������������

������������
����������������������������������� ��

������������������������
��������������������������������������������������������������������������������������������������������

�
2

���� � � � � � ���� � � � � � � � �������

����� � ������ � ���������� � � ������ � � ����������� � � �������������

��

����� � ������ � � � � � ��������� � � � � � � � � � ������������ � � � � �������������

�
3 Reliabilty Engineering

�����������������������������������������������������������������������������������������������
F O R E W O R D ...................................................... 11
Nomenclature .............................................................. 13

Chapter�� 1 Introduction and Basic Concepts ........ 16

Chapter��� 2 Static Reliability Models .................... 136

Chapter��� 3 Reliability in Design + UGF Technique 200

Chapter���� 4 Structural Reliability Analysis ........... 230

Chapter �5 On the combinations of random variables in design ; 282

A glance at the tolerance concept ................................ 282

Chapter �6 Estimation of Mean Lifetime &Reliability+ Experiments

&Tests ......................................................................... 311

Chapter �7 Dynamic Models+ Availability, Application of Markov Chain

..................................................................................... 355

Chapter �8 Enhancement, Optimization & Allocation of Reliability 426

References ................................................449
T A B L E S .........................................455
�����������������������������������������
�

�
4

�
5 Reliabilty Engineering

C O N T E N T S
F O R E W O R D ...................................................... 11
Nomenclature .............................................................. 13

Chap 1 Introduction and Basic Concepts .............. 16

Aims of the chapter ..................................................... 16
1.1 Introduction ........................................................... 16
1-2 Failure ................................................................... 17
1-3 Reliability.............................................................. 17
1-3-1 abbreviations ----------------- 20
1-4 System reliability function: R(t)............................ 22
1-5 Calculation of average lifetime ............................. 23
1-6 Failure rate ............................................................ 26
1-6-1 hazard function for continuous life time distribution 26
1-6-2 Necessary condition for being a hazard function 28
1-6-3 Hazard and relia. functions for discrete distributions ..29
1-6-4 Calculation of� , , , given any of them 30
1-7 The pdf of a part of a distribution ....................... 31
1-8 Some continuous distributions used in reliability theory.. 32
1-8-1 Exponential distribution --------------- 32
1-8-2 Normal(Gaussian) distribution 33
1-8-3 Truncated normal distribution -------- 35
1-8-4 Log- normal distribution(μ,σ) -- 36
1-8-4-1 Calculation of parameters ---- 37
1-8-4-2 Relationship between lognormal & normal 37
1-8-6 Uniform distribution --------------- 41
1-8-6 Weibull distribution ------------------- 42
1-8-7 GEV distribution ------------------------ 46
1-8-8 Gamma distribution ---------------------- 46
1-8-8-1 Erlang distribution ----------------- 47
1-9 Bathtub curve hazard function .............................. 50
1-9-1 Some forms of hazard functions - 53
1-10 Some discrete distributions ................................. 56
1-10-1 Geometric distribution ----------- 56
1-10-2 Binomial distribution ------------ 58
1-10-2 Poisson distribution ------------- 59
1-11 Accelerated life testing + parametric and non-param. relia
analysis ........................................................................ 62
1-11-1 Accelerated life testing(ALT) ------ 62
1-11-2 Parametric reliability analysis ------ 63
1-11-3 Non-parametric reliability analysis 63
1-11-3-1 Non-parametric Estimation of , , , ......63
1-12 pdf &CDF of sample minimum ......................... 77
1-12-1The CDF of the minimum of samples of largish size 79
Fisher -Tippet Theorem ............................................... 91
1-13 Bartlett's goodness of fit test for exponential distribution 93
6

1-14 Q-Q plot ............................................................. 96

1-15 Convolution........................................................ 102
1-15-1 CDF and pdf of sum of variables X and Y 102
1-15-2 n-fold convolution of f with itself 108
1-16 The pdf of the difference of 2 independent random variables 110
1-17 Percentage of a distribution being outside limits 111
Appendix 1 Parameter Estimation Techniques ......... 113
Maximum Likelihood Estimation Method.... ...114
Methods of Moments(MOM) ----------------- ----- 121
Appendix 2: Application of MATLAB in Reliability theory 124
Exercises ..................................................................... 132

