DESIGN

RELIABILITY
Fundamentals and Applications

B. S. Dhillon

Department of Mechanical Engineering

University of Ottawa
Ottawa, Ontario, Canada

CRC Press
Boca Raton London New York Washington, D.C.

Acquiring Editor:
Project Editor:
Cover design:

Cindy Carelli
Susan Fox
Dawn Boyd

Library of Congress Cataloging-in-Publication Data

Dhillon, B. S.
Design reliability : fundamentals and applications / B.S. Dhillon.
p. cm.
Includes bibliographical references (p.
ISBN 0-8493-1465-8 (alk. paper)
1. Engineering design. 2. Reliability (Engineering). I. Title.
TA174.D4929 1999
620.0042--dc21

99-28211
CIP

This book contains information obtained from authentic and highly regarded sources. Reprinted
material is quoted with permission, and sources are indicated. A wide variety of references are listed.
Reasonable efforts have been made to publish reliable data and information, but the author and the
publisher cannot assume responsibility for the validity of all materials or for the consequences of their use.
Neither this book nor any part may be reproduced or transmitted in any form or by any means,
electronic or mechanical, including photocopying, microfilming, and recording, or by any information
storage or retrieval system, without prior permission in writing from the publisher.
The consent of CRC Press LLC does not extend to copying for general distribution, for promotion,
for creating new works, or for resale. Specific permission must be obtained in writing from CRC Press
LLC for such copying.
Direct all inquiries to CRC Press LLC, 2000 Corporate Blvd., N.W., Boca Raton, Florida 33431.
Trademark Notice: Product or corporate names may be trademarks or registered trademarks, and
are only used for identification and explanation, without intent to infringe.
1999 by CRC Press LLC
No claim to original U.S. Government works
International Standard Book Number 0-8493-1465-8
Library of Congress Card Number 99-28211
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
Printed on acid-free paper

Dedication
This book is affectionately dedicate to my mother
Udham Kaur.

1999 by CRC Press LLC

Preface
Today at the dawn of the twenty-first century, we are designing and building new
complex and sophisticated engineering systems for use not only on Earth but to
explore other heavenly bodies. Time is not that far off when humankind will permanently reside on other planets and explore stars outside our solar system. Needless
to say, the reliability of systems being used for space explorations and for use on
Earth is becoming increasingly important because of factors such as cost, competition, public demand, and usage of untried new technology. The only effective way
to ensure reliability of engineering systems is to consider the reliability factor
seriously during their design.
Over the years as engineering systems have become more complex and sophisticated, the knowledge in reliability engineering has also increased tremendously
and it has specialized into areas such as human reliability, software reliability,
mechanical reliability, robot reliability, medical device reliability, and reliability and
maintainability management.
Even though there are a large number of texts already directly or indirectly
related to reliability engineering, there is still a need for an up-to-date book emphasizing design reliability with respect to specialized, application, and related areas.
Thus, the main objective of writing this book is to include new findings as well as
tailor the text in such a manner so that it effectively satisfies the needs of modern
design reliability. Therefore, this book is written to meet this challenge and its
emphasis is on the structure of concepts rather than on mathematical rigor and minute
details. However, the topics are treated in such a manner that the reader needs no
previous knowledge to understand them. Also, the source of most of the material
presented in the book is given in references if the reader wishes to delve deeper into
particular topics. The book contains a large number of examples along with their
solutions, and at the end of each chapter there are numerous problems to test reader
comprehension.
The book is composed of 17 chapters. Chapter 1 presents the historical aspect
of reliability engineering, the need of reliability in engineering design, reliability in
the product design process, and important terms and definitions, and information
sources. Chapter 2 reviews mathematics essential to understanding subsequent chapters. Fundamental aspects of engineering design and reliability management are
presented in Chapter 3. As the failure data collection and analysis are considered
the backbone of reliability engineering, Chapter 4 presents many associated aspects.
Chapter 5 presents many basic reliability evaluation and allocation methods. However, the emphasis of the chapter is on the basic reliability evaluation methods.
Chapters 6 and 7 describe in detail two most widely used methods (i.e., failure
modes and effect analysis and fault tree analysis, respectively) to evaluate engineering design with respect to reliability. Chapter 8 presents two important topics of

1999 by CRC Press LLC

reliability engineering: common-cause failures and three state devices. Two specialized areas of reliability, mechanical reliability and human reliability, are discussed
in Chapters 9 and 10, respectively.
Chapter 11 presents the topics of reliability testing and growth essential in the
design phase of an engineering system. Chapters 12 through 14 present three application areas of reliability, i.e., reliability in computer systems, robot reliability, and
medical device reliability, respectively. In particular, the emphasis of Chapter 12 is
on computer software reliability. Chapter 15 presents important aspects of design
maintainability and reliability centered maintenance. Chapters 16 and 17 describe
three topics directly or indirectly related to design reliability: total quality management, risk assessment, and life cycle costing.
The book is intended primarily for senior level undergraduate and graduate
students, professional engineers, college and university level teachers, short reliability
course instructors and students, researchers, and engineering system design managers.
The author is deeply indebted to many individuals including colleagues, students,
friends, and reliability and maintainability professionals for their invisible inputs
and encouragement at the moment of need. I thank my children Jasmine and Mark
for their patience and intermittent disturbances leading to desirable coffee and other
breaks. And last, but not least, I thank my boss, friend, and wife, Rosy, for typing
various portions of this book and other related materials, and for her timely help in
proofreading.
B.S. Dhillon
Ottawa, Ontario

1999 by CRC Press LLC

The Author
B.S. Dhillon, Ph.D., is a professor of Mechanical Engineering at the University of
Ottawa. He has served as a Chairman/Director of Mechanical Engineering Department/Engineering Management Program for 11 years at the same institution. He has
published over 260 articles on Reliability and Maintainability Engineering and on
related areas. In addition, he has written 20 books on various aspects of system
reliability, safety, human factors, design, and engineering management published by
Wiley (1981), Van Nostrand (1982), Butterworth (1983), Marcel Dekker (1984), and
Pergamon (1986) among others. His books on Reliability have been translated into
several languages including Russian, Chinese, and German. He has served as General
Chairman of two international conferences on reliability and quality control held in
Los Angeles and Paris in 1987. Also, he is or has been on the editorial board of five
international journals.
Dr. Dhillon is a recipient of the American Society of Quality Control Austin J.
Bonis Reliability Award, the Society of Reliability Engineers Merit Award, the Gold
Medal of Honor (American Biographical Institute), and Faculty of Engineering
Glinski Award for excellence in Reliability Engineering Research. He is a registered
Professional Engineer in Ontario and is listed in the American Men and Women of
Science, Men of Achievements, International Dictionary of Biography, Whos Who
in International Intellectuals, and Whos Who in Technology.
At the University of Ottawa, he has been teaching reliability and maintainability
for over 18 years. Dr. Dhillon attended the University of Wales where he received
a B.S. in electrical and electronic engineering and an M.S. in mechanical engineering.
He received a Ph.D. in industrial engineering from the University of Windsor.

1999 by CRC Press LLC

Table of Contents
CHAPTER l: Introduction
l.1
Reliability History
1.2
Need of Reliability in Product Design
1.3
Reliability in the Product Design Process
1.4
Reliability Specialized and Application Areas
1.5
Terms and Definitions
1.6
Reliability Information Sources
1.7
Military and Other Reliability Documents
1.8
Problems
1.9
References

CHAPTER 2: Design Reliability Mathematics

2.1
Introduction
2.2
Boolean Algebra Laws
2.3
Probability Properties
2.4
Useful Definitions
2.4.1
Definition of Probability
2.4.2
Cumulative Distribution Function
2.4.3
Probability Density Function
2.4.4
Reliability Function
2.4.5
Hazard Rate Function
2.4.6
Laplace Transform Definition
2.4.7
Laplace Transform: Final-Value Theorem
2.4.8
Expected Value
2.5
Probability Distributions
2.5.1
Binomial Distribution
2.5.2
Poisson Distribution
2.5.3
Exponential Distribution
2.5.4
Rayleigh Distribution
2.5.5
Weibull Distribution
2.5.6
General Distribution
2.5.7
Normal Distribution
2.6
Hazard Rate Models
2.6.1
Exponential Distribution
2.6.2
Rayleigh Distribution
2.6.3
Weibull Distribution
2.6.4
General Distribution
2.6.5
Normal Distribution

1999 by CRC Press LLC

2.7

2.8
2.9
2.10

Partial Fraction Method and Equation Roots

2.7.1
Partial Fraction Method
2.7.2
Equation Roots
Differential Equations
Problems
References

CHAPTER 3: Engineering Design and Reliability Management

3.1
Introduction
3.2
Engineering Design
3.2.1
Design Failures and Their Common Reasons
3.2.2
The Design Process and Functions
3.2.3
The Design Team and Member Responsibilities
3.2.4
Design Reviews
3.2.5
Design Engineer and Design Review Board Chairman
3.2.6
Designing for Reliability and Design Reliability Check List
Items
3.2.7
Design Liability
3.3
Reliability Management
3.3.1
General Management Reliability Program Related
Responsibilities and Guiding Force Associated Facts for an
Effective Reliability Program
3.3.2
Military Specification Guidelines for Developing Reliability
Programs and a Procedure to Establish Reliability Goals
3.3.3
Reliability and Maintainability Management Tasks in the
Product Life Cycle
3.3.4
Documents and Tools for Reliability Management
3.3.5
Reliability Auditing and Pitfalls in Reliability Program
Management
3.3.6
Reliability and Maintainability Engineering Departments
and Their Responsibilities
3.3.7
Reliability Manpower
3.4
Problems
3.5
References
CHAPTER 4: Failure Data Collection and Analysis
4.1
Introduction
4.2
Failure Data Uses
4.3
Failure Data Collection Soruces in Equipment Life Cycle and Quality
Control Data
4.4
Failure Reporting and Documentation System Design Guidelines and
Failure Data Collection Forms
4.5
External Failure Data Sources
4.6
Failure Data For Selective Items and Tasks
4.7
Hazard Plotting Method
4.7.1
Hazard Plotting Mathematical Theory

1999 by CRC Press LLC

4.8

4.9

4.10
4.11

Underlying Distribution Determination Tests

4.8.1
Bartlett Test
4.8.2
General Exponential Test
4.8.3
Kolmogorov-Smirnov Test
Maximum Likelihood Estimation Method
4.9.1
Exponential Distribution
4.9.2
Normal Distribution
4.9.3
Weibull Distribution
Problems
References

CHAPTER 5:
5.1
5.2
5.3

5.4

5.5

5.6

5.7
5.8

Basic Reliability Evaluation and Allocation

Techniques
Introduction
Bathtub Hazard Rate Curve
General Reliability Analysis Related Formulas
5.3.1
Failure Density Function
5.3.2
Hazard Rate Function
5.3.3
General Reliability Function
5.3.4
Mean Time to Failure
Reliability Networks
5.4.1
Series Network
5.4.2
Parallel Network
5.4.3
r-out-of-n Network
5.4.4
Standby Redundancy
5.4.5
Bridge Network
Reliability Evaluation Methods
5.5.1
Network Reduction Approach
5.5.2
Decomposition Approach
5.5.3
Delta-Star Method
5.5.4
Parts Count Method
5.5.5
Markov Method
Reliability Allocation
5.6.1
Hybrid Method
5.6.2
Failure Rate Allocation Method
Problems
References

CHAPTER 6: Failure Modes and Effect Analysis

6.1
Introduction
6.2
Terms and Definitions
6.3
Types of FMEA and Their Associated Benefits
6.3.1
Design-Level FMEA
6.3.2
System-Level FMEA
6.3.3
Process-Level FMEA

1999 by CRC Press LLC

6.4

6.5

6.6
6.7
6.8
6.9
6.10

Steps for Performing FMEA

6.4.1
Define System and Its Associated Requirements
6.4.2
Establish Ground Rules
6.4.3
Describe the System and Its Associated Functional Blocks
6.4.4
Identify Failure Modes and Their Associated Effects
6.4.5
Prepare Critical Items List
6.4.6
Document the Analysis
Criticality Assessment
6.5.1
RPN Technique
6.5.2
Military Standard Technique
FMECA Information Needs, Data Sources, and Users
FMEA Implementation Related Factors and General Guidelines
Advantages of FMEA
Problems
References

CHAPTER 7: Fault Tree Analysis

7.1
Introduction
7.2
FTA Purposes and Prerequisites
7.3
Fault Tree Symbols
7.4
Fundamental Approach to FTA
7.5
Boolean Algebra Rules
7.6
Analytical Developments of Basic Gates, Repeated Fault Events, and
Minimal Cut Sets
7.6.1
Repeated Fault Events
7.6.2
Algorithm for Obtaining Minimal Cut Sets
7.7
Probability Evaluation of Fault Trees with Repairable and
Nonrepairable Components
7.7.1
Probability Evaluation of Fault Trees with Non-Repairable
Components
7.7.2
Probability Evaluation of Fault Trees with Repairable
Components
7.8
Fault Tree Duality and FTA Approach Benefits and Drawbacks
7.9
Problems
7.10 References
CHAPTER 8: Common Cause Failures and Three State Devices
8.1
Introduction
8.2
Common Cause Failures
8.2.1
Block Diagram Method
8.2.2
Fault Tree Method
8.2.3
Markov Method
8.3
Three State Devices
8.3.1
Reliability Evaluation of a Series Network
8.3.2
Reliability Evaluation of a Parallel Network

1999 by CRC Press LLC

8.4
8.5

8.3.3
Reliability Optimization of a Series Network
8.3.4
Reliability Optimization of a Parallel Network
Problems
References

CHAPTER 9: Mechanical Reliability

9.1
Introduction
9.2
Reasons for the Discpline of Mechanical Reliability and Mechanical
Failure Modes
9.3
General and Gear Failure Causes
9.4
Safety Factor and Safety Margin
9.4.1
Safety Factor
9.4.2
Safety Margin
9.5
Design by Reliability Methodology and Stress-Strength Models
9.6
Mellin Transform Method
9.7
Failure Rate Models
9.7.1
Break System Failure Rate Model
9.7.2
Clutch System Failure Rate Model
9.7.3
Pump Failure Rate Model
9.7.4
Filter Failure Rate Model
9.7.5
Compressor System Failure Rate Model
9.7.6
Bearing Failure Rate Model
9.8
Failure Data Sources for Mechanical Parts
9.9
Optimum Models for Mechanical Equipment Replacement or
Maintenance
9.9.1
Equipment Replacement Model
9.9.2
Equipment Maintenance Model
9.10 Problems
9.11 References
CHAPTER 10: Human Reliability in Engineering Systems
10.1 Introduction
10.2 Terms and Definitions
10.3 Human Error Occurrence Examples and Studies
10.4 Human Error Occurrence Classification and Its Types and Causes
10.5 Human Performance and Stress
10.5.1 Stress Factors and Operator Stress Characteristics
10.6 Human Performance Reliability in Continuous Time and Mean Time
to Human Error (MTTHE) Measure
10.7 Human Reliability Evaluation Methods
10.7.1 Probability Tree Method
10.7.2 Fault Tree Method
10.7.3 Markov Method
10.8 Human Reliability Markov Modeling

1999 by CRC Press LLC

10.8.1
10.8.2

Reliability Analysis of a System with Human Error

Reliability Analysis of a Human Performing a
Time-Continuous Task Under Fluctuating Environment
10.9 Human Error Data
10.9.1 Specific Human Error Data Banks and Sources
10.9.2 Brief Description of Selected Human Error Data Banks
10.9.3 Human Error Data for Selective Tasks
10.10 Problems
10.11 References
CHAPTER 11: Reliability Testing and Growth
11.1 Introduction
11.2 Reliability Testing
11.2.1 Reliability Test Classifications
11.2.2 Success Testing
11.2.3 Confidence Interval Estimates for Mean Time Between
Failures
11.2.4 Accelerated Life Testing
11.3 Reliability Growth
11.3.1 Reliability Growth Program
11.3.2 Reliability Growth Process Evaluation Approaches
11.4 Reliability Growth Models
11.4.1 Duane Model
11.4.2 Army Material System Analysis Activity (AMSAA) Model
11.5 Problems
11.6 References
CHAPTER 12: Reliability in Computer Systems
12.1 Introduction
12.2 Terms and Definitions
12.3 Hardware Reliability Vs. Software Reliability
12.4 Computer Failure Causes
12.5 Software Life Cycle Phases and Associated Error Sources
12.6 Software Reliability Improvement Methods
12.6.1 Reliable Software Design Methods
12.6.2 Fault-Tolerant Software Design Methods
12.6.3 Formal Methods
12.6.4 Testing
12.7 Software Reliability Assessment Methods
12.7.1 Analytical Approaches
12.7.2 Software Metrics
12.7.3 Software Reliability Models
12.8 Fault Masking
12.8.1 Triple Modular Redundancy (TMR)
12.8.2 N-Modular Redundancy (NMR)

1999 by CRC Press LLC

12.9 Problems
12.10 References
CHAPTER 13: Robot Reliability and Safety
13.1 Introduction
13.2 Robot Safety
13.2.1 Robot Accidents
13.2.2 Robot Hazards and Safety Problems
13.2.3 Safety Considerations in Robot Life Cycle
13.2.4 Robot Safeguard Approaches
13.2.5 Human Factor Issues in Robotic Safety
13.3 Robot Reliability
13.3.1 Causes and Classifications of Robot Failures
13.3.2 Robot Reliability Measures
13.3.3 Robot Reliability Analysis and Prediction Methods
13.3.4 Reliability and Availability Analyses of a Robot System
Failing with Human Error
13.3.5 Reliability Analysis of a Repairable/Non-Repairable
Robot System
13.4 Problems
13.5 References
CHAPTER 14: Medical Device Reliability
14.1 Introduction
14.2 Facts and Figures, Government Control and Liability
14.3 Medical Electronic Equipment Classification
14.4 Medical Device Recalls
14.5 Medical Device Design Quality Assurance
14.5.1 Organizations
14.5.2 Specifications
14.5.3 Design Review
14.5.4 Reliability Assessment
14.5.5 Parts/Materials Quality Control
14.5.6 Software Quality Control
14.5.7 Design Transfer
14.5.8 Labeling
14.5.9 Certification
14.5.10 Test Instrumentation
14.5.11 Manpower
14.5.12 Quality Monitoring Subsequent to the Design Phase
14.6 Human Error Occurrence and Related Human Factors
14.6.1 Control/Display Related Human Factors Guidelines
14.6.2 Medical Device Maintainability Related Human Factor
Problems

1999 by CRC Press LLC

14.6.3

14.7

14.8
14.9
14.10
14.11
14.12
14.13

Human Factor Pointers for Already Being Used/to be

Purchased Medical Devices
14.6.4 Issues in Considering the Need for, and Performance of
Human Factors Analysis and Testing of Medical Devices
Medical Device Software
14.7.1 Software Testing for Improving Medical Device Safety
14.7.2 Cardioplegia Delivery System Software Reliability Program
Core Elements and Software Safety Improvement with
Redundancy
Sources for Adverse Medical Device Reportable Events and Failure
Investigation Documentation
A small Instrument Manufacturers Approach to Produce Reliable and
Safe Medical Devices
Aerospace and Medical Equipment Reliability and Reliability
Approach Comparisons
Guidelines for Reliability and Other Professionals Associated with
Health Care
Problems
References

CHAPTER 15: Design Maintainability and Reliability Centered

Maintenance
15.1 Introduction
15.2 Design Maintainability: Maintainability Need and Relationship
Between Reliability and Maintainability
15.2.1 Maintainability and Availability Analyses During Design
and Development Phase and Maintainability Design
Characteristics
15.2.2 Design Reviews
15.2.3 Maintainability Measures
15.2.4 Maintainability Design Guidelines and Common
Maintainability Design Errors
15.3 Reliability Centered Maintenance
15.3.1 Traditional Maintenance, RCM, and Questions for Selecting
Assets for RCM
15.3.2 RCM Process Related Maintenance Task Classification Areas
and Steps
15.3.3 RCM Advantages
15.3.4 Reasons for the RCM Methodology Application Failures
15.4 Problems
15.5 References
CHAPTER 16: Total Quality Management and Risk Assessment
16.1 Introduction
16.2 TQM

1999 by CRC Press LLC

16.2.1
16.2.2
16.2.3
16.2.4
16.2.5
16.2.6

16.3

16.4
16.5

Traditional Quality Assurance Vs. TQM

TQM Principles and Elements
TQM Contributors
TQM Methods
Designing for Quality and Goals for TQM Process Success
Demings Deadly Diseases of American Management
and Common Errors During the Starting Quality Initiative
Phase
Risk Assessment
16.3.1 Risk Analysis Role and Required Expertise Areas
16.3.2 Risk Analysis Objectives in Hazardous System Life Cycle
16.3.3 Risk Analysis Process Steps
16.3.4 Risk Analysis Techniques for Engineering Systems
16.3.5 Advantages of Risk Analysis
Problems
References

CHAPTER 17: Life Cycle Costing

17.1 Introduction
17.2 Reasons and Uses of Life Cycle Costing and Required Inputs
17.3 Life Cycle Costing Steps and Activities
17.4 Skill Requirements Areas of a Life Cycle Costing Analyst and
His/Her Associated Professionals
17.5 Life Cycle Costing Program Evaluation Areas and Life Cycle Cost
Estimate Report
17.6 Time Dependent Formulas for Life Cycle Cost Analysis
17.6.1 Single Payment Compound Amount Formula
17.6.2 Single Payment Present Value Formula
17.6.3 Uniform Periodic Payment Present Value Formula
17.6.4 Uniform Periodic Payment Future Amount Formula
17.7 Life Cycle Cost Estimation Models
17.7.1 General Life Cycle Cost Estimation Models
17.7.2 Specific Life Cycle Cost Estimation Models
17.8 Cost Estimation Models
17.8.1 Cost-Capacity Model
17.8.2 Motor Operation Cost Estimation Model
17.8.3 Corrective Maintenance Labor Cost Estimation Model
17.9 Life Cycle Costing Data
17.9.1 Cost Data Sources
17.10 Life Cycle Costing Advantages and Disadvantages Resistance
Factors and Associated Important Points
17.11 Problems
17.12 References

1999 by CRC Press LLC

Introduction

1.1 RELIABILITY HISTORY

The history of the reliability discipline goes back to the early 1930s when probability
concepts were applied to electric power generation related problems [1-3]. During
World War II, Germans applied the basic reliability concepts to improve reliability
of their V1 and V2 rockets. In 1947, Aeronautical Radio, Inc. and Cornell University
conducted a reliability study of over 100,000 electronic tubes. In 1950, an ad hoc
committee on reliability was established by the United States Department of Defense
and in 1952 it was transformed to a permanent body: Advisory Group on the
Reliability of Electronic Equipment (AGREE) [4].
In 1951, Weibull published a statistical function that subsequently became known
as the Weibull distribution [5]. In 1952, exponential distribution received a distinct
edge after the publication of a paper, presenting failure data and the results of various
goodness-of-fit tests for competing failure distribution, by Davis [6].
In 1954, a National Symposium on Reliability and Quality Control was held for
the first time in the United States and in the following year, the Institute of Electrical
and Electronic Engineers (IEEE) formed an organization called the Reliability and
Quality Control Society. During the following two years, three important documents
concerning reliability appeared:
1956: a book entitled Reliability Factors for Ground Electronic Equipment
[7],
1957: AGREE report [8],
1957: first military reliability specification: MIL-R-25717 (USAF): Reliability Assurance Program for Electronic Equipment [9].
In 1962, the Air Force Institute of Technology of the United States Air Force (USAF),
Dayton, Ohio, started the first masters degree program in system reliability engineering. Nonetheless, ever since the inception of the reliability field many individuals
have contributed to it and hundreds of publications on the topic have appeared [10,11].
This chapter discusses various introductory aspects of the reliability discipline.

1.2 NEED OF RELIABILITY IN PRODUCT DESIGN

There have been many factors responsible for the consideration of reliability in
product design including product complexity, insertion of reliability related-clauses
in design specifications, competition, awareness of cost effectiveness, public demand,
and the past system failures. Some of these factors are described below in detail.
Even if we consider the increase in the product complexity with respect to parts
alone, there has been a phenomenal growth of some products. For example, today a

1999 by CRC Press LLC

typical Boeing 747 jumbo jet airplane is made up of approximately 4.5 million parts,
including fasteners. Even for relatively simpler products, there has been a significant
increase in complexity with respect to parts. For example, in 1935 a farm tractor was
made up of 1200 critical parts and in 1990 the number increased to around 2900.
With respect to cost effectiveness, many studies have indicated that the most
effective for profit contribution is the involvement of reliability professionals with
product designers. In fact, according to some experts, if it would cost $1 to rectify
a design defect prior to the initial drafting release, the cost would increase to $10 after
the final release, $100 at the prototype stage, $1000 at the pre-production stage, and
$10,000 at the production stage. Nonetheless, various studies have revealed that
design-related problems are generally the greatest causes for product failures. For
example, a study performed by the U.S. Navy concerning electronic equipment
failure causes attributed 43% of the failures to design, 30% to operation and maintenance, 20% to manufacturing, and 7% to miscellaneous factors [12].
Well-publicized system failures such as those listed below may have also contributed to more serious consideration of reliability in product design [13-15]:
Space Shuttle Challenger Disaster: This debacle occurred in 1986, in
which all crew members lost their lives. The main reason for this disaster
was design defects.
Chernobyl Nuclear Reactor Explosion: This disaster also occurred in
1986, in the former Soviet Union, in which 31 lives were lost. This debacle
was also the result of design defects.
Point Pleasant Bridge Disaster: This bridge located on the West Virginia/Ohio border collapsed in 1967. The disaster resulted in the loss of
46 lives and its basic cause was the metal fatigue of a critical eye bar.

1.3 RELIABILITY IN THE PRODUCT DESIGN PROCESS

Reliability of the design, to a large extent, is determined by the reliability tasks
performed during the product design. These reliability tasks include: establishing
reliability requirements definition, using reliability design standards/guides/checklists, allocating reliability, predicting reliability, reliability modeling, monitoring
subcontractor/supplier reliability activities, performing failure modes effects and
criticality analysis, monitoring reliability growth, assessing software reliability, environmental stress screening, preparing critical items list, and performing electronic
parts/circuits tolerance analysis [16].
Reliability tasks such as those listed above, if performed effectively, will contribute tremendously to the product design.

1.4 RELIABILITY SPECIALIZED AND APPLICATION AREAS

Ever since the inception of the reliability field, the reliability discipline has branched
into many specialized and application areas such as those listed below [10,11]:

1999 by CRC Press LLC

 Mechanical reliability. This is concerned with the reliability of mechanical items. Many textbooks and other publications have appeared on this
topic. A comprehensive list of publications on mechanical reliability is
given in Reference 17.
Software reliability. This is an important emerging area of reliability as
the use of computers is increasing at an alarming rate. Many books have
been written on this topic alone. A comprehensive list of publications on
software reliability may be found in References 10 and 18.
Human reliability. In the past, many times systems have failed not due
to technical faults but due to human error. The first book on the topic
appeared in 1986 [18]. A comprehensive list of publications on human
reliability is given in Reference 19.
Reliability optimization. This is concerned with the reliability optimization of engineering systems. So far, at least one book has been written on
the topic and a list of publications on reliability optimization is given in
References 20 and 21.
Reliability growth. This is basically concerned with monitoring reliability
growth of engineering systems during their design and development. A
comprehensive list of publications on the topic is available in Reference 10.
Structural reliability. This is concerned with the reliability of engineering structures, in particular civil engineering. A large number of publications including books have appeared on the subject [11].
Reliability general. This includes developments on reliability of a general
nature. Usually, mathematicians and related professionals contribute to
the area [10].
Power system reliability. This is a well-developed area and is basically
concerned with the application of reliability principles to conventional
power system related problems. Many books on the subject have appeared
over the years including a vast number of other publications [22].
Robot reliability and safety. This is an emerging new area of the application of basic reliability and safety principles to robot associated problems. Over the years many publications on the subject have appeared
including one textbook [23].
Life cycle costing. This is an important subject that is directly related to
reliability. In particular, when estimating the ownership cost of the product,
the knowledge regarding its failure rate is essential. In the past, many publications on life cycle costing have appeared including several books [24].
Maintainability. This is closely coupled to reliability and is concerned
with the maintaining aspect of the product. Over the years, a vast number
of publications on the subject have appeared including some books [10].

1.5 TERMS AND DEFINITIONS

There are many terms and definitions used in reliability engineering. Some of the
frequently used terms and definitions are presented below [25-28].

1999 by CRC Press LLC

 Reliability. This is the probability that an item will carry out its assigned
mission satisfactorily for the stated time period when used under the
specified conditions.
Failure. This is the inability of an item to function within the initially
defined guidelines.
Downtime. This is the time period during which the item is not in a
condition to carry out its stated mission.
Maintainability. This is the probability that a failed item will be repaired
to its satisfactory working state.
Redundancy. This is the existence of more than one means for accomplishing a defined function.
Active redundancy. This is a type of redundancy when all redundant
items are operating simultaneously.
Availability. This is the probability that an item is available for application
or use when needed.
Mean time to failure (exponential distribution). This is the sum of the
operating time of given items divided by the total number of failures.
Useful life. This is the length of time an item operates within an acceptable
level of failure rate.
Mission time. This is the time during which the item is performing its
specified mission.
Human error. This is the failure to perform a given task (or the performance of a forbidden action) that could lead to disruption of scheduled
operations or result in damage to property/equipment.
Human reliability. This is the probability of completing a job/task successfully by humans at any required stage in the system operation within
a defined minimum time limit (if the time requirement is specified).

1.6 RELIABILITY INFORMATION SOURCES

There are many different sources for obtaining reliability related information. They
may be classified into various different categories. Three such categories are as follows:
1. Journals
Quality and Reliability Engineering, published by John Wiley & Sons
four times a year.
IEEE Transactions on Reliability, jointly published by the Institute of
Electrical and Electronic Engineers (IEEE) and the American Society
for Quality Control (ASQC) five times a year.
Microelectronics Reliability, published by Pergamon Press 12 times a
year.
Reliability, Quality, and Safety Engineering, published by the World
Scientific Publishing Company four times a year.
Quality and Reliability Management, published by MCB University
Press several times a year.

1999 by CRC Press LLC

 Reliability Engineering and System Safety, published by Elsevier Science Publishers several times a year.
Reliability Review, published by the Reliability Division of ASQC four
times a year.
2. Conference Proceedings
Proceedings of the Annual Reliability and Maintainability Symposium
(U.S.)
Proceedings of the Annual Reliability Engineering Conference for the
Electric Power Industry (U.S.)
Proceedings of the European Conference on Safety and Reliability
(Europe)
Proceedings of the Symposium on Reliability in Electronics (Hungary)
Proceedings of the International Conference on Reliability, Maintainability, and Safety (China)
Proceedings of the International Conference on Reliability and Exploitation of Computer Systems (Poland)
3. Agencies
Reliability Analysis Center
Rome Air Development Center (RADC)
Griffis Air Force Base
New York, NY 13441-5700
Government Industry Data Exchange Program (GIDEP)
GIDEP Operations Center
U.S. Department of Navy
Naval Weapons Station
Seal Beach
Corona, CA 91720
National Aeronautics and Space Administration (NASA) Parts Reliability
Information Center
George C. Marshall Space Flight Center
Huntsville, AL 35812
National Technical Information Center (NTIS)
5285 Port Royal Road
Springfield, VA 22151
Defense Technical Information Center
DTIC-FDAC
8725 John J. Kingman Road, Suite 0944
Fort Belvoir, VA 22060-6218
American National Standards Institute (ANSI)
11 W. 42nd St.
New York, NY 10036
Technical Services Department
American Society for Quality Control
611 W. Wisconsin Ave., P.O. Box 3005
Milwaukee, WI 53201-3005

1999 by CRC Press LLC

 Space Documentation Service

European Space Agency
Via Galileo Galilei
Frascati 00044, Italy
System Reliability Service
Safety and Reliability Directorate
UKAEA
Wigshaw Lane, Culcheth
Warrington, WA3 4NE, U.K.

1.7 MILITARY AND OTHER RELIABILITY DOCUMENTS

Over the years many military and other documents concerning reliability have been
developed. Such documents can serve as a useful tool in conducting practically
inclined reliability studies and related tasks. Some of these documents are as follows:
MIL-HDBK-338, Electronics Reliability Design Handbook, Department
of Defense, Washington, D.C.
MIL-HDBK-251, Reliability/Design Thermal Applications, Department
of Defense, Washington, D.C.
MIL-HDBK-217, Reliability Prediction of Electronic Equipment, Department of Defense, Washington, D.C.
MIL-STD-785, Reliability Program for Systems and Equipment, Development and Production, Department of Defense, Washington, D.C.
MIL-HDBK-189, Reliability Growth Management, Department of Defense,
Washington, D.C.
MIL-STD-1556, Government Industry Data Exchange Program (GIDEP),
Department of Defense, Washington, D.C.
MIL-STD-756, Reliability Modeling and Prediction, Department of
Defense, Washington, D.C.
MIL-STD-721, Definitions of Terms for Reliability and Maintainability,
Department of Defense, Washington, D.C.
MIL-STD-52779, Software Quality Assurance Program Requirements,
Department of Defense, Washington, D.C.
MIL-STD-2167, Defense System Software Development, Department of
Defense, Washington, D.C.
MIL-STD-1629, Procedures for Performing a Failure Mode, Effects and
Criticality Analysis, Department of Defense, Washington, D.C.
MIL-STD-2155, Failure Reporting, Analysis and Corrective Action System (FRACAS), Department of Defense, Washington, D.C.
MIL-STD-790, Reliability Assurance Program for Electronic Parts Specifications, Department of Defense, Washington, D.C.
ANSI/AIAA R-013, Recommended Practice for Software Reliability,
American National Standards Institute, New York.
MIL-STD-337, Design to Cost, Department of Defense, Washington, D.C.

1999 by CRC Press LLC

 MIL-STD-1908, Definitions of Human Factors Terms, Department of

Defense, Washington, D.C.
MIL-STD-1472, Human Engineering Design Criteria for Military Systems, Equipment and Facilities, Department of Defense, Washington, D.C.
MIL-HDBK-H108, Sampling Procedures and Tables for Life and Reliability Testing (Based on Exponential Distribution), Department of Defense,
Washington, D.C.
MIL-STD-781, Reliability Design, Qualification and Production Acceptance Tests: Exponential Distribution, Department of Defense, Washington, D.C.
MIL-STD-2074, Failure Classification for Reliability Testing, Department
of Defense, Washington, D.C.
MIL-STD-690, Failure Rate Sampling Plans and Procedures, Department
of Defense, Washington, D.C.
SAE ARD 50010, Recommended Reliability, Maintainability, Supportability (RMS) Terms and Parameters, Society of Automotive Engineers
(SAE), Warrendale, PA.
MIL-STD-472, Maintainability Prediction, Department of Defense, Washington, D.C.

1.8 PROBLEMS
1. Write an essay on the history of reliability.
2. Discuss the need for reliability in product design.
3. Discuss at least three well-publicized system failures and the reasons for
their occurrence.
4. What are the reliability related tasks performed during the product design
process?
5. Describe the following two specialized areas of reliability:
Human reliability
Software reliability
6. What are the differences between reliability and maintainability?
7. Define the following terms:
Failure rate
Redundancy
Failure
8. Describe the following two agencies for obtaining reliability related information:
Reliability Analysis Center
GIDEP
9. In your opinion, what are the five U.S. military reliability related documents that are the most important during the design process? Give reasons
for your selection.
10. Make comparisons between mechanical reliability and power system reliability.

1999 by CRC Press LLC

1.9 REFERENCES
1. Lyman, W.J., Fundamental consideration in preparing a master system plan, Electrical
World, 101, 778-792, 1933.
2. Smith, S.A., Service reliability measured by probabilities of outage, Electrical World,
103, 371-374, 1934.
3. Dhillon, B.S., Power System Reliability, Safety and Management, Ann Arbor Science
Publishers, Ann Arbor, MI, 1983.
4. Coppola, A., Reliability engineering of electronic equipment: a historical perspective,
IEEE Trans. Reliability, 33, 29-35, 1984.
5. Weibull, W., A statistical distribution function of wide applicability, J. Appl. Mech.,
18, 293-297, 1951.
6. Davis, D.J., An analysis of some failure data, J. Am. Statist. Assoc., 47, 113-150, 1952.
7. Henney, K., Ed., Reliability Factors for Ground Electronic Equipment, McGraw-Hill,
New York, 1956.
8. AGREE Report, Advisory Group on Reliability of Electronic Equipment (AGREE),
Reliability of Military Electronic Equipment, Office of the Assistant Secretary of
Defense (Research and Engineering), Department of Defense, Washington, D.C., 1957.
9. MIL-R-25717 (USAF), Reliability Assurance Program for Electronic Equipment,
Department of Defense, Washington, D.C.
10. Dhillon, B.S., Reliability and Quality Control: Bibliography on General and Specialized Areas, Beta Publishers, Gloucester, Ontario, 1992.
11. Dhillon, B.S., Reliability Engineering Application: Bibliography on Important Application Areas, Beta Publishers, Gloucester, Ontario, 1992.
12. Niebel, B.W., Engineering Maintenance Management, Marcel Dekker, New York,
1994.
13. Dhillon, B.S., Engineering Design: A Modern Approach, Richard D. Irwin, Chicago,
IL, 1996.
14. Elsayed, E.A., Reliability Engineering, Addison Wesley Longman, Reading, MA,
1996.
15. Dhillon, B.S., Advanced Design Concepts for Engineers, Technomic Publishing Company, Lancaster, PA, 1998.
16. Reliability, Maintainability, and Supportability Guidebook, SAE G-11 RMS Committee Report, Society of Automotive Engineers (SAE), Warrendale, PA, 1990.
17. Dhillon, B.S., Mechanical Reliability: Theory, Models and Applications, American
Institute of Aeronautics and Astronautics, Washington, D.C., 1988.
18. Dhillon, B.S., Reliability in Computer System Design, Ablex Publishing, Norwood,
NJ, 1987.
19. Dhillon, B.S., Human Reliability with Human Factors, Pergamon Press, New York,
1986.
20. Tillman, F.A., Hwang, C.L., and Kuo, W., Optimization of Systems Reliability, Marcel
Dekker, New York, 1980.
21. Tillman, F.A., Hwang, C.L., and Kuo, W., Optimization techniques for system reliability with redundancy: A review, IEEE Trans. Reliability, 26, 148-155, 1977.
22. Dhillon, B.S., Power System Reliability, Safety and Management, Ann Arbor Science,
Ann Arbor, MI, 1983.
23. Dhillon, B.S., Robot Reliability and Safety, Springer-Verlag, New York, 1991.
24. Dhillon, B.S., Life Cycle Costing: Techniques, Models, and Applications, Gordon and
Breach Science Publishers, New York, 1989.

1999 by CRC Press LLC

25. MIL-STD-721, Definitions of Effectiveness Terms for Reliability, Maintainability,

Human Factors and Safety, Department of Defense, Washington, D.C.
26. Omdahl, T.P., Ed., Reliability, Availability, and Maintainability (RAM) Dictionary,
ASQC Quality Press, Milwaukee, WI, 1988.
27. Von Alven, Ed., Reliability Engineering, Prentice-Hall, Englewood Cliffs, NJ, 1964.
28. Naresky, J.J., Reliability definitions, IEEE Trans. Reliability, 19, 198-200, 1970.

1999 by CRC Press LLC

Design Reliability
Mathematics

2.1 INTRODUCTION
The history of our current number symbols often referred to as the Hindu-Arabic
numeral system may be traced back over 2000 years to the stone columns erected
by the Scythian Indian emperor Asoka. Some of these columns erected around
250 B.C. contain the present number symbols [1]. It appears the Hindus invented
the numeral system and the Arabs transmitted it to western Europe after their invasion
of the Spanish peninsula in 711 A.D. For example, the numeral symbols are found
in a tenth-century Spanish manuscript [1]. However, zero, the Hindu Sunya meaning
empty or void and Arabic Sifr were only introduced into Germany in the
thirteenth century. Nonetheless, around 2000 B.C. Babylonians solved quadratic
equations successfully.
Even though the traces of the differentiation may be found back to the ancient
Greeks, the idea of modern differentiation comes from Pierre Fermat (16011665).
Laplace transforms often used to find solutions to differential equations were one
of the works of Pierre-Simon Laplace (17491827).
The history of probability may be traced back to the writing of a gamblers
manual by Girolamo Cardano (15011576), in which he considered some interesting
issues on probability [1, 2]. However, Pierre Fermat and Blaise Pascal (16231662)
solved independently and correctly the problem of dividing the winnings in a game
of chance. The first formal treatise on probability based on the PascalFermat
correspondence was written by Christiaan Huygens (16291695) in 1657. Boolean
algebra, which plays a significant role in modern probability theory, is named after
the mathematician George Boole (18151864), who published a pamphlet entitled
The Mathematical Analysis of Logic, Being an Essay towards a Calculus of Deductive Reasoning in 1847 [1, 3].
This chapter presents various mathematical concepts related to design reliability.

2.2 BOOLEAN ALGEBRA LAWS

As Boolean algebra plays on important role in probability theory and reliability
studies, some of its associated laws are presented below [3-5]:

COMMUTATIVE LAW:
XY = YX

1999 by CRC Press LLC

(2.1)

where
X is an arbitrary set or event.
Y is an arbitrary set or event.
Dot () denotes the intersection of sets. Sometime Equation (2.1) is written
without the dot but it still conveys the same meaning.
X+Y=Y+X

(2.2)

where
+ denotes the union of sets.

ASSOCIATIVE LAW:

(XY) Z = X (YZ)

(2.3)

where
Z is an arbitrary set or event.

( X + Y) + Z = X + (Y + Z )

(2.4)

X(Y + Z ) = XY + XZ

(2.5)

X + YZ = ( X + Y) ( X + Z )

(2.6)

X + ( XY) = X

(2.7)

X( X + Y ) = X

(2.8)

XX = X

(2.9)

X+X = X

(2.10)

DISTRIBUTIVE LAW:

ABSORPTION LAW

IDEMPOTENT LAW

1999 by CRC Press LLC

2.3 PROBABILITY PROPERTIES

As the basis for reliability theory is probability, this section presents basic properties
of probability. Some of these properties are as follows [6,7]:
The probability of occurrence of event, say A, is
O P (A) 1

(2.11)

Probability of the sample space S is

P (S) = 1

(2.12)

Probability of the negation of the sample space S is

()

P S =1

(2.13)

where
S is the negation of the sample space S.
The probability of the union of n independent events is
P (A1 + A 2 + A 3 + + A n ) = 1

(1 P (A ))
i

(2.14)

i =1

where
Ai is the ith event; for i = 1, 2, , n.
P (Ai) is the probability of occurrence of event Ai ; for i = 1, 2, , n.
For n = 2, Equation (2.14) reduces to
P (A1 + A 2 ) = P (A1 ) + P (A 2 ) P (A1 ) P (A 2 )

(2.15)

The probability of the union of n mutually exclusive events is

P (A1 + A 2 + A 3 + + A n ) =

1999 by CRC Press LLC

P (A )
i

i =1

(2.16)

 The probability of an intersection of n independent events is

P (A1 A 2 A n ) = P (A1 ) P (A 2 n) P (A

(2.17)

The probability of occurrence and nonoccurrence of an event, say A, is

( )

P (A) + P A = 1

(2.18)

where
P (A) is the probability of occurrence of A.
P (A) is the probability of nonoccurrence of A.

2.4 USEFUL DEFINITIONS

This section presents some definitions that are considered to be useful in performing
various design reliability studies.

2.4.1

DEFINITION

OF

PROBABILITY

This is expressed as [6]:

P ( X) = lim ( N n )
n

(2.19)

where
P (X) is the probability of occurrence of event X.
N is the number of times that X occurs in the n repeated experiments.

2.4.2

CUMULATIVE DISTRIBUTION FUNCTION

For continuous random variables, this is defined by [6]:

t

F (t ) =

f (x) dx

where
t is time.
F (t) is the cumulative distribution function.
f (t) is the probability density function.

1999 by CRC Press LLC

(2.20)

For t = , Equation (2.20) becomes

F () =

f (x) dx

(2.21)

=1
It means the total area under the probability density curve is equal to unity.

2.4.3

PROBABILITY DENSITY FUNCTION

This is given by [6, 8]:

t

d f (x) d x

d F (t )
=
dt
dt

(2.22)

= f (t )

2.4.4

RELIABILITY FUNCTION

This is expressed by:

R( t ) = 1 F( t )
t

= 1

f (x) dx

(2.23)

where
R (t) is the reliability function.

2.4.5

HAZARD RATE FUNCTION

This is defined by:

( t ) = f ( t ) R( t )
= f (t ) (1 F (t ))
where
(t) is the hazard rate function.

1999 by CRC Press LLC

(2.24)

2.4.6

LAPLACE TRANSFORM DEFINITION

The Laplace transform of the function f (t) is defined by:

f (s) = f (t )e st d t

(2.25)

where
s is the Laplace transform variable.
t is the time variable.
f (s) is the Laplace transform of f (t).
Example 2.1
If f (t) = et, obtain the Laplace transform of this function.
Thus, substituting the above function into Equation (2.25), we get

f (s) = e t e st d t
0

s+ t
= e ( ) dt

(2.26)

e ( )
(s + )

1
s+

s+ t

Example 2.2
Obtain the Laplace transform of the function f (t) = 1.
By inserting the above function into Equation (2.25), we get

f (s) = 1 . e st d t
0

e st
s

1
s

(2.27)

Table 2.1 presents Laplace transforms of various functions considered useful for
performing mathematical design reliability analysis [9].

1999 by CRC Press LLC

TABLE 2.1
Laplace Transforms of Some
Commonly Occurring Functions
in Reliability Work
f (t)

f (s)
1/(s + )
c/s
s f (s) f (0)

e
c, a constant
d f (t)/d t
t

f (t ) d (t )

f (s)/s

t f (t)
1 f1 (t) + 2 f2 (t)
tn, for n = 0, 1, 2, 3,
t et

d f (s)/d s
1 f1 (s) + 2 f2 (s)
n!/sn+1
1/(s + )2

2.4.7

LAPLACE TRANSFORM: FINAL-VALUE THEOREM

If the following limits exist, then the final-value theorem may be expressed as
lim f (t ) = lim [s f (s)]

2.4.8

s0

(2.28)

EXPECTED VALUE

The expected value, E(t), of a continuous random variable is expressed by

E( t ) = =

t f (t ) d t

(2.29)

where
t is a continuous random variable.
f (t) is the probability density function.
is the mean value.
Similarly, the expected value, E(t), of a discrete random variable t is defined by
m

E( t ) =

t f (t )
i

i =1

where
m is the number of discrete values of the random variable t.

1999 by CRC Press LLC

(2.30)

2.5 PROBABILITY DISTRIBUTIONS

In mathematical reliability analysis, various types of probability distributions are
used; for example, Binomial, Poisson, exponential, Rayleigh, Weibull, general, and
normal. All these distributions are described below [10].

2.5.1

BINOMIAL DISTRIBUTION

This distribution has applications in many combinational-type reliability problems

and sometimes it is also called a Bernoulli (i.e., after Jakob Bernoulli (16541705))
distribution [1].
This distribution becomes very useful in situations where one is concerned with
the probabilities of outcome such as the total number of failures in a sequence of
k trials. However, it is to be noted that for this distribution, each trial has two possible
outcomes (e.g., success and failure), but the probability of each trial remains constant
or unchanged.
The binomial probability density function, f (y), is defined by
k
f ( y) = p y q k y ,
i

for

y = 0, 1, 2,, k

(2.31)

where

()

k
= k! i! ( k i ) !
i

p is the single trial probability of success.

q is the single trial probability of failure.
y is the number of failures in k trials.

It is to be noted that p + q = 1 always. The cumulative distribution function is given by

F( y ) =

i p

q k i

(2.32)

i=0

where
F(y) is the cumulative distribution function or the probability of y or less
failures in k trials.

2.5.2

POISSON DISTRIBUTION

This is another distribution that finds applications in reliability analysis when one
is interested in the occurrence of a number of events that are of the same type. Each
events occurrence is denoted as a point on a time scale and each event represents
a failure. The Poisson density function is expressed by
f (m) =

1999 by CRC Press LLC

( t )m e t ,
m!

for

m = 0, 1, 2,

(2.33)

where
is the constant failure or arrival rate.
t is the time.
The cumulative distribution function, F, is
m

F=

[( t)

e t

i=0

2.5.3

i!

(2.34)

EXPONENTIAL DISTRIBUTION

This is probably the most widely used distribution in reliability engineering because
many engineering items exhibit constant hazard rate during their useful life [11].
Also, it is relatively easy to handle in performing reliability analysis.
The probability density function of the distribution is defined by
f (t ) = e t

t 0, > 0

(2.35)

where
f (t) is the probability density function.
t is time.
is the distribution parameter. In reliability studies, it is known as the
constant failure rate.
Using Equations (2.20) and (2.35), we get the following expression for the cumulative distribution function:
F( t ) = 1 e t

(2.36)

With the aid of Equations (2.29) and (2.35), the following expression for the expected
value of t was obtained:
E( t ) = 1

2.5.4

(2.37)

RAYLEIGH DISTRIBUTION

This distribution is often used in reliability work and in the theory of sound and is
named after John Rayleigh (18421919), its originator [1]. The probability density
function of the distribution is defined as
2
t

2
f (t ) = 2 t e

1999 by CRC Press LLC

t 0, > 0

(2.38)

where
is the distribution parameter.
Inserting Equation (2.38) into Equation (2.20), we get the following equation for the
cumulative distribution function:
F( t ) = 1 e

t 2

(2.39)

Using Equations (2.29) and (2.38), we get the following expression for the
expected value of t:
E(t ) = (3 2)

(2.40)

where
() is the gamma function and is defined by

(x) =

x 1

e t d t,

for x > 0

(2.41)

2.5.5

WEIBULL DISTRIBUTION

This distribution can be used to represent many different physical phenomena and
it was developed by Weibull in the early 1950s [12]. The probability density function
for the distribution is defined by
b
t

b t b 1
f (t ) =
e
b

t 0, > 0, b > 0

(2.42)

where
b and are the shape and scale parameters, respectively.
Using Equations (2.20) and (2.42), we get the following cumulative distribution
function:
F( t ) = 1 e

t b

(2.43)

Substituting Equation (2.42) into Equation (2.29), we get the following equation for
the expected value of t:
1
E ( t ) = 1 +
b

1999 by CRC Press LLC

(2.44)

For b = 1 and 2, the exponential and Rayleigh distributions are the special cases of
this distribution, respectively.

2.5.6

GENERAL DISTRIBUTION

This distribution can be used to represent the entire bathtub shape hazard rate
curve [13]. Its probability density function is defined by:

f (t ) = k u t u 1 + (1 k ) b t b 1 e t

exp k1 tb u ( k ) e

1
(2.45)

for 0 k 1 and , u, b, > 0

where
u and b are the shape parameters.
and are the scale parameters.
t is time.
Substituting Equation (2.45) into Equation (2.20) yields

)]

F (t ) = 1 exp k t u (1 k ) e t 1
b

(2.46)

The following distributions are the special cases of this distribution:

2.5.7

Bathtub hazard rate curve; for b = 1, u = 0.5

Weibull; for k = 1
Extreme value; for k = 0, b = 1
Makeham; for u = 1, b = 1
Rayleigh; for k = 1, u = 2
Exponential; for k = 1, u = 1

NORMAL DISTRIBUTION

This is one of the most widely known distributions and sometimes it is called the
Gaussian distribution after Carl Friedrich Gauss (17771855), the German mathematician. Nonetheless, its use in reliability studies is rather limited. The probability
density function of the distribution is defined by:
f (t ) =

( t )2
1
n
;
exp
2
2
2

< t < +

(2.47)

where
n and are the distribution parameters (i.e., mean and standard deviation,
respectively).

1999 by CRC Press LLC

Using Equations (2.20) and (2.47), we get the following cumulative distribution
function:
( x )2
n
dx
exp
2 2

1
F (t ) =
2

(2.48)

Substituting Equation (2.47) into Equation (2.29) yields:

E (t ) = n

(2.49)

2.6 HAZARD RATE MODELS

In reliability work, the term hazard rate is also often used. In particular, when the
hazard rate of an item is constant, it is simply referred to as the items failure rate.
Nonetheless, generally the hazard rate of an item varies with the time to failure
distribution. This section presents hazard rates for exponential, Rayleigh, Weibull,
general, and normal distributions.

2.6.1

EXPONENTIAL DISTRIBUTION

By substituting Equations (2.35) and (2.36) into Equation (2.24), we get the following hazard rate function for the exponential distribution:
(t ) =

e t
e t

(2.50)

=
It means the hazard rate of an exponential distribution is constant and it is simply
called the failure rate.

2.6.2

RAYLEIGH DISTRIBUTION

Using Equations (2.24), (2.38), and (2.39), the following hazard rate expression is
for the Rayleigh distribution:
2
2
2
(t ) = 2 t e ( t ) e ( t )

(2.51)

= 2 t 2
Equation (2.51) indicates that the hazard rate of the Rayleigh distribution increases
linearly with time.

1999 by CRC Press LLC

2.6.3

WEIBULL DISTRIBUTION

Inserting Equations (2.42) and (2.43) into Equation (2.24) yields:

b
b
b
(t ) = b t b1 e ( t ) e ( t )

(2.52)

b
= b t b 1

Equation (2.52) indicates that the Weibull hazard rate is a function of time t and at
b = 1 and b = 2, it reduces to the hazard rates for exponential and Rayleigh distributions, respectively.

2.6.4

GENERAL DISTRIBUTION

With the aid of Equations (2.24), (2.45), and (2.46), we get the following expression
for the general distribution hazard rate:

(t ) = k u t u1 + (1 k ) b t b1 e t

exp k1 tb u ( k ) e

)]

exp k t u (1 k ) e t 1
b

= k u t u1 + (1 k ) b t b1 e t

(2.53)

Obviously, in Equation (2.53) the hazard rate, (t), varies with time and its special
case hazard rates are for Weibull, bathtub, extreme value, Makeham, Rayleigh, and
exponential distributions.

2.6.5

NORMAL DISTRIBUTION

Using Equations (2.24), (2.47), and (2.48), we get the following hazard rate expression for the normal distribution hazard rate:
1
( t )2
n

(t ) =
exp
2

2
2

1
n

x ( )
exp
2 2

d x

(2.54)

2.7 PARTIAL FRACTION METHOD

AND EQUATION ROOTS
2.7.1

PARTIAL FRACTION METHOD

This is a useful approach to determine inverse Laplace transforms of a rational

function such as M(s)/Q(s), where M(s) and Q(s) are polynomials, and the degree

1999 by CRC Press LLC

of M(s) is less than that of Q(s). Consequently, the ratio of M(s)/Q(s) may be
expressed as the sum of rational functions or partial fractions as follows:
A (a s + b) , for m = 1, 2, 3,
m

(Bs + C) a s2 + bs + c

(2.55)

, for m = 1, 2, 3,

(2.56)

where
A, B, C, a, b, and c are the constants.
s is the Laplace transform variable.
Heaviside Theorem
This theorem is due to Oliver Heaviside (18501925) [1] and is used to obtain partial
fractions and the inverse of a rational function, M(s)/Q(s). Thus, the inverse of
M(s)/Q(s) is expressed as follows:
M(s)
1
=
Q(s)

M ( i )

Q ( ) e
i =1

it

(2.57)

where
the prime on Q denotes differentiation with respect to s.
m denotes the total number of distinct zeros of Q(s).
i denotes zero i; for i = 1, 2, 3, , m.
t is time.
Example 2.3
Suppose
M(s)
s
=
.
Q(s) (s 5) (s 7)
Find its inverse Laplace transforms by applying the Heaviside theorem.
Hence,
M(s) = s, Q(s) = (s 5) (s 7) = s2 , 12 s + 35 Q (s)
= 2 s 12, m = 2, 1 = 5, and 2 = 7.

1999 by CRC Press LLC

(2.58)

Using the above values in the right-hand side of Equation (2.57) yields:
=

M(1 ) 1 t M ( 2 ) 2
e +
e
Q ( 1 )
Q ( 2 )

M(5) 5t M(7) 7 t
e +
e
Q (5)
Q (7)

=
=

(2.59)

5 5t 7 7t
e + e
2
2

7 7t 5 5t
e e
2
2

Thus, the inverse Laplace transform of Equation (2.58) is given by Equation (2.59).

2.7.2

EQUATION ROOTS

In reliability work, sometimes it may involve finding roots of algebraic equations.

A root is a value of variable say, y, that when substituted into the polynomial equation
results in the value of the equation equal to zero. A polynomial equation is considered
solved when all its roots have been found.
Even though around 2,000 B.C. Babylonians solved quadratic equations and the
ancient Indian mathematicians such as Brahmagupta recognized negative roots, in
the western society prior to the 17th century, the theory of equations was handicapped by the failure of mathematicians to recognize negative or complex numbers
as roots of equations [1]. Nonetheless, formulas to obtain roots of quadratic and
cubic equations are presented below [14, 15].
Quadratic Equation
This is defined by:
a y2 + b y + c = 0

(2.60)

where
a, b, and c are the constants.
Solutions to Equation (2.60) are as follows:
y1, y 2 = b W1 2 2 a

(2.61)

W b2 4 a c

(2.62)

where

1999 by CRC Press LLC

If y1 and y2 are the roots, then

y1 + y 2 = b a

(2.63)

y1y 2 = c a

(2.64)

and

If a, b, and c are real, then the roots are as follows:

Real and equal, if W = 0
Real and unequal, if W > 0
Complex conjugate if W < 0
Cubic Equation
y 3 + a 2 y 2 + a1 y + a 0 = 0

(2.65)

Let
1
1
a a2
3 1 9 2

(2.66)

1
(a a 3 a 0 ) 271 a 23
6 1 2

(2.67)

M=
N=

If y1, y2, and y3 are the roots of Equation (2.65), then we have:
y1 y 2 y3 = a 0

(2.68)

y1 + y 2 + y3 = a 2

(2.69)

y1 y 2 + y1 y3 + y 2 y3 = a1

(2.70)

The following types of roots are associated with Equation (2.65):

All roots real (irreducible case) if M3 + N2 < 0.
All roots real and at least two equal if M3 + N2 = 0.
One real root and one pair of complex conjugate roots if M3 + N2 > 0.
If we let

g1 = N + M3 + N 2

1999 by CRC Press LLC

12 12

(2.71)

and

(2.72)

y1 = (g1 + g2 ) a 2 3

(2.73)

g1 = N M 3 + N 2

12 12

then we have the following roots:

y2 =

1
(g + g ) a 2 3 + i 23 (g1 g2 )
2 1 2

(2.74)

y3 =

1
(g + g ) a 2 3 i 23 (g1 g2 )
2 1 2

(2.75)

2.8 DIFFERENTIAL EQUATIONS

In reliability work, the use of the Markov method results in a system of linear firstorder differential equations. Thus, in order to obtain state probabilities at time t, the
solution to these differential equations has to be found. For this purpose, there are
various methods that can be used, but the Laplace transform approach appears to
be the most attractive.
Finding solutions to a system of differential equations is demonstrated through
the following example:
Example 2.4
Assume that an engineering system can either be in two states: operating or failed.
The following two differential equations describe the system:
d P0 (t )
+ P0 (t ) = 0
dt

(2.76)

d P1 (t )
P0 (t ) = 0
dt

(2.77)

where
Pi (t) is the probability that the engineering system is in state i at time t; for
i = 0 (operating), i = 1 (failed).
is the system failure rate.
At time t = 0, P0 (0) = 1, and P1 (0) = 0.
Find solutions to Equations (2.76) and (2.77).

1999 by CRC Press LLC

In this case, we use the Laplace transform method. Thus, taking Laplace transforms of Equations (2.76) and (2.77) and using the given initial conditions, we get
P0 (s) =

1
s+

P1 (s) = s (s + )

(2.78)

(2.79)

The inverse Laplace transforms of Equations (2.78) and (2.79) are as follows:
P0 (t ) = e t

(2.80)

P1 (t ) = 1 e t

(2.81)

Thus, Equations (2.80) and (2.81) represent solutions to differential Equations (2.76)
and (2.77).

2.9 PROBLEMS
1. Write an essay on the history of mathematics including probability theory.
2. Prove the following Boolean algebra expression:
X + YZ = ( X + Y) ( X + Z )

(2.82)

R (t ) + F (t ) = 1

(2.83)

3. Prove that

where
R (t) is the item reliability at time t.
F (t) is the item failure probability at time t.
4. Obtain the Laplace transform of the following function:
f (t ) =

t k 1 ea t
,
( k 1)!

for k = 1, 2, 3,

(2.84)

5. Take the inverse Laplace transform of the following:

f (s) =

1999 by CRC Press LLC

(s + 4) (s + 3)

(2.85)

6. Obtain an expression for the expected value of the general distribution.

Comment on your result.
7. Discuss hazard rates of exponential and Rayleigh distributions.
8. Prove that the expected value of the Weibull distribution is given by
1
E ( t ) = 1 +
b

(2.86)

where
and b are the scale and shape parameters, respectively.
9. What are the special case distributions of the Weibull distribution?
10. What are the special case distributions of the general distribution?

2.10

REFERENCES

1. Eves, H., An Introduction to the History of Mathematics, Holt, Rinehart, and Winston,
New York, 1976.
2. Owen, D.B., Ed., On the History of Statistics and Probability, Marcel Dekker, New
York, 1976.
3. Lipschutz, S., Set Theory, McGraw-Hill, New York, 1964.
4. Fault Tree Handbook, Report No. NUREG-0492, U.S. Nuclear Regulatory Commission, Washington, D.C., 1981.
5. Ramakumar, R., Engineering Reliability: Fundamentals and Applications, PrenticeHall, Englewood Cliffs, NJ, 1993.
6. Mann, N.R., Schafer, R.E., and Singpurwalla, N.D., Methods for Statistical Analysis
of Reliability and Life Data, John Wiley & Sons, New York, 1974.
7. Lipschutz, S., Probability, McGraw-Hill, New York, 1965.
8. Shooman, M.L., Probabilistic Reliability: An Engineering Approach, McGraw-Hill,
New York, 1968.
9. Oberhettinger, F. and Badic, L., Tables of Laplace Transforms, Springer-Verlag, New
York, 1973.
10. Patel, J.K., Kapadia, C.H., and Owen, D.B., Handbook of Statistical Distributions,
Marcel Dekker, New York, 1976.
11. Davis, D.J., An analysis of some failure data, J. Am. Stat. Assoc., June, 113- 150, 1952.
12. Weibull, W., A statistical distribution function of wide applicability, J. Appl. Mech.,
18, 293-297, 1951.
13. Dhillon, B.S., A hazard rate model, IEEE Trans. Reliability, 29, 150, 1979.
14. Abramowitz, M. and Stegun, I.A., Eds., Handbook of Mathematical Functions with
Formulas, Graphs, and Mathematical Tables, National Bureau of Standards, Washington, D.C., 1972.
15. Spiegel, M.R., Mathematical Handbook of Formulas and Tables, McGraw-Hill, New
York, 1968.

1999 by CRC Press LLC

Engineering Design and

Reliability Management

3.1 INTRODUCTION
For large engineering systems, management of design and reliability becomes an
important issue. Thus, this chapter is devoted to engineering design and reliability
management.
In engineering, the team design/designer may convey different meanings to
different people [1]. For example, to some a design is the creation of a sophisticated
and complex item such as a space satellite, computer, or aircraft and to others a
designer is simply a person who uses drawing tools to draw the details of an item.
Nonetheless, it is the design that distinguishes the engineer from the scientist as
well as it is the design that ties engineers to markets, which give vent to the creativity
of the engineering profession.
Imhotep, who constructed the first known Egyptian pyramid, Saqqara, in 2650 BC
may be called the first design engineer. As the engineering drawings are closely tied
to engineering design, their history could be traced back to 4000 BC when the
Chaldean engineer Gudea engraved upon a stone tablet the plan view of a fortress [2].
The written evidence of the use of technical drawings only goes back to 30 BC
when Vitruvius, a Roman architect, wrote a treatise on architecture. In 1849, in
modern context, an American named William Minifie was the first person to write
a book on engineering drawings entitled Geometrical Drawings.
Today, there are a vast number of publications on both engineering design and
engineering drawings [3]. An engineering design is only good if it is effective and
reliable.
The history of reliability management is much shorter than that of the engineering design and it can only be traced back to the 1950s. Probably the first prime
evidence related to reliability management is the formation of an ad hoc committee
in 1958 for Guided Missile Reliability under the Office of the Assistant Secretary
of Defense for Research and Engineering [4]. In 1959, a reliability program document (exhibit 58-10) was developed by the Ballistic Missile Division of the U.S.
Air Force. The military specification, MIL-R-27542, was the result of Department
of Defense efforts to develop requirements for an organized contractor reliability
program. In 1985, a book entitled Reliability and Maintainability Management was
published [5] and a comprehensive list of publications on the subject is given in
Reference 6.
This chapter discusses engineering design and reliability management separately.

1999 by CRC Press LLC

3.2 ENGINEERING DESIGN

Nowadays, design is increasingly seen as the key to industrial success in all markets.
In simple terms, design is all about producing the right products for the right markets
at the right price, the right time, the right quality and reliability to the right specification and performance [7]. Nonetheless, it may be said that even though a good
design may not guarantee success, it is an extremely powerful contributor to it. Some
of the important characteristics of a good design include long useful life, high
reliability, low cost and maintenance, high accuracy, and attractive appearance.
There would be many reasons for designing engineering products: reducing hazard,
developing a new way, lowering the cost, meeting social changes, meeting competition,
etc. The subsequent sections address the different aspects of engineering design.

3.2.1

DESIGN FAILURES

AND

THEIR COMMON REASONS

History has witnessed many engineering design failures, in particular, in the immediate past two well-publicized design failures were the Chernobl nuclear reactors
and the space shuttle Challenger. Some of the other important design failures are as
follows [8, 9]:
In 1937, a high school in Texas converted from burning methane city gas
to less expensive natural gas. Subsequently an explosion occurred and
455 persons died. An investigation of the accident indicated that the area
through which the pipes ran was poorly ventilated. A recommendation
was made for the installation of malodorants to all natural gas to detect
any possible leaks.
In 1963, a U.S. Navy nuclear submarine, the U.S.S. Thresher, slipped
beneath the Atlantic Ocean surface by exceeding its designed maximum
test depth and imploded.
In 1979, a DC-10 airplane lost an engine during a flight and subsequently
crashed. A follow up investigation indicated that the normal engine service
operation was the cause of the problem. More specifically, as the engines
were periodically dismounted, serviced, and remounted, the mounting
holes became elongated during the servicing process and subsequently
led to the disaster.
In 1980, in the North Sea an offshore oil rig named Alexander L. Kielland
broke up under normal weather conditions. A subsequent study of the
disaster revealed that a 3-in. crack in a part close to a weld joint was the
basic cause of the disaster.
In 1988, a Boeing 737-200 lost its cabin roof during a flight. A preliminary
investigation of the disaster indicated that various metal fatigue cracks
were emanating from rivet holes in the aluminum skin.
Over the years many professionals have studied various design failures and
concluded that there could be many different reasons for product failures ranging
from a disaster to simple malfunction. Most of the common reasons are given in
Table 3.1

1999 by CRC Press LLC

TABLE 3.1
Common Reasons for Product Failures

3.2.2

No.

Reason

1
2
3
4
5
6
7
8
9

Wrong usage by the consumer

Incorrect manufacturing
Faulty reasoning
Poor understanding of the problem to be solved
Unsatisfactory data collection
Incorrect storage
Incorrectly stated problem with respect to basic principles
Wrong or over-extended assumptions
Erroneous data

THE DESIGN PROCESS

AND

FUNCTIONS

The design process is an imaginative integration of scientific information, engineering technology, and marketing for developing a useful and profitable item. The
process may be described in as little as from 5 to 25 steps; for example, in 6 steps
by Dieter [10], 8 steps by Vidosic [11], and 12 steps by Hill [12]. Nonetheless, for
our purpose we will describe it in six steps:
1. Need recognition. This is concerned with understanding and identifying
the needs of the user.
2. Problem definition. This is concerned with developing concise problem
statement, identifying problem related needs and limitations, and obtaining related information.
3. Information gathering. This is concerned with collecting various types
of information from sources including specifications and codes, patent
gazette, Internet, handbooks, technical experts, vendor catalogs, and journals.
4. Conceptualization. The results of conceptualization may take the form
of sketches or free-hand drawings and at this stage, the design process
may be regarded as a creative and innovative activity that generates various
possible alternative solutions to the identified goal. Some of the idea
generation techniques that design professionals can use include group
brainstorming, attribute listing, synectics, and morphological chart.
5. Evaluation. This is concerned with deciding on the best solution out of
all the potential solutions. Some of the evaluations include evaluation
based on feasibility judgment, evaluation based on technology-readiness
assessment, and evaluation based on GO/NO-GO screening.
6. Communication of design. The final solution to the engineering problem
leads to final documents representing the product or the product itself, for
the purpose of communicating design to others. The design documents
include items such as engineering drawings, operation and maintenance
instructions, information on quality assurance, patent applications, and
bills of materials.

1999 by CRC Press LLC

There are basically five functions involved in engineering design: (1) research,
(2) engineering, (3) manufacturing, (4) quality assurance, and (5) commercial. The
research function activities include conducting basic and applied research, developing specifications for quality testing procedures, and preparing process specifications
for the testing of highly stressed parts. The engineering functions are the subcomponents of the design activity and include activities such as developing new design
concepts, estimating cost, developing production design, making provisions of maintenance instructions, and analyzing field problems. The manufacturing functions
include activities such as assembly, manufacturing planning, determining tooling
needs, and purchasing materials. The main objective of the quality assurance function
is to assure the quality of the end product. Some of the activities covered by the
quality assurance function are designing quality related methods and procedures,
and setting up design auditing. The commercial function is concerned with the
relationship of various clients and its activities include conducting market surveys
and tendering, advertising, managing contracts, and arranging delivery and payments.

3.2.3

THE DESIGN TEAM

AND

MEMBER RESPONSIBILITIES

The design engineer is not the only one person who usually develops design. In real
life, there are many other individuals who also participate. For example, representatives from areas such as manufacturing, field services, and quality control are
involved. The nature of the product dictates the degree of their participation. The
design team members responsibilities may be grouped into four areas: (1) design,
(2) manufacturing, (3) quality, and (4) field service [7]. The design related responsibilities include initial design concept, functional integrity, prototype build, test
procedures, documentation issue, specifications, cost estimates, materials, performance goals, bill of materials, and coordination with reliability group, safety group,
standards group, planning group, and model shop.
Some of the responsibilities belonging to the manufacturing area are tooling,
model build co-ordination, assembly procedures, test and set up procedures, vendor
involvement, pre-production builds, procurement schedule, and co-ordination with
quality control, product planning, materials management, and industrial engineering.
The quality related responsibilities include vendor training, quality audits, field
reliability, total quality management, quality circles, supplier approval, and product
verification.
The field service related responsibilities are service procedures, service tools,
service training, customer training, spares list, service documentation, and so on.

3.2.4

DESIGN REVIEWS

The design reviews are an important factor during the design phase of an engineering
product. The basic purpose of such reviews is to assure the application of correct
design principles, in addition to determining whether the design effort is progressing
according to specifications, plans, and so on.
Various types of reviews are performed during the design phase of a product
and the total cost of performing such reviews varies between 1 and 2% of the total
engineering cost of a project [12].

1999 by CRC Press LLC

Different writers and practitioners categorize design reviews differently. For our
purpose, we have divided them into three major categories: (1) preliminary design
review, (2) intermediate design review, and (3) critical design review [12, 13]. Each
of these three category is discussed below.
1. Preliminary design review. This design review is usually conducted prior
to the formulation of the initial design. The primary aim of this review is
to evaluate each and every design specification requirement with respect
to accuracy, validity, and completeness. The areas that could be reviewed
during this design review include cost objective, design alternatives,
present/future availability of materials, applicable legislations, potentional
users/customers, schedule imposed requirements, required functions, customer needs, and critical parts/components.
2. Intermediate design review. This review is conducted before starting the
detailed production drawings. The main objective of this review is to make
comparison of each specification requirement with the design under development. It is to be noted that prior to starting this review, the design
selection process and preliminary layout drawings are complete.
3. Critical design review. This is also known as the final design review and
is conducted after the completion of production drawings. The emphasis
of this review is on areas such as design producibility, value engineering,
review of analysis results, and so on.
Design Review Team
Professionals from various areas participate in design reviews and their number and
type may vary from one project to another. For an effective performance, the size
of the team should not exceed 12 participants [12]. A typical design review team is
made up of professionals such as design engineer, senior design engineer, design
review board chairman, manufacturing engineer, tooling engineer, procurement engineer, customer representative(s) (if any), test engineer, reliability engineer, materials
engineer, and quality control engineer. Each of these individuals evaluates design
from his/her perspective.
Design Review Related Information and Topics
In order to conduct effective reviews, the design review team members must have
access to various information items, as appropriate, including parts list, specifications
and schematic diagrams, acceleration and shock data, circuit descriptions, results of
reliability prediction studies, list of inputs/outputs, results of failure modes and effect
analysis (FMEA), and vibration and thermal tests data.
Usually during design reviews, topics such as mechanical design (i.e., results of
tests, thermal analysis, balance, etc.), specifications (i.e., adherence to specifications,
correctness of specifications, etc.), human factors (i.e., glare, control and display,
labeling and marking, etc.), reproducibility (i.e., reliance of products on a single
part supplier, economical assembly of product in the production shop, etc.), electrical
design (i.e., design simplification, electrical interference, performance, results of

1999 by CRC Press LLC

circuit analysis, etc.), standardization, value engineering, safety, maintainability (i.e.,

interchangeability, maintenance philosophy, etc.), and reliability (i.e., reliability
predictions, reliability allocations, the results of failure modes and effect analysis
(FMEA), etc.) are discussed.

3.2.5

DESIGN ENGINEER

AND

DESIGN REVIEW BOARD CHAIRMAN

Broadly speaking, a design engineer is basically concerned with developing a product

or solving a problem. More specifically, some of the tasks performed by this individual includes designing the product, keeping the design within specified constraints, keeping abreast of changing environments and technology, answering questions concerning the design under consideration, participating in design reviews, and
optimizing the design.
The important qualities of a typical design engineer include creative and innovative mind, good conceptual ability, excellent scientific knowledge, ability to think
logically, good communication skills, and good personal qualities.
The design review board chairman is a key member of the design review team.
Usually, this person belongs to the engineering department and is not in direct line
of authority to the designer whose design is under consideration. Nonetheless, a
configuration manager often is appointed by the design review board chairman [14].
Some of the important qualities of the design review board chairman are general
technical competence, understanding of the problem under consideration, free of
bias for or against the proposed design, appropriate skills to lead a technical meeting,
a pleasant personality, and a high degree of tact and discretion.
Some of the functions performed by the design review board chairman are
determining the type of review to be conducted, scheduling the design reviews,
chairing the meetings of the design review board, establishing the procedure for
selecting particular items for review, evaluating comments from the design review
meetings, supervising the publication of minutes, and providing appropriate assistance to the design organization with respect to preparation of necessary design
review data.

3.2.6

DESIGNING FOR RELIABILITY AND DESIGN RELIABILITY CHECK

LIST ITEMS

A design is only useful if it is reliable. Thus, reliability is a crucial factor in

engineering design. Nonetheless, initial design concepts influenced by the overall
reliability requirement will depend upon factors such as an items/products functional complexity, the cost target, items/products usage requirements, the size
envelope, and the state of development and technical understanding available with
new materials/parts/technologies considered for use.
The design process aspects [7, 15] that can have significant impact on product
reliability include design simplicity, use of parts with known history, failure mode
analysis, understanding of the requirements and the rules, allowance for possible test
house approval requirements in design of electrical, mechanical, safety, and material
selection aspects, procuring outside components with care, making allowance for
rough handling, use abuse, and transportation, simplicity in manufacturability, and

1999 by CRC Press LLC

confirmation of mechanical design choices mathematically as much as possible, e.g.,

bearings, load deflections, etc.
In order to consider various aspects of reliability during the design process
effectively, it is useful to prepare a checklist of reliability related items. Some of
these items are as follows [13]:
Minimization of adjustments
Establishment of requirements for performance/reliability/environment/life/signals for each and every concerned component, interface, and
structure
Adequacy of mechanical support structures
Efficiency of heat transfer devices or designs
Effective usage of failure-indicating devices
Specification of the particular design criteria for reliability and safety of
each and every product component
Identification of components requiring special procurement, testing, and
handling
Provisions for improvements to eradicate any design deficiencies observed
during testing
Establishment of maintenance requirements and maintainability goals
Adequate protection against insertion of similar plugs and connectors in
wrong sockets
Establishment of satisfactory reliability assurance, acceptance, qualification, and sampling tests
Identification of limited life parts
Installation of self-monitoring or self-calibration devices in major products where feasible
Selection of components, interfaces, and structures to meet reliability
objectives
Determination of shelf life of parts chosen for final design
Usage of appropriate safety factors in the application of parts/interfaces/structures
Establishment of stability requirements of all parts and structures associated with each and every adjustment
Utilization of best available methods/techniques for lowering the adverse
effects of operational environments on critical structures
Classification of all critical adjustments with respect to factory, pre-operational, or operator types
Performance of studies considering variability and degradation of
part/structure parameters
Effort expended to make the developmental model as close as the production model
Emphasis of the design to eliminate shock mounts and vibration isolators
where it is feasible
Design of packaging and mechanical layout to facilitate effective maintenance.

