 492
3.2 Vertical coordination
This kind of coordination is hierarchic and there are several levels of abstraction. Each task
of an abstraction level may launch any other task at a lower level, but - on the same
abstraction level one task cannot launch other tasks that belong to the same level (Fig. 5).
With this hierarchical approach the designer may have a global overview of the system and
also, if he intends so, a very detailed local view of the system.
The synchronization process is illustrated in a very detailed way in Figure 5 and the SFC of
higher level coordinates, at the specific order, indicates the evolution of each task. After the
end of each task, the higher level SFC evolutes and, on its next steps, it will launch another
task of the inferior abstraction level - and so on.

a) b)
Fig. 5. Vertical coordination
This approach is easier to systematize and also better adapted to treat more complex
systems because of the use of different abstraction levels.
4. Case study
The case study corresponds to an automatic machine for filling and capping bottles (Fig. 6).
This is divided in three modules, transport and feeding, filling and capping. For increasing
productivity, a conveyor is used with several alveoli for the bottles, allowing the operation
simultaneously at three working stations (modules of the automation system).
The transport and feeding station is composed of a pneumatic cylinder (A) that is
responsible for feeding the bottles to the conveyor and another pneumatic cylinder (B) that
executes the step/incremental advance of the conveyor.
The filling module is composed of a volumetric dispenser, a pneumatic cylinder (C) that
actuates the dispenser and an on/off valve (D) for opening and closing the liquid supply.
The capping station has a pneumatic cylinder (G) to feed the cover, a pneumatic motor (F) to
screw on the cover and a pneumatic cylinder (E) to advance the cover. The cylinder (E)
moves forward until the existent cover retracts with this cover during the retraction of (G),
and it moves forward again with rotation of the motor F to screw on the cover.


Fig. 6. Case study plant
4.1 Base controller behaviour specification
Figure 7 shows the base SFC of the system controller, corresponding only to the normal
production mode. The basic sensors involved are: two end-course-sensors for each cylinder
(example: cylinder A, sensor a0 and a1, respectively, retracted and forward) and a sensor of
pressure e1, which detects the point of contact/stop of the cylinder E in any point of its course.
Valve D and motor F do not have position sensors because they are difficult to implement.
On the other hand, in order to obtain the total SFC controller, which includes all the
operation modes required for the correct operation of the system, the graphic chart of
GEMMA was used because it allows definition of the functioning (operation) modes and
stop states of the machine.

Fig. 7. Base SFC controller specification

Automation 494
4.2 Global controller structure
Figure 8 shows the GEMMA graphic chart developed for the presented case study. The
considered tasks are described as follows:

Fig. 8. GEMMA of the plant controller
A1 The task A1 stop in the initial state represents the task of the machine represented in
Figure 8.
F1 Start of task A1, when the start command of the machine occurs, the change for the task
F1 normal production happens (filling and automatic capping) with the consequent
execution of base SFC presented in Figure 7.
A2 When happens the stop command of the machine happens, the run cycle finishes in
agreement with the condition described at task A2 stop command at the end of cycle.
F2 When the machine is empty (without bottles) it is necessary to feed bottles
progressively, the machine being ready to begin the normal production (task F1) when it has
bottles in the conveyor positions of the production modules 2 and 3, respectively. This
operation is defined by task F2 preparation mode.
F3 The closing mode of task F3 allows the reverse operation, that is, the progressive stop
of the machine with the exit of all of the bottles (emptying of the machine).

D3 When the capping module is out of service it can be decided to produce in another
way, that is, to perform the bottle filling in an automatic way and posterior manual capping,
this is the main purpose of task D3 production in another way.
D1 In the case of a situation emergency, task D1 emergency stop is executed. This stops
all the run actions and closes the filling valve to stop the liquid supply.
A5 After the emergency stop (task D1), cleaning and verification are necessary: this is the
purpose of task A5 prepare to run after failure.
A6 After the procedures of cleaning and verification finish it becomes necessary to perform
the return to the initial task of the machine, as described at the task A6 O.P. (operative
plant) in the initial state.
F4 For example, for volume regulation of the bottle liquid dispenser and adjustment of the
bottle feeder, a separate command for each movement is required, according to task F4
unordered verification mode.
F5 For detailed operation checks, a semiautomatic command (only one cycle) it is
necessary to check the functioning of each module: task F5 ordered verification mode.
To make this possible GEMMA evolution becomes necessary, creating transition conditions
for the run and stop operation modes, as described previously.
These transition conditions will be accomplished using GEMMA, as presented, to proceed:
- To allow the progressive feeding demanded in the preparation way (F2) and the
progressive discharge required in the closing way (F3) it will be necessary to consider
sensors that detect the bottles presence under each one of the modules 1, 2, 3,
respectively, CP1, CP2, CP3 (see Fig. 6);
- Also, it will be necessary a command panel that supplies the transition conditions given
by an operator (Fig. 9).

Fig. 9. Command panel of the system controller
In the command panel, there is a main switch that allows selecting the automatic,
semiautomatic and manual operation modes.

Automation 496
Automatic option corresponds to:
- Two buttons, start and stop, whose actions are memorized in memory M;
- A switch HS3 to put module 3 in service or out of service;
- A switch AA to control the bottles feeding permission (cylinder A), to allow the
emptying of the machine.
These switches/buttons and sensors CP1, CP2 and CP3 are the transition conditions of the
tasks A1, F1, F2, F3, A2 and D3, as shown in Figure 8.
The semiautomatic option corresponds to task F5 ordered verification mode which
allows the actuation of button (m) to check one cycle operation of each module selected by
the semiautomatic switch ,, or .
The manual option corresponds to tasks F4, A5 and A6, which require a separate
command from each movement using a direct command on the directional valves.
Finally, the AU button (emergency stop) allows passing to task D1 which starts from all of
the tasks.
5. Coordination of the case studys complex specification
The implementation of total controller specification, based on GEMMA presented in Figure
8, can be realized using the following two alternative methods, when one SFC for each task
is developed (Multiple SFC):
- Horizontal coordination;
- Vertical coordination.
As shown in section 3, there are several aspects/benefits for each described implementation
(vertical coordination and horizontal coordination). However, it seems to be more
systematic for vertical coordination because two levels of abstraction can be defined and
when the system is really complex, this aspect seems to be very helpful.
Figure 10 shows the schema of the adopted approach for the case study (vertical
coordination).
According Figure 10, GEMMA implementation is performed based on the following main
stages:
1. Elaboration of a high level SFC that directly translates the base GEMMA of the system
behaviour;
2. Elaboration of multiple low level SFCs corresponding to each functioning mode and/or
stop state;
3. Synchronization of the SFCs using the vertical coordination methodology.
5.1 High level SFC
This is the first stage of the vertical coordination implementation of total controller
specification. Figure 11 shows the high level SFC that directly corresponds to the base
GEMMA of the case study plant controller presented previously in the Figure 8.


Fig. 10. GEMMA implementation with vertical coordination of multiple SFC
A1
F2
Auto . start . CP2 . CP3
F1
CP3
F3
D3
A2
D1
A5
A6
F5
F4
Auto . start
auto
AU
=1
manual
=1
=1
manual
AA
Stop
=1
HS3
Stop
auto
Stop
auto AU

Fig. 11. High level SFC
5.2 Low level SFCs
The development of multiple low level SFCs specifications, corresponding to each one of the
functioning modes and/or stop states considered for the case study plant controller, is the
second stage of the vertical coordination implementation of total controller specification.

Automation 498
In this chapter, the SFCs specifications corresponding to each one of the functioning modes
and/or stop states will be shown. As mentioned before, when using the GEMMA approach,
each SFC corresponding to each functioning mode and/or stop state is treated as a task. In
this way, Figure 12 shows the SFC specification for the tasks F1 normal production and F2
preparation mode. The SFC of the task F3 closing mode is not shown because it is
similar to that presented for task F2. Additionally, Figure 13 shows the SFC specification for
tasks F5 ordered verification mode and D3 production in another way.

200
201
203
XF1
202
a -
b +
a1 . b0
b1
205
204
c+ d+
c -
c1
a+ b- 206
g +
207
e +
g1
208
e -
e1
209
g -
e0
210
g0
211
e -
e1
213
XF5 + XD1 + XF3 +
e+ f+
212
=1
F1
a0 . b0 . c0 . e0 . g0 .
(1/2 auto + AU + AA +
+ Stop + HS3)
+ XD3 + XA2
a0 . b0 . c0 . e0 . g0 .
/ (1/2 auto + AU + AA +
+ Stop + HS3)

100
101
AA . CP1
103
XF2
102
a -
b +
(a1 + CP1 + AA) . b0
b1
106
105 c+ d+
c -
c1
a+ b-
108 g +
109
e +
g1
110
e -
e1
111
g -
e0
112
g0
113
e -
e1
a0 . b0 . c0 . e0 . g0 . CP3
115
a0 . b0 . c0 . e0 . g0 . CP3
XF1
104
CP2
107
CP3
CP2
CP3
e+ f+
114
=1
F2

a) b)
Fig. 12. a) Low level SFC for the normal production task, and b) low level SFC for the
preparation task

It is of importance to note that the emergency stop related to GEMMA task D1 is not
treated in this section due to its complexity. The emergency stop controller behaviour
specification will be presented in detail in sections 6 and 7 of this chapter. In particular,
section 7 presents and discusses the implementation of GEMMA task D1 of the same case
study (Fig. 8), with the aim of applying a global approach considering all automation system
emergency stop requirements.
The definition of all functioning modes and all stop states of the automation system were
performed according European standards EN 418 (EN, 1992) and EN 60204-1 (EN, 1997).

300
301
303
XF5
302
a -
a1 . b0
305
304 c+ d+
c -
c1
a+ b- 306 g +
307
e +
g1
308
e -
e1
309
g -
e0
310
g0
311
e -
e1
auto
313
auto + manual
XF1 + XF4
e+ f+
auto 1 . m auto 2 . m auto 3 . m
312
=1
F5

800
801
803
XD3
802
a -
b +

a1 . b0
b1
805
804
c+ d+
c -
c1
a+ b-
807
Stop
XA2
806
D3
Stop

a) b)
Fig. 13. a) Low level SFC for the ordered verification task, and b) low level SFC for the
production in another way task.
5.3 SFCs synchronization
The last stage of the vertical coordination implementation of total controller specification is
related to the synchronization of the low level SFC specifications.

Automation 500
To achieve this as described in section 2, the high level SFC presented in Figure 14 was
completed with the SFC step activity/action (Xi - i step number) that correspond to the low
level SFC execution stop. Figure 14 shows the complete high level SFC obtained for vertical
coordination implementation (the SFC step activities added are represented in red).
All the controller specifications, presented in the previous figure, were simulated on
Automation Studio software. The obtained results led to the conclusion that all the
requirements defined on the Emergency Stop Standards were met.
Further, the specification was translated to Ladder Diagrams according to the SFC algebraic
formalization and implemented on a Programmable Logic Controller (PLC) adopted as the
controller physical device. This part of the developed work is not detailed in this chapter.

A1
F2
Auto . start . CP2 . CP3
F1
CP3 . X115
F3
D3
A2
D1
A5
A6
F5
F4
Auto . start
auto . X213
AU . X213
X551
manual
X602
X401
manual
AA . X213
Stop . X715
X901
HS3 . X213
Stop
auto
Stop .
AU . X501
. X213
. X807
. X313
auto
. X313

Fig. 14. High level SFC completed with low lever SFC step activities (in red).
All the SFC controller specifications, presented in the previous Figure, were simulated on
Automation Studio software (FAMIC, 2003). The obtained results led to the conclusion that
all of the automation system requirements were met.
formalization (IEC, 2002) and implemented on a Programmable Logic Controller (PLC)

adopted as the controller physical device. This part of the developed work is not detailed in
this publication.
6. Emergency stop controller behaviour specification
The Emergency Stop is one of the most important aspects related to the safety of people,
goods and equipments that interact with automated systems.
In order to obtain safe controllers, it must obey some rules (EN, 1992), (EN, 1997):
- A fault in the software of the control system must not lead to hazardous situations;
- Reasonably foreseeable human error during operation must not lead to hazardous
situations;
- The machinery must not start unexpectedly;
- The parameters of the machinery must not change in an uncontrolled way, where such
change may lead to hazardous situations;
- The machinery must not be prevented from stopping if the stop command has already
been given;
- No moving part of the machinery or piece held by the machinery must fall or be ejected;
- Automatic or manual stopping of the moving parts, whatever they may be, must be
unimpeded;
- The safety related parts of the control system must apply in a coherent way to the
whole of an assembly of machinery and/or partly completed machinery;
The above mentioned rules can be seen as very important and they must be accomplished
by the system behaviour and must be guaranteed by the controller program. However, the
ways that designers use to achieve these goals can change. For instance, it can depend on the
complexity of the system: if the system is more complex, then the implementation of the
emergency stop requirements can be harder. Some indications can be done to the designers,
but the final decision depends always of his/her scientific and technical background. This
means that different solutions for application of the emergency stop requirements - can
lead to the same practical results. Also, some indications can be done - according to the
different ways of guaranteeing the emergency stop requirements such as if it is necessary,
or not, a specific SFC for modelling the behaviour of the system after the emergency stop
actuation. Sometimes a specific SFC for modelling the behaviour of the system is necessary
after the emergency stop actuation and sometimes it is not.
From this last point of view, the types of emergency stops are divided in two main groups:
- Without emergency sequence - the actuation of the emergency button stops the
system/automatism through the inhibition of the outputs and/or for stop the evolution
of SFC.
- With emergency sequence - the actuation of the emergency button starts a particular
predefined procedure.
As guarantee that the developed controller will always react according the expected
behaviour, it is only necessary to model the controller and the plant discretely. Indeed, our
system has a hybrid plant, but the behaviour properties that we intend to guarantee for our
system are only related to discrete behaviour.

Automation 502
6.1 Without emergency sequence
The emergency without emergency sequence can be performed in three alternative modes:
- Outputs inhibition;
- Evolution stop;
- Outputs inhibition and evolution stop.
In the case of outputs inhibition the actuation of emergency button does not stop by itself
the evolution of the SFC controller, but it inhibits the outputs associated with their steps, as
shown in Figure 15. The ON outputs (state 1) are turned OFF (state 0), as well as the
evolution of SFC usually being stopped by the non-fulfilment of the logical conditions
associated with SFC transitions.
This can be obtained through the insertion of inhibition functions in the interface with the
machine plant. In this case, after the occurrence of an emergency stop, the actuators
command should be particularly well studied in agreement with the type of expected
response.
For instance, for the cylinders directional valves:
- One stable state valve (single control with spring return), if a cylinder return for a given
position is demanded.
- Two stable state valve (double control), if a stop at the end of the cylinder movement is
demanded.
- Valve with three positions (double control and spring return), if a cylinder stop in the
actual position is demanded.

