2015 - Almeida Et Al. - Multicriteria and Multiobjective Models For Risk, Reliability and Maintenance Decision Analysis

International Series in
Operations Research & Management Science

Adiel Teixeira de Almeida
Cristiano Alexandre Virgínio Cavalcante
Marcelo Hazin Alencar
Rodrigo José Pires Ferreira
Adiel Teixeira de Almeida-Filho
Thalles Vitelli Garcez
Multicriteria and
Multiobjective Models
for Risk,
Reliability and Maintenance
Decision Analysis
International Series in Operations Research
& Management Science
Volume 231
Series Editor
Camille C. Price
Stephen F. Austin State University, TX, USA
Associate Series Editor

Joe Zhu
Worcester Polytechnic Institute, MA, USA
Founding Series Editor

Frederick S. Hillier
Stanford University, CA, USA
More information about this series at http://www.springer.com/series/6161

Adiel Teixeira de Almeida
Cristiano Alexandre Virgínio Cavalcante
Multicriteria and
Multiobjective Models
for Risk,
Reliability and Maintenance
Decision Analysis
Adiel Teixeira de Almeida Cristiano Alexandre Virgínio Cavalcante
Universidade Federal de Pernambuco Universidade Federal de Pernambuco
Recife, PE Recife, PE
Brazil Brazil
Marcelo Hazin Alencar Rodrigo José Pires Ferreira

Brazil Brazil
Adiel Teixeira de Almeida-Filho Thalles Vitelli Garcez

Brazil Brazil
ISSN 0884-8289 ISSN 2214-7934 (electronic)

International Series in Operations Research & Management Science
ISBN 978-3-319-17968-1 ISBN 978-3-319-17969-8 (eBook)
DOI 10.1007/978-3-319-17969-8
Library of Congress Control Number: 2015938699
Springer Cham Heidelberg New York Dordrecht London

© Springer International Publishing Switzerland 2015
This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
The publisher, the authors and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the
authors or the editors give a warranty, express or implied, with respect to the material contained herein
or for any errors or omissions that may have been made.
Printed on acid-free paper
Springer International Publishing AG Switzerland is part of Springer Science+Business Media

(www.springer.com)
To our families for their continuous support
Foreword
Any organization is interested in having a structured decision process for its

strategic success. This is particularly relevant when the decision context involves
technological risk, reliability or maintenance issues. In general these issues may
often be associated with potential threats to human life (e.g. safety) and the
environment. They may also affect the strategic results of any organization. All
these matters may be integrated into a single decision problem in many systems,
for example an electrical supply system. Service interruptions or accidents in this
kind of system may affect health and other emergency services, traffic in big
cities, air traffic control, and many other issues that society has become increasingly
aware of as a result of media reports of major accidents that had or could well
have had a very serious impact on human safety. These interruptions are often
related to the decisions in a system involving Risk, Reliability and Maintenance
(RRM). Usually, these decisions include more than one objective that need to be
dealt with simultaneously, with appropriate support from multicriteria and multi-
objective models. These multicriteria models become even more relevant for the
example of electrical supply system with smart grids conception.
“Multicriteria and Multi-objective Models for Risk, Reliability and Maintenance
Decision Analysis” is a book that enables the reader to have a better understanding
of and guidelines on integrating important application areas of operations research
and management science. This is done by discussing means of structured process
for model building that incorporate RRM issues. This integration is based on the
combination of concepts and foundations related to RRM areas within multicriteria
methods.
The authors represent a group of active members of scientific societies in
operations research and RRM areas. They set out to build a bridge between these
areas with this book; They have had more than 20 years’ experience of engaging
on such research and have had many articles published both in journals of
distinction in the areas of operations research and also in specialized journals
related to risk, reliability and maintenance, since the 1990s. Many of these articles
also consider real problems found in business organizations.
As the current IFORS (International Federation of Operational Research
Societies) President, it is with great pleasure that I present a book that reports
outstanding academic research results in operational research and management
science, thus bridging these relevant areas in order to support the decision process
related to issues of the utmost importance to society.
Nelson Maculan Filho

Prof Emeritus of Universidade Federal do Rio de Janeiro
President of IFORS (International Federation of Operational Research Societies)
vii
Preface
Many decision problems have more than one objective that need to be dealt with
simultaneously. Risk, Reliability and Maintenance (RRM) are contexts in which
decision problems with multiple objectives have been on the increase in recent
years. The importance of having a better structured decision process is essential
for the success of any organization. Additionally, decisions on RRM matters may
affect the strategic results of any organization, as well as, human life (e.g. safety)
and the environment.
RRM influences society and organizations in many ways, since companies and
governments must satisfy several expectations related to the everyday lifestyle
inherent in modern society, such as safeguarding the safety of their employees,
their customers and the community they are part of. Such a lifestyle includes new
paradigms for judging what level of risk is acceptable and this requires multi-
dimensional risks to be evaluated in order to meet society’s and regulatory bodies’
expectations. Reliability and maintenance have become more important also,
since such expectations are extended to the demands that services are constantly
available and that products are of a consistently high quality. Therefore,
companies strive to reduce costs and simultaneously improve their performance
with regard to meeting their strategic objectives. These are affected by reliability
and maintenance and include implications for risk, namely that the analysis of risk
and reliability demands a more conservative approach as do maintenance policies
since failures may have serious implications regarding safety and environmental
losses. As a result, MCDM/A approaches are becoming inevitable when modeling
strategic problems that involve the RRM context.
This book integrates multiple criteria concepts and methods for problems
within the RRM context. The concepts and foundations related to RRM are
considered for this integration with multicriteria approaches. In the book, a
general framework for building decision models is presented and this is illustrated
in various chapters by discussing many different decision models related to the
RRM context.
In general, a decision process or problem in the multicriteria context is related
to the acronyms MCDM (Multi-Criteria Decision Making) and MCDA (Multi-
Criteria Decision Aiding; also known as Multi-Criteria Decision Analysis). The
distinctions between these acronyms are not emphasized in this text. Without loss
of generality, the acronym MCDM/A is applied throughout the text to represent a
variety of approaches associated with MCDM and MCDA (decision making,
decision analysis and decision aiding).
The scope of the book is related to ways of how to integrate Applied
Probability and Decision Making. In Applied Probability, this mainly includes:
decision analysis and reliability theory, amongst other topics closely related to risk
analysis and maintenance. In Decision Making, it includes a broad range of topics
ix
x Preface
in MCDM/A. In addition to decision analysis, some of the topics related to

Mathematical Programming area are briefly considered, such as multiobjective
optimization, since methods related to these topics have been applied to the
context of RRM.
The book addresses the needs of two specific audiences and these include
practitioners and researchers of both areas:
x Those dealing with Risk analysis, Reliability and Maintenance areas, who are
interested in using multicriteria decision methods;
x Those related to multiobjective and MCDM/A, who are interested in making
applications in the contexts of RRM.
Those, who are dealing with decision problems related to the RRM context, in
general need to improve their knowledge of multiobjective and multicriteria
methods so they can build more appropriate decision models. Also, those dealing
with multiobjective and multicriteria decision making area, require to improve
their knowledge of the concepts and methods related to the contexts of RRM, so
that they can approach decision problems on RRM in a more appropriate way.
The book addresses an innovative treatment for the decision making in RRM,
thereby improving the integration of fundamental concepts from the areas of both
RRM and decision making. This is accomplished by presenting an overview of the
literature on decision making in RRM. Some pitfalls of decision models when
applying them to RRM in practice are discussed and guidance on overcoming
these drawbacks is offered. The procedure enables multicriteria models to be built
for the RRM context, including guidance on choosing an appropriate multicriteria
method for a particular problem faced in the RRM context. The book also includes
many research advances in these topics. Most of the multicriteria decision models
that are described are specific applications that have been influenced by this
research and the advances in this field.
The book is not strictly for research and reference by researchers and
practitioners. It has potential for use as an advanced textbook for one of the three
topics: reliability, maintenance and risk management. That is, it could usefully
complement a basic textbook on one of those topics.
The book is implicitly structured in three parts, with 12 chapters. The first part
deals with MCDM/A concepts methods and decision processes (Chaps. 1 and 2).
The second part corresponds to Chap. 3, in which the main concepts and
foundations of RRM are presented. Then, comes the third part, which forms the
greatest section of the book (Chap 4 to Chap. 12) and deals with specific decision
problems in the RRM context approached with MCDM/A models.
Chap. 1 gives a first view on decision problems with multiple objectives, with a
description of the basic elements needed to build decision models. This Chapter is
directly integrated with Chap. 2, which focuses on the decision process and
MCDM/A methods. Although the description and concepts are given in a general
sense, they are focused on the main problems and situations found in the context
Preface xi
that this book explores: risk, reliability and maintenance, although they can be
applied to any other context. Therefore, an explanation is given as to why and how
MCDM/A arises in the RRM context.
Chap. 2 deals with MCDM/A methods and the decision process. A procedure
for building an MCDM/A decision model is presented. Some concerns on the
choice of MCDM/A methods are presented, discussing the compensatory and non-
compensatory approaches. Although this procedure may be applied to any context,
some particular considerations are given to the RRM one. A few MCDM/A
methods are presented, the focus being on deterministic additive methods (MAVT)
and methods for aggregation in probabilistic context, with a focus on MAUT.
Outranking methods are also presented, with some emphasis to ELECTRE and
PROMETHEE methods.
Chap. 3 presents concepts of RRM. These concepts should be considered when
building RRM decision models in order to indicate procedures and techniques that
can be used to calculate and estimate consequences. This allows aspects related to
the state of nature and particularities of RRM to be incorporated when modeling a
decision problem. Chap. 3 includes techniques for dealing with risk analysis
such as the HAZOP, FMEA, FTA, ETA, QRA and ALARP principle; cost effective-
ness; and risk visualization. Reliability and maintenance aspects presented
in Chap. 3 include random failure modeling, reliability and failure functions,
maintenance and reliability interactions, FMEA/FMECA, redundant systems,
repairable and non-repairable systems, maintenance goals and maintenance
management techniques (TPM, RCM). Additionally Chap. 3 presents techniques
for eliciting expert’s prior knowledge.
Chaps. 4 to 12 present an integration of the first and second part when con-
sidering RRM decision problems structured within an MCDM/A approach, for
which formulation and insights for decision problems are given. Chap. 4 presents
a multidimensional risk analysis perspective by introducing a general structure for
building a multidimensional risk analysis decision model. Based on the structure
provided, Chap. 4 presents examples of multidimensional risk evaluation models
for natural gas pipelines and an underground electricity distribution system. Other
contexts are discussed, the purpose of which is to offer insights on how to evaluate
multidimensional risks, such as in power electricity systems, for natural hazards,
counter-terrorism and nuclear power.
Preventive maintenance decisions are presented in Chap. 5 with regard to how
to go about selecting which is the most suitable time interval for scheduling
preventive maintenance actions. This chapter explores the classical optimization
approach for preventive maintenance modeling and gives insights on the
implications of considering an MCDM/A approach by discussing illustrative
applications of two kinds of MCDM/A approaches based on the general procedure
for building MCDM/A models presented in Chap. 2.
xii Preface
Condition-based maintenance (CBM) is tackled in Chap. 6, including a

discussion of MCDM/A models in CBM. An MCDM/A model is presented
including delay time concepts followed by a case study conducted in a power
distribution company, thereby illustrating the advantages of considering an
MCDM/A perspective.
Chap. 7 presents maintenance outsourcing decisions regarding supplier and
contract selection. Throughout this chapter, several criteria for such problems are
discussed and five MCDM/A decision models are presented.
Spare part planning models are discussed in Chap. 8. General aspects of
approaches to sizing spare parts are presented which gives insights into how an
MCDM/A model considers the state of nature over reliability and maintainability,
based on the probability of stockout and cost. Another MCDM/A decision model
grounded on the same objectives is presented for sizing the need for multiple spare
parts for which the case study uses a multiobjective genetic algorithm.
Additionally, a spare parts model integrated with CBM is shown.
The allocation of redundancy is discussed in Chap. 9, and takes the combinatorial
complexity of these problems into account. Therefore, multiobjective formulations
for these problems, found in the literature are presented and the tradeoffs in
redundancy allocation are emphasized. An MCDM/A model is presented for a
standby system in the context of a telecommunications system of an electric power
company with a 2-unit standby redundant system. The model takes interruption
time and cost into account.
Design selection decisions are explored in Chap. 10 with a discussion on the
roles of reliability, maintainability and risk in system design. Based on these
aspects, this chapter includes an MCDM/A model for selecting the design of a car
and an MCDM/A model for risk evaluation in design selection and gives
illustrative applications.
Chap. 11 consists of MCDM/A models for priority assignment in maintenance
planning. An MCDM/A model is presented within the RCM structure to establish
critical failure modes considering a multidimensional perspective and this is
followed by an illustrative example. The second MCDM/A model presented in
this chapter considers the problem of identifying critical devices in an industrial
plant. TPM aspects are also mentioned in this chapter and briefly discussed in
order to emphasize potential MCDM/A problems that may be addressed.
Chap. 12 presents other RRM decision problems including the location of
backup transformers, sequencing of maintenance activities, evaluating the risk of
natural disasters, reliability in power systems, integrated production and maintenance
scheduling, maintenance team sizing and reliability acceptance testing.
Depending on the reader’s background and experience regarding MCDM/A
and RRM concepts, a thorough understanding of the first and second parts of the
book, respectively, may be required in order to understand the decision models
presented in the third part (Chaps. 4 to 12). Otherwise, the reader may dip into
Part 3 directly and choose to read any Chapter (Chaps. 4 to 12) without having read
Preface xiii
the first three Chapters. However, Chap. 2 is required, if the reader wants to use
the procedure for building an MCDM/A decision model even though the reader
has good knowledge of MCDM/A concepts.
We would like to thank our colleagues, students and professionals from
industry, who jointly worked with us on modeling MCDM/A problems in the
RRM context, integrated to the Center for Decision Systems and Information
Development (CDSID). In addition, we are grateful to our sponsors (especially
CNPq - the Brazilian Research Council) and the business organizations that have
supported our research and activities since the 1990s. We would also like to thank
the editors of Springer for their professional help and cooperation, and finally, but
most of all, our families, who constantly supported and encouraged us in our
research work.
Recife, Adiel Teixeira de Almeida

February, 2015 Cristiano Alexandre Virgínio Cavalcante
Contents
Foreword ............................................................................................................ vii

Preface................................................................................................................ ix
Acronyms ........................................................................................................... xxiii
Chapter 1 Multiobjective and Multicriteria Problems

and Decision Models ................................................................... 1
1.1 Introduction .......................................................................................... 1
1.2 Multiobjective and Multicriteria Approaches ...................................... 3
1.3 Decision Models and Methods ............................................................. 4
1.4 Decision Process ................................................................................... 5
1.5 Basic Elements and Concepts of Multiobjective and Multicriteria
Problems ............................................................................................... 8
1.5.1 Basic Ingredients and Related Concepts .................................. 8
1.5.2 Preference Structures ............................................................... 10
1.5.3 Intra-Criterion Evaluation ........................................................ 12
1.5.4 Inter-Criteria Evaluation .......................................................... 13
1.6 Decision Approaches and Classification of MCDM/A Methods ......... 14
1.6.1 Decision Approaches ............................................................... 14
1.6.2 Classification of MCDM/A Methods ...................................... 15
1.6.3 Compensatory and Non-Compensatory Rationality ................ 16
1.7 MCDM/A Models in the Context of Risk, Reliability
and Maintenance................................................................................... 18
1.7.1 Peculiarities of Service Producing Systems
for MCDM/A Models .............................................................. 19
1.7.2 Peculiarities of Goods Producing Systems
for MCDM/A Models .............................................................. 20
1.7.3 Models for RRM Contexts with no Preference
Structure ................................................................................... 20
References...................................................................................................... 21
Chapter 2 Multiobjective and Multicriteria Decision Processes

and Methods ................................................................................ 23
2.1 Introduction .......................................................................................... 23
2.2 Building MCDM/A Models ................................................................. 24
2.3 A Procedure for Resolving Problems and Building
Multicriteria Models ............................................................................. 28
2.3.1 Step 1 - Characterizing the DM and Other Actors................... 30
2.3.2 Step 2 - Identifying Objectives ................................................ 30
2.3.3 Step 3 - Establishing Criteria ................................................... 31
2.3.4 Step 4 - Establishing the Set of Actions and Problematic ....... 34
xv
xvi Contents
2.3.5Step 5 - Identifying the State of Nature ................................... 35

2.3.6Step 6 - Preference Modeling .................................................. 36
2.3.7Step 7 - Conducting an Intra-Criterion Evaluation .................. 38
2.3.8Step 8 - Conducting an Inter-Criteria Evaluation .................... 40
2.3.9Step 9 - Evaluating Alternatives .............................................. 41
2.3.10Step 10 - Conducting a Sensitivity Analysis ........................... 41
2.3.11Step 11 - Drawing up Recommendations ................................ 44
2.3.12Step 12 - Implementing actions ............................................... 45
2.3.13The Issue of Scales and Normalization of Criteria .................. 47
2.3.14Other Issues for Building MCDM/A Models .......................... 50
Psychological Traps ................................................................. 51
The Choice of the MCDM/A Method ..................................... 51
The Intelligence Stage of Simon in the Procedure
for Building Models................................................................. 53
2.3.15 Insights for Building MCDM/A Models
in the RRM Context ................................................................. 54
MCDM/A Models in the Risk Context ................................... 55
Interpretation of an MCDM/A Model or Utility
Function Scores ....................................................................... 55
Paradoxes and Behavioral Concerns Related
to Risk Evaluation ................................................................... 57
2.4 Multicriteria Decision Methods ............................................................ 57
2.4.1 Deterministic Additive Aggregation Methods ......................... 58
Properties for the Additive Model ........................................... 58
Elicitation Procedures for Scale Constants ............................. 60
Avoiding Misinterpretations Regarding
the Scale Constants .................................................................. 61
Some MAVT Additive MCDM/A Methods ........................... 62
Additive-Veto Model .............................................................. 63
Additive Models for the Portfolio Problematic ....................... 63
Methods Based on Partial Information for Elicitation
of Weights ............................................................................... 64
2.4.2 MAUT ...................................................................................... 65
Consequence Space ................................................................. 66
Elicitation of the Conditional Utility Function ....................... 67
Elicitation of the MAU Function ............................................ 68
The Utility Independence Condition ....................................... 68
The Additive Independence Condition ................................... 69
Elicitation of the Scale Constants............................................ 70
Rank-Dependent Utility and Prospective Theory ................... 70
2.4.3 Outranking Methods ................................................................ 70
ELECTRE Methods ................................................................ 72
PROMETHEE Methods .......................................................... 73
PROMETHEE V for Portfolio Problematic ............................ 75
Contents xvii
2.4.5 Other MCDM/A Methods ........................................................ 76

Rough Sets ............................................................................... 76
2.4.6 Mathematical Programming Methods ..................................... 77
2.5 Multiobjective Optimization ................................................................ 77
2.6 Group Decision and Negotiation .......................................................... 79
2.6.1 Aggregation of DMs’ Preferences or Experts’
Knowledge ............................................................................... 80
2.6.2 Types of Group Decision Aggregations .................................. 81
References...................................................................................................... 83
Chapter 3 Basic Concepts on Risk Analysis, Reliability

and Maintenance ........................................................................ 89
3.1 Basic Concepts on Risk Analysis ......................................................... 89
3.1.1 Risk Context............................................................................. 90
3.1.2 Public Perception of Risk......................................................... 92
3.1.3 Risk Characterization ............................................................... 93
3.1.4 Hazard Identification ............................................................... 95
3.1.4.1 FMEA (Failure Mode and Effects Analysis) ........... 95
3.1.4.2 HAZOP (Hazard and Operability Study) ................. 95
3.1.5 FTA (Fault Tree Analysis) ....................................................... 96
3.1.6 Event Tree Analysis (ETA) ..................................................... 98
3.1.7 Quantitative Risk Analysis ...................................................... 101
3.1.8 ALARP..................................................................................... 105
3.1.9 Cost-Effective Approach to Safety .......................................... 109
3.1.10 Risk Visualization .................................................................... 111
3.2 Basic Concepts on Reliability .............................................................. 115
3.2.1 Reliability Perspectives ............................................................ 116
3.2.2 Reliability as a Measure of Performance ................................. 118
3.2.3 Reliability and the Failure Rate Function ................................ 118
3.2.4 Modeling Random Failure ....................................................... 120
3.2.5 Models of Failure Rate Function Dependent
on the Time .............................................................................. 121
3.2.5.1 The Weibull Distribution .......................................... 122
3.2.5.2 Log-Normal Distribution .......................................... 124
3.2.6 Influence of Reliability in Maintenance Activities .................. 125
3.2.7 FMEA....................................................................................... 126
3.2.8 Reliability Management ........................................................... 127
3.2.9 Simulation ................................................................................ 128
3.2.10 Redundant Systems .................................................................. 129
3.2.11 Repairable and Non-Repairable Systems ................................ 130
3.3 Basic Concepts on Maintenance ........................................................... 131
3.3.1 Characteristics of the Maintenance Function........................... 131
3.3.2 Production System and Maintenance /Basic Concepts
on Maintenance ........................................................................ 132
xviii Contents
3.3.3 What is Maintenance Management? ........................................ 133

3.3.4 Do the Functions of Maintenance Activities Depend
on the System? ......................................................................... 134
3.3.5 What are, in Fact, the Objectives of Maintenance? ................. 134
3.3.6 The Aspects that Highlight the Importance
of Maintenance......................................................................... 135
3.3.7 Maintenance Policies ............................................................... 136
3.3.8 Structure of a Decision Problem in Maintenance .................... 140
3.3.8.1 Decision Problems on Maintenance Planning .......... 141
3.3.9 Main Techniques for Maintenance Management .................... 142
3.3.9.1 Total Productive Maintenance (TPM) ...................... 143
3.3.9.2 Reliability Centered Maintenance (RCM) ................ 146
3.4 Prior Knowledge of Experts in Risk, Reliability
and maintenance ................................................................................... 149
3.4.1 Elicitation of Expert’s Knowledge .......................................... 151
3.4.2 Equiprobable Intervals Method................................................ 152
3.4.3 Experts’ Knowledge Aggregation ........................................... 153
References...................................................................................................... 155
Chapter 4 Multidimensional Risk Analysis................................................ 161

4.1 Justifying the Use of the Multidimensional Risk ................................. 161
4.2 Multidimensional Risk Evaluation Model ........................................... 167
4.2.1 Contextualizing the System ..................................................... 170
4.2.2 Identifying the Decision Maker ............................................... 170
4.2.3 Identifying Hazard Scenarios................................................... 171
4.2.4 Defining and Selecting Alternatives ........................................ 172
4.2.5 Estimating the Probability of Accident Scenarios ................... 173
4.2.6 Analysis of Objects Exposed to Impacts ................................. 173
4.2.7 Estimating the Set of Payoffs................................................... 174
4.2.8 Eliciting the MAU Function .................................................... 174
4.2.9 Computing the Probability Functions of Consequences .......... 178
4.2.10 Estimating Multidimensional Risk Measures .......................... 179
4.3 Risk Decision Models ........................................................................... 181
4.3.1 Risk Evaluation in Natural Gas Pipelines Based
on MAUT ................................................................................. 181
4.3.2 Multidimensional Risk Evaluation in Underground
Electricity Distribution System ................................................ 186
on ELECTRE Method and Utility Function ............................ 190
4.4 Other MCDM/A Applications on Multidimensional Risk ................... 197
4.4.1 Power Electricity Systems ....................................................... 197
4.4.2 Natural Hazards ....................................................................... 200
4.4.3 Risk Analysis on Counter-Terrorism ....................................... 203
Contents xix
4.4.4 Nuclear Power .......................................................................... 204

4.4.5 Risk Analysis on Other Contexts ............................................. 205
References...................................................................................................... 208
Chapter 5 Preventive Maintenance Decisions............................................ 215

5.1 Introduction .......................................................................................... 215
5.2 A General MCDM/A Model for Preventive Maintenance................... 216
5.2.1 Classical Optimization Problem of Preventive
Maintenance ............................................................................. 217
5.2.2 MCDM/A Framework for the General Model
for Preventive Maintenance ..................................................... 220
Identifying Objectives and Criteria ......................................... 221
Establishing a Set of Actions and a Problematic..................... 222
Identifying State of Nature ...................................................... 222
Preference Modeling ............................................................... 223
Intra-Criterion Evaluation ....................................................... 223
Inter-Criteria Evaluation.......................................................... 224
Evaluating Alternatives and Sensitivity Analysis ................... 224
Elaborating Recommendation ................................................. 224
5.3 Compensatory MCDM/A Model for Preventive Maintenance ............ 224
5.3.1 The Context, the Set of Alternatives and the Criteria .............. 225
5.3.2 Preference Modeling and Intra-Criteria
and Inter-Criteria Evaluations .................................................. 226
5.3.3 Results and Discussion ............................................................ 227
5.4 A Non-Compensatory MCDM/A Model for Preventive
Maintenance ................................................................................................ 227
5.4.1 First Application ...................................................................... 228
5.4.2 Second Application .................................................................. 230
References...................................................................................................... 231
Chapter 6 Decision Making in Condition-Based Maintenance ................ 233

6.1 Introduction .......................................................................................... 233
6.2 Monitoring and Inspection Activities ................................................... 235
6.3 Delay Time Models to Support CBM .................................................. 237
6.4 Multicriteria and Multiobjective Models in CBM ............................... 238
6.5 A MCDM/A Model on Condition Monitoring ..................................... 239
6.6 Building an MCDM/A Model on Condition Monitoring
for a Power Distribution Company....................................................... 242
References...................................................................................................... 247
Chapter 7 Decision on Maintenance Outsourcing ..................................... 249

7.1 Introduction .......................................................................................... 249
7.2 Selection of Outsourcing Requirements and Contract
Parameters ............................................................................................ 251
xx Contents
7.3 MCDM/A Maintenance Service Supplier Selection ............................ 257

7.3.1 Maintenance Service Supplier Selection
with Compensatory Preferences .............................................. 258
Deterministic Administrative Time Model ............................. 258
Stochastic Administrative Time Model................................... 260
with Non Compensatory Preferences ...................................... 263
with Non Compensatory Preferences Including
Dependability and Service Quality .......................................... 266
with Preference’s Partial Information ...................................... 268
7.4 Other Approaches for Supplier Selection ............................................. 270
References...................................................................................................... 271
Chapter 8 Spare Parts Planning Decisions................................................. 273

8.1 Introduction .......................................................................................... 273
8.2 Some Sizing Approaches for Spare Parts in Repair ............................. 276
8.2.1 Relevant Factors to Sizing Spare Parts .................................... 276
8.2.2 Approach Based on the Risk of Inventory Shortages .............. 278
8.2.3 Approach Based on the Risk of Inventory Shortages
by using Prior Knowledge ....................................................... 279
8.2.4 Approach under the Cost Constraint ........................................ 280
8.2.5 Use of MCDM/A Model .......................................................... 281
8.3 Multiple Spare Parts Sizing .................................................................. 285
8.3.1 The Mathematical Model ......................................................... 286
8.3.2 Case Study................................................................................ 288
8.4 Spare Parts for CBM............................................................................. 290
References...................................................................................................... 294
Chapter 9 Decision on Redundancy Allocation ......................................... 297

9.1 Introduction .......................................................................................... 297
9.2 An MCDM/A Model for a 2-Unit Redundant Standby System ........... 303
References...................................................................................................... 308
Chapter 10 Design Selection Decisions ....................................................... 311

10.1 Introduction .......................................................................................... 311
10.1.1 The Reliability Role in System Design .................................. 313
10.1.2 The Maintainability Role in System Design .......................... 314
10.1.3 The Risk Role in System Design ............................................ 316
10.2 An MCDM/A Model for the Design Selection for a Car ..................... 317
10.3 Risk Evaluation for Design Selection................................................... 321
10.3.1 Risk Assessment Standards .................................................... 322
Contents xxi
10.3.2MCDM Framework for Risk Evaluation in Design

Problems ................................................................................. 323
10.3.3 Illustrative Example of Risk Evaluation in a Design
Problem ................................................................................... 326
10.4 Redesign Required by Maintenance ..................................................... 331
References...................................................................................................... 332
Chapter 11 Decisions on Priority Assignment for Maintenance

Planning ..................................................................................... 335
11.1 Introduction .......................................................................................... 335
11.2 An MCDM/A Model for the RCM Approach ...................................... 337
11.2.1 Traditional RCM Consequence Evaluation ............................ 337
11.2.2 RCM Based on MCDM/A Approach ..................................... 338
11.2.3 Illustrative Example................................................................ 340
11.3 An MCDM/A Vision for the TPM Approach ...................................... 342
11.4 Modeling a Problem for Identifying Critical Devices .......................... 343
References...................................................................................................... 348
Chapter 12 Other Risk, Reliability and Maintenance Decision

Problems .................................................................................... 351
12.1 Introduction .......................................................................................... 351
12.2 Location of Backup Units in an Electric System ................................. 353
12.3 The Sequencing of Maintenance Activities.......................................... 357
12.4 Natural Disasters................................................................................... 361
12.4.1 An MCDM/A Model that Evaluates the Risk
of Flooding ............................................................................. 366
12.5 Operation Planning of a Power System Network ................................. 369
12.6 Integrated Production and Maintenance Scheduling ............................ 371
12.7 Maintenance Team Sizing .................................................................... 375
12.8 Bayesian Reliability Acceptance Test Based on MCDM/A ................ 379
12.9 Some Multiobjective Optimization Models on Reliability
and Maintenance ................................................................................... 382
12.9.1 Approaches in the 1980s and 1990s ....................................... 382
12.9.2 Approaches in the 2000s and 2010s ....................................... 383
References...................................................................................................... 386
Index .................................................................................................................. 391

Acronyms
AHP Analytic Hierarchy Process

ALARP As Low As Reasonably Practicable
Aneel Brazilian government agency responsible for regulating the
generation of electrical power
ANP Analytic Network Process
BLEVE Boiling Liquid Expanding Vapor Explosion
CB Cost benefit ratio
CBM Condition-based maintenance
CBR Case Based Reasoning
CDR Composite dispatching rule
CDRNRGA Non ranking genetic algorithm with composite dispatching rule
CDRNSGA-II Non dominated sort genetic algorithm with composite
dispatching rule
CRA Comparative Risk Assessment
CSE Concept Safety Evaluation
CVCE Confined Vapor Cloud Explosion
DEA Data Envelopment Analysis
DEC Equivalent to System Average Interruption Duration Index
DM Decision Maker
DSS Decision Support System
DT Delay Time
EC-JRC European Commission - Joint Research Centre
EHS Environmental Health Safety
ELECTRE Elimination Et Choix Traduisant la Réalité
EPDC Electric Power Distribution Company
ET Event Tree
ETA Event Tree Analysis
FAR Fatality Accident Rate
FEC Equivalent to System Average Interruption Frequency Index
FFA Functional failure analysis
FMEA Failure Modes and Effects Analysis
FMECA Failure modes, effects, and criticality analysis
FT Fault Tree
FTA Fault Tree Analysis
GA Genetic Algorithm
GD Group Decision Making
GDN Group Decision and Negotiation
GIS Geographic Information System
GIT Geo Information Technology
GPSIA Genetic Pareto set identification algorithm
HAZID Hazard Identification
xxiii
xxiv Acronyms
HAZOP Hazard and Operability Study

I Indifference relation of preference
IEC International Electrotechnical Commission
IEEE Institute of Electrical and Electronics Engineers
ISO International Organization for Standardization
J Incomparability relation of preference
JIPM Japan Institute of Plant Maintenance
LPP Linear Programming Problems
M/M/s A system where arrivals form a single queue, there are s servers
and job service times are exponentially distributed
MACBETH Measuring Attractiveness by a Categorical Based Evaluation
Technique
MAU Multi-Attribute Utility
MAUT Multi Attribute Utility Theory
MAVT Multi-Attribute Value Theory
MCDA Multi-Criteria Decision Aiding; may also be applied to Multi-
Criteria Decision Analysis
MCDM Multi-Criteria Decision Making
MCDM/A Indiscriminately applied to MCDM or MCDA
MOCBA Multiobjective Computing Budget Allocation
MOEA Multiobjective Evolutionary Algorithm
MOGA Multiobjective Genetic Algorithm
MOLP Multi-Objective Linear Problems
MOPSO Multiobjective Particle Swarm Optimization
MTBF Mean Time Between Failures
MTTF Mean Time to Failure
MTTR Mean Time to Repair
Natech REPRESENTS a simultaneous occurrence of a natural disaster event
and a technological accident, both requiring simultaneous
response efforts
NCAP New Car Assessment Program
NORSOK NORSOK standards developed by the Norwegian petroleum
industry
NPD Norwegian Petroleum Directorate
NPRD Non-electronic Parts Reliability Data
NRGA Non ranking Genetic Algorithm
NSGA-II Non dominated Sort Genetic Algorithm
OEE Overall Equipment Effectiveness
OREDA Offshore Reliability Data Handbook
P Strict Preference Relation
PHA Preliminary Hazard Analysis
PHM Proportional Hazards Modelling
PM Preventive Maintenance
PRA Probabilistic Risk Assessment
PROMETHEE Preference Ranking Organization Method for Enrichment
Evaluation
Acronyms xxv
PSA Probabilistic Safety Assessment

PSM Problem Structuring Methods
PSO Particle Swarm Optimization
Q Week Preference Relation
QRA Quantitative Risk Analysis
RCM Reliability Centered Maintenance
RDU Rank-Dependent Utility
ROC Rank Order Centroid
RPN Risk Priority Number
RRM Risk, Reliability and Maintenance
RUL Residual Useful Life
S Outranking Relation
SAFOP Safety and Operability Study
SAIDI System Average Interruption Duration Index
SAIFI System Average Interruption Frequency Index
SEMOPS Sequential Multiple-Objective Problem-Solving Technique
SJA Safe Job Analysis
SMART Simple Multi-Attribute Rating Technique
SMARTER Simple Multi-Attribute Rating Technique Exploiting Ranks
SMARTS Simple Multi-Attribute Rating Technique with Swing
SPEA2 Strength Pareto Evolutionary Algorithm
TPM Total Productive Maintenance
TTR Time To Repair
VCE Vapor Cloud Explosion
VIP Variable Interdependent Parameters
VTTF Variance of Time to Failure
Chapter 1
Multiobjective and Multicriteria Problems
and Decision Models
Abstract: The decision-making process for any organization may be a key factor
for its success. Many decision problems have more than one objective that need to
be dealt with simultaneously. This chapter introduces decision problems with
multiple objectives, with a description of the basic elements needed to build
decision models and focuses on multicriteria methods (MCDM; MCDA;
MCDM/A), in which the DM’s preference structure is considered. An overview
for classification of MCDM/A methods is given, including a discussion on the
DM’s compensatory and non-compensatory rationality and on multi-objective and
multicriteria approaches. The concepts and basic elements of MCDM/A methods
are presented, including preference structures in a multi-attribute context, and
intra-criterion and inter-criteria evaluation. The basic elements of a decision pro-
cess for building decision models and the actors in this process are also presented.
Differences between the descriptive, normative, prescriptive and constructivism
decision approaches are discussed, considering the decision process. Although
these concepts are presented in a general sense, this description deals mainly with
the main context that this book explores: Risk, Reliability and Maintenance
(RRM). Decision problems in a RRM context may affect the strategic results of
any organization, as well as, human life (e.g. safety) and the environment. There-
fore, an explanation is given as to why and how a MCDM/A arises in the RRM
context. In particular, some peculiarities of service producing systems for MCDM/
A models are presented, as well as for goods producing systems.
1.1 Introduction
In order to choose an alternative, from a set of possible alternatives, in a classical

optimization problem, there is an objective function to be maximized or
minimized, whether this function represents gains or losses, respectively. In a
multiobjective or multicriteria problem, there is more than one objective to be
dealt with. In many situations these objectives may be conflicting. These
objectives are associated with the possible consequences (or outcomes) that will
result from choosing an alternative. Therefore, these problems have more than one
© Springer International Publishing Switzerland 2015 1

A.T. de Almeida et al., Multicriteria and Multiobjective Models for Risk, Reliability
and Maintenance Decision Analysis, International Series in Operations Research
& Management Science 231, DOI 10.1007/978-3-319-17969-8_1
2 Chapter 1 Multiobjective and Multicriteria Problems and Decision Models
objective function to be dealt with simultaneously. In some particular situations,

this means that these objectives are comprehensively optimized. Each objective is
represented by a variable, in which its performance for a given alternative can be
evaluated. This variable may be called a criterion or an attribute, depending on the
multicriteria method used.
The acronyms MCDM (Multi-Criteria Decision Making) and MCDA (Multi-
Criteria Decision Aiding) are applied to indicate a decision process or problem in
the multicriteria context. MCDA may also be found as standing for Multi-Criteria
Decision Analysis. Without loss of generality, the acronym MCDM/A is applied
throughout the text to represent a number of approaches associated with MCDM
and MCDA (decision making, decision analysis and decision aiding).
The perception of a decision process involving a tradeoff amongst several
criteria was put forward since centuries ago.
A text of 1722, by Benjamin Franklin, is regularly quoted to indicate the nature
of a multicriteria evaluation for a specific kind of decision problem, which
consists of only one alternative, with either of two options: implement it or do not.
He expressed this in a letter proposing a decision procedure (Hammond et al.
1998; Hammond et al. 1999; Figueira et al. 2005), as follows:
“In the affair of so much importance to you, wherein you ask my advice, .... [...], my way
is to divide half a sheet of paper by a line into two columns; writing over the one Pro, and
over the other Con. [...] When I have thus got them all together in one view, I endeavor to
estimate their respective weights; and where I find two, one on each side, that seem equal,
I strike them both out. If I find a reason pro equal to some two reasons con, I strike out the
three. If I judge some two reasons con, equal to three reasons pro, I strike out the five; and
thus proceeding I find at length where the balance lies; and if, after a day or two of further
consideration, nothing new that is of importance occurs on either side, I come to a
determination accordingly.”
Benjamin Franklin called this procedure prudential algebra. Much later on an

MCDM/A method, called even swaps, was proposed based on this procedure
(Hammond et al. 1998a; Hammond et al. 1999).
An identical perception of evaluation by tradeoff between two set of criteria for
choosing a course of action, was made around 300 B.C., by Plato, a Greek
philosopher, in the Protagoras dialogue. He proposed putting into the balance the
two previous types of criteria Pro (pleasures) and the Con (pains), as follows:
“I should reply: And do they differ in anything but in pleasure and pain? There can be no
other measure of them. And do you, like a skilful weigher, put into the balance the
pleasures and the pains, and their nearness and distance, and weigh them, and then say
which outweighs the other. If you weigh pleasures against pleasures, you of course take
the more and greater; or if you weigh pains against pains, you take the fewer and the less;
or if pleasures against pains, then you choose that course of action in which the painful is
exceeded by the pleasant, whether the distant by the near or the near by the distant; and
you avoid that course of action in which the pleasant is exceeded by the painful. Would
you not admit, my friends, that this is true? I am confident that they cannot deny this.”
1.2 Multiobjective and Multicriteria Approaches 3
These quotations are just two of many others on situations reported long ago
which offered this insight of making tradeoffs amongst criteria in order to evaluate
alternatives in the decision process. The optimization conception is in these views,
when considering the attempt to find the best action, is obtained by means of
combining several objectives.
Historical views and perspectives for the MCDM/A area may be found in
several texts (Koksalan et al. 2011; Edwards et al. 2007).
In this Chapter a first view is given on decision problems with multiple
objectives, with a description of the basic elements needed for building decision
models. This Chapter is directly integrated with Chap. 2, which focuses on the
decision process, MCDM/A methods and multiobjective approaches. Then, an
emphasis is given to the main problems and situations found in the context that
this book explores: risk, reliability and maintenance (RRM), although they can be
applied to any other context.
1.2 Multiobjective and Multicriteria Approaches
Most of the literature makes a distinction between the terms Multiobjective and
Multicriteria. Therefore, one can say that a problem with multiple objectives can
be approached by using either: MCDM/A method or a multi-objective
optimization approach.
An MCDM/A method considers the preference structure of a decision maker
(DM) and involves value judgment. The DM’s preferences will be incorporated in
the decision model in order to support the choice of the alternative, and by doing
so, the multiple criteria will be analyzed simultaneously.
Multiobjective optimization approaches identify the Pareto frontier, the set of
non-dominated alternatives, from the set of alternatives. An alternative A1 is said
to dominate another alternative A2, if the following conditions hold: i) alternative
A1 is not worse than A2 in all criteria, and ii) alternative A1 is better than A2 in at
least one criterion.
The set of non-dominated alternatives consists of those which are not
dominated by any other of the set of alternatives.
In this approach, the DM’s preferences are not taken into consideration. This
means that a specific final solution is not indicated, since a DM’s preferences are
not incorporated into the model for combining objectives.
On the other hand, using an MCDM/A method, the objectives are combined
based on the DM’s preferences. These preferences consist of the DM’s subjective
evaluation of the criteria. This subjectivity is an inherent part of the problem and
cannot be avoided. Otherwise, it means that the model is related to any other
problem, instead of the real problem faced by the DM. Thus, the methodological
issues for dealing with this subjectivity have been one of the main purposes of
research on MCDM/A.
1.3 Decision Models and Methods
The meaning of models and methods may vary amongst texts. In this text, an
important distinction is made between MCDM/A models and MCDM/A methods,
although slight variations may occur in our discussion because of particular
contexts.
As is well known, a model is a simplification of a real situation and it is
expected to deviate (err) to some extent from the real situation. Therefore, when
building a model there is a conflict between its precision and its simplicity. This
precision is related to how close the model is to the real situation (approximation
of the model).
An MCDM/A model is a formal representation of a real MCDM/A problem
faced by a DM. The MCDM/A model incorporates the DM’s preference structure
and particular issues for a specific decision problem. In general, an MCDM/A
model is built based on an MCDM/A method.
An MCDM/A method consists of a methodological formulation, which can be
applied so as to build specific MCDM/A models. A method may consist of a
theoretical formulation based on a well-defined axiomatic structure.
The MCDM/A method has a more general characteristic and may be applied in
order to build a class of MCDM/A models and may be applicable for a variety of
situations related to preference structures. On the other hand, a decision model
incorporates a preference structure of a specific DM. Some decision models may
be built for a specific and unique problem, while others may be built for a more
general and repetitive decision situation.
The use of the term model may appear to be an exception to the above
concepts, when referring to the ‘additive aggregation model’, which indicates a
group of MCDM/A methods. Here this term is associated with the kind of
mathematical model applied for aggregating the criteria in a particular class of
methods. The additive model for aggregation of criteria will be detailed in Chap. 2,
but it is presented below in (1.1) so as to give a first view of an MCDM/A model.
n
v ( ai ) ¦k V (x )
j 1
j j ij (1.1)
1.4 Decision Process 5
where:
v(ai) is the global value of the alternative ai;
kj is the parameter related to inter-criteria evaluation of criterion j; this parameter
is named either as “weight” or “scale constant” of criterion j;
Vj(xij) is the value of consequence for criterion j;
xij is the consequence or outcome of alternative i for criterion j.
1.4 Decision Process
A model for decision process is given by Simon (1960), and consists of three
stages. This model has been adapted, including the addition of new stages, by a
number of posterior contributions, most of them from the area of information
management and decision systems (Bidgoli 1989; Sprague and Watson 1989;
Davis and Olson 1985; Thierauf 1982; Polmerol and Barba-Romero 2000).
Fig. 1.1 shows this updated model. Stages 1 to 3 are in the initial model
proposed by Simon (1960) and consist of Intelligence, Design and Choice. Stages
4 and 5 were added later and are related to revising and implementing the decision
process.
The intelligence stage sets out to search for decision situations, by monitoring
the organization and its environment. This is not a conventional stage for most of
the operational research procedures. In some ways, this stage is related to the view
on structuring a problem given by Keeney (1992) with the Value Focusing Thinking
(VFT) approach, with particular regard to identifying a decision situation. This
stage is also correlated to the vision of strategic management, in which con-
tinuous monitoring and diagnosis of the organization and its environment has to
be done in order to anticipate decision situations in a proactive way (de Almeida
2013).
Conventionally, most operational research procedures consider that there is
already a decision problem to be faced and defining the problem is already part of
working towards finding a solution (Ackoff and Sasieni 1968). Therefore, in most
cases, the decision process starts with the second stage, that of design. This happens
in general in most contexts, especially in the RRM context. However, even in the
RRM context, the organization may derive great benefit by introducing a more
strategic view for dealing with its decision process regarding risk management and
maintenance. For instance, an inadequate maintenance model may affect the
competitive position of any organization, when its clients are adversely affected
by the effects of the unreliability of its products (goods or services).
Searching for decision situations, by

Intelligence monitoring the organization and its
environment
Design Building the decision model
Evaluating the set of alternatives and

Choice producing a recommendation
Revising the previous stages and

Revision introducing a learning process.
Implementation of solution
Implementation
Fig. 1.1 Decision Process
The main focus of the design stage is on building the decision model. This
stage includes generating alternatives and other ingredients of the decision model.
In this stage the feasibility of alternatives are evaluated. Problem Structuring
Methods (PSM) are very useful in this stage in order to ensure that the problem is
clearly defined (Rosenhead and Mingers 2004; Eden 1988; Eden and Ackermann
2004). The mathematical model is worked out in this stage and the parameters of
the model are estimated. The DM has an important role in this stage, with
particular regard to information given through his/her preferences. Also, it is in the
design stage that the MCDM/A method is chosen.
Therefore, this stage has a basic role in the decision process and the model
designed has to be seen to guarantee that it is related as closely as possible to the
real problem faced. As mentioned a model is an approximation of a real situation.
There is a provocative aphorism about models related to this issue: “All models
are wrong but some are useful” (Box 1979). In other words, the aphorism is saying
that all models are approximations of the real situation. In the practical context of
building models the following recommendation is relevant: “Remember that all
models are wrong; the practical question is: how wrong do they have to be to not
be useful?” (Box and Draper 1987).
1.4 Decision Process 7
In the choice stage the alternatives are evaluated according to the model built in
order to produce a recommendation to the DM. The form of this recommendation
depends on the problematic (Roy 1996), which may be, for instance, a selection of
one of the alternative, ranking all alternatives, etc.
Before the recommendation is presented to the DM a revision stage is
conducted, in order to evaluate the assumptions chosen and results obtained in
earlier stages, and also to check for any possible inconsistencies. In this stage the
model building process is evaluated, and takes a comprehensive view, before final
confirmation that is given that the model is in an adequate state. Also, this stage
incorporates an organizational learning process. Actually, this revision may be
done at any time during this whole process, which may be based on a new
perception about aspects dealt with in earlier steps (Davis and Olson 1985).
The implementation stage consists of applying the recommendation in the
organization or in its environment. Communicating the recommendation is an
important action in this stage.
In the decision process there are several actors who play different kinds of role
in the decision process. The literature presents a few possible views on who these
actors should be (Roy 1996; Vincke 1992; Belton and Stewart 2002; Figueira
et al. 2005; Polmerol and Barba-Romero 2000), some of whom are considered in
what follows. The decision maker (DM) plays the central role, but may be
influenced by other actors. The other actors may include: an analyst, a client,
experts, and stakeholders.
The decision analyst (most of the time simply referred to as ‘analyst’) gives
methodological support to the DM in all stages of the decision process, and works
on the problem structuring process and building the decision model.
The client is an actor who acts on behalf of the DM and interacts most of the
time with the analyst, as a surrogate of the DM. In general this actor is a senior
assistant of the DM, who is not available in many situations; or at least for many
steps in the decision process. Perhaps this use of the term ‘client’ came into being
as this person was seen as someone who sought the guidance of the analyst, who,
in most cases, is an external consultant.
There are other actors, called stakeholders, who try to influence the DM’s
behavior in order to obtain a satisfactory result, for themselves or those whom
they represent. In general, these stakeholders are affected by the decision that will
be made by the DM.
The expert is an actor who has specialized knowledge of some part of the
system, which is object of the decision process and who gives factual information
to be incorporated within the model (de Almeida 2013). This information may be
based on prior probabilities related to the state of nature, which represents
variables not under the DM’s control. This actor may be relevant for decision
problems in the context of RRM, since this requires many probabilistic issues to
be modeled, such as that done in the Bayesian Decision Theory framework (Raiffa
1968). This kind of actor is rarely mentioned in the MCDM/A literature, but is
often present in the literature on Decision Analysis (or Decision Theory).
1.5 Basic Elements and Concepts of Multiobjective and

Multicriteria Problems
This section briefly introduces basic ingredients and elements related to

multicriteria problems and also relevant concepts that need to be reflected on the
decision process related to MCDM/A.
1.5.1 Basic Ingredients and Related Concepts
The basic ingredients include the consequences and the set of alternatives.
Concepts related to the family of criteria, the consequence matrix and the
problematic are presented below.
A situation is a decision problem if the DM has at least two alternatives, one of
which he/she must choose. The set of alternatives may be continuous or discrete.
In organizations many managerial decision problems have a set of alternatives
consisting of a discrete set of elements ai, available to the DM. This set may be
represented by A = {a1, a2, a3, ..., an}. A continuous set of alternatives may
be found, in several situations, such as in maintenance planning, in which the
alternatives consist of the time interval tp, within which a preventive maintenance
action should be performed.
In some situations, a continuous set of alternatives may be adapted and
presented as a discrete set of alternatives, when this is an adequate approximation
for the problem. For instance, the time interval for preventive maintenance tp, may
be seen as calendar days, such that the set of alternatives becomes A = {d1, d2, d3,
..., dn}. For any organization, this model is more realistic, since there is no
meaning in considering precisely a continuous time tp, including any time of day
or night. Making a choice of any day di is a reasonable approximation for the
context of preventive maintenance, since a variation in 24 hours does not make
a relevant difference in the consequences related to the decision problem.
The concept of problematic is related to the format of recommendation to be
made for the set of alternatives, which is reflected in the algorithm to be applied
and which will produce the desired result. There are a few types of problematic
found in the literature (Roy 1996; Belton and Stewart 2002) and some of those,
considered the most relevant for this text, are presented below:
x Problematic of choice - In this problematic the result consists of a chosen
subset of alternatives, which should be as small as the procedure can make it.
Normally it is desired to have only one alternative chosen, the optimal one.
This is a particular situation of this problematic, called: optimization. If the sub-
set chosen has more than one alternative, such alternatives are considered
incomparable, since the procedure may not be able to find only one alternative.
Whatever is the size of this subset, only one alternative is implemented in the end.
1.5 Basic Elements and Concepts of Multiobjective and Multicriteria Problems 9
x Ranking problematic - In this problematic the alternatives of the set A are

compared and ranked from the best to the worst.
x Sorting problematic - The alternatives of the set A are classified in categories
or classes. These classes are specified in the model building process and have
a certain order of preference.
x Portfolio problematic - In this problematic there is an interest in choosing a
subset of the set A, in accordance with the objectives of the problem and
subject to some constraints. Unlike the choice problematic, in the portfolio
problematic, all alternatives of this subset are implemented in the end. This
kind of problematic may be implemented based on the knapsack procedure.
A typical example of this kind of problematic is the selection of projects
for a portfolio, in which there is a combination of projects from which there is
a global value of outcomes to be obtained and keeps within some constraint,
such as a limit for the budget.
A fundamental ingredient for the model is the set of consequences, which
consists of the outcomes to be obtained by the DM, when making the decision.
These consequences are associated with the objectives. For each objective there is
a set of possible consequences, which may be the result from the decision process.
The alternatives are evaluated by their consequences. In fact, given that this is
the essential aspect of the decision process, the DM does not choose from amongst
the alternatives. The choice is made from amongst the consequences, which are
informed by the DM’s preference structure. Based on this preference information,
the model will choose the alternative that can supply the most desirable
consequence, according to the DM’s preferences.
At this point it is worth recalling an ancient vision regarding the preceding
role of consequences for evaluating alternatives. It was presented by Pericles,
around 430 B.C., in a Funeral Oration (Thucydides, History of the Peloponnesian
War, II, 40):
“We Athenians, in our own persons, take our decisions on policy and submit them to
proper discussions: for we do not think that there is an incompatibility between words and
deeds; the worst thing is to rush into action before the consequences have been properly
debated. And this is another point where we differ from other people. We are capable at
the same time of taking risks and of estimating them beforehand. Others are brave out of
ignorance; and, when they stop to think, they begin to fear. But the man who can most
truly be accounted brave is he who best knows the meaning of what is sweet in life and
what is terrible, and then goes out undeterred to meet what is to come.”
This has been quoted in many texts related to risk management. There are
many decision problems, in which the consequences are presented in a pro-
babilistic way or there is no information on the frequency of occurrence regarding
the elements of the set of consequences. These situations involve decision
problems under risk or under uncertainty.
Given the nature of the multicriteria problem, a vector of consequences is
considered, since each dimension of this vector is related to each criterion.
For each alternative i there is a possible consequence Xij, given the criterion j.
Let us assume that the set of alternatives is discrete, then, a consequence matrix
may be considered as illustrated in Table 1.1. This consequence may be
represented by a deterministic or probabilistic variable. The Table 1.1 assumes the
deterministic case, in which there is a specific outcome Xij, for each combination
of alternative and criterion. There are situations in which the consequence may be
presented in a probabilistic way. For instance, for repair time t, the consequence
may be represented by a probability density function f(t) over t.
Table 1.1 Consequence matrix
A Criterion 1 Criterion 2 Criterion 3 .... Criterion j ... Criterion n

A1 x11 x12 x13 ... ... x1n
A2 x21 x22 x23 ... ... x2n
... ... ... ... ... ... ... ...
ai ... xij ... ...
... ... ... ... ... ... ... ...
am xm1 xm2 xm3 ... ... ... xmn
1.5.2 Preference Structures
The DM’s preferences are evaluated by means of a preference modeling,

considering basic concepts related to preference relations. These preference
relations are binary relations applied to compare the elements of the set of
consequences X = {x1, x2, x3, ..., xo}.
A binary relation R over a set X = {x1, x2, x3, ..., xo} is a subset of the Cartesian
product RxR. Let x and y be elements of X, then a binary relation is a set of
ordered pairs (x,y). This relation is represented by xRy. If the relation R between
two elements (x, y) does not hold this can be represented as not(xRy). Several
properties may be considered for a binary relation R such as:
Reflexive, if xRx.
Symmetry, if xRy Ј yRx.
Asymmetry, if xRy Ј not(yRx).
Transitivity, if xRy and yRz Ј xRz.
In preference modeling, a relation R is commonly called a preference relation.
The main preference relations to be applied in this text are the following:
x Indifference (I) - xIy indicates that the DM is indifferent between the two
elements x and y. Properties applied: reflexive and symmetry.
x Strict Preference (P) - xPy indicates that the DM clearly prefers the x to y.
Property applied: asymmetry.
x Weak Preference (Q) - xQy indicates that there is some doubt if either the DM
clearly prefers the x to y (xPy) or is indifferent between them (xIy), although it
is clear that not(yPx). Property applied: asymmetry.
x Incomparability (J) - xJy indicates that the DM is not able to compare the two
elements. Any of the following situations may apply, but the DM can not
differentiate amongst them: xIy, xPy, yPx. Properties applied: symmetry and
not reflexive (not(xJx)).
A system of preferences or a preference structure is a collection of preference
relations, applied to a set of consequences, such that, the two following conditions
hold:
1. For each pair of elements (x, y) of X, at least one of the preference relations of
the system of preferences is applied to (x, y);
2. For each pair of elements (x, y) of X, if one of those preference relations is
applied, no other may be applied.
Several preference structures are considered for preference modeling studies.
The following preference structures are the ones most applied in practice:
Structure (P,I);
Structure (P,Q,I);
Structure (P,Q,I,J).
Structure (P,I) has a symmetric preference relation (I) and the other relation is
asymmetric. In this structure it is possible to obtain a complete pre-order or a
complete order for the elements of X. In an order there are no ties (no relation I).
A pre-order may have ties (existence of relation I). For a complete order there is
no incomparability. The Structure (P,I) corresponds to the traditional preference
model, with which many MCDM/A methods are associated. For instance, the
additive model for aggregation of criteria, shown in (1.1) is related to this
structure. Let a and b be elements of X, then, the following conditions hold for this
structure:
aPb Ј v(a) > v(b).
aIb Ј v(a) = v(b).
Structure (P,Q,I) has a symmetric preference relation (I) and two asymmetric
relations (P,Q). In this structure it is possible to obtain a complete pre-order for
the elements of X. For this structure, the previous two conditions hold and the
following may be added:
aQb Ј v(a) v(b).
Structure (P,Q,I,J) has the incomparability relation, which leads to partial pre-
orders for the elements of X. This structure is relevant for situations in which the
DM is not able to give full preference information; for instance the DM may not
be able to compare two elements of X. This is not in agreement with one of the
axioms for the model in (1.1), which is the first axiom of Utility Theory, and
states that the DM is able to make a pre-order of all elements of X. This kind of
situation has been pointed out by Roy (1996) and Simon (1955), who emphasizes
that this may be relevant for MCDM/A situations, in which the DM has to face
several dimensions in a multicriteria evaluation.
An evaluation of the DM’s preference structure is essential for choosing an
MCDM/A method and for building an MCDM/A model.
An arbitrary adoption of any preference structure with a convenient relation for
elements of X, such as a complete pre-order or order, with no considerations for
the DM’s preference may be considered anti-ethical. A situation in which the DM
has any doubt about applying the preference relation P is not a justification to
assume the indifference relation I. For instance, if the DM declares that he/she is
not able to distinguish whether xPy or yPx, and the analyst assumes that this
means an indifference relation I between x and y, this may be a distortion in the
process. Actually, a few elicitation procedures, for obtaining the preference infor-
mation from the DM, may induce this kind of distortion. In this situation it should
be considered whether indifference or incomparability relation should be applied.
1.5.3 Intra-Criterion Evaluation
Before considering the evaluation of consequences amongst criteria, an intra-

criterion evaluation should be conducted. That is, the relative value (performance)
according to the DM’s preference over the outcomes for each criterion should be
considered.
Each criterion represents an objective and can be more formally defined as a
function gj over the set of consequences for criterion j. Let us assume a discrete set
of consequences X. This function gj(x) evaluates the performance obtained by any
consequence x, according to the DM’s preference. This function gj (x) may also be
referred to as a value function vj(x), related to the consequence in the criterion j.
As in the previous discussion related to the decision process, in which a choice
amongst consequences is involved rather than amongst alternatives, normally this
value function is defined over the set of consequences. However, in some
situations, this function may be related to the alternatives, such as in (1.1), since
for each alternative there is a consequence as result of which this alternative
receives its value in (1.1). Therefore, for the sake of simplification the value
function may refer to alternatives or consequences, which does mean the concepts
previously presented are violated.
Therefore, assuming a discrete and deterministic set of consequences, based on
the consequences given for each alternative shown in Table 1.1, the value
functions vj(xj) for each criterion j may be obtained and applied over the
consequences of each alternative i, so that a decision matrix may be obtained,
replacing the elements shown in Table 1.1 by vj(xij). This decision matrix is input
for many MCDM/A methods, which include the intra-criterion evaluation.
In an intra-criterion evaluation a linear or a non-linear value function vj(x) may

be obtained. Linear functions are quite common in MCDM/A problems, although
the possibility of non-linear vj(x), should always be considered. For this reason,
the consideration of normalization procedures is usual for intra-criterion
evaluation, since for linear functions, all criteria and outcomes in Table 1.1 should
have the same scale, in order to apply a model such as the additive shown in (1.1).
In general this normalization uses a scale of between 0 and 1.
It is essential to understand the scales of each criterion and the restrictions of
each MCDM/A method, with regard to this issue, since the kind of normalization
may change the properties of the original scale, in which the outcomes are. These
issues are discussed in Chap. 2.
1.5.4 Inter-Criteria Evaluation
Since the intra-criterion information is available, the following step can be that of
evaluating the inter-criteria, in which all criteria are combined in order to have the
global evaluation of all alternatives. For this evaluation an MCDM/A should be
chosen and applied.
A classification of MCDM/A methods is presented in the next section and a
description of a few methods is given in Chap. 2, but first the concept of a family
of criteria has to be accounted for.
A family F of criteria gj(xj) is the set F = {g1(x1), g2(x2), ..., gm(xm)}. The model
building process should work for a consistent family of criteria (Roy 1996), in
which a few properties has to be followed, such as: being capable of representing
all objectives related to the decision problem and avoiding redundancies.
Since, for each criterion j, the value of the consequences gj(xj) can be produced
for all consequences xj, then, the value of alternatives gj(ai) can be obtained for
each alternative ai.
Given the family of criteria, a dominance relation D between two alternatives
a and b is defined, considering all criteria gj. Then, aDb if gj(a)≥gj(b), given all
j = 1, 2, 3, ..., m, and since the inequality is strict (>) for at least one of the criterion j.
The use of the dominance relation could make the use of an MCDM/A method
unnecessary. However, it is very rare for a solution to be found by applying the
dominance relation. Since, in most situations, many alternatives will not be
dominated by others, then, an MCDM/A method is required in order to evaluate
the inter-criteria.
In Chap. 2 a description of a few MCDM/A methods are given and the
following section gives an overview of their possible classifications.
1.6 Decision Approaches and Classification of MCDM/A Methods
There are four basic decision approaches, which represent perspectives for the
decision process, and which are supported by many methods found in the literature.
These methods may be classified and grouped according to their characteristics.
This grouping process enables common features of such methods to be understood
and facilitates the process of choosing them so as to build particular decision
models. Decision approaches on the other hand will give a perspective on the con-
cepts and the organization of systematic knowledge that supports the decision process.
1.6.1 Decision Approaches
The literature differentiates amongst a few decision approaches, which are pointed
out as perspectives for the study on the decision process. The literature on decision
analysis considers three approaches: descriptive, normative and prescriptive (Bell
1988; Edwards et al. 2007). The literature on MCDM/A also considers a fourth
perspective to the decision process: constructivism (Roy and Vanderpooten 1996).
The descriptive approach focuses on describing how people decide in a real
situation, the concern being to describe how the DM makes judgments and choices
in decision making. This approach is developed by the area of behavioral decision
making (Edwards et al. 2007).
The normative approach focuses on rational choice, based on normative
models, sustained by an axiomatic framework that aims to ensure a logical structure
for decision making. The model in (1.1) is an example of such a normative model,
which imposes a specific rational procedure which a DM may follow. The utility
theory also provides a rational decision model for decisions under uncertainty.
The prescriptive approach consists of procedures that use a model from the
normative perspective, and are structured to support a DM in the decision process.
The prescriptive approach may use the results obtained in the descriptive
approach, in order to deal with the limitation of human judgment. The errors and
inconsistencies examined in the area of behavioral decision are studied in order to
build procedures that can address a consistent way of interacting with DMs so as
to build the preference modeling process and prescribe appropriate models.
The constructivism approach (Roy and Vanderpooten 1996) consists of an
iterative process that uses a learning paradigm (Bouyssou et al. 2006), in which an
analyst interacts with the DM with the support of some method, in order to
construct the recommendation for the problem that the DM faces.
Whereas the prescriptive approach assumes that the DM has a well-defined
preference structure (for instance a utility function to be elicited), in the
constructive approach there is an interactive process that aims to help the DM
reach a more thorough understanding of his/her preference structure.
1.6 Decision Approaches and Classification of MCDM/A Methods 15
1.6.2 Classification of MCDM/A Methods
There are many ways of classifying MCDM/A methods. As first mentioned,

MCDM/A methods may be classified according to the action space, which can be
either discrete or continuous. Both are of interest for the kind of decision problem
analyzed in RRM, especially when a discrete set of alternatives is considered.
A common classification given in the literature (Roy 1996; Vincke 1992;
Belton and Stewart 2002; Pardalos et al. 1995) for methods is that in which three
types are considered:
 Unique criterion of synthesis methods
 Outranking methods
 Interactive methods
The unique criterion of synthesis methods are based on a process of an
analytical combination of all criteria in order to produce a global evaluation or
score for all alternatives and for this reason they are said to have a single criterion
(global score) that synthesizes of all the criteria. The additive model in (1.1) is a
common example of this kind of method and is the basis for many deterministic
additive methods, such as AHP, SMARTS, MACBETH. These are methods for a
deterministic set of consequences and may be referred to as Multi-Attribute Value
Theory (Keeney and Raiffa 1976; Vincke 1992; Belton and Stewart 2002), for
which the acronym MAVT is applied. Also, the Multi-Attribute Utility Theory
(Keeney and Raiffa 1976), very well known by its acronym MAUT, is included in
this group. Most of these methods use the preference structure (P,I), and produce a
complete pre-order.
Outranking methods do not use a unique criterion of synthesis, so many of
these methods produce the final recommendation with no scores for alternatives.
These methods uses the preference structure (P,Q,I,J), considering the incompar-
ability relation, and produce a partial pre-order. The main methods in this group
are the ELECTRE and PROMETHEE methods (Roy 1996; Vincke 1992; Belton
and Stewart 2002).
The unique criterion of synthesis methods and the outranking methods are
representative of several discrete MCDM/A methods.
The interactive methods can be associated with discrete or continuous pro-
blems, although in the majority of cases this class of methods includes the Multi-
Objective Linear Problems (MOLP). Pardalos et al. (1995) include mathematical
programming methods as the third group of methods. A fourth group of methods
is included in their classification for disaggregation methods, which consist of
collecting information from the DM on global evaluation of a few alternatives for
posterior inference on the parameters of an aggregation model. In the end, some of
these methods are related to the unique criterion of synthesis methods.
1.6.3 Compensatory and Non-Compensatory Rationality
The methods may be also classified according to their form of compensation for
aggregating the criteria, which may be considered a kind of rationality. In this
case, two rationalities may be considered leading to: compensatory and non-
compensatory methods (Roy 1996; Vincke 1992; Figueira et al. 2005). Bouyssou
(1986) made remarks on the concepts related to compensation and non-
compensation.
A number of methods may be included in the first type, for instance: MAUT
for uncertainty situations and MAVT, such as the deterministic additive methods,
including AHP, SMARTS, MACBETH, among many others, embracing basic
elicitation procedures; for instance: tradeoff and swing methods (Figueira et al.
2005; Keeney and Raiffa 1976). For non-compensatory methods, lexicographical
and outranking methods, such as PROMETHEE and ELECTRE are included in
this group.
A preference relation P is non-compensatory if the preference between two
elements x and y only depends on the subset of criteria in favor of x and y
(Fishburn 1976). Let P(x,y) = {j: xjPjyj}. That is, P(x,y) is the collection of criteria
for which xjPjyj. Then:
P ( x, y ) P ( z , w) ½
® ¾ [ xPy zPw] (1.2)
¯ P( y, x ) P( w, z ) ¿
In this case, it does not matter what the level of the performance of x or y in
each criterion is. The only information necessary is if one is higher or lower than
the other.
That is, what the value is of the performance (vi(xij)), in decision matrix, of an
alternative for a particular criterion is not taken into account. It is enough to know
if the level of performance (vi(aj)) of an alternative is higher of lower than another.
That is, the only information needed is if vj(az) > vi(ay).This would mean that the
performance of az is higher than the performance of ay and az is preferred to ay.
This is the only information required in (1.2).
Conversely, for a compensatory relation P, it is not enough to know if the level
of performance (vi(aj)) of an alternative is greater or less than another for criterion j.
For the compensatory inter-criteria evaluation process, it matters what the value is
of the performance (vi(aj)) for that criterion j, since that amount will be con-
sidered, in the aggregation model, as the opposite of a non-compensatory model.
That is, for a compensatory method the disadvantage of one criterion may be
compensated for by the advantage in another criterion, as can be done in the
additive model in (1.1).
As remarked by Bouyssou (1986), a preference relation is compensatory if
there are tradeoffs amongst criteria and it is non-compensatory otherwise.
1.6 Decision Approaches and Classification of MCDM/A Methods 17
There are many real situations in which the use of a non-compensatory

rationality is found. Many examples may be found in sports and some of them are
in voting systems.
For instance, in a game of volley-ball, the final result depends on the number of
sets a team has won, rather than the total points it gets. The sets represent the
criteria, with the same weight in the inter-criteria evaluation (de Almeida 2013).
Table 1.2 shows an example of volley game between teams A and B. Team A wins
three sets and is considered the winner, since the team B wins only two sets. It
does not matter how many the teams get in each set. The winner of the set gets all
the set value in the process. On the other hand, if a compensatory rationality is
applied, then team B would be the winner, since it wins a total of 104 points,
against the 93 points team A wins.
Table 1.2 A non-compensatory rationality in a volley-ball game
Team A B Wins set

Set 1 25 23 A
Set 2 25 20 A
Set 3 11 25 B
Set 4 17 25 B
Set 5 15 11 A
Total points 93 104
An interesting example is related to students on a course (Munda 2008),

evaluated with grades in a scale from 0 to 10. A student receives grade 4 for
mathematics and could compensate this grade, by obtaining a grade 10 in
language, for instance, and therefore, passes the final evaluation. This is a
compensatory procedure. Otherwise, if the system considers that each student
should have a minimal performance in each subject, thereby not allowing
compensation amongst different subjects, this evaluation system would be a non-
compensatory one.
There is an interesting example in a voting system (de Almeida 2013), which
concerns the presidential election in the United States of America (USA). In that
system, each state has a symbolic weight, which is related to the number of
senators and congress representatives it may have. This is proportional to the
population of the state (there are a few exceptions that do not change the final
result and for the sake of simplification, are not considered here). Then, the
candidate running in the presidential election, who wins the majority of votes in a
given state, keeps all the weight of that state. In other words, such a candidate
wins all the electoral college votes of that state, no matter the number of electoral
college votes that state has. For instance, California is a state with a high weight,
and has 55 electoral college votes. The winner candidate in California gets all the
55 votes for the final process. Therefore, as in the non-compensatory process, and
volley-ball game illustrated in the Table 1.2, it matters only if the candidate has
the majority of votes cast in that state. At the end of the process, the winner
candidate is the one who gets the states, whose votes sum up to the majority of
weights.
In the presidential election of the USA, the states are equivalent to criteria and
the number of votes obtained in each state corresponds to the score for that
criterion. The combination of criteria, with their weights, plays the role described
for the meaning of the weights in an outranking method (Vincke 1992), which are
combined as a coalition of criteria in order to evaluate the best alternative. The
winner is the one who gets the best coalition of criteria, with the greatest
summation of criteria weights.
It is interesting to note that this non-compensatory rationality means that the
presidential election in the USA is a system with a number of v elections, in which
v = number of states.
1.7 MCDM/A Models in the Context of Risk, Reliability and

Maintenance
The context of risk, reliability and maintenance (RRM) are the focus of this book,
although all the concepts and methodological procedures of MCDM/A are
applicable to any context in general. For this reason a few issues regarding RRM
contexts are discussed below.
In a literature review on MCDM/A models in maintenance and reliability (de
Almeida et al. 2015), more than 180 papers published between 1978 and 2013
were found, which had received more than 4,000 citations. In those studies many
different criteria were found for modeling MCDM/A problems. Amongst the most
common are cost, reliability, availability, time, weight, safety and risk.
Two issues are emphasized in this section regarding MCDM/A models in the
RRM contexts:
x What happens when a decision model does not incorporate the DM’s
preferences;
x The need for MCDM/A models for different kinds of producing systems:
services and goods.
The issue related to whether or not incorporate the DM’s preferences within the
decision model is discussed in the last Section.
There are important issues for MCDM/A models in RRM contexts, which are
related to the peculiarities of two different kinds of producing systems: one for
services and the other for goods, which have different frequencies of demand for
MCDM/A models.
Whatever kind of product it may be, this distinction makes a great difference in
the way that maintenance in general (and preventive maintenance in particular) is
linked to the results of a business. For instance, a system that produces services
1.7 MCDM/A Models in the Context of Risk, Reliability and Maintenance 19
has a feature related to simultaneousness (Slack et al. 2010). This means that at the
time the system is producing the product itself, the customer is being served.
Evidently, in such a context, when a failure in the system occurs, maintenance
definitely has a direct and immediate impact on the competitiveness of the
business (Almeida and Souza 2001). Therefore, preventive maintenance planning
becomes a more strategic decision, linked to highest level of the hierarchical
organizational structure. For the mentioned decision context mentioned above, the
consequences are characterized by multiple and less tangible objectives, which
may require support from an MCDM/A model.
1.7.1 Peculiarities of Service Producing Systems for MCDM/

A Models
In service systems, the output is produced while the customer is being served.
That is, the main feature of this system is its simultaneousness (Slack et al. 2010).
Therefore, the perception of the quality of the service is being created as the
client/user is being served, unlike in goods systems, in which the quality is linked
to the characteristics of the product itself.
The objectives in service producing systems endeavor to reduce costs, when
considered as part of a mix with other objectives, such as: availability, reliability
of the system, time during which the system is interrupted and the quality of the
service.
In service systems, the interruption of the system can be immediately
perceived, since this affects its users. There are many examples of this kind of
system: energy, telecommunications, health, transport, and other public services
(security, defense, water supply).
For this kind of system, interruptions can lead to serious consequences.
Actually, the domain of such consequences is not well defined when compared to
the goods producing system. Another issue that has to be considered is related to
the actors involved in the process. In the case of the service system, the number of
people who are affected by the interruption may be huge. Also, the degree of
impact may vary widely per person. Moreover, it is extremely difficult for a
business organization to trace the totality of damage caused by the disruption of
this kind of product, which is a service.
All things considered, it is easier to understand that failures in these systems
are not only restricted to the financial dimension, so it is of paramount importance
to have MCDM/A support, in order to provide the DM with a broader view about
the problem, and to give to him/her the tools that best take into account the
preferential aspects related to this multidimensional consequence space.
Furthermore, there is an increasing share of service products in the goods
systems, so that the output of this kind of system turns out to be a combination of
goods and services.
1.7.2 Peculiarities of Goods Producing Systems for MCDM/

A Models
In systems that produce goods, losses due to machine downtime can be mitigated
by increasing production beyond normal capacity or by taking some action to
avoid downtime being noticed by clients. In general, failure entails production
delays, re-works, inefficiencies, wastages, overtimes, and/or supply storage
problems, which are easily converted into costs. This would make the problem
change from being one that has multiple objectives to one that has the single
objective of minimizing the total costs. That is why most decision models related
to this context are not based on MCDM/A methods.
However, there are situations, in which, even for systems that produces goods,
the decision context requires an MCDM/A model so that subjective issues can be,
for evaluated. There are two main reasons for this:
x These are more strategic decision contexts which are linked to the highest level
of a hierarchical organizational structure.
x Failures in the production system affect human or social issues, such as safety,
and those to do with the environment.
Moreover, one should be concerned when no DM’s preference is incorporated
into the model, in the modeling process, as subsequently explained.
1.7.3 Models for RRM Contexts with no Preference Structure
Although most studies related to decisions in RRM contexts do not incorporate

DM’s preferences, this has been changing over the recent years. The review
mentioned (de Almeida et al. 2015) shows that the increase in the number of
studies and citations regarding MCDM/A models regarding this area is
considerable. However, most studies on the decision process in RRM contexts still
do not consider the DM’s preferences.
Actually, a ‘decision process’ which does not include the DM’s preferences is
one in which no decision is being made. In such a situation, the model has
whichever preference structure the analyst has introduced explicitly or implicitly,
but this is not the DM’s preference. This may be introduced within the model in
many different ways, such as: arbitrarily or by chance.
In the former an arbitrary preference structure is explicitly (or almost that)
incorporated within the model, in general following a decision previously made by
someone else. Otherwise it may incorporate the analyst’s perception of which
would be the most appropriate preference structure for that context.
References 21
In the latter, some preference structure is implicitly incorporated within the

model, at random, during the model building process. The analyst makes assump-
tions for simplifications or just applies what seems to be usual, following standard
procedures, without properly considering the specific decision context.
For instance, in many situations the intra-criterion evaluation is skipped and a
linear value (or utility) function is applied. This usually happens implicitly. That
is, this is not made as an assumption for simplifying the model, in which process
the approximation consequences are evaluated by the analyst and put forward to
the DM. Actually, most models are built in such a way. In these cases, the
characteristics of non-linearity, such as prone or averse behavior regarding risk,
are not incorporated and may lead to a different solution, which is inappropriate.
The model misinforms the actual decision that should be made. That is why it can
be said that there is no decision being made.
References
Ackoff RL, Sasinieni MW (1968) Fundamentals of operations research. John Wiley & Sons,
New York, p 455
Bell DE, Raiffa H, Tversky A (1988) Decision making: Descriptive, normative, and prescriptive
interactions. Cambridge, UK: Cambridge University Press.
Belton V, Stewart TJ (2002) Multiple Criteria Decision Analysis. Kluwer Academic Publishers
Bidgoli H (1989) Decision support systems: principles and practice. West Pub. Co.
Bouyssou D (1986) Some remarks on the notion of compensation in MCDM. Eur J Oper Res
26(1):150–160
Bouyssou D, Marchant T, Pirlot M, Tsoukis A, Vincke P (2006) Evaluation and decision models
with multiple criteria: Stepping stones for the analyst. Springer Science & Business Media
Box GEP (1979) Robustness in the strategy of scientific model building. Robustness Stat. pp
201–236
Box GEP, Draper NR (1987) Empirical model-building and response surfaces. John Wiley & Sons
Brans JP, Vincke Ph (1985) A preference ranking organization method: the Promethee method
for multiple criteria decision making, Manage Sci 31:647–656
Davis CB, Olson MH (1985) Management Information Systems: Conceptual Foundations,
Structure and Development. McGraw-Hill
de Almeida AT (2013) Processo de Decisão nas Organizações: Construindo Modelos de Decisão
Multicritério (Decision Process in Organizaions: Building Multicriteria Decision Models),
São Paulo: Editora Atlas
de Almeida AT, Ferreira RJP, Cavalcante CAV (2015) A review of multicriteria and multi-
objective models in maintenance and reliability problems. IMA Journal of Management
Mathematics 26(3):249–271
de Almeida AT, Souza FMC (2001) Gestão da Manutenção: na Direção da Competitividade
(Maintenance Management: Toward Competitiveness) Editora Universitária da UFPE. Recife
de Almeida AT, Vetschera R (2012) A note on scale transformations in the PROMETHEE V
method. Eur J Oper Res 219(1):198–200
de Almeida AT, Vetschera R, de Almeida J (2014) Scaling Issues in Additive Multicriteria
Portfolio Analysis. In: Dargam F, Hernández JE, Zaraté P, et al. (eds) Decis. Support Syst.
III - Impact Decis. Support Syst. Glob. Environ. SE - 12. Springer International Publishing,
pp 131–140
Eden C (1988) Cognitive mapping. Eur J Oper Res 36:1–13

Eden C, Ackermann F (2004) SODA. The Principles. In: Rosenhead J, Mingers J (eds) Rational
Analysis for a Problematic World Revisited. Second Edition, Chichester: John Wiley & Sons
Ltd.
Figueira J, Greco S, Ehrgott M (eds) (2005) Multiple Criteria Decision Analysis: State of the Art
Surveys. Springer Verlag, Boston, Dordrecht, London
Fishburn PC (1976) Noncompensatory preferences. Synthese 33:393–403
Hammond JS, Keeney RL, Raiffa H (1998) Even swaps: A rational method for making trade-
offs. Harv Bus Rev 76(2):137–150.
Hammond JS, Keeney RL, Raiffa H (1999) Smart choices: A practical guide to making better
decisions. Harvard Business Press
Keeney RL (1992) Value-focused thinking: a path to creative decisionmaking. Harvard
University Press, London
Keeney RL (2002) Common Mistakes in Making Value Trade-Offs. Oper Res 50:935–945
Keeney RL, Raiffa H (1976) Decisions with multiple objectives: Preferences and Value Trade-
Offs. Wiley Series in Probability and Mathematical Statistics. Wiley and Sons, New York
Koksalan M, Wallenius J, Zionts S (2011) Multiple Criteria Decision Making: From Early
History to the 21st Century, World Scientific, New Jersey
Likert R (1932) A technique for the measurement of attitudes. Arch Psychol 22 140:55.
Munda G (2008) Social multi-criteria evaluation for a sustainable economy. Springer, Berlin
Pardalos PM, Siskos Y, Zopounidis C (eds) (1995) Advances in Multicriteria Analysis. Kluwer
Academic Publishers
Polmerol J-C, Barba-Romero S (2000) Multicriterion Decision in Management: Principles and
Practice. Kluwer
Raiffa H (1968) Decision analysis: introductory lectures on choices under uncertainty. Addison-
Wesley, London
Rosenhead J, Mingers J (eds) (2004) Rational Analysis for a Problematic World Revisited.
Second Edition, John Wiley & Sons Ltd.
Roy B (1996) Multicriteria Methodology for Decision Aiding. Springer US
Roy B, SáowiĔski R (2013) Questions guiding the choice of a multicriteria decision aiding
method. EURO J Decis Process 1:69–97
Roy B, Vanderpooten D (1996) The European school of MCDA: Emergence, basic features and
current works. J Multi-Criteria Decis Anal 5:22–38
Simon HA (1955) A Behavioral Model of Rational Choice. Q J Econ 69:99–118.
Simon HA (1960) The New Science of Management Decision. Harper & Row Publishers, Inc,
New York
Simon, HA (1982) Models of Bounded Rationality. MIT Press
Slack N, Chambers S, Johnston R (2010) Operations management. Pearson Education
Sprague Jr RH, Watson HJ (eds) (1989) Decision Support Systems - Putting Theory into
Practice, Prentice-Hall
Thierauf, RJ (1982) Decision support systems for effective planning and control - A case study
approach. Prentice-Hall, Inc., Englewood Cliffs, New Jersey
Vincke P (1992) Multicriteria Decision-Aid. John Wiley & Sons, New York
Chapter 2
Multiobjective and Multicriteria Decision
Processes and Methods
Abstract: An appropriate decision-making process is relevant for the strategic

success of any organization. Most of the decision problems in these organizations
have multiple objectives that have to be dealt with simultaneously. This chapter
gives a brief description of a few multicriteria (MCDM; MCDA; MCDM/A)
methods, including deterministic additive methods (MAVT), Multi-Attribute Utility
Theory (MAUT), connected with Decision Theory, and outranking methods
(ELECTRE and PROMETHEE). Additionally, Group Decision and Negotiation
process is considered. A procedure for building an MCDM/A decision model is
presented, which enables several factors to be incorporated such as: the DM’s
preference structure and experts’ prior knowledge regarding the state of nature.
The choice of the method is considered. Some concerns related to choosing an
appropriate MCDM/A method are presented, including preference modeling with
the evaluation of the DM’s compensatory and non-compensatory rationality. This
procedure enables an MCDM/A problem to be solved. Several issues concerning
the implementation of this procedure are presented, such as: setting scales and
normalizing criteria, time management in the scheduling of the decision process
(including the procrastination process), and incorporating the intelligence stage of
Simon’s model into the procedure. Although this procedure may be applied to any
context, some particular considerations are given to those of Risk, Reliability and
Maintenance. For instance, a multidimensional risk analysis allows a broader view
and may include the DM’s behavior regarding risk (prone, neutral or averse).
In the reliability and maintenance contexts, the models may include availability,
maintainability, dependability, quality of repair and other aspects besides cost.
2.1 Introduction
In this Chapter two main issues are dealt with. First, considerations are given to
building an MCDM/A model. Then, an overview of MCDM/A methods and
multiobjective optimization approaches are set out.
There are many views for building decision models, since the first propositions
of operational research area. First, some specific issues are emphasized in this
subject in order to establish a basis for the process for building multicriteria

24 Chapter 2 Multiobjective and Multicriteria Decision Processes and Methods
models, which is subsequently presented. Then, regarding the MCDM/A models,

a few concepts and basic issues are presented in order to give a general idea
regarding the main concerns in this topic. Thereafter, a procedure is presented for
dealing with how to tackle resolving MCDM/A decision problems, including the
process for building the associated decision model. Also, a few basic issues related
to the building of MCDM/A models in the RRM context are discussed, with some
practical insights for this process.
The second topic consists of the describing a few MCDM/A methods, found to
be amongst the most relevant for building MCDM/A models for the RRM (Risk,
Reliability and Maintenance) context. There follows an overview of the main
multiobjective optimization approaches, many of which are used in RRM decision
models. Also, an overview of Group Decision and Negotiation (GDN) approaches
is considered, since in some situations there is more than one DM.
2.2 Building MCDM/A Models
In the process for building models the main focus is on simplicity with a view to
finding a degree of approximation that is good enough to make the model useful.
Therefore, when aiming at making a model useful and simple to use several basic
factors have to be borne in mind.
Bouyssou et al. (2006) point out that the use of formal models evokes the
power of hermeneutics, associated with the facility with which a DM’s
preferences can be elicited. They state that the latter depends on the intellectual
and cultural background of the DM. The analyst should be very cautions with
regard to this issue.
On the other hand, the analyst should spend an additional effort in order to
work on the DM’s interpretation difficulties, which are commonly found in
the interactions for preference modeling. Then again, the analyst should avoid the
temptations of choosing easy approaches, that, although keeping away from these
difficulties, deviates the model from the real problem, which it should be
representing, first and foremost (de Almeida 2013a).
Wallenius (1975) states that DMs in general do not trust models, when they
find them to be complex. Considering the observation from Bouyssou et al.
(2006), it may be plausible that this resistance of a DM rather than that associated
with the complexity, it is caused by the DM’s intellectual background for dealing
with the model. A certain complexity of the model may be acceptable, the better
the DM’s intellectual and cultural background are.
Building models is a creative process in nature, which involves intuition and
other spontaneous actions by the analyst, some of them being inspirations driven
in conjunction with the progression of the model (de Almeida 2013a).
2.2 Building MCDM/A Models 25
In spite of the scientific basis of the models, the building process of which may
follow several well-structured steps, as shown in sequence a, in Fig. 2.1, its
creative side does not recommend a rigorously sequential procedure.
Step 1 Step 1
Step 2 Step 2
Step 3 Step 3
Step n Step n
Sequence a Sequence b
Fig. 2.1 Sequence of steps in the decision process
The rigid procedure, in sequence a with consecutive steps, leads to the same
result, if a repetition is made in the process.
A different vision is shown in sequence b of Fig. 2.1, in which the building
process follows a successive refinement procedure (Ackoff and Sasieni 1968). In
this procedure, the analyst can return, at any time, from one step back to any other
previous step, as often as necessary. This return may or not imply the revision of
subsequent steps. This sequence consists of a recursive procedure.
The successive refinement procedure allows any step to be taken in a non
conclusive way, so as to conclude it on returning back, after having a view and
information from subsequent steps. This return makes it possible to enrich the
process with better results for the whole process. Another benefit of this approach
is that the creative modeling process is improved, since this flexibility produces an
environment that is more susceptible to innovative results. In this process the
analyst may get new insights at any time and return to a previous step. In contrast,
the rigid approach of sequence a do not lead the creativity flows for innovation.
Moreover, it should be emphasized that this flexible and creative process does
not hinder the support of the scientific foundations that any model should have.
Also, the process for building models follows basic scientific patterns in order to
avoid misconceptions.
To build decision models there is a strong support from PSM (Rosenhead and
Mingers 2004), whose methods have become vital to understand decision
problems, thus leading to a much closer connection between these problems and
the models. By using PSM, the analyst has adequate support for organizing
information from the actors of the decision process (Franco et al. 2004).
This link between the “real world” and a “model world” is discussed by Keisler
and Noonan (2012). Fig. 2.2 illustrates these ideas, including an adaptation for
considering Simon’s model for a decision process.
Real world Model world
Design the
model
There is a
problem Revision of
model
Action is
implemented Choice
Fig. 2.2 Link between real world and a model world
Figure 2.1 shows that in the “real world”, after recognizing the problem, a
decision process is started, by building the decision model in the “model world”,
which will finally produce the implementation of an action. Comparing this view
with Simons’s model, the stages of design, revision and choice are in the “model
world” and the two other stages are in the “real world”. In this view, there is a
possibility of returning back to reformulate the model, after implementing the
action, since this can still provoke the step of problem recognition.
At this point, it should be observed that the model building process may lead to
many possibilities of models, as illustrated in Fig. 2.3. In Fig. 2.3, at the beginning
of the process, many models are possible. The models are represented by the black
spheres. However, during the modeling process, many modeling decisions are
taken, in which assumptions, choices of approaches and simplifications are
introduced, leading to the elimination (filtering) of some possible models.
2.2 Building MCDM/A Models 27
Funnel in the
modeling process,
selecting the model,
at each assumption Final
taken in the process model
MCDM/A
model
options
Filters of
model
selection
Step a Step b Step c Step d
process for building a model
Fig. 2.3 Funnel in the model building process
The filters in Fig. 2.3 indicate that new assumptions or model definitions are
taken, thus implying the elimination of some possible models, that would prevail
with different assumptions. These modeling decisions also include the preference
modeling information given by the DM. Therefore, during the process parameters
are assigned to the MCDM/A model, thereby reducing the number of alternative
models and leading the process to the final model, as indicated in Fig. 2.3.
A similar illustration with a funnel is given by Slack et al. (1995), for a project
management planning process.
It is interesting to observe that many models may not even be perceived by the
analyst, who eliminates them, by taking directions in the process for building
models. If the analyst has some kind of bias, then, this will be reflected in this
elimination process and perhaps more useful models may not be taken into account.
There are many propositions or general views presented in the literature for
building models in operational research, particularly when using PSM. A few of
these propositions and views have been made for MCDM/A model building
processes.
Roy (1996) presents a view with several stages for building an MCDM/A
model, which includes: establishing the objectives and format of the recommend-
ations; the analysis of consequences and development of criteria; comprehensive
preference modeling and operational aggregation of performances; investigating
and developing the recommendations.
Polmerol and Barba-Romero (2000) propose a few steps for MCDM/A model
building, including: understanding and acceptance of the decision context;
modeling alternatives and criteria; Discussion and model acceptance, refinements
and evaluation of alternatives with a decision matrix; discussion on the choice of
the method, gathering DM information; application of the method; recommend-
ation and sensitivity analysis. They state that this procedure has a linear sequence,
but can be done in a recursive way.
Belton and Stewart (2002) also present their view with the following steps:
identification of the decision problem; problem structuring; model building; use of
model to inform and challenge thinking; developing an action plan.
2.3 A Procedure for Resolving Problems and Building

Multicriteria Models
In this section a procedure for building MCDM/A models is presented, based on

Simons’s model decision process, using the successive refinement procedure for
resolution of MCDM/A problems and the basic ideas presented above.
The procedure for resolution of an MCDM/A problem includes the model
building process, as shown in Fig. 2.4. The full arrows in Fig. 2.4 indicate the
standard sequence to be followed in the process for building models and the
dashed arrows indicate the possibilities of returning to a previous step, as allowed
in the successive refinement process (for the sake of simplification, dashed arrows
are only between two close steps, although the return can be done to any of the
previous steps.
The procedure has three main phases, each one with several steps. The first two
phases are related to the design stage of Simon’s model. First, a preliminary phase
is conducted, in which the main elements of the MCDM/A problem are
approached and PSM may be applied for the problem structuring. The definitions
in this first phase may influence definitively the whole process ahead. In this
phase, many possible models may be eliminated, as illustrated in the filters shown
in Fig. 2.3.
2.3 A Procedure for Resolving Problems and Building Multicriteria Models 29
Preliminary phase
Step 1 Step 2 Step 3 Step 4 Step 5

DM objectives Criteria set of state of
and actions; nature
other problematic
actors
Preference Modeling and method choice
Step 6 Step 7 Step 8

Preference Intra- Inter-
modeling criterion criteria
evaluation evaluation
Finalization
Step 11 Step 12
Step 9 Step 10 Drawing up Implementing
Evaluating Sensitivity recommendations action
alternatives analysis
Fig. 2.4 Procedure for resolving an MCDM/A problem
In the second phase the preference modeling is conducted and the MCDM/A
method is chosen. At the end of this second phase the decision model is ready to
be applied in the third phase, meaning the end of the funnel, illustrated in Fig. 2.3.
The second phase is the most flexible of all of them. In fact the three steps of this
second phase may be done almost at the same time, exploring a richer insightful
process. An already built MCDM/A model is an input to the third phase, although
it still may be changed, due to the possibility of returning to review previous steps
in the successive refinement process.
In the third phase, the choice and implementation stages of Simon’s model are
conducted, for the final resolution of the problem. However, it should be
remembered that it is still possible to return and make revisions and changes in the
built model. In this phase there is a key step of sensitivity analysis, in which this
revision decision is evaluated.
The following sections present details regarding the conception and implement-
ation of each step of this procedure.
2.3.1 Step 1 - Characterizing the DM and Other Actors
In this step is important to describe and typify the DM and other actors in the
decision process. This procedure has an emphasis on decision problems with an
individual DM, although adaptations may easily be conducted in order to
contemplate the situation with a group of DMs.
In this step it has to be clear what the role of the analyst is going to be and the
DM’s participation should be. For instance, the DM may have a more direct or
indirect involvement in the decision process. For the latter, another actor, often
called the “client” may play some important roles in the process and may be very
active in some of the steps of this procedure.
It is relevant to identify how other actors will take part in the process. It is
important to characterize the role of each actor for each of the steps of this
procedure.
Even for a situation of an individual DM, it is the DM who will decide if
decision process may involve many other actors in some steps of the process in
order to collect insights and a broad view regarding some particular issues to be
included in the model. In this case, the analyst may play the role of a facilitator,
who holds meetings with group of actors for a structured discussion of some
issues. In general these meetings are supported by PSM approaches (Rosenhead
and Mingers 2004; Eden 1988; Eden and Ackermann 2004; Ackermann and Eden
2004; Franco et al. 2004).
2.3.2 Step 2 - Identifying Objectives
This step may be considered the most important one, although this can be only
stated in general terms. The most important step for this kind of decision process
depends on the nature of the problem, which demands special attention to one
these steps of the procedure. It may be the intrinsic nature of the problem may
indicate that a particular step has the greatest influence in the quality of the final
decision model. Therefore, only in general terms, it may be stated that this step is the
most important, since the objectives are going to influence every step in this process.
Moreover, the identification of objectives may influence even the process of

establishing the set of alternatives. This may be even more decisive, depending on
the approach applied for creating alternatives. For instance, applying the Value
Focused Thinking (VFT) approach the process for creating alternatives is very
well associated with the structure given for proposing and organizing the
objectives.
Actually, the PSM approaches (Rosenhead and Mingers 2004; Eden 1988;
Eden and Ackermann 2004; Ackermann and Eden 2004; Franco et al. 2004) in
general, amongst which VFT (Keeney 1992), are very useful for conducting this
step.
A clear proposition of objectives may be obtained with the VFT approach, in
which the objectives are characterized by three factors: the decision context, an
object and a preference direction. The objectives are viewed in a hierarchical
structure, including strategic objectives, fundamental objectives, and means
objectives. The determination of a set of objectives in a decision frame is crucial,
since they are the basis of any decision. The insight power of the process is
reduced if the set of objectives are incomplete or vague (Keeney 1992).
2.3.3 Step 3 - Establishing Criteria
For each objective previously established a criterion or attribute has to be

proposed, which will represent those objectives in the decision model. Therefore,
the link between steps 2 and 3 are essential for the representation of the objectives
in the whole decision model.
Keeney (1992) states that the attributes are related to the degree in which their
associated objective is achieved. Therefore, each objective demands a variable in
which this objective can have its degree of performance evaluated. This variable,
usually called a criterion or an attribute, in MCDM/A, may also be called a
measure of effectiveness or measure of performance.
A family of criteria F has to be established with some properties (Roy 1996).
F cannot have redundancy; it must be exhaustive, since all objectives have to be
present and represented by F; and it has to be consistent, in the sense that the
DM’s preferences over the criteria have to be coherent with the global evaluation
of consequences.
A structured view for building attributes or criteria is given by Keeney (1992),
considering three types: natural attributes, constructed attributes and proxy
attributes.
The natural attributes have a common interpretation for all actors in the
decision process, such as the cost, which is presented in monetary units. For the
objective of minimizing human lives, a possible natural attribute is the number of
fatalities per term (annual, for instance). The attributes should be associated with
the decision context and must involve value judgments.
The constructed attributes are applied when it is not possible to use natural
attributes. For instance, an objective that is concerned with improving the image
of a business organization, requires such a type of attribute. Whereas the natural
attributes may be used in any decision context, the constructed attributes are
adequate only for a particular decision context, for which they have been built.
These attributes require the construction of a qualitative scale for evaluation of
the associated objective. These attributes normally are on a discrete scale, which
may be called subjective indices or subjective scales. A Table should be drawn up,
indicating the meaning of each level of this scale, in a clear way (Keeney 1992).
This description should indicate one or several impacts on consequences
associated with each level, and specify the degree of achievement of that
objective. It is quite common to reach a situation in which constructed attributes
are necessary.
If the two previous attributes are not feasible, then a proxy attribute may be
tried. This kind of attribute is an indirect measure of the associated objective. In
general, the proxy attribute of a fundamental objective is the natural attribute of a
mean objective that comprises that fundamental objective.
The criteria should have some properties: measurability, operationality,
understandability (Keeney 1992). Measurability defines the objective with more
details, thereby allowing the value judgment, necessary in the decision process.
An attribute is operational if it describes the possible consequences and provides a
common basis for value judgment, and is thus suitable for the intra-criterion
evaluation. This property has a very close relationship with step 7, in which a
return to this step, for refinement, may be necessary, if the criterion is not properly
operational. Understandability means the attribute may not be ambiguous in the
description of the consequences.
The criterion or attribute may be considered in two ways, regarding its
variability and uncertainty: it may be deterministic or probabilistic. A deterministic
criterion is assumed to have a constant level of performance or fixed outcome.
A probabilistic criterion has a consequence x, which is a random variable and is
specified in terms of its probability density function (pdf): f(x). If a criterion is a
random variable, with a not relevant variability it may be assumed to be
deterministic. In this case, it is assumed that the standard deviation is so small,
that the mean of the variable may represent the consequence x.
For instance, let us consider the time for delivering a product. If the criterion is
assumed to be deterministic, then the establishment of the value function, in step
7, will be that of comparing delivering time such as of 2 or 3 hours, for instance.
Another similar decision context, associated with the maintenance of electricity
supply system, may consider the interruption time (t) of the energy supply. It is
not plausible to assume that this kind of criterion is deterministic, since its
variability is very high, thus it is clearly characterized as a random variable t.
Therefore, the DM has to evaluate this criterion, considering its pdf f(t), since, that
is what the DM gets, as a consequence in the decision process.
Therefore, the evaluation to be conducted on this kind of criterion is related to

comparing alternatives or consequences with different pdfs, as illustrated in Fig. 2.5.
consequence 1
f(t)
consequence 2
interruption time t
Fig. 2.5 Probabilistic consequences
In this case, the DM does not evaluate the difference in preference between 2 or
3 hours of interruption time (t) in the energy supply, since these two consequences
do not really exist. Actually, the comparison would be between the consequences
or alternatives shown in Fig. 2.5. Which of the two pdfs does the DM prefer? f(t1)
or f(t2)? That is, the DM evaluates the difference in preference between f(t1) and
f(t2), shown in Fig. 2.5, related to the interruption time (t) in hours. This may
appear more complicated, at first, although this is actually what the DM gets in the
end in this kind of decision context. Regarding the complexity of the question to
the DM, it should be pointed out that questions put to the DM in the elicitation
procedures are much simpler.
Many problems in the RRM context have this probabilistic characteristic to be
considered. A literature review on maintenance and reliability points out the
nature of MCDM/A models in this context (de Almeida et al. 2015) and the
plausibility of using deterministic representation for criteria, which is discussed in
Sect. 2.3.15.
Thus, the model building process in this step may include a probabilistic
modeling task for this kind of consequence, which goes together with the
preference modeling.
Regarding uncertainty, a criterion or attribute may be found ambiguous in the
representation of its value function, by the DM, and therefore fuzzy numbers
(Pedrycz et al. 2011) could be used to represent them. In this case, a fuzzy
approach may be considered for the decision model, which may influence the
choice of the MCDM/A method. This should be properly evaluated in step 7.
2.3.4 Step 4 - Establishing the Set of Actions and Problematic
This step is related to the set of alternatives for solving the decision problem.
There are four topics to be approached in this step: a) establishing the structure of
the set of alternatives, b) establishing the problematic to be applied to this set,
c) generating the alternatives; and d) establishing the matrix of consequences.
The structure of the set of alternatives has a direct connection with the choice
of the MCDM/A method, since a discrete or continuous set implies completely
different types of methods. For a discrete set of elements ai, A = {a1, a2, a3, ..., an}.
This issue also includes the determination of other features for the set A, which
can be stable or evolutive (Vincke 1992). In the first case, it is known for the
modeling process that the set A is fixed and does not change during the building
process. For the latter, the analyst should be aware of the possibilities of changes
during the decision process, which may represent some kind of constraint.
The set A can be globalized or fragmented. In the former, each element of A
excludes other elements in the resolution process. In contrast, for a fragmented A,
the elements may be combined for the resolution. A portfolio problematic may be
associated to this kind of set. The use of this kind of set is illustrated in Chap. 10.
After establishing the structure of A, then the problematic to be applied to this
set A has to be identified. The problematic may influence the kind of method,
depending on the class of methods to be applied. Some methods may be applied in
more than one problematic; the case of ranking problematic may include the
solution for choice.
After establishment the previous conditions, the generation of the alternatives
can proceed. This is one of the most creative tasks of the whole process.
Analytical insight may be applied in this task, particularly those delineated by the
VFT approach. In this approach the creation of alternatives is based on the value
structure of the objectives. In general PSM can contribute in a considerable way to
this task, and involve a group of experts supported with the guidance of a
facilitator. Depending on the MCDM/A method chosen new alternatives may be
included afterwards, even in an advanced stage of phase three, of finalization.
Some MCDM/A methods assume a fixed set of alternatives and make pairwise
comparisons, for instance. Other MCDM/A methods, build the model and the
preference modeling in a consequence space and may introduce new alternatives
later on.
At this stage, with the criteria and the set of alternatives established the matrix
of consequences can be presented, which consists of the information shown in
Table 1.1. For some problems this matrix can be built very easily, since the
association of alternatives with the corresponding outcome for each criterion can
be made straightforward.
However, for other decision problems this association may not be so
straightforward for some of the criteria. In some cases, the outcome to be achieved
by the alternatives has to be worked out in more complicated procedures.
This possible complexity for establishing the matrix of consequences could

justify that this task would be done as a separate step. However, this association of
each alternative with the outcomes corresponds to the very definition of the
alternatives, including how they are detailed and specified.
2.3.5 Step 5 - Identifying the State of Nature
The state of nature corresponds to one of the ingredients of decision theory (Raiffa
1968; Berger 1985; Edwards et al. 2007; Goodwin and Wright 2004).
The state of nature consists of factors in the system that are not under the DM’s
control and may change randomly, influencing the outcomes of the decision
process. A variable T may represent the state of nature and may be a discrete or
continuous set of elements.
For instance, in a decision problem related to capital investment, regarding new
technologies or machines in an industrial unit, the alternatives are a discrete set of
elements ai, A = {a1, a2, a3, ..., an}, which is a factor under DM’s control. On the
other hand, the demand for the product is the state of nature Ts in this problem,
which is not under the DM’s control. Depending on the nature of the product, it
may be represented by a discrete set of states of nature, 4 = {T1, T2, T3, ..., Tt},
such as for units of computers. Otherwise, the set of Tis continuous, for instance:
liters of juice.
One should be careful with this ingredient of a decision problem, which is in
some situations may be understood as a consequence and represented as a criterion
within the model. This could be a critical modeling error, and affect the decision
process substantially, including a preference modeling on T. Natural consequences
of this kind of problem may lead to two criteria: C: the total cost of the technology
(considering the purchasing and operational costs); and I: the image of the
enterprise as a confident supplier for its costumers.
This ingredient T is integrated in the model, by its association with the
consequences. A consequence function (Berger 1985) makes this association and
may be represented by P(xŇT, a), which for a probabilistic association, such as in
the example of machine purchase. P(xŇT, a) means the probability of obtaining
the consequence x, given that the state of nature is T and the DM chooses the
alternative a.
For a discrete representation of Ts, considering the consequences C and I,
Table 2.1 shows the decision matrix with the state of nature. In this case, the Ts
may represent different scenarios for demand.
Table 2.1 Consequence matrix with the state of nature Ts
A T1 T2 T3 Ts Tt

A1 (C,I)11 (C,I)12 (C,I)13 ... ... (C,I)1t
A2 (C,I)21 (C,I)22 (C,I)23 ... ... (C,I)2t
... ... ... ... ... ... ... ...
Ai ... (C,I)is ... ...
... ... ... ... ... ... ... ...
Am (C,I)m1 (C,I)m2 (C,I)m3 ... ... ... (C,I)mt
The modeling process with this ingredient is approached by decision theory