1999 by CRC Press LLC

3.2.7

DESIGN LIABILITY

This is a very important factor in product design and can simply be stated as the
liability of designers for injury, loss, or damage caused by defects in products
designed by them. According to Roche [16], approximately 40% of causes of product
liability suits are due to defects in design.
In determining the defectiveness of a product, factors such as the marketing of
product, stated instructions, stated warnings, time of sale, and foreseeable application
and intended use are taken into consideration. In situations where injury results, the
important points (i.e., in U.K.) affecting design liability include [17]:
The designer cannot be liable when the risk of injury was not reasonably
foreseeable.
The designer cannot be totally liable when the risk of injury was foreseeable to him/her and he/she took reasonable design precautions.
The designer is liable when the risk of injury was reasonably foreseeable
to him/her and was not obvious to the product user but the designer took
no reasonable measures to warn.
The designer cannot be totally liable when the risk of injury was reasonably foreseeable and obvious to the user.
All in all, there is no simple answer as to how to reduce design-related risks,
but the design reviews are an appropriate vehicle for identifying and evaluating
potential safety hazards.

3.3 RELIABILITY MANAGEMENT

In the 1990s, there has been a major shift in the perception of reliability engineering
by government, manufacturer, and consumer sectors. For example, in the 1960s and
1970s, the reliability engineering function was a discrete unit of a matrix organization;
in todays environment it is an integral component of an engineering design team.
Over the years, as the engineering projects have become more complex and
sophisticated, the application of reliability principles has also become a trying
management task. This has led to the development of various reliability management
related areas [18].

3.3.1

GENERAL MANAGEMENT RELIABILITY PROGRAM RELATED

RESPONSIBILITIES AND GUIDING FORCE ASSOCIATED FACTS
FOR AN EFFECTIVE RELIABILITY PROGRAM

In order to have an effective reliability program, there are various factors in which
the responsibilities of management [19] lie:
Establishing a program to access with respect to reliability of the current
performance of company operations and the product it manufactures;
Establishing certain reliability objectives or goals;

1999 by CRC Press LLC

 Establishing an effective program to fulfill set reliability goals and eradicating current deficiencies. An absolutely effective program should be
able to pay in return many times its establishing cost;
Providing necessary program related authority, funds, manpower, and time
schedule;
Monitoring the program on a regular basis and modifying associated
policies, procedures, organization, and so on, to the most desirable level.
Facts such as the following will be a guiding force for the general management to
have an effective reliability program [20]:
Reliability is an important factor in the management, planning, and design
of an engineering product.
Changes in maintenance, manufacturing, storage and shipping, testing and
usage in the field of the engineering product tend to lower the reliability
of the design.
Planned programs are needed for application in design, manufacturing,
testing and field phases of the engineering product to control reliability.
Reliability is established by the basic design.
It is during the early phases of the design and evaluation testing programs
when high levels of reliability can be achieved most economically.
Improvement in reliability can be through design changes only.
Human error degrades the reliability of the design.
In achieving the desired reliability in a mature engineering product in a
timely manner, deficiency data collection, analysis, and feedback are
important.

3.3.2

MILITARY SPECIFICATION GUIDELINES FOR DEVELOPING

RELIABILITY PROGRAMS AND A PROCEDURE TO ESTABLISH
RELIABILITY GOALS

For the development of reliability programs, the military specification, MIL-R27542, presents the following guidelines [21]:
Assign reliability associated goals for system in question.
To obtain maximum inherent equipment reliability, put in as much effort
as possible during the design phase.
Evaluate reliability margins.
Perform specification review.
Perform design and procedure reviews.
Establish an appropriate testing program.
Conduct a review of changes in specification and drawings with respect
to reliability.
Establish and maintain control during production through inspection, sample testing, and effective production control.
Develop a closed-loop system for failure reporting, analysis, and feedback
to engineering specialists for corrective measures to stop re-occurrence.

1999 by CRC Press LLC

 Assign responsibility for reliability to a single group.

Make sure that the reliability group is reporting to an appropriate authority
so that the reliability work can be performed effectively.
Develop an on-the-job training facility and program.
When working from preestablished requirements, developing reliability goals are
essential and the reliability goal setting basically consists of reducing the overall
requirements to a series of subgoals. A general procedure to establish reliability
goals is as follows [22]:
Clarify requirements.
Review the organizational objective and the organizational units mission
prior to developing the units goals/purposes/missions to pursue the main
goal.
Highlight desired important result areas.
Determine likely highest payoff areas.
Choose most promising result areas to pursue.
Choose goal candidates.
Review resource requirements to pursue each candidate goal to achieve
it successfully.
Highlight any problem area in achieving the goal and proposed solutions.
Rank all candidate goals with respect to ease of payment and the degree
of payoff.
Examine goal interdependencies and make adjustments in goal candidates
for maximum coordination.
Review goals with respect to factors such as attainability, compatibility,
relevance, acceptability, measurability, and supportability.
Make final selection of goals and develop milestones for their successful
achievement. Develop parameters to measure success.
Establish action plans for the goal achievement by considering provisions
for motivational initiatives/supervisions and management support.
Communicate goals in written form to all concerned bodies and review
their progress periodically.
Make adjustments as appropriate.

3.3.3

RELIABILITY AND MAINTAINABILITY MANAGEMENT TASKS

IN THE PRODUCT LIFE CYCLE

To obtain required reliability of an engineering system in the field, the management

tasks with respect to reliability and maintainability has to be coordinated throughout
its life cycle. The life cycle of a system may be divided into four major distinct
phases: (1) the concept and definition phase, (2) the acquisition phase, (3) the operation and maintenance phase, and (4) the disposal phase [5]. Reliability and maintainability management is involved in all four phases. Its tasks in each of these
phases are presented below.

1999 by CRC Press LLC

The Concept and Definition Phase

In this phase, system requirements are established and the basic characteristics are
defined. There are various reliability and maintainability related management tasks
involved in this phase. Some of them are as follows:
Defining all terms used, the system capability requirements, management
controls, and parts control requirements.
Defining a failure.
Defining the reliability and maintainability goals for the system in quantitative terms.
Defining hardware and software standard documents to be used to fulfill
reliability and maintainability requirements.
Defining methods to be followed during the design and manufacturing phase.
Defining constraints proven to be harmful to reliability.
Defining system safety requirements and the basic maintenance philosophy.
Defining data collection and analysis needs during the system life cycle.
Defining management control for documentation and providing necessary
documentation.
Defining system environmental factors during its life cycle.
The Acquisition Phase
This is the second phase of the system life cycle, and it is concerned with activities
associated with system acquisition and installation as well as planning for the
eventual support of the system. During this phase, some of the reliability and
maintainability related management tasks are as follows:
Define the system technical requirements, major design and development
methods to be used, documents required as part of the final system, type of
evaluation methods to assess the system, and demonstration requirements.
Define the reliability and maintainability needs that must be fulfilled.
Define the meaning of a failure or a degradation.
Define the types of reviews to be conducted.
Define the type of data to be supplied by the manufacturer to the customer.
Define the life-cycle cost information to be developed.
Define the kind of field studies to be performed, if any.
Define the kind of logistic support required.
The Operation and Maintenance Phase
This is the third phase of the system life cycle and is concerned with tasks related
to the maintenance, management of the engineering and the support of the system
over its operational life period. Some of the reliability and maintainability related
management tasks involved during the operation and maintenance phase are as
follows:

1999 by CRC Press LLC

Developing failure data banks.

Analyzing reliability and maintainability data.
Providing adequate tools for maintenance.
Providing appropriately trained manpower.
Managing and predicting spare parts.
Developing engineering change proposals.
Preparing maintenance documents.
Reviewing documents in light of any engineering change.

The Disposal Phase

This is the last phase of the system life cycle and it is concerned with activities
needed to remove the system and its nonessential supporting parts. Two of the
reliability and maintainability management tasks involved in this phase are calculation of the final life-cycle cost and the final reliability and maintainability values of
the system in question. The life-cycle cost includes the disposal action cost or
income. The final reliability and maintainability values are calculated for the buyer
of the used system and for use in purchasing of similar systems.

3.3.4

DOCUMENTS

AND

TOOLS

FOR

RELIABILITY MANAGEMENT

Reliability management makes use of various kinds of documents and tools. Some
of the examples of documents used by the reliability management are the in-house
reliability manual; national and international standards, specification or publications;
documents explaining policy and procedures; instructions and plans; reports; and
drawings and contract documents. Similarly, the reliability management makes use
of various kinds of management tools. Some of them are value engineering, configuration management and critical path method or program evaluation and review
technique. Some of these items are discussed in the following.
The reliability manual is the backbone of any reliability organization. Its existence is vital for any organization irrespective of its size. An effective reliability
manual covers topics such as:

Organizational setup and individual responsibilities

Company reliability policy and procedures
Applicable reliability models, reliability prediction techniques, etc.
Procedures for failure data collection and analysis
Methods to be used in equipment reliability testing

Value engineering is concerned with the orderly application of established methods to identify the functions of a product or a service, and also with providing those
identified functions at the minimum cost. Historical evidence indicates that the
application of value engineering concept has returned between $15 to $30 for each
dollar spent [23].
Hundreds of engineering changes are associated with the development of a
complex system. Configuration management assures both the customer and the

1999 by CRC Press LLC

manufacturer that the resulting system hardware and software fully meet the contract
specification. Configuration management is defined as the management of technical
requirements which define the engineering system as well as changes thereto.
Both the critical path method (CPM) and the program evaluation and review
technique (PERT) were developed in the late 1950s to manage engineering projects
effectively. Nowadays, both these methods are used quite frequently. The application
of both these techniques have been found effective in areas such as planning and
smoothing resource utilization, obtaining cost estimates for proposed projects, scheduling projects, and evaluating existing schedules and cost vs. budget, as the work
on the project progresses.

3.3.5

RELIABILITY AUDITING AND PITFALLS

PROGRAM MANAGEMENT

IN

RELIABILITY

The audits are important in reliability work to indicate its strong, weak, and acceptable areas. There are various guidelines in conducting reliability audits. Some of
them are maintain auditing schedule, record the audit results, conduct audits without
prior announcements, perform audits using checklists, choose unbiased persons for
auditing, take follow-up actions, and avoid appointing someone permanently for the
auditing purpose.
Reliability auditing has many advantages. For example, it helps problem areas
to surface, determines if the customer specifications are met satisfactorily, helps to
reduce the complaints received from customers, is useful in predicting the reaction
of the buyer toward reliability, and finds out if the company reliability objectives
are fulfilled.
Pitfalls in reliability program management are the causes for many reliability
program uncertainties and problems. These pitfalls are associated with areas such
as follows [24]:

reliability organization
reliability testing
programming
manufacturing

Two important examples of reliability organization pitfalls are having several

organizational tiers and a number of individuals with authority to make commitments
without internal dialog and coordination. Furthermore, an important example of the
reliability testing pitfalls is the delayed start of the reliability demonstration test.
This kind of delay may compound the problem of incorporating reliability-related
changes into deliverable product to the customer. Programming pitfalls are also
important and must be given full attention. One example of these pitfalls associated
with program requirements is the assumption that each concerned individual with
the program understands the reliability requirements.
There are also pitfalls that occur during the manufacturing phase. For example,
when the parts acquisition lead time is incompatible with the system manufacturing

1999 by CRC Press LLC

schedule, the parts buyers or even the design engineers will (under certain circumstances) authorize substitute parts without paying much attention to their effect on
reliability. These kinds of pitfalls can only be avoided by an effective reliability
management team.

3.3.6

RELIABILITY AND MAINTAINABILITY ENGINEERING

DEPARTMENTS AND THEIR RESPONSIBILITIES

Probably the most crucial factor in the success or failure of reliability and maintainability programs in an organization is the attitude and the thinking philosophy of top
level management toward reliability and maintainability. More clearly, without the
support of the top management, reliability and maintainability programs in a company
will not be effective at all. Once the high-up managements positive and effective
attitude is generated, then appropriate reliability and maintainability organizations
are to be formed. Two distinct departments, i.e., reliability and maintainability, will
be dictated by the need of the company in question. Sometimes the reliability and
maintainability functions may be combined within a single department or assigned
as the responsibility of the quality assurance department. As to the reporting structure
of the reliability and maintainability departments, there is no iron-clad rule as to how
it should be. However, in order to command respect for these programs their heads
should be high enough on the organizational ladder with necessary authority [25].
Reliability Engineering Department Responsibilities
A reliability engineering department may have various kinds of responsibilities.
However, the major ones are as follows:

Establishing reliability policy, plans and procedures

Reliability allocation
Reliability prediction
Specification and design reviews with respect to reliability
Reliability growth monitoring
Providing reliability related inputs to design specifications and proposals
Reliability demonstration
Training reliability manpower and performing reliability-related research
and development work
Monitoring the reliability activities of subcontractors, if any
Auditing the reliability activities
Failure data collection and reporting
Failure data analysis
Consulting

Maintainability Engineering Department Responsibilities

Similarly, as for the reliability engineering department, the maintainability engineering department has many responsibilities as well. Some of them are as follows:

1999 by CRC Press LLC

3.3.7

Establishing maintainability policies and procedures

Training and staffing maintainability manpower
Developing plans for and conducting maintainability demonstration
Collecting and analyzing failure data
Taking part in design and specification reviews
Maintainability assessment
Maintainability allocation
Conducting maintainability tradeoff studies
Monitoring maintainability activities of subcontractors
Providing consulting services to others
Documenting maintainability related information

RELIABILITY MANPOWER

The reliability group is composed of people who have specialties in various branches
of reliability engineering. However, according to some experts [21], the personnel
involved in reliability work should have experience and background in areas such
as quality control, probability theory, project management, system engineering,
operations research, environmental testing, components design, manufacturing methods, data analysis and collection, developing specifications, and test planning. Furthermore, it is to be noted that professions such as engineering physics, mechanical
engineering, statistics, metallurgy, and electronics engineering are related to reliability. Their relationships are briefly discussed in Reference 26. Some areas related
to reliability personnel are described below.
Rules for Reliability Professionals
These rules are useful for the effective implementation of reliability programs in an
organization by the reliability manager or engineer as well as the concerned professional who can expect some advancement in their careers [27]. Some of these rules
are presented below.
Speak as briefly as possible. In other words, get to the core of the problem
by presenting a summary and recommendations as soon as possible.
When talking to others, make sure that you converse in a language they
understand. Avoid using terms and abbreviations with which they are
uncomfortable.
Avoid using statistical jargon as much as possible especially when dealing
with top-level management.
Make sure that all reliability functions are included in the reliability
program plan.
Contribute to the solution by getting directly involved with the problem
investigation effort and work out a group of corrective action options.
Develop for yourself how to be just as comfortable reporting successes
as reporting failures with respect to reliability.
Avoid, whenever it is possible, paralysis of analysis.

1999 by CRC Press LLC

 Develop in yourself the belief that the reliability engineer is a pessimist,

the designer is an optimist, and the combination is a success.
Avoid assuming that the reliability function is responsible for accomplishing reliability goals.
Learn to believe that new reliability techniques are not the final result, but
are only the processes that allow one to achieve product and business goals.
Tasks of a Reliability Engineer
During planning, design and development, manufacturing, and improvement phases
of a system, the reliability engineer performs various tasks. Many of these tasks are
as follows:
Performing analysis of a proposed design
Securing resources for an effective reliability program during the planning
stage
Analyzing customer complaints with reliability
Investigating field failures
Running tests on the system, subsystems, and parts
Examining difficulties associated with maintaining a system
Developing tests on the system, subsystems, and components
Budgeting the tolerable system failure down to the component level
Developing a reliability program plan
Keeping track of parts
Determining reliability of alternative designs
Providing information to designers regarding improving the ease of maintenance and system life
Providing relevant reliability information to management
Ensuring that the relevant information flows to all concerned in an effective manner
Ensuring that the end results of all reliability tasks are organized systematically and effectively
Monitoring subcontractors reliability performance
Participating in developing requirements for new systems
Participating in evaluating requests for proposals
Ensuring that enough consideration is given to reliability in new contracts
Developing reliability prediction models and techniques
Participating in design reviews
Training of Reliability Manpower
This is an important area concerning every reliability manager because for an
effective performance training, developing manpower is a never-ending process.
Need for training has to be planned well in advance in order to avoid any difficulty
at the moment of need. Some of the major objectives [28] of reliability training
indoctrination are as follows:

1999 by CRC Press LLC

 Make all personnel in manufacturing, engineering, reliability, purchasing,

quality control, etc. aware of the specific effect of tasks performed by
them on product reliability;
Assure that the persons involved in reliability work are in a position to
carry out their tasks in an effective and efficient manner;
Enhance reliability consciousness in people associated with the project
under consideration;
Focus on those activity areas deemed to be specifically amenable to a
reliability improvement effort.
There are many useful guidelines with respect to training of reliability manpower
including identifing the weak areas, forecasting the manpower needed, planning in
advance to meet future needs of the organization, making presentations on the latest
reliability state of the art, sponsoring personnel involved with reliability work to
conferences, courses, and seminars, and joining professional reliability societies [27].
Difficulties in Motivating Design Engineers for Reliability
In recent times, during the design phase, reliability has been receiving wider attention
than ever before. However, there still remain several difficulties in motivating design
engineers to accept reliability as one of the design parameters. Some of these
difficulties are as follows [29]:

Understanding product users

Requirement for reliability specifications
Choosing the right person for the job
Understanding an engineering design recipe
Educating educators (i.e., design engineering professors) on reliability
Assisting instead of controlling
Comprehending business considerations

3.4 PROBLEMS
1. Write an essay on important design failures that have occurred over the
past 50 years.
2. List at least five of the most important common reasons for product
failures.
3. Describe a typical design process practiced in the industrial sector.
4. What are the advantages of conducting design review?
5. Discuss the following design reviews:
Preliminary design review
Intermediate design review
Critical design review
6. Describe the functions of the following two professionals:
Design engineer
Reliability engineer

1999 by CRC Press LLC

7. What are the design process aspects that have significant impact on product reliability?
8. List the military specification (i.e., MIL-R-27542) guidelines for developing reliability programs.
9. Discuss the documents used by the reliability management.
10. Briefly discuss the tools used by the reliability management.
11. Describe the reliability program management related pitfalls.
12. What are the typical responsibilities of a reliability engineering department?
13. List the difficulties experienced in motivating design engineers to accept
reliability as one of the design parameters.

3.5 REFERENCES
1. Shigley, J.D. and Mitchell, L.D., Mechanical Engineering Design, McGraw-Hill, New
York, 1983.
2. Farr, M., Design Management, Cambridge University Press, London, 1955.
3. Dhillon, B.S., Engineering Design: A Modern Approach, Richard D. Irwin, Chicago,
IL, 1996.
4. Austin-Davis, W., Reliability management: A challenge, IEEE Trans. Reliability,
R-12, 6-9, 1963.
5. Dhillon, B.S. and Reiche, H., Reliability and Maintainability Management, Van
Nostrand Reinhold Company, New York, 1985.
6. Dhillon, B.S., Reliability and Quality Control: Bibliography on General and Specialized Areas, Beta Publishers, Gloucester, Ontario, 1992.
7. Hurricks, P.L., Handbook of Electromechanical Product Design, Longman Scientific
and Technical, Longman Group UK Limited, London, 1994.
8. Walton, J.W., Engineering Design, West Publishing Company, New York, 1991.
9. Elsayed, E.A., Reliability Engineering, Addison Wesley Longman, Reading, MA,
1996.
10. Dieter, G.E., Engineering Design, McGraw-Hill, New York, 1983.
11. Vidosic, J.P., Elements of Design Engineering, The Ronald Press Co., New York,
1969.
12. Hill, P.H., The Science of Engineering Design, Holt, Rhinehart, and Winston, New
York, 1970.
13. Pecht, M., Ed., Product Reliability, Maintainability, and Supportability Handbook,
CRC Press, Boca Raton, FL, 1995.
14. AMCP 706-196, Engineering Design Handbook, Part II: Design for Reliability, 1976.
Prepared by Headquarters, U.S. Army Material Command, Alexandria, VA.
15. Carter, A.D.S., Mechanical Reliability, MacMillan, London, 1986.
16. Roche, J.G., Design implications of product liability, Int. J. Quality and Reliability
Manage., 2, 1988.
17. Abbot, H., Safer by Design, The Design Council, London, 1987.
18. Dhillon, B.S., Engineering reliability management, IEEE J. Selected Areas Comm.,
4, 1015-1020, 1986.
19. Heyel, C., The Encyclopedia of Management, Van Nostrand Reinhold Company, New
York, 1979.
20. Finch, W.L., Reliability: A technical management challenge, Proc. Am. Soc. Quality
Control Annu. Conf., 851-856, 1981.

1999 by CRC Press LLC

21. Karger, D.W. and Murdick, R.G., Managing Engineering and Research, Industrial
Press, New York, 1980.
22. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Eds., Handbook of Reliability
Engineering and Management, McGraw-Hill, New York, 1996.
23. Demarle, D.J. and Shillito, M.L., Value Engineering, in Handbook of Industrial
Engineering, Salvendy, Ed., John Wiley & Sons, New York, 1982, 7.3.17.3.20.
24. Thomas, E.F., Pitfalls in reliability program management, Proc. Annu. Reliability
Maintainability Symp., 369-373, 1976.
25. Jennings, J.A., Reliability of aerospace electrical equipment: How much does it cost?,
IEEE Trans. Aerospace, 1, 38-40, 1963.
26. Reliability Engineering: A Career for You, Booklet published jointly by the American
Society for Quality Control (ASQC) and the Institute of Electrical and Electronics
Engineers (IEEE). Available from the Director of Technical Programs, ASQC, 230
West Wells St., Milwakee, WI 53203.
27. Ekings, J.D., Ten rules for the reliability professional, Proc. Annu. Am. Soc. Quality
Control Conf., 343-351, 1982.
28. McClure, J.Y., Organizing for Reliability and Quality Control, in Reliability Handbook, Ireson, G., Ed., McGraw-Hill, New York, 16-37, 1966.
29. Bajaria, H.J., Motivating design engineers for reliability, Proc. Annu. Conf. Am. Soc.
Quality Control, 767-773, 1979.

1999 by CRC Press LLC

Failure Data Collection

and Analysis

4.1 INTRODUCTION
Failure data are the backbone of reliability studies because they provide invaluable
information to concerned professionals such as reliability engineers, design engineers, and managers. In fact, the failure data of a product are the final proof of
reliability related efforts expended during its design and manufacture. It would be
impossible to have an effective reliability program without the collection, analysis,
and use of information acquired through the testing and operation of products used
in industrial, military, and consumer sectors.
It may be stated, more specifically, that the basic purpose of developing a formal
system of collecting, analyzing, and retrieving information acquired from past experiences is to allow design and development of better and more reliable products
without repeating the previous efforts concerning research, design, and testing to
achieve the current product reliability. Thus, the fundamental goal of a failure data
collection and analysis system is to convert the relevant information accumulated
in various sources into an effectively organized form so that it can efficiently be
used by individuals with confidence in conducting their assigned reliability related
tasks. There are many different ways to store and present this organized form of
information to the end users. As each such way has advantages and disadvantages,
a careful consideration must be given in their selection.
In particular, the nature of the products and their intended applications are
important factors in deciding the extensiveness, form, etc. of a failure data collection
and analysis system. For example, a company stamping out stainless steel flatware
will require a simple data system as opposed to a complex data system for a company
manufacturing aircraft. All in all, prior to deciding the size, type, etc. of a failure
data system, consider factors such as individuals requiring information, information
required by the individuals to meet their needs, and frequency of their need.
An extensive list of publications on failure data is given in Reference 1. This
chapter presents different aspects of failure data collection and analysis.

4.2

FAILURE DATA USES

There are many areas for the uses of failure data including conceptual design,
preliminary design, test planning, design reviews, manufacturing, quality assurance
and inspection planning, logistic support planning inventory management, procurement, top management budgeting, and field service planning [2]. Nonetheless, some
of the more specific uses of failure data are estimating hazard rate of an item, performing life cycle cost studies, making decisions regarding introduction of redundancy,
predicting an items reliability/availability, performing equipment replacement studies,

1999 by CRC Press LLC

recommending design changes for improving the products reliability, performing

cost vs. reliability studies, determining the maintenance needs of a new item, and
conducting preventive maintenance studies [35].

4.3 FAILURE DATA COLLECTION SOURCES

IN EQUIPMENT LIFE CYCLE AND QUALITY
CONTROL DATA
There are many sources of collecting failure related data during an equipment life
cycle. According to Reference 6, eight of these sources are as follows:
1.
2.
3.
4.
5.
6.
7.
8.

Warranty claims
Previous experience with similar or identical equipment
Repair facility records
Factory acceptance testing
Records generated during the development phase
Customers failure reporting systems
Tests: field demonstration, environmental qualification, and field installation
Inspection records generated by quality control/manufacturing groups

In order to maintain the required quality standards, the quality control groups
within organizations regularly perform tests/inspections on products/equipment. As
the result of such actions, valuable data are generated that can be very useful in
design reviews, part vendor selection, providing feedback to designers on the problems related to production, and so on. The important components of the quality
control data are incoming inspection and test results, quality audit records and final
test results, in-process quality control data, results of machine and process capability
studies, and calibration records of measuring instruments [2].

4.4 FAILURE REPORTING AND DOCUMENTATION

SYSTEM DESIGN GUIDELINES AND FAILURE DATA
COLLECTION FORMS
A failure reporting and documentation system could be very effective if a careful
consideration is given during its design. Some of the associated design guidelines
are simplicity and clarity of the failure reporting form, effective and efficient documentation of the information on failures, clarity in terms and definitions, efficiency
in conducting recorded failure data analysis, user involvements, flexibility, cost
effectiveness, feasibility of prerecording any static information on the data form,
elimination of the needs for memorizing codes, and feedback of the analyzed data
to involved personnel in an effective manner [7].
The backbone of a failure data system is the failure data collection form. Such
forms must be designed so that they require minimum effort to collect failure data
and provide maximum benefits. Also, these forms must be tailored to meet the

1999 by CRC Press LLC

TABLE 4.1
Selective Standard Documents for Failure Data Collection
Document
classification No.

No.
1

MIL-STD-2155

IEC 706 PT3

MODUKDSTAN 00-44

IEEE 500

IEC 362

Document title

Developed by:

Failure Reporting, Analysis and Corrective

Action System (FRACAS)
Guide on Maintainability of Equipment:
Verification and Collection, Analysis and
Presentation of Data
Reliability and Maintainability Data
Collection and Classification

U.S. Department of
Defense
International
Electrotechnical
Commission (IEC)
British Defense
Standards (U.K.
Department of
Defense)
Institute of
Electrical and
Electronics
Engineers (IEEE)

Guide to the Collection and Presentation of

Electrical, Electronic, Sensing Component,
and Mechanical Equipment Reliability
Data for Nuclear Power Generating
Stations
Guide for the Collection of Reliability,
Availability, and Maintainability Data from
Field Performance of Electronic Items
A Reliability Guide to Failure Reporting,
Analysis, and Corrective Action Systems.

International
Electrotechnical
Commission (IEC)
American Society
for Quality Control
(ASQC)

requirements of a specific organization or program effectively. Nonetheless, such

forms are generally designed to include information such as form number, item
description, location of the hardware, serial number of the failed item and manufacturers name, description of the failure, form completion date, dates on which failure
occurred, was detected and corrected, operating hours from the previous failure,
time taken for repair, repair cost, serial number and manufacturer of the replaced
part, repair-persons name and appropriate signatures [8].
Table 4.1 presents useful standard documents for failure data collection.

4.5 EXTERNAL FAILURE DATA SOURCES

There are a large number of organizations, data banks, and documents to obtain
failure data related information. Some of these are as follows [2, 5, 9]:
Organizations
Government Industry Data Exchange Program (GIDEP) [10]. It is
probably the most comprehensive source for obtaining data and is operated and funded by the U.S. government. A company must formally

1999 by CRC Press LLC

become a full or partial participating member of the GIDEP before

making use of it. This means that the company must also submit its
own data to GIDEP in order to receive data from it. The system is
managed by the GIDEP Operations Center, Fleet Missile Systems,
Analysis and Evaluation Group, Department of Defense, Corona, California 91720.
National Technical Information Service (NTIS). This is also a good
source for obtaining various types of failure data and the service is
managed by the U.S. Department of Commerce. The address of the
NTIS is 5285 Port Royal Road, Springfield, Virginia 22161.
Reliability Analysis Center (RAC). RAC provides various types of
information concerning reliability and is managed by the U.S. Air Force
(USAF). In particular, the Nonelectronic Parts Reliability Data (NPRD)
reports are released periodically by the center. The address of the RAC
is Rome Air Development Center (RADC), Griffiss Air Force Base,
Department of Defense, Rome, New York 13441.
Defense Technical Information Center. This center is a useful source
for obtaining various types of failure data related information, particularly concerning the defense equipment. The address of the center is
Defense Logistics Agency, Cameron Station, Alexandria, Virginia 22314.
Parts Reliability Information Center (PRINCE). This center is quite
useful to obtain various types of data related to space systems. The
address of the center is Reliability Office, George C. Marshall Space
Flight Center, National Aeronautics and Space Administration
(NASA), Huntsville, Alabama 35812.
Institute of Electrical and Electronics Engineers (IEEE). This is the
largest professional body in the world and provides failure data concerning various electrical related items. The address of this organization
is 345 East 47th Street, New York, New York 10017.
National Electric Reliability Council (NERC). NERC publishes various types of failure data collected annually from the U.S. power plants.
This organization is located in New York City.
Insurance Information Institute (III). This organization provides
data on disaster impacts and related matters. III is located at 110
William Street, New York, New York 10017.
Data Banks
Nuclear Plant Reliability Data System (NPRDS). This system provides various types of failure data on equipment used in nuclear power
generation. NPRDS is managed by the South West Research Institute,
San Antonio, Texas.
Equipment Reliability Information System (ERIS). This system
provides failure data on equipment used in electric power generation.
The system is managed by the Canadian Electrical Association (CEA),
Montreal, Quebec.
SYREL: Reliability Data Bank. This data bank provides failure data
on equipment used in power generation. The data bank is managed by

1999 by CRC Press LLC

the Systems Reliability Service, Safety and Reliability Directorate,

United Kingdom Atomic Energy Authority (UKAEA), Wigshaw Lane,
Culcheth, Warrington, Lancashire WA3 4NE, England.
ESA Electronic Component Data Bank. This data bank provides data
on various types of electronic parts, especially used in space environment. Its address is Space Documentation Service, European Space
Agency (ESA), Via Galileo, 00044 Frascati, Italy.
Documents
MIL-HDBK-217, Reliability Prediction of Electronic Equipment,
Department of Defense, Washington, D.C.
Component Parts Failure Data Compendium, Engineering Dept., Electronic Industries Association, 11 W. 42nd St., New York, New York
10036.
RADC-TR-85-194, RADC Nonelectronic Reliability Notebook, Reliability Analysis Center, Rome Air Development Center (RADC), Griffiss Air Force Base, Rome, New York 13441-5700, 1985.
IEEE-STD-500-1977, IEEE Nuclear Reliability Data Manual, John
Wiley & Sons, New York, 1977.
WASH-1400 (NUREG 75/014), Appendix 3 and 4, Reactor Safety
Study: An Assessment of Accident Risks in U.S. Commercial Nuclear
Power Plants, U.S. Nuclear Regulatory Commission, Washington,
D.C., 1975.
Dhillon, B.S., Mechanical Reliability: Theory, Models, and Applications,
American Institute of Aeronautics and Astronautics, Washington, D.C.,
1988 (Chapter 9 contains failure data on various mechanical parts).
RAC EMD 1 and 2, Electronic Equipment Reliability Data, Reliability
Analysis Center (RAC), Rome Air Development Center, Griffiss Air
Force Base, Rome, New York.
NUREG/CR-1278, Handbook of Human Reliability Analysis with
Emphasis on Nuclear Power Plant Applications, U.S. Nuclear Regulatory Commission, Washington, D.C.

4.6 FAILURE DATA FOR SELECTIVE ITEMS AND TASKS

For many decades, failure data on engineering items and various tasks performed
by humans have been collected and analyzed. In order to demonstrate the type of
analyzed and available failure data, Tables 4.2 [11] and 4.3 [5, 1215] present failure
rates for some selected electronic and mechanical items, respectively, and Table 4.4
[1619] contains human error data for a few selective tasks.

4.7 HAZARD PLOTTING METHOD

This is a powerful graphical approach to perform failure data analysis. The method
is popular among engineers because of its advantages such as fits the data to a straight
line, results obtained through plotting are convincing, easy to understand, straightforward to visualize the theoretical distribution that fits the field data, and equally

1999 by CRC Press LLC

TABLE 4.2
Failure Rates for Some Electronic Items
No.

Item description

Failure rate
(failures/106 h)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15

Neon lamp
Weld connection
Fuse (cartridge class H or instrument type)
Fiber optic cable (single fiber types only)
Solid state relay (commercial grade)
Vibrator (MIL-V-95): 60-cycle
Terminal block connection
Microwave ferrite devices: isolators and circulatory (100 watts)
Solid state relay (military specification grade)
Single fiber optic connector
Crimp connection
Spring contact connection
Hybrid relay (commercial grade)
Vibrator (MIL-V-95): 120-cycle
Fuse: current limiter type (aircraft)

0.2
0.000015a
0.010a
0.1 (per fiber km)
0.0551a
15
0.062a
0.10a
0.029a
0.1
0.00026a
0.17a
0.0551a
20
0.01a

Use environment: ground, benign.

applicable to analyze data with complete and incomplete observations [2024].

Another important advantage of the approach is that it indicates if the given data set
belongs to the tried distribution. If so, then it estimates the associated parameters.
The complete observations may be described as data in which the times to failure
of all units of a sample are known. On the other hand, when the running times of
unfailed units and the failure times of failed units are given, the data are called
incomplete or censored data/observations. In turn, the running times of unfailed units
are called the censoring times. In the event of having different censoring times of
the unfailed units, the data are known as the multi-censored data. In contrast, when
the censoring times of all the unfailed units are the same and greater than the failed
units failure times, the data are said to be singly censored. Nonetheless, the multicensored data occur due to factors such as follows:
Removing items/units or terminating their use prior to their failure.
Some extraneous causes were responsible for the units/items failure.
Collecting data from active units/items.

4.7.1

HAZARD PLOTTING MATHEMATICAL THEORY

The hazard function of a statistical distribution is the basis for this theory. Thus, the
hazard rate is defined by

1999 by CRC Press LLC

TABLE 4.3
Failure Rates for Some Mechanical Items
No.

Item description

Failure rate
(failures/106 h)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

Pivot
Heat exchanger
Relief valve
Conveyor belt (heavy load)
Heavy duty ball bearing
Piston
Washer
Nut or bolt
Rotating seal
Pipe
Spring (torsion)
Gearbox (reduction)
Bellow
Crankshaft (general)
Cylinder
Knob (general)
Flexible coupling
Axle (general)
Hair spring
Slip ring (general)

1
6.11244.3
0.510
20140
20
1
0.5
0.02
4.4
0.2
14.296a
18.755b
5
33.292b
0.1
2.081a
9.987b
9.539b
1
0.667a

a
b

Use environment: ground, fixed.

Use environment: ground, mobile.

TABLE 4.4
Human Error Rates for Some Tasks
No.

Error description

Error rate

1
2
3
4
5
6
7

Improper servicing/reassembly by the maintenance person

Closing valve incorrectly
Incorrect adjustment by the maintenance person
Procedural error in reading given instructions
Reading gauge incorrectly
Misunderstanding of requirements by the operator
Installation error

0.0153a
1800b
0.0134a
64500b
5000a
0.0076a
0.0401a

a
b

Errors per plant month (for pressurized water reactors).

Errors per million operations.

1999 by CRC Press LLC

z (t ) =

f (t )
R( t )

f (t )
=
1 F( t )

(4.1)

where
z(t)
R(t)
f (t)
F(t)

is
is
is
is

the
the
the
the

hazard rate or instantaneous failure rate.

reliability at time t.
distribution probability density function.
cumulative distribution function.

The cumulative distribution function is defined by

t

F (t ) = f (t ) d t

(4.2)

The cumulative hazard function, zc (t), is defined by

t

zc (t ) = z (t ) d t

(4.3)

Equations (4.1) through (4.3) are used to obtain a straight line expression for
estimating graphically the parameters of failure distributions such as the following
[20, 25]:
Exponential Distribution
This is a single parameter distribution and its probability density function is defined
by
f (t ) = e t

t0

(4.4)

where
is the parameter known as the constant failure rate.
t is time.
Inserting Equation (4.4) into Equation (4.2) yields
F (t ) = 1 e t

1999 by CRC Press LLC

(4.5)

FIGURE 4.1 A hypothetical plot of time to failure, t, against cumulative hazard, zc.

By substituting Equations (4.4) and (4.5) into Equation (4.1), we get

z (t ) =

(4.6)

By inserting Equation (4.6) into Equation (4.3), we obtain the following expression
for the cumulative hazard function:
zc (t ) = t

(4.7)

By letting = 1/, where is the mean time to failure and rearranging Equation (4.7)
to express the time to failure, t, as a function of zc, we get
t (z c ) = z c

(4.8)

Equation (4.8) is the equation of a straight line passing through the origin. In order
to estimate the value of , the time to failure t against the cumulative hazard zc is
plotted. If the plotted field data points fall roughly on a straight line, then a line is
drawn to estimate . At zc = 1 on this plot, the corresponding value of t is the
estimated value of . The value of the also equals the slope of the straight line.
When the plotted data points do not approximately fall along a straight line, this
indicates that the failure data do not belong to the exponential distribution and
another distribution should be tried. A hypothetical plot of Equation (4.8) is shown
in Figure 4.1.
Bathtub Hazard Rate Distribution
This distribution can represent a wide range of failure data representing increasing,
decreasing, and bathtub hazard rates. The probability density function of the distribution is defined by [26, 27]:
f (t ) = ( t )

for , > 0, and t 0

1999 by CRC Press LLC

e( ) ( t ) 1

(4.9)

where
is the scale parameter.
is the shape parameter.
t is time.
Substituting Equation (4.9) into Equation (4.2) yields
F (t ) = 1 e

{e( t ) 1}

(4.10)

The following are the special cases of this distribution:

A t = 1: extreme value distribution
= 0.5: bathtub hazard rate curve
By inserting Equations (4.9) and (4.10) into Equation (4.1), we obtain
z(t ) = ( t )

t
e( ) .

(4.11)

Substituting Equation (4.11) into Equation (4.3) yields

t
z c ( t ) = e( ) 1

(4.12)

Rearranging (4.12) and twice taking natural logarithms of (zc + 1), we get
ln t =

1
ln x ln

(4.13)

where
x ln (zc + 1)
The plot of ln t against ln x gives a straight line; thus, the slope of the line and its
intercept are equal to 1/ and (-ln), respectively. If the plotted field data points fall
roughly on a straight line, then a line can be drawn to estimate the values of and .
On the other hand, when the plotted data do not approximately fall along a straight
line, this obviously indicates that the failure data under consideration do not belong
to the bathtub hazard rate distribution; thus, try another distribution.
Weibull Distribution
This is a widely used distribution in reliability studies and it can represent a wide
range of failure data. The probability density function of the distribution is defined by

1999 by CRC Press LLC

f (t ) =

1 ( t )
t e
,

for t 0

(4.14)

where
t is time.
is the scale parameter.
is the shape parameter.
Inserting Equation (4.14) into Equation (4.2) yields

t
F (t ) = 1 e ( )

(4.15)

At = 1 and 2, the Weibull distribution becomes exponential and Rayleigh distributions, respectively.
By substituting Equations (4.14) and (4.15) into Equation (4.1), we get
z (t ) =

1
t

(4.16)

Inserting Equation (4.16) into Equation (4.3) leads to

z c ( t ) = ( t )

(4.17)

In order to express the time to failure t as a function of zc, we rearrange Equation (4.17)
as follows:
t (z c ) = (z c )

(4.18)

Taking natural logarithms of Equation (4.18) results in

ln t =

1
ln z c + ln

(4.19)

In Equation (4.19), ln t against ln zc plots linearly or in a straight line with slope 1/

and the intercept ln . Also, at zc = 1, the corresponding time ln t equals ln .
Nonetheless, graphically, this fact is used to estimate the value of the scale parameter .
Nonetheless, from the graphical standpoint, if the plotted field data points fall
approximately along a straight line, then a line is drawn to estimate the values of
the parameters and . In the event of such points not falling along the straight
line, try another distribution.
Similarly, the straight line expressions for other failure distributions can be
obtained [20, 25].

1999 by CRC Press LLC

TABLE 4.5
Failure and Running (Censoring) Times
for 16 Identical Units
Unit No.

Failure and running (censoring) times (h)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16

14,092a
3,973
2,037
360
628
3,279
2,965
837a
79
13,890a
184
13,906a
499
134
10,012
3,439a

running (censoring) time

Example 4.1
Assume that 16 identical units, at time t = 0, were put on test and Table 4.5 presents
failure and running times of these units. Determine the statistical distribution fit to
the given data and estimate values for its parameters by using the hazard plotting
steps described below.
Hazard Plotting Steps
The following eight steps are used to construct a hazard plot:
1. Order the data containing m times from smallest to largest without making
any distinction whether these data are running (censoring) or failure times
of units. Use an asterisk (*) or other means to identify the running or
censoring times. In the event of having some running and failure times
equal, mix such times well on the ordered list of smallest to largest times.
For Example 4.1 data given in Table 4.5, in Column 1 of Table 4.6, the
failure and running times of 16 units are ranked from smallest to largest
with the censoring times identified by a superscript a.
2. Label the ordered times with reverse ranks. For example, label the first
time with m, the second with (m-1), the third with (m-2), , and the mth
with 1.
Column 2 of Table 4.6 presents reverse rank labels for the 16 ordered
times of Column 1.

1999 by CRC Press LLC

TABLE 4.6
Ordered Times and Hazard Values
Col. 1
Ordered times t in hours
(smallest to largest)

Col. 2
Reverse rank
labels

Col. 3
Hazard value
(100/Col. 2 value)

Col. 4
Cumulative
hazard (zc )

79
134
184
360
499
628
837a
2,037
2,965
3,279
3,439a
3,973
10,012
13,890a
13,906a
14,092a

16
15
14
13
12
11
10
9
8
7
6
5
4
3
2
1

6.25
6.67
7.14
7.69
8.33
9.09

11.11
12.5
14.29

20
25

6.25
12.92
20.06
27.75
36.08
45.17

56.28
68.78
83.07

103.07
128.07

running (censoring) time

3. Compute a hazard value for each failure using 100/m, 100/(m-1), 100/(m-2),
etc. The hazard value may be described as the conditional probability of
failure time. In other words, it is the observed instantaneous failure rate
at a certain failure time.
For example, as per Columns 1 and 2 of Table 4.6, the first failure
occurred after 79 h out of 16 units put on test at time t = 0. Thus, the
hazard value is
100
= 6.25%
16
After 79 h, only 15 units were operating and 1 failed after 134 h. Similarly,
the hazard value is
100
= 6.67%
15
In a similar manner, the hazard values at other failure times were computed
as shown in Column 3 of Table 4.6. These values are simply 100 divided
by the corresponding number of Column 2.
4. Obtain the cumulative hazard value for each failure by adding its hazard
value to the cumulative hazard value of the preceding failure. For example,

1999 by CRC Press LLC

for the first failure at 79 h (as shown in Column 1 of Table 4.6), its hazard
value is 6.25% and there is no cumulative hazard value of the preceding
failure because this is the first failure. Thus, in this case, the cumulative
hazard simply is 6.25%. However, for the failure at 2037 h, the cumulative
hazard is 11.11 (hazard value) + 45.17 (cumulative hazard value of the
preceding failure) = 56.28. Similarly, Column 4 of Table 4.6 presents
cumulative hazard values for other failures.
It is to be noted that the cumulative hazard values may exceed 100%
and have no physical interpretation.
5. Choose a statistical distribution and prepare times to failure and corresponding cumulative hazard data for use in the selected distribution to
construct a hazard plot.
For our case, we select the Weibull distribution; thus, we take natural
logarithms of the times to failure and of corresponding cumulative hazard
values given in Table 4.6. The processed data for ln t and ln zc are
presented in Table 4.7.
6. Plot each time to failure against its corresponding cumulative hazard value
on a graph paper. Even though the running times are not plotted, they do
determine the plotting points of the times to failure through the reverse
ranks.
In our case, we plotted ln t against ln zc as shown in Figure 4.2 using
Table 4.7 values.

TABLE 4.7
Processed Data for the Weibull Distribution
lnt
(using values for t
from Col. 1Table 4.6)
4.37
4.90
5.22
5.89
6.21
6.44

7.62
8.00
8.10

8.29
9.21

1999 by CRC Press LLC

lnzc
(using values for zc
from Col. 4Table 4.6)
1.83
2.56
3.00
3.32
3.59
3.81

4.03
4.23
4.42

4.64
4.85

FIGURE 4.2 Failure data fit to a Weibull distribution.

7. Determine if the plotted data points roughly follow a straight line. If they
do, it is reasonable to conclude that the selected distribution adequately
fits the data and draw a best fit straight line. If they do not, try plotting
the given data for another statistical distribution.
In our case, the given data roughly follow the straight line; thus, it is
reasonable to conclude that the data are Weibull distributed.
8. Estimate the values of the distribution parameters using the hazard plot.
In our case, using Figure 4.2, we have = 0.58, and = 1.82 h.
Finally, it is emphasized that the results obtained using this method are valid only
when the times to failure of unfailed units are statistically independent of their
running or censoring times, if such units were used until their failure.

4.8 UNDERLYING DISTRIBUTION

DETERMINATION TESTS
Over the years, there have been many tests developed to determine the underlying
statistical distribution of times to failure associated with engineering items when a
set of data are given [5, 2837]. Once the statistical distribution associated with the
given failure data of an item is identified, the technique such as maximum likelihood
estimation (MLE) can be used to estimate the values of distribution parameters. This
section describes three underlying distribution determination tests: (1) Bartlett,
(2) general exponential, and (3) Kolmogorov-Smirnov.

4.8.1

BARTLETT TEST

Often in reliability analysis, it is assumed that an items times to failure are exponentially distributed. The Bartlett test is a useful method to verify this assumption
from a given sample of failure data. The Bartlett test statistic is defined [28] as

1999 by CRC Press LLC

W
Sb m = 12 m 2 ln F

(6 m + m + 1)

(4.20)

ln t

(4.21)

i =1

F=

1
m

(4.22)

i =1

where
m is the total number of times to failures in the sample.
t i is the ith time to failure.
In order for this test to discriminate effectively, a sample of at least 20 times to
failure is required. If the times to failures follow the exponential distribution, then
Sbm is distributed as chi-square with (m-1) degrees of freedom. Thus, a two-tailed
chi-square method (criterion) is used [30].
Example 4.2
After testing a sample of identical electronic devices, 25 times to failure as shown
in Table 4.8 were observed. Determine if these data belong to an exponential distribution by applying the Bartlett test.
By inserting the given data into Equation (4.22) we get
F=

3440
25

= 137.6 h
Similarly, using Equation (4.21) and the given data we have
W = 110.6655

TABLE 4.8
Times to Failure (in hours)
of Electronic Devices
5
6
21
18
33

1999 by CRC Press LLC

34
47
44
61
65

82
87
110
113
140

143
187
183
260
268

270
280
282
350
351

Inserting the above values and the given data into Equation (4.20) yields
110.6655
2
Sb 10 = 12 (25) ln 137.6

25

(6(25) + 25 + 1)

= 21.21
From Table 11.1 for a two-tailed test with 98% confidence level, the critical
values of

0.02
2 , ( m 1) = 2
, (25 1) = 42.98

2
2

where
= 1 (confidence level) = 1 0.98 = 0.2
and

0.02

2 1 , (m 21) = 1
, (25 1) = 10.85

2
2

The above results indicate that there is no reason to contradict the assumption of
exponential distribution.

4.8.2

GENERAL EXPONENTIAL TEST

This test is also used to determine whether a given set of data belong to an exponential
distribution. The test requires calculating the value of , the 2 variate with 2m
degrees of freedom. The is defined by [28, 29, 32]:
m

= 2

T (i )

ln T ()

(4.23)

i =1

where
T () is the total operating time at test termination.
T (i) is the total operating time at the occurrence of ith failure.
m
is the total number of failures in a given sample.
If the value of lies within

2 , 2 m < <2 1 , 2 m
2

1999 by CRC Press LLC

(4.24)

TABLE 4.9
Times to Failure (in hours)
of Electrical Devices
Failure No.

Failure time (h)

1
2
3
4
5
6
7
8
9
10

70
85
25
10
20
40
50
45
30
60

then the assumption of times to failure following an exponential distribution is valid.

The symbol denotes the risk of rejecting a true assumption and is expressed by
= 1 (confidence level)

(4.25)

Example 4.3
Assume that 40 identical electrical devices were tested for 200 h and 10 failed. The
failed devices were not replaced. The failure times of the electrical devices are
presented in Table 4.9. Determine if the times to failure can be represented by the
exponential distribution.
Using the given data, we get
T ( ) = ( 40 10) (200) + 70 + 85 + 25 + 10 + 20 + 40 + 50 + 45 + 30 + 60
T ( ) = 6435 h
Thus, using the above result and the given data in Equation (4.23), we get
(10)( 40)
30 + (25) (38)
10 + (20)(39)
= 2 ln

+ ln
+ ln
6435
6435

6435
55 + (30)(37)
+ ln
+ ln
6435

85 + ( 40)(36)

6435

125 + ( 45)(35)
170 + (50) (34)
+ ln
+ ln

6435
6435

1999 by CRC Press LLC

 220 + (60)(33)
+ ln
+ ln
6435

280 + (70)(32)

6435

350 + (85) (31)

+ ln

6435

= 30.50
For a 98% confidence level from Equation (4.25) we have
= 1 0.98 = 0.02
Using the above value and other given data in relationship (4.24) we get
0.02
0.02
2
, 2 (10) ,2 1
, 2 (10)
2
2

Thus from Table 11.1, we get

2 [0.01, 20] = 37.56
2 [0.99, 20] = 8.26
The above results indicate that the calculated value of falls in between, i.e.,
8.26 < 30.50 < 37.56
It means the times to failure data can be represented by the exponential distribution.

4.8.3

KOLMOGOROVSMIRNOV TEST

This is another test used to determine the underlying statistical distribution of a

given sample of times to failure. The KolomogorovSmirnov test can be applied
regardless of the assumed failure distribution of a given data set. Also, this test is
quite effective regardless of the sample size. However, the test is restricted to
continuous statistical distributions only. The KolmogorovSmirnov test is composed
of the following eight steps [5, 31]:
1. Assume a statistical distribution for the failure data set under consideration.
2. Use the given failure data to estimate the values of the assumed distribution parameters.
3. Determine the level of significance, . In more detail, the level of significance may be described as the risk of not accepting the assumed statistical
distribution for the given data if it happened to be the real representative
distribution.

1999 by CRC Press LLC

4. Compute probability, F(t i) = P (T t i), using the assumed statistical

distribution function and the values of its associated parameters.
) for the ith ordered sample observation by using the given
5. Compute F(t
i
failure data and the following relationship:
F (t i ) = i n, for i = 1, 2, 3, , n

(4.24)

where
n

is the number of data points or observations in the given failure

data sample.
F (t i) denotes the proportion of the sample data observations that are
less than or equal to t i.

Also, shift i to (i 1) and calculate

i 1
F t i 1 =
, for i = 1, 2, 3, , n
n

( )

(4.25)

6. Calculate the statistic:

d=

(} )

max
F ( t i ) F ( t i ) , F ( t i ) F t i 1
i

(4.26)

7. Use Table 4.10 [13, 31, 33, 38] to obtain the value of d for a specified
value of and the sample size.
8. Make comparison of the value of d with the value of d. If d is greater
than d, reject the assumed distribution.
Example 4.4
A medical device manufacturer developed a certain medical device and life tested 12
units of the same device. These units failed after 20, 30, 40, 55, 60, 70, 75, 80, 90, 95,
100, and 110 h of operation. After reviewing the past failure pattern of similar devices,
it was assumed that the times to failure follow the normal distribution. Use the KolomogorovSmirnov method to test this assumption for 0.05 level of significance.
For the specified data, the mean of the normal distribution is given by
m

i =1

= (20 + 30 + 40 + + 100 + )100 12

= 68.75

1999 by CRC Press LLC

TABLE 4.10
KolmogorovSmirnov Critical Values for d
Level of significance,
Sample size, n

= .01

= 0.05

= 0.10

= 0.15

5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
>50

0.67
0.62
0.58
0.54
0.51
0.49
0.47
0.45
0.43
0.42
0.40
0.39
0.38
0.37
0.36
0.36
1.63 n1/2

0.57
0.52
0.49
0.46
0.43
0.41
0.39
0.38
0.36
0.35
0.34
0.33
0.32
0.31
0.30
0.29
1.36 n1/2

0.51
0.47
0.44
0.41
0.39
0.37
0.35
0.34
0.33
0.31
0.30
0.30
0.29
0.28
0.27
0.26
1.22 n1/2

0.47
0.44
0.41
0.38
0.36
0.34
0.33
0.31
0.30
0.29
0.28
0.27
0.27
0.26
0.25
0.25
1.14 n1/2

where
is the mean.
m is the number of medical devices life tested.
t i is the failure time of device i; for i = 1, 2, 3, , 12.
Similarly, for the given data, the standard deviation, , for the normal distribution
is
m

2
=
t i ) (m 1)
(
i =1

12

(20 68.75)2 + (30 68.75)2 + ( 40 68.75)2 + + (110 682 .75)

=

(12 1)

12

= 28.53
Using the above values for and and cumulative normal distribution function
table, the values for F(t i) as shown in Table 4.11 were obtained.
The values for F (t i) and F (t i1) presented in Table 4.12 were obtained using
Equations (4.24) and (4.25), respectively.

1999 by CRC Press LLC

TABLE 4.11
Values for F (t i )
i

ti

(t i )/

F (t i )

1
2
3
4
5
6
7
8
9
10
11
12

20
30
40
55
60
70
75
80
90
95
100
110

1.70
1.35
1.00
0.48
0.30
0.04
0.21
0.39
0.74
0.92
1.09
1.44

0.04
0.08
0.15
0.31
0.38
0.51
0.58
0.65
0.77
0.82
0.86
0.92

TABLE 4.12
Computed Values for F (t i )
and F (t i1)
i

ti

F (t i )

F (t i1)

1
2
3
4
5
6
7
8
9
10
11
12

20
30
40
55
60
70
75
80
90
95
100
110

1/12 = 0.08
2/12 = 0.16
3/12 = 0.25
= 0.33
= 0.41
= 0.5
= 0.58
= 0.66
= 0.75
= 0.83
= 0.91
12/12 = 1

(1 1)/12 = 0
(2 1)/12 = 0.08
(3 1)/12 = 0.16
= 0.25
= 0.33
= 0.41
= 0.5
= 0.58
= 0.66
= 0.75
= 0.83
(12 1)/12 = 0.91

Using the values presented in Tables 4.11 and 4.12 for F(t i), F (t i), and F (t i1),
the absolute values of F(t i) F (t i)  and F(t1) F (t i1)  listed in Table 4.13 were
obtained. By examining Table 4.13 we obtained the maximum absolute difference
of 0.11 (i.e., d = 0.11 for Equation (4.26)).
For given data (i.e., = 0.05 and n = 12) and using Table 4.10, we get d0.05 =
0.38. Thus, we conclude that as d0.05 > d (i.e., 0.38 > 0.11), there is no reason to
reject the normal distribution assumption.

1999 by CRC Press LLC

TABLE 4.13
Absolute Calculated Values for
F(t i ) F (t i ) and F(t i ) F (t i1 )
i

ti

F(t i ) F (t i )

F(t1) F (t i1)

1
2
3
4
5
6
7
8
9
10
11
12

20
30
40
55
60
70
75
80
90
95
100
110

0.04
0.08
0.1
0.02
0.03
0.01
0
0.01
0.02
0.01
0.05
0.08

0.04
0
0.01
0.06
0.05
0.1
0.08
0.07
0.11 max.
0.07
0.03
0.01

4.9 MAXIMUM LIKELIHOOD ESTIMATION METHOD

Once the techniques such as those described in the preceding section identify the
failure distribution of a given set of data or the failure distribution of the data under
consideration is known through other means, the maximum likelihood estimation
method can be used to estimate the values of the distribution parameters. The
maximum likelihood estimation method is the most flexible and powerful of modern
estimation approaches and is described in detail in References 20 and 39 through 41.
The basis for this approach is to assume a random sample of n failure times,
say t1, t2, t3, , tn taken from a population with a probability density function, f (t; );
where t is time and is the distribution parameter to be estimated. The joint density
function of these n variables is called the likelihood function, L, and is defined by
L = f ( t1 ; ) f ( t 2 ; ) f ( t 3 ; ) f ( t n ; )

(4.27)

The value of that maximizes the ln L or the L is known as the maximum likelihood
estimator (MLE) of . Usually, the is estimated by solving the following expression:
ln L
=0

One of the important properties of the MLE is as follows:

1999 by CRC Press LLC

(4.28)

For the large size n, the variance is simply given by

2 ln L
var =
2

(4.29)

where
denotes the estimated value of the .
The application of this method to selected statistical distributions is presented below.

4.9.1

EXPONENTIAL DISTRIBUTION

The probability density function is defined by

f ( t; ) = e t

t0

(4.30)

where
t is time.
is the distribution parameter.
Substituting Equation (4.30) into Equation (4.27) yields
n

L = n e

ti

i =1

(4.31)

By taking natural logarithms of Equation (4.31), we get

n

ln L = n ln

(4.32)

i =1

Substituting Equation (4.32) into Equation (4.28) yields

ln L n
=

t = 0
i

(4.33)

i =1

Rearranging Equation (4.33) leads to

n

= n

t
i =1

1999 by CRC Press LLC

(4.34)

where
is the estimated value of .
By differentiating Equation (4.32) with respect to twice, we get
2 ln L
= n 2
2

(4.35)

For a large sample size n, using Equation (4.35) in Equation (4.29) yields [40]
Var = 2 n

(4.36)

 2 n

4.9.2

NORMAL DISTRIBUTION

The probability density is defined by

f ( t ; , ) =

2
[ ( t )
1
e
2

2 2

]
(4.37)

for < t <

where
and are the distribution parameters.
t is time.
By inserting Equation (4.37) into relationship (4.27) and then taking logarithms, we
get

log L =

n
n
1
ln 2 ln 2
2
2
2 2

(t )

(4.38)

i =1

Differentiating Equation (4.28) with respect to and equating the resulting expression to zero yields
n

i =1

1999 by CRC Press LLC

2 (t i ) ( 1)
=
2 2
=1

ti 1
=

t n = 0
i

i =1

(4.39)

Rearranging Equation (4.39) yields

i =1

ti n

(4.40)

where
is the estimated value of .
Similarly, differentiating Equation (4.38) with respect to 2 and setting the resulting
expression equal to zero, we have
1
2

(t ) n = 0
2

(4.41)

i =1

By rearranging Equation (4.41), we get

1
2 =
n

(t )

(4.42)

i =1

where
2 is a biased estimator for 2.

4.9.3

WEIBULL DISTRIBUTION

The probability density function is defined by

f (t; , b) = b t b1 e t

(4.43)

for t 0
where
is the scale parameter.
b is the shape parameter.
t is time.
By inserting Equation (4.43) into Equation (4.27) and then taking natural logarithms,
we get
n

ln L = n ln + n ln b

i =1

1999 by CRC Press LLC

ln t

t ib + ( b 1)

i =1

(4.44)

Now we set lnL/ = 0, lnL/b = 0 to get

n

b
i

=0

b
i

ln t i +

(4.45)

i =1

and
n

ln t i

i =1

t
i =1

n
=0
b

(4.46)

By rearranging Equations (4.45) and (4.46), we get

n

= n

b
i

(4.47)

i =1

and
n

b =

t
i =1

(4.48)

b
i

ln t i

ln t

i =1

where
is the estimated value of .
b is the estimated value of b.
Equations (4.47) and (4.48) can be solved by using an iterative process [39].

4.10

PROBLEMS

1. List important uses of failure data.

2. Discuss the sources of collecting failure related data during an equipments life cycle.
3. Discuss the typical information to be included on a failure data collection
form.
4. What are the important guidelines to be followed during the design of a
failure reporting and documentation system?
5. Discuss the following U.S. bodies with respect to obtaining failure data:
Reliability Analysis Center (RAC)
National Technical Information Service (NTIS)
Government Industry Data Exchange Program (GIDEP)

1999 by CRC Press LLC

TABLE 4.14
Failure and Censoring Times for 20 Identical
Electronic Devices
Device No.

Failure and censoring times in hours

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20

200
600
1000a
1500
2000
400
600a
2500
3000
2600
1300a
1400
900
800a
500
600
1800a
1000
700
1600

Censoring time.

6. Write an essay on MIL-HDBK-217.

7. Assume that 20 identical electronic devices were put on test at time t = 0
and Table 4.14 presents failure and censoring times of these devices.
Determine the statistical distribution fit to the given data and estimate
values for its parameters by using the hazard plotting method.
8. A sample of 20 identical electrical parts were tested and their times to
failure, in hours, are given in Table 4.15. Determine if these times to
failure follow exponential distribution by employing the Bartlett test.

TABLE 4.15
Failure Times (hours) of Electrical Parts
10
12
15
13
20

1999 by CRC Press LLC

25
37
33
55
60

70
75
95
98
110

115
140
145
180
200

TABLE 4.16
Parts Failure Times
Failure No.

Failure time (h)

1
2
3
4
5
6
7
8

40
100
25
30
110
140
80
150

9. A set of 35 identical parts were tested for 300 h and 8 failed. The failed
parts were not replaced and their times to failure are given in Table 4.16.
Determine if these failure times can be represented by the exponential
distribution.
10. Assume that a sample of 10 identical electrical parts were life tested and
they failed after 50, 60, 70, 90, 100, 115, 120, 125, 130, and 140 h. After
reviewing the failure pattern of similar parts, it was assumed that the times
to failure are normally distributed. Use the KolmogorovSmirnov method
to test this assumption for 0.1 level of significance.
11. The probability density function of the gamma distribution is defined by
f ( t; , ) =

t 1 e t
,
( )

for t > 0

(4.49)

where
t

()

is
is
is
is

time.
the scale parameter.
the shape parameter.
the gamma function. For positive integer , () = ( 1)!.

Develop expressions for estimating distribution parameter values by employing the

maximum likelihood estimation method.

4.11 REFERENCES
1. Dhillon, B.S. and Viswanath, H.C., Bibliography of literature on failure data, Microelectronics and Reliability, 30, 723-750, 1990.
2. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Handbook of Reliability Engineering
and Management, McGraw-Hill, New York, 1996.
3. Kletz, T., The uses, availability and pitfalls of data on reliability, Process Technology,
18, 111-113, 1973.

1999 by CRC Press LLC

4. Mitchell, R.L. and Rutter, R.R., A study of automotive reliability and associated cost
of maintenance in the U.S.A., Soc. Automotive Eng. (SAE) Paper No. 780277, Jan.
1978.
5. Dhillon, B.S., Mechanical Reliability: Theory, Models, and Applications, American
Institute of Aeronautics and Astronautics, Washington, D.C., 1988.
6. Hahn, R.F., Data collection techniques, Proc. Annu. Reliability Maintainability Symp.,
38-43, 1972.
7. Parascos, E.T., A new approach to the establishment and maintenance of equipment
failure rate data bases, in Failure Prevention and Reliability, Bennett, S.B., Ross,
A.L., and Zemanick, P.Z., Eds., American Society of Mechanical Engineers, New
York, 1977, 263-268.
8. A Reliability Guide to Failure Reporting, Analysis, and Corrective Action Systems,
Committee on Reliability Reporting, American Society for Quality Control (ASQC),
Milwaukee, 1977.
9. Dhillon, B.S., Advanced Design Concepts for Engineers, Technomic Publishing Company, Lancaster, PA, 1998.
10. MIL-STD-1556, Government/Industry Data Exchange Program (GIDEP), Department of Defense, Washington, D.C.
11. MIL-HDBK-217F (Notice 2), Reliability Prediction of Electronic Equipment, Department of Defense, Washington, D.C., 1995.
12. TD-84-3, Reliability and Maintainability Data for Industrial Plants, A.P. Harris and
Associates, Ottawa, Canada, 1984.
13. Schafer, R.E., Angus, J.E., Finkelstein, J.M., Yerasi, M., and Fulton, D.W., RADC
Nonelectronic Reliability Notebook, Reliability Analysis Center, Rome Air Development Center, Griffiss Air Force Base, Rome, New York, 1985. Report No. RADCTR-85-194.
14. Rossi, M.J., Nonelectronic Parts Reliability Data, Reliability Analysis Center, Rome
Air Development Center, Griffiss Air Force Base, Rome, New York, 1985. Report
No. NPRD-3.
15. Green, A.E. and Bourne, A.J., Reliability Technology, John Wiley & Sons, Chichester,
England, 1972.
16. Dhillon, B.S., Human Reliability: With Human Factors, Pergamon Press, New York,
1986.
17. Joos, D.W., Sabri, Z.A., and Husseiny, A.A., Analysis of gross error rates in operation
of commercial nuclear power stations, Nuclear Eng. Design, 52, 265-300, 1979.
18. Peters, G.A., Human error: analysis and control, Am. Soc. Safety Eng. J., XI, 9-15,
1966.
19. Recht, J.L., Systems safety analysis: error rates and costs, National Safety News,
February 1966, pp. 20-23.
20. Dhillon, B.S., Reliability Engineering in Systems Design and Operation, Van Nostrand Reinhold Company, New York, 1983.
21. Nelson, W., Hazard plot analysis of incomplete failure data, Proc. Annu. Symp.
Reliability, 391-403, 1969.
22. Nelson, W., Theory and applications of hazard plotting for censored failure data,
Technometrics, 14, 945-966, 1972.
23. Nelson, W., Hazard plotting for incomplete failure data, J. Quality Technol., 1, 27-52,
1969.
24. Nelson, W., Life data analysis by hazard plotting, Evaluation Eng., 9, 37-40, 1970.
25. Nelson, W., Applied Life Data Analysis, John Wiley & Sons, New York, 1982.
26. Dhillon, B.S., A hazard rate model, IEEE Trans. Reliability, 28, 150, 1979.

1999 by CRC Press LLC

27. Dhillon, B.S., Life distributions, IEEE Trans. Reliability, 30, 457-460, 1981.
28. Epstein, B., Tests for the validity of the assumption that the underlying distribution
of life is exponential, Technometrics, 2, 83-101, 1960.
29. Epstein, B., Tests for the validity of the assumption that the underlying distribution
of life is exponential, Technometrics, 2, 327-335, 1960.
30. Lamberson, L.R., An evaluation and comparison of some tests for the validity of the
assumption that the underlying distribution of life is exponential, AIIE Trans., 12,
327-335, 1974.
31. Massey, F., The Kolmogorov-Smirnov test for goodness of fit, J. Am. Stat. Assoc.,
46, 70, 1951.
32. Dhillon, B.S., Quality Control, Reliability, and Engineering Design, Marcel Dekker,
New York, 1985.
33. AMC Pamphlet 706-198, Development Guide for Reliability, Part IV, U.S. Army
Materiel Command, Department of Defense, Washington, D.C., 1976.
34. Reliability and Maintainability Handbook for the U.S. Weather Bureau, Publication
No. 530- 01-1-762, ARINC Research Corporation, Annapolis, MD, April 1967.
35. Ehrenfeld, S. and Mauer, S.B.L., Introduction to Statistical Method, McGraw-Hill,
New York, 1964.
36. Klerer, M. and Korn, G.A., Digital Computer Users Handbook, McGraw-Hill, New
York, 1967.
37. Conover, W.J., Practical Nonparametric Statistics, John Wiley & Sons, New York,
1971.
38. AMCP 702-3, Quality Assurance Handbook, U.S. Army Material Command, Washington, D.C., 1968.
39. Lloyd, M. and Lipow, M., Reliability: Management, Methods, and Mathematics,
Prentice-Hall, Englewood Cliffs, NJ, 1962.
40. Shooman, M.L., Probablistic Reliability: An Engineering Approach, McGraw-Hill,
New York, 1968.
41. Mann, N., Schafer, R.E., and Singpurwalla, N.D., Methods for Statistical Analysis of
Reliability and Life Data, John Wiley & Sons, New York, 1974.

1999 by CRC Press LLC

Basic Reliability
Evaluation and
Allocation Techniques

5.1 INTRODUCTION
As engineering systems become more complex and sophisticated, the importance
of reliability evaluation and allocation techniques during their design is increasing.
Usually, in the design of engineering systems, reliability requirements are specified.
These requirements could be in the form of system reliability, failure rate, mean
time between failures (MTBF), and availability. Normally, in order to determine the
fulfillment of such requirements, various reliability evaluation and allocation techniques are employed.
Over the years, many reliability evaluation techniques have been developed but
their effectiveness and advantages may vary quite considerably. Some of these methods
are known as block diagram, decomposition, delta-star, and Markov modeling [1, 2].
Reliability allocation may be described as the top-down process used to subdivide
a system reliability requirement or goal into subsystem and component requirements
or goals. Its basic objectives are to translate the system reliability requirement into more
manageable, lower level (subsystem and component) requirements and to establish an
individual reliability requirement for each subsystem/hardware designer or supplier.
There are many methods and techniques available to perform reliability allocation [3].
This chapter not only describes basic reliability evaluation and allocation techniques, but also associated areas because a clear understanding of these areas is
considered essential prior to learning the evaluation and allocation methods.

5.2 BATHTUB HAZARD RATE CURVE

The bathtub hazard rate curve shown in Figure 5.1 is used to describe failure rate
for many engineering components. This curve is the result of three types of failures:
(1) quality, (2) stress-related, and (3) wear out. The quality failures decrease with
time, the wear out increase with time, and the stress-related remain constant with
respect to time [4]. Nonetheless, the bathtub hazard rate curve may be divided into
three distinct parts: (1) burn-in period, (2) useful life period, and (3) wear out period.
During the burn-in period, the hazard rate decreases and some of the reasons for the
occurrence of failures during this region are poor quality control, inadequate manufacturing methods, poor processes, human error, substandard materials and workmanship, and inadequate debugging. Other terms used for this decreasing hazard
rate region are infant mortality region, break-in region, and debugging region.

1999 by CRC Press LLC

FIGURE 5.1 Bathtub hazard rate curve.

During the useful life period, the hazard rate remains constant and there are
various reasons for the occurrence of failures in this region: undetectable defects,
low safety factors, higher random stress than expected, abuse, human errors, natural
failures, explainable causes, etc.
During the wear out period, the hazard rate increases and the causes for the
wear out region failures include wear due to aging, corrosion and creep, short
designed-in life of the item under consideration, poor maintenance, wear due to
friction, and incorrect overhaul practices.
The following hazard rate function can be used to represent the entire bathtub
hazard rate curve [5]:
(t ) = b t b1 + (1 = ) c t c1 e t
for
c, b, , > 0
0 1
b = 0.5
c =1
t0
where
t
(t)
b and c
and

is time.
is the hazard rate.
are the shape parameters.
are the scale parameters.

1999 by CRC Press LLC

(5.1)

5.3 GENERAL RELIABILITY ANALYSIS

RELATED FORMULAS
There are a number of formulas often used in conducting reliability analysis. This
section presents four of these formulas based on the reliability function.

5.3.1

FAILURE DENSITY FUNCTION

This is defined by
d R (t )
= f (t )
dt

(5.2)

where
R(t) is the item reliability at time t.
f(t) is the failure (or probability) density function.

5.3.2

HAZARD RATE FUNCTION

This is expressed by
( t ) = f ( t ) R( t )

(5.3)

where
(t) is the item hazard rate or time dependent failure rate.
Substituting Equation (5.2) into Equation (5.3) yields
(t ) =

5.3.3

d R( t )
1

R( t )
dt

(5.4)

GENERAL RELIABILITY FUNCTION

This can be obtained by using Equation (5.4). Thus, we have

1
d R (t ) = (t ) d t
R( t )

(5.5)

Integrating both sides of Equation (5.5) over the time interval [o, t], we get
R(t)

since at t = 0, R (t) = 1.

1999 by CRC Press LLC

1
d R (t ) = (t ) d t
R (t )
o

(5.6)

Evaluating the left-hand side of Equation (5.6) yields

t

ln R(t ) = (t ) d t

(5.7)

Thus, from Equation (5.7) we get

t

R (t ) = e

(t)d t

(5.8)

The above equation is the general expression for the reliability function. Thus, it can
be used to obtain reliability of an item when its times to failure follow any known
statistical distribution, for example, exponential, Rayleigh, Weibull, and gamma.

5.3.4

MEAN TIME

TO

FAILURE

This can be obtained by using any of the following three formulas [6]:

MTTF = E(t ) =

t f (t ) d t

(5.9)

or

R( t ) d t

(5.10)

MTTF = lim R (s)

(5.11)

MTTF =

or
s 0

where
MTTF
E(t)
s
R(s)

is
is
is
is

the
the
the
the

item mean time to failure.

expected value.
Laplace transform variable.
Laplace transform for the reliability function, R (t).

Example 5.1
Assume that the failure rate of a microprocessor, , is constant. Obtain expressions
for the microprocessor reliability, mean time to failure, and using the reliability
function prove that the microprocessor failure rate is constant.
Thus, substituting for (t) into Equation (5.8) yields

1999 by CRC Press LLC

R( t ) = e

dt
o

(5.12)

= e t
By inserting Equation (5.12) into Equation (5.10), we get

MTTF = e t d t
o

(5.13)

Using Equation (5.12) in Equation (5.4) leads to

(t ) =

1
e

( ) e t

(5.14)

=
Thus, Equations (5.12) and (5.13) represent expressions for microprocessor reliability and mean time to failure, respectively, and Equation (5.14) proves that the
microprocessor failure rate is constant.
Example 5.2
Using Equation (5.12), prove that the result obtained by utilizing Equation (5.11) is
the same as the one given by Equation (5.13).
Thus, the Laplace transform of Equation (5.12) is
R (s) = 1 (s + )

(5.15)

Using Equation (5.15) in Equation (5.11) yields

MTTF = lim 1 (s + )
s 0

(5.16)

Equations (5.13) and (5.16) results are identical; thus, it proves that Equations (5.10)
and (5.11) give the same result.
Example 5.3
Assume that the failure rate of an automobile is 0.0004 failures/h. Calculate the
automobile reliability for a 15-h mission and mean time to failure.

1999 by CRC Press LLC

Using the given data in Equation (5.8) yields

R(15) = e ( 0.0004 ) (15)
= 0.994
Similarly, inserting the specified data for the automobile failure rate into
Equation (5.10), we get

MTTF = e ( 0.0004 ) t d t
o

1
(0.0004)

= 2, 500 h
Thus, the reliability and mean time to failure of the automobile are 0.994 and
2,500 h, respectively.

5.4 RELIABILITY NETWORKS

An engineering system can form various different configurations in conducting
reliability analysis. This section presents such commonly occurring configurations.

5.4.1

SERIES NETWORK

This is the simplest reliability network and its block diagram is shown in Figure 5.2.
Each block in the figure represents a unit/component. If any one of the unit fails,
the series system fails. In other words, all the units must work normally for the
success of the series system.
If we let Ei denote the event that the ith unit is successful, then the reliability
of the series system is expressed by
R S = P ( E1 E 2 E 3 n E

(5.17)

where
is the series system reliability.
RS
P (E1 E2 E3 - - - En) is the occurrence probability of events E1, E2, E3, - - -,
and En.

FIGURE 5.2 A series system with n units.