Chap 2 Static Reliability Models ............................ 136

Aims of the chapter ..................................................... 136
2-1 Definition of static reliability models ................... 136
2-2Reliability Block Diagram ..................................... 137
2-3 series configuration ............................................... 137
2-4 Parallel configuration ............................................ 140
2-4-1 Types of parallel configurations ----- 145
2-4-1-1 Two-component system with 1 active and 1 standby-
Perfect switching ---------------------------------- 146
2-4-1-2 n-component system with 1 active and n-1 standby-
Perfect switching ---------------------------------- 147
2-4-1-3 Two-component system with 1 active and 1 standby-
Imperfect switching ------------------------------ 149
2-5 Combination of Series and Parallel Configurations 150
2-5-1 Redundancy Level -------------------- 151
2-6 k-out-of-n configuration ....................................... 157
2-6-1 Reliability of k-out-n configuration 158
2-6-1-1 Upper bound for k-out-of-n reliability 164
2-7 Complex System Analysis .................................... 166
2-7-1 Conditional Probability Method ---- 167
2-7-2 Delta-Star Transformation Approach for Reliability
Evaluation --------------------------------------------- 170
2-7-2-1 Transforming a delta configuration into an equivalent
star configuration --------------------------------- 171
2-7-2-3 Transforming a star configuration into an equivalent delta
configuration ------------------------------------------ 173
2-8 Calculation of upper and lower bounds for complex system using
cut and tie set analysis ................................................. 178
2-8-1 Calculation of reliability upper& lower bounds for complex
systems using auxiliary networks ----------------- 183
2-8-2 An approximate formula for the upper and Lower reliability
bounds of complex systems ------------------------ 187
2-9 Applications of Bays reliability in Design ............ 190
Exercises ..................................................................... 193

�
7 Reliabilty Engineering

Chap 3 Reliability Considerations in Design + UGF Technique 200

Aims of the chapter ..................................................... 200
3.1 Reliability Considerations in Design ..................... 200
3.1-1Reliability considerations in series configurations 201
3.1-2 Reliability considerations in parallel configurations 203
3.1-3 Reliability considerations in series-parallel configurations
----------------------------------------------------------- 206
3-2Universal Generating Function(UGF) analysis of Reliability Systems 211
3.2-1 Moment generating function of discrete random variables 212
3.2-2 Z-transform or probability generating function of discrete
random variables ------------------------------------- 213
3-2-3 The Universal Generating Function(UGF) 215
3-2-4 derivation of the reliability using UGF 222
3-2-4 Reliability Analysis of Binary -State Systems using UGF 223
Exercises ..................................................................... 227

Chap 4 Structural Reliability Analysis ................. 230

Aims of the chapter ..................................................... 230
4.1 Introduction ........................................................... 230
4.2Load-strength Interference Analysis ...................... 231
4-3 System reliability -Load & Strength variable ...... 233
4-3-1 Definition of safety margin(SM)------- 234
4-3-2 Reliability Computation for Probabilistic independent load
and strength ------------------------------------------- 244
4-3-3 Definition of Loading Roughness----- 249
4.3-4 Effect of Safety Margin and Loading Roughness on
Reliability (Multiple Load Applications) -------- 253
4-4 Calculation of structures' reliability Load or Strength deterministic 255
4-5Interrelation between reliability(R) and safety factor(SF) 259
4-6 Determining the structural reliability bounds using nonlinear
programming(NLP) ..................................................... 261
4-6-1 The algorithm for Reliability Lower & Upper Bounds using
NLP ----------------------------------------------------- 264
Appendix : Other definition of safety margin(SM) and its relationship to
safety factor(SF) .......................................................... 275
Exercises ..................................................................... 276

Chap 5 Combinations of variables in design +a glance at tolerance 282

Aims of the chapter ..................................................... 282
5.1Introduction ............................................................ 282
5.2 Certain properties of a function of some random variables 285
5.2.1 The pdf of a function of one random variable 285
5-2-2 Mean of 2 random variables ------------ 288
8

5-2-3 Variance of sum and difference of 2 random variables 288

5-2-3-1 Variance of sum of 2 independent random variables 289
5-2-4 Approximating mean and variance of a function of a random
variable ------------------------------------------------- 289
5-2-5 Approximating the mean of a function of some independent
random variables ------------------------------------- 290
5-2-6 Approximating the variance of a function of some
independent random variables --------------------- 290
5-2-7Approximating the mean and variance of 292
5.3 Statistical Tolerance ............................................ 293
5-3-1 Relationship of assembly tolerance parts tolerance 294
5-3-2 Tolerance in complex systems --------- 297
Exercises ..................................................................... 303