Fig. 15. Functional diagram of output inhibition
In the other hand, in the case of evolution stop the condition AU is present in all logical
conditions associated with SFC transitions(Fig.16-a). With the actuation of emergency
button AU, no logical conditions associated with SFC transitions can be validated and in this
way, the controller SFC cannot step forward. With the AU shutdown, a new cycle evolution
is allowed.
It is of importance to note that in this situation the outputs associated to the active steps stay
validated. This way, the started actions can be maintained, if dangerous situations are not to
occur.
Finally, also it is possible to use the described types of emergency stop simultaneously,
without emergency sequence, outputs inhibition and evolution stop (Fig. 16-b). This

situation is used more in practice, if a specific emergency sequence is not necessary. Seen
that has the advantage of allowing, after the emergency button shutdown, the pursuit of the
evolution of the system starting from the same position at which it was stopped.

a) b)
Fig. 16. a) Evolution stop, and b) evolution stop and outputs inhibition
6.2 With emergency sequence
This type of emergency implies the introduction of an emergency sequence. Through the
activation of the emergency button AU, an emergency sequence can be added to the normal
run SFC (Fig. 17).

Fig. 17. Introduction of an emergency sequence
7. Emergency stop adopted solution: case study
The implementation of total controller specification, based on GEMMA presented in Figure
8, can be realized using the following two main alternative methods:
- Multiple SFC developing one SFC for each task (implementation of horizontal or
vertical coordination);

Automation 504
- Single SFC developing one SFC for all tasks.
Although this last alternative is possible from a theoretical point of view, for most if not all
automation systems practice shows that if the application of emergency stop requirements is
done on a complex system, the first solution (multiple SFC) is better.
However, the specification of a specific behaviour for the emergency stop requirements
(with emergency sequence) and its linking with other specified behaviours for the system (it
doesnt matter if by single or multiple SFCs) is similar. For this reason, this section considers
one single SFC for specification of all the desired behaviours for the system. Although the
Single SFC method was used for the implementation of the emergency stop to allow a
better global visualization and understanding of the implementation of the total controller
specification that includes the emergency stop (GEMMA task D1), it must be highlighted
that step 100 of the SFC presented in Figure 5 corresponds to an emergency stop sequence.
The emergency stop adopted for the case study presented was obtained according the EN
418 and EN 60204-1 standards.
According to the behaviour of the case study, the emergency stop with emergency sequence
was selected. The considered requirements that need to be accomplished by the emergency
sequence are:
- Stop all of the movements;
- Stop the filling operation.
To obtain these procedures the selection of the type of the appropriate directional valves to
accomplish, simultaneously, the requirements of the emergency stop and of the plant
behaviour was crucial.
The directional valve specifications used were the type of control (single solenoid control
with spring return or double solenoid control) and number of ways/ports.
The first security requirement relates to the stop of the movements, obtained by stopping
the air compressed supply to the directional valves of the cylinders A, B, C, E, G and of
motor F. For that, as shown in Figure 6, the air supply will be centralized and controlled
through a directional valve 3/2 way normally closed with spring return (H).
The second security requirement relates to the stopping of the filling operation; this was
performed through the turn OFF of the filling directional valve 2/2 way normally closed
with spring return (D).
Figure 18 shows the total controller SFC specification based on GEMMA implementation
with the single SFC method.
All the SFC controller specifications, presented in Figure 18, were simulated on Automation
Studio software. The obtained results led to the conclusion that all the requirements defined
in the Emergency Stop Standards were met.
formalization and implemented on a Programmable Logic Controller (PLC) adopted as the
controller physical device. This part of the developed work is not detailed on this
publication.


Fig. 18. Total SFC controller specification with emergence sequence
8. Conclusions
A systematic way was presented using the adopted techniques for the implementation of
complex specifications of automation systems.
First the use of GEMMA and SFC for the structure and specification of all the systems
behaviour was explained, considering its stop states and functioning modes.
Further, the vertical coordination implementation of a complex total controller specification,
based on the GEMMA graphical chart, was also presented and discussed.
Also, the adopted techniques for the emergency stop behaviour specification of automation
systems were presented in a systematic way.
The ways to translate the GEMMA graphical chart to low level specification were also
presented and discussed.

Automation 506
The standards (EN418, EN60204-1) related to the stop emergency specifications were
considered and all the requirements were accomplished.
The obtained results, by simulation with Automation Studio software, showed that the
adopted approach is adequate.
9. References
ADEPA (1992). GEMMA - Guide d`tude des Modes de Marches et dÀrrts, Ed.2, In:
Agence Nationale pour le Developpement de la Production Automatise, France
Booch, G.; Jacobson, I. & Rumbaugh, J. (2000). OMG Unified Modeling Language
Specification., In: Object Management Group edition, Object Management Group
Brusamolino, M.; Reina, L. & Spalla, M.F. (1984). An example of microprocessor's
application in minicomputer systems: a copy volume design and implementation,
In: Microprocessing and Microprogramming. Vol.13, Issue 5, pp. 331- 339
EN 418 (1992). Safety of machinery. Emergency stop equipment, functional aspects.
Principles for design, In: European Standard
EN 60204-1 (1997). Safety of Machinery - Electrical Equipment of Machines - Part 1: General
Requirements-IEC 60204-1, In: European Standard
FAMIC (2003). Automation Studio Users Guide, Famic Technologies Inc, Canada.
Harel, D. (1987). Statecharts: a visual formalism for complex systems, In: Science of Computer
Programming, Vol.8, pp 231-274, North Holland, Netherlands
IEC (2002). IEC 60848 - Specification language GRAFCET for sequential function chart, Ed.2,
In: International Electrotechnical Commission
Koornneef, F. & Meulen, M.V.D. (2002). Safety, reliability and security of industrial
computer systems, In: Safety Science, Vol.40, Issue 9, pp. 715-717
Moon I. (1994). Modeling Programmable Logic Controllers for Logic Verification, In: IEEE
Control Systems Magazine, pp. 53-59
Murata, T. (1989). Petri Nets: Properties, Analysis and Applications, Proceedings of the IEEE,
Vol. 77, No. 4, pp. 541-580, 1989
Sohier, C. (1996). Pilotages des Cellules Adaptatives de Production: Apport des Systemes
Multi-Agents, PhD Thesis, cole Normale Suprieure de Cachan, Paris, France
26
Answering Causal Questions and
Developing Tool Support
Sodel Vazquez-Reyes and Perla Velasco-Elizondo
Autonomous University of Zacatecas, Centre for Mathematical Research
Mxico
1. Introduction
People explore the world by asking questions about what is seen and felt. Thus, Question
Answering is an attractive research area as a distinctive combination from a variety of
disciplines, including artificial intelligence, information retrieval, information extraction,
natural language processing and psychology. Psychological approaches focus more on
theoretical aspects, whereas artificial intelligence, information retrieval, information
extraction and natural language processing approaches investigate how practical Question
Answering systems can be engineered.
A simple query to a search engine will return hundreds of thousands of documents. This
raises the need for a new approach to allow more direct access to information. Ideally, a user
could ask any question (open domain) and instead of presenting a list of documents,
Question Answering technology could present a simple answer to the user.
In contrast to Information Retrieval systems, Question Answering systems involve the
extraction of answers to a question rather than the retrieval of relevant documents.
At present, there are textual Question Answering systems working with factual questions.
These systems can answer questions of the type: what, who, where, how many and when.
However, systems working with complex questions, such as how and why, are still under
research, tackling the lack of representations and algorithms for their modelling (Burger et
al., 2001; Maybury, 2003; Moldovan et al., 2003).
One kind of complex question is why (causal). One reason why causal questions have not
been successfully treated is due to their answers requiring more elaboration (explanations)
instead of short answers.
The idea for our research detecting answers to causal questions through automatic text
processing, was born, and the hypothesis is that, whilst why questions appear to require
more advanced methods of natural language processing and information extraction because
their answers involve opinions, judgments, interpretations or justifications, they can be
approached by methods that are intermediate between Information Retrieval and full
Natural Language Understanding. We advocate the use of methods based on information
retrieval (bag-of-words approaches) and on limited syntactic and/or lexical semantic

Automation 508
analysis as a first step towards tackling the problem of automatically detecting answers for
causal questions.
For this research, a why questions, is defined simply as interrogative sentences in which
the interrogative adverb why occurs in the initial position. For example: Why is cyclosporine
dangerous?
The topic of a why question is the proposition that is questioned and it is presupposed to
be true according to the document collection; otherwise, the why question cannot be
answered. The proposition of the why question shown previously is: Cyclosporine is
dangerous.
The topic of the why question that is questioned should be the event that needs an
explanation according to the questioner. Identifying the questions topic and matching it to
an item (event, state, or action) in the text is a prerequisite for finding the answer. The
response to the why question stated previously is:
Why is cyclosporine dangerous?
It can harm internal organs and even cause death.
Inappropriate answers to why questions are mainly due to misunderstanding the
questions themselves (Galambos & Black, 1985). For example, the answer to the following
question varies depending on how the question is understood.
Why did Sodel dine at the Mexican Restaurant?
If we understand that this question concerns Sodels motivation for eating, we could reply
that he ate there because he was hungry. If we understand that the question relates to
why Sodel chose that particular restaurant, we could answer that He had heard that it is a
good restaurant and he wanted to try it. If we understand that the question is about Sodel
going to a restaurant instead of eating at home, we could answer that Sodels wife is out of
town and he cant cook. Neither having the same topic nor being expressed by the same
sentence constitutes a criterion of identity for why questions. In other words, the same
sentence can express different why questions. There are contextual factors and
background knowledge for the description of its interpretation. These answers to why
questions may be subjectively true and each answer have boundary conditions for itself. The
accuracy of the answer is in the mind of the perceiver.
The ultimate goal in textual Question Answering systems should be to answer any type of
question; consequently, the research direction should move towards that goal, and a focus
on causal questions serves this agenda well.
2. Definition of problem
Many members of the information retrieval and natural language processing community
believe that Question Answering is an application in which sophisticated linguistic
techniques will truly shine, owing to it being directly related to the depth of the natural
language processing resources (Moldovan, et al., 2003).
However, Katz and Lin (2003) have shown that the key to the effective application of natural
language processing technology is to employ it selectively only when helpful, without

Answering Causal Questions and Developing Tool Support 509
abandoning the simpler techniques. They proved that syntactic relations enable a Question
Answering system successfully to handle two linguistic phenomena: semantic symmetry
and ambiguous modification. That is, the incorporation of syntactic information in Question
Answering has a positive impact.
Voorhees shows that the research in Question Answering systems has made substantial
advances (Voorhees, 2001, 2002, 2003, 2004), answering factual questions with a high degree
of accuracy. Several forms of definition questions are processed appropriately, and list
questions retrieve sequences of answers with good recall from large text collections,
although it is necessary to answer hard and complex questions too, such as procedural,
comparative, evaluative and causal questions.
On the other hand, (Moldovan, et al., 2003) showed that the main problem with Question
Answering systems is the lack of representations and algorithms for modelling complex
questions in order to derive as much information as possible, and for performing a well-
guided search through thousands of text documents.
Some of the complex questions types seem to need much semantic, world knowledge and
reasoning to be handled properly, e.g. for automatically resolving ambiguities or finding out
which measure or granularity a user would prefer. This is beyond the scope of the current
research and even those in the near future. However, in this research, we advocate the use of
methods based on information retrieval (bag-of-words approaches) and on limited syntactic
and/or lexical semantic analysis as a first step towards tackling the problem of causal
questions.
Our goal was not to implement a functional (fully-fledged) textual Question Answering
system but to investigate how methods based on information retrieval (bag-of-words
approaches) and on limited syntactic and/or lexical semantic analysis can contribute to the
real-world application of causal text and causal questions. This enables us to focus on the
key matter of how the answer is contained in the document collection.
3. Related work
In the literature on Question Answering, the system developed in the Southern Methodist
University and Language Computer Corporation (Harabagiu et al., 2000) has been
considered to have the most sophisticated linguistic techniques due to the depth of its
natural language processing resources. This system classifies questions by expected answer
type, but also includes successive feedback loops that attempt to make progressively larger
modifications to the original questions until they find an answer that can be justified as
abductive proof semantic transformations of questions and answers are translated into a
logical form for being analysed by a theorem prover.
The system of the Southern Methodist University and Language Computer Corporation first
parses the question and recognises the entities contained in it to create a question semantic
form. The semantic form of the question is used to determine the expected answer type by
finding the phrase that is most closely connected to other concepts in the question. The
system then retrieves paragraphs from the corpus, using boolean queries and terms drawn
from the original question, related concepts from WordNet, and an indication of the
expected answer type.