1999 by CRC Press LLC

For independent units, Equation (5.17) becomes

R S = P ( E1 )( E 2 ) ( En3 ) P ( E

(5.18)

where
P (Ei) is the probability of occurrence of event Ei; for i = 1, 2, 3, , n.
If we let Ri = P (Ei) for i = 1, 2, 3, , n in Equation (5.18) becomes
R S = R1 R 2 R 3 nR
(5.19)

Ri

i =1

where
Ri is the unit reliability; for i = 1, 2, 3, , n.
Since normally the value of Ri is between zero and one, the series system reliability
decreases with the increasing value of n.
For constant failure rate, i, of unit i, the reliability of the unit i is given by
Ri = e

i t

(5.20)

where
Ri (t) is the reliability of unit i at time t.
Thus, inserting Equation (5.20) into Equation (5.19) yields
n

RS (t ) = e

i t

i =1

(5.21)

where
Rs (t) is the series system reliability at time t.
By substituting the above equation into Equation (5.10), we get the following expression for the series system mean time to failure:
n

MTTFS =

i t

i =1

(5.22)
n

=1

i =1

1999 by CRC Press LLC

dt

Using Equation (5.21) in Equation (5.4) yields the following expression for the series
system hazard rate:
1

S (t ) =

i t

i =1

i =1

i t
i e i =1

(5.23)

i =1

As the right-hand side of Equation (5.23) is independent of time t, its left-hand

side is simply s, the failure rate of the series system. It means whenever we add
up failure rates of items, we automatically assure they act in series, a worst case
design scenario with respect to reliability.
Example 5.4
A car has four identical and independent tires and the failure rate of each tire is
0.0004 failures per hour. Obviously, when any one of the tires is punctured, the car
cannot be driven. Calculate the following:
Car reliability for a 15-h mission with respect to tires.
Car failure rate with respect to tires.
Car mean time to failure with respect to tires.
Inserting the given data into Equation (5.21) yields
R(15) = e ( 0.0016 ) (15)
= 0.9763
Using the specified failure data in Equation (5.23), we get
S = 4 (0.0004)
= 0.0016 failure h
Substituting the given data into Equation (5.22), we get
MTTFS = 1 4 (0.0004)
= 625 h
Thus, the car reliability, failure rate, and mean time to failure with respect to
tires are 0.9763, 0.0016 failures/h, and 625 h, respectively.

1999 by CRC Press LLC

FIGURE 5.3 Block diagram of a parallel system with n units.

5.4.2

PARALLEL NETWORK

In this case all the n units operate simultaneously and at least one such unit must
work normally for the system success. The n-unit parallel system block diagram is
shown in Figure 5.3. Each block in the figure denotes a unit/component.
If we let Ei denote the event that the ith unit is unsuccessful, then the failure
probability of the parallel system is expressed by
Fp = P ( E1 E 2 E3
n E

(5.24)

where
is the parallel system failure probability.
Fp
P (E1 E2 E3 En) is the occurrence probability of failure events E1, E2, E3,
, En.
For independent units, Equation (5.24) is written as
Fp = P ( E1 ) P ( E 2 )nP ( E3 ) P ( E

(5.25)

where
P (E1) is the probability of occurrence of failure event E1; for i = 1, 2, 3, , n.
Fp
is the failure probability of the parallel system.
If we let Fi = P(E1) for i = 1, 2, 3, , n in Equation (5.25) becomes
Fp = F1 F2 F3 Fn
(5.26)

1999 by CRC Press LLC

where
Fi is the failure probability of unit i; for i = 1, 2, 3, , n.
Substracting Equation (5.26) from unity we get
n

R p = 1 Fp = 1

(5.27)

i =1

where
Rp is the parallel system reliability.
For constant failure rates of units, subtracting Equation (5.20) from unity and then
substituting it into Equation (5.27) yields
n

R p (t ) = 1

(1 e )
i t

(5.28)

i =1

where
Rp (t) is the parallel system reliability at time t.
For identical units, integrating Equation (5.28) over the time interval [0, ] yields
the following formula for the parallel system mean time to failure:

MTTFp =

[1 (1 e ) ] d t
t n

(5.29)
n

1 i
i =1

where
MTTFp is the parallel system mean time to failure.

is the unit failure rate.

Example 5.5
An aircraft has two independent and identical engines. At least one of the engines
must operate normally for the aircraft to fly successfully. Calculate the aircrafts
reliability with respect to engines, if each engines probability of success is 0.95.

1999 by CRC Press LLC

Subtracting the engine probability of success from unity, we get the following
engine failure probability:
Fe = 1 0.95
= 0.05
Inserting the above value and the other given data into Equation (5.27) yields
R p = 1 (0.05)

= 0.9975
Thus, the aircrafts reliability with respect to engines is 0.9975.
Example 5.6
A system is composed of two independent units in parallel. The failure rates of units
A and B are 0.002 failures per hour and 0.004 failures per hour, respectively.
Calculate the system reliability for a 50-h mission and mean time to failure.
Let A be the failure rate of unit A and B the failure rate of unit B. Thus, for
n = 2, using Equation (5.28) we get
+ t
R p (t ) = e A t + e B t e ( A B )

(5.30)

By substituting the given data into Equation (5.30) we get

R p (50) = e ( 0.002 ) ( 50 ) + e ( 0.004 ) ( 50 ) e ( 0.002 +0.004 ) ( 50 )
= 0.9827
Integrating Equation (5.30) over the time interval [0, ], we get the following
result for system mean time to failure:
MTTF =

[e

A t

MTTF =
=

+ t
+ e B t e ( A B ) d t

1
1
1
+

A B A + B
1
1
1
+

(0.002) (0.004) (0.002 + 0.004)

= 583.33 h
Thus, the system reliability and mean time to failure are 0.9827 and 583.33 h,
respectively.

1999 by CRC Press LLC

5.4.3

r-OUT-OF-n NETWORK

This is another form of redundancy in which at least r units out of a total of n units
must work normally for the system success. Furthermore, all the units in the system
are active. The parallel and series networks are special cases of this network for
r = 1 and r = n, respectively.
Using the Binomial distribution, for independent and identical units, the reliability of the r-out-of-n network is given by
n

Rr n =

i R (1 R)
i

n i

(5.31)

i=r

where
n
n!
=
i ( n i )! i !

(5.32)

Rr/n is the r-out-of-n network reliability and R the unit reliability.

For constant failure rates of units, Equation (5.31) becomes
n

R r n (t ) =

n
i e (1 e )

t n i

i t

(5.33)

i=r

By inserting Equation (5.33) into Equation (5.10) we get

MTTFr n =

n n
i t

1 e t
e
i=r r

n i

dt

(5.34)

1 i
i=r

where
MTTFr/n is the mean time to failure of the r-out-of-n system.
Example 5.7
A computer system has three independent and identical units in parallel. At least
two units must work normally for the system success. Calculate the computer system
mean time to failure, if the unit failure rate is 0.0004 failures per hour.

1999 by CRC Press LLC

Substituting the given data into Equation (5.34) yields

3

1 i

1
MTTF2 3 =
(0.0004)
=

i=2

+
(0.0004) 2 3

= 2083.33 h
Thus, the computer system mean time to failure is 2083.33 h.

5.4.4

STANDBY REDUNDANCY

This is another type of redundancy used to improve system reliability. In this case,
one unit operates and m units are kept in their standby mode. The total system
contains (m + 1) units and as soon as the operating unit fails, the switching mechanism detects the failure and turns on one of the standby units. The system fails
when all the standbys fail. For independent and identical units, perfect switching
and standby units, and unit time dependent failure rate, the standby system reliability
is given by
m

Rsd (t ) =

t
(t)d t
(t ) d t e o
i!

i=0

(5.35)

where
Rsd (t) is the standby system reliability.
m
is the number of standby units.
For constant unit failure rate, Equation (5.35) becomes
m

Rsd (t ) =

( t ) e
i

i!

(5.36)

i=0

By substituting Equation (5.36) into Equation (5.10) we get

MTTFsd =

( t )i e t i ! d t
i=0

= (m + 1)

1999 by CRC Press LLC

(5.37)

FIGURE 5.4 A five dissimilar unit bridge network.

where
MTTFsd is the standby system mean time to failure.
Example 5.8
Assume that a standby system has two independent and identical units: one operating,
another on standby. The unit failure rate is 0.006 failures per hour. Calculate the
system reliability for a 200-h mission and mean time to failure, if the switching
mechanism never fails and the standby unit remains as good as new in its standby
mode.
By substituting the specified data into Equation (5.36) yields
l

Rsd (200) =

[(0.006) (200e )]

( 0.006 ) ( 200 )

i!

i=0

= 0.6626
Similarly, using the given data in Equation (5.37) we get
MTTFsd = 2 (0.006)
= 333.33 h
Thus, the standby system reliability and mean time to failure are 0.6626 and
333.333 h, respectively.

5.4.5

BRIDGE NETWORK

In some engineering systems, units may be connected in a bridge configuration as

shown in Figure 5.4. Each block in the figure denotes a unit.
For independent units, Figure 5.4 reliability is given by [7]

1999 by CRC Press LLC

R br = 2

R + R R R
Ri +

i =1

i=2

+ R1 R 4 + R 2 R 5
(5.38)

R R R R R R R R R R
i

i=2

i =1

i =1

i=3

where
Rbr is the bridge network reliability.
Ri is the ith unit reliability; for i = 1, 2, 3, , 5.
For identical units, Equation (5.38) simplifies to
R br = 2 R 5 5 R 4 + 2 R 3 + 2 R 2

(5.39)

For constant failure rates of units, from Equation (5.39) we write

R br (t ) = 2 e 5 t 5 e 4 t + 2e 3 t + 2 e 2 t

(5.40)

where
is the unit failure rate.
By integrating Equation (5.40) over the time interval [0, ], we get the following
formula for the bridge network mean time to failure:
MTTFbr = 49 60

(5.41)

where
MTTFbr is the bridge network mean time to failure.
Example 5.9
Assume that five independent and identical units form a bridge configuration. The
failure rate of each unit is 0.0002 failures per hour. Calculate the configuration
reliability for a 500-h mission.
Substituting the given values into Equation (5.40) yields
R br (500) = 2 e 5( 0.0002 ) ( 500 ) 5 e 4 (.0 0002 ) ( 500 ) + 2 e 3( 0.0002 ) ( 500 ) + 2 e 2 ( 0. 0002 ) ( 500 )
= 0.9806
Thus, the bridge configuration reliability is 0.9806.

1999 by CRC Press LLC

5.5 RELIABILITY EVALUATION METHODS

Over the years many reliability evaluation methods have been developed [2, 8, 9].
This section presents some of those techniques.

5.5.1

NETWORK REDUCTION APPROACH

This is probably the simplest approach to determine the reliability of systems made
up of independent series and parallel subsystems. However, the subsystems forming
bridge configurations can also be handled by first applying the delta-star method [10].
Nonetheless, the network reduction approach sequentially reduces the series and
parallel subsystems to equivalent hypothetical single units until the entire system
under consideration itself becomes a single hypothetical unit. The following example
demonstrates this method.
Example 5.10
An independent unit network representing an engineering system is shown in
Figure 5.5 (i). The reliability Ri of unit i; for i = 1, 2, 3, , 6 is specified. Calculate
the network reliability by using the network reduction approach.
First we have identified subsystems A, B, C, and D of the network as shown in
Figure 5.5 (i). The subsystems A and B have their units in series; thus, we reduce
them to single hypothetical units as follows:
R A = R1 R 3 = (0.4) (0.6) = 0.24
and
R B = R 2 R 4 = (0.5) (0.8) = 0.40
where
RA is the reliability of subsystem A.
RB is the reliability of subsystem B.
The reduced network is shown in Figure 5.5 (ii). Now, this network is composed of
two parallel subsystems C and D. Thus, we reduce both subsystems to single
hypothetical units as follows:
R C = 1 (1 R A ) (1 R B ) = 1 (1 0.24) (1 0.4)
= 0.5440
and
R D = 1 (1 R 5 ) (1 R 6 ) = 1 (1 0.7) (1 0.9)
= 0.97

1999 by CRC Press LLC

FIGURE 5.5 Diagrammatic steps of the network reduction approach: (i) Original network;
(ii) reduced network I; (iii) reduced network II; (iv) single hypothetical unit.

where
RC is the subsystem C reliability.
RD is the subsystem D reliability.
Figure 5.5. (iii) depicts the reduced network with the above calculated values. This
resulting network is a two unit series system and its reliability is given by
R S = R C R D = (0.5440) (0.97)
= 0.5277
The single hypothetical unit shown in Figure 5.5 (iv) represents the reliability
of the whole network given in Figure 5.5 (i). More specifically, the entire network
is reduced to a single hypothetical unit. Thus, the total network reliability is 0.5277.

1999 by CRC Press LLC

5.5.2

DECOMPOSITION APPROACH

This approach is used to determine reliability of complex systems, which it decomposes into simpler subsystems by applying the conditional probability theory. Subsequently, the system reliability is determined by combining the subsystems reliability measures.
The basis for the approach is the selection of the key unit used to decompose a
given network. The efficiency of the approach depends on the selection of this key
unit. The past experience usually plays an important role in its selection.
The method starts with the assumption that the key unit, say k, is replaced by
another unit that is 100% reliable or never fails and then it assumes the key unit k
is completely removed from the network or system. Thus, the overall system or
network reliability is expressed by

()

R n = P ( k ) P (system good k good ) + P k P (system good k fails)

(5.42)

where
Rn
P (k)
P (k)
P ()

is
is
is
is

the
the
the
the

overall network or system reliability.

probability of success or reliability of the key unit k.
failure probability of unreliability of the key unit k.
probability.

This approach is demonstrated by solving the following example.

Example 5.11
An independent and identical units bridge network is shown in Figure 5.6. The letter
R in the figure denotes unit reliability. Obtain an expression for the bridge network
reliability by using the decomposition method.
With the aid of past experience, we choose the unit falling between nodes A
and B, shown in Figure 5.6, as our Key unit, say k.

FIGURE 5.6 An identical unit bridge network.

1999 by CRC Press LLC

Next, we replace this key unit k with a unit that never fails. Consequently, the
Figure 5.6 network becomes a series-parallel system whose reliability is expressed
by

Rsp = 1 (1 R )

= 2 RR

2 2

(5.43)

2 2

Similarly, we totally remove the key unit k from Figure 5.6 and the resulting network
becomes a parallel-series system. This parallel-series system reliability is given by

R ps = 1 1 R 2
=2 R R
2

(5.44)

where
Rps is the parallel-series system reliability.
The reliability and unreliability of the key unit k, respectively, are given by
P (k) = R

(5.45)

()

(5.46)

and
P k = (1 R )
Rewriting Equation (5.42) in terms of our example, we get
R n = R R sp + (1 R ) R ps

(5.47)

Substituting Equations (5.43) and (5.44) into Equation (5.47) yields

Rn = R 2 R R2

+ (1 R ) 2 R 2 R 4

= 2 R 5 R +2 R +2 R
5

(5.48)

The above equation is for the reliability of the bridge network shown in Figure 5.6.
Also, it is identical to Equation (5.39).

1999 by CRC Press LLC

FIGURE 5.7 Delta to star equivalent reliability diagram.

5.5.3

DELTA-STAR METHOD

This is the simplest and very practical approach to evaluate reliability of bridge
networks. This technique transforms a bridge network to its equivalent series and
parallel form. However, the transformation process introduces a small error in the
end result, but for practical purposes it should be neglected [2].
Once a bridge network is transformed to its equivalent parallel and series form,
the network reduction approach can be applied to obtain network reliability. Nonetheless, the delta-star method can easily handle networks containing more than one
bridge configurations. Furthermore, it can be applied to bridge networks composed
of devices having two mutually exclusive failure modes [10, 11].
Figure 5.7 shows delta to star equivalent reliability diagram. The numbers 1, 2,
and 3 denote nodes, the blocks the units, and R() the respective unit reliability.
In Figure 5.7, it is assumed that three units of a system with reliabilities R12,
R13, and R23 form the delta configuration and its star equivalent configuration units
reliabilities are R1, R2, and R3.
Using Equations (5.19) and (5.27) and Figure 5.7, we write down the following
equivalent reliability equations for network reliability between nodes 1, 2; 2, 3; and
1, 3, respectively:
R1 R 2 = 1 (1 R12 ) (1 R13 R 23 )

(5.49)

R 2 R 3 = 1 (1 R 23 ) (1 R12 R13 )

(5.50)

R1 R 3 = 1 (1 R13 ) (1 R12 R 23 )

(5.51)

Solving Equations (5.49) through (5.51), we get

R1 = AC B

1999 by CRC Press LLC

(5.52)

FIGURE 5.8 A five unit bridge network with specified unit reliabilities.

where
A = 1 (1 R12 ) (1 R13 R 23 )

(5.53)

B = 1 (1 R 23 ) (1 R12 R13 )

(5.54)

C = 1 (1 R13 ) (1 R12 R 23 )

(5.55)

R 2 = AB C

(5.56)

R 3 = BC A

(5.57)

Example 5.12
A five independent unit bridge network with specified unit reliability Ri; for i = a,
b, c, d, and e is shown in Figure 5.8. Calculate the network reliability by using the
delta-star method and also use the specified data in Equation (5.39) to obtain the
bridge network reliability. Compare both results.
In Figure 5.8 nodes labeled 1, 2, and 3 denote delta configurations. Using
Equations (5.52) through (5.57) and the given data, we get the following star
equivalent reliabilities:
R1 = AC B
= 0.9633
where
A = B = C = 1 (1 0.8) [1 (0.8) (0.8)] = 0.9280
R 2 = 0.9633

1999 by CRC Press LLC

FIGURE 5.9 Equivalent network to bridge configuration of Figure 5.8.

and
R 3 = 0.9633
Using the above results, the equivalent network to Figure 5.8 bridge network is
shown in Figure 5.9.
The reliability of Figure 5.9 network, Rbr, is

R br = R 3 1 (1 R1 R d ) (1 E 2 R e )
= 0.9126

By substituting the given data into Equation (5.39) we get

R br = 2 (0.8) 5 (0.8) + 2 (0.8) + 2 (0.8)
5

= 0.9114
Both the reliability results are basically same, i.e., 0.9126 and 0.9114. All in all,
for practical purposes the delta-star approach is quite effective.

5.5.4

PARTS COUNT METHOD

This is a very practically inclined method used during bid proposal and early design
phases to estimate equipment failure rate [12]. The information required to use this
method includes generic part types and quantities, part quality levels, and equipment
use environment. Under single use environment, the equipment failure rate can be
estimated by using the following equation [12]:
m

E =

Q ( F )
i

i =1

1999 by CRC Press LLC

q i

(5.58)

where
E is the equipment failure rate, expressed in failures/106 h.
m is the number of different generic part/component classifications in the
equipment under consideration.
g is the generic failure rate of generic part i expressed in failures/106 h.
Qi is the quantity of generic part i.
Fq is the quality factor of generic part i.
Reference 12 presents tabulated values for g and Fq .
Failure Rate Estimation of an Electronic Part
As the design matures, more information becomes available, the failure rates of
equipment components are estimated. Usually, in the case of electronic parts, the
MIL-HDBK-217 [12] is used to estimate the failure rate of electronic parts. The
failure rates are added to obtain total equipment failure rate. This number provides
a better picture of the actual failure rate of the equipment under consideration than
the one obtained through using Equation (5.58).
An equation of the following form is used to estimate failure rates of many
electronic parts [12]:
p = b e q

(5.59)

where
p is the part failure rate.
b is the base failure rate and is normally defined by a model relating the
influence of temperature and electrical stresses on the part under consideration.
e is the factor that accounts for the influence of environment.
q is the factor that accounts for part quality level.
For many electronic parts, the base failure rate, b, is calculated by using the
following equation:
b = C exp [ E kT]
where
C
E
k
T

is
is
is
is

a constant.
the activation energy for the process.
the Boltzmanns constant.
the absolute temperature.

1999 by CRC Press LLC

(5.60)

5.5.5

MARKOV METHOD

This is a powerful reliability analysis tool and it can be used in more cases than any
other method. The method is quite useful to model systems with dependent failure
and repair modes. Markov method is widely used to model repairable systems with
constant failure and repair rates. However, with the exception of a few special cases,
the technique breaks down for a system having time dependent failure and repair
rates. In addition, a problem may occur in solving a set of differential equations for
large and complex systems. The following assumptions are associated with the
Markov approach [8]:
All occurrences are independent of each other.
The probability of transition from one system state to another in the finite
time interval t is given by t, where the is the transition rate (i.e.,
failure or repair rate) from one system state to another.
The probability of more than one transition occurrence in time intervalt
from one state to another is very small or negligible (e.g., (t) (t) 0).
The Markov method is demonstrated by solving the following example.
Example 5.13
Assume that an engineering system can either be in an operating or a failed state.
It fails at a constant failure rate, , and is repaired at a constant repair rate, . The
system state space diagram is shown in Figure 5.10. The numerals in box and circle
denote the system state. Obtain expressions for system time dependent and steady
state availabilities and unavailabilities by using the Markov method.
Using the Markov method, we write down the following equations for state 0
and state 1, respectively, shown in Figure 5.10.
P0 (t + t ) = P0 (t ) (1 t ) + P1 (t ) t

(5.61)

P1 (t + t ) = P1 (t ) (1 t ) + P0 (t ) t

(5.62)

FIGURE 5.10 System transition diagram.

1999 by CRC Press LLC

where
t
t
(1 t)
t
(1 t)
P0 (t + t)

is time.
is the probability of system failure in finite time intervalt.
is the probability of no failure in finite time intervalt.
is the probability of system repair in finite time intervalt.
is the probability of no repair in finite time intervalt.
is the probability of the system being in operating state 0 at time
(t + t).
P1 (t + t) is the probability of the system being in failed state 1 at time (t + t).
Pi (t)
is the probability that the system is in state i at time t, for i = 0, 1.

In the limiting case, Equations (5.61) and (5.62) become

d P0 (t )
+ P0 (t ) = P1 (t )
dt

(5.63)

d P1 (t )
+ P1 (t ) = P0 (t )
dt

(5.64)

At time t = 0, P0 (0) = 1 and P1 (0) = 0.

Solving Equations (5.63) and (5.64), we get
P0 (t ) =

P1 (t ) =

+ t
e ( )

(5.65)

+ t

e ( )
( + ) ( + )

(5.66)

( + ) ( + )

Thus, the system time dependent availability and unavailability, respectively, are
A (t ) = P0 (t ) =

+ t
+
e ( )
( + ) ( + )

(5.67)

and
UA (t ) = P1 (t ) =

( + ) ( + )

+ t
e ( )

where
A (t) is the system time dependent availability.
UA (t) is the system time dependent unavailability.

1999 by CRC Press LLC

(5.68)

The system steady state availability and unavailability can be obtained by using any
of the following three approaches:
Approach I: Letting time t go to infinity in Equations (5.67) and (5.68),
respectively.
Approach II: Setting the derivatives of Equations (5.63) and (5.64)
equal to zero and then discarding any one of the resulting two equations
and replacing it with P0 + P1 = 1. The solutions to the ultimate equations
will be system steady state availability (i.e., A = P0 ) and unavailability
(i.e., UA = P1 ).
Approach III: Taking Laplace transforms of Equations (5.63) and
(5.64) and then solving them for P0(s), the Laplace transform of probability that the system is in operating state at time t, and P1(s), the Laplace
transform of probability that the system is in failed state at time t. Multiplying P0(s) and P1(s) with the Laplace transform variable s and then
letting s in sP0(s) and sP1(s) go to zero result in system steady state
availability (i.e., A = P0 ) and unavailability (i.e., UA = P1), respectively.
Thus, in our case applying Approach I to Equations (5.67) and (5.68), we get
A = lim A (t ) =
t

( + )

(5.69)

and
UA = lim UA (t ) =
t

( + )

(5.70)

where
A is the system steady state availability.
UA is the system steady state unavailability.
Since =

1
MTTF

A=

and =

1
MTTR

, Equations (5.69) and (5.70) become

MTTF
System uptime
=
MTTF + MTTR System uptime + System downtime

where
MTTF is the system mean time to failure.
MTTR is the system mean time to repair.

1999 by CRC Press LLC

(5.71)

and
UA =

MTTR
System downtime
=
MTTF + MTTR System uptime + System downtime

(5.72)

Thus, the system time dependent and steady state availabilities are given by Equations (5.67) and (5.69) or (5.71), respectively. Similarly, the system time dependent
and steady state unavailabilities are given by Equations (5.68) and (5.70) or (5.72),
respectively.

5.6 RELIABILITY ALLOCATION

Reliability allocation may simply be described as the process of assigning reliability
requirements to individual parts or components to achieve the specified system reliability. There are many different methods used to perform reliability allocation [1, 2, 13].
Nonetheless, the reliability allocation problem is not that simple and straightforward
but quite complex. Some of the associated reasons are as follows:
Role the component plays for the operation of the system.
Component complexity.
The chargeable component reliability with the type of function to be
conducted.
Approaches available for accomplishing the given allocation task.
Lack of detailed information on many of the above factors in the early
design phase.
Nonetheless, there are many benefits of the reliability allocation including it forces
individuals involved in design and development to clearly understand and develop
the relationships between reliabilities of components, subsystems, and systems, it
forces the design engineer to seriously consider reliability equally with other design
parameters such as performance, weight, and cost, and it ensures satisfactory design,
manufacturing approaches, and test methods.
Two reliability allocation methods are described below.

5.6.1

HYBRID METHOD

This method is the result of combining two approaches: similar familiar systems
and factors of influence. The resulting method incorporates benefits of the other two
methods; thus, it becomes more attractive.
The basis for the similar familiar systems reliability allocation approach is the
familiarity of the designer with similar systems or sub-systems. In addition, failure
data collected on similar systems from various sources can also be used during the
allocation process. The main drawback of this approach is to assume that reliability
and life cycle cost of previous similar designs were adequate.

1999 by CRC Press LLC

The factors of influence method is based upon the following factors that are
considered to effect the system reliability:
Complexity/Time: The complexity relates to the number of subsystem
parts and the time to the relative operating time of the item during the
entire system functional period.
Failure criticality. This factor considers the criticality of the item failure
on the system. For example, some auxiliary instrument failure in an
aircraft may not be as crucial as the engine failure.
Environment. This factor takes into consideration the susceptibility or
exposure of items to environmental conditions such as temperature,
humidity, and vibration.
State-of-the-Art. This factor takes into consideration the advancement in
the state-of-the-art for a specific item.
In using the above influence factors, each item is rated with respect to each of these
influence factors by assigning a number from 1 to 10. The assignment of 1 means the
item under consideration is least affected by the factor in question and 10 means the
item is most affected by the same influence factor. Ultimately, the reliability is allocated
by using the weight of these assigned numbers for all influence factors considered.
Now, it should be obvious to the reader that the hybrid method is better than
similar familiar systems and factors of influence methods because it uses data from
both of these approaches.

5.6.2

FAILURE RATE ALLOCATION METHOD

This method is concerned with allocating failure rates to system components when
the system required failure rate is known. The following assumptions are associated
with this method:
System components form a series configuration.
System components fail independently.
Time to component failure is exponentially distributed.
Thus, the system failure rate using Equation (5.23) is
n

S =

i =1

where
n is the number of components in the system.
S is the system failure rate.
i is the failure rate of system component i; for i = 1, 2, 3, , n.

1999 by CRC Press LLC

(5.73)

If the system required failure rate is sr , then allocate component failure rate such that
n

*
i

sr

(5.74)

i =1

where
*i is the failure rate allocated to component i; for i = 1, 2, 3, , n.
The following steps are associated with this method:
1. Estimate the component failure rates i for i = 1, 2, 3, , n, using the
past data.
2. Calculate the relative weight, i , of component i using the preceding step
failure rate data and the following equation:
i =

, for i = 1, 2,, n

(5.75)

i =1

It is to be noted that i denotes the relative failure vulnerability of the

component i and
n

= 1
i

(5.76)

i =1

3. Allocate failure rate to component i using the following relationship:

*i = i sr , for i = 1, 2,, n

(5.77)

It must be remembered that Equation (5.77) is subject to the condition that the
equality holds in Equation (5.74).
Example 5.14
Assume that a military system is composed of five independent subsystems in series
and its specified failure rate is 0.0006 failures/h. The estimated failure rates from
past experience for subsystems 1, 2, 3, 4, and 5 are 1 = 0.0001 failures/h, 2 =
0.0002 failures/h, 3 = 0.0003 failures/h, 4 = 0.0004 failures/h, and 5 = 0.0005
failures/h, respectively. Allocate the specified system failure rate to five subsystems.

1999 by CRC Press LLC

Using Equation (5.73) and the given data, we get the following estimated military
system failure rate:
5

s =

i =1

= (0.0001) + (0.0002) + (0.0003) + (0.0004) + (0.0005)

= 0.0015 failures h
Thus, utilizing Equation (5.75) and calculated and given values, we get the
following relative weights for subsystems 1, 2, 3, 4, and 5, respectively:
1 = (0.0001 0.0015) = 0.0667
2 = (0.0002 0.0015) = 0.1333
3 = (0.0003 0.0015) = 0.2
4 = (0.0004 0.0015) = 0.2667
5 = (0.0005 0.0015) = 0.3333
Using Equation (5.77) and calculated and given values, the subsystems 1, 2, 3,
4, and 5 allocated failure rates, respectively, are as follows:
*1 = 1 sr = (0.0667) (0.0006)
= 0.00004 failures h
*2 = 2 sr = (0.1333) (0.0006)
= 0.00007 failures h
*3 = 3 sr = (0.2) (0.0006)
= 0.00012 failures h
*4 = 4 sr = (0.2667) (0.0006)
= 0.00016 failures h
*5 = 5 sr = (0.333) (0.0006)
= 0.00019 failures h

1999 by CRC Press LLC

5.7 PROBLEMS
1. Describe the bathtub hazard rate curve and the reasons for its useful life
region failures.
2. Prove that the item mean time to failure (MTTF) is given by
MTTF = lim R (s)
s 0

(5.78)

where s
is the Laplace transform variable.
R(s) is the Laplace transform of the item reliability.
3. Prove that the mean time to failure of a system is given by
MTTF =

1
1 + 2 + + n

(5.79)

where
n is the number of units.
i is the unit failure rate; for i = 1, 2, 3, , n.
Write down your assumptions.
4. A parallel system has n independent and identical units. Obtain an expression for the system hazard rate, if each units times to failure are exponentially distributed. Compare the end result with the one for the series
system under the same condition.
5. Prove that when an items times to failure are exponentially distributed,
its failure rate is constant.
6. A system has four independent and identical units in parallel. Each units
failure rate is 0.0005 failures/h. Calculate the system reliability, if at least
two units must operate normally for the system success during a 100-h
mission.
7. Compare the mean time to failure of k independent and identical unit series
and standby systems when unit times to failure are exponentially distributed.
8. Calculate the reliability of the Figure 5.11 network using the delta-star
approach. Assume that each block in the figure denotes a unit with reliability 0.8 and all units fail independently.

FIGURE 5.11 An eight unit network.

1999 by CRC Press LLC

9. For the system whose transition diagram is shown in Figure 5.10, obtain
steady state probability equations by applying the final value theorem to
Laplace transforms (i.e., Approach III of Examples 5.13).
10. An aerospace system is made up of seven independent subsystems in series
and it specified failure rate 0.009 failures/h. Subsystems 1, 2, 3, 4, 5, 6, and
7 estimated failure rates from previous experience are 0.001 failures/h,
0.002 failures/h, 0.003 failures/h, 0.004 failures/h, 0.005 failures/h,
0.006 failures/h, and 0.007 failures/h, respectively. Allocate the specified
system failure rate to seven subsystems.

5.8 REFERENCES
1. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Eds., Handbook of Reliability
Engineering and Management, McGraw-Hill, New York, 1996.
2. Dhillon, B.S., Reliability in Systems Design and Operation, Van Nostrand Reinhold
Company, New York, 1983.
3. AMCP 706-196, Engineering Design Handbook: Development Guide for Reliability,
Part II: Design for Reliability, U.S. Army Material Command (AMC), Washington,
D.C., 1976.
4. Reliability Design Handbook, RDG-376, Reliability Analysis Center, Rom Air Development Center, Griffis Air Force Base, Rome, New York, 1976.
5. Dhillon, B.S., A hazard rate model, IEEE Trans. Reliability, 28, 150, 1979.
6. Dhillon, B.S., Mechanical Reliability: Theory, Models, and Applications, American
Institute of Aeronautics and Astronautics, Washington, D.C., 1988.
7. Lipp, J.P., Topology of switching elements vs. reliability, Trans. IRE Reliability
Quality Control, 7, 21-34, 1957.
8. Shooman, M.L., Probabilistic Reliability: An Engineering Approach, McGraw-Hill,
New York, 1968.
9. Dhillon, B.S. and Singh, C., Engineering Reliability: New Techniques and Applications, John Wiley & Sons, New York, 1981.
10. Dhillon, B.S., The Analysis of the Reliability of Multistate Device Networks, Ph.D.,
Dissertation, 1975. Available from the National Library of Canada, Ottawa.
11. Dhillon, B.S. and Proctor, C.L. Reliability analysis of multistate device networks,
Proc. Annu. Reliability Maintainability Symp., 31-35, 1976.
12. MIL-HDBK-217, Reliability Prediction of Electronic Equipment, Department of
Defense, Washington, D.C.
13. Dhillon, B.S., Systems Reliability, Maintainability and Management, Petrocelli
Books, New York, 1983.

1999 by CRC Press LLC

Failure Modes
and Effect Analysis

6.1 INTRODUCTION
Failure modes and effect analysis (FMEA) is a powerful design tool to analyze
engineering systems and it may simply be described as an approach to perform
analysis of each potential failure mode in the system to examine the results or effects
of such failure modes on the system [1]. When FMEA is extended to classify each
potential failure effect according to its severity (this incorporates documenting catastrophic and critical failures), the method is known as failure mode effects and
criticality analysis (FMECA).
The history of FMEA goes back to the early 1950s with the development of
flight control systems when the U.S. Navys Bureau of Aeronautics, in order to
establish a mechanism for reliability control over the detail design effort, developed
a requirement called Failure Analysis [2]. Subsequently, Coutinho [3] coined the
term Failure Effect Analysis and the Bureau of Naval Weapons (i.e., successor to
the Bureau of Aeronautics) introduced it into its new specification on flight controls.
However, FMECA was developed by the National Aeronautics and Astronautics
Administration (NASA) to assure the desired reliability of space systems [4].
In the 1970s, the U.S. Department of Defense directed its effort to develop a
military standard entitled Procedures for Performing a Failure Mode, Effects, and
Criticality Analysis [5]. Today FMEA/FMECA methods are widely used in the
industry to conduct analysis of systems, particularly for use in aerospace, defense,
and nuclear power generation. Reference 6 presents a comprehensive list of publications on FMEA/FMECA. This chapter discusses different aspects of FMEA/FMECA.

6.2 TERMS AND DEFINITIONS

There are many terms used in performing FMEA/FMECA and some of them are as
follows [5]:
Failure cause. The factors such as design defects, quality defects, physical
or chemical processes, or part misapplication are the primary reason for
failure or they start the physical process through which deterioration
progresses to failure.
Failure mode. The notion or manner through which a failure is perceived.
Failure effect. The consequence or consequences a failure mode has on
an items function, operation, or status.

1999 by CRC Press LLC

FIGURE 6.1 Types of FMEA.

Single failure point. An items malfunction that would lead to system

failure and is not compensated through redundancy or through other
operational mechanisms.
Criticality. A relative measure of a failure modes consequences and its
occurrence frequency.
Severity. A failure modes consequences, taking into consideration the
worst case scenario of a failure, determined by factors such as damage to
property, the degree of injury, or ultimate system damage.
Corrective action. A change such as design, process, procedure, or materials implemented and validated to rectify failure cause/design deficiency.
Criticality analysis. An approach through which each possible failure
mode is ranked with respect to the combined influence of occurrence
probability and severity.
Undetectable failure. A postulated failure mode in the FMEA for which
no failure detection approach is available through which the concerned
operator can be alerted of the failure.
Local effect. The consequence or conseqences a failure mode has on the
function, operation, or status of the item currently being analyzed.

6.3 TYPES OF FMEA AND THEIR

ASSOCIATED BENEFITS
FMEA may be grouped under three distinct classifications as shown in Figure 6.1 [7].
These are design-level FMEA, system-level FMEA, and process-level FMEA. Each
of these categories is described below.

6.3.1

DESIGN-LEVEL FMEA

The purpose of performing design-level FMEA is to help identify and stop product
failures related to design. This type of FMEA can be carried out upon componentlevel/subsystem-level/system-level design proposal and its intention is to validate

1999 by CRC Press LLC

the design parameters chosen for a specified functional performance requirement.

The advantages of performing design-level FMEA include identification of potential
design-related failure modes at system/subsystem/component level, identification of
important characteristics of a given design, documentation of the rationale for design
changes to guide the development of future product designs, help in the design
requirement objective evaluation and evaluation of design alternatives, systematic
approach to reduce criticality and risk, accumulated data serve as a useful historical
record of the thought processes and the actions taken during the product development
effort, and a useful tool to establish priority for design improvement actions.

6.3.2

SYSTEM-LEVEL FMEA

This is the highest-level FMEA that can be performed and its purpose is to identify
and prevent failures related to system/subsystems during the early conceptual design.
Furthermore, this type of FMEA is carried out to validate that the system design
specifications reduce the risk of functional failure to the lowest level during the
operational period.
Some benefits of the system-level FMEA are identification of potential systemic
failure modes due to system interaction with other systems and/or by subsystem
interactions, selection of the optimum system design alternative, identification of
potential system design parameters that may incorporate deficiencies prior to releasing hardware/software to production, a systematic approach to identify all potential
effects of subsystem/assembly part failure modes for incorporation into design-level
FMEA, and a useful data bank of historical records of the thought processes as well
as of actions taken during product development efforts.

6.3.3

PROCESS-LEVEL FMEA

This identifies and prevents failures related to the manufacturing/assembly process

for a certain item. The benefits of the process-level FMEA include identification of
potential process-level failure modes at system/subsystem/operation level, identification of important characteristics associated with the process, identification of
potential process shortcomings early in the process planning cycle, development of
priorities for process improvement actions, and documentation of rationale for process changes to help guide the establishment of potential manufacturing processes.

6.4 STEPS FOR PERFORMING FMEA

FMEA can be performed in six steps as shown in Figure 6.2 [4, 8]. These steps are
as follows:

define system and its associated requirements,

establish ground rules,
describe the system and its associated functional blocks,
identify failure modes and their associated effects,

1999 by CRC Press LLC

FIGURE 6.2 Steps for performing FMEA.

prepare critical items list,

document the analysis.
Each of these steps is described below.

6.4.1

DEFINE SYSTEM

AND ITS

ASSOCIATED REQUIREMENTS

This is concerned with defining the system under consideration and the definition
normally incorporates a breakdown of the system into blocks, block functions, and
the interface between them. Usually this early in the program a good system definition does not exist and the analyst must develop his/her own system definition
using documents such as trade study reports, drawings, and development plans and
specifications.

6.4.2

ESTABLISH GROUND RULES

These are established as to which the FMEA is conducted. Usually, developing the
ground rules is a quite straightforward process when the system definition and
mission requirements are reasonably complete. Nonetheless, examples of the ground
rules include primary and secondary mission objectives statement, limits of environmental and operational stresses, statement of analysis level, delineation of mission phases, definition of what constitutes failure of system hardware parts, and the
detail of the coding system used.

1999 by CRC Press LLC

6.4.3

DESCRIBE THE SYSTEM

FUNCTIONAL BLOCKS

AND ITS

ASSOCIATED

This is concerned with the preparation of the description of the system under
consideration. Such description may be grouped into two parts:
Narrative functional statement. This is prepared for each subsystem and
component as well as for the total system. It provides narrative description
of each items operation for each operational mode/mission phase. The
degree of the description detail depends on factors such as an items
application and the uniqueness of the functions performed.
System block diagram. The purpose of this block diagram is to determine
the success/failure relationships among all the system components; thus,
it graphically shows total system components to be analyzed as well as
the series and redundant relationships among the components. In addition,
the block diagram shows the entire systems inputs and outputs and each
system elements inputs and outputs.

6.4.4

IDENTIFY FAILURE MODES

AND

THEIR ASSOCIATED EFFECTS

This is concerned with performing analysis of the failure modes and their effects.
A form such as that shown in Figure 6.3 is used as a worksheet to assure systematic
and thorough coverage of all failure modes. Even though all the terms used in that
form are considered self-explanatory, the terms Compensating provisions and
Criticality classification are described below.
Compensating provisions. These provisions, i.e., design provisions or
operator actions, concerning circumventing or mitigating the failure effect
should be identified and evaluated [5].
Criticality classification. This is concerned with the categorization of
potential effect of failure. For example [4],
People may lose their lives due to a failure.
Failure may cause mission loss.
Failure may cause delay in activation.
Failure has no effect.

6.4.5

PREPARE CRITICAL ITEMS LIST

The critical items list is prepared to facilitate communication of important analysis

results to management. A typical critical items list worksheet is shown in Figure 6.4.

6.4.6

DOCUMENT

THE

ANALYSIS

This is the final step of the FMEA performing process and is concerned with the
documentation of analysis. This step is at least as important as the other previous
five steps because poor documentation can lead to ineffectiveness of the FMEA

1999 by CRC Press LLC

FIGURE 6.3 A typical FMEA worksheet.

FIGURE 6.4 A critical items list worksheet.

process. Nonetheless, the FMEA report incorporates items such as system definition,
FMEA associated ground rules, failure modes and their effects, description of the
system (i.e., including functional descriptions and block diagrams), and critical items
list.

6.5 CRITICALITY ASSESSMENT

The objective of the criticality assessment is to prioritize the failure modes discovered
during the system analysis on the basis of their effects and occurrence likelihood [9].
Thus, for making an assessment of the severity of an item failure, two commonly
used methods are Risk Priority Number (RPN) Technique and Military Standard
Technique. The former is widely used in the automotive industrial sector and the
latter in defense, nuclear, and aerospace industries. Both approaches are described
below [9-11].

6.5.1

RPN TECHNIQUE

This method calculates the risk priority number for a part failure mode using three
factors: (1) failure effect severity, (2) failure mode occurrence probability, and

1999 by CRC Press LLC

TABLE 6.1
Failure Detection Ranking
Item
No.

Likelihood
of detection

1
2
3
4
5
6

Very high
High
Moderate
Low
Very low
Detectability
absolutely uncertain

Rank meaning

Rank

Potential design weakness will almost certainly be detected

There is a good chance of detecting potential design weakness
There is a possibility of detecting potential design weakness
Potential design weakness is unlikely to be detected
Potential design weakness probably will not be detected
Potential design weakness cannot be detected

1, 2
3, 4
5, 6
7, 8
9
10

(3) failure detection probability. More specifically, the risk priority number is computed by multiplying the rankings (i.e., 110) assigned to each of these three factors.
Thus, mathematically the risk priority number is expressed by
RPN = (DR ) (OR ) (SR )

(6.1)

where
RPN is
DR is
OR is
SR is

the
the
the
the

risk priority number.

detection ranking.
occurrence ranking.
severity ranking.

Since the above three factors are assigned rankings from 1 to 10, the value of the
RPN will vary from 1 to 1000. Failure modes with a high RPN are considered to
be more critical; thus, they are given a higher priority in comparison to the ones
with lower RPN. Nonetheless, rankings and their interpretation may vary from one
organization to another. Tables 6.1 through 6.3 present rankings for failure detection,
failure mode occurrence probability, and failure effect severity used in one
organization [12], respectively.

6.5.2

MILITARY STANDARD TECHNIQUE

This technique is often used in defense, aerospace, and nuclear industries to prioritize
the failure modes of the item under consideration so that appropriate corrective
measures can be undertaken [5]. The technique requires the categorization of the
failure-mode effect severity and then the development of a critical ranking.
Table 6.4 presents classifications of failure mode effect severity.
In order to assess the likelihood of a failure-mode occurrence, either a qualitative
or a quantitative approach can be used. This is determined by the availability of
specific parts configuration data and failure rate data. The qualitative method is used

1999 by CRC Press LLC

TABLE 6.2
Failure Mode Occurrence Probability
Item
No.

Ranking
term

Ranking term meaning

1
2

Remote
Low

Occurrence of failure is quite unlikely

Relatively few failures are expected

Moderate

Occasional failures are expected

High

Repeated failures will occur

Very high

Occurrence of failure is almost inevitable

Occurrence
probability
6

<1 in 10
1 in 20,000
1 in 4,000
1 in 1,000
1 in 400
1 in 80
1 in 40
1 in 20
1 in 8
1 in 2

Rank
1
2
3
4
5
6
7
8
9
10

TABLE 6.3
Severity of the Failure-Mode Effect
Item
No.

Failure effect
severity category

Minor

Low

3
4

Moderate
High

Very high

Severity category description

No real effect on system performance and the customer
may not even notice the failure
The occurrence of failure will only cause a slight
customer annoyance
Some customer dissatisfaction will be caused by failure
High degree of customer dissatisfaction will be caused
by failure but the failure itself does not involve safety
or noncompliance with government rules and
regulations
The failure affects safe item operation, involves
noncompliance with government rules and regulations

Rank
1
2, 3
4, 5, 6
7, 8

9, 10

when there are no specific failure rate data. Thus, in this approach the individual
occurrence probabilities are grouped into distinct logically defined levels, which
establish the qualitative failure probabilities. Table 6.5 presents occurrence probability levels [5, 9]. For the purpose of identifying and comparing each failure mode
to all other failure modes with respect to severity, a critical matrix is developed as
shown in Figure 6.3.
The criticality matrix is developed by inserting item/failure mode identification
number values in matrix locations denoting the severity category classification and
either the criticality number (Ki) for the failure modes of an item or the occurrence
level probability. The distribution of criticality of item failure modes is depicted by
the resulting matrix and the matrix serves as a useful tool for assigning corrective
measure priorities. The direction of the arrow, originating from the origin, shown in

1999 by CRC Press LLC

TABLE 6.4
Failure Mode Effect Severity Classifications
Item
No.

Classification
term or name

Catastrophic

Critical

Marginal

Minor

Classification description
The occurrence of failure may result in death or equipment
loss
The occurrence of failure may result in major property
damage/severe injury/major system damage ultimately
leading to mission loss
The occurrence of failure may result in minor injury/minor
property damage/minor system damage, etc.
The failure is not serious enough to lead to injury/system
damage/property damage, but it will result in repair or
unscheduled maintenance

Classification
No.
A
B

C
D

TABLE 6.5
Qualitative Failure Probability Levels
Item
No.

Probability
level

Frequent

II

Reasonably probable

III

Occasional

IV

Remote

Extremely unlikely

Term for the level

Term description
High probability of occurrence during the item
operational period
Moderate probability of occurrence during the item
operational period
Occasional probability of occurrence during the item
operational period
Unlikely probability of occurrence during the item
operational period
Essentially zero chance of occurrence during the item
operational period

Figure 6.5, indicates the increasing criticality of the item failure and the darkened
region in the figure shows the approximate desirable design region. For severity
classifications A and B, the desirable design region has low occurrence probability
or criticality number. On the other hand, for severity classifications C and D failures,
higher probabilities of occurrence can be tolerated. Nonetheless, failure modes
belonging to classifications A and B should be eliminated altogether or at least their
probabilities of occurrence be reduced to an acceptable level through design changes.
The quantitative approach is used when failure mode and probability of occurrence data are available. Thus, the failure mode critical number is calculated using
the following equation:
K fm = F T

1999 by CRC Press LLC

(6.2)

FIGURE 6.5 Criticality matrix.

where
Kfm is the failure-mode criticality number.
is the failure-mode ratio or the probability that the component/part fails
in the particular failure mode of interest. More specifically, it is the
fraction of the part failure rate that can be allocated to the failure mode
under consideration and when all failure modes of an item under consideration are specified, the sum of the allocations equals unity. Table 6.6
presents failure mode apportionments for certain parts [10, 11, 13].
F is the conditional probability that the failure effect results in the indicated
severity classification or category, given that the failure mode occurs.
The values of F are based on an analysts judgment and these values are
quantified according to Table 6.7 guidelines.
T is the item operational time expressed in hours or cycles.
is the item/part failure rate.
The item criticality number Ki is calculated for each severity class, separately.
Thus, the total of the criticality numbers of all failure modes of an item resulting in
failures in the severity class of interest is given by

Ki =

j =1

j =1

(k fm )j = (F T)j

(6.3)

where
n is the item failure modes that fall under the severity classification under
consideration.

1999 by CRC Press LLC

TABLE 6.6
Failure Mode Apportionments for Selective Items
Item
no.

Item
description

Relay

Relief valve

Incandescent lamp

Fixed resistor

Transformer

Diode

Hydraulic valve

Item important failure modes

Contact failure
Open coils
Premature open
Leaking
Catastrophic (filament breakage, glass
breakage)
Degradation (loss of filament emission)
Open
Parameter change
Short
Shorted turns
Open circuit
Short circuit
Intermittent circuit
Open circuit
Other
Stuck closed
Stuck open
Leaking

Approximate probability
of occurrence
(or probability value for )
0.75
0.05
0.77
0.23
0.10
0.90
0.84
0.11
0.05
0.80
0.05
0.75
0.18
0.06
0.01
0.12
0.11
0.77

TABLE 6.7
Failure Effect Probability
Guideline Values
Item
no.

Failure effect
description

Probability
value of F

1
2
3
4

No effect
Actual loss
Probable loss
Possible loss

0
1.00
0.10 < F < 1.00
0 < F 0.10

It is to be noted that when item/part failure mode results in multiple severityclass effects each with its own occurrence probability, then in the calculation of Ki
only the most important is used [14]. This can lead to erroneously low Ki values for
the less critical severity categories. In order to rectify this error, it is recommended
to compute F values for all severity categories associated with a failure mode and
ultimately include only contributions of Ki for category B, C, and D failures [9, 14].

1999 by CRC Press LLC

6.6 FMECA INFORMATION NEEDS, DATA SOURCES,

AND USERS
In order to perform effective FMECA, various types of information are needed.
Design related information required for the FMECA includes system schematics,
functional block diagrams, equipment/part drawings, design descriptions, relevant
company/customer/commercial/military specifications, reliability data, field service
data, effects of environment on item under consideration, configurations management data, operating specifications and limits, and interface specifications.
Some specific FMECA related factors and their corresponding data retrieval
sources in parentheses, are as follows [9]:

Failure modes, causes, and rates (factory database, field experience database).
Failure effects (design engineer, reliability engineer, safety engineer).
Item identification numbers (parts list).
Failure detection method (design engineer, maintainability engineer).
Function (customer requirements, design engineer).
Failure probability/severity classification (safety engineer).
Item nomenclature/functional specifications (parts list, design engineer).
Mission phase/operational mode (design engineer).

Usually, FMECA is expected to satisfy the needs of many groups during the
design process including design, quality assurance, reliability and maintainability,
customer representatives, internal company regulatory agency, system engineering,
testing, logistics support, system safety, and manufacturing.

6.7 FMEA IMPLEMENTATION RELATED FACTORS

AND GENERAL GUIDELINES
The implementation of FMEA is not as easy as it may appear to be. Nonetheless,
prior to its implementation, the following areas must be explored [15, 16]:
Obtaining engineers support and approval.
Examination of every conceivable failure mode by the involved professionals.
Making decisions based on the RPN.
Measuring FMEA cost/benefits.
Use of FMEA to select the concept.
There are a number of guidelines/facts associated with FMEA including: avoid
evaluating every conceivable failure mode, FMEA has limitations, FMEA is not
designed to supersede the engineers work, RPN could be misleading, avoid developing majority of the FMEA in a meeting, a very small RPN may justify corrective
measure, FMEA is not the tool for choosing the optimum design concept, tailor
severity, occurrence, and detection scales to reflect the companys product and

1999 by CRC Press LLC

processes, and applying the Pareto principle to the RPN is a misapplication of the
Pareto principle [15].

6.8 ADVANTAGES OF FMEA

There are many benefits of performing FMEA, including a systematic approach to
classify hardware failures, reduces development time and cost, reduces engineering
changes, easy to understand, serves as a useful tool for more efficient test planning,
highlights safety concerns to be focused on, improves customer satisfaction, an
effective tool to analyze small, large, and complex systems, useful in the development
of cost-effective preventive maintenance systems, provides safeguard against repeating the same mistakes in the future, useful to compare designs, a visibility tool for
manager, a useful approach that starts from the detailed level and works upward,
and useful to improve communication among design interface personnel [8, 15].

6.9 PROBLEMS
1. Write a short essay on the history of FMEA.
2. Define the following terms:
Failure mode
Single point failure
Severity
3. Discuss design-level FMEA and its associated benefits.
4. Describe the FMEA process.
5. Discuss the elements of a typical critical items list worksheet.
6. Describe the RPN method.
7. Identify sources for obtaining data for the following factors:
Item identification numbers
Failure effects
Failure probability/severity classification
8. List at least 10 advantages of performing FMEA.
9. What is the difference between FMEA and FMECA?
10. Who are the users of FMEA/FMECA?

6.10

REFERENCES

1. Omdahl, T.P., Ed., Reliability, Availability and Maintainability (RAM) Dictionary,

American Society for Quality Control (ASQC) Press, Milwaukee, WI, 1988.
2. MIL-F-18372 (Aer), General Specification for Design, Installation, and Test of Aircraft Flight Control Systems, Bureau of Naval Weapons, Department of the Navy,
Washington, D.C., Para. 3.5.2.3.
3. Coutinho, J.S., Failure effect analysis, Trans. NY Acad. Sci., 26, Series II, pp. 564-584,
1963-1964.
4. Jordan, W.E., Failure modes, effects and criticality analyses, Proc. Annu. Reliability
Maintainability Symp., 30-37, 1972.

1999 by CRC Press LLC

5. MIL-STD-1629, Procedures for Performing a Failure Mode, Effects, and Criticality

Analysis, Department of Defense, Washington, D.C., 1980.
6. Dhillon, B.S., Failure mode and effects analysis bibliography, Microelectronics
and Reliability, 32, 719-731, 1992.
7. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Handbook of Reliability Engineering
and Management, McGraw-Hill, New York, 1996.
8. Dhillon, B.S., Systems Reliability, Maintainability, and Management, Petrocelli
Books, New York, 1983.
9. Bowles, J.B. and Bonnell, R.D., Failure mode, effects, and criticality analysis, in
Tutorial Notes, Annual Reliability and Maintainability Symposium, 1994, pp. 1-34.
Available from the Annual Reliability and Maintainability Symposium, C/O Evans
Associates, 804 Vickers Avenue, Durham, NC.
10. AMCP 706-196, Engineering Design Handbook: Development Guide for Reliability,
Part II, Design for Reliability, Army Material Command, Department of the Army,
Washington, D.C., 1976.
11. AE-9, Automotive Electronics Reliability Handbook, Society of Automotive Engineers, Warrendale, PA, 1987.
12. Potential Failure Mode and Effects Analysis in Design (Design FMEA) and for
Manufacturing and Assembly Process (Process FMEA) Instruction Manual, Ford
Motor Company, Detroit, MI, September 1988.
13. Failure Mode/Mechanism Distributions, Report No. FMD-91, Reliability Analysis
Center, Rome Air Development Center, Griffiss Air Force Base, Rome, NY, 1991.
14. Agarwala, A.S., Shortcomings in MIL-STD-1629A guidelines for criticality analysis,
Proc. Annu. Reliability Maintainability Symp., 494-496, 1990.
15. Palady, P., Failure Modes and Effects Analysis, PT Publications, West Palm Beach,
FL, 1995.
16. Mcdermott, R.E., Mikulak, R.J., and Beauregard, M.R., The Basics of FMEA, Quality
Resources, New York, 1996.

1999 by CRC Press LLC

Fault Tree Analysis

7.1 INTRODUCTION
Fault tree analysis (FTA) is widely performed in the industry to evaluate engineering
systems during their design and development, in particular the ones used in nuclear
power generation. A fault tree may simply be described as a logical representation
of the relationship of primary events that lead to a specified undesirable event called
the top event and is depicted using a tree structure with OR, AND, etc. logic gates.
The fault tree method was developed in the early 1960s by H.A. Watson of Bell
Telephone Laboratories to perform analysis of the Minuteman Launch Control
System. A study team at the Bell Telephone Laboratories further refined it and
Haasl [1] of the Boeing company played a pivotal role in its subsequent development.
In 1965 several papers related to the technique were presented at the System
Safety Symposium held at the University of Washington, Seattle [1]. In 1974, a
conference on Reliability and Fault Tree Analysis was held at the University of
California, Berkeley [2]. A paper appeared in 1978 providing a comprehensive list
of publications on Fault Trees [3]. The three books that described FTA in considerable depth appeared in 1981 [46]. Needless to say, since the inception of the fault
tree technique, many people have contributed to its additional developments.
This chapter describes different aspects of FTA in considerable depth.

7.2 FTA PURPOSES AND PREREQUISITES

There are many purposes in performing FTA, including identifying critical areas
and cost-effective improvements, understanding the functional relationship of system
failures, meeting jurisdictional requirements, providing input to test, maintenance,
and operational policies and procedures, understanding the level of protection that
the design concept provides against failures, evaluating performance of systems/equipment for bid-evaluation purposes, providing an integrated picture of some
aspects of system operation, confirming the ability of the system to meet its imposed
safety requirements, and providing input to cost-benefit trade-offs.
There are many prerequisites associated with FTA including thorough understanding of design, operation, and maintenance aspects of system/item under consideration; clear definition of what constitutes system failure: the undesirable event,
clearly defined analysis scope and objectives; a comprehensive review of system/item
operational experience; well-defined level of analysis resolution; clear identification
of associated assumptions; and clearly defined system/item physical bounds and
system interfaces.

1999 by CRC Press LLC

FIGURE 7.1 Commonly used fault tree symbols: (i) rectangle, (ii) circle, (iii) diamond,
(iv) triangle A, (v) triangle B, (vi) AND gate, (vii) OR gate.

7.3 FAULT TREE SYMBOLS

The fault tree method makes use of many different symbols to develop fault trees
of engineering systems. Seven such symbols are shown in Figure 7.1. Each one of
these symbols is described below.
Rectangle. Represents a fault event that results from the logical combination of fault events through the input of the logic gate.
Circle. Denotes a basic fault event or the failure of an elementary part.
The fault events probability of occurrence, failure, and repair rates are
usually obtained from empirical data.
Diamond. Denotes a fault event whose causes have not been fully developed either due to lack of interest or due to lack of information.
Triangle A. Denotes transfer in and is used to avoid repeating segments
of the fault tree.
Triangle B. Denotes transfer out and is used to avoid repeating segments
of the fault tree.
AND gate. Denotes that an output fault event occurs only if all of the
input fault events occur.
OR gate. Denotes that an output fault event occurs if one or more of the
input fault events occur.

1999 by CRC Press LLC

From time to time, there are many other symbols used to perform FTA. Most
of these symbols are discussed in References 6 through 8.

7.4 FUNDAMENTAL APPROACH TO FTA

The development or construction of a fault tree is top-down, in that the undesirable
or top event is the tree root and the logical combination of sub-events are employed
to map out the tree until reaching the basic initiating fault events. Nonetheless, steps
such as those listed below are involved in performing FTA [9]:
Define system, analysis associated assumptions, what constitutes a failure,
etc.
If the simplification of the scope of the analysis is necessary, develop a simple
system block diagram showing relevant inputs, outputs, and interfaces.
Identify undesirable or top fault events to be analyzed and if necessary
develop fault trees for all top-level events.
Identify all the causes that can make the top event occur using fault tree
symbols and the logic tree format. More specifically, using deductive
reasoning highlight event that can lead to the occurrence of the top event.
Assuming the causes of the previous step as intermediate effects, continue
developing the logic tree by identifying the causes of these intermediate
events.
Develop the fault tree to the lowest level of detail as required.
Perform analysis of the completed fault tree with respect to understanding
the logic and the interrelationships among various fault paths, gaining
insight into the unique modes of product faults, etc.
Determine appropriate corrective measures.
Prepare documentation of the analysis process and follow up on identified
corrective measures.
Example 7.1
Assume that a windowless room has a switch and four light bulbs. Develop a fault
tree for the top or undesired fault event dark room (i.e., room without light).
Assume that the room can only be dark if all the light bulbs burn out, there is no
electricity, or the switch fails to close.
A fault tree for this example is shown in Figure 7.2. Each fault event in the
figure is labeled as B1, B2, B3, , B10 .

7.5 BOOLEAN ALGEBRA RULES

A fault tree may be considered as a pictorial representation of those Boolean relationships among fault events that lead to the occurrence of the top event. Furthermore,
a fault tree may be translated into an entirely equivalent set of Boolean expressions.
Usually, prior to the probability evaluation of the top event, the simplification of the
Boolean expressions is required. The Boolean algebra rules play an important role
in this aspect. Selective rules of Boolean algebra are presented below [4, 5, 8].

1999 by CRC Press LLC

FIGURE 7.2 Fault tree for the top event: dark room or room without light.

Idempotent law
AA=A

(7.1)

A+A=A

(7.2)

X (Y + Z ) = XY + XZ

(7.3)

X + YZ = ( X + Y) ( X + Z )

(7.4)

AB = BA

(7.5)

A + B = B+ A

(7.6)

X + XY = X

(7.7)

X ( X + Y) = X

(7.8)

Distributive law

Commutative law

Absorption law

1999 by CRC Press LLC

FIGURE 7.3 An m input OR gate with Boolean expression.

7.6 ANALYTICAL DEVELOPMENTS OF BASIC

GATES, REPEATED FAULT EVENTS,
AND MINIMAL CUT SETS
A fault tree is developed using logic gates such as OR and AND that relate logically
various basic fault events to the undesirable or the top event. Boolean algebra is an
invaluable tool to represent a fault tree diagram in a mathematical form. Boolean
expressions for OR and AND gates are presented below.

OR GATE
An m input fault events A1, A2, A3, , Am OR gate along with its output fault event
A0 in a Boolean expression is shown in Figure 7.3. Thus, mathematically, the output
fault event A0 of the m input fault event OR gate is expressed by
A 0 = A1 + A 2 + A 3 + + A m

(7.9)

where
Ai is the ith input fault event; for i = 1, 2, 3, , m.

AND GATE
A k input fault event X1, X2, X3, , Xk AND gate along with its output fault event
X0 in a Boolean expression is shown in Figure 7.4. Thus, mathematically, the output
fault event X0 of the k input fault event AND gate is expressed by
X 0 = X1 X 2 X 3 X k
where
Xi is the ith input fault event; for i = 1, 2, 3, , k.

1999 by CRC Press LLC

(7.10)

FIGURE 7.4 A k input AND gate with Boolean expression.

Example 7.2
For the fault tree shown in Figure 7.2, develop a Boolean expression for the top
event: Room without light (i.e., fault event B10) using fault event identification
symbols B1, B2, B3, , B9 .
Using relationship (7.10), the Boolean expression for the intermediate fault event
B8 is
B8 = B1 B2 B3 B4

(7.11)

Similarly, utilizing relationship (7.9), the Boolean equation for the intermediate fault
event B9 is given by
B9 = B5 + B6

(7.12)

Again using relationship (7.9), the Boolean expression for the top fault event B10 is
expressed by
B10 = B7 + B8 + B9

7.6.1

(7.13)

REPEATED FAULT EVENTS

A hypothetical fault tree with repeated fault events is shown in Figure 7.5 (i.e., the
basic fault event A is repeated). Thus, in this case, the repetition of A must be
eliminated prior to obtaining the quantitative reliability parameter results for the
fault tree. Otherwise, the quantitative values will be incorrect. The elimination of
repeated events can either be achieved by applying the Boolean algebra properties
such as presented in Section 7.5 or algorithms especially developed for this
purpose [6, 10-12].
One such algorithm is represented in Section 7.6.2.

1999 by CRC Press LLC

FIGURE 7.5 A hypothetical fault tree with repeated events.

Example 7.3
Use Boolean algebra properties to eliminate the repetition of the occurrence of event
A shown in Figure 7.5. Construct the repeated event free fault tree using the simplified Boolean expression for the top event.
Using Figure 7.5, we write down the following Boolean expressions:
X=A+B

(7.14)

Y = A+C

(7.15)

Z = XY

(7.16)

T = Z+D+E

(7.17)

Inserting relationships (7.14) and (7.15) into expression (7.16), we get

Z = (A + B) (A + C)

1999 by CRC Press LLC

(7.18)

FIGURE 7.6 The repeated event free fault tree developed using Equation (7.20).

Utilizing Equation (7.4), we simplify relationship (7.18) to

Z = A + BC

(7.19)

Substituting Equation (7.19) into Equation (7.17) yields

T = A + BC + D + E

(7.20)

The above equation represents the simplified Boolean expression for the top event
of the fault tree shown in Figure 7.5. The repeated event free fault tree constructed
using Equation (7.20) is shown in Figure 7.6.

7.6.2

ALGORITHM

FOR

OBTAINING MINIMAL CUT SETS

One of the difficult problems facing the fault tree method is to obtain minimal cut
sets or eliminate repeated events of a fault tree. This section presents one algorithm
to obtain minimal cut sets of a fault tree [6, 1012].
A cut set may be described as a collection of basic events that will cause the
top fault tree event to occur. Furthermore, a cut set is said to be minimal if it cannot
be further minimized or reduced but it can still ensure the occurrence of the top
fault tree event.
The algorithm under consideration can either be used manually for simple fault
trees or computerized for complex fault trees with hundreds of gates and basic fault
events. In this algorithm, the AND gate increases the size of a cut set and the OR
gate increases the number of cut sets. The algorithm is demonstrated by solving the
following example:

1999 by CRC Press LLC

FIGURE 7.7 An event tree with a repeated event.

Example 7.4
Obtain a repeated event free fault tree of the fault tree shown in Figure 7.7. In this
figure, the basic fault events are identified by the numerals and the logic gates are
labeled as GT0, GT1, GT2, , GT6.
The algorithm starts from the gate, GT0, just below the top event of the fault
tree shown in Figure 7.7. Usually, in a fault tree this gate is either OR or AND. If
this gate is an OR gate, then each of its inputs represents an entry for each row of
the list matrix. On the other hand, if this gate is an AND, then each of its inputs
denotes an entry for each column of the list matrix.
In our case, GT0 is an OR gate; thus, we start the formulation of the list matrix
by listing the gate inputs: 11, 12, GT1, and GT2 (output events) in a single column
but in separate rows as shown in Figure 7.8 (i). As any one input of the GT0 could
cause the occurrence of the top event, these inputs are the members of distinct cut
sets.
One simple rule associated with this algorithm is to replace each gate by its
inputs until all the gates in a given fault tree are replaced with the basic event entries.
The inputs of a gate could be the basic events or the outputs of other gates. Consequently, in our case, in order to obtain a fully developed list matrix, we proceed to
replace the OR gate GT1 of list matrix of Figure 7.8 (i) by its input events as separate
rows, as indicated by the dotted line in Figure 7.8 (ii).

1999 by CRC Press LLC

FIGURE 7.8 List matrix under different conditions.

Next, we replace the OR gate GT2 of the list matrix in Figure 7.8 (ii) by its
input events as indicated by the dotted line in Figure 7.8 (iii).
Similarly, we replace the OR gate GT3 of the list matrix in Figure 7.8 (iii) by
its input events 5 and 6 as identified by the dotted line in Figure 7.8 (iv).
The next gate GT4 to be replaced with its input events is an AND gate. The
dotted line in Figure 7.8 (v) indicates its input events. Again, the next gate, GT5, to
be replaced by its input events is an AND gate.
Its input events are indicated by the dotted line in Figure 7.8 (vi).
The last gate to be replaced by its input events is an OR gate GT6. Its inputs
are indicated by the dotted line in the list matrix of Figure 7.8 (vii). As shown in
the list matrix of Figure 7.8 (vii), the cut set 2 is a single event cut set. As only its
occurrence will result in the occurrence of the top event, we eliminate cut set {3, 4, 2}
from the list matrix because the occurrence of this cut set requires all events 3, 4, and 2
to occur. Consequently, the list matrix shown in Figure 7.8 (viii) represents the
minimal cut sets of the fault tree given in Figure 7.7.
The fault tree of the Figure 7.8 (viii) list matrix is shown in Figure 7.9. Now this
fault tree can be used to obtain the quantitative measures of the top or undesirable event.

7.7 PROBABILITY EVALUATION OF FAULT

TREES WITH REPAIRABLE AND
NONREPAIRABLE COMPONENTS
After obtaining minimal cut sets or the redundancy free events of a given fault tree,
one can proceed to determine the probability of occurrence of the top event. This
probability can be obtained by first estimating the probability of occurrence of the
output events of lower and intermediate logic gates such as OR and AND. The output
event occurrence probability of these gates can be obtained as follows:

1999 by CRC Press LLC

FIGURE 7.9 A fault tree for the minimal cut sets of Figure 7.8 (viii).

OR GATE
Using Figure 7.3 for independent input events, the probability of occurrence of the
output fault event A0 is given by
P (A 0 ) = 1

{1 P (A )}
i

(7.21)

i =1

where
P (Ai) is the probability of occurrence of input fault event Ai; for i = 1, 2, 3,
, m.
For m = 2, Equation (7.21) reduces to
P (A 0 ) = 1

{1 P (A )}
i

i =1

= P (A1 ) + P (A 2 ) P (A1 ) P(A 2 )

1999 by CRC Press LLC

(7.22)

For the probabilities of occurrence of fault events A1 and A2 less than 0.1,
Equation (7.22) may be approximated by
P (A 0 ) P (A1 ) + P (A 2 )

(7.23)

For m fault events, Equation (7.23) becomes

P (A 0 )

P (A )
i

(7.24)

i =1

AND GATE
Using Figure 7.4, for independent input fault events, the probability of occurrence
of the output fault event X0 is
P (X 0 ) =

P (X )
i

(7.25)

i =1

where
P (Xi) is the probability of occurrence of input fault event Xi; for i = 1, 2, 3,
, k.

7.7.1

PROBABILITY EVALUATION OF FAULT TREES

WITH NON-REPAIRABLE COMPONENTS

In this case, the basic fault events, say, representing component failures of a system,
are not repaired but their probabilities of occurrence are known. Under such conditions, the probability evaluation of fault trees is demonstrated through Example 7.4.
Example 7.4
Assume that in Figure 7.2, the probability of occurrence of basic events B1, B2, B3,
B4, B5, B6, and B7 are 0.15, 0.15, 0.15, 0.15, 0.06, 0.08, and 0.04, respectively.
Calculate the probability of occurrence of the top event: room without light.
By inserting the given data into Equation (7.25), the probability of occurrence
of event B8 is
P (B8 ) = P (B1 ) P (B2 ) P (B3 ) P (B4 )
= (0.15) (0.15) (0.15) (0.15)
0.0005

Using the given data with Equation (7.22), we get the following probability of
occurrence of event B9:

1999 by CRC Press LLC

FIGURE 7.10 Calculated fault event probability values of Figure 7.2 fault tree.

P (B9 ) = P (B5 ) + P (B6 ) P (B5 ) P (B6 )

= 0.06 + 0.08 (0.06) (0.08)
= 0.1352
Substituting the above calculated values and the given data into Equation (7.21),
we get the following value for the probability of occurrence of event B10 (i.e., top
event):

][

][

P (B10 ) = 1 1 P (B7 ) 1 P (B8 ) 1 P (B9 )

= 1 [1 0.04] [1 0.0005] [1 0.1352]

= 0.1702
Thus, the probability of occurrence of the top event, room without light, is
0.1702. The Figure 7.2 fault tree is shown in Figure 7.10 with given and calculated
fault event probability values.

7.7.2

PROBABILITY EVALUATION OF FAULT TREES

WITH REPAIRABLE COMPONENTS

In this case, the basic fault events, say, representing component failures of a system,
are repaired and the failure and repair rates of the components are known. Thus,
using the Markov method, the unavailability of a component is given by

1999 by CRC Press LLC

+ t
1 e ( )
+

UA (t ) =

(7.26)

where
UA (t) is the component unavailability at time t.

is the component constant failure rate.

is the component constant repair rate.

For large t, Equation (7.26) reduces to
UA =

(7.27)

where
UA is the component steady state unavailability.
The above equation yields steady state probability of a component (that it will
be unavailable for service) when its failure and repair rates are known. Thus,
substituting Equation (7.27) into Equation (7.21), we get the steady state probability
of occurrence of the output fault event A0 of the OR gate as follows:
m

POS = 1

(1 UA )
i

i =1
m

= 1

i =1

(7.28)

i
1 +

i
i

where
POS is the steady state probability of occurrence of the OR gate output fault
event.
UAi is the steady state unavailability of component i; for i = 1, 2, 3, , m.
i is the constant failure rate of component i; for i = 1, 2, 3, , m.
i is the constant failure rate of component i; for i = 1, 2, 3, , m.
Similarly, inserting Equation (7.27) into Equation (7.25) yields the following result:
k

PaS =

UA

i =1

(7.29)

i =1

1999 by CRC Press LLC

+
i

where
PaS is the steady state probability of occurrence of the AND gate output fault
event.
The probability evaluation of fault trees with repairable components using the above
equations is demonstrated by solving the following example [13]:
Example 7.5
Assume that in Figure 7.2 failure and repair (in parentheses) rates associated with basic
events B1, B2, B3, B4, B5, B6, and B7 are 0.0002 failures/h (4 repairs/h), 0.0002 failures/h
(4 repairs/h), 0.0002 failures/h (4 repairs/h), 0.0002 failures/h (4 repairs/h), 0.0001
failures/h (0.004 repairs/h), 0.00015 failures/h (4 repairs/h), and 0.0001 failures/h
(0.004 repairs/h), respectively. Calculate the steady state probability of occurrence
of the top event: room without light.
Substituting the given data into Equation (7.27), we get the following steady
state probability of occurrence of events B1, B2, B3, and B4:
UA =

0.0002
0.0002 + 4

4.99 10 5

Similarly, inserting the specified data into Equation (7.27), the following steady
state probability of occurrence of events B5 and B7 is obtained:
UA =

0.0001
0.0001 + 0.004

= 0.0244
Finally, substituting the given data into Equation (7.27), we get the following
steady state probability of occurrence of event B6:
UA =

0.00015
0.00015 + 4

= 3.75 10 5
Using the UA calculated value for events B1, B2, B3, and B4 in Equation (7.29),
we get the following steady state probability of occurrence of event B8:

P (B8 ) = 4.99 10 5
= 6.25 10 18

1999 by CRC Press LLC

FIGURE 7.11 Calculated steady state probability values of basic repairable events and of
intermediate and top events for the fault tree of Figure 7.2.

Substituting the UA calculated values for events B5 and B6 into Equation (7.28),
we obtain the following steady state probability of occurrence of event B9:

P (B9 ) = 1 (1 0.0244) 1 3.75 10 5

0.0244

Inserting the UA calculated values for events B7, B8, and B9 into Equation (7.28),
the following steady state probability of occurrence of event B10 is obtained:

][

][

P (B10 ) = 1 1 P (B7 ) 1 P (B8 ) 1 P (B9 )

= 1 [1 0.0244] 1 6.25 10 18 [1 0.0244]

0.0482

Thus, the steady state probability of occurrence of the top event, room without
light, is 0.0482. The above calculated probability values are shown in the Figure 7.11
fault tree.

7.8 FAULT TREE DUALITY AND FTA APPROACH

BENEFITS AND DRAWBACKS
Once a fault tree is developed, it is possible to obtain its dual without much additional
effort. The dual of a fault tree is the success tree. Thus, in order to obtain a

1999 by CRC Press LLC

success tree from a fault tree, replace all OR gates with AND gates in the original
fault tree and vice versa, as well as replace all fault events with success events. For
example, the top fault event, room without light, becomes room lit.
Just like any other reliability analysis method, the FTA approach also has its
benefits and drawbacks. Some of the benefits of the FTA approach are as follows:

It identifies failures deductively.

It serves as a graphic aid for system management.
It provides insight into the system behavior.
It can handle complex systems more easily.
It allows concentration on one particular failure at a time and requires the
analyst to understand thoroughly the system under consideration prior to
the starting of FTA.
It provides options for management and others to conduct either qualitative or quantitative reliability analysis.
On the other hand, some of the drawbacks of the FTA approach include time
consuming, costly, end results difficult to check, and considers components in either
working or failed state (more specifically, the partial failure states of components
are difficult to handle).

7.9 PROBLEMS
1. What are the purposes of performing FTA?
2. Write an essay on the history of the fault tree method.
3. Define the following named symbols used in constructing fault trees:
AND gate
Diamond
Circle
Rectangle
4. Discuss the basic process followed in developing fault trees.
5. What are the advantages and disadvantages of the FTA method?
6. Compare the FTA approach with the block diagram method (i.e., network
reduction technique) used in reliability analysis.
7. Assume that a windowless room has one switch and one light bulb.
Develop a fault tree for the top or undesired event dark room. Assume
the switch only fails to close.
8. Obtain a success tree for the fault tree shown in Figure 7.2.
9. Prove the following Boolean expression:
X + YZ = ( X + Y) ( X + Z )

(7.30)

10. Assume that in Figure 7.2 the basic independent fault events failure and
repair rates are 0.009 failures/h and 0.04 repairs/h, respectively. Calculate
the steady state probability of occurrence of the top event, room without light.