(Raiffa 1968; Berger 1985), which includes MAUT. The decision model may
incorporate prior probabilities S(T) on T. Otherwise, the decision is conducted
under an uncertainty approach, using an appropriate procedure such as MinMax
(Raiffa 1968; Berger 1985).
Thus, if prior probabilities S(T) are incorporated, a probabilistic modeling task
complements the preference modeling. In probabilistic modeling, the analyst
applies an elicitation procedure so as to obtain the S(T). This procedure is usually
applied to an expert on the behavior of T.
2.3.6 Step 6 - Preference Modeling
This is the first step of the second phase of this procedure In this phase the model
is built and the MCDM/A is chosen, although both may be changed by returning
to previous steps.
This step is very connected to the next two steps and all of them are
considerable relevant for choosing the final model, according to the funnel view,
given before.
The preference structure should be evaluated in this step. For instance, the
preference structure (P,I) should be checked with the DM, evaluating if this
structure is appropriate for representing the DM’s preference. If it is, a traditional
aggregation model may be applied, such as the additive model.
However, if (P,I) is not adequate, then, other structures should be checked,
such as the preference structure (P,Q,I,J), in which the incomparability relation is
considered.
The analyst may start this process by checking some basic properties of the
preference structure (P,I), such as the transitivity and if the DM is able to make a
complete pre-order or order in the consequence space. These properties are
essential to the structure (P,I) and can easily be evaluated with the DM, by check-
ing the relations P and I on the consequences. This format is more conceptual than
operational and could be checked as a preliminary procedure, since these
questions in many cases are included in the elicitation procedures of step 8.
That is, these steps 6, 7 and 8, of the second phase, may be conducted in a very
flexible sequence, even simultaneously and integrated. This process should be
conducted under a non-structured approach, in the sense of this management
information systems concept (Bidgoli 1989; Sprague and Watson 1989; Davis and
Olson 1985; Thierauf 1982). That is, the non-structured approach is due to
the extremely interactive nature of the process, which depends on the DM’s
characteristics and availability. The process is recursive, with many moves
forwards and backwards. This is beyond the view of successive refinement shown
in Fig. 2.1. For instance, the evaluation of relations P and I on the consequences,
at step 6, could be done as an anticipation of the elicitation process of steps 7 and 8.
Also, for some decision contexts, the three steps of this phase may be conducted in
a sequential way, with no repetitions or returns. Considering the nature of the
preference modeling process, everything depends on the DM and decision context.
An important issue to be evaluated in this step is the assessment of rationality
regarding compensation amongst criteria, which can be shown in Fig. 3.6.
Evaluating with the DM:

basic preferences properties and preference system
Which type of rationality is the most

adequate to the DM?
Non-compensatory Compensatory
Preliminary selection Preliminary selection

of non- compensatory method; of non- compensatory method;
for instance: outranking methods for instance: MAUT; MAVT
Fig. 2.6 Evaluation of compensatory rationality
This evaluation of compensation is a question for which the number of studies

is still very limited and are of a preliminary nature. Therefore, this evaluation may
be subjected to some improvisation, since, everything depends on the context,
afterwards. This is an important question when choosing the MCDM/A method,
since the main classifications of these methods divide into two representative
groups: compensatory and non-compensatory methods.
Unfortunately, in many situation when modeling MCDM/A problems, this

issue is not even considered. The preference modeling process in most situations
is limited to steps 7 and 8 (only this step in many cases), only to parameterizing a
model, with a method that has been already chosen, since the very beginning. This
is similar to that proverb in which a hammer (method) is always applied, when it
is considered that any problem is a nail.
The notion of compensatory and non-compensatory rationality has already been
presented and it is related to the Fishburn (1976) concept.
Therefore, after the evaluation proposed by the model illustrated in Fig. 2.6,
then, the choice of the MCDM/A method is partially made. Partially, because the
final evaluation of methods, in steps 7 and 8 (mainly in step 8), are based on an
initial method already chosen in a first round. For instance, if the compensatory
rationality is indicated, a method related to the additive model is a natural starting
point. Then, the properties of this first method are evaluated, before making a final
choice.
2.3.7 Step 7 - Conducting an Intra-Criterion Evaluation
This intra-criterion evaluation consists of eliciting the value function vj(x) (may be
referred to as gi(x)), related to the value of different performances of outcomes in
the criterion j. The information given in the decision matrix should be produced in
this step.
This intra-criterion evaluation depends on the preliminary selection of an
MCDM/A method, in the previous section. On the other hand, the results of this
step may influence a revision on the pre-selection of the kind of MCDM/A
method made in the step 6.
Regarding the influence of the previous step, if a non-compensatory method is
found to be the most appropriate, then, an ordinal evaluation for the consequences
may be enough. Therefore, the intra-criteria evaluation may not necessary, if the
preferences of consequences in each criterion j are already ordered. In such a case,
only a normalization for a common scale may be necessary, which is not often the
case.
For a non-compensatory method, such as an outranking method, the
indifference and preference threshold consists of an intra-criterion evaluation and
is conducted in this step. Also the veto and discordance threshold, commonly part
of the ELECTRE method, are evaluated in this step. It should be observed that an
interval scale may be required, depending on the formulation required for veto and
discordance.
For a compensatory method, such as the unique criterion of synthesis type of
method, a cardinal evaluation of outcomes should be considered and so, an
elicitation procedure should be applied for obtaining the value function vj(x). This
procedure may produce either: linear or non-linear value functions vj(x).
For probabilistic consequences, usually the terminology applied is utility

function uj(x), since the value function is usually a term applied for deterministic
consequences. Therefore, one of the available utility function elicitation
procedures (Keeney and Raiffa 1976; Raiffa 1968; Berger 1985) is applied to
obtain uj(x). These procedures consider lotteries, in which a probabilistic con-
sequence is considered, in order to place choice questions between consequences
to the DM. These procedures identify the DM’s behavior regarding risk, which
may be classified into: neutral, averse, or prone to risk. For a neutral risk behavior,
the uj(x) is a linear function. For both averse and prone risk uj(x) is a non-linear
function. In the elicitation procedure, uj(x) is obtained in a scale of 0 to 1.
Therefore, no normalization procedures are necessary for a linear function uj(x). It
should be observed that the utility function uj(x) is given in an interval scale.
For deterministic consequences, there are a few procedures available (Belton
and Stewart 2002), in which approximations may be made very easily and partial
information may be applied to approach the value functions vj(x).
First of all, it should be evaluated if the value functions vj(x) are either: linear
or non-linear. For linear vj(x), one of the normalization procedures should be
applied, verifying the compatibility of scales for the MCDM/A method and the
inter-criteria evaluation procedure applied. For some of the inter-criteria elicitation
procedures related to the additive model, the interval scale is considered.
In many practical situations a linear function for vj(x) may be found to be the
most appropriate. Even when a non-linear vj(x) is indicated, there are many
situations in which a linear function can be applied as a good approximation, as
has been pointed out by Edwards and Barron (1994), highlighting that a deviation
in a model may be better than an elicitation error. A deviation in a model means
the use of a linear model instead of a non-linear one.
At this point, this can illustrate the advantages of the flexible process proposed,
with the possibilities of returning revise of previous steps. The linear approximation
for a non-linear that may be indicated for vj(x), can have its impact evaluated at
the sensitivity analysis step, when the impact of variations in this function vj(x),
may be considered. If variations in vj(x), change the final recommendation, then, a
return to this step in order to replace vj(x) with a non-linear function may be made.
Step 7 can be affected by the way in which step 3 has been conducted, since the
type of attribute (natural, constructed or proxy) may change the process in this
step, and in some cases, it can already bring in the intra-criterion evaluation. This
is very often the case for the constructed attribute. This may include even the non-
linearity of the scale in some cases.
The intra-criterion evaluation may require specific issues depending on the
kind of problematic applied; for instance: sorting or portfolio.
If a sorting problematic is applied, then, this step includes the evaluation of the
profiles for the categories, in which the alternatives will be classified. These
profiles involve an intra-criterion evaluation for the bounds of each category.
For a portfolio problematic the scales of the value function vj(x) should be
considered very carefully. For instance, when using an outranking method, such as
PROMETHEE V, it has been shown that the necessary transformation in the

scales requires additional evaluation (de Almeida and Vetschera 2012; Vetschera
and de Almeida 2012). For the unique criterion of synthesis methods, based on the
additive model, the value function vj(x) should use a ratio scale instead of an
interval scale, which is used by many of the elicitation procedures (de Almeida
et al. 2014).
2.3.8 Step 8 - Conducting an Inter-Criteria Evaluation
In this step, the choice of the MCDM/A is made, at the beginning or it may
already have been made. The inter-criteria evaluation in this step leads to the
parameters of the MCDM/A model, involving the elicitation procedure for the
criteria weights. This evaluation depends strongly on the kind of method chosen.
Since the meaning of weights changes for different methods, the elicitation
procedure depends on the method.
Regarding the additive model, the meaning of the weights, normally called
scale constants kj, does not involve only the importance of the criteria and their
elicitation is related to the scales of the value function vj(x) in each criterion.
Actually, there are quite a few MCDM/A methods related to the additive model
for aggregation of criteria, in which the main differences amongst them are related
to the elicitation procedure applied for kj.
For the additive model there are also indirect procedures, in which an inference
is made, based on the DM’s global evaluation of some alternatives. This kind of
method is usually classified as a disaggregation method.
Regarding outranking methods, the elicitation of weights is completely
different from that for compensatory methods. In this case, the meaning of weights
is closely related to the importance of criteria and can be obtained considering this
issue.
In the group of methods classified as interactive methods, in which MOLP
methods are included, the intra-criteria evaluation is worked out by an interactive
process involving dialog with the DM and a system, in general a DSS (Decision
Support System). The DM gives preference information at each dialog action,
which is alternated by computation action by the system. The DM views the
problem by considering the consequence space related to the decision context in
question.
There are also many adaptations of classical elicitation procedures for the
additive model, in which partial information is required, using interactive
procedures.
For probabilistic consequences, using MAUT, there are very well structured
elicitation procedures for obtaining the scale constants for aggregation of the
utility functions of the criteria (Keeney and Raiffa 1976).
This step concludes the second phase of the process, with two important
results:
x the decision model has been built;
x the MCDM/A method has been chosen.
Now, the third phase is started in order to resolve the problem, recalling that a
return to and revision of previous steps may be made and the model may change.
2.3.9 Step 9 - Evaluating Alternatives
This is the first step of the third phase of the procedure, the finalization. In this
step the set of alternatives is evaluated, according to the problematic proposed.
The decision model is finally applied.
This step is straightforward and consists basically of applying an algorithm in
the decision model in order to evaluating the set of alternatives.
This step will rarely produce a situation that requires a return to a previous step
and the successive refinement has no place in this step, although this may be
represented in the model as a vague possibility.
The output of this step is still not enough for an evaluation, required for
revision of previous steps. Actually the final result concerning the alternatives has
its final consolidation in the next step.
2.3.10 Step 10 - Conducting a Sensitivity Analysis
The result of step 9 consists of a preliminary recommendation, which must be

confronted with an analysis of the robustness of the process, regarding variations
on the parameters of the model and its input data. This step may indicate that the
recommendation is either: robust or sensitive to the input data or to the model
features. Also, this step may show that the results in step 9, should be reevaluated,
after a revision in previous steps, due to some of the assumptions or input data, or
even any inadequate simplification in the model, for instance in the elicitation
process.
That is, this step checks to what extent the result of step 9, the model output, is
sensitive to variations on the input data and parameters of the model. Regarding
data, any organization may have imprecise data with a varied degree of
approximation, the impact of which can be tested in this step. Also the process for
building the model may have some degree of approximation, and the impact of
this can be evaluated in the sensitivity analysis.
Regarding the kind of solution given by the model for each problematic, the
sensitivity analysis checks different questions and may require different procedures.
For each problematic, the following questions (changes in the model output) are
checked:
x For the choice problematic, the output may present alternatives other than those
of step 9, as a solution for the problem. If so, it is desirable to evaluate: how
many alternatives are presented; in which alternatives this happens; and in
which frequency this happens.
x For the ranking problematic, the output may change the position of some
alternatives in the ranking. If so, it is desirable to evaluate: how often this
happens; in which alternatives it happens; and the significance of these
changes.
x For the sorting problematic, the output may present some alternatives in a class
other than that found in step 9. If so, it is desirable to evaluate: how often this
happens; in which alternatives it happens; and the significance of these
changes.
x For the portfolio problematic, the output may present portfolios other than that
of step 9, as a solution. If so, it is desirable to evaluate: how many portfolios
are presented; and in which frequency this happens.
If no changes are observed this indicates that the model is robust for that
particular set of input data. It may happen that a model appears to be robust for a
set of input data and the opposite may happen for another set of input data. It is
important to check the model and its parameters and also the input data.
If changes happen in the model output, then, it is necessary to investigate how
unacceptable this is. Also, the particular input data or parameter that influences
this change is an important piece of information. This may be useful in order to
evaluate if the model should be revised, returning to some previous step. At this
point, it is worthy remembering that there is not a right model; there are useful
models.
The sensitivity analysis may be conducted based on either: analytical analysis
of the mathematical structure of the model or numerical analysis on the model, by
changing the input data. In spite of simplifications of the model, the complexity of
a model may require a numerical analysis.
Many procedures for sensitivity analysis are available in the literature and are
not detailed in this text, the main focus of which is to discuss the role of this
procedure in the model building process. Therefore, for this focus, the following
two kinds of sensitivity analysis are considered:
x for the evaluation of the overall model in a comprehensive process, including
all parameters and input data, at once.
x for a particular analysis of a specific parameter or input data.
The former procedure consists of an evaluation of the overall model, by
changing, simultaneously, a subset or all sets of input data and parameters of the
model. The Monte Carlo simulation procedure may be applied in this case. In this
procedure a random generation of the subset or all set of data is made and applied
in the model to check the results. This procedure is repeated a number of times
(may be hundreds of thousands of times) in order to compare the frequency at
which the output changes, considering the problematic in question. Other
information to be considered is how significant these changes are, by applying
some statistical hypothesis tests, as demonstrated in Daher and de Almeida (2012).
The changes in each piece of data are established, according to a range around
the nominal value considered for the model and applied in step 9. The range is
specified according to the considerations for assumptions and approximations,
given to that particular piece of data in the modeling process. In general, a
percentage around the nominal value is applied; for instance, plus and minus 30%,
20%, or 10 %. A probability distribution should be applied for the random
generation of data, according to the nature of imprecision observed in the modeling
process; for instance, uniform, triangular, normal probability distributions may be
applied.
This first procedure consists of an overall evaluation of the model and may
indicate whether of not there is a need to continue to the second procedure. The
result of this procedure is included in the recommendation to be given to the DM,
which is worked out in step 11.
The second procedure is very simple to implement and consists of changing the
particular variables of awareness. Each variable is evaluated one at a time, in order
to check its specific impact in the model. This procedure may have an important
managerial role in the process of building the model. During this process a
decision may be made to simplify some step in the procedure. This may be
motivated by the time available being limited or the high costs of collecting
information (preferential or factual data).
These simplifications may be made on the following issues:
x general assumptions for the model;
x the elicitation process, with approximations in parameters; for instance in the
criteria weights;
x assumptions regarding specific analytical structures inside the model;
x using partial information for approximate estimation of input data or model
parameters.
For instance, let us suppose that the elicitation process, in step 8, has
considered approximations in the criteria weights, due to limitations on the DM’s
time. Then, in this procedure, the particular impact of changes in the weights may
be evaluated, in order to check whether or not approximations in the criteria
weights were adequate. If there is no relevant variation, then the simplification in
the model may be considered harmless and the results may be accepted.
Otherwise, an evaluation should be made of the possibility of returning to step 8
and repeating the elicitation process.
The DM may consider that other solutions produced in this step are equivalent
to the nominal solution presented at step 9 and therefore, the results of model may be
accepted. The performance proximity of alternatives may lead to such a situation.
A second example may be given for input data, such as estimates for the cost of
implementing each alternative. Estimates of costs for implementing projects are
obtained with a high level of approximation in many situations. In this case this
second procedure for the sensitivity analysis may indicate if the impact of such
approximation is relevant and should be reevaluated, by returning to step 3.
The results obtained in this step are as relevant as the solution given in step 9.
The DM should know not only the alternative indicated by the model, but also the
impact of model simplifications on this result.
There is moreover, an insightful consideration of this step for the whole process
for building models. Since the sensitivity analysis can indicate how the model
simplifications can affect the results, then, this possibility may influence decisions
that the analyst will make as to on simplifying the modeling process. That is, the
possibility for successive refinement may indicate that any step, which is cost or
time consuming may be conducted with approximations in a preliminary way, and
is expected to be repeated, after evaluating the impact of these approximations on
the result of step 9.
This may reveal that a rigorous procedure for some steps may be useless, in the
context of a building process for producing a useful model, as a simplification of
the reality. Therefore, the analyst has to be careful when evaluating the DM and
the organizational contexts, when building models.
2.3.11 Step 11 - Drawing up Recommendations
After the conclusion of the last step, if no return to revise previous steps is
necessary, then, the finalization is approached in this step by analyzing the final
results and producing the report for the DM, with the final recommendations.
The two previous steps produce the main topics to be included in the
recommendations to be given to the DM. Also, the main considerations on
assumptions and simplifications on the model should be included in the report to
the DM.
That is, the DM is not given only the solution indicated in step 9. This is only
part of the recommendation. The DM has to be aware of the simplifications in the
modeling process and its impact in the solution proposed. This kind of report may
be useful for future evaluations regarding results to be achieved by implementing
alternatives.
A good report indicates to the DM the extent to which the solution can be
trusted. The DM should be advised on the nature of the models. The DM should
understand that there is no right model and the usefulness of the model is the main
issue to be evaluated.
2.3.12 Step 12 - Implementing actions
Finally, after the DM has received the recommendation and accepted the proposed
solution, then, its implementation process can start. This may be either: simple and
immediate or complex and time consuming. The latter situation, may require
special attention. Also, the way in which the decision is taken may influence the
implementation process (Brunsson 2007).
A complex implementation process may be as complex as the decision process
and may take much more time to accomplish than the decision process itself. In
such situations, occasionally the implementation process may be conducted by an
actor other than the DM, who may be afraid of changes in the expected outcomes.
For instance, the implementation process for decisions related to public policy
may be so complex and require so much time to be spent on them, that the
complex solution may change in format as time goes by, leading to outcomes that
are different from those expected at the time of the decision process.
Possible changes in the expected outcomes may happen, when the actor
conducting the implementation introduces modifications in the process that may
alter the format of the solution and its expected outcomes. In these cases, the DM
may be concerned with controlling the content of the solution, although in some
cases this cannot be done. The analyst should be aware of this, since this may
influence the DM’s perception on the relations between the consequences and the
alternatives, if the latter may be changed, during the implementation.
There is another issue of time, which is related to the time at which the
implementation process should be started. That is, the deadline for starting the
process may be considerable, compared with the time for the decision process.
This may appear to be controversial, since the time given for producing the
recommendation may be short, thus leading to a stressed model building process,
and at the end, a longer time is available before starting the implementation. This
may be required, when the organization needs to announce the decision made and
there is still some time available before initiating the action.
In this situation, a procrastination process may be introduced in this step. The
procrastination process consists of introducing and managing a delay before
implementing the solution, so that a re-evaluation of the decision may take place.
The procrastination (Partnoy 2012) takes place under the allegation that it is more
important to take the correct action, than to take it sooner. In this case, it would be
wise to procrastinate, taking time to think over the chosen solution. This thinking
time may allow the decision made to be revised and thereby to gain new insights
from the whole process already conducted.
For some situations, managing this delay is more important than other steps of
the decision process. This may suggest the introduction of a sub-step in step 12, in
which the implementation delay is managed.
Prudence is of the utmost importance in a procrastination process, since a delay

beyond the deadline may bring terrible consequences, even making the chosen
solution unfeasible in some situations.
Regarding the scheduling of the decision process (de Almeida 2013a), there are
two main deadlines to be taken into account in the 12 steps of this procedure:
x the deadline for choosing a final solution and having a recommendation, in
step 11;
x the deadline for starting the implementation process.
The whole scheduling and the time managing process may be illustrated in
Fig. 2.7. The first above mentioned deadline has its main effect in phases 1 and 2.
The times for working out phases 1 and 2 are related to building the decision
model, as illustrated in the final part of the funnel of Fig. 2.7, in which the model
is built (chosen). In these phases the deadline is a constraint that obliges the
analyst to simplify the model in phases 1 and 2. A greater deadline allows a more
cautious process for building models, resulting in a more elaborated model. On the
other hand, the first two steps of phase 3 are more technical and take their own
time. In this phase, the analyst is concerned with the application of the model.
The dosage of time management is an important issue, since it suggests a
balance between two opposite and damage tendencies: streamlining the process
too much and detaining the process for unnecessary improvements.
The second deadline is related to the final step, since the decision has already
been made. The concerns with the first deadline are over and concentration is on
the deadline for starting the action. At this time, the procrastination process may
be introduced and this deadline has to be managed carefully. Here the deadline
allows a delay in order to give the DM the opportunity to review the decision
made, before its implementation. This process is much related to the
organizational context and the DM should be very prudent with this delay
management.
Deadlines constraints the process; tendency

for accelerating the choice of the model in the
funnel
Deadlines constraints
the finalization
process by the analyst
Phase 1 Phase 2 (Preferences

(Preliminary) modeling and method
choice)
Phase 3 (Finalization)
Step 9
Model built
Step 10
Solution
chosen
Step 11
Recommendation
Step 12
Deadline for controlling the
procrastination process
Fig. 2.7 Time managing in the scheduling of the decision process
2.3.13 The Issue of Scales and Normalization of Criteria
Just as in the preference modeling so too in the inter-criterion evaluation, the

performance of the consequences may be expressed in terms of numbers. These
numbers are presented in a given scale. The scale on which a criterion is presented
may define the possibilities for choosing an MCDM/A method. For instance, if the
scale of information given in the consequence matrix or in the decision matrix
gives only ordinal information one can identify that a given consequence may be
greater or lesser than other, but by how much cannot be measured. In such a case,
the additive model in (1.1) may not be applied. Therefore, the scales impose
constraints for the kind of method to be applied.
Familiarity with these scales and their associated normalization procedures
(Polmerol and Barba-Romero 2000; Munda 2008) are important issues for dealing
with MCDM/A problems.
First of all, two kinds of scales may be considered: a) a numerical scale; and
b) a verbal scale. Amongst the numerical scales the following are the main interest
of this text: ratio scale, interval scale, and ordinal scale.
The ordinal scale is the one that has a minimal degree of information. In this
scale the numbers only represent the order to be assigned to the elements in a set.
They do not have cardinality in the sense that one can say 4 is twice as much as 2.
Basic arithmetic operations, such as summation, are not allowed when using this
scale. If a decision problem is presented in such a way that some of the criteria are
presented in the ordinal scale, then an ordinal method should be applied. A careful
application of another method is possible, considering an approximation, in which
case one should be careful, when drawing conclusions from the results.
Many verbal and numerical scales are applied for outcomes of criteria,
represented by subjective scales, which in the end present information that is only
consistent with an ordinal scale. Actually, most pieces of information collected
from a DM, by subjective evaluation, using a verbal or numerical scale, are not
consistent with a cardinal scale, unless, an adequate procedure is applied to ensure
that they are.
The ratio scale is the scale with the greatest degree of information. As
suggested by the name, in this scale the cardinality is in the ratio between two
numbers. For instance, the weight of an object is presented on this scale. This
means that 4 kg is twice as much as 2 kg. The ratio scale has unity and e origin,
represented by the zero of the scale, which means absence of property. That is, 0
kg means absence of weight. In this scale a transformation of the following type
may be done and the scale properties are maintained: y = ax, with a > 0. In this
transformation the origin is kept and the unity is changed. That is what happens
when the weight scale is changed between kg and g. Length and the time are other
examples of ratio scales.
In the interval scale, the cardinality is in the interval between two outcomes. In
this scale, the following linear transformation may be applied, keeping the
properties of the interval scale: y = ax + b, with a > 0. In this transformation the
unity and the origin are changed, respectively by a and b. In this scale the zero
does not have the same meaning as in the ratio scale. The zero means just the
minimum value of the scale (as is usual in MCDM/A problems). Temperature is
an example of an interval scale. In this scale, considering the Celsius scale, one
cannot say that 40oC is twice as much than 20oC. On the other hand, one can say
that passing from 30oC to 10oC is twice as much as passing from 40oC to 30oC.
The above linear transformation may be applied for temperature, so that on
changing from Celsius (x) to Fahrenheit (y), one can apply y = (9/5)x + 32).
Verbal scales are applied in many MCDM/A problems and can be transformed
into a numerical scale in order to be incorporated into a decision model. This scale
may be ordinal or cardinal (ratio or interval), depending on the elicitation
procedure applied. However, a simple process of asking a DM to declare a verbal
scale for a set of consequences in most cases will produce an ordinal evaluation. A
verbal scale that is very often applied is the Likert scale (Likert 1932), in which
the number of levels for evaluation is limited to five (there are variations, such as
a four-level scale), due to the limited human cognitive capacity for making
evaluation in a scale of many levels, such as a ten-level scale, from 1 to 10, which
is often applied inadequately.
The type of scale for the consequences of a criterion, as represented in the
consequence matrix, causes constraints for choosing an MCDM/A method. Also,
the type of scale for a value function vj(x), shown in the decision matrix is chosen
according to the necessary degree of information required and the kind of
transformation to be done.
An interval scale is applied in many MCDM/A methods, such as in Utility
Theory, and it is in its axiomatic structure. This scale presents a piece of
information which has a particular relevance for comparing two alternatives. It
shows how much performance is added from one alternative to another. In many
situations the DM wants to know, how much is added to go from one position to
another. Of course the ratio scale also has interval cardinality and, therefore, gives
the same information as the interval scale.
The interval or ratio scale are both applied for methods, such as those based on
the additive model in (1.1). The interval scale includes an additional feature that
may lead it to be the scale preferred by many of those methods, based on the
additive model. In this scale, the minimum value (xmin) of an outcome for a
criterion j is set to be zero, so that the value of vj(xmin) = 0. Since the maximum
(xmax) outcome is set to be 1, so that the value of vj(xmax) = 1, in this scale the range
(xmax - xmin) is reduced to a minimum, for the scale 0 to 1. In contrast, for the ratio
scale the range (xmax - 0) is higher, for this scale of 0 to 1. This makes the interval
scale more precise for estimating subjective values in the preference modeling
process.
There is a specific situation for the model in (1.1), in which the interval scale is
not adequate. When using MCDM/A in the portfolio problematic, the interval
scale may not be applied, since it induces a wrong solution due to a size effect
caused by this scale. In this case a ratio scale should be applied (de Almeida et al.
2014). For other MCDM/A methods similar situations occur (de Almeida and
Vetschera 2012) and additional procedures should be implemented.
If the value functions vj(x) obtained in the intra-criterion evaluation are linear,
then, the information produced in the decision matrix can be obtained by a
normalization procedure. It should be observed that the term normalization in
MCDM/A does not have the same meaning as it has in statistical procedures of
normalization.
A normalization procedure consists of carrying out a scale transformation so as

to change all criteria to the same scale, since some methods, such as the additive
model in (1.1), require this in order to work out the aggregation process. These
procedures may change the unity or the origin of the original scale.
There is a close relationship between setting managerial indices (or managerial
indicators) and the scales and their normalization process for a criterion. If these
indices have to show the level of performance in objectives they should be
associated with the DM’s preferences.
In MCDM/A methods, in general, this transformation for normalization is
made to a scale of 0 to 1. In this case the least preferred (xmin) and the most
preferred (xmax) consequence have the values 0 and 1, respectively.
A few normalization procedures are presented below, considering the discrete
set of consequences such as that presented for Table 1.1 (consequence matrix),
and an increasing preference with the value of x:
x Procedure 1: vj(x) = (x - xmin)/(xmax - xmin).
x Procedure 2: vj(x) = x/xmax.
x Procedure 3: vj(x) = x/ixi.
For all procedures the values of vj(x) are obtained in the interval 0 vj(x) 1.
Procedure 1 uses an interval scale and the values of vj(x) may be interpreted
as the percentage of the range (xmax - xmin). In this procedure the zero means
the minimum value xmin. Of course, this procedure does not maintain the
proportionality of x. That is, the relation vj(xk)/vj(xl) may not be the same as that of
xk/xl.
Procedure 2 maintains the proportionality of x, uses a ratio scale and the values
of vj(x) may be interpreted as the percentage of the maximum value of X (xmax),
indicating the distance to the leader alternative in the consequence matrix. In this
procedure the zero means x = 0.
Procedure 3 maintains the proportionality of x and uses a ratio scale. The
values of vj(x) may be interpreted as the percentage of the summation of all
consequences of X (xi), indicating the distance to the leader alternative in the
consequence matrix. In this procedure the zero means x = 0. This procedure is
widely applied when normalizing weights of criteria.
2.3.14 Other Issues for Building MCDM/A Models
This section deals with a few specific issues for building MCDM/A models, such
as psychological traps, the choice of the method, compensation of criteria, and the
intelligence stage of Simon’s model.
Psychological Traps
There are some psychological traps, discussed in the behavioral decision making
literature that can affect the quality of the information obtained from the DM,
during the elicitation procedures for preference modeling. This is relevant, since
the DM’s preferences to be included in the model are items of subjective-based
information. Simon (1982) discusses the limitation on rationality that people in
general have.
A few of these psychological traps are briefly presented below (Hammond
et al. 1998a):
x Anchoring - People tend to give a strong weight to information received
(impressions, estimates, data) just before making any subjective evaluation.
This should be considered in the way that preference questions are put to the
DM or factual questions to an expert.
x Status Quo - There is a tendency of choosing actions that maintain the Status
Quo. This may lead to confirm and repeat past decisions.
x Estimating and Forecasting - In general people are skilled at making estimates
about time, distance, etc, in a deterministic way. However, making these
estimates considering uncertainty is different. On the other hand, DMs usually
have to make such kinds of estimates for their decisions.
x Overconfidence - DMs tend to be overconfident about their own accuracy, thus
naturally guiding them to errors of judgment in preference elicitation procedures.
This is one of the traps that affect the DM’s ability to assess probabilities
adequately.
With regard to the estimating and forecasting trap, Hammond et al (1998a)
state that DMs rarely get clear feedback about the accuracy of those estimates they
have to make. The feature of successive refinement in the decision procedure
described above can minimize this situation, combined with the results of the
sensitivity analysis, although this does not improve the accuracy for future estimates.
The way in which questions are put forward to the DM may induce errs, in any
of these traps. For instance, the more choices the elicitation procedure gives to the
DM, the more chance there is that the status quo will be chosen (Hammond et al.
1998a).
Suggestions to deal with these difficulties are given by Hammond et al
(1998a).They also present other psychological traps, which include: confirming
evidence, framing, and prudence.
The Choice of the MCDM/A Method
In the literature there are not many studies dealing with the choice of a proper
MCDM/A method for a decision problem. However, this seems to be changing.
The concern with the matching between the method and problem has increased and
may be influencing adaptations in classical methods and even the development

and use of hybrid methods. The latter require many cautions, since the integration
of different axiomatic structures may lead to serious errors. A few studies deal
with this matter. Roy and Sáowinki (2013) put several questions for guiding the
choice of a method.
The above procedure for building an MCDM/A model gives substantial
emphasis to this issue of choosing the MCDM/A method, particularly concerned
with the matching with the decision problem, which is the central issue in this
matter. Phase two of that procedure is devoted to this topic.
Several factors should be observed for the choice of method, which are closely
related to the context of the model building process, and may include:
x The nature of problem analyzed, which is the central feature in the whole process;
x The context in which the problem is faced, which includes organizational
issues, and the time available for the decision to be made;
x The DM’s preference structure;
Unfortunately, the analyst’s preference on the method may play an important
role in this process. This may bring ethical considerations to the process.
Rauschmayer et al. (2009) discuss the ethical issues in the modeling process. They
state that the choice of the method and its parameterization is not neutral and may
bring an ethical problem if:
x Distortions in the results are made for interests other than the DM’s and the
organizational one, in which the problem is faced.
x The assumptions are not shared with the DM.
x The assumptions are selected in a malicious way
It should be noted that the second issue above is carefully considered in step 11
of the above procedure, since all this information should be included in the
recommendation report.
One of the main issues in the choice of an MCDM/A method is the evaluation
of the DM’s preference structure with regard to compensatory and non-
compensatory rationality, as highlighted in step 6 of the procedure for building
MCDM/A models. Simon (1955) pointed out the importance of this issue, before
many of the MCDM/A methods had been developed. Bouyssou (1986) made
remarks on the concepts and notion of compensation and non-compensation and
discussed a few axiomatic issues.
According to Vincke (1992) the choice of a method for aggregating criteria,
such as the additive method, for instance, is equivalent to choosing the type of
compensation amongst those criteria. Roy and Sáowinki (2013) are concerned with
this issue, in the context of choosing a method, when they put the following
question “Is the compensation of bad performances on some criteria by good ones
on other criteria acceptable?”.
Although step 6 of the procedure for building an MCDM/A model includes this
evaluation of the DM’s willingness or otherwise to make compensation, no details
are given on how to deal with this. Indeed, there is still much research work to be
conducted on the evaluation of the DM’s willingness to make compensations,
even though this is an extremely relevant factor for the choice of methods.
The Intelligence Stage of Simon in the Procedure for Building Models
The foregoing procedure for building an MCDM/A model does not include the
intelligence stage of Simon’s model for the decision process (Simon 1960). This pro-
cedure assumes that there is already a problem that has been identified at the start
of the design stage of Simon’s model. Fig 2.8 shows how this intelligence stage can
be integrated with the procedure described above for building a decision model.
This intelligence stage requires a continuous monitoring process on the status
of the organization or the decision context, in which attention to the decision
process is established, and also its external environment.
Intelligence stage of Simon

Model
Monitoring and
collecting
information
Organization SWOT data

or analysis
decision context
Is there any
External environment opportunity or
treats no
yes
Procedure for building the

decision model (Fig. 2.3)
Fig. 2.8 Integrating Simon’s intelligence stage
This monitoring process may, at any moment, indicate a situation requiring

attention and then data collected are analyzed, in order to identify whether or not
there is a problem to be solved, which may include an opportunity to be explored.
If so, then, the above procedure is initialized.
This monitoring process is very well associated to the strategic management

process, in which the diagnosis analysis of the internal and external environment
of the organization is conducted. Also, the VFT approach proposed by Keeney
(1992) can be considered in the model shown in Fig. 2.8. Using the VFT
approach, the specification of values would guide the monitoring process.
2.3.15 Insights for Building MCDM/A Models in the RRM Context
In an MCDM/A model for the RRM Context, uncertainty is usually a certain

thing. That is, a decision under a certainty situation may be possible only as a
simplification of the model. Also, this may be justified either: when the variability
of the random variable is not considerable or when the use of quantiles of the
probability distribution for the variables, such as criteria, may be applied as a good
approximation.
For the former, a deterministic approximation is quite useful and justifiable.
The mean of the random variable can be applied, since the standard deviation is
assumed to be too small.
For the latter formulation, a deterministic approach is usually applied, although
there are many concerns to be taken into account with that approximation. An
alternative to this procedure is the disaggregation of the criterion into two: the
mean and the standard deviation of the random variable. The analyst should
evaluate very carefully, which of these possibilities the DM can better understand.
Even, the choice of the quantile should be considered the best option for the DM’s
understanding; for instance, the quantile could be either: 90% or 80% of the
distribution.
It should be noticed that deterministic MCDM/A methods are largely applied in
reliability and maintenance contexts. Table 2.2 derived from a literature review
shows the percentage use of different MCDM/A approaches in maintenance and
reliability problems (de Almeida et al. 2015).
Table 2.2 MCDM/A approaches applied in reliability and maintenance research
Method Percentage
Pareto Front 48.39
MAUT 10.22
AHP 9.68
MACBETH or other MAVT 8.60
Goal Programming 3.23
ELECTRE 2.69
PROMETHEE 2.15
TOPSIS 1.08
As can be observed, in most cases, it seems that a deterministic model is

applied, since it is not clear what amount of probabilistic adaptations is conducted
in these methods. One can wonder how much this is related to either: a
simplification of model itself or a bias in the analyst’s choice.
This issue is relevant, since reliability and maintenance contexts are very
closely related to risk considerations by their very concepts. An interesting refer-
ence on uncertainties in MCDM/A (Stewart 2005) shows different meanings for
uncertainty and how to deal with them, including a few guidelines for practitioners.
Also, many issues related to a risk analysis of uncertain systems are considered by
Cox (2009). For instance, he discusses the limitations of some quantitative risk
assessment, such as frequency, which is often applied to explain risk, yet which
does not contain enough information for a clear decision to be made.
MCDM/A Models in the Risk Context
With regard to the risk context, there is a variety of concepts in the literature on
risk and also on its perception (Chap. 3 deals with this topic). Some of them
consider only the probability for a specific context. However, if a decision is being
made then the consequences should be considered. Also, the model should
incorporate the DM’s preferences over these consequences. In fact, a ‘decision
process’, in which the DM’s preference is not considered is not a process in which
a decision is actually being made, as discussed at the end of Chap. 1.
According to Cox (2012), the application of utility functions rather than simple
risk formulas – consisting of terms such as exposure, probability and consequence -
allows a DM’s risk attitudes to take into account, thereby improving the effective-
ness of the decision making process to reduce risks. Cox (2009) discusses many
issues related to the decision process in the risk context, including the limitations
of risk assessment using risk matrices and a normative decision framework.
Another classical problem within the risk context is the direct association
between the quality of a decision and the actual consequence obtained at the end.
In fact, at the time in which the decision is being made, the DM cannot assure the
best consequence, since there are uncertainties in the process. Therefore, only
expectations can be evaluated when making the decision. In general, this is
something difficult for many DMs to understood and the analyst should be aware
of how to deal with this by clarifying all these issues to the DM, instead of using
inadequate models for simplifying what is going to be shown. These clarifications
should be made in step 11, when drawing up the recommendations to the DM.
Interpretation of an MCDM/A Model or Utility Function Scores
There are many concerns in the literature with regard to interpreting the scores for
the alternatives given by utility functions. This concern is extended in general to
any MCDM/A method that gives final scores for alternatives, thus representing a
global evaluation, based on the aggregation of multiple criteria. However, these
numbers can be interpreted according to the properties of the scale, for each
particular method, in order to compare alternatives.
If the method uses a ratio scale, it is relatively easy to produce a comparison of
alternatives, considering the ratio of their scores. For instance, in a choice
problematic, a first alternative may be twice as good as the second one, or it could
be 20% better than the second one.
Even for a specific scale, such as the ratio scale, the meaning of this ratio may
be explained, by taking the rationality behind the method into account. For
instance, in the PROMETHE II method, the scores are based on the summation of
criteria weights, within a non-compensatory rationality.
With regard to the interval scale, which is applied for the utility function of
many of the MAVT methods, the alternatives may be compared based on the
properties of this scale.
The interval scale allows an incremental comparison between alternatives. That
is, the differences of the scores of the alternatives are considered. However, a ratio
may also be considered between two differences, as shown in Chap. 4 (see Equations
(4.13) and (4.14)). Therefore, a difference ratio DR may be applied to interpret
v ( a p ) v ( a p 1 )
the values in relation to the alternatives, so that: DR , in
v ( a p 1 ) v ( a p 2 )
which p represents the position in the ranking obtained by alternative ap and v(ap)
represents the score of the alternative. By analyzing these DR results, the DM
can perceive the distance between the pairs of alternatives. This is illustrated in
Table 2.3.
Table 2.3 Analysis of scores of an MCDM/A method with an interval scale
Alternative i Position (p) Value or Utility Interval Ratio of

of the intervals
alternative (DR)
A2 1 0.70 0.10 0.77
A5 2 0.60 0.13 6.50
A1 3 0.47 0.02 0.40
A3 4 0.45 0.05 1.00
A7 5 0.40 0.05 5.00
A8 6 0.35 0.01 0.04
A4 7 0.34 0.24 -----
A6 8 0.10 ----- -----
Table 2.3 presents the position of the alternatives in the second column, their
scores in the third column and their comparisons by the increments of the scores in
the fourth column. The fifth column shows the DR, from which it can be observed
2.4 Multicriteria Decision Methods 57
that the increment of the scores from A1 to A5 is 6.50 times greater than that from
A3 to A1.
Another possible way to explain these results to the DM is to consider the ratio
of differences between two alternatives and the whole range, given by the range
between the best and the worst scores. This difference can be expressed as a
percentage of the whole range. That is, in Table 2.3, the whole range is v(A2)-
v(A6)=0.70-0.10=0.60. Therefore, the difference in scores between alternatives A5
and A1 is 22% of the whole range, while for alternatives A1 and A3 it is 3%.
Applications of these indices are given in Chap 4. The analyst may use any one
of these indices, after evaluating which of them is the most appropriate for a given
DM to understand.
Paradoxes and Behavioral Concerns Related to Risk Evaluation
With regard to the use the expected utility function for models in the risk context,
there are a few paradoxes with which the analyst should be aware of. These
paradoxes have been analyzed by behavioral decision making studies, in the
descriptive perspective context.
There are other approaches that deal with some particular situations, such as
Rank-Dependent Utility (RDU) and Prospective Theory (Edwards et al. 2007;
Wakker 2010).
There are many situations regarding risk which cannot be easily integrated into
decision models. The kind of event known as a ‘black swan’, related to the so
called ‘black swan theory’ may be an example of such a situation. This event is
related to a kind of occurrence that is very unexpected (very low probability), with
very undesirable consequences. These are rare events, which result in a great
damage. In general, their evaluation is not well accepted in the expected value
principle, since the multiplication of the value of such great damage is excessively
reduced by the value of an extremely low probability.
On the other hand, although many concerns with the use of the expected utility
function are clamorously announced in part of the literature, the analyst should be
aware that in many situations these behavioral issues do not matter for many
practical problems. It is necessary to understand their meaning and to evaluate
them when they are relevant. Unfortunately, in many situations, these matters are
inappropriately announced in order to justify other less adequate approaches.
2.4 Multicriteria Decision Methods
A brief overview of MCDM/A methods is given in this section with emphasis on

those most often found in practical application, balanced with the most appropriate
ones for the RRM context.
First the methods related to unique criterion of synthesis are presented, then
some outranking methods are introduced. Interactive methods, related to MOLP
are very briefly mentioned, since most of the problems in the RRM contexts are
non-linear problems. The next section deals with heuristics and evolutionary
multiobjective algorithms for dealing with multiobjective models.
2.4.1 Deterministic Additive Aggregation Methods
This is one of the most applied models for aggregating criteria and it is usually
classified as MAVT (Belton and Stewart 2002), being part of the group of
methods of unique criterion of synthesis. MAVT is distinguished because it
considers deterministic consequences, whereas MAUT (see next subsection) deals
with probabilistic consequences (Keeney and Raiffa 1976).
The additive model, also called a weighted sum model, is recalled from (1.1)
and reintroduced below for prompt reference in (2.1), in which the global value
(v(xi)) is considered for a consequence vector xi = (xi1, xi2, ..., xin), for the
alternative i, which is the same as the global value v(ai) for alternative ai, as
indicated in (1.1).
n
v ( xi ) ¦k v (x )
j 1
j j ij (2.1)
where:
kj is the scale constant (weights) for attribute or criterion j.
vj(xij) is the value of consequence for criterion j, for the alternative i.
xij is the consequence or outcome of alternative i for criterion j.
The scale constant is usually normalized as follows:
¦k
j 1
j 1. (2.2)
Properties for the Additive Model
The additive model has a few properties that should be checked before making a
decision on its application. For practical modeling purposes the main properties
are briefly described.
This model follows the preference structure (P,I), in which it is possible to
obtain a complete pre-order or a complete order. For two consequences xz and
xy, the following conditions hold for this structure: a) xyPxz Ј v(xy) > v(xz);
b) xyIxz Ј v(xy) = v(xz). Therefore, one of the assumptions of this model is that the
DM is able to compare all consequences and order them. Also the transitivity
property holds for the preference relation R, whether it is P or I, so that for three
consequences xw, xy and xz, if xwRxy and xyRxz Ј xwRxz.
Another property of this model is the mutual preference independence
condition amongst the criteria (Keeney and Raiffa 1976). Let Y and Z be two
criteria, the preference independence between Y and Z occurs if and only if the
conditional preference in the Y space (intra-criteria evaluation given, different
levels of y, such as y' and y''), given a certain level of z = z', does not depend on
the level of z. That is, (y',z')P(y'',z') Њ (y',z)P(y'',z), for all z, y' and y''.
This property may be formally presented in the following formulation (Vincke
1992). Let a, b, c and d be four vector of consequences in a consequence space
with two criteria Y and Z. Then, Y and Z are preferentially independent if the
following condition holds: If for criterion Y, vy(a) = vy(b), and vy(c) = vy(d), and
for criterion Z, vz(a) = vz(c), and vz(b) = vz(d), then, aPb Ј cPd. This is illustrated
in Fig. 2.9.
b d
vz(xz)
a c
vz(xz)
vy(xy) vy(xy) Y
Fig. 2.9 Preference independence condition
Therefore, the validation of this model should be done by confirming that the
DM’s preference structure is according to these properties. In some practical
situations a DM may refuse to follow the final recommendation based on this kind
of model, when a violation of one of these properties occurs and there alternatives
close to the solution, in which it is obvious for a global evaluation that a property
is violated. The DM may not be able to perceive which property is being violated,
in such cases, but can recognize the inconsistency of the final result. Although, the
DM can distinguish this kind of inconsistency only in an obvious situation, this
shows that this may not be an issue to be ignored.
Therefore, these properties should be evaluated very carefully, before making a

decision of going through them. Of course, the additive model may be applied, as
a typical simplification procedure for model building, where some property is not
consistent with the DM’s preference. However, the analyst should evaluate
carefully to what extent this is inconsistent with the DM’s preference.
Regarding the preference independence property, it has been observed that in
most practical situations this property is not violated. This may explain, in part,
the broad dissemination of the use of this model, although the other properties
should also be considered. Yet, regarding the preference independence, Keeney
(1992) points out that the preference dependence may indicate that a criterion
may be missing. In this case, a revision of steps 2 and 3 of the above procedure
may allow a better structuring of the problem.
Also, practical applications have shown that the violation of this property is
more likely to happen for a large range of consequences. For a small range of
consequences, the mutual preference independence is more likely to hold. This has
an interesting relation with the kind of scale applied to a criterion. For instance, a
ratio scale tends to be larger than an interval scale. Therefore, one should be
careful, when changing from an interval scale to a ratio scale, for a problematic of
portfolio that requires the latter (de Almeida et al. 2014).
Elicitation Procedures for Scale Constants
There are many elicitation procedures in the literature for the elicitation of the
scale constants (Weber and Borcherding 1993). Amongst these are the tradeoff
and the swing procedures which are described below.
The tradeoff procedure is presented in detail by Keeney and Raiffa (1976).
Weber and Borcherding (1993) consider that this is the procedure with the
strongest theoretical foundation.
This procedure is classified as an indirect procedure (Weber and Borcherding
1993), since the determination of the scale constants is based on inference from
information given by the DM. It is also classified as an algebraic procedure, since
it calculates the n scale constants from a set of n-1 judgments often using a simple
system of equations, which also includes (2.1).
This procedure is based on a sequence of structured questions (Keeney and
Raiffa 1976) put to the DM, in order to obtain preference information, based on
choices between two consequences. A first group of questions obtains the ordering
of the scale constants, then, other questions prepare the DM to understand better
the consequence space and finally, the DM makes choices between pairs of
consequences related to neighboring criteria, in order to make the tradeoffs for the
equations for the algebraic process.
Thus, the procedure is based on the DM making a comparison on two
consequences xb = (x1, x2, ..., xj, ..., xn), which is a vector with the consequences xj
for each criterion j. These consequences have the best outcome bj, for one of the
criteria and the worst outcome wj for the other criteria. For instance, x2 = (w1, b2,
..., wj, ..., wn) has the best outcome for the criterion j = 2, and x3 = (w1, w2, b3, ...,
wj, ..., wn) has the best outcome for j = 3. If the DM’s preference is such that
x3Px2, then, v(x3) > v(x2). Based on (2.1), the value of v(xb) = kb, since v(bj) = 1
and v(wj) = 0. Therefore, if x3Px2, then, k3 > k2. Using these kinds of questions,
the order of the scale constants is obtained.
Next, another pair of consequences is compared in order to find indifference
between them, by decreasing the value of the outcome bj for criterion j which is
the preferred one. For instance, for x3Px2, the consequence b3, of x3, has the
outcome decreased to the level of x3, such that x3'Ix2, in which x3' = (w1, w2, x3, ...,
wj, ..., wn). If the DM can specify the outcome x3', such that x3'Ix2, then, v(x3') =
v(x2). Since, v(xb) = kb and v(xb') = kbvb(xb), by applying (2.1), this leads to k3v3(x3)
= k2. This equation is related to one of the n-1 judgments for the system of
equations necessary in this procedure, in order to obtain all the scale constants kj.
A critical judgment in this procedure is adjustment of the outcome in order to
obtain the indifference between the two consequences above (Weber and
Borcherding 1993).
The swing procedure is included in the SMARTS method (Edwards and Barron
1994). This procedure is classified as an algebraic procedure and also as a direct
procedure (Weber and Borcherding 1993), since the determination of the scale
constants are based on direct information given by the DM, taking the range of the
consequences into consideration.
This procedure is also based on a sequence of structured question (Edwards and
Barron 1994). The first question considers the following consequence w = (w1, w2,
..., wj, ..., wn)., in which all criteria have the worst outcome. Then, the DM is asked
to choose one of the j criterion to improve the outcome of wj to the best outcome
bj. That is, the DM may choose a criterion to ‘swing’ from the worst to the best
outcome. This indicates criterion j for which the scale constant kj has the greatest
value. Then, the DM is asked to choose the next criterion, and so on. At the end
the scale constants of the criteria are ordered. Then, in another step, the criterion
with the largest value of scale constant is arbitrarily assigned 100 points. The other
criteria are assigned points expressed as percentages of the criterion with the
largest scale constant value, considering their range. Finally, these percentages are
normalized to produce the final scale constants.
Avoiding Misinterpretations Regarding the Scale Constants
There is a quite commonly disseminated misconception (for additive models) of

associating the meaning of the scale constants with the degree of importance of
the criteria. This represents a source of one of the main modeling mistakes when
the additive model is used.
In the additive model, this parameter cannot be determined as weights, con-
sidering only the degree of importance of the criterion, which may be appropriate
in other methods, such as in outranking methods. Although the value of a scale

constant of a criterion may be associated with its importance, there are other issues
to be considered. The value of a scale constant is also related to the scale range of
the consequences for the criterion (Edwards and Barron 1994). For instance, in a
decision problem for purchasing a product, in which any five criteria are
considered, including the price, one could state that the price is the most important
criterion, thus with the largest weight. However, if the outcomes related to price
are in a very narrow range of consequences, let us say between $ 99,990 for the
best price and $ 100,005, for the worst price, it does not seem relevant to assign
the highest weight to such a criterion. This is even clearer considering the additive
model in (2.1) and the most usual normalization procedure for the value function
such that the worse outcome is set to 0 and the best outcome is set to 1.
Actually, the scale constants are substitution rates between the criteria (Keeney
and Raiffa 1976; Vincke 1992; Belton and Stewart 2002). Keeney and Raiffa
(1976) point out that it might happen that a criterion may have a scale constant
larger than any other and yet it has less importance. Several practical examples are
discussed on this issue by Keeney and Raiffa (1976) and Keeney (1992).
Finally, one should be aware that changing the normalization procedure or
using different scales (for instance: a ratio or an interval scale) for the value
function completely affects the set of values established for the criteria weights (or
scale constants). In such a case a new set of values for the criteria weights should
be computed. Of course this is valid for the additive model, although it is not valid
for other methods, such as the outranking methods.
Some MAVT Additive MCDM/A Methods
There are quite a few methods incorporating the additive model. The main
difference amongst them is in the elicitation procedures of the parameters,
including both the intra-criterion and inter-criteria evaluations, with emphasis on
the scale constants.
In many situations the use of the additive model is straightforward with the use
of one of the classical elicitation procedures, there being no explicit consideration
of an MCDM/A method. In other cases, an MCDM/A method is considered.
One of the most applied methods that incorporates the additive model is
SMARTS (Simple Multi-Attribute Rating Technique with Swing), in which the
swing procedure is applied (Edwards and Barron 1994). SMARTER (Simple
Multi-Attribute Rating Technique Exploiting Ranks) is a related method that
applies the first step of ordering the scale constants of the criteria and then, uses a
surrogate weight. In these methods the value function for each criterion is assumed to
be linear (Edwards and Barron 1994).
The AHP (Analytic Hierarchy Process) presents a particular procedure for
preference modeling, considering the possibility of a hierarchical structure of
objectives (Saaty 1980). The method uses the additive aggregation model, and
collects information based on pairwise comparison of alternatives. In the literature

there are some complaints that this method does not follow some of the properties
of the additive model and a few other concerns, such as: the possibility of order
reversal, and the interpretation for the criteria weights (Belton and Stewart 2002;
Howard 1992). Howard (1992) points out that it is widely applied, since it does
not demand much effort from the DM.
Macbeth (Measuring Attractiveness by a Categorical Based Evaluation
Technique) is a method based on a qualitative evaluation on the difference of
attractivity (Bana and Costa et al. 2005). They say that this method seeks to be
concerned with constructing the value of outcomes, but does not force the DM to
produce a direct numerical representation of preferences. The DM gives some
preference information that is applied to build a numerical scale, based on a set of
Linear Programming Problems (LPP).
The even swaps are based on the procedure proposed by Benjamin Franklin for
the tradeoff on choosing whether or not implement an action (Hammond et al.
1998a; Hammond et al. 1999).
Additive-Veto Model
The compensatory nature of the additive model may recommend an alternative

with a very low outcome level in one of the criteria, which is compensated by high
outcome levels in one of more of the other criteria. However, it may happen that
the DM may prefer not to select such a kind of alternative, whatever the criterion
with low performance is. Thus, additive-veto models (de Almeida 2013b) may
solve this problem by vetoing alternatives in such situations.
Numerical simulation in such kinds of situations has shown that it may not be
rare for alternatives from a set of alternatives have this kind of characteristic (de
Almeida 2013b), namely, one in which a very low outcome level in one of the
criteria is compensated by high outcome levels in other criteria, thus ranking this
alternative in a high position. This means that, depending on the DM’s preference
structure, if the DM is not willing to accept such a kind of alternative, then, a veto
of the best alternative should occur in the additive model.
Roy and Sáowinki (2013) discuss the choice of MCDM/A methods, considering
several questions, such as this kind of compensation of bad performances in some
criteria by good ones in other criteria. They pointed out that the acceptability of
this situation should be evaluated for a compensatory method.
Additive Models for the Portfolio Problematic
The use of additive models for the portfolio problematic demands some concerns
with the scale to be applied, since there is a size effect that causes the wrong
solution to be selected in the interval scale, which is the one most applied for
elicitation procedures (de Almeida et al. 2014).
The portfolio problematic in the additive model is based on the selection of a
portfolio pr that maximizes the value V pr as given in (2.3).
m § n ·
V pr ¦ ¨¨ x ¦ k v (a ) ¸¸
i 1
i
j 1
j j i (2.3)
© ¹
m
subject to some constraints, such as a budget constraint of ¦x c d B .
i 1
i i
where:
pr >a1, ... , a m @ is the portfolio, which is a vector with the items (projects) ai.
1 if the item (project) x i is included in the portfolio
xi ® .
¯0 if the item (project) x i is not included in the portfolio
C represents the vector of item costs, C >c1 , c2 , ! , cm @ T .
B is the budget or the limit for total cost C.
For portfolio selection, based on additive models, as in (2.3), the interval scale
may not be applied. It has an impact on the result due to the size effect of the
portfolio in this kind of scale, thus causing the wrong portfolio to be selected.
What has been proved to be most appropriate is the ratio scale for this kind of
problem (de Almeida et al. 2014). Most weight elicitation procedures are based on
the interval scale that sets the worst outcome to zero, whereas using a ratio scale
for the portfolio selection, the weights to be applied with the scale should be
changed. The transformation of these scales can be seen at de Almeida et al.
(2014).
Methods Based on Partial Information for Elicitation of Weights
Many behavioral studies have been conducted in order to evaluate the

consistencies of the elicitation procedures. Borcherding et al (1991) have reported
on inconsistencies of 50% and 67% of the time, when using ratio swing tradeoff
procedures.
There has been some justification for using procedures with partial information
instead of those elicitation procedures with complete information, since the
elicitation of weighs can be time-consuming and controversial (Kirkwood and
Sarin 1985; Kirkwood and Corner 1993) and because the DM may not be able to
respond specifically to tradeoff questions (Kirkwood and Sarin 1985).
A few approaches have been proposed to deal with the model in (1.1) using
partial information. One of the ways of dealing with this is to use surrogate
weights. SMARTER (Edwards and Barron 1994) uses this idea, based on the
partial information of the order of criteria weights. Another procedure (Danielson
et al, 2014) increases the precision for surrogate weights by adding numerically
imprecise cardinal information into rank-order methods, such as the ROC (Rank
Order Centroid), also applied in SMARTER.
Other approaches collect more information and use procedures based on
decision rules, formulating linear programming problems (LPP) or simulation
procedures in order to analyze the alternatives. Among these approaches are:
PAIRS (Salo and Hämäläinen, 1992), which uses interval judgments; VIP
Analysis (Dias and Climaco, 2000), based on the progressive reduction of the
number of alternatives; PRIME (Salo; Hämäläinen, 2001) which uses preference
information based on swing method or holistic information; and RICH (Salo and
Punkka, 2005) which uses incomplete ordinal preference statements. Mustajoki
and Hamalainen (2005) integrate preference elicitation in the partial information
framework for the SMART/SWING method.
A flexible elicitation procedure adapts the tradeoff elicitation procedure by
using partial information in an interactive way, and conducts analysis by means of
a set of LPPs (de Almeida 2014a; de Almeida 2014b).
2.4.2 MAUT
MAUT has been developed for MCDM/A problems, from Utility Theory (von
Neumann and Morgenstern 1944), keeping its axiomatic structure (Keeney and
Raiffa 1976). According to Edwards and Barron (1994), Howard Raiffa presented
the fundamental insight for MAUT in 1968, pointing out that there would be more
than one reason to value an object. Raiffa (1968) presented a few considerations
for a multicriteria view in the context of health problems.
This approach gives one of the most classical MCDM/A methods, in which the
most widely applied aggregation approach has been the additive model, for which
the axiomatic structure of the theory indicates a number of properties to be
considered. As mentioned, the main difference from the MAUT additive model to
the model in the previous section is that the probabilistic consequence is
approached in the utility function uj(xj) for each criterion j.
The decision models with MAUT may include the framework of Decision
Theory (Raiffa 1968; Berger 1985; Edwards et al. 2007), also called as Decision
Analysis, which may consider the Bayesian approach to dealing with uncertainties,
incorporating prior probabilities. Therefore, the uncertainties on the state of nature
(T) may be obtained from experts, in the form of prior probabilities S(T). Thus,
T is an additional ingredient to be considered with MAUT, although this may not
be explicit in some models.
For each Ts chosen by nature and each action ai chosen by the DM, a con-
sequence x may be obtained, according to a consequence function (Berger 1985)
P(xŇT,a), which shows the probabilistic association amongst these ingredients,

meaning the probability of obtaining x, given T and a.
Thus, the model building process with MAUT incorporates a probabilistic
modeling task for these ingredients, which complements the preference modeling.
This probabilistic modeling task, in general, may involve another actor in the
decision process, namely an expert. Usually the expert brings knowledge on the
probabilistic behavior of the state of nature, so that the analyst applies elicitation
procedures for obtaining S(T), as subjective probabilities.
Therefore, when applying MAUT, the final model consists of a multi-attribute
utility (MAU) function u(x1, x2, ..., xn) = f>u1(x1), u2(x2), ...., un(xn)@, to be
maximized by the choice of an alternative probabilistically associated with the
consequences (x1, x2, ..., xn). This corresponds to the expected utility function for
the consequences under consideration.
From now on, the main elements of MAUT are going to be presented
considering the case of two criteria x and y leading to the MAU function: u(x, y) =
f>u1(x), u2(y)@.
The choices in Utility Theory consider the concept of lottery, which represents
a probabilistic consequence. For instance, a lottery with two consequences is
represented by [A, p; B, 1–p], which means the possibility of obtaining one of two
consequences A or B, where p is the probability of obtaining A, and 1-p is the
probability of obtaining B.
There has been a set of axioms for Utility Theory, ever since its first
formulation (von Neumann and Morgenstern 1944, Raiffa 1968; Keeney and
Raiffa 1976; Berger 1985), which are applied to MAUT.
Just as in the additive model for MAVT, in MAUT the models follow the
preference structure (P,I). Therefore, the first axiom is related to the ability of the
DM to compare all consequences and order them. The second axiom is the
transitivity preference relations P and I. These two axioms are implicitly related to
probabilistic consequences, so they may apply for lotteries. The other axioms are
explicitly related to lotteries. Let the lotteries with the consequences A, B and C
and the probabilities p and q, then, there are the two following axioms:
x If APB, then there is a probability p, 0<p1, so that for any C, [A,p; C,1–p]P
[B,p;C,1–p]. This is also applied to indifference relation I.
x If APBPC, then, there are p and q, 0<q<p<1, so that [A,p;C,1–
p]PBP[A,q;C,1–q].
Consequence Space
The whole evaluation process for the utility function is made over the
consequence space, with which the DM should be familiar. Fig. 2.10 shows the
consequence space for two criteria x and y.
y
(x0,y*) (x*,y*)
*
y
y1 (x*,y1)
(x0,y0) (x*,y0)
0
y
x0 x* x
Fig. 2.10 Consequence space for two criteria
In the consequence space shown in Fig. 2.10, for each criterion, the most
desirable outcomes are x* and y*, while the least desirable outcomes are x0 and y0.
For the whole space the points (x*,y*) and (x0,y0) represented respectively the most
and least desirable outcomes for the multi-attribute space. The scale for the utility
is arbitrarily set in the interval 0 to 1, so that u(x*,y*) = 1, u(x0,y0) = 0, uj(x*) = 1,
uj(y*) = 1, uj(x0) = 0 and uj(y0) = 0.
Elicitation of the Conditional Utility Function
The utility function uj(xj) for each criterion j, related to the intra-criterion
evaluation, is assessed considering a conditional utility function of criterion j,
which is conditioned to a fixed level of the outcomes in other criteria. For
instance, on the x axis of Fig. 2.10, there is a conditional utility function of
criterion x, given a fixed level of the outcome for criterion y = y0.
The intra-criterion evaluation consists of eliciting this single dimensional utility
function uj(xj). There are several procedures for this elicitation (Raiffa 1968;
Keeney and Raiffa 1976; Berger 1985), many of them use the concept of certain
equivalent of a lottery. This certain equivalent is the consequence B for which
there is a probability p, such that the DM is indifferent between B and a lottery
[A, p; C, 1–p], with consequences A and C.
In general, the consequences of the lottery are the least and the most desirable,
so that the probability p = u(B). Since, u(x*) = 1 and u(x0) = 0, and considering
the indifference between B and [x*,p;x0,1–p], then u(B)=pu(x*)+(1-p)u(x0). From
this, it follows that u(B) = p.
Therefore, the elicitation procedure consists of obtaining the indifference
between this kind of lottery and the consequences x, so that the utility function
u(x) can be obtained. Detailed elicitation procedures are provided in Keeney and
Raiffa (1976).
Elicitation of the MAU Function
For the elicitation of the MAU function u(x1,x2,...,xn)=f>u1(x1),u2(x2),....,un(xn)@,

after obtaining the conditional utility function of each criterion, then the elicitation
procedure is conducted for global utility. Let the two criteria be x and y and
the consequence space in Fig. 2.10. Then, the elicitation seeks to obtain
u(x,y)=f>ux(x),uy(y)@.
For the elicitation of the MAU function there are a few structured procedures
(Keeney and Raiffa 1976). The main process, described below, is based on a
prescriptive approach, in which preference conditions are evaluated with the DM,
and based on these, analytical functions may be applied to u(x,y).
The two main concepts of preference conditions considered for this purpose
are: the additive independence condition and the utility independence condition.
If the mutual additive independence condition is found between x and y, in the
DM’s preference structure, then the additive model, u(x,y)=kxu(x)+kyu(y), may be
applied. (2.4) gives a more general model for n criteria.
n
u( x) ¦k u ( x )
j 1
j j j (2.4)
where:
kj is the scale constant for attribute or criterion j;
uj(xj) is the utility function for criterion j;
xj is the consequence or outcome for criterion j.
The scale constant kj is usually normalized as in (2.1).
If the mutual utility independence condition is found between x and y, in the DM’s
preference structure, then the multilinear model, u(x,y)=kxu(x)+kyu(y)+kxyu(x)u(y),
may be applied. Similar to (2.4), a generalization may be made for a model with n
criteria.
The Utility Independence Condition
This independence preferential condition is associated with the context of utility

functions. This concept may be understood considering the consequence space of
Fig. 2.10. Criterion x is said to be utility independent of criterion y, if the
conditional utility function u(x,y0) is strategically equivalent to any other utility of
x, whatever the outcome for y is. The utility u(x,y0) is the utility for x, given that
y=y0. This means that the certain equivalent of the lottery [(x*,y0),p;(x0,y0),1–p],
whatever the value of p is, is the same for any other lottery [(x*,y),p;(x0,y),1–p],
whatever the outcome for y is.
It is interesting to note that for the strategically equivalent utility function
u(x,y0), a utility u(x,y) may be found by a linear transformation, such as
u(x,y)=a(y)u(x,y0)+b(y), where: a(y)>0 and b(y)>0 are constants, established for
any outcome for y.
Therefore, as shown with this utility independent condition, the utility function
u(x,y) depends only on the particular level of the outcome in criterion y, even so,
by a linear transformation. More details on this concept are given by Keeney and
Raiffa (1976).
The Additive Independence Condition
This independence condition imposes stronger constraints on the additive model.

Let the following consequences of the space in (x,y) be: A, B, C and D,
respectively corresponding to (x1,y1), (x1,y2), (x2,y2), (x2,y1), as illustrated in Fig. 2.11.
B C
y2
A D
y1
y0
x0 x1 x2 x
Fig. 2.11 Additive independence condition
The additive independence condition holds if the DM is indifferent between the

following lotteries: [A,0.5;C,0.5] and [B,0.5;D,0.5], whatever x and y are, in the
consequences A, B, C and D. Since, the same probability p = 0.5 is applied to both
consequences in these lotteries, its representation may be simplified as follows:
[A,C] and [B,D].
Considering the indifference between two lotteries similar to those in Fig. 2.11,
such as those of the values of (x,y) for consequences A, B, C and D, being
[(x0,y0),(x,y)] and [(x0,y),(x,y0)], then the utility of the lotteries has the same value.
Thus, 0.5u(x0,y0)+0.5u(x,y)=0.5u(x0,y)+0.5u(x,y0). Given, the normalized scale for
the extreme values x and y, then u(x,y)=u(x,y0)+u(x0,y).
It can be seen that u(x,y0) and u(x0,y) can be obtained based on the scale
constants kj, such that: u(x,y0)=kxux(x) and u(x0,y)=kyuy(y). This leads to the format
of (2.4). This concept and its development are given in detail by Keeney and
Raiffa (1976).
Elicitation of the Scale Constants
A complete and detailed procedure for the elicitation of the MAU function is
given by Keeney and Raiffa (1976). The elicitation of the scale constants kj is
based on the analytical model obtained, associated with the independence
conditions.
For instance, the scale constants kj for the additive model on the two criteria x
and y correspond to the utility of the two specific consequences (x*,y0) and (x0,y*),
shown in Fig. 2.10. That is, kx=u(x*,y0) and ky=u(x0,y*).
Therefore, the elicitation of kx consists of finding the probability p for which
(x*,y0) is the certain equivalent to the lottery [(x*,y*),p;(x0,y0),1–p]. A similar
evaluation may be made for ky.
Again, as can be seen the scale constants kj for an MAU function are not simply
the relative degree of importance of the criterion. They are related to the scale,
considering the limits for x and y, since the lottery [(x*,y*),p;(x0,y0),1–p] is the
basis for their elicitation.
Rank-Dependent Utility and Prospective Theory
There are quite a few paradoxes related to the use of the expected utility function,
which are presented in the literature. Many of these paradoxes have been analyzed
in a descriptive perspective within the context of behavioral decision making.
In many situations Rank-Dependent Utility (RDU) and Prospective Theory
(Edwards et al. 2007) have been considered as ways of dealing with such
situations (Wakker 2010).
MCDM/A models based on MAUT may be adapted with Rank-Dependent
Utility and Prospective Theory views on modeling risk preferences, which may
have particular relevance for the RRM context.
2.4.3 Outranking Methods
This kind of method has a completely different rationality from the methods in the
two previous subsections. These methods are non-compensatory and may be
applied to a preference structure (P,Q,I,J). The possibility of the incomparability
relation is one of the issues distinguished in this kind of method, and therefore,
only partial pre-orders may be obtained.
Therefore, unlike MAVT and MAUT, this kind of method may be applied in a
situation for which the DM’s preferences are not in agreement with the first two
properties. That is, the DM is not able to compare all consequences and order
them. Also, the transitivity property may not be followed.
This section presents some basic elements of these methods and then,
introduces an overview on the two most widely applied outranking methods:
ELECTRE and PROMETEE.
These methods are based on pairwise comparison of the alternatives, by
exploring an outranking relation between the pairs of alternatives.
There is an important difference between outranking methods and those of
MAVT and MAUT that impacts the preference modeling process, namely the
different meaning for the inter-criteria parameters, which may be called weights.
The meaning of criteria weights corresponds directly to the degree of importance
of the criteria, for outranking methods.
This notion of importance amongst criteria may be compared with votes in a
voting process (Roy 1996; Vincke 1992). Let there be two subsets of criteria G
and H and two alternatives a and b. If the subset of criteria in G is more important
(has more votes) than the criteria in the subset in H, and the following conditions
hold (Vincke 1992):
x a is better than b for all criteria in the subset G;
x b is better than a for all criteria in the subset H; and
x a and b are indifferent for any other criteria;
Then: a is globally better than b.
If this importance (or votes) can be represented by the criteria weights, the
comparison between the subsets of criteria G and H can be based on the
summation of these weights.
That is, the summation of weights for criteria in favor of a is greater than those
in favor of b. This means that a makes a better coalition of criteria than b.
These methods are worked out in two main steps (Roy 1996; Vincke 1992):
x Building the outranking relation, by comparing all pair of alternatives in the set
of alternatives;
x Exploiting the outranking relation by applying an algorithm or procedure for
solving the problem, according to each particular problematic.
These methods may work with different kinds of criteria, depending on their
intra-criterion characteristics. In a true criterion there is no threshold. For a pseudo
criterion there are thresholds that may be one of the following or both: an
indifference threshold and a preference threshold.
The outranking relation S, is applied over all pairs of alternatives of the set of
alternatives, such as a and b. Therefore, aSb means that alternative a outranks
alternative b, which means that a is at least as good as b.
ELECTRE Methods
In the ELECTRE (Elimination Et Choix Traduisant la Réalité) methods the

outranking relation aSb, between two alternatives a and b, is based on concord-
ance and discordance concepts, on which the DM gives preference information in
the form of thresholds.
The family of ELECTRE methods includes the following methods, which
differs from the problematic and the kind of criteria (Roy 1996; Vincke 1992):
x The ELECTRE I method is applied for a choice problematic, considering true
criteria;
x The ELECTRE IS method, which is applied for a choice problematic,
considering pseudo criteria;
x The ELECTRE II method, which is applied for a ranking problematic,
considering true criteria;
x The ELECTRE III method, which is applied for a ranking problematic,
x The ELECTRE IV method, which is applied for a ranking problematic,
x The ELECTRE TRI method, which is applied for a sorting problematic,
considering pseudo criteria.
The ELECTRE I method is subsequently described in order to illustrate the
basic approach followed by these methods. The other methods have some
differences in the parameters for the step of building the outranking relation and
are at their most different in the step of exploiting the outranking relation,
according to their problematic.
For building the outranking relation, ELECTRE I uses the concepts of
concordance and discordance. The former indicates if a considerable subset of
criteria is in favor of an outranking relation S between two alternatives. The latter,
may disagree with this relation S, even if the concordance is in agreement.
Therefore, when evaluating the outranking relation aSb, between two
alternatives a and b, the following indices are applied: the concordance index
C(a,b) and the discordance index D(a,b).
The concordance index C(a,b) is given by (2.5).
(2.5)
C ( a , b) ¦ wj
j:g j ( a )t g j ( b )
where:
wj is the weight for criterion j; the weights are normalized, such that 1.
¦w
j
j
gj(a) and gj(b) is the value of the outcome for criterion j, respectively for
alternatives a and b.
There are a few different formulations for the discordance index (Roy 1996;
Vincke 1992; Belton and Stewart 2002). D(a,b) may be given by (2.6):
§ g j (b) g j ( a ) ·
D ( a , b) max¨ ¸ , j | g j (b) ! g j (a ); j, c, d . (2.6)
¨ max[ g j (c ) g j (d )] ¸
© ¹
A concordance threshold c' and discordance threshold d' should be specified by

the DM in order to build the outranking relation. The outranking relation aSb
between a and b, is established by (2.7).
C(a,b) t c' (2.7)

aSb if and only if ®
¯ D ( a , b) d d '
Having obtained these formulations and parameters, this step for building of
the outranking relation can be finalized, by applying (2.7) for all pair of
alternatives. It may happen with a pair of alternatives that aSb and bSa. In this
case there is a circuit and these alternatives are considered indifferent.
The second step of exploiting the outranking relation can now be worked out.
For the ELECTRE I method, the purpose of this step is to obtain the kernel, which
is the subset of alternatives, in which each of its elements is not outranked by any
other in the kernel. If only one alternative is found in the kernel, the choice
problematic reaches its particular case of optimization. Otherwise, the alternatives
in the kernel have been found to be incomparable.
More details on ELECTRE methods may be found in many basic texts on
MCDM/A methods (Roy 1996; Vincke 1992; Belton and Stewart 2002; Figueira
et al. 2005).
PROMETHEE Methods
PROMETHEE (Preference Ranking Organization Method for Enrichment

Evaluation) is a group of outranking methods, based on a valued outranking
relation (Brans and Vincke 1985; Vincke 1992; Belton and Stewart 2002).
In PROMETHEE methods the DM does not have to specify information on
concordance and discordance regarding the outranking relation. The DM provides
the information on the criteria weights and on the intra-criterion evaluation,
related to the indifference or preference thresholds, if any of them are considered.
This group of methods uses the following formulation for the first step of
building the outranking relation, thereby establishing the outranking degree S(a,b),
for each pair of alternatives a and b, from (2.8).
n
S a, b ¦ w F ( a , b)
j 1
j j (2.8)
where:
wj is the weight for criterion j; the weights are normalized, such that 1.
¦w
j
j
Fj(a,b) is a function of the difference [gj(a)-gj(b)] of the outcomes of the

alternatives for criterion j.
The method has six different patterns for this function Fj(a,b). In the basic form
for Fj(a,b), it does not use either of indifference or preference thresholds for
criterion j. In this case, Fj(a,b) = 1, if gj(a) > gj(b) and Fj(a,b) = 0, otherwise.
Thus, the outranking degree S(a,b), is the summation of all criteria weights for
those criteria, in which a has a better performance than b.
The other five forms for Fj(a,b) considers indifference or preference thresholds,
or both, for criterion j. In these five patterns for Fj(a,b), it has a value between 0
and 1, for criterion j, when the difference [gj(a)-gj(b)] is in the range of the
indifference or preference thresholds. In this range, the outranking degree S(a,b),
adds a partial value of the weights of criterion j, in which a has a better
performance than b, as can be seen in (2.8).
These forms for Fj(a,b), are chosen by the DM, in the context of the intra-
criterion evaluation, and includes the specification of values related to the
indifference or preference thresholds, for that criterion j.
The matrix with the values of the outranking degree S(a,b) for each pair of
alternatives can be available now, thus concluding the first step.
For the second step of exploiting the outranking relation, each alternative a is
evaluated based on the outgoing flow I(a) and on the ingoing flow I(a).
The outgoing flow I(a) indicates the advantage of the alternative a over all
other alternatives b in the set of alternatives A. I(a) is obtained from (2.9).
I(a) = ¦S a, b
bA
(2.9)
where, n-1 gives a normalized scale between 0 and 1, since n is the number of
criteria.
The ingoing flow I(a) indicates the disadvantage of the alternative a compared
with all other alternatives b in the set of alternatives A. I(a) is obtained from
(2.10):
I(a) = ¦S b, a .
bA
(2.10)
Another index for evaluation of the alternatives is the liquid flow I(a), given by
(2.11), which is obtained in a scale of -1 to 1.
I(a)=I(a)-I(a) (2.11)
Now the second step of exploiting the outranking relation, may be concluded
by using these indices on specific procedures for each problematic.
In the PROMETHEE I method two pre-orders are built, based on (2.9) and
(2.10), which indicate the relations of preference (P), indifference (I) and
incomparability (J) between the pairs of alternatives of set A (Brans and Vincke
1985; Belton and Stewart 2002). Therefore, PROMETHEE I outputs a partial pre-
order of the elements of A.
The PROMETHEE II method is based on the liquid flow I(a) from (2.11), in
which each alternative has a score. Therefore, PROMETHEE II outputs a
complete pre-order on the elements of A.
The family of PROMETHEE methods includes other methods: PROMETHEE
III and IV, for a stochastic situation; PROMETHEE V for a portfolio problematic,
as discussed in the following sub-section; and PROMETHEE VI, when the DM
specifies a range for each criterion weight, instead of a precise value of weight.
PROMETHEE V for Portfolio Problematic
The PROMETHEE V method (Brans and Mareschal 1992) is applied for selecting
portfolios using a non-compensatory method for evaluating of alternatives in a
model similar to that in (2.3). The only difference is in computing the value of the
portfolio V pr , which is based on the application of PROMETHEE II for scoring
the items ai (projects) as values vi(ai).
There is also a problem of scale with this method, although different from that
with the additive model. In this case PROMETHEE II presents positive and
negative scores for vi(ai) = I(ai) to be applied in (2.3). Therefore, to work in the
maximization model, the negative scores have to be transformed into positive
scores, thereby changing the properties of the ratio scale (Vetschera and de
Almeida 2012).
This transformation has a similar effect, with the possibility of selecting the
wrong portfolio. Contrary to the case of the additive model in (2.3), the ratio scale
cannot be applied in the PROMETHEE V. In order to overcome this problem, an
analysis should be conducted based on the concept of a c-optimal portfolio
(Vetschera and de Almeida 2012; de Almeida and Vetschera 2012).
2.4.5 Other MCDM/A Methods
There are other approaches and concepts that may be seen either as specific
methods or tools that can be applied in any method, such as those presented above.
Belton and Stewart (2002) consider the latter option for fuzzy sets and rough sets.
A comprehensive view of fuzzy approaches for modeling MCDM/A problems is
given by Pedrycz et al. (2011), while the rough sets approach is briefly described
in the next subsection.
There are a few approaches classified as disaggregation methods, which are
based on holistic (or global) evaluation by the DM, followed by a subsequent step
of inference of the parameters of an aggregation model. Pardalos et al. (1995)
consider these approaches as a fourth group of methods in their classification.
Some of these approaches, such as the UTA method (Jacquet-Lagréze and
Siskos 1982), are related to the single criterion of synthesis methods. However,
inference procedures proposed for the ELECTE TRI method use the same process
of collecting information from the DM on global evaluation for posterior inference
of the parameters of inter-criteria evaluation. The preference learning approach
(Slowinski et al. 2012) uses a similar process.
Rough Sets
This is a kind of MCDM/A method based on preference learning. These methods

consider the DM’s preferences by evaluating a set of decision rules discovered
from preference data, which can be elicited previously from the DM and
afterwards used as an input to establish comparisons among the set of alternatives
(Slowinski et al. 2012).
Rough sets theory has been widely used as an MCDM/A approach based on
preference learning (Pawlak and Slowinski 1994; Greco et al. 2001; Greco et al.
2002; Slowinski et al. 2012).The preference learning approach seeks to avoid the
elicitation of model parameters, such as importance weights or scale constants and
others related to thresholds. It uses information from previous preferences stated
by a DM to establish preference relations among the alternatives based on this
input by assuming that the sample of statements gathered from the DM is enough
to establish decision rules for evaluating the set of alternatives.
This approach may be applied to evaluating risk conditions, for which decision
rules may be built, grounded on preferential information given by the DM. That is,
rough sets could be applied in a similar way to the problem of territorial risk
evaluation (Cailloux et al. 2013), based on ELECTRE TRI method.
2.5 Multiobjective Optimization 77
2.4.6 Mathematical Programming Methods
Several mathematical programming techniques have been proposed to solve

multiobjective problems, such as involving linear (MOLP - Multi-Objective
Linear Programming) and nonlinear programming principles. There is a broad
range of relevant literature on this topic (Korhonen 2009; Korhonen 2005;
Korhonen and Wallenius 2010; Steuer 1986; Ehrgott 2006; Miettinen 1999;
Coello et al. 2007).
Basically, a mathematical programming for solving a multiobjective problem
can be approached in the following ways:
x By considering a preference structure in advance so as to solve the problem by
some approach, such as: transforming multiple objective functions into a single
objective function, solving by an interactive process, and so forth.
x By identifying the non-dominated solutions which together form the set of
Pareto optimal outcomes (more commonly referred to as the Pareto front),
without taking the DM’s preferences into account.
The latter is discussed in the next section. The former may consider the DM’s
preferences by either: collecting information or taking assumptions. In terms of
articulating the DM’s preferences, three classes can be defined: a priori, posteriori
and progressive articulation of the preferences. Some of these approaches are
listed in Table 2.4.
Table 2.4 Summary of MCDA representative methods
Articulation
MCDA methods
of preferences
Global Criterion Method (Osyczka 1984); Goal Programming (Charnes and
Cooper 1961); Goal-Attainment Method (Chen and Liu 1994); Lexicographic
A priori
Method (Rao 1984); Min-Max Optimization (Osyczka 1984); Surrogate Worth
Trade-Off (Haimes et al. 1975).
A posteriori Weighted Sum; İ-constraint Method (Miettinen 1999).
STEP Method (Benayoun et al. 1971); (SEMOPS) Sequential Multiobjective
Progressive
Problem Solving Method (Duckstein et al. 1975)
2.5 Multiobjective Optimization
Multiobjective optimization approaches are related to complex problems and have

spread to the research fields of heuristics and evolutionary algorithms. Two
possible reasons for this evolution are that problems have become more complex
and the ability of these approaches to find Pareto solutions promptly. In terms of
complexity, some problems are classified as NP-Hard and exact methods have not
been successful in finding non-dominated solutions. Therefore, some heuristics

and evolutionary multiobjective algorithms are described.
Multiobjective optimization is based on Pareto-front analysis. In multiobjective
optimization the notion of optimum was generalized by Vilfredo Pareto (in 1896).
It can be said that a vector of decision variables, x*, is Pareto optimal if there is no
other vector of decision variables, x, such that fi(x) < fi(x*) for all i =1, …, k and
fi(x) < fi(x*) for at least one j (Coello et al. 2007).
In multiobjective optimization, all objectives are considered important and all
non-dominated solutions should be found. Thereafter, higher-level information,
generally on non-technical, qualitative and experience-driven matters, can be used
to compare non-dominated solutions before making a choice. This principle is
defined as an ideal multiobjective optimization procedure (Deb 2001).
Several studies are focused only on determining the non-dominated solutions,
assuming that all non-dominated solutions are equally optimum, or that the DM
will provide information on his/her preferences after he/she learns what the Pareto
front is. These assumptions make sense in complex problems where finding non-
dominated solutions is an independent and hard task.
In terms of multiobjective evolutionary algorithms, there are some algorithms
that do not incorporate the concept of Pareto dominance in their selection mechanism.
These are considered first generation methods. They started to become obsolete in
the literature because some algorithms started to rank the population based on
Pareto dominance, which are second generation methods (Coello et al. 2007). It is
important to point out that, in general, multiobjective optimization based on
evolutionary algorithms concentrates its efforts on the first step of the MCDM/A
problem: identifying the Pareto front. Main multiobjective evolutionary algorithms of
these generations are represented in Table 2.5.
Table 2.5 First and second generations of multiobjective evolutionary algorithms (MOEAs)
MOEAs Generation Methods

First Generation GA with Aggregating Functions
VEGA - (Schaffer 1985)
MOGA - (Fonseca and Fleming 1993)
NSGA - (Srinivas and Deb 1994)
NPGA - (Horn et al. 1994)
NPGA 2 - (Erickson et al. 2001)
Second Generation SPEA and SPEA2 - (Zitzler and Thiele 1999)
NSGA-II - (Deb et al. 2002)
PAES - (Knowles and Come 2000)
PESA and PESA II - (Corne et al. 2000)
micro-GA - (Coello Coello and Toscano Pulido 2001)
2.6 Group Decision and Negotiation 79
2.6 Group Decision and Negotiation
In many decision processes there is more than one DM. In such situations a group
decision model or a negotiation process has to be applied in order to come to a
final solution. Therefore, a brief overview is given of Group Decision and
Negotiation (GDN) methods and processes, particularly of those aspects most
closely related to MCDM/A models. The GDN area covers decision problems
with multiple DMs, over a wide range of topics such as: Conflict Analysis (Fraser
and Hipel 1984; Keith et al. 1993; Kilgour and Keith 2005), web-based
negotiation support systems (Kersten and Noronha 1999), evolutionary systems
design (Shakun M.F 1988), connectedness (Shakun 2010), formal consciousness
(Shakun 2006) and fair division (Brams and Taylor 1996).
As stated by Kilgour and Eden (2010) negotiation and group decision contain
both unity and diversity. Regarding the latter, some of the scholars in the field of
GDN understand that it is appropriate to distinguish between Group Decision
(GD) making and negotiation. Kilgour and Eden (2010) explain that in this view
GD making is related to a decision problem shared by more than one DM, who
must make a choice, for which all DMs will have some responsibility. On the
other hand, a negotiation is seen as a process in which two or more DMs, acting in
an independent way, may either: make a collective choice, or not do so. For the
latter, one (or more) of the DMs may give up taking further part in the decision
process and walk away.
Additionally, it can be considered that a GD process involves an analytical
procedure in order to aggregate the preferences of the individual DMs, which
results in a kind of collective representation of the preferences of the group. With
regard to negotiation, this involves a process of interaction between DMs, in order
to find a collective solution for the problem of their mutual interest.
As to using the analytical procedure in order to aggregate the DMs’
preferences, the process for building models pays great attention to following rules
of rationality, related to a normative perspective. Also, there are some concerns
about dealing with some paradoxes, as shown by the descriptive perspective. As
for the negotiation process, the interaction between people invokes other concerns,
such as the accuracy of their communication process.
These issues show some diversity between GD making and a negotiation
process. However, there are some elements of unity between them. For instance,
most negotiation processes are grounded in analytical results and endeavor to
ensure the rationality and fairness of the collective choice. Also, the building
process for the GD model involves agreements with the group of DMs, regarding
several issues and parameters of the model, especially when the problem also
involves multiple objectives, leading to an integrated MCDM/A and GD model.
Therefore, in order to build GD models, some interaction processes may be
necessary between the DMs. The process for building GD models will depend on
the available time of these DMs, and most of all, on how simultaneously their
availability can be made. Also, it should be considered issues related to the

distributive and integrative models (Kersten 2001).
Given the very close relationship between GD making and the MCDM/A
modeling process, a brief description of some aspects of this topic is given below.
Although some studies suggest that MCDM/A models may be straightforwardly
applied for GD aggregation, one should be aware that aggregating people’s
preferences is completely different from aggregating criteria that represent the
objectives of an individual. The area of GDN brings contributions to the concerns
to be dealt with when integrating DMs’ preferences.
2.6.1 Aggregation of DMs’ Preferences or Experts’ Knowledge
While most studies on GD making are related to the aggregation of DMs’

preferences, others are associated with experts’ knowledge. These two GD
procedures are related to aggregating or integrating two substantially distinct
situations. These two kinds of aggregating process have differences in their
foundations. Unfortunately, in some studies this distinction is not clear and may
lead to misconceptions and mislead the decision modeling process. That is, using
an inappropriate foundation to build a decision model will produce a wrong model
and thereby lead to an unsuitable solution.
The aggregation of DMs’ preferences is related to consequences value (Leyva-
Lopez and Fernandez-Gonzalez 2003; Morais and de Almeida 2012). On the other
hand, the aggregation of experts’ knowledge is associated with some specific
subject.
In the former, the process does not seek the true solution. Instead, the process
seeks the most appropriate solution, considering the DMs’ preferences. The
foundations for the aggregation process are concerned with aspects such as
rationality and preference elicitation. This kind of aggregation process considers
the differences in objectives between DMs, and takes into account elements
associated with preferences, such as the DMs’ tradeoffs and the possibilities of
compromising; in other words, the extent to which a DM is wiling to make
concessions in order to reach a final group decision. In this case DMs do not
change their preferences. Instead, they make concessions, always according to
their preferences.
In the latter, the process is focused on seeking the true about some particular
situation, based on the experts’ knowledge. The foundations for this kind of
aggregation process are concerned with aspects such as experts’ knowledge and
their accuracy on evaluating variables in a system. This process considers the
differences in perception among experts, taking into account elements associated
with knowledge, such as the experts’ different backgrounds and experiences. The
experts are not supposed to keep their initial opinion on a subject, unless their
knowledge gives grounds for doing so. An expert may change his/her opinion on a
2.6 Group Decision and Negotiation 81
subject, since they can learn something new from other experts. That is why many
studies are focused on searching for consensus regarding the experts’ perceptions
of that particular topic.
Regardless of these differences on these two kinds of aggregation, some
models are built in order to tackle these two issues mutually, since both are
present in many GDN problems.
Many fuzzy approaches are applied to this kind of problem (Ekel et al. 2008;
Pedrycz et al. 2011), and deal with factors such as ambiguity and uncertainties the
experts have as to describing their perception on the variables that are being
evaluated.
There is a particular kind of situation related to experts’ aggregation of
probabilities, which is related to prior probabilities S(T) on the state of nature T.
There are many studies in the literature on Decision Theory (or Decision Analysis)
related to the elicitation of prior probabilities (Raiffa 1968; Berger 1985) and the
aggregation of a group of experts’ prior probabilities (Edwards et al. 2007). At the
end of Chap. 3 there are more details about this topic.
The following subsection gives a brief description of types of group decision
aggregations regarding DMs’ preferences.
2.6.2 Types of Group Decision Aggregations
Regarding the aggregation of DMs, different actors may play specific roles. For
instance, instead of an analyst, a facilitator or a mediator may act in some
situations. For instance, a facilitator may act so as to intensify the interaction
process between DMs or among other actors in the decision process. With regard
to DMs, the way in which they act and are available for the interaction in the
decision process, for a particular problem, plays an important role when
classifying the types of GD aggregation.
The GD aggregation process consists of reducing the set of individual DMs’
preferences to a collective DMs’ preference. There are some situations in which
one of the actors in the GD process is a supra-DM. This supra-DM makes decision
on final issues, in general, related to global evaluations in the process, such as
evaluating the other DMs’ choices. The supra-DM may have a hierarchical
position above the other DMs in the organization’s structure. Keeney (1976)
considers two types of GD process, with regard to DMs’ interrelationships: the
‘benevolent dictator problem’ and the ‘participatory group problem’. The former
is related to the situation regarding a supra-DM and in the latter, the group acts
jointly in the GD process, with the same power.
Whether or not a supra-DM is present in the process, two kinds of GD
aggregation general procedures may be considered (Kim and Ahn 1999; Leyva-
López and Fernández-González 2003; Dias and Clímaco 2005):
x Aggregation of DMs’ initial preferences.

x Aggregation of DMs’ individual choices, which means the ranking of
alternatives by each DM’s;
These two GD aggregation procedures are illustrated in Fig. 2.12, with the first
kind on the left-hand side and the second on the right-hand side. With regard to the
first steps of preparation for the GD process, there is an integration in the former
procedure, whereas in the latter, the process is completely separate for each DM.
In the former the DMs provide their initial preferences in an integrated way, in
which the aggregation process is considered from the very beginning. Then, the
process produces the final choices for the set of alternatives. This may be given as
a simple ordinal ranking of the alternatives or may include a cardinal score for
each alternative, depending on the method applied, which is the same for all DMs.
The same criteria are considered for all DMs, but the intra-criterion and inter -
criteria evaluations may be different. In most models the former is the same and
the main difference is in the analysis of the criteria weights.
DM1 DM2 ... DMk ranking of

ranking of ranking of
alternatives alternatives alternatives
by by by
DMs’ initial ...
DM1 DM2 DMk
preferences
GD procedure for
aggregating DMs’ GD procedure for aggregating
initial preferences DMs’ individual choices
final DMs’ preferences collective ranking of alternatives

on the alternatives set
Fig. 2.12 Types of GD aggregation procedures
In the latter, each DM provides his/her individual ranking of alternatives. That

is, the individual DMs’ choices produce the final ranking of alternatives or other
results if another problematic, such as choice or sorting, is applied, although in
these cases information on scores of the alternatives is not expected to be
produced, in general. These may be produced by completely different methods,
with different criteria for each DM. It does not matter which objective each DM
considers. The only information that matters is the final individual evaluation of
each alternative by each DM. With regard to the GD process, if a ranking of
alternatives is produced by each DM, then the GD procedure may be conducted by
using a voting procedure, which is based on the foundations of Social Choice
Theory (Nurmi 1987; Nurmi 2002).
References 83
References
Ackermann F, Eden C (2001) SODA - journey making and mapping in practice. In: Rosenhead J,
Mingers J (eds) Rational Analysis in a Problematic World Revisited, 2nd ed. John Wiley &
Sons Inc., United Kingdom, pp 43–61
Ackoff RL, Sasinieni MW (1968) Fundamentals of operations research. John Wiley & Sons,
New York, p 455
Bana e Costa C, De Corte J-M, Vansnick J-C (2005) On the Mathematical Foundation of
MACBETH. Mult. Criteria Decis. Anal. State Art Surv. SE - 10. Springer New York, pp
409–437
Belton V, Stewart TJ (2002) Multiple Criteria Decision Analysis. Kluwer Academic Publishers
Benayoun R, de Montgolfier J, Tergny J, Laritchev O (1971) Linear programming with multiple
objective functions: Step method (stem). Math Program 1(1):366–375
Berger JO (1985) Statistical decision theory and Bayesian analysis. Springer Science & Business
Media, New York
Bidgoli H (1989) Decision support systems: principles and practice. West Pub. Co.
Borcherding K, Eppel T, von Winterfeldt D (1991) Comparison of Weighting Judgments in
Multiattribute Utility Measurement. Manage Sci 37:1603–1619
Bouyssou D (1986) Some remarks on the notion of compensation in MCDM. Eur J Oper Res
26(1):150–160
Bouyssou D, Marchant T, Pirlot M, Tsoukis A, Vincke P (2006) Evaluation and decision models
with multiple criteria: Stepping stones for the analyst. Springer Science & Business Media
Brams SJ, Taylor AD (1996) Fair Division: from cake-cutting to dispute resolution. Cambridge
University Press, New York
Brans JP, Mareschal B (1992) PROMETHEE V: MCDM Problems with Segmentation
Constraints. INFOR 30(2):85-96
Brans JP, Vincke Ph (1985) A preference ranking organization method: the Promethee method
for multiple criteria decision making, Manage Sci 31:647–656
Brunsson N (2007) The consequences of decision-making. Oxford University Press New York,
NY
Cailloux O, Mayag B, Meyer P, Mousseau V (2013) Operational tools to build a multicriteria
territorial risk scale with multiple stakeholders. Reliab Eng Syst Saf 120:88–97
Charnes A, Cooper WW (1961) Management Models and Industrial Applications of Linear
Programming. John Wiley & Sons
Chen Y-L, Liu C-C (1994) Multiobjective VAr planning using the goal-attainment method. IEE
Proc. - Gener. Transm. Distrib. IET, pp 227–232
Coello CAC, Toscano Pulido G (2001) A Micro-Genetic Algorithm for Multiobjective
Optimization. In: Zitzler E, Deb K, Thiele L, Coello Coello CA, Corne D (eds) First
International Conference on Evolutionary Multi-Criterion Optimization: 126-140. Springer-
Verlag. Lecture Notes in Computer Science No. 1993
Coello CC, Lamont GB, Van Veldhuizen DA (2007) Evolutionary algorithms for solving multi-
objective problems. Springer Science & Business Media
Corne DW, Jerram NR, Knowles JD, Oates MJ (2001) PESA-II: Region based Selection in
Evolutionary Multiobjective Optimization. In: Spector L, Goodman ED, Wu A, Langdon
WB, Voigt HM, Gen M, Sen S, Dorigo M, Pezeshk S, Garzon MH, Burke E (eds)
Proceedings of the Genetic and Evolutionary Computation Conference (GECCO’2001): 283-
290, San Francisco, California, Morgan Kaufmann Publishers
Cox LA Jr (2009) Risk analysis of complex and uncertain systems. Springer Science & Business
Media, New York
Cox LA Jr (2012) Evaluating and Improving Risk Formulas for Allocating Limited Budgets to
Expensive Risk-Reduction Opportunities. Risk Anal 32(7):1244–1252
Daher S, de Almeida A (2012) The Use of Ranking Veto Concept to Mitigate the Compensatory
Effects of Additive Aggregation in Group Decisions on a Water Utility Automation
Investment. Group Decis Negot 21(2):185–204
Danielson M, Ekenberg L, Larsson A, Riabacke M (2014) Weighting under ambiguous
preferences and imprecise differences. Int J Comput Int Sys 7(1):105-112
Davis CB, Olson MH (1985) Management Information Systems: Conceptual Foundations,
Structure and Development. McGraw-Hill
de Almeida A, Vetschera R, de Almeida J (2014) Scaling Issues in Additive Multicriteria
Portfolio Analysis. In: Dargam F, Hernández JE, Zaraté P, et al. (eds) Decis. Support Syst. III -
Impact Decis. Support Syst. Glob. Environ. SE - 12. Springer International Publishing, pp
131–140
de Almeida AT (2013a) Processo de Decisão nas Organizações: Construindo Modelos de
Decisão Multicritério (Decision Process in Organizaions: Building Multicriteria Decision
Models), São Paulo: Editora Atlas
de Almeida AT (2013b) Additive-veto models for choice and ranking multicriteria decision
problems. Asia-Pacific J Oper Res 30(6):1-20
de Almeida AT, Almeida JA, Costa, APCS, ALmeida-Filho AT (2014b) A New Method for
Evaluation of Criteria Weights in Additive Models by Interactive Flexible Elicitation.
Working paper, CDSID
de Almeida AT, Costa, APCS, Almeida JA, Almeida-Filho AT (2014a) A DSS for Resolving
Evaluation of Criteria by Interactive Flexible Elicitation Procedure, In: Dargam F, Hernández
J, Zaraté P, Liu S, Ribeiro R, Delibasic B, Papathanasiou J. Decision Support Systems III -
Impact of Decision Support Systems for Global Environments”. LNBIP 184 (Lecture Notes
in Business Information Processing), Springer. 157–166
de Almeida AT, Vetschera R (2012) A note on scale transformations in the PROMETHEE V
method. Eur J Oper Res 219:198–200
Deb K, Pratap A, Agarwal S, Meyarivan T (2002) A fast and elitist multiobjective genetic
algorithm: NSGA-II. IEEE Trans Evol Comput 6:182–197
Dias LC, Climaco JN (2000). Additive aggregation with variable interdependent parameters: The
VIP analysis software. J Oper Res Soc 51:1070–1082
Dias LC, Clímaco JN (2005) Dealing with imprecise information in group multicriteria
decisions: A methodology and a GDSS architecture. Eur J Oper Res 160(2) 291–307
Duckstein L, Monarchi D, Kisicl CC (1975) Interactive Multi-Objective Decision Making Under
Uncertainty. Theor Decis Pract Hodder Stoughtor, London 128–147.
Eden C (1988) Cognitive mapping. Eur J Oper Res 36(1):1–13
Eden C, Ackermann F (2004) SODA. The Principles. In: Rosenhead J, Mingers J (eds) Rational
Analysis for a Problematic World Revisited. Second Edition, Chichester: John Wiley & Sons
Ltd.
Edwards W, Barron FH (1994) SMARTS and SMARTER: Improved Simple Methods for
Multiattribute Utility Measurement. Organ Behav Hum Decis Process 60(3):306–325
Edwards W, Miles Jr RF, Von Winterfeldt D (2007) Advances in decision analysis: from
foundations to applications. Cambridge University Press
Ehrgott M (2006) Multicriteria optimization. Springer Science & Business Media, Berlin
Erickson M, Mayer A, Horn J (2001) The Niched Pareto Genetic Algorithm 2 Applied to the
Design of Groundwater Remediation Systems. In: Zitzler E, Thiele L, Deb K, et al (eds)
Evol. Multi-Criterion Optim. SE - 48. Springer Berlin Heidelberg, pp 681–695
Fishburn PC (1976) Noncompensatory preferences. Synthese 33:393–403
References 85
Fonseca CM, Fleming PJ (1993) Genetic Algorithms for Multiobjective Optimization:

Formulation, Discussion and Generalization. In: Forrest S, (ed) Proceedings of the Fifth
International Conference on Genetic Algorithms, San Mateo, California. University of
Illinois at Urbana-Champaign, Morgan Kaufmann Publishers
Franco LA, Cushman M, Rosenhead J (2004) Project review and learning in the construction
industry: Embedding a problem structuring method within a partnership context. Eur J Oper
Res 152(3):586–601
Fraser NM, Hipel KW (1984) Conflict Analysis: Models and Resolutions. North-Holland, New
York
Goodwin P, Wright G (2004) Decision analysis for management judgment. Wiley London
Greco S, Matarazzo B, Slowinski R (2001) Rough sets theory for multicriteria decision analysis.
Eur J Oper Res 129:1-47
Greco S, Slowinski R, Matarazzo B (2002) Rough sets methodology for sorting problems in
presence of multiple attributes and criteria. Eur J Oper Res 138:247-259
Haimes YY, Hall WA, Freedman HT (1975) Multiobjective optimization in water resources
systems: the surrogate worth trade-off method. Elsevier
Hammond JS, Keeney RL, Raiffa H (1998a) Even swaps: A rational method for making trade-
offs. Harv Bus Rev 76(2):137–150.
Hammond JS, Keeney RL, Raiffa H (1998b) The hidden traps in decision making. Harv Bus Rev
76:47–58.
Hammond JS, Keeney RL, Raiffa H (1999) Smart choices: A practical guide to making better
decisions. Harvard Business Press
Horn J, Nafpliotis N, Goldberg DE (1994) A niched Pareto genetic algorithm for multiobjective
optimization. Evol. Comput. 1994. IEEE World Congr. Comput. Intell. Proc. First IEEE
Conf. IEEE, Orlando, FL, pp 82–87 vol.1
Howard RA (1992) Heathens, Heretics, and Cults: The Religious Spectrum of Decision Aiding.
Interfaces (Providence) 22:15–27
Jacquet-Lagréze E, Siskos J (1982) Assessing a set of additive utility functions for multicriteria
decision making, the UTA method. Eur J Oper Res 10(2):151-164
Keeney RL (1976) A Group Preference Axiomatization with Cardinal Utility. Manage Sci
23(2):140-145
Keeney RL (2002) Common Mistakes in Making Value Trade-Offs. Oper Res 50(6):935–945
Keisler JM, Noonan PS (2012) Communicating analytic results: A tutorial for decision
consultants. Decis Anal 9:274–292
Keith WH, Radford KJ, Fang L (1993) Multiple participant multiple criteria decision making.
IEEE Sys Man Cybern 23(4):1184-1189
Kersten GE (2001) Modeling Distributive and Integrative Negotiations – Review and Revised
Characterization. Group Decis Negot 10(6)493-514
Kersten GE, Noronha SJ (1999) WWW-based negotiation support: design, implementation and
use. Decis Support Syst 25:135–154
Kilgour DM, Eden C. (eds) (2010) Handbook of Group Decision and Negotiation, Advances in
Group Decision and Negotiation 4. Springer Science.
Kilgour DM, Keith WH (2005) The graph model for conflict resolution: past, present, and future.
Group Decis Negot 14(6):441-460
Kim SH, Ahn BS (1999) Interactive group decision making procedure under incomplete
information. Eur J Oper Res 116:498-507
Kirkwood CW, Corner JL (1993) The effectiveness of partial information about attribute weights
for ranking alternatives in multiattribute decision making. Organ Behav Hum Dec 54:456-
476
Kirkwood CW, Sarin RK (1985) Ranking with Partial Information: A Method and an
Application. Oper Res 33:38-48
Knowles JD, Corne DW (2000) Approximating the nondominated front using the Pareto
Archived Evolution Strategy. Evol Comput 8:149–172
Korhonen P (2005) Interactive Methods. Mult. Criteria Decis. Anal. State Art Surv. SE - 16.
Springer New York, pp 641–661
Korhonen P (2009) Multiple objective programming support Multiple Objective Programming
Support. In: Floudas CA, Pardalos PM (eds) Encycl. Optim. SE - 431. Springer US, pp 2503–
2511
Korhonen P, Wallenius J (2010) Interactive Multiple Objective Programming Methods. In:
Zopounidis C, Pardalos PM (eds) Handb. Multicriteria Anal. SE - 9. Springer Berlin
Heidelberg, pp 263–286
Leyva-Lopez JC, Fernandez-Gonzalez E (2003) A new method for group decision support based
on ELECTRE III methodology. Eur J Oper Res 148(1):14-27
Likert R (1932) A technique for the measurement of attitudes. Arch Psychol 22(140):1–55
Miettinen K (1999) Nonlinear multiobjective optimization. Springer Science & Business Media
Morais DC, de Almeida AT (2012) Group Decision Making on Water Resources based on
Analysis of Individual Rankings. Omega 40:42-45
Munda G (2008) Social multi-criteria evaluation for a sustainable economy. Springer, Berlin
Mustajoki J, Hämäläinen RP (2005) Decision Support by Interval SMART/SWING -
Incorporating Imprecision in the SMART and SWING Methods. Decision Sci 36(2):317-339
Nurmi H (2002) Voting Procedures under Uncertainty. Springer Verlag, Berlin-Heidelberg, New
York
Nurmi H (1987) Comparing Voting Systems. Dordrecht: D. Reidel Publishing Company
Osyczka A (1984) Multicriterion optimisation in engineering. Halsted Press
Pardalos PM, Siskos Y, Zopounidis C (eds) (1995) Advances in Multicriteria Analysis. Kluwer
Academic Publishers
Partnoy, F (2012) Wait: The Art and Science of Delay. Perseus Group books
Pawlak Z, Slowinski R (1994) Rough set approach to multiattribute decision-analysis. Eur J
Oper Res 72:443-459
Pedrycz W, Ekel P, Parreiras R (2011) Fuzzy Multicriteria Decision-Making: Models, Methods,
and Applications. John Wiley & Sons, Chichester
Polmerol J-C, Barba-Romero S (2000) Multicriterion Decision in Management: Principles and
Practice. Kluwer
Wesley, London
Rao S (1984) Multiobjective optimization in structural design with uncertain parameters and
stochastic processes. AIAA J 22:1670–1678
Rauschmayer F, Kavathatzopoulos I, Kunsch PL, Le Menestrel M (2009) Why good practice of
OR is not enough—Ethical challenges for the OR practitioner. Omega 37(6):1089–1099
Rosenhead J, Mingers J (eds) (2004) Rational Analysis for a Problematic World Revisited.
Second Edition, John Wiley & Sons Ltd.
Roy B, SáowiĔski R (2013) Questions guiding the choice of a multicriteria decision aiding
method. EURO J Decis Process 1(1-2):69–97
Roy B, Vanderpooten D (1996) The European school of MCDA: Emergence, basic features and
current works. J Multi-Criteria Decis Anal 5(1):22–38
Saaty, TL (1980) The Analytic Hierarchy Process. McGraw-Hill
Salo A, Hämäläinen RP (2001). Preference ratios in multiattribute evaluation (PRIME) -
elicitation and decision procedures under incomplete information. IEEE Sys Man Cybern
31(6):533-545
Salo A, Punkka A (2005) Rank inclusion in criteria hierarchies. Eur J Oper Res 163(2):338-356
References 87
Salo AA, Hämäläinen RP (1992) Preference assessment by imprecise ratio statements. Oper Res
40:1053-1061
Schaffer JD (1984) Multiple Objective Optimization with Vector Evaluated Genetic Algorithms.
PhD thesis, Vanderbilt University
Shakun MF (1988) Evolutionary Systems Design: Policy Making Under Complexity and Group
Decision Support Systems. Holden-Day, Oakland, CA
Shakun MF (2006) ESD: A Formal Consciousness Model for International Negotiation. Group
Decis Negot 15:491–510
Shakun MF (2010) Doing Right: Connectedness Problem Solving and Negotiation. In: Kilgour
DM, Eden C (Eds.) Handbook of Group Decision and Negotiation, Advances in Group
Decision and Negotiation 4. Springer Science
Simon HA (1955) A Behavioral Model of Rational Choice. Q J Econ 69(1):99–118
Simon HA (1960) The New Science of Management Decision. Harper & Row Publishers, Inc,
New York
Simon, HA (1982) Models of Bounded Rationality. MIT Press
Slack N, Chambers S, Harland C, Harrison A, Johnson R (1995) Operations Management,
Pitman Publishing, London
Slowinski R, Greco S, Matarazzo B (2012) Rough set and rule-based multicriteria decision
aiding. Pesq Oper 32:213-269
Sprague Jr RH, Watson HJ (eds) (1989) Decision Support Systems - Putting Theory into
Practice, Prentice-Hall
Srinivas N, Deb K (1994) Multiobjective Optimization Using Nondominated Sorting in Genetic
Algorithms. Evol Comput 2:221–248
Steuer RE (1986) Multiple Criteria Optimization: Theory, Computation, and Application. Wiley,
New York
Stewart TJ (2005) Dealing with uncertainties in MCDA. In: Figueira J, Greco S, Ehrgott M (eds)
Multiple Criteria Decision Analysis: State of the Art Surveys. Springer Verlag, Boston,
Dordrecht, London, 445-470
Thierauf, RJ (1982) Decision support systems for effective planning and control - A case study
approach. Prentice-Hall, Inc., Englewood Cliffs, New Jersey
Vetschera R, de Almeida AT (2012) A PROMETHEE-based approach to portfolio selection
problems. Comput Oper Res 39(5):1010–1020
Von Neumann J, Morgenstern O (1944) Theory of games and economic behavior. Princeton:
Princeton University Press
Wallenius J (1975) Comparative Evaluation of Some Interactive Approaches to Multicriterion
Optimization. Manage Sci 21(12):1387–1396
Weber M, Borcherding K (1993) Behavioral influences on weight judgments in multiattribute
decision making. Eur J Oper Res 67(1):1–12
Zitzler E, Thiele L (1999) Multiobjective evolutionary algorithms: a comparative case study and
the strength Pareto approach. Evol Comput IEEE Trans 3(4):257–271
Chapter 3
Basic Concepts on Risk Analysis, Reliability
and Maintenance
Abstract: Man’s level of dependence on equipment is increasing. This degree of

dependence requires high levels of availability, which has been changing the
impact that disruption of these systems causes. For many systems, an interruption
has consequences that go beyond the dimension of financial loss, thus justifying a
multidimensional consequence approach by using multicriteria (MCDM/A)
models. Thus, understanding the relationships between and among risk, reliability
and maintenance (RRM) is essential in order to offer more comprehensive
solutions to the various problems often treated in isolation from each other, and
which are the most important problems of the competitive market. This chapter
discusses fundamental topics about RRM, including tools for risk analysis and
hazard identification, concepts of reliability, maintenance techniques such as
RCM and TPM and eliciting expert’s knowledge. These topics are presented in
order to provide a basis for structuring different MCDM/A problems that are
addressed in several chapters. Some fundamental aspects could be used as input to
decision models in different forms such as attributes, objectives, criteria, and
problem context.
3.1 Basic Concepts on Risk Analysis
There are many concepts on risk found in the literature and also different
perceptions to it. However, if a decision is being made and risk is involved, then,
the risk concept should combine consequences and probabilities, incorporating the
DM’s preferences over that, as seen in Chap. 2.
Actually, a ‘decision process’ with no DM’s preference has no decision being
made, as discussed at the end of Chap. 1. Instead of that, that process either: a) has
some preference structure incorporated within the model, at random; b) is just
arbitrary following a previous decision of someone else.
Even so, in most of real cases, the consequences are multidimensional, and
therefore, require an MCDM/A approach for building a decision model. The
following topics are mainly based on the basic RRM literature and do not
incorporate the idea of decision support, as given in Chap. 2. That is, DM’s
preferences are not necessarily considered in the model.