Automation 510
Paragraph retrieval is repeated using different term combinations until the query returns a
number of paragraphs in a pre-determined range. The retrieved paragraphs are parsed into
their semantic forms, and a unification procedure is run between the question semantic form
and each paragraph semantic form. If the unification fails for all paragraphs, a new set of
paragraphs is retrieved using synonyms and morphological derivations of the previous
query.
When the unification procedure succeeds, the semantic forms are translated into logical
form, and a logical proof in the form of an abductive backchaining from the answer to the
question is attempted. If the proof succeeds, the answer from the proof is returned as the
answer string. Otherwise, terms that are semantically related to important question concepts
are drawn from WordNet and a new set of paragraphs is retrieved.
While research in Question Answering mainly focussed on responding to factual questions,
definition questions and list questions using stochastic processes, a more recent trend in
Question Answering aims at responding to other types of question that are of great
importance in everyday life or in professional environments such as procedural, causal,
comparative or evaluative questions. These have not yet been studied in depth; they require
different types of methodologies and formalisms, particularly at the level of the linguistic
models, knowledge representation and reasoning procedures.
However, the ideal system does not exist yet although approaches to support that goal have
been created. We will demonstrate a small number of approaches to Question Answering
working with complex questions or advanced methods. The explanations are based on their
general ideas.
In order to answer why questions, the aim of an ideal system should be to address a form
of Question Answering that does not focus on finding facts, but rather on finding the
identification and organisation of opinions, to support information analysis of the following
types: (a) given a particular topic, find a range of opinions being expressed about it; (b) once
opinions have been found, cluster them and their sources in different ways, and (c) track
opinions over time.
Verberne et al. (2007) demonstrate an approach to answering why questions, based on the
idea that the topic of the why question and its answer are siblings in the rhetorical
structure of the document, determined according to Rhetorical Structure Theory (RST)
(Mann & Thompson, 1988), connected by a rhetorical relation that is relevant for why
questions discourse-based answer extraction. They implemented an algorithm that: (a)
indexes all text spans not from the source document but from a manually analysed
representation of it into RST relations that participate in a potentially RST relation relevant;
(b) matches the input question to each of the text spans in the index; and (c) retrieves the
sibling for each of the found spans as the answer. The result is a list of potential answers,
ranked using a probability model that is largely based on lexical overlap. For the purpose of
testing their implementation, they created a test collection consisting of seven texts from the
RST Treebank and 372 why questions elicited from native speakers who had read the
source documents. From this collection, they obtained a recall of 53%, with a mean
reciprocal rank of 0.662. On the basis of the manual analysis of the question-answer pairs,
they argued that the maximum recall that can be obtained for this data set, from the use of
RST relations as proposed, is 58.0%. They declare that, although there are no reference data

for the performance of automatic Question Answering working with why questions, they
considered a recall of 53% (and a maximum recall of 58%) to be mediocre at best.
Waldinger et al. (2004), Benamara and Saint-Dizier (2004), and McGuinness and Pinheiro da
Silva (2004) delve into knowledge-based Question Answering and support inferential
processes for verifying candidate answers and providing justifications. That is, systems that
target the problem of Question Answering over multiple resources have typically taken the
approach of first translating an input question into an intermediate logical representation,
and, in the realm of this intermediate representation, matching parts of the question to the
content supplied by various resources.
Light et al. (2004), provide an empirical analysis of a corpus of questions that enables the
authors to identify examples of reuse scenarios, in which future questions could be
answered better by using information previously available to the system (e.g., in the form of
previously submitted questions or answers already returned to the users). The authors
acknowledge that some of the proposed categories of reuse are very difficult to implement
in working system modules.
Schlaefer (2007) has used ontologies for extracting terms from questions and corpus
sentences and for enriching the terms with semantically similar concepts. In order to
improve the accuracy of Question Answering systems, semantic resources have been used.
Semantic parsing techniques are applied to transform questions into semantic structures and
to find phrases in the document collection that match these structures.
Vicedo and Ferrandez (2000) have demonstrated that their evaluation improvements when
pronominal references are solved for IR and Question Answering tasks. That is, they are
solving pronominal anaphora.
Mitkov (2004) describes that coreference resolution has proven to be helpful in Question
Answering, by establishing coreferences links between entities or events in the query and
those in the documents. The sentences in the searched documents are ranked according to
the coreference relationships.
Castagnola (2002) shows that for the purpose of improving the performance of Question
Answering, he resolves pronoun references via the use of syntactic analysis and high
precision heuristic rules.
Galitsky (2003) introduces the reasoning mechanism as the background of the suggested
approach to Question Answering, particularly, scenario-based reasoning about mental
attitudes. Default logic is used for correction of the semantic representations. He describes
the process of representing the meaning of an input query in the constructed formal
language.
Setzer et al. (2005) address the role that temporal closure plays in deriving complete and
consistent temporal annotations of a text. Firstly, they discuss the approaches to temporal
annotation that have been adopted in the literature, and then further motivate the need for a
closed temporal representation of a document. No deep inferencing, they argue, can be
performed over the events or times associated with a text without creating the hidden
relations that are inherent in it. They then address the problem of comparing the diverse

Automation 512
temporal annotations of the same text. This is far more difficult than comparing, for
example, two annotations of part-of-speech tagging or named entity extent tagging, due to
the derived annotations that are generated by closure, making any comparison of the
temporal relations in a document a difficult task. They demonstrate that two articles cannot
be compared without examining their full temporal content, which involves applying
temporal closure over the entire document, relative to the events and temporal expressions
in the text. Once this has been achieved, however, an inter-annotator scoring can be
performed for the two annotations.
Nyberg et al. (2004) aim to capture the requirements of advanced Question Answering and
its impact on system design and the requirements imposed on the system (e.g., time-
sensitive searches and the detection of obscure relations). The challenges that face the push
towards the development of Question Answering systems of increased complexity are
especially the challenges of practicability and scalability. Indeed, such issues become
important for any system that would actually attempt to perform planning in a broad
domain. Similarly, it may be very challenging to find common linguistic representations to
use across highly modular systems for encoding internal information, as the information
sources themselves can vary widely, from unstructured text at one end of the spectrum to
full-blown knowledge bases at the other end.
Planning structures explicate how a person does certain things, and how he or she normally
tries to achieve some goal. Plans cannot be built from the story itself but have to be taken
from some world knowledge module. Studying instructional texts seems to be very useful
for answering procedural questions how. Aouladomar (Aouladomar, 2005a, 2005b;
Aouladomar & Saint-Dizier, 2005a, 2005b) incorporated concepts from linguistics,
education, and psychology to characterise procedural questions and content to produce an
extensive grammar of the ways in which a procedural text may be organised, a framework
that appears to show much promise. Although her work is on French texts, the procedural
features she identified included general ones, e.g. the distinct morphology of verbs in
procedures. However, her research did not directly address the task of classifying texts as
either procedural or non-procedural.
Aouladomar mentioned that questions beginning with how should not be neglected, since,
according to recent usage data from a highly-trafficked web search engine, queries starting
with how alone is the most popular category of queries beginning with question words.
The approaches to Question Answering mentioned above are likely to become relatively
more language-dependent, as they require larger and more complex resources of various
kinds.
3.1 Challenges of Questions Answering
The ultimate goal in textual Question Answering systems is to answer any type of question.
If the information needs are very simple ones (e.g. factoid, definition or list), then the
answer can be simple word(s), phrase(s) or sentence(s). If the information needs are more
complex, then the answers may come from a deep documentary analysis, or from multiple
documents. Where candidates answer from different corpora, these could be merged or
possibly summarised.

Moreover, if we recognise that users can obtain valuable information through inference and
construction from new material, combined with what they know already, the scope for
Question Answering is far wider. This can be far more than simply deriving the kind of
exact answer that is required due to the rich knowledge or complex inference it requires.
Alternatively, candidate answers from different languages could be translated into the
users native language. For example, if we retrieve answers from Spanish, French and Italian
and translate them into English, we could compare the nature of the answers drawn from
different geographic and cultural contexts.
The research direction is moving gradually towards these goals and it is our hope that the
Question Answering research groups can collaborate in order to achieve these goals.
4. Approach
Most causal questions are of the form Why Q?, where Q is an observation or fact to answer
(which we have identified as an effect). If a why question is an effect, then we are
searching for its explanations (which we have identified as causes). So, we have called the
cause and its effect a causal relation.
The why question (effect) has an infinite number of different answers. Each answer
contains an explanation of a cause for the question. In particular, causes explain their effects.
For this reason, a cause tells us why its effect occurs.
The natural complexity of a question depends on how the question is understood, and the
accuracy of the answer is in the mind of the perceivers, depending mainly on their
knowledge level (contextual factors and background knowledge) for the description of its
interpretation. Some users prefer a more accurate explanation, while others look for
explanations with a broader perspective and better explanatory resources. Any answer that
appeals to a cause is taken to be highly relevant and, therefore, to provide an explanation of
the effect a why question.
In order to get answers to a why question, we should try to detect causal relations.
Although textual Question Answering systems are evolving towards providing exact
answers only, for why questions the answers should be surrounded by some context, with
the purpose of supporting the answer.
4.1 Methodology
We have used the lexico-syntactic classification for why questions proposed by Verberne
et al. (2007). The categories used are existential there questions, process questions,
questions with a declarative layer, action questions and have questions. The result of
question analysis task is not used in the answer candidate extraction task. However, it gives
a category to each why question.
The answer candidate extraction task provides an approach to tackling a subset of causal
questions. We used the following procedure for detecting possible answers to why questions:
Identify the topic of the question.
In the list of sentences of source document, identify the clause(s) that express(es) the same
proposition as the question topic.

Automation 514
Select the best three clauses as answers.
Detect cause-effect information expressed in the answers selected.
Step 1. The topic of the question (which we have identified as an effect) is the observation or
fact that is questioned. In other words, it is the premise of question.
Step 2 and 3. We suggest that decomposition of the complex task of recognizing which
source text expresses the same proposition as the question topic would make a step
towards better understanding the process for answering causal questions. This should
involve making use of set of measures (see 4.1.1), and using each one as a weighting
factor within the whole evaluation for ranking of possible answers. The sum of factors
is the final value. To be precise, each measure is applied to the words belonging to the
question-text pair. The best three answers are selected.
Step 4. We used a rule-based approach to identify and extract cause-effect information
expressed in the answers selected.
4.1.1 Matching formulae
The answer extraction process relies on the computation of four measures:
1. Simple matching. The stop words are not removed; for this reason, non stop words are
weighted with 1.9 and stop words with 0.1. The final weight is calculated as the sum of
all values and normalized dividing it by the length of the question and text (total number
of words). In which, Q is a question and T is a text with possible answer, see (1).
1 2 2 *
jcm
dist IC N IC N IC LCS (1)
However, if simple matching is not possible, we are working with stems. All the occurrences
in the questions stems set that also appear in the texts stems set will increase the
accumulated weight in a factor of one unit. The stems are weighted with 1.9 for non stop
words and 0.1 for stop words.
Longest consecutive subsequence. This process measures the surface structure overlap between
the text with possible answer and the question (only consecutive words). In order to
compute this overlap we extract the longest consecutive subsequence (LoCoSu) between the
question and the text with possible answer, LoCoSu(Q, T), see (2).
_ , structure Overlap LoCoSu Q T (2)
For example, if we have Q= {AA, BB, CC, FF } and T= {AA, BB, DD, FF}
then LoCoSu = {AA, BB} = 2.
In order to calculate LoCoSu, we have used a third party implementation, the longest
common substring tool (Dao, 2005).
This feature indicates the presence of the same word with 1, or otherwise zero. We are
removing stop words. We are using stems if simple matching is not possible.
One should note that this measure assigns the same relevance to all consecutive
subsequences with the same length. Furthermore, the longer the subsequence is, the more
relevant it will be considered.

We have used a threshold of 2, that is, LoCoSu (Q, T) bigger or equal to 2.
Sorensens similarity coefficient. This only considers non stop words. The value is increased by
one per word in the intersection or union. Sorensens similarity coefficient is a distance
measure, see (3).
Sorensen = 1D (3)
Where D is Dice's coefficient, which is always in [0,1] range, see (4).

2 Q T
D
Q T
(4)
Where Q is a set of words of question and T is a set of words of sentence, possible answer.
Note that if simple matching is not possible, we use their stems for evaluation.
WordNet-based Lexical Semantic Relatedness. The measure uses WordNet (Miller, 1995) as its
central resource. Here, we are in fact considering similarities between concepts (or word
senses) rather than words, since a word may have more than one sense. Measures of
similarity are based on information in is-a hierarchy. WordNet only contains is-a hierarchies
for verbs and nouns, so similarities can only be found where both words are in one of these
categories. WordNet includes adjectives and adverbs but these are not organised into is-a
hierarchies, so similarity measures cannot be applied.
Concepts can, however, be related in many ways apart from being similar to each other.
These include part-of relationships, as well as opposites and so on. Measures of relatedness
make use of this additional, non-hierarchal information in WordNet, including the gloss of
the synset. As such, they can be applied to a wider range of concept pairs including words
that are from different parts of speech.
If we want to compute lexical semantic relatedness between pairs of lexical items using
WordNet, we can find that several measures have been reported in the literature. According
to the evaluation of Budanitsky and Hirst (2006), the measure proposed by Jiang and
Conrath (1997) is the most effective. The same measure is found the best in word sense
disambiguation (Patwardhan, Banerjee, & Pedersen, 2003). We confirm that the Jiang and
Conrath measure was the best for the task of patter_induction for information extraction
(Stevenson & Greenwood, 2005).
The Jiang and Conrath metric (jcm) uses the information content (IC) of the least
common subsumer (LCS) of the two concepts. The idea is that the amount of information
two concepts share will indicate the degree of similarity of the concepts, and the amount
of information the two concepts share is indicated by the IC of their LCS. Thus, they take
the sum of the IC of the individual concepts and subtract from that the IC of their LCS,
see (5).
1 2 2 *
jcm
dist IC N IC N IC LCS (5)
Where N1 is the number of nodes on the path from the LCS to concept 1 and N2 is the
number of nodes on the path from the LCS to concept 2.

Automation 516
Since this is a distance measure, concepts that are more similar have a lower score than the
less similar ones. The result with the smallest distance is taken to disambiguate the senses
between two words. In order to maintain consistency among the measures, they convert this
measure to semantic similarity by taking its inverse, see (6).

1
1 2 2 *
jcm
sim
IC N IC N IC LCS

(6)
In order to calculate jcm similarity, we have used a third party implementation, the
WordNet Relatedness tool (Pedersen, Patwardhan, & Michelizzi, 2004).
We use three types of words for calculating this feature in order to discover lexical semantic
relatedness:
a. jcm similarity between question-word against text-word.
b. jcm similarity between question-word against synonyms of text-word. The idea is to
show that if synonyms are different words with identical or at least similar meanings,
then we can use them to calculate jcm similarity in order to provide extra resources for
disambiguation process.
c. jcm similarity between question-word and the synonyms of text-word antonyms. We
think that a word pair where the individual words are opposite in meaning could help
in the disambiguation process, identifying cause-effect (text-question) described with
opposite words. In other words, an additional exploration of potential associative
relations for text-question pairs.
4.2 Implementation
The system architecture is depicted in Fig. 1. As can be seen, it has a base client-server
architecture (Shaw & Garlan, 1996) hosting several components that support different
duties. There are three main actors in the environment surrounding the system: (i) the User
which is the person who issues the questions to be answered by the system, (ii) the
Administrator which is the person whose main duty is to update the systems databases
and (iii) CAFETIERE which is an external tool used to support the query/document
processing work. All the communication among the main architectural parts of the system,
is carried out in a synchronous request-response mode, i.e. a source part submits a request
and waits until the response is returned from the target part. All the components of the
architecture are written in Java.
As depicted in Fig. 1, the client-side of the architecture hosts two user interface (UI)
components: the User UI and Administration UI. These UI components enable the
communication of the User and Administrator with the system.
A User issues a question to the system via the User UI component. This question is passed
up to the server as a user request. In the server side, all user requests are processed by the
Query Processing component. Internally, this component has a Pipe-and-Filter like
architecture (Shaw & Garlan, 1996), which allows splitting the query-processing job into a
series of well-defined low-coupled sequential steps. Three filter components constitute the
Query Processing component: Question Analysis, Answer Candidate Extraction and Cause-
Effect Detection. The Question Analysis filter enables the classification of the issued

question with regard to a category that corresponds to a syntactic pattern. The Answer
Candidate Extraction filter automatically maps the question onto the sentences of document,
mainly by measuring lexical overlapping and lexical semantic relatedness between the
question topic and the sentences of document to detect possible answers for the question
evaluated. Finally the Cause-Effect Detection filter uses a rule-based approach in order to
identify the cause and effect information expressed in the selected answers. Both the
Question Analysis and Cause-Effect Detection filters interact with the CAFETIER tool (Black
et al., 2003). Specifically, they use CAFETIERs lexico-syntactic analysis pipeline.