1999 by CRC Press LLC

7.10

REFERENCES

1. Haasl, D.F., Advanced Concepts in Fault Tree Analysis, System Safety Symposium,
1965. Available from the University of Washington Library, Seattle.
2. Barlow, R.E., Fussell, J.B., and Singpurwalla, N.D., Eds., Reliability and Fault Tree
Analysis, Society for Industrial and Applied Mathematics (SIAM), Philadelphia,
1975. (Conference proceedings publication.)
3. Dhillon, B.S. and Singh, C., Bibliography of literature on fault trees, Microelectronics
and Reliability, 17, 501-503, 1978.
4. Fault Tree Handbook, Report No. NUREG-0492, U.S. Nuclear Regulatory Commission, Washington, D.C., 1981.
5. Henley, E.J. and Kumamoto, H., Reliability Engineering and Risk Assessment, Prentice-Hall, Englewood Cliffs, NJ, 1981.
6. Dhillon, B.S. and Singh, C., Engineering Reliability: New Techniques and Applications, John Wiley & Sons, New York, 1981.
7. Schroder, R.J., Fault tree for reliability analysis, Proc. Annu. Symp. Reliability, 170-174,
1970.
8. Risk Analysis Using the Fault Tree Technique, Flow Research Report, Flow Research,
Inc., Washington, D.C., 1973.
9. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Eds., Handbook of Reliability
Engineering and Management, McGraw-Hill, New York, 1996.
10. Barlow, R.E. and Proschan, F., Statistical Theory of Reliability and Life Testing, Holt,
Rinehart and Winston, New York, 1975.
11. Fussell, J.B. and Vesely, W.E., A new methodology for obtaining cut sets for fault
trees, Trans. Am. Nucl. Soc., 15, 262-263, 1972.
12. Semanderes, S.N., Elraft, a computer program for efficient logic reduction analysis
of fault trees, IEEE Trans. Nuclear Sci., 18, 310-315, 1971.
13. Dhillon, B.S., Proctor, C.L., and Kothari, A., On repairable component fault trees,
Proc. Annu. Reliability Maintainability Symp., 190-193, 1979.

1999 by CRC Press LLC

Common Cause Failures

and Three State Devices

8.1 INTRODUCTION
Over the years, common cause failures have been receiving increasing attention
because of the realization that the assumption of independent failures may be violated
in the real life environment. For example, according to Reference 1, in the U.S.
power reactor industry of 379 components failures or groups of failures arising
from independent causes, 78 involved common cause failures of two or more components. A common cause failure may simply be described as any instance where
multiple items fail due to a single cause. Thus, the occurrence of common cause
failures leads to lower system reliability.
The late 1960s and early 1970s may be regarded as the beginning of the serious
realization of common cause failure problems when a number of publications on
the subject appeared [2-6]. In 1975, a number of methods were proposed to evaluate
reliablity of systems with common cause failures [7, 8]. In 1978, an article presented
most of the publications on common cause failures [9]. The first book that gave an
extensive covereage to common cause failures appeared in 1981 [10]. In 1994, an
article entitled Common Cause Failures in Engineering Systems: A Review presented a comprehensive review of the subject [11]. Nonetheless, ever since the late
1960s many people have contributed to the subject of common cause failures and
over 350 contributions from these individuals can be found in Reference 11.
A device is said to have three states if it operates satisfactorily in its normal
mode but fails in either of two mutually exclusive modes. Two examples of a three
state device are an electronic diode and a fluid flow valve. The two failure modes
pertaining to both these devices are open, short and open, close, respectively. In
systems having these devices, the redundancy may increase or decrease the system
reliability. This depends on three factors: type of system configuration, number of
redundant components, and the dominant mode of component failure.
The history of reliaiblity studies of three state devices may be traced back to
the works of Moore and Shannon [12] and Creveling [13] in 1956. In 1975, the
subject of three state device systems was studied in depth and a delta-star method
to evaluate reliability of complex systems made up of three state devices was
developed [14]. In 1977, an article entitled Literature Survey on Three State Device
Reliability Systems provided a list of publications on the subject [15]. A comprehensive review on the subject appeared in 1992 [16]. Needless to say, many other
individuals have also contributed to the topic and most of their contributions are
listed in Reference 16.
This chapter presents the topics of common cause failures and three state devices,
separately.

1999 by CRC Press LLC

8.2 COMMON CAUSE FAILURES

There have been numerous instances of the occurrence of common cause failures in
industry, particularly in the nuclear power generation. Three examples of the occurrence of the common cause failures in the nuclear power generation are as follows [17]:
1. Two motorized valves were placed in a failed state due to a maintenance
error of wrongly disengaging the clutches.
2. Some spring loaded relays forming a parallel configuration failed simultaneously due to a common cause.
3. A steam line rupture caused mutliple circuit board failures. In this case,
the common cause was the steam line rupture.
In passing, it is to be noted that, in some cases, instead of triggering a total failure
of redundant systems (i.e., simultaneous failure), a common cause may lead to less
severe but common degradation of redundant units. Obviously, this will increase the
joint probability of the system units failure and the redundant unit may fail sometime
later than the first unit failure. Thus, the failure of the second unit is dependent and
coupled to the first unit failure because of the common morose environment.
There are many causes for the occurrence of common cause failures. Some of
them are as follows [10]:
Common external environment. This includes temperature, humidity,
vibrations, dust, and moisture.
Operations and maintenance errors. These are due to factors such as
carelessness, incorrect maintenance, and wrong adjustment or calibration
by humans.
Deficiency in equipment design. This is the result of oversights during
the design phase. For example, failure to consider interdependence
between electrical and mechanical parts.
External catastrophic events. These include natural phenomena such as
earthquake, fire, tornado, and flood. The occurrence of natural events such
as these may cause the simultaneous failure of redundant units.
Functional deficiency. This may result due to poorly designed protective
action or inappropriate instrumentation.
Common external power source. Redundant units fed from the same
power source (directly or indirectly) may fail simultaneously in the event
of the power source failure.
Common manufacturer. The redundant units purchased from the same
manufacturer may have the same fabrication, design, or other defects. Some
examples of the fabrication defects are wiring a circuit board backward,
use of incorrect material, and poor soldering. Thus, the manufacturing
defects such as these may cause the simultaneous failure of redundant units.
Three different methods to handle the occurrence of common cause failures in
reliability analysis of engineering systems are presented below.

1999 by CRC Press LLC

FIGURE 8.1 Block diagram of a parallel system with common cause failures. m is the
number of units and cc is the hypothetical unit representing common cause failures.

8.2.1

BLOCK DIAGRAM METHOD

This is a simple and straightforward approach to analyze redundant systems with

common cause failures. The basis for this approach is to represent all common cause
failures associated with a system as a single hypothetical unit and then connect it
in series with the block diagram of a redundant system representing independent
units or failures. From there on the reliability of the system is determined by using
this overall block diagram and applying the usual network reduction approach.
The method is demonstrated by applying it to parallel and k-out-of-m networks
as discussed below.
Parallel Network
Figure 8.1 shows block diagram of a parallel system with common cause failures.
This figure is composed of two blocks (i.e., blocks A and B) in series. Block A
represents the normal parallel system with independent failures and block B contains
a single hypothetical unit representing all the common cause failures associated with
the parallel system.
Assume that a unit failure rate, , is expressed by
= in + cc

(8.1)

where
in is the unit independent mode failure rate.
cc is the unit or system common cause failure rate.
Thus, the fraction of common cause type unit failures is given by
=

1999 by CRC Press LLC

cc

(8.2)

Therefore,
cc =

(8.3)

By inserting Equation (8.3) into Equation (8.1) and rearranging, we get

in = (1 )

(8.4)

In Figure 8.1 block A represents independent mode failures and block B represents
the common cause failures. The complete system can fail only if the hypothetical
unit representing all system common cause failures fails or all units of the parallel
system fail. The reliability of the independent modified parallel system shown in
Figure 8.1 is given by
R mp = R A R B

= 1 (1 R in )

]R

(8.5)
cc

where
Rmp is the reliability of the modified parallel system (i.e., the parallel system
with common cause failures).
RA is the reliability of block A.
RB is the reliability of block B.
Rin is the units independent failure mode relialibility.
Rcc is the system common cause failure mode reliaiblity or the reliability of
the hypothetical unit representing all system common cause failures.
m is the number of identical units in parallel.
For constant failure rates, in and cc, we get the following reliability equations,
respectively:
R in (t ) = e (1 ) t

(8.6)

R cc (t ) = e t

(8.7)

where t is time.
Substituting Equations (8.6) and (8.7) into Equation (8.5) yields

R mp (t ) = 1 1 e (1 ) t

1999 by CRC Press LLC

e t

(8.8)

1.0
0.9

=1
= 0.75
= 0.5
= 0.25
=0

0.8
0.7

Rmp(t)

0.6
m=2

0.5
0.4
0.3
0.2
0.1
0.0
0.0

0.5

1.0

1.5

2.0

FIGURE 8.2 Reliability plots of a two-unit parallel system with common cause failures.

It can be seen from Equations (8.2) and (8.8) that the parameter takes values from
0 to 1. Thus, at = 0 Equation (8.8) becomes simply a reliability expression for a
parallel system without common cause failures. More specifically, there are no
common cause failures associated with the parallel system. At = 1, it means all
the parallel system failures are of common cause type and the system simply acts
as a single unit with failure rate, .
Figures 8.2 and 8.3 show plots of Equation (8.8) for various specified values of
and for m = 2 and 3, respectively. In both the cases, the system reliability decreases
with the increasing values of .
The system mean time to failure is given by [18]:
MTTFmp =

R
0

mp

(t ) d t

(1)
i =1

i +1

m
{i (i 1) }
i

(8.9)

where
MTTFmp is the mean time to failure of the parallel system with common
cause failures.
m
= m ! i ! ( m i )!
i

1999 by CRC Press LLC

(8.10)

FIGURE 8.3 Reliability plots of a three-unit parallel system with common cause failures.

The system hazard rate is given by [18]:

mp (t ) =

d R mp (t )
1
R mp (t )
dt

)]

(8.11)

= + m (1 ) ( 1) m 1
where

mp (t) is the hazard rate of the parallel system with common cause failures.

= 1 1 e (1 ) t

(8.12)

k-out-of-m Network
Just like in the case of the parallel network, a hypothetical unit respresenting all
system common cause failures is connected in series with the independent and
identical unit k-out-of-m system. More specifically, this modified k-out-of m network
is same as the modified parallel network except that at least k units must function
normally instead of at least one unit for the parallel case. Thus, the k-out-of-m
system with common cause failures reliability is [10]

1999 by CRC Press LLC

 m m

mi
i
Rk m =
R in (1 R in ) R cc
i=k i

(8.13)

where
k is the required number of units for the k-out-of-m system success.
Rk/m is the reliability of the k-out-of-m system with common cause failures.
Inserting Equations (8.6) and (8.7) into Equation (8.13) we get
m

R k m (t ) =

i e

i (1 ) t

i=k

[1 e

i (1 ) t

mi

e t

(8.14)

In a manner similar to the parallel network case, the expressions for system mean
time to failure and hazard rate may be obtained. Also, the block diagram method
can be applied to other redundant configurations such as parallel-series and
bridge [18].
Example 8.1
Assume that a system is composed of two independent, active and identical units.
At least one unit must function normally for the sytem success. The value of the
is 0.5. Calculate the system reliability for a 100-h mission, if the unit failure rate is
0.004 failures/h.
Substituting the given data into Equation (8.8) yields

2
R mp (100) = 1 1 e (10.5) ( 0.004 ) (100 ) e ( 0.5) ( 0.004 ) (100 )

= 0.7918
Thus, the reliability of the two-unit parallel system with common cause failures
is 0.7918.
Example 8.2
If the value of is zero in Example 8.1, calculate the system reliability and compare
it with the system reliability result obtained in that example.
Using the given data in Equation (8.8) we get

R mp (100) = 1 1 e ( 0.004 ) (100 )

= 0.8913

1999 by CRC Press LLC

FIGURE 8.4 Fault tree for Figure 8.1 network.

Thus, in this case the system reliability without common cause failures is 0.8913.
In comparison to the system reliability result obtained in Example 8.1, i.e., 0.7918,
the occurrence of common cause failures leads to lower system reliability.

8.2.2

FAULT TREE METHOD

This is another method that can be used to perform reliability analysis of systems
with common cause failures. The application of this method is demonstrated by
applying it to the network shown in Figure 8.1. In this case, the top event is Network
Failure. The fault tree for the Figure 8.1 network is shown in Figure 8.4. The
probability of the top event occurrence, i.e., network failure, is given by
FN = FA + FB FA FB

(8.15)

where
FN is the network failure probability, i.e., failure propability of the parallel
system with common cause failures.
FA is the block A failure probability, i.e., failure probability of the parallel
system with independent failures.
FB is the block B failure probability, i.e., the failure probability of the hypothetical unit representing all system common cause failures.
In turn,
m

FA =

i =1

1999 by CRC Press LLC

(8.16)

FIGURE 8.5 Fault tree for Example 8.3.

where
Fi is the failure probability of independent unit i; for i = 1, 2, , m.
The network reliability (i.e., the reliability of the parallel system with common
cause failures) is given by
R mp = 1 FN

(8.17)

Example 8.3
Assume that a parallel system with common cause failures is composed of two units
and the system common cause failure occurrence probability is 0.2. In addition, each
units independent failure probability is 0.1. Calculate the failure probability of the
parallel system with common cause failures using the fault tree method.
For m = 2, the fault tree shown in Figure 8.4 will represent this example and
the specific fault tree is given in Figure 8.5. In this case, the top event is the failure
of the parallel system with common cause failures. The occurrence probability of
the top event can be calculated by using Equations (8.15) and (8.16).
Substituting the given data into Equation (8.16) yields
FA = F1 F2
= (0.1) (0.1)
= 0.01

1999 by CRC Press LLC

Using the calculated and the given data in Equation (8.15), we get
FN = (0.01) + 0.2 (0.01) (0.2)
= 0.2080
The given and the calculated probability values are shown in the Figure 8.5 fault
tree. The failure probability of the two-unit parallel system with common cause
failures is 0.2080.
Example 8.4
By using the given data of Example 8.3 and the block diagram method, prove that
the end result obtained in Example 8.3 is the same.
Thus, we have
R CC = 1 0.2
= 0.8
R in = 1 0.1
= 0.9
m=2
By inserting the above values into Equation (8.5) we get

R mp = 1 (1 0.9)

] (0.8)

= 0.7920
Using the above result in Equation (8.17) yields
FN = 0.2080
It proves that Example 8.3 solved through two different methods, fault tree and
block diagram, gives the same end result.

8.2.3

MARKOV METHOD

This is another method that can be used to perform reliability analysis of systems
with common cause failures. The systems can be either repairable or nonrepairable.
The following three models demonstrate the applicability of the method to redundant
systems with common cause failures:

1999 by CRC Press LLC

FIGURE 8.6 State space diagram of a two-unit parallel system with common cause failures.
The numbers and letters denote system state.

Model I
This model represents a two independent unit parallel system with common cause
failures [19]. The system can fail due to a common cause failure when both the units
are active. For the system success, at least one unit must operate normally. The
system transition diagram is shown in Figure 8.6.
The following assumptions are associated with this model:

Both units are identical and active.

Common cause and other failures are statistically independent.
Common cause and other failure rates are constant.
No failed unit is repaired.

The following notation is associated with this model:

i

is the ith system state: i = 0 (means both units working normally), i = 1

(means one unit failed, other operating normally), i = 2 (means both
units failed due to noncommon cause failures), i = cc (means both units
failed due to a common cause failure).
Pi (t) is the probability that the system is in state i at time t; for i = 0, 1, 2, cc.
cc is the system common cause failure rate.

is the unit failure rate.

The following equations are associated with Figure 8.6:

1999 by CRC Press LLC

d P0 (t )
+ (2 + cc ) P0 (t ) = 0
dt

(8.18)

d P1 (t )
+ P1 (t ) = 2 P0 (t )
dt

(8.19)

d P2 (t )
= P1 (t )
dt

(8.20)

d Pcc (t )
= cc P0 (t )
dt

(8.21)

At time t = 0, P0 (0) = 1 and P1 (0) = P2 (0) = Pcc (0) = 0.

Solving Equations (8.18) through (8.21) using Laplace transforms [10], we get
2 + cc ) t
P0 (t ) = e (

2 + cc ) t
P1 (t ) = X e t e (

(8.22)

(8.23)

where
X = 2 (2 + cc )

(8.24)

2 + cc ) t
P2 (t ) = Y + X Z e (
e t

(8.25)

where
Y = 2 (2 + cc )

(8.26)

Z = (2 + cc )

(8.27)

Pcc (t ) =

cc
2 + cc ) t
1 e (
2 + cc

(8.28)

The reliability of the parallel system with common cause failures is

R ps ( t ) = P0 (t ) + P1 (t )

2 + cc ) t
2
=e (
+ X e t e (

+ cc ) t

(8.29)

The mean time to failure of the parallel system with common cause failures is given
by

1999 by CRC Press LLC

MTTFps =

ps

(t ) d t
(8.30)

= 3 (2 + cc )
where
MTTFps is the mean time to failure of the two unit parallel system with
common cause failures.
The probability of the system failure due to a common cause failure is given by
Pcc (t ) =

cc
2 + cc ) t
1 e (
2 + cc

(8.31)

Example 8.5
Assume that a system has two identical units in parallel and it can fail due to the
occurrence of a common cause failure. The common cause failure occurrence rate
is 0.0005 failures/h. Calculate the following:
if the unit failure rate is 0.004 failures/h:
System mean time to failure
System mean time to failure, if the occurrence of common cause failures
is equal to zero.
Compare the above two results.
By substituting the given data into Equation (8.30), we get
MTTF = 3 [2 (0.004) + 0.0005]
= 353 h
For zero common cause failures and the other given data, Equation (8.30) yields
MTTF = 3 2 (0.004)
= 375 h
It means that the occurrence of common cause failures leads to the reduction in
the system mean time to failure (i.e., from 375 h to 353 h).

1999 by CRC Press LLC

FIGURE 8.7 The state transition diagram of the system without repair. CCF means common
cause failure.

Model II
This model represents a system composed of two active parallel units and one
standby subject to two types of common cause failures [20]. The type I common
cause failure leads to total system failure (active plus the standby units) whereas,
type II failures only cause the failure of the active two units. The system units can
also malfunction due to a normal failure.
The system transition diagram is shown in Figure 8.7. The system starts operating (two parallel units and one unit on standby) at time t = 0. As soon as one of
the operating units fails due to a normal failure, it is immediately replaced with the
standby unit. After this, the system may fail due to a common cause failure or one
of the operating units fails because of the normal failure. The single operating unit
may fail either due to a normal failure or a common cause failure.
In the state when two units operate and one is on standby, the system units may
also fail due to type I and type II common cause failures. In the case of the occurrence
of type II common cause failure, the standby unit is immediatly activated. For the
system success, at least one of these units must operate normally.
The following assumptions are associated with the analysis of the Figure 8.7
system transition diagram.
i.
ii.
iii.
iv.

All system units are identical and fail independently.

All failure rates associated with the system are constant.
The switching mechanism for the standby is instantaneous and perfect.
The system starts at time t = 0 with three good units (i.e., two in parallel
and one on standby).
v. The failed system is never repaired.
vi. The standby unit remains as good as new in its standby mode.

1999 by CRC Press LLC

The following symbols are associated with this model:

i

The state of the system shown in Figure 8.7: i = 0 (two units operating
in parallel and one on standby), i = 1 (one of the operating units fails
and the standby is activated), i = 2 (one unit operating, two units
failed), i = 3 (two active parallel units failed due to a type II common
cause failure and the standby unit is activated), i = 4 (system failed,
all three units failed due to a type I common cause failure), i = 5
(system failed due to normal failures), i = 6 (system failed due to a
common cause failure when operating with two or less good units).
Pi (t) The probability that the system is in state i at time t; for i = 0, 1, 2,
3, 4, 5, 6.

Failure rate of a unit.

Type II system common cause failure rate.

c
Type I system common cause failure rate.

Common cause failure rate of the system with one good unit.
R (t) System reliability at time t.
MTTF System mean time to failure.
The system of equations associated with Figure 8.7 is
d P0 (t )
+ (2 + c + ) P0 (t ) = 0
dt

(8.32)

d P1 (t )
+ (2 + ) P1 (t ) = 2 P0 (t )
dt

(8.33)

d P2 (t )
+ ( + ) P2 (t ) = 2 P1 (t )
dt

(8.34)

d P3 (t )
+ ( + ) P3 (t ) = P0 (t )
dt

(8.35)

d P4 (t )
= P0 (t ) c
dt

(8.36)

d P5 (t )
= P2 (t ) + P3 (t )
dt

(8.37)

d P6 (t )
= P1 (t ) + P2 (t ) + P3 (t )
dt

(8.38)

At time t = 0, P0 (0) = 1 and P1 (0) = P2 (0) = P3 (0) = P4 (0) = P5 (0) = P6 (0) = 0.

1999 by CRC Press LLC

Solving Equations (8.32) through (8.38), we get the following state probability
expressions:
P0 (t ) = e A t
P1 (t ) =

(8.39)

2 Bt
e e A t
c

BT

e Bt
e
e A T
P2 (t ) = 4 2
+
+

(B C) (A C) (C B) (A B) (C A) (B A)

P3 (t ) =

Ct
e e A t
D

(8.40)

(8.41)

(8.42)

where
A = 2 + c +
B = 2 +
C=+
D = + c +
P4 ( t ) =

P5 (t ) =

c
1 e A t
A

C e A t A eC t
1+

CA
D

1
e A t
eC t
e Bt
+ 4 3

ABC AD c CDE BE c

(8.43)

(8.44)

where
E = +
P6 (t ) =

1999 by CRC Press LLC

2 Be A t A e Bt
1 +

AB
c

(8.45)

Ce A t A e Ct
1+

A C
D

1
eA t
e Ct
e Bt
+ 4 2
+

ABC A D c CDE BE c
The system reliability is given by
3

R (t ) =

P (t )
i

i=0

= e A t +

2 Bt
e e A t
c

)
(8.46)

Bt

eC t
e
e At
+ 4 2
+
+

(B C) (A C) (C B) (A B) (C A) (B A)

B C t
e e A t
D

The system mean time to failure is

MTTF =

R (t ) d t
0

(8.47)

1 2 4
+
+
1+
A
B BC C
2

Model III
This model is the same as Model II except that the repair process is initiated as soon
as one of the two active units fails and is set equal to zero. Thus, the failed
unit/units in state 1, 2, and 3 of Figure 8.7 are repaired at a constant repair rate .
The additional symbols used in this model are as follows:
s
is the Laplace transform variable.
R(s) is the Laplace transform of the system reliability, R(t).
Pi (s) is the Laplace transform of Pi (t), for i = 0, 1, 2, , 6.
Thus, for this model, the Laplace transforms of the state probabilities are as
follows:

1999 by CRC Press LLC

P0 (s) =

s2 + s G + F
s3 + s2 J (H + GA) + I

(8.48)

where
F = 2 2 + + + 2 +
G = 3 + 2 +
H = 2 2 + + + 2

(8.49)

I = c F + 4 3 + 4 2 + 3 + 2 + 2 + 2
J = 5 + 2 + 2 + c
P1 (s) =

P0 (s) 2 (s + + )
s2 + s G + F

(8.50)

P0 (s) 4 2
s2 + s G + F

(8.51)

P0 (s)
s++

(8.52)

P2 (s) =

P3 (s) =

P4 (s) = P0 (s) c s

(8.53)

P5 (s) = P2 (s) + P3 (s)

(8.54)

P6 (s) =

2 (s + + ) P0 (s)
s s2 + s G + F

(8.55)

The Laplace transform of system reliability with repair is

3

R r (s) =

P (s)
i

(8.56)

i=0

The system mean time to failure with repair is

MTTFr = lim R r (s)
s 0

F
1
= F + 2 ( + ) + 4 2 +
I
+

1999 by CRC Press LLC

(8.57)

FIGURE 8.8 A three state device series system.

8.3 THREE STATE DEVICES

Three state devices may form configurations such as series, parallel, series-parallel,
and bridge. The reliability evaluation mechanism of these configurations is different
than the configurations made up of two state devices (a device is said to have two
states, if it either operates or fails). The ordinary reliability configurations fall under
the two state device category. Nonetheless, the three state device configurations are
more complex to analyze and redundancy may lead to a decrease in the overall
system reliability. This reduction in system reliability is due to factors such as
dominant mode of the device failure, number of redundant units, and the redundant
configuration type.
The reliability evaluation and optimization of series and parallel networks composed of three state devices are presented below.

8.3.1

RELIABILITY EVALUATION

OF A

SERIES NETWORK

The block diagram of a series network composed of m three state devices is shown
in Figure 8.8. Each block represents a three state device i that can operate normally
(Ni ) or fail either open (Oi ) or short (Si ) circuited. The system operates normally
when either all the devices function successfully or until at least one device is
working normally and the remaining ones have failed in their short circuit mode.
This is based on the assumption that the current can still flow through a short circuited
device. On the other hand, the system fails when either any one of the devices fails
in open mode or all the system devices are short circuited. A reliability equation for
this network can be developed by using the probability tree method [21]. A probability tree for a two component three state device network is given in Figure 8.9.
The dot on the extreme left side of Figure 8.9 signifies the point of origin of
paths and the dots on the extreme right side indicate the paths termination points.
The total number of different paths in a three state device probability tree is given
by 3m. The m denotes the number of units (i.e., three-state state devices) in a network.
The following symbols are used in the probability tree analysis:
Ni denotes the normal mode (i.e., success) of the ith unit (i.e., ith path links
success)
Oi denotes the open mode failure state of the ith unit (i.e., ith path links
failure)
Si denotes the short mode failure state of the ith unit (i.e., ith path links
failure)
Pi probability of being in State Ni

1999 by CRC Press LLC

FIGURE 8.9 Probability tree diagram for two non-identical unit three state device network.

qOi probability of being in State Oi

qSi probability of being in State Si
From Figure 8.9, the two non-identical unit series system success paths are: N1N2,
N1S2, and S1N2. With the aid of these paths and the relationships P1 + qO1 + qS1 = 1
and P2 + qO2 + qS2 = 1, the independent series system reliability is
R S2 = P1 P2 + P1 q S2 + P2 q S1
= (1 q O1 ) (1 q O 2 ) q S1 q S2

(8.58)

Similarly, the three non-identical unit parallel system success paths are as follows:
N1 N2 N3, N1 N2 S3, N1 S2 N3, N1 S2 S3, S1 N2 N3, S1 N2 S3, S1 S2 N3
Thus, the reliability of the three independent non-identical unit series system is
given by
R S3 = P1 P2 P3 + P1SP2 q

+ P1 q S2 P3 + P1 q S2 q S3 + q S1 P2 P3

+ q S1 P2 q S3 + q S1 q S2 P3

(8.59)

= (1 q O1 ) (1 q O 2 ) (1 q O3 ) q S1 q S2 q S3
Thus, for an m unit or device series system, we can generalize Equation (8.59) to get
m

R Sm =

i =1

1999 by CRC Press LLC

(1 q Oi )

q
i =1

Si

(8.60)

FIGURE 8.10 State space diagram of a three state device. N, O, and S in rectangle, diamond,
and circle mean device operating normally, device failed in open mode, and device failed in
short mode, respectively.

For identical units or devices, Equation (8.60) simplifies to

R Sm = (1 q O ) q Sm
m

(8.61)

where
qo is the open mode failure probability of the device.
qS is the short mode failure probability of the device.
For constant open and short mode failure rates, we use the Markov method to
obtain the open and short failure probabilities of a three state device. Figure 8.10
presents the state space diagram of a three state device.
The following symbols are associated with the Figure 8.10 model:
i

is the ith state of the three state device: i = N (means the three state
device is operating normally), i = O (means the three state device failed
in open mode), i = S (means the three state device failed in short mode).
Pi (t) is the probability that the three state device is in state i at time t; for
i = N, O, S.
O is the constant open mode failure rate of the three state device.
S is the constant short mode failure rate of the three state device.
The following system of equations is associated with Figure 8.10.
d PN (t )
+ ( O + S ) PN (t ) = 0
dt

(8.62)

d PO (t )
O PN (t ) = 0
dt

(8.63)

d PS (t )
S PN (t ) = 0
dt

(8.64)

At time t = 0, PN (0) = 1 and PO (0) = PS (0) = 0.

1999 by CRC Press LLC

By solving Equations (8.62) through (8.64) we get

+ t
PN (t ) = e ( O S )

PO (t ) = q O (t ) =

(8.65)

O
+ t
1 e ( O S)
O + s

(8.66)

Substituting Equations (8.66) and (8.67) into Equation (8.60) yields

R Sm (t ) =

1
i =1

i =1

Oi
+ t
1 e ( Oi Si )
Oi + Si

Si
+ t
1 e ( Oi Si )
+

Oi
Si

PS (t ) = q S (t ) =

(8.68)

S
+ t
1 e ( O S)
O + s

(8.67)

where
RSm (t) is the three state device series system reliability at time t.
For identical devices, Equation (8.68) becomes

O
+ t
R Sm (t ) = 1
1 e ( O S)
O + S

S
+ t

1 e ( O S)
O + S

(8.69)

Example 8.6
A system is composed of three independent and identical three state devices in series.
Each devices open mode and short mode failure probabilities are 0.05 and 0.03,
respectively. Calculate the system reliability.
Substituting the given data into Equation (8.61) yields
R S3 = (1 0.05) (0.03)
3

= 0.8573
Thus, the series system reliability is 0.8573.

1999 by CRC Press LLC

FIGURE 8.11 Block diagram of a three state device parallel network.

Example 8.7
A three state device network has two independent and identical units in series. Each
units open mode and short mode failure rates are 0.002 and 0.004 failures/h,
respectively. Calculate the network reliability for a 100-h mission.
Substituting the given data into Equation (8.69) yields

0.002
R S2 (100) = 1
e 1
0.002 + 0.004

( 0.002 + 0.004 ) (100 )

0.004

1 e ( 0.002 +0.004 ) (100 )
0.002 + 0.004

= 0.6314
Thus, the network reliability for a 100-h mission is 0.6314.

8.3.2

RELIABILITY EVALUATION

OF A

PARALLEL NETWORK

Figure 8.11 shows the block diagram of a parallel network having m three state
devices. Each block denotes a three state device i that can operate normally (Ni ) or
fail either open (Oi ) or short (Si ) circuited. The system functions normally when
either all the devices work successfully or until at least one device operates normally
and the remaining ones have failed in their open mode. In contrast, the system fails
when all the devices fail in their open mode or any one of them fails in short mode.
As for the series system, the Figure 8.9 probability tree can be used to develop
a reliability expression for the two unit parallel system. Thus, from Figure 8.9 the
two non-identical device parallel system success paths are as follows:

1999 by CRC Press LLC

 N1 N2
N1 O2
O1 N2
With the aid of the above paths and the relationships P1 +qO1 + qS1 = 1 and P2 +qO2 +
qS2 = 1, the independent parallel network reliability is
R p 2 = P1 P2 + P1 q O 2 + P2 q O1
(8.70)

= (1 q S1 ) (1 q S2 ) q O1 q O 2

Similarly, the three non-identical unit parallel system success paths are as follows:
N1 N 2 N 3 , N1 N 2 O3 , N1 O2 N 3 , N1 O2 O3 , O1 N 2 N 3 , O1 N 2 O3 , O1 O2 N 3
With the aid of the above paths, the reliability of the three independent nonidentical unit parallel system is given by
R p 3 = P1 P2 P3 + P1OP2 q

+ P1 q O 2 P3 + P1 q O 2 q O3 + q O1 P2 P3

+ q O1 P2 q O3 + q O1 q O 2 P3

(8.71)

= (1 q S1 ) (1 q S2 ) (1 q S3 ) q O1 q O 2 q O3
Thus, for an m unit or device parallel system, we can generalize Equation (8.71) as
follows:
m

(1 q ) q

R pm =

Si

i =1

(8.72)

Oi

i =1

For identical units or devices, Equation (8.72) simplifies to

R pm = (1 q S ) q Om
m

(8.73)

For a devices constant open and short mode failure rates, O and S, respectively,
we substitute Equations (8.66) and (8.67) into Equation (8.72) to get

R pm (t ) =

1
i =1

i =1

1999 by CRC Press LLC

Si
+ t
1 e ( Oi Si )

+
Oi
Si

Oi
+ t
1 e ( Oi Si )
Oi + Si

(8.74)

where
Rpm (t) is the three state device parallel system reliability at time t.
For identical devices, Equation (8.74) simplifies to

S
+ t
R pm (t ) = 1
1 e ( O S)

+
O
S

S
+ t

1 e ( O S)
O + S

(8.75)

Example 8.8
Assume that in Example 8.6 the system is parallel instead of series and the other
data are exactly the same. Calculate the parallel system reliability.
Inserting the given values into Equation (8.73) we get
R p3 = (1 0.03) (0.05)

= 0.9125
Thus, the parallel system reliability is 0.9125.
Example 8.9
Assume that in Example 8.7 the three state device network is parallel instead of
series and the other given data are exactly the same. Determine the parallel network
reliability for a 100-h mission.
Thus, substituting the specified values into Equation (8.75) yields

0.004
R p2 (100) = 1
e 1
0.002 + 0.004

( 0.002 + 0.004 ) (100 )

0.002

1 e ( 0.002 + 0.004 ) (100 )
+
.
.
0
002
0
004

= 0.4663
Thus, the parallel network reliability for a 100-h mission is 0.4663.

8.3.3

RELIABILITY OPTIMIZATION

OF A

SERIES NETWORK

In this case for independent and identical devices, the series network reliability from
Equation (8.61) is [10]

1999 by CRC Press LLC

R Sm = (1 q O ) q sm
m

(8.76)

In order to obtain the optimum number of three state devices to be connected in

series, we take partial derivatives of Equations (8.76) with respect to m and equate
it to zero. Thus, we have
m
R Sm
= (1 q O ) l n (1 q O ) q Sm l n q S = 0
m

(8.77)

Solving Equation (8.77), we get

m* =

l n l n q s l n (1 q O )

l n (1 q o ) q S

(8.78)

where
m* is the optimum number of devices in series.
Example 8.10
Assume that open and short mode failure occurrence probabilities of independent
and identical three state devices are 0.02 and 0.05, respectively. Determine the
number of devices to be arranged in series to maximize system reliability.
Thus, substituting the given data into Equation (8.78), we get
m* =

l n [l n (0.05) l n (1 0.02)]
l n [(1 0.02) (0.05)]

2 devices

It means two devices should be connected in series to maximize system reliability.

8.3.4

RELIABILITY OPTIMIZATION

OF A

PARALLEL NETWORK

From Equation (8.73), the parallel network, with independent and identical devices,
reliability is
R pm = (1 q S ) q Om
m

(8.79)

In order to determine the optimum number of three state devices to be arranged in

parallel, we take partial derivatives of Equation (8.79) with respect to m and equate
it to zero. Thus, we get

1999 by CRC Press LLC

 R pm
m

= (1 q S ) l n (1 q S ) q Om l n q O = 0
m

(8.80)

Solving Equation (8.80) yields

m* =

l n l n q O l n (1 q S )

l n (1 q S ) q O

(8.81)

where
m* is the optimum number of devices in parallel.
Example 8.11
After examining the failure data of one type of identical three state devices, it is
estimated that their open and short mode failure probabilities are 0.08 and 0.04,
respectively. Calculate the number of independent devices to be placed in a parallel
configuration to maximize system reliability.
Inserting the given data into Equation (8.81) yields

m* =

l n [l n (0.08) l n (1 0.04)]

l n [(1 0.04) 0.08]

devices

Thus, for maximum system reliability two three state devices should be arranged
in parallel.

8.4 PROBLEMS
1. Define the following three terms:
Three state device
Common-cause failure
Short mode failure
2. Discuss at least five causes for the occurrence of common cause failures.
3. Describe the methods that can be used to perform reliability analysis of
redundant networks with common cause failures.
4. Using Equation (8.14), obtain an expression for mean time to failure of
a k-out-of-m network with common cause failures.
5. Assume that in Figure 8.6 the failed unit is reparable from the system
state 1 at a constant repair rate . Develop expressions for system reliability and mean time to failure.

1999 by CRC Press LLC

6. A system is composed of two identical units in parallel and it can malfunction due to the occurrence of a common cause failure. The common
cause failure rate is 0.0001 failures/h. Calculate the system reliability and
common cause failure probability for a 200-h mission, if the unit failure
rate is 0.001 failures/h.
7. An electronic device can fail either in its open mode or short mode or it
simply operates normally. The open and short mode failure rates of the
device are 0.007 and 0.005 failures/h, respectively. Calculate the probability of the device failing in its short mode during a 1000-h operational
period.
8. A series system is composed of two independent and identical three state
devices. The open and short mode failure rates of the device are 0.0003
and 0.007 failures/h, respectively. Calculate the system reliability for a
50-h mission period.
9. A parallel system is composed of three independent and non-identical
three state devices. The open mode failure probabilities of devices A, B,
and C are 0.02, 0.04, and 0.07, respectively. Similarly, the short mode
failure probabilities of devices A, B, C are 0.03, 0.05, and 0.01, respectively. Calculate the open and short mode failure probabilities of the
parallel system.
10. Assume that open and short mode failure occurrence probabilities of
independent and identical three state devices are 0.12 and 0.10, respectively. Determine the number of devices to be connected in parallel to
maximize system reliability.

8.5 REFERENCES
1. Taylor, J.R., A study of failure causes based on U.S. power reactor abnormal occurrence reports, Reliab. Nucl. Power Plants, IAEA-SM-195/16, 1975.
2. Epler, E.P., Common-mode considerations in the design of systems for protection
and control, Nuclear Safety, 11, 323-327, 1969.
3. Ditto, S.J., Failures of systems designed for high reliability, Nuclear Safety, 8, 35-37,
1966.
4. Epler, E.P., The ORR emergency cooling failures, Nuclear Safety, 11, 323-327, 1970.
5. Jacobs, I.M., The common-mode failure study discipline, IEEE Trans. Nuclear Sci.,
17, 594-598, 1970.
6. Gangloff, W.C., Common-mode failure analysis is in, Electronic World, October,
30-33, 1972.
7. Gangloff, W.C., Common mode failure analysis, IEEE Trans. Power Apparatus Syst.,
94, 27-30, 1975.
8. Fleming, K.N., A redundant model for common mode failures in redundant safety
systems, Proc. Sixth Pittsburgh Annu. Modeling Simulation Conf., 579-581, 1975.
9. Dhillon, B.S., On common-cause failures Bibliography, Microelectronics and
Reliability, 18, 533-534, 1978.
10. Dhillon, B.S. and Singh, C., Engineering Reliability: New Techniques and Applications, John Wiley & Sons, New York, 1981.

1999 by CRC Press LLC

11. Dhillon, B.S. and Anude, D.C., Common-cause failures in engineering systems: a
review, Int. J. Reliability, Quality, Safety Eng., 1, 103-129, 1994.
12. Moore, E.F. and Shannon, C.E., Reliable circuits using less reliable relays, J. Franklin
Inst., 9, 191-208, 1956, and 281-297, 1956.
13. Creveling, C.J., Increasing the reliability of electronic equipment by the use of
redundant circuits, Proc. Inst. Radio Eng. (IRE), 44, 509-515, 1956.
14. Dhillon, B.S., The Analysis of the Reliability of Multi-State Device Networks, Ph.D.
Dissertation, 1975, available from the National Library of Canada, Ottawa.
15. Dhillon, B.S., Literature survey on three-state device reliability systems, 16, 601-602,
1977.
16. Lesanovsky, A., Systems with two dual failure modes A survey, Microelectronics
and Reliability, 33, 1597-1626, 1993.
17. WASH 1400 (NUREG-75/014), Reactor Safety Study, U.S. Nuclear Regulatory Commission, Washington, D.C., October 1975. Available from the National Technical
Information Service, Springfield, VA.
18. Dhillon, B.S. and Proctor, C.L., Common mode failure analysis of reliability networks, Proc. Annu. Reliability Maintainability Symp., 404-408, 1977.
19. Dhillon, B.S., Reliability in Computer System Design, Ablex Publishing, Norwood,
NJ, 1987.
20. Dhillon, B.S. and Yang, N., Analysis of an engineering system with two types of
common cause failures, Proc. First Intl. Conf. Quality Reliability, Hong Kong, 2,
393-397, 1995.
21. Dhillon, B.S. and Rayapati, S.N., Reliability evaluation of multi-state device networks
with probability trees, Proc. 6th Symp. Reliability Electron., Budapest, 27-37, 1985.

1999 by CRC Press LLC

Mechanical Reliability

9.1 INTRODUCTION
Usually, the reliability of electronic parts is predicted on the assumption that their
failure times are exponentially distributed (i.e., their failure rates are constant). The
past experience also generally supports this assumption. Nonetheless, this assumption was derived from the bathtub hazard rate concept which states that during the
useful life of many engineering items, the failure rate remains constant. However,
in the case of mechanical items, the assumption of constant failure rate is not
generally true. In fact, in many instances their increasing failure rate patterns can
be represented by an exponential function.
The history of mechanical reliability may be traced back to World War II with
the development of V1 and V2 rockets by the Germans. In 1951, Weibull published
a statistical function to represent material strength and life length [1]. Today it is
known as the Weibull distribution and it has played an important role in the development of mechanical reliability and reliability in general. Also, in the 1950s
Freudenthal reported many advances to structural reliability [2-4].
In 1963 and 1964, the National Aeronautics and Space Administration (NASA)
lost SYNCOM I and Mariner III due to busting of a high pressure tank and mechanical failure, respectively. As a result of these and other mechanical failures, researchers in the field called for the improvement in reliability and longevity of mechanical
and electromechanical parts. In years to come, NASA spent millions of dollars to
test, replace, and redesign various items, including mechanical valves, filters, actuators, pressure gauges, and pressure switches [5].
Some of the major projects initiated by NASA in 1965 were entitled Designing
Specified Reliability Levels into Mechanical Components with Time-Dependent
Stress and Strength Distributions, Reliability Demonstration Using Overstress
Testing, and Reliability of Structures and Components Subjected to Random
Dynamic Loading [6]. After the mid-1960s, many other people [7-9] contributed
to mechanical reliability and a comprehensive list of publications on the subject is
given in Reference 10.
This chapter presents various different aspects of mechanical reliability.

9.2 REASONS FOR THE DISCIPLINE OF MECHANICAL

RELIABILITY AND MECHANICAL FAILURE MODES
There have been many reasons for the development of the mechanical reliability
field. Five such reasons are presented in Figure 9.1 [11]. They are: influence from
electronic reliability, stringent requirements and severe application environments,
lack of design experience, cost and time constraints, and optimization of resources.

1999 by CRC Press LLC

FIGURE 9.1 Basic reasons for the development of the mechanical reliability field.

Influence from electronic reliability for the stimulation of similar developments

in mechanical engineering is an important factor because of vast advances made in
predicting and analyzing reliability of electronic equipment. Stringent requirements
and severe application environments are other important reasons because of many
expensive systems to be used in military and space applications. Cost and time
constraints make it difficult for designers to learn from past mistakes. Lack of design
experience occurs because of rapid changes in technology; consequently, mechanical
designers do not have the time to master the design, in particular design of complex
equipment for use in aerospace or military environment. Optimization of resources
is another good reason as the workable design is no longer considered satisfactory.
Thus, a given design has to be optimized under such constraints as reliability,
performance, weight, size, and cost.
There are many different types of failure modes associated with mechanical
items. Good design practices help to reduce or eliminate the occurrence of these
failure modes. Nonetheless, these failure modes include [12-14]:
Bending failure. This may also be called a combined failure because it
occurs when one outer surface is in tension and the other outer surface is
in compression. The tensile rupture of the outer material is a representative
of the bending failure.
Metallurgical failure. This may also be called a material failure and is
the result of extreme oxidation or operation in corrosive environment. The
environmental conditions such as heat, nuclear radiation, erosion, and
corrosive media accelerate the occurrence of metallurgical failures.

1999 by CRC Press LLC

 Tensile-yield-strength failure. This failure occurs under pure tension and,

more specifically, when the applied stress exceeds the yield strength of
the material. Usually, its consequence is permanent set or permanent
deformation in the structure and seldom catastrophic. The recommended
value of safety factors for this kind of failure could be as low as 1 (i.e.,
design to yield) or a high of 4 or 5 for pressure vessels and bridges.
Instability failure. This type of failure occurs in structural members such
as columns and beams, particularly the ones manufactured using thin
material where the loading is usually in compression. However, instability
failure may also result because of torsion or by combined loading, i.e.,
including bending and compression. Usually this type of failure is crippling or leads to a total failure of the structure. The past experience
indicates that a safety factor value of 2 is usually sufficient but avoid using
less than 1.5.
Ultimate tensile-strength failure. This leads to a complete failure of the
structure at a cross-sectional point and it occurs (i.e., ultimate tensilestrength failure) when the applied stress is greater than the ultimate tensile
strength. The recommended values of safety factors for this kind of failure
could be as low as unity (i.e., design to fail) or a high of 5 or 6 for pressure
vessels and bridges.
Material flaw failure. This type of failure usually occurs due to factors
such as poor quality assurance, fatigue cracks, weld defects, and small
cracks and flaws. These factors reduce the allowable material strength
and, as a consequence, lead to premature failure at the flawed spots.
Compressive failure. This failure is similar to the tensile failures discussed earlier but with one exception, i.e., under compressive loads. The
compressive failure causes permanent deformation/cracking/rupturing.
Fatigue failure. This type of failure occurs because of repeated loading
or unloading (or partial unloading) of an item. In order to prevent the
occurrence of this kind of failure, make sure the proper materials are
selected, e.g., steel outlasts aluminum under cyclic loading.
Bearing failure. This type of failure is similar in nature to the compressive
failure and it usually occurs due to a round cylindrical surface bearing on
either a flat or a concave surface like roller bearings in a race. If repeated
loading occurs, it is important to consider the possibility of the occurrence
of fatigue failures.
Stress concentration failure. This type of failure occurs in circumstances
of uneven stress flow through a mechanical design. Usually, the concentration of stress occurs at sudden transitions from thick gauges to thin
gauges, at sudden variations in loading along a structure, at various attachment conditions, or at right-angle joints.
Shear loading failure. The occurrence of ultimate and yield failure takes
place when the shear stress becomes greater than the strength of the
material when applying high torsion or shear loads. Normally, these failures occur along a 45 axis in relation to the principal axis.

1999 by CRC Press LLC

FIGURE 9.2 Seven possible causes of failure.

Creep/rupture failure. Normally, long-term loads make elastic materials

stretch even when they are less than the normal yield strength of the
material. Nonetheless, material stretches (creeps) if the load is maintained
continuously and usually it ultimately terminates in a rupture. It is to be
noted that at elevated temperatures, creep accelerates.

9.3 GENERAL AND GEAR FAILURE CAUSES

There are many causes of failure. Seven of them are shown in Figure 9.2 [15].
A study conducted over a period of 35 years reported 931 gear failures and their
major classifications were as follows [15]:

Breakage: 61.2%
Surface fatigue: 20.3%
Wear: 13.2%
Plastic flow: 5.3%

Furthermore, the failure causes were grouped into five distinct categories: (1) service
related (74.7%), (2) heat-treatment related (16.2%), (3) design related (6.9%),
(4) manufacturing related (1.4%), and (5) material related (0.8%). In turn, each of
these five categories were further broken into elements as follows:
Service-Related (74.7%)

Continual overloading: 25%

Improper assembly: 21.2%
Impact loading: 13.9%
Incorrect lubrication: 11%
Foreign material: 1.4%
Abusive handling: 1.2%

1999 by CRC Press LLC

 Bearing failure: 0.7%

Operator errors: 0.3%
Heat-Treatment Related (16.2%)

Incorrect hardening: 5.9%

Inadequate case depth: 4.8%
Inadequate core hardness: 2%
Excessive case depth: 1.8%
Improper tempering: 1%
Excessive core hardness: 0.5%
Distortion: 0.2%

Design-Related (6.9%)
Wrong design: 2.8%
Specification of suitable heat treatment: 2.5%
Incorrect material selection: 1.6%
Manufacturing-Related (1.4%)
Grinding burns: 0.7%
Tool marks or notches: 0.7%
Material-Related (0.8%)
Steel defects: 0.5%
Mixed steel or incorrect composition: 0.2%
Forging defects: 0.1%

9.4 SAFETY FACTOR AND SAFETY MARGIN

Safety factor and safety margin are used to ensure the reliability of mechanical items
and, basically, are arbitrary multipliers. These indices can provide satisfactory design
if they are established with care using considerable past experiences. Nonetheless,
in todays environment as new designs involve new applications and new materials,
more consistent approaches are required. Furthermore, the mechanical design totally
based upon safety factors could be misleading and may be costly due to over-design
or catastrophic because of under-design. Both safety factor and safety margin are
described below.

9.4.1

SAFETY FACTOR

There are many different ways of defining a safety factor [9, 10, 1619]. Two different definitions of the safety factor are presented below.

1999 by CRC Press LLC

Definition I
The safety factor is defined by [20]:
SF =

MFGST
1
MFGS

(9.1)

where
SF
is the safety factor.
MFGST is the mean failure governing strength.
MFGS is the mean failure governing stress.
This safety factor is a good measure when both the stress and strength are
described by a normal distribution. However, it is important to note that when the
variation of stress and/or strength is large, this measure of safety becomes meaningless because of positive failure rate.
Definition II
The safety factor is defined by [12, 13]:
SF =

SM
MAWS

(9.2)

where
SM
is the strength of the material.
MAWS is the maximum allowable working stress.
The value of this safety factor is always greater than unity. In fact, its value less
than unity indicates that the item under consideration will fail because the applied
stress is greater than the strength.
The standard deviation of the safety factor is expressed by [12, 13]:

= ( th MAWS) + SM (MAWS) s2
2

12

(9.3)

where
is the standard deviation of the safety factor.
th is the standard deviation of the strength.
s is the standard deviation of the stress.
All in all, there are many factors that must be carefully considered in selecting an
appropriate value of the safety factor: uncertainty of load, cost, failure consequence,

1999 by CRC Press LLC

uncertainty of material strength, degree of uncertainty involved in relating applied

stress to strength, etc. [21].

9.4.2

SAFETY MARGIN

This is defined as follows [9]:

m = SF 1

(9.4)

where
m is the safety margin.
SF is the safety factor.
The value of this measure is always greater than zero and its negative value indicates
that the item under consideration will fail. For normally distributed stress and
strength, the safety margin is expressed by [9]:
m = ( th ms ) th

(9.5)

where
th is the average strength.
ms is the maximum stress.
th is the standard deviation of the strength.
The maximum stress, ms , is given by
ms = s + k s

(9.6)

where
s is the mean value of the stress.
s is the standard deviation of the stress.
k is a factor which takes values between 3 and 6.
It is to be noted that just like the safety factor, the safety margin is a random variable.
Example 9.1
Assume that the following data are known for a mechanical item under design:
th = 30,000 psi,
th = 1200 psi,

s = 15,000 psi
s = 300 psi

For k = 3 and 6, calculate the values of the safety margin.

1999 by CRC Press LLC

Substituting the above data into Equation (9.6) yields

For k = 3
ms = 15, 000 + 3 (300)
= 15, 900 psi
For k = 6
ms = 15, 000 + 6 (300)
= 16, 800 psi
By inserting the above calculated values and the given data into Equation (9.5),
we get
For k = 3
m =

30, 000 15, 900

1200

= 11.75
For k = 6
m =

30, 000 16, 800

= 11
1200

Thus, the values of the safety margin for k = 3 and 6 are 11.75 and 11, respectively.

9.5 DESIGN BY RELIABILITY METHODOLOGY

AND STRESS-STRENGTH MODELS
In order to design a reliable mechanical item, a systematic series of steps are required.
The design by reliability methodology is made up of such steps [8, 22]:
Define the design problem to be solved.
Highlight and list all the relevant design variables and parameters.
Conduct failure modes, effect and criticality analysis (FMECA) as per
MIL-STD-1629 [23].
Verify selection of critical design parameters.
Develop relationship between the critical parameters and the failure governing criteria.

1999 by CRC Press LLC

 Determine the failure-governing stress and strength functions and then

the failure-governing stress and strength distributions.
Calculate the reliability using the failure-governing stress and strength
distributions for each critical failure mode.
Iterate the design until the set reliability goals are satisfied.
Optimize design with respect to factors such as reliability, maintainability,
cost, safety, performance, volume, and weight.
Repeat the design optimization for all associated critical parts.
Calculate item reliability.
Iterate the design until the reliability goals of the item are effectively
satisfied.
Before we discuss stress-strength models, let us define stress and strength:
Stress. This is any factor that tends to produce a failure. Some examples
are mechanical load, electric current, environment, and temperature.
Strength. This is the ability of an item to accomplish its identified goal
adequately without a failure when subject to the external loading and
environment.
When the probability distributions of both stress and strength of an item are known,
the items reliability may be determined through analytical means. Under such
circumstances, the reliability of an item may be defined as the probability that the
failure governing stress will not exceed the failure governing strength. Mathematically, it is expressed as follows:
R = P (x > y) = P ( y < x) = P (x y > 0)

(9.7)

where
R
P
y
x

is
is
is
is

the
the
the
the

item reliability.
probability.
stress random variable.
strength random variable.

Using Reference 10, we rewrite Equation (9.7) as follows:

R=

f ( y) f (x) d x d y
y

where
f (y) is the probability density function of the stress.
f (x) is the probability density function of the strength.

1999 by CRC Press LLC

(9.8)

Equation (9.8) can also be written in different forms and three such forms are as
follows:

R=

f (x) f ( y) d y d x

R=

f (y) 1 f (x) d x d y

R=

f (x) 1 f (y) d y d x

(9.9)

(9.10)

(9.11)

It is to be noted that Equations (9.10) and (9.11) were written using the following
two relationships in Equations (9.8) and (9.9), respectively:

f (x) d x + f (x) d x = 1

f ( y) d y + f ( y) d y = 1

(9.12)

(9.13)

Using the above reliability equations, the following stress-strength models were
developed for defined stress and strength probability density functions of an item:

MODEL I
In this case, stress and strength associated with an item are exponentially distributed.
Thus, we have
f ( y) = e y ,

0y<

(9.14)

f (x) = e x ,

0x<

(9.15)

and

where
and are the reciprocals of the mean values of stress and strength,
respectively.
Inserting Equations (9.14) and (9.15) into Equation (9.8) yields

1999 by CRC Press LLC

R=

e y e x d y
y

+ y
e ( ) dy

(9.16)

=
For = 1 and =
y

1
x

, Equation (9.16) becomes

R=

x
x+y

(9.17)

where
x and y are the mean strength and stress, respectively.
Dividing the top and bottom of the right-hand side of Equation (9.17) by x, we get
R=

1
1+ y x

(9.18)

1
=
1+
where
= y/x.
For reliable design x y, thus 1. For given values of , using Equation (9.18),
we obtained the tabulated values for item reliability as shown in Table 9.1.
Table 9.1 values indicate that as the value of increases from 0 to 1, the item
reliability decreases from 1 to 0.5. More specifically, as the mean stress increases,
the item reliability decreases, accordingly.
Example 9.2
Assume that stress and strength associated with an item are exponentially distributed
with mean values of 10,000 and 35,000 psi, respectively. Compute the item reliability.
Substituting the given data into Equation (9.17) yields
R=

35, 000
35, 000 + 10, 000

= 0.7778
Thus, the item reliability is 0.7778.

1999 by CRC Press LLC

TABLE 9.1
Values for Item Reliability

Item reliability, R

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1

1
0.9091
0.8333
0.7692
0.7143
0.6667
0.6250
0.5882
0.5556
0.5263
0.5

MODEL II
In this case, an items stress and strength are described by normal and exponential
probability density functions, respectively, as
f ( y) =

1 y 2
1
exp

2
2
f (x) = e x ,

< y <

x0

(9.19)

(9.20)

where
is the mean stress.
is the standard deviation associated with stress.
is the reciprocal of the mean strength.
Substituting Equations (6.19) and (6.20) into Equation (9.8), we get

R=

1999 by CRC Press LLC

1
1 y 2
exp

2
2

e x d x dy

1 y

1
exp
+ y dy
2
2

2

(9.21)

since

( y )2 + y = 2 2 2 4 + ( y + 2 )
2 2

2 2

(9.22)

Equation (6.21) yields

1
R = exp 2 2 2

(9.23)

Example 9.3
Assume that the strength of an item follows an exponential distribution with a mean
of 35,000 psi. The stress acting upon the item is normally distributed with mean
10,000 psi and standard deviation 5,000 psi. Calculate the items reliability.
By inserting the given data into Equation (9.23), we get
1 2 (10, 000) (5, 000)2
R = exp

(35, 000)2
2 35, 000
1
= exp {0.5714 0.0204}

2
= 0.7592
Thus, the items reliability is 0.7592.

MODEL III
In this case, the strength of the item is normally distributed and the stress acting
upon it follows the exponential distribution. Thus, we have
1 x 2
1
x
f (x) =
exp
,
x 2
2 x
f ( y) = e y ,
where
is the reciprocal of the mean stress.
x is the mean strength.
x is the strength standard deviation.

1999 by CRC Press LLC

y>0

< x <

(9.24)

(9.25)

Using Equations (9.24) and (9.25) in Equation (9.9) yields

R=

1 x 2 x
1
y
x

exp
d
x
e
d
y

x 2

2 x o

1 x 2

1

x
exp
+ x d x
= 1

x 2

2 x

(9.26)

since

2
2 x x2 + x2 x + x x4 2
1 x x

+
x
=
2 2
2 x2

(9.27)

Equation (9.26) is rewritten to the following form:

1
2 2
R = 1 exp 2 x G
x
2

(9.28)

where

1 2 + x2
1
x
G
exp x
dx
x 2
x
2

(9.29)

Since G = 1 [24], Equation (9.28) simplifies to

1
R = 1 exp 2 x x2 2
2

(9.30)

Example 9.4
An items stress and strength are described by exponential and normal distributions,
respectively. The mean of the stress is 10,000 psi and the mean and standard deviation
of the strength are 30,000 and 3,000 psi, respectively. Calculate the items reliability.
Inserting the specified data into Equation (9.30), we get
1 2 (30, 000) (3, 000)2
R = 1 exp

(10, 000)2
2 10, 000

1999 by CRC Press LLC

1
= 1 exp (6 0.09)
2

= 0.9479
Thus, the items reliability is 0.9479.
Other stress-strength models are presented in Reference 10.

9.6

MELLIN TRANSFORM METHOD

This is a useful method for estimating an items reliability when its associated stress
and strength distributions cannot be assumed, but there is an adequate amount of
empirical data. This approach can also be used when stress and strength distributions
associated with an item are known. Obviously, the method is based on Mellin
transforms and for Equation (9.8) they are defined as follows:

X = f (x) d x
y

(9.31)

= 1 F1 (y)
and

Y = f (y) dy = F2 (y)

(9.32)

where
F1 (y) and F2 (y) are the cumulative distribution functions.
Taking the derivative of Equation (9.32) with respect to y, we have
dY
= f ( y)
dy

(9.33)

d Y = f ( y) d y

(9.34)

Thus, from Equation (9.33), we get

Obviously, it can be easily seen from Equation (9.32) that Y takes values of 0 to 1
(i.e., at y = 0, Y = 0 and at y = , Y = 1). Thus, inserting Equations (9.31) and (9.34)
into Equation (9.8) yields

1999 by CRC Press LLC

R=

X dY

(9.35)

The above equation indicates that the area under the X vs. Y plot represents item
reliability. This area can be estimated by using the Simpsons rule, expressed below.
k

f (z) d z
j

kj
(W0 + 4 W1 + 2 W2 + 4 W3 +) + 2 Wn2 + 4 Wn1 + Wn
3n

(9.36)

where
is a function of z defined over interval (j, k), i.e., j z k.
is the even number of equal subdivided parts of interval (j, k).

f (z)
n
kj
n

= g is the width of the subdivided part.

At j = z0, z1, z2, z3, , zn = k; the corresponding values of W are W0 = f (z0), W1 =

f (z1), W2 = f (z2), , Wn = f (zn).
Example 9.5
Use the Mellin transform method to estimate the item reliability in Example 9.2.
Comment on the end result in comparison to the one obtained in Example 9.2.
Thus, substituting Equation (9.14) and the given relevant data of Example 9.2
into relationship (9.32) yields

Y=

y
1
e 10,000 d y
10, 000

= 1 e

(9.37)

y
10 ,000

Inserting Equation (9.15) and the given relevant data of Example 9.2 into
Equation (9.31), we have

X = 1 1 e
=e

35,000

(9.38)

y
35,000

Table 9.2 presents values of Y and X using Equations (4.37) and (4.38), respectively,
for assumed values of stress y.

1999 by CRC Press LLC

TABLE 9.2
Computed Values of Y and X for
the Assumed Values of Stress y
y (psi)

0
4,000
8,000
12,000
16,000
20,000
24,000
28,000
32,000
36,000
40,000
44,000
48,000

0
0.33
0.55
0.70
0.80
0.87
0.91
0.94
0.96
0.97
0.98
0.99
0.992
1

1
0.89
0.80
0.71
0.63
0.57
0.50
0.45
0.40
0.36
0.32
0.29
0.25
0

Figure 9.3 shows the plot of X vs. Y for data given in Table 9.2. The area under
the Figure 9.3 curve is estimated using Equation (9.36) as follows:
R

(1 0) W + 4 W + 2 W + 4 W + W
[ 0
1
2
3 ]
4
3 ( 4)
1
[1 + 4 (0.95) + 2 (0.84) + 4 (]0.68) + 0
12

0.7667
The above result, i.e., R = 0.7667, is quite close to the one obtained using the
analytical approach in Example 9.2, i.e., R = 0.7778.

9.7 FAILURE RATE MODELS

Over the years, many failure rate estimation mathematical models have been developed to determine failure rates of various mechanical items including brakes,
clutches, pumps, filters, compressors, seals, and bearings [12, 13, 2527]. This section presents failure rate calculation models for such items.

9.7.1

BREAK SYSTEM FAILURE RATE MODEL

The brake system failure rate is expressed by [25]:

bs = bfm + s + b + se + a + h

1999 by CRC Press LLC

(9.39)

FIGURE 9.3 Plot of X vs. Y.

where
bs
bfm
s
b
se
a
h

is
is
is
is
is
is
is

the
the
the
the
the
the
the

break system failure rate, expressed in failures/106 h.

failure rate of brake friction materials.
failure rate of springs.
failure rate of bearings.
failure rate of seals.
failure rate of actuators.
failure rate of brake housing.

The values of bs, bfm, s, b, se, a, and h are obtained through various
means [25, 28]. For example, bfm can be calculated by using the following
equation [25]:
4

bfm = bbfm

(9.40)

i =1

where
bbfm is the base failure rate associated with the brake friction material.
fi
is the ith multiplying factor that considers the effects on the base failure
rate of items such as dust contaminants (i = 1), ambient temperature
(i = 2), brake type (i = 3), and counter-surface roughness (i = 4).

1999 by CRC Press LLC

In turn, the base failure rate of the brake friction material is given by [29]:
bbfm = 3 10 3 . R W . W . n
. . ( A C) ( L T ) . A
2

(9.41)

where
LT
A
RW
W
n

AC

9.7.2

is the lining thickness, expressed in inches.

is the lining area, expressed in inches2.
is the friction materials specific wear rate, expressed in in.3/ft-lbf.
is the vehicle/aircraft weight, expressed in lbf.
is the miles traveled per year.
is the number of brake applications per miles.
is the average change in velocity per brake action, expressed in ft/s.
is the proportion of weight carried by lining during braking action. For
example, in the case of a 4-wheel vehicle, typically each front brake
carries 3/10 of the breaking load.

CLUTCH SYSTEM FAILURE RATE MODEL

The clutch system failure rate is expressed by [25]:

cs = cfm + s + b + se + a

(9.42)

where
cs is the clutch system failure rate, expressed in failure/106 h.
cfm is the failure rate of clutch friction material.
Other symbols used in Equation (9.42) are the same as the ones used in Equation (9.39).
The failure rate of clutch friction materials, cfm, is expressed by
cfm = bcfm f1 f2

(9.43)

where
bcfm is the base failure rate of clutch friction material.
f1
is the factor that considers the effects on the base failure rate of multiple
plates.
f2
is the factor that considers the effects on the base failure rate of ambient
temperature.
The clutch friction material base failure rate, bcfm, is given by [25, 30]:
bcfm = k E a 2 A Wm CFMa L T

1999 by CRC Press LLC

(9.44)

where
k
LT
AWm
CFMa
Ea

9.7.3

is the number of applications per hour.

is the lining thickness, expressed in inches.
is the average wear of the material.
is the area of the clutch friction material on each disc, expressed in
inches2.
is the average energy dissipated per engagement, expressed in ftlbf.

PUMP FAILURE RATE MODEL

The pump failure rate is expressed by [26]:

p = ps + pse + pb + pc + pfd

(9.45)

where
p
ps
pse
pb
pc
pfd

is
is
is
is
is
is

the
the
the
the
the
the

pump failure rate, expressed in failures/106 cycles.

failure rate of the pump shaft.
failure rate of all pump seals.
failure rate of all pump bearings.
failure rate of the pump casing.
failure rate of the pump fluid driver.

The failure rate of the pump shaft, ps, can be calculated by using the following
equation:
6

ps = bps

(9.46)

i =1

where
bps is the base failure rate of the pump shaft.
fi is the ith modifying factor; for i = 1 is associated with material temperature, i = 2 pump displacement, i = 3 casing thrust load, i = 4 shaft
surface finish, i = 5 contamination, and i = 6 material endurance limit.
The pump seal failure rate, pse, is expressed by
7

pse = bpse

i =1

1999 by CRC Press LLC

(9.47)

where
bpse is the pump seal base failure rate.
fi
is the ith modifying factor; for i = 1 is for effects of casing thrust load,
i = 2 surface finish, i = 3 seal smoothness, i = 4 fluid viscosity, i = 5
pressure/velocity factor, i = 6 temperature, and i = 7 contaminates.
Similarly, the values of pb, pc, and pfd can be calculated. Reference 26 presents
procedures to calculate these failure rates.

9.7.4

FILTER FAILURE RATE MODEL

The following equation can be used to predict filter failure rate [26]:
6

f = bf

(9.48)

i =1

where
f is the filter failure rate, expressed in failures/106 h.
bf is the filter base failure rate.
fi is the ith modifying factor; for i = 1 is for temperature effects, i = 2
vibration effects, i = 3 water contamination effects, i = 4 cyclic flow
effects, i = 5 cold start effects, and i = 6 differential pressure effects.

9.7.5

COMPRESSOR SYSTEM FAILURE RATE MODEL

The compressor system failure rate is expressed by the following equation [27]:
com = cs + cse + cva + cdc + cc + cb

(9.49)

where
com
cs
cse
cva
cdc
cc
cb

is
is
is
is
is
is
is

the
the
the
the
the
the
the

compressor system failure rate, expressed in failures/106 h.

failure rate of all compressor shafts.
failure rate of all compressor seals.
failure rate of valve assy (if any).
failure rate due to design configuration.
failure rate of the compressor casing.
failure rate of all compressor bearings.

The compressor shaft failure rate is expressed by

5

cs = bcs

i =1

1999 by CRC Press LLC

(9.50)

where
bcs is the compressor shaft base failure rate.
fi is the ith modifying factor; for i = 1 is for material temperature, i = 2
material endurance limit, i = 3 displacement, i = 4 shaft surface finish,
and i = 5 contamination.
The compressor seal failure rate can be predicted by using the following equation:
10

cse = bcse

(9.51)

i =1

where
bcse is the compressor seal base failure rate.
fi
is the ith modifying factor; for i = 1 is for fluid pressure, i = 2 seal face
pressure and gas velocity, i = 3 seal smoothness, i = 4 seal size, i = 5
temperature, i = 6 allowable leakage, i = 7 contaminants, i = 8 flow rate,
i = 9 fluid viscosity, and i = 10 surface finish and other conductance
parameters.
The compressor valve assembly failure rate is expressed by
cva = so + sa + pa

(9.52)

where
so is the failure rate of the solenoid (if any).
sa is the failure rate of sliding action valve assy (if any).
pa is the failure rate of poppet assy (if any).
Procedures for calculating cdc, cc, and cb are presented in Reference 27.

9.7.6

BEARING FAILURE RATE MODEL

The bearing failure rate is expressed by [12]:

br = pbs (AL SL) (AE 0.006)
x

2.36

(SV OV)0.54 L2c 3

(9.53)

where
br is the failure rate of bearing (using actual conditions), expressed in
failure/106 h of operation.
x is a factor; x = 4.0 for ball bearings and x = 3.33 for roller bearings.

1999 by CRC Press LLC

Lc
AE
SV
OV
AL
SL
pbs

is
is
is
is
is
is
is

the
the
the
the
the
the
the

actual contamination level, g/m 3/60 g/m 3.

alignment error, radians.
specification lubricant viscosity, lb.min/in2.
operating lubricant viscosity, lb.min/in2.
actual load, psi.
specification load, psi.
pseudo failure rate of the bearing and is expressed as follows:
pbs = 10 6 M 0.9 (5.45) RL

(9.54)

where
M is the number of bearings in system.
RL is the rated life in revolutions and is given by:
RL = 16700 rpm (BLR ERL )

(9.55)

where
rpm

BLR
ERL

is the shaft rotating velocity, expressed in rev/min.

is a constant; = 10/3 for roller bearings and = 3 for ball bearings.
is the basic load rating, lb.
is the equivalent radial, load, lb.

9.8 FAILURE DATA SOURCES

FOR MECHANICAL PARTS
Just like in the case of electronic parts, the availability of failure data for mechanical
parts is very important. Usually, it is more difficult to obtain useful failure data for
mechanical parts than for the electronic parts. Reference 10 lists over 50 sources for
obtaining failure data for various mechanical items. Some of the important sources
for obtaining mechanical failure related information are as follows:

RADC Non-Electronic Reliability Notebook [31].

Government Industry Data Exchange Program (GIDEP) [32].
AVCO Corporation Data [33].
Book: Mechanical Reliability [10].
Martin Marietta Corporation Data [34].

Table 9.3 presents failure rates for selected mechanical items. Chapter 4 discusses
the subject of failure data in more depth.

1999 by CRC Press LLC

TABLE 9.3
Failure Rates for Selected Mechanical Parts [10, 12]
No.

Mechanical part

Use environment

Failure rate
(failures/106 h)

1
2
3
4
5
6
7
8
9
10

Hose, pneumatic
Pressure regulator
Fuel pump
Compressor (general)
Clutch (friction)
Filter, gas (air)
Heavy-duty ball bearing
Seal, O-ring
Brake assembly
Guide pin

Not applicable (N/A)

Ground, fixed
Ground, fixed
Ground, mobile
Ground, mobile
Ground, mobile
N/A
N/A
N/A
N/A

29.3
2.4
176.4
33.6
38.2
3.2
14.4
0.2
16.8
13

9.9 OPTIMUM MODELS FOR MECHANICAL

EQUIPMENT REPLACEMENT OR MAINTENANCE
During the mechanical equipment use, its maintenance and replacement are two
important considerations. Over the years, professionals in the field have developed
various types of mathematical models to determine optimum time intervals for
equipment replacement or maintenance. This section presents two such models [10].

9.9.1

EQUIPMENT REPLACEMENT MODEL

This mathematical model is concerned with determining the optimum time interval
between replacements by minimizing the mean annual cost associated with the
equipment with respect to the time between replacements or the equipment life
expressed in years. The model assumes that the equipment mean annual cost is made
up of three components: (1) average operating cost, (2) average maintenance cost,
and (3) average investment cost. Thus, the mean annual cost of the equipment is
expressed by
MC = C0 + C m +

CI (x 1)
+
(IOC + IMC)
x
2

where
MC
C0
Cm
x
IOC

is
is
is
is
is

the
the
the
the
the

1999 by CRC Press LLC

mean annual total cost of equipment.

operational cost for the first year.
maintenance cost for the first year.
equipment life in years.
amount by which operational cost increases annually.

(9.56)

IMC is the amount by which maintenance cost increases annually.

CI is the cost of investment.
Taking the derivatives of Equation (9.56) with respect to x, we get
dMC 1
CI
= (IOC + IMC) 2
2
dx
x

(9.57)

Equating Equation (9.57) to zero and rearranging yields

2 CI

x* =
IOC + IMC

12

(9.58)

where
x* is the optimum replacement time interval.
Example 9.6
Assume that the following data are specified for mechanical equipment:
IMC = $500, IOC = $600, and CI = $15, 000
Calculate the optimum time for the mechanical equipment replacement.
Using the above data in Equation (9.58) yields
2 (15, 000)
x* =

600 + 500

12

= 5.2 years

Thus, the optimum time for the mechanical equipment replacement is 5.2 years.

9.9.2

EQUIPMENT MAINTENANCE MODEL

This mathematical model can be used to calculate the optimum number of inspections per facility per unit time. This information is quite useful to decision makers
because inspections are often disruptive but on the other hand such inspections
normally reduce the equipment downtime because of fewer breakdowns. In this
model, expression for the total equipment downtime is used to obtain the optimum
number of inspections.
The total facility downtime per unit time is given by [35]:
T D T = n (D T I) + (D T B) k n

1999 by CRC Press LLC

(9.59)

where
TDT
k
n
DTI
DTB

is
is
is
is
is

the total facility downtime per unit time.

a constant for specific facility.
the number of inspections per facility per unit time.
the downtime per inspection for a facility under consideration.
the downtime per breakdown for a facility under consideration.

By differentiating Equation (9.59) with respect to n, we get

dTDT
= D T I (D T B) k n 2
dn

(9.60)

Setting Equation (9.60) equal to zero and rearranging yields

n* = [ k (DTB) DTI]

12

(9.61)

where
n* is the optimum number of inspections per facility per unit time.
Inserting Equation (9.61) into Equation (9.59) yields

T D T* = 2 (DTB) (D TI) k

12

(9.62)

where
TDT* is the minimum total facility downtime per unit time.
Example 9.7
The following data are associated with certain mechanical equipment:
DTI = 0.02 month, DTB = 0.1 month, and k = 4.

(9.63)

Calculate the optimum number of inspections per month and the minimum value of
the total equipment downtime by using Equations (9.61) and (9.62), respectively.
Thus, inserting the given data into Equation (9.61) yields
n* = [4 (0.1) 0.02]

12

= 4.5 inspections month

1999 by CRC Press LLC

Similarly, substituting the specified data into Equation (9.62), we have

TDT * = 2 [(0.1) (0.02) 4]

12

= 0.18 months
Thus, the optimum number of inspections per month is 4.5 and the minimum
value of the total equipment downtime is 0.18 month.

9.10

PROBLEMS

1. Give reasons for the development of the mechanical reliability field.

2. Describe the following mechanical failure modes:
Compressive failure
Stress concentration failure
Shear loading failure
Bending failure
3. What are the important causes of gear failures?
4. Give four different ways of defining a safety factor.
5. Define the terms stress and strength.
6. An items stress and strength are normally distributed. Develop an expression for the item reliability.
7. A components stress and strength are distributed exponentially with mean
values of 12,000 and 40,000 psi, respectively. Determine the item reliability.
8. Use the Mellin transform approach to estimate the item reliability in
Problem 7. Comment on the end result in comparison to the one obtained
for Problem 7.
9. List at least 10 important sources for obtaining failure data for mechanical
items.
10. A mechanical components stress and strength are Maxwellian distributed.
Develop an expression for the component reliability.

9.11

REFERENCES

1. Weibull, W., A statistical distribution function of wide applicability, J. Appl. Mech.,

18, 293-297, 1951.
2. Freudenthal, A.M., Safety and probability of structural failures, Trans. Am. Soc. Civil
Eng., 121, 1337-1397, 1956.
3. Freudenthal, A.M., Fatigue in Aircraft Structures, Academic Press, New York, 1957.
4. Freudental, A.M. and Gumbel, E.J., Failure and survival in fatigue, J. Appl. Phys.,
25, 110-120, 1954.
5. Dhillon, B.S. and Singh, C., Engineering Reliability: New Techniques and Applications, John Wiley & Sons, New York, 1981.
6. Redler, W.M., Mechanical reliability research in the national aeronautics and space
administration, Proc. Reliability Maintainability Conf., New York, 763-768, 1966.

1999 by CRC Press LLC

7. Lipson, C., Sheth, N.J., and Disney, R.L., Reliability Prediction-Mechanical

Stress/Strength Inference, Rome Air Development Center, Griffiss Air Force Base,
New York, 1967, Report No. RADC-TR-66-710.
8. Kececioglu, D., Reliability analysis of mechanical components and systems, Nuclear
Eng. Des., 19, 259-290, 1972.
9. Kececioglu, D. and Haugen, E.B., A unified look at design safety factors, safety
margins and measures of reliability, Proc. Annu. Reliability Maintainability Conf.,
522530, 1968.
10. Dhillon, B.S., Mechanical Reliability: Theory, Models and Applications, American
Institute of Aeronautics and Astronautics, Washington, D.C., 1988.
11. Mechanical Reliability Concepts, American Society of Mechanical Engineers
(ASME) Design Engineering Conference, ASME, New York, 1965.
12. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Handbook of Reliability Engineering
and Management, McGraw-Hill, New York, 1996.
13. Doyle, R.L., Mechanical-system reliability, Tutorial Notes, Annu. Reliability Maintainability Symp., 1992.
14. Collins, J.A., Failure of Materials in Mechanical Design, John Wiley & Sons, New
York, 1981.
15. Lipson, C., Analysis and Prevention of Mechanical Failures, Course Notes No. 8007,
University of Michigan, Ann Arbor, June 1980.
16. Howell, G.M., Factors of safety, Machine Design, July 12, 76-81, 1956.
17. Schoof, R., How much safety factor?, Allis-Chalmers Elec. Rev., 21-24, 1960.
18. McCalley, R.B., Nomogram for selection of safety factors, Design News, Sept. 1957,
pp. 138-141.
19. Shigley, J.E. and Mitchell, L.D., Mechanical Engineering Design, McGraw-Hill, New
York, 610-611, 1983.
20. Bompass-Smith, J.H., Mechanical Survival: The Use of Reliability Data, McGrawHill, London, 1973.
21. Juvinall, R.C., Fundamentals of Machine Component Design, John Wiley & Sons,
New York, 1983.
22. Dhillon, B.S. and Singh, C., Engineering Reliability: New Techniques and Applications, John Wiley & Sons, New York, 1981.
23. MIL-STD-1629, Procedures for Performing a Failure Mode, Effects, and Criticality
Analysis (FMECA), Department of Defense, Washington, D.C.
24. Kececioglu, D. and Li, D., Exact solutions for the predication of the reliability of
mechanical components and structural members, Proc. Failure Prevention Reliability
Conf., published by the American Society of Mechanical Engineers, ASME, New
York, 1985, pp. 115-122.
25. Rhodes, S., Nelson, J.J., Raze, J.D., and Bradley, M., Reliability models for mechanical equipment, Proc. Annu. Reliability Maintainability Symp., 127-131, 1988.
26. Raze, J.D., Nelson, J.J., Simard, D.J., amd Bradley, M., Reliability models for
mechanical equipment, Proc. Annu. Reliability Maintainability Symp., 130-134, 1987.
27. Nelson, J.J., Raze, J.D., Bowman, J., Perkins, G., and Wannamaker, A., Reliability
models for mechanical equipment, Proc. Annu. Reliability Maintainability Symp.,
146-153, 1989.
28. Boone, T.D., Reliability Prediction Analysis for Mechanical Brake Systems, NAVA
IR/SYSCOM Report, August 1981, Department of Navy, Department of Defense,
Washington, D.C.