90 Chapter 3 Basic Concepts on Risk Analysis, Reliability and Maintenance
3.1.1 Risk Context
In recent times, undertaking risk studies has become an increasingly complex task,
making it of great importance in different spheres of society. Modern world
facilitates access to information, making people more conscious of decision-
making on risk and its consequences in the social and environmental context. On
the other hand, organizations seek to manage appropriately all risks perceived as
being the most relevant ones in the production of goods or services to ensure that
their final product meets the minimum legal requirements, regulations and
resolutions as well as society’s expectations. However, it is of paramount
importance to emphasize that in the so-called real world, despite organizations
being concerned with identifying and monitoring risks, the restricted availability
of resources is a crucial point, which leads to some risks receiving special
attention with regard to the immediate allocation of resources, while others have
to wait until resources become available.
Although in the literature there are several definitions of the term risk, the basic
concept is associated with uncertainty in an environment and this is related to the
likelihood of an undesirable event occurring and the impact of its consequences.
This is why, according to Theodore and Dupont (2012), risk is defined as a
measure of financial loss or damage to persons, in terms of the likelihood of an
incident occurring and the magnitude of the loss. To Yoe (2012), risk is a measure
of the likelihood and consequences of uncertain future events. It is the chance of
an unwanted result where the lack of information about events that have not yet
occurred is one of the factors inherent in the chance of its happening. Cox (2009)
considers the preferences for consequences.
In the risk context, a change was recently observed by Aven (2012), who states
that traditionally dangerous activities were designed and operated from references
based on codes, standards and hardware requirements. However, what is verified
today is that this trend is more directed towards a functional orientation, where the
focus is associated with what it is sought to achieve. Therefore, the ability to
define risk is the key element in each functional system. Identifying and
categorizing risk are necessary to provide a decision support. The ability to define
what may occur in the future, to evaluate risks and uncertainties and to choose
among alternatives is what guides the decision-making process in the context of
risk.
As seen in Chap. 2, the risk concept in the decision process combines the
consequences with its probabilities, and incorporates the DM’s preferences over
that combination. Even so, in most real cases, the consequences are multi-
dimensional, and therefore, involve an MCDM/A approach, which may involve
tradeoffs as pointed out by Cox (2009), considering dimensions such as: financial,
reliability and health.
Risk Management, Risk Assessment and Risk Analysis are supposed to ensure
proper risk management and control, taking into account aspects such as
3.1 Basic Concepts on Risk Analysis 91
procedures, use of tools, approaches and models. Attention should be paid to

DM’s participation that directly impacts the final results of a risk study. This
requires communications about risks to be properly undertaken among the parties
involved. Finally, detailed analysis to be carried out in risk studies directly impact
the decision-making process. Different authors offer particular insights into these
aspects.
To Modarres et al. (1999), Risk Analysis can be defined as a technique for
hazard identification, characterization, quantification and evaluation. To Theodore
and Dupont (2012), Risk Assessment is the process by which degrees of risk are
estimated. Additionally, Yoe (2012) asserts that Risk Assessment is a qualitative,
quantitative or semi-quantitative systematic process that describes the nature,
probability and magnitude of risk associated with any substance, situation, action
or event that includes uncertainties. Effective risk management requires the
understanding of causes and conditions that contribute to the occurrence of an
undesirable event and to the improvement of the system (Paté-Cornell and Cox
2014).
Regarding to Risk Communication, Fjeld et al. (2007) describe that this is an
interaction process among stakeholders, risk assessors and risk managers. In this
context, the objectives (often set by law), procedures and best practices seek to
ensure that relevant aspects of risk analysis are identified by the stakeholders,
thereby ensuring adequate analysis and a correct understanding of the decisions
taken in relation to managing risk. In decision models, as seen in Chap. 1 and 2,
some of these actors’ role are related a DM.
On the topic of Risk Management, Yoe (2012) defines it as a process during
which problems are identified, information is requested and risks are evaluated,
and some initial definitions should be established to identify, evaluate, select,
implement, monitor and modify actions taken to change the risk levels from
unacceptable to the other two possible levels: acceptable or tolerable. To Aven
and Vinnem (2007), the purpose of risk management is to ensure that appropriate
measures are taken to protect people, the environment and assets from unintended
consequences, as well as to balance different interests, especially with regard to
health, safety, environment and cost. Risk management includes measures to
avoid hazards occurring and to reduce the potential damage from them.
Tweeddale (2003) states that there are three main requirements for risk
management: legal, commercial, moral (or ethical) requirements. Legal require-
ments will depend on the legal structure and the particular legislation in a specific
locality. Commercial requirements are associated with a range of commercial
implications such as loss of income due to production losses and costs related to
damage to equipment, injuries or deaths, environmental damage, legal actions, and
consequences for the company image. Moral or ethical requirements stress the
value of human life, bearing in mind that people’s health should not be measured
monetarily. These requirements draw attention to the complexity of risk and show
that risk has physical, monetary, cultural and social dimensions.
Although, different interests or requirements (criteria or objectives) are

mentioned above, it does not seem to be dealing with multidimensional con-
sequences, involving DM’s preference, which would require an MCDM/A
approach, as seen in Chap. 2. Really, integrating those dimensions (physical,
monetary, cultural and social) may represent a risky complexity, whereas a non
appropriate method is applied. This is identified by Aven and Vinnem (2007),
mentioning MAUT for two attributes, costs and fatalities, although they recognize
the difficulties of the elicitation process for obtaining the DM’s preferences. This
has to be evaluated in a case by case basis. All models have deviation, as seen in
Chap. 2, however, in the purpose of making them useful, the appropriate effort
should be made in the model building process. The successive refinement process
proposed in Chap. 2 may support this evaluation.
3.1.2 Public Perception of Risk
Society deals with risks in everyday life so much so that risk analysis is an
inherent characteristic of human beings. In daily routine activities, risk is always
present e.g. when walking in the street, using public transportation to work, eating
fatty foods, etc. Each person who participates in a hazard/risk analysis gives their
own opinion, memory, attitude and global view of the situation under study.
Moreover, these people are often affected by different types of personal biases
such as their level of education, beliefs, experience, culture, etc. Even experts
come to different conclusions when presented with the same data. The literature
examines the issues that arise by discussing different physical situations and
contexts.
van Leeuwen (2007) supports that perceptions of risk vary among individuals
and the general public, business and other stakeholders, and change over time and
in accordance with the prevailing culture. People continually assess situations and
decide if the risks associated with a particular action can be justified. In some
circumstances, dangerous effects are clearly associated with a particular course of
action. However, in other cases, the impact of each effect can be uncertain and not
immediately obvious.
To Modarres et al. (1999), the perception of risk often differs from the
perception of objective measures, thereby distorting risk management decisions.
Subjective judgments, beliefs and social bias with respect to events with low
probability and high consequence may affect how the results of risk analysis are
understood.
In this context, according to Crowl and Louvar (2001), the general public has
great difficulty with understanding the concept of risk acceptability. The major
problem is related to the involuntary nature of accepting a given degree of risk.
For instance, designers of chemical plants who specify a level of acceptable risk
assume that these risks are satisfactory to those living in the vicinity of the plant.
However, the neighborhood is often unwilling to accept any level whatsoever of

industrial risk especially if the community is aware of there having been an
accident involving a similar plant anywhere else in the world.
Additionally, Theodore and Dupont (2012) state that the lack of connection
between public and experts is of fundamental importance, when addressing the
question of why the public do not trust experts about these matters.
In view of these factors, it is important to pay attention to the fact that a
coherent risk analysis involves people’s perceptions about the risks under study,
and should take into account all aspects that may negatively interfere in the
process.
3.1.3 Risk Characterization
Risk characterization is another important aspect that should be undertaken. The

definition of aspects that directly influence this analysis, the establishment of
standards that ensure risk acceptability, tolerability and unacceptability are issues
that should be considered in risk characterization.
Thus, according to Tweeddale (2003), the nature of the assessed risk will
depend on the answer to two questions: (1) Will the undesirable event impact
people, the environment, property or production? (2) How will the effects of the
event be measured?
The MCDM/A approaches, as seen in Chap. 2, may answer these questions,
which deals with the measurement of desirability by DM’s preference over
multidimensional consequences.
Therefore, Theodore and Dupont (2012) states that risk characterization
estimates the risk associated with the process under investigation. The result of
this characterization is to determine the likelihood of adverse effects which will be
specified and enumerated arising from processes and/or leakages of substances
derived from the process.
According to Smith and Simpson (2010) there is nothing which presents no
risk. Physical assets always have failure rates and humans always make some
kinds of mistake. Hence, this arises up the need of establishing values that qualify
risks within a level considered acceptable by society. But, in practice, what does it
mean when one speaks of a risk being tolerable, acceptable or unacceptable?
Again, the MCDM/A approaches may deal with establishing values for risk,
bringing the DM to centre of the decision process, by means of incorporating
preferences within the model.
For Smith (2011), the term ‘acceptable’ means that the likelihood of fatalities is
accepted as reasonable, taking into account the circumstances and there being no
efforts made to reduce them. The term ‘tolerable’ implies that although prepared
for dealing with a risk level, an effort to tackle the causes of the risk is necessary
in order to reduce them. Cost is an aspect that should be taken into account in this
type of analysis. For Smith and Simpson (2010), the degree of risk considered as
tolerable depends on a number of aspects such as the degree of control under the
circumstances, the nature of risk analysis (intentional or unintentional), the
number of persons subject to risk, etc. Finally, the concept of intolerable risk
consists of not tolerating a specific risk level, thus not allowing activities to be
developed at this level. Additional comments with regard to these definitions can
be verified in the section dealing with ALARP concept.
To Crowl and Louvar (2001), it is impossible to eliminate any kind of risk
completely. At some point in the design stage, someone needs to determine
whether the risks are acceptable or not. In other words, are the risks under analysis
greater than the daily risks that individuals are subject to in their daily lives?
According to Modarres et al. (1999), risk acceptability is a complex and
controversial issue. However, making use of risk assessment results is a common
way to rank the exposure level of risk, where the risk exposure levels that are
socially acceptable should be defined based on risk acceptance thresholds.
In this context, some risk measures can be verified such as Individual Risk,
Societal Risk, Population Risk and Risk Indices. Each of these measures expresses
the risk, taking into account different aspects and contexts.
According to Smith (2011), Individual Risk refers to the frequency of a fatality
for a hypothetical person with respect to a specific hazard scenario, while the
Societal risk reflects the risk measure for a group of people, taking into account
multiple fatalities. Theodore and Dupont (2012) describe Population Risk as the
risk for the entire population, expressed as a certain number of deaths expressed as
thousands or millions of people potentially exposed to danger. Theodore and
Dupont (2012) also define Risk Indices, describing them as measures represented
by a unique number associated with a facility. Some risk indices are quantitative
while others are semi-quantitative, ranking risks in various categories. Risk
indices can also be quantitative average or benchmarkings based on other risk
measures.
In this context, Crowl and Louvar (2001) add that among these risk measures,
losses and accidents based on statistical data are relevant measures. However, they
should be considered with some caution, given that many of these statistics
represent an average, and do not reflect the occurrence of a specific accident with
potential losses. In contrast, no specific method is capable of measuring all aspects
simultaneously. Some of those commonly used are the OSHA incident rate, the
Fatal Accident Rate (FAR) and the Fatality Rate.
More specifically, according to Tweeddale (2003), FAR is a risk measure used
to assess the risks associated with the employees of an industrial plant. FAR is
defined as the number of fatalities, due to accidents at work, per 100 million hours
worked.
In conclusion, the definition of risk measures is necessary so that reference
values are established in risk studies and used in order that objectives are met
regarding monitoring and controlling risk.
3.1.4 Hazard Identification
Nowadays, identifying hazards is a critical factor to ensure that safety

requirements are satisfied, thereby attending to the need for assets, systems and
subsystems to function adequately. Moreover, hazard identification provides data
input for risk analysis in a particular production process (in part or in its entirety).
For better performance, hazards should be identified by using structured tech-
niques, and should involve experts and trained staff. What should always be taken
into account in the planning stage are restrictions on resources (i.e. financial
resources, experts, designers, operational and maintenance manpower, etc.) since
the availability of these will have a direct impact in the outcome of the analysis.
Zio (2007) states that the first step of hazard identification is the output of this
activity which is represented by a list of sources of potential hazards (i.e.
component failures, deviations in processes, external events, operational errors,
etc.) which have a non-zero probability of occurrence and can produce events with
significant consequences.
The methods developed in this step are usually those associated with a
qualitative analysis of systems and their functions, which will be included in a
framework of systematic procedures. Among these methods, FMEA (Failure
Mode and Effects Analysis) and HAZOP (Hazard and Operability Study) will be
highlighted.
3.1.4.1 FMEA (Failure Mode and Effects Analysis)
According to Zio (2007), FMEA is a qualitative method with an inductive nature,

which supports identifying failure modes of components that may disable the
system or initiate accidents that can have considerable consequences.
For FMEA in order to obtain data that is sufficiently detailed, information must
be collected from historical databases as must expert opinion. It is only by using
FMEA in this way that all aspects of a project and system critical components can
be verified. Further details regarding to FMEA, including FMECA (Failure Mode,
Effects, and Criticality Analysis), a derived technique, are given in Sect. 3.2.6.
3.1.4.2 HAZOP (Hazard and Operability Study)
According to Andrews and Moss (2002), HAZOP is a method that was first used
in the chemical industry, where industrial plants are evaluated with regard to
identifying potential hazards to operators and society. These hazards may arise in
a particular system and may be result of interaction among different systems of the
industrial process.
According to MacDonald (2004), HAZOP presents well-defined stages, as

shown in Table 3.1:
Table 3.1 HAZOP stages
HAZOP stages Details

Stage 1: Defining the scope and objectives;
Defining the process Establishing responsibilities;
Forming the team.
Stage 2: Defining planning and implementation schedule;
Preparation Data collection;
Registration methodology;
Stage 3: Systems division;
Verification Identifying deviations;
Establishing causes, consequences and setting protection measures;
Reaching consensus on the actions;
Repeating activities for each element evaluated.
Stage 4: Defining spreadsheets registration;
Registrations and monitoring Preparing reports;
Monitoring actions;
Re-assessing HAZOP periodically;
Producing and distributing final report.
HAZOP is used to identify and assess hazards in production and maintenance

operations. In addition, multidisciplinary teams and expert opinion must be used
in preparatory studies associated with this methodology and the scope and
objectives of projects must be well established. Moreover, people involved in the
process must have a good understanding of the particular terminology. Deviations,
guide words and project intent are some of the terms used.
According to Ericson (2005) some of the disadvantages of HAZOP that have
been reported include: focusing on single events without considering the
combination of more than one event; focusing on specific guide words can result
in some dangers that are unrelated to these guide words not being valued; HAZOP
analysis can be too much time and resources consuming.
According to Zio (2007), while FMEA is mainly based on the structural aspects
of a system, HAZOP processes focus on the plant under analysis.
3.1.5 FTA (Fault Tree Analysis)
The Fault Tree (FT) is a tool widely used in industrial processes within the risk
environment. It can be classified as a qualitative or quantitative tool depending on
the availability of the likelihood values of failure events.
According to Ericson (2005), FTA is defined as a structured deductive

technique which is used to analyze a system so as to identify and describe the root
causes and the likelihood of the occurrence of a particular undesired event. FTA is
applied to evaluate dynamic complex systems, in order to understand and prevent
potential problems. The development of the tree is an iterative process that can be
used preventively or reactively (in this case, after failures have occurred).
The FT is a graphical model built from a top event, also known as an unwanted
event. It is structured in such a way as to identify and combat all possible relevant
causes (root causes) of the event linked with the top event.
This tool can be used in both preventive manner (mitigation) and corrective
manner. The elimination of all root causes produces the elimination of the top
event. Similarly, the elimination of only some root causes results in reducing the
probability of the top event.
According to Andrews and Moss (2002), the fault tree diagram shows two
basic elements: gates and events (both represented by specific symbols depending
on the context). The relations amongst FT events occur through logic gates that
enable or inhibit the passage of failures along the tree, thereby showing the
relations necessary for another event at a top-level of the tree to occur. For each
gate there is a specific gate symbol, a gate name and valid causal relation. The
gates most commonly used are AND and OR gates. For example, the existence of a
gate AND means that the output event occurs if all input events occur
simultaneously (since there are at least 2 input events). On the other hand, the
existence of the gate OR means that the output event occurs if at least one of the
input events occurs simultaneously (since there are at least 2 input events). An
FTA example that shows a top event, AND and OR gates and basic causes, also
known as root causes, is given in Fig. 3.1.
Top Event
AND
gate
Basic OR
event gate
Basic
Basic
event
event
Fig. 3.1 FTA example

FTA is a technique to assist the estimation of failures likelihood (Nwaoha et al.

2013). When FTA is applied as a quantitative approach, the value of the likelihood
of the occurrence of the top event is obtained based on the specific Boolean
properties of the gates.
More specifically, an important matter to be noted is that a FMECA failure
mode can be considered as an input to an FTA top event. Thus, each specific
FMECA failure mode is a top event of a specific FT.
3.1.6 Event Tree Analysis (ETA)
According to Ericson (2005), ETA (Event Tree Analysis) is an analytical

technique to identify and evaluate sequences of events in a potential accident
scenario arising from the occurrence of an initiating event. ETA uses a logical tree
structure known as an event tree (ET). The purpose of ETA is to determine
whether the initial event will unfold in a series of unwanted events or if the event
is sufficiently controlled by security systems and procedures established during
the system design phase. ETA can generate several different results from one
initial event, thereby allowing a specific likelihood for each outcome.
According to Bedford and Cooke (2001), the ET structure starts with an initial
event propagating this event through the system under consideration, taking into
account all the possibilities that can affect the behavior of the system/ subsystem.
ET nodes represent the possible operation (or non-operation) of a system/
subsystem. More specifically, the ET pathway that results in an accident is called
an accident sequence. An example of an ET is shown in Fig. 3.2 (Brito and
Almeida, 2009).
According to Ericson (2005), ETA can be used to model a system entirely,
comprising subsystems, components, software, procedures, environment and
human error. It can also be used at different stages such as the project design
phase, and has been applied to different systems such as nuclear power, aerospace
and chemical plants.
An analyst should guide the ET construction process by identifying and
evaluating all possible outcomes resulting from an initial event. A positive aspect
is that if applied in early stages, ETA helps to identify system security issues, thus
avoiding corrective actions (Andrews and Dunnett 2000).
Initial Immediate Delayed Area Possible

event ignition Ignition confined scenario
yes
Scenario 1
yes
no
Scenario 2
Gas release
(Rupture)
yes
Scenario 3
yes
no
no Scenario 4
no
Scenario 5
Fig. 3.2 Illustrative example of event tree applied to the risk analysis of a pipeline
Regarding the events that comprise ET accident sequences, Zio (2007) states
that they are characterized by: intervention (or not) of protection systems that
should come into operation (or not) to mitigate the accident (System Event Tree);
the running (or not) of security functions (Functional Event Tree); and the
occurrence (or not) of physical phenomena (Phenomenological Event Tree).
According to Zio (2007), these event trees types are applied in different
contexts:
x System Event Tree – this is used to identify accident sequences that have
developed within a plant, involving protection and security systems;
x Functional Event Tree – this is an intermediate step when constructing the
System Event Tree. From the ET initial event, safety functions that need to be
established are identified, and are subsequently replaced by the corresponding
protection and security systems;
x Phenomenological Event Tree – this describes the evolution of a pheno-

menological accident that occurs outside the plant (fire, dispersion ...).
Finally, the integrated use of tools can also be checked in event trees where
the Fault Tree (FT) quantitative approach is applied to obtain a value for the
likelihood that a failed state will occur in any given branch of the ET. An example
is shown in Fig. 3.3. Andrews and Dunnett (2000) presents a comparative analyses
considering ETA and FTA.
Initial Event 1 Event 2 Event 3 Possible

event scenario
Success
Success Scenario 1
Fail
Scenario 2
Initial event
Success
Scenario 3
Success
Fail
Fail
Scenario 4
Fail
Scenario 5
Top Event
AND
gate
Basic Basic
event event
Fig. 3.3 How integrated tools (FTA and ETA) are used to determine failed states
The likelihood of an ETA failure is the same of a top event obtained from FTA,
implemented for each specific failure observed in the ETA. The likelihood of
success is calculated as being the complement of this failure likelihood.
Otherwise, P(Success) = 1 – P(Fail).
3.1.7 Quantitative Risk Analysis
Risk analysis techniques are devoted to supporting managerial decisions regarding

risk reduction in order to achieve and maintain tolerable risk levels and therefore
assuring safety.
According to Vinnem (2014) the abbreviation QRA is also used for Quantified
Risk Assessment, and the context of the analysis defines which of these terms are
more suitable. When an evaluation of the results is combined with the risk
analysis, the term assessment should be used. This nomenclature and the term
QRA are well established for offshore operations and oil and gas and chemical
processes. They are also referred to as Quantitative Risk Assessment (QRA),
Probabilistic Risk Assessment (PRA), Probabilistic Safety Assessment (PSA),
Concept Safety Evaluation (CSE) and Total Risk Analysis (TRA), although the
nuclear industry for example, adopts the terms Probabilistic Risk Assessment or
Probabilistic Safety Assessment (Bedford and Cooke 2001; Vinnem 2014). Some
authors consider that all these terms have almost the same meaning as the tools
considered converge in order to be a scientific analysis of risk.
According to Vinnem (2014), Norway was for many years the only country that
required QRA studies systematically. Norway started doing so in the 1980s.
However, it took the UK almost 10 years before legislation was introduced that
laid down the need for QRA studies, namely when official inquiries due to the
Piper Alpha platform accident in 1988 recommended the adoption of QRA in the
UK similarly to Norway which had done so ten years earlier.
When dealing with risk analysis there are many systematic techniques such as:
x Hazard and Operability Study (HAZOP);
x Safety and Operability Study (SAFOP);
x Safe Job Analysis (SJA);
x Preliminary Hazard Analysis (PHA);
x Failure Model and Effect Analysis (FMEA);
x Quantitative Risk Analysis (QRA).
Despite QRA, most of these approaches are essentially qualitative, although it
is possible to incorporate quantitative information and be performed in a semi-
quantitative way.
However, to perform a QRA, it is necessary initially to identify hazards and
describe risks to personnel, environment and assets in a quantitative manner.
Although the identification of hazards may be obtained from a qualitative study,
the initiating events are evaluated in a quantitative perspective, leading to the
analysis of the causes in terms of probability to estimate the probability of each
scenario.
According to Vinnem (2014), for each scenario, estimates are made of

consequences, effects, facility responses and associated probabilities, which
enables consequences to be quantified in terms of personnel environment and
assets, which represents losses in human, environmental and financial dimensions.
In the MCDM/A approaches, scenarios may be related to the state of nature (T),
which is associated to the probability S(T). Also, the use of MCDM/A approaches
enables multidimensional consequences to be quantified.
Vinnem (2014) describes QRA in five steps, represented by Fig. 3.4. The first
two steps in Fig. 3.4 are mainly qualitative. First of all, events are identified which
may also be called hazard identification (HAZID), and this requires that all
possible hazards and sources of accidents should be investigated to avoid
neglecting any source of accident. During this screening, levels that shall be used
to classify critical and non-critical hazards are defined, providing reports that
register the evaluations made to classify each hazard, in order to have a register of
the reasons why and a demonstration of how a hazard was classified as non-
critical, while assuring that it was safe to state that these hazards were not
considered as critical.
Identification of
Critical Events
Coarse
Consequence
Analysis
Quantitative
Cause Analysis
Detailed
Consequence
Analysis
Risk Calculation
Fig. 3.4 QRA steps

Considering the tools available for hazard identification, such studies are
usually supported by the use of checklists, statistics on failure, a database of
accidents, HAZOP studies and similar risk analysis studies. The experiences
obtained from similar projects are also an important source used to identify
hazards.
After identifying the critical hazards to be considered, it is necessary to identify
the causes of these hazards and which events may lead to an accident scenario
occurring. Identifying the starting point for a potential accident enables the chain
of events that may cause an accident to be established.
During the analysis of the cause analysis (the third step), it is determined which
causes may lead to the initiating events in order to support the assessment of the
probabilities of initiating events. From the cause analysis, it is possible to identify
risk reducing actions that would prevent or interrupt the chain of events that may
cause an accident. In the initial steps of the cause analysis, qualitative techniques
are usually deployed followed by quantitative approaches if there are data
available for quantification. Qualitative approaches are used to identify causes and
conditions for initiating events, thereby establishing the basis for a possible
subsequent quantitative analysis. With regard to the techniques used to identify the
causes, there are: HAZOP, Fault Tree Analysis (FTA), Preliminary Hazard
Analysis (PHA), FMEA and human error analysis techniques, which are also used
in traditional reliability analysis.
Quantitative studies in cause analysis are conducted in order to establish the
probability of the occurrence of initiating events, while using historical statistics to
calculate the frequency of initiating events is one of the most common approaches.
The fourth step in Fig. 3.4 is related to the consequence analysis of accident
scenarios. A consequence analysis considers the existence of barrier functions and
elements to contain hazards and the accident sequences in order to evaluate the
possible function or failure of barriers involved. According to Vinnem (2014), fire
and explosions are two of the main factors evaluated, and both may be assessed by
using the same calculation steps for all scenarios which may involve fire and/or an
explosion. These steps depend on which conditions and sequences are related to
the factors evaluated. Fire and explosions may be a result from a leakage,
punctures or pinhole, any of which may expose a hazardous material which when
associated with a chain of events may result in a Vapor Cloud Explosion (VCE), a
Boiling Liquid Expanding Vapor Explosion (BLEVE), a Flash Fire, a Jet Fire, and
so forth. Thus, the steps of the calculation are used to estimate the amount of
material leaked by considering temperature and pressure conditions in the system
associated with system barriers and mitigation actions. TNO’s Colored Books
present systematic procedures to assist QRA studies, especially consequence
analysis, regarding the estimation of thermal radiation, ignition probabilities,
conditional probabilities for fatalities, damages and other consequences.
Regarding fatalities, probit functions are usually used to calculate the probability
of death due to exposure to toxic substances and / or heat radiation at a given level
of exposure.
The results from a QRA study arise from the risk calculation, the last step of
Fig. 3.4. These results are usually compared and associated with risk tolerance
level. QRA studies are usually performed until barriers and safety actions are
strong enough to assure that any risk is above reference levels. A QRA study aims
to provide a risk picture, which results from the hazard identification, and a cause
and frequency analysis that are combined to express the risk level associated with
all critical hazards.
Although the terms risk calculation, risk analysis and risk assessment can be
easily misunderstood since they have the same general meaning, there are
differences related to the scope of each term. Risk calculation uses information
from consequence analysis and cause analysis thereby providing a risk level
calculated from frequencies and the magnitudes of consequences. While risk
analysis refers to the entire process described in Fig. 3.4, which includes the risk
calculation, Risk assessment is the entire process of risk analysis when the results
are evaluated regarding risk reference levels, which are defined by considering a
notion of risk tolerance.
To ensure the reliability of the results of QRA studies, there are several factors
that must be considered, such as:
x The technical description of the system (activities, operational phases);
x Purpose and target of risk analysis;
x Activity levels on the installation;
x Operation of safety systems;
x Study assumptions: how these are verified and accepted;
x Data Sources.
Thus, QRA is a systematic development of numerical estimates of the expected
frequency and/or consequence of potential accidents associated with a facility or
operation.
According to Arendt and Lorenzo (2000), there are two main misconceptions
about QRA which are to do with the lack of adequate data on equipment failure
and the cost of conducting QRA, i.e. whether it is cheap or expensive.
Regarding the availability of data, there are industry wide databases that can
provide data for frequency rate estimates and regulation authorities that provide
periodical reports, such as:
x The Guidelines for Process Equipment Reliability Data with Data Tables;
x IEEE Guide to the Collection and Presentation of Electrical, Electronic,
Sensing Component, and Mechanical Equipment Reliability Data for Nuclear
Power Generating Stations (IEEE Std 500);
x The OREDA Offshore Reliability Data Handbook;
x Non-electronic Parts Reliability Data 1991 (NPRD-91) and Failure
Mode/Mechanism Distributions 1991;
x Systems Reliability Service Data Bank;
x Nuclear Plant Reliability Data System: Annual Reports of Cumulative System

and Component Reliability;
x Offshore Blowouts Causes and Control;
x UK Health and Safety Executive Reports.
The accuracy of the results is a function of the resources deployed in the
analysis. As the quality of the input into the model improvement, the results will
become more accurate. Thus the availability of resources is the primary constraint
for the quality of QRA results. There is a need to perform a cost-effective analysis,
so managers (or DMs) may balance the value of QRA results compared to the cost
of having such results. Thus, over the years QRA has been considered very cost-
effective.
The QRA results do not show if an installation is safe or unsafe, but they give a
risk picture that has to be evaluated in a risk assessment context. A DM must
decide whether to seek changes and safety improvements in order to reduce risk,
or even if the benefits of these safety improvements would justify the cost of
making them. That is, tradeoffs regarding multiple criteria should be made, which
should be based on MCDM/A methods.
Typically, QRA results report risk in terms of its consequences per year. If
there is an analysis of the consequence of a human dimension, the report will
contain a risk result of the expected number of fatalities and/or injuries per year or
per hours of equipment operation.
If the analysis is with regard to environmental consequences, the report shall
contain risk results in terms of the expected amount of chemical substances spilled
and the extent/size of the affected area on the same basis as for the human
consequences.
The next section tackles risk tolerance, which are used for Risk Assessment in
order to evaluate if a facility or installation is safe or unsafe according to objective
minimum risk targets.
3.1.8 ALARP
According to Bedford and Cooke (2001) the ALARP (As Low as Reasonably
Practicable) principle has guided the setting of tolerance risk levels since the
1980s and 1990s, in order to achieve safety goals. The USNRC policy statement
(NRC 1986) and the UK tolerability of risk document (HSE 1987) seek to convert
the principle of setting this ‘as low as reasonably practicable’ (ALARP) into a
numerical definition, which establishes upper levels for risk intolerance and lower
levels at which risks can be considered as tolerable.
Given that criteria for risk acceptance are generally combined with risk
analysis, some industries and countries have regulations that require such criteria
to be defined prior to the risk analysis (Aven 2012).
In a more practical perspective, ALARP can be understood as a risk goal to be

achieved in order to define investments in safety. Most of the safety standards
indicate that risk evaluation should be conducted until safety improvements result
in tolerable risk levels being reached. One example is the ISO/IEC: Guide 51.
Sutton (2010) describes the idea behind the concept of ALARP as being that
risk should be reduced to a level that is as low as possible without requiring
excessive investment, thus establishing a numerical boundary that determines
whether a risk is definitely acceptable or definitely not acceptable.
Given the tradeoff inherent in considering costs and safety together, Bedford
and Cooke (2001) point out that using the ALARP principle reduces the tradeoff
between safety and costs so as to increase safety by implementing what is
reasonably practicable. However, discussion about the value for a human life is
always an issue that leads to hotly-debated argument. ALARP is usually applied to
support the definition of tolerable limits for human losses.
On using the ALARP principle, it is possible to classify risks into three
categories: negligible risk, tolerable risk and unacceptable risk (Macdonald 2004):
x Negligible risks are risks that fit into a category of being broadly acceptable by
most people in their daily lives. This class of risk considers situations such as
being struck by lightning or having a brake failure in a car;
x Tolerable risks are those risks that one person would rather not have. However
they are deemed to be tolerable in view of the gains obtained by accepting this
situation. For this type of risk the inconvenience in terms of burdens are
balanced against the scale of risk. Thus, a compromise is accepted. An example
of this situation is when a person decides to drive a car or travel by bus.
Usually, in these situations, people accept that accidents can happen but try to
avoid them by minimizing the chances of having an accident;
x Unacceptable risks are those that are at a level of risk that is too high to accept,
and therefore are unacceptable; in other words, they have a tolerance level of
zero. The losses regarding such risks are so high that they cannot be compared
with any possible benefit arising from any situation where there is exposure to
such risk.
The ALARP principle may be understood in the context of MCDM/A methods,
considering the intra-criterion evaluation. This is similar to the constructed
criterion, discussed in Chap. 2. Also, in this kind of approach may be related to a
sorting problematic, in which the consequences or alternatives are classified into
categories.
So this principle is used to guide hazard and risk analysis by setting tolerable
risk goals to be achieved in any hazardous situation. Usually this is the first step
for any assessment of a safety system.
ALARP risk regions can be illustrated as shown in Fig. 3.5, which presents
each risk region according to tolerable risk levels. This is also called a carrot
diagram, which is presented in most of the related literature.
Risk Magnitude
Intolerable The risk of a fatality

region risk is usually higher
than 10 E-4.
ALARP or
Tolerable region
Acceptable The risk of a fatality is

region usually lower than 10
E-6.
Fig. 3.5 ALARP principle: tolerance limits
The definition of ALARP regions are based on everyday risks. Thus, risks that
are considered typical and commonly expected may include risks from all causes,
including bad health.
To measure the risk level, the Fatal Accident Rate (FAR) is used with
particular regard to the employees of some hazardous installations who are usually
exposed to higher risks than those working in less hazardous workplaces.
Aven (2012) adds that in practice the value considered to reflect risk is an
estimation of FAR or the probability p of a certain accident event, since the true
value of FAR or p is unknown. Thus using tolerance limits means to compare an
estimated value with acceptable values. This means that using a best-estimate
approach may not produce clear recommendations, and thus standardized models
and input data may be required. Thus, the acceptance level is a function of such
models and input data.
According to Tweeddale (2003) in some cases there is a subjective opinion and
a potential debate about whether ALARP standards are achieved and this can lead
to such issues being questioned in court. Nevertheless, if the hazardous installation
uses the best technology available and can be set up, and it also uses the best oper-
able and maintainable management systems in order to improve safety by keeping
the equipment maintained to high standards, the risk is usually an ALARP one.
There are some criticisms regarding the use of ALARP as a way to justify risk
exposure. In addition, there is a problem with the term regarding ‘acceptable risk’.
This is because it is commonly used by those who generate the risk to excuse the
fact that others will be exposed to it. And this is why some authors call the very
concept of ALARP into question. Tweeddale (2003) remarks that the level of risk
that an individual accepts is particular to that individual. That is, it is not an
applicable standard to any individual. Moreover, what is regarded as an

‘acceptable risk’ may change over time. It is worthwhile to note that, in the case of
MCDM/A context, since a particular DM’s preference is considered, its result may
not also be an applicable standard to any individual.
Tweeddale (2003) argues that instead of using the term ‘acceptable risk’, terms
such as ‘accepted risk’ or ‘approved risk’ should be used. The former would
denote that the individual involved would wisely or unwisely accept the risk,
independently of whether it is considered low or high compared with everyday
risks. The latter would be used to address exposure to risk that complies with rules
or standards set by an appropriate statutory authority or regulator (that is, a DM)
on behalf of the general community. In this case, the regulator would define what
approved risks are even if those risks were higher than the everyday risks that an
individual is exposed to.
Such risks may include those from many causes that can result in a fatality such
as (Tweeddale 2003):
x Smoking;
x Swimming;
x Travelling by motor vehicle;
x Traveling by train;
x Accidents at home;
x Pedestrian struck by a vehicle;
x Homicide;
x Accidental poisoning;
x Fires and accidental burns;
x Electrocution (non-industrial);
x Storms and flood;
x Lightning strikes;
x Snake bite.
Some individual risks are exemplified in Table 3.2 as FAR and probability
values (Macdonald 2004).
Table 3.2 Example of individual risk and FAR based in UK data
Activity FAR per 108 Individual risk of

death per person
per year x 10-4
Travel
Air 0.02
Train 3-5 0.03
Car 50-60 2
Occupation
Chemical industry 4 0.5
Agriculture 10
Rock Climbing 4,000 1.4
Staying at home 1-4
It is important to remark that these everyday risks may change from country to
country; for example, even in the same country, some regions may have a
significantly higher homicide rate than the rest of the country. Therefore, it is
possible that a given risk level could be considered tolerable in an undeveloped
country and intolerable in a developed country. Thus, the values given by
Macdonald (2004), with some examples given in Table 3.2 reflect the reality of a
developed country, and these might be lower than in undeveloped countries.
3.1.9 Cost-Effective Approach to Safety
After assessing the probability of hazardous events, all possible actions must be
deployed in order to achieve a tolerable risk level. In fact, if the risk to life is so
high that is beyond economic concern, the equipment or plant must be considered
safe, otherwise it must be closed.
However, when the tolerable risk level is reached, investments in risk reduction
are only justified by a cost effective evaluation. Thus, when a risk level is
considered tolerable or an ALARP one, any costs to improve safety must be
followed by a compatible benefit. Otherwise, it should not be implemented.
Usually the cost per life saved with a previously established level is considered.
Aven (2008) defines a similar measurement as the implied value of a statistical
life or the implied cost of averting a fatality, by dividing the cost of the safety
improvement by the number that represents the expected reduction in the number
of fatalities. Therefore this ratio can also be considered by quantities other than
lives saved, if for example, an environmental risk is being considered, the
reference may be to tons of oil spilled.
This allows expending resources in order to improve safety by acting where
one can find the greatest benefits while taking the budget allocated to improve
safety into account. These costs vary according to the type of system, complexity
and regulatory standards regarding the activity. Enterprises usually avoid
disclosing data on levels of cost per life saved. According to Smith (2011), this
value is between £500,000 to £4,000,000, while if the risk has potentially multiple
fatalities, then higher amounts may be considered.
Thus, the more that the number of potential fatalities increases, the more risk
averse the analysis becomes, which leads to choosing a higher cost per life saved
level. It is valuable to observe that utility theory provides an axiomatic structure to
evaluate DM’s behavior regarding risks, including risk aversion, as it seen in
Chap. 2.
As examples of how these values are considered, Smith (2011) points out that
for passenger road transportation, there is a voluntary aspect to the exposure and a
small number of casualties per incident, so the value considered for cost per life
saved is approximately £1,000,000. For the transportation of dangerous material,
where the risk is not under an individual´s personal control which means that there
is an involuntary risk, Smith (2011) presents a cost per life saved of approximately
£2,000,000 to £4,000,000. When considering multiple offshore fatalities, where
there are a large number of fatalities and no personal control by the victims, Smith
(2011) shows that the cost per life saved can be between £5,000,000 to
£15,000,000. Therefore, these values are quite controversial and may change when
they came under scrutiny in the media or are reported as catastrophic accidents,
thereby making the analysis even more risk averse.
Smith (2011) states that the maximum tolerable risk for a single fatality does
not always coincide with the societal risk calculations. Thus, while societal risk
measures the frequency of a fatal event, when considering individual risk, it is the
frequency of individual deaths that is considered. One of the main differences
between estimating individual risk and societal risk is about whether the risk is
voluntary or involuntary. When considering individual risk it is important to
highlight that these individuals are voluntarily exposing themselves to risk, in a
specific place that sets specific conditions for the frequency and risk assessment.
When considering societal risk, what is considered is the involuntary exposure to
risk that may reach random individuals, and it characterizes this concept of an
involuntary risk.
According to Tweeddale (2003) there is no unanimous formal agreement
regarding a specific value that can be considered as denoting a tolerable level of
risk. But in many countries it is typical to consider that an additional risk of 1
chance in a million per year (10-6 per year), due to industrial sources affecting the
person most exposed to these, is a very low risk level compared to everyday risks
that an ordinary person is usually exposed to without questions being raised about
this. Aven (2008) points out that the probability of a fatality for a third person
associated with exposure to risk in an industrial plant is required to be less than
10-5 per year. Therefore, the value that defines whether a risk should be considered
tolerable and therefore accepted by the wider community is the Individual Risk
level. Some of these everyday risks have been exemplified.
When calculating individual risk the focus must be on an event in which one
specific person is seriously injured or killed. Aven (2008) defines individual risk
as the frequency of death for the person or critical group of personnel most at risk
from a given activity due to their location, habits or periods that make them
vulnerable. Thus, individual risk is measured as the annual frequency of an
accident with one or more fatalities over a homogeneous group of people who
voluntarily expose themselves to risk. This is an approximation of the probability
that a random person of a group who conducts a specific voluntary activity will be
killed while he/she is at the industrial facility over the course of the time period
considered, usually a year. This measurement is used to calculate the FAR.
As to the risk to any individual who is involuntarily exposed to some risk,
consideration has to be given to the possibility that more than one person may be
killed due to that risk source. Thus societal risk cannot be measured only by
individual risk, but must include the possibility that there may be 1 to N fatalities.
The more the number of fatalities increases, the more risk averse the analysis is.
Societal risk is usually represented by F-N curves, which show the frequency of
accident events with at least N fatalities.
Tweeddale (2003) recognizes the controversy regarding attempts to put a value
on a human life, since human life can be considered priceless due to the emotional
values that money cannot compensate for. Nevertheless, there is a need to
establish a limit for the amount that can be spent per life saved, otherwise it is
impossible to decide if a choice can be made on economic grounds between
improving safety in order to keep an industrial plant running or closing the plant.
According to Tweeddale (2003), one absolute limit to be established for cost per
life saved would be obtained by dividing the annual gross national product by the
annual number of births. This value represents the amount that may possibly be
spent in order to extend the life expectancy of each new-born baby, if there are no
other expenses in the community. As there are many other requests for financial
resourcing from the wealth derived from the community, the real value of the limit
to the cost per life saved would be less. Therefore the definition of this value will
depend on particular priorities and other characteristics of the problem, such as
those pointed out in the examples which discussed calculating risk values to do
with the transportation of passengers by road, the transportation of dangerous
materials and substances and multiple offshore fatalities.
3.1.10 Risk Visualization
Risk Visualization is a tool used to produce images of risk (i.e. 3D visualization

and risk rich pictures) in order to illustrate and facilitate the risk perception by any
actor (DMs, managers, users, etc.) in a decision making or managerial process.
This subject integrates the concept of information visualization.
The risk visualization may be applied to visualization in risk management
framework, considering visualization in risk identification, visualization in risk
analysis, visualization in risk assessment, visualization in risk communication and
visualization in risk reduction. This support can provide processed information
and better control to making the more appropriate decision making for these
previous modules.
Additionally, the interaction among risks (when it happens) is an important
question that should be treated in risk visualization in decision making process. It
allows a more complete risk appreciation (Ackermann et al. 2014).
According to Bostrom et al. (2008), understand how risk representations affect
judgments and decision making is essential to comprehend the risk management
and the decision-making process. Therefore, graphical representations of risk
seeks to simplify some concepts and constraints related to mathematical, chemical
or physical aspects, making risk management and decision making more
comprehensible to the public (Ale et al. 2015).
In general, the information visualization in risk management process can aid

the perception and understanding of the risk and its several aspects. Therefore, the
risk visualization may be applied in several modules of the risk management
framework, such as in risk identification, risk analysis, risk evaluation, risk
assessment, risk communication and risk reduction. This support can provide
processed information and better control to making the more appropriate decision
making for these previous modules mentioned.
Eppler and Aeschimann (2009) and Horwitz (2004) highlight that the
visualization in risk management is still not a frequent topic in organizations,
probably because there is a difficult to describe and visualize the risk.
Some insights can arise of the answers to the following questions:
x How can the information visualization aid in the various steps of risk
management? For instance, can the information visualization improve the
performance in the risk identification module? Can the information
visualization support the determination of the likelihood and the consequences
estimation?
x How is it possible to handle differences of knowledge and skills, through the
risk visualization, among the various users of the system?
Al-Kassab et al. (2014) emphasize that the way in which information is
‘framed’ and communicated not only helps in interactive decision process, but
also provides a means of knowledge creation. Based on the literature review, they
summarize the information visualization process in five steps: 1) Raw data
collection; 2) Data transformation; 3) Data warehouse; 4) Visual transformation;
5) Viewer interaction.
Firstly, it is need to collect of the quantitative and/or qualitative data from
different sources and store them in one place (database). Based in this set data
collected, is necessary to transform and comprises these data. Then, it is necessary
a visual transformation by mapping of the transformed data, and, therefore, the
creation of a new ‘picture’ of the information that can be seen by DM, through of
visual/graph structures (graphs, tables, maps, etc.). Lastly, the DM can interact
with these visualization structures, allowing the transformation process at different
stages of decision making. Furthermore, DM can adjust their view on the data,
change the visual structure, or even affect the data transformation.
Moreover, Al-Kassab et al. (2014) identified three fundamental managerial
functions of information visualization: a communication medium, a knowledge
management means, and a decision-support instrument. These functions also can
be contextualized within each module of the risk management framework.
The use of visualization as a communication medium function is linked with
the knowledge-based processes, through of the patterns identifying, correlations,
outliners, clusters data and other techniques, mainly when there is a big data. It
adopts several display techniques and approaches aiming to elaborate and analyze
data allowing ‘transmission’ of messages to be interpreted by DM and by

stakeholders. Furthermore, the knowledge created by the information visualization
itself should be shared and interpreted by the DM. Hence, it is essential in order to
make a coherent risk evaluation, risk perceptions, preventive and mitigation
actions or other strategic actions linked to risk management. This information
should be communicated, understood, shared and implemented for all of the
organization, or for all that suffers the impact of the risk.
The information visualization, as a function of a knowledge management
means, can facilitate or obstruct the human brain’s capacity to interpret
information (Al-Kassab et al. 2014). Also, it is highlighted that the information
visualization must take into account the context and purpose of the knowledge
because this interpretation is affected by knowledge and cultural background of
the DM in the risk context. It is important to note that several times in risk
context, the knowledge acquired in risk management is affected harmfully by
absence of information and database.
On one hand, the risk perception is linked with the past experiences of the
individual, producing some biases, that can affect negatively the risk visualization
and consequently the decision making process. For the other hand, if actions are
taken adequately these biases can be minimized or nulled. Thus, any visualization
technique presents pros and cons that need to be addressed clearly with the DM.
Finally, it is discussed the function of information visualization as a decision-
support instrument. The requirement to synthesize and analyze the information in
big problems can be better solved by DMs when they aided by information
visualization. It can improve the process of decision making, when it properly
considers the features of decision making and the characteristics of DM.
In literature, research about spatial and visual perception suggests that,
generally, graphics avoid the inadequate numerical risk representations as well as
countable visuals increasing the accuracy of perceived risks (Bostrom et al. 2008).
In risk map, for instance, one may use means like line thickness, textual
information labels, shapes that varied in size or color and other characteristics.
The color of an enclosed region may represent a ‘concept’ type and the size may
be used to represent the magnitude this ‘concept’. Thus, the reader should quickly
discover the most serious undesirable incidents, since they often represent major
risks.
There are some aspects to be considered when using color-coding. The number
of different colors is limited by the DM’s ability to remember and distinguish the
colors, for the following reason: He/she can present a confused view. The use of
color allows to emphasize the most serious incidents, meaning that the reader
should identify them more quickly compared to using other means. An important
aspect about shapes is to avoid symbols/pictures with similar shapes. When they
are similar, there is an increase for the search time to differentiate them, so that
this is not recommended.
A few more settings about the theme are observed in some literature studies.
Ackermann et al. (2014) present a risk map to engage multiple stakeholders and
build a comprehensive view of risks. The authors use risk map as a dynamic tool
to update information and create knowledge to the decision making process.
Bostrom et al. (2008) presented the foundation for designing and testing
alternative ways to communicate risk and uncertainty for low-probability and
high-consequence events, using the knowledge about the effects of spatial
information, communication of risk, and uncertainty in spatial information and
how these can be tailored effectively for earthquake risk analysis.
Fedra (1998) highlights that technological and environmental risks have an
obvious spatial dimension. Floods, mudslides, and avalanches as much as toxic
spills, explosions, transportation of dangerous goods, or hazardous waste
management are all spatially distributed problems.
Eppler and Aeschimann (2009) present a conceptual framework for risk
visualization in risk management. This framework is based in the answers for the
questions of: ‘why’ (purposes), ‘what’ (contents), ‘for whom’ (target groups),
‘when’ (usage situations), and ‘how’ (formats).
In this context, some applications can be observed in the literature. Brito and
Almeida (2009), Alencar and de Almeida (2010) and Lins and de Almeida (2012)
contextualize the multidimensional risk view in the context of natural and
hydrogen gas pipelines. The multidimensional risk analysis results are presented
by the risk difference between pipeline sections. These risk increments provide to
DMs a different interpretation with regards risk visualization, allowing that the
DM allocates resources according to the risk hierarchy. It also allows the
visualization of the gap size between the risks of two subsequent sections of the
ranking.
Additionally, Garcez and de Almeida (2014) present a multidimensional risk
assessment under an intra-criterion vision in the underground electricity
distribution context. This information view allows the DM to identify the relevant
consequence dimensions for each alternative and thus allocate resources to prevent
and mitigate risk more effectively, prioritizing only those dimensions that impact
the alternative. For example, an alternative that impacts only humans, should not
receive resources that are allocated to the environmental dimension, thus
preventing a misallocation of resources.
Tariq (2013) presents damage curves and maps based on estimated losses and
probabilities of all floods considered. The maps illustrate the flood risk
distribution over the study area, including agricultural land-use zoning and
comparisons over the area before and after crop.
Finally, a specific point that should be mentioned concerns with the application
of Geo Information Technology (GIT), Geo Information Systems (GIS), and
software for visualization of qualitative and quantitative analyses. In this context,
an overview of 3D visualization tools for quantitative analyses could be observed
in Kaufmann and Haring (2014).
As an example of application, Jaedicke et al. (2014) uses a GIS (Geographic
Information Systems) solution to warn avalanches in Norway. Maps are used in
3.2 Basic Concepts on Reliability 115
study showing areas susceptible to occurrence of avalanches providing an

overview on the overall situation.
3.2 Basic Concepts on Reliability
First of all, to study maintenance engineering it is essential to have a thorough

understanding of a key aspect of maintenance that has a strong influence on the
actual effectiveness of maintenance actions. This is nothing more than the aging of
the various devices that make up the system, and it is this that the dynamics of
failure often reveal. Indeed, the purpose of maintenance actions is either to
anticipate or remediate a failure. Thus, note that a better understanding of how
failures occur serves as a starting point for developing effective plans aimed at
anticipating and thus precluding the occurrence of failures.
Any piece of equipment or device that is prone to failure, prior to being
regarded as piece of equipment or device, was first conceived as a design project.
Accordingly, when in the design phase, several requirements are laid down and it
is only after ensuring that these have been met that the final characteristics of such
equipment and devices are achieved and therefore that the project can be said to
have been fully completed. Among these requirements or dimensions that formed
the final design there is the ability to preserve the characteristics and design
features of the equipment/device over time and there is another the ease with
which the device, which has developed a fault, can be returned to its operational
state. These are the two most important characteristics for the process of
maintenance management. The first feature is called reliability; the second
concerns maintainability.
First, reliability is discussed and then the concept of maintainability. Reliability
is an already well-established concept among the main ones outlined here and it
makes an interesting contribution to maintenance procedures. This view leads to
two main approaches towards the study of reliability. The first consists of
formulating a problem in terms of relating it to the aims of a project by establish-
ing systems and structural, technological or organizational measures to ensure that
the standard of reliability required by the production system will meet the require-
ments set by performance issues. Such questions resonate with many problems
that extend right up to the moment prior to using the system (Scarf et al. 2009).
In later chapters, some of the issues that directly affect the reliability of a
project are addressed, either as a result of decisions made when choosing design
requirements in general in order to achieve a certain level of design reliability, or
when taking more specific actions that involve only the allocation of redundancy
so as to guarantee a certain level of reliability.
It is worth mentioning that the development of reliability, as a field of study,
occurred primarily in an attempt to reach a better understanding of the reasons
why equipment and devices fail, and this is done by investigating aspects of
design projects that ended up with products being produced (Rausand and
Høyland 2004). On the other hand, it should also be noted that how pieces of
equipment are operated and maintained may significantly affect the chances of
their developing faults and failures.
This observation, in fact characterizes the second approach which makes use of
common sense and everyday experience to highlight that the effectiveness of a
functioning system depends not only on its “innate” properties, but also on the
quality of its operations, maintenance, repair, or on any activity that interferes
with the operational performance of the equipment. At one extreme, if all
maintenance actions are limited to emergency repairs only after the system has
suffered a failure, then the operational characteristics of the system are likely to be
very low and the system will not operate in an efficient manner (Scarf et al. 2009).
As a result, the second approach deals with numerous issues, the main concerns of
which are related to the system already in operation and its nature and have regard
to proposing measures that will obtain the best possible operational characteristics.
The importance of this point of view for this book is to do with the ease with
which maintenance activity is seen to be related to its proper purpose. Indeed, the
main objective of maintenance is to anticipate failure and consequently to reduce
of the probability of its occurrence, which in turn, contributes to mitigating
possible consequences associated with failure. Therefore, the way in which
maintenance actions affect reliability are also discussed in this chapter and which
actions can be undertaken to ensure the operating performance of equipment is
good.
3.2.1 Reliability Perspectives
Despite the fact that reliability can be tackled over a wide field of study, and may
cover issues not only associated with the project, but also the actions that can be
performed to maximize the performance of the equipment already in place, there
are other narrower views of reliability.
According to Márquez (2007) reliability, as well as risk, are in fact elements
that quantify uncertainties. Thus, as the quantification of uncertainty is not in itself
an end but the means by which it is possible to make better decisions, it can be
said that using risk analysis methods and reliability supports the decision process
under uncertainty.
Viewed from this perspective, reliability would then be a set of methods that
helps in decision making regarding the performance of the system under study.
Commonly reliability is deemed to have three main branches, namely:
 The reliability of hardware;
 The reliability of software;
 Human Reliability.
This chapter refers to the branch of reliability that is associated with the
operation of components and equipment. On the other hand, the existence of
different branches emphasized the need to study the different aspects involved in
socio-technical systems: man; the machine and software (intangible elements that
are used to operate these machines) (Pham 1999).
Indeed, the fact of there being different approaches to dealing, separately, with
the different agents involved in the operation of production systems indicates that
the procedures for using a particular approach, in fact, depict only part of the real
problem. This kind of reductionism in modeling a decision problem eliminates the
influences that might be present from the other actors and by doing so, this makes
it feasible to find solutions.
Thus, it is important to keep always in mind that the domain of the
consequences of a failure is limited to the perspective that is actually used.
Consequently the awareness that reliability analysis may provide incomplete
information about the actual performance of the system in its completeness warns
of the need to keep an open mind and to look for complementary dimensions of
decision making or other issues that were possibly left out by the reliability
method adopted.
As an example, consider the behavior of a piece of equipment that had a
failure. This is done by seeking to discover how the process of wear and tear due
to how the equipment operated occurred which typically entails undertaking an
analysis using a reliability-driven approach to machinery and equipment, in
isolation, and without considering the influence of other elements (human and
software). Yet, at the same time, as equipment becomes older, due to wear and
tear, there is also a set of circumstances that leads operators to misusing
equipment and this can lead to failure. This is not taken into account. Similarly, in
an automated system, the malfunction of the system can lead to failure. Such
irregular operating regimes may be linked to failures in control programs or other
items of software, neither of which is taken into account under a reliability-driven
approach.
Besides reducing the scope of analysis when adopting a reliability-driven
approach, it is very common to limit looking at how the components of a system
are impacted by changes in other parts of the system as a whole or how each of
them may impact the system.
According to Jorgenson et al. (1967) this type of simplification is a way to
overcome the difficulties imposed by the complexity of large systems. Moreover,
reliability analysis at the component level is consistent with the actions that are
carried out in practice. The most frequent failures occur in a component, so it is
unnecessary to replace the complete system. Furthermore, most scheduled
maintenance activities also require software components, not equipment, to be
replaced.
3.2.2 Reliability as a Measure of Performance
When treated as a measure, reliability is a somewhat elusive concept. Its definition

is often associated with different interpretations, such as the confidence level of
operational success and the absence of failure, the durability of an item, security,
etc., all of which are very abstract concepts. These are often easier to understand
when the lack of reliability is considered. The failure of equipment in a production
system may result in the loss of very significant amounts of money. Thus, it is
easy to comprehend what reliability means, when one can visualize what might be
lost in its absence.
For calculation purposes, reliability is defined in scientific texts as the
probability of an item performing a predetermined function for a specific period of
time and under appropriate conditions (Hotelling 1925; Lewis 1987; Barlow and
Proschan 1965). As a result, reliability is a probabilistic concept that relates to the
random variable T lifetime of an item, and therefore its mechanism of failure, as
shown in (3.1).
R(t ) P (T ! t ) . (3.1)
Because it is a probabilistic concept, one resorts to reflecting on some basic

fundamentals of probability so as to be able to construct a sequence of reasoning
that leads to a better understanding of reliability.
3.2.3 Reliability and the Failure Rate Function
As already explained above, reliability is frequently defined as the probability that

a system will perform its specific function satisfactorily, for a determined period
of time, under pre-established conditions. Within this definition, the relationship
of reliability with failure is clear, namely an assessment is made of to what extent
the system is far from satisfactorily performing its function. The most important
variable related to reliability is that of time, and it is for this reason that most
reliability phenomena are understood within the dimension of time (Carter 1986;
Lewis 1987; Finkelstein 2008).
Examination of the dependence of the failure rate in relation to time adds
greatly to an understanding the nature of failures; e.g. investigating whether
failures occur prematurely, are random or are brought about by age. In this
context, it is important to determine what the relationship between reliability and
the failure rate is (Lewis 1987; Kuo and Zuo 2003; Finkelstein 2008; Kuo and Zhu
2012).
f (t )
O (t ) . (3.2)
R (t )
From (3.2), (3.3) may be derived:
1 d
O (t ) R (t ) . (3.3)
R (t ) dt
On solving (3.3), it follows (3.4):
ª t º
R(t ) exp « O (t )dt » .
³ (3.4)
« 0 »
¬ ¼
Even though every care is taken over a set of items, either in the design phase,
or in the phase when the item is already in use, it is observed that failures still
occur. Such failures are characterized into different types depending on what the
predominant mechanisms were that worked most effectively in bringing them
about.
First, there are the failures that occur quite early in the life of a component. The
most likely cause of this type of failure is that equipment parts were defective due
to their having been improperly manufactured or constructed. It is this which leads
to high rates of early failures of engineering devices. Loss of parts, substandard
materials, components that are out of tolerance, and defects caused during
transportation are among the causes of failures. This is indicative of inefficient
quality control and results in excessive failure rates near the beginning of the life-
cycle of the project (Lewis 1987).
The middle part of the bathtub curve contains the lowest levels of failure rate
and shows little variation, behaving approximately as a constant. It is referred to
as useful life. Failures during this period of time are often ratified as chance
failures i.e., they happen irregularly and unexpectedly. They probably arise due to
unavoidable loads. External loads above equipment design capacity can lead to an
increase in failure rate, for example, due to the equipment material fatigue
(Guedes Soares and Garbatov 1996; Garbatov and Guedes Soares 2001).
The part on the right of the bathtub curve is a region in which the failure rate
increases. During this period, failures due to the aging are prevalent and the
cumulative effects of such matters fatigue and as corrosion tend to be the
dominant causes of these. Wear out failures are symptomatic of components aging
(Lewis 1987; Bazovsky 2004). These failures happen only if the item is not
appropriately maintained. In practice, it is the calculation of when the failure rate
will start to increase rapidly that usually forms the basis for determining not only
when parts should be replaced but also for specifying the design life of the
component.
It is important to understand that different devices have different bathtub

curves. This difference is present in the predominance of one of the three failure
mechanisms mentioned above as well as in the different moments that most
emphatically characterize the thresholds of each phase.
In practice, more than one factor or mechanism contributes to a failure
(Brissaud et al. 2010). Therefore, a failure rate curve can be seen as a super-
position of curves for different failure modes, as shown in Fig. 3.6.
O(t)
Fig. 3.6 Bathtub curve
Each failure mode and the consequent behavior of the failure rate can be
represented by an analytical expression, which is associated with the distribution
of the density of probability over time with respect to faults.
3.2.4 Modeling Random Failure
Random failure models are among the most widely used models to describe
reliability phenomena worldwide.
For a device that needs to be free of failure, the magnitude of the effect of early
failures may be limited by controlling product quality and narrowing the
production process, plus a later stage of wear control before its operating life
begins (burn in and debugging). Wear out failures should be limited if there is
careful preventive maintenance with periodic replacement of parts or components
in areas of the production system where the effect of wear is concentrated. Thus,
attention is mainly focused on failures and the chances of preventing, reducing or
completely eliminating consequences.
In order to do so, it is important to model this kind of failure. The lifetime
distribution that describes failures, which occur at random intervals, where the
number of failures is the same for equally long operating periods, is the
exponential distribution. This is given by (3.5) (Bazovsky 2004)
f (t ) Oe Ot . (3.5)
where O is a constant called the chance failure rate. Its cumulative distribution is
given by (3.6).
F (t ) 1 e Ot . (3.6)
From (3.3), the reliability function is given by (3.7).
R(t ) e Ot . (3.7)
This reliability formula could be used on devices which are not subject to early
failures, and which have not yet suffered from aging. In others words, the time
where this formula is valid is the useful life of the device. This interval of time
varies widely for different devices. One of the most important aspects of this kind
of distribution is the fact that the reliability of a device is approximately the same
for operating times of equal length. Thus, the time t in (3.6) measures the
operating hours in an arbitrarily chosen operating period of a device, regardless of
for how many hours the device has already been in operation before this specific
operating period. During its useful life, the device is always as good as new. This
is because its failure rate remains the same.
From (3.3) it follows (3.8).
O (t ) O 1/T . (3.8)
where, T is the expected time E(t) for t, given by (3.9).
f t
1 ( )
E (t ) ³ t e T T. (3.9)
T
0
3.2.5 Models of Failure Rate Function Dependent on the Time
For early failures as well as failures due to the cumulative effect of wear and tear,
also called failures due to age, it is necessary to define the most appropriate
distributions which model the failure time, the context in which time influences
the failure process. Although the log-normal distribution and the standard
distribution are often used to represent the model that demonstrates the effect of
age, the Weibull distribution is the most universally employed. The following
shows some other distributions used to model the behavior of failures due to wear
and tear as well as early failures related to design problems (O’Connor and
Kleyner 2012).
3.2.5.1 The Weibull Distribution
The Weibull distribution is widely used, can assume a very wide variety of forms
and is therefore very flexible and it can be used for various types of data (Nelson
2004; Jiang et al. 2001).
The Weibull Distribution can have two parameters, and the probability density
function is given in (3.10):
ª § ·E º
t
E 1 « ¨¨ ¸¸ »
E ªt º « ©K ¹ »
¼.
f (t ) e¬ (3.10)
K «¬K »¼
where:
E - the shape parameter
K -the scale parameter
One can observe a very important role related to E:
E = 1, the failure rate is constant, in which the exponential function is a special
case.
E > 1, the failure rate is increasing. In this case, the corresponding failure phase
caused by wear and tear can be modeled using the bathtub curve.
E <1, the failure rate is decreasing. In this case, the early failure phase can be
modeled using the bathtub curve.
Fig. 3.7 displays the graph for this function
2
1.6
1.2
f(t)
0.8
0.4
0
0 0.5 1 1.5 2 2.5
Fig. 3.7 Weibull probability density function f(t) for ȕ=3 (____); ȕ=0.5 (…….); ȕ=1 (_ . _)
The reliability function is given by (3.11):

ª § t ·E º
« ¨ ¸ »
« ¨© K ¸¹ »
R(t ) e ¬ ¼ . (3.11)
The graph for this function is shown in Fig. 3.8.

1
0.8
0.6
R(t)
0.4
0.2
0
0 0.5 1 1.5 2 2.5
Fig. 3.8 Reliability for the Weibull Distribution R(t) for ȕ=3 (____); ȕ=0.5 (…….); ȕ=1 (_ . _)
The function concerning the failure rate is a Weibull density which is given by
(3.12).
E 1
E ªt º
O (t ) . (3.12)
K «¬K »¼
The graph for (3.12) is shown in Fig. 3.9.
1.6
1.2
O(t)
0.8
0.4
0
0 0.5 1 1.5 2 2.5
Fig. 3.9 Failure rate function for the Weibull distribution O(t) for ȕ=3 (___); ȕ=0.5 (….); ȕ=1 (_ . _)
3.2.5.2 Log-Normal Distribution
Using a Log-Normal distribution curve is appropriate for the situation where it is

early failures which predominantly occur i.e. they conform to the bathtub curve
law. This is known as the period of infant mortality (Martz and Waller 1982).
However, it can model many types of data, due to its ability to assume several
formats.
This distribution, commonly used in modeling certain types of life data, is also
widely known in modeling equipment repair times.
Its density distribution can be given by (3.13):
1 (ln t [ ) 2
f (t ) exp[ ], 0tf (3.13)
Vt 2S 2V 2
where [ = E(ln T) and V2 = Var (ln T).

The graph for (3.13) is shown in Fig. 3.10.
0.06
0.045
f(t)
0.03
0.015
0
0 0.5 1 1.5 2 2.5
Fig. 3.10 Density function for Log-Normal Distribution for ȝ=3; ı=0.5 (____);ı =1 (_ . _); ı =1.5
(…….)
Equation (3.14) is the pdf, based in the standardized normal distribution.
§ ln t [ · 1
f (t ) I¨ ¸ , 0t f. (3.14)
© V ¹ Vt
Given the logarithmic relationship with Normal distribution, then the

Reliability measure can be obtained as (3.15):
§ ln t [ ·
R (t ) 1 )¨ ¸. (3.15)
© V ¹
0.75
R(t)
0.5
0.25
0
0 0.5 1 1.5 2 2.5
Fig. 3.11 Reliability for Log-Normal distribution for ȝ=3; ı=0.5 (____);ı =1 (_ . _); ı =1.5 (…….)
The failure rate function is given (3.16).
§ ln t [ ·
I¨ ¸
O (t ) © V ¹ . (3.16)
§ ln t [ ·
Vt Vt)¨ ¸
© V ¹
3.2.6 Influence of Reliability in Maintenance Activities
With regard to maintenance actions, it is interesting to explain how reliability can

guide the process for planning maintenance.
In drawing up a maintenance plan there are different actions that must be
previously defined and analyzed in order to compile it. It is easy to see that for a
production system, the more complex the system under study is, the more diverse
the set of actions is. Moreover, despite the great diversity of actions that make up
maintenance plans, there are some similarities, especially with regard to its
purpose.
For preventive maintenance actions, there are two main objectives that can
commonly be identified: (1) actions that are performed to ensure the functioning
of the system within the design conditions, and (2) actions that are undertaken to
restore the operational condition of the project.
For the first category of preventive maintenance actions, there are routine
actions, such as: cleaning, lubricating, adjusting, retightening and any others that
may contribute to the permanence of the design conditions. The implementation of
such actions is very important, considering that during the use of a particular
device, it is possible to identify periods in which the equipment was used outside
the design conditions. Should such times be long, or even if they are short but
frequent, the chances are that the aging process will change. Thus, the distribution
of failure times can be modified.
For the second category of preventive maintenance actions, the main goal is to
control the level of wear and tear that arises from using the equipment, whether or
not this is due to the conditions of use set in the design project being met. Thus, by
replacing a part or component, it is expected that the condition of the device gets
close to that of the original design and consequently the probability of failure is
reduced.
In practice it is common to say that a device that undergoes routine intervention
type 1 is only undergoing corrective actions, when no preventive action type 2 is
performed. This is because a device, even when its design conditions are assured,
it still degrades and ages.
Moreover, the implementation of actions, aimed only to ensure operation
within the design conditions, is really the basic assumption, in order to enable
estimates of reliability to be made, given that the vast majority of reliability
models assume that such conditions are indeed ensured.
Thus, as a result of not taking actions in advance to avoid failure due to wear,
the equipment will be doomed to fail sooner or later. Understanding these issues is
essential for effective maintenance planning.
3.2.7 FMEA
FMEA (Failure Mode and Effects Analysis) emerged in the 40s, and was derived
from standards set for U.S. military systems. It is a qualitative method, used to
identify potential failure modes and their effects and to make recommendations
regarding measures to be taken to mitigate risks which can impact the reliability of
a system.
The FMEA is structured in tabular form, usually on spreadsheets, where
knowledge and experience of those involved are considered as input for a
historical database. Information can be extracted, for example, from drawings,
process specifications, technical manuals, flow and operational procedures. The
aim of applying FMEA is to identify design issues, the critical process and
maintenance components.
Ericsson (2005) argues that a more detailed version of FMEA is known as:
FMECA (Failure Mode, Effects and Criticality Analysis), where three criteria are
usually defined order to calculate the RPN (Risk Priority Number): severity (S),
occurrence (O) and detectability (D). These three criteria define the RPN (Risk
Priority Number) given by (3.17):
RPN  SxOxD . (3.17)

Despite many critical questions related to RPN being raised in the literature
(Zammori and Gabbrielli 2012; Yang et al. 2008; Dong 2007; Braglia et al. 2003;
Puente et al. 2002; Braglia 2000; Chang et al. 1999), the RPN value is used in
many studies as a comparison measure for analysis and investigation.
Although many authors emphasize the importance of differentiating between
FMEA and FMECA, Rausand (2011) states that where the frontier between them
lies is rather vague and there is no reason to distinguish between them.
However, some negative aspects of FMEA deserve attention: a significant time
period is required if they are to be applied effectively and they do not take human
factors into account (Stephans 2004); and FMEA is not useful in the process for
identifying combined failures (Nolan 2011). According to Assael and Kakosimos
(2010), each individual failure is considered as an independent event which is not
related to other system failures, except with regard to subsequent effects that may
arise. However, it can be applied in conjunction with other techniques such as
HAZOP (Hazard and Operability Study) when special investigations into complex
systems are made, for example.
On the other hand, in the literature, many of its positive points are highlighted:
FMEA principles are easy to understand (Stephans 2004); FMEA’s description of
failures provides analysts with a basis for making changes to improve a system
(Assael and Kakosimos 2010); it is a useful tool for making analyses and
recording recommendations for design changes (Ericsson 2005).
3.2.8 Reliability Management
According to Birolini (2014), reliability is a characteristic of an item expressed by

the probability that the item will perform the function required at a stated time
interval. From a qualitative standpoint, reliability can be understood as the ability
of an item to remain functional. Quantitatively, reliability specifies the probability
that no operational interruptions will occur during a given time interval.
In this context, how best to use reliability engineering management is a crucial
issue for organizations. This should be incorporated into the strategic level to
ensure that, by using appropriate methodologies and procedures, equipment/system
reliability levels are maintained within the standards laid down.
Calixto (2013) points out that the success of reliability management depends
primarily on four factors: organizational culture, organizational structure, the
availability of resources and work routines. With respect to organizational culture,
two aspects are important: satisfactory financial results and making decisions
based on quantitative data. In other words, one of the minimum requirements for
effective management is the availability of a reliable historical database of faults
and repairs. Reliability management when considering age-dependent models are
subjected to the maintenance and working conditions (Martorell et al. 1999).
Taking these aspects into account, it is observed that the wording of a suitable
model definition to address questions relating to the reliability of equipment
and/or system depends on both the context and the problem type being analyzed.
It is important to note that the objectives of reliability studies can affect the
modeling in different ways. Different goals require different approaches and
methods for modeling and analysis. Furthermore, goals can also directly impact
the choice of the computational approach to be used in the analysis (Aven and
Jensen 2013).
Considering the multiple factors and objectives above mentioned, as dealt with
in Chap. 2, this kind of decision making process and analysis combines the
multiple factors or objective and may incorporate the DM’s preferences over those
factors, therefore, an MCDM/A approach shall be applied.
3.2.9 Simulation
An important aspect in reliability management concerns simulation, which is used

to investigate which can occur in uncertain environments, depending on the
problem type being evaluated. A simulation can be used in complex environments
where a more detailed analysis of a particular parameter can provide very valuable
information for a particular test model.
Additionally, Yoe (2012) points out that quantitative and probabilistic methods
are divided into analytical and numerical methods. Analytical methods are used
when explicit equations are solved, while numerical methods have wide
applicability and the flexibility to categorize the effects of natural variation and
knowledge uncertainty. Among the numerical methods, the Monte Carlo
simulation stands out. This basically consists of two steps: generating artificial
random numbers and transforming random numbers into useful values using a
frequency distribution of the variable under study.
Andrews and Moss (2002) emphasize that simulation seeks to analyze the
interaction among components. The result is usually presented in terms of selected
measures of system performance. The simulation should be regarded as a
statistical experiment where each run of the model is an observation. In this case,
the experiment is conducted entirely on a computer.
Wang and Pham (2006) emphasize the importance of simulation in assessing
the reliability, availability and optimal maintenance of complex large-scale
networks. The Reliability Monte Carlo Simulation generates random failure times
from the failure distribution of each component.
However, Smith (2011) points out that some complicating factors can be
observed in the evaluated environment, thus making it a complex one. As an
example, it is noted that there are complex failure and repair scenarios where the
effects of failures and redundancy depend on aspects such as the number of repair
teams. Furthermore, there is the possibility of failure rates and downtimes

occurring that are not constant.
The appropriate use of the Monte Carlo simulation in environments that
involve an uncertainty context, such as maintenance management, is of great
importance, since this enables the modeling of important events and a more
accurate analysis to be made of possible outcomes of the parameters evaluated.
3.2.10 Redundant Systems
Redundant systems are used in different industrial plants so that the systems
continue operating for longer, even if a failure in a system unit occurs when it is in
operational mode. According to Calixto (2013), two redundancy types can be
observed: passive and active redundancy. In passive redundancy, the redundant
equipment (in standby mode) is for most of the time in a passive state. In other
words, these passive devices operate only when the active equipment fails.
Modarres et al. (1999) add that passive redundant systems are also called
standby redundant systems. The units of this system type remain out of operation
until activated by a sensing and switching device. This process continues to be
carried out until all standby units have been brought into operation and failed. In
this last case, the system is considered failed. Calixto (2013) states that systems
can provide active redundancy, in addition to passive redundancy. This occurs
when similar pieces of equipment perform in conjunction the same function in a
system, in an environment where there is a condition that defines production
losses when several pieces of equipment fail. In some cases the charge distribution
effect may occur, where some items of equipment fail and other pieces of
equipment maintain the same level of production in the system, despite their
degrading faster than usual due to this overload. Since in active redundancy the
components operate constantly, it is expected that the mean time between failures
will be lower than in the case of passive redundancy.
According to Modarres et al (1999), a reliability function for a redundant
system with a standby unit is defined by the following mathematical equation:
f
R p (t ) R I (t ) ³f I (t I )dt I R pp (t I ) R II` (t I ) R II (t t I ) . (3.18)
0
where:
fI(t)I = pdf for failure time (unit I);
Rpp (t)I = sensing and switching device reliability;
R`II (tI) = unit II reliability in standby mode operation;
RII(t- tI) = unit II reliability after coming into operation in time tI.
Calixto (2013) states that the most part of the redundancy increases projects
and maintenance costs, introducing in many types of organizations risk systems,
such as pipelines and tanks.
However, redundancies are often essential systems designed according to the
requirements and specifications of industrial plant systems (Kuo and Zuo 2003;
Tian and Zuo 2006; Kuo and Zhu 2012). There is an extensive literature regarding
redundant systems (Kuo and Prasad 2000).
3.2.11 Repairable and Non-Repairable Systems
Under reliability management, understanding the definition of repairable and non-

repairable systems is crucial for a proper analysis of systems reliability, since they
have very different characteristics with respect to the lifetime of the device/
system and the number of possible failures.
According to O’Connor and Kleyner (2012), the reliability of non-repairable
systems is defined as the survival probability over the expected life of the item or
asset, or a range of its expected lifetime when only one single failure may occur.
The non-repairable items can be either individual item or systems composed of
several parts. Calixto (2013) adds that the availability of non-repairable equipment
is defined by the same equation of reliability, where the term repair means
replacement. The faulty piece of equipment in this particular case is replaced by
another one. The repair time of non-repairable equipment is similar to the repair
time of repairable equipment. System unavailability can be caused in both cases,
with associated losses.
O’Connor and Kleyner (2012) state that after a failure occurs, the reliability of
repairable items is defined as the probability that failure will not occur within the
time period of interest, taking into account in this case the possibility that more
than one failure can occur. Additionally, the availability of repairable items is
affected by the rate at which failures occur and the maintenance period, with
corrective or preventive actions.
Moreover, according to O’Connor and Kleyner (2012), in some specific
situations, an item under review can be seen at different times as repairable and
non-repairable. Guided projectiles, such as missiles, used in military missions are
considered in the first instance as belonging to a repairable system (while stored
and subjected to planned tests), and are regarded as belonging to a non-repairable
system when launched towards a real target. In this case, the reliability analysis
must take into account these two states in different moments.
Consequently, a system reliability study should always consider whether a
system is repairable (or non-repairable) so as to ensure that appropriate actions are
taken effectively at the appropriate time interval.
3.3 Basic Concepts on Maintenance 131
3.3 Basic Concepts on Maintenance
Today it is almost impossible to think of living without the conveniences brought

by the technological development. In fact, our dependence on social infra-
structures is so deep that the absence of them is unimaginable living for an interval
of time without those technological artifacts. Consequently, the greater the
importance that these devices have in our lives, the higher is the relevance of the
production system responsible for producing them. The maintenance of them is
equally relevant.
This is why maintenance does not stand apart from this process, due to the
simple fact that all over the world, among all the humankind achievements, there
is nothing that is indestructible. In some sense, everything comes to an end - from
the simplest product to the most complex system. Therefore, the role of
maintenance is twofold: 1) to postpone this outcome for as long as possible by
undertaking activities, the objectives of which are to maintain the product or
system in a working condition, and 2) to restore anything to the operational state,
when was not possible to avoid the fault.
3.3.1 Characteristics of the Maintenance Function
The maintenance function has some very specific characteristics, which differ
from those of the project function. For example, there is a clearly defined
beginning and end of each project for which this function is responsible.
Maintenance function does not have a defined period during which each item
or equipment of a system will be under the care of maintenance. Maintenance
function seems to be timeless, since its objective is associated with the perform-
ance of a specific system, so whether the system is supposed to be working or not,
there is no time at or during which maintenance activities can be left aside.
The fact that there is a strong demand for maintenance activities does not mean
that there is a right time at which they should take place. Any time can be the time
to do maintenance and these ranges from corrective maintenance that could
happen randomly due to a failure, to preventive maintenance. Even when a system
is not working, it is possible that some maintenance is being undertaken. Indeed,
for some kinds of systems to which accessibility is very difficult, maintenance
activities can only be done when the system stops operation.
Another interesting feature of the maintenance function that further enhances
the first feature of timelessness is the fact that maintenance is used to cope
with and counteract some natural processes, which never cease interfering in
operations. For example, a set of actions has to be frequently undertaken in order
to reduce the consequences of the aging process, the influence of which, most of
the time, is reflected in bringing about changes in failure behavior. Therefore, if
any device is under the influence of one or more natural processes, such as,
damage, corrosion, and wear and tear, it is not possible to stop doing maintenance
without increasing the risk of serious consequences. It is necessary to conduct
maintenance actions continuously to achieve system’s desirable performance.
Currently, the maintenance function has received the importance that it
deserves. The point to be made about this is that this recognition of its importance
was not given immediate. Maintenance departments spread around the world
faced critical battles until it was recognized that maintenance is a strategic ally in
the struggle to remain competitive. Indeed, even today, there are, right now, some
companies where these battles are still being waged, and where the maintenance
function is viewed only as an unwelcome and burdensome source of costs. In fact,
even although it is no longer acceptable not to acknowledge the importance of
maintenance, since the impact of poor maintenance on a company´s production
targets is quickly evident, there are still a great many companies that only do the
minimum in terms of maintenance, i.e. they correct what has failed.
The approach of maintenance and the level of importance given to maintenance
activities are related to many different issues. Therefore, the way in which the
maintenance is approached may be specific for each company due to distinct
characteristics of the productive system, and the different levels of development of
the maintenance function. Thus, despite the fact that problems related to
maintenance are similar, the particular features of each company´s production
system, and a set of different matters, such as company culture, make the problem
unique for each company.
3.3.2 Production System and Maintenance /Basic Concepts

on Maintenance
Although maintenance is a supporting function, depending on the type of system

that calls for maintenance, its role can range from simple support to a central role
within the plant. The truer this is, the greater the effect of the results of the
maintenance action on the company´s revenue and operating costs.
Thus, whether maintenance has a central or supporting role, the challenge is to
carry out maintenance actions in order to make sure that they will have a positive
effect on the system. In other words, the challenge is to guarantee the effective-
ness of maintenance management. The problem is that between the desired and
achieved outcomes, there is often a wide gap and there are many alternative routes
that could be taken to narrow it. Most of them do not close it or get near to doing
so. This is why the effectiveness of maintenance is not a trivial matter. The
amount spent on maintenance is not directly associated with improving production
performance (Scarf 1997). This finding should be one of the most important
guidelines for managing maintenance. It warns that maintenance activities require
structuring and planning in order to enable the system to attain levels of operational
availability at the lowest possible cost by reducing the inappropriate use of

resources, and by the “inappropriate use of resources” is meant using them
excessively or insufficiently. This discussion raises some important questions that
make think more structurally about maintenance:
1. What is maintenance management?
2. Do the functions of maintenance activities depend on the system?
3. What are, in fact, the objectives of maintenance?
4. What are the aspects that highlight the importance of maintenance?
A discussion of these points is subsequently presented.
3.3.3 What is Maintenance Management?
Maintenance can be defined as the set of activities that aims to ensure the levels of
performance necessary to guarantee the achievement of production targets. This
could be by avoiding failure or by restoring the operating condition when the
failure has already occurred. In the first case, this is only possible by means of
planned maintenance actions; in the second case, corrective actions are addressed
and their purpose is to change the state of failure so as to restore the operational
status of equipment, fast enough in order to ensure losses will not be high.
By a closer examination of the maintenance problems, there is a key event,
around which a number of different actions must be performed in order to safe-
guard the competitive existence of different production systems. A failure is, in
fact, the non-operating condition of a device or a condition of productive disability
and this often reveals itself with different consequences, which can often be
summarized as monetary losses but at other times result in very negative outcomes
that are difficult to convert into objective monetary values, such as the loss of a
human life and serious damage to the environment.
This diversity in the nature of consequences, coupled with the behavior of
equipment failure, in which the uncertainties governing different fault events, are
major complicating aspects when attempting to establish systematic maintenance
actions. This makes it very difficult to adopt standard procedures, with regard
to dealing with failures when they have already occurred, or anticipating such
failures. They demand a more rigorous treatment, with an emphasis on using
mathematical maintenance models (Dekker 1996).
3.3.4 Do the Functions of Maintenance Activities Depend

on the System?
Sufficient differences which justify different levels of maintenance being carried

out in different processes can be observed. What degree of rigor, for example,
should be set for maintenance standards related to aircraft and aircraft engines or
turbines? It can be said that these standards must be much higher than the
standards set out in, for example, a small plant. Indeed, although the actions of
maintenance engineering are different in each plant, being influenced by its size,
type, company policy, and many other factors; it is essential to know the scope of
activities of the maintenance engineering department (Corder 1976).
In general, they can be grouped into two general categories: primary and
secondary functions. Primary functions are often very similar, regardless of where
they are put into practice. The intention is to ensure the proper performance that is
demanded of equipment. In fact, it is these functions which justify the existence of
the maintenance engineering department. With regard to secondary functions,
these differ greatly from company to company and are carried out by this
department because of the convenience, or because of whatever other reasons that
are different from those associated with the primary functions.
3.3.5 What are, in Fact, the Objectives of Maintenance?
The definition of maintenance strategies must be aligned with business goals and
therefore the characteristics of the production system (Pinjala et al. 2006). Thus,
although it is possible for there to be variations in the objectives of maintenance
due to the peculiarities of the system, the main and common objectives in various
sectors in which maintenance is conducted can be identified, such as (Corder
1976):
1. To extend the useful life of assets;
2. To ensure satisfactory levels of availability;
3. To ensure operational readiness of systems, and;
4. To safeguard people who use the facilities.
The first three objectives are, in fact, directly associated with the way, whether
this is good or bad, that the maintenance activities are being performed. On the
other hand, the last objective is rather indirect. Actually, maintenance is not in
charge of safety. But, obviously, each time that a failure with dangerous
consequences for humans and for the integrity of the system is avoided as a result
of maintenance activities, these have contributed to making the plant safer.
There is no doubt that these four aspects are in fact the most important
objectives of maintenance. But it may be observed in the literature some variations
in these common objectives of maintenance. For example, Dekker (1996) summarizes
maintenance objectives under four headings, namely: ensuring that the system
functions (availability, efficiency and production quality); ensuring the system’s
life (asset management); ensuring safety; and, ensuring human well-being.
Despite there being some slight differences in the main objectives of maintenance,
the simultaneous achievement of these objectives is not a trivial job. In fact, due to
objectives conflicting with each other, it is very common to adopt only one
objective under the supposition that the one chosen is the one that is the most
closely associated with the strategic objective of the business (Rosqvist et al.
2009; Khazraei and Deuse 2011).
The problem with this approach is the fact that by reducing maintenance
objectives to only the main one, the DM’s view of the problems in the maintenance
field is considerably restricted.
This restriction is even more serious if it is observed the contemporary aspects
that invite the maintenance managers to think more broadly about maintenance
problems. Some of these aspects are listed above (Levitt 2003; Newbrough and
Ramond 1967).
3.3.6 The Aspects that Highlight the Importance of Maintenance
Following aspects of the importance of maintenance are highlighted (Newbrough

and Ramond 1967):
1. the increase in mechanization. This has reduced the direct cost of manual labor,
but has increased the importance of giving due regard to the maintenance of
equipment;
2. the increase in the complexity of equipment. This affects the demand for highly
specialized skills when conducting maintenance activities;
3. the growth of the parts and supplies inventory. In fact, this is a direct
consequence of the first two factors;
4. Stricter control of production;
5. Programming stricter deliveries. This has reduced the inventory of finished
products and has improved customer service. On the other hand, it has also
increased the effects of disruptions in the production process;
6. Increasing quality requirements. While providing an increase in the sales
potential by increasing the attractiveness of products, this also emphasizes the
need for a more immediate response to any abnormality of the product or of the
production process.
7. The increase in concern about environmental damage and risk of human deaths
associated with failures of devices;
8. The widening of the consequence domain, and the diversification of its nature.
It is quite impossible, at a failure event of some kinds of systems, to track all
the agents affected by one failure, or to determine the nature of this affect;
9. The dangers that arise from managerial mistakes with regard to maintenance
activities that are being currently emphasized (Levitin 2000; Wang and Pham
2006);
10.Finally the fact of the business scenario being so competitive that the aim is to
avoid all failures;
All these aspects emphasize the importance of decision-making in the
maintenance context. To achieve the maintenance objective, the decision maker
develops or follows maintenance policies that are the most appropriate for his/ her
objectives and the characteristics of the production system.
The main challenge for the maintenance manager is to structure the maintenance
procedures and activities to be undertaken in such a way that the strategic
objectives associated with them are achieved. This means that the manager has to
plan the maintenance actions with these objectives in mind.
According to Márquez (2007), a maintenance plan is a structured set of tasks
that includes activities, procedures, resources and time required to perform
maintenance tasks. The implementation of maintenance planning in practice leads
to establishing maintenance policies. Maintenance policy is the process of
coordinating maintenance activities with the particular characteristics of each
system, as well as with the goals that the decision-makers wish to reach, which
reflect the company’s strategic objectives.
To define a maintenance policy a mathematical model is associated with it in
order to make sure that the policy achieves best results. The model, by using a
performance function, defines the levels of each action and these should be used to
optimize this function. The most appropriate action can be defined before defining
its frequency. However, there are models in which not only the activity, but also
the frequency, is defined simultaneously (Scarf et al. 2009).
The next section presents a more structured discussion about how mathematical
models contribute to maintenance management, emphasizing the contribution of
the maintenance policy to the maintenance management process.
3.3.7 Maintenance Policies
For a better understanding of the mathematical models on maintenance, and

changes made to them over time, one can go back to the past and describe some
important aspects that had to be taken into consideration at the time that these
models were proposed.
According to Jorgenson et al. (1967), two distinct classes of problems are
involved in asset management: inventory management and management of
durable equipment. Inventories provide items for the production process; durable
equipment provides services. The management of durable equipment, however,
imposes two additional problems, choosing appropriate levels of service for the
equipment and keeping up these services.
Years ago, choosing an appropriate level of service was discussed, while
maintenance costs were assumed to be constant or nonexistent. In the 1960s, how-
ever, an entire theory on the maintenance of equipment started to be constructed.
Optimal maintenance policies were proposed and characterized for a wide variety
of situations.
The first studies on maintenance policies treated the problems as being
deterministic (Taylor 1923; Hotelling 1925), i.e., problems in which the result of
each maintenance action is non-random. Some years later, however, different
studies properly faced up to the stochastic aspects of maintenance problems
(Barlow and Proschan 1965; Barlow and Hunter 1961; Barlow and Proschan
1975; Barlow and Hunter 1960; Glasser 1969), for which the consequences of
the maintenance actions would be random. What most motivated developing
maintenance policies emerged largely and importantly from tackling practical
problems of maintaining complex electronic equipment, i.e., aircraft, missiles,
spacecraft, communications equipment, computers, and so on.
The methodology and the theoretical development related to stochastic
maintenance have a striking resemblance to the stochastic theory of inventory
management. Both have their roots in simple deterministic models. Stochastic
inventory theory models usually assume that for a particular item, demand per unit
time and delivery time are random variables. The corresponding stochastic
elements in the theory of maintenance are the time to failure of equipment as well
as repair time.
Indeed, from a broader perspective, Jorgenson et al. (1967) state that no
distinction is made between the principles of management of inventory control
and of durable goods. For durable equipment, the outputs of conservation
activities are services rather than individual items. The level of conservation
activity depends not only on acquiring a spare part of productive assets, but also
various other inputs of materials and services that represent the maintenance
activity. The output of service equipment can be fed into other activities.
Terborgh (1949) refers to the same issue. According to him, the hand of time
lies heavily on the work or the deeds of men. He also notes that it is a fact that
more practical consequences confront the owner of the item with two problems:
the first is to distinguish the speed of death, or, in other words, to say if an asset
that has not been exhausted physically still has a life of economic usefulness,
either generally or for the particular function it performs. The second is to make
the financial provision, in order to be possible to prevent wear of durable goods
over its service life. Jardine (1973) classifies maintenance policies into two
general classes: probabilistic models of maintenance policies and deterministic
models. A striking difference between these classes of problem, besides the
existence of a stochastic process that governs the events processed within the
policies, is to check whether or not there has been a complete failure on an item.
For example, for complex systems where the probability of its main function
being stopped is very low. However, since with the passage of time there is a
considerable increase in operating costs, it is quite appropriate to use a deter-
ministic approach which consists of observing a cost function and has features
very similar to inventory control models.
Therefore, Jardine (1973) establishes different subgroups in each class, which
consist of models with similar features:
1. A class of Deterministic Models: models for replacing equipment when
operating costs and use increase; models for replacing equipment when
operating costs use, for use with finite time horizons; models for replacing
equipment considered capital investment, taking into account the discounted
net benefit; models for replacing equipment considered capital investment,
taking into account the technological improvement.
2. Class of Probabilistic Models: Age Replacement based models, taking into
account the time of repair and replacement, and finally, the block replacement
model.
Sherif (1982), discusses different maintenance policies in an article that
summarizes several studies on the subject. He also makes the same classification
as Jardine (1973), although he makes different divisions existing within these
subgroups of two broad classes.
McCall (1965) presents a survey of maintenance policies which is in a quite
different format from those of Jardine (1973) and Sherif (1982)
McCall (1965) does not consider any deterministic model and makes an in-
depth analysis of maintenance policies for systems with stochastic failure.
According to McCall (1965), the development of such policies is based on a
variety of mathematical techniques. This foundation, along with a variety of
applications, sometimes obscures the underlying structure common to all policies.
The first author’s purpose is to identify this common structure, and thus clarify the
relationships between the various maintenance policies.
McCall (1965) classifies models into two categories. The first corresponds
to the class of models called (preparedness) preparation or readiness, when
equipment fails stochastically. Its state, in fact, is not known with certainty.
Alternative maintenance actions for such equipment include inspection and
replacement. Preventive maintenance models constitute the second class of
maintenance models. In these models, the machine is subject to stochastic failures,
and machine status is always known with certainty. If the equipment displays an
increasing rate of failures and, moreover, if it is more costly to repair the fault
when the system is in operation than to replace the equipment before it fails, then
it may be advantageous to replace the equipment before it fails. The problem is to
determine a suitable replacement plan (Nakagawa 1984; Nakagawa 1989).Currently
the first class of models has gained considerable development. Investigating the
state of equipment is now supported by a large number of technological tools, in
addition to which the research field has increased and diversified in recent years.
This class has been referred to as condition-based maintenance and has been
declared as a new milestone of a new generation of approaches in the practice of
production and maintenance management (Ahmad and Kamaruddin 2012; Wang
2012; Baker and Christer 1994).
In contrast with this, for the second class, the current preventive maintenance
models are almost the same as the earliest ones. Basically, these models deal with
the process of failure, by observing the lapse of time since the last preventive
maintenance activity (Chang 2014). Despite the small evolution of preventive
maintenance models per se, one of the most important contributions, currently, is
the combination of the two distinct strategies, for instance, checking the state and
replacing preventively after some time. These kinds of combination sometimes are
known as hybrid policies, where both actions could be taken following different
rules. For some examples of this policy (Scarf and Cavalcante 2012; Scarf and
Cavalcante 2010).
Another very important class of policies that is very useful for systems
comprising more than one part or multi-component is known as an opportunistic
policy. The main distinguishing feature of these policies is that maintenance actions
for a piece of equipment depend on the state of the rest of the equipment. The con-
nection established between the states of the components allows more favorable
outcomes to occur when compared with those that could arise from individual
policies for each component. By restricting the vision to one device at a time, the
opportunity to observe actions for multiple components simultaneously is lost, as
are the savings that would be made by dealing with opportunities in the most
intelligent way.
Just as in the preventive maintenance policies, in which the most recent major
contributions are associated with the combination of different actions (Drapella and
Kosznik 2002; Jiang and Jardine 2007; Thangaraj and Rizwam 2001), the combi-
nation of actions was also used in order to improve the opportunistic maintenance
policies. The advantages arising from the combined use of the activities inherent
to the two most typically known groups: preventive and corrective maintenance.
This combination was systematically studied by representatives of the RAND
Corporation in the early sixties (Radner and Jorgenson 1963; McCall 1965;
Jorgenson and McCall 1963). As a result, opportunistic maintenance basically
refers to the situation where preventive maintenance is performed arising from
opportunities related to the choice of a date or constraints due to the impossibility
of postponement, given a failure event. In many cases, it is assumed that the
process of generating opportunity is completely independent of the fault (Dekker
1996). On the other hand, it is common to consider the opportunities that coincide
with the time of failure of individual components. Due to economies of scale in
the cost maintenance function, the undesirable event of a fault in a component is
also considered as an opportunity for preventive maintenance of other components.
One must note that in many situations, a combination of preventive and
corrective maintenance repairs is not realistic. The need for corrective
maintenance arises unexpectedly, while preventive maintenance can be planned.

Thus, if there is a combination of both types of activity, the character of
schedulable preventive maintenance is lost, or is forced to ignore that a device is
flawed for some period of time. Nevertheless, there are situations in which this
loss and non-action are acceptable, particularly when the corrective repair of a
single component requires the entire system to be disassembled. Thus, combining
a corrective repair of a component with the pre-emptive repair of its neighboring
components can be lucrative.
As previously mentioned, there are two options for making a combination. On
the one hand, preventive maintenance can be brought forward when a failure
occurs, and thus when repairs cannot be postponed. On the other hand, when
faulty components can be kept idle for a limited period of time, one can opt to
delay corrective action until to the next preventive maintenance instant. There are
several studies that further develop and refine this type of policy (McCall 1965;
Radner and Jorgenson 1963; Woodman 1967; Jorgenson and McCall 1963; Zheng
and Fard 1991; Zheng 1995).
Even at different times, several authors such as (Barlow and Proschan 1965;
McCall 1965; Dekker 1995; Dekker 1996) reported on the growing interest in
developing and implementing maintenance policies for systems with stochastic
failure. Undoubtedly, this interest was caused by the high costs and the extra-
ordinary demands arising from more complex equipment such as jet aircraft,
electronics, computers, etc.. It was also observed that, unlike the consideration of
maintenance as an expensive nonsense - a concept that has prevailed for a long
time - its real importance has been identified in the face of operational require-
ments that are achieved as a result of implementing relatively sophisticated
maintenance policies.
3.3.8 Structure of a Decision Problem in Maintenance
Specific literature on maintenance try to give a vision regarding to the structure of

a decision problem in maintenance, although it do not use the basic principles of
decision making area, particularly concerning to the MCDM/A methods.
According to McCall (1965), the general structure of these problems has
elements that are characteristic of decision theory models. While in operation, the
equipment in question may take one of several states, with the two extreme states
being as good as new and the faulty state. Between these two state-limits there is a
set of intermediate states, which denote different degrees of deterioration (Grall
et al. 2002; Bérenguer et al. 2003; Fouladirad and Grall 2014). The move from
state to state is governed by a stochastic mechanism the behavior of which could
be unknown, partially known or completely known by the DM. A neglected piece
of equipment moves stochastically from one state to another in a natural way, to
reach the state of absorption that corresponds to failure. The behavior of the
device can, however, be regulated by choosing a particular action at each decision

point. These actions include doing nothing, conducting an inspection, carrying out
repairs and replacing different types, or performing a complete overhaul thereby
renewing the equipment. The sequence of actions chosen by a DM reflects the
maintenance policy and the difference between the controlled and non-inhibited
degradation process of the equipment. It is a measure of the influence of policies.
The performance of the policy can be measured in terms of costs, by associating
an occupancy cost to each state and a cost of intervention with each action. The
goal of the DM is to choose maintenance actions such that the cost per unit time of
operation of the equipment is minimized (McCall 1965; Jorgenson and McCall
1963; Jardine 1973; Radner and Jorgenson 1963; Dekker and Scarf 1998).
Regarding to the approach of decision maintenance problems using MCDM/A
methods, there is an extensive work found on the literature, which is given in
subsequent chapters. The following sub-sections give an idea of the kind of
maintenance problems to be approached in this vision.
3.3.8.1 Decision Problems on Maintenance Planning
The ever increasing need for higher productivity, in the face of growing
competition, has demanded of the various sectors of the economy a constant
search for tools that will enable them to acquire competitive advantages.
In order for these organizations to meet these requirements, it is essential that
their production systems are able to operate under normal conditions; in other
words they must be reliable and available. It is the maintenance function that is in
charge of ensuring the normal operation of these systems. To be successful in this
objective, paying due attention to the maintenance structure is the best way to deal
with common problems related to the management of maintenance.
According to Kelly (1983), maintenance planning is a traditional practice,
recommended for the maintenance of machinery, equipment and tools, and should
be conducted by preparing work plans and setting norms and standards for their
conduct. Márquez (2007) states that what is needed for any level of maintenance is
a structured set of tasks that includes activities, procedures, resources and defines
the time required to perform maintenance tasks. These definitions explain the
scope of maintenance planning:
x What should be done?
x When should it be done?
x Which resources should be employed?
The more correct the answers are to these three points, the more efficient the
planning of maintenance resulting from these issues is. In this sense, effective
maintenance planning enables managers to take actions using the correct equip-
ment, at the right time, and with the proper tools. The successful implementation
of maintenance activities is directly related to precise preplanning.
The answers to these questions usually follow a hierarchical order. So, initially,
it is essential to specify which activity or activities to conduct on each device;
subsequently establish the frequency with which each of the activities should be
undertaken on each of the items of equipment, and finally define the set of
resources that will be used.
3.3.9 Main Techniques for Maintenance Management
Maintenance used to be defined as a simple task of restoring the original condition

of equipment and systems and it is currently conceived, in a broad and modern
way, as a process that ensures reliability and the availability of the function of the
equipment and facilities for a production process or the provision of services, with
security, while preserving the environment and being conducted at appropriate
cost.
In accordance with British Standard BS EN 13306 (2010), maintenance is the
combination of all technical, administrative and managerial actions during the life
cycle of an item, which are intended to retain it or restore it to a state in which it
can perform the required function.
Maintenance is the term used to address the way in which organizations try to
avoid the failures of their assets. It is an important part of production systems,
particularly when it is critical to the company’s business. For example, this applies
to power plants, airlines, refineries and petrochemical plants.
Although the paradigm of the past dictated that maintenance professionals
should perform a good repair service when prompted, now maintenance work is
being given more recognition. Skills and technologies have been developed to
prevent failures instead of correct them.
Maintenance professionals are increasingly required to have several core
competencies such as:
x Sizing and integrating physical, human and financial resources in maintaining
systems, and doing so efficiently and at least cost, while considering the
possibility of continuous improvement;
x Using management methodologies, mathematical and statistical tools to support
the planning and control of maintenance systems and thus to aid decision
making;
x Incorporating quality concepts and techniques into maintaining production
systems, in technological and organizational aspects, improving processes, and
producing standards and procedures for control and audit;
x Using performance indicators, costing systems, and assessing the economic and
financial viability of projects;
x Information management in companies using appropriate technologies;
Significant research work has been conducted in various subfields of

maintenance, based on more specific aspects of maintenance. Such research
includes issues such as data analysis and fault repair, preventive maintenance
models, reliability models, asset management, human reliability, accelerated
testing, diagnosis and prognosis models in predictive maintenance, performance
evaluation of maintenance policies. This specificity and focus is essential for
developing and validating contributions to scientific research.
On the other hand, there is a set of management approaches and a more
systemic and generalist view of maintenance management as a process that
involves resources such as human, material and financial resources to develop
better performance and thereby greater plant availability. Among this set of
approaches there are TPM (Total Productive Maintenance) and RCM (Reliability
Centered Maintenance).
Many organizations have adopted managerial maintenance approaches such as
TPM and RCM, since these approaches are committed to the long-term improve-
ment of maintenance management. Several authors have reported maintenance
management as a strategic management activity that can contribute significantly to
the success of business (Reis et al. 2009). In following sections, an overview of
some managerial techniques used in the field of maintenance management is
given.
3.3.9.1 Total Productive Maintenance (TPM)
Total Productive Maintenance (TPM) is defined as the productive maintenance

performed by all employees through small group activities in which productive
maintenance is the form of maintenance management that recognizes the
importance of reliability, maintenance and economic efficiency in the design of
plants (Nakajima 1988).
The term productivity in TPM is related to the goal of the maximum overall
efficiency of equipment, which is a measurement of the capacity of machines
versus the amount actually produced in time. Availability, quality and labor saving
because of plant modifications are essential aspects of TPM. This maximum
efficiency can be achieved through quality management, which has the function of
controlling the possible defects that may occur during the process. TPM seeks to
eliminate losses and achieve zero defects, zero breakdowns and zero accidents, so
that the length of time that the production line is available is longer and therefore
it can produce at maximum capacity. TPM is a management philosophy that
promotes change in the organizational culture towards greater quality and pro-
ductivity at all levels in the company. TPM tries to eliminate the different losses
that adversely impact the effective operation of the system (Pintelon and Gelders
1992).
Tajiri and Gotoh (1992) and Shirose (1992) state that a definition of TPM
contains the following five points:
1. It aims at getting the most efficient use of equipment.
2. It establishes a total (company-wide) planned maintenance system, (preventive
maintenance, and improvement related maintenance).
3. It motivates the participation of department workers, equipment operators, and
equipment designers.
4. It involves everyone from top management down.
5. It promotes and implements planned maintenance based on autonomous, small
group activities.
In other words, the goal of TPM is to redesign the system of the company, by
seeking to improve the performance of people and equipment. Improving staff
performance is based on training employees (operators and maintenance workers)
so that they can maintain the machines working as per their specifications, and
when an abnormality occurs, the operator himself is able to identify it and solve
the problem, whenever possible.
Improving equipment consists of structural modifications that represent some
kind of benefit to the yield of machine and operator. Another relevant point is to
reduce future maintenance costs when evaluating the purchase of new machines.
Companies want to increase their productivity and reduce losses. TPM is one of
the tools used to eliminate such losses, which can be classified, according to
Shirose (1992), as:
1. Breakdown losses – these can be failures because of stoppage in the operation,
which is caused suddenly, or by deterioration in function, which is a partial
reduction in the capacity and function of the equipment compared to the
original state. This loss is related to the loss of function of the equipment, and
leads to both chronic failures and sporadic faults, resulting in the loss of time
and productivity;
2. Setup and adjustment losses – these happen when one device produces different
products, so it may take excessively long to adjust the equipment so that it is
able to produce another product with the desired quality;
3. Idling and minor stoppage losses – occur when short breaks are not taken into
consideration, but when added together can result in a high loss of time, and
empty operations;
4. Reduced speed losses – occur when the machine operates for any reason at a
slower speed than normal;
5. Quality defects and rework – occur when there are defects that can lead to
disposing of the product and so time and materials for production are lost,
while defects need to be corrected and to fix them an additional amount of
operating and labor time will be necessary which entails sustaining a loss;
6. Startup / yield losses (reduced yield between machine startup and stable
production) – correspond to the period in which the performance conditions of
the equipment after it has been triggered do not reach stable production.
So that TPM may develop in organizations, it is necessary that the foundations,
called pillars, are constructed in teams and coordinated by managers or leaders of
each team. The eight pillars of TPM philosophy form a support system that targets
ensuring productive efficiency for the entire organization. Nakajima (1988) lists
the eight pillars of TPM as being:
1. Autonomous Maintenance - places responsibility for routine maintenance, such
as cleaning, lubricating, and inspection, in the hands of operators;
2. Planned Maintenance - schedules maintenance tasks based on predicted and/or
measured failure rates;
3. Quality Maintenance – detects design errors and prevents them from entering
into production processes. It applies root cause analysis to eliminate recurring
sources of quality defects;
4. Focused Improvement - Small groups of employees work together pro-actively
to achieve regular, incremental improvements in the operation of equipment;
5. Early Equipment Management - directs practical knowledge and understanding
of manufacturing equipment gained through TPM towards improving the
design of new equipment;
6. Training and Education – fill in gaps in knowledge which is required to achieve
TPM goals. Training and educational opportunities are given to operators,
maintenance personnel and managers;
7. Safety, Health, Environment – These are about maintaining a safe and healthy
working environment;
8. TPM in Administration - Applying TPM techniques to administrative functions.
The concept of overall equipment effectiveness (OEE) is an important TPM
topic. It is calculated by multiplying the availability of equipment by its perform-
ance efficiency and by its quality rating. OEE gives a useful measure for tracking
the progress and improvements from the TPM program; but it does not give enough
detail to determine why the equipment is better or worse (Mobley et al. 2008).
According to Nakajima (1989), the results from TPM are an increase in the
machine availability index by decreasing the number of breakdowns; a decrease in
the number of failures in the process and thus a decrease in the number of customer
complaints; a reduction in production costs; and a decrease in the number of
workplace accidents. All this is possible by preparing and developing people,
combined with greater integration between man and machine that operates so as to
improve productivity and increase the competitiveness of the entire organization.
There is a trend for many companies to adopt TPM as a tool since they are
interested in the potential success of this methodology. It is also true that many of
the targets are quite challenging, which is why it is important to motivate people
to seek continuous improvement in order to achieve zero losses in the production
environment and equipment. Some companies have not been successful in

implementing the TPM, and this is due to several causes:
x No support is given from upper management and implementation does not
follow the “top down” direction recommended. This is a key point since it is
necessary to change the culture of staff so they will adopt new practices, and to
invest in improvements in equipment, since, without the support of top
management, this challenge becomes more difficult;
x The internalization required for autonomous maintenance is missing, in which
the minimum requirements are often not guaranteed and the pillar performs
tasks that are more aesthetic than about implementing techniques;
x Without there being an effective program of planned maintenance, there is a
change of attitude in the maintenance sector and the environment remains the
same as it was before implementing TPM
x Without systematic measurements and the monitoring of losses that
compromise the performance of the equipment, it becomes difficult to manage
the improvement process
x Without changing the practices of how new systems and spare parts are
acquired, maintenance performance may not be effective.
There are implementation procedures for TPM in the literature (Manzini et al.
2009). TPM recommends deployment steps to be followed, and indicates that the
maintenance plan must choose which of the various types of policies will be more
profitable, but does not explain how to do so in detail. Thus, TPM leaves a gap in
supporting decision making about the best maintenance policy and there are
different interpretations as to how to implement TPM.
3.3.9.2 Reliability Centered Maintenance (RCM)
RCM is a methodology for identifying maintenance needs in physical or industrial

processes. It came from the aeronautics industry in the 1970s, and was adopted by
the American defense industry. Then it was extended to the nuclear energy area,
and several industrial sectors. RCM is widely used in various industries. (Nowlan
and Heap 1978). The process involves the assessment of a structured set of
questions that sequentially identify some aspects of the equipment: Main
functions, functional failures; Failure modes; effects of failures and consequence
of failures.
RCM is a program that integrates various engineering techniques that aim to
ensure the functioning of industrial equipment. This program has been recognized
as a very efficient way of addressing maintenance issues, since it uses a rational
and systematic approach to solve problems (Moubray 1997). Moreover, according
to Ben-Daya (2000), RCM is an approach used to optimize a preventive
maintenance strategy and its main focus is on maintaining the function of a system
rather than wanting to restore it to its optimal condition.
To be at its most effective, RCM needs to be based on certain factors such as:
x The involvement of engineers, operators and maintenance technicians;
x Due importance being given to the study of the consequences of failures that
drive maintenance tasks;
x The scope of the analysis, as this should include safety issues, the environment,
and operation costs;
x Suitable importance given to proactive activities that involve predictive and
preventive tasks;
x Avoiding hidden failures that reduce system reliability;
According to Moubray (1997) there are seven basic questions that should be
used by an RCM program:
1. What are the functions and associated performance standards of the asset in its
present operating context?
2. In what ways does it fail to fulfill its functions?
3. What causes each functional failure?
4. What happens when each failure occurs?
5. In what way does each failure matter?
6. What can be done to predict or prevent each failure?
7. What should be done if a suitable proactive task cannot be found?
In RCM, the four most important terms are: system, subsystem, functional
failure and mode of failure:
x System: This is the plant as a whole or a subdivision thereof which is identified
in the RCM analysis;
x Subsystem: This is a group of items of equipment and/or components which
together perform one or more functions and can be considered as a separate
functional unit within the system;
x Functional failure: Every subsystem performs a certain function. A functional
failure describes how each subsystem failure occurs;
x Failure Mode - Identifies each specific condition related to a specific piece of
equipment which causes loss of function of a subsystem.
RCM provides functional requirements and standards for the desirable
performance of equipment; For each function, functional failures are defined, and
the failure modes and effects of failures analyzed using FMEA (Failure Modes
and Effects Analysis). Consequences of each failure are analyzed for impacts
arising from the effects of the failure modes. Errors fit into one of four categories:
hidden; associated with safety or the environment; operational; and non-operational.
According to Rausand and Vatn (2008), the RCM analysis process can be
carried out over the following 12 steps:
1. Study preparation;
2. System selection and definition;
3. Functional failure analysis (FFA);
4. Critical item selection;