Fig. 1. The architecture of the system.
The Administration UI component is the means by which the Administrator maintains the
systems databases. As shown in Fig. 1, there are three databases: (i) Newspaper Documents
which contains a list of candidate text passages (possible answers) that, in all likelihood,
match the original question, (ii) Rules which contains a set of lexico-syntactic and basic
semantic rules for the English language that are used to produce phrasal and conceptual
annotations as well as representations of elements of interest, events and relations and (iii)
Ontologies which are lists of known names of places, people, organizations, artifacts, etc.;
that help to assign conceptual classes to single and/or multi word phrases as additional
information for the information extraction analysis.
In the following sections we will focus on describing the elements of the architecture
supporting Query Processing.

Automation 518
4.2.1 Question Analysis
As mentioned before, we use CAFETIEREs lexico-syntactic analysis pipeline for the
classification of why questions. This pipeline is constituted by: sentence splitter, tokenizer,
orthography tagger, stemmer, POS tagger, gazetteer lookup (single and multi-word names
and terms), and rule-based analyzer (context sensitive rule-based analysis).
The Question Analysis filter that implements the required logic to interact with CAFETIERE
in the following way:
1. Modifying the resources for an already existing analysis engine. For example, we have done
the following:
- added words to the lexicon used by the part-of-speech tagger.
- added rules to the file used by the part-of-speech tagger.
- added patterns to the file used by the part-of-speech tagger.
The previous three points have been executed for improving the result of part-of-speech
tagger to our research.
2. Creating a new instance of an existing analysis engine type, with its own set of resources. This
does not involve changing any of the code of the analysis engine, only declaring what
specific resource instance(s) it will use. We have for example:
- created one lookup analysis engine for using our six gazetteers: list of cue words, modal
verbs, auxiliaries, process verbs, declarative verbs and agentive nouns. The objective
was to support the phrase-level analysis of our research.
- created three different rule-based analysis engines, each with its own set of rules for
performing higher level analysis up to the clause and sentence level. They have been
divided into three levels: tags, phrase level and clause level.
3. Creating a new aggregate analysis engines to run modules in different sequences, or to run
different permutations of modules that are used in the question analysis. Basically, the analysis
engines integrating the aggregate are: sentence splitter, tokenizer, orthography tagger,
stemmer, POS tagger, gazetteer lookup, rule-based analyzer and concept collector.
This question analysis process relies on CAFETIERE tool in order to assign a question
category that corresponds to the types of entities, which constitute the category each
question category corresponds to a syntactic pattern. The question focus is its premise.
The process is a rule-based approach, which uses hand-crafted rules that look for lexical and
syntactic clues in the question. We have sets of rules for tags, phrases, and clauses (which
include the question types). The rules use six gazetteers as knowledge source: cue words,
modal verbs and auxiliaries, process verbs, declarative verbs and agentive nouns. We have 143
rules, which are constituted by 32 rules for tags, 76 rules for phrases and 35 rules for clauses.
The output generated for this filter, which is the annotated and original question, are passed
up to the answer candidate extraction filter.
4.2.2 Answer candidate extraction
We implemented an answer candidate extraction from text to identify the three best answers
to the question. In order to do that, this component implements the logic to perform the

computation of four matching formulae: simple matching, longest consecutive subsequence,
Sorensens similarity coefficient and WordNet-based lexical semantic relatedness. This
process matches the question against the sentences in the text to identify the clauses that
express the same proposition as the question.
Within the entire set of measures, each one of them is considered as a feature with the same
weight. For selecting the best three possible answers for each question, we have combined
the four metrics, see (7).
_
4
sm lcs ssc lsr
f f f f
matching formulae

(7)
The extraction of possible answers is accomplished by selecting the top three possible
answers for each question. The answers are saved into a file, which is passed up to the
Cause-Effect Detection filter to do the corresponding processing.
4.2.3 Cause-Effect Detection
The detection of cause-effect information expressed in the identified best three possible
answers is done by the Cause-Effect Detection filter. As depicted in Fig. 1, this filter contains
the required logic to use the CAFETIERE tool to support this job.
We have investigated how cause-effect information could be extracted from newspaper text
using rule-based approach without full parsing of sentences. A set of rules that usually
indicate the presence of a causal relationship was constructed and used for the extraction of
cause-effect information.
No inferencing from common sense knowledge or domain knowledge was used.
Knowledge-based inferencing of causal relationships requires a detailed knowledge of the
domain, and newspaper text covers a very wide range of topics. Only linguistic clues were
used to identify causal relationships. For example, we are using explicit linguistic
indications of cause and effect, such as because, however, due to, so, therefore, but, as a result of
this, and so on, instead of inferencing from common sense knowledge or domain
knowledge.
We identified the following two ways of explicitly expressing cause-effect:
1. Using causal links to link two phrases, clauses or sentences.
2. Using causative verbs.
The detection of cause-effect information (rule-based approach) expressed in the identified
answers was implemented in the same way that our question analysis (see Section 4.2.1), (a)
modifying the resources for an already existing analysis engine; (b) creating a new instance
of an existing analysis engine type, with its own set of resources; and (c) creating a new
aggregate analysis engine.
Rules were created to identify sentences containing causal links and causative verbs. The cause-
effect information was then extracted. The implementation can identify causal relations in
newspaper text when it focuses on the causal relations that are explicitly indicated in the
text using linguistic means.

Automation 520
A complete example is the question, Why did researches compare changes in oxygen
concentration? which belongs to the document wsj_0683. It is an action question. The best
three possible answers detected by our answer candidate extraction task are:
1. Researchers at Ohio State University and Lanzhou Institute of Glaciology and
Geocryology in China have analyzed samples of glacial ice in Tibet and say
temperatures there have been significantly higher on average over the past half-century
than in any similar period in the past 10,000 years.
2. According to greenhouse theories, increased carbon dioxide emissions, largely caused
by burning of fossil fuels, will cause the Earth to warm up because carbon dioxide
prevents heat from escaping into space.
3. To compare temperatures over the past 10,000 years, researchers analyzed the changes
in concentrations of two forms of oxygen.
The correct answer is the third one. The verb phrase to compare temperatures over the past
10,000 years is the cause, and the clause researchers analyzed the changes in concentrations of two
forms of oxygen is the effect.
5. Evaluation
We have shown previously that a why question has more than one answer since there
does not exist only one correct way of explaining things; therefore, it is quite difficult to
determine whether a string of text provides the correct answer. Human assessors have
legitimate differences of opinions in determining whether a response actually answers a
question. If human assessors have different opinions, then eventual end-users of the
Question Answering technology have different opinions as well because some users prefer a
more accurate explanation, while others look for explanations having a broader perspective
and better explanatory resources.
We can highlight that the time and effort required to manually evaluate a Question
Answering application is considerable, owing to the need for human judgment. This issue is
compounded by the fact that there is no such thing as a canonical answer form. Assessors
decision on the correctness of an answer makes resulting scores comparative, not absolute.
For our evaluation, we used the collection of Verberne et al. (2007), It has a relative
preponderance of questions (typically expressed in the past tense) about specific actions of
their motivations or relations to them, because their source documents are Wall Street
Journal articles (news) and they describe a series of events that are specific to the topic, place
and time of the text. The collection consists of seven texts from the RST Treebank of 350-550
words each and 372 why questions.
Throughout the evaluation of the answers detected, we adopt a manual approach whereby
an assessor determines if a response is suitable for a question, with two possible outcomes.
The response is either correct (i.e. the answer string must contain exactly the information
required by the question), or incorrect (i.e. it is a wrong answer or no answer at all). To
evaluate a question, an assessor was required to judge each answer string in that question's
answer pool (set of three answers).

The kind of evaluation executed for our research is a post-hoc evaluation on analyzed data
question-answer pairs. It is divided in three sections: question classification, which is a
lexico-syntactic classification where each question category corresponds to a pattern, answer
candidate extraction, which maps the question onto the correct source text in order to detect
possible answers for the question evaluated, and the detection of cause-effect information
expressed in the answers selected.
5.1 Question classification
We classified the 372 why questions of the collection using the question analysis filter
(rule-based approach). The lexico-syntactic classification is constituted of 5 categories:
existential there questions, process questions, questions with a declarative layer, action questions
and have questions.
The following questions are examples that were classified with the existential there
category:
- Why is there resistance to the Classroom Channel?
- Why is there a reference to the musical "The Music Man"?
- Why is there controversy in this Dallas suburb?
Examples of questions in the process question category are:
- Why did Cincinnati Public Schools reject the subscription offer?
- Why did the US Coast Guard close part of the Houston Ship Channel?
- Why did the petroleum plant explode?
The following questions are examples of declarative layer category:
- Why does Whittle think he can reach subscription goals within one year?
- Why did the report say that advertisers were showing interest?
- Why does Mr Hogan think the company will be successful?
Examples of questions in the action question category are:
- Why do researches conclude that the earth is warming?
- Why did Dr. Starzl advise against buying Fujisawa stock?
- Why were town officials embarrassed?
The following questions are examples that were classified with the have question category:
- Why did firefighters have difficulties getting the fire under control?
- Why does Whittle have reason for concern?
- Why did the research team have no financial stake in the drug?
For some categories, the question analysis filter only needs fairly simple cues for choosing a
category. For example, the presence of the word there with the syntactic category EX leads to
an the category existential there question.
For deciding on questions with a declarative layer, action questions and process questions,
complementary lexical-syntactic information is needed. In order to decide whether the
question contains a declarative layer, the filter checks whether the main verb is in

Automation 522
declarative verbs list, and whether it has a subordinate clause. The distinction between
action and process questions is made by looking up the main verb in a list of process verbs.
This list contains the 529 verbs from Levin verb index (1993). If the main verb is not
determined to be process, declarative or have, it is assigned to the action verb category.
Questions with a declarative layer need further analysis because they are ambiguous. For
example the question Why did they say that migration occurs? can be interpreted in two
ways: Why did they say it? or Why does migration occur?. Our answer candidate
extraction filter should try to find out which of these two questions is supposed to be
answered. In other words, the filter should decide which of the clauses contains the question
focus. For this reason, questions with a declarative layer are most difficult to answer.
Table 1 shows the results of our question classification. We observe that the five categories
of classification (existential there questions, process questions, questions with a
declarative layer, action questions and have questions) had a performance of 54.56%. In
other words, for the 372 questions of collection, 203 questions were classified correctly. We
want to highlight that question classification uses only questions, without answers.

Category
Question
per
Category
Questions
Classified
Correctly
Performance
Existential there
questions
11 5 45.45%
Process questions 145 102 70.34%
Questions with a
declarative layer
93 17 18.27%
Action questions 102 65 63.72%
Have questions 21 14 66.66%

TOTAL 372 203 54.56%
Table 1. Lexico-syntactic classification for the 372 why questions of collection.
We have assigned correctly the existential there category to 45.45% of the questions;
70.34% were labelled as process questions; 18.27% of the questions had a declarative layer;
the category of action questions was assigned to 63.72% of the questions because if the main
verb of the questions is not a process, declarative or a have verb, then we are assumed that
its type is action. And 66.66% were labelled, as have questions.
We observe that question with a declarative layer are most difficult to identify because of
clausal object, that is, a subordinate clause must be detected after declarative verb.
The general rules (only a lexico-syntactic analysis) for categories of classification are:
- Existential there category:
WRB + VP + EX + NP + [ADVP|PP|NP] + ?

- Process question category:
WRB + VP + + NP + Process-VERB + [ADVP|PP|NP] + ?
- Declarative layer category:
WRB + VP + NP + Declarative-VERB + S + ?
- Action question category:
WRB + VP + NP + Action-VERB + [NP] + [ADVP|PP|NP] + ?
- Have question category:
WRB + VP + NP + Have-VERB + NP + [ADVP|PP|NP] + ?
If the lexico-syntactic pattern of the question did not correspond to any category, then it was
assigned to the default category. However, we detected that questions can also be classified
in the default category due to (a) errors of part-of-speech tagger; and (b) errors of rule-based
analyzer. The default category contained 45.43% of the questions. A few examples of
questions assigned to the default category are:
- Why is the sago expensive?
- Why is the Sago a pricey lawn decoration?
- Why is rowdy behavior unlikely at the Grand Kempinski?
5.2 Answer candidate extraction
In order to evaluate the answer candidate extraction filter, most previous Question
Answering work has been evaluated using traditional metrics as recall, and mean reciprocal
rank (Voorhees, 2003, 2004). We follow the standard definition of recall, see (8).

c
recall
t
(8)
Where c is the number of correct annotations produced, and t is the total number of
annotations that should have been produced.
Using this formula, the recall obtained by our lexical overlapping and lexical semantic
relatedness approach is 36.02%. In our test corpus, t=372, the number of why questions in
the collection, and c=134, the number that were answered correctly by those techniques.
We hypothesized some of the why questions could have been unanswered because the
collections questions were created by native speakers who might have been tempted to
formulate why questions that did not address the type of argumentation that one would
expect of questions posed by persons who needed a practical answer to a natural why
questions.
For this reason, working from the premise that our lexical overlapping and lexical semantic
relatedness approach can only answer why questions with explicit and ambiguous
causation because it uses basic external knowledge for disambiguation, we recalculated
recall for the 218 why questions with explicit and ambiguous causation. The rate of our
recall increased to 61.46% of the former. The rate of recall thus increases considerably and

Automation 524
we reach similar results as the RST method (Verberne, et al., 2007), which relies on texts
where all causal relations have been pre-analysed.
The second metric used was mean reciprocal rank (MRR), see (9). The original evaluation
metric used in the Question Answering tracks TREC 8 and 9 (Voorhees, 2000) was mean
reciprocal rank (MRR), which provides a method for scoring systems which return multiple
competing answers per question. We used MRR because our implementation returns a list
with the best 3 possible answers that have been found to each question.