1999 by CRC Press LLC

29. Minegishi, H., Prediction of Brake Pad Wear/Life by Means of Brake Severity Factor
as Measured on a Data Logging System, Society of Automotive Engineers (SAE)
Paper No. 840358, 1984, SAE, Warrendale, PA.
30. Spokas, R.B., Clutch Friction Material Evaluation Procedures, SAE Paper No. 841066,
1984, SAE, Warrendale, PA.
31. Shafer, R.E., Angus, J.E., Finkelstein, J.M., Yerasi, M., and Fulton, D.W., RADC
Non-Electronic Reliability Notebook, Report No. RADC-TR-85-194, 1985, Reliability Analysis Center (RAC), Rome Air Development Center (RADC), Griffiss Air
Force Base, Rome, NY.
32. GIDEP Operations Center, U.S. Department of Navy, Naval Weapons Station, Seal
Beach, Corona Annex, Corona, CA.
33. Earles, D.R., Handbook: Failure Rates, AVCO Corporation, Massachussetts, 1962.
34. Guth, G., Development of Nonelectronic Part Cycle Failure Rates, Rept. No. 17D/A050678, December 1977, Martin Marietta Corporation, Orlando, FL. Available from
the National Technical Information Service (NTIS), Springfield, VA.

1999 by CRC Press LLC

10

Human Reliability in
Engineering Systems

10.1 INTRODUCTION
The failure of engineering systems is not only caused by hardware or software
malfunctions, but also by human errors. In fact, the reliability of humans plays a
crucial role in the entire system life cycle: design and development phase, manufacturing phase, and operation phase.
Even though in modern times the history of human factors may be traced back
to 1898 when Frederick W. Taylor performed studies to determine the most appropriate designs for shovels [1], it was not until 1958 when Williams [2] recognized
that human-element reliability must be included in the overall system-reliability
prediction; otherwise, such a prediction would not be realistic. In 1960, the work of
Shapero et al. [3] further stressed the importance of human reliability in engineering
systems by pointing out that human error is the cause for 20 to 50% of all equipment
failures.
In 1962, a database known as Data Store containing time and human performance
reliability estimates for human engineering design features was established [4]. Also,
in the 1960s two symposia concerning human reliability/error were held [5, 6]. In
1973, IEEE Transactions on Reliability published a special issue on human
reliability [7]. In 1980, a selective bibliography on human reliability was published
covering the period from 1958 to 1978 [8]. The first book on human reliability
entitled Human Reliability: With Human Factors appeared in 1986 [9].
Over the years many professionals have contributed to the field of human reliability. A comprehensive list of publications on the subject is given in Reference 10.
This chapter presents different aspects of human reliability.

10.2 TERMS AND DEFINITIONS

There are numerous terms and definitions used in the human reliability area and
some are presented below [9, 11-13].
Human factors: This is a body of scientific facts pertaining to the characteristics of humans.
Human error: This is the failure to perform a given task (or the performance of a forbidden action) that could result in the disruption of scheduled operations or damage to property and equipment.
Human performance: This is a measure of human-functions/actions
subject to specified conditions.

1999 by CRC Press LLC

 Continuous task: This is a task that involves some sort of tracking

activity; for example, monitoring a changing situation.
Human reliability: This is the probability of accomplishing a task successfully by the human at any required stage in system operations within
a stated minimum time (if such time requirement is stated).

10.3 HUMAN ERROR OCCURRENCE EXAMPLES

AND STUDIES
In the past human error has been responsible for many accidents and equipment
failures. Numerous studies have reported a significant proportion of equipment and
other failures due to human error. Results of some of those studies are given below.
Human operator accounted for over 90% of the documented air traffic
control system errors [14].
A study of 135 vessel failures occurring during the period from 19261988
discovered that 24.5% of failures were directly due to humans [15].
Over 50% of all technical medical equipment problems are due to operator
errors [16].
Up to 90% of accidents both generally and in medical devices are caused
by human mistakes [17, 28].
A study of 23,000 defects in the production of nuclear components revealed
that approximately 82% of the defects were due to human error [19].
During the period from June 1, 1973 to June 30, 1975, 401 human errors
occurred in U.S. commercial light-water nuclear reactors [20].

10.4 HUMAN ERROR OCCURRENCE CLASSIFICATION

AND ITS TYPES AND CAUSES
Human errors can occur in many different ways and their occurrence may be
classified into six distinct categories as shown in Figure 10.1. These are decision
errors, action errors, transmission errors, checking errors, diagnostic errors, and
retrieval errors.
Decision errors occur when the wrong decision is made after considering the
situation. Action errors are the result of no action, incorrect action, or the performance of correct action on the wrong object when required. Transmission errors
occur when information that must be passed to others is not sent, sent incorrectly,
or sent to the wrong destination. Checking errors occur when the systems require
checks, the incorrect checks are made, checks are omitted, or correct checks are
made on the wrong objects. Diagnostic errors are the result of misinterpreting the
actual situation when an abnormal event occurs. Retrieval errors occur when required
information either from an individual, an individuals memory, or from any other
reference source is not received or the incorrect information is received.

1999 by CRC Press LLC

FIGURE 10.1 Categories of errors that humans make.

Human errors may be broken into many distinct types as follows [9, 21, 22]:
Design errors. These types of errors are the result of inadequate design.
The causes of these errors are assigning inappropriate functions to
humans, failure to implement human needs in the design, failure to ensure
the man-machine interaction effectiveness, and so on. An example of
design errors is the placement of controls and displays so far apart that
an operator is unable to use them in an effective manner.
Operator errors. These errors are the result of operator mistakes and the
conditions that lead to operator errors include lack of proper procedures,
complex tasks, poor personnel selection and training, poor environment,
and operator carelessness.
Assembly errors. These errors occur during product assembly due to
humans. Assembly errors may occur due to causes such as poorly designed
work layout, inadequate illumination, excessive noise level, poor blueprints and other related material, excessive temperature in the work area,
and poor communication of related information.
Inspection errors. These errors occur because of less than 100% accuracy
of inspectors. One example of an inspection error is accepting and rejecting out-of-tolerance and in-tolerance parts, respectively. Nonetheless,
according to Reference 23, an average inspection effectiveness is close to
85%.
Maintenance errors. These errors occur in the field due to oversights by
the maintenance personnel. As the equipment becomes old, the likelihood
of the occurrence of those errors may increase because of the increase in
maintenance frequency. Some of the examples of maintenance errors are

1999 by CRC Press LLC

calibrating equipment incorrectly, applying the wrong grease at appropriate points of the equipment, and repairing the failed equipment incorrectly.
Installation errors. These errors occur due to various reasons including
using the wrong installation related blueprints or instructions, or simply
failing to install equipment according to the manufacturers specification.
Handling errors. These errors basically occur because of inadequate
storage or transportation facilities. More specifically, such facilities are
not as specified by the equipment manufacturer.
In general, there could be numerous causes for the occurrence of human error. Some
of those are listed below [9, 21].

Poor motivation of involved personnel.

Poor training or skill of concerned personnel.
Poor equipment design.
Inadequate or poorly written equipment operating and maintenance procedures.
Poor job environment: poor lighting, high/low temperature, high noise
level, crowded work space, etc.
Inadequate work tools.
Complex tasks.
Poor work layout.

10.5 HUMAN PERFORMANCE AND STRESS

Human performance depends on many different factors; thus, it may be said that
humans perform differently under different conditions. More specifically, the important
factors that will affect a persons performance include reaction to stress, time at work,
fatigue, group interaction and identification, social pressure, repetitive work, supervisors expectations, morale, social interaction, crew efficiency, and idle time [24].
In particular, stress is probably the most important factor that affects human
performance. Over the years, many researchers have studied the relationship between
human performance effectiveness and stress/anxiety. The resulting curve of their
effort is shown in Figure 10.2 [25, 26]. This curve indicates that a moderate level
of stress is necessary to increase human performance effectiveness to its maximum.
At a very low stress level, the task becomes unchallenging and dull and most humans
performance effectiveness is not at the maximum. In contrast, as the stress bypasses
its moderate level, the human performance effectiveness begins to decline. This
decline is caused by factors such as fear, worry, and other types of psychological
stress. In the highest stress region, human reliability is at its lowest.
Another important point to note from Figure 10.2 is that as the stress increases
in Region A, human performance effectiveness improves. However, in Region B,
human performance effectiveness decreases with the increasing stress.

1999 by CRC Press LLC

FIGURE 10.2

A hypothetical curve of human performance effectiveness vs. anxiety/stress.

10.5.1 STRESS FACTORS

AND

OPERATOR STRESS CHARACTERISTICS

There are many factors that increase the stress on a human and in turn decrease
his/her reliability in work and other environments. Some of these factors are listed
below [25].

Dissatisfied with the current job.

Faced with serious financial problems.
Working with individuals having unpredictable temperaments.
Low chance of promotion from current position/job.
Facing a possible work layoff.
Current job/position below ability and experience.
Conducting tasks under extremely tight time schedules.
Having health problems.
Excessive demands of superiors.
Inadequate expertise to conduct the tasks required in the current job/position.
Often takes work home to meet deadlines.
Having difficulties with spouse/children.

A human operator has certain limitations in conducting certain tasks. The error
occurrence probability increases when such limitations are exceeded. In order to
improve operator reliability, such operator limitations or stress characteristics must
be considered carefully during the design phase.
The operator stress characteristics include [12]: very short decision-making time,
several displays difficult to discriminate, requirement to perform steps at high speed,
poor feedback for the determination of accuracy of actions taken, requirement for

1999 by CRC Press LLC

prolonged monitoring, very long sequence of steps required to perform a task, requirement to make decisions on the basis of data obtained from various different sources,
and requirement to operate at high speed more than one control simultaneously.

10.6 HUMAN PERFORMANCE RELIABILITY

IN CONTINUOUS TIME AND MEAN TIME
TO HUMAN ERROR (MTTHE) MEASURE
As humans perform time-continuous tasks such as aircraft maneuvering, scope
monitoring, and missile countdown, a general expression for human performance
reliability can be developed the same way as for the development of the general
classical reliability function.
Thus, the human error occurrence probability in time intervalt is [27-29]
P ( X 2 X1 ) = h ( t ) t

(10.1)

where
X1
X2

is an error-free human performance event of duration time t.

represents an event that the human error will occur during time
interval [t, t + t].
h (t) is the human error rate at time t.
The joint probability of the error-free human performance is expressed by

P X 2 X1 P ( X1 ) = P ( X1 ) P ( X 2 X1 ) P ( X1 )

(10.2)

where

represents an event that human error will not occur during interval [t,
t + t].

More specifically, Equation (10.2) denotes an error-free human performance

probability over intervals [o, t] and [t, t + t].
Equation (10.2) may be rewritten to the following form:
R h (t + t ) = R h (t ) R h (t ) P ( X1 X 2 )
where
is the human performance reliability at time t.
Rh (t)
Rh (t + t) is the human performance reliability at time t + t.

1999 by CRC Press LLC

(10.3)

By inserting Equation (10.1) into Equation (10.3) we have

R h (t + t ) R h (t )
= R h (t ) h (t )
t

(10.4)

In the limiting case, expression (10.4) becomes

d R h (t )
= R h (t ) h (t )
dt

(10.5)

By rearranging Equation (10.5) we get

h (t ) d t =

1
d R h (t )
R h (t )

(10.6)

By integrating both sides of Equation (10.6) over the time interval [o, t] results in
t

h (t ) d t =
o

R (t ) d R (t )
1

(10.7)

Since at t = 0, Rh (t) = 1 Equation (10.7) becomes

Rh (t )

h (t ) d t =
o

R (t ) d R (t )
1

(10.8)

After evaluating the right-hand side of Equation (10.8), we get

t

l n R h (t ) = h (t ) d t

(10.9)

Thus, from Equation (10.10) we write

t

R h (t ) = e

h (t )d t
0

(10.10)

The above equation is the general human performance reliability function. It can be
used to predict human reliability at time t when times to human error are described
by any known statistical distribution. A study reported in Reference 27 collected
human error data for time-continuous tasks under laboratory conditions. The Weibull,
gamma, and log-normal distributions fitted quite well to these data.

1999 by CRC Press LLC

In order to obtain a general expression for mean time to human error (MTTHE),
we integrate Equation (10.10) over the interval [0, ]:

MTTHE =

R h ( t ) d t = exp h (t ) d t d t

0
o

(10.11)

The above equation can be used to obtain MTTHE when times to human error are
governed by any known distribution function.

EXAMPLE 10.1
Assume that the time to human error associated with a time-continuous task is
exponentially distributed and the human error rate is 0.02 errors per hour. Calculate
the following:
Human performance reliability for a 5-h mission.
MTTHE.
Thus, in this case we have
h (t ) = h = 0.02 errors h
Inserting the above value into Equation (10.10) yields
t

R h (t ) = e
=e

( 0.02 ) d t
o

(10.12)

( 0.02 ) t

At t = 5 h, from Equation (10.12) we get

R h (5) = e ( 0.02 ) ( 5)
= 0.9048

(10.13)

By substituting Equation (10.12) into Equation (10.11) we get

MTTHE =

(0.02)

= 50 h
It means the human performance reliability is 0.9048 and a human error can
occur after every 50 h.

1999 by CRC Press LLC

EXAMPLE 10.2
A person is performing a time-continuous task and his/her time to human error is
described by Weibull distribution. Obtain expressions for human reliability and
MTTHE.
In this case, as the time to human error is Weibull distributed, the human error
rate at time t is given by
h ( t ) = ( t )

(10.14)

where
is the shape parameter.
is the scale parameter.
By inserting Equation (10.14) into Equation (10.10) we have
t

R h (t ) = e

[( t )

dt

(10.15)

= e ( t )

Using Equation (10.15) in Equation (10.11) yields

MTTHE = exp (t )
o

] dt
(10.16)

1
= + 1

where
() is the gamma function.
The human reliability and MTTHE are given by Equations (10.15) and (10.16),
respectively.

10.7

HUMAN RELIABILITY EVALUATION METHODS

Over the years, many techniques to evaluate human reliability have been developed [9].
Each such technique has advantages and disadvantages. Some of these methods are
presented below.

1999 by CRC Press LLC

10.7.1 PROBABILITY TREE METHOD

This is used to perform task analysis by diagrammatically representing critical
human actions and other events associated with the system. Usually, this method is
utilized to perform task analysis in the technique for the Human Error Rate Prediction
(THERP) approach [30]. Nonetheless, the branches of the probability tree represent
diagrammatic task analysis. More specifically, the outcomes (e.g., success of failure)
of each event are represented by the branching limbs of the tree and each tree branch
is assigned an occurrence probability.
There are many benefits of this method including a visibility tool, simplified
mathematical computations which in turn reduce the error occurrence probability
due to computation, and estimate conditional probability readily, which may otherwise be obtained through complex probability equations.
This approach can also be used to calculate reliability of standard networks [31].
The method is demonstrated by solving the following example.

EXAMPLE 10.3
A person performs two quality control tasks: and . Each of these two tasks can
either be performed correctly or incorrectly. The task is performed before task
and both the tasks are independent of each other. In other words, the performance
of task does not affect the performance of task or vice-versa. Develop a
probability tree for this example and obtain an expression for probability of not
successfully completing the overall mission.
In this case, the person first performs task correctly or incorrectly and then
proceeds to task which can also be performed correctly or incorrectly. This scenario
is depicted by the probability tree shown in Figure 10.3. In this figure, and with
bars denote unsuccessful events and without bars denote successful events. Other
symbols used in the solution to the example are as follows:
Ps is the probability of success of the overall mission (i.e., performing both
tasks and correctly).
Pf is the failure probability of the overall mission.
P is the probability of performing task correctly.
P is the probability of performing task incorrectly.
P is the probability of performing task correctly.
P is the probability of performing task incorrectly.
Using Figure 10.3, the probability of success of the overall mission is
Ps = P P

(10.17)

Similarly, using Figure 10.3, the failure probability of the overall mission is given by
Pf = P P + P P + P P

1999 by CRC Press LLC

(10.18)

FIGURE 10.3 Probability tree for performing tasks and .

Since P + P = 1 and P + P = 1, by subtracting Equation (10.18) from unity yields

the same result as Equation (10.17), i.e.,

Ps = 1 Pf = 1 P P + P 1 P + P (1 P )

Ps = 1 P + P P P

(10.19)

(10.20)

The above equation is exactly the same as Equation (10.17), if we write the right-hand
term of Equation (10.17) in terms of failure probabilities. Nonetheless, the probability
of not successfully completing the overall mission is given by Equation (10.18).
Example 10.4
Assume that in Example 10.3 the probabilities of performing tasks and incorrectly are 0.15 and 0.2, respectively. Determine the following:
Probability of successfully completing the overall mission.
Probability of not successfully completing the overall mission.
The sum of the above two results is equal to unity.

1999 by CRC Press LLC

Inserting the given data into Equation (10.20) yields

Ps = 1 [0.15 + 0.2 (0.15) (0.2)]
= 0.68
Since P + P = 1, we get
P = 1 P
= 1 0.15
= 0.85
Similarly, since P + P = 1, we have
P = 1 P
= 1 0.2
= 0.8
By substituting the above two values and the specified data into Equation (10.18)
we get
Pf = (0.15) (0.2) + (0.15) (0.8) + (0.85) (0.2)
= 0.32
Thus,
Ps + Pf = 0.68 + 0.32 = 1
The probabilities of successfully/not successfully completing the overall mission
are 0.68/0.32, respectively, and the sum of these two values is equal to unity.

10.7.2 FAULT TREE METHOD

This approach is described in detail in Chapter 7 and it can also be used to perform
human reliability analysis. The following examples demonstrate the application of
this method in human reliability work.
Example 10.5
A person is required to do a certain job, J, composed of two independent tasks X
and Y. These two tasks must be performed correctly for the successful completion

1999 by CRC Press LLC

FIGURE 10.4 Fault tree for the performance of Job J incorrectly.

of the job. Task X is made up of two subtasks X1 and X2. If any one of these two
subtasks is performed correctly, Task X can be completed successfully. Task Y is
composed of subtasks Y1, Y2, and Y3. All of these three subtasks must be performed
correctly for the success of Task Y. Both subtasks, X1 and Y1, are composed of two
steps each, i.e., x1, x2 and y1, y2, respectively. Both steps for each of these two
subtasks must be completed correctly for subtask success. Develop a fault tree for
the event that job J will be performed incorrectly by the person.
Figure 10.4 presents the fault tree for the example.
Example 10.6
Assume that the probability of occurrence of the basic events in Figure 10.4 is 0.04.
Calculate the probability of occurrence of the top event, Job J will be performed
incorrectly by the person. Assume in your calculations that the given fault tree is
independent.
The probability of performing subtask X1 incorrectly is given by
P ( X1 ) = P (x1 ) + P (x 2 ) P (x1 ) P (x 2 )
= 0.04 + 0.04 (0.04) (0.04)
= 0.0784
where
P (x1) is the probability of performing step x1 incorrectly, for i = 1, 2.

1999 by CRC Press LLC

Similarly, the probability of performing subtask Y1 incorrectly is

P (Y1 ) = P (y1 ) + P (y 2 ) P (y1 ) P (y 2 )
= 0.04 + 0.04 (0.04) (0.04)
= 0.0784
where
P(yi ) is the probability of performing step yi incorrectly, for i = 1, 2.
The probability of performing task X incorrectly is given by
P ( X) = P ( X1 ) + P ( X 2 )
= (0.0784) (0.04)
= 0.0031
where
P(Xi) is the probability of performing subtask Xi incorrectly, for i = 1, 2.
The probability of performing task Y incorrectly is expressed by

)(

)(

P (Y) = 1 1 P (Y1 ) 1 P (Y2 ) 1 P (Y3 )

= 1 (1 0.0784) (1 0.04) (1 0.04)
= 0.1507
where

P(Yi ) is the probability of performing subtask Yi incorrectly, for i = 1, 2, 3.

The probability of performing job J incorrectly by the person is
P ( J ) = P ( X ) + P (Y) P ( X ) P (Y)
= 0.0031 + 0.1507 (0.0031) (0.1507)
= 0.1534
Figure 10.5 shows the fault tree with given and calculated probability values.
The probability that job J will be performed incorrectly by the person is 0.1534.

1999 by CRC Press LLC

FIGURE 10.5 Fault tree with given and calculated probability values.

FIGURE 10.6 State space diagram for a human performing a time continuous task.

10.7.3 MARKOV METHOD

This is a powerful reliability engineering tool that can also be used to perform time
continuous human reliability analysis. The following assumptions are associated
with this method:
The probability of a transition occurrence from one state to another in
finite time t is h t. The parameter h in our case is the constant human
error rate.
The probability of two or more transitional occurrences in finite time t
is negligible [i.e., (h t) (h t) 0].
All occurrences are independent of each other.
The state space diagram of a human performing a time-continuous task is given in
Figure 10.6. In this case, the system has two states (i.e., 0 and 1). The state 0
represents human performing the time continuous task normally and state 1 denotes
human committed error. By using the Markov method, we can obtain the probabilities of the human being in states 0 and 1 at time t for known human error rate.

1999 by CRC Press LLC

The numerals in the circles of Figure 10.6 denote system states. The following
symbols are associated with Figure 10.6:
P0 (t) is the probability that the human is performing his/her assigned task
normally at time t.
P1 (t) is the probability that the human has committed error at time t.
h
is the constant human error rate.
Using the Markov approach, we write down the following equations for Figure 10.6.
P0 (t + t ) = P0 (t ) (1 h t )

(10.21)

P1 (t + t ) = P0 (t ) ( h t ) + P1 (t )

(10.22)

where
P0 (t + t) is the probability that the human is performing his/her assigned
task normally at time t + t.
P1 (t + t) is the probability that the human has committed error at time t.
(1 h t) is the probability of the occurrence of no human error in finite
time t.
Rearranging Equations (10.21) and (10.22) and taking the limits, we get
d P0 (t )
= h P0 (t )
dt

(10.23)

R h (t + t ) = R h (t )

(10.24)

At time t = 0, P0 (0) = 1 and P1 (0) = 0.

Solving Equations (10.23) and (10.24) using Laplace transforms [9] leads to
Po (t ) = e h t

(10.25)

P1 (t ) = 1 e h t

(10.26)

Thus, the human reliability, Rh (t), is given by

R h (t ) = P0 (t ) = e h t

1999 by CRC Press LLC

(10.27)

By integrating Equation (10.27) over the time interval [0, ], we get the following
equation for MTTHE:

MTTHE =

R (t ) d t
h

= eh t d t

(10.28)

= 1 h
Example 10.7
Assume that a person is performing a time continuous task and his/her human error
rate is 0.006 errors per hour. Calculate the probability that the person will commit
an error during an 8-h mission.
Substituting the given data into Equation (10.26) yields
P1 (8) = 1 e ( 0.006 ) (8)
= 0.0469
There is an approximately 5% chance that the person will commit an error during
the 8-h mission.

10.8 HUMAN RELIABILITY MARKOV MODELING

As demonstrated in the preceding section, the Markov method is a useful tool to
perform various types of human reliability analysis. This section presents analysis
of two mathematical models using the Markov method.

10.8.1 RELIABILITY ANALYSIS

OF A

SYSTEM

WITH

HUMAN ERROR

This mathematical model represents a system which can fail either due to a hardware
failure (i.e., a failure other than a human error) or to a human error. The system
human/non-human failure rates are constant. The system state space diagram is
shown in Figure 10.7. By using the Markov approach we can obtain equations for

FIGURE 10.7 System transition diagram.

1999 by CRC Press LLC

system reliability, system failure probability due to human error, system failure
probability due to non-human error, and system mean time to failure. The numerals
in the boxes of Figure 10.7 denote system state.
The following symbols are associated with this model:
h is the constant human error rate.
nh is the constant nonhuman error rate (i.e., hardware failure rate, etc.)
Pi (t) is the probability that the system is in state i at time t; for i = 0 (means
system operating normally), i = 1 (means system failed due to nonhuman error), and i = 2 (means system failed due to human error).
By applying the Markov method and using Figure 10.7, we get the following equations:
d P0 (t )
+ ( h + nh ) P0 (t ) = 0
dt

(10.29)

d P1 (t )
nh P0 (t ) = 0
dt

(10.30)

d P2 (t )
h P0 (t ) = 0
dt

(10.31)

At time t = 0, P0 (0) = 1, P1 (0) = P2 (0) = 0.

Solving the above equations using Laplace transforms [9] results in
+ t
P0 (t ) = e ( h nh )

(10.32)

P1 (t ) =

nh
+ t
1 e ( h nh )
( nh + h )

(10.33)

P2 (t ) =

( nh + h ) [

(10.34)

+ t
1 e ( h nh )

The system reliability, Rs (t), with human error is

+ t
Rs (t ) = P0 (t ) = e ( h nh )

The system mean time to failure (MTTF) is given by

1999 by CRC Press LLC

(10.35)

MTTF =

R (t ) d t
S

+ t
= e ( h nh ) d t

(10.36)

= 1 ( h + nh )
Example 10.7
A system can fail either due to a hardware failure or a human error and its hardware
failure and human error rates are 0.0005 failures/h and 0.0001 errors/h, respectively.
Calculate the system MTTF and failure probability due to human error for a 12-h
mission.
Inserting the given data into Equation (10.36) yields
MTTF = 1 (0.0001 + 0.0005)
= 1666.7 h
Similarly, we substitute the specified data into Equation (10.34) to get
P2 (12) =

(0.0001) e 1
(0.0001 + 0.0005) [

( 0.0001+ 0.0005 ) (12 )

= 0.0012
Thus, the system failure probability due to human error for a 12-h mission is
0.0012.

10.8.2 RELIABILITY ANALYSIS OF A HUMAN PERFORMING A TIMECONTINUOUS TASK UNDER FLUCTUATING ENVIRONMENT
This mathematical model represents a human performing a time continuous task
under fluctuating environment. The environment fluctuates from normal to abnormal
(stress) and vice versa [32]. Human error can occur under both environments. However, it is logical to assume that the human error rate under the abnormal environment
will be greater than under the normal environment because of increased stress. In
this model, it is also assumed that the rate of change from normal to abnormal
environment and vice versa is constant along with human error rate under either
environment. The model state space diagram is shown in Figure 10.8. The numerals
in the boxes of this figure denote system state. The following symbols were used to
develop equations for the model:

1999 by CRC Press LLC

FIGURE 10.8 State space diagram of a human performing a time continuous task under
fluctuating environment.

Pi (t) is the probability of the human being in state i at time t; for i = 0

(means human performing task correctly under normal environment),
i = 1 (means human committed errors under normal environment), i = 2
(means human performing task correctly under abnormal environment),
i = 3 (means human committed error under abnormal environment).
h is the constant human error rate from state 0.
ah is the constant human error rate from state 2.
n
is the constant transition rate from normal to abnormal environment.
a
is the constant transition rate from abnormal to normal environment.
By using the Markov method, we get the following system of equations from
Figure 10.8:
d P0 (t )
+ ( h + n ) P0 (t ) = P2 (t ) a
dt

(10.37)

d P1 (t )
P0 (t ) h = 0
dt

(10.38)

d P2 (t )
+ ( ah + a ) P2 ( t ) = P0 (t ) n
dt

(10.39)

d P3 (t )
P2 (t ) ah = 0
dt

(10.40)

At time t = 0, P0 (0) = 1, and P1 (0) = P2 (0) = P3 (0) = 0.

Solving the above equations using Laplace transforms we obtain
P0 (t ) = (s2 s1 )

1999 by CRC Press LLC

[(s

+ ah + n ) es2 t 1 (s + ah +1 n ) es t

(10.41)

where

]2

(10.42)

]2

(10.43)

s1 = a1 + a12 4 a 2

s2 = a1 a12 4 a 2

12

12

a1 h + ah + n + a

(10.44)

a 2 h ( ah + a ) + n ah

(10.45)

P1 (t ) = a 4 + a 5 es2 t a 6 es1t

(10.46)

a 3 = 1 (s2 s1 )

(10.47)

a 4 = h ( ah + a ) 1s s2

(10.48)

a 5 = a 3 ( h + a 4 s1 )

(10.49)

a 6 = a 3 ( h + a 4 s2 )

(10.50)

where

P2 (t ) = n a 3 es2 t es1t

P3 (t ) = a 7 (1 + a 3 ) s1 es2 t s2 es1t

(10.51)

)]

(10.52)

where
a 7 ah n s1 s2

(10.53)

The reliability of the human is expressed by

R h (t ) = P0 (t ) + P2 (t )

1999 by CRC Press LLC

(10.54)

Thus, the MTTHE is

MTTHE =

R (t ) d t
h

[P (t) + P (t)] d t
0

(10.55)

= ( ah + n + a ) a 2
Example 10.8
Assume that a person is performing a time continuous task under normal and
abnormal environments with error rates 0.002 errors/h and 0.003 errors/h, respectively. The transition rates from normal to abnormal environment and vice versa are
0.02 per hour and 0.04 per hour, respectively. Calculate the MTTHE.
Substituting the given data into Equation (10.55) we get
MTTHE =

(0.003 + 0.02 + 0.04)

0.002 (0.003 + 0.04) + (0.02) (0.003)

= 431.51 h
Thus, the mean time to human error is 431.51 h.

10.9 HUMAN ERROR DATA

Human reliability predictions are only as effective as the body of quantitative data on
which they are made. More specifically, the human error data are the backbone of any
human reliability prediction. Thus, the collection and maintenance of human error data
are at least as important as making various types of human reliability predictions.
Human error data can be collected through several means including [3335]:

Experimental studies
Expert judgments
Self-made error reports
Human data recorder
Automatic data recorder
Published literature

The data collected from experimental studies is usually generated under the
laboratory conditions. These conditions may not be the true representatives of actual
conditions. In addition, the method is time consuming and rather expensive. The

1999 by CRC Press LLC

main advantage of data collected through experimental means is that the data are
probably the least influenced by the subjective elements that may procure some
error. One example of data based on the experimental findings is the Data Store [4].
Expert judgments are another approach to obtain human reliability data. This
approach is used quite often by human reliability methodologists and has two
attractive features: (i) it is comparatively easy to develop because a large amount of
data can be collected from a small number of expert respondents, and (ii) it is
relatively cheaper to develop. Some of the drawbacks of this method are less reliable
than data obtained through other means and frequent use of less experienced experts
than required.
In the case of self-made error reports, the person who makes the error also
reports that error. One of the drawbacks of this method is that people are generally
reluctant to confess making an error. The human data recorder approach calls for
the physical presence of a person to observe task performance and document events
as necessary. Some of the disadvantages of this approach are that it is expensive and
the observer may fail to recognize committed errors. In operational system testing
of human-machine systems, the human data recorder approach is used quite frequently. In the case of the automatic data recorder method, the use of instrumentation
permits the automatic recording of all operator actions. Two typical examples are the
Operational Performance Recording and Evaluation Data System (OPREDS) [35]
and the Performance Measurement System. The latter system was developed by the
General Physics Corporation for recording actions in nuclear power plant simulators.
The published literature approach calls for collecting the data from publications such
as journals, conference proceedings, and books.

10.9.1 SPECIFIC HUMAN ERROR DATA BANKS

AND

SOURCES

Over the years many data banks for obtaining human error related information have
been developed [9, 37, 38]. Nine such data banks listed in Table 10.1 are reviewed
in Reference 39.

TABLE 10.1
Human Error Data Banks
No.

Data bank title

1
2
3
4
5
6
7

Data Store [4]

Operational Performance Recording and Evaluation Data System (OPREDS) [36]
Nuclear Plant Reliability Data System [40]
Aerojet General Method [41]
Aviation Safety Reporting System [42]
Bunker-Ramo Tables [43]
Air Force Inspection and Safety Center Life Sciences Accident and Incident
Reporting System [44]
Safety Related Operator Action (SROA) Program [39]
Technique for Establishing Personnel Performance Standards (TEPPS) [39]

8
9

1999 by CRC Press LLC

Other important sources for obtaining human error related data are References 45
through 48. Over 20 sources for obtaining human reliability data are listed in
References 9 and 34.

10.9.2 BRIEF DESCRIPTION

DATA BANKS

OF

SELECTED HUMAN ERROR

This section briefly describes three important data banks for obtaining human error
related information [39].
Data Store. It has served as an important source for obtaining human
reliability data [49]. The Data Store was established in 1962 by the American Institute for Research, Pittsburgh [4] and it contains estimates for
time and human performance reliability. More specifically, this data bank
possesses data relating human error probabilities to design features. All
in all, Data Store is probably the most developed data bank for application
during the design process.
Aviation Safety Reporting System. Originally this data bank was developed by the National Aeronautics and Space Administration (NASA) and
contains information on civil aircraft accidents/incidents. However, the
data is based on voluntary reporting and it receives roughly 400 human
error reports per month.
Operation Performance Recording and Evaluation Data System. This
system was developed to collect data on operational human performance
by the U.S. Navy Electronics Laboratory, San Diego. The system permits
automatic monitoring of human operator actions and in the late 1960s and
early 1970s it was used to collect data on various U.S. Navy ships. Nonetheless, the system was restricted to switch turning and button-pushing.

10.9.3 HUMAN ERROR DATA

FOR

SELECTIVE TASKS

In order to give examples of the types of available human reliability data, Table 10.2
presents human error data for selective tasks taken from published sources [34].

10.10

PROBLEMS

1. Define the following terms:

Human reliability
Human error
Human factors
2. Give some examples of well-publicized catastrophic disasters due to
humans.
3. What are the important categories of errors that humans make?

1999 by CRC Press LLC

TABLE 10.2
Human Error Data for Selective Tasks

Number

Error/task description

Finding maintenance (scheduled)

approaches in maintenance manual
Reading gauge incorrectly
Requirement misinterpretation/
misunderstanding
Adjusting incorrectly
Incorrect servicing
Connecting hose incorrectly
Turning rotary selector switch to
certain position
Installation o-ring incorrectly

2
3
4
5
6
7
8

Performance
reliability

Errors per
plant-month
(boiling water
reactors)

Errors per
million
operations

0.997

0.0074

5000

0.9996

0.026
0.03

4700

66700

4. Describe the following terms:

Operator errors
Maintenance errors
Assembly errors
5. Discuss the human performance effectiveness vs. stress relationship.
6. List at least 12 available sources for obtaining human error data.
7. Assume that a person performs three independent tasks associated with
a maintenance job, i.e., tasks x, y, and z. Each task can either be performed
correctly or incorrectly. The task x is performed before task y and task y
is performed before task z. Obtain an expression for probability of not
successfully completing the maintenance job by developing a probability
tree.
8. A person is performing a time-continuous task with exponentially distributed time to human error. Calculate the following, if the persons error
rate is 0.05 errors/h:
Persons reliability over an 8-h period.
Persons mean time to human error.
9. An engineering system can fail either due to a human error or a nonhuman error. Both the human error and non-human error rates are 0.006
and 0.009 failures per hour, respectively. Calculate the system failure
probability due to a human error over a 100-h period.
10. A person is performing a time-continuous task with Rayleigh distributed
time to human error. Develop an expression for mean time to human error.

1999 by CRC Press LLC

10.11

REFERENCES

1. Chapnis, Man-Machine Engineering, Wadsworth Publishing Company, Belmont, CA,

1965.
2. Williams, H.L., Reliability evaluation of the human component in man-machine
systems, Elect. Manufact., April, 78-82, 1958.
3. Shapero, A., Cooper, J.I., Rappaport, M., Shaeffer, K.H., and Bates, C.J., Human
Engineering Testing and Malfunction Data Collection in Weapon System Programs,
WADD Technical Report, No. 60-36, Wright-Patterson Air Force Base, Dayton, OH,
February 1960.
4. Munger, S.J., Smith, R.W., and Payne, D., An Index of Electronic Equipment Operability: Data Store, American Institute for Research, Pittsburgh, PA, Report No. AIRC43-1/62 RP(1), 1962.
5. Proc. Symp. Quantification of Human Performance, sponsored by the Electronic Industries Association and the University of New Mexico, Albuquerque, NM, August 1964.
6. Askren, W.B., Ed., Proc. Symp. Reliability of Human Performance in Work, Report
No. AMRL-TR-67-88, Aerospace Medical Research Laboratories, Wright-Patterson
Air Force Base, Ohio, May 1967.
7. Regulinski, T.L., Ed., Special issue on human reliability, IEEE Trans. Reliability, 22,
August 1973.
8. Dhillon, B.S., On human reliability-bibliography, Microelectronics and Reliability,
20, 371-373, 1980.
9. Dhillon, B.S., Human Reliability: With Human Factors, Pergamon Press, New York,
1986.
10. Dhillon, B.S. and Yang, N., Human reliability: a literature survey and review, Microelectronics and Reliability, 34, 803-810, 1994.
11. Hagen, E.W., Ed., Human reliability analysis, Nuclear Safety, 17, 315-326, 1976.
12. Meister, D., Human factors in reliability, in Reliability Handbook, Ireson, W.G., Ed.,
McGraw-Hill, New York, 1966, 12.2-12.37.
13. Definitions of Effectiveness Terms for Reliability, Maintainability, Human Factors
and Safety, MIL-STD-721B, August 1966. Available from the Naval Publications and
Forms Center, 5801 Tabor Avenue, Philadelphia, PA.
14. Kenney, G.C., Spahn, M.J., and Amato, R.A., The Human Element in Air Traffic
Control: Observations and Analysis of Performance of Controllers and Supervisors
in Providing Air Traffic Control Separation Services, METREK Div., MITRE Corp.,
Report No. MTR-7655, December 1977.
15. Organizational Management and Human Factors in Quantitative Risk Assessment,
Report No. 33/1992 (Report 1), 1992, British Health and Safety Executive (HSE),
London.
16. Dhillon, B.S., Reliability technology in health care systems, Proc. IASTED Int. Symp.
Comp. Adv. Technol. Med., Health Care, Bioeng., 84-87, 1990.
17. Nobel, J.L. Medical devices failures and adverse effects, Pediat-Emerg. Care, 7,
120-123, 1991.
18. Bogner, M.S., Medical devices and human error, in Human Performance in Automated
Systems: Current Research and Trends, Mouloua, M. and Panasuraman, R., Eds.,
Lawrence Erlbaum Associates, Hillsdale, NJ, 1994, pp. 64-67.
19. Rook, L.W., Reduction of Human Error in Industrial Production, Report No. SCTM
93-63 (14), June 1962, Sandia Naitonal Laboratories, Albuquerque, NM.
20. Joos, D.W., Sabril, Z.A., and Husseiny, A.A., Analysis of gross error rates in operation
of commercial nuclear power stations, Nuclear Eng. Design, 52, 265-300, 1979.

1999 by CRC Press LLC

21. Meister, D., The problem of human-initiated failures, Proc. Eighth Natl. Symp. Reliability and Quality Control, 234-239, 1962.
22. Cooper, J.I., Human-initiated failures and man-function reporting, IRE Trans. Human
Factors, 10, 104-109, 1961.
23. McCornack, R.L., Inspector Accuracy: A Study of the Literature, Report No. SCTM
53-61 (14), 1961, Sandia Corporation, Albuquerque, NM.
24. Lee, K.W., Tillman, F.A., and Higgins, J.J., A literature survey of the human reliability
component in a man-machine system, IEEE Trans. Reliability, 37, 24-34, 1988.
25. Beech, H.R., Burns, L.E., and Sheffield, B.F., A Behavioural Approach to the Management of Stress, John Wiley & Sons, Chichester, 1982.
26. Hagen, E.W., Ed., Human reliability analysis, Nuclear Safety, 17, 315-326, 1976.
27. Regulinski, T.L. and Askern, B., Mathematical modeling of human performance
reliability, Proc. Annu. Symp. Reliability, 5-11, 1969.
28. Askern, W.B. and Regulinski, T.L., Quantifying human performance for reliability
analysis of systems, Human Factors, 11, 393-396, 1969.
29. Regulinski, T.L. and Askren, W.B., Stochastic modelling of human performance
effectiveness functions, Proc. Annu. Reliability Maintainability Symp., 407-416,
1972.
30. Swain, A.D., A Method for Performing a Human-Factors Reliability Analysis, Report
No. SCR-685, Sandia Corporation, Albuquerque, NM, August 1963.
31. Dhillon, B.S. and Rayapati, S.N., Reliability evaluation of multi-state device networks
with probability trees, Proc. Sixth Symp. Reliability in Electron., Hungarian Academy
of Sciences, Budapest, August 1985, pp. 27-37.
32. Dhillon, B.S., Stochastic models for predicting human reliability, Microelectronics
and Reliability, 21, 491-496, 1982.
33. Meister, D., Human Reliaiblity, in Human Factors Review, Muckler, F.A., Ed., Human
Factors Society, Santa Monica, CA, 1984, pp. 13-53.
34. Dhillon, B.S., Human error data banks, Microelectronics and Reliability, 30, 963-971,
1990.
35. Dhillon, B.S. and Singh, C., Engineering Reliability: New Techniques and Applications, John Wiley & Sons, New York 1981.
36. Urmston, R., Operational Performance Recording and Evaluation Data System
(OPREDS), Descriptive Brochures, Code 3400, Navy Electronics Laboratory Center,
San Diego, CA, November 1971.
37. Kohoutek, H.J., Human centered design, in Handbook of Reliability Engineering and
Management, Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Eds., McGraw-Hill,
New York, 1996, pp. 9.19.30.
38. LaSala, K.P., Human reliability: An overview, tutorial notes, Annu. Reliability Maintainability Symp., 145, 1992.
39. Topmiller, D.A., Eckel, J.S., and Kozinsky, E.J., Human Reliability Data Bank for
Nuclear Power Plant Operations: A Review of Existing Human Reliability Data
Banks, Report No. NUREG/CR2744/1, U.S. Nuclear Regulatory Commission, Washington, D.C., 1982.
40. Reporting Procedures Manual for the Nuclear Plant Reliability Data System
(NPRDS), South-West Research Institute, San Antonio, TX, December 1980.
41. Irwin, I.A., Levitz, J.J., and Freed, A.M., Human Reliability in the Performance of
Maintenance, Report No. LRP 317/TDR-63-218, Aerojet-General Corporation, Sacramento, CA, 1964.
42. Aviation Safety Reporting Program, FAA Advisory Circular No. 00-46B, Federal
Aviation Administration (FAA), Washington, D.C., June 15, 1979.

1999 by CRC Press LLC

43. Hornyak, S.J., Effectiveness of Display Subsystems Measurement Prediction Techniques, Report No. TR-67-292, Rome Air Development Center (RADC), Griffis Air
Force Base, Rome, NY, September 1967.
44. Life Sciences Accident and Incident Classification Elements and Factors, AFISC
Operating Instruction No. AFISCM, 127-6, U.S. Air Force, Washington D.C., December 1971.
45. Stewart, C., The Probability of Human Error in Selected Nuclear Maintenance Tasks,
Report No. EGG-SSDC-5580, Idaho National Engineering Laboratory, Idaho Falls,
ID, 1981.
46. Boff, K.R. and Lincoln, J.E., Engineering Data Compedium: Human Perception and
Performance, Vols. 1-3, Armstrong Aerospace Medical Research Laboratory, WrightPatterson Air Force Base, Ohio, 1988.
47. Gertman, D.I. and Blackman, H.S., Human Reliability and Safety Analysis Data
Handbook, John Wiley & Sons, New York, 1994.
48. Swain, A.D. and Guttmann, H.E., Handbook of Human Reliability Analysis with
Emphasis on Nuclear Power Plant Applications, Report No. NUREG/CR-1278, The
United States Nuclear Regulatory Commission, Washington, D.C., 1983.
49. Meister, D., Human reliability data base and future systems, Proc. Annu. Reliability
Maintainability Symp., 276-280, 1993.

1999 by CRC Press LLC

11

Reliability Testing
and Growth

11.1 INTRODUCTION
Just like in the case of any other reliability activity, reliability testing is an important
reliability task. In fact, it may be called one of the most important reliability activities
of a reliability program. The main purpose of reliability testing is to obtain information regarding failures, in particular, the product/equipment tendency to fail as
well as the failure consequences. This type of information is extremely useful in
controlling failure tendencies along with their consequences. A good reliability test
program may be classified as the one providing the maximum amount of information
concerning failures from a minimal amount of testing [1]. Over the years, many
important publications on reliability testing have appeared; in particular, two such
publications are listed in References 2 and 3.
In the design and development of new complex and sophisticated systems, the
first prototypes usually contain various design and engineering related deficiencies.
In fact, according to References 4 and 5, the reliability of revolutionary design
product/systems could be very low, i.e., 15 to 50% of their mature design capability.
It means without the initiation of various corrective measures during the development
stage to improve reliability, such products/systems reliability could very well
remain at the low initial value. Nonetheless, correcting the weaknesses or errors in
design, manufacturing methods, elimination of bad components, etc. leads to a
products reliability growth [6]. The term reliability growth may be defined as the
positive improvement in a reliability parameter over a time period because of changes
made to the product design or the manufacturing process [7]. Similarly, the term
reliability growth program may be described as a structure process for discovering
reliability related deficiencies through testing, analysis of such deficiencies, and
implementation of corrective measures to lower their occurrence rate.
The serious thinking concerning reliability growth may be traced back to the
late 1950s. In 1964, a popular reliability growth monitoring model was postulated
by Duane [8]. Comprehensive lists of publications up to 1980 on reliability growth
are given in Refences 9 and 10.
This chapter discusses reliability testing and growth.

11.2 RELIABILITY TESTING

A wide range of topics can be covered under this subject, including classification
of reliability tests, success testing, confidence interval estimates for mean time
between failures, and accelerated life testing. Each of these topics is presented below.

1999 by CRC Press LLC

11.2.1 RELIABILITY TEST CLASSIFICATIONS

Basically, reliability tests may be grouped under three classifications: (1) reliability
development and demonstration testing, (2) qualification and acceptance testing, and
(3) operational testing [11].
Reliability development and demonstration testing is concerned with meeting
objectives such as to indicate if any design changes are required, to determine if the
design is to be improved to satisfy the reliability requirement, and to verify improvements in design reliability. The nature of this type of testing depends on factors such
as the level of complexity under consideration, and the type of system/subsystem
being investigated. For example, in the case of electronic parts, the reliability development and demonstration testing could take the form of life tests to evaluate if the
part can satisfy its reliability goals; if not, what measures are necessary?
In order to meet reliability development and demonstration objectives effectively,
the accumulated test data must be of a kind that allows insight into the failure
probabilities/failure effects for a certain design under consideration. Furthermore,
these data serve as a good basis for reliability analysis and assessment for two
particular items: design under consideration and subsequent related programs.
Qualification and acceptance testing is concerned with meeting two basic objectives: to arrive at decision if a part/assembly/end item is to be accepted or not, and
to determine if a certain design is qualified for its projected use. These objectives
differ from the objectives of other reliability tests, particularly with regard to the
accept/reject mechanism. Qualification and acceptance testing incorporates the usual
testing and screening the quality-control function of incoming parts. With respect
to the materials and components to be used in the system/equipment under development, the qualification and acceptance testing starts early in the program.
Operational testing is concerned with objectives that include verifying the results
of reliability analysis conducted during the system/equipment design and development, providing data indicating desirable changes to operating policies and procedures with respect to reliability/maintainability, and providing data for subsequent
activities. All in all, the operational testing provides the feedback from practice to
theory.

11.2.2 SUCCESS TESTING

This type of testing is sometimes practiced in receiving inspection and in engineering
test laboratories where a no-failure test is specified. Usually, the main goal for this
test is to ensure that a certain reliability level has been achieved at a given confidence
level.
In this case, for zero failures, the lower 100 (1 )% confidence limit on the
desired reliability level can be written as [12]
R Low = 1 k

1999 by CRC Press LLC

(11.1)

where
k is the number items/units placed on test.
is the level of significance or consumers risk.
Thus, with 100 (1 )% confidence, it may be stated that
R Low R T

(11.2)

where
RT is the true reliability.
Taking the natural logarithms of the both sides of Equation (11.1) leads to
ln R Low =

1
ln
k

(11.3)

Thus, from Equation (11.3), we get

k=

ln
l n R Low

(11.4)

The desired confidence level, C, is expressed by

C = 1

(11.5)

= 1 C

(11.6)

Rearranging Equation (11.5), we get

Using relationships (11.2) and (11.6) in Equation (11.4), we get

k=

l n (1 C)
l n RT

(11.7)

Equation (11.7) can be used to determine the number of items to be tested for
specified reliability and confidence level.
Example 11.1
Assume that 95% reliability of a television set is to be demonstrated at 85% confidence level. Determine the number of television sets to be placed on test when no
failures are allowed.

1999 by CRC Press LLC

Inserting the above given data into Equation (11.7) yields

k=

l n (1 0.85)
l n 0.95

= 37
Thus, 37 television sets must be placed on test.

11.2.3 CONFIDENCE INTERVAL ESTIMATES

BETWEEN FAILURES

FOR

MEAN TIME

In many practically inclined reliability studies, the time to item failure is assumed
to be exponentially distributed. Thus, the item failure rate becomes constant and, in
turn, the mean time between failures (MTBF) is simply the reciprocal of the failure
rate (i.e., 1/, where is the item constant failure rate).
In testing a sample of items with exponentially distributed times to failures, a
point estimate of MTBF can be made but, unfortunately, this figure provides an
incomplete picture because it fails to give any surety of measurement. However, it
would probably be more realistic if we say, for example, that after testing a sample
for t hours, m number of failures have occurred and the actual MTBF lies somewhere
between specific upper and lower limits with certain confidence.
The confidence intervals on MTBF can be computed by using the 2 (chi-square)
distribution. The general notation used to obtain chi-square values is as follows:
2 ( p, d f )

(11.8)

where
p is a quantity, function of the confidence coefficient.
df is the degrees of freedom.
The following symbols are used in subsequent associated formulas [11, 13]:

C = 1-
k
m
m*

is the mean life or mean time between failures.

is the acceptable error risk.
is the confidence level.
is the number of items that were placed on test at zero time (i.e.,
t = 0).
is the number of failures accumulated to time t*, where t* denotes
the life test termination time.
is the number of preassigned failures.

There are two different cases for estimating confidence intervals:

Testing is terminated at a preassigned time, t*.
Testing is terminated at a preassigned number of failures, m*.

1999 by CRC Press LLC

Thus, for the above two cases, to compute upper and lower limits, the formulas that
can be used are as follows [11, 13]:
Preassigned truncation time, t*
2X
2X

2
, 2 m + 2 1 , 2 m

2
2

(11.9)

Preassigned number of failures, m*

2X
2X

2
, 2 m 1 , 2 m

2
2

(11.10)

The value of X is determined by the test types: replacement test (i.e., the failed unit
is replaced or repaired), non-replacement test.
Thus, for the replacement test, we have
X = k t*

(11.11)

Similarly, for the non-replacement test, we get

m

X = ( k m) t * +

(11.12)

i =1

where
t i is the ith failure time.
In the case of censored items (i.e., withdrawal or loss of unfailed items) the value
of X becomes as follows:
For replaced failed units but non-replacment of censored items
s

X = ( k s) t * +

t
j =1

where
s is the number of censored items
t j is the jth, censorship time.

1999 by CRC Press LLC

(11.13)

 For non-replaced failed and censored items

s

X = ( k s m) t * +

t
tj +

j =1

(11.14)

i =1

The tabulated values of 2 (p, df) are presented in Table 11.1.

Example 11.2
Assume that 16 identical medical devices were placed on test at time t = 0 and none
of the failed units were replaced and the test was terminated after 100 h. Four medical
devices failed after 10, 30, 50, and 90 h of operation. Estimate the medical devices
MTBF and their upper and lower limits with 90% confidence level.
Substituting the specified data into Equation (11.12) yields
X = (16 4) (100) + (10 + 30 + 50 + 90)
= 1380 h
Thus, the medical devices MTBF is given by
=

1380
= 345 h
4

Inserting the given and other values into relationship (11.9) and using Table 11.1
yields the following upper and lower limit values for the MTBF:
Upper limit =
=

2 (1380)
2 (0.95, 8)
2 (1380)
2.73

= 1010.9 h
Lower limit =
=

2 (1380)
(0.05,10)
2

2 (1380)
18.30

= 150.8 h

1999 by CRC Press LLC

TABLE 11.1
Values of Chi-Square
Probability

Degrees of
freedom

0.99

0.95

0.9

0.5

0.1

0.05

0.01

2
4
6
8
10
12
14
16
18
20
22
24
26
28
30

0.02
0.29
0.87
1.64
2.55
3.57
4.66
5.81
7.01
8.26
9.54
10.85
12.19
13.56
14.95

0.1
0.71
1.63
2.73
3.94
5.22
6.57
7.96
9.39
10.85
12.33
13.84
15.37
16.92
18.49

0.21
1.06
2.2
3.49
4.86
6.3
7.79
9.31
10.86
12.44
14.04
15.65
17.29
18.93
20.59

1.38
3.35
5.34
7.34
9.34
11.34
13.33
15.33
17.33
19.33
21.33
23.33
25.33
27.33
29.33

4.6
7.77
10.64
13.36
15.98
18.54
21.06
23.54
25.98
28.41
30.81
33.19
35.56
37.91
40.25

5.99
9.44
12.59
15.5
18.3
21.02
23.68
26.29
28.86
31.41
33.92
36.41
38.88
41.33
43.77

9.21
13.27
16.81
20.09
23.2
26.21
29.14
32
34.8
37.56
40.28
42.98
45.64
48.27
50.89

Thus, we can state with 90% confidence that the medical devices true MTBF
will lie within 150.8 h and 1010.9 h or 150.8 1010.9.
Example 11.3
Assume that 20 identical electronic parts were put on test at zero time and at the
occurrence of the tenth failure, the testing was stopped. The tenth failure occurred
at 150 h and all the failed parts were replaced.
Determine the MTBF of the electronic parts and upper and lower limits on
MTBF at 80% confidence level.
Inserting the given data into Equation (11.11) we get
X = (20) (150) = 3000 h
Thus, electronic parts MTBF is
=

3000
= 300 h
10

Substituting the specified and other values into relationship (11.10) and using
Table 11.1, we get the following values of MTBF upper and lower limits:

1999 by CRC Press LLC

2 (3000)
(0.90, 20)

Upper limit =
=

6000
12.44

= 482.3 h
Lower limit =
=

2 (3000)
2 (0.1, 20)
6000
28.41

= 211.1 h
At 80% confidence level, the electronic parts true MTBF will lie within 211.1 h
and 482.3 h or 211.1 482.3.
Example 11.4
On the basis of data given in Example 11.3, determine the following:
Probability of an electronic parts success for a 200-h mission.
Upper and lower limits on this probability at 80% confidence level.
Thus, the electronic parts reliability is
R (200) = e

200

300

= 0.5134
The upper and lower limit reliability are as follows:
Upper limit = e

200

482.3

= 0.6606
Lower limit = e

200

211.1

= 0.3877
Thus, the reliability of the electronic part is 0.5134 and its upper and lower
values at 80% confidence level are 0.6606 and 0.3877, respectively.

1999 by CRC Press LLC

11.2.4 ACCELERATED LIFE TESTING

This is used to obtain quick information on items reliabilities, failure rates, life
distributions, etc. by subjecting the test items to conditions such that the malfunctions
occur earlier. Thus, the accelerated life testing is a useful tool to make long term
reliability prediction within a short time span. The following two approaches are
employed to perform an accelerated life test [1417]:
Approach I. This is concerned with accelerating the test by using the
item under consideration more intensively than its normal usage. Usually,
the products such as a high bulb of a telephone set and a crankshaft of a
car used discretely or non-continuously can be tested by using this
approach. However, it is not possible to use this approach for an item
such as a mainframe computer in constant use. Under such a scenario,
the next approach can be used.
Approach II. This is concerned with conducting the test at very high
stress (e.g., temperature, humidity, and voltage) levels so that failures can
be induced in a very short time interval. Typical examples of items for
application under this approach are air traffic control monitors, components of a power-generating unit, and communication satellites. Usually,
accelerated failure time testing over the accelerated stress testing is preferred because there is no need to make assumptions regarding the relationship of time to failure distributions at both normal and accelerated
conditions. Nonetheless, the accelerated-stress testing results are related
to the normal conditions by using various mathematical models. One such
model will be presented subsequently.
Relationship Between the Accelerated and Normal Conditions
This section presents relationships between the accelerated and normal conditions
for the following items [14, 18]:
Probability density function
The normal operating condition failure probability density function is
expressed by
fn (t ) =

1 t
f
s

(11.15)

where
t
is time.
fn (t) is the normal operating condition failure probability density
function.

is the acceleration factor.

t is the stressful operating condition failure probability density
fs
function.

1999 by CRC Press LLC

 Cumulative distribution function

The normal operating condition cumulative distribution function is
t
Fn (t ) = Fs

(11.16)

where
Fs

t is the stressful operating condition cumulative distribution

function.

Time to failure
The time to failure at normal operating condition is given by
t n = ts

(11.17)

where
tn is the time to failure at normal operating condition.
ts is the time to failure at stressful operating condition.
Hazard rate
The normal operating condition hazard rate is given by
h n (t ) =

fn (t )
1 Fn (t )

(11.18)

Substituting Equations (11.15) and (11.16) into Equation (11.18) yields

1 t
f
n
h n (t ) =
t
1 Fs

(11.19)

Thus from Equation (11.19), we have

h n (t ) =

1 t
h
s

where
hs

t is the stressful operating condition hazard rate.

1999 by CRC Press LLC

(11.20)

Acceleration Model
For an exponentially distributed time to failure at an accelerated stress, s, the
cumulative distribution function is
Fs ( t ) = 1 e s t

(11.21)

where
s is the parameter or the constant failure rate at the stressful level.
Thus, from Equation (11.16) and (11.21), we get
t
Fn (t ) = Fs = 1 e s t

(11.22)

Similarly, using relationship (11.20), we have

n = s

(11.23)

where
n is the constant failure rate at the normal operating condition.
For both non-censored and censored data, the failure rate at the stressful level can
be estimated from the following two equations, respectively [14]:
Noncensored data
k

s = k

(11.24)

j =1

where
k is the total number of items under test at a certain stress.
tj is the jth failure time; for j = 1, 2, 3, 4, , k.
Censored data

s = q

1999 by CRC Press LLC

kq

t
tj +

j =1

j =1

(11.25)

where
q is the number of failed items at the accelerated stress.
tj is the jth censoring time.
Example 11.5
Assume that a sample of 40 integrated circuits were accelerated life tested at 135C
and their times to failure were exponentially distributed with a mean value of 7500 h.
If the value of the acceleration factor is 30 and the integrated circuits normal
operating temperature is 25C, calculate the integrated circuits, operating at the
normal conditions, failure rate, mean time to failure, and reliability for a 5000-h
mission.
In this case, the failure rate of the integrated circuits at the accelerated temperature is given by
s = 1 (integrated circuits' mean)life under accelerated testing
=

1
0.000133 failure h
7500

Inserting the above result and the specified data into Equation (11.23), we get
n

0.000133
30

4.4333 10 6 failure h

Thus, the integrated circuits mean time to failure (MTTFn) at the normal operating condition is
MTTFn =

1
n
225, 563.9 h

The integrated circuits reliability for a 5000-h mission at the normal operating
condition is

({

R (5, 000) = exp 4.4333 10 6 (5, 000)

0.9781

Thus, the integrated circuits failure rate, mean time to failure and reliability at
the normal operation are 4.4333 x 10-6 failure/h, 225,563.9 h, and 0.9781, respectively.

1999 by CRC Press LLC

FIGURE 11.1 Reliability growth program basic structure main factors.

11.3 RELIABILITY GROWTH

Over the years many advances have been made in the area of reliability growth.
These are in management, modeling, technical standards, etc. This section presents
different aspects of reliability growth.

11.3.1 RELIABILITY GROWTH PROGRAM

As described earlier, the reliability growth program is a structured process used to
discover reliability deficiencies through testing, analyzing such deficiencies, and
implementation of corrective measures to lower the rate of occurrence. Some of the
important advantages of the reliability growth program include assessments of
achievement and projecting the product reliability trends. Nonetheless, the fundamental structure of a reliability growth program consists of three basic factors as
shown in Figure 11.1. These are management, testing, and failure reporting, analysis,
and corrective action system (FRACAS) [5]. The management is concerned with
making it happen and the testing provides opportunity to discover deficiencies in
areas such as design, components, manufacturing, and procedures. FRACAS is
primarily a process used for determining failure root causes and recommending
appropriate corrective measures.
Some of the major management considerations concerning a reliability growth
program are as follows:
The level of the estimated initial reliability without reliability growth
testing and its acceptability. Obviously, this information is related to how
much of the system/product is new technology.
Determination of the reliability growth slot within the system/product
development program.
Reliability growth programs impact on cost and schedule.

1999 by CRC Press LLC

11.3.2 RELIABILITY GROWTH PROCESS EVALUATION APPROACHES

A manager can basically use two approaches to evaluate the reliability growth
process. Each of these is described below [7].
Approach I. This approach makes use of assessments (i.e., quantitative
evaluations of the current reliability status) based on data from the detection
of failure sources; thus, it is a results oriented approach. In order to evaluate
reliability growth progress, a comparison between the assessed and the
planned values is made. Subsequently, the appropriate decisions are taken.
Approach II. This approach monitors various process activities to ensure
that such activities are being performed in a timely manner and the effort
and work quality are as per the program plan. Thus, this approach is
activity oriented and is practiced to supplement the assessments. Sometimes early in the program it is relied upon entirely. The main reason for
this is the lack of adequate objective data in the initial program stages.

11.4 RELIABILITY GROWTH MODELS

Over the years many mathematical models for reliability growth have been developed. Ten such models are described in detail in Reference 10. Furthermore,
Reference 7 presents an overview of 17 reliability growth models by grouping them
into two categories: discrete and continuous. Two such models are presented below.

11.4.1 DUANE MODEL

In 1964, J.T. Duane was the first person to report the most commonly accepted pattern
for reliability growth. The Duane model basically is a graphical approach to perform
analysis of reliability growth data and is simple and straightforward to understand.
Nonetheless, the two important benefits of this approach are as follows [2]:
The straight line used by the Duane plot can be fitted by eye to the data
points.
Various facts can be depicted by the Duane plot which otherwise could
be hidden by statistical analysis. For example, even though the application
of a goodness-of-fit test may conclude the rejection of a certain reliability
growth model, it will not provide any possible reasons for the rejection.
On the other hand, a plot of the same data might provide some possible
reasons for the problem.
In contrast, the drawbacks of this model are the reliability parameters cannot be
estimated as well in comparison to a statistical model and no interval estimates can
be calculated. Nonetheless, with respect to the development of this model Duane
collected data for several systems developed by General Electric to determine if any
systematic changes in reliability improvement took place during their development.
After performing analysis of these data, he concluded that a plot of cumulative hours

1999 by CRC Press LLC

vs. cumulative failure rate fell close to a straight line on log-log paper, under the
maintenance of a continued reliability effort. Thus, he defined the cumulative failure
rate of his model as follows:
k =

f
= T
T

(11.26)

where
k is the cumulative failure rate.
is a parameter denoting the growth rate.
is a parameter determined by circumstances such as product complexity,
design margin, and design objective.
T is the total test hours.
f denotes failures during T.
In order to estimate the values of the parameters, we take the logarithms of both
sides of Equation (11.26) to get
log k = log log T

(11.27)

Equation (11.27) is the equation for a straight line. Thus, the plot of the logarithm
of the cumulative failure rate, k, against the logarithm of cumulative operating
hours, T, can be used to estimate the values of and . The slope of the straight
line is equal to and at T = 1, is equal to the corresponding cumulative failure rate.
The least-squares method can be used to have a more accurate straight line fit
in estimating and [10].

11.4.2 ARMY MATERIAL SYSTEM ANALYSIS

ACTIVITY (AMSAA) MODEL
This is another model that can be used to track reliability growth within test
phases [7, 19, 20]. The model allows, for the purpose of reliability growth tracking,
the development of rigorous statistical procedures. The following two assumptions
are associated with this model:
Reliability growth can be modeled as a nonhomogeneous Poisson process
(NHPP) within a test phase.
On the basis of failures and test time within a test phase, the cumulative
failure rate is linear on log-log scale.
The following symbols are associated with this model:
t
M(t)

1999 by CRC Press LLC

is the test time from the start of the test phase.

is the number of system failures by time t.

E [M(t)] is the expected value of M (t).

and are the parameters.
Thus, the expected value of M(t) is expressed by
E [M(t )] = t

(11.28)

By differentiating Equation (11.28) with respect to t, we get the following expression

for the intensity function:
(t ) = t 1

(11.29)

where
(t) is the intensity function.
The instantaneous mean time between failures is expressed by
m (t ) = 1 (t )

(11.30)

where
m (t) is the instantaneous mean time between failures.
The model is described in detail in References 2 and 7.

11.5 PROBLEMS
1. Discuss the following:
Demonstration testing
Qualification and acceptance testing
Success testing
2. What is the difference between reliability demonstration testing and reliability growth modeling?
3. Assume that 99% reliability of a medical device is to be demonstrated at
60% confidence level. Calculate the total number of medical devices to
be placed on test when no failures are allowed.
4. What are the two approaches used to perform an accelerated life test?
Describe them in detail.
5. What are the benefits of reliability growth testing?
6. Twenty-five identical electrical devices were put on test at zero time, none
of the failed units were replaced, and the test was stopped after 200 h.
Seven devices malfunctioned after 14, 25, 30, 65, 120, 140, and 150 h of
operation. Calculate the electrical devices mean time between failures
and their associated upper and lower limits with 80% confidence level.

1999 by CRC Press LLC

7. Define the following two terms:

Reliability growth
Reliability growth program
8. A sample of 60 identical electronic parts were accelerated life tested at
140C and their times to failure were represented by an exponential
distribution with a mean value of 4000 h. The normal operating temperature of these parts is 30C and the value of the acceleration factor is 28.
Determine the mean time to failure, failure rate, and reliability for a 100-h
mission under the normal operating conditions.
9. What are the major management considerations concerning a reliability
growth program?
10. A sample of 30 identical mechanical devices were placed on test at time
t = 0 and at the occurrence of the seventh failure, the testing was terminated. The last or the seventh failure occurred at 200 h and all the failed
devices were replaced. Calculate the MTBF of the mechanical devices
and the upper and lower limits on MTBF at 70% confidence level.

11.6 REFERENCES
1. AMC Pamphlet AMCP 702-3, Quality Assurance, U.S. Army Material Command,
Washington, D.C., 1968.
2. MIL-HDBK-781, Reliability Test Methods, Plans and Environments for Engineering
Development, Qualification and Production, Department of Defense, Washington,
D.C.
3. MIL-STD-781, Reliability Design Qualification and Production Acceptance Test:
Exponential Distribution, U.S. Department of Defense, Washington, D.C.
4. Benton, A.W. and Crow, L.H., Integrated reliability growth testing, Proc. Annu.
Reliability Maintainability Symp., 160-166, 1989.
5. Crow, L.H., Reliability growth management, models, and standards, in tutorial notes,
Annu. Reliability Maintainability Symp., 1-12, 1994.
6. Mead, P.H., Reliability growth of electronic equipment, Microelectronics and Reliability, 14, 439-443, 1975.
7. MIL-HDBK-189, Reliability Growth Management, Department of Defense, Washington, D.C.
8. Duane, J.T., Learning curve approach to reliability monitoring, IEEE Trans. Aerospace, 563-566, 1964.
9. Dhillon, B.S., Reliability growth: A survey, Microelectronics and Reliability, 20,
743-751, 1980.
10. Dhillon, B.S., Reliability Engineering in Systems Design and Operation, Van Nostrand Reinhold Company, New York, 1983.
11. Von Alven, W.H., Ed., Reliability Engineering, Prentice-Hall, Englewood Cliffs, NJ,
1964.
12. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Handbook of Reliability Engineering
and Management, McGraw-Hill, New York, 1996.
13. Dhillon, B.S., Systems Reliability, Maintainability and Management, Petrocelli
Books, New York, 1983.
14. Elsayed, E.A., Reliability Engineering, Addison Wesley Longman, Reading, MA,
1996.

1999 by CRC Press LLC

15. Nelson, W., Accelerated Testing, John Wiley & Sons, New York, 1980.
16. Bain, L.J. and Engelhardt, M., Statistical Analysis of Reliability and Life-Testing
Models: Theory, Marcel Dekker, New York, 1991.
17. Meeker, W.Q. and Hahn, G.J., How to Plan an Accelerated Life Test: Some Practical
Guidelines, American Society for Quality Control (ASQC), Milwaukee, WI, 1985.
18. Tobias, P.A. and Trindade, D., Applied Reliability, Van Nostrand Reinhold Company,
New York, 1986.
19. Crow, L.H., Estimation procedures for the Duane model, Proc. U.S. Army Mater.
Syst. Anal. Act. (AMSAA) Reliability Growth Symp., Aberdeen Proving Ground,
Maryland, September 1972.
20. Crow, L.H., Reliability analysis for complex repairable systems, in Reliability and
Biometry, Proschan, F. and Serfling, R.J., Eds., Society for Industrial and Applied
Mathematics (SIAM), Philadelphia, PA, 1974, pp. 379-410.

1999 by CRC Press LLC

12

Reliability in
Computer Systems

12.1 INTRODUCTION
In recent years, the applications of computers have increased quite dramatically
ranging from personal use to control space systems. Todays computers are much
more complex and powerful than their infant versions. As the computers are made
up of both the software and hardware components, the percentage of the total
computer cost spent on software has changed quite remarkably from the days of the
first generation computers. For example, in 1955 software (i.e., including software
maintenance) accounted for 20% of the total computer cost and in 1985, the software
cost increased to 90% [1].
For an effective performance of a computer, both its hardware and software must
function with considerable reliability. In fact, according to various past studies, the
software is many times more likely to fail than hardware. Nonetheless, as computers
are becoming more complex and sophisticated, the demand on their reliability has
increased exponentially. For example, NASAs Saturn V Launch computer (circa
1964) had a mean time to failure (MTTF) goal of 25,000 h and the SIFT and FTMP
avionic computers designed and developed in the late 1970s to control dynamically
unstable aircraft were expected to have a MTTF of 1 billion hours [2, 3]. The early
history of computer hardware reliability may be traced back to the works of
Shannon [4], Hamming [5], Von Neumann [6], and Moore and Shannon [7]. For
example, the triple modular redundancy (TMR) scheme was first proposed by Von
Neumann [6] to improve system reliability in 1956. Since then, many other people
have contributed to the computer hardware reliability and a comprehensive list of
publications on the subject is given in Reference 8.
The serious effort on software reliability appeared to have started at Bell Laboratories in 1964 [9]. An evidence of this effort is a histogram of problems per month
concerning switching system software. In 1967, Floyd [10] considered approaches
for formal validation of software programs and Hudson [11] proposed Markov
birthdeath models. Since those years, a large number of publications on the subject
have appeared [8].
This chapter presents important aspects of computer hardware and software
reliability.

12.2 TERMS AND DEFINITIONS

As there are certain terms specifically concerned with computer hardware and
software reliability, this section presents some of those terms along with their
definitions [8, 12-15].

1999 by CRC Press LLC

 Software error. This is a clerical, conceptual, or syntactic discrepancy

which leads to one or more faults in the software.
Fault. This is an attribute which adversely affects an items reliability.
Covered fault. This is a fault from which the concerned item can recover
automatically.
Uncovered fault. This is a fault from which the concerned item cannot
recover automatically.
Debugging. This is the process of eradicating and isolating errors.
Fault-tolerant computing. This is the ability to execute specified algorithms successfully irrespective of computer hardware failures and software errors [16].
Software reliability. This is the probability of a given software operating
for a stated time interval, without an error, when used under the designed
conditions on the stated machine.
Software testing. This is the process of executing software to determine
if the results it produces are correct [15].

12.3 HARDWARE RELIABILITY VS.

SOFTWARE RELIABILITY
In order to have a better understanding of the differences between computer hardware
and software reliability, Table 12.1 presents comparisons of the important concerned
areas [8, 1719].

12.4 COMPUTER FAILURE CAUSES

Computers fail due to many different causes. The major sources of computer failures
are shown in Figure 12.1 [20, 21]. These are human errors, communication network
failures, peripheral device failures, environmental and power failures, processor and
memory failures, mysterious failures, saturation, and gradual erosion of the data
base. The first six of these system failure sources are described below.
Human errors are, in general, the result of operator oversights and mistakes.
Often operator errors occur during starting up, running, and shutting down the
system. Communication network failures are associated with inter module communication and, usually, most of them are of a transient nature. The use of vertical
parity logic can help to detect over two-thirds of errors in communication lines.
Peripheral device failures are important but they seldom cause a system shutdown.
The frequently occurring errors in peripheral devices are transient or intermittent
and the usual reason for their occurrence is the electromechanical nature of the
devices. The causes, such as electromagnetic interference, air conditioning equipment failure, earthquakes, and fires, are responsible for the occurrence of environmental failures. Power failures can occur due to factors such as transient fluctuations
in voltage or frequency and total power loss from the local utility company.
Usually, processor errors are catastrophic but they occur quite rarely, as there
are occasions when the central processor fails to execute instructions properly due

1999 by CRC Press LLC

TABLE 12.1
Comparison of Hardware and Software Reliability
No.

Hardware reliability

Software reliability

1
2

Wears out.
Many hardware parts fail according to the
bathtub hazard rate curve.
A hardware failure is mainly due to physical
effects.
The failed system is repaired by performing
corrective maintenance.
Interfaces are visual.
The hardware reliability field is well established,
particularly in the area of electronics.
Obtaining good failure data is a problem.
Usually redundancy is effective.
Potential for monetary savings.
Preventive maintenance is performed to inhibit
failures.
It is possible to repair hardware by using spare
modules.
Hardware reliability has well-established
mathematical concepts and theory.
Has a hazard function.
Mean time to repair has significance.

Does not wear out.

Software does not fail according to the bathtub
hazard rate curve.
A software failure is caused by programming
error.
Corrective maintenance is really redesign.

3
4
5
6
7
8
9
10
11
12
13
14
15

Reliability is time-related with failures

occurring as a function of operational/storage
time.

Interfaces are conceptual.

The software reliability field is relatively new.
Obtaining good failure data is a problem.
Redundancy may not be effective.
Potential for monetary savings.
Preventive maintenance has no meaning in
software.
It is impossible to repair software failures by
using spare modules.
Software reliability still lacks well-established
mathematical concepts and theory.
Has an error rate function.
Mean time to repair does not have any
significance.
Reliability is not time-related and failures occur
when a program step/path containing the fault
is executed.

FIGURE 12.1 Major sources of computer failures.

1999 by CRC Press LLC

to a dropped bit. Nowadays, memory parity errors are very rare because of an
impressive improvement in hardware reliability and also they are not necessarily
fatal. Mysterious failures occur unexpectedly and thus in real-time systems such
failures are never properly categorized. For example, when a normally functioning
system stops operating suddenly without indicating any problem (i.e., software,
hardware, etc.), the malfunction is called the mysterious failure.