5. Data collection and analysis;
6. Failure modes, effects, and criticality analysis (FMECA);
7. Selection of maintenance actions;
8. Determination of maintenance intervals;
9. Preventive maintenance comparison analysis;
10. Treatment of non-critical items;
11. Implementation;
12. In-service data collection and updating.
RCM training involves the basic concepts, functional failures, failure patterns,
block diagram, concepts of reliability, redundancy, FMEA, predictive, corrective
and preventive maintenance, an RCM decision diagram and deployment steps. In
the selection phase of relevant maintenance activities, more effort is devoted to the
most critical components. Maintenance tasks can be predictive, which are based on
wear; preventive when time-based; and reactive, when equipment runs until failure.
In terms of documenting maintenance activities, RCM suggests a worksheet
that contains diagrams of the system, subsystem, components, description of
activity, its frequency and person responsible for. When you have quantitative
data you can base the study on reliability, or when such data are scarce, the work
team must define the periodicity of maintenance. It is important that activities are
documented, and many of them are carried out by maintenance staff, but can also
be performed by staff from operations, by engineers or by a third party.
In the implementation of RCM, the establishment of targets and indicators is
fundamental for a successful application. Initially, indicators are defined and then
the current situation is set, to then be able to develop goals that are coherent and
feasible to achieve, yet while challenging, they are not impossible. The indicators
need to be monitored so that feedback is given to work teams.
The review of the RCM program should be performed regularly because
implementation is an evolutionary process. The conditions of the equipment and
the resources of maintenance change so often that it is necessary to review
maintenance procedures in order to be up-to-date. Furthermore, it is important to
note that work teams´ knowledge is always increasing and if used in a good way,
this can contribute to the continued development of the RCM program.
Finally, it may be added that according to Ben-Daya (2000), if RCM is
implemented in combination with TPM, better results can be achieved. He states
that RCM offers a framework for optimizing the maintenance effort and getting
the maximum out of the resources committed to the planned maintenance program
and he argues that RCM can help achieve better results from implementing TPM.
Moubray (1997) states that RCM can achieve greater safety and environmental
integrity, improve operating performance further (output, product quality and
customer service) and lead to maintenance being even more cost-effective, to
prolonging the useful life of expensive items, to making the database more
comprehensive, to motivating individuals more and to better teamwork.
3.4 Prior Knowledge of Experts in Risk, Reliability and maintenance 149
3.4 Prior Knowledge of Experts in Risk, Reliability and

maintenance
In reliability, risk analysis and maintenance models, it is essential to incorporate

uncertainty into the modeling. These uncertainties are usually derived from natural
variation (random pattern), lack of knowledge or lack of understanding of cause-
effect relationships in the present or future condition. Therefore, uncertainty may
arise from the uncertain knowledge of some aspect, such as the inaccuracy of the
measurement techniques, lack of data, lack of detail, and other factors that directly
affect the measurement of uncertainty.
Furthermore, the variability associated with estimates and uncertainties may
come from the lack of a clear specification of what is required; the lack of
experience in certain activities; complexity in terms of the factors of influence and
interdependence of variables; limited analysis of the processes involved in the
activities; and, the possibility of particular and rare events or conditions occurring
that may affect the activity under analysis.
Besides, under the aspect of the uncertainty of the data, in the context of
decision-making, there are uncertainties related to the objectives, priorities and
acceptable tradeoffs that decision-makers have to deal with. There must be a
complete understanding between the parties involved (clarifying the goals and the
reasons for them). Therefore, the various parties involved introduce uncertainties
due to ambiguities with respect to: specifying responsibilities; their perception of
roles; communication interfaces; contractual conditions and their effects; and with
respect to mechanisms for coordination and control.
According to Berger (1985), an important element of many decision problems
is the prior information concerning the state of nature ș. A convenient way to
quantify each medium of information is by using probability distribution (ʌ(ș)),
also known as prior probability distribution. Therefore, the experience acquired by
experts about a variable can be used in the form of a probability distribution
(Martz and Waller 1982).
The use of measures of probability/possibility/occurrence in risk management,
maintenance and reliability models and decision analysis models is a very strong
requirement. To estimate these measures, information is needed about the several
events (failure mode, incidents, accidents, etc.). It happens that in many situations,
there are items of information that are unthinkable or extremely unlikely to occur,
i.e., they form a set of rare events. For example, rare events are those that might be
included in an analysis of systems judged to be highly reliable (e.g. nuclear
systems, aircraft systems, space systems, etc.), or also of the occurrence of rare
events (catastrophic events by natural disasters, nuclear accidents, accidents in
new technologies, certain conjunctions of causes and effects, etc.).
Hence, on those occasions it becomes quite hard to determine the precise
values of probability of failure or outcome of an accident. Yet, even in ordinary
circumstances, in an industrial system (in which events occur with a higher
frequency), determining likelihood is affected by the lack of a comprehensive

database of failures, failure modes and accidental events and their consequences.
Therefore, managers require alternative means of acquiring knowledge about
the context. With a view to minimizing this obstacle, in the literature there are
methodologies aimed at eliciting experts’ prior knowledge, all of whom should be
familiar with the theoretical approach and have experience of the context
analyzed. Thus, this section will focus on one brief context: addressing the main
characteristics of the use of expert’s knowledge, and ways to aggregate them.
According to Walley (2002), theories of statistical inference can be divided into
two broad classes: those that satisfy the principles of probability and those in
which the inferences are grounded on interpreting what would happen if historical
events were to be repeated and on data sampling (the frequentist approach) .
This approach is very useful for solving decision problems. As its name
suggests, it uses historical data or data obtained from trials on which to base its
claims.
In risk analysis methodologies, a frequentist approach may be used to tackle
failure modes analysis. However, as previously mentioned, a purely frequentist
concept of probability cannot always be applied due to the fact that there are some
rare events, the repetition of which are almost impossible (very hard) to predict,
especially when considering the operations of a small production system or unique
system, or when the accumulated amount of historical data is small (insufficient).
Thus, it becomes impractical to establish a probability, based on the past
experience of the company, due to the absence of such data (Garcez et al. 2010).
According to Garcez et al. (2010), one way to overcome the lack of an internal
database is to use an external database (e.g. the database of other local companies
or international organizations). However, the simple use of an external statistical
database as a benchmark may be mistaken because some characteristics that
directly influence this probability represented in the external database, such as
regulations, operational structures, levels of technology employed, safety, societal
culture, etc. may not reflect the environment of the system that will be analyzed,
thus generating differences in statistics.
Therefore, it is necessary to correlate the factors influencing the probabilities
with the technical characteristics of the system being analyzed and its nearby
systems. To do this, all the experience gained by experts in the field is applied (by
the Bayesian Approach), using their expertise and also knowledge about the
operating system analyzed, thereby providing valuable information to the decision
process (Clemen and Winkler 1999).
The analysis of the data from the database allows a better view of the historical
statistical relationships and their relationship to accidents, while the Bayesian
approach enables a realistic representation of the expert’s knowledge about the
dynamics of operation and failure modes in the systems analyzed (O’Hagan 1998).
As an alternative way to determine the rate at which accidents are caused by
failure, the methodology defined in Raiffa (1968) can be used. This calls for prior
knowledge (using the Bayesian hypothesis) to be elicited along with an analysis of
historical data on accidents and failures, coming from external (local or

international) databases, internal data of the company itself or similar companies,
and thus may enjoy the advantages of each approach.
3.4.1 Elicitation of Expert’s Knowledge
According to Kadane and Wolfson (1998), the purpose of eliciting prior

knowledge is to capture the main characteristics of an expert’s opinion, and
thereby to integrate their experience and their academic knowledge. For O’Hagan
and Oakley (2004), frequentist inference only enables probability to be
interpreted, while Bayesian statistical methods are based on a personal (or
subjective) interpretation of probability.
Subjective probability is the degree of belief of the expert in the chance of a
particular event occurring, i.e., there is not a correct (accurate) probability, but
there is a probability distribution that can be assigned to an event, following all the
basic postulates of probability theory (Berger 1985).
For Keeney and von Winterfeldt (1991), formal elicitation of an expert’s view
of probability elicitation consists of the following steps:
x Identifying and selecting problems;
x Identifying and selecting experts;
x Discussing and refining the problematic;
x Training experts on why and how knowledge is elicited;
x Elicitation process;
x Analyzing, aggregating (outcomes) and resolving disagreements;
x Documenting and reporting results.
According to Garthwaite et al. (2005), the procedure for eliciting the prior
knowledge of the expert can be separated into four stages:
1. Arranging for (setup), selecting and training experts and identifying aspects of
the problem to be elicited;
2. Eliciting, interaction with experts;
3. This relates to adjusting the probability distribution of the result of the
elicitation;
4. The last step is linked to assessing the adequacy of the elicitation process.
In order to elicit an expert´s prior knowledge properly, Kadane and Wolfson
(1998) list some important points, namely: there must be consensus on the
elicitation procedures: it is only expert opinion that should be elicited; experts
should be questioned only on observable quantities; experts should not be asked to
estimate moments of distribution (in the first instance), they should be asked to
review quantiles or probabilities of predictive distribution; frequent feedback
should be given to the experts during the elicitation procedure; and, experts should
be asked to evaluate hypothetical data, unconditionally and conditionally.
3.4.2 Equiprobable Intervals Method
This section discusses a methodology for eliciting an expert’s prior knowledge,

given by Raiffa (1968), who uses the method of equiprobable intervals. Subjective
probability refers to the degree of belief in a proposition. At one extreme, there is
P(A)=1 if event A is trusted to be completely true; and at the other, there is
P(A)=0 if event A is trusted to be completely false, so the points in the interval
[0,1] express beliefs that lie between P(A)=1 and P(A)=0.
Therefore, this method is based on successive subdivisions of equiprobable
intervals (intervals with equal probability), i.e., percentiles, about which the
interview with the expert takes place. This methodology is structured as follows:
1. Explain the process to the expert in general terms, warning him/her of the fact
that the goal is to estimate the most likely value for T and not its exact real
value;
2. Establish a range of possible values of T. Define the minimum expected value
of T (the minimum value of the event that is unlikely to occur – a false
event) , and the maximum expected value of T (the maximum value of the
entire event that is likely to occur – a true event);
3. Start subdivision into equiprobable intervals, initially obtaining the value T,
for which P(T ;
4. Divide the interval between T and T, thus obtaining T where
P(T ;
5. Divide the interval between T and T, thus obtaining T, where
P(T ;
6. Repeat the procedure for the division of other percentiles that need analysis
(TTT, T,TT
7. In the final step, apply a consistency test on the expert, by asking him/her:
What is the range in which ș is most likely to fall? Is it within or outside the
range T0.25 and T0.75? For this question the expert may only give only one of
three answers: within, outside, or indifferent. In this case, the correct answer
would be indifferent, because, if there is consistency in the elicited values, the
probability of being within or outside the range is 0.5. Should the expert answer
either within or outside, one must reevaluate the points with the expert because
either answer appears to be inconsistent, i.e., there was probably some
inconsistency.
After having determined the percentiles and checked the consistency thereof, a
statistical analysis will be undertaken in order to fit the points to a given
probability distribution function.
3.4.3 Experts’ Knowledge Aggregation
In the context of decision-making and risk assessment, the required information is

not always complete or available (Zio 1996), or when there is a need to consider
the uncertainty, experts must quantify their knowledge and generate a distribution
of subjective probability.
Should the DMs require as much more information as possible, they can
consult other experts who have more information or knowledge, and preferably
those who have skills in and knowledge of the area of interest, and thus several
experts can be used.
However, the absence of any knowledge based on data, models, analogies,
theories, physical principles, etc. to assist the experts, can result in judgments that
are mere “assumptions” (Garcez et al. 2011).
For Fischer (1981), assessments of subjective probabilities can improve
substantially when the opinions of a group of experts are aggregated, so that more
than just a probability distribution is considered. However, the expert must be
rational when evaluating the uncertainty of the results, and the expert’s views
must be internally consistent with the theory of probability.
Winkler et al. (1992) list several reasons why the knowledge of multiple
experts should be combined:
1. The combined probability distribution produces a better overview than a single
probability distribution, both from the perspective of a psychological
standpoint (as in the idiomatic expression: two heads are better than one) or a
statistical standpoint (when representation by the average of several samples is
better than the average of a single sample);
2. The set of probability distributions may be considered as a form of agreement
between the various expert’s knowledge;
3. It is more reasonable and practical to use a single probability distribution than
several distributions. Therefore the analysis is more complete.
When the probability distributions represent the judgments of several experts, a
distribution can be obtained that will represent the consensus between them. Thus,
the problem of determining this distribution may be treated as a probability
distribution agreement/aggregate/combined problem (Winkler and Cummings
1972; Hampton et al. 1973; Ekel et al. 2009). This probability distribution must
fully reflect the information provided by these experts (Winkler 1981; Kaplan
1992).
To justify using an aggregate of expert’s knowledge, Fischer (1981) argues that
the general individual probability forecasts tend to be too radical, i.e., events that
are considered highly likely to occur are much less frequent than expected; and
events that are considered extremely unlikely to occur, occur much more fre-
quently than expected. Thus, the evaluation of the opinions of multiple experts
enables a less radical view of the probability of the event to be reached.
In contrast, Clemen and Winkler (1999) argue that a group of experts can
defend a course of action that is more risky than that by an individual or a group
of experts reached without discussion. This is probably because experts rely on
information provided by others, or there is a sharing of responsibilities among
experts.
In order to choose which procedure (method) should be used to aggregate
experts’ knowledge, it is necessary to consider pragmatic issues such as cost and
acceptance. Cost considerations generally favor using simpler procedures, such as,
the statistical average. However, when there are considerations that affect
acceptability, it is likely that more complex aggregation procedures for interaction
between experts are more favorable, such as face-to-face procedures or using the
Delphi methodology, for example (Fischer 1981).
Clemen and Winkler (1999) list some general guidelines to determine what
approach to aggregating knowledge from experts should be considered:
x What information is provided by experts? Is the probability distribution
complete? It is not, if there is only partial information on some of these
distributions (e.g., means, variances, etc.);
x Who is involved? A single or a group of experts?;
x What degree of modeling should be performed?;
x What type of aggregation rules are to be used?;
x What parameters necessary for the aggregation method? (e.g., setting weights),
and;
x What is the level of complexity of the aggregation process to be adopted?
In the literature, there are two main approaches to aggregating experts’
knowledge (opinions), when it is represented by a probability distribution: the
mathematical approach and behavioral approach.
The mathematical aggregation procedures consist of analytical models that
work on each individual probability distribution so as to produce a combined
probability distribution. Aggregation in the behavioral approach tries to generate
associations between experts by their interacting with each other and reaching
agreement. This can be face-to-face or may involve the exchange of information
without direct contact. This approach considers the quality of individual
information and dependence between these (Garcez et al. 2011).
Instead of probabilities aggregation, there are other approaches related to fuzzy
logic, which may be found in the literature. Ekel et al. (2009) specify two main
approaches to reach a consensus: first, expert’s opinions are combined into a
collective opinion, using weighted aggregation. The disadvantage of this approach
is when there is an expert who has a deep knowledge of the problem and there is a
discrepancy between her/him and other experts. Also, there is a disadvantage
when an expert can be neglected due to reducing the weight to his/her opinions,
and also defining the set of weights may require significant computational effort.
The second approach, described by Ekel et al. (2009), is to maintain the
weights of each expert constant. To reach consensus, the weight given to the
References 155
expert who most disagrees with the rest of the group is reevaluated. A
disadvantage of this approach is that an expert who disagrees may have to change
his/her opinion drastically (perhaps unjustifiably), or this expert may be repeatedly
asked to revise the opinions of his/her initial position, which requires greater
intellectual effort.
Furthermore, decisions typically require the multiple views of different experts,
as a single person may not have sufficient knowledge about the problem and
therefore cannot solve it alone (Ekel et al. 2009; Parreiras et al. 2010).
As to aggregating experts’ knowledge, even though experts can agree what the
relevant variables to be analyzed are, this does not mean that they have a
consensus on the probability distribution. If they do not disagree on any point,
there is no need to consult more than one expert and therefore there would be no
need to make expert aggregation (Clemen and Winkler 1999).
In other words, the members of the group would have a uniform opinion and
therefore their knowledge would be the same as the one that would have been
made, if there had been only one expert. Although such a situation rarely occurs,
in cases where the consequences of taking wrong decisions are potentially very
serious, when experts are selected, it may be valuable to make a preliminary effort
to determine whether they do disagree over the probability distribution of the
variables.
To increase the overall satisfaction level of the solution (collective opinion),
experts should have the chance to influence the consensus by providing
information about their individual knowledge.
References
Ackermann F, Howick S, Quigley J, et al. (2014) Systemic risk elicitation: Using causal maps to
engage stakeholders and build a comprehensive view of risks. Eur J Oper Res 238:290–299
Ahmad R, Kamaruddin S (2012) An overview of time-based and condition-based maintenance in
industrial application. Comput Ind Eng 63(1):135–149
Ale B, Burnap P, Slater D (2015) On the origin of PCDS – (Probability consequence diagrams).
Saf Sci 72:229–239
Alencar MH, de Almeida AT (2010) Assigning priorities to actions in a pipeline transporting
hydrogen based on a multicriteria decision model. Int J Hydrogen Energy 35(8):3610–3619
Al-Kassab J, Ouertani ZM, Schiuma G, Neely A (2014) Information visualization to support
management decisions. Int J Inf Technol Decis Mak 13(2):407–428
Andrews JD, Dunnett SJ (2000) Event-tree analysis using binary decision diagrams. Reliab IEEE
Trans 49(2):230–238
Andrews JD, Moss TR (2002) Reliability and risk assessment. Wiley-Blackwell, Suffolk
Arendt JS, Lorenzo DK (2000) Evaluating process safety in the chemical industry: a user’s guide
to quantitative risk analysis. American Chemistry Council
Assael MJ, Kakosimos KE (2010) Fires, explosions, and toxic gas dispersions: Effects
calculation and risk analysis. CRC Press Taylor & Francis Group, Florida
Aven T (2008) Risk analysis: assessing uncertainties beyond expected values and probabilities.
John Wiley & Sons, Chichester
Aven T (2012) Foundations of Risk Analysis. 2nd ed. John Wiley & Sons, p 198, Chichester
Aven T, Jensen U (2013) Stochastic Models in Reliability. Stochastic Modelling and Applied
Probability. Springer, New York
Aven T, Vinnem JE (2007) Risk Management with applications from the offshore petroleum
industry. Springer Series in Reliability Engineering. Springer-Verlag, London
Baker RD, Christer AH (1994) Review of delay-time OR modelling of engineering aspects of
maintenance. Eur J Oper Res 73(3):407–422
Barlow R, Hunter L (1960) Optimum Preventive Maintenance Policies. Oper Res 8:90–100
Barlow RE, Hunter LC (1961) Reliability Analysis of a One-Unit System. Oper Res 9:200–208
Barlow RE, Proschan F (1965) Mathematical theory of reliability. John Wiley & Sons,
New York
Barlow RE, Proschan F (1975) Statistical theory of reliability and life testing: probability
models. DTIC Document
Bazovsky I (2004) Reliability theory and practice. Dover Publications, Mineola
Bedford T, Cooke R (2001) Probabilistic Risk Analysis: Foundations and Methods. Cambridge
BeńDaya M (2000) You may need RCM to enhance TPM implementation. J Qual Maint Eng
6(2):82–85
Bérenguer C, Grall A, Dieulle L, Roussignol M (2003) Maintenance policy for a continuously
monitored deteriorating system. Probab Eng Informational Sci 17(2):235–250
Media, New York
Birolini A (2014) Reliability Engineering. Theory and Practice. Springer Berlin Heidelberg
Bostrom A, Anselin L, Farris J (2008) Visualizing Seismic Risk and Uncertainty. Ann. N. Y.
Acad. Sci. Blackwell Publishing Inc, pp 29–40
Braglia M (2000) MAFMA: multíattribute failure mode analysis. Int J Qual Reliab Manag
17(2):1017–1033
Braglia M, Frosolini M, Montanari R (2003) Fuzzy TOPSIS approach for failure mode, effects
and criticality analysis. Qual Reliab Eng Int 19(5):425–443
Brissaud F, Charpentier D, Fouladirad M, et al. (2010) Failure rate evaluation with influencing
factors. J Loss Prev Process Ind 23(2):187–193
Brito AJ, de Almeida AT (2009) Multi-attribute risk assessment for risk ranking of natural gas
pipelines. Reliab Eng Syst Saf 94(2):187–198
BS EN 60706-2 (2010) Maintenance. Maintenance terminology, British Standards Institution
Calixto E (2013) Gas and Oil Reliability Engineering: Modeling and Analysis. Gulf Professional
Publishing, Oxford
Carter ADS (1986) Mechanical reliability. Macmillan London
Chang C, Wei C, Lee Y (1999) Failure mode and effects analysis using fuzzy method and grey
theory. Kybernetes 28(9):1072–1080
Chang C-C (2014) Optimum preventive maintenance policies for systems subject to random
working times, replacement, and minimal repair. Comput Ind Eng 67:185–194
Clemen RT, Winkler RL (1999) Combining Probability Distributions from Experts in Risk
Analysis. Risk Anal 19:187–203
Corder AS (1976) Maintenance management techniques. McGraw-Hill
Media, New York
Crowl DA, Louvar JF (2001) Chemical Process Safety: Fundamentals with applications. Prentice
Hall, Boston
Dekker R (1995) On the use of operations research models for maintenance decision making.
Microelectron Reliab 35(9):1321–1331
Dekker R (1996) Applications of maintenance optimization models: a review and analysis.
Reliab Eng Syst Saf 51(3):229–240
References 157
Dekker R, Scarf PA (1998) On the impact of optimisation models in maintenance decision

making: the state of the art. Reliab Eng Syst Saf 60(2):111–119
Dong C (2007) Failure mode and effects analysis based on fuzzy utility cost estimation. Int J
Qual Reliab Manag 24(9):958–971
Drapella A, Kosznik S (2002) Combining preventive replacement and burn-in procedures. Qual
Reliab Eng Int 18(5):423–427
Ekel P, Queiroz J, Parreiras R, Palhares R (2009) Fuzzy set based models and methods of
multicriteria group decision making. Nonlinear Anal Theory, Methods Appl 71:e409–e419
Eppler MJ, Aeschimann M (2009) A systematic framework for risk visualization in risk
management and communication. Risk Manag 11(2):67–89
Ericson CA (2005) Hazard analysis techniques for system safety. John Wiley & Sons
Fedra K (1998) Integrated risk assessment and management: overview and state of the art.
J Hazard Mater 61:5–22
Finkelstein M (2008) Failure rate modelling for reliability and risk. Springer Science & Business
Media, London
Fischer GW (1981) When oracles fail – A comparison of four procedures for aggregating
subjective probability forecasts. Organ Behav Hum Perform 28:96–110
Fjeld RA, Eisenberg NA, Compton KL (2007) Quantitative environmental risk analysis for
human health. John Wiley & Sons
Fouladirad M, Grall A (2014) On-line change detection and condition-based maintenance for
systems with unknown deterioration parameters. IMA J Manag Math 25(2):139–158
Garbatov Y, Guedes Soares C (2001) Cost and reliability based strategies for fatigue
maintenance planning of floating structures. Reliab Eng Syst Saf 73(3):293–301
Garcez TV, Almeida-Filho AT de, de Almeida AT (2011) Procedures for aggregating experts’
knowledge and group decision model approaches. In: Bérenguer C, Grall A, Soares CG (eds)
20th European Safety and Reliability (ESREL 2011) annual conference, Troyes, September
2011. Safety, Reliability and Risk Management. 2012. Taylor and Francis, London, p 3076
Garcez TV, Almeida-Filho AT de, de Almeida AT, Alencar MH (2010) Experts’ elicitation of
prior knowledge on accidental releases in a natural gas pipeline. In: Bris R, Soares CG,
Martorell S (eds) European safety and reliability conference, Prague, September 2009.
Reliability, Risk, and Safety: Theory and Applications, Vol. 1-3. 2009. Taylor and Francis,
London, UK, p 2480
Garcez TV, de Almeida AT (2014) Multidimensional Risk Assessment of Manhole Events as a
Decision Tool for Ranking the Vaults of an Underground Electricity Distribution System.
Power Deliv IEEE Trans 29(2):624–632
Garthwaite PH, Kadane JB, O’Hagan A (2005) Statistical Methods for Eliciting Probability
Distributions. J Am Stat Assoc 100:680–701
Gertsbakh IB (1977) Models of preventive maintenance. North-Holland, New York
Glasser GJ (1969) Planned replacement- Some theory and its application (Probability theory
applied to age and block replacement models in preventive maintenance of parts, noting
inspection cost distribution). J Qual Technol 1:110–119
Grall A, Bérenguer C, Dieulle L (2002) A condition-based maintenance policy for stochastically
deteriorating systems. Reliab Eng Syst Saf 76(2):167–180
Guedes Soares C, Garbatov Y (1996) Fatigue reliability of the ship hull girder accounting for
inspection and repair. Reliab Eng Syst Saf 51(3):341–351
Hampton JM, Moore PG, Thomas H (1973) Subjective probability and its measurement. J R Stat
Soc Ser A 136:21–42
Horwitz R (2004) Hedge Fund Risk Fundamentals: Solving the risk management and
transparency challenge. John Wiley & Sons
Hotelling H (1925) A General Mathematical Theory of Depreciation. J Am Stat Assoc
20(151):340–353
HSE (1987) The Tolerability of Risk from Nuclear Power Stations. HMSO - Health and Safety
Executive, London
Jaedicke C, Syre E, Sverdrup-Thygeson K (2014) GIS-aided avalanche warning in Norway.

Comput Geosci 66:31–39
Jardine AKS (1973) Maintenance, Replacement and Reliability. John Wiley, New York
Jiang R, Jardine AKS (2007) An optimal burn-in preventive-replacement model associated with
a mixture distribution. Qual Reliab Eng Int 23:83–93
Jiang R, Murthy DNP, Ji P (2001) Models involving two inverse Weibull distributions. Reliab
Eng Syst Saf 73(1):73–81
Jorgenson DW, McCall JJ (1963) Optimal Replacement Policies for a Ballistic Missile. Manage
Sci 9(3):358–379
Jorgenson DW, McCall JJ, Radner R (1967) Optimal Replacement Policies. Rand McNally
Kadane JB, Wolfson LJ (1998) Experiences in Elicitation. J R Stat Soc Ser D The Stat 47:3–19.
Kaplan S (1992) “Expert information” versus “expert opinions”. Another approach to the
problem of eliciting/combining/using expert knowledge in PRA. Reliab Eng Syst Saf 35:61–72
Kaufmann R, Häring I (2014) Comparison of 3D visualization options for quantitative risk
analyses. In: Steenbergen RDJM, VanGelder PHAJM, Miraglia S, Vrouwenvelder ACWMT
(eds) 22nd Annual Conference on European Safety and Reliability (ESREL), Amsterdam,
2013. Safety, Reliability and Risk Analysis: Beyond the Horizon. Taylor & Francis Group,
London, UK, p 758
Keeney RL, von Winterfeldt D (1991) Eliciting probabilities from experts in complex technical
problems. IEEE Trans Eng Manag 38:191–201
Kelly A (1983) Maintenance planning and control. Butterworths, London
Khazraei K, Deuse J (2011) A strategic standpoint on maintenance taxonomy. J Facil Manag
9(2):96–113
Kuo W, Prasad VR (2000) An annotated overview of system-reliability optimization. Reliab
IEEE Trans 49(2):176–187
Kuo W, Zhu X (2012) Importance Measures in Reliability, Risk, and Optimization: Principles
and Applications. John Wiley & Sons, Chichester
Kuo W, Zuo MJ (2003) Optimal Reliability Modeling: Principles and Applications. John Wiley
& Sons, New Jersey
Levitin G, Lisnianski A (2000) Optimization of imperfect preventive maintenance for multi-state
systems. Reliab Eng Syst Saf 67(2):193–203
Levitt J (2003) Complete guide to preventive and predictive maintenance. Industrial Press
New York
Lewis EE (1987) Introduction to reliability engineering. Wiley, New York
Lins PHC, de Almeida AT (2012) Multidimensional risk analysis of hydrogen pipelines. Int J
Hydrogen Energy 37:13545–13554
Macdonald D (2004) Practical – Hazops, Trips and Alarms. Newnes – Elsevier, Oxford
Manzini R, Regattieri A, Pham H, Ferrari E (2009) Maintenance for Industrial Systems. Springer
London
Márquez AC (2007) The maintenance management framework: models and methods for
complex systems maintenance. Springer Science & Business Media
Martorell S, Sanchez A, Serradell V (1999) Age-dependent reliability model considering effects
of maintenance and working conditions. Reliab Eng Syst Saf 64(1):19–31
Martz HF, Waller RA (1982) Bayesian Reliability Analysis. John Wiley & Sons, New York
McCall JJ (1965) Maintenance Policies for Stochastically Failing Equipment: A Survey. Manage
Sci 11(5):493–524
Mobley RK, Higgins LR, Wikoff DJ (2008) Maintenance engineering handbook. McGraw-Hill
Modarres M, Kaminskiy M, Krivtsov V (1999) Reliability Engineering and Risk Analysis:
A Practical Guide. CRC Press
Moubray J (1997) Reliability-centered maintenance. Industrial Press Inc., New York
Nakagawa T (1984) A summary of discrete replacement policies. Eur J Oper Res 17(3):382–392
Nakagawa T (1989) A replacement policy maximizing MTTF of a system with several spare
units. Reliab IEEE Trans 38:210–211
References 159
Nakajima S (1988) Introduction to TPM: total productive maintenance. Productivity Press

Nelson WB (2004) Applied life data analysis. John Wiley & Sons
Newbrough ET, Ramond A (1967) Effective maintenance management: organization,
motivation, and control in industrial maintenance. McGraw-Hill, New York
Nolan DP (2011) Handbook of fire and explosion protection engineering principles for oil, gas,
chemical and related facilities. Gulf Professional Publishing, Oxford
Nowlan FS, Heap HF (1978) Reliability-centered Maintenance. Dolby Access Press
NRC (1986) Safety Goals for Nuclear Power Plants. US Nuclear Regulatory Commission
NUREG-0880
Nwaoha TC, Yang Z, Wang J, Bonsall S (2013) A fuzzy genetic algorithm approach for analysing
maintenance cost of high risk liquefied natural gas carrier systems under uncertainty. J Mar
Eng Technol 12(2):57–73
O’Connor P, Kleyner A (2012) Practical reliability engineering. John Wiley & Sons, Chichester
O’Hagan A (1998) Eliciting expert beliefs in substantial practical applications. J R Stat Soc Ser
D The Stat 47:21–35
O’Hagan A, Oakley JE (2004) Probability is perfect, but we can’t elicit it perfectly. Reliab Eng
Syst Saf 85:239–248
Parreiras RO, Ekel PY, Martini JSC, Palhares RM (2010) A flexible consensus scheme for
multicriteria group decision making under linguistic assessments. Inf Sci (Ny) 180:1075–
1089
Paté-Cornell E, Cox LA Jr (2014) Improving Risk Management: From Lame Excuses to
Principled Practice. Risk Anal 34(7):1228–1239
Pham H (1999) Software reliability. John Wiley & Sons
Pinjala SK, Pintelon L, Vereecke A (2006) An empirical investigation on the relationship
between business and maintenance strategies. Int J Prod Econ 104(1):214–229
Pintelon LM, Gelders LF (1992) Maintenance management decision making. Eur J Oper Res
58(3):301–317
Puente J, Pino R, Priore P, Fuente D de la (2002) A decision support system for applying failure
mode and effects analysis. Int J Qual Reliab Manag 19(2):137–150
Radner R, Jorgenson DW (1963) Opportunistic Replacement of a Single Part in the Presence of
Several Monitored Parts. Manage Sci 10(1):70–84
Wesley, London
Rausand M (2011) Risk assessment. Theory, Methods, and Applications. Wiley, New Jersey
Rausand M, Høyland A (2004) System reliability theory: models, statistical methods, and
applications, vol 396. John Wiley & Sons, New Jersey
Rausand M, Vatn J (2008) Reliability Centred Maintenance. Complex Syst. Maint. Handb.
SE - 4. Springer London, pp 79–108
Reis ACB, Costa APCS, de Almeida AT (2009) Planning and competitiveness in maintenance
management: An exploratory study in manufacturing companies. J Qual Maint Eng 15:259–
270
Rosqvist T, Laakso K, Reunanen M (2009) Value-driven maintenance planning for a production
plant. Reliab Eng Syst Saf 94(1):97–110
Scarf PA (1997) On the application of mathematical models in maintenance. Eur J Oper Res
99(3):493–506
Scarf PA, Cavalcante CAV (2010) Hybrid block replacement and inspection policies for a multi-
component system with heterogeneous component lives. Eur J Oper Res 206(2):384–394
Scarf PA, Cavalcante CAV (2012) Modelling quality in replacement and inspection
maintenance. Int J Prod Econ 135(1):372–381
Scarf PA, Cavalcante CAV, Dwight RA, Gordon P (2009) An Age-Based Inspection and
Replacement Policy for Heterogeneous Components. Reliab IEEE Trans 58(4):641–648
Sherif YS (1982) Reliability analysis: Optimal inspection and maintenance schedules of failing
systems. Microelectron Reliab 22:59–115
Shirose K (1992) TPM for Workshop Leaders. Productivity Press, New York
Smith DJ (2011) Reliability, Maintainability and Risk. Practical methods for engineers. BH
(Elsevier), Oxford
Smith DJ, Simpson KGL (2010) Safety Critical Systems Handbook. A straightforward guide to
functional safety, IEC 61508 (2010 Edition) and related standards. BH, Oxford
Stephans RA (2004) System safety for the 21st century. The update and revised edition of system
safety 2000. Wiley-Interscience, New Jersey
Sutton I (2010) Process Risk and Reliability Management Operational Integrity Management,
William Andrew – Elsevier, Oxford
Tajiri M, Gotoh F (1992) TPM implementation, a Japanese approach. McGraw-Hill Companies
Tariq M (2013) Risk-based flood zoning employing expected annual damages: the Chenab River
case study. Stoch Environ Res Risk Assess 27:1957–1966
Taylor JS (1923) A Statistical Theory of Depreciation. J Am Stat Assoc 18(144):1010–1023
Terborgh GW (1949) Dynamic equipment policy. McGraw-Hill Book Co
Thangaraj V, Rizwam U (2001) Optimal replacement policies in burn-in process for an
alternative repair model. Int J Inf Manag Sci 12(3):43–56
Theodore L, Dupont RR (2012) Environmental Health and Hazard Risk Assessment. Principles
and Calculations. CRC Press, Boca Raton
Tian Z, Zuo MJ (2006) Redundancy allocation for multi-state systems using physical
programming and genetic algorithms. Reliab Eng Syst Saf 91(9):1049–1056
Tweeddale M (2003) Managing Risk and Reliability of Process Plants. Gulf Professional
Publishing. Burlington
Van Leeuwen CJ, Vermeire TG (2007) Risk Assessment of Chemicals. An introduction.
Springer, Dordrecht
Vinnem J-E (2014) Offshore Risk Assessment. Principles, Modelling and Applications of QRA
Studies Vol. 2, Springer-Verlag, London
Walley P (2002) Reconciling frequentist properties with the likelihood principle. J Stat Plan
Inference 105:35–65
Wang H, Pham H (2006) Reliability and Optimal Maintenance. Springer-Verlag, London
Wang W (2012) An overview of the recent advances in delay-time-based maintenance
modelling. Reliab Eng Syst Saf 106:165–178
Winkler RL (1981) Combining Probability Distributions from Dependent Information Sources.
Manage Sci 27:479–488
Winkler RL, Cummings LL (1972) On the choice of a consensus distribution in Bayesian
analysis. Organ Behav Hum Perform 7:63–76
Winkler RL, Hora SC, Baca RG (1992) The quality of expert judgment elicitations. San Antonio,
TX: Center for Nuclear Waste Regulatory Analyses
Woodman RC (1967) Replacement policies for components that deteriorate. OR 18:267–280.
Yang Z, Bonsall S, Wang J (2008) Fuzzy Rule-Based Bayesian Reasoning Approach for
Prioritization of Failures in FMEA. Reliab IEEE Trans 57:517–528
Yoe C (2012) Principles of Risk Analysis – Decision Making under uncertainty. CRC Press,
Boca Raton
Zammori F, Gabbrielli R (2012) ANP/RPN: a multi criteria evaluation of the Risk Priority
Number. Qual Reliab Eng Int 28:85–104
Zheng X (1995) All opportunity-triggered replacement policy for multiple-unit systems. Reliab
IEEE Trans 44(4):648–652
Zheng X, Fard N (1991) A maintenance policy for repairable systems based on opportunistic
failure-rate tolerance. Reliab IEEE Trans 40(2):237–244
Zio E (1996) On the use of the analytic hierarchy process in the aggregation of expert judgments.
Reliab Eng Syst Saf 53:127–138
Zio E (2007) An Introduction to the basics of Reliability and Risk Analysis. Series in Quality,
Reliability and Engineering Statistics vol 1. World Scientific, Singapore
Chapter 4
Multidimensional Risk Analysis
Abstract: Accidents involve critical consequences that require an appropriate and

efficient form of risk management. A multidimensional risk analysis allows a
broader view. MCDM/A approaches enable more consistent decision-making,
taking into account the DM’s rationality (compensatory or non-compensatory),
DM’s behavior regarding risk (prone, neutral or averse) and the uncertainties
inherent in the risk context. This chapter presents numerical applications
illustrating the use of multicriteria models in two different contexts: a natural gas
pipeline and an underground electricity distribution system. Two different
MCDM/A approaches are considered: MAUT (Multiattribute Utility Theory) and
the ELECTRE TRI outranking method. In the numerical applications, MCDM/A
approach steps for building decision models are presented: identifying hazard
scenarios, estimating the set of payoffs, eliciting the MAU function (Multi-attribute
Utility function), computing the probability function of consequences and estimating
multidimensional risk. Loss functions are introduced in the models to calculate the
probability distribution functions over the multiple criteria such as impact on
humans, and environmental and financial losses. Therefore, Decision Theory con-
cepts are applied to estimate risk in industrial plants and modes of transportation.
Finally, other decision problems related to multidimensional risk analysis, using
MCDM/A, are considered in different contexts, such as: power electricity systems,
natural hazards, risk analysis on counter-terrorism, nuclear power plant.
4.1 Justifying the Use of the Multidimensional Risk
The perceived level of risk is directly linked to the perceived intensity of

consequences to people and society as well as to issues related to the level of
probability. These consequences are multidimensional and are associated to the
objectives, represented by criteria and can be approached with an MCDM/A or a
multiobjective method (see Chap. 2). Many studies show that using a single
dimension of risk may not be realistic (Morgan et al. 2000; Willis et al. 2005;
Apostolakis and Lemon 2005; Brito and de Almeida 2009; Garcez et al. 2010;
Alencar et al. 2010; Alencar and de Almeida 2010; Brito et al. 2010; Garcez and
de Almeida 2014b; Lins and de Almeida 2012; Garcez and de Almeida 2014c).
The perception of risk and its tolerability is highly affected by recent events.
For example, in the maritime risk context after accidents such as Amoco Cadiz

162 Chapter 4 Multidimensional Risk Analysis
(in 1978), Derbyshire (in 1980), Herald Free Enterprise (in 1987) and Piper Alpha
(in 1988) many maritime sectors started to seek for the improvement and
application of risk modeling and decision making techniques (Wang 2006). Such
risk evaluations and safety concerns has to be observed not only by companies
involved with the specific context, as maritime transportation and offshore
operations, but also other companies related with the sector, in this case, ship
designers and shipbuilders in order to improve safety (Guedes Soares and Teixeira
2001).
Since the widely accepted notion of risk (see Chap. 3) is also based on
consequences, there is a need to estimate consequences/loss/severity. For many
authors, risk assessment deals with estimating possible losses and is an essential
procedure, whose outcome is the foundation on which the DM justify his/her
decisions. Thus, the consequences of an event can be represented based on some
of these aspects, e.g., the number of fatalities, the number of people injured,
financial loss, damage to property, environmental losses, etc. (Alencar et al. 2014;
Alencar and de Almeida 2010; Brito et al 2010; Luria and Aspinall 2003).
Furthermore, both Individual Risk and Societal Risk concepts only consider the
scale of human loss. Cox (2009) argues that, in the risk context, a rational decision
making seeks to ensure that a risk analysis builds evaluations and comparisons of
proposed risk management actions and interventions, not merely describing the
current situation.
In some studies, under a more conservative view, the risk of human loss is
assessed from the perspective of a result of the occurrence of injuries, and not only
fatalities. Therefore, precisely how people are injured (for example, first or second
degree burns) are considered as a consequence for the calculation of risk (Brito
and de Almeida 2009). In this context, Cox (2009) reinforces that dominated
actions should be eliminated, choosing the best option among non-dominated
alternatives and guaranteeing that those alternatives are not ignored. An evaluation
of the total consequences is necessary, in order to provide an effective risk
management. For each alternative, the overall consequences are calculated taking
into account the summing of all the impacts of proposed alternatives on humans
exposures.
Although, there is a need for a multidimensional risk view, under the relevant
aspects of various ways of analyzing the result of accidents in industrial plants and
modes of transportation in various parts of the world, most of the studies consider
only issues related to a single dimension. Doing so makes a single dimension
approach inadequate or incomplete when the issues involved are complex.
Furthermore, nowadays, analyzing the consequences must satisfy the expectations
of society, the state (public sector) and private companies. The magnitude and
severity of the consequences make it essential to develop a more appropriate and
efficient form of risk management, which provides for positive outcomes. In this
sense, Beaudouin and Munier (2009) present a critic on industrial risk manage-
ment techniques based on procedures derived from health, safety, and environ-
ment within quality management programs, and draw attention that decision
4.1 Justifying the Use of the Multidimensional Risk 163
analysis techniques derived from experiments and theoretical foundations are

more efficient practices for risk management. Hence, there is a need for a
multidimensional assessment, which enables more consistent decision-making to
be made and which takes into account the DM’s preferences and the context of
uncertainty.
Almeida-Filho and de Almeida (2010b) emphasize that risk has been a topic of
interest for many years, however the majority of studies avoid considering
multiple risk dimensions, in most of the cases there is multiple risk dimensions
and it’s evaluated through different indexes that are difficult to aggregate into a
joint evaluation. There have been used in the literature risk evaluation frameworks
from NORSOK and ISO to evaluate risk in an oil and gas context, however, these
approaches do not provide a multiple dimension evaluation, only seeks to achieve
tolerable risk levels, disregarding decision maker’s judgment about the relation
between different risk dimension levels and the level difference in each risk
dimension. Thus they presented a framework based on well-established risk
evaluation framework in the literature (NORSOK and ISO) considering the
multidimensional risk aspects.
Still under the one-dimensional view of risk, many studies and risk analyzes
take as a basis the financial aspect associated with a monetary value as a criterion
of the loss to be used. This approach could appear to be broader, because it
regards risk from a more managerial point of view and by analyzing costs.
However, considering only the financial aspect is not always an appropriate
measure. This can be verified, for example, when the monetary value is not the
only measure of value or when certain considerations cannot or should not be
converted into an equivalent financial value, e.g. fatalities.
Tweeddale (2003) points out that establishing which economic factors are
associated with risk is a point widely discussed in the literature, in which different
approaches are taken. A critical point in this context is the attempt to associate a
financial value with the loss of a human life. For many, a human life is priceless.
For others there is the question of the emotional value of life for friends and
family that cannot be compensated by any amount of money. Moreover, according
to Hobbs and Meier (2000), some value judgments of interest such as the value of
a human life are made by analysts and cannot be properly dealt with in calculations.
Hobbs and Meier (2000) states that other aspects may also be considered with
respect to the monetization of criteria, such as the issue that some techniques
associated with monetization may be difficult to implement, or even impossible to
apply in practice, thus increasing the time required to attempt to do so or this may
lead to less suitable methods being used.
In this same perspective, Bedford and Cooke (2001) state that cost-benefit
analysis is a well-established method where monetary values are defined for a
particular unit (for example, human life). Cost-benefit analysis is used to guide the
decision-making process in the area of the ALARP principle (see Chap. 3). Thus,
cost-benefit analysis reflects how society prioritizes the various attributes
considered, which in principle will be the dimensions of human and financial loss.
For most of the contexts, besides risk, the cost-benefit analysis is appropriate
for capturing the society priorities, rather than individual preferences, which are
better captured by MCDM/A methods. The former is related to Societal Risk
concept, whereas the latter is related to Individual Risk concept. This issue is
related to the comparison between the use of cost-benefit analysis and MCDM/A
methods (Almeida-Filho and de Almeida 2010a).
Another important point to be highlighted is that financial losses cannot always
be measured with complete accuracy. This occurs due to firms having to take
account of pressure groups in society who are well informed of possible dangers
that their companies may present to society at large. To counter this, several
companies have a strategy for differentiating themselves from their competitors.
This includes creating an image that the care for the environment, and that their
first priority is to safeguard the safety of their employees, their customers and the
community they form part of. Nevertheless, when any kind of accident occurs that
bring losses to any of the “users” of the system, there is pressure from society not
to consume the products of this company. These results in losses to the company
that are not only brought about by the accident itself (in the monetary perspective)
but also because they lose customers and suppliers; contracts may be broken; their
business image is damaged. These losses cannot be “easily” or completely
(precisely) measured in financial terms.
Hence, traditional approaches to risk analysis do not consider the multiple
dimensional impacts (consequences) that industrial accidents may cause. How-
ever, a multidimensional risk view, in many different contexts is necessary.
Furthermore, nowadays, analyzing the consequences must satisfy the
expectations of society, the state (public sector) and private companies. The
magnitude and severity of the consequences make it essential to develop a more
appropriate and efficient form of risk management, which provides for positive
outcomes. In other words, the results must be at an acceptable level of safety and
also from an economic point of view, the survival of the company is necessarily
called into question in the sense that the cost of taking measures that prevent and
mitigate risks has to be balanced against the likelihood of accidents happening to
people or extensive damage to the environments exposed to them.
As already shown in several studies, an approach to risk assessment that uses
only a single dimension of risk cannot be sufficiently comprehensive to ensure
that the most realistic and efficient assessment of risk is made (Alencar and de
Almeida 2010; Apostolakis and Lemon 2005; Brito and de Almeida 2009; Brito et
al. 2010; Garcez and de Almeida 2014b; Lins and de Almeida 2012; Morgan et al.
2000; Willis et al. 2005). Additionally, for Brito and de Almeida (2009), even if
other effects are not as important as the risks to human beings, they also require
substantial attention from DMs.
Hence, there is a need for a multidimensional assessment, which enables more
consistent decision-making to be made and which takes into account the context of
uncertainty risk. In many decision problems, more than one factor influences the
4.1 Justifying the Use of the Multidimensional Risk 165
DM’s preferences with respect to possible outcomes (Montiel and Bickel 2014;
Bedford and Cooke 2001).
According to Salvi et al. (2005), if environmental assessment and risk
management is given more importance, all stakeholders will take part in the
decision-making process. Probably, this feature results from the development of
society, which has increasing access to global information, and this is combined
with people’s concerns related to the sustainable development of society. Society
has also taken a cautious attitude due to experiences caused by industrial disasters
(e.g., Flixborough, Chernobyl, Bhopal, and more recently Fukushima on 11 March
2011).
The occurrence of these and other disasters have shown that there must be
public consultation with various stakeholders, and that this dialogue should not
occur independently of the risk management process, the main objective of which
is to ensure the long term security of populations. Therefore, the maintenance and
consent of an industrial activity is strongly dependent on society accepting the
risks that the activity generates.
Hence, companies are coming to recognize the need to take the different
opinions and preferences of the various stakeholders into account in the risk
decision-making process. MCDM/A methodologies can be extremely useful to
aggregate these different opinions (criteria, preference, weights) so the most
appropriate decision may be taken both at the national and local level (Roy 1996).
Therefore, according to Yoe (2012), what is observed is that a process of
decision making can be simple or complex depending on a few factors that need to
be considered. When the analysis is of a single dimension of the problem and
there is only one DM, the process is simpler. The same is not true under the risk
management process. It is considered to be a complex process due to there being a
number of aspects, such as, the views of interested and involved parties, the
processes of identification, analysis and risk assessment, and the analysis of
consequences, not merely financial impacts.
The process of risk management involves managers and stakeholders with
different values, priorities and objectives. In this process, consideration is given to
such aspects as tradeoffs between risks, costs, benefits, social values and other
impacts of conflicts of values as a result of many perspectives represented by
stakeholders in the decision making process.
Hobbs and Meier (2000) affirm that MCDM/A methods present many positive
as well as negative aspects. The positive points are:
x Emphasis on learning and understanding by the users;
x Tradeoffs more explicit as to the interests involved;
x Values obtained directly by the stakeholders;
x Reject dominated alternatives.
They argue on some the critical points:

x Large amount of information regarding alternatives and criteria (often not
properly interpreted by stakeholders);
x Possible failure of priority of stakeholder groups;
x Improper application of MCDM/A methods, generating distortions of the DM’s
preferences, as well as inconsistencies in value judgments.
However, if care is taken in the definition, study and use of MCDM/A methods,
the potential occurrence of these negative issues can be avoided. Furthermore, this
issue has to be seen for each case. It should be reminded that for the purpose of
making the model useful, the appropriate effort should be made in the model
building process.
Hence, the justification for the use of MCDM/A approaches associated with
managing risk is that is made of a set of techniques, methodologies and models,
with the goal of working in a better way with aspects associated with uncertainty,
understanding conflicts and the tradeoffs involved. Another point highlighted by
Cailloux et al. (2013) is that a multicriteria decision aiding approach helps with
the subjective part of risk assessment.
In strategic risk management, DMs usually have to consider various conflicting
objectives under uncertain decision parameters (Comes et al. 2011). Since
MCDM/A methods are easy to use in structuring complex problems and building
consensus, they have often been used successfully to support DM in emergency
management (Geldermann et al. 2009).
According to Hobbs and Meier (2000), the aim of MCDM/A methods is to
improve the quality of decisions involving multiple criteria by making decisions
more explicit, rational and efficient. Some aspects of this should be considered:
structuring the decision problem; tradeoffs among the criteria; value judgments of
the people involved in the process; helping people develop more consistent
assessments with respect to risk and uncertainty; facilitating negotiation and;
documenting how decisions are made.
For Linares (2002), risk analysis also presents some advantages when
combined with a multicriteria decision approach: it allows including the DM’s
preferences in relation to risk, and may also be taken consistently with a
compromise programming approach.
In the multicriteria approach, there is a multidimensional value because
multiple criteria are taken into consideration. Thus, instead of considering a single
dimension (aspect) such as human or financial loss, other dimensions are taken
into consideration depending on the context studied and persons (entities) that are
part of the decision process.
Some loss dimensions that can be considered in this context are:
x The human dimension, which can take into account the damage to people
affected by the consequences of a failure event which can be estimated by the
number of people affected (injuries and/or fatalities);
4.2 Multidimensional Risk Evaluation Model 167
 The environmental dimension, which may include, e.g., areas affected as a

result of the event (Alencar et al. 2014; Brito et al. 2010; Alencar et al. 2010;
Brito and de Almeida 2009);
 The financial dimension where we can consider monetary losses arising from
events occurring;
 The operational dimension that considers the influence of the consequences of
the event and the behavior of the production system;
 Several others that somehow express the needs or preferences that DMs wish to
consider.
Given the existence of the uncertainty associated with risk analysis, the use of
MAUT to develop models for multicriteria decision is quite appropriate in this
context of risk analysis (Keeney and Raiffa 1976; de Almeida 2007; Brito et al.
2010; de Almeida et al. 2015).
In utility theory, measures are obtained based on multiple attributes, where the
DM establishes the degree of preference for possible multidimensional results
(Keeney and Raiffa 1976; Berger 1985; Bedford and Cooke 2001).
MAUT is used because it presents a well-structured protocol, supported by a
very solid and consistent axiomatic framework for decisions involving multiple
criteria. Moreover, according to Keeney and Raiffa (1976), in the modeling step,
probabilistic uncertainties are inserted within the axiomatic structure, thereby
enabling a more consistent approach to the application of MAUT in multicriteria
decision problems under conditions of uncertainty. Furthermore, the probabilistic
modeling is a complement to modeling the DM’s preferences.
In this context, the two next sections present risk evaluations and decision
models built with the use of MCDM/A methods and the next section presents a
procedure for building models for risk evaluation, using MCDM/A.
4.2 Multidimensional Risk Evaluation Model
This section presents an MCDM/A procedure for building risk evaluation and
decision models, which is adapted from that of Chap. 2, incorporating a specific
situation, in which the DM’s behavior regarding to risk (prone, neutral, averse)
can be approached via utility theory. According to Cox (2012), the application of
utility functions rather than simple risk formulas – composed by terms such as
exposure, probability and consequence - allows to take into account DM’s risk
attitudes, improving the effectiveness of the decision making process to reduce
risks. This procedure has been applied in several contexts described in next
section (Brito and de Almeida 2009; Brito et al. 2010, Alencar and de Almeida
2010; Lins and de Almeida 2012; Garcez and de Almeida 2014a; Garcez and de
Almeida 2014b).
According to Geldermann et al. (2009), emergency situations caused by

humans or by Nature require effective and consistent management, and always
involve complex decisions. Many conflicting objectives need to be solved;
priorities need to be set, while the various perspectives of different stakeholders
should converge towards a consensus.
Brito et al. (2010) state that risk management is a critical activity for many
processes and systems, especially for systems that transport hazardous materials.
The consequences of accidents highlight the importance of developing a proper
and effective risk management technique for this type of process. Additionally, the
complexity inherent in the process of decision making on risks, which involves
considering technical, economic, environmental, political, psychological and
social issues, is an increasingly important aspect of risk management that requires
to be tackled more thoroughly.
The decision model presented in this section uses a multicriteria approach
based on MAUT, and incorporates the DM’s behavior in the decision making
process. The model enables the decision maker (DM) to define actions in priority
classes in order to mitigate risks in the context under consideration. MAUT
provides a well-structured protocol, supported by a solid and consistent axiomatic
framework for making decisions involving multiple criteria. Moreover, in the
probabilistic modeling step, uncertainties are inserted within the axiomatic
structure, thereby enabling a more consistent approach to be taken to a MAUT
application in multicriteria decision problems under uncertainty. This stage of
probabilistic modeling can be understood as a complement to that of modeling the
DM’s preferences. The model to be presented takes into account aspects of
Decision Theory which will be presented in more detail.
According to Berger (1985), during the decision-making process, it is of great
importance to take the possible states of nature into consideration. T is used to
denote the set of all possible states. Typically, when procedures are developed to
obtain information about T, experiments are designed so that the observations are
distributed according to some probability distribution that presents T as a
parameter of uncertainty.
In Decision Theory, there is an attempt to combine information from samples
with other relevant aspects of the problem, thereby enabling the best decision be
made. In addition to this information from samples, two other information types
are relevant. The first is knowledge of the possible consequences of decisions.
Commonly, this knowledge can be quantified by defining the loss (or gain) that is
expected to occur for each possible decision and possible T values. The second
refers to a priori knowledge. Generally, these items of information are derived
from past experiences in similar situations involving similar T.
The model presented is a quantitative model that incorporates the DM’s
preferences and his/her behavior with respect to risk, thus enabling alternatives to
be prioritized by making a hierarchical ranking of the risks, which allows a
multidimensional view to be taken of risks from the perspective of different

consequences. To illustrate the stages of the model, the structure of a decision
model in the context of a natural gas pipeline is shown in Fig. 4.1.
Who is the DM?
What are the What are the hazard What are the
alternatives? scenarios (state of hazard scenarios
nature)? probabilities?
Estimation of the hazard What are the

area/size payoffs set?
Estimation of probabilities of Elicitation of the

consequence functions utility functions
Calculation of risk value
Risk ranking
Fig. 4.1 Structure of decision model
Another important aspect that has prompted developing the model is that it uses
Decision Support Systems (DSS) to assist in routing between steps in order to
make the process more dynamic, thus making it possible for the DM to make a
more detailed study of all stages of the steps of the risk analysis. Furthermore, use
of DSS aims to support the decision making process, and takes into account both
technical aspects such as its stochastic nature and the variety of the parameters
which will be entered into the model as well as factors related to the decision
making process on risk analysis (Lopes et al. 2010).
Finally, it is worth mentioning that the steps of the proposed methodology are
not static. In other words, there is a transition between steps which allows the DM
to return to the previous steps to adjust a parameter in order to make the result
more dynamic and realistic. Further details of these aspects will be observed
throughout the text.
4.2.1 Contextualizing the System
In this step, the system should be contextualized, since it is necessary to describe

the general characteristics of the system. The reason for this is that it is only by
questioning the purposes of the system and why, in overall terms, it is structured
the way it is that some methodological approaches can be more fully understood.
Therefore it is necessary to become familiar in overall terms with technical,
environmental, social and external environment issues that impact the system and
to determine the extent to which each of these, separately, or by interacting with
each other affect the performance of the system. The answers to such questioning
will guide the decision making process within the type of multidimensional risk
analysis that will be selected and applied.
4.2.2 Identifying the Decision Maker
This is the stage used to define who will be responsible for the decision, since it is
this DM’s preference structure which will be adopted. It is extremely important to
identify the DM correctly because decision making in complex environments
(such as transport systems for hazardous products, electric power systems, nuclear
systems, critical infrastructure, etc.) involves potential severely adverse impacts
on society, the environment, economic losses, etc.
Therefore, it is necessary that the DM is thoroughly familiar with the context of
the risk analysis. For example, he/she must be fully alert to possible accident
scenarios, be fully aware of the consequence dimensions of accidents, and be able
to draw up and implement protective and mitigation measures. In other words, not
only must the DM be knowledgeable about the context in which decisions about
risk may have to be taken but also about the needs of the various stakeholders
involved in the decision making process.
It is worth mentioning that the DM’s preferences should reflect the interests
and goals of the organization (company) and also of the managers who are
responsible for any consequences arising from the decision. In some situations it is
necessary to include the preferences of various DMs. This process is characterized
as a group decision, which may involve three main actors: the company
representative of the system considered, the government representative (regulators)
and the representative of the community in which the system is located.
In this model, it is assumed that there is a single DM who fully meets the
requirements of having the necessary experience, the required level of
responsibility and thorough knowledge of the system. This DM is responsible for
seeing to it that public safety (regulatory body) standards are met, and as DM
assumes appropriate responsibilities to society.
Additionally, is worth noting that the information from the risk management
should serve as input that should be passed on to other managers with a view to
guiding them on how to perform their functions more adequately. This applies to
such managers as those in charge of maintenance, health, environment and safety,
or even the production manager. The DM can also be the planning or project
manager, where there is an already established system or new systems are being
implemented. Thus, the proposed model can be applied to systems that are not yet
in operation or those that will be developed. It will determine which alternatives
will require most attention in the project design or project execution stage. This
should then lead to preventive and mitigation measures being drawn up and taken
so as to minimize risks at the project level.
Apart from the DM, another person who has an important role to play in the
decision-making process is the expert. Experts provide technical and theoretical
support to assist the DM with any questions or issues that may influence the
decision-making process. Since this model is intended as a tool that assists risk
management. Some experts with relevant knowledge who perform important
functions in the organization can be included.
On some specific occasions, the DM plays the role of DM and an expert at the
same time, due to his/her having technical knowledge regarding some related to
such matters as likelihood, repair times, failure rates, and the characteristics of the
system. The DM’s preference structure is also incorporated into the problem since
it reflects the preference structure of the company, represented by managers’
decisions. However, this is not necessarily a requirement of the model. The model
allows preference aggregation, when the DM is aided by several specialists. This
occurs when the DM does not have the necessary knowledge about specific
information.
4.2.3 Identifying Hazard Scenarios
This step consists of defining all the possible scenarios which have resulted from
system/subsystem failure modes. These scenarios describe the set of states of nature
 = {11, 12, …, 21, 22, …, jk } related to the failure mode j and the resulting
hazard scenario k.
Hazard scenarios do not define the causes of the failure mode or accidents, but
rather the phenomena or accidents associated with the failure mode, which are
influenced by the type of failure mode and by the existence of other interacting
factors (e.g. there is immediate or delayed ignition and a confined space).
In this context, Crowl and Jo (2007) state that accidents originate from
incidents. An incident can be defined as a loss of control over a material or form
of energy. Many incidents are followed by a series of events which propagate
accidents. This can include fire, explosions and toxic gas leaks. According to the
authors, a single section of equipment may have dozens of scenarios, each of

which must be identified.
A widely used technique to determine possible accident scenarios is Event Tree
Analysis. This technique enables the sequencing of initial events to be analyzed as
well as their interactions with the factors that affect the evolution of the event to
its final result. This analysis is conducted based on a failure mode.
Once every possible hazard scenario 4 = {T11, T12, …, T21, T22, …, Tjk} is
known, the DM must indicate which scenarios the model will consider.
4.2.4 Defining and Selecting Alternatives
At this stage of the model, the alternatives are defined for the DM. The
multicriteria decision model produces a risk hierarchy related to the company’s
systems or subsystems, and it is these which are the alternatives.
In an alternative, the features must be homogeneous, and take into
consideration both technical and social issues as well as aspects that influence the
probability of a hazard scenario occurring. Expert opinion is important, because it
is the expert who has prior knowledge about the behavior of the system. For
example, for technical issues related to a natural gas pipeline system, extremely
important characteristics include the diameter of the pipe, gas pressure, age of the
pipe, characteristics of the soil, composition of the pipe material, the corrosion
protection used, etc. These factors along the sections (alternatives) impact on the
variation in failure rates and the consequences of accidental releases of natural gas
from the pipeline (Jo and Ahn 2002; Jo and Ahn 2005; Sklavounos and Rigas
2006; Jo and Crowl 2008; Brito and de Almeida 2009; Garcez et al. 2010; Alencar
et al. 2010; Brito et al. 2010)
Regarding environmental dimension, characteristics that could be considered
include the type of the surrounding vegetation, the presence of wildlife exposed to
risk, the importance degree of the environment, environmental impact, etc. As to
the human dimension, characteristics that should be considered include land use,
population density and community type.
Returning to the context of natural gas pipelines, Henselwood and Phillips
(2006) assert that these factors may influence the likelihood of an accidental
ignition of a natural gas leak. As an example, in an industrial region, the ignition,
due to the presence of large numbers of ignition sources, of leaking gas is more
likely than in a rural area, where the population density and the presence of
ignition sources are low. More details about these aspects are given in Brito and
de Almeida (2009), Alencar and de Almeida (2010) and Lins and de Almeida
(2012).
Finally, it is important to emphasize that the uniformity of the characteristics
listed above in each system/subsystem comprises a distinct discrete set (A = {a1,
a2, …, an}), where the final system is the sum of all the subsystems analyzed.
4.2.5 Estimating the Probability of Accident Scenarios
Risk analysis enables system failures to be anticipated, thereby helping to identify

potential causes and possible consequences. They can be anticipated by analyzing
accidents that have previously occurred in similar facilities and which have been
recorded in the specialized literature or databases. This analysis allows a statistical
evaluation to be made of the most common causes and local conditions which
favored the occurrence of claims (Garcez et al. 2010).
In this step a priori probabilities (Sai(Tjk)) of accidental scenarios defined in the
previous step are estimated for each alternative i established. According to Raiffa
(1968), the Bayesian approach has become important in situations where there are
few or even no data. In these situations, it does not make sense discard a priori
knowledge that a specialist has about a variable (or variables) in question. A priori
knowledge is a result of variables interacting with the structure, conditioning
factors and intervening aspects of the problem and its details, and it is these which
make it possible to explain this knowledge using a probability distribution. These
probabilities can be obtained from different procedures. One of the best-known is
that of eliciting an expert’s prior knowledge (Bayesian hypothesis).
4.2.6 Analysis of Objects Exposed to Impacts
At this stage objects that are exposed to impacts due to an accident scenario
having occurred Tjk will be analyzed in a particular alternative i, and in the
different consequence dimensions (C = {c1, c2,…, cr,…, cm}) considered. As
mentioned earlier, these consequence dimensions may consider impacts on human
health, environmental impacts, financial loss, company image losses, operating
loss, etc.
For each hazard scenario and alternative, mathematical models are used and
numerical applications made on several features of the objects in the surroundings
exposed to hazard. Through this mathematical study, possible impacts are
estimated on the different consequence dimensions considered.
However, in the first place, it is necessary to determine what the area or danger
zone (Si) is that results from each scenario and each specific alternative. Having
done so, estimates can be made of the impacts and consequences in the
dimensions considered in a particular alternative. The danger zone, according to
DziubiĔski et al. (2006), is a region where impacts exceed critical limits, causing
injury to persons, property and environment losses.
4.2.7 Estimating the Set of Payoffs
During this stage possible impacts (consequences) or payoffs that arise from
accident scenarios (Tjk), are verified, in a danger zone (Si) which has been defined
in the previous step.
The model consists of a set of multidimensional consequences involving risks.
For each consequence dimension considered, the maximum impacts (losses)
resulting from an accident should be defined.
4.2.8 Eliciting the MAU Function
According to Brito and de Almeida (2009), the traditional representation of risk

considers probabilities or the multiplication of probabilities and consequences that
do not reflect people’s aversion to harmful events with low-probability and high
(often catastrophic) consequences. An approach that considers the DM’
preferences is required. The consequence utility function is a way to incorporate a
DM’s preference in the context of risk where consideration is given to losses due
to accidents.
MAUT can be used to aggregate preference values and consequences with
respect to multiple dimensions taking into account the DM’s preferences and
behavior, considering cases with uncertainty (Brito and de Almeida 2009; Alencar
and de Almeida 2010).
In MAUT, compensation between criteria implies the use of a synthetic
function that aims to aggregate all criteria in a single analytic function. Thus, the
structure of the DM’s preferences should be based on a compensatory notion.
Moreover, MAUT incorporates utility theory axioms. The basic idea of utility
theory is to quantify the DM’s desire, by assigning values to assets such that these
values represent a rule of choice for the DM.
Keeney and Raiffa (1976) break the MAU function elicitation procedure down
into five stages that should be used when modeling a problem:
x Introduction to terminology and ideas;
x Identifying the independence assumptions;
x Evaluating the conditional utility functions;
x Evaluating the scale constant;
x Checking and validating consistency.
The first step consists of ensuring that the DM understands the purpose of the
utility function and the consequence space. Therefore, one of the most important
insights the DM can have is the issue that there is no great preference to be
defined, but rather a set of consequences in which the DM demonstrates his/her
preferences. As preferences are the DM’s subjective representations, there is not a
correct choice.
Before engaging with the utility elicitation procedures, it is essential to
familiarize the DM with concepts such as: decision analysis, utility functions, and
lotteries. Details of these concepts can be found in Keeney and Raiffa (1976), Roy
(1996) and Vincke (1992).
Another relevant aspect, according to Keeney and Raiffa (1976), concerns the
Von Neumann-Morgenstern expected utility that can be used to characterize an
individual risk attitude through simple lotteries.
The concept of a simple lottery can be seen in the following example where the
DM maker has, certainly, an amount of money to gamble (e.g. $t.00) and needs to
set the probability value p that makes him indifferent towards two situations:
keeping the money or making the lottery bet. In other words, the DM remains
indifferent between having $t.00 with certainty and risks in a lottery with two
possible outcomes: receiving an X amount with probability p or losing the game
with probability 1 – p. Graphically, this may be represented by Fig. 4.2.
$x
p
$t
1-p $y
$t a $x, p, $y! x ! t ! y, p >@
Fig. 4.2 Graphical representation of a payoff lottery
When the DM understands the concepts, the structure of the decision problem
and the consequence space are established. To reach a better understanding of this,
an example with three consequence dimensions (c1, c2, c3) will be presented (Brito
and de Almeida 2009; Alencar et al 2010; Garcez et al. 2010; Brito et al 2010),
where c1 represents losses in the human dimension (e.g.: the number of people
exposed to fatality), c2 represents losses in the environmental dimension (e.g.:
a vegetation area exposed to fire) and c3 represents the losses in the financial
dimension (e.g.: the maximum monetary amount disbursed). A graphical
representation of this is given in Fig. 4.3.
Fig. 4.3 Graphical representation of the consequences space of the MAU function
Eliciting the utility function occurs over a closed interval of consequences,

where the maximum value is limited to a null result (no impact). In other words,
 
the most desirable utility is u c11 , c12 , c13  1 . The minimum utility value is linked
to the scenario of the worst consequences estimated by the alternatives. Thus,
 
u c10 , c 20 , c 30  0 is the least desirable consequence, since we are dealing with
losses.
It is worth mentioning that, although it is possible verify discrete and
quantifiable consequence values (e.g. the number of people injured), the
consequence sets in each dimension can be considered continuous for the purposes
of evaluating the utility function.
Therefore, the following values of the consequences space are observed:
0 1
 c1  c1  c1 (e.g.: 100 dead people  x  0 dead people);
0 1 2 2
 c 2  c 2  c 2 (e.g.: 156m burnt vegetation  y  0m burnt vegetation);
 c 30  c 3  c13 (e.g.: loss of $3,000,000  z  $0.00 ).
To confirm the DM’s understanding with respect to the limits of the

consequence space and his/her preferences, he/she is asked regarding to define
his/her preferences with respect to the points Sc1 and Tc1, Sc2 and Tc2 and finally Sc3
and Tc3 defined in Fig. 4.3. What consequence points does the DM prefer:
 Sc1 or Tc1?
 Sc2 or Tc2?
 Sc3 or Tc3?
If there is any inconsistency in the DM’s answers (the DM must state his/her
highest preference for one of these points: Tc1, Tc2 or Tc3), the DM must be given a
new explanation that will lead him/her to a correct understanding of the limits of
the consequence space and the conceptual basis of utility theory.
According to Keeney and Raiffa (1976), some independence utility
assumptions should be verified after defining the limit values of the utility
functions and checking that the DM understands them correctly.
According to Alencar and de Almeida (2010), an attribute c1 is additively
independent of an attribute c2 if two lotteries are equally preferable for all (c1 and
c2) and for a ‘c1 and c2’ arbitrarily chosen, as presented in Fig. 4.4.
cc cc’
p p
and
p c’c’ p c’c
Fig. 4.4 Lotteries to check the additive independence
According Figueira et al. (2005) when attributes (from the perspective of the
Von Neumann-Morgenstern utility model) and the DM’s preferences are
consistent with the conditions of utility independence, then u(c1, c2, …, cr, , …, cm)
can be decomposed into additive, multiplicative or another well-defined structure
in order to simplify the evaluation of these relations.
The MAUT can be expressed in an additive form, if and only if, cr attributes
are mutually independent in utility and the additive independence between the
attributes is observed. Then:
m
u ¦ k uc
r 1
r r (4.1)
where ur represents the one-dimensional utility functions [0,1]; and, kr represents

the scale constants estimated by the elicitation process based on the comparison of
lottery payoffs. The sum of the scale constants must be equal to one
§ m k 1·.
¦
¨ r 1 r
©
¸
¹
On continuing with the utility function elicitation process, it is necessary to
estimate the functions that depict one-dimensional utility functions on the m
consequence sets analyzed by the model. The procedures for eliciting the one-
dimensional utility function are also described in Keeney and Raiffa (1976).
According to Keeney and Raiffa (1976), to evaluate the scale constants, a
structured set of questions should be applied in which the DM makes probabilistic
choices of lotteries involving payoffs in the dimensions analyzed.
Returning to the three-dimensional example, the DM is asked to find the p

value where the DM is indifferent between the certainty of having consequence

c11, cw0 (in this case p = 1) or playing the lottery c30 , cw0 , p, c13 , c1w where the
value of c w0
corresponds to the consequence c10 ,c20 and the value c1w is
equivalent to the consequence c11,c12 .
Once the p value is defined, the DM is asked about the value of q in which
he/she is indifferent between the certainty of having consequence c10 ,c20 or
playing the lottery , q, . Having obtained the p and q estimated
c10 , c20 c11 , c12
m
values and the condition ¦ k 1 , the following may be defined: k = p, k = 1 2
r 1 r
(1 – p)q and k3 = (1 – p)(1 – q).
The last step consists of verifying the consistency and the variability of the
results if some parameters are modified. Due to the associated uncertainty related
to the parameters of the model, this phase can capture the impact of the results by
using sensitivity analysis on the model.
4.2.9 Computing the Probability Functions of Consequences
Several uncertainties are present in scenarios (Tjk) and estimating hazard zones
(Sn), as shown in the earlier stages of the model. These uncertainties are
undesirable, because it becomes impossible to define deterministically which
multidimensional consequences can occur due to an accident scenario. For this, it
is necessary to estimate the probability distributions of the consequences,
represented by a consequence function P, defined by the probability of obtaining a
consequence p, since a scenario Tjk occurred in alternative ai.
In this step of the model, there is a need to estimate the joint probability
distribution over the possible values in “m” consequence dimensions P(c1,…,m |Tjk ,ai)
for each alternative and hazard scenario adopted.
According to Brito and De Almeida (2009), in some contexts it may be
considered that different consequence dimensions can have small or even
negligible correlations between them. This is because the hazard radius covers
several dozen meters. The combination of these consequence dimensions occur
randomly and independently, depending on the specific characteristics of each
alternative, so that the probabilities P(c1|Tjk ,ai),…,P(cr|Tjk ,ai),…,P(cm|Tjk ,ai) can
be estimated independently.
However, in some risk analysis contexts, the probability distributions of these
consequences are not treated independently, as is the case of risk analysis
regarding petroleum extraction platforms, nuclear power plants, etc. where the
danger zones usually extend over a wide area and the size of impact interferes
non-randomly in various consequence dimensions.
In the case of probability distributions independent of consequences it is
possible to define mathematical formulations to model the consequence functions
for each loss independently.
There are several models in the context of natural gas pipelines considering
consequence functions for estimating the human, environmental and financial risk
dimensions (Brito and de Almeida 2009; Garcez et al. 2010; Alencar et al. 2009;
Brito et al. 2010). Similarly, for the context of hydrogen gas pipelines with
required adaptations, same approach is considered for estimating risk dimensions
(Alencar and de Almeida 2010; Lins and de Almeida 2012). A model for risk
evaluation in underground vaults of an electricity distribution system considers the
same decision analysis principles for assessing risk dimensions of human impacts,
financial losses, operating losses and disturbance on the local transit vehicles
(Garcez and de Almeida 2014b).
4.2.10 Estimating Multidimensional Risk Measures
In the context of decision making, the DM must choose an action in order to

ensure that the consequences are those that are the most favorable ones possible
for him.
Decision Theory is a mathematical formalization of this paradigm. It allows
rational decisions under uncertainty. According to Berger (1985), Decision Theory
involves the following aspects:
x Analyzing past and current information of the system under study, based on the
objective and / or subjective information available;
x Eliciting probability distributions to model uncertainties;
x Developing a mathematical model that describes the system and its revision
level, which considers the level of accuracy required;
x Eliciting the DM’s preferences and values;
x Identifying or designing alternative actions that lead to the desired goals;
x Using mathematical logic to combine alternative actions, utilities and
probabilities with the mathematical model of the system in order to identify the
best action course for the DM;
x Implementing the action(s) chosen in the previous step;
x Returning to the first step and restarting the process to correct errors and
distortions regarding the data, probabilities, utilities and action alternatives.
According to Berger (1985), by Decision Theory, the loss function can be
defined as the negative of the utility function of the expected consequence,
expressed by:
cr
u c r | T jk , ai (4.2)
It can be considered that the consequences are results of the impact dimension
of a given action, which can be estimated by using a probability distribution
function P(c1,…,m |Tjk ,ai).
Keeney and Raiffa (1976) point out that if an appropriate utility is assigned
to each possible consequence and the expected utility of each alternative is
calculated, what is observed as the best course of action is an alternative with the
highest expected utility. Thus, the consequence utility is the expected value of the
utility:
cr E >u c r @ ³ Pc uc dc
r r r (4.3)
cr
Therefore, the utility function u(cr) can be calculated by:

u T jk , ai u Pcr | T jk , ai ³ Pcr | T jk , ai ucr dcr (4.4)
cr
After having obtained knowledge about an a priori probability distribution of

the states of nature Sai(Tjk), which depends on the characteristics/conditions of
each system (alternative) analyzed, it is possible to calculate the risk associated
with each alternative, using a risk perspective such as a consequence/damage/
severity added with the uncertainty, as can be seen in the following equation:
m § § ··
¨ ¨ ¸¸
r ai ¦¦ ¨ S ai T ¨ u c r P c r | T jk , ai dc r ¸ ¸ 1S ai T N
³ (4.5)
rT 1¨ ¨ cr
¸¸
© © ¹¹
where r represents the various dimensions (attributes) of the analysis. In other

words, these are the consequence dimensions (c1, c2,…, cr,…, cm), after having
considered the occurrence of all hazard scenarios 4= {T11,…,Tjk} and alternatives
ai analyzed. The value of Sai(Tjk) depends on the characteristics/conditions of each
system analyzed.
The state of nature TN represents the normality scenario of the system, where
the system operates under normal conditions, without any dangerous scenario
occurring, thus justifying the loss function value equal to –1. The risk values could
be found in the range [–1,0], where the value –1 is related to the lowest risk and
the value 0 to the highest risk.
4.3 Risk Decision Models 181
Thus, the risk concept based on Decision Theory assesses the consequences (cr)
of the hazard scenarios (Tjk), by combining both uncertainties associated with: (i)
the consequences P(cr|Tjk ,ai); and (ii) the hazard scenarios Sai(Tjk).
Additionally, the risk measure used considers the DM’s preference structure in
the set of expected consequences, through utility functions u(cr), representing the
“desirability” that the DM has about property losses (in this particular case, the
consequences of an accident scenario occurring) and allowing a probabilistic
evaluation of the consequences under uncertainty.
These risk measures comprise a descending risk hierarchy of several of the
alternatives (ai) evaluated. Consequently, the results of this hierarchy serve as
input to the decision-making process and risk management.
4.3 Risk Decision Models
Several applications of a multidimensional risk evaluation and decision models

have been conducted, based on the previous procedure, adapted from Chap. 2.
These applications incorporate the situation in which the DM’s behavior regarding
to risk is represented by a utility function. This procedure has been applied in
several contexts: natural gas pipeline (Brito and de Almeida 2009; Brito et al.
2010), hydrogen gas pipeline (Alencar and de Almeida 2010; Lins and de Almeida
2012) and electricity distribution system (Garcez and de Almeida 2014a; Garcez
and de Almeida 2014b).
In this section, three applications of a multidimensional risk evaluation model
are presented. The first application is made in the context of risk analysis in
natural gas pipelines and is based on Brito and de Almeida (2009). The second
application concerns the context of an underground electricity distribution system.
This application is based on Garcez and de Almeida (2014b). The third application
considers a different MCDM/A method, taking into account a non-compensatory
rationality, according to the procedure presented in Chap. 2 (Brito et al. 2010).
4.3.1 Risk Evaluation in Natural Gas Pipelines Based on MAUT
Natural gas is a fossil fuel with reserves available in many parts of the world. Its
use has grown over the last 30 or so years due to a number of factors, including,
for example, economic and environmental aspects. The high demand for it in
widely scattered different locations requires a mode of transportation to convey
large amounts of gas from its source to its destination, quickly and safely. Thus,
among the existing modes of transportation, pipelines stand out. Although using
pipelines is considered a safe system, some accidents have occurred over the
years, some of which have had critical consequences.
In this context, this subsection will present a numerical application of a

Multidimensional risk evaluation, taking into account the characteristics of the
model that have been presented earlier in this chapter, well as some additional
points, specific to the context of natural gas pipelines.
Thus, multidimensional risk analysis in natural gas pipelines is conducted using
hazard scenarios, in order to estimate the probability of the occurrence of a hazard
scenario and the possible consequences that might result from pipeline failure.
Additionally, the model presents a ranking of pipeline sections in a multi-
dimensional risk hierarchy, in which three dimensions of risk are considered,
namely the human, financial and environmental dimensions. These dimensions are
the main ones to be considered that arise from the operation of the pipeline
sections under analysis. A ranking of these segments under a risk hierarchy is
presented so as give insights into the process of managing pipeline risk, thereby
contributing to defining mitigating actions according to the risks associated with
each section analyzed. A single DM was considered.
The total length of the pipeline analyzed in this application is 18,000m divided
into 9 sections that comprise a discrete set X = {x1, …, x9}, where each element
presents specific features.
Probabilities of each scenario are obtained as per procedures presented by Brito
and de Almeida (2009). These authors use a conservative risk assessment for each
scenario and pipeline extension, and include the most critical danger zone for each
segment associated with the worst accident scenario that may occur in that specific
extension.
A conservative estimate of the radius of maximum danger CDR is given in
(4.6), considering the operating pressure Po, the diameter d of the pipe and length
of the pipeline L from the compressor station. More details can be found in Jo and
Ahn (2002).
Po1 2 d 5 4
CDR # 1,512 1 4
(4.6)
L
Settled danger areas for each section, and the human, environmental and
financial consequences should be defined. This set of consequences will be
included in the analysis using the model, for which the most pessimistic values in
each consequence dimension will be input.
The proposed model seeks to assess risks considering three risk dimensions in
natural gas pipelines: Human Risks (rh), Financial Risks (rf) and Environmental
Risks (rm). The reasons why it is primarily these dimensions that are considered
are based on values that are normally found in both productive organizations and
in other organizations or institutions involved. These will be translated into
principles of social and environmental responsibility and ethical aspects of human
relationships. These aspects should influence company actions that seek to secure
the financial return aimed at.
As to the human dimension, Brito and de Almeida (2009) assume that human
consequences are estimated by the number of people affected physically due to a
particular accident scenario, and who receive at least second degree burns, and not
necessarily by the number on fatalities.
With regard to the environmental dimension, the area of vegetation affected is
used as a measure for the environmental consequences, taking into account the
extent of environmental impacts caused by this type of accident (Alencar et al.
2010; Garcez et al. 2010; Brito et al. 2010; Alencar et al. 2014).
Finally there is the financial dimension for which disbursements on foregone
income, contractual fines for supply disruptions, fines and other indemnifications
for harm caused to people, environment or organizations and companies are
considered. Additionally there are expenses related to maintenance and operational
actions taken with a view to re-establishing the operational conditions of the
pipeline.
The next step corresponds to eliciting a MAU function U(h, f, m), which it is
considered an additive function. The property of additive independence implies
that there is preferential Independence among the payoff sets. U(h, f, m) can be
expressed by the following (4.7).
U h, f , m   k h Ph |  , xi U h dh  k f

  P f |  , x U  f df
i
h f
(4.7)
 k m P m |  , xi U m dm

m
The calculation of the average radiation flux (due to a hazardous scenario of

deflagration) is obtained from (4.8) (Jo and Crowl 2008).
I
  a  Qeff  H c  (4.8)
4 CDR 2
where I is the average radiation flux, a is the atmospheric transmissivity,  is the

ratio of the irradiated heat over the total heat released, Hc is the combustion heat
of the natural gas, CDR is the critical danger radius and Qeff is the effective rate of
gas leak.
The estimate of risk is based on Decision Theory principles. According to
Berger (1985), risk is considered as the expected value of the loss and can be
defined by (4.9) verified in Alencar and de Almeida (2010).
r  xi     L
i jk jk , xi  (4.9)
Knowing that:

L T jk , xi
u P p | T jk , xi (4.10)
In this way, losses associated with each scenario and section are summed in the
three dimensions discussed, multiplied by accident scenario probabilities and
added to the losses associated with a normal scenario (TN), as shown in (4.11).
r xi >
ET L T jk , xi @ ¦¦ LT jk ,x S i T jk 1S i T N
i
(4.11)
j k
Due to the additive independence properties of the MAU function and the
independence in probability of the probability distributions over the consequences,
the risk r(xi) is given by (4.12).
ª º
«k h Ph | T , xi u h dh »
« ³ »
« h »
r xi ¦¦ « f ³ i » i jk i N
« k P f | T , x u f df »S T 1S T (4.12)
j k « f »
« »
« k m Pm | T , xi u m dm»
³
«¬ m »¼
Using the risk values obtained from (4.12), pipeline sections can be ordered in
descending order, thereby obtaining a ranking of pipeline sections that should be
used as input for risk management activities.
The MAUT interval scale allows an incremental comparison between the risk
sections in line with the utility value between the alternatives. Thus, (4.13) and
(4.14) are applied to analyze the relationship between alternatives, showing
respectively the absolute difference between alternatives and the difference ratio
between alternatives. The difference ratio DR is used to interpret the values in
relation to the calculated risks.
DA rb xi rb 1 xi (4.13)
rb xi rb1 xi
DR (4.14)
rb1 xi rb 2 xi
where the index (b) represents the position in the ranking of the section and rb(xi)
represents the risk value related to a specific section. Through the analysis
obtained from the results of these equations, the DM can define which sections
should be included given the resources available, thus representing how much
more a section adds to the risk when compared to another section placed further
down in the ranking provided by the risk model.
Therefore, taking into account all the calculation steps described earlier in this
section, Table 4.1 presents the sections prioritized based on comparisons of the
increments of risk. The values listed in rb(xi) – rb+1(xk) column must be multiplied
by (10-5).
Based on Table 4.1, some interpretations may be made. A descending ranking
of values is applied for risk assessment, where S1 shows the highest value of risk
among the sections evaluated. The highest losses associated with the likely
consequences of accidents are expected for S1. Additionally, it is observed that the
increment in the risk values from S4 to S1 is 1.3098 times greater than that from S7
to S4. In the same way, the increment in the risk values from S9 to S6 is almost 14
times greater than that from S8 to S9.
Table 4.1 Ranking Positions, DA and DR of the analysis
Ranking Position Section DA DR

(E) (xi)
1 S1 0.7277 1.3098
2 S4 0.5556 0.0450
3 S7 12.3355 0.5135
4 S6 24.0237 13.5551
5 S9 1.7723 1.8107
6 S8 0.9788 1.9436
7 S2 0.5036 1.4291
8 S3 0.3524 -
9 S5 - -
According to Brito and de Almeida (2009), given financial, technical and

manpower constraints, the ranking obtained helps to prioritize the most critical
pipeline sections in order to allocate a greater amount of resources for mitigating
actions to those sections deemed most critical in the DM’s view, bearing in mind
that his/her preferences were incorporated throughout the development of the
model, based on different risk dimensions. The DR analysis enables the DM to
analyze the sections considered more consistently, making it possible for him/her
to establish better planning actions, as well to allocate resources better.
In conclusion, all these improvements observed by using a MAUT application
in this multidimensional risk model provided consistent results that can support
managers in planning activities. Additionally, the ranking of risk values enables
managers to analyze the existing context better, leading the organization to
consider these aspects of mitigating risks and to consider preventive actions linked
to the risk mitigation process.
4.3.2 Multidimensional Risk Evaluation in Underground

Electricity Distribution System
Typically, energy distribution systems are big and complex. These systems are
considered as being among the main elements of the critical infrastructure. Several
other external systems such as systems of water supply, telecommunications,
traffic, public transport, health, food supply, gas distribution, and others are
dependent on this system. Therefore, several impacts other systems can be caused
by small faults in the power system, and to generate a chain of consequences,
which is why it is a critical part of the infrastructure for society.
Greater initial investment is required by installation of the infrastructure of an
underground system. In generally, it is more complex than overhead systems.
There are some disadvantages by use of the underground systems, such as it incurs
higher costs associated with maintenance; it is also difficult to access underground
networks; to upgrade the system (physical and limited space configuration); and,
to operate and maintain auxiliary ventilation systems, etc.
Though, this system have advantages: the operation of underground systems is
more safer and reliable than overhead systems for the population; more immune
to interference from nature (storms, winds, storms, falling trees, etc.); better
accessibility of disabled people, low visual pollution in the city and presenting less
impact on the occurrence of traffic accidents.
Regardless of being safer than overhead systems, many underground vaults
events have occurred. Hundreds of accidents in vaults occur every year in Ney
York, such as smoke, explosions, fires, etc. (Radeva et al. 2009; Rudin et al. 2010;
Rudin et al. 2011; Rudin et al. 2012).
The low frequency of the occurrence of accident scenarios, its magnitude of
their consequences and the complex environment surrounding the hazard zone
make the risk management becoming even more complex and uncertain (Garcez
and de Almeida 2014a; Garcez and de Almeida 2014b). Also, the large number of
subsystems, with each having particular characteristics, and there is a lack of (or
incomplete) historical data of accidents and its failure modes and past events make
the decision process even more complex.
Hazard scenarios can produce various consequences, for instance, fatalities and
injuries to people, blackout, disruptions to local vehicular traffic, explosions and
fires in nearby locations, impact of the company image, the population being
afraid (on account of the uncertainty of when and where an accident will occurs),
affect the system reliability and safety and other consequences which cannot be in
financial terms (Garcez and de Almeida 2014a; Garcez and de Almeida 2014b).
Hence, these consequences can disturb directly or indirectly the sector of the
society, the public sector and business.
According to Garcez and de Almeida (2014b), assessing the risks comprehensively
and realistically is extremely important. It may generate knowledge that can be
applied to assist a DM to choose and implement preventive and mitigating measures.
Furthermore, the several resources available by company, such as: money,

time, work teams, technology, safety equipment, etc. are limited and scarce. For
optimization the use this resources, it is necessary to use decision-making tools
that assess the consequences and uncertainties. Moreover, it is necessary to
evaluate risks together with the DM’s preference structure, thereby solving the
problem more adequately (Garcez and de Almeida 2014a; Garcez and de Almeida
2014b; Garcez and de Almeida 2014c).
Therefore, it is necessary a decision making tool to aid the DM, generating a
hierarchy of the multidimensional risks from the several underground vaults. The
aim is to prioritize available resources to implement actions (preventive and
mitigate actions) that increase system safety.
As seen, the MCDM/A, MAUT, permits the use of multiple value judgments;
thereby incorporating the uncertainty and subjectivity inherent in the problem of
estimating and evaluating different dimensions of the risks involved; and
aggregating the DM’s preferences.
According Berger (1985), a good decision should be a logical consequence of
what one wants, what one knows and what one can do, so that the DM can choose
an action (or actions) in order to bring about the most favorable consequences/
results for the DM. In this context, the Decision Theory is a mathematical
formalization of this paradigm. It allows for rational decision-making under
uncertainty, where the loss function is established as the negative of the utility
function of the expected consequence.
The consequences are the result of the impact of the accident, which can be
estimated using a probability distribution function P(c|T,Vq), where T are the states
of nature (hazard scenarios); c is the consequences; and, Vq is the underground
vault analyzed.
By MAUT concepts, Decision Theory and probabilistic independence, the risk
measure can be expressed by (4.15).
§ § § ··
¨ § · · ¸
r §¨Vq ·¸ ¦ ¨ ¦ ¨¨ S T ¨ ³ u c P¨ c T , Vq ¸dc ¸ ¸¸ ¸ 1S T N (4.15)
© ¹ ¨ © ¹ ¸¹ ¸
i ¨© T © © c ¹¹
where i represents different dimensions of consequences and the state of nature TN

is the normal setting of the system (there are no consequences – justifying that the
value of the loss function is -1, the operation of company is normal without any
accident occurrence). S(T) is the probability of the hazard scenario. These risk
values r(Vq) are in the range [-1,0], where the value -1 is related to the lowest risk
and the value 0 to the greatest risk.
This section presents a numerical application based on the study realized by
(Garcez and de Almeida 2014b). The hazard scenario, internal explosion caused
by an arc flash, was considered. It is regarded as having the greatest impact and
causes the manhole cover to be blown off and projected. The study evaluated the
consequences (c) from four dimensions: operational impacts (cO), financial

impacts (cF), disruptions to vehicular traffic (cT) and human impacts (cH).
The cO corresponds to the impact on the supply operation of the electricity
distribution company (downtime). The cT is evaluated by the process of how
traffic jams form on the streets around the accident area. The cH deals with injuries
caused by the projection of manhole covers and burns of at least the second degree
due to exposure to incident energy from an arc flash. Lastly, the cF is about any
kind of monetary compensation related to an accident occurring.
Equiprobable Intervals method (Keeney and Raiffa 1976), based on results of
Walsh and Black (2005), were used to estimate the distance projection of the
manhole cover. Other hazard zone, calculated by IEEE Standard 1584 (IEEE1584
2002), also known as the Flash Protection Boundary, can be calculated as the
minimum distance from the arc flash at which people could be safely exposed to
incident energy without suffering second-degree burns. Estimates of the risk
measures are made from the perspective of DM by Eq. (4.15).
As it is supposed that the DM’s preference structure is additive independent
between the criteria, the utility functions from the perspective of a one-
dimensional utility (U(cO),U(cT),U(cH),U(cF)), can be elicited separately. To do so,
the procedures described in Keeney and Raiffa (1976) were followed. It was
considered that the DM is risk averse in the human dimension and risk prone in
the remaining dimensions. The values of the scale constants obtained were: kcO =
0.12; kcT = 0.16; kcH = 0.29; and kcF = 0.43.
Hence, the multidimensional risk measure is calculated (4.15). The ranking of
the multidimensional risk assessment is shown in Table 4.2.
The risk difference is calculated by (4.16).
 
ri Vq  ri 1 Vq   (4.16)
The risk ratio is calculated by (4.17).
ri Vq   ri1 Vq  r1 Vq   ri Vq 

st
n
(4.17)
As the conclusion, Vq3 is ranked as first underground vault and Vq2 as second.
Furthermore, it is observed that the difference between these risk values
corresponds to approximately 44% of the total range of risk. Therefore, it is
evident that is necessary to allocate more resources as a priority to preventive and
mitigating actions on the first vault.
After the risk of the first alternatives (Vq3 and Vq2) has been attend, there is
another gap between the alternative ranked second Vq2 and the one in third place
Vq5 (14% of the total range of the risk). Again, one prioritizes additional actions to
prevent and mitigate the risk addressed in the first two vaults.
Another relevant information, it is that there is a homogeneous group of

alternatives with similar risk values (Vq5, Vq1, Vq6). This information is important
to the DM, because the DM can direct different and additional resources to
preventive and mitigation actions to these alternatives, since they have very
similar risk values.
Table 4.2 Results of ranking the risk
Rank Vq Risk Difference Risk Ratio
1st Vq3 1.66E-03 44%

2nd Vq2 5.64E-04 15%
3rd Vq5 5.64E-05 1.5%
4th Vq1 2.12E-04 5.6%
5th Vq6 5.74E-05 1.5%
6th Vq4 1.24E-03 32.8%
7th Vq7 - -
Other issues (criteria) can be considered by the DM to choose which under-

ground vault. DM will tackle first within this homogeneous group of risk. Another
aspects can be considered by DMs when a decision making is taken: which actions
and what alternatives will generate benefits more earlier? Additionally, in what
alternative could be more efficiently? Finally, in another view that could be taken
into account is decision-making for policy issues.
Under an inter-criteria approach, as shown in Fig. 4.5, on analyzing the risk
values, it is concluded that: the first alternative shows that the traffic impact is the
major one, while in the last-placed alternative the human impact is nonexistent.
Furthermore, all alternatives have a financial impact and the only major value of
the impact of these last-placed alternatives is on the financial dimension.
The comparison among the increments in risk, in inter-criteria analysis, is a
different strategic information (Garcez and de Almeida 2012). This analysis
allows identify the criterion that contributes to the greatest difference in risk
between alternatives. By analysis, as shown in Fig. 4.6, the comparison pair-to-
pair of the alternatives Vq2 and Vq5 can conclude that there are major impacts
between the consequences of the financial, operational and human loss dimension.
Therefore, the DM can conclude that preventive and mitigating actions that
direction on disturbances to traffic loss dimension will not produce any impact in
the difference in risk between these two alternatives. However, focusing on
preventive and mitigating actions in the operational or human loss dimension of
alternative Vq2, would result in reducing the amount of global risk compared to
alternative Vq5. Thus, resources of the company can be reallocated to a manage
risk more effectively.
Fig. 4.5 Analysis of the measures of inter-criteria risk
Fig. 4.6 Analysis of the intra-criteria of the risk differences of alternatives Vq and Vq
2 5

on ELECTRE Method and Utility Function
This section presents the application (Brito et al. 2010) of a different MCDM/A
method, which is integrated with utility function, the ELECTRE TRI method.
Three main issues should be highlighted, when compared with the two previous
models. First, it is a non-compensatory approach, taking into account a specific
kind of DM’s rationality. Second, the problem consists of a sorting problematic,
since the managerial issues in this application are distinct from the two previous.
Third, it integrates the ELECTRE method with utility theory, in order to
incorporate the DM’s behavior regarding to risk (prone, neutral or averse) into
ELECTRE.
As details given subsequently, this application illustrates the step 6 in the
decision process given in Chap. 2, which involves the identification of DM’s
rationality (compensatory or non-compensatory).
In several situations, it is quite difficult (or even incoherent) for the DM to

confront directly or indirectly monetary losses on non-monetary losses such as
loss of life, injury to people, environmental damage, company image losses (Faber
and Stewart, 2003) and social impacts. Therefore, it is considered that the DM
feels more comfortable using a non-compensatory rationality approach, due this
kind of procedure does not demand the condition of full comparability as must be
done in the compensatory approach.
Specifically, in the risk management context, according to the DM, a low risk
in a given criterion (with higher weight) does not compensate directly a high
risk in another criterion, as should happen in an aggregation procedure with
compensation. Therefore, for these cases, a non-compensatory approach for inter-
criterion evaluation is more appropriate for representing the DM’s structure of
preferences.
Several gas pipeline problems, including new projects and concessions might
be related to other DMs linked to other private or public institutions. Thus, one
can be admitted that the DM wishes indirectly to consider his perception regarding
the opinion of other actors (stakeholders, including population, government
authorities and regulatory agency) in the decision process and this may change his
final structure of preferences.
Moreover, one can consider some incomparability that may arise in the process
of inter-criteria evaluation, due to a particular context (Brito et al. 2010).
As specified in step 6, in the decision process shown in Chap. 2, the decision
model assumes a DM’s non-compensatory structure of preferences for inter-
criterion evaluation (among each risk dimensions). Hence, the outranking
approach, including methods of ELECTRE’s family, is more appropriated in the
inter-criterion evaluation of risks to natural gas pipelines.
Another important point, as it was highlighted at the beginning of this section,
is related to the problematic applied. In the two previous models, the ranking
problematic was applied, based on MAUT. These models provide a comparison of
alternatives with information on how large the difference in risk evaluation is
between two alternatives. Differently, in this model under discussion, the DM
faced different challenges related to maintenance and risk management, where for
some situations a sorting problematic may be more appropriate (Brito et al. 2010).
The classification (sorting) of the natural gas pipeline sections into categories
allows the DM to organize particular management approaches for each risk
category.
The ELECTRE TRI method, more detailed in the Chap. 2, deals with a sorting
problematic, assigning each alternative si from a set S to a category or class Ck.
For the context of this model, si represents sections of natural gas pipeline to be
sorted, and the profiles b are comparison sections for the categories of risk.
The model application makes an evaluation of several sections of pipeline
according to their multiple risk dimensions, which allows a comparison of these
sections with the risk profiles in order to classify the sections into risk categories
defined by the natural gas transportation/distribution company’s management.
In this context, the profiles b that define the particular risk categories, depend
essentially on the perception that the DM has on different risk levels related to his
system, the availability of resources, the occurrence of previous accidents, society
pressures, as well as being dependent on the number of different strategies, policies,
and measures that the company possesses to deploy among the categories.
The highest risk category contains an alternative with higher probabilities of
occurrence of financial, environmental and human consequences. This category
demands relatively urgent actions that often require changes in some aspects of
the project, and that demand a major financial investment in order to obtain
significant reductions of these risks. Similarly, a lower class of risk presents
sections of pipeline with lower levels of risk, thus allowing a little longer planning
time to find effective solutions and at satisfactory costs (Brito et al. 2010).
Brito et al. (2010) highlight that the manner of determining the reference
profiles b for the each risk categories must be carried out very carefully by the
DM, since the sorting process is fundamentally guided by comparisons with these
profiles.
A procedure to aid DM to infer theses profiles is proposed by Mousseau and
Slowinski (1998). It enables the inference by means of a sample of alternatives
directly sorted by the DM.
The third point highlighted on beginning of this section is the integration
between the ELECTRE TRI method and utility theory, in order to incorporate the
DM’s behavior regarding to risk (prone, neutral, averse). The utility theory
presents an axiomatic approach that can assess the DM’s behavior with regard to
the risk (Keeney and Raiffa 1976) when there are accidents consequences.
Let D be the set of all outcomes in a given accident impact dimension.
Uncertainties are related to the states of nature T, the resulting accidental scenarios
of a pipeline accident, and to its impacts under a given dimension of outcomes.
For dealing with uncertainties on D, it is necessary to use a probabilistic approach,
represented by a probability distribution over the deterministic consequences and
by the elicitation of the utility functions for these consequences (Brito et al. 2010).
This procedure is applied in the intra-criterion assessment process (for each risk
dimension) with the aim of risk evaluation for human, environmental and financial
dimensions posed by each section of pipeline.
As defined in (4.11), the risk is assessed as the expected loss, which is
estimated for each section of pipeline. The loss is given by combining the
probability over the deterministic consequences p in D, named by P(p|T,si), and
the utility function (U(p), where p D ) over these consequences, as shown in
(4.18). It is used the traditional notation for decision analysis (Utility Theory),
where p (from payoff) denotes an element of the set of outcomes D, whereas P
(capital P) refers to a probability (which is a probabilistic payoff).
LT , s i ³ P p | T , s U p dp
i (4.18)
p
Therefore, the expected risk can be calculated for each section of pipeline
under each criterion, applying (4.18) in (4.11), then (4.19) is obtained.
r si ¦
T
S T ³ P p | T , s u p dp
i
p
i (4.19)
As previously discussed, the ELECTRE TRI method is more appropriate than

