Assembly Line Design

MANUFACTURING ENGINEERING
AND MATERIALS PROCESSING

A Series of Reference Books and Textbooks

SERIES EDITORS

Geoffrey Boothroyd George E. Dieter

Chairman, Department of Industrial Dean, Col/sge of Engineering
and Manufacturing Engineering UnitfBrsity of Maryland
Univsrsity of Rhode Island_ College Park, Maryland
Kingston, Rhode Island

1. Computers in Manufacturing, U. Rembold, M. Seth and J. S. Weinstein

2. Cold Rolling of Steel, William L. Roberts
3. Strengthening of Ceramics: Treatments, Tests and Design Applications,
Henry P. Kirchner
4. Metal Forming: The Application of Limit Analysis, Betzalel Avitzur
5. Improving Productivity by Classification, Coding, and Data Base
Standardization: The Key to Maximizing CAD/CAM and Group
Technology, William F. Hyde
6. Automatic Assembly, Geoffrey Boothroyd, Corrado Poli, and Laurence
E. Murch
7. Manufacturing Engineering Processes, Leo Alting
8. Modern Ceramic Engineering: Properties, Processing, and Use in Design,
David W. Richerson
9. Interface Technology for Computer-Controlled M~_nufacturi ng Processes,
Ulrich Rembold, Karl Armbruster, and Wolfgang U/zmann
10. Hot Rolling of Steel, William L. Roberts
11. Adhesives in Manufacturing, edited by Gerald L. Schneberger
12. Understanding the Manufacturing Process: Key to Successful CAD/ CAM
Implementation, Joseph Harrington, Jr.
13. Industrial Materials Science and Engineering, edited by Lawrence E. Murr
14. Lubricants and Lubrication in Metalwork ing Operations, Elliot S.
Nachtman and Serape Kalpakjian
15. Manufacturing Engineering: An Introduction to the Basic Functions,
John P. Tanner
16. Computer-Integrated Manufacturing Technology and Systems, Ulrich Rembold,
Christian Blume, and Ruediger Oil/mann
17. Connections in Electronic Assemblies, Anthony J. Bilotta
18. Automation for Press Feed Operations: Applications and Economics,
Edward Walker
19. Nontraditional Manufacturing Processes, Gary F. Benedict
20. Programmable Controllers for Factory Automation, David G. Johnson
21. Printed Circuit Assembly Manufacturing, Fred W. Kear
22 . Manufacturing High Technology Handbook, edited by Oonatas
Tijunelis and Keith E. McKee
23. Factory Information Systems : Design and Implementation for CIM
Management and Control, John Gaylord
24. Flat Processing of Steel, William L. Roberts
25. Soldering for Electronic Assemblies, Leo P. Lambert
26. Flexible Manufacturing Systems in Practice : Applications,
Design, and Simulation, Joseph Talavage and Roger G. Hannam
27 . Flexible Manufacturing Systems : Benefits for the low Inventory
Factory, John E. Lenz
28. Fundamentals of Machining and Machine Tools, Second Edition,
Geoffrey Boothroyd and Winston A. Knight
29. Computer-Automated Process Planning for World-Class Manufacturing,
James Nolen
30. Steel-Rolling Technology : Theory and Practice, Vladimir 8. Ginzburg
31. Computer Integrated Electronics Manufacturing and Testing, Jack Arabian
32 . In-Process Measurement and Control, Stephan 0. Murphy
33. Assembly line Design: Methodology and Applications, We-Min Chow
34. Robot Technology and Applications, edited by Ulrich Rembold
35. Mechanical Deburring and Surface Finishing Technology, Alfred F.
Scheider

OTHER VOLUMES IN PREPARATION

Assembly Line Design
Methodology and Applications

We-Min Chow
IBM Corporation
San Jose, California

0 CRC Press
Taylor & Francis Group
Boca Raton London New York

CRC Press is an imprint of the

Taylor & Francis Group, an informa business
CRC Press
Taylor & Fra ncis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

© 1990 by Taylor & Fra ncis Group, LLC

CRC Press is an imprint of Taylor & Francis Group, an Informa business

No claim to original U.S. Government wo rks

This book contains information obtained from authentic and hi ghly regarded sources. Reaso nable efforts have bee n made to
publish reliable data and information, but th e author and publisher ca nnot assume res ponsibility fo r the va lidity of all materials
or the co nsequ ences of th eir use. The authors an d publishers have attempted to trace the copyr ight holders of all material repro·
duced in this publication and apologize to copyright holders if permission to publish in this fo rm has not been obta ined. If any
copyright material has not bee n ac knowledged please write and let us kn ow so we may rec tify in any future reprint.

Except as permi tted under U.S. Copyri ght Law, no part of t his book may be reprinted, reproduced, transmi tted, or utilized in any
form by any electronic, mechanical, or other means, now known or hereafter inve nted, including photocopying, microfilmin g,
and recording, or in any informat io n storage or ret rieval system, wit hout written permission from the publishers.

For permission to photocopy or use material elec tronic ally from this wo rk, please access www.copy ri ght. co m (http://www.co py·
right.corn /) or co ntac t the Co pyright Clea rance Center, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400.
CCC is a not·fo r·pro fit organi zation th at provides licenses and regist ration for a va riet y of use rs. For organizations that have been
granted a photocopy license by the CCC, a separate sys tem of pay ment has been arranged.

Trademark Notice: Product or corporate names may be trademark s or registered trademarks, and are used only for identifica·
tion and explanation wit hout intent to infringe.
Visit the Taylor & Francis We b site at
http: //www.taylorandfrancis.com

and the CRC Press Web site at

http://www.crcpress.com
To 1-Fllng Yao and Chien Tseng
Preface

Design of a manufacturing assembly line is an old problem and yet it is

also a new problem. In the past several decades, an enormous amount
of knowledge in design methodology and related subjects has been devel-
oped. Innovations in product and in manufacturing technology often result
in new design problems. On the one hand, technological advances lead
to more sophisticated manufacturing systems. On the other hand, mar-
ket competition causes a short product lifetime, which forces line change
or reduces the service years of a manufacturing line. Line design, there-
fore, plays an increasingly important role in manufacturing. Given such a
complex situation, we are unlikely to reach an economical solution for line
design by invoking intuition, rule-of-thumb, or trial-and-error approaches.
Design optimization requires a good knowledge of methodology and a fa-
miliarity with the manufacturing environment. Design work should be
guided by a systematic approach, with a careful evaluation of the design
concept from both cost and line-performance viewpoints. To a practic-
ing engineer, problems of line design may come from several different
sources:

1. Evaluation techniques and design methods are often derived from

advanced mathematics, and are not always easily accessible.
2. Much of the existing literature is limited to specific assumptions
that may not be compatible with the designer's environment. In
particular, books and articles dealing with isolated manufacturing
components cannot be used directly for line design. The designer
must often integrate isolated pieces of information.
v
vi Preface

3. For a new product or process, information for the design task is rarely
available. A recent emphasis on manufacturability in industry de-
mands early involvement of manufacturing in design functions. This
means overlapping of product development, process definition, and
line design. Consequently, the problem of information availability
becomes even more acute.
4. Due to uncertainty in market demand or engineering changes, the
line structure may have to be altered after installation, by adding
or deleting workstations. This can be a very upsetting problem,
particularly if it causes line disruption.