12.5 SOFTWARE LIFE CYCLE PHASES AND

ASSOCIATED ERROR SOURCES
A software life cycle may be divided into many different phases: requirements
definition and analysis, preliminary design, detailed design, code and unit testing,
integration and system testing, acceptance testing, maintenance and operation, and
retirement. Within each of these phases, there are various sources of error, thus
requiring careful consideration. All of these life cycle phases are described
below [12].
The requirements definition and analysis phase defines and documents the engineering requirements for the software to be developed and its end product is a
software requirement specification that allows the development organization to proceed with the design. Some of the sources that may generate errors in this phase
include incomplete requirements, conflicting requirements, inadequately described
user needs, and nonfeasible requirements.
In the preliminary design phase, the software product is decomposed into a
hierarchal structure consisting of computer software configuration items. In turn,
these configuration items are decomposed further into computer software components and modules. During the detailed design phase, the computer software component design specifications are progressively described in more detail until the
identification of a set of primitive (i.e., non-decomposable) computer software units.
Some of the sources that may induce errors in the detailed design phase are inconsistent data definitions, input data range errors, inadequate validity checking, unsatisfactory error recovery, and wrong error analysis for algorithms.
In the code and unit testing phase, the programmer translates the detailed software design into the programming language and the unit testing is performed to
verify unit functionality. As coding is a fault-prone process, a number of standard
bugs that may occur during this phase include missing code, incomplete initialization
or resetting of variables and parameters, unreachable code, improper logic for branch
conditions and loops, out-of-range calculations and infinite loops, and improper or
incomplete validity checks.
At the completion of the unit tests, the integration testing starts and the separately
tested units are assembled into logical combinations and tested again to demonstrate
that the assembly adequately satisfies product requirements. The process is repeated
by combining larger and larger classes of units until the full integration of all units
of a computer software configuration item. The purpose of the system testing is to
demonstrate that the software and hardware work together to satisfy all system
functions as well as that the end product is ready for release to production. The

1999 by CRC Press LLC

errors/faults introduced in the test phase significantly influence the test programs
capability to uncover faults in the software and those errors/faults include test
plans/procedures that have wrongly interpreted software requirements and
errors/faults in code written for the test program.
In the acceptance testing phase, the test team determines if the product under
consideration meets its original requirements. The acceptance test plan is developed
and the phase terminates when all the tests in the acceptance plan are successfully
executed.
In the maintenance and operation phase, primarily the attention is focused on
rectifying errors that appear in the exercise of the software or on fine-tuning the
software to improve its performance.
In the retirement phase, the user may decide not to use the software any more
and discard it altogether. The primary reason is that the software may have become
difficult to maintain because of recurrent changes.

12.6 SOFTWARE RELIABILITY

IMPROVEMENT METHODS
There are a number of methods to improve reliability of software products. The
classifications of these methods include reliable software design, fault-tolerant
design, formal methods, and testing.

12.6.1 RELIABLE SOFTWARE DESIGN METHODS

Under this classification, there are a number of techniques that can help programmers
to systematically derive the software design from its detailed specification [22].
These include structured programming, top-down programming, and data structure
design.
The clarity of the design is the direct aim of the structured programming. More
specifically, the structured programming calls for easy to understand design and a
need to develop a program structure in which the vital software elements and their
inter-relationships are easily identifiable. Some of the easy to use rules associated
with structure programming include restricted use of GO TO statements in modules,
ensuring that each module has only one entry and one exit, using a language and
compiler with structured programming capabilities for the implementation purpose,
and ensuring that all statements, including subroutine calls, are commented [23].
There are many advantages associated with the structured programming: increase
in the productivity of programmers with respect to instructions coded per man-hour,
a useful tool in localizing an error, maximum amount of code can be reused in
redesign work, and an effective tool in understanding the program design by the
designer and others. The top-down programming or design is also sometimes referred
to as hierarchic decomposition because of the cascading nature of the process [24].
Nonetheless, the top-down design is a decomposition process concerned with directing attention to the program control structure or to the program flow of control.
Furthermore, the top-down design is also concerned with the production of code at
a later stage.

1999 by CRC Press LLC

The top-down programming begins with a module representing the program in

its entirety and it decomposes the module into subroutines. In turn, each subroutine
is broken down further and this process continues until all the broken down elements
are easy to comprehend and work with. The program structure is analogous to a
tree; the trunk of the tree represents the module containing the entire program and
its leaves represent the computer language statements. Some of the benefits of the
top-down structured programming are reduction in the cost of testing, better quality
software, software in a more readable form, lower cost of software maintenance,
increase in confidence in software, etc.
The data structure design is inclined toward the development of a detailed
program design and is also referred to as Jackson Structured Programming. The
structure of the input files drive the methodology and its resulting design and from
these files software is read and by the structure of the output files it is generated.
According to this approach, an effective software design reproduces the data structure in the input/output files. Currently, this is the most systematically available
design method and is simple, teachable, consistent (e.g., two totally different programmers will end up writing the same program), and anti-inspirational. The downside is that the method is not applicable to scientifically based programs because
their contents are essentially algorithmic (i.e., functional).

12.6.2 FAULT-TOLERANT SOFTWARE DESIGN METHODS

It would never be possible to produce error-free software, irrespective of the degree
of testing or other measures [25, 26]. Under such circumstances, one approach to
improve software reliability is to create a fault-tolerant architecture to recover from
software failures. The expensiveness of the approach usually limits applications to
high-risk areas, for example, embedded software for controlling satellites. Three
methods of fault-tolerant software design are recovery-block design, N-version programming, and consensus recovery block. The recovery-block design is concerned
with developing versions of software under the same set of requirements on the
assumption that the pieces of software being developed are independent, and thus
the likelihood of their failing simultaneously is negligible. The versions are ordered
from the most reliable to the least reliable. The most reliable software is executed
first and after the acceptance test run, if its output is rejected then a recovery block
is entered. The recovery is conducted in the following three steps:
Recover input.
Execute most reliable software and restore input.
Submit the output to the same acceptance test as before. In the event of
rejection, the next recovery block is entered and the process continues.
In the case of N-version programming, many versions of the software are developed independently, N programs are executed in parallel or simultaneously. Subsequently the resulting outputs are compared and when at least programs have the
same output, it is taken as the correct result and the process continues. This approach
is not suitable where multiple correct outputs can be generated and also it discriminates correct results or solutions affected by rounding errors.

1999 by CRC Press LLC

The consensus recovery block method combines attributes of the preceding two
approaches; thus, it attempts to discard the weaknesses of those two methods.
Nonetheless, this approach requires the development of N-versions of a software
and of an acceptance test voting procedure. The reliability factor is used to rank the
different versions of the software after execution of all software versions. The
resulting outputs are submitted to the voting mechanism. If no agreement is reached,
the order of reliability is used in submitting each output successively to the acceptance test. As soon as one of the resulting outputs passes the test, the process
terminates and the software under consideration continues with its operation. All in
all, it may be said that this method is more reliable than recovery-block design and
N-version programming approaches.

12.6.3 FORMAL METHODS

These include formal specification approaches and formal verification [12, 27]. The
formal specification approaches consist of a detailed specification of the software
requirements using a formal language. More specifically, the nature language of
traditional programming is not used, but rather a formal substitute having exactly
defined semantics and syntax. A typical example of such a language is mathematics.
Nonetheless, the formal specification approaches help to enforce clarity in requirement definitions and leave little room for ambiguity and misinterpretation, which is
usually a concern when using a natural language.
The formal verification makes use of inference on the formal specification in
order to demonstrate that the software meets its objectives [28]. All in all, it may
be said that the formal approaches are not properly understood by the involved
professionals because of their complexity and need for considerable training.

12.6.4 TESTING
This may simply be described as the process of executing a software program to
uncover errors. There are many different types of software testing including module
testing, top-down testing, bottom-up testing, and sandwich testing [8, 2931]. Each
of these testing methods is discussed below.
Module Testing. This is concerned with testing each module usually in
isolation from the rest of the system, but subject to the environments to
be experienced in the system. Usually, a general module test program is
developed instead of writing an entirely new program for each module to
be tested. Because it is easy to correct uncovered errors and more grave
consequences of errors can be discovered at a later stage, it is advisable
to conduct module testing thoroughly [32].
Top-Down Testing. This is concerned with integrating and testing the
software program from the top end to the bottom end and in the program
structure, the top module is the only module that is unit tested in isolation.
At the conclusion of the top module testing, this module calls the remaining modules one by one to merge with it. In turn, each combination is
tested and the process continues until the completion of combining and

1999 by CRC Press LLC

testing of all the modules [31]. The advantages of the top-down testing
are efficiency in locating problem areas at the higher program levels, easy
representation of test cases after adding the input/output functions, and
early demonstrations because of the early skeletal program. Similarly,
some of the drawbacks of the top-down testing are that stub modules must
be written, stub modules may be complex, test conditions could be impossible or difficult to create, and the representation of test cases in stubs can
be difficult prior to adding the input/output functions [12].
Bottom-Up Testing. This is concerned with integrating and testing the
software program from the bottom end to the top end and the terminal
modules are module (unit) tested in isolation. These modules do not call
any other modules and at the conclusion of the terminal module testing,
the modules that directly call these tested modules are the ones in line
for testing. The modules are not tested in isolation; in fact, the testing is
conducted together with the earlier tested lower level modules. The process continues until reaching the top of the program. Some of the benefits
of the bottom-up testing are ease in creating test conditions, efficiency in
locating problem areas at the lower levels of the program, and ease in
observating test results. Similarly, the bottom-up testing drawbacks
include: driver modules must be produced, and the program does not exist
as an entity until the adding of the final module [12].
Sandwich Testing. This method is the result of combining the top-down
testing and bottom-up testing methods. Obviously, the idea behind developing this approach was to extract benefits of both top-down testing and
bottom-up testing and eliminate some of their drawbacks. As both topdown testing and bottom-up testing are performed simultaneously, the
program is integrated from both sides (i.e., top and bottom). Even though
the resulting meeting point from both sides depends upon the program
being tested, it can be predetermined by reviewing the structure of the
program.

12.7 SOFTWARE RELIABILITY ASSESSMENT METHODS

There are many qualitative and quantitative software reliability assessment methods
available in the published literature. Such methods may be grouped into three distinct
classifications [12]:
1. Analytical approaches
2. Software metrics
3. Software reliability models

12.7.1 ANALYTICAL APPROACHES

Two important techniques belonging to the analytical approaches are failure mode
and effect analysis (FMEA) and fault tree analysis (FTA). Both of these methods
are widely used to assess hardware reliability and they can equally be applied to

1999 by CRC Press LLC

assess reliability of software. FMEA and FTA are described in detail in Chapters 6 and
7, respectively.

12.7.2 SOFTWARE METRICS

These may be described as quantitative indicators of the degree to which a software
item or process possesses a specified attribute. Often, the application of the software
metrics is made to determine the status of a trend in a software development activity
as well as to determine risk of going from one phase to another. Two software
measures are described below.
Defect Density Measure
This is an important measure and can be applied to determine the reliability growth
during the design phase. The measure requires establishing defect severity categories
and possesses some ambiguity, since its low value may mean either an inadequate
review process or a good product. Nonetheless, the cumulative defect ratio for design
is expressed by

cd =

(12.1)

i =1

where
cd is the cumulative defect ratio for design.
is the total number of reviews.
Ni is the total number of unique defects, at or above a given severity level,
discovered in the ith design review.
L is the number of source lines of design statement expressed in thousands,
in the design phase.
In the event of having estimates of defect density greater than the ones for
comparable projects, review the development process to determine if poor training/practices are responsible or if the requirements are ambiguous or incomplete.
Under such circumstance, it may be the correct course to delay development until
such time the corrective measures can be taken. In contrast, if estimates of defect
density are less than for comparable projects, review the methodology and the review
process itself. If the assessed review process is considered to be satisfactory, it is
quite reasonable to conclude that the development phases are generating low-defect
software products.
Code and Unit Test Phase Measure
For this phase, another form of the defect density measure is more appropriate and
again this form requires the establishment of defect severity classifications. Thus,
the cumulative defect ratio for code is expressed by

1999 by CRC Press LLC

cd =

(12.2)

SL

i =1

where
cd is the cumulative defect ratio for code.
is the total number of reviews.
Mi is the total number of unique defects, at or above a given severity level,
discovered in the ith code review.
SL is the number of source lines of code reviewed, expressed in thousands.

12.7.3 SOFTWARE RELIABILITY MODELS

Even though over the years many software reliability models have been developed
[33-35], the first one appeared in 1967 [36]. Most of the current software reliability
models may be grouped into four classifications as presented in Table 12.2 [12]. The
two important assumptions associated with classification I models are indigenous
and seeded faults have equal probability of detection and the seeded faults are
distributed randomly in the software program under consideration. Mills seeding
model falls under this category [37].
The key assumptions associated with the classification II models include independent test intervals, homogeneously distributed testing during intervals, and independent faults discovered during non-overlapping time intervals. Examples of the
models belonging to this classification are Musa model [38] and the Shooman
model [39].
There are many key assumptions associated with the classification III models:
independent embedded faults, equal probability of exposure of each fault, correction
process does not introduce faults, and independent times between failures. Two

TABLE 12.2
Classification of Software Reliability Models
No.

Classification

Fault seeding

II

Failure count

III

Times between failures

IV

Input domain based

1999 by CRC Press LLC

Description
This incorporates those models that determine the number of faults in
the program at zero time via seeding of extraneous faults.
This includes models counting the number of failures/faults occurring
in given time intervals.
This incorporates models providing the time between failure
estimations.
This incorporates models that determine the program/software
reliability under the circumstance the test cases are sampled randomly
from a known operational distribution of inputs to the
program/software.

examples of the models belonging to this classification are the Jelinski and Moranda
model [40] and the Schick and Wolverton model [21].
The classification IV models, i.e., input domain based models, have three key
assumptions: (1) inputs selected randomly, (2) input domain can be partitioned into
equivalence groups, and (3) known input profile distribution. Two examples of this
classification model are the Nelson model [42] and the Ramamoorthy and Bastani
model [43].
Some of the software reliability models are presented below.
Mills Model
This is a different and more pragmatic approach to software reliability prediction
proposed by Mills [37] in 1972. Mills argues that an assessment of the faults remaining in a given software program can be made through a seeding process that makes
an assumption of a homogeneous distribution of a representative category of faults.
Prior to starting the seeding process, a fault analysis is required to determine the
expected types of faults in the code and their relative occurrence frequency. Nonetheless, an identification of both seeded and unseeded faults is made during reviews
or testing and the discovery of seeded and indigenous faults allows an assessment
of remaining faults for the fault type under consideration. It is to be noted that the
value of this measure can only be estimated, provided the seeded faults are discovered.
The maximum likelihood of the unseeded faults is expressed by [12]

N1 = Nsf m fu msf

(12.3)

where
N1
Nsf
mfu
msf

is
is
is
is

the
the
the
the

maximum likelihood of the unseeded faults.

number of seeded faults.
number of unseeded faults uncovered.
number of seeded faults discovered.

Thus, the number of unseeded faults remaining in a given program is

N = N1 m fu

(12.4)

Example 12.1
A software program under consideration was seeded with 35 faults and, during
testing, 60 faults of the same type were found. The breakdowns of the faults uncovered were 25 seeded faults and 35 unseeded faults. Estimate the number of unseeded
faults remaining in the program.
Substituting the given data into Equation (12.3) yields
N1 = (35) (35) 25 = 49 faults

1999 by CRC Press LLC

Inserting the above result and the other specified data value into Equation (12.4),
we get
N = 49 35 = 14 faults
It means 14 unseeded faults still remain in the program.
Musa Model
This model belongs to classification II models of Table 12.2 and is based on the
premise that reliability assessments in the time domain can only be based upon
actual execution time, as opposed to calender or elapsed time. The reason for this
is that only during execution does the software program become exposed to failureprovoking stress. Nonetheless, this model should only be used after the completion
of integration or, more specifically, when all the relevant modules are in place [44].
Some of the important assumptions associated with the model are as follows:
Execution time between failures is piecewise exponentially distributed.
Failure intervals are statistically independent and follow Poisson distribution.
Failure rate is proportional to the remaining defects.
A comprehensive list of assumptions may be found in Reference 8. Musa developed the following simplified equation to obtain the net number of corrected faults:

m = M 1 exp ( kt M Ts )

(12.5)

where
m
M
Ts
t
k

is the net number of corrected faults.

is the initial number of faults.
is the mean time to failure at the beginning of the test.
is time.
is the testing compression factor defined as the mean ratio of detection
rate of failures during test to the rate during normal use of the software
program.

Mean time to failure, T, increases exponentially with execution time and is

expressed by
T = Ts exp ( kt M Ts )

(12.6)

Thus, the reliability at operational time t is

R (t ) = exp ( t T )

1999 by CRC Press LLC

(12.7)

From the above relationships, we obtain the number of failures that must occur to
improve mean time to failure from, say, T1 to T2 [44]:
1
1
m = M Ts
T1 T2

(12.8)

The additional execution time required to experience m is given by

t =

M Ts
ln (T2 T1 )
k

(12.9)

Example 12.2
A newly developed software is estimated to have approximately 450 errors. Also,
at the beginning of the testing process, the recorded mean time to failure is 4 h.
Determine the amount of test time required to reduce the remaining errors to 15, if
the value of the testing compression factor is 5. Estimate reliability over a 100-h
operational period.
Substituting the given data into Equation (12.8), we get
1
4

(450 15) = (450) (4)

1
T2

(12.10)

Rearranging Equation (12.10) yields

T2 = 120 h
Inserting the above result and the other specified data into Equation (12.9), we get
( 450) ( 4)
t =
ln (120 4)
5

= 1224.4 h
Thus, for the given and calculated values from Equation (12.7), we get
R (100) = exp ( 100 120) = 0.4346
Thus, the required testing time is 1224.4 h and the reliability of the software for
the specified operational period is 0.4346.
Shooman Model
This is one of the earliest software reliability models and has influenced the development of several others over the time period [43, 44]. The model does not require

1999 by CRC Press LLC

fault collection during debugging on a continuous basis and it can be used for
software programs of all sizes. The key assumptions associated with this model are
as follows [8]:
Total machine instructions remain constant.
Debugging does not introduce new errors.
The hazard function is proportional to residual or the remaining software
errors.
The residual errors are obtained by subtracting the number of cumulative
rectified errors from the total errors initially present.
The total number of errors in the software program remains constant at the
start of integration testing and they decrease directly as errors are corrected.
The models hazard function is expressed by
(t ) = C Fr (x)

(12.11)

where
(t)
t
C
x
Fr (x)

is the hazard rate.

is the system operating time.
is the proportionality constant.
is the elapsed or debugging time since the start of system integration.
is the number of software faults remaining in the program at time x.

In turn, Fr (x) is defined by

Fr (x) =

Fz
Fc (x)
I

(12.12)

where
Fz
is the number of initial faults at time x = 0.
I
is the total number of machine language instructions.
Fc (x) is the cumulative number of faults corrected in interval x.
Inserting Equation (12.12) into Equation (12.11) yields
F

(t ) = C z Fc (x)
I

(12.13)

In reliability theory, the reliability of an item at time t is expressed by [8]

t

R (t ) = exp (t ) d t

1999 by CRC Press LLC

(12.14)

Thus, inserting Equation (12.13) into Equation (12.14), we get

F

R (t ) = exp C z Fc (x) t

I

(12.15)

By integrating Equation (12.15) over the interval [0, ], the following expression
for mean time to failure (MTTF) results:

F

MTTF = exp C z Fc (x) t d t
I


0

(12.16)

= 1 C z Fc (x)
I

In order to estimate the constants, C and Fz, we use the maximum likelihood
estimation approach to get [44-46]

=
C

i =1

Mi

i =1

F z

Fc (x i ) Wi
I

(12.17)

and

=
C

i =1

Mi

F z

Fc (x i )
I =1

(12.18)

where
N (N 2) is the number of tests following the debugging intervals (0, x1),
(0, x2), , (0, xN).
Wi
is the total time of successful and unsuccessful (i.e., all) runs in
the ith test.
is the total number of runs terminating in failure in the ith test.
Mi
Power Model
This model is also known as the Duane model because Duane [47] originally proposed it for hardware reliability in 1964. For software products, the same behavior
has been observed. The reason for calling it a power model is that the mean value
function, m(t), for the cumulative number of failures by time t is taken as a power
of t, i.e.,
m (t ) = t , for > 0, > 0

1999 by CRC Press LLC

(12.19)

For = 1, we get the homogeneous Poisson process model. The key assumption
associated with this model is that the cumulative number of failures by time t, N (t),
follows a Poisson process with value function described by Equation (12.19). In
order to implement the model, the data requirement could be either of the
following [48]:
Elapsed times between failures, i.e., y1, y2, y3, , yn, where yi = t i t i1
and t 0 = 0.
Actual times the software program failed, i.e., t1, t2, t3, , tn.
If T is the time for which the software program was under observation, then we can
write
m (T) T Expected number of faults by T
=
=
T
T
Total testing time, T

(12.20)

Taking the natural logarithms of Equation (12.20), we get

D = ln + ( 1) ln T

(12.21)

The above equation plots as a straight line and it is the form fitted to given data.
Differentiating Equation (12.19) with respect to t we get the following expression
for the failure intensity function:
d m (t )
= t 1 = (t )
dt

(12.22)

For > 1, Equation (12.22) is strictly increasing; thus, there can be no growth in
reliability [48].
Using the maximum likelihood estimation method, we get [49]

= n t n

(12.23)

and
n 1

= n

ln (t
i =1

ti )

(12.24)

The maximum likelihood estimation for the MTTF [i.e., for the (n + 1)st failure] is
given by [48, 49]

= t n n
MTTR

1999 by CRC Press LLC

(12.25)

Air Force Model

This model is the result of the U.S. Air Force Rome Laboratorys [50] effort to
predict software reliability in the earlier phases of the software life cycle. In this
model, the predictions of fault density are developed and subsequently they can be
transformed into other reliability measures such as failure rates [48]. The factors
related to fault density at the initial phases include development environment, nature
of application, traceability, program size, complexity, modularity, language type,
extent of reuse, quality review results, standards review results, and anomaly management. Thus, the initial fault density is expressed by [48, 50]
= (DA) ( NA) (T) ( PS) (C) (M) ( LT) ( ER ) (QR ) (SR ) (AM)

(12.26)

where

DA
NA
T
PS
C
M
LT
ER
QR
SR
AM

is
is
is
is
is
is
is
is
is
is
is
is

the initial fault density.

the development environment.
the nature of application.
traceability.
the program size.
complexity.
modularity.
the language type.
the extent of reuse.
the quality review results.
the standards review results.
the anomaly management.

The initial failure rate, I, is expressed by

I = ( LEF ) ( FER) ( NIF )
= ( LEF ) ( FER) ( LSC)

(12.27)

where
LSC is the number of lines of source code.
LEF is the programs linear execution frequency.
FER is the fault expose ratio (1.4 x 10-7 FER 10.6 10-7).
The linear execution frequency, LEF, of the program is expressed by
LEF = (MIR) (OIP)
where
MIR is the mean instruction rate.
OIP is the number of object instructions in the program.

1999 by CRC Press LLC

(12.28)

The object instructions in the program, OIP, are given by

OIP = (SI)(CER )

(12.29)

where
SI is the number of source instructions.
CER is the code expansion ratio. More specifically, this is the ratio of
machine instructions to source instructions and, normally, its average
value is taken as 4.
The number of inherent faults, IF, is expressed by
IF = () ( LSC)

(12.30)

Inserting Equations (12.28) through (12.30) into Equation (12.27) yields

I = (MIR) ( FER) (IF ) (SI) (CER )

(12.31)

12.8 FAULT MASKING

In fault tolerant computing the term fault masking is used in the sense that a system
having redundancy can tolerate a number of failures prior to its failure. The implication of the term masking is that a problem has surfaced somewhere within a digital
system, but because of the design, the problem does not effect the overall system
operation. Probably, the best known fault masking technique is modular redundancy.

12.8.1 TRIPLE MODULAR REDUNDANCY (TMR)

In this scheme, three identical redundant units or modules perform the same task
simultaneously and the voter compares the outputs of all the tree modules and sides
with the majority. The TMR system only experiences a failure when more than one
module/unit fails or the voter fails. In other words, this type of redundancy can
tolerate failure of a single module/unit. Figure 12.2 shows a block diagram of the
TMR scheme.
The TMR scheme was first proposed by Von Neumann [6] in 1956 and one
important example of its application was the SATURN V launch vehicle computer.
This computer used TMR with voters in the central processor and duplication in the
main memory [51]. The reliability of the TMR system with voter and the independent
units is

R TMRV = 3 R m2 2 R m3 R V

1999 by CRC Press LLC

(12.32)

FIGURE 12.2 TMR system with voter.

where
RTMRV is the reliability of the TMR system with voter.
RV
is the voter reliability.
Rm
is the module/unit reliability.
With perfect voter, i.e., RV = 1, Equation (12.32) reduces to
R TMR = 3 R m2 2 R m3

(12.33)

where
RTMR is the reliability of the TMR system with perfect voter.
The improvement in reliability of the TMR system over a single unit system is
determined by the single units reliability and the reliability of the voter. In the case
of perfect voter, the reliability of the TMR system given by Equation (12.33) is only
better than the single unit system when the reliability of the single unit or module
is greater than 0.5. At RV = 0.9, the reliability of the TMR system is only marginally
better than the single unit system reliability when the reliability of the single unit
is approximately between 0.667 and 0.833 [12]. Furthermore, when RV = 0.8, the
TMR system reliability is always less than a single units reliability.
For the single unit systems reliability, RS = Rm, and the perfect voter, i.e., RV = 1,
the ratio of RTMR to Rc is expressed by [3]
r=

R TMR 3 R m2 2 R m3
=
= 3 R m 2 R m2
RS
Rm

(12.34)

Differentiating Equation (12.34) with respect to Rm and equating it to zero, we get

dr
= 3 4 Rm = 0
dRs

1999 by CRC Press LLC

(12.35)

Thus, from Equation (12.35), we have

R m = 0.75
This means that the maximum value of the reliability improvement ratio, r, and
the maximum reliability of the TMR system occur at Rm = 0.75. Thus, inserting this
value for Rm into Equations (12.33) and (12.34) we get the following maximum values
of the TMR system reliability and the reliability improvement ratio, respectively:
R TMR = 3 (0.75) 2 (0.75)
2

0.84

and
r = 3 (0.75) 2 (0.75)

= 1.125
Example 12.3
The reliability of the TMR system with perfect voter is given by Equation (12.33).
Determine the point where both the single unit or simplex system reliability is equal
to the TMR system reliability. Assume that the simplex system reliability is given by
Rs = R m

(12.36)

In this case, we equate Equations (12.33) and (12.36) to get

3 R m2 2 R m3 = R m

(12.37)

2 R m2 3 R m + 1 = 0

(12.38)

Therefore,

The roots of the above quadratic equation are

Rm =

3 + [9 4 (2)(1)]
2 (2 )

12

=1

and
3 [9 4 (2)(1)]
Rm =
2 (2 )

1999 by CRC Press LLC

12

=1 2

Thus, at Rm = 1, the reliability of the simplex and TMR systems is the same.
This means that the TMR system reliability will only be better than the reliability
of the simplex system when Rm > 0.5.
Example 12.4
Assume that the reliability of the single unit in Example 12.3 is time dependent and
is expressed by
R m (t ) = e m t

(12.39)

where
t is time.
m is the unit/module constant failure rate.
Determine the point where both the single unit or simplex system reliability is equal
to the TMR system reliability.
Since from Example 12.3 we have Rm = 1, we can write
em t = 1

(12.40)

em t = 1 2

(12.41)

and

that is
m t = 0.6931
Equations (12.40) and (12.41) indicate that at t = 0 or m = 0 and at m t = 0.6931,
respectively, the reliability of the simplex and TMR systems is the same. The
reliability of the TMR system will only be better then the simplex system reliability
when the value of the t is less than 0.6931.
TMR System Time Dependent Reliability and Mean Time
to Failure (MTTF)
For constant failure rates of units, the TMR system with voter reliability using
Equation (12.32) is

R TMRV (t ) = 3 e 2 m t 2 e 3 m t e V t
=3e

1999 by CRC Press LLC

(2 m + V ) t

2 e

(3 m + V ) t

(12.42)

where
t
RTMRV
m
V

is
is
is
is

time.
the TMR system with voter reliability at time t.
the module/unit constant failure rate.
the voter constant failure rate.

Integrating Equation (12.42) over the interval from 0 to , we get the following
expression for the MTTF of the TMR system with voter:

MTTFTMRV =

[3 e

(2 m + v ) t

3 + t
2 e ( m v ) dt

(12.43)

(2 m + V ) (3 m + V )

For the perfect voter, i.e., V = 0, Equations (12.42) and (12.43) simplify to
R TMR (t ) = 3 e 2 m t 2 e 3 m t

(12.44)

and
MTTFTMR =

3
2
5

=
2 m 3m 6 m

(12.45)

where
RTMR (t) is the TMR system with perfect voter reliability at time t.
MTTFTMR is the MTTF of the TMR system with perfect voter.
Example 12.5
Assume that the failure rate of a unit/module belonging to an independent TMR
system with voter is m = 0.0001 failures/106 h. Calculate the TMR system MTTF,
if the voter failure rate is V = 0.00002 failures/106 h. Also, compute the system
reliability for a 100-h mission.
Inserting the specified data into Equation (12.43) yields
MTTFTMRV =

3
2

2 (0.0001) + 0.00002 3 (0.0001) + 0.00002

7, 386 h

1999 by CRC Press LLC

Similarly, substituting the given data into Equation (12.42) yields

2 0.0001) + 0.00002 ](100 )
3 0.0001) + 0.00002 ](100 )
R TMRV (100) = 3 e [ (
2 e [ (

0.9977

Thus, the reliability and MTTF of the TMR system with voter are 0.9977 and
7386 h, respectively.
Example 12.6
Repeat the Example 12.5 calculations for a TMR system with perfect voter. Comment
on the end results.
Thus, in this case we have V = 0.
Substituting the remaining given data into Equations (12.44) and (12.45), we get
R TMR (100) = 3 e ( 2 )( 0.0001)(100 ) 2 e (3)(. 0 0001)(100 )

0.9997

and
MTTFTMR =

5
6 (0.0001)
8, 333 h

In this case, the reliability and MTTF of the TMR system with perfect voter are
0.9997 and 8333 h, respectively. The perfect voter helped to improve the TMR
system reliability and MTTF, as can be observed by comparing Examples 12.5 and
12.6 results.
TMR System with Repair and Perfect Voter Reliability Analysis
In the preceding TMR system reliability analysis, all units were considered nonrepairable. This is quite true in certain applications like space exploration but in
others it is possible to repair the failed units. Thus, in this case we consider an
independent and identical unit repairable TMR system with perfect voter. As soon
as any one of the TMR system units fails, it is immediately repaired. When more
than one unit fails, the TMR system is not repaired. The TMR system state space
diagram is shown in Figure 12.3.
Markov technique is used to develop state probability equations [8] for the state
space diagram shown in Figure 12.3. The following assumptions are associated with
this TMR system model:

1999 by CRC Press LLC

FIGURE 12.3 TMR system with repair state space diagram.

The system is made up of three independent and identical units with a

perfect voter.
All system units are active and the system fails when more than one unit
fails.
Failure and repair rates of a unit are constant.
The system is only repaired when one unit fails and a repaired unit is as
good as new.
The numerals in the box, diamond, and circle of Figure 12.3 identify
specific system states.
The following symbols were used to develop equations for this model:
j

is the jth state of the TMR system shown in Figure 12.3; i = 0 (three
units up), i = 1 (one unit failed system up), i = 2 (two units failed
system failed).
Pj (t) is the probability that the TMR system is in state j at time t; for j = 0,
1, and 2.
m is the unit/module failure rate.

is the unit repair rate.

The following differential equations are associated with Figure 12.3:
P (t ) = P1 (t ) 3 m P0 (t )

(12.46)

P1 (t ) = 3 m P0 (t ) (2 + ) P1 (t )

(12.47)

P2 (t ) = 2 m P1 (t )

(12.48)

where the prime denotes differentiation with respect to time t. At time t = 0, P0 (0) = 1
and P1 (0) = P2 (0) = 0.
By solving Equations (12.46) through (12.48), we get the following expression for
the TMR system reliability, with repair and perfect voter:

1999 by CRC Press LLC

R TMRr (t ) = P0 (t ) + P1 (t )
=

) ]

1
(5 m + ) ex1 t ex2 t + x1 ex1 t x2 ex2 t
x1 x 2

(12.49)

where

x1, x 2 = (5 m + ) 2 + 2 + 10

12

(12.50)

Integrating Equation (12.49) over the interval [0, ] yields

MTTFTMRr =

TMRr

(t ) d t

(12.51)

5
=
+
6 m 6 2m
where
MTTFTMRr is the TMR system, with repair and perfect voter, mean time to
failure.
For no repair facility, i.e., = 0, Equation (12.51) becomes the same as Equation
(12.45).
Example 12.7
A unit/module of an independent TMR system, with repair and perfect voter, has a
failure rate of 0.005 failures/h. The unit repair rate is 0.6 repairs/h. Calculate the
following:
System with repair mean time to failure.
System mean time to failure, if there is no repair.
Comment on the end results.
Substituting the given data into Equation (12.51) yields
MTTFTMRr =

1999 by CRC Press LLC

5
0.6
+
2
6 (0.005) 6 (0.006)
4167 h

Similarly, for no repair, inserting the given data into Equation (12.45) yields
MTTFTMR =

5
6 (0.005)

167

It means the introduction of repair helped to increase the TMR system MTTF
from 167 h to 4167 h.

12.8.2 N-MODULAR REDUNDANCY (NMR)

This is a logical extension of TMR and it contains N identical modules/units. The
number N is any odd number and is defined by N = 2 m + 1. The NMR system will
be successful if at least (m+1) units operate normally. As the voter acts in series
with the N-module system, the total system fails if the voter fails. For independent
units, the reliability of the NMR system is

R NMRV

m N

i
Ni
= RV
R m (1 R m )
i=0 i

N
N ! ( N i )! i !
i

(12.52)

(12.53)

where
RNMRV is the reliability of the NMR system with voter.
RV
is the voter reliability.
Rm
is the module reliability.
The time dependent and other reliability analysis of this system can be performed
in a manner similar to the TMR system analysis. Additional redundancy schemes
may be found in Reference 8.

12.9 PROBLEMS
1. Make a comparison of hardware reliability and software reliability.
2. What are the major sources of computer failures? Describe them in detail.
3. Define the following terms:
Software error
Debugging
Software reliability
Fault-tolerant computing

1999 by CRC Press LLC

4. Discuss software life cycle phases and associated error sources.

5. Describe the following software design methods:
Structure programming
Top-down programming
Jackson structured programming
6. Describe the following techniques associated with software reliability
improvement:
N-version programming
Top-down testing
Bottom-up testing
7. What are the major classifications of software reliability models? Describe
them briefly.
8. A software under development was seeded with 30 faults and during the
testing process 70 faults of the same kind were discovered. The breakdowns of the faults uncovered were 25 seeded faults and 45 unseeded
faults. Estimate the number of unseeded faults remaining in the software
under consideration.
9. The failure rate of a unit/module belonging to an independent TMR system
with voter is 0.0001 failures/h and the voter failure rate is 0.00003 failures/h. Calculate the TMR system MTTF and its reliability for a 200-h
mission.
10. Develop state probability expressions for the TMR system state space
diagram shown in Figure 12.3. Also, prove that the system reliability is
given by Equation (12.49).

12.10 REFERENCES
1. Keene, S.J., Software reliability concepts, Annu. Reliability and Maintainability
Symp. Tutorial Notes, 1-21, 1992.
2. Pradhan, D.K., Ed., Fault-Tolerant Computing Theory and Techniques, Vols. 1 and
2, Prentice-Hall, Englewood Cliffs, NJ, 1986.
3. Shooman, M.L., Fault-tolerant computing, Annu. Reliability and Maintainability
Symp. Tutorial Notes, 1-25, 1994.
4. Shannon, C.E., A mathematical theory of communications, Bell System Tech. J., 27,
379-423 and 623-656, 1948.
5. Hamming, W.R., Error detecting and error correcting codes, Bell System Tech. J., 29,
147-160, 1950.
6. Von Neumann, J., Probabilistic logics and the synthesis of reliable organisms from
reliable components, in Automata Studies, Shannon, C.E. and McCarthy, J., Eds.,
Princeton University Press, Princeton, NJ, 1956, pp. 43-98.
7. Moore, E.F. and Shannon, C.E., Reliable circuits using less reliable relays, J. Franklin
Inst., 262, 191-208, 1956.
8. Dhillon, B.S., Reliability in Computer System Design, Ablex Publishing, Norwood,
NJ, 1987.
9. Haugk, G., Tsiang, S.H., and Zimmerman, L., System testing of the no. 1 electronic
switching system, Bell System Tech. J., 43, 2575-2592, 1964.
10. Floyd, R.W., Assigning meanings to program, Math. Aspects Comp. Sci., XIX, 19-32,
1967.

1999 by CRC Press LLC

11. Hudson, G.R., Programming Errors as a Birth-and-Death Process, Report No. SP3011, System Development Corporation, 1967.
12. Pecht, M., Ed., Product Reliability, Maintainability, and Supportability Handbook,
CRC Press, Boca Raton, FL, 1995.
13. Lipow, M., Prediction of software failures, J. Syst. Software, 1, 71-75, 1979.
14. Anderson, R.T., Reliability Design Handbook, Rome Air Development Center, Griffiss Air Force Base, Rome, NY, 1976.
15. Glass, R.L., Software Reliability Guidebook, Prentice-Hall, Englewood Cliffs, NJ,
1979.
16. Avizienis, A., Fault-tolerant computing: An overview, Computer, January/February,
5-8, 1971.
17. Kline, M.B., Software and hardware reliability and maintainability: What are the
differences?, Proc. Annu. Reliability and Maintainability Symp., 179-185, 1980.
18. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Handbook of Reliability Engineering
and Management, McGraw-Hill, New York, 1996.
19. Dhillon, B.S., Reliability Engineering in Systems Design and Operation, Van Nostrand Reinhold Company, New York, 1983.
20. Yourdon, E., The causes of system failures Part II, Modern Data, 5, 50-56, 1972.
21. Yourdon, E., The causes of system failures Part III, Modern Data, 5, 36-40, 1972.
22. Bell, D., Morrey, L., and Pugh, J., Software Engineering: A Programming Approach,
Prentice-Hall, London, 1992.
23. Wang, R.S., Program with measurable structure, Proc. Am. Soc. Quality Control
Conf., 389-396, 1980.
24. Koestler, A., The Ghost in the Machine, Macmillan, New York, 1967.
25. Neufelder, A.M., Ensuring Software Reliability, Marcel Dekker, New York, 1993.
26. Scott, R.K., Gault, J.W., and McAllister, D.G., Fault tolerant software reliability
modeling, IEEE Trans. Software Eng., 13(5), 1987.
27. Neuhold, E.J. and Paul, M., Formal description of programming concepts, Int. Fed.
Info. Process. (IFIP) Conf. Proc., 310-315, 1991.
28. Galton, A., Logic as a formal method, Comp. J., 35(5), 213-218, 1992.
29. Beizer, B., Software System Testing and Quality Assurance, Van Nostrand Reinhold
Company, New York, 1984.
30. Myers, G.J., The Art of Software Testing, John Wiley & Sons, New York, 1979.
31. Myers, G.J., Software Reliability: Principles and Practices, John Wiley & Sons, New
York, 1976.
32. Kopetz, H., Software Reliability, Macmillan, London, 1979.
33. Musa, J.D., Iannino, A., and Okumoto, K., Software Reliability, McGraw-Hill, New
York, 1987.
34. Sukert, A.N., An investigation of software reliability models, Proc. Annu. Reliability
Maintainability Symp., 478-484, 1977.
35. Schick, G.J. and Wolverton, R.W., An analysis of competing software reliability
models, IEEE Trans. Software Eng., 4, 104-120, 1978.
36. Hudson, G.R., Program Errors as a Birth and Death Process, Report No. SP-3011,
System Development Corporation, December 4, 1967.
37. Mills, H.D., On the Statistical Validation of Computer Programs, Report No. 72-6015,
1972. IBM Federal Systems Division, Gaithersburg, MD.
38. Musa, J.D., A theory of software reliability and its applications, IEEE Trans. Software
Eng., 1, 312-327, 1975.
39. Shooman, M.L., Software reliability measurement and models, Proc. Annu. Reliability Maintainability Symp., 485-491, 1975.

1999 by CRC Press LLC

40. Jelinski, Z. and Moranda, P.B., Software reliability research, in Proceedings of the
Statistical Methods for the Evaluation of Computer System Performance, Academic
Press, 1972, pp. 465-484.
41. Schick, G.J. and Wolverton, R.W., Assessment of software reliability, Proceedings of
the Operations Research Physica-Verlag, Wurzburg-Wien, 1973, pp. 395-422.
42. Nelson, E., Estimating software reliability from test data, Microelectronics and Reliability, 17, 67-75, 1978.
43. Ramamoorthy, C.V. and Bastani, F.B., Software reliability: status and perspectives,
IEEE Trans. Software Eng., 8, 354-371, 1982.
44. Dunn, R. and Ullman, R., Quality Assurance for Computer Software, McGraw-Hill,
New York, 1982.
45. Craig, G.R., Software Reliability Study, Report No. RADC-TR-74-250, Rome Air
Development Center, Griffiss Air Force Base, Rome, NY, 1974.
46. Thayer, T.A., Lipow, M., and Nelson, E.C., Software Reliability, North-Holland
Publishing Company, New York, 1978.
47. Duane, J.T., Learning curve approach to reliability monitoring, IEEE Trans. Aerospace, 2, 563-566, 1964.
48. Lyu, M.R., Ed., Handbook of Software Reliability Engineering, McGraw-Hill, New
York, 1996.
49. Crow, L.H., in Reliability Analysis for Complex Repairable Systems, Reliability and
Biometry, Proschan, F. and Serfling, R.J., Eds., Society for Industrial and Applied
Mathematics (SIAM), Philadelphia, PA, 1974, pp. 379-410.
50. Methodology for Software Reliability Prediction and Assessment, Report No. RLTR-92-52, Volumes I and II, Rome Air Development Center, Griffiss Air Force Base,
Rome, NY, 1992.
51. Mathur, F.P. and Avizienis, A., Reliability analysis and architecture of a hybrid
redundant digital system: Generalized triple modular redundancy with self-repair,
1970 Spring Joint Computer Conference, AFIPS Conf. Proc., 36, 375-387, 1970.

1999 by CRC Press LLC

13

Robot Reliability
and Safety

13.1 INTRODUCTION
Robots are increasingly being used in the industry to perform various types of tasks:
material handling, arc welding, spot welding, etc.
The history of robots/automation may be traced back to the ancient times when
the Egyptians built water-powered clocks and the Chinese and Greeks built waterand steam-powered toys. Nonetheless, the idea of the functional robot originated in
Greece in the writings of Aristotle (4th century B.C.), the teacher of Alexander The
Great, in which he wrote: If every instrument could accomplish its own work,
obeying or anticipating the will of others [1].
In 1920, Karl Capek (18901938), a Czechoslovak science-fiction writer first
coined the word robot and used it in his play entitled Rossums Universal Robots
which opened in London in 1921. In the Czechoslovakian language, robot means
worker. In 1939, Isaac Asimov wrote a series of stories about robots and in 1942,
he developed the following three laws for robots [2]:
A robot must not harm a human, nor, through inaction, permit a human
to come to harm.
Robots must obey humans unless their orders conflict with the preceding
law.
A robot must safeguard its own existence unless it conflicts with the
preceding two laws.
In 1954, George Devol [3] developed a programmable device that could be
considered the first industrial robot. Nonetheless, in 1959, the first commercial robot
was manufactured by the Planet Corporation and Japan, today the world leader in
the use of robots, imported its first robot in 1967 [3]. In 1970, the first conference
on industrial robots was held in Chicago, Illinois, and five years later (i.e., in 1975),
the Robot Institute of America (RIA) was founded. Today, there are many journals
and a large number of publications on robots.
In 1983, the estimated world robot population was around 30,000 [4] and by the
end of 1998, it is forecasted to be around 820,000 [5].
A robot may be defined as a reprogrammable multi-functional manipulator
designed to move material, parts, tools, or specialized items through variable programmed motions for the performance of various tasks [6]. As a robot has to be safe
and reliable, this chapter discusses both of these topics.

1999 by CRC Press LLC

13.2 ROBOT SAFETY

As in the case of general system safety, robot safety is an important factor in the
effectiveness of robot systems. It is expected that the rapidly increasing robot
population will further emphasize the importance of robot safety in a significant
manner.
The review of published literature indicates that in the 1970s, robot safety
problems became quite apparent to the professionals working in the area. Consequently, 1980 witnessed a flood of publications on the topic. In fact, some of the
important publications of the 1980s and the early 1990s were as follows:
In 1985, the Japanese Industrial Safety and Health Association (JISHA)
developed a document entitled An Interpretation of the Technical Guidance on Safety Standards, etc. of Industrial Robots [7].
In 1985, a book entitled Robot Safety, edited by Benny and Yong [8],
appeared and it contained many useful articles on the topic.
In 1986, a safety standard entitled Industrial Robots and Robot Systems:
Safety Requirements was published jointly by the American National
Standards Institute and the Robotics Industries Association (RIA) [9].
In 1987, a journal article [10] entitled On Robot Reliability and Safety:
Bibliography, appeared and it listed most of the publications on the
subject.
In 1991, a book [2] entitled Robot Reliability and Safety was published
and it presented a comprehensive list of publications on robot safety in
addition to the important aspects of robot safety.
Since the early 1990s, there have been many more publications on the topic.

13.2.1 ROBOT ACCIDENTS

Over the years, there have been many accidents involving robots [2, 1114]. These
accidents occurred in many robot user countries: Japan, U.S., Sweden, France, U.K.,
etc. [11]. According to published sources, there have been at least five fatal accidents;
four in Japan and one in the U.S.. Some examples of these fatal accidents are as
follows [2]:
A worker turned on a welding robot during the presence of a person in
its work zone. Consequently, the robot pushed that person into the positioning fixture and subsequently the person died.
During a temporary stoppage cycle of a robot, a repair person climbed
over a safety fence without turning off the power to the robot. As the
robot commenced its operation, it pushed that person into a grinding
machine and, consequently, the person died.
During the momentary stoppage of a line servicing robot, a worker
climbed onto a conveyor belt in motion to recover a faulty part. As the
robot started its operation, the person was crushed to death.

1999 by CRC Press LLC

TABLE 13.1
National Institute for Occupational Safety and Health Recommendations
for Minimizing Robot Related Incidents
Recommendation
No.
1
2
3
4
5
6

Recommendation description
Allow sufficient clearance distance around all moving parts of the robot system.
Provide adequate illumination in control and operational areas of robot system.
Provide physical barrier incorporating gates with electrical interlocks, so that when
any one gate opens, the robot operation is terminated.
Ensure that floors or working surfaces are marked so that the robot work area is
clearly visible.
Provide back-up to devices such as light curtains, motion sensors, electrical
interlocks, or floor sensors.
Provide remote diagnostic instrumentation so that system maximum
troubleshooting can be conducted from outside the robot work area.

Subsequent investigation revealed that in all of the above cases, the humans put
themselves in harms way. The robot alone did not kill the human.
Some of the causal factors for robot accidents include robot design, workplace
design interfacing, and workplace design guarding. In order to minimize the risk of
robot incidents, the National Institute for Occupational Safety and Health proposed
various recommendations [15]. In fact, all of these recommendations were directed
at three strategic areas: (1) worker training, (2) worker supervision, and (3) robot
system design. The robot system design related recommendations are presented in
Table 13.1

13.2.2 ROBOT HAZARDS

AND

SAFETY PROBLEMS

There are various hazards associated with robots. The three basic types of hazards
are shown in Figure. 13.1 [16, 17]. These are impact hazards, the trapping point

FIGURE 13.1 Basic robot hazards.

1999 by CRC Press LLC

hazards, and those that develop from the application itself. The impact hazards are
concerned with an individual(s) being struck by a robots moving parts, by items
being carried by the robot, or by flying objects ejected or dropped by the robot. The
trapping point hazards usually occur because of robot movements with respect to
fixed objects such as machines and posts in the same space. Furthermore, other
possible factors include the movements of auxiliary equipment such as work carriages and pallents. The hazards that are generated by the application itself include
burns, electric shocks, exposure to toxic substances, and arc flash.
There could be many different causes of the above three basic types of robot
hazards but the most prevalent causes include unauthorized access, human error, and
mechanical related problems from the robot system itself or from the application.

13.2.3 SAFETY CONSIDERATIONS

IN

ROBOT LIFE CYCLE

A robot life cycle could be divided into four distinct phases: (1) design, (2) installation, (3) programming, and (4) operation and maintenance. The overall robot safety
problems can only be minimized if careful consideration is given to safety throughout
the robot life cycle. Some safety measures that could be taken during each of the
four robot life cycle phases are discussed below [2, 7, 1821]:
Design Phase
The safety measures associated with this phase may be categorized into three distinct
groups: electrical, software, and mechanical. The electrical group safety measures
include eliminating the risk of an electric shock, minimizing the effects of electromagnetic and radio-frequency interferences, designing wiring circuitry capable of
stopping the robots movement and locking its brakes, and having built-in hose and
cable routes using adequate insulation, sectionalization, and panel covers. Some of
the safety measures belonging to the software group include having built-in safety
commands, periodically examining the built-in self-checking software for safety,
having a stand-by power source for robots functioning with programs in random
access memory, using a procedure of checks for determining why a failure occurred,
and prohibiting a restart by merely resetting a single switch.
The mechanical group related safety measures include providing dynamic brakes
for a software crash or power failure, providing drive mechanism covers, providing
several emergency stop buttons, providing mechanisms for releasing the stopped
energy, and putting guards on items such as gears, belts, and pulleys.
Installation Phase
There are a large number of safety measures that can be taken during the robot
installation phase. Some of these measures include installing interlocks, sensing
devices, etc.; distancing circuit boards from electromagnetic fields; installing electrical
cables according to electrical codes; identifying the danger areas with the aid of codes;
placing robot controls outside the hazard zone; providing an appropriate level of
illumination to humans concerned with the robot; and labeling stored energy sources.

1999 by CRC Press LLC

Programming Phase
During this phase, programmers/setters are the people particularly at risk. For example, according to a study of 131 cases [21], such people were at the highest risk
57% of the time. Some of the safety measures that can be taken during the programming phase include designing the programmer work area so that unnecessary stress
with respect to stand accessibility, visibility/lighting, and forced postures, is eliminated; installing hold-to-run buttons; turning off safety-related devices with a key
only; planning the programmers location outside the movement only; and marking
the programming position.
Operation and Maintenance Phase
There are a large number of safety related measures that belong to this phase. Some
of these are developing appropriate safety operations and maintenance procedures,
ensuring functionality of all emergency stops, providing appropriate protective gear
to all concerned individuals, minimizing the potential energy of an unexpected
motion by having the robot arm extended to its maximum reach position, blocking
out all concerned power sources during maintenance, posting the robots operating
weight capacity, observing all government codes and other regulations concerned
with robot operation and maintenance, and ensuring that only authorized and trained
personnel operate and maintain robots.

13.2.4 ROBOT SAFEGUARD APPROACHES

In order to improve robot safety, over the years there have been many safeguard
approaches developed [22]. These include flashing lights, warning signs, intelligent
systems, electronic devices, infrared light arrays, and physical barriers. Each of these
approaches is described below.
Flashing Lights
This approach calls for the installation of flashing lights at the perimeter of the robot
working zone or on the robot itself. The primary aim of these lights is to alert
concerned individuals that the robot is active or could be in motion at any moment.
In the event of pursuing this approach, it is useful to ensure that the flashing lights
are energized at all times during the active drive power period.
Warning Signs
This approach is concerned with the use of warning signs. Use of these signs may
be considered where robots, because of their speed, size, and inability to impart
significant amount of force, cannot injure people. Usually, the warning signs are
used for robots used in laboratories and for small-part assembly robots. All in all
warning signs are useful for all robot applications, irrespective of the robots ability
to injure people or not.

1999 by CRC Press LLC

Intelligent Systems
This approach makes use of intelligent control systems that make decisions through
remote sensing, hardware, and software. In order to have an effective intelligent
collision-avoidance system, the robot operating environment has to be restricted.
Also, special sensors and software should be used. Nonetheless, usually in most
industrial settings it is not possible to restrict environment.
Electronic Devices
This approach makes use of ultrasonic for parameter control to seek protection from
intrusions. The parameter control electronic barriers use active sensors for intrusion
detection. Usually, the use of ultrasonic is considered in those circumstances where
unobstructed floor space is a critical issue. An evaluation of robot parameter control
devices is presented in Reference 23.
Infrared Light Arrays
This approach makes use of linear arrays of infrared sources known as light curtains.
The light curtains are generally reliable and provide an excellent protection to
individuals from potential dangers in the robots operating area. False triggering is
the common problem experienced with the use of light curtains. It may occur due
to various factors including smoke, flashing lights, or heavy dust in situations where
the system components are incorrectly aligned.
Physical Barriers
Even though the use of physical barriers is an important approach to safeguard humans,
it is not the absolute solution to a safety problem. The typical examples of physical
barriers include safety rails, plastic safety chains, chain-link fences, and tagged-rope
barriers. Some of the useful guidelines associated with physical barriers include:
Consider using safety rails in areas free of projectiles.
Consider using fences in places where long-range projectiles are considered a hazard.
Safety rails and chain-link fences are quite effective in places where
intrusion is a serious problem.
When considering the use of a peripheral physical barrier, it is useful to
ask questions such as the following:
What is being protected?
Is it possible that it can be bypassed?
How effective is the barrier under consideration?
How reliable is the barrier under consideration?

13.2.5 HUMAN FACTOR ISSUES

IN

ROBOTIC SAFETY

One of the single most important components in robotic safety is the issue of human
factors. Just like in the case of other automated machines, the problem of human
factors in robotics is very apparent and it has to be carefully considered. This section

1999 by CRC Press LLC

FIGURE 13.2 Areas of human factors discussion for robotic safety.

presents human factor aspects of robotic safety in five distinct areas as shown in
Figure 13.2 [24]. These are human-robot interface design, documentation preparation, methods analysis, future considerations, and miscellaneous considerations.
The human-robot interface design is one important area which requires careful
consideration of human factors. Here, the primary objective is to develop humanrobot interface design so that the probability of the human error occurrence is at
minimum. There are various steps that help to prevent the occurrence of human
error, including analyzing human actions during robotic processes, designing hardware and software with the intention of reducing human errors, paying careful
attention to layouts, and considering factors such as weight, layout of buttons,
readability of the buttons functional descriptions, connectors flexibility, and handheld devices shape and size.
The document preparation is concerned with the quality of documentation for
use by robot users. These documents must be developed by considering the qualifications and experience of the target group. Some examples of target groups are
operators, maintainers, programmers, and designers. Nonetheless, during the preparation of such documents factors, such as easily understandable information, inclusion of practical exercises, completeness of information, and inclusion of pictorial
descriptions at appropriate places must be carefully considered.
The classical industrial engineering approach to methods analysis is found to
be quite effective in improving robotic safety with respect to human factors. Flow
process charts and multiple-activity process charts are two important examples.
Some of the future considerations to improve robotic safety with respect to
human factors include the application of artificial intelligence to the worker-robot
interface, and considering human factors in factory design with respect to possible
robot applications.

1999 by CRC Press LLC

The miscellaneous considerations include the analysis of ambient conditions and

layout, the preproduction testing of robotic systems, and examination of factors such
as noise, lighting, and temperature.

13.3 ROBOT RELIABILITY

A robot not only has to be safe, but it also has to be reliable. An unreliable robot
may become the cause of unsafe conditions, high maintenance costs, inconvenience,
etc. As robots make use of electrical, mechanical, electronics, pneumatic, and
hydraulic parts, their reliability and maintainability problems are very challenging
because of many different sources of failures.
Even though there is no clear cut definitive point in the beginning of robot
reliability and maintainability field, Engelberger [25] could be regarded as one of
the first persons to think in the terms of robot mean time between failures (MTBF)
and mean time to repair (MTTR). In 1987, an article entitled On Robot Reliability
and Safety: Bibliography presented most of the published literature on robot reliability and maintainability [26]. In 1991, a book entitled Robot Reliability and Safety
covered robot reliability and maintainability in a significant depth along with presenting a comprehensive list of publications on these two areas [27].

13.3.1 CAUSES

AND

CLASSIFICATIONS

OF

ROBOT FAILURES

Over the years, there have been various studies performed to determine the causes
of robot failures. As a result of such studies, some of the frequently occurring failure
causes highlighted were oil pressure valve problems, servo valve malfunctions, noise,
printed circuit board troubles, and human errors. In particular, References 28 and
29 pointed out that the robot problems occurred in the following order: control
system problems, jig and other tool incompatibility, robot body problems, programming and operation errors, welding gun problems and problems of other tooling
parts, deterioration and precision deficiency, runaway, and miscellaneous.
A robots reliability and its safe operation are primarily affected by four types
of failures [30, 31]: (1) random component failures, (2) systematic hardware faults,
(3) software faults, and (4) human errors.
The failures that occur unpredictably during the useful life of a robot are known
as random failures. The reasons for the occurrence of such failures include undetectable defects, unavoidable failures, low safety factors, and unexplainable causes.
The systematic hardware faults occur because of the existence of unrevealed mechanisms in the robot design. One example of the protection against robot hardware
faults is to use sensors in the system to detect the loss of pneumatic pressure/line
voltage/hydraulic pressure. Reference 32 presents various useful techniques to
reduce the occurrence of systematic hardware faults/failures.
The occurrence of software faults is an important factor in the malfunctioning
of robots and they may occur due to reasons such as embedded software or the
controlling software and application software. Some studies have revealed that over
60% of software errors are made during the requirement and design phases as
opposed to less then 40% during the coding phase. The measures such as the

1999 by CRC Press LLC

performance of failure mode and effects analysis (FMEA), fault tree analysis, and
testing help to reduce robot software faults.
During the design, manufacture, test, operation, and maintenance of a robot,
various types of human errors may occur. Some studies reveal that the human error
represents a significant proportion of total equipment failures. There could be many
causes for the occurrence of human error including poor equipment design, poorly
written operating and maintenance procedures, poorly trained operation and maintenance manpower, task complexity, inadequate lighting in the work zone, and
improper tools used by maintenance personnel.

13.3.2 ROBOT RELIABILITY MEASURES

There are various types of reliability-related measures associated with a robot: mean
time to robot failure, mean time to robot-related problems, and robot reliability.
Mean Time to Robot Failure (MTRF)
The MTRF can be obtained by using either of the following two formulas:

MTRF =

R (t ) d t
b

(13.1)

MTRF = Trp Tdrf N

(13.2)

where
Rb (t)
Trp
Tdrf
N

is
is
is
is

the
the
the
the

robot reliability at time t.

robot production time expressed in hours.
downtime due to robot failures, expressed in hours.
total number of robot failures.

Relationship (13.2) can be used in real life environments to determine mean

productive robot time before robot failure. On the other hand, relationship (13.1)
can be used to calculate mean time to failure of a single or of redundant robots with
defined reliability function.
Example 13.1
Assume that at a robot facility the total robot production hours were 20,000 h with
the downtime due to robot failure of 500 h. There have been a total of 15 robot
failures. Calculate the robot mean time to failure.
Substituting the above data into Equation (13.2) yields
MTRF = (20, 000 500) 15
= 1300 h
Thus, the robot mean time to failure is 1300 h.

1999 by CRC Press LLC

Example 13.2
Assume that the failure rate of a robot is 0.0004 failure/h and its reliability is
expressed by the following equation:
R b (t ) = e b t

(13.3)

where
Rb (t) is the robot reliability at time t.
b
is the robot failure rate.
Substituting Equation (13.3) into Equation (13.1) yields

MTRF = e b t d t
0

(13.4)

1
=
b
Inserting the robot failure rate value into Equation (13.4), we get
MTRF =

1
0.0004

= 2500 h
Thus, the robot mean time to failure is 2500 h.
Mean Time to Robot-Related Problems
The mean time to robot-related problems (MTRP) simply is the average productive
robot time prior to the occurrence of a robot-related problem and is expressed by

MTRP = Trp Tdrp K

(13.5)

where
Tdrp is the downtime due to robot-related problems, expressed in hours.
K is the total number of robot-related problems.
Example 13.3
Assume that at an industrial robot installation the total robot production hours were
25,000 h. Furthermore, downtime due to robot-related problems was 800 h and there
have been a total of 30 robot-related problems. Calculate the MTRP.

1999 by CRC Press LLC

Substituting the specified data into Equation (13.5), we get

MTRP = (25, 000 800) 30
= 806.67 h
Thus, the mean time to robot-related problems is 806.67 h.
Robot Reliability
This may be expressed as the probability that a robot will perform its designated
mission adequately for the specified time period when used under the designed
conditions. Mathematically, the robot reliability may be expressed as follows:
t

R b (t ) = exp b (t ) d t

(13.6)

where
b (t) is the robot time-dependent failure rate or hazard rate.
It is to be noted that Equation (13.6) can be used to obtain robot reliability for any
given robot time to failure distribution (e.g., Weibull, gamma, or exponential).
Example 13.4
A robot time to failure is exponentially distributed; thus, its failure rate, b, is 0.0004
failure/h. Calculate the robot reliability for a 15-h operating mission.
Thus, in this case, we have
b (t ) =

(13.7)

Inserting Equation (13.7) into Equation (13.6) yields

t

R b (t ) = exp b d t

(13.8)

= eb t
Substituting the specified data into Equation (13.8) we get
R b (15) = e ( 0.0004 ) (15)
= 0.994
Thus, there is a 99.4% chance that the robot will operate successfully during
the specified mission.

1999 by CRC Press LLC

13.3.3 ROBOT RELIABILITY ANALYSIS

AND

PREDICTION METHODS

In the field of reliability engineering, there are many techniques and methods available that can be used effectively to perform robot reliability and availability analysis
and prediction studies. Some of these methods are network reduction technique,
fault tree analysis (FTA), failure modes and effect analysis (FMEA), and Markov
method [32].
In particular, the two widely used practical methods, i.e., part stress analysis
prediction and part count reliability prediction, can also be used to predict robot
system reliability during the design phase. Both of these methods are described in
Reference 33.
The Markov method is also a widely used approach in reliability engineering
and it can equally be applied to perform various kinds of robot reliability studies.
Two of its applications are presented in the next section.

13.3.4 RELIABILITY AND AVAILABILITY ANALYSES

SYSTEM FAILING WITH HUMAN ERROR

OF A

ROBOT

This section presents reliability and availability analyses of a mathematical model

representing a robot system by using the Markov method [27]. The robot system
can fail either due to a human error or other failure. The failed robot system is
repaired to its original operating state. The system state space diagram is shown in
Figure 13.3. This figure shows three states of the robot system: state 0 (the robot
system operating normally), state 1 (the robot system failed due to a human error),
and state 2 (the robot system failed due to failures other than human error).
The following assumptions are associated with this model:
The robot system can fail due to a human error.
The robot system failures are divisible into two categories: failures resulting from human error, and failures resulting from other than human error.
The robot system human error and other failures are exponentially distributed or their occurrence rates are constant.
The failed robot system repair times are exponentially distributed or its
repair rates are constant.
The repaired robot system is as good as new.

FIGURE 13.3 Robot system state space diagram.

1999 by CRC Press LLC

The following symbols were used to develop Equations for the model:
i

is the state of the system: i = 0 means the robot system is operating

normally; i = 1 means the robot system malfunctioned because of a
human error; i = 2 means the robot system malfunctioned because of
a failure other than human error.
t
is time.
Pi (t) is the probability that the robot system is in state i at time t, for i = 0,
1, 2.
h is the constant robot system human error rate.

is the constant robot system non-human error failure rate.

h is the constant robot system repair rate from failed state 1.

is the constant robot system repair rate from failed state 2.

Using the Markov technique, from Figure 13.3, we write the following set of differential equations:
P0 (t ) + ( + h ) P0 (t ) = h P1 (t ) + P2 (t )

(13.9)

P1 (t ) + h P1 (t ) = h P0 (t )

(13.10)

P2 ( t ) + P2 ( t ) = P0 (t )

(13.11)

At time t = 0, P0 (0) = 1 and P1 (0) = P2 (0) = 0.

After solving Equations (13.9) through (13.11), we get the following state
probability expressions:
P0 (t ) =

h ( k1 + ) ( k1 + h ) k1 t ( k 2 + )2( k 2 + h ) k
+
e
e
k1 k 2 k1 ( k1 k 2 )
k 2 ( k1 k 2 )

where

k1 , k 2 =

b b 2 4 ( h + h + h )
2

b = + h + + h
k1 k 2 = h + h + h
k1 + k 2 = ( + h + + h )

1999 by CRC Press LLC

(13.12)

P1 (t ) =

h h k1 + h k1 t ( + k 2 )2 h k
+
e
e
k1 k 2 k1 ( k1 k 2 )
k 2 ( k1 k 2 )

P2 (t ) =

h k1 + h k1 t ( h + k 22) k
+
e
e
k1 k 2 k1 ( k1 k 2 )
k 2 ( k1 k 2 )

(13.13)

(13.14)

The robot system availability, AR (t), is expressed by

AR (t ) = P0 (t ) =

h ( k1 + ) ( k1 + h ) k1 t
+
e
k1 k 2 k1 ( k1 k 2 )
(k + ) (k 2 + h ) k t
2
e 2
k 2 ( k1 k 2 )

(13.15)

As time t becomes very large, the robot system steady-state availability and other
state probabilities become
AR = lim A R (t ) =
t

h
k1 k 2

(13.16)

P1 = lim P1 (t ) =

h
k1 k 2

(13.17)

P2 = lim P2 (t ) =

h
k1 k 2

(13.18)

where
AR is the robot system steady state availability.
P1 is the robot system steady state failure probability due to human error.
P2 is the robot system steady state failure probability due to failures other
than human error.
In order to perform robot system reliability analysis, we set = h = 0, in Equations
(13.9) through (13.11) and then solve the resulting equations to get
+ t
P0 (t ) = e ( h )

1999 by CRC Press LLC

(13.19)

(13.20)

(13.21)

P1 (t ) =

h
+ t
1 e ( h )

( h)

P2 (t ) =

+ t
1 e ( h )
( + h )

The robot system reliability at time t from Equation (13.19) is given by

+ t
R r (t ) = P0 (t ) = e ( h )

(13.22)

Similarly, the probability of the robot system failure, Ph (t), due to human error from
Equation (13.20) is
Ph (t ) = P1 (t ) =

h
1 R (t )

( + h ) [ r ]

(13.23)

Mean time to robot system failure (MTTRr) is expressed by

MTTR r =

R (t ) d t

(13.24)

+ t
= e ( h ) dt

(13.25)

1
h +

The robot system overall hazard rate or its total failure rate is expressed by

r =

d R r (t )
R r (t ) d t
1

(h + ) t

= h +

1999 by CRC Press LLC

+ t
d e ( h )

dt

(13.26)

Example 13.5
Assume that a robot system can fail either due to human error or other failures and
the human error and other failure times are exponentially distributed. Thus, the values
of human error and other failure rates are 0.0005 error per hour and 0.002 failure/h.
The failed robot system repair times are exponentially distributed with the repair
rate value of = h = 0.004 repair per hour. Calculate the robot system steady state
availability and its reliability for a 100-h mission.
Thus, we have
= 0.002 failure/h, h = 0.0005 error/h, = h = 0.004 repair/h, and t = 100 h.
Substituting the specified values into Equation (13.11) yields robot system steady
state availability
AR =
=

h
h
=
k1 k 2 h + h + h

(0.004)2
(0.004)2 + 0.004 (0.002 + 0.0005)

= 0.6154
Inserting the given data into Equation (13.22) we get
R r (100) = e ( 0.0005+0.002 ) (100 )
= 0.7788
Thus, the robot system steady state availability and reliability are 0.6154 and
0.7788, respectively.

13.3.5 RELIABILITY ANALYSIS

ROBOT SYSTEM

OF A

REPAIRABLE/NON-REPAIRABLE

This section presents reliability analysis of a mathematical model representing a

repairable/non-repairable robot system, by using the Markov technique [34]. The
robot system is composed of a robot and an associated safety system/unit. The
inclusion of safety systems with robots is often practiced because of robot accidents
involving humans. The robot system transition diagram is shown in Figure 13.4.
As shown in Figure 13.4, the normally working robot and its associated safety
unit may move to two mutually exclusive states: robot working normally and the
safety system has failed, safety system working normally and the robot has failed.
From the state robot working normally and its safety system failed, the working
robot system may fail safely or with an incident. The following assumptions are
associated with the model shown in Figure 13.4:

1999 by CRC Press LLC

FIGURE 13.4 Robot system transition diagram.

All failures are statistically independent.

All failure rates are constant.
The robot system is composed of a robot and a safety system.
The robot system fails when the robot fails.

The following notation is associated with the model shown in Figure 13.4:
i

ith state of the system: i = 0 (robot and its associated safety system
operating normally), i = 1 (robot operating normally, robot safety
system failed), i = 2 (robot failed with an incident), i = 3 (robot failed
safely), i = 4 (robot failed but its associated safety system operated
normally).
Pi (t) The probability that the robot system is in state i at time t; for i = 0,
1, 2, 3, 4.
i
ith constant failure rate; i = s (state 0 to state 1), i = ri (state 1 to state 2),
i = rs (state 1 to state 3), i = r (state 0 to state 4).
Using the Markov approach, the following system of differential equations is
associated with Figure 13.4:

1999 by CRC Press LLC

d P0 (t )
+ ( s + r ) P0 (t ) = 0
dt

(13.27)

d P1 (t )
+ ( ri + rs ) P1 (t ) = P0 (t ) s
dt

(13.28)

d P2 (t )
= P1 (t ) ri
dt

(13.29)

d P3 (t )
= P1 (t ) rs
dt

(13.30)

d P4 (t )
= P0 (t ) r
dt

(13.31)

At time t = 0, P0 (0) = 1, and P1 (0) = P2 (0) = P3 (0) = P4 (0) = 0.

Solving Equations (13.27) to (13.31), we get the following state probability equations:
P0 (t ) = e A t
P1 (t ) =

(13.32)

s Ct
e e At
B

(13.33)

where
A = s + r
B = s + r rs ri
C = ri + rs
P2 (t ) =

ri s
1 A e Ct C e At B
AC

) ]

(13.34)

P3 (t ) =

rs s
1 A e Ct C e At B
AC

(13.35)

r
1 e At .
A

(13.36)

[ (

[ (

P4 ( t ) =

) ]

The reliability of both robot and its safety system working normally is given by
R rsu (t ) = P0 (t ) = e At

(13.37)

The reliability of the robot working normally with or without the safety system
functioning successfully is:
Rss ( t ) = P0 (t ) + P1 (t ) = e At +

1999 by CRC Press LLC

s Ct
e e At .
B

(13.38)

The mean time to failure of the robot with the safety system up is expressed by:

MTTFrsu =

rsu

(t ) d t =

1
.
A

(13.39)

Similarly, the mean time to failure of the robot with safety system up or down is
expressed by:

MTTFss =

R
0

ss

(t ) d t =

1
(1 + s C) .
A

(13.40)

If the failed safety system shown in Figure 13.4 (State 1) can be repaired, the
robot system reliability will improve considerably. Thus, for constant failed safety
system repair rate, , from robot system state 1 to 0, the system of differential
equations for the new scenario is
d P0 (t )
+ A P0 (t ) = P1 (t )
dt

(13.41)

d P1 (t )
+ D P1 (t ) = P0 (t ) s
dt

(13.42)

where
D = ri + rs +
d P2 (t )
= P1 (t ) ri
dt

(13.43)

d P3 (t )
= P1 (t ) rs
dt

(13.44)

d P4 (t )
= P0 (t ) r
dt

(13.45)

At time t = 0, P0 (0) = 1, and P1 (0) = P2 (0) = P3 (0) = P4 (0) = 0.

Solutions to Equations (13.41) through (13.45) are as follows:
2

e At
e r1 t
er t
+
+
P0 (t ) = e At + s
(13.46)
( r1 + A) ( r2 + A) ( r1 + A) ( r1 r2 ) ( r2 + A) ( r2 r1 )

1999 by CRC Press LLC

where
r1, r2 =

E E2 4 F
2

E = A+C+
F = ri s + rs s + ri r + rs r + r

[(

P1 (t ) = s e r1t e1r2 t

P4 (t ) =

) (r r )]

(13.47)

[ (

) (r r ])

(13.48)

[ (

) (r r ])

(13.49)

P2 (t ) =

ri s
1 + r1 e r2 t r2 e r1t
r1 r2

P3 (t ) =

rs s
1 + r1 e r2 t r2 e r1t
r1 r2

1
r
e At
1 e At + s r

A
r1 r2 A A ( r1 + A) ( r2 + A)

(13.50)

e
e
+
+

r1 ( r1 + A) ( r1 r2 ) r2 ( r2 + A) ( r2 r1 )
r1t

rt

The reliability of both robot and its associated safety system working normally with
safety system repair facility from Equation (13.46) is
2

e At
e r1t
er t
R rsr (t ) = e At + s
+
+
(13.51)
( r1 + A) ( r2 + A) ( r1 + A) ( r1 r2 ) ( r2 + A) ( r2 r1 )

The reliability of the robot operating normally with or without the safety system
operating (but having the safety system repair facility) from Equations (13.46) and
(13.47) is
Rssr (t ) = e At +

s e r1t e r2 t

(r1 r2 )

) +
s

e At
( r1 + A) ( r2 + A)

e
e
+
+
(r1 + A) (r1 r2 ) (r2 + A) (r2 r1 )
r1t

1999 by CRC Press LLC

rt

(13.52)

The mean time to failure of the robot with repair and with safety system operating
is expressed by

MTTFrsr =

rsr

(t ) d t

(13.53)

1
(1 + s F) .
A

Similarly, the mean time to failure of the robot with repair and with or without
the safety system operating is given by

MTTFssr =

ssr

(t ) d t

(13.54)

1
1 + s F) + s .
(
A
A

Example 13.6
Assume that a robot system is composed of a robot and a safety system/unit and
the operating robot with failed safety unit can either fail with an incident or safely.
The failure rates of robot and safety system/unit are 0.0004 failure per hour and
0.0002 failure per hour, respectively. In addition, the failures rates of the robot failing
with an incident or safely are 0.0001 failure per hour and 0.0003 failure per hour,
respectively. Calculate the following and comment on the end results:
The robot system mean time to failure when the safety system is operating.
More specifically, use Equation (13.39).
The robot system mean time to failure with repair. More specifically, use
Equation (13.53).
Thus, in this example, we have r = 0.0004 failure per hour, s = 0.0002 failure per
hour, ri = 0.0001 failure per hour, rs = 0.0003 failure per hour, and = 0.005 repair
per hour. Substituting the specified data into Equation (13.39) yields
MTTFrsu =
=

1
1
=
A s + r
1

(0.0002) + (0.0004)

= 1666.67 h

1999 by CRC Press LLC

Inserting the given data into Equation (13.53) we get

MTTFrsr = (1 + s F ) A
= (1 + (0.0005) (0.0002) F ) A
= 2410.7 h
where
F = (0.0001) (0.0002) + (0.0003) (0.0002) + (0.0001) (0.0004) + (0.0003)
(0.0004) + (0.005) (0.0004)
= 2.24 10-6
A = (0.0002) + (0.0004)
= 0.0006
The above results indicate that with the introduction of repair, the robot system mean
time to failure improved from 1666.67 h to 2410.7 h.

13.4 PROBLEMS
1. Discuss the historical developments in robotics with emphasis on robot
reliability and safety.
2. Discuss the basic hazards associated with robots.
3. State safety considerations in various phases of the robot life cycle.
4. Describe the following robot safeguard methods:
Intelligent systems
Infrared light arrays
Flashing lights
5. What are the important areas for human factors attention in robotic safety?
6. What are the main causes of robot failures?
7. Define the terms robot reliability and robot mean time to failure.
8. Assume that the times to failure of a robot are Weibull distributed. Obtain
an expression for the robot reliability.
9. What is the important difference between the following two terms:
Mean time to robot failure
Mean time to robot-related problems
10. A robot system can fail either due to human error or other failures. The
estimated values of human error and other failure rates are 0.0007 error
per hour and 0.004 failure per hour. The repair rate of the failed robot
system either due to human error or other failures is 0.01 repair per hour.
Calculate the steady state probability of the robot system being in the
failed state due to human error.