MAUT for undertaking the inter-criterion pipeline risk evaluation. Another issue
related to the DM’s structure of preferences is the observation that not all
hypotheses required by MAUT are always accepted in the case of inter-criterion
evaluation (among risk dimensions). This may happen even when these hypo-
theses are appropriate in the intra-criterion evaluation. To be precise, the DM
accepts the Utility Theory hypotheses when he evaluates separately each risk
dimension.
The use the utility functions is justified because the model can incorporate the
DM’s behavior regarding risk (averse, prone or neutral). The utility function is
also appropriate because the results occur in an interval scale rather that an ordinal
scale for comparison with the profiles categories in the sorting problematic.
Furthermore, this interval scale is explored in the process of eliciting preferential
parameters for ELECTRE TRI method, including the profile for each category
defined and the thresholds. In other words, the DM knows the amount of risk
differences to be considered in the ELECTRE TRI method for building the credibility
index. In this manner, the integration of the utility theory and the ELECTRE is
seen as a useful (Brito et al. 2010; de Almeida 2005; de Almeida 2007).
The decision model proposed by Brito et al. (2010) presents the procedure steps
for problem resolution and to construct multicriteria models, as shown in Chap. 2.
This application aims to build an MCDM/A model for the multicriteria risk
assessment of pipeline sections and for their assignment into risk categories.
Initially, the pipeline system was segmented into 12 different sections. These
sections were divided according to several technical factors such as age of the
pipeline section, pressure, land occupation, soil characteristics, degree of third-
party interference and demographic concentration on the surface area surrounding
each section.
In addition, it was considered 10 hazard scenarios T: Detonation/Deflagration;
Fireball/Jet Fire; Confined Vapor Cloud Explosion (CVCE); Flash Fire; Gas
Dispersion to both failure modes: rupture and puncture.
The accidental scenario probabilities, SiT, were based on EGIG report,
because of its ability to distinguish between pipeline failures modes, and also
because it gives more conservative estimates for the scenario of probabilities than
other databases, such as those from the United States Department of Transportation
(Brito et al. 2010).
The payoffs used in this application involve the human (H), environmental (M)
and financial (N) consequences of an accident caused by the release of gas. The
payoff of the human consequences considers injuries to human beings. Generally,

it is dealt as the number of fatalities due to thermal radiation (Jo and Ahn 2005).
The use of monetary values for estimating this type of consequence is not
appropriate to represent the consequence in a decision-making problem (Brito et
al. 2010). Therefore, this model adopts a more conservative criterion for analyzing
the human consequences (H) than monetary estimates or the number of deaths.
These consequences are estimated as the number of people exposed, at least, to
second degree burns. According to Brito et al. (2010), although very conservative,
this reasoning is appropriate when dealing with impacts on human beings,
assuming that any type of physical harm to the population should be avoided.
The environmental impacts (M) are given by the area that is exposed to the
atmospheric pollution and to the effects of scorched vegetation on animal and
vegetable species. Similarly, as in the case of human consequences, it cannot be
expressed by monetary values. Therefore, it is used the area of the vegetation
destroyed (in square meters) as measurement (Alencar et al. 2014). According to
Brito et al. (2010), although this is not a very complete way to interpret these
types of consequences, this measurement is useful and is reasonably related to the
extent of environmental impacts caused by natural gas pipeline accidents.
The financial consequences (N) are associated to operational losses that a
pipeline accident may cause, such as: expenses on labor, equipment and raw
material to substitute pipes, expected loss in revenues from supply interruptions,
refunds to customers for interrupted production, and compensation for damage
caused to others.
The one-dimensional utility functions U(h), U(m) and U(n) may be obtained
from the elicitation of some utility values in each dimension, using a lottery
procedure (Keeney and Raiffa 1976). Thus, a regression curve over plotted values
may be adjusted. Exponential functions are among functions that often present a
best fit for utility functions (Berger 1985), as per (4.20).
P p p
U p e (4.20)
where p = h, m or n. The parameter Pp is obtained by means of curve fitting. The

following parameters were obtained for the utility functions, as given in (4.21): for
U(h): Ph = 0.12 (R2 = 0.91); for U(m): Pm = 0.0017 (R2 = 0.89); and for U(n):
Pn = 3.5x10-7 (R2 = 0.94).
The calculation of consequence probabilities P(p|T,si) is obtained for each pair
(T,si) of scenario and section of pipeline. In other words, this function is the
probability of obtaining a consequence p given that T happened. Depending on the
mathematical models used, these consequence functions may assume different
forms (Arnaldos et al. 1998; Jo and Ahn 2002). For Brito et al. (2010), this
modeling can consider any type of probability distribution obtained for consequence
functions, simply by adjusting the calculations of the consequences functions to
another context or system. Thus, it is not limited to a single application.
Based on expected loss function (4.18), the combination of the probability

density functions to the one-dimensional utility functions U(h), U(m) and U(n)
was undertaken in order to estimate the one-dimensional losses.
Next, it is necessary to estimate the risk values for each pipeline section.
Whereas there is a state of nature (scenario) in which there is a probability
associated with it of no failures occur (named by TN), then this section pipeline
suffers no damage (L(T,si) = 1). Therefore, the human, environmental and
financial risk values for each section of pipeline are given by (4.21). A linear scale
transformation, r’p(si) = 100rp(si) + 100, was used to facilitate the handling of
values by the DM. These risk values are shown in the Table 4.3 (Brito et al. 2010).
§ P p p ·
r p si ¦ S T ¨ ³ P p | T , s e
i i dp ¸ (1)S i T N (4.21)
© p ¹
T
Table 4.3 Human, environmental and financial risk values
Section pipeline Human risk Environmental risk Financial risk

s1 0.0093 0.0142 0.0080
s2 0.0180 0.0199 0.0326
s3 0.0249 0.0265 0.0101
s4 0.0085 0.0270 0.0521
s5 0.0104 0.0113 0.0282
s6 0.0293 0.0181 0.0237
s7 0.0379 0.0152 0.0242
s8 0.0081 0.0128 0.0345
s9 0.0104 0.0070 0.0233
s10 0.0205 0.0245 0.0554
s11 0.0565 0.0440 0.0467
s12 0.0190 0.0201 0.0738
Subsequently, the DM wishes to sort those sections pipelines in risk categories,

ordered by decreasing levels of risk, these are: High Risk (C1), Medium Risk (C2)
and Low Risk (C3). For each defined category, the reference profiles (ELECTRE
TRI parameters) are determined, as shown in Table 4.4.
Table 4.4 ELECTRE TRI parameters employed in the analysis
Parameter rh rm rn
b1 (divides the High Risk from the Medium Risk category) 0.025 0.025 0.05
b2 (divides the Medium Risk from the Low Risk group) 0.013 0.01 0.02
weight 0.60 0.10 0.30
q (indifference threshold) 0.001 0.001 0.005
p (strict preference threshold) 0.005 0.009 0.007
By analysis of the DM, the sections in the first category demand higher states
of alert, and thus financial resources would be assigned preferentially to this
category in order to increase measures of physical protection and to intensify the
monitoring of the high risk sections. The Medium Risk category involves pipeline
sections which, although they do not lay claim to such intensive care as those in
the previous class, do demand more thorough planning for preventive measures in
order to avoid neglect in relation to maintaining their safety levels. Finally, as to
the sections assigned to the Low Risk category, the maintenance of routine
inspection actions is planned in order to keep these sections with low risk levels
within the human, environmental and financial dimensions of possible outcomes
(Brito et al. 2010).
The analyst has to explain the meaning of ELECTRE TRI parameters in order
to obtain the proper specification. It was decided not to use a veto threshold for
any risk dimension. With regard to the cutting level, k = 0.65 has been applied.
After applying the sorting model for each individual section of pipeline, the results
in Table 4.5 were obtained.
Table 4.5 Final sorting
Section pipeline Category

s1 C3
s2 C2
s3 C1
s4 C3
s5 C2
s6 C2
s7 C2
s8 C3
s9 C2
s10 C2
s11 C1
s12 C2
It was observed that, for this application under study, the results were intensely
influenced, but not completely controlled, by the human risks, given their high
weight value. Among the segments under study, 7 out of the total of 12 sections
were assigned to the Medium Risk category (C2), for which more rigorous
preventive measures should be established within 6 months. Sections s3 and s11
were assigned to the High Risk category (C1), for they present risk levels worse
than or very close to the profile b1 in a more significant proportion of impact
dimensions. Finally, sections s1, s4 and s8 were assigned to the Low Risk category
(C3) because they had more satisfactory performances than those presented by
profile b2.
4.4 Other MCDM/A Applications on Multidimensional Risk 197
A sensitivity analysis was conducted in order to analyze responses and opinion

from the DM and to evaluate the robustness of the results with respect to
imprecise data, and the way in which the model can be used by the DM.
The parameters were varied by 10% of the initial value specified by the DM.
It was concluded to be robust for the majority of parameters, such as weights and
profiles for environmental and financial risk criteria.
Nevertheless, it was observed changes for parameters related to weight and
profiles for the human risk criterion (rh). A particular change was found for the
specification of the cutting level k, which is related to the weight for rh.
A reduction of less than 10% in k makes it less than the weight for rh, which should
be avoided. As a result, sections s5 and s6 change from category C2 (Medium
Risk) to C1 (High Risk). According Brito et al. (2010), this happens precisely
because the risk for human criterion is greater than the profile b1 for this criterion.
Since this analysis, the DM decided to maintain the previous results, classifying
sections s5 and s6 as category C2 (Medium Risk).
Another sensitivity was observed when k is increased by 10%. Only section s3
changes to a lesser risk category. Into a more safety view, it was also decided to
maintain the previous classification, so s3 remained in C1.
4.4 Other MCDM/A Applications on Multidimensional Risk
In the next sub-sections, several other decision problems in the related to

multidimensional risk analysis, using MCDM/A, are presented. These problems
are grouped by its context, such as: power electricity systems and natural hazards.
4.4.1 Power Electricity Systems
The generation of electrical energy can be from various sources. Each energy
source will generate different risks inherent in its own production and supply.
Regős (2012) compared the general risk of the four most important energy chains
(coal, nuclear, gas, hydro). For this, he applied an MCDM/A approach, and chose
severe accidents, terrorism, environmental and health risks, risk of price changes
as risk criteria.
Normally, generation and power supply systems are large and complex systems
which society considers form a critical part of the infrastructure. Typically, several
other systems or subsystems, such as water supply systems, telecommunication,
traffic, health, food supply, etc. are dependent on power supply systems. Thus,
failures in the electricity system can impact other systems and generate a chain of
consequences, which is why it is critical for the infrastructure.
Moreover, the system of transmitting and distributing energy consists of

networks in different settings, such as networks in rings, radial or redundant
networks. These settings are intended: to distribute the loads, this creating
redundancy in the system; to increase reliability; to minimize the loss in case
faults occur; or to minimize the occurrence of failures in chains, which can cause
multiple impacts. Therefore, analysis and risk management becomes very complex
since several aspects have to be considered.
There are several reasons for failure in power systems. The most common
technical failures are those which originate from: inadequate maintenance of the
system; system overload; using design (dimensioning) and unsuitable equipment;
conducting maneuvers in the wrong networks (human error); dimensioning loads
poorly, etc.
Besides these factors, one of the causes of failures is due to the occurrence of
extreme natural events such as storms, hurricanes, floods and earthquakes.
Furthermore, there is an external pressure causing stress on the network because
of the need to integrate new public services and the joint use of renewable energy,
and hence, increasingly, power systems are operated closer to their stability limits
(Haidar et al. 2010).
In order to evaluate risk and manage risk effectively, there must be a clear
analysis. Consequently, in order to facilitate the process of decision-making,
various aspects analyzed in this context need to be taken into consideration.
Faced with increasing pressure from society in general for a higher level of
safety, risk management has become an arduous, complex and uncertain task. This
because it can involve all of the following: a large number (hundreds or even
thousands) of primary and secondary power systems with particular characteristics;
the absence or incomplete historical data on failure modes and accidental events
that have already occurred; the rarity of occurrence of accident scenarios; the
magnitude of consequences; and the complexity of the area surrounding the
hazard zone, etc. (Garcez and de Almeida 2014a; Garcez and de Almeida 2014b,
Garcez and de Almeida 2014c).
Therefore, effective risk management plays a role of great importance to
society, the public sector and the electricity distributors, since the impacts caused
by accidents can adversely affect all three areas, directly or indirectly.
The importance of evaluating the risks comprehensively and realistically
generates knowledge that can be applied to assist the distributor power company
in choosing what preventive and mitigating actions to take, thus resulting in risk
management that is effective and efficient (Garcez and de Almeida 2014b).
Furthermore, since the available resources (monetary, time available, work teams,
technology, etc.) of power energy companies are limited and scarce, and regulators
require power systems to demonstrate greater availability and system reliability, it
is necessary to use decision-making that adds in the effects and uncertainties from
multidimensional risks and to evaluate these together with the preference structure
of the company. It is only by doing so that the problem will be dealt with more
adequately (Garcez and de Almeida 2014a, Garcez and de Almeida 2014b).
In the area of asset management of energy companies in general, it is

recognized that there is a need to use a more formal and structured analysis of the
increasingly complex decisions. This is challenging. Current asset management
practices focus primarily on risk quantification in monetary terms, and on the
reliability of the system, combined with estimates of the condition of the
components (estimated lifetime, etc.). The analysis of other aspects of risk, such as
the risk to personal safety, the risk of environmental damage or the risk of a
negative public response are usually “decoupled” from quantitative risk analysis.
So for Catrinu and Nordgård (2011), it is necessary to improve the current practice
of asset management, by making the best use of knowledge and data available
from experts and adopting new methods of risk analysis and decision support, and
moreover, the best ways to document decisions.
For this, Catrinu and Nordgård (2011) integrate the methods of risk analysis
and decision support for advanced management under uncertainty in the assets of
a power distribution system. The focus of this study was to incorporate different
business objectives of risk analysis in a structured framework so as to decide how
to deal with the physical assets of the electricity distribution network.
The growing importance of environmental issues at the global and regional
levels including water and air pollution, the use of non-renewable energy sources,
as well as outcomes such as global warming and climate change, have led to it
being considered essential to take environmental factors into account when
planning how and from where to generate and distribute power (Jozi and Pouriyeh
2011; Rezaian and Jozi 2012). Therefore, in the process for planning energy
systems, uncertainties should be more carefully handled because of the increasing
concern about the environmental impact of electricity generation and because this
market sector is highly competitive.
Linares (2002) presents a multi-criteria model for planning electricity, which
deals with uncertainty and risks associated with minimizing the environmental risk
and performs a risk analysis (in a multicriteria view) to apply classical decision-
making rules and therefore to select the best planning strategy under uncertainty.
Linares emphasizes that incorporating additional criteria leads to more flexible
and efficient strategies, which greatly reduces the environmental risk at a small
incremental cost, while the process of risk analysis selects flexible and robust
strategies for the scenarios analyzed.
In this context of risk management, the need to generate a risk hierarchy of the
various subsystems of the electricity supply system is seen. Garcez and de
Almeida (2014b) propose a form of risk assessment in an underground electricity
distribution system under a multidimensional view (multicriteria), in which they
generate risk measures, which can be ordered. The aim is to generate a priority list
of issues to be considered when allocating additional resources to prevent and
mitigate risks, such as conducting inspections and maintenance; modifying
projects in order to increase safety; developing preventive and mitigating actions;
modernizing and improving the subsystems (upgrade) (Garcez and de Almeida
2014c).
A mitigation measure widely used in case of faults in power systems to prevent

failures in chain is Load Reduction (LR). It is considered a very effective
emergency measure for stabilizing the power system (Dong et al. 2008). To
implement load reduction it is necessary to disconnect certain areas of the power
grid, so this technique generates direct impacts on the population, economy and
local industry. Therefore, normally it is among the last measures to be applied and
is usually only used to prevent the total collapse of the network.
However, to implement LR, it is first necessary to setup which areas should be
disconnected. That choice alone is already a decision process, because not only
operational aspects of systems are considered but also aspects of consequence
covering a multidimensional view of the problem. LR has also been successfully
implemented in Europe and USA. More recently, LR was applied successfully to
manage: the impacts of Hurricane Sandy 2012; The 2006 European Blackout (Van
der Vleuten and Lagendijk 2010); The 2003 Northeast Blackout (Andersson et al.
2005), and The Italian 2003 Blackout (Berizzi 2004).
For the LR method, from a decision analysis point of view, the areas of energy
supply represent the alternatives of the model. To analyze the potential con-
sequences resulting from the uncoupling of these areas, the vulnerability of each
area must be analyzed.
4.4.2 Natural Hazards
According to natural hazard theory, risk appears wherever and whenever assets are
subjected to hazards; it is usually defined as ‘the expected potential loss due to a
particular hazard for a given area and reference period’ and can be mathematically
defined as the combination of hazard and vulnerability (Merad et al. 2004).
For Nefeslioglu et al. (2013), it is fully acceptable for a natural event, such as a
flood or an earthquake to become a natural hazard when people are affected by a
natural hazard. Since the world population is increasing, the need to find habitable
areas has increased considerably, which has led to people having to being caught
up in these natural events more often.
Consequently, Viscusi (2009) states that the occurrence of natural disasters
often generates a cluster of fatalities rather than just a single fatality. Hundreds or
sometimes thousands of people could die from the occurrence of a single event.
Additionally, another important point is that deaths perceived to occur due to
the probability of a natural disaster is a very heterogeneous concept and often, and
its probability is much lower when compared to other risks associated other causes
with fatality. However, the risk management of natural disasters should not just
stick to the issue of people being killed or injured.
Depending on the type and scale of natural event occurring and its impact on
society, other points may be incorporated in the analysis, such as the issue of
safety, security and public health, population migration, cost estimation, information
sharing, planning public and environmental aspects.
For example, when examining the risk of a flood, a check should be made on
aspects that are part of this context, such as the question of the complexity of the
event, broad spatial scales, intervals of time between events, vulnerability and
social-psychological aspects such as depressions, anxieties and conflicts of
interest, in addition to several conflicting aspects between them.
Additionally, taking into account the study described in Levy (2005) for the
operation and management of reservoirs, there is a complex analysis of the
tradeoffs between protection against flooding (i.e., minimizing the discharge of
reservoirs during the peak periods of flooding) and energy production (meeting
the goal of producing pre-defined levels of energy). On the one hand, flood
protection means that the tank must be maintained at the lowest possible level so
that the reservoir can accommodate the excess water coming from the period of
flooding. On the contrary, the production of energy requires that there be the
largest possible amount of water in the reservoir. In this case, the decision-making
process will directly affect risk management, so what is needed is a more
structured analysis that provides satisfactory results.
According to Nefeslioglu et al. (2013) the evaluation of the interaction between
natural and human events in terms of hazards and risks has become a common
topic for analysis in the last 20 years. Modeling consequences and probabilities is
one of the main tools for assessing the impacts of natural hazards.
Given the uncertainty associated with the environmental context, Parlak et al.
(2012) state that analysis based on multicriteria decision methods provides a
systematic approach to managing the complexities and uncertainties associated
with the occurrence of natural disasters, since multicriteria methods make use of
stochastic approach which help to develop this modeling.
Levy (2005) points out the use of MCDM/A has increased in the last three
decades due to a number of factors, including dissatisfaction with conventional
methods that use only a single criterion, as well as ease of access to software and
algorithms that enable the solution of complex environmental problems to be
found. Thus, in the aforementioned study for the operation and management of
reservoirs, MCDM/A is useful for eliciting and modeling stakeholders’ pre-
ferences and to improve coordination between state agencies, organizations and
the affected population in such a way as to minimize the risks associated with
floods e.g., death of or injury to persons, damage to property and possible environ-
mental impacts.
The need for multicriteria approaches can also be observed in planning the
response to a disaster, where, according to Parlak et al. (2012) such planning
requires the engagement of multiple disciplines such as engineering (infrastructure),
management emergencies, health care, mass communication, water supply and
food logistics. Planning the integration scenario by using multicriteria analysis,
according to the authors, enables initiatives to be prioritized and that this

contributes to plans (the response to a disaster) being better understood.
Several applications of multicriteria decision methods are to be found in the
area of risk management for natural disasters, as will be explored in the next
paragraphs.
For addressing territorial risk evaluation considering a group of DM’s, Cailloux
et al. (2013) proposed an MCDM/A model based on ELECTRE TRI method to
evaluate the level of risk for territorial zones surrounded by a given industrial
plant considering a natural hazard, such as flooding. Scawthorn (2008) indicates
how to assess assets at risk in risk areas and in particular, the impact of an
earthquake on social cohesion and peace; public confidence; political unity;
education; and the mental health of the population affected. He includes physical
assets and non-physical assets that can be given a monetary value. Subjective
judgments may be necessary to compare the vulnerability of these different assets
before finally obtaining an overall assessment of risk.
Nefeslioglu et al. (2013) propose a derivation of the AHP (Analytical
Hierarchy Process) MCDM/A method called M-AHP to support decision-making
problems in natural hazard areas, specifically snow avalanches in mountainous
regions.
Karvetski et al. (2011) consider principles of MCDM/A to define a methodo-
logy that measure the impact of possible scenarios for engineering systems in the
context of climate change.
Stefanidis and Stathis (2013) evaluate hazard areas associated with floods by
using AHP and GIS (Geographic Information Systems) to assess the danger from
both natural and anthropogenic aspects, thereby creating of two indices for floods.
Tamura et al. (2000) deal with a process of decision analysis to mitigate risks
associated with natural disasters, which consider events of low probability and
high consequence. The authors propose the use of a function value at risk (Value
Function under Risk) instead of the expected utility theory.
In the context of landslides, the use of GIS and spatial multicriteria evaluation
is widely used. Multiple indicators are processed, analyzed and weighted
according to their contribution to the risk and vulnerability. To reduce losses from
disasters, existing planning on being prepared for a disaster and the immediate
response to it needs to be improved as does planning on how to reduce risks from
disasters. This should be based on a multidimensional evaluation of risk at all
levels of management. Abella and Westen (2007) in their study used four key
indicators for a study on vulnerability from a landslide:
x living conditions and transportation indicators (physical vulnerability);
x population (indicator of social vulnerability);
x production (indicator of economic vulnerability), and;
x protected areas (indicator of environmental vulnerability).
Abella and Westen (2007) applied these indicators and the results obtained
from the analysis led to the development of a plan for mitigating risks from
landslides at the national level in Cuba, and this information being linked to the
national system, which gives early warning of hurricanes, and warns and
evacuates people from areas prone landslides.
Another context of natural hazard together with man’s intervention arises from
ending mining operations in populated regions. According to Merad et al. (2004),
in the Lorraine region of France many landslides and subsidence have occurred,
which led to the need to develop a specific methodology for risk zoning of the
area. The authors propose a methodology based on a multicriteria decision support
tool (ELECTRE-TRI), with the aim of assigning risk zones in predefined classes
of inhabited regions. This approach enabled the knowledge of experts, multiple
qualitative and quantitative criteria and uncertainties to be considered.
4.4.3 Risk Analysis on Counter-Terrorism
In recent decades, the fight against terrorism has been the focus of constant
analysis worldwide. Security measures have been strengthened and new anti-
terrorism policies are presented to the world by nations periodically to society.
One goal of these policies is to establish the benefits of preventing a terrorist
attack, such as reducing the number of deaths and injuries associated with the
human dimension. More specifically, the risk management of terrorist attacks has
intensified, especially after the terrorist attack on the World Trade Center on
September 11, 2001 in the United States.
Risk management, according to Aven and Renn (2009) seeks to ensure that
adequate measures are established to protect people, the environment and assets
from harmful consequences arising from human activities or natural events. The
extent to which risk reduction measures are justified depends on the balance
between costs and benefits in terms of the security gain. Furthermore, several PRA
models have been applied taking into account aspects such as infrastructure, food
supply chains, population, etc., and considering risk as a product of three
components: threat, vulnerability and consequence (Greenberg et al 2012).
Due to terrorism threats, several models and approaches have been proposed in
order to mitigate the risks associated with such events (Merrick and Leclerc 2014;
Shan and Zhuang 2014; Haphuriwat and Bier 2011; Ezell et al. 2010; Parnell et al.
2010; Ngange et al 2008; Leung et al 2004). Among these models, there are
several studies considering MCDM/A approaches (Akgun et al. 2010; Sri
Bhashyam and Montibeller 2012; Koonce et al. 2008; Patterson and Apostolakis
2007).
In their paper, Akgun et al. (2010) stress that assessing the vulnerability of
critical assets (e.g.: airports, dams, chemical plants, nuclear power plants) to
terrorist attacks is a highly complex strategic activity, requiring a methodology
structured to support the decision-making process in defense planning. Their
approach seeks to define the vulnerability of each critical defense asset against
terrorist attacks taking into account multiple criteria. They use SMART in
conjunction with Fuzzy Set Theory and Fuzzy Cognitive Maps in a group decision
environment. Their model seeks to identify hidden vulnerabilities and to define the
roles and most critical (or active) components of each system, and five criteria
were established:
x Deterrence (implemented method of defense, perceived by terrorists as hard to
penetrate);
x Detection (of a terrorist attack);
x Delay (the time during which an element of a physical protection system is
designed to prevent terrorist invasions);
x Response (the time taken to respond to a threat), and;
x Recovery (the time taken to return the areas and people affected to their
existing status prior to the event).
In the second study, Sri Bhashyam and Montibeller (2012) propose a
framework that can be used to infer how the priorities of the terrorists may change
over time and the impact that these changes may have on the choice of a harmful
action. This is done based on a multicriteria model that uses MAUT. The
objectives were visualized in three categories: revenge, reputation and reaction.
The alternatives of the decision problem were established by: strikes by the
terrorists, improvised explosive devices in a public place, explosions of portable
nuclear devices in modes of mass transit, detonating bombs and biological
weapons or dirty bombs that combine explosives and radioactive materials.
Therefore, the aim of modeling terrorists’ priorities is to define the objectives
that terrorists will use to evaluate the attack, providing the best tradeoff between
the operational side of an attack (costs) and benefits (if goals are achieved).
4.4.4 Nuclear Power
Risk analysis in power systems is a crucial activity so as to ensure adequate

security for society, especially with regard to the operation of power plants. More
specifically, in recent years, different sectors of society have insisted on new
discussions with regard to safety in nuclear power plants due to the Fukushima
accident in Japan in 2011. In this context, according to Rogner (2013), accidents
like Fukushima have created a greater climate of distrust with respect to society’s
view of nuclear energy. In contrast, industries have tried to increase their security
level. Additionally, several issues have begun to dominate public debate on energy
policy such as: energy security; the price of fossil fuels; climate change; the
increase in the demand for electricity. As nuclear power has a mitigating role in
several of these points, the societies of some countries once again have a higher
level of tolerance for nuclear technology.
In this context, Papamichail and French (2012) points out that radioactive
accidents have emphasized the requirement to provide support for all emergency
management phases. Several decision support tools are currently being developed
to prevent and mitigate the effects of radioactive accidents. Among these tools,
multicriteria decision techniques stand out.
The literature describes some recent applications of multicriteria decision
methods in the context of nuclear energy. Examples include the following:
x Atmaca and Basar (2012) use an Analytic Network Process (ANP) to evaluate
6 different alternatives of nuclear power plants taking into account criteria such
as technological aspects and sustainability, economic viability, quality of life
and socio-economic impacts.
x Hong et al. (2013) use a multi-criteria decision analysis to assess future
scenarios for generating electricity in Japan electricity which take economic,
environmental and social impacts into consideration. Their study is a response
to the nuclear crisis caused by the Fukushima accident.
x Erol et al. (2014) define the location problem of a nuclear power plant in
Turkey as a multicriteria decision problem using fuzzy logic, and consider
qualitative and quantitative criteria. The primary criteria that they establish are:
proximity to the existing electrical infrastructure; proximity to the transportation
infrastructure; access to large amounts of cooling water. The authors also
consider a number of secondary criteria: population density; geological issues;
atmospheric conditions; cost factors; and risk factors.
x Beaudouin (2015) proposes an MCDM/A model that supports debate about
nuclear power plants safety choices. Therefore, six safety criteria are considered in
combination with cost-effectiveness analyses to point out the best portfolio of
power plant design modifications, satisfying security requirements.
Thus, MCDM/A tools can be used at various stages in the context of nuclear
power production in order to contribute to risk management, thereby making the
decision-making process an important aspect when planning safety measures for
nuclear power plants.
4.4.5 Risk Analysis on Other Contexts
A requirement for the building industry, both with regard to permission to build
and to certification that the finished building meets regulatory requirements, is
necessary an environmental management, where the identification and evaluation
of risk to human and environmental health are the first stages. Topuz et al. (2011)
propose an approach that integrates the assessment of risk to humans with environ-
mental health in industries using hazardous materials, to support environmental
DMs with quantitative and directive results. For this, the methodologies used
multicriteria and fuzzy logic to deal with the problems arising from the complexity
of the environment and uncertain data.
Specifically, in the context of bridges, the use of bridges is among the most
important structural elements that reduce traffic problems. Risk management of
bridges serves to determine the best allocation of resources. According to Adey et
al. (2003), these systems are usually evaluated against the structural deterioration
which bridges may suffer from as a result of traffic loads. However, these systems
are affected by various other hazards, such as floods and earthquakes, not only the
traffic load.
The destruction of big bridges are usually important and significant events, and
may result in loss of lives, property and economic losses. According to Shetty et
al. (1997), the consequences of the destruction of bridges can be summarized as:
 Human elements that impact the number of deaths and injuries, such as the
high rate of vehicular traffic, the flow of pedestrians that pass over or under the
bridge;
 Environmental consequences resulting from spills of hazardous substances, due
to the intersection of transport between road, rail, etc.;
 Formation of traffic jams, increasing the volume of traffic at a particular site,
causes overload on other transport routes;
 Economic factors, including the cost of taking construction material residuals
away; reconstruction; indemnities payable on the destruction of vehicles; the
environmental catharsis; and legal costs.
According to Wang et al. (2008), the risk assessment of bridges is essentially a
multicriteria problem, which involves multiple assessment criteria such as safety
(safety of the public), functionality (effects on the level of service/availability of
the network for use), sustainability (expenditure and workload) and environment
(effects on the environment, including the (aesthetic) appearance of the
structures). Wang et al. (2008) propose an integrated AHP–DEA methodology to
evaluate risks to or from bridges of hundreds or thousands of bridge structures,
based on which maintenance priorities for the bridge structures can be drawn up.
Environmental risk assessment and decision-making strategies in recent
decades have become increasingly sophisticated, and use intensive and complex
information, including approaches such as expert opinions, cost-benefit analyzes
and evaluation of the toxicological risk. According to Linkov et al. (2006), a tool
that has been used to support environmental decision-making is comparative risk
assessment (CRA), but CRA lacks a structured process to arrive at an alternative
optimal design method. The approach of using multicriteria decision analysis fills
this need by providing methods that give better support to comparing alternatives
and also provides a structure which incorporates input from stakeholders of the
project, the aim of which is to rank alternatives.
In the context of the hazard from forest fires, over past decades, in several
regions, especially in tropical and Mediterranean regions, these fires are due to
several underlying factors, which have received increasing attention because of
the wide range of ecological, economic, social and political impacts. The more
complex fire models require spatial information, which is done by remote sensing
and GIS (Vadrevu et al. 2010; Arianoutsou et al. 2011).
According to Vadrevu et al. (2010), the integration of MCDM/A methods in
the spatial domain provides a new framework for addressing many environmental
problems, including quantifying “fire hazards”. These authors conducted a study
in a thickly-forested area (Indian region), where most of the stakeholders are the
local people, and their dependence on forest resources is immense.
Moreover, the problem of forest fires in the study area is spatially diverse in
nature and involves both biophysical and socioeconomic parameters, providing an
ideal place to use an MCDM/A methodology. Combining these multiple para-
meters using decision-making methods in a collaborative framework may yield
good results, so, the risk of fires in tropical deciduous forests, in India, was
quantified as a function of topographic, vegetation, climatic, and socioeconomic
attributes in order to evaluate the fire risk in the study area.
Still in the environmental context, the contamination of water resources on land
has been a major environmental concern during the last decades, mainly due to
public health concerns. According to Khadam and Kaluarachchi (2003), traditionally,
environmental decision-making scenarios of subsurface contamination are guided
by means of cost-benefit analysis.
This context, the risk assessment includes quantification of the risk to human
health, as well as evaluating the importance of this risk. When the risk is deter-
mined unacceptable, potential remedial alternatives are identified and decision
analysis is performed to choose the best corrective action. There is a tradeoff
between individual risk and societal risk, the tradeoff between the residual risk
and the cost of reducing this risk, and cost-effectiveness as a justification for
remediation. The authors propose an integrated approach for the management of
contaminated ground water using a multicriteria decision framework to assess the
risk to health and to make an economic analysis.
Another current context to be analyzed is in the newish field of nanotechnology
which is increasingly being embedded in innovations that can benefit humanity
(Siegrist et al. 2007). However, there is a variety of factors involved in managing
the development of nanomaterial, ranging from the technical specifications of the
material to possible adverse effects in humans. Therefore, it is important to assess
the benefits and risks inherent in issues of Environmental Health and Safety (EHS)
related to nanotechnology. According to Linkov et al. (2007), there is currently no
structured approach for making justifiable and transparent decisions with explicit
trade-offs among the many factors.
Linkov et al. (2007) conceptualize the use of the MCDM/A as a powerful
analytical framework and scientifically sound decision tool for assessing and
managing risk when using nanomaterial. They seek a balance between social
benefits and unintended side effects and risks. They also investigate how to gather
multiple lines of evidence to estimate the likely toxicity and risks of nanomaterial,
given limited information on its physical and chemical properties. An essential
contribution of MCDM/A, highlighted by the authors, is to link this information

on performance with decision criteria and weightings triggered from scientists and
managers, thus enabling the trade-offs involved in the decision making process to
be visualized and quantified.
Luria and Aspinall (2003) use expert opinions, complementary skills and
expertise from different disciplines in conjunction with quantitative traditional
analysis, in an approach to major industrial hazard assessment, based on a multi-
criteria approach (Analytic Hierarchy Process - AHP). According to these authors,
this approach is in line with the main concepts proposed by the European directive
on major hazard accidents, which recommends increasing the participation of
operators, taking the other players into account and, moreover, paying more
attention to the concepts of urban control, subjective risk (risk perception) and
intangible factors.
References
Abella EAC, Van Westen CJ (2007) Generation of a landslide risk index map for Cuba using
spatial multi-criteria evaluation. Landslides 4:311–325
Adey B, Hajdin R, Brühwiler E (2003) Risk-based approach to the determination of optimal
interventions for bridges affected by multiple hazards. Eng Struct 25:903–912
Akgun I, Kandakoglu A, Ozok AF (2010) Fuzzy integrated vulnerability assessment model for
critical facilities in combating the terrorism. Expert Syst Appl 37:3561–3573
Alencar MH, Cavalcante CAV, de Almeida AT, Silva Neto CE (2010) Priorities assignment for
actions in a transport system based on a multicriteria decision model. In: Bris R, Soares CG,
London, UK, p 2480
Alencar MH, de Almeida AT (2010) Assigning priorities to actions in a pipeline transporting
hydrogen based on a multicriteria decision model. Int J Hydrogen Energy 35(8):3610–3619
Alencar MH, Krym EM, Marsaro MF, de Almeida AT (2014) Multidimensional risk evaluation
in natural gas pipelines: Environmental aspects observed through a multicriteria decision
model. In: Steenbergen RDJM, VanGelder PHAJM, Miraglia S, Vrouwenvelder ACWMT
(eds) 22nd Annual Conference on European Safety and Reliability (ESREL), Amsterdam,
2013. Safety, Reliability and Risk Analysis: Beyond the Horizon. Taylor & Francis Group,
London, UK, p 758
Almeida-Filho AT de, de Almeida AT (2010a) Cost-effectiveness analysis and multicriteria
approaches: two irreplaceable paradigms for different problems in risk and safety problems.
In: Proceedings of the European Safety and Reliability Annual Conference, Rhodes, 2010.
Reliability, Risk and Safety: Back to the Future, p 2293
Almeida-Filho AT de, de Almeida AT (2010b) Multiple dimension risk evaluation framework.
In: Bris R, Soares CG, Martorell S (eds) European Safety and Reliability Conference
(ESREL), Prague, Czech Republic, 2009. Reliability, Risk and Safety: Theory and
Applications. CRC Press-Taylor & Francis Group, p 1049
Andersson G, Donalek P, Farmer R, et al. (2005) Causes of the 2003 major grid blackouts in
North America and Europe, and recommended means to improve system dynamic
performance. Power Syst IEEE Trans 20:1922–1928
References 209
Apostolakis GE, Lemon DM (2005) A Screening Methodology for the Identification and
Ranking of Infrastructure Vulnerabilities Due to Terrorism. Risk Anal 25:361–376
Arianoutsou M, Koukoulas S, Kazanis D (2011) Evaluating Post-Fire Forest Resilience Using
GIS and Multi-Criteria Analysis: An Example from Cape Sounion National Park, Greece.
Environ Manage 47:384–397
Arnaldos J, Casal J, Montiel H, et al. (1998) Design of a computer tool for the evaluation of the
consequences of accidental natural gas releases in distribution pipes. J Loss Prev Process Ind
11:135–148
Atmaca E, Basar HB (2012) Evaluation of power plants in Turkey using Analytic Network
Process (ANP). Energy 44:555–563
Aven T, Renn O (2009) The Role of Quantitative Risk Assessments for Characterizing Risk and
Uncertainty and Delineating Appropriate Risk Management Options, with Special Emphasis
on Terrorism Risk. Risk Anal 29:587–600
Beaudouin F (2015) Implementing a Multiple Criteria Model to Debate About Nuclear Power
Plants Safety Choices. Gr Decis Negot 1–29
Beaudouin F, Munier B (2009) A revision of industrial risk management: Decisions and
experimental tools in risk business. Risk Decis Anal 1:3–20
Bedford T, Cooke R (2001) Probabilistic Risk Analysis: Foundations and Methods. Cambridge
Media, New York
Berizzi A (2004) The Italian 2003 blackout. Power Eng Soc Gen Meet 2004 IEEE 1673–1679
Vol. 2
Brito AJ, de Almeida AT, Mota CMM (2010) A multicriteria model for risk sorting of natural
gas pipelines based on ELECTRE TRI integrating Utility Theory. Eur J Oper Res 200:812–
821
Cailloux O, Mayag B, Meyer P, Mousseau V (2013) Operational tools to build a multicriteria
territorial risk scale with multiple stakeholders. Reliab Eng Syst Saf 120:88–97
Catrinu MD, Nordgård DE (2011) Integrating risk analysis and multi-criteria decision support
under uncertainty in electricity distribution system asset management. Reliab Eng Syst Saf
96:663–670
Comes T, Wijngaards N, Hiete M, et al (2011) A Distributed Scenario-Based Decision Support
System for Robust Decision-Making in Complex Situations. Int J Inf Syst Cris Response
Manag 3:17–35
Cox LA Jr (2009) Risk analysis of complex and uncertain systems. Springer Science & Busi-
ness Media, New York
Cox LA Jr (2012) Evaluating and Improving Risk Formulas for Allocating Limited Budgets to
Expensive Risk-Reduction Opportunities. Risk Anal 32(7):1244–1252
Crowl DA, Jo Y-D (2007) The hazards and risks of hydrogen. J Loss Prev Process Ind 20:158–
164
de Almeida AT (2005) Multicriteria Modelling of Repair Contract Based on Utility and
ELECTRE I Method with Dependability and Service Quality Criteria. Ann Oper Res
138:113–126
de Almeida AT (2007) Multicriteria decision model for outsourcing contracts selection based on
utility function and ELECTRE method. Comput Oper Res 34(12):3569–3574
Dong M, Lou C, Wong C (2008) Adaptive Under-Frequency Load Shedding. Tsinghua Sci
Technol 13:823–828
DziubiĔski M, Frątczak M, Markowski AS (2006) Aspects of risk analysis associated with major
failures of fuel pipelines. J Loss Prev Process Ind 19:399–408
Erol ø, Sencer S, Özmen A, Searcy C (2014) Fuzzy MCDM framework for locating a nuclear
power plant in Turkey. Energy Policy 67:186–197
Ezell BC, Bennett SP, Von Winterfeldt D, et al. (2010) Probabilistic Risk Analysis and
Terrorism Risk. Risk Anal 30(4):575–589
Faber MH, Stewart MG (2003) Risk assessment for civil engineering facilities: critical overview
and discussion. Reliab Eng Syst Saf 80:173–184
Garcez TV, de Almeida AT (2012) Multiple Dimension Manhole Explosion in an Underground
Electrical Distribution System. In: Proceedings of the 11th International Probabilistic Safety
Assessment and Management Conference and the Annual European Safety and Reliability
Conference 2012. Curran Associates, Inc. Helsinki, Finland, p 4893–4899
Garcez TV, de Almeida AT (2014a) A risk measurement tool for an underground electricity
distribution system considering the consequences and uncertainties of manhole events. Reliab
Eng Syst Saf 124:68–80
Garcez TV, de Almeida AT (2014b) Multidimensional Risk Assessment of Manhole Events as a
Power Deliv IEEE Trans 29:624–632
Garcez TV, de Almeida AT (2014c) Multidimensional risk assessment of underground electricity
distribution systems based on MAUT. In: Steenbergen RDJM, VanGelder PHAJM, Miraglia
S, Vrouwenvelder ACWMT (eds) 22nd Annual Conference on European Safety and
Reliability (ESREL), Amsterdam, 2013. Safety, Reliability and Risk Analysis: Beyond the
Horizon. CRC Press-Taylor & Francis Group, p 2009
Garcez TV, de Almeida-Filho AT, de Almeida AT, Alencar MH (2010) Multicriteria risk
analysis application in a distribution gas pipeline system in Sergipe. In: Bris R, Soares CG,
Martorell S (eds) Reliability, risk and safety: theory and applications vols 1-3. European
safety and reliability conference (ESREL 2009), Prague, September 2009. Taylor and
Francis, 1043-1047
Geldermann J, Bertsch V, Treitz M, et al. (2009) Multi-criteria decision support and evaluation
of strategies for nuclear remediation management. Omega 37:238–251
Greenberg M, Haas C, Cox LA Jr, et al. (2012) Ten Most Important Accomplishments in Risk
Analysis, 1980–2010. Risk Anal 32(5):771–781
Guedes Soares CG, Teixeira AP (2001) Risk assessment in maritime transportation. Reliab Eng
Syst Saf 74(3):299–309
Haidar AMA, Mohamed A, Hussain A (2010) Vulnerability control of large scale interconnected
power system using neuro-fuzzy load shedding approach. Expert Syst Appl 37:3171–3176
Haphuriwat N, Bier VM (2011) Trade-offs between target hardening and overarching protection.
Eur J Oper Res 213(1):320–328
Henselwood F, Phillips G (2006) A matrix-based risk assessment approach for addressing linear
hazards such as pipelines. J Loss Prev Process Ind 19:433–441
Hobbs BF, Meier P (2000) Energy Decisions and the Environment. A guide to the use of
multicriteria methods (International Series in Operations Research & Management Science).
Kluwer Academic Publisher, Norwell
Hong S, Bradshaw CJA, Brook BW (2013) Evaluating options for the future energy mix of
Japan after the Fukushima nuclear crisis. Energy Policy 56:418–424
IEEE1584 (2002) IEEE Guide for Performing Arc-Flash Hazard Calculations. IEEE Std 1584-
2002
Jo Y-D, Ahn BJ (2002) Analysis of hazard areas associated with high-pressure natural-gas
pipelines. 2 J Loss Prev Process Ind 15:179–188
Jo Y-D, Ahn BJ (2005) A method of quantitative risk assessment for transmission pipeline
carrying natural gas. J Hazard Mater 123:1–12
References 211
Jo Y-D, Crowl DA (2008) Individual risk analysis of high-pressure natural gas pipelines. J Loss
Prev Process Ind 21:589–595
Jozi A, Pouriyeh A (2011) Health-safety and environmental risk assessment of power plants
using multi criteria decision making method. Chem Ind Chem Eng Q 17:437–449
Karvetski CW, Lambert JH, Keisler JM, Linkov I (2011) Integration of Decision Analysis and
Scenario Planning for Coastal Engineering and Climate Change. Syst Man Cybern Part A
Syst Humans, IEEE Trans 41(1):63–73
Khadam IM, Kaluarachchi JJ (2003) Multi-criteria decision analysis with probabilistic risk
assessment for the management of contaminated ground water. Environ Impact Assess Rev
23:683–721
Koonce AM, Apostolakis GE, Cook BK (2008) Bulk power risk analysis: Ranking infrastructure
elements according to their risk significance. Int J Electr Power Energy Syst 30:169–183
Leung M, Lambert JH, Mosenthal A (2004) A Risk-Based Approach to Setting Priorities in
Protecting Bridges Against Terrorist Attacks. Risk Anal 24(4):963–984
Levy J (2005) Multiple criteria decision making and decision support systems for flood risk
management. Stoch Environ Res Risk Assess 19:438–447
Linares P (2002) Multiple criteria decision making and risk analysis as risk management tools
for power systems planning. Power Syst IEEE Trans 17:895–900
Linkov I, Satterstrom FK, Kiker G, et al. (2006) From comparative risk assessment to multi-
criteria decision analysis and adaptive management: Recent developments and applications.
Environ Int 32:1072–1093
Linkov I, Satterstrom FK, Steevens J, et al. (2007) Multi-criteria decision analysis and
environmental risk assessment for nanomaterials. J Nanoparticle Res 9:543–554
Lopes YG, de Almeida AT, Alencar MH, Wolmer Filho LAF, Siqueira GBA (2010) A Decision
Support System to Evaluate Gas Pipeline Risk in Multiple Dimensions. In: Bris R, Soares
CG, Martorell S (eds) European Safety and Reliability Conference (ESREL), Prague, Czech
Republic, 2009. Reliability, Risk and Safety: Theory and Applications. Crc Press-Taylor &
Francis Group, p 1043
Luria P, Aspinall PA (2003) Evaluating a multi-criteria model for hazard assessment in urban
design. The Porto Marghera case study. Environ Impact Assess Rev 23:625–653
Merad MM, Verdel T, Roy B, Kouniali S (2004) Use of multi-criteria decision-aids for risk
zoning and management of large area subjected to mining-induced hazards. Tunn Undergr Sp
Technol 19:125–138
Merrick JRW, Leclerc P (2014) Modeling Adversaries in Counterterrorism Decisions Using
Prospect Theory. Risk Anal. doi: 10.1111/risa.12254
Montiel LV, Bickel JE (2014) A Generalized Sampling Approach for Multilinear Utility
Functions Given Partial Preference Information. Decis Anal 11(3):147–170
Morgan MG, Florig HK, DeKay ML, Fischbeck P (2000) Categorizing Risks for Risk Ranking.
Risk Anal 20:49–58
Mousseau V, Slowinski R (1998) Inferring an ELECTRE TRI Model from Assignment
Examples. J Glob Optim 12:157–174
Nefeslioglu HA, Sezer EA, Gokceoglu C, Ayas Z (2013) A modified analytical hierarchy
process (M-AHP) approach for decision support systems in natural hazard assessments.
Comput Geosci 59:1–8
Nganje W, Bier V, Han H, Zack L (2008) Terrorist Threats to Food: Guidance for Establishing
and Strengthening Prevention and Response Systems. Am J Agric Econ 90(5):1265–1271
Papamichail KN, French S (2013) 25 Years of MCDA in nuclear emergency management. IMA
J. Manag. Math. pp 481–503
Parlak AI, Lambert JH, Guterbock TM, Clements JL (2012) Population behavioral scenarios
influencing radiological disaster preparedness and planning. Accid Anal Prev 48:353–62
Parnell GS, Smith CM, Moxley FI (2010) Intelligent Adversary Risk Analysis: A Bioterrorism
Risk Management Model. Risk Anal 30:32–48
Patterson SA, Apostolakis GE (2007) Identification of critical locations across multiple
infrastructures for terrorist actions. Reliab Eng Syst Saf 92:1183–1203
Radeva A, Rudin C, Passonneau R, Isaac D (2009) Report Cards for Manholes: Eliciting Expert
Feedback for a Learning Task. Mach Learn Appl 2009 ICMLA ’09 Int Conf 719–724
Wesley, London
RegĘs G (2013) Comparison of power plants’ risks with multi criteria decision models. Cent Eur
J Oper Res 21:845–865
Rezaian S, Jozi SA (2012) Health- Safety and Environmental Risk Assessment of Refineries
Using of Multi Criteria Decision Making Method. APCBEE Procedia 3:235–238
Rogner H-H (2013) World outlook for nuclear power. Energy Strateg Rev 1:291–295
Rudin C, Passonneau R, Radeva A, et al. (2010) A process for predicting manhole events in
Manhattan. Mach Learn 80:1–31
Rudin C, Passonneau RJ, Radeva A, et al. (2011) 21st-century data miners meet 19th-century
electrical cables. Computer (Long Beach Calif) 44:103–105
Rudin C, Waltz D, Anderson RN, et al. (2012) Machine Learning for the New York City Power
Grid. Pattern Anal Mach Intell IEEE Trans 34:328–345
Salvi O, Merad M, Rodrigues N (2005) Toward an integrative approach of the industrial risk
management process in France. J Loss Prev Process Ind 18:414–422
Scawthorn C (2008) A Brief History of Seismic Risk Assessment. In: Bostrom A, French S,
Gottlieb S (eds) Risk Assessment, Model. Decis. Support. Springer Berlin Heidelberg, Berlin,
Shan X, Zhuang J (2014) Subsidizing to disrupt a terrorism supply chain-a four-player game.
J Oper Res Soc 65(7):1108–1119
Shetty NK, Chubb MS, Halden D (1997) An overall risk-based assessment procedure for
substandard bridges. In: Das PC (ed) Saf. Bridg. Telford, London, Uk, pp 225–235
Siegrist M, Keller C, Kastenholz H, et al. (2007) Laypeople’s and Experts’ Perception of Nano-
technology Hazards. Risk Anal 27:59–69
Sklavounos S, Rigas F (2006) Estimation of safety distances in the vicinity of fuel gas pipelines.
J Loss Prev Process Ind 19:24–31
Sri Bhashyam S, Montibeller G (2012) Modeling State-Dependent Priorities of Malicious
Agents. Decis Anal 9:172–185
Stefanidis S, Stathis D (2013) Assessment of flood hazard based on natural and anthropogenic
factors using analytic hierarchy process (AHP). Nat Hazards 68:569–585
Tamura H, Yamamoto K, Tomiyama S, Hatono I (2000) Modeling and analysis of decision
making problem for mitigating natural disaster risks. Eur J Oper Res 122:461–468
Topuz E, Talinli I, Aydin E (2011) Integration of environmental and human health risk
assessment for industries using hazardous materials: A quantitative multi criteria approach
for environmental decision makers. Environ Int 37:393–403
Tweeddale M (2003) Managing Risk and Reliability of Process Plants. Gulf Professional
Publishing. Burlington
Vadrevu K, Eaturu A, Badarinath KVS (2010) Fire risk evaluation using multicriteria analysis—
a case study. Environ Monit Assess 166:223–239
Van der Vleuten E, Lagendijk V (2010) Transnational infrastructure vulnerability: The historical
shaping of the 2006 European “Blackout.” Energy Policy 38:2042–2052
Viscusi WK (2009) Valuing risks of death from terrorism and natural disasters. J Risk Uncertain
38:191–213
References 213
Walsh BP, Black WZ (2005) Thermodynamic and mechanical analysis of short circuit events in
an underground vault. Power Deliv IEEE Trans 20:2235–2240
Wang J (2006) Maritime Risk Assessment and its Current Status. Qual Reliab Eng Int 22(1):3–19
Wang Y-M, Liu J, Elhag TMS (2008) An integrated AHP–DEA methodology for bridge risk
assessment. Comput Ind Eng 54:513–525
Willis HH, DeKay ML, Fischhoff B, Morgan MG (2005) Aggregate, Disaggregate, and Hybrid
Analyses of Ecological Risk Perceptions. Risk Anal 25:405–428
Yoe C (2012) Principles of Risk Analysis – Decision Making under uncertainty. CRC Press,
Boca Raton
Chapter 5
Preventive Maintenance Decisions
Abstract: Technological advances in equipment and the increase in process

automation has led to the maintenance function having a role in business
competitiveness. The contribution of preventive maintenance is discussed, as an
important part of this function, with some emphasis on methods for planning
replacement, in the sense of time interval of preventive maintenance. The classical
optimization approach is used to illustrate the original preventive maintenance
problem, thereby enabling insights and discussion of the main features that require
the use of MCDM/A approaches for these decisions, and thus considering the
multidimensional consequence space. A structured framework to build a multi-
criteria decision model for supporting the selection of time interval is presented.
Two different MCDM/A methods are applied depending on the decision maker’s
(DM) preferences. The first illustrates the application of Multi-attribute Utility
Theory (MAUT) as an example of compensatory method and; the second details
the application of a non-compensatory PROMETHEE method, which considers
outranking relations.
5.1 Introduction
In the face of growing competition, leading to an ever increasing need for higher
productivity, there is a need for methods, tools and technologies that enable the
producing systems to acquire competitive advantages. Preventive maintenance
decisions are quite relevant to the strategic results of any business organization, in
which a producing systems has to make products, may them be goods or services.
The type of product makes a great difference in the way that maintenance in
general (and preventive maintenance in particular) is linked to business results.
For instance, a service producing system has a feature of simultaneousness (Slack
et al. 2010), which means that at the time the system is producing the product, the
customer is being served. In such a context, when a failure in the system occurs,
the maintenance has an immediate impact on the business competitiveness (de
Almeida and Souza 2001). Therefore, preventive maintenance planning becomes a
more strategic decision that is linked to a higher level of the hierarchical organizational
structure. For given decision context, the consequences are characterized by

216 Chapter 5 Preventive Maintenance Decisions
multiple and less tangible objectives, which may require an MCDM/A support.
These issues are discussed in Chap. 1, presenting the peculiarities of two different
types of systems, service and goods producing systems.
This chapter addresses preventive maintenance planning by the selection of the
preventive maintenance time interval. This kind of decision is applied to a
component or item (or device) and it is not applicable to a system, unless this
system is replaced as whole, using the failure behavior of the system.
The next section presents a classical optimization approach, followed by a
general MCDM/A preventive maintenance model. The last two sub-sections deal
with two different MCDM/A approaches to support the preventive maintenance
time interval.
It is important to note that there are many different mathematical models
related to preventive maintenance (Shafiee and Finkelstein 2015), although similar
process to build MCDM/A models can be applied considering the required
adaptations.
5.2 A General MCDM/A Model for Preventive Maintenance
One of the most important problems in the maintenance area is the definition of
the frequency at which preventive maintenance actions should be performed.
In both producing systems, this decision has a great impact.
In a literature review on MCDM/A models in maintenance, around 22 % of
research found is related to preventive maintenance (de Almeida et al. 2015).
Multiobjective optimization in preventive maintenance has been considered since
the late 1970. Inagaki et al. (1978) considered three objectives: mission reliability,
total cost, system weight in a multiobjective nonlinear mixed-integer problem and
proposed a procedure based on interactive optimization and a nonlinear pro-
gramming algorithm, called ICOM (Interactive Coordinatewise Optimization
Method). Hwang et al. (1979) formulated a scheduled-maintenance policy problem
and set three objectives: minimum replacement cost-rate, maximum availability,
and lower-bound on mission reliability. Four multicriteria methods were analyzed:
strictest-selection; lexicographic; Waltz lexicographic, and the sequential multiple-
objective problem-solving technique (SEMOPS). Jiang and Ji (2002) consider
four attributes: cost, availability, reliability, and lifetime, via a multiple attribute
value theory (MAVT).
Before the presentation of a general MCDM/A preventive maintenance model,
a classical optimization approach is presented in the next subsection.
5.2 A General MCDM/A Model for Preventive Maintenance 217
5.2.1 Classical Optimization Problem of Preventive Maintenance
Glasser (1969) presents age replacement and block replacement as two methods of
planning replacement in a program of preventive maintenance. Although these
methods have been previously described by other authors (Barlow and Hunter
1960; Cox 1962), the main contribution of Glasser (1969) is the focus on the
managerial impacts of these methods. These issues have been presented in many
other texts in the literature (Scarf et al. 2005).
According to Glasser (1969), the main problem of preventive maintenance is
associated with the uncertainty about the exact time at which an item will fail.
This uncertainty establishes a difficulty in guarantee the effectiveness of the
replacement, which in some times could happen earlier than failure, in others only
after a failure takes place.
Glasser (1969) structures a two-phase process to model the problem of
replacement planning. The phases consist of: 1) the description of the pattern of
failures of the item over time, in terms of a probability density function f(t); and 2)
the development of an equation that describes the expected cost per time of
following a particular policy of planned replacement.
A general description of this model is given by (5.1).
E ( c(t ))
cr (t ) (5.1)
E ( v (t ))
where:
cr(t) is the cost rate;
E(c(t)) is the expected cost;
E(v(t)) is the expected cycle length.
The final expression of cr(t) depends on the assumptions of the model, which
may be related to the influence of the action on the system.
A great number of papers deal with these different aspects, almost all following
the general structure of Glasser (1969). Although these models have a great potential
to support the maintenance manager, they may not be sufficient to describe the
consequence space of failures. So, the rate cost as a criterion should be considered
together with other criteria. This is described in a general framework that could be
used to address the problem via the MCDM/A approach, given in the next sub-
section.
The assumptions related to the simplest case of the replacement model (Cox
1962) are valid for the MCDM/A models in the subsequent sub-sections. The
assumptions of the simplest replacement age-based model are (Cox 1962):
1. The state of the item is known;
2. The alternatives set is defined as opportunistic intervals, which can be days,
weeks, months, or other period;
3. The failure probability density function f(t) of the item is IFR (an increasing
failure rate);
4. The system can only be in one of two states, failed or operational;
5. Replacement prior a failure is worthwhile, there are saving in avoiding a failure
by doing preventive replacements;
6. The item replacement restores the system to the as good as new state;
7. The equipment failure times can be modeled by a known probability density
function f(t);
8. The time necessary to perform a replacement is negligible compared to the time
between failures, so it is not considered into a cycle.
From these assumptions (5.1) becomes (5.2).
ca (1 R(t )) cb R(t )
cr (t ) t
(5.2)
³ xf ( x)dx t( R(t ))
0
where
cr(t) is the cost rate;
R(t) is the reliability function;
f(t) is the density probability function;
ca is the cost replacement after a failure;
cb is the cost replacement before a failure.
As already stated, most of the models restrict the analysis to only the cost rate
criterion cr(t). Therefore, it is important to understand the behavior of this aspect.
To illustrate the different behaviors of the cost rate, a Weibull function is
assumed to f(t). Also, different values for its parameters may be applied to give an
idea of how the cost rate may change with different failure data patterns.
Assuming a Weibull distribution in (5.2), (5.3) can be obtained.
E E
ª § t ·º ª § t ·º
« ¨¨ ¸¸ » « ¨¨ ¸¸ »
K K
ca (1 e «¬ © ¹ »¼ ) cb e «¬ © ¹ »¼
cr (t ) E E
(5.3)
t E 1 ª« §¨ x ·¸ º» ª § t ·º
« ¨¨ ¸¸ »
E ªxº ¨ ¸
¬« © K ¹ ¼» ¬« © K ¹ ¼»
³ x K «¬K »¼
0
e dx te
It is clear that preventive maintenance is only effective if the f(t) function is

IFR. In practice, this means that a time-based preventive maintenance action, is
effective only if the failure mechanism is associated with time.
Some behaviors of the cost rate function (5.3) for different values of parameter
ȕ, of the Weibull density function f(t) are shown in Fig. 5.1. This parameter is
associated with the intensity at which the failure rate function increases.
2.5
2
Cr(t)
1.5
0.5
0
0.5 2.5 4.5 6.5 8.5 10.5 12.5
t (age)
Fig. 5.1 The cost rate function for ca=10, cb=1, Ș =10 and different values of ȕ: ȕ =1 (____); ȕ=2
(…….); ȕ=3 (_ . _); ȕ=4 (_ Ƒ _); ȕ=5 (_ × _)
In a particular case, when ȕ =1, the Weibull distribution corresponds to an

exponential distribution. In this case, Fig. 5.1 shows that there is not optimum
point for age replacement. In other words, a preventive maintenance plan is not
indicated, and the replacement only should happen when the item fails.
Another possibility is that the cost rate function presents a flat curve, as for
small values of ȕ. Although there are advantages in doing preventive maintenance
at the optimum point, there is not a great difference in terms of cost, versus when
this action is taken at points other than the optimum.
Variations in costs (cb and ca) affect the time (t*) of the minimum cost, as
shown in Fig. 5.2. The greater the ratio ca/cb, the smaller is the time t*. This is
exactly what is necessary to guide the activities of maintenance manager. To avoid
failures, the management guideline should mandate conducting preventive
maintenance actions more often as the cost ratio increases.
2.5
2
Cr(t)
1.5
0.5
0
0.5 2.5 4.5 6.5 8.5 10.5 12.5
t (age)
Fig. 5.2 The cost rate function for cb=1, Ș =10 and different values of ca: ca=1(____); ca=3
(…….);ca=10 (_ . _); ca=50 (_ Ƒ _); ca=100 (_ × _)
It is important to note that in some situations the cost rate does not provide
information about the best time to carry out preventive maintenance because the
different alternatives (t ages) have almost the same evaluation in terms of cost
rate. Therefore, for the purpose of making a decision, this aspect does not help the
DM. It should not be considered, for example, in the curve for ca=3 (…….) when
considering t > 4.5.
Alternatives with almost the same evaluation according to one specific criterion
could have very different evaluations in terms of others. That is why it is essential
to make sure that the DM has as broad a view as possible, to make consistent
decisions.
In the next sub-section, other criteria are introduced into the decision problem.
This includes one step from the MCMD/A framework to build the multicriteria
decision problem to support the selection of the preventive maintenance interval.
5.2.2 MCDM/A Framework for the General Model for Preventive

Maintenance
This sub-section is organized in accordance with the decision structure presented

in Chap. 2. For the sake of clarity some of the steps discussed in detail in that
chapter are omitted or superficially considered in this section.
Identifying Objectives and Criteria
As stated in Chap. 2, this is one of the most important steps, as the objectives
influence every step in the decision process. In the preventive maintenance
context, the cost is only part of the maintenance objective. As discussed in Chap. 3,
the main objectives of the maintenance function are: to extend the useful life of
assets, to ensure satisfactory levels of availability, to ensure operational readiness
of systems, and to safeguard the people who use the facilities. These objectives are
pursued by the maintenance function as a whole.
It is not necessary to emphasize that for service producing systems, the system
availability is even more important. When failures lead to interruptions of these
systems, they are easily perceived by the customer. Thus, an increase in the
availability may increase the level of user satisfaction. The downtime provides an
indirect measure of this objective. The availability is also related to the reliability,
the capability of the system to work without interruption. The behavior of the
reliability function is shown in Fig. 5.3.
1
0.8
0.6
R(t)
0.4
0.2
0
0 2 4 6 8 10 12
t (age)
Fig. 5.3 Reliability function for, Ș =10 and different values of ȕ: ȕ=1 (____); ȕ=2 (…….); ȕ=3
(_ ___); ȕ=4 (_ Ƒ _); ȕ=5 (_ × _)
Sometimes, the reliability is used as a constraint. However, it may be useful to

distinguish the alternatives even beyond the constraint level. The DM’s preference
structure with respect to this aspect should also be considered to be reflected in the
MCDM/A results.
Availability, cost rate, downtime, mean-time between operational failures are
possible criteria related to the decision context in preventive maintenance. In a
recent literature review on MCDM/A models in maintenance, several criteria are
described as having been considered in previous works, including those discussed
above (de Almeida et al. 2015).
Establishing a Set of Actions and a Problematic
As presented in Chap. 2, this step addresses four topics: a) establishing the

structure of the set of alternatives, b) establishing the problematic to be applied to
this set, c) the generation of alternatives; and d) establishing the matrix of
consequences. In preventive maintenance, some of these topics are either not
necessary or straightforwardly defined. In this decision problem, the solution is
related to the time interval for preventive maintenance, and therefore the
problematic is a straightforward choice. The generation of alternatives also need
not be considered. Therefore, only two topics need to be discussed, the structure of
the set of alternatives and the matrix of consequences.
The kind of the set of alternatives may completely change the MCDM/A
methods to be applied. For the selection of the preventive maintenance interval,
the two kinds of sets of alternatives (discrete or continuous) require different
methods. As already mentioned, a set of alternatives consists of the different possible
intervals of time at which the maintenance activities within a maintenance policy
could be performed.
This problem is associated with the classic optimization problem, in which the
set of alternatives is already well defined and consists of a continuous set of time
interval for preventive maintenance t. This time interval t may be seen as days in a
calendar, such that the set of alternatives becomes discrete: A = {d1, d2, d3, ..., dn}.
This model is more realistic because there is no need to use a continuous time t
that includes any time by day or night. Making a choice of day di is a reasonable
approximation for the context of preventive maintenance because a variation of 24
hours does not have a relevant difference in the consequences related to the
decision problem, as shown in Fig. 5.1.
At this stage, with the criteria and the set of alternatives established, the matrix
of consequences can be built, collecting the necessary cost data and other relevant
data associated with the criteria. The construction of this matrix, for this particular
problem, is somewhat straightforward.
Identifying State of Nature
As stated in Chap. 2 the state of nature (T) corresponds to aspects that could not be
controlled by the DM and influence the outcome. In fact, they may change
randomly, and consequently, may deeply influence the consequences of the
decision process. The modeling process for this ingredient uses decision theory,
which includes MAUT.
A typical T is the reliability (de Almeida and Souza, 2001), which influences
the outcomes, such as the availability, a usual criterion in the preventive maintenance
decision problem. That is, the reliability is not a consequence, although it may be
considered as such, as a simplification of the model (Cavalcante and de Almeida
2007; de Almeida 2012).
Similar to the set of alternatives, the set of states of nature may be discrete or
continuous and incorporate prior probabilities S(T) on T.
In the maintenance preventive problem, as the data on failure are scarce, the
probability density function that models the time to failure may be completely
unknown or otherwise have undetermined parameters. Thus, if prior probabilities
S(T) are incorporated, a probabilistic modeling task complements the preference
modeling.
Preference Modeling
This step provides information for choosing the MCDM/A method, aligned with
the DM’s preference structure, which may consider, among other factors,
compensatory or non-compensatory rationality. The main question of this factor
considers which of these classes of methods would be more appropriate for a
particular problem. This process could use the model building procedure in Chap.
2. The analysis of the DM’s rationality is essential to ensure that the results from
the MCDM/A model truly reflect the DM’s preferences.
In the next two sections, applications illustrate the use of compensatory and
non-compensatory methods to support the problem of selecting intervals of
preventive maintenance.
Intra-Criterion Evaluation
For a specific problem, this step consists of the elicitation of the value function
vj(x), or utility function uj(x), related to the values of different performances of
outcomes of criterion j, for any j = 1, 2, ..., n.
For a non-compensatory method, an ordinal scale is enough, so the intra-
criterion evaluation is easily quantified. Furthermore, for an outranking method,
the parameters related to indifference, preference and the discordance threshold
may be addressed.
For the compensatory methods, the usual results consist of an overall value for
each alternative that reflects a synthesis of all the criteria for that alternative. This
overall value arises from the aggregation of the utility functions related to each
criterion uj(x). The assessment of the uj(x) relies on the elicitation procedure. The
utility function reflects the preferential structure from the DM for uncertainty
contexts, considering his behavior with respect to risk. The DM could be risk
neutral, averse and prone. Each of these standard behaviors is reflected by a
specific form of the utility function uj(x).
Inter-Criteria Evaluation
The inter-criteria evaluation is a fundamental step in the MCDM/A problem. The

inter-relationship among criteria is what distinguishes the results from any other
approaches even when multiple aspects are considered, such as availability and
cost. The essence of the MCDM/A approach is how the conflicts between
availability and cost criteria, for instance, are reflected in the preference domain.
The inter-criteria evaluation includes the process of defining the criteria
weights by means of an elicitation procedure. This process and the meaning of the
criteria weights depend on the type of method.
Evaluating Alternatives and Sensitivity Analysis
For the preventive maintenance interval selection, the alternative evaluation

results in the time interval to be applied.
To evaluate how the results provided by the model vary with the parameters
and whether the assumed simplifications affect the results, a sensitivity analysis is
essential.
This step provides further insight to the DM. Some non-obvious behaviors may
be identified during this process providing the DM with the broad view that is
needed for a consistent decision.
Elaborating Recommendation
Given the insights and the view achieved by the application of an MCDM/A
method, a complete report should include the essentials of the whole decision
process, as well as, the main aspects that came up during this process. It should
provide any detail that would be requested during the explanation of the results
and the recommended decisions.
Assumptions, simplifications and changes to the original problem should be
explicit and clear, to aid in transmitting an understanding of the results of the
model and their limitations.
5.3 Compensatory MCDM/A Model for Preventive Maintenance
A compensatory method deals with the DM’s preference structure by means of a

tradeoff amongst criteria, with features that where discussed in Chap. 2. In this
section, a compensatory method is applied to illustrate an MCDM/A model for
selecting preventive maintenance intervals. This model is based on MAUT
(de Almeida and Souza 2001; de Almeida 2012), and illustrates a real study in an
electric power company.
5.3 Compensatory MCDM/A Model for Preventive Maintenance 225
One insight of this model is the analysis of the consequence space for the
preventive maintenance decision problem. When this consequence space cannot
be reduced to only one dimension, the classical optimization approach is not
useful. Additionally, when the consequences are multidimensional, the DM’s
preference for each criterion has to be treated very carefully because any mis-
conception or mistake in this process may waste the effort to bring the DM to the
center of the problem.
5.3.1 The Context, the Set of Alternatives and the Criteria
The context of this problem is an electric power company and it considers the cost
rate and reliability criteria (de Almeida 2012). The underlying model that this
application takes as its base is the model of age based replacement, so all assumptions
and expressions that were presented before are valid in this application.
The set of alternatives corresponds to a discrete set of time intervals. For
instance, months or days may be applied as usual intervals. For a month interval
any alternative is a multiple of 30 days, so that an element of the set of alternatives
could be represented by 30i, where i is any positive integer from 1 to N and N is
the number of alternatives. A quantile of the probability distribution of the time
interval could be used to choose N, although it is not explicit in the model.
It is worth noting that for the age replacement based, whenever a first failure
happens, the timing counting should be restarted. This means that the planning of
the preventive maintenance actions should be performed very carefully, because
the calendar time is not useful to help the manager schedule a particular
preventive maintenance action, as once the schedule is put into action, a failure
can force the schedule to be rearranged. In this way, the use of the base time does
not mean that the action will necessarily happen each 30i days, but rather it means
that 30i is the maximum number of days that a specific item will run until it is
replaced by another. The calendar logic is valid for the block replacement based
policy, in which it is not necessary to keep a register of times to failure.
As already stated the criteria are the cost rate cr(t) and reliability R(t), and their
parameters are presented in Table 5.1.
Table 5.1 Data of the cost and reliability functions
Weibull ȕ 3
Ș 1200
Costs Replacement Cost Cb 600
Failure Cost Ca 1200
It is possible to build a consequence matrix applying the models for cr(t) and
R(t), as shown in Fig. 5.4.
0.8
R(t), cr(t)
0.6
0.4
0.2
0
30 280 530 780 1030
t (age)
Fig. 5.4 The criteria in function of t: R(t) (____) and cr(t) (…….)
5.3.2 Preference Modeling and Intra-Criteria and Inter-Criteria

Evaluations
For the intra-criteria evaluation, a logistic function was found for the reliability
attribute U(R) and an exponential function for the cost U(cr) attribute. The logistic
utility function for reliability shows that the DM considers the variation at R > 0.9
to be small, and views only changes at R < 0.8 to be important. However, the
higher the cost is, the less the utility is, which reveals the risk averse behavior of
the DM.
With regard to the inter-criteria evaluation, the elicitation process (Keeney and
Raiffa 1976) includes the validation of some axioms about the DM’s preferential
structure. The mutual utility independence between the two attributes was
confirmed, and a multilinear utility function is therefore applied, as given by (5.4).
U ( cr , R ) K1U ( cr ) K 2U ( R ) K 3U ( cr )U ( R ) (5.4)
where:
U(cr) is the utility function for the cost rate criterion;
U(R) is the utility function for the reliability criterion;
U(cr,R) is the multiattribute utility function;
K1, K2 and K3 are the scale constants with K1+K2+K3 = 1.
Following the elicitation procedure, the values obtained for these scale
constants were K1 = 0.35, K2 = 0.45, and K3 = 0.20.
5.4 A Non-Compensatory MCDM/A Model for Preventive Maintenance 227
5.3.3 Results and Discussion
The results show the highest value for the overall utility function for t = 600 days,
corresponding to 20 months. Applying the classical optimization model, which
considers only the cost rate criterion, the time of the minimum cost rate (t*) is
t=780, which corresponds to 26 months.
The reliability for the time of the minimum cost rate (t*), R(780) is around 0.75.
Therefore, it may be too risky to follow the policy that minimizes the cost rate,
because the probability of a failure of this item may be considered too high to risk
interrupting the supply of electricity. Consequently, the reliability for this item
should not be neglected.
For a large time interval [540, 660] the overall utility function varies over a
range of less than 0.009. In practical terms, this means that the DM has flexibility
regarding the time interval for preventive maintenance without a considerable
decrease in the overall utility value.
Another interesting insight can be recognized when analyzing the change on
the utility form of the cost criterion. This analysis shows that when a linear
function is used to model the cost rate instead of an exponential function, the
overall utility is affected and has its highest value at t*=360 days. In this case, the
DM is not averse to risk to slightly increase the cost, so the smaller time intervals
that were judged unfavorable when using the exponential functions have improved
results using a linear utility function.
The DM may view one of the characteristics of the compensatory method not
suitable. In this approach, alternatives with very poor performance in some criteria
can compensate by good performance in other criteria, and this could happen in
unlimited way. This feature does not apply to the non-compensatory methods.
However, this would not be a reason to change the approach, which should be
based only on the DM’s compensatory rationality. Another way to address this
issue is using the compensatory method with veto (de Almeida 2013).
5.4 A Non-Compensatory MCDM/A Model for Preventive

Maintenance
Among non-compensatory approaches, the outranking methods are the main

group of methods following this rationality. In these methods, an outranking
relation is built by a pairwise comparison between alternatives, and incomparability
may be considered. The methods of the PROMETHEE family have been applied
in this case (Chareonsuk et al. 1997; Cavalcante and de Almeida 2007; Cavalcante
et al. 2010).
Two applications are presented at this section. There are some similarities to
previous models thus some steps of the modeling process are omitted.
5.4.1 First Application
The criteria considered were cost rate and reliability, as in the previous study. Let
the set of alternatives be A={ti}, where ti= 720 i, for i = 1…12.
The parameters for the criteria are given in Table 5.2.
Table 5.2 Cost and reliability data
Weibull ȕ 1.4
Ș 1800
Costs Replacement Cost Cb 300
Failure Cost Ca 1800
The consequence matrix is given in Table 5.3.
Table 5.3 Consequence matrix for the decision problem
Alternatives T R(t) Cm(t)

T1 720 0.9638 0.2526
T2 1440 0.9072 0.1633
T3 2160 0.8421 0.1401
T4 2880 0.7733 0.1327
T5 3600 0.7038 0.1315
T6 4320 0.6354 0.1333
T7 5760 0.5075 0.1413
T8 7200 0.3957 0.1522
T9 7920 0.3466 0.1582
T10 8640 0.3022 0.1644
The intra-criterion evaluation, for the PROMETHEE method, as described in

Chap. 2, produces the preference functions Pj(a,b), which leads to Ȇ(a,b). Because
in this section S(T) represents the prior probability function, the notation Ȇ(a,b) is
used for the preference index, although in Chap. 2, as in the general literature it is
represented by S(a,b).
Ȇ(a,b) is based on Pj(a,b), as shown in (5.5).
3 ( a , b) N
° ¦ Pj ( a, b) w j ½°
j 1
® N ¾ (5.5)
°3 (b, a )
¯ ¦ j
P (b, a ) w j °
1 j ¿
The weight for a criterion j (wj.) has to be established for each criterion based
on the DM’s preferences.
5.4 A Non-Compensatory MCDM/A Model for Preventive Maintenance 229
The scores of the alternatives are based on the outcome and income flows, as
shown in Chap. 2, and recalled in (5.6) and (5.7).
k k
1 1
I (a ) ¦
n 1 x A
3(a, x ) ¦¦
n 1 x A j 1
Pj ( a , x ) w j ¦I
j 1

j (a ) w j (5.6)
k k
1 1
I (a ) ¦
n 1 x A
3 ( x, a ) ¦¦
n 1 x A j 1
Pj ( x, a ) w j ¦I
j 1

j (a ) w j (5.7)
The weights and intra-criterion parameters are given in Table 5.4.
Table 5.4 Preference function and criteria characteristics
Characteristics R Cm
Max/Min Max Min
Weight 0.34 0.66
Preference function Type V Type V
Indifference threshold 0.001 0.00062
Preference threshold 0.07 0.032
The next step consists of building the outranking relations, as shown in Table 5.5.
Table 5.5 Alternatives flows
Alternatives T I I
T1 720 0.242142 0.660000
T2 1440 0.307792 0.318536
T3 2160 0.444011 0.022754
T4 2880 0.514990 0.037727
T5 3600 0.505728 0.092916
T6 4320 0.459229 0.164855
T7 5760 0.323753 0.248657
T8 7200 0.180039 0.409705
T9 7920 0.111208 0.533040
T10 8640 0.073333 0.674034
The PROMETHEE I method is applied as given by (5.8), in which PI, II, and RI
correspond to preference, indifference and incomparability, respectively.
aS b e aS b
°°
aP I b ®aS b e aI b
°
aI b e aS b
¯°
aI I b aI b e aI b (5.8)
aR I b on the other cases
The best alternatives have been found to be T3 and T4, which correspond to
replacing the components every 2160 or 2880 hours, respectively. These
alternatives are not comparable, indicating that the DM must reflect further when
choosing between them, given that there is not sufficient information or reason,
through the comparison, to have a particular preference for one or the other.
T4 T5 T6 T7 T2 T1 T10
T3 T8 T9
Fig. 5.5 Partial pre-ranking among the alternatives for actions
As explained in Chap. 2, the PROMETHEE II method provides a complete pre-

order, forcing a comparison between T3 and T4. It should be noticed that the
result of using PROMETHEE I is more informative than that of using
PROMETHEE II because the incomparability is known. Fig. 5.6 shows the result
of PROMETHEE II.
T4 T3 T5 T6 T2 T8 T1 T9 T10
Fig. 5.6 Complete pre-ranking among the alternatives for action
5.4.2 Second Application
This model takes different assumptions than the previous application, as following
(Cavalcante et al. 2010):
x The time spent on maintenance actions, whether preventive replacement, or
corrective replacement, is non-negligible and known;
x The distribution of the time to failure is known, but its parameters are not.
References 231
With these two basic changes the expressions for R(t) and cr(t) are different.
Despite this increasing in complexity, this model is realistic because it is common
the absence of time to failure data, which makes the parameters of the distribution
of time unknown.
The alternatives that are considered are times multiples of 100; inside the
interval [200, 3000]. These time intervals in units of days.
The data for this problem are given in Table 5.6.
Table 5.6 Preference function and criteria data
Weibull ʌ(ȕ) ȕ1 =3.4

Ș1=4.5
ʌ(Ș) ȕ2 =2.8
Ș2=2200
Costs Replacement Cost Cb $250
Failure Cost Ca $1000
Times Preventive replacement 0.5 day
Corrective replacement 3 days
The preference function for the criteria are both type V, as presented in Chap. 2,
which corresponds to the case in which the difference in performance increases as
the difference of evaluation in a criterion increase in a linear relationship. In
addition there are two thresholds: the indifference and preference threshold.
Different values of weight are used to give to the DM more information about
the sensitivity of these values. From the variations applied the first solution
indicated by the PROMETHEE II rank changed from 600 days to 800 days. The
sensitivity analysis provides the DM with more information, indicating the level
of variation that is expected (Cavalcante et al. 2010).
References
Barlow R, Hunter L (1960) Optimum Preventive Maintenance Policies. Oper Res 8:90–100
Cavalcante CAV, Almeida AT de (2007) A multi-criteria decision-aiding model using
PROMETHEE III for preventive maintenance planning under uncertain conditions. J Qual
Maint Eng 13:385–397
Cavalcante CAV, Ferreira RJP, de Almeida AT (2010) A preventive maintenance decision
model based on multicriteria method PROMETHEE II integrated with Bayesian approach.
IMA J Manag Math 21:333–348
Chareonsuk C, Nagarur N, Tabucanon MT (1997) A multicriteria approach to the selection of
preventive maintenance intervals. Int J Prod Econ 49:55–64
Cox DR (1962) Renewal theory, vol 4. Methuen & Co, London
de Almeida AT (2012) Multicriteria Model for Selection of Preventive Maintenance Intervals.
Qual Reliab Eng Int 28:585–593
de Almeida AT (2013) Additive-veto models for choice and ranking multicriteria decision
(Maintenance Management: Toward Competitiveness) Editora Universitária da UFPE, Recife
Glasser GJ (1969) Planned replacement- Some theory and its application (Probability theory
applied to age and block replacement models in preventive maintenance of parts, noting
inspection cost distribution). J Qual Technol 1:110–119.
Hwang CL, Tillman FA, Wei WK, Lie CH (1979) Optimal Scheduled-Maintenance Policy
Based on Multiple-Criteria Decision-Making. Reliab IEEE Trans R-28:394–399
Inagaki T, Inoue K, Akashi H (1978) Interactive Optimization of System Reliability Under
Multiple Objectives. Reliab IEEE Trans R-27:264–267
Jiang R, Ji P (2002) Age replacement policy: a multi-attribute value model. Reliab Eng Syst Saf
76(3): 311-318
Scarf PA, Dwight R, Al-Musrati A (2005) On reliability criteria and the implied cost of failure
for a maintained component. Reliab Eng Syst Saf 89:199–207
Shafiee M, Finkelstein M (2015) An optimal age-based group maintenance policy for multi-unit
degrading systems. Reliab Eng Syst Saf 134:230-238
Slack N, Chambers S, Johnston R (2010) Operations management, 6th ed. Pearson Education,
Harlow
Chapter 6
Decision Making in Condition-Based
Maintenance
Abstract: (G:9>8I>K: B6>CI:C6C8: BD9:A>C< >H 6 IDDA I=6I 86C EGDK>9: B6CN
7:C:;>IH ID I=: B6>CI:C6C8: B6C6<:B:CI 6G:6 ,=>H 8=6EI:G HJ<<:HIH 6 BJAI>
8G>I:G>6 6EEGD68= ;DG BD9:A>C< 8DC9>I>DC 76H:9 B6>CI:C6C8: 9:8>H>DCH /=:G:6H
I=: (G:9>8I>K: B6>CI:C6C8: JH:H 8DC9>I>DC >C;DGB6I>DC D; 6HH:IH I=: EG:K:CI>K:
B6>CI:C6C8: >H I>B:76H:9 DCAN ,=>H 8=6EI:G 9>H8JHH:H 9:8>H>DC B6@>C< >C
8DC9>I>DC76H:9 B6>CI:C6C8: % 6C9 EG:H:CIH HDB: JH:;JA 6EEGD68=:H ;DG
7J>A9>C<BJAI>EA:D7?:8I>K:BD9:AH>CI=>H8DCI:MI !C>I>6AAN6HJBB6GN>H<>K:CD;
I=: ;JC96B:CI6AH D; % 6C9 H:K:G6A 8DC8:EIH D; BDC>IDG>C< 6C9 >CHE:8I>DC
68I>K>I>:H 6H>8 8DC8:EIH D; 9:A6N I>B: 6G: EG:H:CI:9 6C9 9>H8JHH:9 L>I=>C 6C
%%
6EEGD68= ,=:C 6 HIGJ8IJG: D; 6 BJAI>8G>I:G>6 BD9:A ID 9:I:GB>C:
>CHE:8I>DC>CI:GK6AHD;8DC9>I>DCBDC>IDG>C<76H:9DC%JAI>6IIG>7JI:JI>A>INI=:DGN
%-, >H >CIGD9J8:9 ,=: 6HE:8IH D; EG:;:G:C8: BD9:A>C< H86A: 8DCHI6CI
:A>8>I6I>DCJI>A>INI=:DGN6C96%PH7:=6K>DGIDG>H@EGDC:C:JIG6A6K:GH:6G:
>C8AJ9:9>CI=:9:8>H>DCBD9:A C>AAJHIG6I>K::M6BEA:D;6C%%
BD9:A>H
<>K:C >C I=: 8DCI:MI D; 6C :A:8IG>8 EDL:G 9>HIG>7JI>DC HNHI:B ,=JH I=:
BJAI>8G>I:G>6 BD9:A EG:H:CI:9 L=>8= >H 76H:9 DC %-, 6C9 L=>8= =6H 6C
6M>DB6I>8HIGJ8IJG:6>BHID6CHL:GI=:C::9H>9:CI>;>:96C9:C67A:HI=:IG69:D;;
6BDC<I=:8DHIH9DLCI>B:6C9;G:FJ:C8ND;7G:6@9DLCHID7:9:6AIL>I=
6.1 Introduction
FJ>EB:CI B6CJ;68IJG:GH D;I:C HI6I: I=6I E:G>D9>8 >CHE:8I>DC 6C9 EG:K:CI>K:

B6>CI:C6C8:68I>K>I>:HBJHI7:9DC:>C688DG96C8:L>I=8:GI6>CG:8DBB:C96I>DCH
HD I=6I I=: L6GG6CIN G:B6>CH K6A>9 6C9 I=: :FJ>EB:CI DE:G6I:H EGDE:GAN
%6>CI:C6C8:I:6BH>CK6G>DJH>C9JHIG>:H=6K:ID;DAADLEG:9:I:GB>C:9H8=:9JA:H
;DGBDHID;I=:>GB6>CI:C6C8:68I>K>I>:H ,=:69K6CI6<:D;I=:H:EGD<G6BH>HI=6I
I=:N 6G: H>BEA: DL:K:G >C;DGB6I>DC DC I=: 8DC9>I>DC D; I=: :FJ>EB:CI >C
DE:G6I>DC >H CDI 8DCH>9:G:9 L>I= G:<6G9 ID BD9>;N>C< I=: B6>CI:C6C8: H8=:9JA:
,=JH HDB: ;68>A>I>:H 6G: CDI B6>CI6>C:9 DEI>B6AAN DCH:FJ:CIAN B6>CI:C6C8:
EDA>8>:HI=6I6G:76H:9DC8DC9>I>DCG6I=:GI=6CI=:6<:D;I=::FJ>EB:CI=6K:7::C
9:K:ADE:9L>I=I=:6>BD;>BEGDK>C<I=::;;>8>:C8ND;B6>CI:C6C8:68I>DCH6C9D;
:MI:C9>C<I=:JH:;JAA>;:D;I=:6HH:IH
© Springer International Publishing Switzerland 2015

=6EI:G :8>H>DC%6@>C<>CDC9>I>DC6H:9%6>CI:C6C8:
&DLA6C 6C9 :6E HI6I:9 I=6I 6C >I:B >H H6>9 ID 7: B6>CI6>C:9 7N
8DC9>I>DC BDC>IDG>C< >; >I >H E:GB>II:9 ID G:B6>C >C H:GK>8: L>I=DJI EG:K:CI>K:
B6>CI:C6C8: JCI>A 6 ;JC8I>DC6A ;6>AJG: D88JGH DC9>I>DC6H:9 %6>CI:C6C8:
%>H8DCH>9:G:967GD69:GHJ7?:8I6C9>C8AJ9:H8DC9>I>DCBDC>IDG>C< %>H
HI6I:97N"6G9>C::I6A 6H7:>C<6B6>CI:C6C8:EGD<G6BI=6IG:8DBB:C9H
B6>CI:C6C8: 68I>DCH 76H:9 DC I=: >C;DGB6I>DC 8DAA:8I:9 JH>C< 8DC9>I>DC
BDC>IDG>C< :H>9:H % 6II:BEIH ID 6KD>9 JCC:8:HH6GN B6>CI:C6C8: I6H@H 7N
I6@>C<B6>CI:C6C8:68I>DCHDCANL=:CI=:G:>H:K>9:C8:D;67CDGB6A7:=6K>DGHD;
6 E=NH>86A 6HH:I ,=>H IDE>8 L6H ;DJC9 >C D; EJ7A>86I>DCH 8DCH>9:G>C<
%%
6EEGD68=:H>C6A>I:G6IJG:G:K>:L9:AB:>96:I6A
/6C< 6C9 6D >C9>86I: I=6I I=:G: >H 6C >C8G:6H:9 C::9 ;DG BDG:
:;;:8I>K: 6C9 :;;>8>:CI I:8=C>FJ:H I=6I BDC>IDG B68=>C: 8DC9>I>DCH >C G:6A I>B:
9:I:8I I=: >C8:EI>DC 6C9 EGD<G:HH>DC D; 9:;:8IH 6C9 :C67A: ;A:M>7A: B6>CI:C6C8:
H8=:9JA>C<7:;DG:69:;:8IG:HJAIH>CJC:ME:8I:9B68=>C:9DLCI>B: "6G9>C::I6A
HI6I:I=6I9>6<CDHI>8H6C9EGD<CDHI>8H6G:ILD>BEDGI6CI6HE:8IH>C6%
EGD<G6B /=>A:9>6<CDHI>8H9:6AL>I=I=:9:I:8I>DC>HDA6I>DC6C9>9:CI>;>86I>DCD;
9:;:8IH L=:C I=:N D88JG EGD<CDHI>8H 9:6A L>I= I=: EG:9>8I>DC D; 9:;:8I 7:;DG: >I
D88JGH
>6<CDHI>8 I:8=C>FJ:H 6G: IDDAH I=6I =6K: 6C >C8G:6H>C< 6EEA>867>A>IN >C
8DBE6C>:H 9J: ID I=: EDI:CI>6A ;>C6C8>6A G:IJGCH :HE:8>6AAN L=:C 8DBE6G:9 L>I=
I=: EDA>8>:H D; 8DGG:8I>K: 6C9 EG:K:CI>K: B6>CI:C6C8: !C8AJ9:9 6BDC< I=: B6>C
I:8=C>FJ:H6G:EGD8:9JG:H;DGB:6HJG>C<DG6C6ANO>C<.>7G6I>DC8DJHI>8:B>HH>DC
'>A 6C6ANH>H -AIG6HDC>8 ,=:GBD<G6E=N ,:BE:G6IJG: +E::9 (:G;DGB6C8:
DGGDH>DC'JIEJIEDL:G(G:HHJG:A:8IG>8JGG:CI J>9:A>C:HDC8DAA:8I>C<6C9
6C6ANO>C<8DC9>I>DCBDC>IDG>C<96I686C7:;DJC9>C!+'
,=:9>6<CDHI>8HEGD7A:B86C7:9:H8G>7:96H6=NEDI=:H>HI:HIEGD7A:BH>C8:
I=: HI6I: D; I=: HNHI:B >H JC@CDLC C :;;>8>:CI 9>6<CDH>H >H DC: I=6I =6H 6
B>C>B6A:GGDGG6I: H>C=NEDI=:H>HI:HI>C<I=:G:B6N7:ILD INE:HD;:GGDGH>C
9>6<CDH>H !I86CB>HI6@:CAN8A6>B:9I=6I6EGD9J8I>H9:;:8I>K:L=:C>C;68I>I>H
CDI6C9I=>HINE:D;:GGDG>H6AHD86AA:96;6AH:C:<6I>K: 'CI=:DI=:G=6C9>I86C
7:H6>9I=6I6EGD9J8I>HCDI9:;:8I>K:L=:C>C;68II=:G:>H67J<6AHD86AA:9;6AH:
EDH>I>K: :GG69::I6A EG:H:CI:9ILDBD9:AH>CDG9:GID9:6AL>I=:GGDG>C
I=:>CHE:8I>DCEGD8:HHHJ8=6H;6AH:EDH>I>K:H;6AH:6A6GBH6C9;6AH:C:<6I>K:HD;
EGDI:8I>DCHNHI:BH
%6GI>C 9>HI>C<J>H=:H 7:IL::C =6G9 6C9 HD;I ;6JAIH ,=:H: ;6JAIH 6G:
H=DLC>C>< ,=:HD;I;6JAIA:69HID6EG:9>8I67A:H>IJ6I>DC>IA:C9H>IH:A;ID
8DC9>I>DC BDC>IDG>C< L=>A: I=: =6G9 ;6JAI I6@:H EA68: >CHI6CI6C:DJHAN : HI6I:H
I=6I EG:9>8I>K: B6>CI:C6C8: >CKDAK:H E:G>D9>8 BDC>IDG>C< DC I=: =:6AI= D; I=:
B68=>C: 6C9 H8=:9JA>C< B6>CI:C6C8: DCAN L=:C 6 ;JC8I>DC6A ;6>AJG: >H 9:I:8I:9
,=>H 6AADLH ;DG IG:C9H D; I=: B68=>C: 8DBEDC:CI ID 7: 8DCHIGJ8I:9 6C9 I>B: ID
;6>AJG:ID7::HI>B6I:9
(GD<CDHI>8I:8=C>FJ:H6>BID:HI>B6I:I=:G:H>9J6AA>;:D;ID6E>:8:D;:FJ>E
B:CII6@>C<8DC9>I>DCBDC>IDG>C<>CID8DCH>9:G6I>DC ,=:8DC8:EID;EGD<CDHI>8H
<D:H7:NDC99>6<CDHI>8H >GHII=:EGD7A:B>H9:I:8I:9I=:C69>6<CDH>H>HB69:
%DC>IDG>C<6C9!CHE:8I>DC8I>K>I>:H
67DJII=:;6>AJG:BD9:6C9>IHH:K:G>IN !I>H6AHD>BEDGI6CIIDEG:9>8II=::KDAJI>DC
D; I=: ;6>AJG: >C DG9:G ID :HI>B6I: I=: G:B6>C>C< JH:;JA A>;: D; I=: B68=>C:
:C6N6:I6A D.6C6C9TG:C<J:G
Hard failure
Good
Condition
Soft failure
Failure
Time t2
t1

Fig. 6.1DC9>I>DC:KDAJI>DC;GDB=6G96C9HD;I;6>AJG:H
,=: 9:K:ADEB:CI D; BDC>IDG>C< 6C9 9>6<CDHI>8 HNHI:BH ;DG 6>G8G6;I 6C9 DI=:G
8DBEA:MHNHI:BH=6HA:9IDI=:G:8D<C>I>DCI=6IEG:9>8I>K:EGD<CDH>H>H9:H>G:96C9
I:8=C>86AAN EDHH>7A: ,=: BDHI JH:9 EGD<CDHI>8H I:8=C>FJ: >H 8DC8:GC:9 L>I=
EG:9>8I>C<=DLBJ8=I>B:>HA:;I7:;DG:6;6>AJG:D88JGH<>K:CI=:8JGG:CIB68=>C:
8DC9>I>DC 6C9 E6HI DE:G6I>DC EGD;>A: JHJ6AAN 86AA:9 >IH G:B6>C>C< JH:;JA A>;:
*-$ "6G9>C: :I 6A 8A6HH>;>:9 G:H:6G8= DC EGD<CDHI>8H >C I=G:: 6G:6H
G:B6>C>C< JH:;JA A>;: EGD<CDHI>8H >C8DGEDG6I>C< B6>CI:C6C8: EDA>8>:H 6C9
8DC9>I>DCBDC>IDG>C<>CI:GK6A
88DG9>C< ID /6C< 8JGG:CI EGD<CDHI>8 6EEGD68=:H 86C 7: 8A6HH>;>:9
>CID I=G:: 76H>8 <GDJEH 6 BD9:A76H:9 6EEGD68= 6 96I69G>K:C 6EEGD68= 6C9 6
=N7G>9 6EEGD68= *:H>9J6A A>;: BD9:AA>C< K>6 HID8=6HI>8 ;>AI:G>C< >H 6 G:A:K6CI
BD9:AA>C<I:8=C>FJ:JH:9>C%/6C<6C9=G>HI:G "6G9>C::I6A
EGDEDH:9 EGDEDGI>DC6A =6O6G9H BD9:AA>C< ( % >C DG9:G ID >C8DGEDG6I>C<
:MEA6C6IDGN K6G>67A:H >CID 6 BD9:A ;DG :HI>B6I>C< I=: ;6>AJG: G6I: .AD@ :I 6A
JH:9 6 /:>7JAA EGDEDGI>DC6A=6O6G9H BD9:A ID 9:I:GB>C: I=: DEI>B6A
G:EA68:B:CI ;DG 6C >I:B L=>8= >H HJ7?:8I ID K>7G6I>DC BDC>IDG>C< /6C<
DJIA>C:H6H:B>HID8=6HI>8;>AI:G>C<76H:9G:H>9J6AA>;:EG:9>8I>DC6EEGD68=;DGI=:
>I:BHBDC>IDG:9>C% !C<:C:G6AEGD<CDHI>8H>C;DGB6I>DC>H6@:N:A:B:CI>C
BD9:A>C<I=:9:8>H>DCB6@>C<6HE:8ID;%
6.2 Monitoring and Inspection Activities
6GADL6C9(GD8=6CBD9:A:9>CHE:8I>DCEDA>8>:HL=>8=6HHJB:I=6I;6>AJG:
>H9>H8DK:G:9DCAN7N68IJ6A>CHE:8I>DC6C9>C<:C:G6ADCAN6;I:GHDB:I>B:=6H
:A6EH:9H>C8:I=:D88JGG:C8:D;I=:;6>AJG:6C9:K6AJ6I:9H8=:9JA:HD;>CHE:8I>DC
I>B:HL=>8=B>C>B>O:I=:IDI6A:ME:8I:98DHIG:HJAI>C<;GDB7DI=>CHE:8I>DC6C9
;6>AJG: ,=:N6HHJB:I=6I;6>AJG::KDAK:;GDB69:I:G>DG6I>DCEGD8:HH ,=>HEGD8:HH
>H 6HHJB:9 HID8=6HI>8 6C9 I=: 8DC9>I>DC D; I=: HNHI:B >H @CDLC DCAN I=GDJ<=
>CHE:8I>DC
!C >CHE:8I>DC EDA>8>:H CD G:EA68:B:CI DG G:E6>G >H G:8DBB:C9:9 7:;DG:
9:I:8I>DCD;;6>AJG: 68=>CHE:8I>DC=6H68DHII=6I>BEA>:H7:>C<JC;:6H>7A:>CHE:8I
K:GND;I:C DL:K:G6ADC<A6EH:D;I>B:7:IL::C;6>AJG:6C99:I:8I>DC>BEA>:H6
E:C6AIN8DHI ,=:B6>C8=6AA:C<:>HID;>C9I=:7:HI>CHE:8I>DCEDA>8N>CDG9:GID
B>C>B>O::ME:8I:9IDI6A8DHI DG:M6BEA:I=:G:6G:HNHI:BHHJ7?:8IIDEG:8:9:C8:
8DCHIG6>CIH6C96H:FJ:C8:D;>CHE:8I>DCHH=DJA97:9:I:GB>C:9=>J:I6A
6GADL 6C9 (GD8=6C EGDEDH:9 6 BD9:A I=6I B>C>B>O:H :ME:8I:9 8DHI
JCI>A9:I:8I>DCD;;6>AJG:6BD9:AI=6IB>C>B>O:H:ME:8I:98DHI6HHJB>C<G:C:L6A
6I9:I:8I>DCD;;6>AJG:6C96CDEEDGIJC>HI>8G:EA68:B:CIHIG6I:<ND;6H>C<A:E6GI>C
I=:EG:H:C8:D;H:K:G6ABDC>IDG:9E6GIH
:C6N6 :I 6A ED>CI:9 DJI I=6I ;DG HDB: HNHI:BH 6 8DCI>CJDJH
BDC>IDG>C<D;I=:>GDE:G6I>C<HI6I:H>HCDI:8DCDB>86AAN?JHI>;>67A:6C9>CHE:8I>DCH
6G: JH:;JA >C BDC>IDG>C< I=: 8DC9>I>DC D; I=: HNHI:B 6I EG:9:I:GB>C:9 I>B:H >C
DG9:GIDG:9J8:I=:EGD767>A>IND;>IHB6A;JC8I>DC>C< 'C8:6CDJID;8DCIGDAHI6I:
>H9:I:8I:96G:E6>G>H86GG>:9DJIIDG:HIDG:I=:HNHI:BID>IH>C8DCIGDAHI6I:
,=:G: 6G: HNHI:BH I=6I HNBEIDBH D; ;6>AJG: 6G: CDI 6EE6G:CI 6C9 I=: A:K:A D;
9:<G696I>DCDG9:I:G>DG6I>DC86C7:@CDLCDCANI=GDJ<=>CHE:8I>DCDJA69>G69
6C9G6AA =:A7>6C9>I#69> G6AA:I6A JNC=:I6A
+DB: :M6BEA:H 6G: 6A6GB 6C9 HI6C97N HNHI:BH C >CHE:8I>DC HIG6I:<N :HI67A>H=:H
I=:I>B:6IL=>8=DC:DGBDG:DE:G6I>C<E6G6B:I:GH=6K:ID7:8DCIGDAA:9>CDG9:G
ID9:I:GB>C:>;I=:HNHI:B>H>C6CDE:G6I>C<DG6;6>AJG:HI6I:=:A7>6C9>I#69>

,LD<:C:G6AH>IJ6I>DCHL:G:>9:CI>;>:97N=:A7>6C9>I#69> >GHIAN
>CHE:8I>DCH8DCH>HIH>BEAN>C6HH:HH>C<>;I=::FJ>EB:CI>HLDG@>C<DG>C6;6>A:9
HI6I: +:8DC9ANI=::FJ>EB:CI8DC9>I>DC86C7:6HH:HH:9I=GDJ<=9>G:8IDG>C9>G:8I
8DCIGDA6C9EG:K:CI>K:68I>DCH86C7:9DC:7:;DG:;6>AJG:D88JGG:C8:>I>H@CDLC
6H%
76H>8 >CHE:8I>DC BD9:A HJ<<:HI:9 7N =:A7> 6C9 >I#69> 6HHJB:9
I=6I:FJ>EB:CI>CHE:8I:96I>CHI6CIHxi. !;6C>CHE:8I>DCG:K:6AHI=6II=::FJ>EB:CI
>H>C6;6>A:9HI6I:6C:L>9:CI>86ADC:>BB:9>6I:ANG:EA68:H>I ,=:H:FJ:C8:D;
>CHE:8I>DC>CHI6CIH>HH=DLC>C><
Failure
0 x1 x2 … xi-1 xi xi+1
Inspection Times

Fig. 6.2,=:H:FJ:C8:D;>CHE:8I>DCI>B:H
:A6N,>B:%D9:AHID+JEEDGI%
6.3 Delay Time Models to Support CBM
=G>HI:G 6C9 /6AA:G >CIGD9J8:9 6C >CHE:8I>DC BD9:A 76H:9 DC ILD HI6<:
;6>AJG: EGD8:HH 86AA:9 :A6N ,>B: , ,=:N 86AA:9 I=: >C>I>6A ED>CI u D; I=:
9:;:8I I=: ;>GHI DEEDGIJC>IN L=:G: I=: EG:H:C8: D; 6 9:;:8I B><=I G:6HDC67AN 7:
:ME:8I:9ID7:G:8D<C>O:97N6C>CHE:8I>DC6C9I=:I>B:hID;6>AJG:;GDBu86AA:9
I=: 9:A6N I>B: D; I=: 9:;:8I 6H H=DLC >C >< :A6N I>B: G:EG:H:CIH 6 I>B:
L>C9DL;DGEG:K:CI>C<6;6>AJG:6;I:G69:;:8ID88JGG:9/6C<

h
u failure

Fig. 6.3:A6N,>B:hD;6;6>AJG:
!C,BD9:AHI=:;6>AJG:EGD8:HH>H6HHJB:9ID7:6CDC=DBD<:C:DJH(D>HHDC
EGD8:HH ,=: DE:G6I>DC6A 8DHI D; 6EEAN>C< 6C >CHE:8I>DC EDA>8N 86C 7: B:6HJG:9
,=:>CHE:8I>DC8DHI9:CDI:97NCiG:EG:H:CIHI=:K6AJ:D;I=:G:HDJG8:HC::9:9ID
E:G;DGB6C>CHE:8I>DCI6H@ ,=:>CHE:8I>DCG:E6>G8DHI9:CDI:97NCr8DCH>HIHD;
I=: 8DHIH C:8:HH6G>AN >C8JGG:9 ID G:E6>G 6 ;6JAI >9:CI>;>:9 >C I=: >CHE:8I>DC
6H>86AANI=:8DHID;67G:6@9DLC>H6E:C6AIN8DHI6C9>H6HHD8>6I:9L>I=I=:8DHI
D; I=: 8DCH:FJ:C8:H 86JH:9 7N 6 ;6>AJG: L=>8= >H 6I 7DIIDB I=: 8DHI G:A6I:9 ID
ADHH D; EGD9J8I>DC ,=: 7G:6@9DLC G:E6>G 8DHI >H 9:CDI:9 7N Cb 6H:9 DC I=:
9:A6N I>B: 8DC8:EI 6C9 69B>II>C< I=6I I=: BDHI >BEDGI6CI 6HHJBEI>DC I=6I >H
7GDJ<=I;GDBI=>H;JC96B:CI6A6EEGD68=>H8DHI688DG9>C<ID=G>HI:G6C9/6AA:G
I=::ME:8I:98DHID;6C>CHE:8I>DCEDA>8N;DG676H>8>CHE:8I>DCBD9:A>H
<>K:C>C
OT ^CbbT Cr > bT @` Ci

C T
T d
L=:G:
TQI>B:7:IL::C>CHE:8I>DCH
f(h) Q I=:EGD767>A>IN9:CH>IN;JC8I>DCD;9:A6NI>B:
O>HI=:6GG>K6AG6I:D;9:;:8IHE:GJC>II>B:
CbQI=:6K:G6<:7G:6@9DLCG:E6>G8DHI
CrQI=:6K:G6<:G:E6>G8DHI
CiQI=:6K:G6<:>CHE:8I>DC8DHI
dbQI=:6K:G6<:9DLCI>B:IDG:E6>G67G:6@9DLC
dQI=::ME:8I:99JG6I>DCD;I=:>CHE:8I>DCI>B:9,
b(T)QI=:EGD767>A>IND;6;6JAI86JH>C<67G:6@9DLC
,=: >C>I>6A >CHI6CI 6I L=>8= 6 9:;:8I B6N 7: 6HHJB:9 ID ;>GHI 6G>H: L>I=>C I=:
EA6CI>HJC>;DGBAN9>HIG>7JI:9DK:GI>B:H>C8:I=:A6HI>CHE:8I>DC6C9>C9:E:C9:CI
D;h6H<>K:C7N
T
§T h·
bT ³ ¨© ¸ f h dh
T ¹

!C I:GBH D; 6K6>A67>A>IN I=>H 8G>I:G>DC >H G:A6I:9 ID I=: CDCBDC:I6GN 6HE:8IH
6HHD8>6I:9 L>I= I=: ;6>AJG:H ,=>H G:;A:8IH I=: 67>A>IN D; I=: HNHI:B ID E:G;DGB
JC9:GI=:>C;AJ:C8:D;6C>CHE:8I>DCEDA>8N K6>A67>A>IN>H67DJII=:E:G8:CI6<:D;
I>B:I=6II=:HNHI:B>H6K6>A67A: +D>CI:GBHD;6H:GK>8:HNHI:BI=>H6K6>A67>A>IN
8DGG:HEDC9H ID I=: E:G8:CI6<: D; I>B: I=6I I=: H:GK>8: >H EGDK>9:9 ID I=: 8A>:CI
,=JHI=:ADL:GI=:6K6>A67>A>IN>HI=:<G:6I:GI=:8A>:CIPH9>HH6I>H;68I>DC
99>I>DC6AAN 6 8G>I>86A ;68IDG ;DG 6HH:HH>C< I=: E:G;DGB6C8: D; 6C >CHE:8I>DC
EDA>8N>HHNHI:B6K6>A67>A>IN ,=:9DLCI>B:86C7::K6AJ6I:9>CDG9:GIDG:EG:H:CI
I=>H;68IDG 88DG9>C<ID=G>HI:G6C9/6AA:GI=::ME:8I:99DLCI>B:;DG6
76H>8>CHE:8I>DCBD9:A>H<>K:C7N
O T d b bT d
DT
T d
6.4 Multicriteria and Multiobjective Models in CBM
>6<CDHI>8 I:8=C>FJ:H =6K: 7::C 9:K:ADE:9 L=>8= 6G: 67A: ID >9:CI>;N I=: BDHI
>C8>E>:CI9:;:8IH6C9EGD<CDHI>8I:8=C>FJ:H6G:67A:ID:HI>B6I:G:H>9J6AA>;:BDG:
688JG6I:AN I=:H: ILD 6G:6H D; % =6K: 7::C 76H:9 DC B6I=:B6I>86A 6C9
HI6I>HI>86A :HI>B6I>DC I:8=C>FJ:H "6G9>C: :I 6A EG:H:CI:9 96I6 68FJ>H>I>DC
96I6 EGD8:HH>C< 6C9 B6>CI:C6C8: 9:8>H>DCB6@>C< 6H I=G:: @:N HI:EH D; 6 %
EGD<G6B +JEEDH: 6 H:I D; :FJ>EB:CI D; I=: H6B: INE: =6K: I=: H6B: G:H>9J6A
JH:;JA A>;: :E:C9>C< DC I=: >BE68I D; I=: ;6>AJG: 6 E>:8: D; :FJ>EB:CI 86C 7:
>CHE:8I:9 BDG: ;G:FJ:CIAN DG EG:K:CI>K: B6>CI:C6C8: 68I>K>IN 86C 7: 6CI>8>E6I:9
9J:IDI=:EGD767>A>IND;;6>AJG: ,=:%PHEG:;:G:C8:H;DGI=:E:G;DGB6C8:D;6
B6>CI:C6C8:EDA>8N86C7:BD9:A:99JG>C<I=:9:8>H>DCEGD8:HH /=:CBDG:I=6C
DC: D7?:8I>K: >H 8DCH>9:G:9 6 %%
BD9:A 86C 7: 9:K:ADE:9 ID HJEEDGI 6
% EGD<G6B +DB: :M6BEA:H D; %%
BD9:AH >C % EGD<G6BH 6G:
9>H8JHH:9
6GC:GD EGDEDH:9 6C :K6AJ6I>DC HNHI:B D; H:II>C< JE 6 EG:9>8I>K:
B6>CI:C6C8: EGD<G6B JH>C< I=: C6ANI>8 >:G6G8=N (GD8:HH ( 6N:H>6C
%%
%D9:ADCDC9>I>DC%DC>IDG>C<
I:8=C>FJ:H 6C9 9:8>H>DC GJA:H +6HB6A 6C9 *6B6C?6C:NJAJ 6C6ANO: 6

B:I=D9DAD<N ;DG 6HH:HH>C< I=: 8DC9>I>DC D; 7G>9<:H JH>C< I=: ( EGD8:HH >C 6
;JOON :CK>GDCB:CI .>HJ6A <:C:G6A 6C9 9:I6>A:9 6HH:HHB:CIH L:G: I=: 8G>I:G>6
8DCH>9:G:9 ,6C6@6 :I 6A EG:H:CI 6 EGD8:9JG: ;DG 6HH:HH>C< I=: =:6AI= D;
:FJ>EB:CI ;DG HJ7HI6I>DC B6>CI:C6C8: 6C9 ;DG EA6CC>C< JE<G69:H 76H:9 DC I=:
(6C9I=:NHJEEANG:A>67>A>IN=6G9L6G:>CI:<G>IN6C9G:<JA6I>DC6H8G>I:G>6;DG
I=:BD9:A !CI:GBHD;>CHE:8I>DCEA6CC>C<>CI=::A:8IG>8EDL:G>C9JHIGN6;JOON
BD9:AL6H9:;>C:97N+:G<6@>6C9#6A6>IO6@>H ;DGG6C@>C<I=:8G>I>86A>IND;
8DBEDC:CIH 6C9 I=:N >C8DGEDG6I:9 8G>I:G>6 8DC8:GC>C< 6HE:8IH D; H6;:IN 6C9
G:A>67>A>IN :8DCDBN K6G>67A: DE:G6I>DC6A 8DC9>I>DCH 6C9 :CK>GDCB:CI6A >BE68IH
:GG:>G6:I6A EGDEDH:96BD9:A;DGBJAI>D7?:8I>K:DEI>B>O6I>DC76H:9DC
I=:9:A6NI>B:8DC8:EI>CL=>8=8DHI6C99DLCI>B:6G:D7?:8I>K:;JC8I>DCHD;I=:
BD9:A
#>B 6C9 G6C<DEDA 9:K:ADE:9 6 BJAI>D7?:8I>K: BD9:A L>I= ILD
D7?:8I>K:H BDC>IDG>C< 8DHI 6C9 6K6>A67>A>IN ;GDB L=>8= (6G:ID HDAJI>DCH
6HHD8>6I:9 L>I= I=: 9JG6I>DC D; BDC>IDG>C< 6C9 D; EG:9>8I>DCH 6G: D7I6>C:9 $>J
6C9G6C<DEDA JH:96BJAI>D7?:8I>K:<:C:I>86A<DG>I=B>CDG9:GID76A6C8:
I=: D7?:8I>K:H D; I=: B6>CI:C6C8: 8DHIH D; I=: A>;:8N8A: L>I= I=: 8DC9>I>DC 6C9
H6;:IN A:K:AH D; 9:I:G>DG6I>C< 7G>9<:H %6GH:<J:GG6 :I 6A EG:H:CI 6 BJAI>
D7?:8I>K: DEI>B>O6I>DC 6EEGD68= 76H:9 DC <:C:I>8 6A<DG>I=BH ,=:N 8DCH>9:G:9
I=: EGD767>A>IN D; HNHI:B ;6>AJG: 6C9 >IH K6G>6C8: 6H D7?:8I>K:H %6GIDG:AA :I 6A
9:BDCHIG6I:69DJ7A:ADDE%JAI>EA:D7?:8I>K:<:C:I>86A<DG>I=BIDE:G;DGB
I=: H>BJAI6C:DJH DEI>B>O6I>DC D; E:G>D9>8 ,:HI !CI:GK6AH ,! 6C9 ,:HI (A6CC>C<
,(>CDEI>B>O>C<HJGK:>AA6C8:G:FJ>G:B:CIHL=>8==6K:I=:B:6CJC6K6>A67>A>IN
I=: B6M>BJB I>B:9:E:C9:CI JC6K6>A67>A>IN 6C9 I=: 8DHI D; I=: HNHI:B 8DHI 6H
D7?:8I>K:;JC8I>DCH
(D9D;>AA>C>:I6A EGDK>9:6BJAI>D7?:8I>K:<:C:I>86A<DG>I=BIDDEI>B>O:
>CHE:8I>DC 6C9 B6>CI:C6C8: EGD8:9JG:H L>I= G:HE:8I ID 7DI= I=: :8DCDB>8 6C9
H6;:ING:A6I:9 6HE:8IH D; G6>AL6N IG68@H :E:C9:CI EGD767>A>IN D; ;6>AJG: DC
9:B6C9HEJG>DJHIG>EG6I:6C9A>;:8N8A:8DHIL:G:I=:D7?:8I>K:;JC8I>DCHJH:97N
,DGG:H8=:K:GG>6:I6A >C6CDEI>B>O6I>DCBD9:A;DGEGDD;I:HI>C<EDA>8>:H
;DG H6;:IN >CHIGJB:CI:9 HNHI:BH ,=:>G BD9:A L6H >CI:<G6I:9 L>I= I=: &+!!
<:C:I>8 6A<DG>I=B 2>D 6C9 .>696C6 9:K:ADE:9 6 %JAI>D7?:8I>K: >;;:G:CI>6A
KDAJI>DC HD 6H ID DEI>B>O: I=: >CHE:8I>DC >CI:GK6AH D; 6 ><= (G:HHJG: !C?:8I>DC
+NHI:B -C6K6>A67>A>IN 8DHI :MEDHJG: I>B: L:G: I=: D7?:8I>K:H D; I=: BD9:A
EG:H:CI:9
6.5 A MCDM/A Model on Condition Monitoring
,=:BDHI8DBBDC6HHJBEI>DC>HI=6I6CN;6>AJG:>H9:I:8I:96II=:I>B:D;I=:C:MI
8=:8@ 6C9 6 G:EA68:B:CI >H >BB:9>6I:AN B69: 6GADL 6C9 (GD8=6C
&6@6<6L6 9:G>K6I>K: 6EEGD68= 6C6ANO:H I=: 9:A6N I>B: :A6N I>B: 6
ILDHI:E;6>AJG:EGD8:HH>HI=:I>B:A6EH:;GDBL=:C6HNHI:B9:;:8I8DJA9;>GHI
=6K:7::CCDI>8:9JCI>AI=:I>B:L=:C>IHG:E6>G86CCDADC<:G7:9:A6N:97:86JH:
D;JC688:EI67A:8DCH:FJ:C8:HHJ8=6H6H:G>DJH86I6HIGDE=:L=>8=B><=I6G>H:9J:
ID;6>AJG:=G>HI:G ,=:>BEDGI6C8:D;9:A6NI>B:>CB6>CI:C6C8:B6C6<:
B:CI 6EEA>86I>DCH L6H >CK:HI><6I:9 7N /6C< :A6N ,>B: BD9:AH =6K:
7::C6EEA>:9>CH:K:G6A8DCI:MIH !C6B6CJ;68IJG>C<>C9JHIGN"DC:H:I6A
:K6AJ6I: 6 HJ7?:8I>K: B:6HJG: D; I=: ;6>AJG: 8DCH:FJ:C8:H 76H:9 DC 9:A6N I>B:
6C6ANH>H >C I:GBH D; 8DHI ID I=: :CK>GDCB:CI >C BDC:I6GN K6AJ: ID I=: 8DBE6CN
6C9 I=: 96B6<>C< :;;:8I ID I=: 8DBE6CN >B6<: :A6N ,>B: BD9:A >H 6AHD
EGDEDH:9ID9:I:GB>C:>CHE:8I>DC>CI:GK6AHDC;>H=>C<K:HH:AH(>AA6N:I6A
!CHE:8I>DC I6H@H 6G: 67A: ID >9:CI>;N >CI:GB:9>6I: HI6I:H 7:;DG: ;6>AJG:
%%
9:8>H>DC BD9:A >C DG9:G ID 6>9 B6>CI:C6C8: EA6CC>C< >C >CHE:8I>DC
BD9:AHL6H9:K:ADE:976H:9DC%-,7N:GG:>G6:I6A ,=>HBD9:AI6@:H
I=:%PHEG:;:G:C8:H>CID688DJCI6HL:AA6HI=:BDHI>BEDGI6CI6HE:8IHL=:C
8DCH>9:G>C<H:II>C<I=:>CHE:8I>DC>CI:GK6AH;DGE:G>D9>88DC9>I>DCBDC>IDG>C<6C9
I=:H:6G:I=:8DHI6C99DLCI>B:6HHD8>6I:9L>I=I=:>CHE:8I>DCEDA>8N
!CHE:8I>DC 86C 7: 9:;>C:9 6H 6 I6H@ D; :M6B>C>C< 6C9 D7H:GK>C< >C DG9:G ID
8A6HH>;N6C>CHE:8I:9>I:B>CI:GBHD;>IH;:6IJG:H6C9EGDE:GI>:H /=:C>CHE:8I>DCH
6G: L:AA 9:;>C:9 6 8DBE6CN 86C B>C>B>O: B6>CI:C6C8: 8DHIH 6C9 >BEGDK: I=:
6K6>A67>A>IN D; HNHI:BH !C <:C:G6A I=: >CI:G:HI >H >C 9>H8DK:G>C< I=: HI6I: D; 6C
6HH:I> : L=:I=:G>I>H9:;:8I>K:DGCDI
%6C6<:GH 6G: >CI:G:HI:9 >C 76A6C8>C< I=: 8DHIH D; I=: >CHE:8I>DC EDA>8N L>I=
H6K>C<H6G>H>C<;GDB>BEGDK>C<I=:E:G;DGB6C8:D;I=:HNHI:B !C;68I9:E:C9>C<
DC I=: EGD9J8I>DC EGD8:HH I=6I >H HJEEDGI:9 7N 6 8DBEA:M HNHI:B 6 K:GN H=DGI
>CI:GGJEI>DC9J:ID6;6>AJG:86C86JH:6HJ7HI6CI>6A;>C6C8>6AADHH
+>C8:I=:BD9:A>H76H:9DCI=:8DC8:EID;9:A6NI>B:I=:8DCHIGJ8I>DCD;I=:
BD9:AH=DJA98DCH>9:GI=:8DBBDC6HHJBEI>DCH9:;>C:9>CI=:9:A6NI>B:6EEGD68=
%-, BD9:A >H EGDEDH:9 L=>8= 8DCH>9:GH I=6I 6IIG>7JI:H D; 8DHI 6C9
6K6>A67>A>IN6G:699>I>K:>C9:E:C9:CI>;6C9DCAN>;I=:ILD6IIG>7JI:JI>A>IN;JC8I>DC
>H699>I>K: DGI=:H:8G>I:G>6I=:699>I>K:;DGBB6N7:LG>II:C6H
B6M u C T D T k c u c C T k d u d DT
L=:G:
kcQH86A:8DCHI6CI;DGI=:8DHI8G>I:G>DC
uc(C(T))Q8DC9>I>DC6AJI>A>IN;JC8I>DC;DGI=:8DHI8G>I:G>DC
kdQH86A:8DCHI6CI;DGI=:9DLCI>B:8G>I:G>DC
ud(D(T))Q8DC9>I>DC6AJI>A>IN;JC8I>DC;DGI=:9DLCI>B:8G>I:G>DC
88DG9>C< ID #::C:N 6C9 *6>;;6 I=: 6HH:HHB:CI EGD8:HH ;DG 6 %-
JC8I>DC8DCH>HIH76H>86AAND;;>K:HI:EH>CIGD9J8>C<I=:I:GB>CDAD<N6C9>9:6
>9:CI>;N>C<G:A:K6CI>C9:E:C9:C8:6HHJBEI>DCH6HH:HH>C<8DC9>I>DC6AJI>A>IN
;JC8I>DCH 6HH:HH>C< H86A>C< 8DCHI6CIH 8=:8@>C< ;DG 8DCH>HI:C8N 6C9
G:>I:G6I>C<
%%
%D9:ADCDC9>I>DC%DC>IDG>C<
,=:;>GHIHI:E8DCH>HIHD;B6@>C<I=:%JC9:GHI6C9I=:B6>CEJGEDH:D;I=:
JI>A>IN;JC8I>DC6C9:HE:8>6AANHDI=6I=:
H=:JC9:GHI6C9HI=:8DCH:FJ:C8:HE68:
DG DJG E6GI>8JA6G 86H: >I >H >BEDGI6CI ID CDI>8: I=6I t I>B: ID >CHE:8I>DC >H
8DCH>9:G:9 6 ;:6H>7A: 6AI:GC6I>K: 6C9 T >H I=: H:I D; 6AA I>B:D;>CHE:8I>DC
6AI:GC6I>K:H !CI=>H86H:I=:H:ID;6AAI>B:HD;>CHE:8I>DC6AI:GC6I>K:H>H9:;>C:9
7NT3 DG:68=>CHE:8I>DCI>B:tI=:G:>H68DCH:FJ:C8:>CI:GBHD;8DHI
C(t) 6C9 9DLCI>B: D(t) DG :M6BEA: I=: ED>CI C(t1) D(t1) 7:ADC<H ID I=:
8DCH:FJ:C8: HE68: C :M6BEA: D; I=: 8DCH:FJ:C8: HE68: D; I=>H EGD7A:B >H
H=DLC>C><
D(T)
D0(T)
D*(T)
C*(T) C0(T) C(T)