This book attempts to treat line design and its related subjects in a
cohesive manner, with an emphasis on design applications. Mathematics
in the book is at about the level of first-year college calculus. Theoretical
results, when needed, are derived by heuristic approaches, rather than by
rigorous mathematical manipulations. Design procedures and evaluation
techniques are usually followed by examples from real-world cases. General
guidelines for setting up assumptions and determining line performance
parameters are discussed, based on empirical data from either literature
reports or line observations. Solutions to the problem of uncertainty are
also discussed.
Chapter 1, "Fundamental Concepts," gives a brief overview of assembly
systems and characterizes the nature of the design problem. Individual
components and their relationships are reviewed. The design and analysis
work can be greatly simplified by looking at one component at a time. On
the other hand, the procedure also keeps individual components subject
to an integrated design concept. Cost and performance measures, which
reappear in the later chapters, are introduced.
Chapter 2, "Empirical Data and Statistical Analysis," begins with a brief
review of elementary probability theory and statistics, a prerequisite for
data analysis and performance evaluation. Characteristics of line parame-
ters are discussed, based on empirical data. Subjects include arrival process,
operator behavior, machine reliability, product yield, and improvement (or
learning) processes.
Chapter 3, "Queuing Analysis and Optimization Techniques," provides a
mathematical background for design analysis and optimization. In addition
to an introduction to the theory, practical queuing models, computational
algorithms, and their applications are discussed.
Chapter 4, "Process and Operation Design," discusses functional design
for the assembly process. The first part of the chapter reviews techniques
of line balancing and the impacts of the length of assembly cycle time. To
simplify material-flow and material-handling effort, an assembly process is
Preface vii

often organized in a sequential manner. A dynamic programming proce-

dure for sequential lines is developed for the minimal cost solution. Cost
items considered in the procedure are direct labor, workstation tooling,
building space, factory overhead, and work-in-process. The last part of the
chapter deals with strategies for product testing and rework.
Chapter 5, "Material-Handling Systems," compares material-handling
systems by looking at their cost, performance, and integration with other
manufacturing components. Considered are three basic material-handling
systems: manual, conveyor (continuous type), and automated storage and
retrieval system (discrete type). Performance models, operating character-
istics, and selection criteria for material-handling systems are discussed.
Chapter 6, "Work-in-Process Management," first deals with parts logis-
tics problems, such as ordering, delivery, kitting, and line feeding. Two
different work-in-process management policies are considered: pull and
closed loop. Performance comparisons between the two policies and de-
termination of design parameters under each policy are discussed. The last
part of the chapter presents a procedure for buffer-capacity analysis and
design optimization.
Chapter 7, "Capacity Planning and Human Resource Management,"
reviews the concept of the improvement process and its empirical models.
Knowledge in this area will lead to useful results in line scale-up, capacity
planning, job training, and cost estimating. The second part of the chapter
discusses line capacity planning and adjustment. Finally, problems of
operator selection for both training and assignment are investigated. An
optimal assignment policy is obtained by using the matching algorithm,
based on operator behavior, work-in-process distribution, and workstation
availability.
Chapter 8, "Line Integration," discusses issues in line cost, sizing, and
configuration. Since many line-design decisions are based on cost analysis,
both cost estimation methods and cost items are considered. This discussion
will give the reader a basic understanding about cost structure of a manufac-
turing line. For a fixed demand, the line cost can be regarded as a function
of line size. A minimal cost solution can be found by using a dynamic pro-
gramming technique. For a variable demand pattern, a design procedure,
based on the concept of line flexibility, is developed. This design concept
will reduce the risk and the cost resulting from uncertainty in market de-
mand as well as engineering changes.
Chapter 9, "Line Simulation," introduces the Monte Carlo method and
its applications for manufacturing lines. After design work has been laid
out, performance evaluation should be the next step. At this stage, the
line is usually very complex. Analytic methods may not provide detailed
analysis, and therefore the designer has to rely on a simulation approach for
viii Preface

performance evaluation. Methodology and common pitfalls of simulation

models are discussed.
Finally, Chapter 10, ·~Analysis of an Assembly Line Design," presents
a case study, using the concepts of the previous chapters. Line assumptions
and the values of line parameters are obtained from real-life assembly lines.
Much of the material in this book was developed over the past few years,
when I participated in a manufacturing line design project. It is my wish
that this book may encourage more theoreticians to look into practical
manufacturing problems, and more practicing engineers to learn how to
convert theoretical results into powerful design tools.
I am grateful for the generous support that was given to this writing
project by my management in the IBM General Products Division, San Jose,
California. My thanks are due particularly to F. H. Fitch and Jack 0. Hilde-
brand for their encouragement. I am also indebted to my IBM associates
who helped me in many ways in the preparation of the book. Ronald Q.
Lewton and Edward A. MacNair read the manuscript and offered many sug-
gestions. Kenneth R. Sweet did all the graphic work. Finally, I thank my
wife for her understanding and patience.