1
1
Q
i i
r
MRR
Q
(9)
Where Q is the question collection and r
i
the rank of the first correct answer to question i or
0 if no correct answer is returned.
We showed that our answer candidate extraction filter found answers for 134 why
questions on undifferentiated texts. The distribution for 134 why questions is 48 correct
answers located in the first position, 29 correct answers located in the second position and
57 correct answers located in the third position. Consequently, the MRR is 0.219
Working from the previously mentioned premise that our lexical overlapping and lexical
semantic relatedness approach can only answer why questions with explicit and
ambiguous causation because it uses basic external knowledge for disambiguation, then for
the 218 why questions with explicit and ambiguous causation, the rate of our MRR
increases to 0.373 of the former.
5.3 Cause-effect information
The cause-effect information in the answers selected is mainly expressed by causal links.
The reason for this could be the fact that the events discussed in newspaper texts use
connectives between two adjacent clauses. We detected the following causal links: before,
after, where, due to, because of, but, about, because, for, and from. Three examples are presented:
Why are the Mayor and two members of the Council worried?
Mayor Lynn Spruill and two members of the council said they were worried about
setting a precedent that would permit pool halls along Addison's main street.
Why would the interior regions of Asia heat up first?
Some climate models project that interior regions of Asia would be among the first to
heat up in a global warming because they are far from oceans, which moderate
temperature changes.
Why could the number of people known injured increase?
Nearby Pasadena, Texas, police reported that 104 people had been taken to area
hospitals, but a spokeswoman said that toll could rise.
The 89.01 % of question-answer pairs of the collection contain causal links.

In order to evaluate cause-effect information, we manually identified 50 question-answer
pairs in the collection. After that, we used the answer candidate extraction filter for
evaluating the cause-effect information detected, using the same 50 question-answer pairs
manually identified. Table 2 shows we detect 34% of cause-effect information.

Total questions-
answer pairs
% of questions-
answer pairs
Question-Answer pairs analysed
manually
50 100%
Questions for which we identified
a text (possible answer)
39 78%
Questions for which the identified
text is a correct answer
17 34%
Table 2. Outcome of cause-effect evaluation.
Questions, which contain modals, constitute 7.52% of the total of collection. The function of
modals is important in defining the semantic class of question. We cannot solve this issue
because our answer candidate extraction filter works with lexico-syntactic level. For
example:
1. Why did Nandos not use actors to represent chefs in funny situations?
2. Why can Nandos not use actors to represent chefs in funny situations?
Answer to Question 1 is a motivation, and answer to question 2 is a cause.
6. Conclusion
We introduced an approach which draws on linguistic structures, enabling the classification
of why questions and the retrieval of answers for why questions from a newspaper
collection. The steps to summarize our approach are:
1. Assign one category to each why question, using lexico-syntactic analysis. Each
question corresponds to a syntactic pattern (rule-based approach).
The lexico-syntactic classification is constituted of 5 categories: existential there
questions, process questions, questions with a declarative layer, action questions and have
questions.
2. Detecting three possible answers to each why questions:
2.1. Identify the topic of the question (effect).
2.2. In the list of sentences of source document, identify the clause(s) that express(es)
the same proposition as the question topic (making use of a set of measures).
2.3. Select the best three clauses as answers.
2.4. Detect cause-effect information expressed in the answers selected (rule-based
approach).
The output for each question is a question category and three possible answers.

Automation 526
6.1 Contributions
We have hypothesised that these methods will also work for why questions, and have
attempted to discover to what extent methods based on information retrieval (bag of
words approaches), and on limited syntactic and/or lexical semantic analysis can find
answers to why questions. So, our research contributes new knowledge to the area of
automatic text processing by the following:
We have developed an analysis component for feature extraction and classification
from questions (a rule-based approach as a first step towards tackling the problem of
question analysis of why questions using an Information Extraction Analyzer.
An original answer candidate extraction filter has been developed that uses an approach
that combines lexical overlapping and lexical semantic relatedness (lexico-syntactic
approach) to rank possible answers to causal questions. On undifferentiated texts, we
obtained an overall recall of 36.02% with a mean reciprocal rank of 0.219, indicating that
simple matching is adequate for answering over one-third of why questions. We
analyzed those question-answer pairs where the answer was explicit, ambiguous and
implicit, and found that if we can separate the latter category, the rate of recall increases
considerably. When texts that contain explicit or ambiguous indications of causal
relations are distinguished from those in which the causal relation is implicit, recall can
be calculated as 61.46% of the former, with a mean reciprocal rank of 0.373, which is
comparable to results reported for texts where all causal relations have been pre-
analyzed. This plausible result shows the viability of our research for automatically
answering causal questions with explicit and sometimes ambiguous causation.
We have found that people have conflicting opinions as to what constitutes an
acceptable response to a why question. Our analysis suggests that there should be a
proportion of text in which the reasoning or explanation that constitutes an answer to
the why question is present, or capable of being extracted from the source text.
Consequently, the complexity of a why question depends on the knowledge level of
users. While some users prefer a more accurate explanation, others look for
explanations with a broader perspective and better explanatory resources. Any answer
that appeals to a cause is taken to be highly relevant and, therefore, to provide an
explanation of the effect a why question. In order to provide a context with which to
support the answer, the paragraph from which the answer was extracted should be
returned as the answer.
We conclude that this research offers a greater understanding of why. It provides an
approach to tackling a subset of why questions (with explicit and sometimes ambiguous
causation) which combines lexical overlapping and lexical semantic relatedness. It further
considers the detection of cause-effect information that is explicitly indicated in the text
using causal links and causative verbs. For these reasons, this research contributes to a
better understanding of automatic text processing for detecting answers to why questions
and to the development of future applications for answering causal questions.
6.2 Further work
To improvements the answer candidate extraction, we could experiment with ambiguous
and implicit causation since our lexico-syntactic approach has not been successful for these

types of causation. In order to generate correct answers, we would need to go beyond the
co-occurrence of terms and lexical semantic relatedness due to the mismatch between the
expressions used in the question and the expressions used in the source text.
We should contribute to the implementation and evaluation of fundamental techniques
representing knowledge and reasoning.
When we use a causal relation to describe the interaction between two sentences, it would
be more interesting and informative if we could present an answer that offers chain of
explanations connecting the two events, that is, the entire answer to the why question
should be a chain of explanation of its causal relation.
When considering the relevance of answers to causal questions, we should involve carrying
out inferences (we view inference as the process of making implicit information explicit) to
arrive at the required answer. A relevant answer requires the provision of an appropriate
explanation, according to the questioner. The explanation should increase the questioners
existing knowledge rather than duplicate it. A filter for explanation generation that takes
into account the descriptions already presupposed by the question could be implemented by
using descriptions to generate explanations, which are themselves answers to the why
question. The algorithm could begin to make use of the information explicitly encoded by
the lexical and syntactic analysis.
Three questions need to be considered in order to advance our reasoning about the
descriptions embedded in why question, (1) Where can we find the descriptions
presupposed by the question?, (2) How can we recognise the descriptions presupposed by
the question?, and (3) How can we represent and compute the descriptions presupposed by
the question?
In order to understand this process in more detail, an analysis of epistemology (the theory of
knowledge) would be necessary. This is the branch of philosophy concerned whit the
nature, origin, and scope of knowledge. It addresses the question "how do you know what
you know?"
7. References
Aouladomar, F. (2005a). A Preliminary Analysis of the Discursive and Rhetorical Structure
of Procedural Texts, Proceedings of the Symposium on the Exploration and Modelling of
Meaning, pp. 21-24, Biarritz, France, November 14-15, 2005.
Aouladomar, F. (2005b). Towards Answering Procedural Questions, Proceedings of the
Workshop on Knowledge and Reasoning for Answering Questions, pp. 21-31, Edinburgh,
Scotland, July 30, 2005.
Aouladomar, F., & Saint-Dizier, P. (2005a). An Exploration of the Diversity of Natural
Argumentation in Instructional Texts, Proceedings of the 5th International Workshop on
Computational Models of Natural Argument, pp. 12-15, Edinburgh, Scotland, July 30,
2005.
Aouladomar, F., & Saint-Dizier, P. (2005b). Towards Generating Procedural Texts: an
exploration of their rhetorical and argumentative structure, Proceedings of the 10th
European Workshop on Natural Language Generation, pp. 156-161, Aberdeen, Scotland,
August 8-10, 2005.

Automation 528
Benamara, F., & Saint-Dizier, P. (2004). Advanced Relaxation for Cooperative Question
Answering. In: New Directions in Question Answering, M. Maybury, (Ed.), 263-274,
ISBN 0-262-63304-3, Cambridge, Massachusetts USA.
Black, W., McNaught, J., Vasilakopoulos, A., Zervanou, K., Theodoulidis, B., & Rinaldi, F.
(2003). CAFETIERE: Conceptual Annotations for Facts, Events, Terms, Individual
Entities, and Relations, Parmenides Technical Report No. TR-U4.3.1, In: The
University of Manchester.
Budanitsky, A., & Hirst, G. (2006). Evaluating WordNet-based Measures of Lexical Semantic
Relatedness. Computational Linguistics, Vol. 32, No. 1, (March 2006), pp. 13-47, ISSN
0891-2017.
Burger, J., Cardie, C., Chaudhri, V., Gaizauskas, R., Harabagiu, S., Israel, D., et al. (2001).
Issues, Tasks and Program Structures to Roadmap Research in Question & Answering,
Technical Report, In: NIST, Available from
http://www-nlpir.nist.gov/projects/duc/roadmapping.html.
Castagnola, L. (2002). Anaphora Resolution for Question Answering, Master Degree Thesis,
Department of Electrical Engineering and Computer Science, Massachusetts
Institute of Technology, USA, June, 2002.
Dao, T. (2005). The Longest Common Substring with Maximal Consecutive, In: The Code
Project, Retrieved on 30/09/2011, from
http://www.codeproject.com/KB/recipes/lcs.aspx?display=Print
Galambos, J. A., & Black, J. B. (1985). Using Knowledge of Activities to Understand and
Answer Questions. In: The psychology of questions, A. C. Graesser & J. B. Black (Eds.),
148-169, ISBN 0898594448, Lawrence Erlbaum Associates.
Galitsky, B. (2003). Natural Language Question Answering system, Advanced Knowledge
International Pty Ltd, ISBN 0868039799, London, UK.
Harabagiu, S., Moldovan, D., Pasca, M., Mihalcea, R., Surdeanu, M., Bunescu, R., Girju, R.,
Rus, V. and Morarescu, P. (2000). FALCON: Boosting Knowledge for Answer
Engines. Proceedigns of the Ninth Text REtrieval Conference (TREC-9), pp. 479-488,
online publication, 06/10/11,
http://trec.nist.gov/pubs/trec9/t9_proceedings.html, Gaithersburg, Maryland,
USA, November 13-16, 2000.
Jiang, J., & Conrath, D. (1997). Semantic similarity based on corpus statistics and lexical
taxonomy. Proceedings of the International Conference on Research in Computational
Linguistics, pp. 19-33, Taipei, Taiwan, August, 1997.
Katz, B., & Lin, J. (2003). Selectively Using Relations to Improve Precision in Question
Answering, Proceedings of EACL-2003 Workshop on Natural Language Processing for
Question Answering, pp. 43-50, Budapest, Hungary, April 12-17, 2003.
Levin, B. (1993). English Verb Classes and Alternations: a preliminary investigation, University of
Chicago Press, ISBN 0226475336, USA.
Light, M., Ittycheriah, A., Latto, A., & McCracken, N. (2004). Reuse in Question Answering:
A Preliminary Study. In: New Directions in Question Answering, M. Maybury, (Ed.),
162-182, ISBN 0-262-63304-3, Cambridge, Massachusetts USA.
Mann, W. C., & Thompson, S. A. (1988). Rhetorical Structure Theory: Toward a functional
theory of text organization. Text, Vol. 8, No. 3, (December 1998), pp. 243-281.

