Proceedings of the 1999 Winter Simulation Conference

P. A. Farrington, H. B. Nembhard, D. T. Sturrock, and G. W. Evans, eds.

SIMULATION OF MANUFACTURING SYSTEMS

Averill M. Law
Michael G. McComas
Averill M. Law & Associates, Inc.
P.O. Box 40996
Tucson, AZ 85717, U.S.A.

ABSTRACT
This paper discusses how simulation is used to design new
manufacturing systems and to improve the performance of
existing ones. Topics to be discussed include:
manufacturing issues addressed by simulation, simulation
software for manufacturing applications, techniques for
building valid and credible models, and statistical
considerations. A comprehensive example will be given in
the conference presentation.
1

Performance evaluation

INTRODUCTION

One of the largest application areas for simulation

modeling is that of manufacturing systems, with the first
uses dating back to at least the early 1960s. In this paper
we present an overview of the use of simulation in the
design and analysis of manufacturing systems. Detailed
discussions of simulation, in general, may be found in
Banks, Carson, and Nelson (1996) and Law and Kelton
(1999). A practical discussion of the steps in a sound
simulation study is given in Law and McComas (1990).
This paper is a synopsis of a three-day short course with
the same title as this paper, which the first author has given
more than fifty times since 1985.
2

Production scheduling
Inventory policies
Control strategies [e.g., for an automated
guided vehicle system (AGVS)]
Reliability analysis (e.g., effect of preventive
maintenance)
Quality-control policies

The following are some of performance measures

commonly estimated by simulation:

The following are some of the specific issues that

simulation is used to address in manufacturing:
The need for and the quantity of equipment and
personnel

Throughput analysis
Time-in-system analysis
Bottleneck analysis

Evaluation of operational procedures

MANUFACTURING ISSUES
ADDRESSED BY SIMULATION

Location and size of inventory buffers

Evaluation of a change in product volume or
mix
Evaluation of the effect of a new piece of
equipment on an existing manufacturing
system
Evaluation of capital investments
Labor-requirements planning
Number of shifts

Number, type, and layout of machines for a

particular objective
Requirements for transporters, conveyors,
and other support equipment (e.g., pallets and
fixtures)

56

Throughput
Time in system for parts
Times parts spend in queues
Queue sizes
Timeliness of deliveries
Utilization of equipment or personnel

Law and McComas

3

infeasible or not cost-effective. Thus, it is important for a

simulation analyst to determine whether the simulation
model is an accurate representation of the system being
studied, i.e., whether the model is valid. It is also important
for the model to be credible; otherwise, the results may
never be used in the decision-making process, even if the
model is valid.
The following are some important ideas/techniques for
deciding the appropriate level of model detail (one of the
most difficult issues when modeling a complex system),
for validating a simulation model, and for developing a
model with high credibility:

SIMULATION SOFTWARE FOR

MANUFACTURING APPLICATIONS

Historically, simulation packages were classified to be of

two major types, namely, simulation languages and
applications-oriented simulators. Simulation languages
were general in nature and model development was done
by writing code. Simulation languages provided, in
general, a great deal of modeling flexibility, but were often
difficult to use. On the other hand, applications-oriented
simulators were oriented specifically toward a particular
class of applications and a model was developed by using
graphics, dialog boxes, and pull-down menus. Simulators
were sometimes easier to learn and use, but might not have
been flexible enough for some problems.
However, in recent years vendors of simulation
languages have attempted to make their software easier to
use by employing a graphical model-building approach. A
typical scenario might be to have a palette of modelbuilding icons located on one side of the computer screen.
The icons are selected from the palette with a mouse and
placed on the work area. The icons are then connected to
indicate the flow of entities through the system of interest.
Finally, one double-clicks on an icon to bring up a dialog
box where detail is added. On the other hand, vendors of
simulators have attempted to make their software more
flexible by allowing programming in certain model
locations using an internal pseudo-language. In at least
one simulator, it is now possible to modify existing
modeling constructs and to create new ones. Thus, the
distinction between simulation languages and simulators
has really become blurred.
Based on the above discussion, we will now say that
there are two types of simulation packages. A generalpurpose simulation package can be used for any
application, but might have special features for certain ones
(e.g., for manufacturing or process reengineering).
Examples of general-purpose simulation packages are
Arena, AweSim, Extend, GPSS/H, Micro Saint, MODSIM
III, SIMPLE++, SIMUL8, SLX, and Taylor Enterprise
Dynamics Developer. On the other hand, an applicationsoriented simulation package is designed to be used for a
certain class of applications such as manufacturing, health
care, or call centers. Examples of manufacturing-oriented
simulators are Arena Packaging Edition, AutoMod,
AutoSched, Extend + MFG, ProModel, QUEST, Taylor
Enterprise Dynamics Logistics Suite, and WITNESS.
A much more detailed discussion of the topics in this
section may be found in Chapter 3 of Law and Kelton (1999).
4

State definitively the issues to be addressed

and the performance measures for evaluating
a system design at the beginning of the study.
Collect information on the system layout and
operating procedures based on conversations
with subject-matter experts (SMEs).
Delineate all information and data summaries
in an assumptions document,
which
becomes the major documentation for the
model.
Interact with the manager (or decision-maker)
on a regular basis to make sure that the
correct problem is being solved and to
increase model credibility.
Perform a structured walk-through (before
any programming is performed) of the
conceptual simulation model as embodied in
the assumptions document before an audience
of SMEs, managers, etc.
Use sensitivity analyses [see Chapter 5 of
Law and Kelton (1999)] to determine
important model factors, which have to be
modeled carefully.
Simulate the existing manufacturing system
(if there is one) and compare model
performance measures (e.g., throughput and
average time in system) to the comparable
measures from the actual system.