We-Min Chow
Contents

Preface v

1 Fundamental Concepts 1

1.1 The Nature of the Line Design Problem 2

1.2 Line Components 5
1.3 Measures of Effectiveness 8
1.4 Remarks 15

2 Empirical Data and Statistical Analysis 17

2.1 Random Variable, Probability and Measures 18

2.2 Some Theoretical Distributions 20
2.3 Sample Distribution and Statistical Inference 29
2.4 Poisson Process 34
2.5 Operation Process Time 40
2.6 Yield Factor 47
2.7 Equipment Reliability 51
2.8 Remarks 56
References 57
Jx
X Contents

3 Queuing Analysis and Optimization Techniques 59

3.1 Basic Queuing Concepts 59

3.2 M/M/kQueue 63
3.3 M/G/kQueue 69
3.4 Network of Queues 81
3.5 Optimization Problems and Basic Concepts 87
3.6 Classic Optimization and Gradient Method 89
3.7 Computational Algorithms 95
3.8 Remarks 109
References 110

4 Process and Operation Design 111

4.1 Line Balance 111

4.2 Other Considerations for Line Balancing 121
4.3 Assembly Cycle Time 125
4.4 Design Optimization for Sequential Lines 128
4.5 A Case Study 136
4.6 Test/Inspection Strategy 140
4.7 Rework Strategies . 150
4.8 Remarks 153
References 155

5 Material-Handling Systems 157

5.1 The Role of Material Handling Systems 157

5.2 Automated Storage and Retrieval Systems 159
5.3 Conveyor System 178
5.4 Comparison of AS/RS and Conveyor 197
5.5 Case Study I 202
5.6 Case Study II 205
5.7 Remarks 212
References 213

6 Work-in-Process Management 215

6.1 Parts Ordering and Delivery 216

6.2 Parts Preparation and Feeding 228
Contents xi

6.3 In-Process Inventory Control 237

6.4 Buffer-Capacity Analysis 245
6.5 Remarks 261
References 262

7 Capacity Planning and Human Resource Management 265

7.1 Improvement Process 266

7.2 Line Capacity Planning 268
7.3 Training Program 291
7.4 Job Assignment 298
7.5 Remarks 317
References 318

8 Line Integration 319

8.1 Line Cost 319

8.2 Line Size 330
8.3 Line Configuration 337
8.4 Design for Line Flexibility 340
8.5 Case Study-A Flexible Line 351
8.6 Remarks 355
References 358

9 Line Simulation 361

9.1 Basic Concept of Monte Carlo Method 362

9.2 Random-Number Generation 364
9.3 Simulation Procedure 372
9.4 Run Length and Confidence Interval 379
9.5 Applications 390
9.6 Remarks 395
References 396

10 An Analysis of an Assembly Line Design 399

10.1 Problem and Approach 399

10.2 Process and Operation Design 400
XII Contents

10.3 Material Handling and WIP Management 402

10.4 Workstation Design and Facility Planning 404
10.5 Human Resource 405
10.6 Capacity and Resource Planning 406
10.7 Line Performance Analysis 409
References 416

Author Index 417

Subject Index 421

Assembly Line Design
1
Fundamental Concepts

Assembly work has a very long history. Ancient people already knew how to
create a useful object composed of multiple parts. However, the objective of
modern assembly processes is to produce high quality and low cost products.
A number of important ideas have been developed to facilitate assembly
work. First of all, assembly parts are standardized. Parts of the same type
must be subject to the same specifications. This also implies uniform parts
quality. Parts from different sources then can be put together as a finished
product.
Another innovation in the history of assembly manufacturing is the
division of the assembly job. If an assembly task has a long process time or
involves too many parts, the work may be broken into a number of smaller
tasks. Each task builds a part of the assembly. By progressively adding parts
to an assembly, a finished item is produced. Since each task has a relatively
limited content, skill can be developed in a short time. Thus, assembly
speed may be increased and quality improved. In many cases, nonassembly
operations may be needed. For instance, parts preparation, inspection, and
testing operations may be introduced to assure the product quality level,
and facilitate assembly work.
Since an assembly is composed of multiple parts and parts are assem-
bled by multiple operations, material flow becomes relatively complex.
On the other hand, moving physical material is a rather simple and me-
chanical task, as compared with an assembly task. Operator productivity
can be further improved by introducing a dedicated material-handling
system and letting operators concentrate their efforts on assembly tasks
where skills are needed. As a consequence, automated material-handling
1
2 Chapter 1

systems have become very popular in modern assembly lines. Manu-

facturing cost is reduced due to prompt material delivery, smaller space
requirement, better inventory accountability, less handling damage, and
less labor.
Engineers and scientists have engaged in multidisciplinary analyses,
learning that proper work conditions (e.g., job content, tooling and fixtures,
workstations, etc.) provide operators with safer and more productive jobs.
Time and motion study, analysis of human performance, and ergonomics
have been introduced to industry. At this stage, assembly job design begins
to integrate human behavior into workstation design. Consequently, effi-
ciency at the workstation level is greatly improved.
As more and more components are included, line efficiency eventually
becomes a problem. Efficiency improvement at the component level does
not guarantee overall performance efficiency. Line-integration concepts
therefore are introduced. The line designer must take a system view and
a structural approach. First, a cost objective must be defined based on
both market analysis and manufacturing conditions. Then the product
cost structure must be understood. Usually, this is defined by the product
characteristics and the manufacturing environment. By comparing the cost
objective and the cost structure, the line designer may conduct a study
for production feasibility and affordability. At this stage, questions such
as resource availability, production capacity, the speed of scale-up, and
engineering skills must be answered. Derived from this study is a line-
design concept that involves a number of interrelated subjects, e.g., tooling
strategy, material-handling system, line size, line configuration, flexibility
needed for future engineering changes or line-capacity adjustment, and
space strategy. The mission of line design is then to convert the design
concept into a physical line.

1.1 The Nature of the Une Design Problem

Although it is not difficult to develop a logical procedure for line design,
problems may still come from several different sources. First, all the
information needed for line design may not be available; the line designer
must deal with unknowns. Then, among all the available information, some
items are subject to change. Third, even if all design parameters have
been firmed up, stochastic phenomena in an assembly line are inevitable.
Due to lack of deterministic regularity, line problems can be very complex.
Finally, problem complexity may also come from the organization of the
design team. Poor communications between design groups responsible
for different functions will cause schedule delay, poor design, and high
cost.
Fundamental Concepts 3

None of these problems have simple solutions. However, it is not

impossible to find solutions, given effective design guidance. The rest of
this section will attempt to give such guidance.