Maybury, M. T. (2003). Toward a Question Answering Roadmap. In: New Directions in
Question Answering, M. Maybury, (Ed.), 3-14, ISBN 0-262-63304-3, Cambridge,
Massachusetts USA.
McGuinness, D., & Pinheiro da Silva, P. (2004). Trusting Answers from Web applications. In:
New Directions in Question Answering, M. Maybury, (Ed.), 275-286, ISBN 0-262-
63304-3, Cambridge, Massachusetts USA.
Miller, G. A. (1995). WordNet: A Lexical Database for English. Communication of the ACM,
Vol. 38, No. 11, (November 1995), pp. 39-41, ISSN 0001-0782.
Mitkov, R. (2004). Anaphora Resolution. In: The Oxford Handbook of Computational Linguistics,
R. Mitkov (Ed.), 265-283, Oxford University Press, ISBN 0198238827, USA.
Moldovan, D., Pasca, M., Harabagiu, S., & Surdeanu, M. (2003). Performance Issues and
Error Analysis in an Open-Domain Question Answering System. ACM Transaction
on Information Systems, Vol.21, No.2, (April 2003), pp. 133-154, ISSN 1046-8188.
Nyberg, E., Burger, J., Mardis, S., & Ferrucci, D. (2004). Software Architectures for Advanced
QA, In: New Directions in Question Answering, M. Maybury, (Ed.), 19-30, ISBN 0-262-
63304-3, Cambridge, Massachusetts USA.
Patwardhan, S., Banerjee, S., & Pedersen, T. (2003). Using Measures of Semantic Relatedness
for Word Sense Disambiguation, Proceedings of the Fourth International Conference on
Intelligence Text Processing and Computational Linguistics, pp. 241-257, Mexico City,
Mexico, February 16-22, 2003.
Pedersen, T., Patwardhan, S., & Michelizzi, J. (2004). WordNet::Similarity - Measuring the
Relatedness of Concepts, Proceedings of the 9th National Conference on Artificial
Intelligence, (Intelligent Systems Demonstration), pp. 38-41, Stroudsburg, PA, USA,
May 2-7, 2004.
Schlaefer, N. (2007). A Semantic Approach to Question Answering, VDM Verlag Dr. Mueller
e.K., ISBN 3836450739.
Setzer, A., Gaizauskas, R., & Hepple, M. (2005). The Role of Inference in the Temporal
Annotation and Analysis of Text, Journal Language Resources and Evaluation, Vol. 39,
No. 2-3, (May 2005), pp. 243-265, ISSN 1574020X.
Shaw, M., & Garlan, D. (1996). Software Architecture: Perspectives on an Emerging Discipline,
Prentice Hall, ISBN 0131829572, USA.
Stevenson, M., & Greenwood, M. (2005). A semantic approach to IE pattern induction,
Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics,
pp. 379-386, Stroudsburg, Pensilvania, USA, June 25-30, 2005.
Verberne, S., Boves, L., Oostdijk, N., & Coppen, P. (2007). Discourse-based answering of
why-questions, Journal Traitement Automatique des Langues. Special issue on
Computational Approaches to Discourse and Document Processing, Vol. 47, No. 2,
(Octuber 2007), pp. 21-41.
Vicedo, J., & Fernandes, A. (2000). Applying Anaphora Resolution to Question Answering
and Information Retrieval Systems, Proceedigns of Web-Age Information Management,
pp. 344-355, ISBN 3-540-67627-9, Shanghai, China, June 21-23, 2000.
Voorhees, E. M. (2000). Overview of the TREC-9 Question Answering Track, Proceedings of
the Ninth Text REtrieval Conference. pp. 1-13, online publication, 06/10/11,

Automation 530
Voorhees, E. M. (2001). Overview of the TREC 2001 Question Answering Track, Proceedings
of the Tenth Text REtrieval Conference. pp. 1-15, online publication, 05/10/11,
of the Eleventh Text REtrieval Conference. pp. 1-15, online publication, 04/10/11,
of the Twelfth Text REtrieval Conference. pp. 1-13, online publication, 03/10/11,
of the Thirteenth Text REtrieval Conference, pp. 1-12, Special Publication 500-261,
Gaithersburg, Maryland, USA, November 16-19, 2004.
Waldinger, R., Appelt, D., Dungan, J., Fry, J., Hobbs, J., Israel, D., et al. (2004). Deductive
Question Answering from Multiple Resources, In: New Directions in Question
Answering, M. Maybury, (Ed.), 253-262, ISBN 0-262-63304-3, Cambridge,
Massachusetts USA.
27
An Intelligent System for
Improving Energy Efficiency in Building Using
Ontology and Building Automation Systems
Hendro Wicaksono, Kiril Aleksandrov and Sven Rogalski
FZI Research Centre for Information Technology
Germany
1. Introduction
Energy efficiency is a keyword that can be found nowadays in all domains in which energy
demand exists. The steadily rising energy demand, the consequent energy scarcity and
rising prices of energy resources are forcing companies and people to redefine their
activities in a more energy efficient way. Besides industry and transportation, the building
sector is the most important energy consumer, as a study for European countries showed
that up to 40% of the total energy consumption in the EU can be ascribed to this segment.
The residential use represents around 3% of the total energy consumption in the building
sector (Balaras et al., 2007).
There are many technical possibilities to improve energy usage efficiency. The European
Union has issued the Directive 2002/91/EC concerning the overall energy efficiency of
buildings. The directive aims to improve energy efficiency by taking into account outdoor
climatic and local conditions, as well as indoor climate requirements and cost-effectiveness
(Cox & Fischer Boel, 2002). The most widespread energy-saving measures in buildings are
conventional and are based on improved thermal insulation, energy-efficient air
conditioners and appliances as well as the usage of energy saving light bulbs and smart
meters. Yet there is more potential for energy saving in buildings than the abovementioned
measures. Based on the directive, a new DIN V 18599 Energetische Bewertung von
Gebuden describing an integral calculation method for energy requirements for heating,
acclimatisation and lighting has been published. Furthermore, in the future, energy savings
in buildings can be increased by utilizing a building automation system. This kind of
method is at least as important as the conventional thermal insulation of walls or insulating
glazing for improving energy efficiency in buildings (Spelsberg, 2006a, 2006b)
A look at the operation of residential and commercial buildings has also shown that there is
a wide range of technical possibilities to reduce energy consumption, as demanded by the
European Union through the European Directive 2002/91/EC on the energy efficiency in
buildings. An important aspect that can improve the energy efficiency in buildings is the use
of building automation systems. However, building automation systems are usually not
considered for energy conservation, as they are mostly used for comfort and safety. This
often causes great problems due to an inefficient use of these systems and unawareness of

Automation

532
energy consumption. It is therefore crucial that the existing system solutions are adapted to
focus on energy conservation.
This book chapter describes our research approach in developing an intelligent system to
improve energy efficiency in buildings, which takes into account the different technical
infrastructures of buildings, the occurring events in the buildings and the relevant
environmental factors with the help of building automation systems and ontology. By using
different building automation technologies, data in a variety of detail and quality could be
collected. Thus, it allows the user to have an overall view of the energy consumption inside
the entire building. Ontology offers the generic model for different building automation
technologies and allows logical inference as well.
Modern building automation solutions for the private or commercial sector offer a great
potential for the comprehensive improvement of energy efficiency. Building automation
systems that are composed of computer aided networked electronic devices are mostly used
for improvement of the individual quality of life, and safety. Building automation
technology offers an easy way to control and monitor the building conditions. However, the
existence of building automation systems does not imply an optimal usage of energy. They
are predominantly applied for comfort and security reasons and do not lead to an
improvement in energy consuming behaviour.
Low cost and low energy consuming building automation technologies have already been
developed in recent years. Technologies such as digitalSTROM
1
and WPC
2
offer energy
measurement and sensors by using small chips that consume less than 10 mW (Watteco,
2009). These chips communicate through power line communication (PLC). Therefore
modifications of walls and other building structures as well as extra wiring in the building
are not necessary for installation. By using these devices, extra energy consumption can be
avoided. Thus, an improvement of energy efficiency can be achieved.
In this paper we proposed a framework of intelligent system for energy management in
buildings by utilizing building automation systems. The framework uses a knowledge-
based approach to providing an intelligent analysis based on the relations between energy
consumption, activities and events in the building, building related information and
surrounding factors, such as temperature and weather conditions.
2. State of the art and related works
The importance of the use of building automation systems in assisting energy saving in
buildings, especially households, has been proven in recent years. In 2009 the Electric Power
Research Institute USA conducted research about the impact of energy consumption
feedback information on energy conservation levels in households. They categorized the
feedback mechanism based on information availability into standard billing, enhanced
billing, real-time feedback, real-time plus feedback, etc. (Neenan et al., 2009). The research
showed that real-time plus feedback leads to the biggest improvement of energy
conservation compared to the other feedback mechanisms, despite its higher cost of
implementation. Real-time plus feedback allows users to monitor their energy consumption

1
http://www.digitalstrom.org
2
http://www.watteco.com
Improving Energy Efficiency in Building Using Ontology and Building Automation Systems

533
and/or control appliances in their home in real-time by utilizing building automation
system (BAS) and home area network (HAN).
In Europe, there are several projects that aim to improve energy efficiency in buildings by
using building automation technologies. Most of the projects do not have the functionality
to intelligently synchronize the energy consumption with other energy consumption
impacting parameters, such as temperature, occupancy or the state of the building. Some of
the projecs developed knowledge based approachs to enable the intelligent analysis of the
collected data. In the project IntUBE, a knowledge base containing semantic building
objects, their properties and their relationships, is developed in their energy management
integration platform (Madrazo et al., 2009). The knowledge base to provide an intelligent
system is not only developed in energy management system for buildings. The project AmI-
MoSeS develops an (ambient) intelligent monitoring system for energy consumption,
dedicated to manufacturing SMEs, to provide comprehensive information about the energy
efficiency and a knowledge-based decision support system for the optimisation of energy
efficiency (ATB Institut fr angewandte Systemtechnik Bremen GmbH et al., 2008). In these
projects the knowledge bases are mostly generated manually. This prevents the system
from being able to react to and solve problems accordingly by a given specific situation,
since all the situation-reaction causal relations are given by a human, based only on his
knowledge and experience. This might be too general.
Currently there are some research projects that investigate and develop ontology based
approaches to the building automation domain. Ontology is used as the generic application
model facilitating an integration of heterogeneous building automation networks. The
ontology provides a single access point for configuration and maintenance tasks, as well as
automatic calculation of gateway configuration data (Reinisch et al., 2008). Ontology is also
used to build an expert system to control home automation devices by incorporating SWRL
rules that regulate the system behaviour (Valiente-Rocha & Lozano-Tello, 2010). In
requirement engineering of building automation systems, ontology provides a semantic
representation of the requirements, containing possible and specific requirements and their
causal dependencies. By means of the ontology, consistent requirements for an individual
building automation system can be achieved (Runde et al., 2009). Ontology that allows a
representation of a sophisticated model and an intelligent reasoning is used to support
complex interoperation, generalization and validation tasks in the building automation
environment (Bonino et al., 2008).
Ontologies represent knowledge in a particular domain. They are commonly created
manually by experts in the corresponding domain. For example biology ontologies are
created by biologists. However, a manual ontology creation is considered a very time-
consuming task. To overcome this problem, researchers have developed methods to
generate ontology components in a semi-automatic way. Researchers from the University of
Leipzig, Germany have developed a semi-automatic OWL generation from XML files by
converting XSD items to ontologys classes and attributes. To perform this they developed
four XSLT instances to transform XML files to OWL without any semantic and structural
intervention during the transformation (Bohring & Auer, 2005). Dragan Gasevic et al. have
also developed an XSLT based approach to automatically generate OWL ontology from
UML. The resulting ontology still needs to be refined using ontology development tools.
However, the approach helps software engineers who want to participate in ontology

Automation

534
development without having a deep knowledge of ontology development tools. Ontology
can be generated semi-automatically from a set of free text documents. It was proven by
research conducted by the University of Karlsruhe, Germany in 2001 (Maedche & Staab,
2001). They developed a tool called TextToOnto
3
for semi-automatic ontology construction
encompassing ontology import, extraction, pruning, refinement, and evaluation. The
framework helps ontology engineers to construct ontologies.
3. Overview of the approach
In this paper we propose an IT framework for intelligent energy management in a building
using building automation technologies. The framework considers different aspects of the
building that influence the energy consumption in the building and are to be considered in
the analysis of the energy consumption. The framework allows the user to have a holistic
and integrated view of the energy consumption in their apartment, office, as well as in entire
buildings. The building automation system offers the possibility to perform an energy
consumption evaluation in different levels of detail and quality, for instance, energy
consumption per appliance, per group of appliances, per zone in the building, or per user
event. These evaluations are carried out by considering the relation between building
elements, energy consuming appliances, surrounding circumstances such as temperature
and weather conditions, as well as user events.

Fig. 1. Functional architecture of SERUM-iB
The energy consumption can be evaluated based on activities or events that occurred in the
building, for example breakfast, cooking, sleeping, work day and holiday. This kind of

3
http://texttoonto.sourceforge.net

535
evaluation offers the occupants the possibility to get a view of their activity-specific energy
consumption. Therefore, they can consider, whether their activities are energy-efficient
enough and how can they reorganize their activities to improve the energy efficiency. The
framework offers the functions for energy monitoring, energy data analysis and manual and
automatic controlling of energy consuming appliances using building automation systems.
Fig. 1 shows the designed functional architecture of the framework whose main modules are
explained briefly in the following sub sections. We named the developed framework
SERUM-iB (Smart Energy and Resource Usage and Management in Building)
Framework. The work described in this paper is part of our research project KEHL
4
.
3.1 2D-drawing interpretation module
CAD software applications that find application in construction engineering are commonly
used to draw building layouts in two dimensional graphics. But unfortunately, the two
dimensional drawings only contain geometric information of the objects, for instance lines,
points, curves, circles, symbols, etc. They are not able to describe the semantics of the
building components contained in the drawings themselves. People could still understand
the semantics of the drawing, if they had the knowledge in interpreting the semantics of the
symbols, shapes and other geometric information, for example symbols of doors, double
lines representing walls, rectangles constructed by four double lines representing rooms, etc.
We have developed a method to semi-automatically interpret the semantics of 2D-drawings
to support the life-cycle management of building automation systems (Krahtov et al., 2009).
In this module we use configurable JavaScript based rules to semi-automatically interpret
2D-CAD-drawings. These rules define the relation between CAD-symbols and the semantics
of building components and energy consuming appliances. The results of the 2D-drawing
interpretation are stored as ontology individuals. This is explained further in section 5.
In existing energy analysis systems, users have to manually configure the system in order to
perform an energy consumption analysis for each room or appliances. Another
configuration mechanism is through automatic discovery of appliances. But it has the
drawback that users have to manually relate the appliance to the building environment. The
mechanism is dependent on a specific building automation technology as well. The
interpretation of 2D-drawings helps users to configure the software framework, so they can
have a model that represents their building layout with minimal effort.
3.2 Sensor data acquisition module
This module collects data from different building automation systems. Building automation
technologies have different bus systems to exchange data. Most of them provide gateways
to IP networks. This module contains an interface to communicate with different building
automation logic control units or gateways via Simple Object Access Protocol (SOAP),
which uses Extensible Markup Language (XML) for its message format and relies on
Hypertext Transfer Protocol (HTTP) for message negotiation and transmission. The
collected data, for instance energy consumption and the relevant sensor data such as that
from the contact sensor, occupancy sensor, and weather station are converted into a uniform