Fig. 6.4DCH:FJ:C8:HE68:;DG8DHI6C99DLCI>B:
,=: H:8DC9 HI:E 8DCH>HIH D; >9:CI>;N>C< HDB: >C9:E:C9:C8: 6HHJBEI>DCH I=6I
6G: IGJ: ;DG I=: % L>I= G:<6G9 ID I=: 8G>I:G>6 I=6I 6G: 7:>C< 8DCH>9:G:9 -I>A>IN
>C9:E:C9:C8: C::9H ID 7: 8=:8@:9 >C DG9:G ID :K6AJ6I: I=: =NEDI=:H>H D; I=:
I=:DGN !; I=>H >C9:E:C9:C8: D88JGH I=:C I=: %- ;JC8I>DC 86C 7: H>BEA:
DI=:GL>H:BDG:8DBEA:M;JC8I>DCH6G:C:8:HH6GNIDG:EG:H:CII=>H;JC8I>DC
'C8: 699>I>K: >C9:E:C9:C8: >H D7H:GK:9 I=: HIG6I:<N D; 9>K>9: 6C9 8DCFJ:G
8DJA97:I=DGDJ<=AN:MEADG:9L=:C6HH:HH>C<I=:%-;JC8I>DC ,=:G:;DG::68=
DC:9>B:CH>DC6A JI>A>IN ;JC8I>DC ;DG :68= G:HE:8I>K: 6IIG>7JI: H=DJA9 7: :A>8>I:9
AI:GC6I>K:AN>CHDB:86H:H6HE:8>;>86C6ANI>8;JC8I>DC8DJA97:JH:9L=:G:>IH
H=6E:<>K:H6C>CI:G:HI>C<9:H8G>EI>DCD;6HE:8>;>8>CHI6C8:D;I=:%PH7:=6K>DG
;DG6<>K:C6IIG>7JI:
,=: I=>G9 HI:E 8DCH>HIH 76H>86AAN D; :A>8>I>C< 8DC9>I>DC6A JI>A>IN ;JC8I>DCH ;DG
:68= 8G>I:G>DC ,=: 8DC9>I>DC6A JI>A>IN ;JC8I>DCH 86C 7: E:G;DGB:9 76H>86AAN 7N
9>G:8I6HH:HHB:CIDG:HI>B6I>DCD;I=:JI>A>IN;JC8I>DC !CHDB:86H:HI=:G:>H6C
6C6ANI>86A:MEG:HH>DCL=>8=>H@CDLCID7:6<DD9BD9:A;DGG:EG:H:CI>C<JI>A>IN
;JC8I>DCHD;HDB:HE:8>;>88G>I:G>6
H ID I=: ;DJGI= HI:E 6HH:HH>C< I=: H86A>C< 8DCHI6CIH 9:E:C9H B6>CAN DC I=:
H:8DC9HI:E !;699>I>K:>C9:E:C9:C8:>H8DC;>GB:9L:86CJH:I=:ADII:GNH=DLC
>C >< ID >9:CI>;N I=: K6AJ: D; I=: 8DCHI6CI >C6AAN I=: %- ;JC8I>DC
u(C(T),D(T))H=DJA97:B6M>B>O:9>CDG9:GID;>C9I=:7:HIDEI>DC;DG>CHE:8I>DC
I>B:>CI:GBHD;8DHI6C99DLCI>B:
(C*(T), D*(T))
p
(C*(T), D0(T)) ~
1-p (C0(T), D0(T))

Fig. 6.5$DII:GNID;>C9I=:H86A>C<8DCHI6CIkc
!C >< I=: %-, 6EEA>86I>DC E=6H: >H HJBB6G>O:9 H=DL>C< E6GI D; I=:
BD9:A>C< EGD8:9JG: L=>8= >H 76H:9 DC I=: <:C:G6A EGD8:9JG: EGDEDH:9 ;DG
7J>A9>C<I=:%%
BD9:A6H<>K:C>C=6E E:8JA>6G>IN>CI=>HHE:8>;>8
EGD8:9JG:>HI=:JH:D;I=:(6G:ID;GDCI>9:CI>;>86I>DC;DGI=:<:C:G6I>DCD;I=:H:I
D;6AI:GC6I>K:H
DM identification and
parameters estimation for C(T) and D(T)
Pareto front identification and

assessment of utility and additive independences
Conditional utility functions assessment

of uc(C(T)) and ud(D(T))
Scaling constants assessment
Maximization of the
Multiattribute utility function u(C(T),D(T))

Fig. 6.6+IGJ8IJG:D;I=:9:8>H>DCBD9:A;DG>CHE:8I>DC>CI:GK6AHD;8DC9>I>DCBDC>IDG>C<
6.6 Building an MCDM/A Model on Condition Monitoring for a

Power Distribution Company
%6>CI:C6C8: B6C6<:B:CI >H 6 7JH>C:HH ;JC8I>DC I=6I 6>BH ID :CHJG: I=:
6K6>A67>A>IND;EGD9J8I>DCG:HDJG8:HID:C67A:I=:DE:G6I>DCD;6CDG<6C>O6I>DC !C
I=:8DCI:MID;:A:8IG>8EDL:G9>HIG>7JI>DC8DBE6C>:H(HI=:6K6>A67>A>IND;
:A:8IG>8 EDL:G >H :HH:CI>6A ;DG HD8>:IN DG I=>H G:6HDC <DK:GCB:CI G:<JA6IDGN
6.6 Building an MCDM/A Model on Condition Monitoring…

6<:C8>:H E:G;DGB 6C :HH:CI>6A GDA: >C 8DCIGDAA>C< I=: FJ6A>IN D; H:GK>8: D; I=:H:
8DBE6C>:H *:<JA6IDGN 6<:C8>:H EGDK>9: ;6KDG67A: 8DC9>I>DCH ;DG I=: :A:8IG>8>IN
B6G@:I ID 9:K:ADE >C 6 76A6C8:9 :CK>GDCB:CI 6BDC<HI 6<:CIH ;DG I=: 7:C:;>I D;
HD8>:IN !CI=>HH:8I>DC686H:HIJ9N>H86GG>:9DJIID:K6AJ6I:6C96EEANI=:E:G;DGB
6C8: D; 6 9:8>H>DC BD9:A 76H:9 DC 96I6 ;GDB 6 G6O>A>6C EDL:G 9>HIG>7JI>DC
8DBE6CN:GG:>G66C99:AB:>96
!CH:K:G6A8DJCIG>:H:A:8IG>8EDL:G9>HIG>7JI>DC8DBE6C>:H6G:>C6C:CK>GDCB:CI
8DCIGDAA:9 7N <DK:GCB:CI G:<JA6IDGN 6<:C8>:H .6G>DJH E:G;DGB6C8: >C9>86IDGH
L:G:9:K:ADE:9L>I=I=:6>BD;:CHJG>C<I=:FJ6A>IND;H:GK>8:L=>8=>HBDC>IDG:9
7N G:<JA6IDGN 6<:C8>:H ,=: EGD;>I67>A>IN D; 8DBE6C>:H >H 9>G:8IAN G:A6I:9 ID HJ8=
<D6AH ,LD G:A:K6CI 9>HIG>7JI>DC G:A>67>A>IN >C9>8:H I=6I B:6HJG: I=: 9JG6I>DC 6C9
;G:FJ:C8ND;I=:6K:G6<:>CI:GGJEI>DCD;6HNHI:B6G:@CDLC6H+NHI:BK:G6<:
!CI:GGJEI>DCG:FJ:C8N!C9:M+!!6C9+NHI:BK:G6<:!CI:GGJEI>DCJG6I>DC
!C9:M+!!ý:E>C ,=:G:;DG:6B:6HJG:I=6I86C7:JH:9ID6HH:HHI=:
CJB7:GD;8JHIDB:GH6;;:8I:97N6CDJI6<:>H9:G>K:9;GDBI=::ME:8I:9CJB7:G
D;;6>AJG:HNf(T)6+!!:FJ>K6A:CI:HI>B6I:;DG6C>CHE:8I>DCEDA>8N6H9:;>C:9
7N/6C< >C
T
E3 N f T 4 ³ OF t dt
L=:G:
F(t)Q8JBJA6I>K:9>HIG>7JI>DC;JC8I>DC D;I=:9:A6NI>B:
6H:9 DC I=: 9:A6N I>B: 8DC8:EI I=: :FJ>K6A:CI :HI>B6I: I=6I G:EG:H:CIH I=:
SAIDI>HI=:9DLCI>B:D(T)<>K:C>C
d f E3 N f T 4 d s
D T
T ds
L=:G:
6>AJG:L>AA7:G:E6>G:9>BB:9>6I:AN6I6C6K:G6<:8DHIcf6C99DLCI>B:df
An inspection takes place every T time units, costs cs units and requires ds time
units, where ds << T.
,=:8DHID;6C>CHE:8I>DCEDA>8N86C7:9:I:GB>C:97NI=:9:A6NI>B:8DC8:EI
JH>C< =G>HI:G
c f E 3 N f T 4 c s
C T
T ds
/=:C B6@>C< 9:8>H>DCH 67DJI I=: ;G:FJ:C8N D; >CHE:8I>DCH >C 6C :A:8IG>8
:C:G<N HNHI:B H:K:G6A ;68IDGH H=DJA9 7: I6@:C >CID 688DJCI HJ8= 6H I=:
6K6>A67>A>IND;I=:HNHI:B6C9I=:CJB7:GD;>CI:GGJEI>DCH !CI=::A:8IG>86A:C:G<N
H:8IDG 8DCH:FJ:C8:H 6HHD8>6I:9 L>I= ;6>AJG: H=DJA9 7: 6KD>9:9 9J: ID I=: =><=
>BE68IDCI=:8DC8:HH>DCD;I=:H:GK>8:
BJAI>8G>I:G>6 BD9:A 86C 7: JH:9 >C DG9:G ID 9:;>C: HIG6I:<>:H D; >CHE:8I>DC
>CI:GK6AH I=6I L>AA B::I I=G:: D7?:8I>K:H C6B:AN ID B>C>B>O: I=: CJB7:G D;
:ME:G>:C8:H68JHIDB:G=6HD;HJHI6>C:9>CI:GGJEI>DCDK:G6EG:9:;>C:9E:G>D9D;
I>B:I=:A:C<I=D;6C>CI:GGJEI>DC6C9I=:8DHIIDI=:HNHI:B
%-,#::C:N6C9*6>;;6L6H8=DH:CIDBD9:AI=:EGD7A:B ,=:B6>C
G:6HDC;DGI=>H8=D>8:>H76H:9DCI=:6HHJBEI>DCI=6II=:%PHG:6HDC>C<;DGI=>H
EGD7A:B 86C 7: G:EG:H:CI:9 7N I=: 6M>DB6I>8 HIGJ8IJG: D; I=>H I=:DGN !C I=>H
I=:DGN I=: 8DBE:CH6I>DC 7:IL::C I=: 8G>I:G>6 >BEA>:H I=: JH: D; 6 HNCI=:H>H
;JC8I>DC I=: <D6A D; L=>8= >H ID 6<<G:<6I: 6AA 8G>I:G>6 >C DC: 6C6ANI>8 ;JC8I>DC
,=:G:;DG: I=: %PH EG:;:G:C8: HIGJ8IJG: H=DJA9 7: 76H:9 DC I=: CDI>DC D;
8DBE:CH6I>DC
!I>H6HHJB:9I=6II=:G:>H6B6>CI:C6C8:B6C6<:GL=DH:G:HEDCH>7>A>IN>I>HID
9:I:GB>C: I=: >CHE:8I>DC >CI:GK6AH D; 8DC9>I>DC BDC>IDG>C< ,=JH I=: BD9:A
EGDEDH:9 I=GDJ<= %-, >H 9:K:ADE:9 >C DG9:G ID B::I =>H
=:G G:FJ>G:B:CIH HD
6H ID 6HH:HH I=: %- JC8I>DC !C I=>H 86H: I=: H:I D; 6AA I>B:HD;>CHE:8I>DC
6AI:GC6I>K:H >H 9:;>C:9 7N T 3 DG :68= >CHE:8I>DC I>B: I I=:G: >H 6
8DCH:FJ:C8: >C I:GBH D; 8DHI C(t) 9DLCI>B: D(t) 6C9 ME:8I:9 &JB7:G D;
;6>AJG:H Nf(t) DG :M6BEA: I=: ED>CI 3Ct1 Dt1 Nft14 7:ADC<H ID I=: 8DC
H:FJ:C8:HE68:
C :M6BEA: D; I=: 8DCH:FJ:C8: HE68: D; I=>H EGD7A:B >H H=DLC >C ><
,=: % L>H=:H ID B>C>B>O: I=: I=G:: 9>B:CH>DCH D; I=: 8DCH:FJ:C8: HE68:
H>BJAI6C:DJHAN ,=: EGD7A:B 8DCH>HIH D; 8=DDH>C< I=: 7:HI >CHE:8I>DC I>B: 7JI
I=:G:>H68DC;A>8I7:IL::CI=:I=G::9>B:CH>DCHD;I=:8DCH:FJ:C8:
D(T)
C(T)
Nf(T)

Fig. 6.7DCH:FJ:C8:HE68:;DGI=:I=G::8G>I:G>69DLCI>B:8DHI6C9CJB7:GD;;6>AJG:H

6.6 Building an MCDM/A Model on Condition Monitoring…
,=:8DCH:FJ:C8:HE68:9:;>C:9>C>< H=DLH6AAEDHH>7A:8DB7>C6I>DCHD;
I=:I=G::8DCH:FJ:C8:H7JI<:C:G6AANI=>HHE68:EG:H:CIH6ADID;JC;:6H>7A:ED>CIH
HJ8= 6H I=: DEI>B6A ED>CI >C :68= 9>B:CH>DC I=: ED>CI C*(T), D*(T), Nf*(t)). !;
I=>H ED>CI >H ;:6H>7A: I=:C >I >H CDI C:8:HH6GN ID BD9:A I=: EGD7A:B JH>C< 6
BJAI>8G>I:G>66EEGD68=7:86JH:I=>HED>CI9DB>C6I:H6AADI=:G6AI:GC6I>K:H6C9HD>I
H=DJA97:8=DH:C
(G:;:G:CI>6AJI>A>IN6C9699>I>K:>C9:E:C9:C8:H6G:I=G::>BEDGI6CI8DC8:EIHID
7: :MEADG:9 >C I=: JH: D; %-, -I>A>IN >C9:E:C9:C8: >H 6 8DC8:EI >C %-,
:FJ>K6A:CIIDI=6ID;EGD767>A>HI>8>C9:E:C9:C8:>CBJAI>K6G>6I:EGD767>A>INI=:DGN
-I>A>IN>C9:E:C9:C8:8DC9>I>DCH>BEANI=6II=:%-;JC8I>DCBJHI7:D;6HE:8>;>:9
;DGB !C<:C:G6A>C9:E:C9:C8:6HHJBEI>DCH<G:6IANH>BEA>;NI=:6HH:HHB:CID;I=:
DG><>C6AJI>A>IN;JC8I>DC ,=:I=G::6IIG>7JI:H6G:699>I>K:>C9:E:C9:CI>;I=:E6>G:9
EG:;:G:C8: 8DBE6G>HDC D; 6CN ILD ADII:G>:H 9:;>C:9 7N ILD ?D>CI EGD767>A>IN
9>HIG>7JI>DCH 9:E:C9H DCAN DC I=:>G B6G<>C6A EGD767>A>IN 9>HIG>7JI>DCH #::C:N
6C9*6>;;6
6H:9DCI=:699>I>K:JI>A>IN8DC8:EI>I86C7:8DC8AJ9:9I=6II=:6IIG>7JI:H6G:
699>I>K:>C9:E:C9:CI>;6C9DCAN>;I=:I=G::6IIG>7JI:JI>A>IN;JC8I>DC>H699>I>K:
DGI=:H:8G>I:G>6I=:699>I>K:;DGBB6N7:9:;>C:96H
u C T D T Nf T k c uc C T k d ud D T k n un Nf T
L=:G:
kcQH86A:8DCHI6CI;DGI=:8DHI8G>I:G>DC
uc(C(T))Q8DC9>I>DC6AJI>A>IN;JC8I>DC;DGI=:8DHI8G>I:G>DC
kdQH86A:8DCHI6CI;DGI=:9DLCI>B:8G>I:G>DC
ud(D(T))Q8DC9>I>DC6AJI>A>IN;JC8I>DC;DGI=:9DLCI>B:8G>I:G>DC
knQH86A:8DCHI6CI;DGI=:CJB7:GD;;6>AJG:H8G>I:G>DC
unNf,Q8DC9>I>DC6AJI>A>IN;JC8I>DC;DGI=:CJB7:GD;;6>AJG:H8G>I:G>DC

,=>H 6EEA>86I>DC >H 76H:9 DC 6 8DC;>9:CI>6A 86H: HIJ9N >C 6C :A:8IG>8 EDL:G
9>HIG>7JI>DC 8DBE6CN AI=DJ<= I=: ;><JG:H 6C9 DI=:G 6HE:8IH D; I=>H 6EEA>86I>DC
6G:CDII=:G:6A96I6I=:N=6K:7::C6EEGDEG>6I:AN6AI:G:9>CDG9:GIDG:EG:H:CI6
G:6A>HI>86C98DCH>HI:CI8DCI:MI
,=: D7?:8I>K: D; I=: 8DBE6CN >H ID B>C>B>O: I=: 8DHI 9DLCI>B: 6C9 I=:
:ME:8I:9 CJB7:G D; >CI:GGJEI>DCH D; I=: HNHI:B 7GDJ<=I 67DJI 7N 6C >CHE:8I>DC
EDA>8N 6H:9DCI=:BD9:AEGDEDH:9>CI=:EG:K>DJHH:8I>DCI=:;>GHIHI:EIDL6G9H
6EEAN>C<I=:BD9:A>HID>9:CI>;NI=:% B6>CI:C6C8:B6C6<:GG:HEDCH>7A:;DG
B6@>C<HJ8=9:8>H>DCH>CI=:8DBE6CNL6H>9:CI>;>:9 ,=:C:8:HH6GNE6G6B:I:GH
D;I=:BD9:AL:G::HI>B6I:96C96G:>AAJHIG6I:9>C,67A:

Table 6.1(6G6B:I:GHD;I=:BD9:A
(6G6B:I:G .6AJ:
O ;6JAIHE:G6N
h /:>7JAA
ds 96NH
db 96NH
Cb -+
Ci -+

H6G:HJAID;I=:%-;JC8I>DCI=:B6M>BJBJI>A>IND;I=:>CHE:8I>DCI>B:>H
;DG96NH !C>< I=:JI>A>IND;I=:>CHE:8I>DC>CI:GK6AH>HH=DLC
,=>HH:8I>DCEG:H:CI:96%-,IDHJEEDGII=:EA6CC>C<D;6C>CHE:8I>DCEDA>8N
>C 6C :A:8IG>8 EDL:G 9>HIG>7JI>DC 8DBE6CN ,=: CJB7:G D; I>B:H 6 8JHIDB:G
:ME:G>:C8:96HJHI6>C:9>CI:GGJEI>DCDK:G6EG:9:;>C:9E:G>D9D;I>B:I=:A:C<I=
D;>CI:GGJEI>DC6C98DHID;HNHI:B6G:I=G::D7?:8I>K:H8DCH>9:G:9 ,=:8DC8:EID;
9:A6NI>B:L6HJH:9IDBD9:AI=:;6>AJG:EGD8:HH

Fig. 6.8%-;JC8I>DC
References 247
MAUT was chosen to model a DM’s preferences for the cost, SAIDI and
SAIFI criteria in accordance with regulatory laws of this sector and in a suitable
way to deal with the tradeoff of probabilistic consequences. This model was
evaluated and validated by managers from a Brazilian company.
The modeling of predictive maintenance and monitoring is a tool that can
provide many benefits to the area of maintenance management. This chapter
suggests a multicriteria approach for modeling CBM decisions. Thus, the pro-
posed multicriteria model aimed to answer this need based on the MAUT which
has an axiomatic structure and allows to deal with the conflict between the expected
and the cost of an inspection policy downtime.
References
Barlow RE, Proschan F (1965) Mathematical theory of reliability. John Wiley & Sons, New York
Ben-Daya M, Duffuaa S, Raouf A (eds) (2000) Maintenance, Modeling, and Optimization.
Kluwer Academic Publishers, Norwell
Berrade MD, Cavalcante CAV, Scarf PA (2012) Maintenance scheduling of a protection system
subject to imperfect inspection and replacement. Eur J Oper Res 218:716–725
Carnero MC (2006) An evaluation system of the setting up of predictive maintenance
programmes. Reliab Eng Syst Saf 91:945–963
Čepin M (2011) Assessment of Power System Reliability: Methods and Applications. Springer
London
Chelbi A, Ait-Kadi D (2009) Inspection Strategies for Randomly Failing Systems. In: Ben-Daya
M, Duffuaa SO, Raouf A, et al. (eds) Handb. Maint. Manag. Eng. SE - 13. Springer London,
pp 303–335
Chiu SY, Cox LA Jr, Sun X (1999) Optimal sequential inspections of reliability systems subject
to parallel-chain precedence constraints. Discret Appl Math 96 - 97:327–336
Christer AH (1999) Developments in delay time analysis for modelling plant maintenance.
J Oper Res Soc 50:1120–1137
Christer AH, Waller WM (1984) Delay time models of industrial inspection maintenance
problems. J Oper Res Soc 35:401–406
Do Van P, Bérenguer C (2012) Condition-Based Maintenance with Imperfect Preventive Repairs
for a Deteriorating Production System. Qual Reliab Eng Int 28(6):624–633
Ferreira RJP, de Almeida AT (2014) Multicriteria model of inspection in a power distribution
company. Reliab Maintainab Symp (RAMS), 2014 Annu 1–5
Ferreira RJP, de Almeida AT, Cavalcante CAV (2009) A multi-criteria decision model to
determine inspection intervals of condition monitoring based on delay time analysis. Reliab
Fouladirad M, Grall A (2014) On-line change detection and condition-based maintenance for
systems with unknown deterioration parameters. IMA J Manag Math 25(2):139–158
Grall A, Bérenguer C, Dieulle L (2002) A condition-based maintenance policy for stochastically
deteriorating systems. Reliab Eng Syst Saf 76(2):167–180
Huynh KT, Castro IT, Barros A, Bérenguer C (2012) Modeling age-based maintenance strategies
with minimal repairs for systems subject to competing failure modes due to degradation and
shocks Eur J Oper Res 218(1):140-151
!+' DC9>I>DCBDC>IDG>C<6C99>6<CDHI>8HD;B68=>C:H:C:G6A<J>9:A>C:H

!CI:GC6I>DC6A'G<6C>O6I>DC;DG+I6C96G9>O6I>DC
"6G9>C: #+ $>C 6C?:K>8 G:K>:L DC B68=>C:GN 9>6<CDHI>8H 6C9 EGD<CDHI>8H
>BEA:B:CI>C<8DC9>I>DC76H:9B6>CI:C6C8: %:8=+NHI+><C6A(GD8:HH Q
"6G9>C:#+*6AHIDC(*:>9&+I6;;DG9"(GDEDGI>DC6A=6O6G9H6C6ANH>HD;9>:H:A:C<>C:
;6>AJG:96I6 )J6A*:A>67C<!CI Q
"DC:H ":C@>CHDC ! /6C< " %:I=D9DAD<N D; JH>C< 9:A6NI>B: 6C6ANH>H ;DG 6
B6CJ;68IJG>C<>C9JHIGN *:A>67C<+NHI+6;Q
#::C:N*$*6>;;6 :8>H>DCHL>I=BJAI>EA:D7?:8I>K:H(G:;:G:C8:H6C9.6AJ:,G69:
';;H />A:N+:G>:H>C(GD767>A>IN6C9%6I=:B6I>86A+I6I>HI>8H />A:N6C9+DCH&:L1DG@
#>B + G6C<DEDA DHI;;:8I>K: $>;:I>B: +IGJ8IJG6A :6AI= %DC>IDG>C< 6H:9 DC
K6>A67>A>IN "+IGJ8IC< Q
$>J%G6C<DEDA %JAI>D7?:8I>K:%6>CI:C6C8:(A6CC>C<'EI>B>O6I>DC;DG:I:G>DG6I>C<
G>9<:HDCH>9:G>C<DC9>I>DC+6;:IN6C9$>;:N8A:DHI "+IGJ8IC<Q
%6GH:<J:GG6%2>D(D9D;>AA>C>$ 'EI>B6AG:A>67>A>IN
6K6>A67>A>IND;JC8:GI6>CHNHI:BH
K>6BJAI>D7?:8I>K:<:C:I>86A<DG>I=BH *:A>67!,G6CH Q
%6GI>C # G:K>:L 7N 9>H8JHH>DC D; 8DC9>I>DC BDC>IDG>C< 6C9 ;6JAI 9>6<CDH>H >C
B68=>C:IDDAH !CI"%68=,DDAH%6CJ; Q
%6GIDG:AA + 6GADH + .>AA6CJ:K6 " :I 6A -H: D; BJAI>EA: D7?:8I>K: :KDAJI>DC6GN
6A<DG>I=BH>CDEI>B>O>C<HJGK:>AA6C8:G:FJ>G:B:CIH *:A>67C<+NHI+6; Q
&6@6<6L6, %6>CI:C6C8:,=:DGND;*:A>67>A>IN +EG>C<:G$DC9DC
&DLA6C+ :6E *:A>67>A>IN8:CI:G:9%6>CI:C6C8: DA7N88:HH(G:HH
(>AA6N/6C<"/6AA*JMIDC, B6>CI:C6C8:HIJ9ND;;>H=>C<K:HH:A:FJ>EB:CI
JH>C<9:A6NI>B:6C6ANH>H ")J6A%6>CIC<Q
(D9D;>AA>C>$2>D.6IC" *>H@>C;DGB:9DEI>B>H6I>DCD;G6>AL6NIG68@H>CHE:8I>DC6C9
B6>CI:C6C8:EGD8:9JG:H *:A>67C<+NHI+6; Q
+6HB6A+*6B6C?6C:NJAJ# DC9>I>DC:K6AJ6I>DCD;:M>HI>C<G:>C;DG8:98DC8G:I:7G>9<:H
JH>C<;JOON76H:96C6ANI>8=>:G6G8=N6EEGD68= ME:GI+NHIEEAQ
+:G<6@>#6A6>IO6@>H# ;JOON@CDLA:9<:76H:9B:I=D9;DGB6>CI:C6C8:EA6CC>C<>C6
EDL:GHNHI:B *:A>67C<+NHI+6;Q
,6C6@6 ,HJ@6D + 16B6H=>I6 :I 6A %JAI>EA: G>I:G>6 HH:HHB:CI D; +J7HI6I>DC
DC9>I>DCH 7N (6>G/>H: DBE6G>HDC D; C6ANI>8 >:G6G8=N (GD8:HH (DL:G :A>K !
,G6CH Q
,DGG:H8=:K:GGV6%6GIDG:AA+,=DBEHDC %D9:AA>C<6C9DEI>B>O6I>DCD;EGDD;
I:HI>C<EDA>8>:H;DGH6;:IN>CHIGJB:CI:9HNHI:BH *:A>67C<+NHI+6;Q
.AD@("D:IO::"$6C?:K>8:I6A 'EI>B6A8DBEDC:CIG:EA68:B:CI9:8>H>DCHJH>C<
K>7G6I>DCBDC>IDG>C<6C9I=:EGDEDGI>DC6A=6O6G9HBD9:A "'E:G*:H+D8Q
/6C< $ 6D *0 DC9>I>DC %DC>IDG>C< 6C9 DCIGDA ;DG !CI:AA><:CI %6CJ;68IJG>C<
+EG>C<:G+:G>:H>C9K6C8:9%6CJ;68IJG>C<+EG>C<:G&:L1DG@
/6C< / :A6N I>B: BD9:AA>C< !C #D7768N # %JGI=N &( :9H DBEA:M +NHI
%6>CI 6C97 + +EG>C<:G$DC9DCEEQ
/6C< / 'K:GK>:L D; 6 H:B>HID8=6HI>8 ;>AI:G>C< 6EEGD68= ;DG G:H>9J6A A>;: :HI>B6I>DC
L>I=6EEA>86I>DCH>C8DC9>I>DC76H:9B6>CI:C6C8: (GD8!CHI%:8=C<(6GI'"*>H@*:A>67
Q
/6C< / C DK:GK>:L D; I=: G:8:CI 69K6C8:H >C 9:A6NI>B:76H:9 B6>CI:C6C8:
BD9:AA>C< *:A>67C<+NHI+6;Q
/6C< / =G>HI:G ,DL6G9H 6 <:C:G6A 8DC9>I>DC 76H:9 B6>CI:C6C8: BD9:A ;DG 6
HID8=6HI>89NC6B>8HNHI:B "'E:G*:H+D8 Q
2>D.>696C6 'EI>B>O6I>DCD;I=:>CHE:8I>DC>CI:GK6AHD;6H6;:INHNHI:B>C6CJ8A:6G
EDL:G EA6CI 7N %JAI>'7?:8I>K: >;;:G:CI>6A KDAJI>DC %' *:A>67 C< +NHI +6;
Q
Chapter 7
Decision on Maintenance Outsourcing
Abstract: This chapter presents key aspects of multicriteria (MCDM/A) approaches

for decisions on maintenance outsourcing regarding maintenance contract, which
includes contract selection (e.g. repair contract) and supplier selection. Contract
design is a multi-objective task that leads the maintenance manager (or decision
maker - DM) to decide amongst a combination of contracts and suppliers’ bids for
the service. Given the multiple objective nature of this kind of problem, this
chapter presents models that include maintainability, dependability, quality of
repair and other aspects besides cost. The decision models presented consider
methods such as Multi-attribute utility theory (MAUT) to address compensatory
preferences and ELECTRE for preferences that require an outranking method. The
DM’s behavior to risk (prone, neutral and averse) is considered by using Utility
Theory and Decision theory foundations in order to include the state of nature in
decision models. Thus, most of the problems are related to supplier and contract
selection, which may be modeled into a single problem when considering all
combinations of contracts and suppliers as alternatives, including the possibility
of in-house maintenance being undertaken by a maintenance service supplier.
Depending on the organization and in how strategic its maintenance function may
be, decisions in maintenance outsourcing may be approached in different stages.
Thus, a key performance indicator (KPI) for such problems are defined depending
on the type of organization, its capabilities and the number of maintenance
activities, while the tradeoff amongst strategic objectives is balanced in order to
assure the system’s availability.
7.1 Introduction
Ever since management theory took shape, there has been extensive discussion
with regard to downsizing, core competences, business process re-engineering and
other managerial trends, that are deployed into general outsourcing. Such discussion
is also applicable to maintenance. According to Buck-Lew (1992), a company
outsource when it requests the services of an outside party to fulfill a function or
functions in the organization. Decisions on outsourcing are very close related to
contract selection (de Almeida 2001b), contract design and supplier selection
decisions.

250 Chapter 7 Decision on Maintenance Outsourcing
According to de Almeida (2007) the outsourcing decisions are requiring more

and more attention since the contract price is not the only aspect to be considered
by a DM. Therefore, MCDM/A techniques include the most appropriate tools for
evaluating the costs of the contract and the associate service performance. This
topic was found in 2.7% of publications reviewed by de Almeida et al. (2015),
considering MCDM/A approaches for reliability and maintenance.
As noticed in other organizational areas such as Information Systems (IS), the
trend towards outsourcing in maintenance has been greatly affected by rapid
changes in technology. Murthy and Jack (2008) pointed out the role of technological
advances, which have resulted in more complex and expensive equipment, have
increased the level of specialties and techniques needed to repair such equipment
and have led to a variety of work force specialties and diagnostic tools that require
constant upgrading.
Murthy and Jack (2008) also remind that the maintenance of governmental
infrastructures was traditionally maintained in-house. This changed in line with
these managerial trends in order to have a second party performing activities such
as road or rail maintenance services, for example.
Thus, outsourcing is a trend followed by many organizations that wish to focus
on their core competences, and due to technological advances, it has increased and
inspired a wide range of articles in the literature on maintenance outsourcing
decisions, especially because most of the organizations do not view maintenance
as a core business activity.
Based on the literature, this chapter presents some of the main MCDM/A
maintenance outsourcing decision problems, and criteria that might be considered
to guide these decisions. Also, repair outsourcing decision problems are approached,
including contract selection.
Outsourcing decisions are strategic and most of them include defining which
functions and activities are candidates for outsourcing and which should be kept
in-house. Secondly, there is a need to establish the criteria and key performance
indices to be followed by the maintenance service supplier, which should be
structured in the outsourcing contract.
Moreover, there is a choice problem, when selecting the service supplier. These
decisions are based in multiple factors that emphasize an MCDM/A approach
inherent in such decisions.
These decisions can be modeled into a single decision problem, by evaluating
all combinations of available contracts and suppliers as alternatives, including
the in-house maintenance service if the organization is capable to carry out its
maintenance activities. Thus, in-house maintenance service is also one of the
service supplier alternatives, therefore the final recommendation will reflect if the
activity should be outsourced or not, and also which contract should be selected.
In these decisions it is clear that there is a need to consider MCDM/A,
especially when facing problems with characteristics discussed in Chap. 2 that
turns maintenance problems more strategic and relevant. First of all, when
selecting if an activity should be outsourced, strategic aspects are observed as key
7.2 Selection of Outsourcing Requirements and Contract Parameters 251
performance indices regarding the impacts of outsourcing such activity considered.

With regard to the outsourcing requirements established in outsourcing contracts,
there are several attributes that are considered in terms of the objectives that shall
be used to evaluate these contracts. Same applies for deciding upon a list of
service suppliers. Traditionally, costs are considered to be among these criteria.
The need to consider an MCDM/A approach arises from several factors related
to these decisions, which require methodological support to ensure that the manager
(DM) will be supported in order to evaluate these factors properly according to
his/her preferences. During the following sections, specific problems are tackled,
and the criteria found in the literature to address outsourcing decisions.
This chapter focuses on the maintenance and repair contracts problems with an
MCDM/A perspective, which can be adapted to supplier selection. There are other
problems in maintenance outsourcing decisions related literature considering
warranties (Wu 2013), extended warranties and maintenance contract design
(Wang 2010). These topics are not addressed in this chapter.
7.2 Selection of Outsourcing Requirements and Contract

Parameters
The strategy for outsourcing maintenance goes together with the management of
contracts. The relationship between the contractor (the company outsourcing one
or more of its services – client company) and contracted firms (suppliers of such
services) is regulated by a contract in which the parties involved define the rules
of the service to be performed for an agreed length of time.
Service contracts in the area of maintenance typically emphasize the legal
aspects in clauses (terms) that deal with price, forms of readjusting price, payment
terms, quality and warrant provisions of the service to be provided, technical aspects,
transfers of responsibilities to third parties, retention/fines/damages, termination,
period (deadline), exchanges of information (communication channel) and other
important aspects of this relationship.
According to Brito et al. (2010), selecting contracts is a very important stage in
the process of outsourcing maintenance given the current trend towards reducing
costs and increasing competitiveness by focusing on core competences. Many
studies have been carried out on outsourcing and maintenance contracts, most of
which deal with qualitative aspects (Kennedy 1993; de Almeida 2005). Thus,
MCDM/A, plays an important role supporting DMs to deal with multiple and
conflicting criteria, and associated uncertainties in the process for selecting out-
sourcing contracts (Brito et al. 2010).
Wideman (1992) suggests that when companies consider outsourcing, they
need to make prior enquiries about bidding companies at the start of the hiring
process. According to Martin (1997), maintenance contractors are very interested in
developing new types of contracts that promise to offer them higher profitability
and increased flexibility and lower maintenance costs to them.
From a historical standpoint, the initial practice of the industrial sector was to
hire maintenance services in the form of manpower, i.e., paid for in terms of man-
hours worked. In this type of contract, it is the sole responsibility of the
maintenance service providers to ensure the presence of their staff in the industrial
plants of their customers, and therefore the suppliers is paid for the total number
of hours their staff worked. Main weakness of manpower contract are: less-skilled
personnel; low productivity of services; low quality of services; higher accident
rates; noncompliance with labor legislation.
Although still widely applied, this type of contract does not require the
commitment of outsourced staff to produce good results and, invariably, the
consequences for the industry may be negative in the medium and long term.
Therefore, this type of contract results in a relatively high business risk and
should not be entered into if the reason is the company’s vision is one of global
optimization. This is because although there may be an apparent reduction in the
cost of maintenance, undesired effects on the overall results can be generated.
This type of contract is practically a unilateral relationship. From the
perspective of game theory, one can assume that the policy contract is “win-lose”
in the short term, but in reality, in the medium or long term it can become a policy
of “lose-lose”. Therefore, sometimes this model proves to be bad for both the con-
tractor and the supplier.
Due to the problems discussed above, industry developed a different type of
contract: hiring for specific maintenance jobs or for special maintenance servicing
of specified equipment and machinery. This type of contract occurs in an isolated
form or as part of a hybrid contract (more than one contract type), the latter being
widely used in the industrial sector. Some advantages of this type of contract are:
better-qualified manpower; increased productivity; better quality of work.
The process of outsourcing maintenance activities evolved into hiring a single
supplier or a few suppliers, who are highly specialized and qualified and are made
responsible for the overall maintenance process. At this stage, the relationships
that have been established by the partnership between companies and their sub-
contractors in the maintenance area mature.
In this type of contract, contractors must give support to the activities out-
sourced, and make the staff of the contracted company feel a bond with the
contracting company and as if they were an integral part of a single organism,
which for its best performance needs to keep its basic functions operating in a
healthy way. Achieving this maximum mutual commitment remains the greatest
challenge for obtaining the best results in the process of maintenance outsourcing.
Another type of contract, with emphasis on both the client and the supplier, is
the type of contract that includes tracking results for performance. Typically, this
type of contract involves greater commitment from both sides of the contract, and
formalizes partnerships in the medium and long term (Wideman 1992).
Alternatively, Tsang (2002) discusses an alternative form of contracts by lease,

in which the contractor is a user of the final product produced by the supplier (be-
sides the active maintenance company being an investor).
Wideman (1992) discusses various types of contracts in the area of project
management, which can be widely replicated for the reality of maintenance
management. He considers that there are four main areas of risk in different types
of contracts from the customer’s perspective:
x Lump Sum (Global price) - the final price is based on the sum of all costs
involved, considering the contingencies, risks, overheads, profit margin, or any
parameter that can be expected to help form the contract price;
x Unit Price - should consider all direct and indirect costs involved, as well as the
overall price, and divided by number of events occurred;
x Target Cost (Based on a goal of total cost) - costs are defined transparently
between the parties involved and the final price contract is established together
with the target contract value;
x Reimbursable Costs (Variable remuneration) - pays the actual costs involved
and is based on full transparency and trust (partnership) between the parties to
the contract. A strategic alliance and a high level of maturity between the client
and the supplier need to be established.
Alternatively, Martin (1997) develops an analysis of the types of maintenance

contract in terms of operational criteria and knowledge retention. He divides them
into three classes: work package contract, performance contract and facilitator
contract, described in Table 7.1:
Table 7.1 Features of maintenance contracts
Type of contract Description

Most basic type of contract. The contract is simple, in which the payment
of the contracted services is based on the unit rate or lump sum. The service
Work package request is made by the client. The contractor can focus on the supplier
contract selection of the cheapest. The level of relationship with the contractor is
minimal. The knowledge about the operation system remains almost
entirely with the contractor.
Based on performance targets, the contractor and the contracted company
assume shared responsibility. The complexity of the contract is high, due
Performance clause contracts that are defined to assess the outcome of the contract (the
contract conflicts of interests of performance indicators). The relationship between
the parties should be close and usually long-term. The knowledge is shared
between both parties.
It is a type of contract where the service supplier is fully responsible for the
Facilitator contract result to be achieved, consequently the complexity of the contract is less.
It is also known as a lease contract.
The relationships of these contract types to each other and issues of contract
complexity, client-contractor relationship and client maintenance knowledge base
are shown in Fig. 7.1. The three types of contract discussed in Martin (1997) are
extreme cases. However, contractors may develop different (hybrid) contracts for
different sets of production systems, the skills involved and to split the financial
risk between the client and supplier.
The potential impact of maintenance on equipment and systems in terms of
quality, flexibility, cost, availability, and safety is increasingly evident within the
maintenance management system. Therefore, the need for measuring the performance
of maintenance is evident, and this means that maintenance as a function generates
profitability for the firm.
Therefore, what is critical is the process of setting performance indicators,
which will serve as regulatory elements of quality, variable remuneration (depending
on the chosen type of contract) or other indicators.
Fig. 7.1 Relationships of the various contract types and contract complexity, client-contractor
relationship and client’s knowledge of maintenance
It is noteworthy that there is no single standardization for the development

of indicators for the different segments of production systems. Their ways of
evaluating ‘productivity’ are different from each other, e.g.: one can consider
‘productivity’ can be considered as only about improving profits or as improve-
ments in such matters as availability or production rate or products or inventory
management or safety, or a combination of several of these.
In addition to selecting which performance indices will be used, the ranges of
performance using these indices must be considered by the parties involved. For
example, both parties to the contract must negotiate on the implications of having
ranges for performance indicators. For example, they must agree on from what
point in the range of reduced costs (cost of contract) that a high handling time
(availability of resources for the outsourced activities) and from what point in the
range that the high cost of the contract will result in it taking a short time to
complete the outsourced service.
Within several contract templates, one can identify the type of partnership
agreement based on indicators of availability and of the reliability of the pro-
duction system (using the Mean Time Between Failure – MTBF - and Mean Time
to Repair – MTTR - indicators), where the company to which services have been
outsourced increases its profitability as it improves the availability and reliability
of the client enterprise system (de Almeida and Souza 2001). Therefore, this type
of contract no longer remunerates services (grants bonuses to), but rather solutions
that will improve the levels of availability and reliability of systems.
However, some factors can disturb this type of contract. One that stands out is
the alignment amongst the strategic objectives of the interested parties. In fact,
there is a conflict of interest because the company to which services have been
outsourced also has difficulties of surviving in the competitive market, and it
needs to be competitive. For outsourced companies this type of service is a core
activity, while for contracting companies, it is a means of supporting an end
activity, so the maintenance services performed by subcontractors is the only
source of funds. This conflict is quite evident when contracts are mainly short and
medium term in length.
Therefore, assuming that the aspects of reliability were properly dealt with in
the phase of the designing the production system (i.e., both parties are aware that
the maintenance service will not have the ability to improve system reliability be-
yond that already specified in the design), it would remain for a maintainability
study to be included in maintenance agreements (de Almeida 2002). In this regard,
one has the administrative time (TD), the effective time to repair (TTR), the
availability of spare parts and the level of training of the outsourced teams.
However, all these previous aspects have a cost (C) associated with obtaining the
levels desired.
TD is the time that it takes to notify a maintenance company of a failure and the
time it takes this company to go to the client to deal with it. TD basically consists
of: the time spent in selecting and making the technical staff ready to perform the
service; the time taken to provide tools and the budget necessary to perform the
service; and the commuting time between the service provider company and the
location of the system to be repaired.
Therefore, TD can be a negotiated in a contract because it directly affects the
interruption time (TI) of the client system. As a counterpoint to this, it has to be
remembered that since outsourced firms have many clients, they try to keep the
idle time of their work teams within certain levels in order to meet the demands of
their diverse clients and to satisfy the times of visits agreed to by contract.
The time taken for the maintenance team from the start of the repair process to
putting the production system back into a normal state of operation is the TTR .
Normally, this time is directly related to the technical skills of the team, team
training, the team’s learning curve, the modularity of the system/equipment, the
availability of repair spare parts and other variables.
Therefore, when the contract is modeled as a function of the TD and TTR, the
maintenance contract must adequately compensate for the cost of the maintenance
structure of the company providing services that ought to be in a state of readiness
to meet sudden demands from the client company. However, keeping a large
contingent of maintenance staff available to the client and a high level of inventory
of spare parts, so as to guarantee an adequate level of system availability, becomes
very and does not fit the competitive market model that both the client and the
supplier find themselves in.
In reality, what is required is that the company providing maintenance services
has a firm commitment to ensuring the availability of the system (hence the need
for the outsourced team to be available at short notice) and not increasing the cost
of service. Thus, the best choice would to make a contract using a decision model
that incorporates the DM’s preference structure represented by the utility function
of the attributes cost C , of the interruption time (TI=TD+TTR) and of the
maintainability of the system as modeled in probabilistic terms.
Thus, the problem faced by the manager or DM becomes how to proceed with a
decision process that allows the various performance indices considered in the
contract to be optimized, since these various indices can conflict with each other.
Brito et al. (2010) state that contracts that present a lower cost might present a less
satisfactory performance concerning criteria related to quality and availability,
which creates a complex frame of trade-offs.
The DM faces several options for maintenance contracts, each implying
different system performances and related costs. de Almeida (2001a) identifies
that the selection of repair contracts is a non-trivial process since the consequences
of a wrong choice may be critical, for instance, in services where availability is
fundamental, as in telecommunications and electric power distribution services.
With regard to selecting contracts, little work has been conducted on exploring
a multi-criteria decision-making approach. de Almeida (2001b) has presented
MCDM/A models based on MAUT for selecting repair contracts, which aggregate
interruption time and related cost through an additive utility function. A different
approach can be found in de Almeida (2002), where the ELECTRE I method has
been combined with utility functions regarding a repair contract problem.
Brito and de Almeida (2007) and Brito et al. (2010) propose a MCDM/A
methodology to support the selection of maintenance contracts in a context where
in-formation is imprecise, when DMs are not able to assign precise values to the
importance parameters of criteria used for contract selection. Utility theory is
combined with the Variable Interdependent Parameters method (VIP) to evaluate
alternatives using an additive value function regarding interruption time, contract
cost and maintenance service supplier’s dependability.
7.3 MCDM/A Maintenance Service Supplier Selection 257
In general, in the context of drawing up outsourced maintenance contracts, the

DM should choose the option most preferred, the one with the best combination of
contract conditions (de Almeida 2002). For de Almeida (2005) what variables are
used may vary depending on the market that the company is in and its strategy,
and may involve: delivery speed or response time, quality, flexibility, depend-
ability and obviously, cost.
From an overview of the types of outsourcing maintenance contracts, it is
important to emphasize that there is no model for an optimal contract.
In reality, there are certain types of contract that are best suited to certain types
of relationships and partnerships between client and supplier, the type of service
contract, financial relations, economic issues, etc.
Therefore, in order to draw up complex contracts, it is necessary to have a large
amount of consistent information and knowledge. Wideman (1992) recommends
starting with the model for a simple and traditional maintenance contract. Later,
when a closer and systematic relationship between the parties has been established,
more advanced analysis of contracts involving performance evaluation criteria and
evaluator can be used.
Another factor that must be taken into account in entering into a contract is the
exchange of cultures between those involved. There is always resistance to a
change in culture when the culture between the parties is initially quite divergent.
The adaptation process can be time consuming and have a direct adverse effect on
the expected results from the contract.
Furthermore, companies tend to think that maintenance contracts can be
compiled quickly and easily which often leads to their being entered into
precipitately as a result of which invalid assumptions are made that will disturb
relationships in the partnership between the parties in the long term. The parties
should strive to reach a common point of view in order to generate a win-win
game, which will lead to their enjoying a transparent long-lasting relationship with
a high level of satisfaction for both parties. Therefore, permanent maintenance
contracts (with shared responsibilities) should be regarded as exemplifying the
strategic alliance between the two parties.
The next section presents some of the literature regarding maintenance service
supplier selection based on multiple criteria.
7.3 MCDM/A Maintenance Service Supplier Selection
This is an important decision problem for the outsourcing process; therefore, it

should be a compromise between costs and the performance required from service
suppliers. Specifically there is a need to address such problems with tools that
enable conflicting criteria to be dealt with that are usually followed by uncertainties
when referring to the consequences of maintenance decisions.
There are some decision models and applications in the literature that consider
MCDM/A techniques which will be discussed in this section.
7.3.1 Maintenance Service Supplier Selection with Compensatory

Preferences
To address a maintenance service supplier selection problem when the DM has a

compensatory preference structure, the literature presents two decision models
based in MAUT (de Almeida 2001a; de Almeida 2001b).
These decision models are based on the MCDM/A approach described in Chap. 2,
dealing with the following objectives: Interruption time and Cost.
Although both models deal with the same objectives, different assumptions
characterize each model, reflecting different situations that may be faced by a DM.
As described in Sect. 7.2, the interruption time is represented by the time spent
with administrative activities and the time spent executing the repair.
The first model assumes that during the interruption time the administrative
time is deterministic (de Almeida 2001a), while the second model assumes that
the administrative time follows an exponential distribution (de Almeida 2001b).
Deterministic Administrative Time Model
Despite considering the administrative time deterministic, the model presented by

de Almeida (2001a) follows the MCDM/A approach described in Chap. 2 and
considers the uncertainties related to the states of nature inherent to this problem.
de Almeida (2001a) considered the following assumptions:
x TI is explained by TD and TTR; where TI=TD+TTR;
x TTR follows an exponential distribution for all service suppliers in all contracts,
given by (7.1), where u is MTTR-1, therefore this parameter u represents the
state of nature.
f (TTR ) ue uTTR (7.1)
x There is prior knowledge S(u) about u that can be assessed from experts.
x TD is deterministic and assumes different values according to the service
supplier and contract.
x DM’s preference structure fits MAUT axiomatic requirements to be represented as
an additive utility function U(TI,C), given by (7.2), where kTI and kC are the
respective scale constants:
U (TI , C ) k TI U TI (TI ) k C U C (C ) (7.2)
x DM’s preference structure fits in both attributes an exponential utility function,

UTI(TI) and UC(C), to represent DM’s one-dimensional preferences, given by
(7.3) and (7.4). It means that for such a DM higher values of time or cost are
undesirable, which is a reasonable assumption, one of the reasons for assuming
this kind of utility function in many practical applications.
U TI (TI ) e A1TI (7.3)
U C (C ) e A2C (7.4)
Considering the uncertainties referring to the states of nature (MTTR), the DM

shall maximize his/her expected utility value EuU(u,ai), where U(u,ai) is the utility
of the state of the nature u and the action ai,, which refers to a specific
maintenance service supplier and contract representing the consequence (TI,C),
consequently U(u,ai) is obtained. The value of EuU(u,ai) is given by (7.5) (de
Almeida 2001a).
EuU (u , ai ) ³ U (u, a )S (u )du

i
(7.5)
u
In order to maximize the expected utility from (7.5), it is required to obtain

U(ui,ai) from U(TI,C). By considering the assumption that TD deterministic, is
possible to include TD as a constant into TTR, then, TI is reduced to TTR, thus,
U(TI,C) is equivalent to U(TTR,C), and U(TI) becomes U(TTR).
Thus, as pointed by de Almeida (2001a), U(u,a) is the expected value of
U(TTR,C), given by (7.6):
U (u , ai ) ³ U (TTR, C ) Pr(TTR | u , a )dTTR i

(7.6)
TTR
Since Pr(TTR | u , ai ) corresponds to f (TTR ) , then (10.6) can be rewritten

as (7.7):
f
³ >k U (TI ) k U (C )@u e

u ( TTR )
U (u , a )
i TI TI C C
dTTR (7.7)
0
By replacing (7.3) into (7.7), (7.8) is obtained:

k TI u
U (u , a i ) k CU C (C ) (7.8)
A1 u
Finally, replacing (7.4) in (7.8) and (7.5), there is (7.9):
ª k TI u º
EuU (u , ai ) ³«A u k e C
A2 C
»S i (u )du (7.9)
u
¬« 1 ¼»
Thus, for each distribution of TTR there will be an implied cost for the
respective service supplier contract, which means that the DM is deciding upon
the TTR pdf and its respective cost (C) in order to maximize his/her multi attribute
utility function. The alternatives for this problem are the existing combination of
maintenance service suppliers and its contract. Solving this problem consists
in solving (7.9) for all alternatives, which are all the existing combination of
maintenance service suppliers and its contract. Therefore, kTI and kC represents the
tradeoff between cost and time to repair according to DM’s preferences.
Stochastic Administrative Time Model
The model proposed by de Almeida (2001b) enables consideration to be given to

different types of contract, thereby seeking to select the best alternative in terms of
cost and system performance given the decision maker’s preferences represented
also by an additive function.
This model differs from the model presented in the last section for considering
the TD as a stochastic variable. This feature allows incorporating specific
conditions that appears in many real problems, that includes significant variation
and uncertainty on TI.
de Almeida (2001b) exemplifies situations that require to consider TD as a
stochastic variable, such as those associated with spares provisioning.
The assumptions considered by de Almeida (2001b) for this model are:
x TI is explained by TD and TTR; where TI=TD+TTR.
given by (7.1).
x TD follows an exponential distribution for all service suppliers in all contracts,
given by (7.10), where Z is a parameter defined according to the service
supplier contract service level and spare provisioning.
f (TD ) Ye YTD (7.10)

x TD and TTR are independent random variables.

x DM’s preference structure fits MAUT axiomatic requirements to be
represented as an additive utility function as given by (7.2).
x DM’s preference structure fits in both attributes an exponential utility function
to represent DM’s one-dimensional preferences, as given by (7.3) and (7.4).
Since this model considers a stochastic TD, TI is now the sum of two
independent random variables. The pdf of TI is obtained by (7.11):
f f
f (TI ) ³ f (TD) f (TI TD)dTD ³ f (TTR ) f (TI TTR )dTTR (7.11)

f f
Thus, from (7.1) and (7.10), follows that (7.12) results in (7.13), considering
that this result would be positive if, and only if TD t 0 and TI t TD , thus
TI t TD t 0 , from this result the integer from (7.12) turns into (7.13):
f
f (TI ) ³Ye ue
YTD u ( TI TD )
dTD (7.12)
f
TI
f (TI ) Yue uTI

³e
TD (Y u )
dTD (7.13)
0
Hence, developing (7.13), it is possible to find (7.14) for all TI t 0 as:
Yu
f (TI ) (e YTI
e uTI
) (7.14)
u Y
Thus, each maintenance service supplier contract ai is associated with a cost ci

and a specific probability function for TD represented by the parameter Zi .
Therefore, as in the previous model, the expected utility is given by (7.5) and shall
be maximized considering the prior knowledge S(u) over (7.14) instead of (7.1)
(de Almeida 2001b). Thus, similar to the previous model, U(u,a) is the expected
value of U(TI,C), given by (7.15) by applying the utility functions linearity
property (de Almeida 2001b):
U (u , a ) i ³ U (TI , C ) Pr(TI | u , a )dTI i

(7.15)
TI
Given that Pr(TI | u , ai ) corresponds to (7.14), then (7.15) can be rewritten

as (7.16):
Yu
U (u , a ) ³ >k e k U (C )@ (e e )dTI (7.16)
A1TI YTI uTI
i TI C C
TI u Y
By developing (7.16) into (7.17), and then replacing (7.14) in (7.17), (7.18) is
obtained, and developed into (7.19), (7.20) and (7.21):
Yu
U (u , a ) i ³ e (e e
Y
k TI
A1TI TI uTI
)dTI
u Y TI
(7.17)
Yu
k U (C ) ³ (e Y e )dTI TI uTI
u Y
C C
TI
Yu f f
U (u , a ) i
k TI ®³ e
( A1 Y ) TI
dTI ³ e ( A1 u ) TI
dTI ½¾
u Y ¯ 0 0 ¿ (7.18)
f
k U (C ) ³ F (TI )dTI
C C
0
Yu f f
e ( A1 Y ) TI dTI e ( A1 u ) TI dTI ½
U (u , a i ) k TI ®³ ³0 ¾
u Y ¯0 ¿ (7.19)
k CU C (C )
f f
Yu ° e Y ( A1
e ) TI ( A1 u ) TI ½°
U (u , a ) k ® ¾
u Y °̄ ( A Y ) ( A u) (7.20)
i TI
1 0 1 0 °¿
k U (C )
C C
Yu 1 1 ½
U (u , a i ) k TI ® ¾ k CU C (C ) (7.21)
u Y (
¯ 1 A Y ) ( A1
u) ¿
Hence, by developing (7.21) and applying (7.4), U(u,ai) is given by (7.22):
Yu
U (u , ai ) k TI k e A2 Ci
(7.22)
( A Y )( A u )
C
1 1
Thus, each service supplier contract (alternative or action) will be characterized

by the distribution of TI, considering the random variables TD and TTR , and the
respective implied cost.
Therefore the DM is deciding upon the a TI distribution and its respective cost
(C) in order to maximize the expected value of (7.22), given S(u) according to
(7.5).
Applying (7.22) in (7.5) gives the expression of EuU(u,ai) for the stochastic TD
model as given in (7.23):
ª Yu º
EuU (u , ai ) ³ «k ( A Y )( A u ) k e
TI C
A2 Ci
»S (u )du (7.23)
u
¬ 1 1 ¼
Hence, for the stochastic TD model (7.23) should be maximized, similarly to

(7.9) in the deterministic TD model.
The main assumptions of these models previously presented are that:
x Time has an exponential distribution.
x DM’s preferences fits MAUT requirements for an additive utility function
regarding system’s performance and cost.
These are very realistic assumptions, since there are many practical situations
in which both assumptions are confirmed during applications (de Almeida 2001a,
de Almeida 2001b).
The application of such decision models allows to reasonably measure the
response time of a maintenance service supplier contract, allowing also to consider
in-house maintenance service to be compared with other maintenance service
suppliers, and evaluate which activities would be better performed if outsourced
by considering an additive utility function for modeling the DM’s preferences
with regard to cost and the performance of the system.
Depending on the context of the problem the MCDM/A framework given in
Chap. 2 should be applied in order to build a more accurate decision model
by considering different MCDM/A methods and/or different probabilistic
assumptions, hence the choice among these will depend on the context of the
problem as discussed in Chap. 2.
7.3.2 Maintenance Service Supplier Selection with Non

Compensatory Preferences
In order to provide a more suitable model for a DM that has non-compensatory

rationality, a decision model considering non compensatory preferences is
presented. It adapts the decision model based in MAUT for using a compatible
method with non compensatory preferences for the maintenance service supplier
selection problem (de Almeida 2002). This illustrates a situation related to the step
6 of the building model procedure presented in Chap. 2.
This decision model associates Utility Theory with the ELECTRE I method.
The use of the ELECTRE I adds to the MCDM/A decision model a pairwise
dominance approach based on concordance and discordance indices that builds
outrank preference relations for selecting the best maintenance service supplier
contract. Thus, this decision model uses one-dimension utility functions values as
the performance of alternatives for each criterion.
This decision model was built for a repairable system considering the
implications of each alternative in terms of two aspects: Interruption time (or
response time) and Costs.
Similarly to the decision model previously presented, this decision model
evaluates the benefit of the maintenance service supplier contract in terms of
maintainability and the associated cost of the service.
The maintenance service supplier contract performance in the response time
reflects its specific condition for spare provisioning and repair capability.
The assumptions of the decision model are (de Almeida 2002):
x TI is explained by TD and TTR; where TI=TD+TTR.
given by (7.1).
x TD follows an exponential distribution for all service suppliers in all contracts,
given by (7.10).
x TD and TTR are independent random variables.
x DM’s preference structure fits a non compensatory rationality and requires an
MCDM/A approach compatible with outranking relation preferences according
to Chap. 2.
x DM’s preference intra criterion preference structure fits for both attributes an
exponential utility function for representing DM’s one-dimensional preferences
as in the deterministic administrative time model (de Almeida 2001a) and in
the stochastic administrative time model (de Almeida 2001b), these functions
are given by (7.3) and (7.4).
Since this model considers TTR and TD as two independent random variables,
given by (7.1) and (7.10) respectively, TI is given by (7.14).
Despite the fact of this decision model is not considering tradeoffs, the DM
behavior facing subjected to uncertainties is being modeled in the intra criterion
valuation by the utility functions given by (7.3) and (7.4).
Due to the assumptions of this particular decision model, the costs are not
affected by the state of nature, therefore each maintenance service supplier
contract has its particular cost definition not affected by uncertainties, thus the
cost criterion is evaluated directly by (7.4) according to each alternative’s cost (de
Almeida 2001b).
In the other hand, the response time represented by TI cannot be evaluated
directly for each alternative as the cost, due to the interference of state of the
nature uncertainties over its consequences.
For dealing with this situation, de Almeida (2002) considered the parameter Z,
related to TD. Applying the utility function linearity property as in the previous
models, UTI(Z) is given by (7.24).
U TI (Y ) ³ U (TI ) Pr(TI | Y )dTI

TI
(7.24)
TI
Pr(TI | Y ) corresponds to (7.14), then (7.24) can be rewritten as (7.25):
uY
U (Y )
TI
(7.25)
( A Y )( A u )
1 1
Based in the intra criterion alternatives evaluation given by (7.4) and (7.25), the
ELECTRE I method builds outranking relations based in concordance index
C(a,b) and in a discordance index D(a,b).
The concordance index is given by (7.26), and measures the relative advantage
of each alternative a compared with an alternative b (Vincke 1992).
C ( a, b)
¦ (W 0.5W )

, (7.26)
¦ (W 0.5W W )

where W+ corresponds to the sum of weights in which a is preferable to b, W= is

the sum of the weights in which a is equal to b, and W- is the sum of the weights in
which b is preferable than a.
The discordance index is given by (7.27) for measuring the relative
disadvantage of each alternative a compared to an alternative b (Vincke 1992).
ª ( Z Z ak ) º
D ( a, b) max « bk* », (7.27)
¬ (Z k Z k ) ¼

where Z ak is the evaluation of alternative a related to the criteria k, Z bk is the

*
evaluation of alternative b related to the criteria k, Z k is the best degree of
evaluation obtained for criteria k, and Z k is the worst degree of evaluation

obtained for criteria k.

Due to DM’s preference structure assumed for this decision model, it was
necessary to change the approach for evaluating the maintenance service suppliers
contracts on the response time, differently than the previous approaches using
MAUT.
Besides the adaptations required since different assumptions are made, is
important to highlight that the set of parameters representing DM’s preferences for
each decision model has different meanings, thus the measurements to represent
DM’s preferences such as the weights used for ELECTRE method are incompatible
with the required “weights” for MAUT, namely scale constants.
Therefore, building a decision model considering a different preferential

paradigm is important to improve the accuracy of the available decision models
for this class of problems, in order to give more flexibility for different types of
DM as discussed in Chap. 2.
7.3.3 Maintenance Service Supplier Selection with Non

Compensatory Preferences Including Dependability and Service
Quality
A maintenance service supplier selection problem using a non compensatory

MCDM/A approach is addressed (de Almeida 2005). This model approaches a
situation that includes three criteria besides cost, namely the repair time,
dependability and service quality using the ELECTRE I method.
The definition of dependability given by Slack and Lewis (2002) is that
dependability is related to measuring the performance of the promised deliveries
accomplishments. Therefore it represents a measurement about the chances of a
service supplier succeed in keep its service level beneath pre established limits.
Thus, for a maintenance service supplier it is associated with the probability di
of succeeding to perform the service under a response time faithful to the contract
proposal i.
Service quality may have several definitions. The definition adopted (de
Almeida 2005) for the decision model is that the service quality reflects the degree
of mistakes introduced once a repair has been performed. Thus, it is represented
by the probability qi that no fault has being introduced during the repair service
according to the expected conditions defined in the contract i.
With these extensions, this decision model was built to address the
maintenance service supplier selection problem including these four criteria:
x Interruption time or response time (TI);
x Cost (C);
x Dependability (di);
x Service quality (qi).
Similarly to the decision model presented in the previous sections, this decision
model (de Almeida 2005) evaluates the benefits of a maintenance service supplier
contract in terms of these three criteria and the cost related to the service contract.
Therefore, the maintenance service supplier contract performance now includes
not only the response time as a reflect of its specific condition for spare
provisioning and repair capability, but also the reliability of the maintenance
service team in order to avoid introducing failures in the system, and also that its
sizing would be enough to provide the service under the response time settled in
the contract.
Thus, the model considers (de Almeida 2005) the following assumptions:
x TI is explained only by TTR, following the maintainability approach given by
Goldman and Slattery (1977);
given by (7.1);
x Although there is prior knowledge S(u) about u that can be assessed from
experts, it is assumed that there is an uncertainty about the real value of ui,
withregard to the respective contract i;
x Based on the last assumption, di is defined as the probability that ui t uie for
action ai, therefore uie is the value committed by contract for ui . Therefore, di
is given by (7.28):
di ³ S (u )du
i i
(7.28)
uie
x DM’s preference structure fits a non compensatory rationality and requires an

MCDM/A approach compatible with outranking relation preferences according
to Chap. 2.
x DM’s preference intra criterion preference structure fits for an exponential
utility function for the attributes repair time and cost, given by (7.3) and (7.4),
respectively. Once the higher is di , higher is Ud(di) , the DM utility function for
dependability. Thus, a logarithm utility function is assumed for the depend-
ability given by (7.29) (de Almeida 2005). Same applies to service quality,
therefore assuming also a logarithm utility function, given by (7.30).
U d (d i ) B3 C 3 ln( A3 d i ) (7.29)
U q (q i ) B4 C 4 ln( A4 q i ) (7.30)
From the assumptions of this decision model costs are also not affected by the
state of nature as in the previous section. Therefore the cost criterion is evaluated
directly by (7.4) according to each alternative’s cost (de Almeida 2005). Same
applies to dependability and service quality, evaluated respectively by (7.29) and
(7.30). From the assumption of the prior knowledge over TTR , the state of nature
must be considered for evaluating the consequences on repair time by considering
the parameter ui instead of TTR. Similarly to the previous, the decision model
proposed by de Almeida (2005) uses the linearity property to obtain UTI(ui) from
(7.31).
U TI (u i ) ³ U (TTR ) Pr(TTR | u )dTTR

TI i
(7.31)
Since Pr(TTR | ui ) is given by (7.1), then, applying (7.1) and (7.3) to (7.31),
(7.32) is achieved.
ui
U TI (u i ) (7.32)
( A1 u i )
By assuming that TI is explained only on TTR is a simplification that may be

adopted if necessary for a particular organizational condition. It depends on the
specificities of each application. Such simplification in the decision model allows
to deal with more accuracy with the parameters included in the evaluation for
addressing the preferences over the attributes dependability and service quality of
the maintenance service supplier contracts.
Another point to emphasize is that for the response time one may be interested
in assessing directly over ui given (7.32), however it is easier for a DM to have its
preferences elicited directly over TTR than in ui.
7.3.4 Maintenance Service Supplier Selection with Preference’s

Partial Information
In some situations, the DM may not feel comfortable about setting precise values
for the decision model parameters, and thus an approach suitable for dealing with
this situation should be used.
The decision models presented by Brito and de Almeida (2007) and Brito et al.
(2010) for selecting maintenance service supplier contracts addressed such a
particular situation, using an approach that enabled a recommendation to be made
based on imprecise statements with regard to the decision maker’s preferences and
this was supported by an elicitation procedure.
Many DMs have difficulties to fix constant values for criteria ‘weights’ that
must represent not only the importance of the criteria but also the compensation
rates between criteria in additive value functions.
There may be several reasons for a decision maker to avoid precise statements.
One of these may be that he is unsure if a parameter should be 0.75 or 0.7. Thus if
a range of values can be used for such parameters, the decision maker can give
more confident statements regarding the decision problem.
Brito and de Almeida (2007) considered three basic criteria: interruption time,
applicant’s dependability and contract cost in this model.
The dependability criterion is used to assess alternatives of contract in relation
to “deadlines” being met. It is a measure related to keeping delivery promises,
which it is represented by the probability of the company selected achieving the
time to repair under a specified probability distribution, as set out in the contract
proposal of the maintenance service supplier, similarly as in the model presented
in last section.
To Brito and de Almeida (2007), these three criteria may be conflictive among
alternatives. Usually, lower interruption times (times to repair) are related to better
resource conditions, better spares provisioning and higher professional skills, and
they often imply higher costs.
Besides, the dependability of the alternative is not directly related to the
proposal conditions associated with interruption time, but it is assessed by
the contracting company taking into consideration other aspects such as the
applicant’s reputation, previous services, the structure of repair facilities, etc.
The approach used in the decision model by Brito et al. (2010) considers utility
functions aggregated by variable interdependent parameters for an additive
function and uses the following criteria for evaluation:
x Mean time to repair (MTTR).
x Service supplier cost.
x Geographical spread of the service supplier network.
x Service supplier reputation.
x Compatibility of company cultures.
The specific problem considered by Brito et al. (2010) was related to power
distribution services, which may also be extended to the telecommunications
context.
The service supplier’s performance on MTTR indicates its structure and
capabilities, thereby reflecting its maintenance staff’s skills, transportation
resources, facilities and spares inventory.
The geographical spread of the service supplier reveals its logistical network
structure, and relates to the number and spread of local branch offices, which
gives flexibility and speed with regard to performing repairs. This is an important
point for companies with widespread local branch offices, and it is directly related
to the speed of service response and flexibility offered to the contracting
organization or its several units
The service supplier’s reputation is another important factor to be considered
since this may avoid bad experiences from past services or even service level
inconsistencies being repeated during the time span of the contract. Evaluating the
service supplier in this respect may be from external sources, such as other
companies that had previous experiences with the service supplier, verifying if
payment of taxes to the government is up-to-date and possession of the due
certifications in quality and/or safety norms.
Cultural compatibility is an issue that has become more and more relevant,
since many organizations are seeking to establish long-term relations by building
strategic partnerships. Allied to such strategic factors, many companies have
added undertaking social and environmental responsibility activities to their
organizational objectives. This includes their seeking sustainability and requiring
this commitment also from their partners and suppliers.
By using variable interdependent parameters, Brito et al. (2010) considered the
range for each parameter, assessing a lower and an upper bound. Another kind of
imprecise information was the order (ranking) of the parameters. The assessment
of these imprecise statements given by decision maker enabled dominance
relations among the service suppliers to be established, based on the decision
maker’s assessed preferences.
According to Brito et al. (2010), in order to assess the performances of
alternatives of contracts for the first two criteria, since MTTR and contract cost
can be directly represented by values, utility values should be elicited using the
due procedures. However, the last three criteria present a less objective feature; in
this case, each candidate may be evaluated after completion of a questionnaire,
which is constructed so as to obtain all the information required by the contracting
organization in order to assess the candidates on each of the three criteria.
7.4 Other Approaches for Supplier Selection
The problem of supplier selection has been studied in many contexts, rather than
the RRM context. Also, studies have been found in a broader way, therefore
MCDM/A and other approaches to supplier evaluation and selection problem have
been widely studied.
There are various decision making approaches proposed in the literature. Ho
et al. (2010) presented a literature review on this topic emphasizing which
approaches were frequently applied, which criteria were most considered, and
investigated inadequacy with regard to the studies of the approaches found in the
literature within their review of articles published in international journals from
2000 to 2008.
Among other approaches widely used for supplier evaluation and selection, Ho
et al. (2010) highlight the use of: Analytic hierarchy process (AHP), Analytic
network process (ANP), Case-based reasoning (CBR), Data envelopment analysis
(DEA), Fuzzy set theory, Genetic algorithm (GA), Mathematical programming,
Simple multi-attribute rating technique (SMART), and hybrid approaches.
Ho et al. (2010) evaluated MCDM/A approaches versus traditional cost based
approaches. The advantages of applying MCDM/A approaches enable con-
sideration to be given to important and relevant factors for the decision process
other than cost.
Another recent literature review on supplier selection was presented by Chai et
al. (2013), considering articles published in journals from 2008 to 2012 that
presented applications of decision making techniques for supplier selection.
From the literature review conducted by Chai et al. (2013) many decision
making approaches have been applied to these problems recently. Chai et al.
(2013) identified twenty six decision making techniques applied for supplier
evaluation and selection, and grouped these techniques into three categories:
MCDM/A, mathematical programming and artificial intelligence techniques.
References 271
Supplier selection is an important topic, studied and tackled with many

approaches, although most of them are not related to RRM context. The specific
techniques applied to supplier selection problems are listed below by each of the
categories considered by Chai et al. (2013):
 MCDM/A: AHP, ANP, ELECTRE, PROMETHEE, TOPSIS, VIKOR,
DEMATEL, SMART, Multiobjective programming, Goal programming.
 Single Objective Mathematical programming: DEA, Linear programming,
Nonlinear programming, Stochastic programming.
 Artificial intelligence: Genetic algorithm, Grey system theory, Neural net-
works, Rough set theory, Bayesian networks, Decision tree, Case-based
reasoning, Particle swarm optimization, Support vector machine, Association
rule, Ant colony algorithm, Dempster-Shafer theory of evidence.
The choice of a maintenance service supplier may be addressed by using
different criteria and different techniques depending on the decision context,
although on the particular context of maintenance there still a scarce literature, this
is an important complex decision problem which includes strategic organizational
objectives and consequences subjected to different kinds of states of nature.
References
Brito AJ de M, Almeida-Filho AT de, de Almeida AT (2010) Multi-criteria decision model for

selecting repair contracts by applying utility theory and variable interdependent parameters.
IMA J Manag Math 21:349–361
Brito AJ de M, de Almeida AT (2007) Multicriteria decision model for selecting maintenance
contracts by applying utility theory and variable interdependent parameters. In: Carr M, Scarf
P, Wang W (eds) Model. Ind. Maint. Reliab. Proc. Mimar. 6th IMA Int. Conf. Manchester,
United Kingdom, pp 74–79
Buck-Lew M (1992) To outsource or not? Int J Inf Manage 12:3–20
Chai J, Liu JNK, Ngai EWT (2013) Application of decision-making techniques in supplier
selection: A systematic review of literature. Expert Syst Appl 40:3872–3885
de Almeida AT (2001a) Repair contract decision model through additive utility function. J Qual
Maint Eng 7:42–48
de Almeida AT (2001b) Multicriteria decision making on maintenance: Spares and contracts
planning. Eur J Oper Res 129:235–241
de Almeida AT (2002) Multicriteria modelling for a repair contract problem based on utility and
the ELECTRE I method. IMA J Manag Math 13:29–37
de Almeida AT (2005) Multicriteria Modelling of Repair Contract Based on Utility and
ELECTRE I Method with Dependability and Service Quality Criteria. Ann Oper Res
138:113–126

Goldman AS, Slattery TB (1977) Maintainability: a major element of system effectiveness.
Robert E. Krieger Publishing Company, New York
Ho W, Xu X, Dey PK (2010) Multi-criteria decision making approaches for supplier evaluation
and selection: A literature review. Eur J Oper Res 202:16–24
Kennedy WJ (1993) Modeling in-house vs. contract maintenance, with fixed costs and learning
effects. Int J Prod Econ 32:277–283
Martin HH (1997) Contracting out maintenance and a plan for future research. J Qual Maint Eng,
3:81–90
Murthy DNP, Jack N (2008) Maintenance Outsourcing. In: Kobbacy KAH, Murthy DNP (eds)
Complex Syst. Maint. Handb. SE - 15. Springer London, pp 373–393
Slack N, Lewis M (2002) Operations Strategy. Prentice Hall, London
Tsang AHC (2002) Strategic dimensions of maintenance management. J Qual Maint Eng 8:7–39
Wang W (2010) A model for maintenance service contract design, negotiation and optimization.
Eur J Oper Res 201:239–246
Wideman RM (1992) Project and program risk management: a guide to managing project risks
and opportunities. Project Management Institute
Wu S (2013) A review on coarse warranty data and analysis. Reliab Eng Syst Saf 114:1–11
Chapter 8
Spare Parts Planning Decisions
Abstract An important issue related to maintenance management is the problem

of sizing the amount of spare parts. An excess number of spare parts results in
financial losses. However, a lack of spare parts is also negative, because this may
result in a loss of production due to the increased downtime of equipment.
Therefore, spare parts should be available in quantity and at the right time. Spare
part planning decisions need to evaluate multidimensional objectives, such as
costs, profitability, reliability, availability and probability of stockout. Typically,
these objectives are conflicting. Unlike a single objective approach, which often
implies the poor performance of other objectives desired by the decision maker
(DM), a multicriteria (MCDM/A) approach provides a spectrum of compromise
solutions, which reflect the tradeoffs represented by DM’s preference structure, by
using a multi-attribute utility function. Another relevant aspect is the management
of uncertainties about the reliability or maintainability of the system, using the
concepts of Decision Theory and a Bayesian approach, which incorporate experts’
prior knowledge. This chapter presents a model, based on Multi-attribute Utility
Theory (MAUT), for spare parts sizing that considers aspects of the risk of
inventory shortages and cost. Furthermore, an NSGA-II multi-objective model for
multiple spare parts sizing is discussed. Finally, a model considering condition-
based maintenance (CBM) is presented.
8.1 Introduction
Management of spare parts certainly has a positive influence on maintenance

management, since this leads to the higher reliability and availability of equipment
and therefore has a direct impact on business profitability. Therefore, one of the
most important issues related to maintenance management is the problem of sizing
the number of spare parts to be held in stock, bearing in mind that this affects the
performance of maintenance, because the number of spare parts available directly
affects the downtime or interruption to the full operation of a given piece of
equipment (system). Spare parts should be available in quantity and at the right
time. Just as stocking an excess number of spare parts results in losses or
foregoing funds that a company could have applied elsewhere, a lack of spare

274 Chapter 8 Spare Parts Planning Decisions
parts is also negative, as this may well result in a loss of production due to
increased downtime of equipment while awaiting delivery of the spare parts
needed. Therefore, sizing the number of spare parts that optimally need to be held
strongly influences a company´s costs and profitability. Consequently, the
management of this resource is one of the most critical tasks in maintenance
management (de Almeida and Souza 2001). This topic is relevant for many
contexts, may they be related to individual plants, such as refinery (Porras and
Dekker 2008), or to a logistic network (Syntetos et al. 2009).
When comparing with other types of inventory models, such as raw material
for manufacturing processes, sizing and managing spare parts inventory is a far
more complex task, considering that manufacturing inputs are usually is easier to
forecast its demand, especially if comparing its turnover. Production inventories
usually follows market rules, but spare parts are required based on failure rates
and the system reliability design.
Thus, spare parts are sized according to its relative importance according to the
system reliability. A bad decision on spare part sizing may lead to high losses,
compromising company’s profitability as well as its system availability.
According to British Standards 3843-1:1992, Terotechnology is the study that
allows the maintenance of assets in optimal manner by the combination of
management, financial, engineering and other practices applied to physical assets
such as equipment considering its life cycle costs.
The number of spare parts sized must consider time to repair or time of service
disruption. This decision must assure that parts required will be available when
requested. Thus, the spare parts sizing problem has conflicting goals that are
the increase of spare parts available as a contribution to increase equipment
availability by the reduction of service disruption time. As in the other hand, the
goal to reduce the inventory and purchase costs of spare parts.
Considering that manufacturing inventory model are not suitable to manage
spare parts inventory due to the differences in the items demand for both cases,
spare parts inventory considers that the demand for each item follows a stochastic
process, represented by the random variable of equipment failure.
According to Marseguerra et al. (2005), to avoid risks to plant and costly plant
being unavailable due to a shortage of spares parts, the latter are often over-
stocked, thus leading to huge losses due to having invested unnecessarily in an
excessive number of them or to too many of them becoming obsolete.
For Roda et al. (2014), spare parts management plays a relevant role for
equipment-intensive companies. They review the type of criteria applied to spare
parts classification. An important step of such a process is that of classifying spare
parts (criticality) with a view to enabling different items to be properly managed
by taking into account their peculiarities. Many advantages can be achieved by
proper classification, e.g. an organization may align its policy for the stock
management system with the criticality of the need for holding spare parts
(Macchi et al. 2011); demand forecasting may be driven by data collected on the
parts for different classes while improving the performance of equipment and the
8.1 Introduction 275
system overall may concern critical classes, thus making the work of the analyst
easier by allowing him/her to concentrate on tests of inventory control policies
currently in force. Forecasting spare parts demand is an important issue (Boylan
and Syntetos 2010) for building related decision models.
Some studies have addressed the problem of determining the optimal spare
parts inventory, such as by using gradient methods, dynamic programming,
integer programming, mixed integer and nonlinear programming. Unfortunately,
as mentioned by Marseguerra et al. (2005), such optimization techniques typically
entail the use of simplified plants ‘or systems’ models about which predictions
may be of questionable realism and reliability.
In general, spare parts management has at least two main objectives that are
conflicting: to contribute to increased system availability by acquiring and
stocking spare parts; i.e., ensuring the supply of spare parts in the proper amount
to reduce interruption times and; to reduce the cost of buying and stocking spare
parts (de Almeida 2001).
A decision model on provisions for spares assumes that at least one spare item
is held in stock (de Almeida 2001; de Almeida 1996). Normally, when a failure
occurs, in due time, the failed item is replaced by a spare part, which should be
available in the depot. The faulty item is sent out for repair and after being
repaired (good-as-new), it is shipped back to the plant depot where it serves as a
spare. This decision problem uses an MCDM/A model in order to define how
many additional spares should be provided in accordance with what criteria. de
Almeida (2001) applies a Bayesian approach, based on prior probability
distributions. Aronis et al. (2004) also uses prior distributions of the failure rates
to forecast demand.
Some techniques for planning and inventory control were developed for the
context of manufacturing systems (goods producing systems), and were later
extended to the service producing systems. An example of this is Just-In-Time,
which aims to meet the instantaneous demand i.e. only the amount necessary for
the customer at that moment of need. These techniques are suitable for systems
that have predictable demand and are determined by the client. However, in
studies on reliability (in the maintenance context), demand is a probabilistic event,
represented by the number of failures (a random variable). For this reason, the
literature on sizing stocks of spare parts in the maintenance area addresses the
question in very specific ways.
Other noteworthy conditions that make them different from production
inventories are (Kennedy et al. 2002; Macchi et al. 2011): the number of spare
parts in stock is often too large; the sourcing of spare parts is often limited to one
or a few suppliers, causing constraints regarding procurement lead time and the
costs; or in the opposite case of multiple sourcing, the related risk of variations in
the quality of materials supplied can occur; obsolescence may be a problem;
indeed, it is difficult to determine how many units of a spare part to stock for an
obsolescent machine; the high variety in the characteristics of spare parts can
normally be observed (the rates of consumption for some parts are very much
higher than for others; some parts are characterized by being cheap to buy, while
others are very expensive; often, procurement lead times vary greatly and may be
lengthy, especially in the case of specific parts or those that have to be placed on
order); and, the management process often lacks information visibility, due to
poor inventory data record-keeping, inefficient or ineffective ordering processes
and inventory management information being hidden in separated “silos”, these
being only some typical reasons for such low visibility.
Duchessi et al. (1988) propose a top-down methodology, which classifies
spares into distinct categories and associates appropriate controls with each
category. This methodology identifies spare parts that do not have to be stocked.
By eliminating these spare parts from the inventory, the manager can reduce costs
and thus improve profits. Thereafter, it identifies critical spare parts that, if not in
stock when needed, result in excessive downtime costs. Moreover, avoidance of
downtime reduces production lead time and improves performance regarding on
time delivery to customers. Finally, it displays a logical framework so that the
need for and stock of spare parts can be matched with formal control policies,
procedures and techniques.
Molenaers et al. (2012) propose a spare part classification method based on the
criticality of an item, using an MCDM/A model. Starting from a multicriteria
analysis, the proposed model converts relevant criteria on such criticality into a
single score which then is considered the level of criticality of the item. This level
is used to rationalize the efficiency of the spare parts inventory policy.
A literature review on MCDM/A approaches in reliability and maintenance
shows work conducted related to spare parts sizing (de Almeida et al. 2015).
8.2 Some Sizing Approaches for Spare Parts in Repair
This text highlights some approaches to the problem of sizing the need for spare
parts (de Almeida and Souza 2001):
x An approach based on the risk of inventory shortages;
x An approach based on the risk of inventory shortages by using prior knowledge;
x An approach under the cost constraint;
x An approach according to an MCDM/A model.
8.2.1 Relevant Factors to Sizing Spare Parts
The system type, whether repairable or not repairable, will influence how to size
the need for spare parts. For non-repairable systems, the desired lifecycle of the
system should be considered as a variable time T (de Almeida 1996; de Almeida
8.2 Some Sizing Approaches for Spare Parts in Repair 277
and Souza 2001). It should be noted, therefore, that the size of the stock is defined
by the difficulty in acquiring spare parts (price, delivery time, availability of more
than one supplier, etc.) and by issues directly to inventory management (available
space, cost storage, etc.).
As to repairable systems, the variable T is equal to the time at which the item
will be repaired, i.e., the system is restored when the defective item is replaced
with a similar one that is already in stock. The number of spare parts in this case is
equal to N s N 1 , since the defective item returns to stock after being repaired
(de Almeida 1996; de Almeida and Souza 2001).
Another issue that will influence spare parts management is related to the
behavior of the failure rate over time. As seen, the number of items available for
spare parts is directly related to the number of failures, which in turn is directly
related to the reliability of the equipment (system). Therefore, the problem is
directly related to the behavior that the variable deemed the number of failures is a
function of time. One should also consider the independence of failures among the
items that make up the system.
Under the analysis of the bathtub curve, spare parts management, in the repair
context, is usually dealt with only in the second life stage that matches the useful
life or the operational phase of the equipment. At this stage, it is assumed that the
failure rate O(t) has a constant behavior as a function of time (the reliability
function is represented by an exponential probability function).
In the first phase of the bathtub curve, in which the predominant faults are
classified as early failures, these are usually covered by the equipment manu-
facturer’s warranty, with no need for the user to direct efforts to solve this
problem, i.e., it is not necessary to have spare parts in stock to cover this period in
the lifecycle of the equipment. In some specific kinds of contract, it is interesting
to analyze the possibility of having spare parts. From the manufacturer point of
view, the sizing decision for this stage has to be made and may follow the model
presented in this section, with proper assumptions.
In the third phase of the bathtub curve, the equipment is at the end of its useful
life. Therefore it might not make much sense to study the problem of dimension-
ing the need for spare parts, in the context of repair, because at this stage the
failure rate is high due to wear and tear. The failure rate O(t) increases with time
so that repair is not sufficient to change the behavior of degenerative equipment,
so the equipment has reached its use limit at this stage. At this stage, what remains
is to consider the policy for preventive maintenance, replacement, reconstruction
or overhaul. If economically feasible, this period may be prolonged as necessary
until the equipment is deemed obsolescent and can then be discarded.
The spare parts sizing in this different context should use the information
collected for the maintenance decision in that particular context. For instance, if a
preventive maintenance model, such as one of those in Chap. 5, is applied, then,
the information from the decision model regarding to the amount of replacements
necessary in the planning time horizon is related to the sizing of the spare parts.
Another relevant factor to be considered for sizing the need for spare parts is
technological outdating, which can be a limiting factor in the lifetime of a piece of
equipment (system) and thus, may well shorten its life expectancy. Therefore, when-
ever the equipment becomes technologically outdated before the end of its useful
life, the spare parts for it that are in stock lose their functionality in short periods,
there being an economic loss (obsolete inventory) that need to be written off.
Furthermore, what to do about perishable goods (spare parts subject to
degradation while held in stock) should be made of the Wilson model defined in
Rezg et al. (2008) and Ben-Daya et al. (2009). Gopalakrishnan and Banerji (2013)
point out that perishable spares, with a short shelf life, must be identified, and the
First in First out method must be practiced. Therefore, the optimal sizing of the
total quantity of each spare part has to be determined, and must take into
consideration the objectives of minimizing the cost to the system and wastage
(loss of materials due to deterioration) as investment constraints (Padmanabhan
and Vrat 1990).
Van Volkenburg et al. (2014) develop a model which addresses the effects of
the shelf-life of spare parts (perishable items) on optimizing the stocking of spare
parts because certain conditions exacerbate their deterioration, thereby affecting
the reliability of the system being supported or the spare part being found to be
unserviceable when required. This is especially evident in non-repairable
components that are stored for extended periods.
8.2.2 Approach Based on the Risk of Inventory Shortages
This approach involves determining a number of spare parts N, for a given value
of risk of stock shortages D within a particular time value T. Thus, cost is
considered in an indirect way, because as the desire is to reduce the risk D, there is
a resultant increase in cost and vice versa. So the cost is obtained at the instant that
defines what level of risk to run (i.e., will be determined at the time of choosing
the value of D).
The risk of stock shortages D means the probability that the number of spare
parts in stock is less than the number of failures x, namely, Px ! N (Probability
of Stockout of the spare parts (PS)). Thus, the Margin of Safety (MOS) is defined
as MOS 1 D , i.e. MOS 1 PS , which is a measure of the probability that
the stock will not fall outside the range considered (de Almeida 1996; de Almeida
and Souza 2001). Therefore,
MOS 1 D Px d N (8.1)
where N is the number of spare parts kept in stock. Notice that the MOS
corresponds to the cumulative probability distribution of the number of failures.
Assuming a Poisson Process, for a system comprising n items:
N
nOT k e nOT
MOS P x d N ¦
k 0
k!
(8.2)
As O s nO :
N
O S T k e OST
MOS Px d N ¦ k!
(8.3)
k 0
where N is the number of items held in stock; O s rate is the failure system and T
is the time interval.
Finally, there is a procedure for calculating the number of spare parts of N, for
some Drisk of stockout or the MOS, so that, respectively:
Px ! N D (8.4)
or
Px d N t MOS (8.5)
Therefore, the procedure consists of finding each of the possible values of N,

starting from N 0 (no spare parts) until the first value of N is found that meets
the condition of keeping the risk within the limit established.
8.2.3 Approach Based on the Risk of Inventory Shortages by using

Prior Knowledge
There are practical situations where it is not possible to obtain the values for the
parameters of reliability and/or maintainability of a system. This approach
provides a procedure for sizing the need for spare parts where at least one of these
parameters is not known (de Almeida 1996; de Almeida and Souza 2001).
In such cases, prior knowledge is used (as discussed in Chap. 3) with respect to
the reliability and/or maintainability of the system. Therefore, the prior probability
is applied to obtain the expected values of risk or MOS in order to determine what
the appropriate number of spare parts to be held in stock should be.
For this study, three scenarios are considered:
x Lack of knowledge about the failure rate O ;
x Lack of knowledge about the MTTR (Mean Time to Repair);
x Lack of knowledge about the parameter O and MTTR.
In the first case, the absence of O, one obtains the prior probability of O: S O ;
in the second case, one should obtain the prior probability on the MTTR:
S MTTR . I.e., these two functions of prior probabilities are required. Therefore,
to address the problem of sizing the need for spare parts in the absence of data, it
is considered the expected value of MOS data as being derived from previously
defined situations, respectively:
§ N nOT k e nOT ·
EO >MOS @ ³ MOS S O dO ³¦¨ ¸S O dO (8.6)
¨ k! ¸
O O ©k 0 ¹
E MTTR >MOS @ ³ MOS S MTTRdMTTR

MTTR
(8.7)
§ N nO MTTR k e nO MTTR ·
³ ¦ ¨
¨
MTTR © k 0
k!
¸S MTTR dMTTR
¸
¹
ª º
EO , MTTR >MOS @ ³ ³ « MOS S O dO »S MTTR dMTTR (8.8)
«
MTTR ¬ O
»¼
As shown in Chap. 3, the procedures for eliciting prior knowledge about

parameters of interest, based on the Equiprobable Intervals Method (Raiffa 1968),
is a very viable alternative.
8.2.4 Approach under the Cost Constraint
In this approach, the attribute of value is treated directly. The cost criterion is seen
as a limiting factor, since for a given cost limit, one tries to minimize the risk of
breakage of stock shortages, i.e., starting from the amount of (monetary) resources
that have been allocated in order to determine the optimal number of inventory
items that should be held (Goldman and Slatery 1977).
The decision process is to determine the threshold value of cost, which depends
on the availability of resources. This determines what the number of spare parts of
N is that minimizes the risk of stock shortages.
Therefore, an expression for calculating the number of spare parts of N needed,
so that:
CT d C0 (8.9)
where CT is the final total cost and C 0 is the amount of resources available is:
CT N .C (8.10)
where N is the number of spare parts and C is the unit cost of each item. The
procedure consists of finding the value of N, starting from N 0 (no spare parts),
that minimizes the risk of inventory shortages that meets the condition previously
established by the budget constraint.
In more complex situations, such as a modularized system that has equipment
with a number of J different types of modules, where each module has its own Oj
failure rate. The final total cost is obtained by summing the total final costs of
each module:
J
CT ¦ N j .C j (8.11)
j 0
where Cj is the individual cost of the module type j; and, Nj is the number of
modules (items) of type j.

For this approach, a MOS in which N N 1 ,! , N j is considered to represent
the probability that there will be no stock shortages of N, i.e., this is given by the
product operator between the MOS modules:
J
MOS N Px j d N j (8.12)
j 0
Therefore the solution is to maximize MOS N , such that the total cost is less
than or equal to the initial cost imposed as a constraint, CT N d C 0 . For this, one
needs to use non-linear optimization.
8.2.5 Use of MCDM/A Model
This approach comes from the perspective of multidimensionality (de Almeida

2001; de Almeida 1996). Multiple objectives can be aggregated to decision
models such as by taking into account maximizing system revenues and
minimizing the volume of total spares. However, the fact remains that when
attempting to optimize any design aspect of an engineered system, the analyst is
frequently faced with the demand of achieving several targets (e.g. low costs, high
revenues, high reliability, low accident risks), some of which may very well be in
conflict with each other. At the same time, several peculiar requirements (e.g. in
spacecraft systems, maximum allowable weight, volume, etc.) should also be
satisfied (Marseguerra et al. 2005).
Unlike a single objective approach, which often implies the poor performance
of other desired objectives, the set identified by a multi-objective approach
provides a spectrum of ‘acceptable’ solutions and attempts made to find a
compromise. This is one of the advantages of working under the multidimensional
aspect (multiobjective).
According to Jajimoggala et al. (2012), a systematic evaluation of the criticality
of spare parts is the key to effective control of spare parts recorded by inventory
systems. There are many factors to be considered which will measure the
criticality of spare parts for maintenance activities, and these evaluation pro-
cedures involve several objectives and it is often necessary to make compromises
among possibly conflicting tangible and intangible factors. In their study,
Jajimoggala et al. (2012) use an MCDM approach to solve this kind of problem,
namely by using a three-phase hybrid model: the first stage involves identifying
the criteria; the second is to prioritize the different criteria using fuzzy ANP; and,
finally in the third phase, the criticality of spare parts is ranked using fuzzy
TOPSIS.
Molenaers et al. (2012) propose a spare parts classification method based on
the criticality of items. Starting from a multicriteria analysis, the proposed model
converts relevant criteria impacting the criticality of an item into a single score
which thus represents the criticality level. This level is used to rationalize the
efficiency of the spare parts inventory policy. They consider the following criteria:
x The criticality of equipment;
x The probability of an item failing;
x Replenishment time;
x The number of potential suppliers;
x The availability of technical specifications, and;
x Maintenance type.
de Almeida (2001) considers two criteria (risk and cost), which are combined
through a multi-attribute utility function in a decision model for provisioning
spares, i.e., spares provisioning can also be modelled by the multicriteria utility
function (by the MAUT method) based on the need for spares and the risk of no
supply.
MAUT has been rarely used for the spares provisioning problem. Several
criteria such as: availability, risk, and cost are used to estimate the volume of
spares needed. Risk is a common criterion used in Mickel and Heim (1990). Other
models optimize a single criterion such as availability or risk subject to costs
(Goldman and Slattery 1977; Barlow et al. 1996).
The combination of these two attributes will be made through the utility
function (multiattribute function). The decision to be adopted in this approach is to
determine values for the attributes of cost C and risk D in order to maximize
the multi-attribute utility function of consequence U C , D .
By concepts of Decision Theory, one has the space of action a , which
consists of the possible quantities of spare parts N, which is the element on which
the decision maker can act in order to achieve the desired goal, in which case it is
the maximization of the multiattribute utility function U C , D .
Furthermore, the state of nature T , the reliability of the system and the
maintainability of its structure need to be considered. They can be represented by
the parameters of reliability and maintainability, which can be obtained by using a
statistical procedure or the use of experts’ prior knowledge, as previously
mentioned (de Almeida 2001; de Almeida 1996).
The observation data (obtained by an analysis of likelihood) concerning the
reliability and maintainability of the system under study allows some
considerations about the behavior of the state of nature T . The state of nature
has a direct influence on the results of the consequences of the decision made by
the decision maker, but the decision maker does not have any control or influence
over the state of nature.
The consequence space is given by the expected utility value E >u C , D @ . The
function u C , D is obtained using the procedure for eliciting a multi-attribute
utility function (described in Keeney and Raiffa (1976)), which defines the DM’s
preference structure with respect to the values of cost and risk of stock shortages.
Finally, the goal of the approach is to determine the number of N spare parts,
which maximizes u C , D . The mathematical model is given by:
u T , a E p|T ,a u p ³ u p P p | T , a dp ³ uC, D PC, D | T , a dp (8.13)
where PC , D | T , a is a consequence function given the decision-maker adopted

an action a (defined by the combination of a certain value and risk D and cost C)
and the state of nature T that had occurred.
It is emphasized that the cost of spare parts depends exclusively on the action
chosen to maximize the utility function, and there is not, for this case, dependence
on the state of nature C O , T .
On the other hand, the Į risk depends on the state of nature T and action a
to be adopted by the decision maker, and thus there is no dependency between the
attributes, thereby allowing the use of the conditional probability function
P p | T , a as follows:
P p | T , a PC , D | T , a PC | T , a PD | T , a (8.14)
For every action ai determined by the decision maker there is an associated

cost, so having a deterministic view of the result of the cost function:
PCi | T , a 1 iff , a ai (8.15)
Hence,
P p | T , a PC | T , a PD | T , a 1 PD | T , a PD | T , a (8.16)
For PD | T , a , risk D 1 PO , N , T ; therefore, with the values from O, N

and T, the value of risk D can be determined. Thus similarly:
N
nOT k e nOT
° P D | T , a
°
P D 1 MOS | T , a 1 iff, MOS ¦
k 0
k!
® (8.17)
N
° nOT k e nOT
° P D | T , a
¯
P D 1 MOS | T , a 0 iff, MOS z ¦
k 0
k!
The behavior of the random number x to a system fault variable is represented

by a Poisson probability distribution, due to the fact that the failure rate parameter
shows a constant behavior in time function, given that the reliability function is
represented by the function exponential probability. Likewise, in a deterministic
view, PD 1 MOS | T , a 1 .
Therefore, the maximization of u C , D is obtained through a deterministic
approach, where u T , ai u p | T ,a i that consists in determining the number of
N spare parts to be available in stock.
Among the criteria of Decision Theory for maximizing a utility function one
highlights the Bayesian method, which consists of choosing the action a i , which
is the number of spare parts available in stock in order to maximize expected
utility, u T , ai , depending on the prior probability S T according to the
following formulation:
max u ai , T S T dT
³T (8.18)
ai
In this model, one considers that the state of nature has two dimensions: one
that matches the reliability of the equipment comprising the system, represented
by the rate of system failures (Os); and the second dimension is the maintainability
8.3 Multiple Spare Parts Sizing 285
of a repairable system, represented by the mean time to repair (MTTR). Therefore,

the probability distribution of the state of nature S T is defined as S MTTR . So
the expected utility can be expressed by:
Tmax O max
E p |T , a >u p @ EC ,D |T , a >u C , D @ ³ O³ uO , T ; N S O S T dOdT (8.19)

T0 0
Therefore, maximizing the multi-attribute utility function is obtained by

maximizing the expected utility function, depending on the number of spare parts:
>
max EO ,T >u O , T ; N @
N
@ (8.20)
Formulated mathematically as:
ªTmax O max º
max « ³ ³ u O , T ; N S O S T dOdT », iff,
N « »
¬ T0 O 0 ¼ (8.21)
N
nOT k e nOT
D 1 ¦
k 0
k!
and C a i Ci
8.3 Multiple Spare Parts Sizing
Spare parts management usually considers issues of a single item and an

independent decision problem of other system items. However, many items that
require spare parts to be available compete for the same resources. For example,
someone may evaluate the possibility of decreasing the number of spare parts of a
given item to balance the increase in the number of spare parts of another item. It
should be noted that these alternatives can have distinct global performance
measures.
When considering the modeling of multiple spare parts simultaneously, the
maintenance manager is not only interested in defining the optimal number of
each item. In this case he is interested in finding an optimized allocation of
resources, given that distributing a limited amount of resources among various
items is considered a typical portfolio problem.
In this context, several papers address issues of spare parts policy using multi-
objective genetic algorithms. Marseguerra et al. (2005) explore the possibility of
using genetic algorithms to optimize the number of spare parts in a
multicomponent system. The objectives considered are the maximization of
system revenues and the minimization of the total volume of spares. A Monte
Carlo simulation approach was defined to deal with system failure, repair and
replacement stochastic processes. Ilgin and Tunali (2007) propose an approach
using genetic algorithms to optimize preventive maintenance and spares policies
of a manufacturing system operating in the automotive sector, while Lee et al.
(2008) develop a framework that integrates a multi-objective evolutionary
algorithm (MOEA) with a multi-objective computing budget allocation (MOCBA)
method for the multi-objective simulation optimization problem of allocating
spare parts for aircraft.
In general, the maintenance manager is interested in minimizing the total cost
of spare parts and also minimizing the probability of stockout. In this section, a
multi-objective genetic algorithm is proposed to tackle the multi spare parts
problem. Firstly, it is important to point out that this model assumes the ‘fixed’
shape of the failure rate. Finkelstein and Cha (2013) state that this assumption is
well founded for the spare parts setting. This feature of a failure rate makes sense
only for spare parts used in corrective maintenance, rather than parts used in
preventive maintenance for which consumption can be defined by a periodic
replacement strategy. This setting justifies the use of the Poisson distribution for
the computations of the probability of stockout of an item. It is also assumed that
each item has a failure rate and purchase cost. Each item can be classified into two
levels of importance to the system, in order to manage different levels of criticality
of items to the system. There are critical and non-critical items which compete for
the same resources of a limited budget.
A multi-objective model based on NSGA-II is developed to aid the
management of multiple spare parts. The model was tested in an urban passenger
bus transport company.
8.3.1 The Mathematical Model
The mathematical model proposed for the spare parts inventory problem was a
multi-objective optimization model, where the objectives are to optimize the
average of the probability of stockout, and the total cost of the spare parts
purchased, which should be minimized.
The model aims to answer the main question inherent in any process of
inventory management: what is the ideal inventory level for a spare part which can
be obtained at minimum cost and provide maximum availability.
As shown by Kennedy et al. (2002) and Bevilacqua et al. (2008), the Poisson
distribution is the most widely-used mathematical-statistical model in the
literature for optimizing inventories of spare parts, and is premised on modelling
the behavior of demand for the item by a probability distribution, which is widely
used to describe rare random events. The Poisson distribution is represented by
(8.22):
(Ot ) x e Ot
Px (t ) (8.22)
x!
where x represents the consumption of replacement parts by time interval for

which the wish is to estimate the probability; t is the time interval considered; Ȝ is
the historical consumption rate of the replacement parts by unit of time; and, Px(t)
is the probability of there being x requests for replacement parts during time
interval t.
The model can be represented as follows:
Ni
( O it ) i e Oit
PS i P ( x ! N i ) 1 Px d N 1 MOS 1 ¦ i!
(8.23)
i 0
where PS i is the probability of stockout of the i-th spare part associated with the
quantity N i ; i represents each spare part (critical or non-critical); N i is the
amount in stock of each spare part; Oi is the monthly rate of consumption of the i-
th spare part; and, C i is the unit cost of the i-th spare part.
Hence, the problem consists of minimizing the objective functions (8.24) and
(8.25).
ª n º
¦
min « PS i »
«¬ i 1 »¼
(8.24)
ª n º
¦
min « C i »
«¬ i 1 »¼
(8.25)
The genetic algorithm used to solve the spare parts inventory problem was an
adaptation of the elitist multi-objective genetic algorithm proposed by Deb et al.
(2002), NSGA II. The algorithm is based on sorting the chromosomes based on
non-dominance to find the Pareto front of multi-objective problems, as well as to
maintain the good solutions during the evolutionary process, since it is an elitist
algorithm. In the proposed algorithm, the length of the chromosome is equal to the
number of different spare parts, where each gene represents the amount to be
purchased.
The selection of parents for crossover operation to generate offspring is
random, and, for each set of parents chosen, two descendants are generated using
the genetic operators. The crossover operator is based on random selection of a
position of the chromosome that is the cutoff point. The offspring 1 inherit the
genes of parent 1 until the cutoff position and, from there on, they inherit the
genes of parent 2. Additionally, the offspring 2, thus, inherit the genes of the
parent 2 until the cutoff point, and, from that point on, inherit the genes of parent 1.
8.3.2 Case Study
For the case study, the procedure followed was: (1) to define the components of
the replacement spare parts of the buses to be studied, such that 14 critical items
used were defined in corrective maintenance actions alone, and (2) to define the
parameters to be quantified, which in this case were the consumption rate Ȝ, the
unit price and the initial stock. These parameters are shown in Table 8.1. Items
which cause transport service failure were determined as critical items.
The data were collected from an urban collective public transport company that
has been operating buses for more than 25 years and is regarded as anonymous in
this study. This company has a fleet of 83 buses, the average age of which is 5.69
years, which run 600,000 km per month.
Table 8.1 Initial data from the critical items
Part Ȝ Unit Cost (C) Initial Stock

P1 0.636 168.00 0
P2 0.364 660.00 0
P3 0.727 2,700.00 0
P4 1.0 1,843.00 0
P5 0.364 23.00 0
P6 2.273 882.00 0
P7 2.727 1,176.00 1
P8 1.364 136.00 0
P9 0.909 200.00 0
P10 18.0 1,180.00 13
P11 7.636 380.00 4
P12 85.455 14.49 74
P13 1.273 268.00 0
P14 3.0 30.00 1
Initially, the algorithm was run for the critical items. 99.9% was established as
the upper limit of the average of probability of stockout, which ensures a high
quality of service obtained by the purchase of critical items. As mentioned in the
previous section, the initial solution of the genetic algorithm was generated as the
result of applying the model based on cost benefit ratio (CB), which is obtained
through the cost ratio by varying the level of service caused by the purchase of
spares. The algorithm based on cost-benefit presented a total of 142 solutions in
the Pareto front. The population size chosen for use in NSGA II was twice the
amount of solutions obtained by the CB model, i.e., 284. The first 142 chromo-
somes of the initial solution are the same chromosomes obtained by the CB
model, and the other half of the chromosomes is random generated, so that the
diversity in the solutions is preserved.
After 250 iterations of the genetic algorithm, a total of 276 solutions on the
Pareto front are obtained. Of this number, only 21 coincide with the solutions
generated by the CB model, which shows that the genetic operators have
diversified the initial solution a lot. If analyzed together, the two models generated
a total of 397 different solutions, of which 363 are non-dominated. A comparative
graph of the solutions of the model based on cost-benefit and NSGA II for critical
items is shown in Fig. 8.1 and Fig 8.2.
Total cost x probability of stockout of spare parts - CB

0.8
0.7
0.6
Probability of Stockout
0.5
0.4
0.3
0.2
0.1
0
0 10000 20000 30000 40000 50000 60000 70000
Total Cost
Fig. 8.1 Total cost versus probability of stockout for critical items (Cost-benefit)
Total cost x probability of stockout of spare parts - NSGA II

0.8
0.7
0.6
Probability of Stockout
0.5
0.4
0.3
0.2
0.1
0
0 10000 20000 30000 40000 50000 60000 70000
Total Cost
Fig. 8.2 Total cost versus probability of stockout for critical items (NSGA-II)
In Fig. 8.1 and Fig. 8.2, it can be realized that from an PS (Probability of
Stockout) of 10% ahead, it is soon seen that there is a “saturation” in the curve,
thus reversing the prevailing logic, i.e. there are then high investments for little
return (low reduction in the probability of stockout), which clearly it is not worth
the company’s spending resources on, in this situation.
The option to deal separately with the critical items allows the manager to have
greater flexibility in managing the contingency element of his/her budget, and
certainly yields a better result for inventory management as it allows the logic of
the program, based on the typical problem of a portfolio of assets, in which several
items, within their group of criticality, compete for resources simultaneously, thus
gaining the one that presents the lowest cost-benefit index, which brings a gain to
the operation as a whole.
It can be concluded that the model developed and applied in a real situation
reached its objective, as it allowed important parameters for controlling the
inventory of replacement spare parts to be monitored efficiently, thus contributing
to the management of an urban bus company. It is further understood that this
model can be replicated in any other company which has replacement spare parts
in its inventory and consumes them when carrying out corrective maintenance.
8.4 Spare Parts for CBM
Probability of failure, inspection period, holding cost and obsolescence are crucial
factors in modeling spare parts inventories. In terms of the maintenance policy,
one can argue that condition monitoring may well give a better forecast of the
residual life of the system monitored and can support better decisions about
8.4 Spare Parts for CBM 291
acquiring spares, in the context of Condition-Based Maintenance (CBM). The

demand for spare parts is commonly generated by the need for preventive
maintenance actions and by failures. Besides, maintenance costs are influenced by
the availability of spare parts. It needs to be borne in mind that penalties due to
spare parts being unavailable usually consist of the cost of, for example, extended
downtime and the high costs of acquiring spare parts in emergency situations.
Technical advances in condition monitoring techniques have provided a means to
ensure high availability and to reduce scheduled and unscheduled production
shutdowns (Ferreira and Wang 2012; Wang 2012; Wang 2008).
Studies on spare parts dealing with failure based maintenance, age- or block-
based replacement policies have been of interest to several researchers. A review
of the literature on spare parts inventories was conducted by Kennedy et al.
(2002). They set out how research directions were conducted on this theme,
although no CBM model was found to be used in spare part inventory control.
In terms of age-based replacement, a comparative study between optimal
stocking policy and the Barlow–Proschan age replacement policy shows the cost
effectiveness of the former. Joint stocking and age-based replacement policy were
studied by Zohrulb Kabir and Al-Olayan (1996). Barabadi et al. (2014) evaluated
reliability models with covariates in the field of spare part predictions. Van
Horenbeek et al. (2013) proposed a joint maintenance and inventory policy model
based on predictive information in order to evaluate the added value of predictive
information (RUL) for multi-component systems.
Rezg et al. (2008) proposed a joint optimal inventory control and preventive
maintenance policy subject to a required minimum level of availability. Diallo
et al. (2008) suggested a mathematical model which aims at maximizing the
availability of a system under a budget constraint where the parameters for placing
orders and the intervals of preventive maintenance are derived, based on the
lifetime distribution of the system. Vaughan (2005) assumed that the demands for
spare parts due to regularly scheduled preventive maintenance and the random
failure of units in service are independent. Chang et al. (2005) proposed an
inventory model for spare parts taking into account the criticality of the pro-
duction equipment. Aronis et al. (2004) applied a Bayesian approach to forecast
demand, based on prior distributions of the failure rates, where the number of
spare parts is determined for a required level of service.
CBM strategies should be integrated with traditional models to indicate when
and how many spares are needed. A hybrid of simulation and analytical models is
proposed taking into account the residual life of equipment estimated by using
condition monitoring techniques. The advantages of CBM include reducing the
cost of the inventory, making better predictions of and planning for the volume of
spares required, since the residual life can be better predicted by condition infor-
mation, which can lead to better forecasting of the quantity of spare parts needed.
In CBM modelling, it is important recognize two fundamental classes of
problems. Wang (2008) explains the concepts of direct and indirect monitoring. In
direct monitoring, the actual condition of the item can be observed, and a critical
level can be set up. While in indirect monitoring, one can only collect measure-
ments related to the actual condition of the item monitored in a stochastic manner.
Some enhancements to direct monitoring have been made. Rausch and Liao
(2010) develop a model for joint production and spare part inventory based on
CBM, where the condition monitored can be observed directly. Wang et al. (2009)
present the concept of condition-based replacement and spares provisioning
policy, and through the simulation method and the genetic algorithm, the decision
variables were jointly optimized for minimizing the cost rate. Linear and expo-
nential degradation models is evaluated by Elwany and Gebraeel (2008) in order
to support the dynamic decisions of replacement and inventory based on the
physical condition of the equipment. Ferreira et al. (2009) propose a multicriteria
decision model to determine inspection intervals of condition monitoring based on
delay time analysis.
Ferreira and Wang (2012) assume that there are a number of identical compo-
nent items used in a system, which are condition monitored periodically. For
example, there may be many critical and identical bearings installed on a paper
machine and proper maintenance of these bearings should lead to better availability
and lowering the operating costs of the machine as a result of having condition
monitoring information. Having the appropriate volume of spare parts available at
the right time is a relevant issue when managing maintenance activities.
Opportunities for maintenance actions such as condition monitoring and
preventive maintenance times, likewise the order time and arrivals of acquisitions,
are illustrated in Fig. 8.3 in order to represent the main features of the problem.
CM1 CMi t
PM1 PMk t
OT1 OTn t
t
AT1 ATn
Fig. 8.3 Intervals of condition monitoring (CM), preventive maintenance (PM), order time (OT),
order arrival (AT) and lead time (W) of spare parts
8.4 Spare Parts for CBM 293
Thus there is a decision problem at each replacement opportunity as shown in

Fig. 8.4. Basically there are alternatives:
1. Alternative 1 – Replacements at present moment, subject to a stock level;
2. Alternative 2 – Replacements at the next condition monitoring opportunity
(CMi), subject to probability of failures and stock level;
3. Alternative 3 – Replacements at the next order arrival time opportunity (ATn),
subject to probability of failures and stock level;
4. Alternative 4 – Replacements at the next preventive maintenance opportunity
(PMk), subject to probability of failures and stock level;
Immediate replacement
Stock level > 0
Stock level = 0
Replacement
Stock level > 0
at the next CMi
Failure before CMi
Stock level = 0
No Failures before CMik
Stock level > 0
Stock level = 0
Replacement
Stock level > 0
at the next ATm
Failure before ATm
Stock level = 0
No Failures before ATm
Stock level > 0
Stock level = 0
Replacement
Stock level > 0
at the next PMk
Failure before PMk
Stock level = 0
No Failures before PMik
Stock level > 0
Stock level = 0
Fig. 8.4 Decision tree at each replacement opportunity

From the decision tree structure of Fig. 8.4, it is possible to evaluate dynamically
the performance of a given maintenance policy by comparing the results and
analyzing the replacement times. Based on the monitored information, the structure
of the risk of stockout, costs and estimates of the residual life are derived. These
estimates may vary which implies that the need for spare parts may change.
This section addresses a spare part problem by using condition monitoring
information. CBM is a more cost effective maintenance policy than time-based
maintenance since it can avoid premature maintenance or replacement while
making better forecasts of the need for spare parts. In traditional CBM models,
there is a strong assumption that spare parts are always available when needed,
and in several practical situations this is not true.
References
Aronis K-P, Magou I, Dekker R, Tagaras G (2004) Inventory control of spare parts using a
Bayesian approach: A case study. Eur J Oper Res 154:730–739
Barabadi A, Barabady J, Markeset T (2014) Application of reliability models with covariates in
spare part prediction and optimization - A case study. Reliab Eng Syst Saf 123:1–7
New York
Ben-Daya M, Duffuaa SO, Raouf A, et al. (2009) Handbook of Maintenance Management and
Engineering. Springer, London
Bevilacqua M, Ciarapica FE, Giacchetta G (2008) Spare parts inventory control for the
maintenance of productive plants. 2008 IEEE Int. Conf. Ind. Eng. Eng. Manag. IEEM 2008.
IEEE, Singapore, pp 1380–1384
Boylan JE, Syntetos AA (2010) Spare parts management: A review of forecasting research and
extensions. IMA J Manag Math 21:227–237
Chang PL, Chou YC, Huang MG (2005) A (r,r,Q) inventory model for spare parts involving
equipment criticality. Int J Prod Econ 97:66–74
de Almeida AT (1996) Multicriteria for spares provisioning using additive utility function.
In: International Conference on Operational Research for Development, IFORS-ICORD II.
Rio de Janeiro, RJ, pp 1414–1418.
de Almeida AT (2001) Multicriteria decision making on maintenance: Spares and contracts
planning. Eur J Oper Res 129:235–241
Diallo C, Ait-Kadi D, Chelbi A (2008) (s,Q) Spare parts provisioning strategy for periodically
replaced systems. IEEE Trans Reliab 57:134–139
Duchessi P, Tayi GK, Levy JB (1988) A Conceptual Approach for Managing of Spare Parts. Int
J Phys Distrib Logist Manag 18:8–15
Elwany AH, Gebraeel NZ (2008) Sensor-driven prognostic models for equipment replacement
and spare parts inventory. IIE Trans 40:629–639
References 295
Ferreira RJP, de Almeida AT, Cavalcante CAV (2009) A multi-criteria decision model to
determine inspection intervals of condition monitoring based on delay time analysis. Reliab
Ferreira RJP, Wang W (2012) Spare parts optimisation subject to condition monitoring. In: 11th
International Probabilistic Safety Assessment and Management. Conference and the Annual
European Safety and Reliability Conference, Helsinki, Finland, 25-29 June 2012
Finkelstein M, Cha JH (2013) Stochastic Modeling for Reliability: Shocks, Burn-in and
Heterogeneous populations. Springer, London
Gopalakrishnan P, Banerji AK (2013) Maintenance and Spare Parts Management. PHI Learning,
New Delhi
Ilgin MA, Tunali S (2007) Joint optimization of spare parts inventory and maintenance policies
using genetic algorithms. Int J Adv Manuf Technol 34:594–604
Jajimoggala S, Rao VVSK, Beela S (2012) Spare parts criticality evaluation using hybrid
multiple criteria decision making technique. Int J Inf Decis Sci 4:350
Kabir ABMZ, Al-Olayan AS (1996) A stocking policy for spare part provisioning under age
based preventive replacement. Eur J Oper Res 90:171–181
Kennedy WJ, Wayne Patterson J, Fredendall LD (2002) An overview of recent literature on
spare parts inventories. Int J Prod Econ 76:201–215
Lee LH, Chew EP, Teng S, Chen Y (2008) Multi-objective simulation-based evolutionary
algorithm for an aircraft spare parts allocation problem. Eur J Oper Res 189:476–491
Macchi M, Fumagalli L, Pinto R, Cavalieri S (2011) A Decision Making Framework for
Managing Maintenance Spare Parts in Case of Lumpy Demand: Action Research in the
Avionic Sector. In: Altay N, Litteral LA (eds) Serv. Parts Manag. pp 89–104
Marseguerra M, Zio E, Podofillini L (2005) Multiobjective spare part allocation by means of
genetic algorithms and Monte Carlo simulation. Reliab Eng Syst Saf 87:325–335
Mickel LS, Heim RL (1990) The spares calculator: a visual aid to provisioning. Annu. Proc.
Reliab. Maintainab. Symp. IEEE, Los Angeles, CA, pp 410 – 414
Molenaers A, Baets H, Pintelon L, Waeyenbergh G (2012) Criticality classification of spare
parts: A case study. Int J Prod Econ 140:570–578
Padmanabhan G, Vrat P (1990) Analysis of multi-item inventory systems under resource
constraints: A non-linear goal programming approach. Eng Costs Prod Econ 20:121–127
Porras E, Dekker R (2008) An inventory control system for spare parts at a refinery: An
empirical comparison of different re-order point methods. Eur J Oper Res 184:101–132
Wesley, London
Rausch M, Liao H (2010) Joint production and spare part inventory control strategy driven by
condition based maintenance. IEEE Trans Reliab 59:507–516
Rezg N, Dellagi S, Chelbi A (2008) Joint optimal inventory control and preventive maintenance
policy. Int J Prod Res 46:5349–5365
Roda I, Macchi M, Fumagalli L, Viveros P (2014) A review of multi-criteria classification of
spare parts: From literature analysis to industrial evidences. J Manuf Technol Manag 25:528–
549
Syntetos A, Keyes M, Babai M (2009) Demand categorisation in a European spare parts logistics
network. Int J Oper Prod Manag 29:292–316
Van Horenbeek A, Scarf P, Cavalcante CAV, Pintelon L (2013) On the use of predictive
information in a joint maintenance and inventory policy. In: Steenbergen RDJM, VanGelder
PHAJM, Miraglia S, Vrouwenvelder ACWMT (eds) 22nd Annual Conference on European
Safety and Reliability (ESREL), Amsterdam, 2013. Safety, Reliability and Risk Analysis:
Beyond the Horizon. Taylor & Francis Group, London, UK, p 758
Van Volkenburg C, Montgomery N, Banjevic D, Jardine A (2014) The effect of deterioration on

spare parts holding. 2014 Reliab. Maintainab. Symp. IEEE, Colorado Springs, CO, pp 1–6
Vaughan TS (2005) Failure replacement and preventive maintenance spare parts ordering policy.
Eur J Oper Res 161:183–190
Wang L, Chu J, Mao W (2009) A condition-based replacement and spare provisioning policy for
deteriorating systems with uncertain deterioration to failure. Eur J Oper Res 194:184–205
Wang W (2008) Condition-based maintenance modelling. In: Kobbacy KAH, Prabhakar Murthy
DN (eds) Complex Syst. Maint. Handb. Springer London, pp 111–131
Wang W (2012) A stochastic model for joint spare parts inventory and planned maintenance
optimisation. Eur J Oper Res 216:127–139
Chapter 9
Decision on Redundancy Allocation
Abstract: Redundancy allocation is a decision that involves assessing and choosing

where to locate additional components or subassemblies, above the minimum
required for an existing system to operate, in order to promote the system’s
reliability. Specifically, the field of multi-objective redundancy allocation has
received several contributions since the 1970s and the combinatorial complexity
of these problems has mainly encouraged researchers to develop search algorithms
focused on the Pareto front definition, the most frequent approach in this literature.
Finding a set of non-dominated solutions based on heuristics is a step that demands
much computational effort to solve the problem. Despite these difficulties, the
DM’s preferences should be evaluated in order to recommend a solution that
represents the best compromise among the criteria considered, such as reliability,
cost and weight. This chapter covers redundancy allocation problems from a multi-
criteria perspective. Therefore, basic concepts related to the typical criteria and
tradeoff in redundancy allocation problems are presented and a brief review of the
literature on MCDM/A redundancy allocation is given. To illustrate the MCDM/A
approach for redundancy allocation, a decision model considering a standby
system based on Multi-attribute Utility Theory (MAUT) is presented including the
DM’s behavior to risk (prone, neutral and risk averse). The problem approached in
this chapter involves a question about how to select a suitable maintenance
strategy in order to evaluate the tradeoff between a system’s availability and cost,
including experts’ prior knowledge to deal with the uncertainty of failure and
repair rate parameters.
9.1 Introduction
Redundancy allocation is a decision that involves assessing and choosing where to

locate additional components or subassemblies, above the minimum required for
an existing system to operate, in order to promote system reliability. This theme is
one of the classic issues in reliability theory, in which the system design seeks to
balance fundamental factors such as reliability, cost and weight. The balance amongst
these factors has been the subject of research since the classic publications on
reliability theory, such as that by Barlow and Prochan (1965).