1. Unknowns To shorten the product cycle and achieve optimum design

for manufacturability, the line design and the product design func-
tions must work in parallel and interact with each other. However,
this implies that information for line design, at least at an early stage,
is not totally available due to lack of a completed product definition.
In reality, since a physical line can only work in a given environment,
the line designer must often make assumptions for unknowns. To
minimize the risk of using wrong assumptions, the designer should
consider the generality of assumptions. That is, either the design so-
lution is not sensitive to the assumptions, or the assumptions have
a good coverage of the reality. In many cases, this can be done by
conducting mathematical analysis or line simulation against differ-
ent sets of line parameters.
2. Uncertainty Even if all the necessary information were available, it
is still subject to another problem-change or uncertainty. It is
too naive to believe that all the available information would remain
unchanged during the line-design process. In some cases, a change
is so drastic that the entire design concept must be revised.
Two major problem areas are production demand and engineering
change. The design concept of an assembly line is heavily influenced
by the desired production volume. For example, automation is a
common strategy only for high-volume production. Unfortunately,
production demand is usually based on the market situation which is
difficult to understand and well beyond the control of the manufac-
turer. On the other hand, engineering change is quite common when
dealing with a new product. Any changes in product design, manu-
facturing process, tooling, fixtures, or even work methods may result
in a need for line adjustment.
Although a line can be modified by adding, deleting or rearranging
workstations and altering material-handling methods, it can be a very
costly and time-consuming solution without a careful plan. It is the
line designer's responsibility to make sure that the line has flexibility
to cope with any changes easily.
Solution methods should be computerized, if possible. Whenever
a change is made, the computer package can be reinvoked with a new
set of parameters to find the new solution rapidly.
3. Variation Most line parameters that can be accurately estimated by
engineers are available in terms of their average values. Examples
4 Chapter 1

are the mean process time, the average yield, the mean time between
failures, the mean repair time, and so on. From a capacity-planning
point of view, however, variability of a parameter is as important as
its average value. This thought can be clarified by a simple example
as follows:

Consider a sequential line with two workstations. Both work-

stations have the same process time, say 2 minutes per opera-
tion cycle. If the process times are constant, the two stations
start and complete their operation cycles at the same time
epochs. The line throughput is clearly 0.5 per minute. Sup-
pose that the process times are subject to a simple distribution:
with an equal probability, the process time is either 1 or 3 min-
utes. The mean process time remains the same. If both stations
are in a 1-minute or a 3-minute cycle, they can still start and
complete their cycle at the same moments. But if their pro-
cess times happen to be different, then one station must wait
for the other. In this case, the production cycle becomes 3 min-
utes. Since the probability of a 1-minute cycle is 0.5 x 0.5 = 0.25,
the line throughput is equal to 1/(0.25 x 1 + 0.75 x 3) = 0.4 per
minute. If the line has more than 10 workstations, the produc-
tion cycle is almost always 3 minutes and the line throughput is
reduced to 0.333 per minute.

Compared to unknowns or uncertainty, the problem of variation

is relatively easy to deal with. The line designer may try to reduce
the variability, for instance, by introducing automation or improved
work methods. In many cases, an adaptive line operation strategy
can be invoked to adjust the line operation mode. One common
strategy is to regulate line flow by introducing buffers for work-
in-process between operations, so that operations become nearly
independent of each other. Sometimes, assigning a fast operator
to the operation with high variability may also help to increase line
productivity. The last resort is to plan for more resources such as
adding more workstations.
4. Complexity A line-design problem often has a complex structure due
to multiple line components, e.g., tooling, operators, parts, material-
handling facility, yield management, information system, and so on.
For each single component, a number of design alternatives may
exist . . The problem can easily become unmanageable if the line
designer has to consider all the possible combinations of these al-
ternatives. Therefore, the problem of complexity must be handled
Fundamental Concepts 5

Table 1.1 Problem Solving Strategy

Problem Solution
Unknown Generality
Uncertainty Flexibility
Variation Adaptability
Complexity Analysis, Communication

by a structural approach. For a given product and a given manufac-

turing environment, the design objective and constraints should be
clearly defined. Then the problem may be partitioned into a number
of subproblems. Since the scope of each subproblem is limited, a
complete analysis becomes possible. Results of analysis should help
to reduce the number of alternatives at the subproblem level and,
consequently, to simplify the overall design problem.
The line design problem can also be complicated due to the
requirement of multiple disciplines. For any sizable line, the de-
sign project usually involves people from different groups such as
material control, quality assurance, production control, facility en-
gineering, and information system. Consequently, communication
between people is often a major obstacle. Solutions to this problem
include: (i) Keep the design team small. (ii) Effectively use meeting
and other communication means. (iii) Document design assump-
tions, and (iv) to make the direction clear to all team members.

The above discussion is summarized in Table 1.1.

1.2 Line Components

To alleviate complexity of a design job, an assembly line may be subdivided
into a number of logical or physical components. The design team usually
includes members from several departments; each is responsible for one
or more components. The subdivision method and the definitions of com-
ponents are application-dependent. One possible breakdown is given as
follows:

1. Process design
2. Line balance
3. Test strategy
4. Yield management
6 Chapter 1

5. Material handling
6. Maintenance policy
7. Work-in-process management
8. Parts procurement
9. Parts feeding
10. Human resource
11. Line size
12. Line layout
13. Information system

The first thing that one should consider is how a complete product can
be assembled. Assembly process design is a very critical area. A good
process design should lead to a well-balanced line and appropriate job
content for each operation. Often assemblies must be tested or inspected
for quality assurance. Through testing or inspection, assembly problems
will be detected and analyzed. The results then can be used for problem
corrections. Hence, a good testing strategy should reduce the amount of
rework and scrap, and increase line productivity by improving product yield.
Material handling is another area that deserves intensive study. Al-
though this function does not add value to a product, it facilitates assembly
process flow. If an assembly line has two or more workstations, products
must be moved from one to another. Furthermore, assembly parts also need
to be delivered to workstations. Any material-handling delay or damage will
degrade line performance and may become costly.
Since both workstations and material-handling systems may be subject
to failure, a maintenance function becomes necessary. Compared to many
other functions, a maintenance operation is relatively simple ( e.g.fewer
parts and fewer persons involved), but may require a high skill level.
The material-handling function is closely related to work-in-process
(WIP) management. The latter is responsible for determination of in-
process inventory, from the parts-staging area to product shipment. A WIP
management policy also regulates line flow and parts-feeding speed.
In many assembly lines, labor is still indispensable. Problems of manual
operations include the learning process, ergonomic considerations, human
inconsistency, performance deviations, and job assignment. An effective
training program and a good job-assignment policy are essential to efficient
production.
Since market demand is based on forecasts, it is difficult to match the
line capacity with the demand. To avoid the problem of overinvestment or
insufficient capacity, design strategy should emphasize line flexibility, i.e.,
the ability to quickly adjust capacity at a minimal cost. Both line size and
layout are important subjects in this area.
Fundamental Concepts 7