1999 by CRC Press LLC

13.5 REFERENCES
1. Heer, E., Robots in modern industry, in Recent Advances in Robotics, Beni, G. and
Hackwood, S., Eds., John Wiley & Sons, New York, 1985, pp. 11-36.
2. Dhillon, B.S., Robot Reliability and Safety, Springer-Verlag, New York, 1991.
3. Zeldman, M.I., What Every Engineer Should Known About Robots, Marcel Dekker,
New York, 1984.
4. Worldwide Robotics Survey and Directory, Robot Institute of America, P.O. Box
1366, Dearborn, MI, 1983.
5. Rudall, B.H., Automation and robotics worldwide: reports and surveys, Robotica, 14,
243-251, 1996.
6. Parsad, H.P., Safety standards, in Handbook of Industrial Robots, Nof, S.Y., Ed., John
Wiley & Sons, New York, 1988.
7. An Interpretation of the Technical Guidance on Safety Standards in the Use of
Industrial Robots, Japanese Industrial Safety and Health Association, Tokyo, 1985.
8. Bonney, J.F. and Yong, J.F., Eds., Robot Safety, Springer-Verlag, New York, 1985.
9. Industrial Robots and Robot Systems: Safety Requirements, American National Standards Institute, New York, 1986.
10. Dhillon, B.S., On robot reliability and safety: Bibliography, Microelectronics and
Reliability, 27, 105-118, 1987.
11. Dhillon, B.S., Robot accidents, Proc. First Beijing Int. Conf. Reliability, Maintainability, and Safety, International Academic Publishers, Beijing, 1992.
12. Altamuro, V.M., Working safely with the iron collar worker, National Safety News,
38-40, 1983.
13. Nicolaisen, P., Safety problems related to robots, Robotics, 3, 205-211, 1987.
14. Study on Accidents Involving Industrial Robots, prepared by the Japanese Ministry
of Labour, Tokyo, 1982. NTIS Report No. PB 83239822, available from the National
Technical Information Service (NTIS), Springfield, VA.
15. Request for Assistance in Preventing the Injury of Workers by Robots, prepared by
the National Institute for Occupational Safety and Health, Cincinnati, OH, December
1984, NTIS Report No. PB 85236818, available from the National Technical Information Service (NTIS), Springfied, VA.
16. Ziskovsky, J.P., Working safety with industrial robots, Plant Eng., May, 81-85, 1984.
17. Ziskovsky, J.P., Risk analysis and the R3 factor, Proc. Robots 8 Conf., 2, 15.915.21,
1984.
18. Bellino, J.P. and Meagher, J., Design for safeguarding, Robots East Sem., Boston,
MA, October 9-11, 1985, pp. 24-37.
19. Russell, J.W., Robot safety considerations: A checklist, Professional Safety, December
1983, pp. 36-37.
20. Nicolaisen, P., Ways of improving industrial safety for the programming of industrial
robots, Proc. 3rd Int. Conf. Human Factors in Manufacturing, November 1986,
pp. 263-276.
21. Jiang, B.C., Robot Safety: Users Guidelines in Trends in Ergonomics/Human Factors
III, Karwowski, W., Ed., Elsevier, Amsterdam, 1986, pp. 1041-1049.
22. Addison, J.H., Robotic Safety Systems and Methods: Savannah River Site, Report
No. DPST-84-907 (DE 35-008261), December 1984, issued by E.I. du Pont de
Nemours & Co., Savannah River Laboratory, Aiken, SC.
23. Lembke, J.R., An Evaluation of Robot Perimeter Control Devices, Topical Report No.
705349, Report No. DBX-613-3031, Bendix (Kansas City Division), January 1984.

1999 by CRC Press LLC

24. Zimmers, E.W., Human factors aspects of robotic safety, Proc. Robotic Industries
Assoc. (RIA) Robot Safety Sem., Chicago, April 24, 1986, pp. 1-8.
25. Engelberger, J.F., Robotics in Practice, Kogan Page, London, 1980.
26. Dhillon, B.S., On robot reliability and safety, Microelectronics and Reliability, 27,
105-118, 1987.
27. Dhillon, B.S., Robot Reliability and Safety, Springer-Verlag, New York, 1991.
28. Sato, K., Case study of maintenance of spot-welding robots, Plant Maintenance, 14,
28, 1982.
29. Sugimoto, N. and Kawaguchi, K., Fault tree analysis of hazards created by robots,
Proc. 13th Int. Symp. Industrial Robots, 1983, pp. 9.139.28.
30. Khodabandehloo, K., Duggan, F., and Husband, T.F., Reliability of industrial robots:
A safety viewpoint, Proc. 7th British Robot Assoc. Annu. Conf., 233-242, 1984.
31. Khodabandehloo, K., Duggan, F., and Husband, T.F., Reliability assessment of industrial robots, Proc. 14th Int. Symp. Industrial Robots, 209-220, 1984.
32. Dhillon, B.S., Reliability Engineering in Systems Design and Operation, Van Nostrand Reinhold Company, New York, 1983.
33. MIL-HDBK-217F, Reliability Prediction of Electronic Equipment, U.S. Department
of Defense, Washington, D.C.
34. Dhillon, B.S. and Yang, N., Reliability analysis of a repairable robot system, J. Quality
Maintenance Eng., 2, 30-37, 1996.

1999 by CRC Press LLC

14

Medical Device
Reliability

14.1 INTRODUCTION
Even though the history of Reliability Engineering goes back to World War II, the
serious thinking of the applications of reliability engineering concepts to medical
devices or equipment is not that old. Nowadays, there are encouraging signs that
both the engineers and manufacturers in the medical field are breaking away from
their conventional outlook. There could be various reasons for this new phenomenon;
government requirements, cost effectiveness, public demand, advances made in other
areas, etc. For example, because of the dependance of human life on the performance
and reliability of the myriad of medical devices, in the U.S. the two basic important
sources of regulation and control are the government (e.g., Food and Drug Administration (FDA)) and the legal system (e.g., product liability) [1]. Basically, both of
these factors have played an instrumental role in driving the medical device industry
to implement more and stringent controls on its products.
Therefore, today an engineer involved in the design of various medical products
must address the issue of the degree of reliability desired as a function of the specific
application. This is very crucial because a consequence of a failure may vary from
one application area to another. For example, harm to the patient could result from
a failure in a defibrillator; however, on the other hand, failure of a monitoring device
may be tolerated without directly endangering the patient.
The latter part of the 1960s may be regarded as the real beginning of the medical
device reliability field as this period witnessed the start of various publications on
the subject [26]. In 1980 an article entitled Bibliography of Literature on Medical
Equipment Reliability provided a comprehensive list of publications on the
subject [7]. In 1983, a book on reliability engineering devoted a chapter to medical
equipment reliability [8]. Since the early 1980s, many more publications on the field
have appeared and this chapter discusses various aspects related to the medical device
reliability.

14.2 FACTS AND FIGURES, GOVERNMENT CONTROL

AND LIABILITY
The modern health care system has been in existence for over a century; for example,
the American Hospital Association (AHA) was founded in 1896 and various types
of medical devices have been designed and manufactured for a long time. With
advances in modern technology, the sophistication of medical devices has also
increased at a significant rate. For example, Greatbatch invented and patented the
implantable cardiac pacemaker somewhere between 1958 and 1960 and the first

1999 by CRC Press LLC

successful human implant was performed in Buffalo in 1960 [9]. Today, total artificial heart and assist devices are under development and they are expected to play
an important role in the improvement of health care throughout the world. Nonetheless, some of the facts and figures directly or indirectly associated with the
medical devices are as follows:
There are approximately 5000 institutional, 600 associate, and 40,000
personal members of the AHA. The institutional members include short
and long term hospitals, headquarters of health care systems, hospital
affiliated educational programs, etc. and the associate members are made
up of organizations such as commercial firms, consultants, and suppliers.
The individuals working in the health care field, health care executive
assistants, full-time students studying hospital administration, etc. are
classified under the personal member category.
In 1974 dollars, the total assets of the AHA registered hospitals were in
the order of $52 billion [10].
Today, modern hospitals use over 5000 different medical devices ranging
from tongue depressors to sophisticated devices such as pacemakers.
Medical devices account for approximately $120 billion worldwide
market [11] and in 1984, medical devices using electronics accounted for
over $11 billion annual worldwide sales [1214].
In 1997, there were 10,420 registered medical device manufacturers in
the U.S. [15].
In the middle of the 1990s, 93% of medical devices had markets worth
less than $150 million [16].
The committee on hospitals (NFPA) stated 1200 deaths per year due to
faulty instrumentation [17, 18].
In 1990, an FDA study reported that approximately 44% of the quality
related problems that resulted in voluntary medical device recall during
the period from October 1983 to September 1989 were attributable to
deficiencies/errors that could have been eradicated by effective design
controls [19].
During the early half of the 1990s, over a period of five years, the top 10
orthopedic companies in the U.S. increased their regulatory affairs staff
by 39% [20].
Operator errors account for well over 50% of all technical medical equipment problems [21].
According to the findings of the Philadelphia Emergency Care Research
Institute, from 4 to 6% of hospital products were sufficiently dangerous
to warrant immediate correction. These findings were based on testing of
a sample of 15,000 different hospital products [21].
In 1990, the U.S. Congress passed the Safe Medical Device Act (SMDA)
that empowered the FDA to implement the Preproduction Quality Assurance program. This program requires/encourages medical device manufacturers to address deficiencies in product design contributing to failures.

1999 by CRC Press LLC

As medical devices can range from a complexity of a modern magnetic resonance

imaging (MRI) scanner to the simplicity of dental floss, many countries around the
world regulate them and impose restrictions on their manufacture and import including quality and reliability. As the competition is a good policeman of economy, the
market imposes significant expectations of performance and reliability of the medical
devices. In fact, in many times such expectations surpass the regulatory requirements
quite significantly.
In the U.S., the FDA regulates medical devices to ensure their safety and
effectiveness [22]. For example, two years after the enactation of the Medical Device
Amendments of 1976, the FDA issued the medical device good manufacturing
practices (GMPs) regulation. This document included requirements prescribing the
facilities, techniques/methods, and controls to be employed in the medical device
manufacture, packaging, and storage [23]. However, the U.S. Congress Subcommittee on Oversight and Investigations held hearings on March 13, 1984 concerning
medical device failures, particularly deaths resulting from cardiac pacemaker leads
that malfunctioned due to design defects [19]. It concluded that the FDA was unable
to address such problems under the 1976 amendments.
Nowadays, the FDA requires reports concerning incidents of death, serious
injury or serious illness, or failure of a certain device likely to cause or contribute
to a death or serious injury if the failure were to recur [24]. Manufacturers, distributors, and user facilities of medical devices have certain reporting requirements.
Noncompliance to such requirements can lead to FDA regulatory actions including
warning letters, product seizure, injunction, or civil penalties. Also, it is mandatory
that any organization (including non-U.S. manufacturers) that manufactures medical
devices for the U.S. market must establish a quality system that is in line with the
GMPs [22].
A different strategy to regulate medical devices is followed by the European
Union (EU). The Medical Device Directive (MDD) of the European Parliament
defines the requirements for medical devices in the EU countries. After May 1998,
a firm (including non-EU companies) manufacturing medical devices to market in
the EU countries must establish a quality system in conformance to the MDD. The
MDD outlines a set of requirements relating to the medical device safety covering
a broad range, including electromagnetic compatibility, biological compatibility, and
electrical safety [22].
Currently, product liability is another factor for manufacturing safe and reliable
medical devices. For example on June 26, 1996, the U.S. Supreme Courts ruling
on the Medtronic, Inc. vs. Lohr case handed out a sweeping victory to American
consumers by rejecting the Medtronics claim that federal law provides immunity
to medical device manufacturers from all civil liability, including when individuals
are injured or killed due to defective products [25]. This case was filed by a Florida
woman named Lora Lohr who was implanted with a cardiac pacemaker to rectify
her abnormal heart rhythm. The pacemaker manufactured by Medtronic failed and
resulted in requiring emergency surgery plus additional medical procedures. The
lawsuit filed by Lohr alleged that the failure of the pacemaker was the result of
defective design, manufacturing, and labeling.

1999 by CRC Press LLC

In delivering the judgment on the case, Justice John Paul Stevens wrote
Medtronic argument is not only unpersuasive, it is implausible [25]. The Supreme
Courts decision has put additional pressure on manufacturers to produce safe and
reliable medical devices.

14.3 MEDICAL ELECTRONIC

EQUIPMENT CLASSIFICATION
There is a large variety of electronic equipment used in the health care system.
Basically, such equipment may be grouped under three broad categories [3, 21].
Category A
Category B
Category C
Category A includes equipment requiring high reliability because it is directly and
immediately responsible for the patients life or which may become so in emergencies. When such equipment fails, there is seldom much time for even emergency
repairs; thus, it must always function at the moment of need. Some examples of
such equipment are respirators, cardiac defibrillator, electro-cardiographic monitors,
and cardiac pacemakers.
A vast majority of medical electronic equipment fall under Category B. Such
equipment is used for routine or semi-emergency diagnostic or therapeutic purposes.
Failure of this equipment does not create the same emergency as in the case of
Category A equipment; in addition, there is time for repair. Some examples of this
category of equipment are electronyography equipment, electrocardiograph and electroencephalograph recorders and monitors, ultrasound equipment, gas analyzers,
colorimeters, diathermy equipment, and spectrophotometers.
Equipment belonging to Category C is not essential to a patients life or welfare
but serves as a convenience equipment. Some of the examples are wheel chairs,
electric beds, and bedside television sets. One should note here that there is considerable overlap, especially in Categories A and B. More specifically, say some equipment of Category B may be used as Category A equipment. One example of such
equipment is electrocardiograph recorder or monitor.

14.4 MEDICAL DEVICE RECALLS

In the U.S., defective medical devices are subject to recall and repair. The passage
of the Medical Device Amendments of 1976 has helped to increase the publics
awareness of these activities and their corresponding number. The FDA Enforcement
Report publishes data on medical device recalls. For example, during the period of
19801982, there were a total of 230 medical device related recalls [26]. These
recalls were classified into nine problem areas as follows [21]:
Faulty product design
Defective components

1999 by CRC Press LLC

Defects in material selection and manufacturing

Missassembly of parts
Electrical problems
Contamination
Mislabeling
Radiation (X-ray) violations
No premarket approval and failure to comply with GMPs.

The number of recalls associated with each of these categories was 40, 13, 54, 10,
17, 40, 27, 25, and 4, respectively. It is interesting to note that faulty product design,
product contamination, mislabeling, and defects in material selection and manufacturing accounted for 70% of the medical device recalls. The components of the faulty
product design category were premature failure, alarm defects, potential for malfunction, electrical interference, failure to perform as required, and potential for leakage
of fluids into electrical components. Four subcategories of the product contamination
were defective package seals, non-sterility, other package defects, and other faults.
The problem areas of the mislabeling category were incomplete labeling, inadequate
labeling, disparity between label and product, and misleading or inaccurate labeling.
There were five specific problems in the material selection and manufacturing category: inappropriate materials, manufacturing defects, separation of bonded components, actual or potential breakage/cracking, and material deterioration.

14.5 MEDICAL DEVICE DESIGN

QUALITY ASSURANCE
Just like in the life cycle of other engineering devices, the design phase is the most
important phase in the life cycle of a medical device. The inherent safety and
effectiveness of the medical devices can only be improved through design enhancement. Careful planning and management of the design process can help in establishing effectiveness, reliability, and safety in a device.
It is emphasized that a deficiency in a device design could be very costly once
the device is manufactured and released for field use. The costs may not only be
made up of replacement and redesign but also of liability and loss of customer faith
in the device. Nonetheless, according to FDA studies, one of the major causes of
device failure is defective design. Thus, the design quality assurance is very important to produce reliable and safe medical devices.
In 1989, for the first time the FDA published a notice of availability for design
control recommendations entitled Preproduction Quality Assurance Planning: Recommendations for Medical Device Manufacturers [23]. It encouraged all medical
device manufacturers to follow these recommendations by establishing a program
for assessing the reliability, safety, and effectiveness of device design before releasing the design to manufacturing. The FDA preproduction or design quality assurance
program is composed of 12 elements:
1. organization
2. specifications

1999 by CRC Press LLC

3.
4.
5.
6.
7.
8.
9.
10.
11.
12.

design review
reliability assessment
parts/materials quality control
software quality control
design transfer
labeling
certification
test instrumentation
manpower
quality monitoring subsequent to the design phase.

Each of these elements is described below.

14.5.1 ORGANIZATION
The formal documentation should contain the organizational elements and authorities
desirable to develop the preproduction quality assurance (PQA) program, to execute
program requirements, and to achieve program goals. Also, it is essential to formally
assign and document the responsibility for implementing the entire program and
each program element. Audit the PQA program periodically and update it as experience is gained.

14.5.2 SPECIFICATIONS
After establishing desired device design characteristics such as physical, chemical,
performance, etc., translate them into written design specifications. Specifications
should address performance characteristics, as applicable, such as reliability, safety,
stability, and precision, important to meet the devices end objective. It is important
to remember that these preliminary specifications provide the vehicle with which to
develop, control, and evaluate the device design.
Evaluate and document changes made to the specifications during research and
development, in particular, ensure that the products safety or effectiveness is not
compromised by these changes. All in all, for the sake of effectiveness of specifications, the qualified individuals from areas such as reliability, quality assurance,
manufacturing, and marketing should be asked to review the specification document.

14.5.3 DESIGN REVIEW

The purpose of design review is to detect and correct deficiencies during the design
phase of a medical device. As the cornerstone of the PQA program, the design
reviews should be conducted at various stages of device design to assure conformance to design criteria and to detect weak design spots. The sooner a design review
is initiated, the better, as the design weaknesses can be detected earlier and in turn
it will cost less to rectify the problem. The extent and frequency of design reviews
depend on the significance and complexity of the device under consideration. The
items to be assessed during these reviews include subsystems (including software,

1999 by CRC Press LLC

if applicable), components, drawings, test results and specifications, packaging,

analysis results, and labeling.
Usually, the members of the design review team belong to areas such as engineering, quality assurance, manufacturing, purchasing, servicing, and marketing.
In particular, the design reviews should include, as applicable, failure mode and
effect analysis. In addition, such analysis should be performed at the initial stage of
the design effort and as part of each design review to detect design flaws, especially
with respect to device safety, reliability, and performance. Usually, the following
two failure mode analysis techniques are employed:
Failure Mode Effects and Criticality Analysis (FMECA)
Fault Tree Analysis (FTA)
FMECA is an inductive bottom-up approach and assumes the existence of
basic defects at the component level and then evaluates the effects on assembly
levels. Furthermore, it is a structured approach for proceeding part-by-part through
the system to determine effects of failures and is described in detail in Reference 26.
FTA is another approach to failure mode analysis, but it is a deductive and topdown method. The approach is especially useful for application to medical devices
because human/device interfaces can be taken into consideration. The approach is
described in detail in the chapter devoted to the method.

14.5.4 RELIABILITY ASSESSMENT

This is the prediction and demonstration process for estimating the basic reliability
of a device, and it can be used for both new and modified designs. More specifically,
reliability assessment may be described as a continuous process involving reliability
prediction, demonstration, data analysis, then reliability re-prediction, re-demonstration, and re-data analysis continually.
Reliability assessment can be started by theoretical/statistical approaches by first
determining each parts reliability, then progressing upward, establishing each subassembly and assembly reliability, until the establishment of the entire devices reliability.
Even though a good estimate of reliability can be made by testing complete devices
under simulated operating environments, the most meaningful reliability data can only
be obtained from field use. By making the reliability assessment an essential component of the PQA program, it will certainly help to produce reliable medical devices.

14.5.5 PARTS/MATERIALS QUALITY CONTROL

In order to assure that all parts/materials (P/M) used in device designs have the
desirable reliability, it is essential to have an effective P/M quality assurance program. The essential elements of the program are P/M selection, specification, qualification, and ongoing quality verification, i.e., irrespective of whether P/M was
fabricated in-house or supplied by vendors.
It is useful to develop a preferred P/M list during the preliminary design stage
and refine it as the design progresses.

1999 by CRC Press LLC

14.5.6 SOFTWARE QUALITY CONTROL

The establishment of a software quality assurance program (SQAP) is essential when
a medical device design incorporates software developed in-house. The basic purpose
of this program is to outline a systematic approach to software development, particularly including a protocol for the device software formal review and validation
to meet overall functional reliability. Nonetheless, the primary goals of the software
quality assurance are reliability, maintainability, correctness, and testability.
Reference 27 is a useful document for writing software quality assurance plans.

14.5.7 DESIGN TRANSFER

Once the medical device design is transformed into a physical entity, its technical
adequacy, reliability, and safety verifications are sought through testing under the real
life operating or simulated use conditions. It is to be noted that in going from laboratory
to scaled-up production, it is quite possible that standards or techniques and procedures
may not be accurately transferred or additional manufacturing processes are added.
Thus, this factor requires careful attention. All in all, the clinical trials of a device under
consideration can only begin after the verification of its safety and reliability under the
simulated operating environments, specifically at the expected performance limits.

14.5.8 LABELING
Labeling includes items such as display labels, manuals, inserts, software for cathode
ray tube (CRT) displays, panels, charts, and recommended test and calibration
protocols. Design reviews are a useful vehicle to ensure that the labeling is in
compliance with laws and regulations as applicable and the directions for the device
usage can be understood easily by the user community.

14.5.9 CERTIFICATION
After passing the preproduction qualification testing by the initial production units,
it is essential to perform a formal technical review to assure adequacy of the design,
production, and quality assurance approaches/procedures. In addition, the review
should incorporate a determination of items such as those listed below.
Overall adequacy of quality assurance plan.
Resolution of any discrepancy between the standards/procedures
employed to construct the design during the research and development
phase and those identified for the production phase.
Adequacy of specifications.
Suitability of test approaches employed for evaluating compliance with
the approved specifications.
Adequacy of specification change control program.
Resolution of any discrepancy between the approved specifications for
the device and the end produced device.

1999 by CRC Press LLC

14.5.10 TEST INSTRUMENTATION

This is concerned with effectively calibrating and maintaining, as per the formal calibration and maintenance program, the equipment used in qualification of the design.

14.5.11 MANPOWER
Only qualified and competent individuals must be allowed to perform device associated design activities such as design review, analysis, and testing.

14.5.12 QUALITY MONITORING SUBSEQUENT

TO THE DESIGN PHASE
The quality assurance effort does not end at the design phase of a medical device,
but it continues in manufacturing and field use phases. In order to meet this challenge,
it is expected that manufacturers of medical devices establish an effective program
for activities such as follows:
Identifying trends or patterns associated with device failures.
Performing analysis of quality related problems.
Initiating appropriate corrective measures to stop recurrence of the above
failures or problems.
Reporting in a timely manner the problems found either in-house or in
the field use.
The device manufacturers should also establish an appropriate mechanism to
assure that failure data resulting from user complaints and service records relating
to design are reviewed by the individuals involved with the design of the device.

14.6 HUMAN ERROR OCCURRENCE AND RELATED

HUMAN FACTORS
Human errors are universal and committed each day; some are trivial and others are
serious or fatal. In the medical field, human errors can result in serious injuries and
even deaths [28, 29]. Some examples and facts concerning human error occurrence
in the health care system are as follows [29, 30]:
Up to 90% of accidents both generally and in medical devices are due to
human error [31, 32].
A human error caused a fatal radiation overdose accident involving the
Therac radiation therapy device [33].
A study of anesthesiology errors revealed that 24% of the anesthesiologists
responded positively to committing an error with fatal results [34].
A patient was seriously injured by over-infusion because a nurse mistakenly read the number 7 as a 1 [30].

1999 by CRC Press LLC

 During the treatment of an infant patient with oxygen, a physician set the
flow control knob between 1 and 2 liters per minute without realizing the
fact that the scale numbers represented discrete, instead of continuous,
settings. As the result of the physicians action, the patient became
hypoxic [30].

14.6.1 CONTROL/DISPLAY RELATED HUMAN FACTORS GUIDELINES

As many medical devices usually contain large consoles with rows of controls and
displays, they could be the important cause for many human errors if designed
without seriously considering human factors. The guidelines such as those listed
below have proven quite useful for displays and controls [30]:
Ensure that controls, displays, and workstations are designed considering
user capabilities such as reach, memory, strength, hearing, vision, and
dexterity.
Ensure that the design of switches and control knobs correspond to the
user conventions outlined in current medical device standards and so on.
Ensure that all design facets are consistent with user expectations as much
as possible.
Ensure that all displays and labels are made so that they can be read with
ease from normal distances and viewing angles. In particular, pay attention
to factors such as color, symbol size, display depth, and controls.
Ensure that the design of control and display arrangements is well organized and uncluttered.
Ensure that tactile feedback is provided by controls.
Ensure that knobs, switches, and keys are arranged and designed so that
the likelihood of inadvertent activation is reduced to a minimum.
Ensure that color and shape coding, as appropriate, are used to facilitate
the efficient identification of controls and displays.
Ensure that the intensity and pitch of auditory signals are such that they
can be easily heard by the device users.
Ensure that visual signals brightness is such that it can be easily perceived
by users performing their tasks under various ambient illumination conditions.
Ensure that items such as acronyms, abbreviations, text, and symbols
placed on or displayed by the device are consistent with the instructional
manual.

14.6.2 MEDICAL DEVICE MAINTAINABILITY RELATED HUMAN

FACTOR PROBLEMS
As poor maintenance can affect safe and reliable operation of medical devices, proper
emphasis is required during design to simplify their maintenance. Some of the
problems frequently encountered by the maintainers include poor self-diagnostic

1999 by CRC Press LLC

capability, ambiguous part arrangements, poor design for easy cleaning, difficulty
in reaching or manipulating screws/parts, inadequate labeling, coding, or numbering
of parts, difficulty in locating parts visually or by touch, and requirements for
difficult-to-find tools [30].

14.6.3 HUMAN FACTOR POINTERS FOR ALREADY BEING

USED/TO BE PURCHASED MEDICAL DEVICES
The recognition of human factor design related problems with already being used
or to be purchased medical devices is important for their safe, reliable, and effective
operation. When the design of a medical device has serious deficiencies with respect
to human factors, patterns may emerge. Experienced users, maintenance personnel,
and others can be a valuable source of information. Specifically, the symptoms of
problems with already being used devices may be categorized in four groups [30]:
1.
2.
3.
4.

Complaints
Observations
Installation Problems
Incidents

Some examples of the problems under the complaint group are poorly located
or labeled controls, illogical and confusing device operation, difficult to read or
understand displays, annoying device alarms, and difficult to hear or distinguish
alarms. The items belonging to the observation group include slow and arduous
training, refusal of the staff to use the device, and only a few individuals seem to
be able to use the device correctly. Some of the installation related problems are
accessories installed incorrectly, frequently parts become detached, involved individuals find installation of accessories difficult, confusing, or overly time-consuming,
and alarms and batteries fail often. One important element of the incident category
is that highly competent people are involved in accidents or near-misses. More
specifically, this could be a warning.
It is important to consider the means of assessing usability of medical devices
prior to their procurement, particularly if they are to be used for life-sustaining or
life-supporting purposes. Some of the associated steps include the following:
Determine if the manufacturer has performed human factors/usability
testing of the device under consideration.
Review published evaluation data concerning the device under consideration.
Determine from staff and other sources about the human factor related
performance concerning predecessor device models produced by the same
manufacturer.
Negotiate a trial period prior to the actual procurement of the device under
consideration.
Make contact with facilities that have already used the device under
consideration.

1999 by CRC Press LLC

14.6.4 ISSUES IN CONSIDERING THE NEED FOR, AND PERFORMANCE

OF, HUMAN FACTORS ANALYSIS AND TESTING
OF MEDICAL DEVICES
There are many issues that could be raised in considering the need for, and performance of, human factors analysis and testing of medical devices. Some of these
issues are consequences of error for the patient/user, fulfillment of user requirements,
requirement of user interaction with the device with respect to operation, maintenance, cleaning, or parts installation, reviewing of the literature and company files
for useful human factors information, likelihood of the combination of user interface,
user population, and operating conditions leading to errors, types of studies, analyses,
and tests being conducted, testing in simulated and/or actual environments by the
project team, user involvement in actual testing, individuals belonging to the design
team focusing on the user-related issues during actual testing, existence of a test
plan, development of user requirements and their update, and coordination of hardware and software designers, technical writers, and the efforts of others with respect
to human factors [30].

14.7 MEDICAL DEVICE SOFTWARE

Nowadays many medical devices use software and its reliability is as important as
the hardware reliability of such devices. For example, as per References 35 and 36,
software failures have lead to patient fatalities, and the potential for more such
hazards is increasing as software becomes complex and increasingly crucial in the
manufacture of medical devices. There are primarily two reasons for medical device
software failures: inadequacies in the functionality of the software and errors in its
implementation [37]. Before we examine the subject of software quality assurance/reliability in detail, let us first explore the definition of a medical device because
the FDA medical device statutory and regulatory provisions application is subject
to items falling under the Federal Food, Drug, and Cosmetic Act (FD&C Act)
definition.
The definition of a medical device under this Act basically is any instrument/apparatus/machine/etc. or other similar or related article, including any part,
component, or accessory intended for application in the diagnosis of disease/prevention of disease/etc. or intended to affect any function or the structure of
body [38, 39]. Any software satisfying this definition is subject to concerned FDA
rules and regulations. FDA classifies software into two categories: stand-alone software (i.e., it can be a device itself) and software that is a part, component, element,
or accessory to a device. Some examples of the stand-alone software are as
follows [39]:
Blood bank software systems controlling donor deferrals and the release
of blood products.
Software that helps diagnose osteoporosis.

1999 by CRC Press LLC

 Software that analyzes potential therapeutic interventions for a certain

patient.
Information systems used in hospitals.
The typical examples of component or part software are the software used in
pacemakers, diagnostic X-ray systems, ventilators, infusion pumps, magnetic resonance imaging devices, etc. Similarly, examples of the accessory software are the
softwares used for the conversion of pacemaker telemetry data, computation of bone
fracture risk from bone densitometry data, and calculation of rate response for a
cardiac pacemaker.

14.7.1 SOFTWARE TESTING

DEVICE SAFETY

FOR IMPROVING

MEDICAL

Medical device safety with respect to software failures could be significantly

improved through software testing. The basic purpose of software testing is to ensure
that a device performs its assigned task and its potential for hazard is minimized.
The testing accounts for a very large proportion of the total software development
process. Nonetheless, software testing may be grouped into two areas: manual
software testing and automated software testing. Each of these areas is described
below [36].
Manual Software Testing
There are many traditional approaches to manual software testing including functional testing, safety testing, free-form testing, white box testing, and error testing
as shown in Figure 14.1.

FIGURE 14.1 Some of the traditional approaches to manual software testing.

1999 by CRC Press LLC

Functional testing is the heart of traditional validation and verification procedures

and focuses test efforts on the performance expected from the device or, more
specifically, the device must perform as per its design specification. This type of
testing usually receives the most attention and the greatest test cases because each
function must undergo exhaustive testing. Even though functional testing is very
time-consuming, it is an orderly process. However, it is extremely boring, and
boredom frequently leads to human error in the tests and omissions in the documentation, irrespective of the testers effort in accuracy.
The purpose of the safety testing is that a device must operate safely. Thus, this
type of testing is focused on areas of potential harm to people such as patients or
users. Safety testing is useful for verifying that software-associated hazards have
been mitigated [40].
As no formal testing can find each and every important software bug, it is
advisable to conduct free-form testing on top of the formal approaches such as
described earlier. Free-form testing is not performed according to device specifications; thus, it defies formalization and subjects the device to additional stresses. This
form of testing may simply be described as a process in which testers with varying
knowledge levels operate the device in unexpected and unconventional ways to
provoke failures. Even though free-form testing can discover a large percentage of
the errors or bugs, there is no guarantee that it will cover all potential problem areas.
The white-box testing approaches are employed in situations where the tester is
required to verify the internal operations of the device software in order to test it
satisfactorily. This type of testing allows the tester to look into the interior of the
device and develop tests to discover weak spots in the software programs internal
logic. The white-box testing techniques can be incorporated into other manual test
approaches such as functional testing, free-form testing, and safety testing.
As a device is expected to handle error conditions effectively, the error testing
is a useful tool to assure that the device under consideration operates correctly when
things do go wrong. Some examples of the things that can go wrong are power
fluctuations or outages, nonsense values generated by peripheral devices, and internal
components failures. Tests for errors such as these are rather cumbersome to create
and execute because of the difficulty involved in recreating the errors in a systematic
and repeatable manner.
Automated Software Testing
This makes use of computer technology to test software with the purpose to make
the test process more efficient, accurate, and complete, in addition to backing it up
with better documentation. More specifically, automated software test systems make
use of the computer technology for the following four purposes:
1.
2.
3.
4.

To stimulate the test target.

Monitor associated response from the device.
Record the end results or conclusions.
Control the total process.

1999 by CRC Press LLC

The approaches such as functional testing, error testing, free-form testing, safety
testing, and white-box testing can be improved through automation.

14.7.2 CARDIOPLEGIA DELIVERY SYSTEM SOFTWARE RELIABILITY

PROGRAM CORE ELEMENTS AND SOFTWARE SAFETY
IMPROVEMENT WITH REDUNDANCY
The cardioplegia delivery system is used to provide for multiple combinations of
blood, arresting agent, crystalloids, and additive solutions under a perfusionists
programmable control. During the development of the operating software for a
cardioplegia delivery system, the core elements of the software reliability program
that were implemented included design simplification, documented performance
requirements, code inspections, periodic audits and audit trails, safety hazard analysis, design and coding standards, requirement tracing, a clearly defined user interface, quality practices and standards, reliability modeling to predict operational
reliability, extensive testing beginning at the module level, and documentation and
resolution of all defects [41].
It is considered that knowledge about these core elements will provide a useful
input in developing software for other medical devices.
The past experience indicates that errors in software will still occur even with
the best of development methodologies and intentions. However, it is possible to
improve software safety quite significantly by designing with redundant safety
mechanisms [37]. The process starts by conducting hazard analysis [42]. After the
identification of important hazards and their associated severity and potential causes,
it is feasible to devise electromechanical systems for detecting such conditions and
overriding the devices processor. If such systems can be developed orthogonal to
the software (i.e., the unique point of common functionality is in the discovery of
the error), then it is possible to have true redundancy and improvement in safe
response to hazardous situations.

14.8 SOURCES FOR ADVERSE MEDICAL DEVICE

REPORTABLE EVENTS AND FAILURE
INVESTIGATION DOCUMENTATION
In order to ensure the safety of patients and others, directly or indirectly, the medical
device reporting (MDR) rules and regulations in both the U.S. and EU countries
state that adverse events concerning medical devices must be tracked and reported
in an effective manner. The FDA requires medical device manufacturer/user facilities
such as hospitals, nursing homes, distributors/importers, and ambulatory surgical
facilities to report MDR-type incidents effectively. Failure to comply with its requirements may lead to product seizure, injunction, warning letters, etc.
In particular, it is necessary that the manufacturers of medical devices must
examine all possible sources for MDR-type reportable incidents. Some of those
sources are as follows [43]:

1999 by CRC Press LLC

Servicing/repair reports
User/distributor records
Published/unpublished literature sources
The manufacturers product complaint-handling mechanism
In-house research/testing evaluation/etc. records
Legal records
Sales representative contacts
Technical service customer contacts.

The documentation of an investigation of a device, process, or quality system failure

is very important and requires careful consideration. Thus, the main objective of a
medical device failure investigation documentation should be to provide all necessary
information about the problem as well as to serve as an information repository that
can be utilized meaningfully by other individuals including external auditors and
assessors. Nonetheless, an effective documentation of a failure investigation should
include, at a minimum, proper information on items such as unique identification
of the failure investigation, problem description, level of investigation, investigation
method/procedures/tests/inspections, equipment used in the investigation and their
unique idenfication, reviewed documentation/components/parts/labeling/packaging/processes descriptions, the results of all inspections/testing examinations, analysis methods and analysis of the results, cross-references to other failure investigations/complaints/etc., investigators signature and dates of test and examinations
performed, actual results of the testing and examinations, report of the conclusions
to demonstrate adequate closure, recommendations for corrective actions to prevent
recurrence of the problem, and formal acceptance and closure of the failure
investigation [44].

14.9 A SMALL INSTRUMENT MANUFACTURERS

APPROACH TO PRODUCE RELIABLE AND SAFE
MEDICAL DEVICES
Over the years many companies have developed their own approaches to manufacture
safe and reliable medical devices. One such approach was developed by Bio-Optronics. The Bio-Optronicss product development plan is composed of 13 steps listed
below [45].
1.
2.
3.
4.
5.
6.
7.
8.

Analyze existing medical problems.

Develop a product concept to find a solution to a specific medical problem.
Evaluate environments under which the medical device is operating.
Evaluate the people expected to operate the product under consideration.
Build a prototype.
Test the prototype under laboratory environment.
Test the prototype under the field operating environment.
Make changes to the device design as appropriate to satisfy the actual
field requirements.

1999 by CRC Press LLC

9.
10.
11.
12.
13.

Perform laboratory and field tests on the modified device.

Build pilot units to conduct appropriate tests.
Invite impartial experts to test pilot units under the field environment.
Release the product/device design for production.
Study the field performance of the device and support with device maintenance.

14.10 AEROSPACE AND MEDICAL EQUIPMENT

RELIABILITY AND RELIABILITY
APPROACH COMPARISONS
The development of the reliability discipline was basically associated with the
manufacture of aerospace equipment. In the manufacture of medical devices, the
application of the reliability concepts is relatively new. Nonetheless, comparisons
of a number of factors associated with aerospace and medical product reliability are
presented below [8, 46].
Aerospace Product Reliability
Costly equipment.
A well-developed field for reliability.
Human well being directly or indirectly is a factor.
Large manufacturing concerns with well-developed reliability engineering departments with many years of experience.
Individuals with previous reliability related experience with aerospace
equipment use sophisticated reliability methods.
Medical Product Reliability
Relatively less expensive products.
A relatively new area for reliability concept applications.
Patients well being or lives are involved.
Relatively smaller manufacturing concerns with relatively recent interest with the application of reliability principles in product development.
Individuals with less previous reliability related experience with medical devices use relatively simple reliability approaches.
The driving force in the aerospace industry to consider and analyze reliability by
imposing a complex set of standards and specifications that control all aspects of
equipment design has been the U.S. Department of Defense. In contrast, the medical
device industry driven by factors such as market forces, limited government regulation, and safety considerations, has been lagging in imposing reliability requirements in the device design and development process. A rough comparison of the
approaches followed and controls imposed by both aerospace and medical device
industries with respect to some important factors affecting the product reliability is
given in Table 14.1 [47].

1999 by CRC Press LLC

TABLE 14.1
Aerospace and Medical Industries Approach to Selected Reliability Factors
No.

Reliability
factor

Aerospace industrys
approach

Stress screening

Part quality

Predictions

Design margins

MIL-HDBK-2164 [48] is
used for control.
MIL-STD-883 [49] is used
for control.
MIL-HDBK-217 [50] is
used for control.
MIL-HDBK-251 [52] is
used for control.

Complexity

Testing and
verification

MIL-HDBK-338 [53] is
used for control.
MIL-STD-781 [54] and
MIL-HDBK-781 [55] are
used for control.

Medical industrys approach

None.
Left to individual firm.
Bell Communication Research
(BELLCORE) [51] is used for control.
Safety standards produced by bodies such
as International Electro-Technical
Commission (IEC) and American
Society of Mechanical Engineers
(ASME) are used for control.
Left to individual firm.
None for reliability.

14.11 GUIDELINES FOR RELIABILITY AND OTHER

PROFESSIONALS ASSOCIATED WITH
HEALTH CARE
There are various types of professionals involved in the manufacture and use of
medical devices. Reliability engineers are one of them. References 8 and 46 present
some suggestions for all of these professionals. Thus, suggestions for reliability and
other professionals associated with health care are given below.
Reliability Professionals
Focus your attention on critical device failures as not all device failures
are equally important.
Pay proper attention to cost effectiveness and remember that some
reliability improvement decisions need very little or no additional
expenditure.
In order to obtain immediate results employ approaches such as design
review, parts review, qualitative FTA, and FMEA.
Use simple reliability techniques instead of some sophisticated methods
used by the aerospace industry. Examples of the techniques that can be
quite useful in medical device industry are FMEA and qualitative FTA.
Remember that the device reliability responsibility rests with the manufacturer during the design and manufacturing phase, but in the operational phase it is basically the responsibility of the user.

1999 by CRC Press LLC

 Other Professionals
Keep in the back of your mind that probably the largest single expense
in a business organization is the cost of failures. Such failures could
be associated with people, equipment, business systems, and so on.
The cost of business can be decreased quite significantly by reducing
such failures.
The manufacturer is responsible for device reliability during the design
and manufacturing phases, but during field use the user must use the
device according to design conditions. More specifically, both parties
are expected to accept their responsibilities for the total success.
Make comparison of the human body and medical device failures. Both
of them need positive actions from reliability engineers and doctors to
improve device reliability and extend human life, respectively.
Recognize that failures are the cause of poor medical device reliability
and positive thinking and actions can improve device reliability.
Remember that the applications of reliability principles have successfully improved the reliability of aerospace systems and such applications to medical devices will generate dividends of a similar nature.

14.12 PROBLEMS
1. Make a comparison of medical device reliability with aerospace system
reliability.
2. Write an essay on medical device failures.
3. Discuss at least five typical important reasons for the medical device
recalls in the U.S.
4. State measures proposed by the FDA to produce reliable devices.
5. Give at least five examples of human error in the health care system.
6. Discuss human factors with respect to medical devices.
7. What are the two software classifications used by the FDA? Describe both
of them in detail.
8. Discuss the following types of manual software testing:
Functional testing
Safety testing
Error testing
9. Compare manual software testing with automated software testing.
10. Discuss the possible sources of MDR-type reportable incidents or events.

14.13 REFERENCES
1. Bell, D.D., Contrasting the medical-device and aerospace-industries approach to
reliability, Proc. Annu. Reliability Maintainability Symp., 125127, 1995.
2. Johnson, J.P., Reliability of ECG instrumentation in a hospital, Proc. Annu. Symp.
Reliability, 314-318, 1967.
3. Crump, J.F., Safety and reliability in medical electronics, Proc. Annu. Symp. Reliability, 320-322, 1969.

1999 by CRC Press LLC

4. Gechman, R., Tiny flaws in medical design can kill, Hosp. Top., 46, 23-24, 1968.
5. Meyer, J.L., Some instrument induced errors in the electrocardiogram, J. Am. Med.
Assoc., 201, 351-358, 1967.
6. Taylor, E.F., The effect of medical test instrument reliability on patient risks, Proc.
Annu. Symp. Reliability, 328-330, 1969.
7. Dhillon, B.S., Bibliography of literature on medical equipment reliability, Microelectronics and Reliability, 20, 737-742, 1980.
8. Dhillon, B.S., Reliability Engineering in Systems Design and Operation, Van Nostrand Reinhold Company, New York, 1983.
9. Eberhard, D.P., Qualification of high reliability medical grade batteries, Proc. Annu.
Reliability Maintainability Symp., 356362, 1989.
10. Fairhurst, G.F. and Murphy, K.L., Help wanted, Proc. Annu. Reliability Maintainability Symp., 103-106, 1976.
11. Murray, K., Canadas medical device industry faces cost pressures, regulatory reform,
Medical Device and Diagnostic Industry Magazine, 19(8), 30-39, 1997.
12. Bassen, H., Sillberberg, J., Houston, F., and Knight, W., Computerized medical
devices: Trends, problems, and safety, IEEE Aerospace and Electronic Systems (AES)
Magazine, September 1986, pp. 20-24.
13. Dickson, C., World medical electronics market: An overview, Medical Devices and
Diagnostic Products, May 1984, pp. 53-58.
14. World Medical Electronics Markets 1985 Yearbook, Market Intelligence Research
Company, Palo Alto, CA, March 1985.
15. Allen, D., California home to almost one-fifth of U.S. medical device industry,
Medical Device and Diagnostic Industry Magazine, 19(10), 64-67, 1997.
16. Bethune, J., The cost effectiveness bugaboo, Medical Device and Diagnostic Industry
Magazine, 19(4), 12-15, 1997.
17. Walter, C.W., Electronic News, January 27, 1969.
18. Micco, L.A., Motivation for the biomedical instrument manufacturer, Proc. Annu.
Reliability Maintainability Symp., 242-244, 1972.
19. Schwartz, A.P., A call for real added value, Medical Industry Executive, February/March, 1994, pp. 5-9.
20. Allen, R.C., FDA and the cost of health care, Medical Device and Diagnostic Industry
Magazine, 18(7), 28-35, 1996.
21. Dhillon, B.S., Reliability technology in health care systems, Proc. IASTED Int. Symp.
Comp. Adv. Technol. Med. Health Care Bioeng., 84-87, 1990.
22. OLeary, D.J., International standards: Their new role in a global economy, Proc.
Annu. Reliability Maintainability Symp., 17-23, 1996.
23. Hooten, W.F., A brief history of FDA good manufacturing practices, Medical Device
and Diagnostic Industry Magazine, 18(5), 96, 1996.
24. Cady, W.W. and Iampietro, D., Medical Device Reporting, Medical Device and
Diagnostic Industry Magazine, 18(5), 58-67, 1996.
25. Bethune, J., Ed., On product liability: Stupidity and waste abounding, Medical Device
and Diagnostic Industry Magazine, 18(8), 8-11, 1996.
26. MIL-STD-1629A, Procedures for Performing a Failure Mode, Effects, and Criticality
Analysis, Department of Defense, Washington, D.C.
27. ANSI/IEEE-STD-730-1984, IEEE Standard for Software Quality Assurance Plans,
American National Standards Institute (ANSI), New York, 1984.
28. Bogner, M.S., Ed., Human Error in Medicine, Lawrence Erlbaum Associates, Hillsdale, NJ, 1994.

1999 by CRC Press LLC

29. Maddox, M.E., Designing medical devices to minimize human error, Medical Device
and Diagnostic Industry Magazine, 19(5), 166-180, 1997.
30. Sawyer, D., Do It By Design: Introduction to Human Factors in Medical Devices,
Center for Devices and Radiological Health (CDRH), Food and Drug Administration
(FDA), Washington, D.C., 1996.
31. Nobel, J.L., Medical devices failures and adverse effects, Pediat-Emerg. Care, 7, 120123, 1991.
32. Bogner, M.S., Medical devices and human error, in Human Performance in Automated
Systems: Current Research and Trends, Mouloua, M. and Parasuraman, R., Eds.,
Lawrence Erlbaum Associates, Hillsdale, NJ, 1994, pp. 64-67.
33. Casey, S., Set Phasers on Stun: and Other True Tales of Design Technology and
Human Error, Aegean, Inc., Santa Barbara, CA, 1993.
34. McDonald, J.S. and Peterson, S., Lethal errors in anaesthesiology, Anesthesiol, 63,
A 497, 1985.
35. Wood, B.J. and Ermes, J.W., Applying hazard analysis to medical devices, Part II:
Detailed hazard analysis, Medical Device and Diagnostic Industry Magazine, 15(3),
58-64, 1993.
36. Weide, P., Improving medical device safety with automated software testing, Medical
Device and Diagnostic Industry Magazine, 16(8), 66-79, 1994.
37. Levkoff, B., Increasing safety in medical device software, Medical Device and Diagnostic Industry Magazine, 18(9), 92-101, 1996.
38. Federal Food, Drug, and Cosmetic Act, as Amended, Sec. 201 (h), U.S. Government
Printing Office, Washington, D.C., 1993.
39. Onel, S., Draft revision of FDAs medical device software policy raises warning flags,
Medical Device and Diagnostic Industry Magazine, 19(10), 82-92, 1997.
40. Mojdehbakhsh, R., Tsai, W., and Kirani, S., Retrofitting software safety in an implantable medical device, Trans. Software, 11, 41-50, 1994.
41. Heydrick, L. and Jones, K.A., Applying reliability engineering during product development, Medical Device and Diagnostic Industry Magazine, 18(4), 80-84, 1996.
42. IEEE-STD-1228, Standard for Software Safety Plans, Institute of Electrical and
Electronic Engineers (IEEE), New York, 1994.
43. Cady, W.W. and Iampietro, D., Medical device reporting, Medical Device and Diagnostic Industry Magazine, 18(5), 58-67, 1996.
44. Thibeault, A., Documenting a failure investigation, Medical Device and Diagnostic
Industry Magazine, 19(10), 68-74, 1997.
45. Rose, H.B., A small instrument manufacturers experience with medical instrument
reliability, Proc. Annu. Reliability Maintainability Symp., 251-254, 1972.
46. Taylor, E.F., The reliability engineer in the health care system, Proc. Reliability and
Maintainability Symp., 245-248, 1972.
47. Bell, D.D., Contrasting the medical-device and aerospace-industries approach to
reliability, Proc. Annu. Reliability Maintainability Symp., 125-127, 1995.
48. MIL-STD-2164, Environment Stress Screening Process for Electronic Equipment,
Department of Defense, Washington, D.C.
49. MIL-STD-883, Test Methods and Procedures for Microelectronics, Department of
Defense, Washington, D.C.
50. MIL-HDBK-217, Reliability Prediction of Electronic Equipment, Department of
Defense, Washington, D.C.
51. Reliability Prediction Procedure for Electronic Equipment, Report No. TR-NWT000332, issue 4, Bell Communication Research, New Jersey, 1992.

1999 by CRC Press LLC

52. MIL-HDBK-251, Reliability/Design Thermal Applications, Department of Defense,

Washington, D.C.
53. MIL-HDBK-338, Electronic Reliability Design Handbook, Department of Defense,
Washington, D.C.
54. MIL-STD-781, Reliability Design Qualification and Production Acceptance Tests:
Exponential Distribution, Department of Defense, Washington, D.C.
55. MIL-HDBK-781, Reliability Test Methods, Plans and Environments for Engineering
Development, Qualification and Production, Department of Defense, Washington,
D.C.

1999 by CRC Press LLC

15

Design Maintainability
and Reliability
Centered Maintenance

15.1 INTRODUCTION
Maintainability is a design and installation characteristic that imparts to an item a
greater inherent ability to be maintained, and consequently results in factors such
as a decrease in required maintainability, tools, skill levels, facilities, and manhours.
The history of maintainability can be traced back to 1901 in the Army Signal
Corps Contract with the Wright brothers to develop an airplane. This document
contained a clause that the aircraft under consideration be simple to operate and
maintain [1]. Various studies conducted during the period between World War II
and the early 1950s by the U.S. Department of Defense indicated startling results
concerning the state of reliability and maintainability of equipment used by the three
services [2, 3]. For example, an Army study revealed that approximately between
two-thirds and three-fourths of its equipment were either under repair or out of
service.
In 1956, a series of articles appeared in Machine Design covering areas such as
design electronic equipment for maintainability, recommendations for designing
maintenance access in electronic equipment, design recommendations for test points,
and factors to consider in designing displays [4]. In 1959, the U.S. Department of
Defense released a document (i.e., MIL-M-26512) on maintainability specification.
In fact, an appendix to this document contained an approach for planning a maintainability demonstration test.
In 1960, the first commercially available book entitled Electronic Maintainability
appeared [5]. Since the 1960s a large number of publications on various aspects of
maintainability has appeared [6, 7].
Reliability centered maintenance (RCM) is a systematic methodology employed
to highlight the preventive maintenance tasks necessary to realize the inherent
reliability of an item at the lowest resource expenditure. As per the published sources,
the original development of the RCM concept could be traced back to the late 1960s
in the commercial aircraft industry. For example, in 1968 a document (MSG 1)
entitled Maintenance Evaluation and Program Development was prepared by the
U.S. Air Transport Association (ATA) for use with the Boeing 747 aircraft [8]. At a
later date, MSG 1 was revised to handle two other wide-body aircrafts: L-1011 and
DC-10. The revised version became known as MSG 2 [9].
In 1974, the U.S. Department of Defense commissioned United Airlines to
prepare a document on processes used by the civil aviation industrial sector to

1999 by CRC Press LLC

develop maintenance programs for aircraft [10]. The resulting document was entitled
Reliability Centered Maintenance by United Airlines.
MSG 2 was revised by the ATA in 1980 to include maintenance programs for
two more aircraft: Boeing 756 and 767. The revised version was named MSG 3 [11].
In Europe, MSG 3 became known as European MSG 3 because it included
maintenance programs for two more aircraft: A-300 and Concorde [12]. In the 1970s,
U.S. forces were attracted by the RCM methodology and in 1985 the military
published two documents entitled Guide to Reliability Centered Maintenance [13]
and Reliability Centered Maintenance for Aircraft, Engines, and Equipment [14].
Today, RCM methodology is actively being practiced in many different parts of the
world.
This chapter describes design maintainability and RCM.

15.2 DESIGN MAINTAINABILITY: MAINTAINABILITY

NEED AND RELATIONSHIP BETWEEN
RELIABILITY AND MAINTAINABILITY
There are several factors responsible for the need of product maintainability. In
particular, two important factors are the alarmingly high operating and support costs
because of failures and subsequent maintenance. For example, one study conducted
by the U.S. Air Force in the 1950s reported that one third of all Air Force operating
cost was due to maintenance, and one third of Air Force manpower was directly or
indirectly associated with maintenance [15]. Another study performed by the Army
reported that 60 to 75% of the electronic equipment total life cycle cost was due to
its operation and maintenance [16, 17].
The main objectives of both reliability and maintainability include: to assure
that the system/equipment manufactured will be in the readiness state for operation
when required, capable of performing effectively its designated functions, and meeting all the required maintenance characteristics during its life time.
In order to understand the relationship between reliability and maintainability,
some of the important aspects of both reliability and maintainability are described
below.

RELIABILITY
This is a design characteristic that results in durability of the system/equipment to
carry out its designated function under a stated condition and time period. It is
accomplished through various actions including controlling processes, selecting
optimum engineering principles, testing, and adequate component sizing. Nonetheless, there are many specific general principles of reliability: design for simplicity,
use less number of parts to perform multiple functions, design to minimize the
occurrence of failures, maximize the use of standard parts, provide fail safe designs,
use parts with proven reliability, minimize stress on parts, provide satisfactory safety
factors between strength and peak stress values, provide redundancy when required,
and provide for simple periodic adjustment of parts subject to wear [18].

1999 by CRC Press LLC

MAINTAINABILITY
This is a built-in design and installation characteristic that provides the end product/equipment an inherent ability to be maintained, thus ultimately resulting in lower
maintenance cost, required skill levels, man-hours required, required tools and equipment, and improved mission availability.
Some of the specific general principles of maintainability are reduce mean time
to repair (MTTR), lower life cycle maintenance costs, provide for maximum interchangeability, consider benefits of modular replacement vs. part repair or throwaway
design, reduce or eliminate altogether the need for maintenance, establish the extent
of preventive maintenance to be performed, reduce the amount, frequency, and
complexity of required maintenance tasks, and lower amount of supply supports
required [18].

15.2.1 MAINTAINABILITY AND AVAILABILITY ANALYSES DURING

DESIGN AND DEVELOPMENT PHASE AND MAINTAINABILITY
DESIGN CHARACTERISTICS
There are various types of maintainability and availability analyses performed during
a products design and development phase. Such analyses serve a number of purposes
including evaluating the design performance with respect to both qualitative and
quantitative requirements, developing most appropriate ways and means to reduce
the need for maintenance, generating maintainability related data for application in
maintenance planning and in conducting support analysis, and quantifying maintainability requirements at the design level [19].
There are a number of maintainability-related basic characteristics that must be
carefully considered during the product design phase. The maintainability design
characteristics include those features/factors that help to lower product downtime
and unavailability (e.g., maintenance ease). Nonetheless, some of the particular
features that affect the maintainability aspects of engineering design include standardization, safety, interchangeability, equipment packaging, and human factors.
Each of these items is described below [1, 20].
Standardization
This is basically concerned with imposing limitations on the variety of different
items that can be used to satisfy the product requirements. Some of the specific
goals of the standardization are to minimize the use of different types of parts, use
the maximum number of common parts in different products, and maximize the use
of interchangeable parts. The benefits of standardization include reduction in item
life-cycle acquisition and support costs and improvement in item reliability and
maintainability.
Safety
This is an important maintainability design factor because individuals performing
various types of maintenance tasks may be exposed to hazards and accident situations.

1999 by CRC Press LLC

Hazards and accidents could be the result of unsatisfactory attention given to safety
during the design phase. Nonetheless, the human safety guidelines include installing
appropriate fail-safe devices, carefully studying the potential sources of injury by
electric shock, fitting all access openings with appropriate fillets, providing appropriate emergency doors, installing items requiring maintenance such that hazard in
accessing them is minimized, and providing adequate amount of tracks, guides, and
stops to facilitate equipment handling.
Interchangeability
It means that a given component can be replaced by any similar part, and the
replacing part can carry out the required functions of the replaced part effectively.
There are various factors that must be considered carefully in determining needs for
interchangeability including cost effectiveness of manufacture and inspection, and
field conditions. Maximum interchangeability can only be assured if the design
professionals consider factors such as providing adequate level of information in the
task instructions, physical similarities (include shape, size, mounting, etc.), and nonexistence of physical interchangeability in places where functional interchangeability
is not expected.
Equipment Packaging
This is an important maintainability factor and is basically concerned with the
manner in which equipment is packaged: ease of parts removal, item layout, access,
and so on. In order to achieve the effectiveness of equipment packaging, careful
attention must be given to factors such as accessibility requirements, modularization
needs, manufacturing requirements, reliability and safety factors, environmental
factors, and standardization needs.
In particular, modularization and accessibility are very important and are
described in more detail. Modularization is the division of a product into distinct
physical and functional units to assist removal and replacement. The advantages of
modular construction include reduction in maintenance time and cost, usually easily
maintainable product, shorter design times because of simplified design, and requires
relatively low skill levels and fewer maintenance tools.
On the other hand, accessibility is the relative ease with which an item can be
reached for actions such as repair, inspection, or replacement. The accessibility is
affected by factors such as types of tools required, access usage frequency, type of
task to be performed, hazards involved with respect to access usage, and distance
to be reached.
Human Factors
As maintainability depends on both the maintainer and the operator, human factors
are very important. For example, one aircraft maintenance study reported that over
a period of 15 months, human error resulted in 475 accidents and incidents in flight
and ground operations [1, 20]. Furthermore, most of the accidents happened shortly

1999 by CRC Press LLC

after periodic inspections and 95 aircraft were damaged or destroyed with a loss of
14 lives. In human factors, environment plays an important role since it can vary
quite considerably from one application to another. Environment can be classified
into three categories: physical, operational, and human. The physical category
includes factors such as temperature, noise, vibration, radiation, pressure, and dust.
The components of the operational category are work duration, illumination, maintenance workspace arrangement, acoustics, ventilation, etc. The factors belonging
to the human category include physical, psychological, human limitations, and
physiological.

15.2.2 DESIGN REVIEWS

Design reviews are an important aspect of equipment development and they allow
parties involved in the development process to assess factors such as the adequacy
of program planning and the maturity of the design effort. There are various types
of design reviews but, in particular, two important reviews are preliminary design
review and critical design review [19]. The preliminary design review is performed
after the availability of hardware development specifications and is a formal review
of the basic design approach for a system or a functionally related group of lowerlevel components. The issues addressed during the preliminary design reviews from
the maintainability aspect include quantitative maintainability requirements, repair
time data sources, repair rate prediction methods, system/product maintainability
characteristics, conformance to the maintainability ground rules and design criteria,
plans for maintainability demonstrations, preventive maintenance schedules, and
comparison between quantitative maintainability requirements and available preliminary predictions.
The critical design review is a formal review for a system/functionally related
group of lower-level elements or parts, performed before releasing fabrication or
production blue prints to ensure that detailed design solutions effectively meet the
requirements established by the hardware development specification. The maintainability related issues addressed during this review include quantitative maintainability requirements and their comparison with available preliminary predictions, details
of plans for maintainability demonstrations, methods for automatic, semiautomatic,
and manual recovery from failure, conformance to the maintainability ground rules
and design criteria, preventive-maintenance schedules for compatibility with system
requirements, and system conformance to the planned maintenance concept and
unique maintenance procedures (if any) [19].
On all of the above issues, various questions are raised during design reviews.
One typical example of such questions is as follows:
What are the major failure modes of the item under consideration and
how will they affect the system operation? Two examples of the item are
a replaceable unit and a module.
In addition, for each failure mode, questions on areas such as those listed below
could be asked.

1999 by CRC Press LLC

Mechanism for operator or user to know if the item has failed.

Mechanism employed by the built-in test system to identify failure.
Remove and replace sequence for the failure.
Training or experience necessary for performing the maintenance action.
Required tools or equipment.
Hazards involved in performing the maintenance action.
Actions for reducing or eliminating failure mode frequency.
Measures to reduce the maintenance time needed to accomplish the maintenance action.

15.2.3 MAINTAINABILITY MEASURES

As quantitative maintainability specifications are based on desired limiting conditions imposed on product/equipment down time, maintenance labor-hours, etc., the
knowledge of various maintainability parameters during the design phase is important for an effective end product maintainability. Some of the typical maintainability
measures are mean time to repair (MTTR), medium corrective maintenance time,
maximum corrective maintenance time, mean preventive maintenance time, and the
probability of completing repair in a given time interval (i.e., the maintainability
function). Each of these measures is presented below.
Mean Time to Repair
In maintainability analysis, this is probably the most widely used maintainability
measure or parameter. MTTR is also referred to as mean corrective maintenance
time and a typical corrective maintenance or repair cycle is made up of steps such
as fault/failure detection, fault/failure isolation, disassemble to gain access, repair,
and reassemble. Usually probability distributions such as exponential, log-normal,
and normal are used to represent corrective maintenance times. In particular, the
exponential distribution is assumed for electronic equipment having an effective
built-in test capability along with a rapid remove and replace maintenance concept.
However, it must be remembered that the exponential assumption for corrective
maintenance times may result in incorrect conclusions as most repair actions consume some repair time.
The lognormal distribution is frequently used to represent corrective maintenance times associated with electronic equipment without having a built-in test
capability. In addition, it can also be used for electromechanical systems possessing
widely individual repair times.
The normal distribution is often assumed for mechanical/electro-mechanical
equipment with a remove and replace maintenance concept.
The system MTTR is expressed by

MTTR =

1999 by CRC Press LLC

j =1

j Tj

j =1

(15.1)

where
Tj is the corrective maintenance or repair time required to repair unit j; for
j = 1, 2, 3, , k.
k is the total number of units.
j is the constant failure rate of unit j; for j = 1, 2, 3, , k.
Example 15.1
An engineering system is made up of six replaceable subsystems 1, 2, 3, 4, 5, and
6 with constant failure rates 1 = 0.0002 failures/h, 2 = 0.0002 failures/h, 3 =
0.0001 failures/h, 4 = 0.0004 failures/h, 5 = 0.0005 failures/h, and 6 = 0.0006
failures/h, respectively. Corrective maintenance times associated with subsystems 1,
2, 3, 4, 5, and 6 are T1 = 1 h, T2 = 1 h, T3 = 0.5 h, T4 = 2 h, T5 = 3 h, and T6 = 3.5 h,
respectively. Estimate the system MTTR.
Inserting the given values into Equation (15.1) yields
MTTR =

(0.0002)(1) + (0.0002)(1) + (0.0001) (0.5) + (0.0004)(2)

(0.0002) + (0.0002) + (0.0001) + (0.0004)
+ (0.0005) (3) + (0.0006) (3.5)
+ (0.0005) + (0.0006)

= 2.425 h
Thus, the system mean time to repair is 2.425 h.
Median Corrective Maintenance Time
It is the time within which 50% of all corrective maintenance actions can be
performed. The measure is dependent upon the probability density function describing the times to repair (e.g., exponential, lognormal). Thus, the median corrective
maintenance time (MCMT) for the exponential distribution is expressed by
MCMT = 0.69

(15.2)

where
is the constant repair rate, thus the reciprocal of the MTTR.
Similarly, the MCMT for the lognormal distribution is given by

MCMT = MTTR exp 2 2

(15.3)

where
2 is the variance around the mean value of the natural logarithm of repair
times.

1999 by CRC Press LLC

Maximum Corrective Maintenance Time

This is another maintainability measure and is used to estimate the time to accomplish a specified percentage of all potential repair actions. The most commonly
specified percentiles are 90th and 95th. For example, the 90th percentile means that
no greater than 10% of the maintenance activities may exceed the maximum corrective maintenance time.
The estimation of maximum corrective maintenance time depends on the probability distribution representing the times to repair. The maximum corrective maintenance times for exponential, normal, and lognormal distributions are given below.
Exponential
The maximum corrective maintenance time is defined by
Tmax = C (MTTR)

(15.4)

where
Tmax is the maximum corrective maintenance time.
C is equal to 2.312 or 3 for the 90th and 95th percentiles, respectively.
Normal
The maximum corrective maintenance time is expressed by
Tmax = MTTR + m n

(15.5)

where
m is equal to 1.28 or 1.65 for the 90th and 95th percentiles, respectively.
n is the standard deviation of the normally distributed maintenance time.
Lognormal
The maximum corrective maintenance time is given by
Tmax = antilog (t a + m l )

(15.6)

where
ta is the average of the logarithms of the repair times.
l is the standard deviation of the logarithms of the repair times.
Mean Preventive Maintenance Time
Various preventive maintenance related activities such as inspections, calibrations,
and tuning are performed to keep the item or equipment at a specified performance

1999 by CRC Press LLC

level. A well-planned and well-executed preventive maintenance program can be

quite useful to reduce equipment/item downtime and improve its performance.
The mean preventive maintenance time is expressed by

MPMT =

j =1

Tpj fpj

pj

(15.7)

j =1

where
MPMT
n
fpj
Tpj

is the mean preventive maintenance time.

is the total number of preventive maintenance tasks.
is the frequency of preventive maintenance task j; for j = 1, 2, 3, , n.
is the elapsed time for preventive maintenance task j; for j = 1, 2, 3,
, n.

In computing the value of MPMT, it is to be remembered that if the frequencies,

fpj, are given in maintenance tasks per hour, then the Tpj must be expressed in hours.
Maintainability Function
This is used to calculate probability of completing repair in a specified time interval.
After the identification of the repair time distributions, the corresponding maintainability functions can be obtained by using the following relationship [1]:
t

M (t ) = frp ( t ) d t

(15.8)

where
t
is the time.
M (t) is the maintainability function.
frp (t) is the probability density function of the repair times.
Maintainability functions for exponential, lognormal, and normal repair time distribution are obtained below.
Exponential
The probability density function representing corrective maintenance times is defined
by
frp ( t ) = exp ( t )
where
t is the variable repair time.
is the constant repair rate or the reciprocal of the MTTR.

1999 by CRC Press LLC

(15.9)

Inserting Equation (15.9) into relationship (15.8) yields

t

M (t ) = exp ( t ) d t

(15.10)

= 1 exp ( t )
Since = 1/MTTR, Equation (15.10) becomes
M (t ) = 1 exp ( t MTTR)

(15.11)

Example 15.2
After performing analysis of repair actions associated with an electric generator, it
is established that the generators repair times are exponentially distributed with a
mean value of 4 h. Calculate the probability of accomplishing a repair in 5 h.
By inserting the given data into Equation (15.11), we get
M (5) = 1 exp ( 5 4)
= 0.7135
It means that there is an approximately 72% chance that the repair will be
completed within 5 h.
Lognormal
This is a widely used probability distribution in maintainability work and its probability density function representing corrective maintenance times is given by
frp (t ) =

(t )

2
1
exp {ln (t ) }
2

(15.12)

where
is a constant denoting the shortest time below which no maintenance
activity can be performed.
is the mean of the natural logarithms of the maintenance times.
is the standard deviation with which the natural logarithms of the maintenance times are spread around the mean .
The following relationship defines the mean:

= ln t1 + ln t 2 + ln t 3 + + ln t k

1999 by CRC Press LLC

(15.13)

where
tj is the maintenance time j; for j = 1, 2, 3, , k.
k is the number of maintenance times.
The standard deviation, , is expressed by the following relationship:

(ln t )
j

j =1

( k 1)

12

(15.14)

The maintainability function, M(t), for this distribution is expressed by

M (t ) =

tf

rp

(t ) d t

(15.15)

1 ln t
dt

exp 2

1
2

Normal
Corrective maintenance times can also be represented by the normal distribution.
The distribution repair time probability density function is defined by
frp ( t ) =

1 t 2
1
exp

2
2

(15.16)

where
is the mean of maintenance times
is the standard deviation of the variable maintenance time t around the
mean .
By substituting Equation (15.16) into Equation (15.8), we get
1
M (t ) =
2

1 t 2
exp
dt
2

(15.17)

The mean of the maintenance times is expressed by

k

t
j =1

1999 by CRC Press LLC

(15.18)

where
tj is the maintenance time j; for j = 1, 2, 3, , k.
k is the total number of maintenance times.
The standard deviation is

(t )
j

j =1

( k 1)

15.2.4 MAINTAINABILITY DESIGN GUIDELINES

MAINTAINABILITY DESIGN ERRORS

12

AND

(15.19)

COMMON

Over the years professionals involved with maintainability have developed many
guidelines that can be used to produce effective design with respect to maintainability. These guidelines include design for minimum maintenance skills, provide for
visual inspection, use standard interchangeable parts, design for minimum tools and
adjustments, avoid the use of large cable connectors, design for safety, provide test
points, use color-coding, label units, provide handles on heavy parts for the ease of
handling, use captive-type chassis fasteners, provide troubleshooting techniques,
group subsystems, and use plug-in rather than solder-in modules [21].
During the design phase of an equipment, there are many errors committed that
adversely affect its maintainability. Some of the common design errors are placing
an adjustment out of arms reach, omitting handles, using access doors with numerous small screws, placing removable parts such that they cannot be dismantled
without taking the entire unit from its case, locating adjustable screws close to a hot
component/an exposed power supply terminal, placing adjustable screws in locations
cumbersome for maintenance personnel to discover, providing unreliable built-in
test equipment, locating fragile parts just inside the bottom edge of the chassis where
the maintenance personnel are expected to place their hands, using chassis and cover
plates that drop when the last screw is taken out, placing low-reliability parts beneath
other parts, placing screwdriver adjustments beneath modules, and providing insufficient space for maintenance personnel to get their gloved hands into the unit to
perform required adjustment [21].

15.3 RELIABILITY CENTERED MAINTENANCE

The commercial airline industry developed the reliability centered maintenance
(RCM) concept after becoming fully aware that the traditional approach to maintenance with its heavy emphasis on scheduled inspection, servicing, and removal was
not impacting the equipment reliability as per expectations. In fact, it became very
apparent after the introduction of more complex aircraft systems that the relationship
between reliability and preventive maintenance tended to disappear [12]. RCM is
based on the premise that the inherent reliability or safety of a system cannot be

1999 by CRC Press LLC

improved through maintenance and good maintenance practices can only preserve
such characteristics. The RCM philosophy calls for the performance of scheduled
maintenance on critical components only under the following circumstances:
It will stop a decrease in reliability and/or degradation of safety to unacceptable levels or
It will decrease equipment life cycle cost.
In contrast, the RCM philosophy also calls for not performing scheduled maintenance on noncritical items unless it will lower equipment life cycle cost.

15.3.1 TRADITIONAL MAINTENANCE, RCM,

FOR SELECTING ASSETS FOR RCM

AND

QUESTIONS

In order to provide a better understanding of RCM, here we briefly describe both

traditional maintenance and RCM. Websters Encyclopaedic Dictionary defines the
words maintain and maintenance, respectively, as follows [22]:
To cause to remain unaltered or unimpaired.
A maintaining or being maintained.
In contrast, RCM may be described as a process employed to determine the maintenance needs of any physical facility in its operating context [10].
Prior to the start of performing analysis of maintenance needs of the facilities,
it is essential to fully know about these facilities and to select the ones to be subjected
to the RCM review process. Nonetheless, the RCM process entails asking various
questions concerning the chosen facilities on areas such as functions and associated
performance standards of the facility in its current operating context, the ways the
facility fails to meet its required functions, each functional failures causes, consequences of each failure, ways each failure matter, preventive measures to stop each
failure, and measures to be taken if an appropriate preventive task cannot be found.

15.3.2 RCM PROCESS RELATED MAINTENANCE TASK

CLASSIFICATION AREAS AND STEPS
The RCM process calls for maintenance tasks to be categorized into three distinct
areas: hard-time maintenance, on-condition maintenance, and condition monitoring
[23].
The hard-time maintenance is for those failure modes that need scheduled
maintenance at pre-determined time or usage intervals.
The on-condition maintenance is for those failure modes that need scheduled
inspections or tests for measuring the level of item deterioration so that on the basis
of this deterioration level either corrective maintenance can be carried out or the
item can still remain in service.
The condition monitoring is for those failure modes that need unscheduled tests
or inspection on parts where it is possible to tolerate a failure during system operation

1999 by CRC Press LLC

or where it is possible to detect an impending failure through the means of regular

monitoring during normal operations.
The RCM process is made up of a number of steps and each such step is an
integral part of RCM engineering planning and analysis. Nonetheless, the application
of this process as well as the resulting preventive maintenance program ensure that
systems or equipment are kept in good working condition and, consequently, are
quite effective to support their operational requirements. However, to attain such
results, it is absolutely necessary that the RCM process be initially practiced during
equipment design and development phase to develop maintenance plans and procedures, and continued during the operational phase utilizing actual experience data
to make changes, as appropriate, to the maintenance plans.
The seven basic steps of the RCM process are as follows [23]:
1.
2.
3.
4.
5.
6.
7.

Determine maintenance-important items.

Collect necessary failure data.
Conduct fault tree analysis.
Use decision logic to important failure modes.
Record maintenance categories.
Implement RCM decisions.
Apply sustaining engineering based on real-life experience information.

Each of the above steps is described below.

Determine Maintenance-Important Items
Traditionally to identify maintenance-important items the failure mode, effects, and
criticality analysis (FMECA) approach has been used.
Also, the approach provided the basis for developing detailed preventive/corrective maintenance requirements to be used during the system/equipment operation
phase. However, for more effective results, a better method is required that uses the
results of data collection programs and field operating experience to develop new
and upgrade existing maintenance programs. Fault tree analysis (FTA) is considered
to be a quite effective approach for this purpose because it can use field failure data
to determine maintenance-importance items.
Collect Necessary Failure Data
This step is concerned with obtaining required data for each basic fault associated
with the fault tree. For assessing criticality and determining occurrence probabilities,
the essential inputs include part failure rate, operator error probability, and inspection
efficiency data. Some of the sources of obtaining part failure rate data are the actual
real life experience, MIL-HDBK-217 [24], and generic failure data banks. Many of
the sources for obtaining human error related data are listed in Reference 25.
Conduct Fault Tree Analysis
This step is concerned with calculating the occurrence probabilities of all fault events
(i.e., basic, intermediate, and top) on the basis of combinatorial properties of the

1999 by CRC Press LLC

logic elements in the fault tree. The sensitivities (sensitivity or conditional probability
may be defined as the probability that an occurrence of a basic fault event will lead
to a safety incident) are calculated by assigning a probability of unity to an elementary or basic fault event and then determining a safety incidents resultant probability.
In turn, the criticality of each basic fault event is calculated by using these sensitivity
results.
Criticality may simply be described as a measure of the relative seriousness or
impact of each and every fault event on the undesired or top fault event. It (i.e.,
criticality) involves both qualitative and quantitative fault tree analyses and provides
a base mechanism to rank the fault events in their order of severity.
In quantitative terms, the criticality is expressed below [26].
C = P ( y) P ( F y)

(15.20)

where
P(y) is the probability of occurrence of the basic fault event, y.
P(F  y) is the sensitivity or conditional probability that an occurrence of a
basic fault event will lead to a safety incident.
C
is the criticality.
Use Decision Logic to Important Failure Modes
This step is concerned with the RCM decision logic designed to result in, by asking
standard assessment questions, the most effective preventive maintenance task combinations. Each and every critical failure mode highlighted by the FTA is reviewed
by applying the decision logic for each maintenance-important item. Judgments are
passed as to the need for various maintenance tasks as each and every failure mode
is processed. The tasks considered necessary along with the intervals established to
be appropriate, form the overall scheduled preventive maintenance program.
The decision logic is composed of the following two levels:
Level 1. This is concerned with the evaluation of each failure mode for
determining consequence category: hidden safety, evident safety, operational economic, non-operational economic, or non-safety/economic. Four
questions associated with this level are as follows:
1. Does the failure or fault cause a direct adverse effect on operating
performance?
2. Does the failure/secondary failure lead to a safety incident?
3. Can operator(s) detect failures or faults when they occur?
4. Does the hidden failure itself, or the combination of a hidden failure
plus an additional failure of a system-related or back-up function, lead
to a safety incident?
Level 2. This is concerned with taking into consideration the failure causes
for each failure mode for choosing the specific type of task. Twenty-one

1999 by CRC Press LLC

questions are asked in this level [23]. Some examples of those questions
are as follows:
Is an operator monitoring task effective and applicable?
Is a check for verifying operation/detecting impending failure effective
and applicable?
Is a servicing task effective and applicable?
Record Maintenance Categories
This step is concerned with applying the decision logic of the previous step to group
maintenance requirements into three categories (i.e., hard-time maintenance requirements, on-condition maintenance requirements, and condition monitoring maintenance requirements) and defining a maintenance task profile. This task profile basically highlights part number and failure mode, and the preventive-maintenance taskselection for RCM logic questions with answer yes. Subsequently, the maintenance-task profile is used for determining applicable preventive maintenance tasks
to each part under consideration.
Implement RCM Decisions
Subsequent to the establishment of the maintenance-task profile, the task frequencies/intervals are established and enacted as part of the overall maintenance plan.
Apply Sustaining Engineering Based on Real-life
Experience Information
As the RCM process has a life-cycle perspective, the main objective in this step is
to reduce scheduled maintenance burden and support cost while keeping the equipment/systems in a desirable operational-readiness state. After the system is operational and field data start to accumulate, one of the most pressing steps is to reassess
all RCM default decisions.
All in all, it must be remembered that the key objective is to eliminate all
excessive maintenance related costs while maintaining established and required
reliability and safety levels.

15.3.3 RCM ADVANTAGES

Over the years many companies have applied the RCM methodology and have
experienced many advantages to a varying degree. Some of these advantages are as
follows [10, 23]:
Better safety and environmental integrity. Prior to considering the
impact of each failure mode on operations, the RCM methodology first
considers its effects on safety and environments. It means appropriate
steps are taken to minimize or eliminate all identifiable equipment-associated safety and environmental hazards.