298 Chapter 9 Decision on Redundancy Allocation
Barlow and Prochan (1965) presented models involving redundancy and revealed
how to allocate redundancy among the various subsystems under linear constraints
on weight, volume and cost, in order to maximize system reliability. The problem
of maximizing the reliability of a series system subject to one or more constraints
on total cost, weight, volume, has a number of variations depending on whether the
redundancy is parallel or standby. In parallel systems, redundant units operate
simultaneously, and they are subject to failure. In standby redundancy, redundant
units are place on stand-by as spares and used successively for replacement, and they
are not subject to failure while in standby condition. In some active parallel redundant
configurations, it may be required that k out of the n units must be working for the
system to function. The reliability of a k out of n system, with n independent
components in which all the unit reliabilities are equal, is expressed by the
binomial reliability function (O’Connor and Kleyner 2012; Kuo and Zuo 2003).
Even though Barlow and Prochan (1965) did not use the terminology of an
MCDM/A problem, they indicated a class of problems in which no specific set of
constraints is provided. In this case, one may wish to generate a family of non-
dominated allocations in terms of reliability and cost both in parallel as well as
standby redundancy.
The concept of non-dominated solutions is defined by Barlow and Prochan
(1965) as: x 0 ( x10 , x20 ,..., xn0 ) is non-dominated if R(x)>R(x0) implies cj(x)>cj(x0)
for some j, whereas R(x) = R(x0) implies either cj(x)>cj(x0) for some j or
n
cj(x)=cj(x0) for j, where c j ( x) ¦c x ij i .
i 1
This property is the same as the classical definition of the Pareto Front used in
multiobjective formulations presented in Chap. 2. Barlow and Prochan (1965)
stated that if the set consisting of all non-dominated redundancy allocations is
obtained (the complete family of non-dominated redundancy allocations), then the
solution of a redundancy allocation problem with a set of constraints must be a
member of this family. In other words, Barlow and Prochan (1965) realized that
the mono-objective formulation is a particular case of the MCDM/A formulation
and the solution of a mono-objective case is one from the Pareto Front set.
They presented a procedure that is able to generate an incomplete family of
non-dominated allocations in a single cost factor. The procedure is based on the
principle of adding the most reliability obtained per dollar spent in each iteration,
starting with no redundancy in the system. For a multiple cost factor case, a simple
weighted function of reliability is proposed and arbitrarily chosen values of
weights are recommended. They also suggested a procedure to find a complete
family of non-dominated allocations based on the dynamic programming algorithm
of Kettelle Jr (1962).
A literature review of the optimal redundancy allocation models was carried
out by Tillman et al. (1977). They classified early references in the field in terms
of optimization techniques and system configurations. Among the optimization
techniques, no MCDM/A approach was cited.
Kuo and Prasad (2000) updated the literature review of Tillman et al. (1977)
and this included identifying if an MCDM/A approach had been analyzed as a
way to help optimize system-reliability. They found twelve papers within this
scope. They stated that an MCDM/A approach was an important but not widely
studied problem in reliability optimization. Although some exact methods can be
used to solve redundancy allocation problems, heuristics used include: ant colony
optimization method; hybrid genetic algorithm and tabu search.
Kuo and Wan (2007) cited multiobjective optimization as a recent topic and
indicated eleven references to this, which have published since 2000 in this field.
They defined four problem structures, namely: 1) The traditional reliability-
redundancy allocation problem; 2) The percentile life optimization problem; 3)
Multi-state system optimization; 4) Multiobjective optimization.
Some kinds of system configuration are defined as shown in Fig. 9.1 and Fig. 9.2.
1 2 N1
IN OUT
1 2 N2
… … …
1 2 NM
Fig. 9.1 Mixed series-parallel system, N components are connected in series, and M such series
connections are connected in parallel to form the system
IN
1 2 3
4 5
OUT
Fig. 9.2 Non series-parallel system

A simple version of the redundancy allocation problem is shown in Fig. 9.3.

It is a series system which regards system reliability as an objective function R(xj).
The system has n Stages in series with xj + 1 independent identical distributed
units in parallel in Stage j. xj is the number of parallel redundant components in
Stage j. cij is the cost of type i of each component in Stage j. The cost types
include monetary values, weight and volume. pj is the reliability of each
component in Stage j. cj is some specified cost of each component in Stage j. r is
the number of cost types considered. It is assumed that all units fail independently.
Stage 1 Stage 2 Stage 3 ... Stage n
... ... ... ...

x1 + 1 x2 + 1 x3 + 1 xn + 1
units in paralel units in paralel units in paralel units in paralel
Fig. 9.3 Structure of a simple redundancy allocation problem
The problem is mathematically stated as follows in (9.1):
n
x j 1
maximize R( x j ) (1 (1 p )
j 1
j )
n
subject to ¦c x
i 1
ij j d c j , j 1,..., r (9.1)
0 d xj d uj, x j integer, j 1,..., n
In mono-objective formulations, it is possible to maximize the reliability of a

system subject to the constraints on the amount of available resources or to
minimize the cost of some resource subject to the constraint that the reliability of
the system must meet a specified reliability target. Cost, weight and volume can
be limited by constraints.
Kuo and Zuo (2003) presented some measures for the importance of a
component, such as structural importance, reliability importance, criticality
importance and relative criticality. These factors can be useful in order to compare
components in terms of their importance to a system.
Kuo and Zhu (2012) defined three types of standby redundancy: hot standby,
warm standby, and cold standby. A hot standby has the same failure rate as the
active component. A cold standby has a zero failure rate. Warm standby implies
that inactive components have a failure rate that is between zero and the failure
rate of active components. A warm standby and a hot standby may fail while in
the standby condition, but a cold standby will not fail.
Some limitations of redundancy allocation problems are important to be

considered. For example, redundant components can be subjected to the same
external loads and common failures modes that limit the effectiveness of the
redundancy (Paté-Cornell et al. 2004).
In terms of classifying models, there are redundancy models which assume that
only two component states are possible: the operating and failed states. But there
are some models assuming more than two component states. These are called
multi-state systems.
According to a literature review (de Almeida et al. 2015) on reliability and
maintenance models based on MCDM/A approaches, 18.8% of the of publications
are related to redundancy allocation. A set of relevant publications of the
MCDM/A redundancy allocation problems is presented in Table 9.1. Most articles
use Reliability, Cost and Weight as optimization objectives. Among the tech-
niques for finding solutions, a diverse range of proposals has been suggested.
Table 9.1 A list of publications on MCDM/A redundancy allocation problems

Reliability
References Other Criteria Search method

Weight
Cost
Multiobjective particle
Khalili-Damghani et al. (2013) X X X
swarm optimization
Fuzzy multiobjective
Garg and Sharma (2013) X X
particle swarm optimization
Decomposition-based
Cao et al. (2013) X X X
approach
Tchebycheff;
Sahoo et al. (2012) X X Lexicographic; Genetic
Algorithms
Safari (2012) X X NSGA-II
Genetic Pareto set
Okafor and Sun (2012) X X
identification algorithm
Khalili-Damghani and Amiri İ-constraint method and
X X X
(2012) data envelopment analysis
Zio and Bazzo (2011a); Zio and Clustering procedure; Level
X X Availability
Bazzo (2011b) Diagrams and MOGA
NSGA-II and data
Li et al. (2009) X X X
envelopment analysis
Multiobjective hierarchical
Kumar et al. (2009) X X genetic algorithm; SPEA2
and NSGA-II
Physical programming;
Tian et al. (2008) X System utility
Genetic algorithms
(continued)
Table 9.1 (continued)
Limbourg and Kochs (2008) X Life distribution Feature models; NSGA-II

Multiobjective multi-state
Taboada et al. (2008) X X Availability
genetic algorithm
Multiobjective ant colony
Zhao et al. (2007) X X
system;
Taboada et al. (2007) X X X NSGA
Variable neighbourhood
Liang et al. (2007) X X
search
Availability and Simulated annealing and
Chiang and Chen (2007) X
net profit genetic algorithms
Physical programming;
System
Tian and Zuo (2006) X X genetic algorithms and
performance utility
fuzzy theory
Salazar et al. (2006) X X NSGA-II
Multiple weighted objective
Subsystem
Coit and Konak (2006) heuristic; linear
reliability
programming
Reliability Genetic algorithms and
Marseguerra et al. (2005) X
estimated variance Monte Carlo simulation
Reliability
Coit et al. (2004) X Weighted sum;
estimated variance
Weighted sum; Genetic
Elegbede and Adjallah (2003) X Availability
algorithms
Fuzzy and multiobjective
Huang (1997) X X
optimization;
de Almeida and Souza (1993);
X Interruption time Multi-attribute utility theory
de Almeida and Bohoris (1996)
Fuzzy goal programming
Gen et al. (1993) X X X
model
Dhingra (1992); Rao and Fuzzy goal-programming
X X X
Dhingra (1992) and goal-attainment
Efficient search
multiobjective
Misra and Sharma (1991) X X X
programming; min-max
concept
Surrogate Worth Trade-off
Sakawa (1980); Sakawa (1981) X X X Volume method and dual
decomposition method
Surrogate Worth Trade-off
Sakawa (1978) X X
method
Inagaki et al. (1978) X X X Interactive Optimization
9.2 An MCDM/A Model for a 2-Unit Redundant Standby System 303
From this set of 35 publications listed in Table 9.1, 23 used Reliability as

objective function (65.7%); 32 used Cost (91.4%) and 17 used Weight (48.6%) on
multiobjective redundancy allocation problems.
Redundancy allocation problems are complex by nature. Chern (1992)
evaluated the computational complexity of allocating reliability redundancy in a
series system and proved that some reliability redundancy optimization problems
are Non-deterministic Polynomial-time hard (NP-hard).
Due to the complexity of the problem, there is a focus of research with
emphasis on use of heuristic methods to find solutions of Pareto fronts. However,
an absence of a preference modeling is relevant shortcoming in the selection
process of alternatives. There is a real need on the part of DM in choosing which
of the set of the Pareto solutions provides the best balance for a given preferences
structure.
9.2 An MCDM/A Model for a 2-Unit Redundant Standby

System
In this section, a decision model (de Almeida and Souza 1993) for a standby
system based on the MAUT is presented. This model addresses the waiting time to
call a repair facility when the first piece of equipment of a 2-unit standby system
fails. The first failure implies only a reliability reduction, not system failure, since
the other unit is still operating. This scheme of waiting-time when the first fault
occurs avoids overtime costs in the repair facility. An expert prior knowledge
approach is lead in order to deal with the uncertainty of the parameters failure and
repair rates. Another decision model (de Almeida and Bohoris 1996) extends this
first model, introducing a Gamma distribution to the repair time. The possible
states for a 2-unit redundant standby system are shown in Fig. 9.4.
The problem involves a question about how to select a suitable maintenance
strategy in order to combine system availability and cost preferences. There is an
assumption that the capacity of repair is limited, and instantaneous repair is not
applicable. An MCDM/A approach can solve the conflicting requirements of
system availability and cost through a multi-attribute utility function taking into
account DM’s preferences over these requirements. In this way, MAUT can also
deal with uncertainty of the consequences.
e0 - 2 units working
failure repair
e1 - 1 unit working
and 1 unit failed
failure repair
e2 - 2 units failed
Fig. 9.4 States for a 2-unit redundant standby system
It is noteworthy that several redundancy allocation models assume that the

system configuration is fixed for a given time horizon, which reflects an emphasis
on design aspects and system reliability, corresponding to a planning stage, prior
to system operation. Moreover, the maintenance actions define a strategy that
would have a balanced way in terms of cost and availability on the system operation
phase. Assuming the design phase the system was planned with redundant units
operating in standby, the time limit in which a repairman perform the repair or
replacement of a failed unit needs to be established. Clearly, there is a conflict
between the cost of maintenance and system availability. The parameters of the
model are given in Table 9.2 (de Almeida and Souza 1993).
Table 9.2 Model parameters
Parameters Description
Ȝ Failure rate of the equipment
ȝ Repair rate of the equipment
a An action, element of the action space, representing the maintenance strategy
e0, e1, e2 State of the system when [0, 1, 2] of the units failed
Ta Decision variable representing the repair delay corresponding to a
T0 Time at which the first failure occurs
T1 Time at which the second failure occurs
Time at which the first-failed unit resumes operation, which could be returning the
T2
system to e0
TTR T2 - Ta
ʌ1(Ȝ) Prior knowledge distribution about Ȝ
ʌ2(ȝ) Prior knowledge distribution about ȝ
Ai Scale parameter of ʌi
(continued)
Bi Shape parameter of ʌi
Ci Cost for ai
FCi Fixed cost for ai
CRi Repair cost-rate for ai
MCRi Mean CRi
U{TI,C} Multi-attribute utility function for interruption time and cost
The assumptions of the model (de Almeida and Souza 1993) are:
1. The probability distribution of failure of two units are identically distributed;
2. Each unit has two states: good and failed;
3. The system is down when no unit is available for operation;
4. There is one repair facility;
5. Failure rate (Ȝ) is constant and the number of failures follows a Poisson
distribution;
6. A unit repaired becomes as good as new;
7. Repair rate is constant (ȝ);
8. If during a repair of a failed unit, the other unit also fails, the latter unit waits
for repair until the first unit is repaired;
9. There is prior knowledge about Ȝ and ȝ represented as prior probability
distributions over these parameters;
10.Failure and repair states are s-independent;
11.The DM has a structure of preferences over the consequence space (TI,C)
according to the axiomatic preferences of the utility theory;
12.C and TI are s-independent;
13.The objective function is to maximize the multi-attribute utility function
U{TI,C}
The decision model building was based on the context of a telecommunication
system of an electric power company with a 2-unit standby redundant system. The
DM’s preference elicitation over consequences (interruption time and cost)
produces a multi-attribute utility function, which is introduced into the decision
model, according to (9.5). The expected utility of alternatives is given by (9.2).
Om Pm
E( O , P ) {U {(O , P ), ai }} ³O ³P S 1 (O ) S 2 ( P ) U {(O , P ), ai }dOdP (9.2)
0 0
where:
S 1 (O ) ( B1 / A1 ) (O / A1 ) B1 1 exp[(O / A1 ) B1 ] (9.3)
S 2 ( P ) ( B2 / A2 ) ( P / A2 ) B2 1 exp[( P / A2 ) B2 ] (9.4)
U {TI , C} K t exp( K kt TI ) Kc U {Ci } (9.5)
Ci FCi (O / P ) CRi (9.6)
§ K P · ª § K kt · º
U {(O , P ), ai } K c U {Ci } ¨¨ t ¸¸ «1 ¨¨ ¸¸ exp( O Tai )» . (9.7)
© K kt P ¹ ¬ © O P ¹ ¼
The problem is solved by applying (9.2) into (9.8).
Maxai ( E(O , P ) {U {(O , P ), ai }}) (9.8)
Prior knowledge about the states of nature can be obtained from prior
distributions of Ȝ and ȝ. There are several prior probability elicitation procedures
available in the literature, such as that given by Winkler (1967). The elicitation
procedure applied is based on equal probable intervals. Based on experts on the
equipment and the system maintainability, respectively, these ʌ1(Ȝ) and ʌ2(ȝ) were
obtained according to (9.3) and (9.4), which are illustrated in Fig. 9.5 and Fig. 9.6.
Prior know ledge about ȝ - ʌ 1 (Ȝ)
5.00E+04
4.50E+04
4.00E+04
3.50E+04
3.00E+04
pdf of ʌ Ȝ)
2.50E+04
2.00E+04
1.50E+04
1.00E+04
5.00E+03
0.00E+00
0 0.00001 0.00002 0.00003 0.00004 0.00005 0.00006 0.00007 0.00008 0.00009 0.0001
Ȝ
Fig. 9.5 Prior knowledge about Ȝ, ʌ1(Ȝ) with A1 = 18.06·10-6 and B1 = 1.68
Prior know ledge about ȝ - ʌ 2 (ȝ)
4.00E+01
3.50E+01
3.00E+01
2.50E+01
pdf of ʌ ȝ)
2.00E+01
1.50E+01
1.00E+01
5.00E+00
0.00E+00
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1
ȝ
Fig. 9.6 Prior knowledge about ȝ, ʌ2(ȝ) with A2 = 0.028 and B2 = 2.57
There are three possible situations in the state e1 (de Almeida and Souza 1993):
x T1 > Ta, and T1 > T2; therefore TI = 0;
x T1 > Ta, and T1 < T2; therefore TI = T2 - T1 > 0;
x Otherwise, there is an emergency, and the repair facility is called immediately,
so that Ta is set equal to T1.
Then, (9.9) represents the interruption time formulation.
TI max(0, min(Ta TTR T1 , TTR )) (9.9)
The set of alternatives for this problem is represented by maintenance strategies

in terms of repair delay, as follows:
x a1 - There is no repair delay, then Ta1 = 0. It is assumed that maintenance
department have infra-structure and resources to repair a unit immediately upon
a failure.
x a2 - There is zero repair-delay only during usual work hours and there is repair-
delay during non-usual work hours, then Ta2 is a random variable between 0
and 14 hours. It is assumed that the expected value of Ta2 is equal to 7 hours.
x a3 - Zero repair-delay is only in the usual work hours, with a cheaper structure,
but the accessibility is lower, then Ta3 is a random variable between 0 and 62
hours. It is assumed that the expected value of Ta3 is equal to 31 hours.
x a4 - A repair delay is allowed so that the resources are shared with other tasks.
Thus, Ta4 = 360 hours.
Fixed cost and repair cost rate (FCi and CRi) were obtained from the company
for these four alternatives and the alternative a3 got the best performance of the
multi-attribute utility function.
References
New York
Cao D, Murat A, Chinnam RB (2013) Efficient exact optimization of multi-objective redundancy
allocation problems in series-parallel systems. Reliab Eng Syst Saf 111:154–163
Chiang C-H, Chen L-H (2007) Availability allocation and multi-objective optimization for
parallel–series systems. Eur J Oper Res 180:1231–1244
Coit DW, Jin T, Wattanapongsakorn N (2004) System optimization with component reliability
estimation uncertainty: a multi-criteria approach. Reliab IEEE Trans 53:369–380
Coit DW, Konak A (2006) Multiple weighted objectives heuristic for the redundancy allocation
problem. IEEE Trans Reliab 55:551–558
de Almeida AT, Bohoris GA (1996) Decision theory in maintenance strategy of standby system
with gamma-distribution repair-time. Reliab IEEE Trans 45:216–219
de Almeida AT, Souza FMC (1993) Decision theory in maintenance strategy for a 2-unit
redundant standby system. Reliab IEEE Trans 42:401–407
Dhingra AK (1992) Optimal apportionment of reliability and redundancy in series systems under
multiple objectives. Reliab IEEE Trans 41:576–582
Elegbede C, Adjallah K (2003) Availability allocation to repairable systems with genetic
algorithms: a multi-objective formulation. Reliab Eng Syst Saf 82:319–330
Garg H, Sharma SP (2013) Multi-objective reliability-redundancy allocation problem using
particle swarm optimization. Comput Ind Eng 64:247–255
Gen M, Ida K, Tsujimura Y, Kim CE (1993) Large-scale 0–1 fuzzy goal programming and its
application to reliability optimization problem. Comput Ind Eng 24:539–549
Huang H-Z (1997) Fuzzy multi-objective optimization decision-making of reliability of series
system. Microelectron Reliab 37:447–449
Inagaki T, Inoue K, Akashi H (1978) Interactive Optimization of System Reliability Under
Multiple Objectives. Reliab IEEE Trans R-27:264–267
Kettelle Jr JD (1962) Least-Cost Allocations of Reliability Investment. Oper Res 10:249–265
Khalili-Damghani K, Abtahi A-R, Tavana M (2013) A new multi-objective particle swarm
optimization method for solving reliability redundancy allocation problems. Reliab Eng Syst
Saf 111:58–75
Khalili-Damghani K, Amiri M (2012) Solving binary-state multi-objective reliability redundancy
allocation series-parallel problem using efficient epsilon-constraint, multi-start partial bound
enumeration algorithm, and DEA. Reliab Eng Syst Saf 103:35–44
Kumar R, Izui K, Yoshimura M, Nishiwaki S (2009) Multi-objective hierarchical genetic
algorithms for multilevel redundancy allocation optimization. Reliab Eng Syst Saf 94:891–
904
Kuo W, Prasad VR (2000) An annotated overview of system-reliability optimization. Reliab
IEEE Trans 49:176–187
Kuo W, Wan R (2007) Recent Advances in Optimal Reliability Allocation. In: Levitin G (ed)
Comput. Intell. Reliab. Eng. SE - 1. Springer Berlin Heidelberg, pp 1–36
Kuo W, Zhu X (2012) Importance measures in reliability, risk, and optimization: principles and
applications. John Wiley & Sons, New York
Kuo W, Zuo MJ (2003) Optimal reliability modeling: principles and applications. John Wiley &
Sons, New York
Li Z, Liao H, Coit DW (2009) A two-stage approach for multi-objective decision making with
applications to system reliability optimization. Reliab Eng Syst Saf 94:1585–1592
References 309
Liang Y-C, Lo M-H, Chen Y-C (2007) Variable neighbourhood search for redundancy allocation
problems. IMA J Manag Math 18:135–155
Limbourg P, Kochs H-D (2008) Multi-objective optimization of generalized reliability design
problems using feature models - A concept for early design stages. Reliab Eng Syst Saf
93:815–828
Marseguerra M, Zio E, Podofillini L, Coit DW (2005) Optimal design of reliable network
systems in presence of uncertainty. Reliab IEEE Trans 54:243–253
Misra KB, Sharma U (1991) An efficient approach for multiple criteria redundancy optimization
problems. Microelectron Reliab 31:303–321
Okafor EG, Sun Y-C (2012) Multi-objective optimization of a series–parallel system using
GPSIA. Reliab Eng Syst Saf 103:61–71
Paté-Cornell ME, Dillon RL, Guikema SD (2004) On the Limitations of Redundancies in the
Improvement of System Reliability. Risk Anal 24(6):1423–1436
Rao SS, Dhingra AK (1992) Reliability and redundancy apportionment using crisp and fuzzy
multiobjective optimization approaches. Reliab Eng Syst Saf 37:253–261
Safari J (2012) Multi-objective reliability optimization of series-parallel systems with a choice of
redundancy strategies. Reliab Eng Syst Saf 108:10–20
Sahoo L, Bhunia AK, Kapur PK (2012) Genetic algorithm based multi-objective reliability
optimization in interval environment. Comput Ind Eng 62:152–160
Sakawa M (1978) Multiobjective reliability and redundancy optimization of a series-parallel
system by the Surrogate Worth Trade-off method. Microelectron Reliab 17:465–467
Sakawa M (1980) Reliability design of a standby system by a large-scale multiobjective
optimization method. Microelectron Reliab 20:191–204
Sakawa M (1981) Optimal Reliability-Design of a Series-Parallel System by a Large-Scale
Multiobjective Optimization Method. Reliab IEEE Trans R-30:173–174
Salazar D, Rocco CM, Galván BJ (2006) Optimization of constrained multiple-objective
reliability problems using evolutionary algorithms. Reliab Eng Syst Saf 91:1057–1070
Taboada HA, Baheranwala F, Coit DW, Wattanapongsakorn N (2007) Practical solutions for
multi-objective optimization: An application to system reliability design problems. Reliab
Taboada HA, Espiritu JF, Coit DW (2008) MOMS-GA: A Multi-Objective Multi-State Genetic
Algorithm for System Reliability Optimization Design Problems. Reliab IEEE Trans 57:182–
191
Tian Z, Zuo MJ (2006) Redundancy allocation for multi-state systems using physical
programming and genetic algorithms. Reliab Eng Syst Saf 91:1049–1056
Tian Z, Zuo MJ, Huang H (2008) Reliability-Redundancy Allocation for Multi-State Series-
Parallel Systems. Reliab IEEE Trans 57:303–310
Tillman FA, Hwang C-L, Kuo W (1977) Optimization Techniques for System Reliability with
Redundancy: A Review. Reliab IEEE Trans R-26:148–155
Winkler RL (1967) The Assessment of Prior Distributions in Bayesian Analysis. J Am Stat
Assoc 62:776–800
Zhao J-H, Liu Z, Dao M-T (2007) Reliability optimization using multiobjective ant colony
system approaches. Reliab Eng Syst Saf 92:109–120
Zio E, Bazzo R (2011) Level Diagrams analysis of Pareto Front for multiobjective system
redundancy allocation. Reliab Eng Syst Saf 96:569–580
Zio E, Bazzo R (2011a) A clustering procedure for reducing the number of representative
solutions in the Pareto Front of multiobjective optimization problems. Eur J Oper Res
210:624–634
Chapter 10
Design Selection Decisions
Abstract: The design selection problem in the RRM context considers long-term
performance, and represents higher additional costs if unforeseen features that
should have been included during the project design phase had to be implemented
afterwards. Design decision involves multiple aspects and may be more critical
depending on the kind of item, such as consumer appliances, industrial equipment
or projects that have to consider safety aspects (airplanes or facilities). Reliability
has an essential role for design selection although other aspects have to be
considered such as maintainability and risk depending on the specific design
problem. Therefore, a multidimensional approach is usually required. In this
chapter, all these aspects are discussed in order to illustrate the importance of a
broader perspective when facing design decision problems. The fundamental
requirements are to consider reliability, maintainability and risk aspects so as to
establish features in the design project, including the definition of material,
redundancies, control systems and safety barriers. To illustrate these decisions,
aspects such as reliability (e.g. MTBF), maintainability (e.g. MTTR), safety, cost,
service life, efficiency, are discussed as criteria for these problems. Multi-attribute
utility theory (MAUT) is applied in this chapter to illustrate how reliability,
maintainability and risk aspects are included in an MCDM/A model for design
selection incorporating states of nature. The decision regarding the selection of
which features to include in a design project may be considered as an MCDM/A
portfolio problem. Finally, an introductory view is given of how the redesign
problem arises in the maintenance context with multicriteria approaches.
10.1 Introduction
The term design may have different meanings, such as project, and aesthetical
conception. The main issue in this chapter is related to the former, although the
latter is specifically applied as one of the criteria decision for a car project,
subsequently presented.
Decisions about the design of a product are determinant to its reliability. Errors
in the design process can increase considerably the costs during the product
development cycle. In this way, reliability is highly connected with problems of
engineering design (O’Connor and Kleyner 2012).

312 Chapter 10 Design Selection Decisions
In this field, performance capability and cost are two of the most important
factors of the design. While performance capability means the adequacy of a
design to perform required functions, cost means how much these performance
requirements of a design in monetary units are. To deal with a three-way trade-off
among performance capability, cost and reliability, into the design process, the
designer (the DM) have to examine how reliability requirements are set to build
reliability into a design (Lewis 1987). A literature review on MCDM/A approaches
in reliability and maintenance, shows that 16.7% of the work conducted is related
to design (de Almeida et al. 2015). It is important to note that this decision has
different implications depending on the mode and probability of failures.
Reliability requirements can be defined by several ways, such as by the designer,
by the buyer of the product and by government agencies.
Products that are more reliable imply higher capital costs and lower costs for
maintenance and repair. For the classical optimization approach, it is possible to
describe a function that represents the total cost including capital and repair costs.
Thus, an optimal solution can be found when this function is minimized. How-
ever, DM can have a preference for other solutions instead of the optimal that
minimizes the single objective cost function. This occurs in several practical
situations when the DM has some additional considerations about the trade-off
between capital and maintenance costs. The DM’s preferences are illustrated for
three examples: a mobile phone, an automated industrial machine and an aircraft
turbine.
A mobile phone is a consumer appliance in which reliability requirements rely
on consumer expectations. In this case, the reliability of this product can increase
the prices, reduce repair costs, provide elements to offer a longer warranty than a
competitor and boost sales. While some negative consequences are due to
excessive reliability such as lower sales, some inconveniences are due to the
opposite, such as company reputation for poor design. Design decisions in this
context are made by the manufacturer taking into account public’s preferences
about price and reliability through market surveys.
In a design of an automated industrial machine, or other equipment directed to
large organizations, the DM’s preference structure is completely different. In this
case, the trade-off between capital cost and production lost through breakdowns
should be evaluated. Thus, design decisions have to consider aspects of reliability
and maintainability.
Finally, in a design of an aircraft turbine, failure consequences are so severe
that a higher level of reliability is required. It means that the increase of a turbine
cost can be justifiable due to the level of reliability required to this product.
Additionally, aspects of delays and safety in airline maintenance are also relevant
(Sachon and Paté-Cornell 2000). In this case, insurance underwriters and
government agencies are responsible to define reliability specifications and risk
analysis performs an important role in the design.
10.1.1 The Reliability Role in System Design
Reliability estimation of a new product, before it is manufactured, is attractive for

designers. This information can allow accurate forecasts of support costs, spares
requirements, warranty costs and marketability (O’Connor and Kleyner 2012).
Reliability is a key aspect in the design selection. Different standards establish a
comprehensive design specification, general requirements and descriptions of
activities in order to guide the industry and government for developing reliable
products and systems (IEEE 1998; US MIL-STD-785B 1980; IEC 61160 2005;
BS 5760-0 2014).
According to Ren and Bechta Dugan (1998), the design requirements typically
consider reliability, cost, weight, power consumption, physical size, and other
system attributes. In order to meet these requirements, the DM should observe the
whole system as a set of components, which each design component can be
chosen from a set of design alternatives. Fu and Frangopol (1990) deal with the
problem of optimal structural design from a multi-objective perspective that taking
into account weight, system reliability and system redundancy.
Designers are experts in the creative process to provide reliable products. They
must know the types and extent of the loading, and the range of environmental
conditions under which the product operates. Additionally, they must know the
physics of the potential failure modes to ensure the required level of reliability.
Design margins, redundancy allocation and protection against strength degradation
are frameworks that can help designers to enhance reliability (Lewis 1987;
O’Connor and Kleyner 2012).
Design margin is a framework that considers that the reliability can be
increased by ratio of capacity of components and loads applied to them. In Fig. 10.1,
the probabilistic mechanism of the failure function is illustrated for different levels
of loading, loadi, where load1<load2. That is, the failure rate decreases as the
component load is reduced for a given operating features. For instance, a pave-
ment design for a road network can reduce drastically the probability of failures
and maintenance costs by means of this analysis.
load2
Ȝ(t)
load1
Fig. 10.1 Failure rate function at different levels of loading

In other words, a product is designed to have a reliability performance in

excess of that stated in its specifications. This often leads to an overdesign, early
in the development phase of a design. Hurd (1966) presents some examples of this
overdesign, for instance in structural design, a factor of 10 is used. A structure is
designed to hold 10,000 lb, while the maximum specification is 1,000 lb.
A similar situation is illustrated in an electronic context, when electronic parts are
designed to be used to 10 or 15 per cent of their rated capabilities.
Redundancy allocation in design allows the increasing of system reliability by
means of addition of components in parallel. It means that one or more
components can fail without result a system failure. Design decisions of multi-
state weighted k-out-of-n systems is a example of a relevant redundancy allocation
problem (Li and Zuo 2008). Decisions on redundancy allocation problems are
discussed in Chap. 9.
Strength degradation includes several complex mechanisms such as fatigue in
metals, corrosion and wear. Based on these mechanisms, the designer can specify
a fatigue limit of operation. Tests can provide the required data by generating
failures under known loading conditions, and reliability estimation can be carried
out. The designer must specify maintenance procedures for inspection, lubrication
or scheduled replacement when a suitable protection is not possible to be
determined by the design (O’Connor and Kleyner 2012). Inspections and
maintenance need to be planned at the design stage when fatigue failures are
present. Economic criteria to minimize lifecycle maintenance costs but satisfying
a minimum reliability level must be considered in design decisions. Reliability-
based maintenance strategies can help designers to deal with the conflict of
minimizing maintenance costs and maximizing reliability levels, specifically when
some fatigue mechanisms need to be inspected (Guedes Soares and Garbatov
1996; Garbatov and Guedes Soares 2001).
According to Sahoo et al. (2012) while most reliability optimization problems
have been formulated into the single objective optimization approach, it can be
recognized that most real-world design problems involving reliability optimization
require a broader perspective, and that is done by optimizing simultaneously more
than one objective function.
10.1.2 The Maintainability Role in System Design
A contribution to the problems associated with maintainability comes from the

engineering design specification. It should specify the engineering requirements,
including reliability and maintainability issues. Designer should comprehend the
international standards and their specific content. There are international standards
that deal with maintainability issues in design phase (IEC 60706-2 2006; BS EN
60706-2 2006).
As stated in the last section, the reliability has a central role in system design.
Additionally, the overall performance of a plant is also associated with
maintainability. The availability of a plant depends on the frequency and the
downtime of interruptions. Maintainability is a design characteristic that reflects
the probability of an item be restored in a given period to specified conditions by
maintenances actions that meet some procedures and resources requirements
(Goldman and Slattery 1977).
The fundamental function of equipment depends on its availability or readiness.
Thus, some trade-offs between reliability and maintainability could be made in
order to meet the availability requirements. Different levels of reliability R(t) and
maintainability M(t) can result an availability A(t) level.
As stated by Goldman and Slattery (1977), maintainability is a concept related
to different aspects from the basic physical characteristic of the design to the
strategic level of the maintenance function. Amongst the different issues
comprised on the spectrum of aspects related to maintainability, it is possible to
highlight design requirements, selection of maintenance strategy, logistic
provisioning, and so on.
Despite these different issues, this section focus on the role of that maintain-
ability plays, providing guidelines on the design specification, in order to provide
to the overall equipment some features related with effectiveness and lifetime
support cost.
In a design selection problem, while the designer has to develop the system,
taking into account various aspects, including the maintainability, and different
constraints imposed by budget project and standards; the user has to accept the
final configuration of the design and handle with the challenges related to
effective operational use of this built design. From the difference between user
and designer perspectives, sometimes, feedbacks have to be considered in both
directions: from the operation to design and from the design to operation.
Design development is complex and non-failure-free task. Thus, besides the
problems that could exceed the design phase, some other problems could arise
related to the discordance between designer and user views. Therefore, analysis of
the design process is essential not only as an alternative to handle with the views
discordances, but also to reduce the number of failures that comes from the project
phase. In this process of improvement, the feedbacks are essential. A simple
reason is that some problems come up after the launch of the product. As any item
is subject to failure, maintenance could be a quite frequent activity. In this case,
the maintenance should be done in order to restore the system for the operational
state, as quickly as possible, so that interruptions resulting from failures do not
affect the production targets.
In order to act as quick as possible, maintenance teams have to deal with some
barriers related with design selection issue considering the difficulty associated to
the maintenance activity. In some cases, the maintainability attributes might be too
restrictive resulting in difficulties to reduce maintenance times. For example, the
downtime reflects relations among three main groups of elements: design

decisions, maintenance policies and technician requirements. Therefore, after the
phase of design, the decisions related with the other elements taking into account
the constraints of the project (Goldman and Slattery 1977).
The maintainability is not only a criterion at the design selection problem, but it
should be considered in other problems such as downtime reduction or availability
increase. In some design problems, the maintainability formulated as state of
nature. The downtime is a non-controllable variable by a configuration of the
design, once there are human aspects such as motivation and ability that have
great impact on the downtime (time to repair).
10.1.3 The Risk Role in System Design
According to Lewis (1987), a range of consequences can be produced from a

failure of a system, affecting people and the environment. With the intention of
reduce this possibility, several levels of the acceptability of risk can be
established, based on specific characteristics such as procedures, resolutions and
standards, type of industry, industry or design location, etc.
Two risk aspects are involved in systems’ design, the first is the level of
acceptability defined in different sectors (such as civil, energy and chemical
engineering) can be affected by a specific accident, for instance, the catastrophe at
Chernobyl in 1986. The second is how technological levels impact directly in the
occurrence of a hazard situation (Vrijling et al. 1998).
Thus, in the context of systems’ design, an effective risk management (see
Chap. 3) is necessary to mitigate or prevent the risk occurrence, such a way that a
minimum risk level is reached. The observance of procedures, resolutions and
standards is a fundamental question to minimize the chance of the risk occurrence.
In Sect. 10.3, it is illustrated the use of standards as input to design selection.
A conflicting question in the context of system or equipment design should be
observed considering the following perspectives: reliability and safety. From the
safety perspective, if a hazardous event occurs, the risk to the public should be
minimized by the plant shutdown. From the reliability point of view, the plant
should stand in operation waiting for a failure occurrence before a shut down take
place or, as a last resort; the repair of the plant should be performed if the shut
down it is not possible. Thus, the challenge of an effective risk management in
system’s design context is to reduce the possibility of an accident, reducing the
probability to very low levels through design and safety detailed analysis (Lewis
1987).
Another interesting aspect with regard to safety and design is the increase in the
requirements of safety barriers, once each safety feature added to the project
increases the project costs, it also reduces the facility profitability. Nuclear power
plants faced such situation after the Fukushima accident in 2011. After the
10.2 An MCDM/A Model for the Design Selection for a Car 317
accident, safety requirements has been revised, compromising the economic

viability of specific project designs that would have to raise safety in order to meet
risk acceptance levels more conservative.
10.2 An MCDM/A Model for the Design Selection for a Car
In this section is presented an illustrative example of the design selection for a car
according to the MCDM/A approach presented in Chap. 2.
According to the first step discussed in Chap. 2, it is important to make some
remarks and observations to characterize a DM involved in such problem. In this
illustrative example, the DM is the senior engineer responsible for defining which
car project should be selected as the best design from the product development
stage.
At this point is important to notice that all standards and requirements have
already been achieved by each project, thus, the decision regards on evaluating
these alternatives considering the factors or objectives related to this decision,
which recalls the second step pointed in Chap. 2.
As the objectives for this kind of decision, one may list the following
objectives: Maintainability, Reliability, Safety, Cost, Service life, Efficiency,
Aesthetic.
For simplification purposes this illustrative example will not consider the last
three objectives. Focusing on the first four objectives allows to explore more
deeply the RRM perspective of such problem.
Other objectives may be more emphasized during different phases of the
product development, taking the initial list of objectives as a reference or by the
addition of other objectives depending on specific aspects related to the problem
context.
The definition of criteria related to each objective refers to the third step given
in Chap. 2 with regard to establishing criteria.
In order to measure maintainability, the concepts given in previous sections and
in Chap. 3 shall be used, thus as the maintainability concept relates to the time
spent during repairs.
The maintainability function for any probability distribution may be
represented as in (10.1) (Dhillon 1999; Stapelberg 2009), where t represents time
and fr(t) is the pdf of the repair time.
t
m(t ) ³ f (t )dt
r
(10.1)
0
An Exponential, Lognormal, Weibull, Normal, and others may represent the

repair time depending on the equipment considered.
A few indices are applied for analyzing maintainability in a deterministic

context. For instance, Dhillon (1999) presents some measures for maintainability:
x Mean time to repair (MTTR).
x Mean active preventive maintenance time and median active corrective
maintenance time.
x Maximum corrective maintenance time.
x Mean maintenance downtime.
In a probabilistic context, considering the step 5 of the procedure for building a
MCDM/A model (see Chap. 2), the pdf of the maintainability is introduced into
the model, as a state of nature, such as in the decision model of Chap. 7 and Chap.
9. The MTTR represents the expected value of t, given fr(t). In other models, the
maintainability may be modeled as a consequence, such as in this subsection.
Simplifying with deterministic indices, it could be applied for the maintain-
ability either: a quantile of the maintainability distribution or use the standard
deviation with the MTTR, as explained in Chap. 2.
As previously observed in Chap. 3, the objective reliability is related to a
probabilistic concept with regard to the time whenever the equipment shall fail.
In order to provide an easier scale for measuring the reliability of each car
project for a DM, the mean time between failures (MTBF) concept may be used
for evaluating each alternative’s reliability, as pointed by O’Connor and Kleyner
(2012) as an alternative for measuring reliability of repairable items. The
definition of MTBF is given by (10.2), from the reliability function (Stapelberg
2009).
f
MTBF ³ R(t )dt (10.2)

0
In order to standardize the reliability measurements of each car project, one

may consider it for each car project measured by the number of hours of highway
driving to evaluate its operation.
There are many procedures for assessing a car project safety. Most of these
procedures involve a crash test. Thus, there is an evaluation of the entire project
and the outcomes of safety items such as seatbelts, airbags, anti-intrusion bars
(side protection), laminated windshields, crumple zones, cargo barriers, safety
cell, and others.
For each new car project there is a study based in crash tests to assess its safety
through a New Car Assessment Program (NCAP). Depending on the region, the
Euro or US NCAP, for example, may assess a new car project.
Considering the US NCAP, the crash tests includes the evaluation of a frontal
crash, a side crash, and the risk of rollover in a five star safety rating, from 1 to 5
stars, with 5 being the best rating. Therefore, the criteria adopted for measuring
safety shall be the lowest estimate rating of the car project in the frontal crash and
side crash test.
10.2 An MCDM/A Model for the Design Selection for a Car 319
The objective cost measures and evaluates all aspects that may be converted
into a monetary value scale. It is important to emphasize that MCDM/A
approaches provide methodological support to understand and value alternatives
among different objectives scales and provide a global evaluation that includes all
objectives.
Thus, when measuring cost, all factors that may be represented by monetary
scale shall be summed in order to evaluate the respective cost of each car project.
In this example, it shall be assumed that all these factors have been considered for
estimating each car project cost.
The elements of the set of alternatives are the car projects. The problematic
relies on the choice of the best car project according to the corresponding criteria.
The states of nature in this illustrative problem corresponds to factors that are
not under the DM’s control and are subject to uncertainty, influencing the decision
outcomes, such as when a failure may occur or the time for a repair service. Such
uncertainties may be represented by probability distributions that provide the
performance estimates for each car project.
Assuming that the DM preferences fit the axiomatic structure required by
MAUT, the next step regards the intra criterion evaluation. Establishing intra
criterion evaluation consists in the definition of a utility function for each criterion
by assessing its shape and parameters from the DM’s evaluation of the outcomes
in each criterion through lotteries or other elicitation procedure.
For a DM with additive independence condition, the additive utility function is
represented by (10.3):
ua k m E a [um ( m )] k r E a [ur ( r )] k s Ea [u s ( s )] k c Ea [uc ( c )] , (10. 3)
where ua is the expected utility of a, based on the additive function over the
expected utility of the attributes maintainability (m), reliability (r), safety (s) and
cost (c); km, kr, ks and kc are the respective scale constants.
The scale constants in MAUT are elicited through lotteries as given in Chap. 2.
For this illustrative problem, consider km, kr and ks equal to 0.3 and kc equals to 0.1.
The respective one-dimensional utility functions are um(m), ur(r), us(s) and
uc(c). For maintainability (m) and reliability (r), the random variable repair time
(m) and time to failure (r) are considered consequences, with its respective pdfs
fa(m), fa(r), for each alternative a. Their expected utilities are respectively given as
follows by (10.4) and (10.5).
f
Ea [um ( m )] ³u m ( m ) f a ( m ) dm (10.4)
0
f
Ea [ur ( r )] ³ u (r) f
r a ( r ) dr (10.5)
0
Similar formulation are given to safety (s) and cost (c).

Considering seven non-dominated car project alternatives for this design
selection problem, named as Alt1, Alt2, Alt3,..., Alt7, Table 10.1 presents each
alternative expected utility over the considered attributes and its correspondent
additive utility value.
In a single attribute perspective, alternatives 4, 6, 3 and 5 gives the best
performance for maintainability, reliability, safety and cost, respectively.
Alternatives 1, 2 and 7 provide a more distributed performance among the
attributes, as a result, an MCDM/A approached is required to evaluate tradeoffs
and value the overall value of these alternatives in order to provide a
recommendation for the selection problem.
Table 10.1 Car project alternatives evaluation for a design selection problem
u(m) u(r) u(s) u(c) ua

Alt1 0.558 0.812 0.126 0.453 0.494
Alt2 0.419 0.750 0.106 0.818 0.464
Alt3 0.626 0.586 0.976 0.600 0.716
Alt4 0.892 0.761 0.760 0.200 0.744
Alt5 0.139 0.508 0.091 0.941 0.315
Alt6 0.739 0.881 0.626 0.105 0.685
Alt7 0.861 0.563 0.765 0.606 0.717
From the results given by Table 10.1 the best car project for such DM would be
the alternative 4, which achieved the highest value in the additive utility function.
By evaluating alternative 4 individual utilities is interesting to observe that it has
the best performance on the maintainability attribute, although presenting one of
the worst values for the attribute cost, only better than alternative 6.
Thus, for a DM with such preferences, the alternative 4 values for
maintainability, reliability and safety are compensating its bad outcome for the
cost attribute. A different DM may have different preferences and make different
tradeoffs, leading to the selection of a different alternative. As pointed in Chap. 2,
the sensitivity analysis is an important step in order to evaluate the robustness of
the preliminary recommendation, allowing to increase the accuracy of the
elicitation process when it is required.
For a DM with a non-compensatory rationale, the selected alternative could be
other due to a different kind of preference structure. Such DM with a different
preference structure, would set different model parameters and establish different
comparisons, such that in many times would be reflected as the selection of a
different alternative. The use of the MCDM/A approach enriches the decision
process by allowing to incorporate these particularities of the DM preferences
with accuracy, by incorporating it in the decision model.
10.3 Risk Evaluation for Design Selection 321
10.3 Risk Evaluation for Design Selection
During the design phase, it is possible to improve system reliability and risk
barriers in order to reduce risks and avoid unnecessary costs to adjust and match
the project to the required risk standards. During this phase, accident rates can be
influenced when deciding on what material and components to use in the project.
Many studies consider that risk evaluation follows a trend, which reflects
probabilities separately from consequences. As a result of this trend, many
decisions are taken which consider only probabilities or consequences, without
aggregating these two important factors when evaluating the overall risk.
This may occur due to the difficulties in estimating and/or simulating these
processes to quantify probabilities and the magnitude of the consequences.
However, these two measures can be considered together with human judgment if
utility functions are used, as do Baron and Paté-Cornell (1999), Brito and de
Almeida (2009), Brito et al. (2010), Almeida-Filho and de Almeida (2010) and
Garcez and de Almeida (2014).
The concept of ALARP has been questioned by many authors in the literature.
Melchers and Stewart (1993) shows that each individual can have a different level
that he/she finds acceptable for different types of risk and also that this can change
also from one culture to another.
Aven and Vinnem (2005) presented a different risk analysis regime that is not
based on risk acceptance criteria at all. They argue that a rule based on cost-
effectiveness should do better than pre-defined risk acceptance limits. In some
situations, it is possible to achieve risks below ALARP levels. Thus a methodo-
logy is required that can consider a DM’s tradeoffs among costs and other loss
dimensions such as environmental and potential losses of life.
Aven and Kristensen (2005) presented a discussion on several perspectives of
risk, establishing a common basis for the different perspectives, emphasizing how
important it is to consider all possible consequences associated with their uncertainties.
When considering the context of risk analysis in the literature on the oil and gas
industry, there are two main models for risk management with a specific focus on
risk evaluation and risk reduction that can guide the selection and design decisions
by evaluating risk levels. Besides these models there are other models in the
literature that can be used after start-up at the facility, such as the framework
proposed by Øien (2001) for structuring risk indicators for risk control during
operation.
Khan et al. (2002) present an example of design selection in order to implement
safety measures. Khan et al. (2002) present an offshore oil and gas facility and
design alternatives that may reduce risk.
The approach of inherent safety design was presented initially by Kletz (1985),
and detailed later in Kletz (1998). Khan and Amyotte (2002) presented a study
showing that safety measures should be a concern from the design stage for a
facility in order to reduce costs throughout its life span.
10.3.1 Risk Assessment Standards
One of the main models for risk evaluation can be found in ISO/IEC Guide 51:
2014 and another in the NORSOK Standard Z-013. The model described in
ISO/IEC Guide 51: 2014 updates the 1999 version regarding this subject. It can
also be used combined with IEC Guide 73: 2009, which refers to the vocabulary
and meanings regarding risk management, so as to consolidate terminologies.
The NORSOK Standard Z-013 is a standard edited by the NPD (Norwegian
Petroleum Directorate), which is a Norwegian agency in charge of regulating oil
industry activities in the North Sea.
Brandsæter (2002) describes the implementation and uses of risk analysis using
quantitative and qualitative methodologies for the oil and gas offshore industry
with contributions to the EC-JRC International Workshop on “Promotion of
Technical Harmonisation on Risk-Based Decision Making (2000)”, which is
formatted as if it were a response to a set of questions prepared by workshop
organizers, and in which both models mentioned are discussed.
In ISO/IEC Guide 51: 2014, risk evaluation is defined as a wide process of
estimation and analysis. For ISO/IEC Guide 51: 2014, risk analysis terminology is
defined as the systematic use of information to identify hazards and estimate risk, and
risk estimation is defined as a procedure to determine if the risks are tolerable or not.
Thus the ISO/IEC Guide 51: 2014 model is represented by an iterative process
to evaluate and reduce risks that can be applied to qualitative and quantitative risk
evaluations. In ISO/IEC Guide 51: 2014, it is clear that some tolerance criteria
have to be defined, as in ALARP. However, it does not suggest any procedure to
deal with the situation where the limits have already been satisfied (or not) even
how to choose between non-dominated alternatives considering multiple risk
dimensions.
This iterative process considers that each hazard must be considered and must
satisfy a tolerable risk level. According to ISO/IEC Guide 51, it is necessary to
identify each hazardous situation and event by anticipating stages and conditions
for the system, including installation, operation, maintenance, repair and destruction/
disposal. This iterative process considers an entire process of risk assessment.
The ISO/IEC Guide 51: 2014 presents a “three-step method” from the design
phase and additional measures at the use phase. The risk reduction process starts
from the design of the installation, beginning with inherently safe design as a way
to start the risk reduction process.
Additional risk reduction alternatives have to be implemented after the design
stage such as training and procedures that will reduce residual risks after all
protective measures have been deployed.
The NORSOK Standard Z-013 model for the process of assessing risk and
emergency preparedness describes this in a similar way to that of the ISO/IEC,
although NORSOK includes an assessment of preparedness for emergencies in its
process.
The previous version of the process for this standard already gave more
emphasis to a typical quantitative risk analysis methodology when considering the
risk assessment process. It emphasized the importance of the estimation, analysis
and evaluation approach for typical quantitative risk analysis methodologies when
applied to offshore oil and gas structures. It defines risk as a probability or an
expected frequency, and requires a risk acceptance criterion to be defined, which
should consider the probability or frequency of an associated consequence,
thereby establishing a risk index and an acceptable risk limit. For a human
dimension, for example, the risk for an individual can be used and/ or the FAR
(Fatality Accident Rate) which should be compared with acceptable limits by the
adopted standards in order to establish the risk picture.
10.3.2 MCDM Framework for Risk Evaluation in Design

Problems
Throughout the risk analysis process there is no specific framework about how to
aggregate preferences amongst multiple risk dimensions, especially when there is
a decision problem where some of the alternatives have already reached the
acceptable risk levels defined in standards.
Thus, all dimensions are considered in terms of constraints that must be
respected and considered in terms of cost benefit evaluation. Nevertheless, during
the design process, there are opportunities to improve safety and prioritize safety
alternatives.
To address this issue, this section presents a framework with a numerical
application that aggregates preferences by using metrics, which consider
probability, give values to human judgment on consequences and their behavior
with regard to risk (prone, neutral and risk averse). These metrics are provided by
Utility Theory.
Based on the literature, this procedure supports the structuring of decision
problems in the context of evaluating multidimensional risk, based on the frame-
work given in Chap. 2.
The approach given in this section addresses decision problems regarding risk
reduction and safety improvements for the design of hazardous facilities. These
kinds of decision problems can refer to a choice, ranking, sorting or a portfolio
decision problem, and depending on the type of problematic (Roy 1996) a specific
methodology should be used to aggregate the DM’s preferences and doing so
should be amongst its objectives.
While due consideration should be given to the models for risk estimation,
analysis and evaluation from both standards (NORSOK and ISO/IEC), this
MCDM/A procedure can be used as a framework to evaluate risk reducing
alternatives. This can be done even if some of these risks have already achieved
the acceptable risk levels and assuming there are still safety improvements that
could be implemented. Also decisions can be taken about which risk reduction
measures should be implemented, according to the priority of each action and the
risk involved in different parts of the process.
This kind of decision has two main actors who, in general terms, can be
identified as a DM and an analyst who will give methodological support to the
DM. Both of these actors will exert influence during the decision process, in
which the former, a DM, will influence the decision because of cognitive aspects
and his/her preference structure. The analyst will influence it in such way that he
may bring bias to the process due to his/her own opinion about the subject and/or
because of the use of his/her preferred methodological approaches (Almeida-Filho
and de Almeida 2010).
Figure 10.2 presents the steps to consider an MCDM/A approach in order to
choose or prioritize design alternatives for risk reduction based on the procedure
for building MCDM/A model presented in Chap. 2.
Design Problematic Design alternatives

decision definition identification
Objective Value Uncertainty evaluation /

consolidation states of nature
Design
Risk Aggregation
alternatives
Analysis procedure
evaluation
Results Sensitivity
Analysis analysis
Fig. 10.2 Multi dimensional risk evaluation in design problems
It starts with a Decision Situation that represents the phase when a decision
problem has been identified or has appeared and needs to be outlined. This
includes determining the risk level achieved and the risk reduction alternatives to
be considered in design. The step ahead is to define the kind of problematic (Roy
1996; Vincke 1992) that addresses the problem itself.
The process of identifying alternatives should be extensive and exhaustive with

a view to this set comprising as many alternatives as possible, except for those,
which can be previously considered as dominated alternatives. This is a very
important step since it seeks to avoid a situation where good alternatives are
neglected.
After the set of alternatives is well defined, it is necessary to evaluate
uncertainties regarding each alternative (action) and their possible states of nature.
For this stage, the same QRA techniques can be applied for its estimation as
suggested by NORSOK Standard Z-013 or the ISO/IEC Guide 51: 2014, including
simulation and estimation models for damage radii (DR) of different propensities,
for example. To estimate probability, the same QRA methods discussed in Chap. 3
can be applied such as fault trees analysis, event tree analysis and expert’s
knowledge elicitation amongst other techniques.
Afterwards it is necessary to establish the DM’s preferences, in order to
evaluate alternatives. The use of utility theory to evaluate risk in each criterion
enables the probability and the consequence value for the DM of each possible
outcome to be considered together, thereby providing a metric for each risk
dimension and also the DM’s behavior (probe, neutral and risk averse) in each risk
dimension. This often considers financial aspects, human potential losses and
environmental damages as the three dimensions usually considered. Thus, what is
required is to elicit the DM’s utility for consequences in each risk dimension.
(Brito and de Almeida 2009; Brito et al. 2010; Alencar et al. 2010; Lopes et al.
2010; Garcez et al. 2010; Almeida-Filho and de Almeida 2010; Garcez and de
Almeida 2014). These consequences result from the combination of alternatives
and the possible states of nature, as shown in Chap. 2 (Table 2.1).
To aggregate all risk dimensions evaluations, an aggregation method must be
considered. To make this choice some aspects will have to be taken into account,
as which kind of preferences structure the DM has and his/her preferences should
be modeled to determine if he/she has a compensatory or non-compensatory
rationality.
As pointed in Chap 2, the rationality behind the DM’s preferences guides the
choice of a compatible aggregation method. As to non-compensatory methods
there are, for instance, the ELECTRE family of methods (Roy 1996) and the
PROMETHEE family of methods (Brans and Mareschal 2002).
As to compensatory approach, there are several methods that can be used, of
which MAUT (Keekey and Raiffa 1976) is amongst the most used methods that
considers the risk evaluation structure.
The evaluation of alternatives is the phase where the MCDM/A method chosen
is applied and its parameters should be obtained through an elicitation process that
may change according to the nature of each kind of aggregation methodology.
These steps are detailed in Chap. 2, thereby it is possible to consider multiple
risk dimensions and aggregate them from the perspective of the DM’s preferences
in order to reduce risks and improve safety conditions of a hazardous installation
by considering different design alternatives.
This MCDM/A framework allows the DM not only to use acceptance levels as
references but also to evaluate them according to his/her preferences and risk
behavior (prone, neutral, averse).
10.3.3 Illustrative Example of Risk Evaluation in a Design

Problem
In this section, an illustrative example is presented based on a realistic problem of

implementing a safety project to illustrate an application of MCDM/A for risk
evaluation in facilities design. Thus, there is a set of safety projects that can be
implemented and a DM has to define which sub-set of safety projects to
implement, with regard to an offshore oil and gas platform, specifically in the
primary process (Khan et al. 2002).
Oil and gas are well-known hazardous materials and when they are extracted,
there are several sources of hazard, one of which is the primary process where the
crude oil from the wellhead (a mixture of oil, gases and water) is separated before
it is processed.
In a general way, the primary process on an offshore oil and gas platform
consists of a first separator to separate the crude oil from the gases and water,
which it then sends to a transportation line. A second separator is used to separate
the residual water from the gas and send it to other subsequent units to separate
the wet gas and the dry gas. The other units comprise two compressor units, a
flash drum unit and a drier unit.
The decision situation consists of improving safety throughout the primary
process on an offshore oil and gas platform by choosing whether or not to implement
the design of some safety features. Therefore, the set of alternatives may be global-
ized or fragmented. The former consists that each alternative exclude the others,
while the latter considers the combinations of the set of alternatives (Vincke
1992). Thus, if considering different safety features, the set of the alternatives may
be the combination of all features that may be considered for the design project.
The problematic involved in this decision situation is about choosing to
implement one or more features. In this particular case, the model could use a
choice or a portfolio problematic. Although the decision relies in maximizing
safety and minimizing costs simultaneously, while technical aspects are
formulated as constraints into the model, for instance, acceptable risk limits.
Therefore, a knapsack problem (Martello and Toth 1990) would be a model
representation considering to objectives and technical aspects to define the design
as presented in the model given by (10.6).
max E >U x1 , x2 , x3 ,..., xi ,..., xn 1 , xn @

(10.6)
s.t. rj x1 , x2 , x3 ,..., xi ,..., xn 1 , xn d AL j for each j 1 to m.
By a knapsack problem representation the set of alternatives would be

fragmented and each design project alternative would be represented by the vector
(x1, x2, x3,..., xi,..., xn-1, xn) which is all the possible combinations of the n safety
features considered for the design problem. By maximizing the expected MAU
function value, E[U(x1, x2, x3,..., xi,..., xn-1, xn)], subjected to the technical
constraints, represented in the model given by (10.6) only by the m risk
acceptance levels (ALj) for simplification purposes as other technical aspects may
be included in this formulation. Thus, the design project alternative recommended
by this model is in compliance with the technical aspect considered. For this
illustrative example, the alternative risk level rj(x1, x2, x3,..., xi,..., xn-1, xn) in
dimension j has to be lower than ALj.
For this application, five features were considered that could be implemented in
the design of the facility in order to improve safety throughout the primary process
on the offshore oil and gas platform. The first feature (A) introduces improve-
ments in the first separator; the second feature (B) introduces improvements in the
second separator; the third feature (C) introduces improvements in the compressor
units; the fourth feature (D) introduces improvements in the flash drum unit; and
the fifth feature (E) introduces improvements in the drier unit.
If all these features were implemented, they would improve safety throughout
the process by reducing risk. In other words, they would reduce the probabilities
of events that would generate different accident scenarios. This could be to
substitute some kinds of materials for stronger and more reliable ones or it could
also be to implement different control procedures and automation throughout the
unit, for example.
Given the structure of MAUT, when considering an MCDM/A portfolio
analysis there are some issues that must be observed. There are effects associated
with the different utility scales on the results of an MCDM/A portfolio, especially
non linearity as occurs in the utility scale for evaluating the consequences (de
Almeida et al. 2014). When such aspects are involved, a different approach has to
be used in order to avoid the bias due to utility scales issues into the aggregation
procedure for the MCDM/A portfolio analysis. Thus, to avoid misleading results,
a complete enumeration approach for the portfolio problem given the five safety
features is used to avoid possible bias effects associated with the utility scales.
Thus, it allows illustrating this portfolio problem as a choice given all possible
design project combinations considering the safety features. Enumeration schemes
are an alternative approach to solve knapsack problems (Yanasse and Soma 1987;
Martello and Toth 1990).
Therefore, a choice problematic is used for modeling, and thus all possible
alternatives are enumerated for considering all five safety features combinations in
order to provide all the design projects that represent the set of alternatives. So,
from the choice problematic definition given in Chap. 2, a DM can choose a
subset of this, which, in this case, is one of the design projects.
The identification of alternatives considers the existence of features A, B, C, D

and E only, and that the DM can choose all of them if he/she thinks that this is
worthwhile. Thus, the set of alternatives consists of all the combinations of
implementing (1) or not (0) each project. This can be summarized in 32 alternatives
as shown in Table 10.2.
Table 10.2 Set of alternatives
Alternative Action A B C D E
1 No feature implemented 0 0 0 0 0
2 Implement E 0 0 0 0 1
3 Implement D 0 0 0 1 0
4 Implement D and E 0 0 0 1 1
5 Implement C 0 0 1 0 0
6 Implement C and E 0 0 1 0 1
7 Implement C and D 0 0 1 1 0
8 Implement C, D and E 0 0 1 1 1
9 Implement B 0 1 0 0 0
10 Implement B and E 0 1 0 0 1
11 Implement B and D 0 1 0 1 0
12 Implement B, D and E 0 1 0 1 1
13 Implement B and C 0 1 1 0 0
14 Implement B, C and E 0 1 1 0 1
15 Implement B, C and D 0 1 1 1 0
16 Implement B, C, D and E 0 1 1 1 1
17 Implement A 1 0 0 0 0
18 Implement A and E 1 0 0 0 1
19 Implement A and D 1 0 0 1 0
20 Implement A, D and E 1 0 0 1 1
21 Implement A and C 1 0 1 0 0
22 Implement A, C and E 1 0 1 0 1
23 Implement A, C and D 1 0 1 1 0
24 Implement A, C, D and E 1 0 1 1 1
25 Implement A and B 1 1 0 0 0
26 Implement A, B and E 1 1 0 0 1
27 Implement A, B and D 1 1 0 1 0
28 Implement A, B, D and E 1 1 0 1 1
29 Implement A, B and C 1 1 1 0 0
30 Implement A, B, C and E 1 1 1 0 1
31 Implement A, B, C and D 1 1 1 1 0
32 Implement all features 1 1 1 1 1
The consequence evaluation considered the most credible scenarios for these
units (Khan et al. 2002) are summarized in Table 10.3.
Table 10.3 Consequences for the most credible scenarios
DR 100% 3rd. Degree of Burn (m)
DR 50% 3rd. Degree of Burn (m)

DR 100% Fatality / Damage (m)
DR 50% Fatality / Damage (m)
Possibility of Spills
Units
First Separator 230 288 333 428 yes

Second Separator 53 74 69 78 yes
Compressor Units 24 35 44 57 no
Flash Drum 25 42 56 77 yes
Drier 73 92 106 136 yes
The most credible scenario for the first separator is a BLEVE followed by fire;
for the second separator it is VCE followed by fire; for the compressor units it is a
gas release possibly turning into a jet fire; for the Flash Drum unit it is a VCE
followed by fire; and for the Drier unit the most credible scenario is a BLEVE
followed by fire.
These scenarios, which are the most credible ones, have a higher probability in
the present situation and a lower probability after implementing safety features,
for each scenario (Khan et al. 2002). These probabilities are illustrated in Table 10.4.
Table 10.4 Probabilities considering design implementation
Probability after implementing

Scenario Probability in present situation the design
Normality 0.9990804690 0.9999998688
Accident in First Separator 0.0000107000 0.0000000179
Accident in Second Separator 0.0009474000 0.0000000155
Accident in Compressor Unit 0.0136400000 0.0000013110
Accident in Flash Drum Unit 0.0009060000 0.0000000786

Accident in Drier Unit 0.0000028310 0.0000000347
The first objective to be considered by a DM would be the potential number of

lives that could be saved by simply choosing a specific alternative for the facility
design. Another concern, of a DM in this situation, would be the environmental
dimension, which would be affected if an accident scenario occurs. There are also
many monetary or financial aspects to be evaluated, such as property losses,
downtime in production and several financial compensations and fines, which
would have to be paid, and also the costs of any safety improvement. The
aggregation procedure can be more extensive depending on the methodology used
to model and elicit DM’s preference structure, as given in Chap. 2, it may also
consider a value focused thinking approach (Keeney 1992) for structuring DM’s
objectives in order to inspire design features and so, creating design project
alternatives.
With regard to the criteria or objectives for this problem, these can be
summarized by a human objective, which implies minimizing loss of human life;
an environmental objective, which implies minimizing environmental losses; and
a last objective which is a financial objective, that of minimizing any expected
financial loss and also minimizing the costs of implementing safety improvement
(actions) considering that these costs will occur if these actions are chosen.
For each dimension, it is necessary to elicit the conditional utility functions.
These multiple risk dimensions are aggregated considering MAUT as an MCDM/A
approach. Thus, the alternatives are evaluated by using a MAUT to provide a
complete rank of all alternatives considered using an additive MAU function, such
as (10.7), where kh, ke and kf are the scale constants of the additive utility function,
which represents the trade-offs amongst these objectives, i.e. human (h),
environmental (e) and financial (f).
u (h, e, f ) k h u h 1 ( h ) k e u e (e) k f u f ( f ) (10.7)
The elicitation of these scale constants considers the range (variability) in

consequences and the importance of each criterion, so this measure represents
these two figures. As to the values for scale constants, these are considered as 0.5
for the Human dimension, 0.49 for the Environmental dimension and 0.01 for the
Monetary or Financial dimension. These values reflect the difference between the
range of best and worst consequences in each dimension and the importance
relation for a DM of changes in values among dimensions. This also reflects a DM
who would be more inclined to spend by some proportion if this will reduce the
probabilities of injures to people or to the environment.
After evaluating the alternatives, it is possible to provide a complete rank of all
the alternatives considered. The ranking of 10 alternatives is presented in Table
10.5, with the main result in order to compare these alternatives, which is the
difference ratio between them.
Since the utility scale is highly affected by the huge difference between
normality and any of the accident scenarios, the analysis of the differences ratio of
the alternatives shows more information than the utility scale itself. This measure
10.4 Redesign Required by Maintenance 331
is used due to the nature of the utility measure, which is based in an interval scale,
so what really matters is the size of the utility difference ratio instead of absolute
differences between them.
Table 10.5 The ranking of the design alternatives
Difference Ratio
Rank Alternative
( ui u /u u
i 1 i 1 i 2
)
1 5 0.65
2 7 1.65
3 6 12.07
4 13 0.08
5 8 2.45
6 15 1.70
7 1 0.28
8 21 5.09
9 3 2.25
10 14 0.07
Thus the last column of Table 10.5 shows that the difference between
alternative 5 and 7 is 65.39% bigger than the difference between alternative 7 and
6, and that the difference between 8 and 15 is 245.48% bigger than the difference
between 15 and 1 (implement none of safety features). This measure gives to DMs
a clear idea about the difference of these alternatives considering probabilities,
consequences, their individual preferences and their behavior towards risk.
By conducting a sensitivity analysis, it was possible to observe that the first
positions of the ranking would not change if the values chosen for the scale
constants were changed. It is also interesting to highlight that according to the
results, safety will be improved, whereas the alternative that represents no invest-
ment in safety appears only in 7th position in the ranking.
10.4 Redesign Required by Maintenance
From the perspective of the maintenance function, redesign is the action that is
done when the status quo is not acceptable. Some reasons for that are: higher
performance requirements due to competitiveness; more conservative standards
related to environment and safety; and more severe degradation not covered by the
initial design.
According to Moubray (1997), when a failure of a device implies safety and

environmental losses and, there is no effective maintenance activity to reduce
these consequences, the redesign may be undertaken with at least one objective,
such as: Reducing the probability of failure modes; Mitigating the consequences
of failures; Reducing the downtime.
The probability of critical failure modes can be reduced by increase the quality
of the components or making changes that affect specifically the reliability. For
the second objective, mitigating the consequence of failures usually is made by
addition of protective devices that reduce the chance of serious consequences
happen. Finally, the third objective can be achieved by design changes that make
the maintenance actions faster.
In this way, the selection of the parts or equipment that need to be redesigned,
can be defined as an MCDM/A problem, in which the set of alternatives is
composed by equipment and criteria are related to the attributes of maintainability,
and others associated with the cost of the redesign and possible consequences of
failures due to the permanence of the status quo.
Efforts to redesign should be planned based on the potential gain in reducing or
increasing the frequency of the occurrence of specific operating systems. Thus, for
a plant with distinct redesign demands, a ranking of these demands based on this
expected gains can be useful in order to manage resources efficiently (Heins and
Roling 1995).
The redesign process is usually an expensive process and the probability that it
will not solve the performance problem can be high. When the design provides
opportunity of improvement, maintenance actions should help to achieve the
desirable performance. However, when the desired performance is beyond what
the design could provide, maintenance actions are ineffective.
References
Alencar MH, Cavalcante CAV, de Almeida AT, Silva Neto CE (2010) Priorities assignment for
actions in a transport system based on a multicriteria decision model. In: Bris R, Soares CG,
London, UK, p 2480
Almeida-Filho AT de, de Almeida AT (2010) Multiple dimension risk evaluation framework. In:
Bris R, Soares CG, Martorell S (eds) European safety and reliability conference, Prague,
September 2009. Reliability, Risk, and Safety: Theory and Applications, Vol. 1-3. 2009. Taylor
and Francis, London, UK, p 2480
Aven T, Kristensen V (2005) Perspectives on risk: review and discussion of the basis for
establishing a unified and holistic approach. Reliab Eng Syst Saf 90:1–14
Aven T, Vinnem JE (2005) On the use of risk acceptance criteria in the offshore oil and gas
industry. Reliab Eng Syst Saf 90:15–24
Baron MM, Paté-Cornell ME (1999) Designing risk-management strategies for critical engineering
systems. Eng Manag IEEE Trans 46:87–100
Brandsæter A (2002) Risk assessment in the offshore industry. Saf Sci 40:231–269
References 333
Brans JP, Mareschal B (2002) Prométhée-Gaia: une méthodologie d’aide à la décision en

présence de critères multiples. Éditions de l’Université de Bruxelles
Brito AJ, de Almeida AT, Miranda CMG (2010) A Multi-Criteria Model for Risk Sorting of
Natural Gas Pipelines Based on ELECTRE TRI integrating Utility Theory. Eur J Oper Res,
200:812-821
BS 5760-0 (2014) Reliability of systems, equipment and components. Guide to reliability and
maintainability. British Standard.
BS EN 60706-2 (2006) Maintainability of equipment- Part 2: Maintainability requirements and
studies during the design and development phase, British Standards Institution
de Almeida AT, Vetschera R, de Almeida JA (2014) Scaling Issues in Additive Multicriteria
Portfolio Analysis. In: Dargam F, Hernández JE, Zaraté P, et al. (eds) Decis. Support Syst.
III - Impact Decis. Support Syst. Glob. Environ. SE - 12. Springer International Publishing,
pp 131–140
Dhillon BS (1999) Engineering Maintainability: How to Design for Reliability and Easy
Maintenance. Gulf Professional Publishing
Fu G, Frangopol DM (1990) Balancing weight, system reliability and redundancy in a multi-
objective optimization framework. Struct Saf 7:165–175
Garbatov Y, Guedes Soares C (2001) Cost and reliability based strategies for fatigue
maintenance planning of floating structures. Reliab Eng Syst Saf 73(3):293–301
Garcez TV, Almeida-Filho AT de, de Almeida AT, Alencar MH (2010) Multicriteria risk
analysis application in a distribution gas pipeline system in Sergipe. In: Bris R, Soares CG,
Martorell S (eds) Reliability, risk and safety: theory and applications vols 1-3. European
safety and reliability conference (ESREL 2009), Prague, September 2009. Taylor and Francis,
1043-1047
Garcez TV, de Almeida AT (2014) Multidimensional Risk Assessment of Manhole Events as a
Guedes Soares C, Garbatov Y (1996) Fatigue reliability of the ship hull girder accounting for
inspection and repair. Reliab Eng Syst Saf 51(3):341–351
Hurd Jr W (1966) Engineering design and development for reliable systems. In: Ireson W (ed)
Reliab. Handb. McGraw-Hill, New York, pp 10–33
IEC 60706-2 (2006) Maintainability of equipment - Part 2: Maintainability requirements and
studies during the design and development phase, International Electrotechnical Commission
IEC 61160 (2005) Design review. International Electrotechnical Commission.
IEEE (1998) Standard Reliability Program for the Development and Production of Electronic
Systems and Equipment. IEEE Std 1332-1998:i.
ISO/IEC (2014) Guide 51: Safety aspects – Guideline for their inclusion in standards. ISO/IEC
Khan FI, Amyotte PR (2002) Inherent safety in offshore oil and gas activities: a review of the
present status and future directions. J Loss Prev Process Ind 15:279–289
Khan FI, Sadiq R, Husain T (2002) Risk-based process safety assessment and control measures
design for offshore process facilities. J Hazard Mater 94:1–36
Kletz TA (1985) Inherently safer plants. Plant/Operations Prog 4:164–167
Kletz TA (1998) Process plants: A handbook of inherently safer design. 2nd ed, Taylor &
Francis, Philadelphia
Lewis EE (1987) Introduction to reliability engineering. Wiley, New York
Li W, Zuo MJ (2008) Optimal design of multi-state weighted k-out-of-n systems based on
component design. Reliab Eng Syst Saf 93(11):1673–1681
Lopes YG, de Almeida AT, Alencar MH, Wolmer Filho LAF, Siqueira GBA (2010) A Decision
Support System to Evaluate Gas Pipeline Risk in Multiple Dimensions. In: Bris R, Soares
CG, Martorell S (eds) European Safety and Reliability Conference (ESREL), Prague, Czech
Republic, 2009. Reliability, Risk and Safety: Theory and Applications. CRC Press-Taylor &
Francis Group, p 1043
Martello S, Toth P (1990) Knapsack problems: algorithms and computer implementations. John
Wiley & Sons, Chichester
Melchers RE, Stewart MG. (1993) Probabilistic risk and hazard assessment. Balkema, Rotterdam
NORSOK (2010) NORSOK Z-013: Risk and emergency preparedness analysis. Rev. 2,
Norwegian Technology Centre
Øien K (2001) A framework for the establishment of organizational risk indicators. Reliab Eng
Syst Saf 74:147–167
Polovko AM, Pierce WH (1968) Fundamentals of reliability theory. Academic press New York
Rathod V, Yadav OP, Rathore A, Jain R (2013) Optimizing reliability-based robust design model
using multi-objective genetic algorithm. Comput Ind Eng 66:301–310
Ren Y, Bechta Dugan J (1998) Design of reliable systems using static and dynamic fault trees.
Reliab IEEE Trans 47:234–244
Sachon M, Paté-Cornell E (2000) Delays and safety in airline maintenance. Reliab Eng Syst Saf
67(3):301–309
Sahoo L, Bhunia AK, Kapur PK (2012) Genetic algorithm based multi-objective reliability
optimization in interval environment. Comput Ind Eng 62:152–160
Stapelberg RF (2009) Handbook of reliability, availability, maintainability and safety in engineering
design. Springer-Verlag, London
US MIL-STD-785B (1980) Reliability Program For System and Equipment Development and
Production, US Military Standard
Vanem E, Endresen Ø, Skjong R (2008) Cost-effectiveness criteria for marine oil spill preventive
measures. Reliab Eng Syst Saf 93:1354–1368
Vrijling JK, van Hengel W, Houben RJ (1998) Acceptable risk as a basis for design. Reliab Eng
Syst Saf 59:141–150
Yanasse HH, Soma NY (1987) A new enumeration scheme for the knapsack problem. Discret
Appl Math 18:235–245
Chapter 11
Decisions on Priority Assignment
for Maintenance Planning
Abstract: This chapter presents multicriteria (MCDM/A) models to classify and

assign maintenance priorities in order to allow maintenance planning to be more
effective. From traditional maintenance planning techniques such as RCM
(Reliability Centered Maintenance), TPM (Total Productive Maintenance) and
others, a common aspect of these techniques is the definition of maintenance
priorities, based on a criticality classification for RCM, for example. As
maintenance planning has to satisfy multiple objectives, such as availability,
maintainability, detectability, safety and reliability besides cost, the maintenance
manager is a decision maker (DM) who has to establish tradeoff amongst multiple
criteria. This chapter presents an MCDM/A model integrated with the RCM
structure using Utility Theory principles to include states of nature and the DM’s
behavior to risk (prone, neutral and averse) in a decision model based on Multi-
attribute utility theory (MAUT). To illustrate situations when a DM has a non-
compensatory rationality, and requires an outranking method, a decision model
based on ELECTRE TRI is applied. In addition, TPM aspects are discussed in
order to emphasize potential MCDM/A problems that may be approached.
11.1 Introduction
Among the decisions regarding maintenance planning, one of the most important
decisions is related to define which kind of maintenance actions are more
appropriated. This decision involves subjective and technical aspects in order to
evaluate the consequences of failures. This chapter presents MCDM/A models
considering the assignment of priorities before establishing a maintenance plan.
A maintenance plan can be defined with different approaches. Selective
maintenance is an example for building a maintenance plan. This approach
includes the specification of each action that should be done for each item in a
multicomponent system, for an interval longer than one cycle, observing the
constraints for optimizing a single objective. Originally, this problem was formulated
considering a fixed time window (Lust et al. 2009).

336 Chapter 11 Decisions on Priority Assignment for Maintenance Planning
In practice, this problem requires the observation of multiple aspects and an

MCDM/A approach enhances the solution by considering multiple aspects, such
as: system performance, costs, total time spent at maintenance, number of repaired
components and availability of spare parts.
Note that, the selective maintenance approach can also be used to build the
annual maintenance plan for items without considering time window constraints.
In the other hand, as the selective maintenance approach requires accurate
information about the changes that occur in each cycle, it may demand too many
information, turning the maintenance plan definition a complex task. Thus, due to
the challenge that is building an annual maintenance plan, it is not revised as much
as it should be.
There are other approaches for building a maintenance plan, which are based
on the definition of the maintenance strategy (Bashiri et al. 2011). It considers that
there is a most appropriate action for each component in order to optimize a
specific criterion. Some authors considered MCDM/A approaches for defining a
maintenance strategy in order to build a maintenance plan (Gómez de León Hijes
and Cartagena 2006; Zaeri et al. 2007; Bevilacqua and Braglia 2000). A literature
review considers MCDM/A models in maintenance (de Almeida et al. 2015) and
points out the increasing number of research dealing with these models.
The maintenance literature considers that while in the selective maintenance
approach it can be difficult to establish a maintenance plan due to its information
requirements, the maintenance strategy selection problem may be too simplistic
and inconsistent with some realities, especially when there is no interest in
updating the strategy, and consequently the maintenance plan defined. Therefore,
practical situations show that maintenance plans should be updated and revised
continuously (Berrade et al. 2013; Berrade et al. 2012; Scarf and Cavalcante
2012).
From this perspective, priority assignment is an important step before
establishing, updating or revising maintenance plans. The definition of which
systems (subsystems, items) or failure modes are more critical for the producing
system mission is an important decision when considering approaches such as
RCM (Reliability Centered Maintenance), used to assist maintenance planning. As
a result, a maintenance plan or strategy is only defined after the defining the
critical systems (subsystems, items) or failure modes in the system. Considering
this principle, the maintenance will be more effective and maintenance plans/
strategy shall be more accurate.
Based on this perspective, this chapter presents different MCDM/A models to
establish criticality before building a maintenance plan. In addition, considerations
are given to traditional approaches, such as RCM and Total Productive Maintenance
(TPM) structure and its integration with MCDM/A models.
11.2 An MCDM/A Model for the RCM Approach 337
11.2 An MCDM/A Model for the RCM Approach
In this section, a quantitative MCDM/A model for evaluating the consequences of

failure is presented (Alencar and de Almeida 2011). The procedure for resolution
of problems and building MCDM/A models presented in Chap. 2 is referenced in
some stages of the model. The model enhance the RCM approach features by
providing a structured decision making process, taking into account uncertainties
and the DM’s preferences.
11.2.1 Traditional RCM Consequence Evaluation
Two important aspects of the traditional RCM approach are presented in this
subsection in order to provide a better understanding of the MCDM/A model
built: the procedure steps and the evaluation of failure consequences.
From the twelve steps introduced in Chap. 3, Moubray (1997) emphasizes these
following steps:
x Establish the functions of each asset within the operating context considering
the associated desired standards of performance;
x Define failures that may occur in the physical asset;
x Identify the failure modes;
x The fourth step involves checking the effects of failure;
x The fifth step is to verify and analyze the consequences of failure;
x Finally, the last step is to establish maintenance actions that could be verified
by applying two techniques: proactive tasks and default actions.
Additionally, RCM approach classifies the consequences of failure into four
categories (Moubray 1997):
x Hidden failure consequences: when it does not present a direct impact, but can
expose the organization to multiple failures with serious consequences
(including catastrophic);
x Safety and environmental consequences: when it presents safety consequences
considering the possibility of injury or death. The environmental consequences
might mean that an organization has violated a national or international
environmental standard;
x Operational consequences: failures which affects only production;
x Non-operational consequences: failures in this category do not affect either
production or safety, involving only the direct cost of repair.
According to the facilities analyzed, a failure could produce irrelevant con-
sequences or compromise essential systems for the organization or society or
safety. In RCM approach, the consequences are evaluated by verifying the impacts
of the effects of a failure mode on system operation, physical security, the

environment, and the economy of the process. Clemente et al. (2012) state that
RCM, when used with other approaches can offer a more complete understanding
of the operational context, providing financial and management information for
decision making.
11.2.2 RCM Based on MCDM/A Approach
Some terms are relevant to build the model such as: the observed context, the
availability of information and its degree of accuracy, the rationality required the
DM’s preference structure and the problematic. An important aspect is the rationality
for the DM in the problem under study that involves a non-compensatory or
compensatory approach. In this sense, the decision model presented in this
subsection sets out to improve the RCM approach by incorporating contributions
from MAUT.
According to de Almeida (2007), MAUT considers that the DM’s preferences
are modeled for computing the MAU function in which the aggregation of uni-
dimensional utility functions must respect MAUT axiomatic structure. Additionally,
Brito and de Almeida (2009) state that MAUT can be applied to aggregate valued
preferences and uncertain consequences related with multiple criteria, providing
results that can be used as input in the process of maintenance management.
The stages of this RCM MCDM/A model are shown in Fig. 11.1. Traditional
RCM steps that remains are: define the functions of assets; identify functional
failures; define failure modes; identify the effects of failure; and, establish
maintenance actions. Therefore, this subsection focuses mainly on the MCDM/A
model built for evaluating the consequences of failures.
For each objective defined, a dimension of consequences is proposed, representing
the objectives of this decision model. Thus, a set of consequence dimensions are
established.
The consequences of failures are evaluated based on five categories defined as
the dimensions of the consequences, in which some of the characteristics differ
from those established by the traditional RCM approach, as follows:
x Human dimension (h): considers the damage with respect to people affected by
the consequences of failures;
x Environmental dimension (e): considers the area affected due to a failure;
x Financial dimension (f): considers the financial losses due to a failure;
x Operational dimension:
– Operational dimension I (o’): considers failures that do not interrupt the
producing system operation;
– Operational dimension II (o’’): considers failures that interrupt the
producing system operation.
Defining the
functions of
physical assets
Establishing
functional
failures
Identifying
failure modes
Establishing
failure effects MAUT quantitative stages (Failure
consequences):
x Establishing consequences
Defining failure dimensions;
consequences x Analyzing the consequences;
x Probabilistic modeling;
x Computing the overall indices
of utility for each failure mode;
x Ranking the alternatives;
Identifying
maintenance
actions
Fig. 11.1 Stages of an MCDM/A model
The identification of state of nature is based on the step 5 of procedure for

resolution of problems and building MCDM/A models presented in Chap. 2.
For evaluating the consequences, elements of decision theory are applied, in
which ș is established as the state of nature. It is used to express the uncertainty
related with the problem. The consequences are represented by c and the set of all
actions under study is represented by A.
A probabilistic approach is applied to incorporate the associated uncertainties
in A considering a probability distribution over consequences and by eliciting
utility functions for these consequences. The probability of each state of nature is
defined as ʌ(ș). U(ș,ai) is the utility when ș and action ai are considered (Berger
1985).
The utility values are defined in an interval scale between [0, 1], where 0 is
associated to the least preferred while the extreme 1 is related to the most
preferred (Keeney and Raiffa 1976). The utility function of these consequences is
shown by (11.1) when the set of consequences is discrete.
U (T , ai ) ¦ P(c T , a )U (c)
c
i (11.1)
Finally, (11.2) shows the utility function of these consequences for continuous
cases.
U (T , ai ) ³ P(c T , a )U (c)dc
i (11.2)
c
Following the step 6, the preference modeling is structured (Vincke 1992;

Keeney 1992). Step 7 includes intra-criteria evaluation, mandatory to define the
functions of the consequences considered.
Step 8 consists of the inter-criteria evaluation, for establishing each criterion
scale constant, ki , and the overall utility function (Keeney and Raiffa 1976).
Assuming an additive MAU function, (11.3) represents the overall utility.
U (h, e, f , o' , o' ' ) k1U (h) k 2U (e) k3U ( f ) k 4U (o' ) k5U (o' ' ) (11.3)
where:
ki is a scale constant that represents the value of the tradeoff;
When the DM’s preferences require limiting tradeoff effects, a model that
considers veto can be incorporated (de Almeida 2013).
The final results are presented by the obtained ranking, established by multi-
attribute utility values found for each failure mode.
The interval scale of utility function allows the incremental value to be
compared to the failure modes Keeney and Raiffa (1976). Applying the interval
scale, it may be affirmed that the difference U(MFx)ȕx – U (MFy)ȕx+1 is M times
greater than the difference U(MFy)ȕx+1 – U (MFz)ȕx+2. This can be seen from the
increment ratio IR of these differences since IR = (U(MFx)ȕx –
U(MFy)ȕx+1)/(U(MFy)ȕx+1) – U(MFz)ȕx +2).
11.2.3 Illustrative Example
For this illustrative example, 16 failure modes are considered, FMx, x=1,2,…,16.
Each FMx is associated with human (h), environmental (e), financial (f),
operational I (o’) and operational II (o’’) consequence dimensions (Alencar and de
Almeida 2011).
There is a prior probability ʌ(șx) associated to each FMx, as can be observed
from Table 11.1.
Table 11.1 A prior probability of failure modes
Component Failure Mode Prior Probability

X1 FM1 0.0766
X2 FM2 0.0256
X3 FM3 0.0578
X4 FM4 0.0333
X5 FM5 0.0835
X6 FM6 0.0259
X7 FM7 0.0768
X8 FM8 0.0493
X9 FM9 0.0876
X10 FM10 0.0087
X11 FM11 0.07
X12 FM12 0.0563
X13 FM13 0.0367
X14 FM14 0.0154
X15 FM15 0.0958
X16 FM16 0.0757
The scale constants k1=0.19, k2=0.13, k3=0.27, k4=0.11, k5=0.30, are elicited
from the DM adopting structured protocols (Keeney and Raiffa 1976).
The interval scale of the utility function allows comparison of the differences in
utility among failure modes. These differences are verified in Table 11.2 (fourth
column).
Table 11.2 Comparisons of differences in utility among failure modes
Ranking position (ȕx) Failure Mode FMi U(FMx)ȕx U(FMx)ȕx - U(FMy)ȕx+1 Difference ratio
ȕ01 FM15 0 0.08788 0.51200
ȕ02 FM9 0.08788 0.17164 4.15593
ȕ03 FM1 0.25952 0.0413 0.40249
ȕ04 FM3 0.30082 0.10261 1.46064
ȕ05 FM5 0.40343 0.07025 0.85598
ȕ06 FM7 0.47368 0.08207 0.70308
ȕ07 FM14 0.55575 0.11673 14.70151
ȕ08 FM8 0.67248 0.00794 0.41966
ȕ09 FM11 0.68042 0.01892 1.09745
ȕ10 FM4 0.69934 0.01724 0.19334
(continued)
ȕ11 FM13 0.71658 0.08917 2.21540

ȕ12 FM02 0.80575 0.04025 0.99187
ȕ13 FM12 0.84600 0.04058 1.33399
ȕ14 FM16 0.88658 0.03042 0.36651
ȕ15 FM6 0.91700 0.083 -
ȕ16 FM10 1 - -
The values presented in Table 11.2 provide important information for the DM.
The difference between the values of the utilities associated with the failure modes
FM14 and FM8 is 0.11673; and the difference between the values of the utilities
associated with the failure modes FM8 and FM11 is 0.00794. The ratio among
differences (fifth column) allows the DM to understand the relative difference
among each FM quantified by the utility scale.
This measure allows to state that the relative difference between FM14 and FM8
is approximately 15 times greater than the difference between FM8 and FM11. It is
important to highlight that these values presented in Table 11.2 reflects the DM’s
preferences among four consequence dimensions. From the numerical example
given, it is possible to observe how undesirable is the differences among such
failure modes for DM.
11.3 An MCDM/A Vision for the TPM Approach
There is no doubt of how the quality of maintenance activities affects the

performance of a producing system. In some cases, system failures occurrence is
affected predominantly from the influence imposed by personnel, whereas the
ageing is a secondary failure mechanism (Levitin and Lisnianski 2000; Wang and
Pham 2006; Scarf and Cavalcante 2012).
Furthermore, the implementation of any model, technique or procedure
developed to support the maintenance effectiveness relies on the maintenance
personnel effort. Thus, tools which get personnel involved and with high
commitment become essential in the operational level.
Total Productive Maintenance (TPM) is a technique in which one of the main
goals is to keep people engaged and motivated to participate in process of
improvements related to maintenance issues. A TPM principle is to bring the
attention of the operator for the signals of non-regular operations, in order to find
and fix some small problems in the system. On the absence of a sophisticated
monitoring system, this is a way to provide the continuous inspection. Therefore,
personnel become a kind of monitoring system.
The Japan Institute of Plant Maintenance (JIPM) created TPM in the 1970s,
during the Japanese quality improvement movement. It considers basic pillars that
11.4 Modeling a Problem for Identifying Critical Devices 343
are divided by topics. Furthermore, TPM has an evolutionary structure, which

makes this technique more flexible to be implemented. Therefore, the attention
can be focused in one phase per time.
Although the importance of the TPM, there are few researches conducted using
an MCDM/A approach in maintenance problems under the TPM framework
(de Almeida et al. 2015). Thus, there still a niche to be explored addressing
MCDM/A decision problems that arise under a TPM program.
Some potential MCDM/A models to explore decision problems in the TPM
context include:
x Maturity evaluation of specific TPM pillars, taking into account multiple
aspects involved for assessing the maintenance program in the organization;
x Priorities assessment of TPM pillars, for the budget allocation and team effort
to improve the potential results from TPM;
x Overall Equipment Efficiency measure, including multiple dimensions through
an MCDM/A approach in order to consider DM’s preferences;
These problems can be addressed using the framework described in Chap. 2 to
build an MCDM/A decision model.
As an example, consider the problem related with assigning priorities among
TPM pillars, the focus is to point out which pillars should receive more attention
during the implementation in order to maximize the chances of the TPM to
succeed. During different moments, the organizations are subjected to an environ-
ment and constraints that would require improvements in different directions.
Similar dynamic environment is considered when considering Goldratt’s Theory
of Constraints, thus specific efforts are deployed to achieve the goals required in
the actual state of the system, in other words, the actual constraint that represents a
bottleneck.
Therefore, the set of alternatives would be related to different pillars com-
binations, which will lead to a portfolio of actions to be prioritized considering the
resources available for the maintenance function. TPM literature recommends a
top-down implementation, which means that the DM would represent the board of
the company.
The criteria considered for such problem includes each one of the pillars, with
their strategic considerations, highlighting the gaps between the status quo and the
organization goals.
11.4 Modeling a Problem for Identifying Critical Devices
This section presents a model for identifying critical devices, which classifies the
items from an industrial plant into predetermined category of criticality. Using the
general procedure proposed in Chap. 2, an MCDM/A model is introduced. For
simplification purposes, only some steps of the Chap. 2 procedure are highlighted
in the model presentation.
Maintenance planning requires a thorough understanding of the system and of
the goals, as well as of the consequence dimensions associated with the failure of
the item. Deciding “what to do” may be based on technical, environmental
and financial aspects. Most of the models try to establish a severity index for
representing a measure in different dimensions in order to support DM for
deciding “what to do”. The weakness of these approaches is that DM’s preferences
are not considered when building such measures.
Facing a large set of piece of equipment, the DM seeks to organize this set into
classes of criticalities. This classification helps the DM to specify the most
appropriated set of actions for each class, considering the adequate resources to be
deployed in a more effective way.
For example, in a power distribution network there are several similar items,
however their location in the network, despite its similarity adds the specific
branch characteristics resulting in different criticality levels, which will point for
different maintenance actions or polices depending on this specific item.
Depending on the item which fails, multidimensional consequence will arise. The
specific location of the item can characterize the number of affected customers,
public services, and result in losses for business supplied in the distribution branch
affected. Depending on the kind of failure, safety aspects also may arise. If such
failure occurs in an underground distribution network, for example, and has the
potential to cause explosions in a high density area, such as in large cities (Garcez
and de Almeida 2014a, Garcez and de Almeida 2014b).
Considering the specific problem presented in this section for assigning devices
into priority classes, by means of an MCDM/A sorting model, based on
ELECTRE TRI, using information about the characteristics of device and the
multidimensional consequences associated with its failures.
ELECTRE TRI was designed to describe actions in ordered categories, it
enables a set of alternatives pre-defined and ranked categories based on multiple
criteria to be classified, as illustrated in Fig. 11.2, where each device xi (x = 1 … n)
is classified according to the device’s criticality.
Class 1 Class 2 .... Class n
X1 X2 X3 X1 X2 X9 X10 X5
X4 X5 X3 .. X6
X6 .. X7 X8
X7 X8
Fig. 11.2 Criticality classification of devices

This MCDM/A classification process can be replicated in different levels,

starting from the equipment, device, component and failure modes level,
successively, as soon as the need information becomes available.
Thus, the MCDM/A model sorts each device into priority classes that supports
the maintenance management, giving an initial filter before addressing the
elaboration of a maintenance plan for the entire plant. Therefore, an application is
presented with an illustrative example of MCDM/A model.
The criteria considered are:
x Safety and environment losses (g1): it refers to the possibility of someone being
injured; or an environmental damage caused by the device failure;
x Financial losses (g2): it considers monetary losses resulting from a device
failure, including repair costs and other costs from downtime;
x Frequency of the device faults (g3);
x Delay-time (g4): expected time elapsed since the arrival of the defect until a
device failure;
x Detectability (g5): it represents the level of difficulty of the fault detection.
The set of alternatives is formed by 10 generic devices {x1, x2, x3, …x10}.
All these criteria are measured considering a semantic scale from 1 to 5. These
scales are detailed for the evaluation of the performance for each device according
to the Tables 11.3-7, respectively for each criterion.
Table 11.3 Safety and environment damage scale
Description Scale
Catastrophic consequence 5
Major consequence 4
Severe Consequence 3
Minor Consequence 2
Trivial Consequence 1
Table 11.4 Financial losses scale
Description Scale
Loss of more than 20,000 monetary unities 5
Loss of 15,001 a 20,000 monetary unities 4
Loss of 0 a 5,000 monetary unities 1
Table 11.5 Frequency of the device faults scale
Description Scale
Failed more than 15 times on interval 5
Failed from 12 to 15 times on interval 4
Table 11.6 Delay-time scale
Description Scale
Mean Delay-time of 0 a 10 time unities 5
Mean Delay-time bigger than 40 time unities 1
Table 11.7 Detectability scale
Description Scale
Almost Impossible detection 5
Difficult detection 4
Moderate detection 3
Easy detection 2
Immediate detection 1
The matrix of consequences should be elicited from a multidisciplinary team,

including experts. It is shown in Table 11.8.
Table 11.8 Matrix of consequences
Alternatives\Criterion g1 g2 g3 g4 g5
x1 1 1 2 1 3
x2 4 5 1 3 4
x3 3 2 3 4 2
x4 3 4 3 4 1
x5 5 5 1 5 1
x6 4 3 2 4 3
x7 1 2 5 2 2
(continued)
x8 2 3 5 3 5
x9 1 1 3 2 2
x10 2 3 4 5 3
Besides the scale defined for each criterion, the definition of the parameters of
the preference functions in order for the ELECTRE TRI. It was defined that the
values of the limits of preference and indifference are both equal to 0. Disregarding
the veto threshold and setting Ȝ = 0.5.
Related to the intra-criteria evaluation, the weights defined by the DM are
given by Table 11.9.
Table 11.9 The weights of the criteria
Criterion Weight
g1 0.25
g2 0.35
g3 0.18
g4 0.12
g5 0.1
In this study, five categories are considered, ordered according to their degree
of importance concerning the priority of planning and conduct of maintenance
actions. The classes considered are:
x Highly critical device: the occurrence of a fault in any device belonging to this
class will bring serious damage to the organization;
x High priority devices;
x Intermediate priority devices;
x Low-priority devices;
x Extremely low-priority devices: i.e., one can, in a way, neglect the maintenance
of equipment belonging to this class so as to direct more concentrated efforts to
the most critical equipment;
The equivalence classes serve as standards by which the devices will be
classified. The equivalence classes adopted for this study are defined by lower and
upper (“Profiles”) limits, as shown in Table 11.10.
Table 11.10 Classes of equivalence and their lower and upper limits
Class Lower Limit Upper Limit

C1 4.5 -
C2 3.5 4.5
(continued)
C3 2.5 3.5
C4 1.5 2.5
C5 - 1.5
The results are presented in Table 11.11.