4
http://kehl.actihome.de

Automation

536
format and unit, for example, the energy consumptions are converted into kWh, the values
from occupancy sensors are converted to true (room is occupied) and false (room is empty).
These data are then aggregated and stored in a relational database.
In order to allow the visual representation of energy consumption data, we perform the
necessary data pre-processing such as removing erroneous values, data transformation and
data selection. The data are prepared to enable an energy consumption analysis of different
criteria based on relation between rooms, appliances, time, and user events as mentioned
before. This module provides data in such a form as to enable the execution of a data mining
algorithm for finding the energy usage pattern.
3.3 Module for energy analysis
The energy consumption and other related data which are collected and stored in the SQL
database as mass data are analysed in this module. Through this module, energy
consumption can be related to device levels, room and time, which in addition, can be
combined with the relation to user events and surrounding conditions. In this module we
consider two methods of energy analysis, i.e. data-driven analysis and knowledge-driven
analysis. Data-driven analysis means the analysis is conducted directly on the collected data
by performing SQL-query, simple calculation, or visualization, for instance, energy
consumption per time unit and each appliance. Knowledge-driven analysis is not conducted
directly on the data, but by utilizing ontology that represents the knowledge.
The module consists of three sub-modules. They are a learning engine, a knowledge base
consisting of ontology and rules and an energy analyser. The learning engine is responsible
for extracting the knowledge hidden in the database. The learning engine utilizes data
mining algorithms to get the association rules representing causal relations between the
data. These rules are used, for example, for detecting energy usage anomalies and energy
wasting. Furthermore, the rules are used to detect the activities that occur in the building
based on data from different sensors, for instance, occupancy sensors, windows and door
contact sensors, pressure sensors, etc. The data mining algorithms are also used for
appliance classification based on their energy consumption. They are also used to detect the
operation states of appliances. These functions will be explained further in section 6.
The knowledge base is the central component that provides the intelligent mechanism of the
system. The knowledge base containing OWL ontology and SWRL rules is initially created
by an expert. The knowledge base is then enriched with knowledge components (rules and
instances), which are resulted from the 2D-Drawing Interpretation module and data mining
algorithm.
The analysis module evaluates energy consumption data that were collected and aggregated
by the data acquisition module, which is then presented in the presentation module.
Through this module, energy consumptions can be related to device levels, room and time.
In addition, they can be combined with the relation to user events and surrounding
conditions.
In this module we consider two methods of energy analysis, i.e. data-driven analysis and
knowledge-driven analysis. Data-driven analysis means the analysis is conducted directly
on the collected data by performing SQL-query, simple calculation, or visualization, for

537
instance energy consumption per time unit and each appliance. Knowledge-driven analysis
is not conducted directly on the data, but by utilizing ontology that represents the
knowledge. The ontology is created manually by an expert and enriched with ontological
elements generated from the data through the ontology generation process. This is
explained further in section 5 and 6.
3.4 Presentation module
This module gives the users a visual overview of their energy consumption in the building
related to the building environment and their behaviour. Within the module, the energy
consumption and other relevant information are presented intuitively. The energy
consumptions in a building and in single rooms or for appliances are visualized via a
terminal or smart client placed in the building. At the same time the configuration and
control of the building automation systems can be performed within the developed GUI.
4. Ontology for knowledge representation
In the SERUM-iB framework we use ontology to build the knowledge base. It contains
concepts representing rooms, energy consuming appliances, building automation devices,
environment parameters, events and their relations. It is represented in OWL (Web
Ontology Language), a W3C
5
specified knowledge representation language (Smith et al.,
2004). In our work, we use the knowledge-driven analysis to find energy wasting conditions
and to recognize energy usage anomalies in buildings. We also use it for identifying the
operational state of an appliance, to identify appliance classes based on their energy
consumption and to identify the user event or consumption behaviour based on the sensor
data.
Within the SERUM-iB framework, we develop the three following methods to generate the
ontology components:
1. manual ontology generation by an expert,
2. semi-automatic through interpretation of 2D-drawings, and
3. semi-automatic by data mining algorithms.
The ontological classes as well as their attributes and relations representing building
automation devices, for instance sensor and energy meter, building environments such as
door, window, room, are created manually by experts. The ontology containing these hand-
crafted elements builds the knowledge called building domain knowledge in the knowledge
base. It contains only the ontological classes or Tbox components that terminologically
describe the knowledge. It provides a common conceptual vocabulary of building domains.
It does not contain any ontological individuals or Abox components. The building domain
knowledge represents the meta model of buildings, therefore it owns the validity for all
buildings. It does not contain any building instance specific information.
As occupants, we usually do not realize that our current energy usage leads to energy
wasting. For instance, the light on the balcony is still on, even though the sun has already
risen. In our work, we try to resolve this problem by modelling in the ontology the general

5
http://www.w3.org

Automation

538
situations in the building that lead to energy wasting. Thus, a computer supported reaction
can be performed.
Energy wasting situations are modelled by experts in the base ontology. These situations are
described in a rule representing language called SWRL (Semantic Web Rule Language).
SWRL is a language combined by OWL (Web Ontology Language) and RuleML (Rule Mark-
up Language) that enables the integration of rules in OWL ontology (Horrock et al, 2004).
Here we provide examples of modelling energy waste situation with SWRL. We assume that
a smart meter attached to an appliance can give information on whether the appliance is
currently in use. Rule (1) models the situation that turning the heating appliance on and
opening the window located in the same room leads to energy wasting.

e(?h) EnergyWast ?r) isIn(?h, ?r) isIn(?w, Room(?r)
?h) To(?em, isAttached ) true" " em, hasState(? Meter(?em) E ) true" " h, hasState(?
Heater(?h) ) true" " cs, hasState(? sor(?cs) ContactSen ?w) To(?cs, isAttached Window(?w)

nergy (1)
The ontology representing the building domain knowledge is enriched with ontological
individuals or Abox components from building specific information. The individuals are
created by configuring the framework based on building specific information. They are also
created automatically by interpreting the 2D construction drawing of the building layout
depicting building structures, such as room, floor, door, windows and contained energy
consuming appliances, as well as building automation devices, for instance temperature
sensors and occupancy sensors. The semantic information of building environments such as
rooms, lights, and appliances is obtained from geometric data through the interpretation of
2D-drawings. The concept of 2D-drawing interpretation is further presented in section 5.
Kitchen Corridor
Bathroom
Livingroom
Bedroom
C
C
C
C
Kitchen Corridor
Bathroom
Livingroom
Bedroom
C
C
C
C
Kitchen Corridor
Bathroom
Livingroom
Bedroom
C
C
C
C

Fig. 2. Creating building specific knowledge from building domain knowledge
The building domain knowledge is added with building specific SWRL rules. These SWRL
rules are created semi-automatically from the data mining algorithm. The data mining
algorithm generates the association and classification rules as the hidden knowledge that is
extracted from the collection of sensor data. The rules are converted to SWRL rules and
integrated into the building domain knowledge. Both processes create the new version of

539
knowledge called building specific knowledge. Fig. 2 depicts the concept of creating
building specific knowledge from building domain knowledge using 2D-drawing
interpretation and the data mining algorithm.
5. Knowledge acquisition through interpretation of 2D construction drawings
For its correct usage, this approach requires geometric and semantic information about
building layout and appliances as an initial knowledge base. Geometric information could
include the exact shape and measurements of rooms in a building, their topological relations
to each other (and/or to the energy appliances) as well as the position of the appliances and
their dimensions. Semantic information is identification keys, names, consumption values,
etc. Manually inputting this type of data, especially for large buildings or facilities, could be
a time-consuming, error-prone and inefficient task. For these reasons we adopted an
approach for rule-based, semi-automatic interpretation of 2D construction drawings to
create the initial knowledge base.
In order for the intelligent system to improve the energy efficiency in buildings, the
following domain elements are considered important:
1. Zones that describe various parts of a building like rooms, corridors, etc.
2. Zones can also be nested (i.e. apartment, floor, building), as they can comprise other
zones. They have geometric information about their layout and their size as well as
semantic information (names, ids etc.). Zones can be topologically connected to energy
consumption devices, furniture, etc.
3. Appliances such as any electrical device or energy consumption device like heaters.
4. Sensors and smart meters.
A layout of a building is commonly created as a two dimensional drawing or sketch using
CAD software applications, such as AutoCAD
6
. Further AutoCAD-based software tools are
used to plan and model many domains of a building, such as ventilation, heating, access
controls and photovoltaic (Krahtov et al., 2009). The number of elements in a sketch and its
complexity may vary (Donath, 2008). Some examples of variations are the degree of the
details drawn (information density Fig. 3 a, b) or their representation (representation depth
Fig. 3 c, d).
The diversity of the 2D-drawings creates problems in the energy management approach
described in this chapter. There are currently two open issues linked to knowledge
acquisition from 2D building plans. The first issue is their overall high complexity and the
lack of explicit semantics of drawn objects, due to the fact that 2D building plans are
mainly used by professional designers and engineers who have the background knowledge
to interpret them. The second issue is the difference in the structure of the various CAD-
drawings and in the representation styles used in it (Kwon et al., 2009). This high
variability obstructs any attempts at an automatic interpretation of an arbitrary building
plan.
To handle high complexity we need an abstraction from the unimportant elements of a
building plan that would allow us to concentrate on the energy consumption-related objects

6
http://usa.autodesk.com/autocad

Automation

540
of a building. For the second problem we need a semi-automatic solution that allows us to
use our background knowledge about the structure and representation style of a 2D
building drawing.

Fig. 3. Variety in the degree of information density and level of representation depth in 2D-
CAD-drawings
We developed an approach that deals with the described issues and allows us to semi-
automatically interpret 2D drawings with an abstraction level defined by the user. The
workflow of the approach consists of the following steps:
1. First step is the identification of relevant entities from a 2D drawing. Those can be
building elements as well as building automation devices.
2. The second step is to define a diagram for the relevant elements from step one.
3. The third step is the definition of a mapping between the visual representations in the
2D drawing and the elements of the diagram via pattern based recognition rules.
4. The fourth step is the execution of the defined recognition rules.
5. The last step is the transforming of the recognized elements to knowledge base
individuals.
We implemented the approach, creating the software framework presented on Fig. 4. The
framework consists of several parts (i.e. AutoCAD export function, diagram editor,
interpretation rule engine, transformation utility). These framework parts aim to solve the
described issues, at the same time ensuring the modularity of the approach and its extensibility.

541

Fig. 4. Simplified architecture of the 2D drawing interpretation framework
The AutoCAD export function is the first component of the developed framework. This
component was needed to prepare the 2D drawings for the interpretation process. As part of
it, an XML-based transfer format was developed that can be used directly in the
interpretation process. This is a key feature of the interpretation because it allows the
development of a generic recognition engine that works with the neutral format and that
can be extended by additionally exporting modules for various CAD applications. For
verification an exemplary conversion function was implemented that exports a native
AutoCAD-file (.dwg format) into the defined intermediate data format. The function was
developed using ObjectARX, a programming interface provided by Autodesk. The export
function allows a variable degree of detail for the exported geometric primitives.
The second part of the developed framework is the diagram editor. It is developed as an
Eclipse RCP
7
program and provides an intuitive configuration process for the source
domain of the interpretation framework. Via the diagram editor, the user can define all the
important building parts and elements from the given 2D drawing, their semantic
properties and their topological relations. They all build up a diagram also called a domain
model. By the definition the user can configure the abstraction level for the interpretation
process, ignoring all irrelevant drawing details. He can also define various semantic
properties, which are not drawn on or not directly derivable from the 2D drawing (i.e.
names and ids of the appliances, etc.). The most important part of the diagram that is
defined through the editor is the interpretation rule base. These rules are defined as
JavaScript routines, which can read and manipulate elements from the neutral transfer
format, created by the AutoCAD export function. They describe patterns in the XML format
that represent elements of the 2D construction drawing. They are based on the layers,
attribute information and relations between AutoCAD primitives that build up a
construction element. By defining rules and attaching them to the elements of the domain
model created by the diagram editor, the user creates a mapping between the drawing
elements and the elements of the domain.

7
http://www.eclipse.org/home/categories/rcp.php

Automation

542
The rules are used in the third major component of the interpretation framework the
interpretation rule engine. This is a Java based application that can execute the defined
interpretation rules over the extracted intermediate XML data, recognizing the building
elements and building automation appliances. For the execution of the rules we use the
Rhino
8
framework, which is an open-source implementation of JavaScript for Java and
enables the end user to use scripts inside of Java applications.
The last part of the interpretation framework is the transformation service. It is used
subsequently after the end of the interpretation process, to transform the recognized
elements into ontology individuals, creating the initial state of the knowledge base. The
transformation service provides a configurable mapping of recognized building elements,
building automation or other electronic elements to zones, appliances or sensors and smart
meters.
The following models participate in the interpretation process:
1. The domain model for information about the environment, defined by the knowledge
experts familiar with the application domain and the specifics of the given 2D drawing.
The domain model is the diagram for the recognition process and is defined through
the diagram editor.
2. The logical model is the instance of the domain model, created via the recognition
process. It represents an abstraction of the 2D drawing.
3. The knowledge model is generated from the logical model through the transformation
service.
The described approach for interpretation of two-dimensional building drawings
successfully handles the issues that were presented earlier in this book chapter, providing
the following benefits in context of the knowledge acquisition for the creation of an initial
knowledge data base:
1. No manual input of geometric and topological information should be done.
The building-related information as well as the appliance positions and their
topological relations to the environment or to each other are provided by the
interpretation process.
2. User defined level of abstraction is guaranteed. All unnecessary elements of a two
dimensional drawing can be ignored.
3. The approach provides flexibility in both dimensions of types and attributes. That
means that new energy appliances (new building automation or household devices)
that havent been considered yet can easily be defined and integrated into an existing
energy efficiency system.
6. Knowledge generation through data mining algorithm
In our work, data mining algorithms are used to identify energy consumption patterns and
their dependencies. Data mining is defined as the entire method-based computer application
process with the purpose to extract hidden knowledge from data (Kantardzic, 2003). In the
SERUM-iB framework we used different data mining procedures to generate knowledge for
recognition of energy usage anomalies, determining classification of appliances based on

8
http://www.mozilla.org/rhino/

543
their energy consumption, and identifying operation states of an appliance based on energy
usage. We illustrate these procedures in the following sub sections.
6.1 Recognition of energy usage anomalies
In this work, we relate the energy consumption with the occupants behavioural pattern in
the building. We aim to recognize energy consumptions that do not occur normally in the
building. To perform this task, we have to know how energies are consumed normally
regarding occupant events and surroundings. For example, normally when an occupant is
sleeping and the outside temperature is comfortable, let us say greater than 20 degrees
Celsius, total energy consumption in the building is low, for instance lower than 10 kWh. If
in the same pre condition total energy consumption in building is more than 10 kWh, then it
is considered a usage anomaly.
It is difficult for users if they always have to log their activities. In our work we use simple
sensors to automatically recognize user activities. Simple sensors can provide important
hints about user activity. For instance, an occupancy sensor in a kitchen can strongly give a
clue whether somebody is currently cooking. Of course it should be combined with
information of appliance states in the kitchen. The method for automatic recognition of user
activity is explained in section 6.2.
The rules representing normal energy consumptions are obtained through a data
mining classification rules algorithm. The algorithm is based on a divide-and-conquer
approach. The created rule (2) shows the probability of 67% of how often it could
happen if the event is sleeping while outside temperature is greater than 20 degree with
total energy consumption is lower than 10. This value is called confidence. The rules
described in (2) represent a condition that normally occurs. The rule is transformed to
(3), in order to represent an anomaly condition, by negating the consequent part of the
rule.