The rapid development of modern data processing also has a significant

impact on line design. Sophisticated software programs can assist line
management, including data collection, process control, failure analysis,
parts tracing, and shop flow control. Information systems play a key role
for both yield management and flow control.
From an analytic point of view, it is convenient to subdivide a line into a
number of components, and to deal with individual components separately.
On the other hand, components must be designed in an integrated manner.
To achieve design integration, interrelated components must be considered
under the same design function. Five functions are identified:

1. Line flow and configuration

2. Line operation
3. Materiallogistics
4. Product yield and quality
5. Information management

The first function deals with the physical line, where the key prob-
lems are (i) compatibility of line layout with process flow, (ii) interface
between material handling-system and workstations, (iii) equipment utiliza-
tion, (iv) compatibility of material handling system with WIP management
policy, and (v) line flexibility.
For a given assembly line, the next problem is how the line should
be operated. A common problem is that line operation policy has not
been carefully analyzed during the design phase. Consequently, the actual
line throughput may be considerably lower than the targeted capacity.
Operation management problems, such as line scheduling, parts feeding,
WIP management, job dispatching, flow control, and maintenance policy,
should be regarded as a part of line design problem.
Since assembly involves multiple parts, material logistics can be a major
problem. Typical logistics problems are parts ordering, delivery, parts
staging, and feeding. The right kind of parts should be delivered in the
right quantity to the right place at the right time in the right manner. In
some cases, parts preparation serves an important function. For clean
room operations, assembly parts must be cleaned before assembly. For
automated stations, parts may have to be placed in a kit with the appropriate
orientation.
The fourth function consists of components that may affect product yield
and quality. First, a high-quality product must have high-quality parts.
Next, during the assembly process, testing or inspection functions may be
established to examine product quality. Test results should be used for
process control so that both product quality and yield can be improved.
 8 Chapter 1

Table 1.2 Functional Relations Between Line Components

Flow and Line Material Yield and Information
configuration operation logistics quality management

Process design X X
Line balance X
Test strategy X X X X
Yield management X X X
Material handling X X
Maintenance policy X X X
WIP management X X X X X
Parts procurement X X X
Parts feeding X X X
Human resource X X X
Line size X
Line layout X
Information system X X X X X

In addition, tool calibration and maintenance can be important for high-

yield production. Finally, all operators must be appropriately trained and
supervised to follow the process instructions.
Management of line information can be vital to the success of a mod-
ern manufacturing business. Manufacturing information systems have
been widely used for many purposes. Problems in this area involve data-
requirement definition, computer and network capability, information re-
trieval or data base system, and line management support.
The relations between line components and design functions are sum-
marized in Table 1.2. These relations define a "map" for line integration.
Each function links a number of interrelated components together, while
each component may be responsible for one or more function. This map
will help the line designer make sure that components are both logically
and physically compatible and that design assumptions at function levels
are consistent and realistic.

1.3 Measures of Effectiveness

An important problem sometimes overlooked is design optimization. For
example, in high-technology industries, a company can profit by its product
innovations. Even if its manufacturing efficiency is poor, the company may
still have a high profit margin because it is offering a unique product. How-
Fundamental Concepts 9

ever, once losing its competitive edge, the company can quickly encounter
financial problems due to high product cost. For this reason, line design
should be integrated with the business plan.
A sound business plan should consider three things: (i) market, (ii) prod-
uct, and (iii) production. As illustrated in Figure 1.1, among these three
items the key factors are product performance, cost, and manufacturabil-
ity, respectively. A product is designed not only for good performance but
also for manufacturing simplicity. If a product is properly designed, an ef-
fective production line will produce it at low cost and with high quality. The
relations among product performance, manufacturability, and cost should
be carefully evaluated.
To assure that a line is appropriately designed, line measurements should
be established to understand design effectiveness. Measurements can be
classified into two categories: cost and line performance. A cost item usu-
ally measures the total amount of spending needed to produce a given
amount of products. This may include product design, development, labo-
ratory testing, and all spending to support production activity such as direct
and indirect labor, equipment, facilities, and other overhead. The ratio
of the total spending to the total amount of products is then the product
unit cost. From an investment point of view, however, the time factor is
important. To a business person, a dollar today has more value than a dol-
lar tomorrow. Consequently, cost estimation may be based on a discount
method; that is, the present value of a future cost item is estimated by mul-
tiplying the cost by a predetermined discount factor.

PRODUCT PERFORMANCE
MARKET PRODUCT

BUSINESS
PLAN

PRODUCTION

Figure 1.1 Three components of a business plan and their relationships.

10 Chapter 1

Different cost estimation methods have been devised, few of them mea-
sure intangible costs such as line flexibility, job enrichment, product yield,
parts quality, line balance, process time variation, and so on. Although not
all these intangibles can be understood easily, their costs may be measured
by indirect methods. In many cases, a cost estimation method can be derived
from performance measures. For example, line flexibility would impact the
capital investment plan. Yield and quality are related to line capacity and
material handling cost. Process time variation may cause workstation uti-
lization or in-process inventory problems.
Performance measures are much more sophisticated than cost measures.
Often advanced mathematical tools are required. A manufacturing line can
be regarded as a service system of multiple types of resources. Line perfor-
mance results from resource contention due to production activities. The
line designer may use analysis tools to characterize the line performance
problems and optimization tools to resolve the problems. A good design
should reduce contention to a minimal level and yet achieve high resource
utilization. Hence, the fundamental problems are (i) to plan for an ap-
propriate amount of resources, and (ii) to schedule and control production
activities properly. These two subjects are the theme of this book and will
be discussed in detail in the subsequent chapters. Basic concepts ofline per-
formance and measurement are briefly introduced in the following pages.
Line performance is determined by many parameters, such as work-
station availability, operation process time, yield, process flow pattern,
operator skill level, line operation management policy, and line layout. This
can be explained by using the diagram shown in Figure 1.2.
First, let us consider a workstation. When a workstation breaks down,
the waiting time for repair action is dependent on interfailure time, repair
time, the size of the repair staff, the number of workstations served by the
repair crew, and the dispatching rule of repair jobs. The sum of waiting
time and repair time is then equal to workstation downtime. A workstation
operating cycle is composed of an interfailure time and a downtime. The
ratio of the expected interfailure time to the average operating cycle is the
reliability of the workstation.
In addition to workstation failure, scheduled downtime may exist such
as preventive maintenance, engineering changes, setup, and calibration. If
it is a manual workstation, further production time may be lost because of
operator's fatigue, personal delay, breaks and lunch time. The proportion
of time that can be used for production is called the availability of a
workstation.
For each operation, the process time is a function of job complexity,
workstation layout, tools and fixtures, and operator skill level. The average
process time may be estimated by direct measurement or by a synthetic
Fundamental Concepts 11