Table 11.11 Results
Equipment Pessimistic Optimist

x1 C5 C5
x2 C2 C2
x3 C3 C3
x4 C3 C3
x5 C1 C1
x6 C3 C3
x7 C4 C4
x8 C3 C3
x9 C5 C5
x10 C3 C3
Analyzing the results from Table 11.11, only one device was sorted as
extremely critical (x5). This is due to the fact that, if a failure occurs in this device,
there will be catastrophic losses for the company, with regard to matters relating to
the financial, human and environmental dimensions; and besides, this device has a
short delay-time, which deserves careful attention.
The simulation is therefore useful since it enables the manager of maintenance
to drive the maintenance actions in such a way as to focus on the most critical
devices, while it reinforces that equipment considered less important can be neglected.
The impact from the application of an MCDM/A approach on the maintenance
management process may be reflected in the improved operating performance of
the device, due to more efficient maintenance planning for each class of device
having been adopted.
References
Alencar MH, de Almeida AT (2011) Applying a Multicriteria Decision Model So as to Analyse

the Consequences of Failures Observed in RCM Methodology. In: Takahashi RC, Deb K,
Wanner E, Greco S (eds) Evol. Multi-Criterion Optim. SE - 41. Springer Berlin Heidelberg,
pp 594–607
References 349
Bashiri M, Badri H, Hejazi TH (2011) Selecting optimum maintenance strategy by fuzzy

interactive linear assignment method. Appl Math Model 35:152–164
Media, New York
Berrade MD, Cavalcante CAV, Scarf PA (2012) Maintenance scheduling of a protection system
subject to imperfect inspection and replacement. Eur J Oper Res 218:716–725
Berrade MD, Scarf PA, Cavalcante CAV, Dwight RA (2013) Imperfect inspection and replace-
ment of a system with a defective state: A cost and reliability analysis. Reliab Eng Syst Saf
120:80–87
Bevilacqua M, Braglia M (2000) The analytic hierarchy process applied to maintenance strategy
selection. Reliab Eng Syst Saf 70:71–83
Clemente T, Almeida-Filho AT de, Alencar MH, Cavalcante CAV (2013) A Decision Support
System Based on RCM Approach to Define Maintenance Strategies. In: Poels G (ed) Enterp.
Inf. Syst. Futur. SE - 9. Springer Berlin Heidelberg, pp 122–133
de Almeida AT (2013) Additive-veto models for choice and ranking multicriteria decision
Garcez TV, de Almeida AT (2014a) A risk measurement tool for an underground electricity
distribution system considering the consequences and uncertainties of manhole events. Reliab
Garcez TV, de Almeida AT (2014b) Multidimensional Risk Assessment of Manhole Events as a
Gómez de León Hijes FC, Cartagena JJR (2006) Maintenance strategy based on a multicriterion
classification of equipments. Reliab Eng Syst Saf 91(4):444–451
Keeney RL (1992) Value-focused thinking: a path to creative decision making. Harvard
Levitin G, Lisnianski A (2000) Optimization of imperfect preventive maintenance for multi-state
systems. Reliab Eng Syst Saf 67:193–203
Lust T, Roux O, Riane F (2009) Exact and heuristic methods for the selective maintenance
problem. Eur J Oper Res 197:1166–1177
Scarf PA, Cavalcante CAV (2012) Modelling quality in replacement and inspection maintenance.
Int J Prod Econ 135(1):372–381
Zaeri MS, Shahrabi J, Pariazar M, Morabbi A (2007) A combined multivariate technique and
multi criteria decision making to maintenance strategy selection. Ind. Eng. Eng. Manag. 2007
IEEE Int. Conf. IEEE, Singapore, pp 621–625
Chapter 12
Other Risk, Reliability and Maintenance
Decision Problems
Abstract: In this chapter, specific problems in risk, reliability and maintenance

context are described, such as location of backup units, sequencing of maintenance
activities, natural disasters, operation planning of a power system network,
integrated production and maintenance scheduling, maintenance team sizing and
reliability acceptance tests. This chapter presents a multicriteria decision model
with an illustrative application for most of these problems. Amongst the MCDM/A
approaches considered for the illustrative applications in this chapter are: Multi-
attribute utility theory (MAUT), PROMETHEE II, NSGA-II. Regarding the reli-
ability acceptance test an MCDM/A Bayesian approach is presented. For these
problems, several aspects have been considered such as: size of population, degree
of industrialization, the extent of health services (location of backup units); degree
of damage, consumption, electric load, special clients, healthcare services, SAIDI
and SAIFI (sequencing of maintenance activities); human, environmental,
financial and infrastructure concerns (natural disasters); expected tardiness and
maintenance costs (integrated production and maintenance scheduling); waiting
time and cost of personnel (maintenance team sizing); probability of accepting
equipment not in accordance with the reliability specified by the manufacturer;
and delaying the project conclusion (reliability acceptance test). Finally, some
aspects of multiobjective optimization are discussed.
12.1 Introduction
A literature review found 186 papers related to maintenance and reliability

problems based on MCDM/A published between 1978 and 2013. Studies from
various countries contributed to this subject. In fact, more than 30 countries were
identified (de Almeida el al. 2015). This spread around the world is shown in
Fig. 12.1, in which the size of circles indicates the number of such studies found
per country relative to each other.

352 Chapter 12 Other Risk, Reliability and Maintenance Decision Problems
Fig. 12.1 World map of publications on the use of MCDM/A in maintenance and reliability
research
Figure 12.2 shows that there has been an exponential trend in citations which
explains the growth of publications on this subject. The ever increasing amount of
publications indicates the relevance of the topic and the perspectives in the area.
Number of articles per year 19

20 18
17 17
18
15
16
Number of articles
14 12 12
11
12
9
10
7
8 6
5 5
6 4 4 4
3
4 2 2 2 2 2 2
1 1 1 1 1 1 1
2
0
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Year
Fig. 12.2 Number of articles per year on MCDM/A in maintenance and reliability research
Furthermore, the 170 articles considered until 2012 had received 4,306
citations from 1996 to 2013 according to the Scopus database, which represents an
average of 25.33 citations per paper. Fig. 12.3 reflects the impact of this research
area, measured by citations per year since 1996 (de Almeida el al. 2015). For
instance, the articles received 831 citations in 2012.
12.2 Location of Backup Units in an Electric System 353
Number of citations per year

900 831
800 737
700 665
Number of citations
576
600
500 453
400
315
279
300
200 161
92
100 44 55
8 8 12 15 18 14 23
0
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
Year
Fig. 12.3 Number of citations per year on MCDM/A in maintenance and reliability research
From this perspective, this chapter presents RRM problems that arise in
different particular contexts not mentioned in previous chapters.
Amongst the many specific problems that require an MCDM/A approach, this
chapter illustrates RRM problems that require appropriate modeling in order to
allow a DM to consider multidimensional consequences.
Thus, this chapter covers topics related to the following problems:
x Location of backup units;
x Sequencing of maintenance activities;
x Natural disasters;
x Reliability in power systems;
x Integrated production and maintenance scheduling;
x Maintenance team sizing;
x Reliability acceptance testing.
12.2 Location of Backup Units in an Electric System
One of the main objectives of the maintenance function is to minimize the

occurrence of failure, i.e., reduce its frequency. This can be achieved by design
improvements, proper use of assets, preventive maintenance, and condition
monitoring. There is also an interest in minimizing the time spent on corrective
actions when failures occur in order to maximize system availability.
Two significant portions of time are usually considered in corrective actions.
Each has a different impact on the total time spent on maintenance actions. First,
there is the time needed for the logistics i.e. from identifying the failure, placing
the work order, obtaining and preparing the resources needed to perform the
maintenance such as tools, labor, parts; and dislocating the maintenance staff to
the place of the service. When the maintenance team is ready to conduct the
service, there is an elapse of time from service startup to completion, and then the
asset returns to operational status.
In many practical applications, is possible to observe that the maintenance time
needed for a repair has been considered one of the most relevant. On the other
hand, when the resources available to perform maintenance are scarce, some other
factors need to be addressed to minimize the elapsed time for corrective
maintenance actions. In addition to this scarcity, the location of some assets may
be geographically dispersed and may have a strong influence on the time needed
to conduct the maintenance actions.
The example given in this section for location of backup units considers the
context of an electric power distribution company. For this kind of companies,
there are several geographically dispersed systems throughout the distribution
network. Given the geographically dispersion, the maintenance function has to
overcome the logistics obstacles to cope with the required performance standards
that require the availability of equipment located in each of the electric power
substations distributed along the network.
The equipment considered includes high-tension power transformers, which are
heavy, expensive assets and have a high useful life. Such equipment costs millions
of dollars and has a useful life around 30 years. Moreover, the lead-time for
ordering such equipment can take several months, besides the time referring the
logistics of installing the equipment.
Although expensive equipment with a low failure rate cannot justify investment
in a high amount of redundancy for electric power substations, the best use of
resources should be evaluated to deal with an emergency, especially when the
impact on system unavailability is high.
In the case of power transformers, it is known that a limited number of backup
units are available to electric power substations for possible replacements due to a
failure. The decision problem is to define the locations of backup transformers to
minimize the overall consequences of a failure and the need for emergency
replacement.
The consequences of these equipment failures can be characterized as in Chap. 1,
when considering service producing systems. In this particular case, the number of
users affected can escalate from thousands to millions, depending on the con-
sequences of the failure and the blackout effects. Such consequences vary
depending on the equipment location, similar to the example given in Chap. 11
regarding the implications to the priority assignment for maintenance planning
decisions.
To prevent and minimize such consequences, it is essential to plan the location
of backup units in order that it can quickly restore the system in case of failure.
Therefore, the location of backup transformers involves many factors, which
directly influence the operation of a power distribution system. Such process
involves objectives beyond costs related to service interruptions prolonged due to
the absence of a backup unit. These factors have a direct influence on system’s
12.2 Location of Backup Units in an Electric System 355
availability and maintainability (de Almeida et al. 2006; Ferreira et al. 2010;
Ferreira and Ferreira 2012).
Thus, the decision model for this problem seeks to ensure that customers are
minimum affected by the inconvenience of service interruption and the losses
associated. Although a similar problem structure is addressed in classical facility
location (Drezner and Hamacher 2004), different objectives are considered.
A failure in each particular location affects multidimensional consequences.
For this particular problem, three dimensions are considered: number of customers,
health services and local economy affected.
Brandeau and Chiu (1989) give an overview of location problems, which have
been previously studied, with the emphasis on models that were developed in the
field of operations research, formulated as optimization problems, such as the
p-median. The p-median is a classical problem in the field of combinatorial
optimization problem. An algorithm with re-optimization procedures for multi-
objective combinatorial optimization problems is proposed by Bornstein et al.
(2012).
The decision model consists of an MCDM/A p-median model based on three
criteria:
x The size of the population (popi);
x Degree of industrialization (indi);
x The extent of health services (hsi).
The p-median model was adapted to minimize three criteria as shown in (12.1).
The distance factor (dij) represents the distance amongst the electric power
substations. The distance factor works as a multiplier weight in relation to pop, ind
and hs.
ns ns ª K1 U ( pop) ij K 2 U (ind ) ij K 3 U (hs ) ij º

Max ¦¦ «¬« K U ( pop)
i 1 j 1 ij U ( hs ) ij
» xij
¼»
ns
s. t. ¦x
i 1
ij 1; j N
ns , (12.1)
¦x
j 1
jj nb
xij d x jj ; i, j N
xij {0,1}; i, j N
where:
K1, K2, K3, and K are scale constants related to the respective attributes;
N is a set of electric power substations, N = {1, …,ns};
nb is the number of back-up transformers;
popi is the size of population served by the substation i;
indi is the degree of industrialization served by the substation i;
hsi is the extent of health services served by the substation i;
xij is a decision matrix variable, where xij = 1 if the backup transformer of the
substation i is allocated to substation j, and xij = 0, otherwise. And xjj=1 if the
substation j is allocated to store a back-up transformer (a median) and xjj=0,
otherwise.
For the specific application, regarding the inter-criteria evaluation all attributes
are found to be preference independent, except health service and population
criteria. Thus, DM’s preferences over health are affected when varying values of
the size of the population. The model, which represents these conditions, corresponds
to a multi-linear model expressed in the MAU function, given by (12.1).
The objective function represents the utility for substation k, if the backup is
located in substation j. In order to have an indicator for the location of the backup
in substation j, the sums of the utilities of location for all the electric power
substations k should be calculated.
By calculating the maximum utility for each substation, the recommendation
will be to locate a backup transformer in the power substation for which this
transformer will provide the highest maximum utility.
The considered company has to decide which substation should host a backup
transformer from 19 options considered.
As to the multi-attribute p-median model, the results are shown in Fig. 12.4.
The maximum value of the multi-attribute function for nb = 6 is 17.071. The
constant scales represent the preferences of the manager of the current project.
The parameters K1 = 0.2, K2 = 0.5, K3 =0.2, K = 0.1 were obtained using a
structured process suggested by Keeney and Raiffa (1976). The electric power
substations chosen are S = {3, 6, 11, 8, 13, 19}.
1
3 2
8 7
9
6 4
11
13 15 10
5
14
12 16
19
17
18
Fig. 12.4 Example of a solution of the multi-attribute p-median model (S = {3, 6, 11, 8, 13, 19})
12.3 The Sequencing of Maintenance Activities 357
The robustness of this model is verified by a sensitivity analysis of the scale

constants K1, K2, K3 and K .
This model recommendation suggests the best alternative in terms of the trade-
off between the multidimensional consequences and the logistics for restoring the
system availability by using a backup unit.
12.3 The Sequencing of Maintenance Activities
The sequencing of maintenance activities is an important problem although not

necessarily an issue for most maintenance systems. Depending on the size of the
system, the strategies and priorities that have been established are sufficient to
define the sequence of maintenance activities, and so too, of course, are the
technical constraints.
If considering large systems such as an electric power distribution network, a
rail network or a water supply network for example, there are several maintenance
services such as inspection and repairs that must be performed if personnel are
available to do so.
If compared with the sequencing and scheduling of manufacturing orders, there
is a different set of criteria besides cost that shall be considered, such as system’s
availability, quality, service dependability, degradation effects on product quality
and other factors that should be considered due to the nature of the system.
This section describes an MCDM/A decision model according to the general
procedure for building an MCDM/A given in Chap. 2. This MCDM/A model
deals with the planning of maintenance activities by establishing the most
appropriate sequence among a large number of maintenance services. This model
for sequencing maintenance activities has been applied in an electrical power
distributor assisted by a Decision Support System (DSS) (Almeida-Filho et al.
2013).
As the model considers a real situation, the contextual factors related to this
situation have been taken into account when formulating the problem and defining
the model. Considering the size an electrical power distribution network, the
number of repair and inspection services to be performed represents a large
sequencing problem.
This model for sequencing maintenance activity was built from data taken from
a specific Brazilian electrical power distribution network, which extends over
128,412.5 km in order to supply almost two hundred towns, in an area of about
98,500 square kilometers. It has almost 3.1 million customers who consume
12,266,246 MWh per year.
Throughout this network, there are several components such as voltage
transformers, isolators and so forth, which are exposed to severe weather
conditions that degrade these pieces of equipment and age components more
quickly.
The maintenance database for this power distribution network is updated from
data from inspections scheduled on a calendar that covers the entire power
distribution network over a period of ten years, with periodical activities that take
place at one, two, five and then ten year intervals.
The maintenance strategy adopted is that of immediately restoring the system
when a failure that disrupts the service supply occurs or even if this failure is
reported as not disrupting the service but it does expose the population to risk.
The maintenance culture adopted in this context is similar to that described by
Moubray (1997), who set out three typical states related to an equipment mode:
the normal state, a defect and a failure. Depending on the effect of a failure on the
functioning of the item, the failure is classified either as:
x A potential failure, which is an observable condition, which implies there will
be a functional failure if no preventive action is taken (Moubray 1997);
x A functional failure, which means the inability of an item of equipment to
perform a specific function within desirable operational limits (Moubray 1997).
The focus of this problem on sequencing maintenance activities is to do with
the potential failures, which are identified by inspections included in the calendar
and recorded on the maintenance information system. Thus, these potential
failures are prioritized to avoid a disruption to the service and its consequences for
strategic and operational objectives. The sequence of maintenance services is
based on MCDM/A, which define the order among services to be performed.
In the particular problem addressed, there were about 25 thousand potential
failures identified by inspections. Given the capacity of the maintenance division
workforce and the annual budget set aside for preventive maintenance, only four
thousand potential failures can be tackled per year. The practical meaning is that
potential failures are sequenced taking their priority into account and are corrected
over the year, which leaves a set of potential failures to be reevaluated by
maintenance services in the following year together with other new potential
failures identified in inspections undertaken as per the inspection calendar. This
preventive maintenance budget is usually defined in the sense that the potential
failures identified and pending repairs are still at a tolerable level in the light of
the organizational targets. There are several models regarding the definition of
preventive maintenance time interval (Jiang and Li 2002; Shafiee and Finkelstein
2015), although there are practical situations when the DM has to consider also the
resources available and the production scheduling to define the exact time of
a maintenance repair or replacement. Sect. 12.6 illustrates this problem with
a decision model.
Some of these organizational targets are defined in order to meet regulatory
aspects, such as those defined by ANEEL, which is the Brazilian government
agency responsible for regulating the generation of electrical power, its
transportation and the distribution companies involved. This agency defines
operational and service levels for these companies, and has the power to levy fines
in accordance with the regulatory rules. Another important issue is that the
12.3 The Sequencing of Maintenance Activities 359
electrical power tariff is proportional to the service quality provided. Thus,

improving the service level reflects directly on the company´s revenue. There are
two main measures of quality of service considered by ANEEL: DEC and FEC.
DEC is related to the duration of service disruptions whenever these occur and
FEC considers the frequency of disruption to the service (ANEEL 2012).
The reliability indices used by ANEEL are similar to those defined by IEEE
(2012), where DEC corresponds to the System Average Interruption Duration
Index (SAIDI) and FEC to the System Average Interruption Frequency Index
(SAIFI).
Given the MCDM/A nature of this problem, the decision model requires a
method that allows the preference among criteria to be elicited in order to find the
most adequate sequence of maintenance repairs to be performed. For this specific
decision model, the PROMETHEE II method was used. PROMETHEE II is one
of the methods of the PROMETHEE family which have been evolving since 1982
(Brans and Mareschal 1984; Brans and Mareschal 2002).
The choice of this method is justified as it can provide a complete ranking
order that considers a wide range of value functions and has an easy-to-understand
elicitation procedure to assess the DM’s preferences. Thus, an important factor for
choosing this MCDM/A method is related to the simplicity with which it elicits
and requires parameters. This is important as it consolidates the DM’s readiness to
understand the recommendations provided from the decision model.
Another important issue is the calculation process. Given that there are about
25 thousand alternatives and that this number may grow, an MCDM/A method
needs to be able to give a response within an appropriate interval of time so DMs
may build scenarios and conjectures and use sensitivity analysis.
With regard to PROMETHEE II, the literature raises questions regarding rank
reversal when new alternatives are added to the sets of alternatives, which is a
frequent issue when using methods based on a pair-wise comparison process.
Mareschal et al. (2008) presented conditions when this situation may occur, which
is restricted to very limited situations. This is one of the reasons for choosing this
method rather than other outranking methods for the decision model.
The PROMETHEE II method allows the DM to choose between six different
value functions, namely, defining each criterion as the usual criterion; a u-shape
criterion; a v-shape criterion; a level criterion; a v-shape with an indifference
criterion; or a Gaussian criterion (Brans and Mareschal, 1984; Brans and
Mareschal, 2002).
PROMETHEE II uses pairwise comparisons throughout its process to aggregate
preference indices and outranking flows. Equation (12.2) represents the preference
indices, and expresses to what degree a is preferred to b over all the criteria, where
k
W ¦ w , and wj represents the weight of criterion j, w t 0 .

j j
j 1
1 k
S ( a, b) ¦ w j Pj (a, b) (12.2)
W j1
Equation (12.3) represents the net outranking flow, which consists of the
difference between the positive and the negative flow of an alternative a. Based on
the net outranking flow, a complete pre-order is provided that ranks all
alternatives.
1 ªn n º
I (a) «¦ S (a, b) ¦ S (b, a )» (12.3)
n 1 ¬« bbz1a b 1
bza ¼»
The criteria to be considered in the MCDM/A model were assessed together

with a DM when structuring the problem. Thus, each type of potential failure had
specific characteristics due to the location of the equipment. This has to be
considered since similar failures, which are located in different segments along the
distribution network, would cause different consequences and damages.
Thus, the set of criteria identified for this decision model is:
x Degree of Damage (to installation and people, a verbal scale is used);
x Average Affected Consumption;
x Electric Load;
x Percentage of Regional Network Electric Load (considering the network
branch);
x Special Clients Affected (subjected to regulatory special rules);
x Healthcare Services;
x Slack on DEC/SAIDI (difference between branch DEC/SAIDI and Aneel target
for DEC/SAIDI);
x Slack on FEC/SAIFI (difference between branch FEC/SAIFI and Aneel target
for FEC/SAIFI);
x Political Consequences of a Failure.
The main screen of the DSS is presented in Fig. 12.5. MCDM/A concepts and
the company’s maintenance culture have been combined into a DSS. Therefore,
the DSS draws on Moubray’s RCM critical levels and uses verbal scales to
determine the level of degradation of the equipment.
The DSS allows scenario and application notes to be recorded, so information
regarding the decision process for the maintenance activities planning, including
personnel involved may be retrieved later, and also to be compared with different
scenarios when such evaluation is required. The top section of the screen shown in
Fig. 12.5 represents such input information.
The lower section of the screen in Fig. 12.5 shows the interface between the
parameters of the MCDM model. On the left, there is the list of criteria, followed
by the type of preference function chosen for each criterion and a scroll button for
setting the parameters of the preference function. On the right, the input of
weights is displayed numerically and graphically.
12.4 Natural Disasters 361
Fig. 12.5 DSS decision model parameters- main screen
After inputting the MCDM/A model parameters into the required fields for the
decision model and performing the PROMETHEE II method, the sequencing of
the maintenance orders is obtained by considering their priority according to the
set of criteria under which all preventive maintenance orders were evaluated.
The DM can also generate reports on preventive maintenance orders which
takes account of the budgetary constraints and of an analysis of the sensitivity
analysis report. The DSS also enables the DM to perform a scenario analysis
supported by graphs so that he/she can compare the effectiveness of each action
while considering costs and the managerial objectives (DEC and FEC), and can
evaluate the cost levels incurred to carry out preventive maintenance orders as
against losses from potential failures (such as, in revenue and because of fines),
these being the consequences if the failure became a functional one.
It is interesting to observe that some prioritized maintenance actions may
not prove to be financially effective. However, they do prevent losses in other
dimensions, such as service quality, which is monitored by the regulatory agency
(ANEEL) or any special clients affected, for example. This illustrates the
importance of considering the MCDM/A nature of such problems; if this is not
done, these factors would not be given appropriate consideration.
12.4 Natural Disasters
It is well-known that there has been an increase in research studies on natural

hazards and the relationship of the latter with the climatic changes that have been
occurring around the world in recent years. Additionally, human migration to

urban areas and the consequent growth in and/or density of the population in
urban areas increases the impact of natural disasters significantly.
Urban settlement worldwide is becoming more and more evident. Half of the
world’s population now resides in urban areas. There is an expectation that these
numbers will increase in the coming decades (Linnekamp et al. 2011). A large part
of the urban population lives in coastal areas, where the impacts of specific effects
of climate change potentially have the most critical consequences. In this context,
Li et al. (2014) states that knowledge about future extreme events is important to
support actions in order to define more appropriate safety levels for the society.
Thus, the increase of population density in many regions and cities has a direct
impact by provoking the occurrence of events that lead to financial losses (billions
of dollars). Keller and DeVecchio (2012) state that natural disasters affect the
lives of millions of people around the world as a result of events such as flooding,
earthquakes and hurricanes, which lead to an annual loss of around 80,000 human
lives, besides economic losses of approximately 50 billion dollars per year.
Considering these facts, natural disaster risk management is crucial if the most
appropriate mitigating actions are to be appropriately planned and taken. Solecki
et al. (2011) point that climate change has a direct impact on such risks. Climate
change such as temperature variations and oscillations in precipitation patterns can
have a direct impact on the probability of extreme events taking place. Changes in
the intensity and distribution of rainfall might well increase the occurrence of
flooding or water rationing. High temperatures and melting glaciers may well lead
to the sea level being raised, thus increasing the chance of severe flooding in
coastal regions. Models in flooding context have also considered risk analysis
(Hansson et al. 2013; Vari et al. 2003).
However, some observations need to be made about some obstacles that hinder
the better management of risks from natural disasters and of assessing such risks.
One of them concerns the availability of a reliable data base, since the dynamics
of the social context (e.g. significant changes in the demographic occupation and
use of land) associated with climate change often make data collected in previous
periods of little or no current value. Keller and DeVecchio (2012) assert that,
nowadays, there is a need for effective risk assessment under different scenarios
for hazards that need to be associated with the analysis of natural disasters. Due to
the occurrence of climate change past events often fail to provide adequate
information on what may happen today or in future.
Moreover, population migration itself sees to it that different aspects, in distinct
time intervals and in the same locality, should be recorded as well as different
aspects related to vulnerability, an aspect that should be taken into account in the
risk management process regarding natural disasters. Vulnerability, according to
Pelling (2003), can be defined from the degree of exposure to natural hazards, and
the ability of the area and community affected to prepare for and recover from
given negative impacts.
Pine (2009) states that the changes in disasters frequency may be the result of
natural climatic variations that occur over a time period or arise from changes of
variables that impact the frequency or severity of environmental change. The
intensification of human activity in hazardous areas such as the construction of
residences without planning permission on hills subject to landslides or on land
known to be occasionally subject to severe flooding are examples. Additionally,
changes to the environment (such as those caused by buildings, technology and
the infrastructure to support human habitation) that lead to the degradation of
natural systems can also increase the severity of the hazard.
Thus, it is of fundamental importance to consider the dynamic nature of risk
and vulnerability. Karimi and Hüllermeier (2007) reinforce this idea by stating
that, due to there being all manner of uncertainty types, evaluating the risk of
losses due to natural disasters is a complex activity, mainly for lack of sufficient
physical knowledge and inadequate statistical data with respect to the origin,
characteristics and consequences of each disaster that has actually taken place.
Bobrowsky (2013) states that risks related to natural hazards and climate
change are not autonomous or externally generated. Therefore, society ought to be
able to react, adapt or respond to them. These risks are the result of the interaction
between society and the natural or built environment. Consequently, risk
management requires a better understanding of this relationship and the factors
influencing it.
Unlike the aspects considered in traditional risk management, in natural
disaster environments, some additional concepts are important so that DMs can
evaluate the situation more adequately. Hence there is a need to consider aspects
such as vulnerability and resilience, besides the concept of risk. The concepts of
risk, vulnerability and resilience are important in studies on natural disasters, and
are used as an approach to understand the dynamics of natural disasters (Paul
2011).
According to Field et al. (2012), vulnerability is the result of different conditions
and processes, which must include considering historical, social, political, cultural,
institutional and environmental matters and natural resources. Resilience is
defined in the context of natural disasters as a means of promoting sustainable
livelihoods, which enables individuals or systems to be able to cope with an
extreme event without using all the available resources (Paul 2011). Resilient
systems tend to reduce physical damage, thereby providing time for the environ-
ment to recover after an extreme event has occurred. Therefore, it reflects the
interest in improving the capacity of human and physical systems to respond to
natural events that occur.
According to Field et al. (2012), the risk of a disaster can be understood as the
possibility of adverse effects in the future arising from interaction between social
and environmental processes, and a combination of physical hazards and
vulnerabilities in the exposed elements. The simultaneous consideration of risk,
vulnerabilities and dynamic changes in the different phases of crises and disasters
produces a complex scenario out of which the degree of risk and vulnerability that
this contains needs to be identified and assessed, as should the measures that need
to be taken to mitigate risk and to adapt strategies. An understanding of extreme
events and disasters is a prerequisite for drawing up adaptation strategies related to
climate change and reducing risk in disaster risk management.
Natural disasters manifest themselves independent of the pre-existing states of
economic, social and physical environmental. Therefore, infrastructure, services
and organizations are prone to being affected by an event triggered by a natural
phenomenon (such as an earthquake or a flood) or a technical event like an
explosion or gas leakage (De-Leon 2006; Guikema 2009). Thus, it is observed that
a disaster is preceded by at least two aspects: the possibility that an initiated event
occurs, usually termed a danger from this potential state, and; a pre-existing
vulnerability. In other words, there is a predisposition for people, processes, infra-
structure, services, organizations or systems to be affected, damaged or destroyed
when an event occurs.
Considering the coupled with vulnerability and danger as a prerequisite of there
being a risk of a disaster, the exposure can be considered as another prerequisite.
The exposure is understood as the number of people and/or other elements at risk
that may be affected by a particular event (Thywissen 2006). Among other
definitions, risk must be understood as a function of hazard, vulnerability,
exposure and resilience.
Another issue discussed in studies related to natural disasters is consequence
analysis. The occurrence of an event or combination of two or more events can
cause different impacts in different dimensions.
The potential effects of climate change on natural hazards are an input to the
formulation of strategies to adapt risk management practices using knowledge
developed about the risks associated with people and with economic impacts
(Zischg et al. 2013).
Extreme natural events can induce higher losses especially when they occur in
vulnerable and/or areas that are densely populated (Huttenlau and Stotter 2011).
Risk analysis in natural hazards is used for estimating the consequences in order to
provide information to the public and DM. It is observed in this type of analysis
that different perceptions of the concept of risk are seen to have different goals
and to take different approaches, as given in Chap. 3, there are different
approaches when considering individual or social risks. Therefore, depending on
the approach given to the risk evaluation, different perspectives may be considered
and depending on how the consequences are evaluated an MCDM/A approach is a
more suitable manner to address the problem of evaluating multidimensional
consequences. Considering the complexity of consequence evaluation activity,
MCDM/A approaches allow to include different types of losses.
There are some aspects that may be considered for measuring the effects of
natural disasters, amongst which are economic and social disruption and
environmental impacts. Social disruption can, for example, include the number of
people made homeless or the incidence of crime such as the number of homicides,
arrests, the extent of civil disorder, including riots and street-fighting (Pine 2009).
Economic disruption may be associated with unemployment, lost work days,

loss of production volume, decrease in sales and in tax collection. The environ-
mental impacts can be evaluated for cost recovery, re-establishing water or sewer
systems, the number of days of unhealthy air, or the number of warnings that
involve not eating fish or restrict the use of water (Pine 2009).
These aspects can be adjusted depending on the type of situation analyzed,
taking into account, for example, the kind of loss resulting from the natural
disaster having occurred. Therefore, losses can be classified into: direct tangible
losses and indirect losses. In the first type of loss, the losses considered are those
that occur immediately after the event such as deaths, injuries and repair costs.
Indirect losses involve loss of income due to unemployment, sales losses, pro-
ductivity losses, disease and increase in the crime rate (Pine 2009).
Impacts can also be broadly classified by making distinctions between social
and physical impacts, where physical impacts include property damage, deaths
and injuries. Social impacts can be more difficult to measure once it develops over
a long time run. A better understanding of social impacts is important to enable
appropriate contingency plans to be drawn up to prevent and/or minimize adverse
effects from extreme events. The social impacts of natural disasters are often
broken down into demographic, economic, political, institutional, psychological
and health impacts (Paul 2011).
A more critical situation is the possibility of a natural disaster occurring in
industrial areas which can increase the chance of occurrence of events with
extremely catastrophic consequences. According to Krausmann et al. (2011) the
threat of natural disasters impacting on chemical industries, refineries, nuclear
power plants and pipelines and the consequent leakage of hazardous substances
have been recognized as an emerging risk in today’s society. Industrial accidents
from natural events such as earthquakes and floods are mentioned in many studies
on Natech accidents. Natech accidents can generate leaks of hazardous substances
leading to deaths, injury to persons, environmental pollution and economic losses.
Natech risks differ from technological and natural risks by requiring a risk
management approach that is integrated and more complex. One of the main
problems with this type of scenario is the simultaneous occurrence of a natural
disaster and a technological accident, both requiring simultaneous response efforts.
In addition, the leakage of hazardous materials can be inducted by a single source
or multiple sources simultaneously, from various hazardous installations in the
area impacted by a natural disaster.
According to Krausmann and Cruz (2013), a practical example can be found in
the earthquake and tsunami that hit Japan on March 11, 2011, damaging and
destroying many industrial plants and killing more than 16,000 people, with more
evidence due to the effects of the Fukushima’s nuclear power plant facility. This
event shows that even well-prepared countries are subject to the occurrence of
Natech events. In the case of natural disasters that hit a wide impact area, multiple
and simultaneous leaks of hazardous materials may occur, these being more severe
in areas which are close to residential areas.
Girgin and Krausmann (2013) highlight that Natech risks is likely to be more
frequent in the future due to industrial growth, changes in the patterns of
occurrence (due to climate change) of natural disasters and the fact that society is
becoming increasingly vulnerable, the more interconnected it becomes –
something that is happening with each passing day.
In conclusion, organizations and countries should regard the search for more
effective risk management as a fundamental goal. Adequate control of risk and
monitoring should be done in the best way as possible, thereby minimizing the
occurrence of catastrophic consequences.
To illustrate how an MCDM/A model may address such aspects, the following
section structures an MCDM/A model for multidimensional risk evaluation
considering flooding as an example of natural disaster.
12.4.1 An MCDM/A Model that Evaluates the Risk of Flooding
There are several natural hazards related to climate changes and global warming.
In this section, an illustrative example is presented of using MCDM/A to evaluate
risk considering multidimensional consequences specifically for one of the most
frequent natural disasters, namely flooding.
Therefore, an MCDM/A model is described (Priori Jr. et al. 2015) focusing on
specific aspects, including the occurrence of different events/scenarios, the choice
of criteria, different methods and the distinct rationality required from the DM`s
preference structure.
Some steps of the general procedure for building MCDM/A models proposed
in Chap. 2 are mentioned throughout the presentation of the model.
This risk evaluation considers urban areas located in coastal regions at sea
level, or even below sea level, for example, in the Netherlands.
In underdeveloped countries, there are poor communities that are more exposed
than others to flooding due to the lack of infrastructure. This vulnerability puts
people at risk from landslides as a consequence of rainfall even without flooding
affecting the safety of such communities.
A probabilistic background is necessary for this type of evaluation. Thus, a risk
hierarchy can be built, based on Utility Theory, for the most critical areas by
assigning priorities to risks with a view to reducing or mitigating them in order to
allocate the available resources better and to a level above the local safety
standards applied. Therefore, when considering step 2 of the general procedure for
building MCDM/A models, the overall objective is to assign priorities to risk so as
to guide how resources will be allocated (Lins and de Almeida 2012). According
to step 3, human, environmental, financial and infrastructure are considered
consequence dimensions. The hierarchical structure of these dimensions and
attributes is shown in Fig. 12.6.
Natural hazards
(Consequences)
Human Environmental Financial Infrastructure
Communication
Impact area
No fatalities
Transport
Buildings
Drainage
Fatalities
Energy
Cost
Fig. 12.6 Hierarchical structure of the consequence dimensions and attributes
Specifically, it is important to note that for the infrastructure dimension there

are different attributes. The definition of these attributes is based on a World Bank
report (Jha et al. 2013). Additional comments are presented for each consequence
dimension and attributes considered from after the occurrence of the natural
hazard:
x Human consequences (h): This dimension considers fatalities and injuries (no
fatalities) as possible consequences;
x Financial consequences (f): This dimension considers the financial losses that
arise from the occurrence of the event such as the production losses;
x Environmental consequences (e): This dimension considers the area impacted
(including the area covered by vegetation area, fauna and flora);
x Infrastructure consequences (s): This dimension considers different attributes.
In the building attribute, the current state of the edification is considered, taking
into account the possibility of its structural collapse. In the Energy attribute, the
physical structure of the power grid is analyzed. With regard to the drainage
attribute, aspects of the drainage facility are taken into account. As to the
communication attribute, the communication facility is studied. Finally, with
respect to the transport attribute, the operation of existing transports systems is
evaluated.
Based on step 4, there is a discrete set of elements A = {a1, a2, a3, ..., an},
defined as limited urban areas, in which the extent of the urban area is established
including some factors, such as infrastructure, number of inhabitants, topography
and climate.
Step 5 deals with identifying the state of nature. The DM has to consider a set
of controllable and uncontrollable inputs that impacts the problem analyzed.
Additionally, the DM makes decisions that take account of states of nature and
probabilities for several consequences. Therefore, utility functions are defined so
as to represent the DM’s preferences for different consequences (Cox 2009).
Elements of decision theory are used to evaluate the consequences. The
consequences are represented by c and the set of alternatives by A. The state of
nature ș, represents the uncertainty related to the problem, measured by the
magnitude of the rainfall. The state of nature is represented by a continuous set
denoted by real numbers regarding the rate of rainfall in a given region per hour
(mm/h), in a determined rainfall event. Lognormal or Gamma probability density
functions could be applied to represent the mm/h for a determined rainfall event in
a specific location (Cho et al. 2004).
As to considering consequences of a rainfall, a probabilistic approach can be
introduced to incorporate the associated uncertainties in A, considering a
probability distribution over consequences given the state of nature. By eliciting
utility functions for these consequences, DM’s preferences are represented in the
model. The prior probability ʌ(ș) is introduced as the probability of each state of
nature. Therefore, the expected utility E[U(ș,ai)] is used to represent the risk
associated with each given alternative (Berger 1985).
The utility is calculated by combining the probability of the consequences c in
A, the consequence function P(c|ș,ai). The effects of the tide on run-off rainwater
can be included in this probabilistic mechanism. Therefore, the expected utility
E[U(ș, ai)] of these consequences is represented by (12.4).
E >U (T , ai )@ ³ P(c T , a )U (c)dc

i (12.4)
c
According to Berger (1985), the loss function is defined as the negative of the
utility function L(ș, ai) = - E[U (ș, ai)]. Thus, the losses are computed for each
criterion (L(h); L(f); L(e); L(s)) considering the urban area analyzed and the state of
nature. Moreover, the risk to an urban area is defined by (12.5).
r (ai ) ³T S (T ) L(T , a )dT

i i (12.5)
Step 6 from Chap. 2 on preference modeling is required in order to evaluate the

DM’s preferences. This model considers that the DM’s preferences satisfy MAUT
axiomatic requirements for an additive utility function.
Thus, the intra-criterion and inter-criteria evaluations consider steps 7 and 8
respectively. The intra-criterion evaluation is based on the conditional utility
function, defined for each dimension. The inter-criteria evaluation relies on the
12.5 Operation Planning of a Power System Network 369
additive utility function, achieved by elicitation procedures through lotteries

(Keeney and Raiffa 1976). Therefore, (12.6) represents the additive utility function.
U (ai ) k hU (h) k eU (e) k f U ( f ) k sU ( s ) (12.6)
where kh, ke, kf, ks are scale constants for the human, environmental, financial and
infrastructure dimensions respectively.
From the hierarchical structure of the attributes, the infrastructure dimension s
and the human dimension h present specific attributes. These attributes are
considered in the MAU function as given by (12.7).
U (ai ) k h >k h1U (h1 ) k h2U (h2 )@ keU (e)

(12.7)
k f U ( f ) k s >k s1U (h s1 ) ... k s5U (h s5 )@
where:
kh, ke, kf, ks are a scale constants that represent the value of the tradeoff
(dimensions);
kh1, kh2, ks1,…, ks5 are a scale constants that represent the value of the tradeoff
(specific attributes).
Step 9 consists of applying an algorithm in the decision model in order to
evaluate the set of alternatives. In this model, the interval scale of the utility
function is applied to provide additional information based on comparing the
utility values and ratios of the increments of utility between alternatives.
Finally, step 10 of the procedure for resolving problems and building MCDM/A
models presented in Chap. 2 consolidates step 9 by conducting a sensitivity
analysis to verify the robustness of the model, incorporating the data and
parameters analyzed.
12.5 Operation Planning of a Power System Network
The demand for electric power has increased and all current forecasts indicate
even higher growth due to improvements in the quality of life, besides population
growth. As mentioned in Sect. 12.4, when discussing the effects and trends of
changes in the climate, consequences in generating and consuming electric power
can also be observed. The effects of climate changes can limit power generation
capacity, when considering renewable sources, and in terms of environmental
constraints for generation when considering other sources such as coal and oil. On
the other hand, the demand for electric power rises due to severe weather such as
colder winters or higher temperatures in summer. Therefore, the impact of power
outages becomes more and more critical due to the great dependence of modern
society on power always functioning.
Thus, since there are many social and economic issues that are rising in
importance, there are many MCDM/A problems regarding decisions on the
reliability of power systems.
From this perspective, due to the threat of irreversible damages to the
environment, there is a need to consider various forms of energy production in
order to satisfy the aspects related to demand, environment and cost (Jebaraj and
Iniyan 2006).
Besides environmental and social changes, electric power systems have
evolved and have been restructured according to the country’s needs, and these
depend on each country’s specific energy policies. Therefore, there is a need to
emerge from the unique view of provision of power at “minimum cost” to a
broader perspective that allows consideration of multiple aspects, and this may
consider the different interests of the actors involved in dealing with planning
energy systems (Diakoulaki et al. 2005).
For power systems predominantly based on hydropower generation, such as
Canada and Brazil, additional complexity is introduced when depending on this
power source. Such complexity concerns planning the power system due to the
dynamics of river flows and precipitation patterns in such energy producing
systems.
As to step 5 of the general procedure for building MCDM/A models given in
Chap. 2, the state of nature T reflects the potential energy generation stored in
water reservoirs over time. Depending on the context, different levels of uncertainty
are found. For example, while in Canada river flows are more predictable due to
the relation with the accumulated volume of snow layers; in Brazil the prediction
of potential energy generation stored in water reservoirs is more difficult since
there is no such relation (Albuquerque et al. 2009).
Considering the importance of potential energy generation forecasts, the
operational planning of these power systems is much more complex than in power
systems higher percentages of coal, oil and nuclear generation sources. The
operational planning of generating power in such systems, which are not
predominantly dependent on hydropower, is not highly associated with matching
temporal to spatial data (Diniz and Maceira 2008).
Therefore, many MCDM/A problems arise in order to assure the supply
reliability. Another aspect that must be considered is how the electric power
system is designed, as this may result in different kinds of constraints and
consequences over the system.
The objective of the system operator is to assure that the demand will be met
with minimum cost and the maximum reliability of system supply. Thus the
planning of such systems must be formulated with an MCDM/A paradigm that
takes into account besides cost and the reliability of system supply, other aspects
related both to the quality of service (such as voltage, power-frequency and
harmonics) and to environmental impacts.
Thus, a reliability decision problem in power systems includes several
objectives and a variety of constraints which reflect the physical system (Pinto
12.6 Integrated Production and Maintenance Scheduling 371
et al. 2013). In the classical optimization approach, it is usual to remove important

objectives and consider doing so as one of the problem constraints. The MCDM/A
approach enriches the decision process by allowing a compromise solution,
beyond constraint levels defined when modeling an objective as a constraint, to be
evaluated. Therefore a DM can make tradeoffs and find the most suitable
recommendation if he/she uses an appropriate methodology for modeling.
What is more and more observed is the need to consider other criteria regarding
environmental issues due to the atmosphere being increasingly polluted which has
already been experienced in China. As a result, the emission of greenhouse gases
is another aspect that has been covered when considering MCDM/A approaches
for power system planning (Diakoulaki et al. 2005; Batista et al. 2011).
Therefore, in order to meet the exhortations of environmental regulations,
environmental aspects become a new criterion for planning and operating of
power systems (Farag et al. 1995; Yokoyama et al., 1988; Wong et al., 1995).
12.6 Integrated Production and Maintenance Scheduling
Production scheduling models can support decision making on allocating jobs in

manufacturing systems in order to optimize a given objective function. Generally,
objective functions are related to the productivity of the system, such as:
maximum tardiness, total tardiness, total weighted tardiness, total weighted
completion time, maximum lateness, number of tardy jobs and makespan (Pinedo
2012). However, machine breakdowns can result in losses in terms of the
productivity performance measured by these objective functions. In other words,
the solution found for a specific problem, which assumes that failures are not
possible, can be unrealistic when breakdowns occur.
In order to deal with breakdowns, a maintenance policy can recommend
preventive actions with the objective of reducing the probability of machine
failures. This means that performing preventive maintenance necessarily incurs a
cost and time must be set aside for this. Besides having to take stochastic features
of failures and repair of machines into account, maintenance performance can be
in conflict with production performance. In this context, the main issue of this
problem is how to balance maintenance and production objectives.
According to Aghezzaf and Najid (2008) most of the time, a contingency
review of the production plan, due to a failure, is very expensive and also impacts
the quality of products. Therefore, preventive maintenance has an essential role to
play, not only to ensure the production plan is fulfilled by reducing the number of
failures, but also to ensure quality and service within appropriate levels.
Independently of the kind of system, an appropriate production scheduling
enables production systems to achieve strategic objectives, which might range
from achieving minimum cost and tardiness. Therefore, it seems that, in most
cases, dealing separately with the production schedule and maintenance plan does
not work in practice. Indeed, it is not possible to ensure long lasting results for the
production perspective, since production scheduling does not last for long due to
failures. Nevertheless, a very common hypothesis considers that equipment is
always available during the scheduling period, even though the probability of
failure in intensively used production systems has a significant value (Allaoui
et al. 2008).
Thus, it is interesting to include the activity of preventive maintenance in a
production schedule in an integrated form (Ángel-Bello et al. 2011). According to
Allaoui et al. (2008), in the literature, there are two particularly prominent
approaches with regard to the problem of integrating production and preventive
maintenance. For the first kind, the optimum maintenance schedule in the
production system can be determined. The second approach comprises optimizing
the scheduling of production by considering a preventive maintenance plan. By
doing so, the maintenance schedule decision could be drawn up in advance of the
production schedule. The problem with this approach is that the dynamic nature of
the problem is overlooked.
Despite there being some interesting papers dealing simultaneously with a
maintenance and a production schedule, most of them consider only one decision
criterion (Alardhi et al. 2007; Benmansour et al. 2011; Ji et al. 2007; Sortrakul and
Cassady 2007; Su and Tsai 2010). In fact, since these integrated models derive
from the original problem of the production schedule, some of these papers still
consider only the original objectives such as total weighted expected tardiness.
Therefore, maintenance features that influence joint scheduling are dealt with as
secondary aspects, mostly as elements of constraint.
It is worth stating that maintenance aspects are completely different from the
common criteria used to define the production schedule. Instead of simply having
some rules based on the strategy of client satisfaction, such as, expected tardiness
and makespan; maintenance aspects are related to the performance of the
equipment, such as availability, probability of finishing by the end of the schedule,
the total cost, considering preventive maintenance and interruption; and so on.
Thus, it is not difficult to realize that the operational and maintenance aspects are
complementary.
An integrated decision model that takes into account two objectives to be
optimized simultaneously: minimizing total weighted expected tardiness and
minimizing expected maintenance costs is developed. The conflicts of the
maintenance function and production are dealt with under the approach of
MCDM/A. Some results give evidence that on applying NSGA-II (Deb et al.
2002), satisfactory solutions can be found for the integrated scheduling problem.
It is assumed that there are a number of jobs to be scheduled in a single
machine in a production system. Each job has a fixed processing time, due date
and importance weight. In addition to production scheduling, it is assumed that
this machine may be unavailable due to preventive maintenance or repairs that are
needed due to failures. These features imply a conflict between the production and
maintenance objectives. Whereas the production objective may be related to
12.6 Integrated Production and Maintenance Scheduling 373
minimizing tardiness in finishing jobs, the maintenance objective may be related

to minimizing time losses incurred by unnecessary maintenance actions and is
mainly characterized by the expected cost of maintenance. To estimate the latter, it
is assumed that the time to failure of this machine is governed by a Weibull
probability distribution (Sortrakul and Cassady 2007). Replacements should be
recommended when the expected cost of replacement is lower than the cost of
preventive maintenance and additional costs including production losses.
It is assumed that jobs cannot be preempted by preventive maintenance
activity, and only one failure can occur during the processing of a job. The basic
decision variables to be determined are: what the sequence of the jobs and when
preventive maintenance actions should be performed with the objective of
minimizing the total weighted expected tardiness and expected maintenance cost
(Cassady and Kutanoglu 2003; Sortrakul and Cassady 2007).
The mathematical model is defined by (12.8), in order to minimize two
objective functions. Let F1 be the total weighted expected tardiness and F2, the
total expected cost of maintenance.
n i
minimize F1( xij , y i ) ¦
i 1
w[i ] ( ¦T
k 0
[ i ,k ]S [i ,k ] )
(12.8)
n i
minimize F2 ( xij , y i ) ¦¦ cm
i 1 k 0
[i ,k ]S [i ,k ]
The maintenance cost is given by (12.9), job completion time is given by

(12.10) and tardiness is given by (12.11).
i
cm[i ,k ] cb ¦y
l 1
[l ] ca k k 0,1,..., i; i 1,..., n (12.9)
i i
c[i ,k ] tp ¦
l 1
y[l ] ¦p
l 1
[l ] kt r k 0,1,..., i; i 1,..., n (12.10)
T [i ,k ] max(0, c[i ,k ] d [i ] ) k 0,1,..., i; i 1,..., n (12.11)
where:
n - Total number of jobs to be scheduled;
p[i] - Processing time for the i-th job performed;
d[i] - Due date for the i-th job performed;
w[i] - Weight for the i-th job performed;
c[i] - Completion time for the i-th job performed;

ș[i] - Tardiness for the i-th job performed;
ȕ - Shape parameter of the Weibull distribution;
Ș - Scale parameter of the Weibull distribution;
a[i] - The age of machine immediately after finishing the i-th job;
a [i-1] - The age of machine immediately before processing the i-th job;
y[i] - Binary variable decision when preventive maintenance is performed prior to
the i-th job;
ʌ[i,k] - probability mass function of k failures during the i-th job;
tp - Duration of preventive maintenance action;
tr - Duration of corrective repairs to a machine;
cb - Cost of preventive maintenance action;
ca - Cost of corrective repairs to a machine.
The result of a simulation of NSGA-II is presented in Fig. 12.7. As can be seen,
nine solutions obtained with the algorithm were found in PFtrue.
Integrated Production and Maintenance Scheduling
205
200
195
F2 - Maintenance Cost
190
185
180
175
170
165
1350 1400 1450 1500 1550 1600 1650 1700
F1 - Tardiness
True Pareto Front Pareto Front Know n by NSGA-II
Fig. 12.7 Result of NSGA-II in the integrated production and maintenance scheduling
The integrated model proposed may well be of interest to industry in order to

tackle production and maintenance needs which take into account two conflicting
objectives related to tardiness and maintenance cost with regard to scheduling
production jobs.
12.7 Maintenance Team Sizing 375
12.7 Maintenance Team Sizing
Maintenance team sizing is a topic that involves various types of methodologies.

Simulation and queuing theory are examples of approaches that can be used to
determine the best number of maintenance personnel. Such approaches aim to
minimize the waiting time and service costs due to losses associated with an
inappropriate team size, high investments and strategic reasons. What should be
taken into account are both: the cost of hiring personnel and the estimated cost of
the consequences of the unavailability of the system, which can be represented by
the cost of production losses. Hillier (1963) proposes economic models that
minimize the total cost, which comprise of the expected costs and service costs.
Queuing theory allows the DM to analyze the problem using a structure that
can incorporate the probabilistic mechanism present in the reliability and
maintainability of systems. A maintenance system can consist of several queues.
Customers are devices that need repairs and the servers are personnel who perform
repair services, which may eventually form a virtual queue waiting for service.
Some queuing system indicators show stochastic features that can support
decisions about maintenance team sizing. Examples are the utilization factor of the
personnel; the probability of finding n customers in the system, the probability
that all the servers are busy, the average number of items in the queue and the
average time spent in the queue waiting for the equipment and system.
Two maintenance models are investigated in a flexible manufacturing system
using queuing theory to study features of the system utilization. The failure of a
machine requires the activation of a stand-by, while the failed unit goes to repair.
A stand-by is required to perform a certain level of service. For these types of
systems to minimize the cost of loss of production, which includes the cost of
customer dissatisfaction, these costs can be minimized by maximizing system
availability (Lin et al. 1994). A bi-objective formulation to solve a maintenance
workforce sizing problem is found by using a branch and bound algorithm
(Ighravwe and Oke 2014).
The sizing of the maintenance team can be defined for corrective or preventive
maintenance. In some situations, estimating failure and repair rates from historical
data may be difficult and an expert’s knowledge could be useful. The use of prior
knowledge deals with the uncertainty in a more appropriate way. Therefore, a
decision model for maintenance team sizing with use of prior knowledge is
developed.
The model considers a system with p maintenance teams, denoted by MTi,
where i = 1, ..., p. Each maintenance team MTi is responsible for the repair of qi
different items of the equipment j. Each piece of equipment, denoted by Eqij, has
nij items. Each MTi has team size si. Each maintenance team has similar
characteristics with respect to repair time. A generic representation of the
maintenance system is shown in Fig. 12.8.
Eq11 n11
MT1 s1
...
Eq12 n12 ...
...
Eq1q n1q
...
Fig. 12.8 Example of a maintenance system
In general, a combination of reliability, maintainability and cost directly

influences the system performance measures. In some situations, it may be
interesting to acquire more reliable items than to increase the size of maintenance
teams. Reliability and maintainability features of the system are represented by the
failure and repair rate, respectively.
Assuming an exponential probability distribution for the equipment reliability
function, Ȝ represents the equipment failure rate, with different values for each
type of equipment and this corresponds to the arrivals of the customers in queuing
system of the maintenance team.
The repair times are modeled by an exponential distribution, where the constant
ȝ is the repair rate, which may have different values for each maintenance team.
The objective of this problem is to determine the required number of
maintenance personnel to achieve satisfactory levels in the performance indicators
and associated costs.
The system can be represented by p models of one-queue, s servers and an
infinite population. Based on the Kendall notation, the decision model developed
for the maintenance team sizing problem is represented by type M/M/s. A
structure of the decision model proposed is shown in Fig. 12.9.
12.7 Maintenance Team Sizing 377
Identifying types and quantities of equipment
Estimating the prior probability distribution ʌ(ȝ) and ʌ(Ȝ)
Determining the minimum and maximum number of personnel smin and smax
Computing the expected value of the loss function for the interval [smin, smax]
Establishing the trade-off between the system cost and the waiting time
Carrying out sensitivity analysis of the parameters of the model
Selecting the maintenance team size
Fig. 12.9 Structure of decision model with use of prior knowledge, ș = [Ȝ, ȝ]
The objective of the decision model is to define the team size while dealing
with the tradeoff between service level and cost. Thus, the problem is solved by
maximizing the MAU function given by (12.12).
§f f ·
U ( si ) k1U 1 ¨
¨ ³³ L(O , P , s ) S (O )dO S ( P )dP ¸ k 2U 2 c p s
¸
(12.12)
© f f ¹
The expected queue length, L(Ȝ, ȝ, s), is given by (12.13) and the maximum
number of servers, smax , is given by (12.14).
s
§O·
¨¨ ¸¸ O P
©P¹ 1 §O·
L (O , P , s ) ¨¨ ¸¸ (12.13)
s 1!( s P O ) 2 s 1
1 §O·
n
1 §O·
s
§ sP · ©P ¹
¦ n! ¨¨© P ¸¸¹
n 0
¨ ¸
s! ¨© P ¸¹
¨¨ ¸¸
©sP O ¹
1 1 4O
smax 1 (12.14)
2 2 P 'U critical
where:
Ȝ - failure rate;
ȝ - repair rate;
s - number of servers;
cp - cost of personnel
ʌ(Ȝ) - prior probability on failure rate;
ʌ(ȝ) - prior probability on repair rate;
ǻȡcritical - critical value of difference in terms of factor utilization.
Considering an illustrative example with ʌ(Ȝ) and ʌ(ȝ) given by Weibull
probability distributions with ȕ=11.3 and Ș=0.05154 for ʌ(Ȝ); and ȕ=10.7 and
Ș=0.01153 for ʌ(ȝ). In addition, an additive function with scale constants k1 = 0.10
and k2 = 0.90 for the MAU function given by (12.12).
The minimum number of servers smin to satisfy the stationary condition of 99%
is found by considering the Ȝmax and ȝmin from the inverse Weibull function.
Therefore, smin = 8.
Based on (12.14), the maximum number of servers is defined as smax = 23.
Thus, for this scenario, when maximizing the MAU function, the best maintenance
team size would be s*=10. This result is illustrated in Fig. 12.10.
Maintenance team sizing
0,98
0,96
MAU function
0,94
0,92
0,9
0,88
0,86
8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
Num ber of servers
Fig. 12.10 U(si) for si as [smin, smax]
To evaluate the robustness of the solution recommended, a sensitivity analysis

is presented in Table 12.1. From the achieved results, it is possible to realize that
the scale parameters of prior distributions are the most sensitive parameters in the
model.
12.8 Bayesian Reliability Acceptance Test Based on MCDM/A 379
Table 12.1 Sensitivity analysis of parameters of the model
ȕ of ʌ(Ȝ) Ș of ʌ(Ȝ) ȕ of ʌ(ȝ) Ș of ʌ(ȝ) k1 k2

+20% +5% +20% +20% +20% +20%
-20% -20% -20% -5% -20% -20%
The result observed from the sensitivity analysis indicates that the analyst
should give more attention to the elicitation process that obtained the scale
parameters from the experts. As these are the most sensitive parameters, this
elicitation process has to be more accurate than other parameters not so sensitive.
12.8 Bayesian Reliability Acceptance Test Based on MCDM/A
The operations and maintenance planning of systems requires the use of

information on the reliability of equipment, which is usually provided by the
manufacturers. A concern about the state of nature (TO) that reflects the real
reliability of this equipment has triggered the need to ensure, by contract, that
reliability acceptance testing takes place (de Almeida and Souza 2001; de Almeida
and Souza 1986).
The number of equipment failures during the phase of operation trials that can
occur in the set of equipment ordered is limited, so that the actual failure rate TO is
in accordance with the specified Ȝ0.
The problem considered in this section regards the decision about the
acceptance of TO for given equipment during the phase of operation trials, in order
to decide whether or not to return the equipment. If it is, this may delay the start of
the industrial plant project. If the decision is to return the equipment to the
manufacturer, this implies delaying the completion time of the project. Therefore,
the DM has to consider the tradeoff between delivering the industrial plant project
on time which may adversely affect the reliability of the project by accepting
equipment not in accordance with the specified Ȝ0; and delaying the project
conclusion so as to assure the reliability requirements of the project. Therefore,
this is an MCDM/A problem with two clear objectives (de Almeida and Souza
2001; de Almeida and Souza 1986).
Thus, the decision is more than just testing the hypothesis on TO with regard to
Ȝ0 , thus requiring the DM’s preferences to be evaluated for each specific situation
which may lead to different decisions depending on the specific priorities or
aspects involved. Thus a DM may decide to delay the conclusion of a project if
safety requirements are compromised due to the TO , for example, or else he/she
may decide to accept equipment with lower reliability in order to conclude the
project on time.
As an illustrative example, consider a study consisting of certain sampling

restrictions considering N=36 new items and an observation time, ¨t, of 3 months
in the phase of operation trials. Thus, for this observation time, x failures are
observed in the 36 items, representing TO. This considers that the unit failure rate
Ȝu specified by the equipment’s manufacturer as Ȝu= 5.88x10-6 per hour; Ȝ0= Ȝu N
¨t = 0.457 for the set of 36 items, in a time interval of 3 months.
The function P(x|TO) corresponds to the probability of x failures occurring,
given a true failure rate TO. Thus, the number of failures x is explained by a
Poisson process.
Thus, the DM seeks to find out the number of failures that can occur in the
population of 36 items during the phase of operation trials, so that the actual
failure rate TO is compatible with the specified Ȝ0.
There are three approaches for addressing this problem: Hypothesis testing
under the Neymann-Pearson approach; a decision problem formulation with a
Bayesian criterion; and a definition of minimax estimators with a Bayesian
estimator for the failure rate TO (de Almeida and Souza 2001; de Almeida and
Souza 1986).
A model using the specified MTBF considers that the x failures observed are
random, and not caused by improper operation or by external effects or by faults
in the manufacturing process. Furthermore, it is considered that in the beginning
of the phase of operation trials, premature failure have been removed after burn-in
testing and debugging. Thus, it is possible to conclude that the failure rate is in the
operational phase of the bathtub curve that is constant over time.
Considering a hypothesis test, the state of nature becomes the equipment’s
failure rate, which can be represented by a discrete set such as ș = {ș0 and ș1}, in
which ș0 means that TO Ȝ0 and ș1 means that TO > Ȝ0. Moreover, it is assumed that
there is no initial knowledge about the state of nature TO, and the Neyman-Pearson
approach can be applied.
By using this Neyman-Pearson approach, the problem is reduced to the choice
of the best decision rule that minimizes the risk Rb for a given ș, subject to the
constraint that the Rb risk to the other ș is less than or equal to a predetermined
level Į as given by (12.15).
min Rb (T1 )
(12.15)
s.t. Rb (T 0 ) d D
This formulation corresponds to testing the null hypothesis H0: that the
equipment has a lower failure rate than or equal to a specified against the
alternative hypothesis H1. This means that H0: TO Ȝ0 and H1: TO > Ȝ0.
There are two errors involved in the hypothesis test that should be minimized.
The probability Į of rejecting the null hypothesis when it is true, Rb(ș0), and the
probability ȕ of incorrectly accepting the null hypothesis, Rb (ș1). These errors are
known as error Type I and error Type II, respectively.
12.8 Bayesian Reliability Acceptance Test Based on MCDM/A 381
While the DM prefers to increase the probability Į to reduce the ȕ probability,

the equipment manufacturer seeks to reduce probability Į in order to increase the
ȕ probability.
Usually, when considering statistical hypothesis tests a value of 0.05 is adopted
by convention. However, for this specific case, if a DM chooses an Į level by
convention, that DM’s preferences have not been considered and therefore, neither
has the context of the problem.
Using the Bayesian approach the DM considers an Į level that meets his/her
expectations. To solve this problem a decision rule has to be defined (de Almeida
and Souza 2001; de Almeida and Souza 1986) in order to minimize risk rd as
given by (12.16).
° i 1 b ½°
min ® ¦ ³ S (T )[ L(T , a0 ) L(T , a1 )] e T e x dT ¾ (12.16)
i
°̄ x 0 x! a °¿
where:
ș - state of nature;
ʌ(ș) - prior probability distribution;
L(ș,ai) - Loss function.
The interval [a, b] corresponds to the given range of TO in a prior distribution.
The solution that minimizes (12.16) represents the maximum number of failures
that there are for not rejecting the null hypothesis. If during the phase of operation
trials, there occurs a greater number of failures than the solution of (12.16), the
null hypothesis must be rejected.
This procedure using a Bayesian approach can support a DM to find the
maximum number of failures that would be acceptable, considering the objectives
and knowledge available.
Thus, when considering reliability acceptance for equipment there are con-
flicting objectives as discussed previously regarding the system’s reliability and
the project being delivered on time.
Therefore, in some situations a DM can be concerned with evaluating of the
tradeoff between the error Type I dimension and the delay in delivering the project
due to rejecting the equipment.
For example, if a DM is interested in the evaluating whether the reliability
defined in the contract is consistent with the actual reliability of the purchased
items, he/she may consider a tolerance level. He/she does so in order to include an
upper limit greater than the equipment’s nominal reliability value, which is usually
the one set out on the contract, for the null hypothesis. Thus, this tolerance level
represents how much the DM is willing to tradeoff in terms of reliability in order
to succeed in delivering the project on time. Therefore, this means that the DM
may decide to accept the null hypothesis since the reliability is lower than the one
defined by contract, but respects the tolerance level accepted by the DM in order
to deliver the project on time.
This decision may involve a substantial delay in project execution time from
several months to years due to the specificities of the items that may be rejected.
For these kinds of items, the lead time involved with the purchasing process and
delivery is sufficiently long to compromise delivering the project on time. This is
because the manufacturer usually adopts a make-to-order strategy for this kind of
equipment which means that in most cases, it is not available in the short term i.e.
immediately after ordering the equipment.
Thus, for dealing with a decision that has such relevant and strategic objectives,
the MCDM/A approach provides techniques and methods for modeling the DM’s
preferences in order to give a recommendation on the reliability acceptance test
considering the broader aspects involved in the problem. Therefore, the general
procedure defined for building MCDM/A models in Chap. 2 gives the directions
to build a suitable decision model for a problem with these features.
12.9 Some Multiobjective Optimization Models on Reliability

and Maintenance
This section is divided into two topics based on the generation of MOEAs
(Multiobjective Evolutionary Algorithms).
12.9.1 Approaches in the 1980s and 1990s
The first papers using multiobjective formulation to find Pareto solutions were
published in the 1980s. Dewispelare (1984) formulated a non-linear multiple
objective problem with regard to a pre-production decision on an airborne tactical
missile where the reliability, survivability, combat effectiveness, cost and flight
area were considered as objective functions. Feasible space was explored for all
non-dominated solutions obtained by a constrained optimization technique.
Although non-dominated solutions should be found, a scalar scoring function is
recommended when the DM is not able to make a choice due to the incomplete
ordering of the Pareto solutions set.
Soltani and Corotis (1988) constructed a trade-off curve of a design for
structural systems as a result of using multiobjective linear programming obtained
and a constrained optimization technique to formulate objective functions of cost
of failure versus initial cost.
Fu and Frangopol (1990) found Pareto optimal solutions in a multiobjective
formulation of structural systems considering three objectives: weight, system
reliability and redundancy. They used the H-constraint method to find Pareto
solutions.
12.9 Some Multiobjective Optimization Models on Reliability and Maintenance 383
Misra and Sharma (1991), Dhingra (1992) and Rao and Dhingra (1992) used
MOEAs for redundancy allocation, as discussed in Chap. 9.
12.9.2 Approaches in the 2000s and 2010s
With the development of the Second Generation of MOEAs close to 2000, several
studies have developed with a view to evaluating the effectiveness of these
techniques and more often from this point on, in the field of maintenance and
reliability, such as: NSGA-II, SPEA2 and other approaches.
The use of the second generation of MOEAs in Reliability and Maintenance
problems has become one of the most common approaches across the Pareto-front
approaches. Some cases from the literature are highlighted in Table 12.2.
Table 12.2 Some Pareto-front approaches used in Reliability and Maintenance problems
Optimization Method References

Dewispelare (1984), Soltani and Corotis (1988), Fu and
Constrained optimization technique; Frangopol (1990), Dhingra (1992), Rao and Dhingra
H-constraint method; goal (1992), Barakat et al. (2004), Azaron et al. (2009),
programming; goal-attainment Moghaddam (2013)
Min-max concept; Exact algorithm; Misra and Sharma (1991), Certa et al. (2011), Chou and Le
PSO; GPSIA (2011)
Ramirez-Rosado and Bernal-Agustin (2001), Marseguerra
et al. (2002), Marseguerra et al. (2004), Kumar et al.
(2006), Kumar et al. (2008), Cadini et al. (2010), Moradi
et al. (2011), Wang and Hoang (2011), Chiang (2012),
MOEA; MOGA; NSGA-II
Torres-Echeverria et al. (2012), Zio et al. (2012), Gjorgiev
et al. (2013), Jin et al. (2013), Li et al. (2013), Lins et al.
(2013), Rathod et al. (2013), Trivedi et al. (2013), Zidan
et al. (2013)
These approaches have been applied to several reliability and maintenance

problems, such as:
x Design selection (Ramirez-Rosado and Bernal-Agustin 2001; Marseguerra
et al. 2004; Barakat et al. 2004; Azaron et al. 2009; Chiang 2012; Torres-
Echeverria et al. 2012; Rathod et al. 2013);
x Maintenance strategy selection (Marseguerra et al. 2002);
x Service restoration (Kumar et al. 2006; Kumar et al. 2008);
x Power system planning (Cadini et al. 2010; Zio et al. 2012; Gjorgiev et al.
2013; Jin et al. 2013; Li et al. 2013; Trivedi et al. 2013; Zidan et al. 2013);
x Preventive maintenance (Certa et al. 2011; Chou and Le 2011; Moradi et al.
2011; Wang and Hoang 2011; Moghaddam 2013).
Ramirez-Rosado and Bernal-Agustin (2001) applied a multiobjective evolutionary

algorithm to determine the set of non-dominated solutions in the design of
distribution systems for two objective functions: economic costs and reliability.
Marseguerra et al. (2002) considered a continuously monitored multi-
component system and used a genetic algorithm and Monte Carlo simulation to
determine the optimal degradation level beyond which preventive maintenance
has to be performed in order to optimize two objective functions: profit and
availability. A multiobjective genetic algorithm approach was also applied in
nuclear safety system by Marseguerra et al. (2004). They considered two
objectives: unavailability and the variance of its estimate.
Barakat et al. (2004) proposed the use of an H-constraint method when
designing pre-stressed concrete beams, and set minimizing the overall cost and
maximizing the reliability of the system and of its flexural strength as objectives.
The H-constraint method decomposes the multiobjective optimization into a series
of single objective optimizations. The procedure involves minimizing a primary
objective, and expressing the other objectives in the form of inequality constraints.
Consequently, the entire Pareto set can be obtained by varying the H value.
Kumar et al. (2006) introduced an NSGA-II model for service restoration in a
distribution system using three objectives: out-of-service area, number of switch
operations and losses. Kumar et al. (2008) used an NSGA-II model for service
restoration considering various practical operational issues in a distribution
system, such as priority customers, presence of remotely controlled, as well as
manually controlled switches. The same objective functions as those defined in
Kumar et al. (2006) were used.
Azaron et al. (2009) found Pareto solutions in a cold-standby redundancy
scheme using genetic algorithms and the goal attainment method in order to
minimize the initial purchase cost of the system, to maximize its MTTF (mean
time to failure), to minimize its VTTF (variance of time to failure) and also to
maximize its reliability during the mission time.
Cadini et al. (2010) studied the optimal expansion of an existing electrical
power transmission network using multiobjective genetic algorithms with two
objectives: maximizing reliability and minimizing cost.
Certa et al. (2011) evaluated when maintenance actions should be undertaken
in order to assure the required reliability level until the next fixed stop for
maintenance, thereby minimizing the global maintenance cost and the total
maintenance time. They proposed an exact algorithm that is able to find the whole
optimal Pareto frontier.
Chou and Le (2011) used a multiobjective particle swarm optimization
(MOPSO) technique in order to optimize the reliability and cost of roadway
pavement maintenance.
Moradi et al. (2011) investigated an integrated flexible job shop problem with
preventive maintenance activities, thereby optimizing two objectives: minimizing
makespan and system unavailability. Four evolutionary algorithms are compared,
12.9 Some Multiobjective Optimization Models on Reliability and Maintenance 385
NSGA-II, NRGA, CDRNSGA-II and CDRNRGA. A composite dispatching rule

(CDR) was included in the last two.
Wang and Hoang (2011) used an NSGA-II approach in order to optimize
availability and the cost of an imperfect preventive maintenance policy for
dependent competing risk systems with hidden failure.
Chiang (2012) discussed a multiobjective genetic algorithm integrated with a
DEA approach to create an optimal design chain partner combination with total
expected cost, total expected time for product development and product reliability
as objective functions.
Konak et al. (2012) dealt with a multi-state multiple sliding window system
problem and used NSGA-II where each failure type constitutes a minimization
objective.
Torres-Echeverria et al. (2012) used a multiobjective genetic algorithm
approach to design and test safety instrumented systems using NSGA-II and set
three objectives: those of calculating the average probability on demand of
dangerous failure, the spurious trip rate and the lifecycle cost.
Zio et al. (2012) analyzed the vulnerability of the Italian high-voltage electrical
transmission network in which the most critical groups of links were identified.
A multiobjective genetic algorithm approach was carried out. Two objective
functions are considered: the betweenness centrality of a group of edges and the
cardinality of the group of edges.
Gjorgiev et al. (2013) recommended a multiobjective genetic algorithm for
scheduling the optimal generation from a power system for which they set three
objectives: those of minimizing cost, emissions and unavailability.
Jin et al. (2013) proposed a multicriteria model based on genetic algorithms to
design and operate a wind-based distributed generation with two objective
functions: cost and reliability.
Li et al. (2013) formulated a multiobjective optimization model for protecting
against cascading failures in complex networks based on the principles of NSGA-
II with three objective functions: those of minimizing global connectivity loss,
local connectivity loss, number of lines switched-off.
Lins et al. (2013) evaluated a multiobjective genetic algorithm to select the
design for a security system which had two objectives: those of calculating
the probability of a successful defense and of minimizing the acquisition and
operational costs.
Moghaddam (2013) used a goal programming technique integrated with a
Monte Carlo simulation to determine Pareto-optimal preventive maintenance and
replacement schedules for a repairable multi-workstation manufacturing system
which had been experiencing an increasing rate of the occurrence of failures.
Three objective functions were evaluated: costs, reliability and availability.
Rathod et al. (2013) proposed a multiobjective genetic algorithm for a reliability-
based robust design optimization problem where seven specific objective
functions were defined using the first version of the NSGA.
Trivedi et al. (2013) addressed day-ahead thermal generation based on genetic

algorithms using three objective functions: scheduling operation cost, emission
cost and reliability. The population is ranked using the constrained-domination
principle of the constrained NSGA-II.
Zidan et al. (2013) modeled how to plan a distribution network using NSGA II
with two objective functions: an economic function involving costs of line upgrades,
energy losses, switching operations required for network reconfiguration, and
distributed generation capital, operation and maintenance costs, and an environ-
mental function involving emissions from both grid and distributed generation
units. Decision variables are defined such as switch status, line to be upgraded,
distributed generation size, location and type, and year in which each decision is
to be implemented.
References
Aghezzaf E-H, Najid NM (2008) Integrated production planning and preventive maintenance in
deteriorating production systems. Inf Sci (Ny) 178:3382–3392
Alardhi M, Hannam RG, Labib AW (2007) Preventive maintenance scheduling for
multícogeneration plants with production constraints. J Qual Maint Eng 13:276–292
Albuquerque LL de, Almeida AT de, Cavalcante CAV (2009) Aplicabilidade da programação
matemática multiobjetivo no planejamento da expansão de longo prazo da geração no Brasil
(Multiobjective mathematical programming applicability in long-term expansion planning of
generation in Brazil). Pesqui Operacional 29 :153–177
Allaoui H, Lamouri S, Artiba A, Aghezzaf E (2008) Simultaneously scheduling n jobs and the
preventive maintenance on the two-machine flow shop to minimize the makespan. Int J Prod
Econ 112:161–167
Almeida-Filho AT de, Ferreira RP, de Almeida AT (2013) A DSS Based on Multiple Criteria
Decision Making for Maintenance Planning in an Electrical Power Distributor. In: Purshouse
R, Fleming P, Fonseca C, et al. (eds) Evol. Multi-Criterion Optim. SE - 58. Springer Berlin
ANEEL (2012) Agência Nacional de Energia Elétrica, Brazil. Qualidade do serviço (Quality of
service). Available at http://www.aneel.gov.br/area.cfm?idArea=79&idPerfil=2. Accessed 05
Mai 2012
Ángel-Bello F, Álvarez A, Pacheco J, Martínez I (2011) A heuristic approach for a scheduling
problem with periodic maintenance and sequence-dependent setup times. Comput Math with
Appl 61:797–808
Azaron A, Perkgoz C, Katagiri H, et al. (2009) Multi-objective reliability optimization for
dissimilar-unit cold-standby systems using a genetic algorithm. Comput Oper Res 36:1562–
1571
Barakat S, Bani-Hani K, Taha MQ (2004) Multi-objective reliability-based optimization of pre-
stressed concrete beams. Struct Saf 26:311–342
Batista FRS, Geber de Melo AC, Teixeira JP, Baidya TKN (2011) The Carbon Market
Incremental Payoff in Renewable Electricity Generation Projects in Brazil: A Real Options
Approach. Power Syst IEEE Trans 26:1241–1251
Benmansour R, Allaoui H, Artiba A, et al. (2011) Simulatiońbased approach to joint
production and preventive maintenance scheduling on a failuréprone machine. J Qual
Maint Eng 17:254–267
References 387
Berger J (1985) Statistical Decision Theory and Bayesian Analysis (Springer Series in Statistics).
Springer, New York
Bobrowsky PT (ed) (2013) Encyclopedia of Natural Hazards. Springer, Dordrecht
Bornstein CT, Maculan N, Pascoal M, Pinto LL (2012) Multiobjective combinatorial
optimization problems with a cost and several bottleneck objective functions: An algorithm
with reoptimization. Computers and Operations Research, 39 (9): 1969-1976
Brandeau ML, Chiu SS (1989) An Overview of Representative Problems in Location Research.
Manage Sci 35:645–674
Brans JP, Mareschal B (1984) PROMETHEE: a new family of outranking methods in
multicriteria analysis. Operational Research 84. Brans JP (eds). Amsterdam: North-Holland,
pp. 408–421
Brans JP, Mareschal B (2002) Prométhée-Gaia: une méthodologie d’aide à la décision en
présence de critères multiples. Éditions de l’Université de Bruxelles
Cadini F, Zio E, Petrescu CA (2010) Optimal expansion of an existing electrical power
transmission network by multi-objective genetic algorithms. Reliab Eng Syst Saf 95:173–181
Cassady CR, Kutanoglu E (2003) Minimizing Job Tardiness Using Integrated Preventive
Maintenance Planning and Production Scheduling. IIE Trans 35:503–513
Certa A, Galante G, Lupo T, Passannanti G (2011) Determination of Pareto frontier in multi-
objective maintenance optimization. Reliab Eng Syst Saf 96:861–867
Chiang T-A (2012) Multi-objective decision-making methodology to create an optimal design
chain partner combination. Comput Ind Eng 63:875–889
Cho H-K, Bowman KP, North GR (2004) A Comparison of Gamma and Lognormal
Distributions for Characterizing Satellite Rain Rates from the Tropical Rainfall Measuring
Mission. J Appl Meteorol 43:1586–1597
Chou J-S, Le T-S (2011) Reliability-based performance simulation for optimized pavement
maintenance. Reliab Eng Syst Saf 96:1402–1410
Media
de Almeida AT, Cavalcante CAV, Ferreira RJP, et al. (2006) Location of Back-up Transformers.
Eng. Manag. Conf. 2006 IEEE Int. IEEE, Salvador, Bahia, pp 300–302
de Almeida AT, Souza FMC (1986) Bayes-Like Decisions in Reliability Engineering.
In: Proceedings of International Conference on Information Processing and Management of
Uncertainty in Knowledge-Based Systems, Paris, 30 June-4 July, 87-90
De León JCV (2006) Vulnerability: A conceptual and methodological review. ‘Studies of the
University: Research, Counsel, Education. No. 4/2006, Institute for Environment and Human
Security (UNU-EHS), Bonn, Available at https://www.ehs.unu.edu/file/get/8337.pdf.
Accessed 16 Jan 2014
DeWispelare AR (1984) A computer based application of non-linear multiple objective
optimization. Comput Ind Eng 8:143–152
Dhingra AK (1992) Optimal apportionment of reliability and redundancy in series systems under
multiple objectives. IEEE Trans Reliab 41:576–582
Diakoulaki D, Antunes CH, Gomes Martins A (2005) MCDA and Energy Planning. Mult.
Criteria Decis. Anal. State Art Surv. SE - 21. Springer New York, pp 859–890
Diniz AL, Maceira MEP (2008) A Four-Dimensional Model of Hydro Generation for the Short-
Term Hydrothermal Dispatch Problem Considering Head and Spillage Effects. Power Syst
IEEE Trans 23:1298–1308
Drezner Z, Hamacher HW (2004) Facility location: applications and theory. Springer Science &
Business Media, New York
Farag A, Al-Baiyat S, Cheng TC (1995) Economic load dispatch multiobjective optimization
procedures using linear programming techniques. Power Syst IEEE Trans 10:731–738
Ferreira RJP, de Almeida AT, Ferreira HL (2010) Multi-attribute p-median model for location of
back-up transformers. Brazilian J Oper Prod Manag 7(2):09–28
Ferreira RJP, Ferreira HL (2012) Decision support system for location of back-up transformers
based on a multi-attribute p-median model. Syst. Man, Cybern. (SMC), 2012 IEEE Int. Conf.
IEEE, Seoul, pp 629–631
Field CB, Barros V et al. (eds) (2012) IPCC – Managing the risks of extreme events and
disasters to advance climate change adaptation. A Special Report of Working Groups I and II
of the Intergovernmental Panel on Climate Change. Cambridge University Press, New York
Fu G, Frangopol DM (1990) Balancing weight, system reliability and redundancy in a multi-
objective optimization framework. Struct Saf 7:165–175
Girgin S, Krausmann E (2013) RAPID-N: Rapid natech risk assessment and mapping
framework. J Loss Prev Process Ind 26(6):949–960
Gjorgiev B, Kanþev D, ýepin M (2013) A new model for optimal generation scheduling of
power system considering generation units availability. Int J Electr Power Energy Syst
47:129–139
Guikema SD (2009) Natural disaster risk analysis for critical infrastructure systems:
An approach based on statistical learning theory. Reliab Eng Syst Saf 94:855–860
Hansson K, Danielson M, Ekenberg L, Buurman J (2013) Multiple Criteria Decision Making for
Flood Risk Management. In: Amendola A, Ermolieva T, Linnerooth-Bayer J, Mechler R
(eds) Integr. Catastr. Risk Model. SE - 4. Springer Netherlands, pp 53–72
Hillier FS (1963) Economic Models for Industrial Waiting Line Problems. Manage Sci 10:119–
130
Huttenlau M, Stötter J (2011) The structural vulnerability in the framework of natural hazard risk
analyses and the exemplary application for storm loss modelling in Tyrol (Austria). Nat
Hazards 58:705–729
IEEE (2012) Std 1366-2012: IEEE Guide for Electric Power Distribution Reliability Indices.
IEEE, New York
Ighravwe DE, Oke SA (2014) A non-zero integer non-linear programming model for
maintenance workforce sizing. Int J Prod Econ 150:204–214
Jebaraj S, Iniyan S (2006) A review of energy models. Renew Sustain Energy Rev 10:281–311
Jha AK, Miner TW, Stanton-Geddes Z (2013) Building Urban Resilience: Principles, Tools, and
Practice. 1–180. © World Bank. http://elibrary.worldbank.org/doi/book/10.1596/978-0-8213-
8865-5
Ji M, He Y, Cheng TCE (2007) Single-machine scheduling with periodic maintenance to
minimize makespan. Comput Oper Res 34:1764–1770
Jiang R, Ji P (2002) Age replacement policy: a multi-attribute value model. Reliab Eng Syst Saf
76(3):311–318
Jin T, Tian Y, Zhang CW, Coit DW (2013) Multicriteria Planning for Distributed Wind
Generation Under Strategic Maintenance. Power Deliv IEEE Trans 28:357–367
Karimi I, Hüllermeier E (2007) Risk assessment system of natural hazards: A new approach
based on fuzzy probability. Fuzzy Sets Syst 158:987–999
Keller EA, DeVecchio DE (2012) Earth’s Processes as Hazards, Disasters, and Catastrophes.
Natural Hazards. Pearson Prentice Hall, New Jersey
Konak A, Kulturel-Konak S, Levitin G (2012) Multi-objective optimization of linear multi-state
multiple sliding window system. Reliab Eng Syst Saf 98(1):24–34
Krausmann E, Cozzani V, Salzano E, Renni E (2011) Industrial accidents triggered by natural
hazards: an emerging risk issue. Nat Hazards Earth Syst Sci 11:921–929
References 389
Krausmann E, Cruz A (2013) Impact of the 11 March 2011, Great East Japan earthquake and
tsunami on the chemical industry. Nat Hazards 67:811–828
Kumar Y, Das B, Sharma J (2006) Service restoration in distribution system using non-
dominated sorting genetic algorithm. Electr Power Syst Res 76:768–777
Kumar Y, Das B, Sharma J (2008) Multiobjective, multiconstraint service restoration of electric
power distribution system with priority customers. Power Deliv IEEE Trans 23:261–270
Li F, van Gelder PHAJM, Ranasinghe R, et al. (2014) Probabilistic modelling of extreme storms
along the Dutch coast. Coast Eng 86:1–13
Li YF, Sansavini G, Zio E (2013) Non-dominated sorting binary differential evolution for the
multi-objective optimization of cascading failures protection in complex networks. Reliab
Lin C, Madu CN, Chien TW, Kuei C-H (1994) Queueing Models for Optimizing System
Availability of a Flexible Manufacturing System. J Oper Res Soc 45(10):1141–1155
Linnekamp F, Koedam A, Baud ISA (2011) Household vulnerability to climate change:
Examining perceptions of households of flood risks in Georgetown and Paramaribo. Habitat
Int 35:447–456
Lins ID, Rêgo LC, Moura M das C, Droguett EL (2013) Selection of security system design via
games of imperfect information and multi-objective genetic algorithm. Reliab Eng Syst Saf
112:59–66
Mareschal B, De Smet Y, Nemery P (2008) Rank reversal in the PROMETHEE II method: Some
new results. Ind Eng Eng Manag 2008 IEEM 2008 IEEE Int Conf 959–963
Marseguerra M, Zio E, Podofillini L (2002) Condition-based maintenance optimization by
means of genetic algorithms and Monte Carlo simulation. Reliab Eng Syst Saf 77:151–165
Marseguerra M, Zio E, Podofillini L (2004) A multiobjective genetic algorithm approach to the
optimization of the technical specifications of a nuclear safety system. Reliab Eng Syst Saf
84:87–99
Misra KB, Sharma U (1991) An efficient approach for multiple criteria redundancy optimization
problems. Microelectron Reliab 31:303–321
Moghaddam KS (2013) Multi-objective preventive maintenance and replacement scheduling in a
manufacturing system using goal programming. Int J Prod Econ 146:704–716
Moradi E, Fatemi Ghomi SMT, Zandieh M (2011) Bi-objective optimization research on
integrated fixed time interval preventive maintenance and production for scheduling flexible
job-shop problem. Expert Syst Appl 38:7169–7178
Paul, BK (2011) Environmental hazards and disasters: contexts, perspectives and management.
Wiley-Blackwell, Chichester
Pelling M (2003) Vulnerability of cities. Earthscan Publications Ltd., London
Pine JC (2009) Natural hazards analysis: reducing the impact of disasters. CRC Press, Taylor &
Francis Group, Florida
Pinedo ML (2012) Scheduling: theory, algorithms, and systems. Fourth Edition. Springer,
New York.
Pinto RJ, Borges CLT, Maceira MEP (2013) An Efficient Parallel Algorithm for Large Scale
Hydrothermal System Operation Planning. Power Syst IEEE Trans 28:4888–4896
Priori Jr L, Alencar MH, de Almeida AT (2015) Adaptations to possible climate change impacts:
problem structuring based on VFT methodology. CDSID Working report
Ramirez-Rosado IJ, Bernal-Agustin JL (2001) Reliability and costs optimization for distribution
networks expansion using an evolutionary algorithm. Power Syst IEEE Trans 16:111–118
Rao SS, Dhingra AK (1992) Reliability and redundancy apportionment using crisp and fuzzy
multiobjective optimization approaches. Reliab Eng Syst Saf 37:253–261
Rathod V, Yadav OP, Rathore A, Jain R (2013) Optimizing reliability-based robust design model
using multi-objective genetic algorithm. Comput Ind Eng 66:301–310
Shafiee M, Finkelstein M (2015) An optimal age-based group maintenance policy for multi-unit
degrading systems. Reliab Eng Syst Saf 134:230–238
Solecki W, Leichenko R, O’Brien K (2011) Climate change adaptation strategies and disaster
risk reduction in cities: connections, contentions, and synergies. Curr Opin Environ Sustain
3:135–141
Soltani M, Corotis RB (1988) Failure cost design of structural systems. Struct Saf 5:239–252
Sortrakul N, Cassady CR (2007) Genetic algorithms for total weighted expected tardiness
integrated preventive maintenance planning and production scheduling for a single machine.
J Qual Maint Eng 13:49–61
Su L, Tsai H (2010) Flexible preventive maintenance planning for two parallel machines
problem to minimize makespan. J Qual Maint Eng 16:288–302
Thywissen K (2006) Components of Risk: A Comparative Glossary. Institute for Environment
and Human Security (UNU-EHS), Bonn
Torres-Echeverría AC, Martorell S, Thompson HA (2012) Multi-objective optimization of
design and testing of safety instrumented systems with MooN voting architectures using a
genetic algorithm. Reliab Eng Syst Saf 106:45–60
Trivedi A, Srinivasan D, Sharma D, Singh C (2013) Evolutionary Multi-Objective Day-Ahead
Thermal Generation Scheduling in Uncertain Environment. Power Syst IEEE Trans 28:1345–
1354
Vari A, Linnerooth-Bayer J, Ferencz Z (2003) Stakeholder Views on Flood Risk Management in
Hungary’s Upper Tisza Basin. Risk Anal 23:585–600
Wang Y, Pham H (2011) A Multi-Objective Optimization of Imperfect Preventive Maintenance
Policy for Dependent Competing Risk Systems with Hidden Failure. Reliab IEEE Trans
60:770–781
Wong KP, Fan B, Chang CS, Liew AC (1995) Multi-objective generation dispatch using
bi-criterion global optimisation. Power Syst IEEE Trans 10:1813–1819
Yokoyama R, Bae SH, Morita T, Sasaki H (1988) Multiobjective optimal generation dispatch
based on probability security criteria. Power Syst IEEE Trans 3:317–324
Zidan A, Shaaban MF, El-Saadany EF (2013) Long-term multi-objective distribution network
planning by DG allocation and feeders’ reconfiguration. Electr Power Syst Res 105:95–104
Zio E, Golea LR, Rocco S. CM (2012) Identifying groups of critical edges in a realistic electrical
network by multi-objective genetic algorithms. Reliab Eng Syst Saf 99:172–177
Zischg A, Schober S, Sereinig N, et al. (2013) Monitoring the temporal development of natural
hazard risks as a basis indicator for climate change adaptation. Nat Hazards 67:1045–1058
Index
A of failures, 316, 337–340, 358

Acceptable level, 106, 164, 323, 326 matrix, 10, 34, 36
Actors, 7, 19, 26, 30, 81, 170, 191, 324 Constructivism perspective, 14
Additive MAU function, 65, 68, 177, 183, Contract, 249
258, 260, 319, 330, 340 design, 251
Administrative time, 255, 258, 260 parameters, 251
Age-based replacement, 138, 217, 291 selection, 250–251, 257, 258, 263, 266,
Aggregation of DMs’, 80 268, 270
individual choices, 81 C-optimal portfolio, 75
initial preferences, 82 Corrective maintenance, 288
preferences, 79, 171, 191, 312, 338, 344 Cost, 64, 105, 109, 237, 243, 259, 266,
Aggregation of experts’ knowledge, 80, 287, 301, 373, 377
153–155 effective, 109
As low as reasonably practicable effectiveness, 321
(ALARP), 105, 163, 321 rate, 217–219
Asset management, 136, 199 Cost-benefit ratio, 109, 163–164, 289, 290
Assignment of priorities, 164, 335 Counter-terrorism, 203
Availability, 192, 221, 224, 238, 240, Critical devices, 115, 129, 343
254–255, 273, 280, 286, 292, 303, Critical infrastructure, 170, 186, 367, 369
315, 357 Criticality, 95, 126, 274, 276, 290, 335, 336
B D
Bathtub curve, 119–120, 277 Danger zone, 173–174, 178–179, 182, 329
Bayesian approach, 65, 151, 173, 275, 339, Debugging, 120
379 Decision
Blackout, 186, 200, 354 analysis, 2, 7, 81, 149, 175, 179, 192,
Block-based replacement, 291 200, 202, 205–206
Burn in, 120 matrix, 12, 16, 35, 38, 49
method, 57, 201
model, 1–21, 26, 53, 167–169, 172,
C 181–197, 258, 263, 266, 268,
Catastrophic losses, 174, 337, 365 275, 318, 338, 357
Choice of MCDM/A method, 14, 28, 33, perspectives, 14
38, 51, 63, 164, 166, 325, 366 theory, 7, 35, 168, 179, 181, 183, 187,
Choice of method, 325, 359 258, 260, 283
Climate changes, 199, 202, 204, 366 Decision maker’s
Compensatory, 16–18, 37, 63, 174, 181, preference, 3, 4, 9–10, 20–21, 31, 36,
190–191, 223, 258, 263, 266, 338 51, 59, 89–90, 92–93, 128, 163,
Condition-based maintenance, 139, 234, 166–168, 170–171, 174, 181,
290 187, 256, 258–265, 267–268,
Condition monitoring, 234, 239, 242, 270, 283
290–291 rationality, 16–18, 37, 51, 181, 190–191,
Consequences, 31, 117 263, 267, 325, 338, 366
analysis, 5, 8, 12, 33, 89, 112, 161–162, risk behavior, 7, 21, 39, 154, 161,
168–169, 173–182, 192, 259, 167–168, 172, 174, 181, 190,
267, 283 192–193, 264, 277, 283–284, 286

& Management Science 231, DOI 10.1007/978-3-319-17969-8
392 Index
Decision-making process, 2, 5–7, 14, 20, F

25, 47, 53, 90, 111, 163, 171, 181, Facility location, 355
190–191, 256, 270, 280 Failure mode, 95, 98, 126, 147, 171, 186,
Decision support system (DSS), 40, 169, 193, 198, 313
357, 361 Failure Mode Effects and Criticality
Delay time, 237, 240, 292 Analysis (FMECA), 95, 98,
Dependability, 256, 266, 269, 357 126–127, 148
Descriptive perspective, 14, 57, 70, 79 Failure Modes and Effects Analysis
Design, 115 (FMEA), 95, 126–127, 147
decisions, 5–6, 312, 321 Failure rate, 119–121, 129, 171–172, 274,
selection, 311, 315, 317, 321 277, 279, 281, 284, 313, 354, 376,
selection problem, 313, 315 379
Detectability, 127 Fatal Accident Rate (FAR), 94, 107, 323
Deterministic additive method, 15, 16, 58 Fatalities, 31, 109, 111, 162, 166, 175,
Diagnosis, 5, 234 183, 186, 194, 200, 367
Difference ratio, 56, 184, 331 Fault Tree Analysis (FTA), 96–98, 100,
Downtime, 20, 188, 221, 238, 243, 273–274, 110
276, 291 Financial loss, 19, 162–164, 166, 167, 173,
DSS. See Decision support system (DSS) 175, 179, 182, 185, 188–189, 192,
240, 254, 257, 274, 330
Flooding, 198, 206, 362, 366
E FMEA. See Failure Modes and Effects
Early failures, 121, 277 Analysis (FMEA)
Effectiveness of maintenance, 132, 291 FMECA. See Failure Mode Effects and
Effects of failures, 128, 146 Criticality Analysis (FMECA)
Electricity distribution, 161, 179, 181–182, FTA. See Fault Tree Analysis (FTA)
186–190, 199, 256, 357 Fuel transport, 181
Electric power, 170, 242, 256, 305, 357, 369 Functional failures, 146, 339
Electric power distribution, 256, 357
company, 242, 354
network, 357 G
Elimination Et Choix Traduisant la Réalité Gas pipeline, 114, 169, 172, 179, 181–185,
(ELECTRE), 15, 72, 190, 263, 190–197
265–266, 271 Genetic algorithm, 285–287, 289, 292, 384
ELECTRE TRI, 72, 190–193, 195–196, Greenhouse gas emissions, 371
202–203 Group Decision and Negotiation (GDN),
Engineering design specification, 314 79, 170
Environmental Group decision process, 79, 170
consequences, 105, 114, 162, 167,
172–173, 175, 179, 182–183,
191, 199, 201–202, 205–207 H
issues, 109, 168, 170, 199 Hand labor contract, 253
loss, 162, 191, 330 Hazard and Operability Study (HAZOP),
Event tree analysis (ETA), 98–100, 172, 95–96, 127
325 Hazard scenario, 102, 169, 171–174, 176,
Expert knowledge, 80–81, 153–155, 172, 178, 180–184, 186–187, 327
325 HAZOP. See Hazard and Operability
Experts’ prior knowledge, 149–155, 172, Study (HAZOP)
258, 260, 261, 264, 267, 276, Hidden failure, 147, 337
279–280, 283, 303, 375 Human loss, 106, 162–163, 166, 172, 175,
Exponential distribution, 120, 194, 258, 179, 182, 188, 192, 203, 205, 325
260, 263, 267, 376 Hydropower generation, 43, 370
Extreme events, 198, 362, 364–365 Hypothesis test, 43, 234, 380
Index 393
I Managerial indicators, 249, 361

Identifying critical devices, 343 Managerial indices, 250, 361
Image loss, 32, 35, 173, 191 MAUT. See Multi-attribute utility theory
Individual risk, 94, 108, 162, 164, 175, 207 (MAUT)
Information visualization, 111–113 MAVT. See Multiple attribute value theory
Inherent safety design, 321 (MAVT)
In-house maintenance, 250, 263 MCDA. See Multi-criteria decision aiding
Inspection, 145, 196, 199, 233, 290, 292, (MCDA)
357 MCDM. See Multi-criteria decision
Integrated production and maintenance making (MCDM)
scheduling, 371 MCDM/A, 89, 93, 165–166, 205, 336–339,
Intelligence stage of Simon’s model, 53 366
Interruption time, 32, 255, 264, 275, 305 methods, 2–4, 6, 8, 11–21, 57, 161,
Intra-criterion evaluation, 12–13, 21, 38, 164, 166, 181, 187, 190, 193,
62, 67, 82, 114, 192–193, 369 197, 250, 251, 256, 257, 270,
ISO/IEC Guide 51: 2014, 106, 322 275, 276, 281
ISO-risk, 163 methods, classification, 14–18, 37
Mean time, 221
between failure, 129, 255, 318
K to repair, 255, 258–259, 269–270, 279,
Key performance indicator, 249 280, 285, 318
Knowledge management, 113 Method choice, 29
K out of n system, 298 Monte Carlo simulation, 42, 128–129, 286
Multi-attribute utility theory (MAUT), 15,
65, 167–168, 174, 177, 181,
L 184–185, 187, 191, 204, 240, 244,
Location of backup units, 353 256, 258, 261, 263, 265, 282, 303,
Log-Normal distribution, 124 338–339
Lump sum, 253 Multi-component, 285, 291
Multicomponent system, 285
Multicriteria approach, 3–4, 57, 166, 168,
M 201
Maintainability, 125, 255–256, 264, 267, Multi-criteria decision aiding (MCDA), 2
279, 283, 306, 314, 315, 317, 375 Multi-criteria decision making (MCDM),
Maintainability role, 314 2
Maintenance Multidimensional
actions, 115, 125–126, 133, 288, consequences, 19, 93, 102, 174, 178,
291–292, 304, 332, 337–339, 355, 364, 366
353, 361 risk, 114, 161–208, 323
activities, 125, 250, 252, 282, 292, 357 Multi-linear model, 68, 356
contract, 249, 251, 253, 256–257 Multiobjective
function, 131, 132, 249, 331, 353, 372 approach, 3, 77
management, 115, 129, 133, 242, 254, optimization, 3, 77, 382
273, 338 Multiple attribute value theory (MAVT),
outsourcing, 249, 250, 257 15, 16, 58
plan, 335–336, 371 Multiple spare parts, 285–290
planning, 8, 19, 126, 335–336, 379
policy, 136, 290–291, 294, 371
service supplier, 249, 256–271 N
strategy, 336 Natech, 365
supplier selection, 249, 253, 257–271 Natural disasters, 200–202, 362, 364–367
team sizing, 375 Natural gas pipeline, 169, 172, 179,
Management risk, 316, 322 181–185, 190–197
394 Index
Natural hazards, 200–203, 364, 366, 367 Priorities assessment, 361

Non-compensatory, 16–18, 37, 70, 181, Priority Assignment, 335, 354
190–191, 263, 320, 325 Priority classes, 344
Non-repairable system, 130, 276 Prior knowledge, 149–157, 172–173, 258,
Normalization, 13, 38, 47 279–280, 303, 375
Normative perspective, 14, 55, 79 Prior probability, 149, 275, 279–280, 284,
NORSOK Standard Z-013, 163, 322 340–341, 368
NSGA-II, 286, 290, 372, 383 Procedure for resolving problems, 28,
Nuclear power, 98, 101, 178, 204–205, 369
316 Process for building models, 24
Procrastination, 45
Production scheduling, 271
O Prognostics, 234
Operational loss, 167, 188, 194, 200, PROMETHEE. See Preference Ranking
337–338, 340 Organization Method for
Operation planning, 369 Enrichment Evaluation
Outranking method, 15, 37, 70, 191, 264, (PROMETHEE)
265, 359
Outsourcing
contract, 250 Q
requirements, 251–257 Quality of repair, 266
Overall equipment effectiveness (OEE), 145 Quantitative risk analysis (QRA), 101
Queuing theory, 375
P
Parallel-series system, 298 R
Parallel system, 298 Rainfall, 366
Perception of risk, 92, 111, 161, 192, 208 RCM. See Reliability Centered
Periodic replacement, 217 Maintenance (RCM)
Perishable items, 278 Redesign, 144, 331, 332
Petrochemical, 142 Redundancy allocation, 297, 313
Planning replacement, 217 Redundant systems, 129–130, 303
P-median, 355 Reliability, 18, 115–117, 126–130, 243,
Portfolio of actions, 285, 290 255, 276, 283, 284, 300, 313, 336,
Portfolio problem, 9, 34, 63, 285, 326 380
Power Reliability acceptance test, 379
generation, 369 Reliability Centered Maintenance (RCM),
outages, 369 146, 336–341, 358, 360
system, 170, 186, 198, 200, 370 Repairable system, 130, 264, 277, 285
transformers, 354 Repair contract selection, 249, 251, 256
Predictive maintenance, 233 Replacement, 130, 216–217, 277, 286–288,
Preference 290–294
modeling, 10, 14, 24, 36, 340, 368 Residual useful life, 234, 290
structure, 3, 9–12, 20, 36, 58, 66, 68, Resilience, 363
170, 171, 181, 188, 198, 244, Response time, 257, 263–266
256, 258, 261, 264, 283, 338 Risk
Preference Ranking Organization Method acceptability, 92–94, 106, 323, 326
for Enrichment Evaluation analysis, 90–95, 101, 112, 114, 161,
(PROMETHEE), 15, 16, 73, 271, 164, 167, 169–170, 173, 178,
359 197, 203, 321
Prescriptive perspective, 14, 68 assessment, 55, 91, 94, 101, 111, 162,
Preventive maintenance, 8, 18, 126, 215, 164, 182, 199
238, 277, 286, 291, 292, 358, 371 categories, 106, 191–193, 195
Index 395
characterization, 93 Societal risk, 94, 110, 162, 164, 207

communication, 91, 111–112 Spare parts, 255, 269, 273–276, 285, 290,
difference, 114, 188, 331 313, 336
dimensions, 163, 179, 182, 185, 191, 323 Standby, 129, 298, 303
evaluation, 57, 106, 113, 163, 167–182, State of nature, 7, 29, 35, 65, 81, 102, 149,
186, 190–193, 321 180, 187, 195, 222, 258, 264, 267,
identification, 111 283, 284, 339, 368, 370, 379
indices, 94, 321 Stock out, 278–279, 286–287, 289–290,
of inventory shortages, 276, 278–281 294
management, 9, 90–91, 111–113, Strategic result, 1
162–166, 168, 171, 181, 184, Supplier selection 249, 251, 253, 257–271
186, 191, 198, 199, 202, 205, Supply reliability, 257
316, 322 System Average Interruption Duration
map, 113 Index (SAIDI), 243, 359
measures, 94, 179–181, 199 System Average Interruption Frequency
perception, 113, 208 Index (SAIFI), 243, 359
picture, 104, 323 System of preferences, 11, 31, 36, 51, 59
priorities, 126
prone, neutral or averse, 39, 167, 188,
190, 192–193, 323 T
visualization, 111–114 Target cost, 253
Role of maintenance, 131 Technological risk, 316
Root causes, 97 Telecommunication system, 256, 269, 305
3D visualization, 112, 114
Time to repair, 255, 260, 268, 269, 274,
S 279, 316, 318
Safety, 18, 20, 99, 101, 105, 109, 162, 164, Top event, 97–98, 100, 110
171, 186, 197, 204, 206, 254, 316, Total productive maintenance (TPM), 143,
323, 337 342
SAIDI. See System Average Interruption Total weighted expected tardiness, 372
Duration Index (SAIDI)
SAIFI. See System Average Interruption
Frequency Index (SAIFI) U
Scales, 13, 32, 39, 47, 64 Underground electricity distribution
Selective maintenance, 336 system, 179, 181, 186–190
Sequencing of maintenance activities, 357 Urban passenger bus transport company,
Series-parallel system, 299 286
Series system, 298
Service contracts, 251, 351
Service supplier, 249–251, 253, 256–271 V
Severity, 126, 162, 164, 180, 344 Vulnerability, 200–203
Simulation, 65, 128–129, 291–292, 325,
374
Sizing spare parts, 276–285 W
Social disruption, 364 Wear out failures, 119–120
Social risk, 110, 364 Weibull distribution, 121–123, 317

2015 - Almeida Et Al. - Multicriteria and Multiobjective Models For Risk, Reliability and Maintenance Decision Analysis

Uploaded by

Copyright:

Available Formats

2015 - Almeida Et Al. - Multicriteria and Multiobjective Models For Risk, Reliability and Maintenance Decision Analysis

Uploaded by

Document Information

Original Title

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

2015 - Almeida Et Al. - Multicriteria and Multiobjective Models For Risk, Reliability and Maintenance Decision Analysis

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics are covered in the book?

What topics are covered in the book?

International Series in

Operations Research & Management Science

Associate Series Editor

Founding Series Editor

More information about this series at http://www.springer.com/series/6161

Marcelo Hazin Alencar Rodrigo José Pires Ferreira

Adiel Teixeira de Almeida-Filho Thalles Vitelli Garcez

ISSN 0884-8289 ISSN 2214-7934 (electronic)

Springer Cham Heidelberg New York Dordrecht London

Printed on acid-free paper

Springer International Publishing AG Switzerland is part of Springer Science+Business Media

Any organization is interested in having a structured decision process for its

Nelson Maculan Filho

in MCDM/A. In addition to decision analysis, some of the topics related to

Condition-based maintenance (CBM) is tackled in Chap. 6, including a

Recife, Adiel Teixeira de Almeida

Foreword ............................................................................................................ vii

Chapter 1 Multiobjective and Multicriteria Problems

Chapter 2 Multiobjective and Multicriteria Decision Processes

2.3.5Step 5 - Identifying the State of Nature ................................... 35

2.4.5 Other MCDM/A Methods ........................................................ 76

Chapter 3 Basic Concepts on Risk Analysis, Reliability

3.3.3 What is Maintenance Management? ........................................ 133

Chapter 4 Multidimensional Risk Analysis................................................ 161

4.4.4 Nuclear Power .......................................................................... 204

Chapter 5 Preventive Maintenance Decisions............................................ 215

Chapter 6 Decision Making in Condition-Based Maintenance ................ 233

Chapter 7 Decision on Maintenance Outsourcing ..................................... 249

7.3 MCDM/A Maintenance Service Supplier Selection ............................ 257

Chapter 8 Spare Parts Planning Decisions................................................. 273

Chapter 9 Decision on Redundancy Allocation ......................................... 297

Chapter 10 Design Selection Decisions ....................................................... 311

10.3.2MCDM Framework for Risk Evaluation in Design

Chapter 11 Decisions on Priority Assignment for Maintenance

Chapter 12 Other Risk, Reliability and Maintenance Decision

Index .................................................................................................................. 391

AHP Analytic Hierarchy Process

HAZOP Hazard and Operability Study

PSA Probabilistic Safety Assessment

In order to choose an alternative, from a set of possible alternatives, in a classical

© Springer International Publishing Switzerland 2015 1

objective function to be dealt with simultaneously. In some particular situations,

Benjamin Franklin called this procedure prudential algebra. Much later on an

1.2 Multiobjective and Multicriteria Approaches

1.3 Decision Models and Methods

1.4 Decision Process

Searching for decision situations, by

Design Building the decision model

Evaluating the set of alternatives and

Revising the previous stages and

Fig. 1.1 Decision Process

1.5 Basic Elements and Concepts of Multiobjective and

This section briefly introduces basic ingredients and elements related to

1.5.1 Basic Ingredients and Related Concepts

x Ranking problematic - In this problematic the alternatives of the set A are