) 67 . 0 : ( 10 20 " " conf on yConsumpti TotalEnerg perature OutsideTem Sleeping Event
(2)

) 67 . 0 : ( 10 20 " " conf on yConsumpti TotalEnerg perature OutsideTem Sleeping Event
(3)
The rules created by the data mining algorithm are stored in ontology as SWRL. SWRL rule
(4) represents the transformed rule (3), which is stored in ontology.

) (? ) 10 , (? :
) ? , (? ) (? ) 20 , (? :
) ? , (? ) (? ) " " , (? ) (?
e ly UsageAnoma smv nOrEqual greaterTha swrlb
smv sm hasValue sm SmartMeter otv nOrEqual greaterTha swrlb
otv ot hasValue ot rmometer OutsideThe Sleeping e hasName e Event

(4)
The rules resulting from the data mining algorithm do not always have 100% confidence.
Therefore we represent the rules as SWRL in order to enable verification by using a rule
editor provided in the Presentation Module or using a SWRL editor such as Protg. The
editor enables users to add, modify and delete the resulted rules. Fig. 5 illustrates the whole
process to enrich the knowledge base with SWRL rules coming from data mining to
recognize the energy usage anomalies.

Automation

544

Fig. 5. Data mining process for recognizing an energy usage anomaly
6.2 Recognition of user activities in the building
In our work, we use occupancy sensors, window contact sensors, door contact sensors, as
well as energy meter to collect the data that will be used to recognize user activities in a
building. An energy meter attached to an appliance can give a clue as to the appliances
state. It means we can get the information, for example whether the appliance is on or off.

Fig. 6. Data mining process for recognizing user activities in a building
The data from these different sensors are collected and stored in database. Similar to the
process for recognizing energy usage anomalies, the data mining algorithm is executed on
the data. The algorithm generates rules that it uses to recognize the user activities. Fig. 6
depicts the whole process to generate these rules and store them in a knowledge base. The
simplified rule (5) is an example of a rule used to recognize the user activity Holiday
based on the data depicted in Fig. 5.

) , (? ) " " , (?
) (? ... ) " " , (? ) (? ) " " , (? ) (?
true u ace isTakingPl Holiday u hasName
u ty UserActivi off m hasState m Microwave off s hasState s Stove

(5)

545
6.3 Determining classification of appliances
Classification is the process of finding a set of models or functions that describe and
distinguish data classes or concepts for the purpose of being able to use the model to predict
the class of objects whose attribute is unknown (Han & Kamber, 2001). The model is based
on the analysis of a set of training data, i.e. data objects whose attribute is known.
In our work, appliances are classified based on their energy consumption by using
classification algorithms, for instance Naive Bayes. Such a grouping determines which
devices consume more energy or if a household appliance works properly due to its energy
demand. The grouping even allows the occupants to understand which of their activities
need more energy. The resulted classes are transformed into ontological individuals of class
Energy_Cosumption_Class. A relation is_member_of is added. It relates instances of
appliance classes created by the interpretation of 2D-drawings module and corresponding
instances of class Energy_Consumption_Class.
6.4 Generating knowledge to identify operation states of appliances
In many cases an appliance consumes energy differently depending on its operational states.
We are often not able to determine in which state our appliances consume more energy.
Therefore we cannot react respectively, for example, to reduce or even to avoid using
appliances operating modes that consume more energy.
We use cluster analysis algorithms to recognize operation states of appliances, for instance, a
washing machine has the operation states standby, water heating, washing and dry-
spinning. A clustering algorithm analyses and puts data objects in certain clusters so that
objects have high similarity to the others within a cluster but dissimilarity in comparison to
objects in other clusters. In our case, we measure the similarity of energy consumption of an
appliance. Fig. 7 shows the approach to generate the knowledge to identify operation states
of appliances using data mining algorithm.

Fig. 7. Data mining process to identify operation states of appliances
7. Intelligent analysis through ontology processing
In the knowledge base represented in ontology, all conditions of energy wasting and
anomalies are represented as SWRL. Periodically the data acquisition module requests real-

Automation

546
time data from a building automation gateway. These data contain states of all the installed
building automation devices. SWRL rules are used to decide whether these incoming data
correspond to an energy waste or usage anomaly condition. We develop a rule engine based
on SWRLJessBridge to support the execution of SWRL rules combined with Protg API
that provides functionality in managing OWL ontology.
First the attribute values of relevant ontological instance are set to values corresponding to
incoming data. For example, if contact sensor attached to window gives a state Open then
the attribute hasState of the corresponding ontology instance of concept ContactSensor is
set to Open. After that the rule engine executes the SWRL rules and automatically assigns
individuals to the ontology classes defined in the rules consequent. For example for rule (1)
the instance of concept Heater is additionally assigned to EnergyWaste class and for rule
(4) the instance of class UserActivity to class UsageAnomaly.
SPARQL (SPARQL Protocol and RDF Query Language) is used to evaluate whether energy
wasting conditions or energy usage anomalies occurs. Which appliances cause the energy
wastage can be retrieved as well. It is performed by querying all the individuals of
EnergyWaste or UsageAnomaly class. If individuals of these classes are found, a predefined
automatic notification to the user such as sending an SMS, email or visually on the GUI can
be performed. With this mechanism, the user can react more quickly in order to avoid more
energy wasting. SPARQL is also used to get appliances belonging to a certain energy
consumption class. The SPARQL processing is implemented in analysis module.
8. Conclusion
In this paper we have presented a system of comprehensive intelligent energy analysis in
buildings. In the developed system, we combined classical data-driven energy analysis with
novel knowledge-driven energy analysis that is supported by ontology. The analysis is
performed on information collected from building automation devices. The ontology
supported analysis approach provides intelligent assistance to improve energy efficiency in
households, by intensively considering individual user behaviour and current states in the
building. Users do not have to read the whole energy consumption data or energy usage
profile curves in order to understand their energy usage pattern. The system will
understand the energy usage pattern, and notify user when energy inefficient conditions
occur.
We have also presented the methods to semi-automatically generate ontological components
i.e. individuals and rules, in building automation domains through the interpretation of 2D-
drawing and data mining approaches. The approaches can reduce the ontology generation
process, since the manual ontology components creation process is very time-consuming.
However, the base ontology representing terminological knowledge in building automation
is created manually.
The approach described in this paper can be adopted in more complex systems such as
production systems by integrating the base ontology with other ontologies in production
system domains. Additional relative ontological components representing knowledge in
production systems, for example product, production tasks, machine related information
should be created.

547
9. Acknowledgment
This work has been partially funded by the German government (BMBF) through the
research project KEHL within the program KMU-Innovativ. We would like to thank Enrico
Kusnady, Ariel Guz, Mariusz Kosecki, and Youssef Aitlaydi for their contribution in
implementing the prototype of this work.
10. References
ATB Institut fr angewandte Systemtechnik Bremen GmbH; Fundacin LABEIN-Tecnalia;
RIFOX Hans Richter GmbH Spezialarmaturen; Vicinay Cadenas Sociedad
Anonima; MB Air Systems Limited; Reimesch Kommunikationssysteme GmbH;
Disseny I Subministres Tecnics, S.A.; Solvera Lynx d.d. & Valtion Teknillinen
Tutkimuskeskus (2008). Ami-MoSES Concept, In: AmI-MoSES Deliverable 7.4,
08.09.2011, Available from
http://www.ami-moses.eu/fileadmin/templates/amimoses/files/AmI-
MoSES_D7.4_Concept_Public_v.1.0.pdf
Balaras, C.A.; Gaglia, A.G.; Georgopoulou, E.; Mirasgedis, S.; Sarafidis, Y. & Lalas,
D.P. (2007). European Residential Buildings and Empirical Assessment of the
Hellenic Building Stock, Energy Consumption, Emissions and Potential Energy
Savings. Journal Building and Environment, Vol. 42 No. 3, pp. 1298-1314
Bohring, H. & Auer, S. (2005). Mapping XML to OWL Ontologies, In: Leipziger Informatik-
Tage 2005, pp. 147-156
Bonino, D.; Castellina, E. & Corno, F. (2008). DOG: an Ontology-Powered OSGi Domotic
Gateway, In: Proceeding 20th IEEE International Conference on Tools with Artificial
Intelligence, Vol. 1, pp. 157-160, ISBN 978-0-7695-3440-4, IEEE Computer Society
Washington, DC, USA
Cox, P. & Fischer Boel, M. (2002). DIRECTIVE 2002/91/EC OF THE EUROPEAN
PARLIAMENT AND OF THE COUNCIL of 16 December 2002 on the energy
performance of buildings, In: Official Journal of the European Communities,
13.10.2011, Available from
http://eur-
lex.europa.eu/LexUriServ/LexUriServ.do?uri=OJ:L:2003:001:0065:0065:EN:PDF
Donath, D. (2008) Bauaufnahme und Planung im Bestand, pp. 35-36, Vieweg+Teubner Verlag,
Wiesbaden
Gasevic D.; Djuric, D.; Devedzic, V. & Damjanovic, V. (2004). From UML to Ready-To-Use
OWL Ontologie, In: Proceedings of the IEEE International Conference Intelligent
Systems
Han, J. & Kamber, M. (2001). Data Mining Concepts and Techniques, Morgan Kaufmann
Publishers, ISBN 978-155-8609-01-3, San Francisco, USA
Horrock, I.; Patel-Schneider, P.; Boley, H.; Tabet, S.; Grosof, B. & Dean, M. (2004). SWRL: A
Semantic Web Rule Language Combining OWL and RuleML, In: W3C Member
Submission, May 2004
Kantardzic, M. (2003). Data Mining: Concepts, Models, Methods, and Algorithms, John Wiley &
Sons, ISBN 978-047-1228-52-3, New Jersey, USA
Krahtov, K.; Rogalski, S.; Wacker, D.; Gutu, D. & Ovtcharova, J. (2009). A Generic
Framework for Life-Cycle-Management of heterogenic Building Automation

Automation

548
Systems, In: Proceedings to Flexible Automation and Intteligent Manufacturing, 19th
International Conferemce (FAIM 2009), July 6th-8th, 2009
Kwon, D.Y.; Gross, M.D. & Do, E.Y.L. (2009), ArchiDNA: An interactive system for creating
2D and 3D conceptual drawings in architectural design, Computer-Aided Design,
Vol. 41, No. 3, pp. 159-172
Madrazo, L.; Sicilia, A.; Cantos, S.; Bhms, M.; Chervier, B.; Decorme, R.; Dawood, N.;
Karhela, T. & Plokker W. (2009). IntUBE Integration Platform Specification, In:
IntUBE Executive Summary of Deliverable 6.2, 08.09.2011, Available from
http://www.intube.eu/wp-
content/uploads/COI/2010/executive_summary_D6.2.pdf
Maedche, A. & Staab, S. (2001). Ontology Learning for the Semantic Web, In: IEEE Intelligent
System, vol. 16, pp. 72-79
Neenan, B.; Robinson, B. & Boisvert R. N. (2009). Residential Electricity Use Feedback: A
Research Synthesis and Economic Framework. EPRI, Palo Alto, CA. 1016844
Reinisch, C.; Granzer, W.; Praus, F. & Kastner, W. (2008). Integration of Heterogeneous
Building Automation Systems using Ontologies, In: Proceedings of 34th Annual
Conference of the IEEE Industrial Electronics Society (IECON '08), pp. 2736-2741
Runde, S.; Dibowski, H.; Fay, A. & Kabitzsch K. (2009). A Semantic Requirement Ontology
for the Engineering of Building Automation Systems by means of OWL, In:
Proceedings of the 14th IEEE international conference on Emerging technologies & factory
automation, pp. 413-420, ISBN 978-1-4244-2727-7, IEEE Press Piscataway, NJ, USA
Smith, M.; Welly, C. & McGuinness D. (2004). OWL Web Ontology Language Guide, In:
W3C Recommendation, February 2004
Spelsberg, J. (2006). Chancen mit dem Gebudeenergiepass. Energie sparen durch Gebu-
deautomation. Technik am Bau, Vol. 37, pp. 74-77
Spelsberg, J. (2006). Knstiliche Intelligenz im Bro senkt Energiekosten. UmweltMagazin,
Oct-Nov 2006, pp. 62-63
Valiente-Rocha, P.A. & Lozano-Tello, A. (2010). Ontology-Based Expert System for Home
Automation Controlling, In: Proceeding IEA/AIE'10 Proceedings of the 23rd
international conference on Industrial engineering and other applications of
applied intelligent systems, Volume Part I, pp. 661-670, ISBN 3-642-13021-6 978-3-
642-13021-2, Springer-Verlag Berlin, Heidelberg

Automation

Uploaded by

Copyright:

Available Formats

Automation

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Automation

Uploaded by

Copyright:

Available Formats

What is the main topic discussed in the document?

What is the main topic discussed in the document?

What are some applications of automation mentioned?

What are some applications of automation mentioned?

AUTOMATION

326, ISSN 0001-4966.

- immune colonies, - infected colonies

g(t )u()d = g(t) u(t),

that is, the "transposed matrix" of K(A

for certain coefcients a

transform the initial matrices into the control canonical form.

we get the control canonical form

we get the control canonical form

(2.ii) Reasoning in the same way, it results:

(2) rank K(A, B) = rank

No solutions exist because

whose solution is: u

The solutions can be parameterized by u

193 Learning Automation to Teach Mathematics

(2.3) rank K(A

where the last equality follows from Cayley-Hamilton theorem.

(2) The computations are analogous to those in (1).

is a function of many parameters, but for the boiler of a specific construction, it

, which assures that the evaporator is

, which shall be delivered to the

absorbed by the walls of the boiler furnace chamber, expressed in

absorbed by the walls of the furnace chamber

, in kW, delivered to the evaporator of the OP-210M boiler, as a

, as the heat flux is proportional to temperature

is reduced, and the heat flux

calculations, determined for various saturated steam mass flow rates,

to be delivered to the evaporator.

, which shall be delivered from the

delivered to the evaporator of the boiler in the furnace chamber for

, is determined on the basis of the

cos cos sin sin ,

sin sin cos

sin cos sin cos sin cos sin

cos cos sin cos sin sin

LOC `50 sinp

LOC are the average values of Halstead Volume, McCabes Cyclomatic

= mgl sin ml z cos c

C], mean radiant

You might also like