REPAIR TIMES

H
NUMBER OF WORKSTATIONS
NUMBER OF REPAIRMEN WAITING I
TIME I
I UNSCHEDULED
l DOWNTIME
I
DISPATCHING RULE
INTERFAILURE TIMES RELIABILITY I

H
PREVENTIVE MAINTENANCE
ENGINEERING CHANGES SCHEDULED
DOWNTIME
AVAILABILITY I
CALIBRATION I SETUP
PERSONAL FATIGUE I DELAY
LUNCH I BREAK
WORKSTATION LAYOUT _I OPE RAT I ON
JOB COMPLEXITY
I PROCESS
TIME
r----\ CAPACITY
TOOLS
SKILL LEVEL
PARTS QUALITY
I DISCIPLINE 'j
YIELD
ENVIRONMENT
~
ASSEMBLY PROCESS I YIELD ,,
I CAPACITY

SCHEDULING
JOB DISPATCHING
PARTS AVAILABILITY LINE
r-----~~ PERFORMANCE
LAYOUT I PROCESS
MATERIAL HANDLING

Figure 1.2 Line performance as a function of line parameters.

12 Chapter 1

method based on a time and motion study. The maximum number of

products that a workstation can build per unit of time (e.g., an hour or a
day) is its capacity. The hourly capacity, for instance, is simply the ratio of
availability to the average process time in hours.
Another important line parameter is the product yield, which is depen-
dent on production tools, operator skill, parts quality, working environ-
ment, test strategy, and line discipline. For a given manufacturing process,
operation process time and yield determine the yield capacity-the maxi-
mum number of good products per time unit that an operation can support.
Line measures may be derived from workstation capacity, yield, line
scheduling policy, job dispatching policy, parts availability, line layout, and
material-handling method. Several different measures, as follows, should
be used to complete line performance evaluation.

1. Line capacity
Line capacity is equal to the maximum expected number of good
products that the line can produce per time unit. This number is
usually close to but less than the yield capacity of the bottleneck
operation, i.e., the smallest yield capacity among all the operations.
2. Production lead time
This is the total amount of time that a product spends in the
line, including waiting and rework. Lead time is also an important
parameter for material-requirements planning.
3. Work-in-process
The total number of products in the line is a measure of in-process
inventory. During a sufficiently long time interval, the average
amount of work-in-process is approximately equal to the product of
the average lead time and line throughput. This relation is usually
a good indicator of line performance. An ideal condition is high
throughput, short lead time, and low work-in-process.
4. Workstation utilization
This is the proportion of time that a station is busy doing useful
work. For a well-balanced line, all workstations should have ap-
proximately the same utilization. If the line is running at full speed,
workstation utilization will be close to 1. The average workstation
utilization is defined by

1 n
W = -n~"
'"""T·S· (1.1)
i=l
Fundamental Concepts 13

where n =number of workstations

si ==average process time at workstation i
ri ==throughput at workstation i

5. Operator utilization
This is similar to workstation utilization, except that operators are
more flexible. Sometimes, even for an unbalanced case, workload
may be evenly distributed to operators. Operator utilization is a
measure for the efficiency of a job assignment policy. The operator
utilization is defined by

v=NS/mT (1.2)

where S =expected total manual process time (including rework, if

any)
N ==production volume per day
m =number of operators
T ==total productive time per day

6. Line utilization
This is the ratio of production volume to line capacity. Line
utilization is given by

u == ~ ['o r(t) dt (1.3)

t0 Jo c(t)

where c(t) =line capacity at t

r(t) =line throughput at t
t 0 = product life time

Since 0:::; r(t):::; c(t), then 0:::; u:::; 1. A line is fully utilized when
u=l.
7. Line flexibility
Line flexibility measures the difference between line capacity and
production demand:

f = ~ ['" min{c(t),D(t)} dt (1.4)

to Jo max{c(t),D(t)}
14 Chapter 1

where D(t) is the demand at t.

The value off is always between 0 and 1. If c(t) = D(t) for all t,
then/= 1. In this case, the maximum flexibility is achieved.
8. Quality
A conventional measure for quality is the defect rate, e.g., the av-
erage number of defective products per lot. From a line design point
of view, this measure provides little information. First, product qual-
ity is a cost factor and, therefore, should be measured against cost or
resource consumption rate. Next, quality should be appropriately
maintained during the product-building process. Quality control
must be a part of the manufacturing process. Strategies for testing,
inspection, failure analysis, and rework are important design consid-
erations. Finally, quality is closely related to product yield, and may
influence productivity.
Measures for quality include:

1. Scrap rate
2. ·Average total rework time per product
3. Average test time per product
4. Average yield over the product life time

Clearly, all these measures are related to line capacity and re-
source consumption requirements.
9. Line scale-up capability
This measures how fast a line capacity can be increased. Suppose
that line capacity reaches its peak at time y. The line scale-up
capability can be measured by looking at its scale-up index:

d =~ r c(t) dt
Y lo c(y)
(1.5)

When c(t) = c(y) for all tin (O,y), d takes its maximum value at
1. This means that the line can reach its highest capacity on the first
production day. Line capacity is mainly determined by process time
and yield. As long as the process times are finite and yield is pos-
itive, capacity can be increased by adding more resources such as
operators, workstations, etc. Therefore, this measure is meaningless
without considering the affordability issue. The line designer should
also examine the market demand. Even if affordability is not an is-
sue, sometimes the demand may limit the expansion of line capacity.
Fundamental Concepts 15

The scale-up index is not only a measure for the merit of line work,
but also for the entire business.