1999 by CRC Press LLC

 Improved motivation of individuals. The RCM effort is especially useful

to improve the motivation of the individuals involved in the review process. This occurs because of factors such as clearer understanding of the
functions of the facility under consideration, clear general understanding
of the issues involved, and the knowledge that each group member played
a role in formulating goals.
Better operating performance. As the RCM philosophy dictates that all
types of maintenance have value to a certain degree, thus it (RCM) outlines
rules for deciding which is the most effective in every circumstance.
Consequently, the most effective form of maintenance is chosen for each
machine and appropriate actions are pursued in cases where maintenance
can be effective.
Improved maintenance cost-effectiveness. In most of the industrial sector, the maintenance activity is the third biggest component of operating
costs. Thus, controlling maintenance costs is very important. The application of RCM helps to lower these costs in several ways including less
routine maintenance, less need to acquire services of costly experts, clearer
guidelines for acquiring new maintenance technology, and better buying
of maintenance services.
Improved team effort. The application of the RCM methodology helps
to provide a common and easily comprehensible language for all individuals involved with the maintenance effort. This, in turn, not only fosters
teamwork within the review groups themselves but also improves communication and cooperation among production and maintenance departments,
management, repair persons, operators, designers, vendors, and so on.
Better maintenance database. RCM information and decision worksheets are a comprehensive maintenance database, which in turn has
advantages such as reducing the effects of staff turnover, more accurate
drawings and manuals, and adapting to changing circumstances.
Improved useful life of costly items.

15.3.4 REASONS FOR THE RCM METHODOLOGY

APPLICATION FAILURES
When the RCM methodology is applied correctly and effectively, it can provide fast
results and associated benefits. However, its every application may not yield its full
potential. In fact, some of the applications may only achieve very little or nothing
at all. Some of the major possible reasons for the RCM methodology application
failures are too hurried or too superficial application, too much emphasis on failure
data (e.g., on parameters such as mean time between failures (MTBF) and mean
time between to repair (MTTR), and the analysis carried out at too low a level [10].

1999 by CRC Press LLC

15.4

PROBLEMS

1. Define the term Reliability Centered Maintenance and discuss its historical developments.
2. Discuss the following terms:
Maintainability
Interchangeability
Standardization
3. Write an essay on design maintainability.
4. What are the important factors that affect the maintainability aspects of
an equipment design?
5. What are the issues addressed during the preliminary design review of an
equipment from the stand point of maintainability?
6. An electronic system is composed of four subsystems 1, 2, 3, and 4. The
time to failure of each subsystem is exponentially distributed. The failure
rates of subsystems 1, 2, 3, and 4 are 1 = 0.01 failure/h, 2 = 0.03 failure/h,
3 = 0.04 failure/h, and 4 = 0.05 failure/h, respectively. In addition, the
corrective maintenance times associated with subsystems 1, 2, 3, and 4
are T1 = 2 h, T2 = 4 h, T3 = 1 h, and T4 = 3 h, respectively. Calculate the
electronic system mean time to repair.
7. Obtain a maintainability function when the probability density function
of the repair times is described by the Weibull distribution.
8. Compare traditional maintenance with RCM.
9. Describe the RCM methodology process.
10. What are the benefits of applying the RCM methodology?

15.5 REFERENCES
1. AMCP 706-133, 1976, Engineering Design Handbook: Maintainability Engineering
Theory and Practice, prepared by the Army Material Command, Department of the
Army, Washington, D.C.
2. Shooman, M.L., Probabilistic Reliability: An Engineering Approach, McGraw-Hill,
New York, 1968.
3. Moss, M.A., Minimal Maintenance Expense, Marcel Dekker, New York, 1985.
4. Retterer, B.L. and Kowalski, R.A., Maintainability: A historical perspective, IEEE
Trans. Reliability, 33, 56-61, 1984.
5. Akenbrandt, F.L., Ed., Electronic Maintainability, Engineering Publishers, Elizabeth,
NJ, 1960.
6. Dhillon, B.S., Reliability Engineering in Systems Design and Operation, Van Nostrand Reinhold Company, New York, 1983.
7. Dhillon, B.S., Reliability and Quality Control: Bibliography on General and Specialized Areas, Beta Publishers, Gloucester, Canada, 1993.
8. MSG 1, Maintenance Evaluation and Program Development, 747 Maintenance Steering Group Handbook, Air Transport Association, Washington, D.C., July 1968.
9. MSG 2, Airline/Manufacturer Maintenance Program Planning Document, Air Transport Association, Washington, D.C., September 1980.
10. Moubray, J., Reliability Centered Maintenance, Industrial Press, New York, 1992.

1999 by CRC Press LLC

11. MSG 3, Airline/Manufacturer Maintenance Program Planning Document, Air Transport Association, Washington, D.C., September 1980.
12. Anderson, R.T., Reliability Centered Maintenance: Management and Engineering
Methods, Elsevier Applied Science Publishers, London, 1990.
13. U.S. AMC Pamphlet 750-2, Guide to Reliability Centered Maintenance, U.S. Army,
Department of Defense, Washington, D.C., 1985.
14. MIL-STD-1843, Reliability Centered Maintenance for Aircraft, Engines, and Equipment, Department of Defense, Washington, D.C., 1985.
15. Hall, A.C., Relationship between operational effectiveness of electronic systems and
maintenance minimization and training, Proc. Symp. Electron. Mainten., Washington,
D.C., 1955.
16. AMC Memo AMCRD-ES, Subject: Economy vs. Life Cycle Cost, signed by Maj.
Gen. Guthrie, J.R., Director, Research, Development, and Engineering, Headquarters,
Washington, D.C., January 21, 1971.
17. Ankenbrandt, F.L., Ed., Electronic Maintainability, Vol. 3, Engineering Publishers,
Elizabeth, NJ, 1960.
18. AMCP-706-134, Maintainability Guide for Design, prepared by the Department of
the Army, Department of Defense, Washington, D.C., 1972.
19. Grant Ireson, W., Coombs, C.F., and Moss, R.Y., Handbook of Reliability Engineering
and Management, McGraw-Hill Companies, New York, 1996.
20. Dhillon, B.S., Engineering Design: A Modern Approach, Richard D. Irwin, Chicago,
IL, 1996.
21. Pecht, M., Ed., Product Reliability, Maintainability, and Supportability Handbook,
CRC Press, Boca Raton, FL, 1995.
22. Websters Encyclopaedic Dictionary, Lexicon Publication, New York, 1988.
23. Brauer, D.C. and Brauer, G.D., Reliability centered maintenance, IEEE Trans. Reliability, 36, 17-24, 1987.
24. MIL-HDBK-217, Reliability Prediction of Electronic Equipment, Department of
Defense, Washington, D.C.
25. Dhillon, B.S., Human Reliability: With Human Factors, Pergamon Press, New York,
1986.
26. NURG-0492, Fault Tree Handbook, prepared by the U.S. Nuclear Regulatory Commission, Washington, D.C., 1981.

1999 by CRC Press LLC

16

Total Quality Management

and Risk Assessment

16.1 INTRODUCTION
Total quality management (TQM) is an enhancement to the conventional approach
of conducting business and it may simply be stated as a philosophy of pursuing
continuous improvement in each and every process through the integrated efforts of
all concerned persons associated with the organization.
The roots of the total quality movement may be traced back to the early 1900s
in the works of Frederick W. Taylor, the father of scientific management, concerning
the time and motion studies [1, 2]. In 1924, to control product variables, Walter A.
Shewhart, working as a quality control inspector for Bell Telephone Laboratories,
developed a statistical chart. This development is probably considered as the beginning of the statistical quality control.
In the late 1940s, the efforts of quality gurus such as W.E. Denning, J. Juran,
and A.V. Feigenbaum played an instrumental role in the strengthening of the TQM
movement [3]. In 1950, Deming lectured on the principles of statistical quality
control to 230 Japanese engineers and scientists at the request of the Japanese Union
of Scientists and Engineers (JUSE) [4]. In turn, in 1951, JUSE established the
Deming prize to be awarded to a company demonstrating the most effective implementation of quality measures and policies [5]. However, the term Total Quality
Management was not coined until 1985 and, in fact, it is credited to an American
behavioral scientist, Nancy Warren [5].
After witnessing the success of the Deming prize in Japan, the U.S. government
established the Malcolm Baldrige Award in 1987 for companies demonstrating
effective implementation of quality assurance policies and measures. In 1988, the
Cellular Telephone Division of Motorola was the first recipient of the Malcolm
Baldrige Award for its achievements in reducing defects from 1060/106 to 100/106
during the time frame from 1985 to 1988 [6, 7].
Since the late 1980s many other important events related to TQM have occurred.
Risk is present in all human activity and it can either be health and safety related
or economic. An example of the economic related risk is loss of equipment/production due to accidents involving fires, explosions, and so on. Nonetheless, risk may
simply be described as a measure of the probability and security of a negative effect
to health, equipment/property, or the environment [8]. Insurance is one of the oldest
strategies for dealing with risks and its history can be traced back to about 4000 years
in Mesopotamia (modern Iraq) when the Code of Hamurabi, in 1950 BC, formalized
bottomry contracts containing a risk premium for the chance of losing ships and
their cargo [9]. In the same area (i.e., Tigris-Euphrates Valley) around 3200 BC a

1999 by CRC Press LLC

group of people known as the Asipu served as risk analysis consultants to other
people involved in making risky, uncertain, or difficult decisions.
In the fourth century BC a Greek doctor, Hippocrates, correlated occurrence of
discases with environmental exposures and in the sixteen century AD Agricola
identified the correlation between occupational exposure to mining and health [9].
The basis of modern quantitative risk analysis was developed by Pierre Laplace
in 1792 by calculating the probability of death with and without smallpox vaccination. In the twentieth century, the conceptual development of risk analysis is due to
factors such as the development of nuclear power plants and concerns about their
safety and the establishment of such U.S. bodies as the Environmental Protection
Agency (EPA), Occupational Safety and Health (OSHA), and National Institute for
Occupational Safety and Health (NIOSH). A more detailed history of risk analysis
and risk management is provided in Reference 10.
Both TQM and risk assessment are described below.

16.2 TQM
A key to success in the corporate world has been the ability to improve on business
processes and operational tasks. Today, the playing field for company products has
changed forever because competition is no longer based solely on price, but rather
on a combination of price and quality. It means the combination must surpass the
competition offered by the competitive products for the ultimate success. Consequently, it may be said that companies must continually improve and upgrade their
capabilities in order to satisfy changing customer needs. Thus, quality goals and
measuring attainment of such goals is a dynamic process. In other words, achieving
quality is a journey without end.
A survey of 100 U.S. executives conducted in 1989 revealed that only 22% of
them stated that their company had done all it could to create a quality-fostering
environment [11, 12]. This was probably an important factor for picking up steam
by the TQM movement in the U.S. Nonetheless, according to Reference 11, the
rationale for embarking upon TQM varied among the U.S. companies and the
principal reasons included increased competition, customer demands, employees,
and internal crisis.
It is important to remember that many times there could be good scientific
innovations but due to an effective practice of a TQM methodology, the organizations/nations may lose the competitive advantage. For example, during the period
between the mid-1950s to the mid-1980s, British researchers won 26 Nobel Prizes
for new discoveries in comparison to only 4 by the Japanese [12]. But during the
same time period, the Japanese industry successfully took a major share of the world
market.

16.2.1 TRADITIONAL QUALITY ASSURANCE VS. TQM

In order to effectively practice TQM, it is important to understand basic differences
between the traditional quality assurance approach and TQM. The main areas of differences are shown in Figure 16.1. Table 16.1 presents differences in these areas [5, 7].

1999 by CRC Press LLC

FIGURE 16.1 The main areas of differences between the traditional quality assurance
program and TQM.

TABLE 16.1
Differences Between the Traditional Quality Assurance Approach (TQAA) and
TQM
Area
Objective
Definition
Cost
Customer
Quality defined
Quality
responsibility
Decision making

TQAA

TQM

Find errors
Product driven
Improvements in quality
increase cost
Ambiguous understanding of
customer needs
Products meet specifications
Inspection center/quality
control department
Practiced top-down approach

Prevent errors
Customer driven
Improvements in quality lower cost and increase
productivity
An effective approach defined to comprehend and
meet customer requirements
Products suitable for consumer use
All employees in the organization involved

16.2.2 TQM PRINCIPLES

AND

Practiced team approach with team of employees

ELEMENTS

Two fundamental principles of TQM are customer satisfaction and continuous

improvement. An organization involved in manufacturing or providing goods or
services could have either internal or external customers. The internal customers
belong to the organization itself, whereas the external customers are not part of the
organization. The practice of the market in concept allows a healthy customer
orientation and recognizes that every work process is made up of various stages [14].
Thus, at each stage, inputs from customers are sought for the purpose of determining
the changes to be made to better satisfy their needs.

1999 by CRC Press LLC

Continuous improvement is another fundamental principle and an important

factor to meet the quality challenge facing an organization. Two important functions
performed by management include continuously making improvements to techniques and procedures currently used, through process control and directing efforts
to make important breakthroughs in concerned processes. Seven important elements
of TQM are management commitment and leadership, customer service, supplier
participation, team effort, cost of quality, training, and statistical techniques and
methods [14].
For the success of the application of the TQM concept, the management commitment and leadership role is very essential. A clear understanding of the TQM
concept by the senior management is very important in order to count on its commitment and leadership role. Subsequently, the management can establish new goals
and policies for the entire organization and then play a leadership role in their
realization.
As the demand for better quality increases, the involvement of customers is
becoming increasingly important. The main purpose of such involvement is to
represent customer interests and ultimately reflect them in goals, plans, and controls
developed by the quality team.
A companys ability to produce a good quality product to a certain degree
depends upon the type of relationship among the parties involved in the process: the
processor, the customer, and the supplier. Even though mutual trust and partnership
are the essential ingredients for supplier responsiveness, some organizations require
their suppliers to have formal TQM programs [15].
The purpose of the team effort or approach is to involve all concerned parties
in the TQM program including customers, vendors, and subcontractors. The quality
team membership is on a voluntary basis and 3 to 15 individuals form the team.
However, the team members possess skills in areas such as cost-benefit analysis,
brainstorming approaches, statistics, planning and controlling projects, public relations, flow charting, and presentation methods.
The cost of quality is a primary quality measurement tool and is applied to select
quality improvement projects, monitor the effectiveness of the TQM process, and
justify the cost to doubters. The two important advantages of the cost of quality are
to communicate to management the benefit of applying the TQM concept in terms
of dollars and to raise quality awareness.
A Japanese axiom states that quality begins with training and ends with training
[16]. As under the TQM program, quality is the responsibility of each employee in
the organization; thus, it is essential that the training effort must be targeted for each
hierarchy level of the firm.
Statistical techniques and methods are used to solve various TQM related problems. Some examples of the statistical techniques and methods are control charts,
scatter diagrams, graphs and histograms, cause and effect diagrams, and Pareto
diagrams [5, 17-20]. Such statistical approaches may be used for purposes such as
verifying, repeating, and reproducing measurements based on data, identifying and
separating quality problem causes, making decisions on facts based on data rather
than the opinions of individuals, and communicating the information in a language
that can easily be understood by all quality team members.

1999 by CRC Press LLC

16.2.3 TQM CONTRIBUTORS

Over the years, there have been many people who directly or indirectly contributed
to the development of the TQM concept. Three of these people who played important
roles were W.E. Deming, J.M. Juran, and P.B. Crosby. Each of these individuals lay
down series of points/steps to improve quality. Such points/steps are presented
below [1].
Deming Points
W.E. Deming, a graduate in engineering, mathematics, and physics, outlined his
philosophy in 14 points to improve quality. His 14 points are as follows [1, 5, 21, 22]:
Create constancy of purpose for improving products/services. This means
develop a mission statement by considering issues such as quality philosophy,
investors, long term corporate objectives, employees, and profit distribution.
Adopt the new philosophy or lead to promote change. More specifically,
the current acceptable levels of defects/mistakes/delays are unacceptable
and every concerned body is alerted to determine reasons for their (i.e.,
defects/mistakes/delays) existence. Ultimately, the team effort is practiced
to rectify problems under consideration.
Stop depending on mass inspection to achieve quality and build quality
into the product.
Stop awarding business or contracts on the basis of low bids and develop
long-term relationships on the basis of performance.
Continuously improve product, quality, service.
Institute training that includes modern methods, techniques, and approaches.
Take advantage of modern supervision techniques and approaches.
Eliminate fear so that everyone may perform effectively.
Break down barriers between departments/units and demand team effort.
Eradicate slogans, numerical goals, etc. for the workforce because they
generate adversarial relationships.
Remove numerical quotas and management by objectives (MBO) and
instead substitute leadership.
Remove obstacles to employee pride in workmanship.
Encourage a vigorous education and self-improvement program.
Make the transformation every individuals task and make every individual
work on it.
Juran Steps
J.M. Juran, another guru of quality, stated his quality philosophy in a number of
steps. His steps to quality improvement are the following:
build awareness of both the need for improvement and opportunities for
improvement
develop goals for acquired improvements

1999 by CRC Press LLC

organize to achieve the set goals

impart training as appropriate
implement projects directed at finding solutions to problems
report progress periodically
give proper recognition
communicate results to all concerned bodies
keep scores
sustain momentum by making improvements to the organizations regular
systems.

Jurans quality philosophy is described in detail in References 1 and 23.

Crosby Steps
P.B. Crosby, another major contributor to quality, outlined his quality philosophy in
14 steps as follows [1, 23]:

Ensure that management is committed to quality for the long period.

Establish cross-departmental teams to improve quality.
Highlight areas with current and potential problems.
Estimate the cost of quality and also describe how it is being utilized by
management.
Enhance the quality awareness and all employees personal commitment
to quality as much as possible.
Take corrective measures as soon as possible to rectify highlighted problems.
Develop program aim at zero defects.
Provide appropriate training to supervisors so that they can perform their
responsibilities in the quality program effectively.
Hold a zero defects day for the purpose of ensuring that everyone in the
organization is aware of the new direction.
Encourage both individuals and teams to develop personnel and team
improvement goals with respect to quality.
Encourage all employees to make management aware of difficulties
encountered by them in achieving set quality goals.
Effectively recognize participating employees.
Establish quality councils to promote communication on an on-going
basis.
Repeat everything to demonstrate that quality improvement is a journey
without an end.

16.2.4 TQM METHODS

Various appropriate methods or tools are needed to implement TQM. The success of
such implementation will depend on the degree of understanding of these methods by
the management as well as adopting the most suitable of such methods. Nonetheless,

1999 by CRC Press LLC

TABLE 16.2
Selective TQM Methods Belonging
to the Analytical Classification
Method
No.
1
2
3
4
5
6
7
8
9
10

Method name
Taguchi methods
Cause and effect analysis
Force field analysis
Domainal mapping
Solution effect analysis
Failure mode and effect analysis
Tolerance design
Fault tree analysis
Minute analysis
Paired comparisons

the primary role of the TQM methods in problem-solving for improving quality is
to be effective in satisfying customer needs. Furthermore, the methods can also be
useful to identify possible root causes and their potential solutions, in addition to
utilizing data/information to choose the most suitable alternatives for managing
quality.
There are a large number of methods that can be used in TQM work. In fact,
100 such methods are described in Reference 20. This reference classifies the TQM
methods into four distinct categories:

Analytical
Management
Idea generation
Data collection, analysis, and display

Tables 16.2 through 16.5 present selective methods belonging to each of the above
four categories, respectively. Some of these TQM methods are described below
[20-22].
Pareto Analysis
This is performed to separate the most important causes of a problem from the trivial
ones. It is named after an Italian economist called Vilfredo Pareto (18481923) and
its popular use in quality work is due to J.M. Juran who emphasized that 80% of
quality problems are the result of only 20% of the causes. Thus, Pareto analysis is
an extremely useful TQM method to highlight areas for a concerted effort.
Sometime Pareto analysis is also called the 80/20 rule. Nonetheless, the following steps are used to perform such analysis [20]:

1999 by CRC Press LLC

TABLE 16.3
Selective TQM Methods Belonging to
the Management Classification
Method
No.

Method name

1
2
3
4
5
6
7
8
9
10

Kaizen
Pareto analysis
Quality circles
Quality function deployment (QFD)
Potential problem analysis
Deming wheel (PDCA)
Error proofing (pokayoke)
Benchmarking
Arrow diagram
Mystery shopping

TABLE 16.4
Selective TQM Methods Belonging
to the Idea Generation Classification
Method
No.
1
2
3
4
5
6
7
8
9
10

Method name
Brainstorming
Morphological forced connections
Opportunity analysis
Mind mapping
Nominal group technique
Imagineering
Snowballing
Buzz groups
List reduction
Multi-voting

List activities/causes in a tabular form and count their occurrences.

Arrange them in descending order.
Compute the total for the entire list.
Determine for each cause/activity the percentage of total.
Develop a Pareto diagram by showing percentages vertically and their
corresponding causes horizontally.
Conclude from the end results.

1999 by CRC Press LLC

TABLE 16.5
Selective TQM Methods Belonging to the Data
Collection, Analysis, and Display Classification
Method
No.
1
2
3
4
5
6
7
8
9
10

Method name
Histograms
Process analysis
Box and whisker plots
Checksheets
Scatter diagrams
Hoshin Kanri (quality policy deployment)
Flowcharts
Spider web diagrams
Statistical process control (SPC)
Dot plots

Hoshin Kanri (Quality Policy Deployment)

This method is used to delight customers through the manufacturing and servicing
process by implementing the company goals. The method is applied when through
top-down and bottom-up consultation the objectives are identified at each organizational level and the overall organizational goals have been established as particular
targets. The following factors are associated with Hoshin Kanri [20, 24]:

Establish long and short term organizational goals.

Identify the goals that can be measured.
Identify the important processes involved in achieving the above goals.
Direct the team members to reach consensus on performance indicators
at appropriate process stages.
Challenge each process level to force organization to make changes to its
quality culture.
Use organizational goals as measurable goals for the purpose of making
employees comprehend the criticality of the quality improvement process.
Cause-and-Effect Diagram
This is also known as the Ishikawa diagram, named after its Japanese originator K.
Ishikawa, who developed it in the early 1950s. Another name used to describe the
cause-and-effect diagram is Fishbone diagram because of its resemblance to the
skeleton of a fish as shown in Figure 16.2.
The cause-and-effect diagram method is an extremely useful approach to determine the root causes of a given problem and to generate relevant ideas. Pictorially,
the right-hand side of the diagram or the fish head represents effect and the left-hand

1999 by CRC Press LLC

FIGURE 16.2 A general cause-and-effect diagram.

side represents all the possible causes which are connected to the central line known
as the Fish spine.
In particular, the cause-and-effect diagram with respect to TQM can be described
as follows: the customer satisfaction could be the effect, and manpower, materials,
methods, and machinery are the major causes. The following steps are useful in
developing a cause-and-effect diagram:
Develop problem statement.
Brainstorm to highlight all possible causes.
Establish main cause classifications by stratifying into natural groupings
and process steps.
Develop the diagram by connecting the causes by following essential
process steps and fill in the problem or the effect in the diagram box on
the extreme right.
Refine cause classifications through questions such as follows:
What causes this?
Why does this condition exist?
There are many advantages of the cause-and-effect diagram, including useful to
highlight root causes, a useful tool to generate ideas, an effective mechanism to
present an orderly arrangement of theories, and a useful vehicle to guide further
inquiry.
Quality Function Deployment (QFD)
This is a method used for optimizing the process of developing and manufacturing
new products as per customer need. The method was developed by the Japanese in
1972 [22, 25] to translate customer needs into appropriate technical requirements.
The approach can be applied in areas such as research, product development, engineering, manufacturing, and marketing. The technique makes use of a set of matrices
to relate customer needs to counterpart characteristics expressed as technical specifications and process control requirements. The important QFD planning documents
include customer needs planning matrix, process plan and quality control charts,
operating instructions, and product characteristic deployments matrix.

1999 by CRC Press LLC

The customer need planning matrix is used to translate the consumer requirements into product counterpart characteristics and the purpose of the process plan
and quality control charts is to identify important process and product parameters
along with control points. The operating instructions are useful to identify operations
that must be accomplished to achieve critical parameters and on the other hand the
product characteristic deployment matrix is used to translate final product counterpart characteristics into critical component characteristics.
A QFD matrix is frequently called the House of Quality because of its resemblance to a house. The following steps are necessary to build the house of quality
[22, 25]:
Identify the needs of customers.
Identify the essential product/process characteristics that will meet the
customer requirements.
Establish the relationship between the customer needs and the counterpart
characteristics.
Conduct evaluation analysis of competing products.
Develop counterpart characteristics of competing products and formulate
appropriate goals.
Identify counterpart characteristics to be utilized in the remaining process.
One clear cut advantage of the QFD is that it encourages companies to focus
on the process itself rather than focusing on the product or service. Furthermore,
the development of correlations between what is required and how it is to be acquired,
the important areas become more apparent, help in making decisions.
One important limitation associated with the application of QFD is that the exact
needs must be identified in complete detail.

16.2.5 DESIGNING FOR QUALITY

PROCESS SUCCESS

AND

GOALS

FOR

TQM

It may be said that quality starts with the product design specification writing phase
and continues to its operational phase. Nonetheless, there are a number of product
design elements that can adversely affect quality. Some of these elements are poor
part tolerances, lack of design refinement resulting in the use of more parts than
required to carry out the desired functions, difficulty in fabricating parts because of
poor design features, and the use of fragile parts.
There are various measures that can be useful to improve quality during the
product design phase including eliminating the need for adjustments, using repeatable
and well-understood processes, using parts that can easily withstand process operations, reducing parts and part numbers, eliminating engineering changes on released
products, simplifying assembly and making it foolproof, designing for robustness
using Taguchi methods, and designing for efficient and satisfactory testing [26].
There are a number of goals that must be satisfied in order to achieve the
effectiveness of the TQM process. These goals include:

1999 by CRC Press LLC

 A clear understanding of internal/external customer needs by all organization employees.

Meeting of control guidelines as per customer requirements by all significant process/systems.
Use of a system to continuously improve processes that better fulfill the
current and future needs of customers.
Establishment of real rewards and incentives for employees when process
control and customer satisfaction results are achieved.

16.2.6 DEMINGS DEADLY DISEASES OF AMERICAN

MANAGEMENT AND COMMON ERRORS DURING
THE STARTING QUALITY INITIATIVE PHASE
W.E. Deming not only put forward 14 points to improve quality, but he also pointed
out seven deadly diseases of American Management that are counter-productive to
quality. These deadly diseases are excessive medical costs (e.g., Blue Cross is the
highest paid supplier of General Motors), excessive cost of liabilities (e.g., the U.S.
is the leading country in the world in lawsuits), lack of constancy of purpose, job
hopping by management personnel (this works against understanding and long-term
efforts), performance reviews (they are detrimental to team effort and build fear),
running an organization on visible data and information alone (effects of a dissatisfied customer cannot be measured), and emphasis on short-term profits [1, 27].
Beside putting in half-hearted implementation efforts and setting unrealistic
expectation goals, many companies also make various common errors when starting
quality initiatives. These common errors are as follows [1, 28]:
Delegation of authority and poor management leadership. In some
organizations, the management delegate responsibility to a hired expert
when starting quality initiatives rather than taking the appropriate leadership role to get everyone on board.
Team mania. Ultimately, all employees should be involved with teams.
Also, supervisors and employees must learn how to be effective coaches
and team players, respectively. The team effort can only succeed after
cultural change in the organization.
Practicing a narrow, dogmatic approach. Some companies follow Deming philosophy, Juran philosophy, or Crosby philosophy without realizing
that none of these philosophies is truly a one-size-fits-all proposition.
More specifically, the quality programs must be tailored to satisfy their
individual specific needs.
Deployment process. Some companies develop quality initiatives without
simultaneously developing effective mechanisms to integrate them into
all different aspects of the organization; for example, marketing, operations, and budgeting.
Fuzziness about the differences among factors such as inspiration,
awareness, skill building, and education.

1999 by CRC Press LLC

16.3 RISK ASSESSMENT

Risk assessment may simply be described as the process of risk analysis and risk
evaluation [8, 29]. In turn, risk analysis is concerned with the utilization of available
data for determining the risk to human, environment, or property/equipment from
hazards and usually is made up of steps such as scope definition, hazard identification, and risk determination. The stage at which values and judgments make entry
to the decision process is known as risk evaluation.
The complete process of risk assessment and risk control is called risk management. The term risk control is simply the decision making process concerned with
managing risk as well as the implementations, enforcement, and re-evaluation of its
effectiveness periodically, using risk assessment end results as one of the input
factors.

16.3.1 RISK ANALYSIS ROLE

EXPERTISE AREAS

AND

REQUIRED

Risk can only be managed effectively after its comprehension analysis. Risk analysis
serves as a useful tool in identifying health and safety problems and approaches to
uncover their solutions, satisfying regulatory requirements, and facilitating objective
decisions on the risk acceptability.
A multi-disciplinary approach is often required to conduct risk analysis and it
may require adequately sufficient knowledge in areas such as probability and statistics, engineering (electrical, mechanical, chemical, or nuclear), systems analysis,
health sciences, social sciences, and physical, chemical, or biological sciences.

16.3.2 RISK ANALYSIS OBJECTIVES

SYSTEM LIFE CYCLE

IN

HAZARDOUS

The life cycle of hazardous systems may be divided into three major phases as shown
in Figure 16.3.

FIGURE 16.3 Major phases of hazardous systems.

1999 by CRC Press LLC

The major phases shown in the figure are as follows [8]:

Phase I: Concept and definition, design, and development phase.
Phase II: Construction and production, operation, and maintenance phase.
Phase III: Disposal phase.
The risk analysis objectives concerning Phase I include: identify major contributors
to risk, provide input to design process and assess overall design adequacy, provide
input to help in establishing procedures for normal and emergency conditions, and
provide appropriate level of input to the evaluation of the acceptability of proposed
potentially hazardous facilities or activities.
Some of the specific risk analysis objectives belonging to Phase II are to gauge
and assess experience for the purpose of making comparisons between actual performance and relevant requirements, to update information on major risk contributors, to provide input on plant risk status in operational decision making, and to
provide input to the optimization of normal and emergency procedures.
The two important risk analysis objectives of Phase III are to provide input to
disposal policies and procedures, and to assess the risk associated with process
disposal activities so that appropriate requirements can be satisfied effectively.

16.3.3 RISK ANALYSIS PROCESS STEPS

The risk analysis process is composed of six steps as shown in Figure 16.4.

FIGURE 16.4 Risk analysis process steps.

1999 by CRC Press LLC

The establishing scope definition is the first step of the risk analysis process and
the risk analysis scope is defined and documented at the beginning of the project
after a thorough understanding of the system under consideration. Nonetheless, the
following five basic steps are involved in defining the risk analysis scope:
1. Describe the problems leading to risk analysis and then formulate risk
analysis objectives on the basis of major highlighted concerns.
2. Define the system under consideration by including factors such as system
general description, environment definition, and physical and functional
boundaries definition.
3. Describe the risk analysis associated assumptions and constraints.
4. Highlight the decisions to be made.
5. Document the total plan.
Identifying hazards is the second step of the risk analysis process and is basically
concerned with the identification of hazards that will lead to risk in the system. This
step also calls for the preliminary evaluation of the significance of the identified
hazardous sources. The main purpose of this evaluation is to determine the appropriate course of action.
Estimating risk is the third step of the risk analysis process and risk estimation
is conducted in the following steps:
Investigate sources of hazards to determine the probability/likelihood of
occurrence of the originating hazard and associated consequences.
Conduct pathway analysis to determine the mechanisms and likelihood
through which the receptor under consideration is influenced.
Choose risk estimation methods/approaches to be used.
Identify data needs.
Discuss assumptions/rationales associated with methods/approaches/data
being utilized.
Estimate risk to evaluate the degree of influence on the receptor under
consideration.
Document the risk estimation study.
Documenting is the fourth step of the risk analysis process and is basically concerned
with effectively documenting the risk analysis plan, the preliminary evaluation, and
the risk estimation. The documentation report should contain sections such as title,
abstract, conclusions, table of contents, objectives/scope, assumptions/limitations,
system description, analysis methodology description, results of hazard identification, model description and associated assumptions, quantitative data and associated
assumptions, results of risk estimation, references, appendices, discussion of results,
and sensitivity analysis.
The fifth step of the risk analysis process is basically concerned with verifying
the end results. More specifically, it may be stated that verification is a review process

1999 by CRC Press LLC

used to determine the integrity and accuracy of the risk analysis process. Verification
is conducted at appropriate times by person(s) other than the involved analyst(s).
The final step of risk analysis is concerned with periodically updating the
analysis as more up-to-date information becomes available.

16.3.4 RISK ANALYSIS TECHNIQUES

FOR

ENGINEERING SYSTEMS

Over the years many different methods to conduct risk analysis have been
developed [8, 30-32]. It is important to carefully consider the relevance and suitability of these techniques prior to their proposed applications. Some of the factors
to be considered specifically in this regard are appropriateness to the system under
study, scientific defensibility, format of the results with respect to improvement in
understanding of the risk occurrence and risk controllability, and simplicity.
The additional factors that should be used as the basis for selecting risk analysis
technique(s) by the analyst include study objectives, development phase, level of risk,
system type and hazard being analyzed, information and data needs, manpower requirement, level of expertise required, updating flexibility, and resource requirement.
The risk analysis methods used for engineering systems may be grouped into
two categories: (1) hazard identification and (2) risk estimation. Examples of the
techniques belonging to the hazard identification category are hazard and operability
study (HAZOP), event tree analysis (ETA), and failure modes and effect analysis
(FMEA). Similarly, the examples of the techniques belonging to the risk estimation
category include frequency analysis and consequence analysis.
Four of these techniques are discussed below.
Consequence Analysis
Consequence analysis is concerned with estimating the impact of the undesired event
on adjacent people, property, or the environment. Usually, for risk estimations
concerning safety, it consists of calculating the probability that people at different
distances (and in different environments) from the undesired event source will suffer
injury/illness. Some examples of the undesired event are explosions, projection of
debris, fires, and release of toxic materials. It means there is a definite need to use
consequence models for predicting the extent and probability of casualties due to
such undesired events. Nonetheless, it is appropriate that consequence analysis takes
into consideration the factors such as analysis based on chosen undesirable events,
measures to eradicate consequences, explanation of any series of consequences
resulting from the undesirable events, measures to eradicate consequences, explanation of any series of consequences resulting from the undesirable events, outlining
of the criteria used for accomplishing the identification of consequences, and immediate and aftermath consequences.
Hazard and Operability Study (HAZOP)
This is a form of FMEA and was developed for applications in chemical industries.
HAZOP is a systematic approach used to identify hazards and operational problems
throughout a facility. Over the years, it is proven to be an effective tool for identifying

1999 by CRC Press LLC

unforeseen hazards designed into facilities due to various reasons, or introduced into
existing facilities due to factors such as changes made to process conditions or
operating procedures.
The approach has primarily threefold objectives: develop a full facility/process
description, systematically review each and every facility/process part to identify
how deviations from the design intentions can occur, and pass judgment on whether
such deviations can result in hazards/operating problems.
HAZOP can be used to analyze design at various stages as well as to perform
analysis of process plants in operation. However, the application of HAZOP in the
early design phase often leads to a safer detailed design. The following basic steps
are associated with HAZOP [8]:
Develop study objectives and scope.
Form HAZOP team by ensuring that its membership is comprised of
persons from design and operation.
Collect appropriate drawings, process description, and other relevant documentation; for example, layout drawings, process control logic diagrams,
operation and maintenance procedures, and process flowsheets.
Conduct analysis of all major pieces of equipment, supporting equipment,
etc.
Effectively document the study.
Frequency Analysis
This is basically concerned with estimating the occurrence frequency of each undesired event or accident scenario. Two commonly used approaches for performing
frequency analysis are as follows:
Making use of the frequency data of past relevant undesired events to
predict frequency of their future occurrences.
Using methods such as ETA and fault tree analysis (FTA) to calculate the
occurrence frequencies of undesired events.
All in all, each of the above two techniques has strengths where the other has
weaknesses; thus, each such approach should be used to serve as a check for the
other wherever it is feasible.
Event Tree Analysis (ETA)
This is a bottom-up approach used to identify the possible outcomes when an
initiating events occurrence is known. The approach is useful for analyzing facilities
having engineered accident-mitigating characteristics to identify the events
sequence that follows the initiating event and generate given sequences. Usually, it
is assumed that each sequence event is either a success or a failure. It is important
to note that often the ETA approach is used to perform analysis of more complex
systems then the ones handled by the FMEA approach [30, 33, 34].

1999 by CRC Press LLC

Because of the inductive nature of ETA, the fundamental question asked is What
happens if ? ETA highlights the relationship between the success or failure of
various mitigating systems as well as the hazardous event that follows the initiating
event. Additional factors associated with ETA include
A comprehensive risk assessment requires identification of all potential
initiating events.
An excellent tool to identify events that require further investigation using
FTA.
It is rather difficult to incorporate delayed success or recovery events when
performing ETA.
ETA application always leaves room for missing some important initiating
events.

16.3.5 ADVANTAGES

OF

RISK ANALYSIS

There are many benefits to performing risk analysis. Some of those are as follows [8]:

It
It
It
It
It
It
It

identifies potential hazards and failure modes.

provides in-depth understanding of the system.
helps to make comparisons of risks to those of other similar systems.
is useful to suggest modifications to reduce risks.
provides quantitative risk statements.
helps to identify major contributors to risk.
is useful to develop priorities for expenditures on improved safety.

16.4 PROBLEMS
1. Write an essay on TQM.
2. Define the following terms:
Risk
Risk management
Risk assessment
Risk control
3. Compare traditional quality assurance program with TQM.
4. Discuss Demings 14 points concerning TQM.
5. List 10 most important methods of TQM. Provide a short discussion on
each of these methods.
6. What are the Demings deadly diseases of American management with
respect to quality?
7. What are the common errors made when starting quality initiatives?
8. Describe the risk analysis process.
9. What are the benefits of risk analysis?
10. Discuss the following two risk analysis methods:
HAZOP
ETA

1999 by CRC Press LLC

16.5 REFERENCES
1. Goetsch, D.L. and Davis, S., Implementing Total Quality, Prentice-Hall, Englewood
Cliffs, NJ, 1995.
2. Rao, A., Carr, L.P., Dambolena, I., Kopp, R.J., Martin, J., Raffi, F., and Schlesinger,
P.F., Total Quality Management: A Cross Functional Perspective, John Wiley & Sons,
New York, 1996.
3. Gevirtz, C.D., Developing New Products with TQM, McGraw-Hill, New York, 1994.
4. Dobyns, L. and Crawford-Mason, C., Quality or Else, Houghton Mifflin, Boston, 1991.
5. Schmidt, W.H. and Finnigan, J.P., The Race without a Finish Line: Americas Quest
for Total Quality, Jossey-Bass Publishers, San Francisco, CA, 1992.
6. Van Ham, K., Setting a total quality management strategy, in Global Perspectives on
Total Quality, The Conference Board, New York, 1991.
7. Muadu, C.N. and Chu-hua, K., Strategic total quality management (STQM), in
Management of New Technologies for Global Competitiveness, Madu, C.N., Ed.,
Quorum Books, Westport, CT, 1993, pp. 3-25.
8. Risk Analysis Requirements and Guidelines, CAN/CSA-Q6340-91, prepared by the
Canadian Standards Association, 1991. Available from Canadian Standards Association, 178 Rexdale Boulevard, Rexdale, Ontario, Canada.
9. Molak, V., Ed., Fundamentals of Risk Analysis and Risk Management, CRC Press,
Boca Raton, FL, 1997.
10. Covello, V.T. and Manpower, J., Risk analysis and risk management: A historical
perspective, Risk Analysis, 5, 103-120, 1985.
11. Farquhar, C.R. and Johnston, C.G., Total Quality Management: A Competitive Imperative Report No. 60-90-E, 1990, The Conference Board of Canada, Ottawa, Ontario,
Canada.
12. Caropreso, F., Making Total Quality Happen, The Conference Board, New York, 1990.
13. Spenley, P., World Class Performance Through Total Quality, Chapman and Hall,
London, 1992.
14. Burati, J.L., Matthews, M.F., and Kalidindi, S.N., Quality management organizations
and techniques, J. Construction Eng. Manage., 118, 112-128, 1992.
15. Matthews, M.F. and Burati, J.L., Quality Management Organizations and Techniques,
Source Document 51, The Construction Industry Institute, Austin, Texas, 1989.
16. Imai, M., Kaizen: The Key to Japans Competitive Success, Random House, New
York, 1986.
17. Ishikawa, K., Guide to Quality Control, Asian Productivity Organization, Tokyo,
1982.
18. Kume, H., Statistical Methods for Quality Improvement, The Association for Overseas
Technology Scholarship, Tokyo, 1985.
19. Perisco, J., Team up for quality improvement, Quality Progress, 22, 33-37, 1989.
20. Kanji, G.K. and Asher, M., 100 Methods for Total Quality Management, Sage Publications Ltd., London, 1996.
21. Heizer, J. and Render, B., Production and Operations Management, Prentice-Hall,
Upper Saddle River, New Jersey, 1995.
22. Mears, P., Quality Improvement Tools and Techniques, McGraw-Hill, New York, 1995.
23. Uselac, S., Zen Leadership: The Human Side of Total Quality Team Management,
Mohican Publishing Company, Londonville, OH, 1993.
24. Akao, Y., Hoshin Kanri: Policy Deployment for Successful TQM, Productivity Press,
Cambridge, MA, 1991.
25. Yoji, K., Quality Function Deployment, Productivity Press, Cambridge, MA, 1991.

1999 by CRC Press LLC

26. Daetz, D., The effect of product design on product quality and product cost, Quality
Progress, June 1987, pp. 63-67.
27. Coppola, A., Total quality management, in Tutorial Notes, Annu. Reliability Maintainability Symp., 1992, pp. 1-44.
28. Clemmer, J., Five common errors companies make starting quality initiatives, Total
Quality, 3, 4-7, 1992.
29. Kunreuther, H. and Slovic, P., Eds., Challenges in risk assessment and risk management, in Annu. Am. Acad. Political Soc. Sci., 545, 1-220, 1996.
30. Wesely, W.E., Engineering risk analysis, in Technological Risk Assessment, Rice, P.F.,
Sagan, L.A., and Whipple, C.G., Eds., Martinus Nijhoff Publishers, The Hague, 1984,
pp. 49-84.
31. Covello, V. and Merkhofer, M., Risk Assessment and Risk Assessment Methods: The
State-of- the-Art, NSF report, 1984, National Science Foundation (NSF), Washington,
D.C.
32. Dhillon, B.S. and Rayapati, S.N., Chemical systems reliability: A survey, IEEE Trans.
Reliability, 37, 199-208, 1988.
33. Cox, S.J. and Tait, N.R.S., Reliability, Safety, Risk Management, Butterworth-Heinemann Ltd., Oxford, 1991.
34. Ramakumar, R., Engineering Reliability: Fundamentals and Applications, PrenticeHall, Englewood Cliffs, NJ, 1993.

1999 by CRC Press LLC

17

Life Cycle Costing

17.1 INTRODUCTION
Today in the global economy and due to various other market pressures, the procurement decisions of many products are not entirely made on initial acquisition
costs but on their total life cycle costs; in particular is the case of expensive products.
Many studies performed over the years indicate that the product ownership costs
often exceed procurement costs. In fact, according to Reference 1, the product
ownership cost (i.e., logistics and operating cost) can vary from 10 to 100 times the
procurement cost. Even the assertion of the ownership cost being high could be
detected from the overall annual budgets of various organizations. For example, in
1974, the operation and maintenance cost accounted for 27% of the total U.S.
Department of Defense budget as opposed to 20% for procurement [2].
Life cycle cost of a product may simply be described as the sum of all costs
incurred during its life span, i.e., the total of procurement and ownership costs. The
history of life cycle costing goes back to the mid-1960s when Logistics Management
Institute prepared a document [3] entitled Life Cycle Costing in Equipment Procurement for the U.S. Assistant Secretary of Defense, Installations and Logistics.
Consequently, the Department of Defense released a series of three guidelines for
life cycle costing procurement [4-6].
In 1974, Florida became the first U.S. state to formally adopt the concept of life
cycle costing and in 1978, the U.S. Congress passed the National Energy Conservation Policy Act. The passage of this Act made it mandatory that every new federal
government building be life cycle cost effective. In 1981, Reference 7 presented a
list of publications on life cycle costing. Over the years many people have contributed
to the subject of life cycle costing and Reference 8 presents a comprehensive list of
publications on the subject. This chapter discusses different aspects of life cycle
costing.

17.2 REASONS AND USES OF LIFE CYCLE COSTING

AND REQUIRED INPUTS
Life cycle costing is increasingly being used in the industry to make various types
of decisions. Some of the reasons for this upward trend could be increasing maintenance cost, competition, increasing cost effectiveness awareness among product
users, budget limitations, costly products (e.g., aircrafts, military systems), and
greater ownership costs in comparison to procurement costs.
The life cycle costing concept can be used for various different purposes including selecting among options, determining cost drivers, formulating contractor incentives, choosing the most beneficial procurement strategy, assessing new technology
application, forecasting future budget needs, making strategic decisions and design

1999 by CRC Press LLC

trade-offs, providing objectives for program control, deciding the replacement of

aging equipment, improving comprehension of basic design associated parameters
in product design and development, optimizing training needs, and comparing logistics concepts [9, 10].
A life cycle costing project can only be accomplished effectively if the required
information is available. Nonetheless, the professionals involved in conducting life
cycle cost studies should seek information on general items such as those listed
below prior to the start of the project under consideration [8, 9, 11]:

Estimate goal
Time schedule
Required data
Involved individuals
Required analysis format
Required analysis detail
Treatment of uncertainties
Ground rules and assumptions
Analysis constraints
Analysis users
Limitations of funds
Required accuracy of the analysis

Nonetheless, the specific data required to conduct life cycle cost studies for an item
include useful life, acquisition cost, periodic maintenance cost, transportation and
installation costs, discount and escalation rates, salvage value/cost, taxes (i.e., investment tax credit, tax benefits from depreciation, etc.), periodic operating costs (e.g.,
energy cost, labor cost, insurance, cost of materials, and cost of supplies) [12].

17.3 LIFE CYCLE COSTING STEPS AND ACTIVITIES

A life cycle costing study can be conducted in seven steps as shown in Figure 17.1
[13]. There are many activities associated with life cycle costing. Some of those
activities are as follows [14]:

Develop an accounting breakdown structure.

Define activities that generate ownership costs of products.
Establish cause-and-effect relationships.
Identify cost drivers.
Determine or define the life cycle of items/products.
Develop cost estimate relationships for every element in the life cycle
cost breakdown structure.
Conduct sensitivity analysis.
Develop escalated and discounted life cycle costs.

1999 by CRC Press LLC

FIGURE 17.1 Life cycle costing steps.

17.4 SKILL REQUIREMENT AREAS OF A LIFE

CYCLE COSTING ANALYST AND HIS/HER
ASSOCIATED PROFESSIONALS
In order to perform an effective life cycle cost analysis, the life cycle cost analyst
must possess skills in many areas including engineering, finance and accounting,
estimating, statistical analysis, reliability and maintainability engineering, logistics,
and contracting. Nonetheless, the professionals such as those listed below provide
input to life cycle cost studies:

Design engineers
Maintenance engineers
Tooling engineers
Reliability and maintainability engineers
Planning engineers
Test engineers
Quality control engineers
Manufacturing engineers

1999 by CRC Press LLC

17.5 LIFE CYCLE COSTING PROGRAM EVALUATION

AREAS AND LIFE CYCLE COST ESTIMATE REPORT
In order to keep a life cycle costing program functioning effectively, it is evaluated
periodically. There are many areas in which questions could be raised by the equipment manufacturer/user management. These areas include [9, 15]:

Inclusion of life cycle cost requirements into subcontracted designs.

Cost performance review of subcontractors.
Validity of cost-estimating approaches being used.
Identification of cost drivers.
Effectiveness of inflation and discounting factors being used.
Cost estimating databases broadness.
Buyers awareness of top ten cost drivers.
Managements defining of cost priority in relation to factors such as
product delivery schedule and performance.
Planning and design groups awareness of life cycle cost requirements
concerning them.
Coordination of life cycle costing effort.
Life cycle cost requirements compatibility with reliability, maintainability, and safety programs.
Qualifications of professionals managing the life cycle costing program.
Cost estimates validation by an independent appraisal.
Cost model construction.
Performance of trade-off studies.

The documentation of the results of the life cycle cost analysis is as important as
the analysis itself. Poor documentation can dramatically decrease the effectiveness
of actual analysis. Needless to say, the writing of life cycle cost estimate reports
requires careful consideration. Even though such reports may vary from one
project/organization to another, they should include information on items such as
those listed below [8].

Data sources and their validity/accuracy

The life cycle cost model
Ground rules and assumptions
Sensitivity analyses
Cost-estimating relationships used
Important cost breakdown estimates
Contractual cost incentives/requirements (e.g., reliability improvement
warranty and design to cost)
Comparisons of life cycle cost estimate with other similar estimates (if
applicable)
Significance and conclusions of end estimates.

1999 by CRC Press LLC

17.6 TIME DEPENDENT FORMULAS FOR LIFE CYCLE

COST ANALYSIS
As the value of money changes with the passage of time in estimating the life cycle
cost of a product, in particular its ownership cost has to be discounted by using
various time dependent formulas before it is added to the procurement cost. In
modern society, interest and inflation rates are used to take into consideration the
time dependent value of money. Nonetheless, the history of the interest may be
traced back to 2000 B.C. in Babylon where interest on borrowed commodities (e.g.,
grain) was paid in the form of grain or through other means. Around 575 B.C., the
concept of interest reached to a such level that it led to the establishment of a firm
of international bankers with home offices in Babylon [16].
This section presents selective formulas to take into consideration the time value
of money for use in life cycle cost analysis [8, 17, 18].

17.6.1 SINGLE PAYMENT COMPOUND AMOUNT FORMULA

In this case, P amount of money is invested for m interest periods (i.e., normally
years) at compound interest rate i (i.e., interest rate per interest period). Thus, the
future amount, FA, of the invested money is
FA = P (1 + i)

(17.1)

Example 17.1
A person deposited $700 dollars in a bank at a compound interest rate of 5% per
year. Calculate its future amount after 7 years.
By inserting the given data into Equation (17.1) we get
FA = 700 (1 + 0.05)

$985

Thus, the future amount after 7 years will be $985.

17.6.2 SINGLE PAYMENT PRESENT VALUE FORMULA

This formula is concerned with determining the present value of a single amount of
money to be received or spent after m interest periods or years at compound interest
rate i.
Thus, from Equation (17.1), we get
P = FA (1 + i)

1999 by CRC Press LLC

(17.2)

where
P is the present value of FA.
Example 7.2
A company sold a personal computer and the buyer agreed to pay $5000 after 5 years.
The estimated interest rate is 6% per year. Calculate the present value of the amount
to be paid by the buyer.
Inserting the specified data into Equation (17.2) yields
P = 5, 000 (1 + 0.06)

$3736

Thus, the present value is $3736.

17.6.3 UNIFORM PERIODIC PAYMENT PRESENT VALUE FORMULA

In this case X amount of equal payments are made at the end of m interest periods
and all these payments are deposited at interest rate i compounded after each interest
period. Thus, the present value, PV, of all these payments is expressed by
PV = PV1 + PV2 + PV3 + + PVm
=

X
X
X
X
+
+
+ +
(1 + i) (1 + i)2 (1 + i)3
(1 + i)m

(17.3)

where
PVi is the present value of payment X made at the end of interest period i;
for i = 1, 2, - - - -, m.
In order to find the sum of the geometric series represented by Equation (17.3), we
multiply both sides of this equation by 1/(1 + i) to get
PV
X
X
X
X
=
2 +
3 +
4 + +
1 m+
(1 + i) (1 + i) (1 + i) (1 + i)
(1 + i)

(17.4)

Subtracting Equation (17.3) from Equation (17.4) yields

PV

(1 + i)

1999 by CRC Press LLC

PV =

(1 + i)m +1 (1 + i)

(17.5)

By rearranging Equation (17.5) we get

1 (1 + i) m
PV = X

(17.6)

Example 17.3
A machines expected useful life is 10 years and its use will generate an estimated
revenue of $20,000 at the end of each year. Calculate the present value of the total
revenue to be generated by the machine, if the annual interest rate is expected to be
5%.
Substituting the given data into Equation (17.6) yields
1 (1 + 0.05)10
PV = (20, 000)

0.05

$154, 435

Thus, the present value of the total machine revenue is $154,435.

17.6.4 UNIFORM PERIODIC PAYMENT FUTURE AMOUNT FORMULA

This represents the same situation as for Section 17.6.3 except that instead of finding
the total present value of the periodic payments, the total future amount is sought.
Thus, the total future amount, FAt, of all the periodic payments is given by
FA t = X + X (1 + i) + + X (1 + i)

m2

+ X (1 + i)

m 1

(17.7)

As Equation (17.7) is a geometric series, its sum can be found. Thus, we multiply
both sides of Equation (17.7) by factor (1 + i) to get

(1 + i) FA t = X (1 + i) + X (1 + i)2 + + X (1 +1 i)m + X (1 + i)m

(17.8)

Subtracting Equation (17.7) from Equation (17.8) yields

(1 + i) FA t FA t = X (1 + i)m X

(17.9)

By rearranging Equation (17.9) we get

FA t = X (1 + i) 1 i

1999 by CRC Press LLC

(17.10)

Example 17.4
In Example 17.3, instead of calculating the present value of the total income, find
the total future amount.
Substituting the given data into Equation (17.10) yields

FA t = (20, 000) (1 + 0.05) 1 0.05

10

$251, 558

Thus, the total future amount is $251,558.

17.7 LIFE CYCLE COST ESTIMATION MODELS

Over the years there have been many mathematical models developed to estimate
life cycle cost of various items. A life cycle cost estimation model of an item usually
requires inputs such as mean time between failures, mean time to repair, acquisition
cost, training cost, warranty coverage period, labor cost per corrective maintenance
action, average material cost of a failure, cost of carrying spares, cost of installation,
and time spent for travel [19]. Nonetheless, the life cycle cost estimation models
may be divided into two major classifications as follows:
General models
Specific models
Both these classifications are discussed below.

17.7.1 GENERAL LIFE CYCLE COST ESTIMATION MODELS

These models are not specifically tied to any particular system or equipment but
they may vary from one organization to another or within an organization. For
example, the life cycle cost estimation model used by the U.S. Navy is different
than the one used by the Army [14]. Nonetheless, the basic idea is to sum up all the
costs associated with an item over its life span. Three general life cycle cost estimation models are presented below [8, 11, 17].
General Life Cycle Cost Model I
In this case, the item life cycle cost is divided into two categories, i.e., recurring
cost (RC) and nonrecurring cost (NRC). Thus, the item life cycle cost (LCC) can
be expressed by
LCC = RC + NRC

(17.11)

In turn, RC has five major components: maintenance cost, operating cost, support
cost, labor cost, and inventory cost.

1999 by CRC Press LLC

Similarly, the elements of the NRC are procurement cost, training cost, LCC
management cost, support cost, transportation cost, research and development cost,
test equipment cost, equipment qualification approval cost, installation cost, and
reliability and maintainability improvement cost.
General Life Cycle Cost Model II
In this case, the equipment life cycle cost is categorized into four major classifications: research and development cost (RDC), production and construction cost
(PCC), operation and support cost (OSC), and disposal cost (DC). Thus, mathematically, the equipment LCC is given by
LCC = RDC + PCC + OSC + DC

(17.12)

The major elements of the RDC are engineering design cost, software cost, life
cycle management cost, product planning cost, test and evaluation cost, research
cost, and design documentation cost.
PCC is made up of five important elements: construction cost, manufacturing
cost, initial logistics support cost, industrial engineering and operations analysis
cost, and quality control cost.
The major components of the OSC are product distribution cost, product operation cost, and sustaining logistic support cost.
Finally, DC is expressed by
DC = URC + [CF () (IDC RV)]

(17.13)

where

CF
URC
IDC
RV

is
is
is
is
is

the
the
the
the
the

number of unscheduled maintenance actions.

condemnation factor.
ultimate retirement cost of the equipment under consideration.
equipment disposal cost.
reclamation value.

General Life Cycle Cost Model III

In this case, the equipment life cycle cost is made up of three major components:
initial cost (IC), operating cost (OC), and failure cost (FC). Thus, mathematically,
the equipment LCC can be expressed as follows [20]:
LCC = IC + OC + FC

(17.14)

The average cost of a failure is given by

AC = [HC (MTTR + RT) + ALC] + [LSC (MTTR + RT) + MAC
] + RPC

1999 by CRC Press LLC

(17.15)

where
RPC
ALC
MTTR
RT
HC
MAC
LSC

is the cost of replaced parts.

is the administrative loss cost due to a failure.
is the mean time to repair.
is the repaired item response time.
is the cost of the repair organization per hour.
is the mission-abort loss cost which occurs with each interruption of
service, and is also called the event-related, one-time cost.
is the loss of service cost per hour.

Thus, the failure cost is expressed by

FC = (AC) ( LT) MTBF

(17.16)

where
MTBF is the mean time between failures in hours.
LT
is the life time of the item in hours.
The failure loss cost rate per hour, FCR, is defined by
FCR = AC MTBF

(17.17)

Substituting Equations (17.16) and (17.17) into Equation (17.14) yields

LCC = IC + OC + ( FCR) ( LT)

(17.18)

17.7.2 SPECIFIC LIFE CYCLE COST ESTIMATION MODELS

These models are developed to estimate life cycle cost of specific items or equipment.
Three such models are presented below.
Switching Power Supply Life Cycle Cost Model
This was developed to estimate life cycle cost of switching power supplies [21]. The
life cycle cost is expressed by
LCC = IC + FC

(17.19)

FC + ( RC + SC)

(17.20)

In turn, FC is given by

1999 by CRC Press LLC

where

RC
SC

is
is
is
is

the
the
the
the

constant unit failure rate.

products expected life.
repair cost.
cost of spares.

The cost of spares is given by

SC = (USC) Q

(17.21)

where
USC is the unit spare cost.
Q
is the fractional number of spares for each active unit.
Inserting Equations (17.20) and (17.21) into Equation (17.19) yields
LCC = IC + [RC + (USC) Q]

(17.22)

Early Warning Radar Life Cycle Cost Model

This model was developed to estimate life cycle cost of an Early Warning Radar [8].
The life cycle cost of this equipment was divided into three parts: acquisition cost
(AC), operation cost (OC), and logistic support cost (LSC). Mathematically, this is
expressed as follows:
LCC = AC + OC + LSC

(17.23)

The predicted breakdown percentages of the LCC for AC, OC, LSC were 28%, 12%,
and 60%, respectively.
The four major components of the acquisition cost were fabrication cost, installation and checkout cost, design cost, and documentation cost. Their predicted
breakdown percentages were 20.16%, 3.92%, 3.36%, and 0.56%, respectively.
OC was made up of three principal elements (each elements predicted breakdown percentage is given in parentheses): cost of personnel (8.04%), cost of power
(3.84%), and cost of fuel (0.048%).
The logistic support cost was expressed by
LSC = RLC + ISC + AC + RMC + ITC + RSC
where
RLC is the repair labor cost.
ISC is the initial spares cost.

1999 by CRC Press LLC

(17.24)

AC
RMC
ITC
RSC

is
is
is
is

the
the
the
the

age cost.
repair material cost.
initial training cost.
replacement spares cost.

The predicted breakdown percentages of the LCC for each of these six cost elements
were 38.64%, 3.25%, 1.186%, 5.52%, 0.365%, and 11.04%, respectively.
Appliance Life Cycle Cost Model
This model is proposed to estimate life cycle cost of major appliances [22]. In this
case, the life cycle cost of an appliance is expressed by
n

LCC = APC +

FC (1 + FER)k
(1 + i)k

EC
k

k =1

(17.25)

where
APC is the appliance acquisition cost in dollars.
ECk is the appliance energy consumption in year k, expressed in million
BTUs.
n
is the appliance useful life in years.
FC is the fuel cost in year one, expressed in constant dollars per million
BTUs.
FER is the annual fuel escalation rate (%) in constant dollars.
i
is the discount rate (%) in constant dollars.
If ECk and FER are constant over the useful life span of the appliance, Equation
(17.25) becomes
FC (1 + FER)

(1 + i)
k =1
n

LCC = APC + ( EC) ( FC)

(17.26)

The typical useful life of appliances such as refrigerators, freezers, ranges and ovens,
electric dryers, and room air conditioners are 15, 20, 14, 14, and 10 years, respectively.
Example 17.5
A company using a machine to manufacture certain engineering parts is contemplating replacing it with a better one. Three different machines are being considered
for its replacement and their data are given in Table 17.1. Determine which of the
three machines should be purchased to replace the existing one with respect to their
life cycle costs.

1999 by CRC Press LLC

TABLE 17.1
Data for Three Machines Under Consideration
No.

Description

Machine A

Machine B

Machine C

1
2
3
4
5
6

Procurement price
Expected useful life in years
Annual failure rate
Annual interest rate
Annual operating cost
Cost of a failure

$175,000
12
0.02
5%
$5,000
$4,000

$160,000
12
0.03
5%
$7,000
$3,000

$190,000
12
0.01
5%
$3,000
$5,000

The expected failure costs of Machines A, B, and C, respectively, are given by

FC A = (0.02) ( 4, 000)
= $80
FCB = (0.03) (3, 000)
= $90
and
FCC = (0.01) (5, 000)
= $50
Using the given and calculated data in Equation (17.6), we obtain the following
present value of Machine A failure cost over its useful life:
1 (1 + 0.05)12
PVFC A = 80

0.05

$709

where
PVFCA is the present value of Machine A failure cost over its life span.
Similarly, the present values of Machines B and C failure costs over their useful
lives are as follows:

1999 by CRC Press LLC

1 (1 + 0.05)12
PVFCB = 90

0.05

$798

and
1 (1 + 0.05)12
PVFCC = 50

0.05

$443

The present values of Machines A, B, and C operating costs over their useful
life span, using Equation (17.6) and the given data, are
1 (1 + 0.05)12
PVOC A = 5, 000

0.05

$44, 316

1 (1 + 0.05)12
PVOCB = 7, 000

0.05

$62, 043

and
1 (1 + 0.05)12
PVOCC = 3, 000

0.05

$26, 590

where
PVOCA is the present value of Machine A operating cost over its useful life.
PVOCB is the present value of Machine B operating cost over its useful life.
PVOCC is the present value of Machine C operating cost over its useful life.
The life cycle costs of Machines A, B, and C are
LCC A = 175, 000 + 709 + 44, 316
= $220, 025
LCCB = 160, 000 + 798 + 62, 043
= $222, 841

1999 by CRC Press LLC

and
LCCC = 190, 000 + 443 + 26, 590
= $217, 033
As the life cycle cost of Machine C is the lowest, thus it should be purchased.

17.8 COST ESTIMATION MODELS

Over the years, various types of mathematical models have been developed to
estimate costs for use in LCC models. Many such models are presented in
Reference 8. This section presents three cost estimation models taken from published
literature.

17.8.1 COST-CAPACITY MODEL

This is a useful model to obtain quick approximate cost estimates for similar new
equipment, plants, or projects of different capacities. Mathematically, the costcapacity relationship is expressed as follows:
C n = Cod (KPn KPod )

(17.27)

where
Cn
KPn
Cod
KPod

is the cost of the new item under consideration.

is the capacity of the new item.
is the cost of the old but similar item.
is the capacity of the old but similar item.
is the cost-capacity factor and its frequently used value is 0.6. For items
such as pumps, heat exchanger, tanks, and heaters, the proposed values
of this factor are 0.6, 0.6, 0.7, and 0.8, respectively [8, 23-25].

Example 17.6
An electric utility spent $1.5 billion to build a 3,000 megawatt nuclear power
generating station. In order to meet the growing demand for electricity, the company
plans to construct a 4,000 megawatt nuclear station. Estimate the cost of the new
station, if the value of the cost-capacity factor is 0.8.
Substituting the given data into Equation (17.27) yields
C n = 1.5 ( 4, 000 3, 000)

0.8

= $1.89 billion
Thus, the new station will cost $1.89 billion to construct.

1999 by CRC Press LLC

17.8.2 MOTOR OPERATION COST ESTIMATION MODEL

This model is concerned with estimating operating cost of an alternating current
(AC) motor. Thus, the cost to operate a motor is expressed by [26]
Com = (0.746) ( hp) Tom K e

(17.28)

where
Com
hp
Tom
Ke

is
is
is
is
is

the
the
the
the
the

cost to operate the motor per year.

motor size in horsepower.
motor annual operating time in hours.
electricity cost expressed in dollars per kilowatt hour ($/KWH).
motor efficiency.

Example 17.7
A 40-hp AC motor is operated for 3500 h per year and the cost of electricity is
$0.08/kWh. Calculate the annual operating cost of the motor, if the motor efficiency
is 90%.
Substituting the given data into Equation (17.28) yields
Com = (0.746) ( 40) (3500) (0.08) 0.90
$9, 284

Thus, the annual operating cost of the motor is $9,284.

17.8.3 CORRECTIVE MAINTENANCE LABOR COST

ESTIMATION MODEL
This model is concerned with estimating the annual labor cost of corrective maintenance associated with an item when its (item) mean time between failures (MTBF)
and mean time to repair (MTTR) are known. Thus, the annual labor cost of the
corrective maintenance is expressed by
Ccm = (SOH) C ml (MTTR MTBF )
where
Ccm
SOH
MTTR
MTBF
Cm 

is
is
is
is
is

1999 by CRC Press LLC

the
the
the
the
the

annual labor cost of corrective maintenance.

scheduled operating hours of the item per year.
mean time to repair of the item.
mean time between failures of the item.
cost of the corrective maintenance labor per hour.

(17.29)

Example 17.8
An electric motor is scheduled for 4500 h of operation in one year and its estimated
MTBF is 1500 h. Whenever it fails, it takes on the average 15 h to repair. Calculate
the annual cost for the motor corrective maintenance labor, if the maintenance labor
cost is $30 per hour.
Inserting the given data into Equation (17.29) we get
Ccm = ( 4, 500) (30) (15 1500)
= $1350
Thus, the annual cost for the motor corrective maintenance labor is $1350.

17.9 LIFE CYCLE COSTING DATA

In order to perform an effective life cycle costing study, the availability of good cost
data are essential. This can only be achieved through the existence of good cost data
banks. Thus, whenever a cost data bank is being developed, careful attention must
be paid to factors such as flexibility, uniformity, ready accessibility, orientation,
expansion or contraction capability, size, responsiveness, and comprehensiveness
[27]. Furthermore, at a minimum, a life cycle costing database should include
information such as cost records, user pattern records, procedural records: operation
and maintenance, and descriptive records: hardware and site.
Nevertheless, prior to embarking on a life cycle costing study, factors such as
data availability, data applicability, data obsolescence, data bias, data comparability
to other existing data, and data orientation towards the problem must be carefully
examined.

17.9.1 COST DATA SOURCES

Even though there are many sources for obtaining cost related information for use
in life cycle cost analysis, their quality and amount may vary quite considerably.
Some of the sources for obtaining cost related data are as follows [8]:
American Building Owners and Mangers Association (BOMA) handbook
Unit price manuals: Marshall and Swift, Means, Dodge, Richardson, and
Building Cost File
Programmed Review of Information for Costing and Evaluation (PRICE)
Model [28]
Cost Analysis Cost Estimating (CACE) Model [28, 29]
Budgeting Annual Cost Estimating (BACE) Model [28, 29]
Cost for Pressure Vessels [30]
Costs for Motors, Storage Tanks, Centrifugal Pumps, etc. [31, 32]
Costs for Heat Exchangers [33-35]
Costs for Varieties of Process Equipment [36-39]
Costs for Solid Waste Shredders [40].

1999 by CRC Press LLC

17.10 LIFE CYCLE COSTING ADVANTAGES AND

DISADVANTAGES, RESISTANCE FACTORS
AND ASSOCIATED IMPORTANT POINTS
There are many advantages to using the life cycle costing concept, including useful
to control programs, making effective equipment replacement decisions, making a
selection among the competing contractors, useful in reducing the total cost, and
comparing the cost of competing projects.
In contrast, some of the disadvantages of life cycle costing are time consuming,
expensive, and doubtful data accuracy.
Despite many advantages of the life cycle costing concept, there are many factors
that still resist its usage. These factors include failure of some past procurement
policies resembling life cycle costing, doubts about the life cycle costing methodology, data accuracy, data reliability, and contractor reluctance to guarantee concerned estimates [9]. Nonetheless, some of the important points associated with life
cycle costing are as follows [8, 9]:
Life cycle costing objective is to obtain maximum advantages from limited
resources.
Expect some surprises, irrespective of the excellence of the life cycle cost
estimator.
Life cycle costing is most beneficial if it is started early in the program.
The group performing life cycle costing should be part of the finance or
logistics organization.
Ensure that all costs associated with the program are included in the life
cycle cost model.
The risk management is the essence of life cycle costing.
Both the manufacturer and the user must organize effectively to control
life cycle cost.
Various data related difficulties may be overcome by a highly competent
cost analyst.
Perform trade-offs between life cycle cost, performance, and design to
cost throughout the program.
For good life cycle cost estimates, good data are indispensable.
Management is key to making the life cycle costing effort worthwhile.
In the event of life cycle costing based acquisition, ensure factors such
as existence of written performance requirements, capable buyer and seller
management to use life cycle cost results, sufficient time to conduct
analysis, and availability of good historical cost data.

17.11

PROBLEMS

1. What are the advantages and disadvantages of life cycle costing?

2. Describe the steps associated with life cycle costing.
3. What are the reasons and uses of the life cycle costing concept?

1999 by CRC Press LLC

TABLE 17.2
Data for Engineering Systems Offered by Two Different Manufacturers
No.

Description

Manufacturer As system

Manufacturer Bs system

1
2
3
4
5
6

Purchasing price
Cost of failure
Annual interest rate
Annual operating cost
Annual failure rate
Expected useful life in years

$80,000
$10,000
6%
$4,000
0.004
15

$60,000
$5,000
6%
$5,500
0.004
15

4. What are the inputs required to conduct life cycle costing?

5. What are the life cycle cost associated activities?
6. The salvage value of a computer system after a 10-year period will be
$1500. Calculate the present value of this salvage value if the estimated
annual interest rate is 5%.
7. A company plans to purchase an engineering system and two manufacturers are bidding for the system. Table 17.2 presents data associated with
each manufacturers system. Determine the system to be procured with
respect to life cycle cost.
8. A 50-hp AC motor is operated for 2500 h per year and the electricity cost
is $0.07/kWh. Calculate the annual operating cost of the motor if its
efficiency is 85%.
9. List at least 10 important points associated with life cycle costing.
10. What are the important factors to be considered in developing a cost data
bank for use in life cycle costing studies?

17.12

REFERENCES

1. Ryan, W.J., Procurement views of life cycle costing, Proc. Annu. Symp. Reliability,
164-168, 1968.
2. Wienecke-Louis, E. and Feltus, E.E., Predictive Operations and Maintenance Cost
Model, Report No. ADA078052, 1979. Available from the National Technical Information Service (NTIS), Springfield, VA.
3. Life Cycle Costing in Equipment Procurement, Report No. LMI Task 4C-5, prepared
by Logistics Management Institute (LMI), Washington, D.C., April 1965.
4. Life Cycle Costing Procurement Guide (interim), Guide No. LCC1, Department of
Defense, Washington, D.C., July 1970.
5. Life Cycle Costing in Equipment Procurement-Casebook, Guide No. LCC2, Department of Defense, Washington, D.C., July 1970.
6. Life Cycle Costing for System Acquisitions (interim), Guide No. LCC3, Department
of Defense, Washington, D.C., January 1973.
7. Dhillon, B.S., Life cycle cost: A survey, Microelectronics and Reliability, 21, 495-511,
1981.
8. Dhillon, B.S., Life Cycle Cost: Techniques, Models, and Applications, Gordon and
Breach Science Publishers, New York, 1989.

1999 by CRC Press LLC

9. Robert-Seldon, M., Life Cycle Costing: A Better Method of Government Procurement,

Westview Press, Boulder, CO, 1979.
10. Lamar, W.E., Technical Evaluation Report on Design to Cost and Life Cycle Cost,
North Atlantic Treaty Organization (NATO) Advisory Group for Aerospace Research
and Development (AGARD) Advisory Report No. 165, May 1981. Available from
the National Technical Information Service (NTIS), Springfield, VA.
11. Reiche, H., Life cycle cost, in Reliability and Maintainability of Electronic Systems,
Arsenault, J.E. and Roberts, J.A., Eds., Computer Science Press, Potomac, MD, 1980,
pp. 323.
12. Brown, R.J., A new market tool: Life cycle costing, Ind. Market. Manage., 8, 109-113,
1979.
13. Coe, C.K., Life cycle costing by state governments, Public Administration Rev.,
September/October 1981, pp. 564-569.
14. Eddins-Earles, M., Factors, Formulas, and Structures for Life Cycle Costing, EddinsEarles, Concord, MA, 1981.
15. Bidwell, R.L., Checklist for Evaluating LCC Program Effectiveness, Product Engineering Services Office, Department of Defense, Washington, D.C., 1977.
16. Paul-Degarmo, E., Canada, J.R., and Sullivan, W.G., Engineering Economy, Macmillan Publishing Co., New York, 1979.
17. Dhillon, B.S., Reliability Engineering in Systems Design and Operation, Van Nostrand Reinhold Company, New York, 1983.
18. Newman, D.G., Engineering Economic Analysis, Engineering Press, San Jose, CA,
1988.
19. Siewiorek, D.P. and Swarz, R.S., The Theory and Practice of Reliable System Design,
Digital Press, Digital Equipment Corporation, Bedford, MA, 1982.
20. Dickinson, D.B. and Sessen, L., Life cycle cost procurement of systems and spares,
Proc. Annu. Reliability Maintainability Symp., 282-286, 1975.
21. Monteith, D. and Shaw, B., Improved R, M, and LCC for switching power supplies,
Proc. Annu. Reliability Maintainability Symp., 262-265, 1979.
22. Turiel, I., Estrada, H., and Levine, M., Life cycle cost analysis of major appliances,
Energy, 6, 945-970, 1981.
23. Desai, M.B., Preliminary cost estimating of process plants, Chem. Eng., July 1981,
pp. 65-70.
24. Dieter, G.E., Engineering Design, McGraw-Hill, New York, 1983.
25. Jelen, F.C. and Black, J.H., Eds., Cost and Optimization Engineering, McGraw-Hill,
New York, 1983.
26. Brown, R.J. and Yanuck, R.R., Life Cycle Costing: A Practical Guide for Energy
Managers, Fairmont Press, Atlanta, GA, 1980.
27. Bowen, B. and Williams, J., Life cycle costing and the problems of data, Industrial.
Forum, 6, 21-24, 1975.
28. Marks, K.E., Garrison-Massey, H., and Bradley, B.D., An Appraisal of Models Used
in Life Cycle Cost-Estimation for U.S. Air Force (USAF) Aircraft Systems, Report
No. R-2287-AF, October 1978. Prepared by the Rand Corporation, Santa Monica, CA.
29. USAF Cost and Planning Factors, Report No. AFR 173-10, Department of the Air
Force, Washington, D.C., February 6, 1975.
30. Mulet, A., Corripio, A.B., and Evans, L.B., Estimate cost of pressure vessels via
correlations, Chem. Eng., 88(20, 456), 1981.
31. Corripio, A.B., Chrien, K.S., and Evans, L.B., Estimate costs of heat exchangers and
storage tanks via correlations, Chem. Eng., 89(2, 125), 1982.

1999 by CRC Press LLC

32. Corripio, A.B., Chrien, K.S., and Evans, L.B., Estimate costs of centrifugal pumps
and electric motors, Chem. Eng., 89(4), 115, 1982.
33. Purohit, G.P., Cost of double pipe and multi-tube heat exchangers, Chem. Eng., 92,
96, 1985.
34. Woods, D.R., Anderson, S.J., and Norman, S.L., Evaluation of capital cost data: Heat
exchangers, Can. J. Chem. Eng., 54, 469, 1976.
35. Kumana, J.D., Cost update on specialty heat exchangers, Chem. Eng., 91(13), 164,
1984.
36. Hall, R.S., Mately, J., and McNaughton, K.J., Current costs of process equipment,
Chem. Eng., 89(7), 80, 1982.
37. Klumpar, I.V. and Slavsky, S.T., Updated cost factors: Process equipment, commodity
materials, and installation labor, Chem. Eng., 92(15), 73, 1985.
38. Humphreys, K.K. and Katell, S., Basic Cost Engineering, Marcel Dekker, New York,
1981.
39. Peters, M.S. and Timmerhaus, K.D., Plant Design and Economics for Chemical
Engineers, McGraw-Hill, New York, 1980.
40. Fang, C.S., The cost of shredding municipal solid waste, Chem. Eng., 87(7), 151,
1980.

1999 by CRC Press LLC