STATISTICAL ISSUES IN SIMULATING

MANUFACTURING SYSTEMS

Since random samples from input probability distributions

drive a simulation model of a manufacturing system
through time, basic simulation output data (e.g., times in
system of parts) or an estimated performance measure
computed from them (e.g., average time in system from the
entire simulation run) are also random. Therefore, it is
important to model system randomness correctly and also
to design and analyze simulation experiments in a proper
manner. These topics are briefly discussed in this section.

DEVELOPING VALID AND

CREDIBLE SIMULATION MODELS

A simulation model is a surrogate for actually

experimenting with a manufacturing system, which is often

57

Simulation of Manufacturing Systems

amount of time, the warmup period, before the output data
are actually used to estimate the desired performance
measure. Use of the warmup-period data would bias the
estimated performance measure.
A comprehensive treatment of simulation output-data
analysis can be found in Chapter 9 of Law and Kelton
(1999).

5.1 Modeling System Randomness

The following are some sources of randomness in
simulated manufacturing systems:

Arrivals of orders, parts, or raw materials

Processing, assembly, or inspection times
Machine times to failure
Machine repair times
Loading/unloading times
Setup times

In the actual conference presentation, we will give a

detailed analysis of a manufacturing system. We will
address the following issues:

In general, each source of system randomness needs to

be modeled by an appropriate probability distribution, not
what is perceived to be the mean value. Note that sources
of randomness encountered in practice are rarely normally
distributed. A detailed discussion of simulation input
modeling is given in Chapter 6 of Law and Kelton (1999).

5.2 Design and Analysis of Simulation Experiments

Because of the random nature of simulation input, a

simulation run produces a statistical estimate of the (true)
performance measure not the measure itself. In order for an
estimate to be statistically precise (have a small variance)
and free of bias, the analyst must specify for each system
design of interest appropriate choices for the following:

SIMULATION ANALYSIS OF A
MANUFACTURING SYSTEM

Evaluating different machine and forklifttruck resource levels

Sizing of work-in-process buffers
Determining the impact of random machine
downtimes
Determining the effect of different logic for
the forklift trucks

REFERENCES
Banks, J., J. S. Carson, and B. L. Nelson. 1996. DiscreteEvent System Simulation. 2d ed. Upper Saddle River,
New Jersey: Prentice-Hall.
Law, A. M. 1998. How to Select Simulation Software.
Tucson, Arizona: Averill M. Law & Associates.
Law, A. M., and W. D. Kelton. December, 1999.
Simulation Modeling and Analysis. 3d ed. New York:
McGraw-Hill.
Law, A. M., and M. G. McComas. Secrets of Successful
Simulation Studies. Industrial Engineering 22: 47-48,
51-53, 72.

Length of each simulation run

Number of independent simulation runs
Length of the warmup period, if one is
appropriate

We recommend always making at least three to five

independent runs for each system design, and using the
average of the estimated performance measures from the
individual runs as the overall estimate of the performance
measure. (Independent runs means using different random
numbers for each run, starting each run in the same initial
state, and resetting the models statistical counters back to
zero at the beginning of each run.) This overall estimate
should be more statistically precise than the estimated
performance measure from one run. Note that independent
runs (as compared to one very long run) are required to
obtain legitimate and simple variance estimates and
confidence intervals.
For most simulation studies of manufacturing systems,
we are interested in the long-run (or steady-state) behavior
of the system, i.e., its behavior when operating in a
normal manner. On the other hand, simulations of these
kinds of systems generally begin with the system in an
empty and idle state. This results in the output data from
the beginning of the simulation run not being
representative of the desired normal behavior of the
system. Therefore, simulations are often run for a certain

AUTHOR BIOGRAPHIES
AVERILL M. LAW is President of Averill M. Law &
Associates, Inc. (Tucson, Arizona), a company specializing
in simulation consulting, training, and software. He has
been a simulation consultant to more than 100 organizations,
including General Motors, IBM, AT&T, General Electric,
Nabisco, Xerox, NASA, the Air Force, the Army, and the
Navy. He has presented more than 315 simulation short
courses in 17 countries, and delivered more than 100 talks
on simulation modeling at technical conferences.
He is the author (or coauthor) of three books and more
than 35 papers on simulation, manufacturing,
communications, operations research, and statistics,
including the textbook Simulation Modeling and Analysis
that is used by more than 60,000 people worldwide. His
series of papers on the simulation of manufacturing systems
won the 1988 Institute of Industrial Engineers' best

58

Law and McComas

publication award. He is the codeveloper of the ExpertFit
software package for selecting simulation input probability
distributions, and he has developed several simulation
videotapes. Dr. Law wrote a regular column on simulation
for Industrial Engineering magazine from 1990 through
1991.
He has been a tenured faculty member and has taught
simulation at the University of Wisconsin and the University
of Arizona. Dr. Law has a Ph.D. in industrial engineering
and operations research from the University of California at
Berkeley.
MICHAEL G. MCCOMAS is Vice President for
Consulting Services of Averill M. Law & Associates, Inc.
He has considerable simulation modeling experience in such
manufacturing industries as food processing, paper products,
microcomputers, aerospace materials, medical components,
electronic components, pet-care products, and basic metal
processing. He is also the coauthor of seven published
papers on simulation. His educational background includes
an M.S. in Systems and Industrial Engineering from the
University of Arizona.

59