1.4 Remarks
Due to the existence of unknowns and uncertainty, line design often be-
comes a philosophical problem. On the other hand, line design is also a
very practical problem which can greatly influence the product cost. Since
a company's ultimate goal is to make profit, line design cannot be treated
as an isolated engineering problem. It must be considered as a part of the
entire business plan. Three things particularly important to line design are:

1. The business goal and strategy

2. Intensive cost and performance analysis
3. Line integration

The line designer must behave as a business person, an engineer, a

financial analyst, a system analyst, and a line manager.
2
Empirical Data and Statistical
Analysis

The analysis of line performance behavior is a key area in line design. Dif-
ferent design concepts and trade-offs between manufacturing components
should be carefully evaluated. Unfortunately, information needed for anal-
ysis is seldom totally available during the line design process. Even when
some information does become available, more often than not line design-
ers can only deal with average values such as mean time between failures,
mean repair time, average operation process time, average product yield,
and average throughput. Analysis in an "average" sense, however, is by no
means sufficient for line performance evaluation; more detailed informa-
tion is needed for accurate and thorough analysis. This dilemma can be one
of the major causes of using unverified or even improper assumptions for
line studies. For instance, many line simulation studies incorrectly assume
either uniform or normal assembly times. If the purpose of simulation is to
investigate a detailed buffer placement policy, neither of these assumptions
may honestly reflect reality.
If information is not available, line designers do not always need to
"invent" information for their studies. It would be more logical to base the
assumptions on experiences or observations from existing lines. Although
different manufacturing lines have different performance behaviors, some
commonalities can be identified and made useful to a very broad range of
applications. There are two major sources for this kind of information in
today's environment. One may search the literature to explore reported
results, or one may collect and analyze data from existing lines. In the
latter case, often the data do not directly provide useful information, either
because the data volume is too large or because the data are not presented
17
18 Chapter 2

in an understandable format. Raw data must be processed by data reduction

or analysis tools. Efficient tools for data analysis can be developed by using
statistical techniques. The first part of this chapter gives a brief introduction
to statistical concepts and techniques with which line designers should be
familiar. Discussions about commonality constitute the second part of the
chapter. Major parameters will be discussed, such as job or event arrival
process, operation process time, product yield and equipment reliability.

2.1 Random Variable, Probability and Measures

Since the real world environment is changing constantly, it is unlikely to
predict future events exactly. A convenient way to resolve this problem
is to postulate some regularity that follows a statistical law. For example,
process times of two operation cycles may not be identical, but the chances
are identical that both take values within the same range. Outcomes of
experiments (such as the process times), when measured in real numbers,
are called random variables. There are two types of random variables:
discrete and continuous. The former only takes values at isolated points on
a real line.
The probability of event A is the proportion of time that A can be
observed in repeated experiments. The value of a probability is between
(0, 1]. For example, A may stand for the event that a random variable X is
between a and b. The probability that A occurs is equal to P(A] =P[a S
X S b ]. If X is discrete, then an event can be {X =a} and its probability
is P[X =a]. When dealing with assembly line problems, random variables
usually have nonnegative values. IfX is nonnegative,P[X ~0] = 1.
The conditional probability of A, given B, is defined by

P(A IB] =P(AB]IP[B]

where AB is the event that both A and B occur. As an example, let T

be the time interval between machine failures, A = {T S 200 hours} and
B = {T > 100 hours}. Then P(A I B] is the conditional probability that
the machine failure would occur within 200 hours, given that the machine
has been operated for 100 hours (without failure). Clearly, P(A I C] = 1 if
C = {T < t} for all t S 200 hours, and P[A ID] = 0 if D = {T > t} for all
t ~ 200 hours.
The probability distribution function of X is defined by F(x) = P[X Sx].
If X is continuous,f(x) =dF(x)ldx is called the probability density function
of X. If X is discrete, it has a probability mass function, p(x) = P[X =
x]. Since the distribution function is also uniquely determined by the
Empirical Data and Statistical Analysis 19

density (or mass) function, statistical properties of a random variable can be

investigated by using either its density (or mass) function or its distribution
function. Bothf(x) andp(x) are nonnegative and F(x) is a nondecreasing
function of x and is valued between 0 and 1.
It is not always convenient to deal with these probability functions. Some
of them may have rather complex algebraic expressions. Properties of
randomness sometimes can be characterized by expectations or moments.
The rth moment of X is given by

E[Xr] = 1xrf(x)dx ifX is continuous

(2.1)
= l::)rp(x) ifX is discrete
X

Note that moments are averaged values with a weight functionf(x) or

p(x). When r = 1, m = E[X] is the mean or the expectation of X. The
formal definition of the expectation of X is given by (2.1) with r = 1. It
can be shown, however, that the expectation ofg(X) for a functiong can be
obtained by using the distribution of X (instead of the distribution ofg(X))
and is equal to

E[g(X)] = 1 g(x)f(x)dx if X is continuous

= L)(x)p(x) if X is discrete
X

The r th central moment of X for r > 1 is given byE[(X - m t]:

mr = 1(x-mYf(x)dx ifX is continuous

(2.2)
= l::)x -mYp(x) if X is discrete
X

The second central moment is called the variance, Var[X]. The square
root of the variance, denoted by SD[X], is the standard deviation. Both
the variance and the standard deviation are measures for dispersion. The
third and fourth central moments are measures for skewness and degree
of peakness, respectively. A distribution is symmetric if and only if its
20 Chapter 2

third central moment is zero. A distribution is positive (negative) skew, if

this moment is positive (negative). Normally positive (negative) skewness
implies that the distribution has a right-hand (left-hand) tail. Commonly
used relative measures are given by moment ratios: .

Coefficient of variation: CV[XJ = SD[XJ! E[XJ

Skewness index: SK[XJ =m 3/(SD[X]) 3 (2.3)
Kurtosis index: KR[X] = m 4!(SD[XJ) 4

Statistical relation between random variables can be investigated by

looking at their joint distribution. The joint distribution of X and Y is
defined by

F(x,y) =P[X ::;x,Y ::=;y]

Their correlation is measured by their covariance and co"elation coeffi-

cient, respectively:

Cov[X, Y] = E[XY]- E[X]E[Y]

Cov[X,Y] (2.4)
p(X,Y) = SD[X]SD[Y]

X andY are said to be unco"elated if and only if Cov[X, Y] = p(X, Y) = 0.

If X andY are independent, thenF(x,y) =P[X ::=;x]P[Y ::=;y] = F(x)F(y). It
is easy to show that X andY are independent if and only if E[g(X)h(Y)] =
E[g(X)]E[h(Y)] for almost all functions, g and h. Since independence
implies thatE[XY] =E[X]E[Y], then Cov[X,Y] = 0 and p(X,Y) = 0. But
two uncorrelated random variables are not necessarily independent of each
other.

2.2 Some Theoretical Distributions

Some frequently used distributions and their statistical properties are dis-
cussed in this section. Considered are three continuous cases: (i) uniform,
(ii) exponential and (iii) normal; and three discrete cases: (i) geometric,
(ii) binomial and (iii) Poisson.
Empirical Data and Statistical Analysis 21

Uniform (or Rectangular) Distribution

A random variable is uniformly distributed over [a,b], if its density is
identical for all x between a and b.

1
f(x)=-, fora ~x ~b
b-a
(2.5)
x-a
F(x)=- . ' fora ~x ~b
b -a

Plots of the density and the distribution functions lJ.re given in Figure 2.1.
Some moments and moment ratios are given by

E[X]=a+b
2
b-a)' 1
m,= ( -2- (r+l)

CV[X] = b -a__!__
b+aJ3
SK[X] =0
KR[X] = 1.8

This distribution can be used to characterize randomness of event oc-

currence times. For example, the arrival epochs of a random stream are
uniformly distriquted over a given time interval.

Exponential Distribution
A random variable has an exponential distribution with a rate >., if

f(x) = >.e--U for .t > 0

(2.6)
F(x)=1-e--U forx>O

Plots of the density and the distribution functions are given in Figure 2.2.
Mean and moment ratios are

E[X] = 1/>.
22 Chapter 2

f ( X)

(a) 0 0 b X

----- - ------,-----

F(x)

(b) 0 0 b X

Figure 2.1 (a) Uniform density and (b) distribution functions.

CV[X] =1
SK[X] =2
KR[XJ=9

Note that the value of CV is one. The density function has a right-hand
tail, i.e., the distribution is positively skewed. The degree of peakness is
much higher than that of uniform distribution.
Empirical Data and Statistical Analysis 23

(a) 0 X

------- -- -- ------- - -

F( x)

(b) 0 X

Figure 2.2 (a) Exponential density and (b) distribution functions.

An important feature of exponential distribution is its memoryless prop-

erty.Given {X > s }, the conditional probability that X > t + s is identical
to the unconditional probability of {X > t }. To see this, first note that
P[X >t] = 1-F(t) = exp(-.\t), and {X >t +s} implies {X >s}. Then

P[X >t +sIX >s] =P[X >t +s]/P[X >s]

=e-)l(t+s)/e->.s
=e->.J
24 Chapter 2

= 1-F(t)
=P[X >t]

Suppose that the interfailure time of a machine is exponentially dis-

tributed and is denoted by X. If the machine has been operated for y time
units, the elapsed time until the next failure is independent ofy and is statis-
tically identical to a new interfailure time. This property greatly simplifies
our analysis work, because it asserts that what may happen in the future is
independent of the past history. The distribution of remaining life of a ma-
chine is identical to an ordinary exponential lifetime. It is also possible to
prove that the exponential random variable is the only continuous random
variable that has the memoryless property.
If X 1, X 2 , ••• , Xn are independent, identical exponential random vari-
ables, then Y = L:Xi has an Erlang distribution with a density function

fory 2:0 (2.7)

where r(n) = (n -1)! is the gamma function.

If the interarrival time is exponential, the nth arrival time is the sum
of n exponential random variables and has an Erlang distribution. For
noninteger n, (2.7) becomes a gamma distribution. In this case, r(n) has
a value that makes Jf (y )dy = 1. For both gamma and Erlang distributions,
E[Y] =n!>.. and Var[Y] =n/>..2 •
Let Z = 2>..Y. The density function of Z is given by

1
f(z) = --e-z12(z/2)"-1 for z 2:0 (2.8)
2r(n)

In case that 2n is a positive integer, this form of the gamma distribution is

called the chi-square distribution with 2n degrees offreedom. The chi-square
distribution has been widely used in statistical tests of hypotheses.

Geometric Distribution
A counterpart of the exponential random variable in the discrete case is the
geometric random variable. Its mass function is defined by

p(x)=q(1-qf O<q < l,x = 0,1,2, ... (2.9)

Empirical Data and Statistical Analysis 25

where q can be interpreted as a probability.

Suppose that the outcome of an experiment is either a success or fail-
ure. Let q be the probability of success. (The failure probability is equal to
1-q.) Then the number of failures before the first success is geometrically
distributed. Given k failures have been observed, the conditional probabil-
ity ofx additional failures before the first success is independent of k and is
identical to the unconditional probability of x failures before the first suc-
cess. The geometric distribution is the only discrete distribution that has
the memoryless property. For a geometric random variable,

E[X] = (1-q)/q
Var[X] = E[X]Iq
SK[X]=(2-q)/~
.KR[X] = 9 + (1/Var[X])

The sum of n independent, identical geometric random variables is

known as the negative binomial random variable. Let Y be such a variable.
Its mass function has two parameters, q and n, and is defined by

p(y) = n +y
(
n
-1) qn(l-q'j' O<q < 1,y = 0,1,2, .... (2.10)

Thus Y is the number of failures observed before the nth success. Since
theX's are independent and identical,

E[Y] = LE[Xd =n(1-q)/q

Var[Y] = L Var[Xd = n(1-q)/q 2

Binomial Distribution
A binomial distribution has two parameters: nand q. The mass function is
given by

O<q < 1,x = 0,1,2, ... ,n (2.11)

26 Chapter 2

If q is the probability of success, then X is the number of successes

observed in n trials. Let Z be the outcome of an experiment. Z = 1 (or
0) if the outcome is a success (or failure). Z is usually referred to as the
Bemoulli random variable with P[Z = 1] = 1-P[Z = 0] = q. A binomial
random variable is the sum of n independent, identical Z's.
Moments and moment ratios are

E[X] =nq
Mzr[X] =nq(1 - q)
SK[X] = (1-2q)/SD[X]
KR[X] = 3 + [1- 6q(l-q)]/Var[X]

Clearly if q = 0.5, the distribution is symmetric about its mean. .

Poisson Distribution
A Poisson distribution is a limiting case of binomial with a fixed nq, but
n --too and q --t 0. Let X be a binomial random variable. The mass function
of X can be rewritten as

p(x) = n(n-1) .. ·(n -x + 1)tf( 1 -q)n-x

x!

Introduce a factor nx, let (J =nq, and rearrange the terms,

p(x)=1 ( 1-n1) ···( 1----;z

X - 1) (J) n (1-q) -x x!1
(Of ( 1-n

Letting n --too and q --t 0, and using the fact that (1 +x/n )n --teXas n --too,
we have a Poisson distribution with a mass function:

p(x) = e- 9 fJX fJ > O,x = 0, 1, 2, .... (2.12)

x!

The mean and other measures are

E[X] = fJ