Chap 6 Estimation of Mean Lifetime &Reliability and 311

Related Experiments &Tests ....................................... 311
Aims of the chapter ..................................................... 311
6.1 Introduction ........................................................... 311
6.2 Estimation of product mean life given a lifetime sample of size n 312
6.3 Tests for Estimating Mean Life ............................ 312
6.3.1 Time censoring (Type-I) ----------------- 313
6.3.2 failure censoring (Type-II) -------------- 313
6.4 Estimation of mean life ......................................... 314
Calculation of T for = = = 315
6-4-1 Type I censoring life test --------------- 316
6-4-1-1Type I censoring life test with replacement 316
6-4-1-2Type I censoring life test without replacement 318
6-4-2 Type -II censoring life test ------------- 320
6-4-2-1Type II censoring life test without replacement 320
6-4-2-2Type II censoring life test with replacement 321
6-5 On the Accelerated life testing(ALT) .................. 323
6-6 Confidence interval for mean lifetime-exponential distribution case 324
6-6-1 Lower-bound confidence interval(CI) for mean of
exponential distribution ----------------------------- 330
6-6-2 The confidence interval for the time during which fraction p
of exponentially-distributed-life products fail --- 330
6-7 Reliability Acceptance Sampling Plans ................ 331
6-7-1Type I&II errors of Sampling plans 333
6-7-2 Design of single plans using Table 6-2 333
6-7-3 The operating characteristic (OC) curve of single sampling
plans ---------------------------------------------------- 336
6-7-3-1 OC curve for single sampling plans 336
6-8 statistical hypothesis testing on mean and minimum lifetime 341
6-8-1 Test of hypothesis on minimum life of an exponentially
distributed lifetime(K&L page263) --------------- 342

�
9 Reliabilty Engineering

6-8-2 Test of hypothesis on mean lifetime and failure rate

concerning exponential distribution(K&L page263) 344
Exercises ..................................................................... 348

Chap 7Dynamic Models+ Availability, Applic. of Markov Chain 355

Aims of the chapter ..................................................... 355
7.1 Dynamic Models in Reliability ............................ 355
7-1-1 Series Systems --------------------------- 355
Mean lifetime and reliability function of a series system of
identical components with failure rate --------- 359
7-1-2 Series Chain Model or Weakest Link Model 359
7-1-3 Parallel systems --------------------------- 364
7-1-3-1 Reliability in active redundancy - 365
7-1-3-2Reliability and MTBF in active system with
exponentially- distributed-lifetime components 366
7-1-3-3 Reliability of standby parallel system 370
Life pdf & reliability for Perfect switch: 2-component standby
system----------------------------------------------- 382
Special case: Constant failure rate & perfect switch( = ) 383
Life pdf & reliability for Perfect switch: 3-component standby
system----------------------------------------------- 384
Reliability of n-component standby system :Perfect Switch 385
Imperfect switch ,2-component standby system 387
2-component standby , constant failure rate - 387
Imperfect switch with reliability ---------- 387
2-component standby , constant failure rate - 388
Imperfect switch with reliability function ( ) 388
2-component standby system with similar units, imperfect switch
and warm redundant unit ---------------------------- 394
7-1-4 Shared load parallel configuration ---- 399
Special case:2-component constant failure rates 400
7-2 System Effectiveness Measures ........................... 402
Serviceability ------------------------------------------------ 402
Maintainability ---------------------------------------------- 403
Maintainability function ----------------------- 403
Operational readiness(OR) -------------------- 405
7-3 Availability ........................................................... 407
7-3-1 Intrinsic Availability --------------------- 407
7-3-2 Availability function --------------------- 408
7-3-3 availability function for nonrepairable systems 410
7-3-4 Steady state availability ----------------- 410
7-3-5 Intrinsic availability in long rage ------ 413
7-3-6 Mission Availability --------------------- 415
7-3-7 System Availability in terms of components' Availability 417
7-3-8 The steady-state availability in Preventive Maintenance419
7-3-9 Definition of Unavailability Function 420
10

7-4Application of Markov Chains to System Reliability Analysis 421

Exercises ..................................................................... 424

Chap 8 ...................................................................... 426

Enhancement, Optimization & Allocation of Reliability 426
Aims of the chapter ..................................................... 426
8-1 Enhancement(Improvement) of system reliability 426
8-1-1 Improving Component reliability ------ 426
8-1-2 Hot and standby redundancy----------- 427
8-1-2 Active redundancy ---------------------- 428
8-1-2 Standby redundancy -------------------- 428
8-2 Reliability Optimization ........................................ 430
8-2-1 Methods for the solution of The above problems 434
8-3��Reliability Allocation ........................................ 435
8-3-1 Equal Apportionment Technique ------ 438
8-3-2 The ARINC technique ------------------ 441
8-3-3 The AGREE allocation method ------ 443
Exercises ..................................................................... 448
References ........................................... 449
Table A������Crtical values of F distribution Fm,n ............ 456
Table �B Values of CDF of Poisson Distribution ..... 460
Table D Critical values of standard normal ( ) .... 467
Table E Critical values of Chi2 Distribution-----.... 469
Table F Chracteristic continuous distributions ....... 471
Table G Chracteristic of discrete distributions ...... 474
Table H MATLAB commands for distributions ..... 475
�

�
11 Reliabilty Engineering

�����������������
This book, whose Persian version written by the same author
is published by our univerisity, is the outcome of teaching a
course on reliability for several years to graduate students
using many books especially the book written by Dr Kapur and
Lamberson (abbreviated by K&L throughout the book) . The
main prerequisite for understanding the materials of the book is
probability. As evident from the chapter titles, the book
introduces readers with reliability and availability of products. I
would like to the students who helped the author in some phases
of editing.
At the end of this foreword a software and� a symposium are
introduced.
Software
ReliaSoft's reliability software tools facilitate a set of
reliability engineering modeling and analysis techniques.
(downloable from http://www.reliasoft.com)
A symposium: RAMS associated to IEEE
Reliability and Maintainability www.rams.org
The proceedings are available from
http://ieeexplore.ieee.org/xpl/mostRecentIssue.jsp?punumber=6516162
It is suggested to the readers, especially those working in
industry, to read books on design for reliability after reading this
book.
Thanks God for making me successful to present this work
which I hope to be useful in both academic and industrial
environments.
The author would be pleased if the readers write him about
any kind of deficiencies in the book.
Hamid Bazargan
Jan 2023
College of Engineering,
Shahid Bahonar University of Kerman, Iran
[email protected]
The wise is one who puts everything
in its right place
 12

������������� �������������
Symbols �� Description ��

A,B,C n� 1)sample size 2)the number

of components in a series or
parallel configuration
A 1)steady-state
� Average safety factor
availability
2)Location parameter Number of components in
of Weibull distribution�
�
ith subsystem
Initial size of the sample
� Intrinsic Availability
�t) The number of survivor
a.r.t actual repair time components (or the number
still working adequately)at
B 1)a parameter of some time t
distributions (Weibull ,
GEV, ��� Operational readiness
GPD) 2) Bartlet's test
C t ti parameter
1)A ti of some
distributions (Weibull ,
P,Q
GEV,
GPD) p Failure probability of a
device
CDF Cumulative
distribution
function
pdf Probability density function

The probability that a

P
F s� switch operates well

Fx CDF of random P� The probability of

variable X a accepting a lot by a
FY CDF of random sampling plan
variable Y
� pdf of system's lifetime Pi (t ) The probability that the
� system is in state i at time t
� pdf of random variable
X Fraction defective in a
� population
CDF of strength
F Q Failure (unreliability)
probability of a device
pdf of stress(load)
f S (s ) Failure probability of ith
component
CDF of strength
F (s ) R
f  ( )
pdf of strength
∗( )� The optimum reliability of
the system at time t
� ( )
pdf of sample min.
The required reliability of
� ∗F convolution of FX,FY the system
∗ The reliability allocated to
critical value of F �
F,mn, distribution for n &m
ith unit
The given reliability for ith
degrees of freedom at unit
given sig. lev. α n The reliability of n-
R (t) component system(a active)
F critical value of F sy s
, 2 r 2  2 , 2 r1  2
2 distribution for 2 The reliability of 2-
2 r2  2 & 2 r1  2 R component system
deg. of f. at lev. α/2
standby
� ( 1 active 1 standby)
2 The reliability of 2-

H,L,N,O R
active �
component system( both
active)
ℎ( )� Instantaneous failure rate
������ The reliability function of
function the switch
���� Lower Specification Level System reliability function
�������

�
13 Reliabilty Engineering

������������� �������������
� ) The reliability function of The failure rate of
ith unit component a
� The failure rate of
component b�

S,T �
In load sharing:the failure
rate for the component
S� Stress, Load
under half load�
Smax� Maximum of stress(load),
upper extreme value of S �
In load sharing:The failure
rate for the component
under full load�
Smean Average stress value
The failure rate of the
�
Smin Minimum of stress(load),
lower extreme value of S
switch
The failure rate of the
t* Duration of experiment
� standby component
while in standby mode
T� Total operating time of all

items under the lifetest The mean of a population

� Average load
U,V,W,X Average stength
The standard devaition of
U� The upper specification
level of a dimention of a
� apopulation
product Standard devation of load
sistribution
� − ) The difference of U al L
standard devation of
in a part
strength distribution�
δ lower bound of strength
� − ) � The difference of U al L in
the assembled unit
σ part � the standard devaition of a
� � Coefficient of variation of measurement in a part
safety factor(SF)
In AGREE Method: the standard devaition of a
�
The importance factor
of the ith subsystem
 sum � measurement
in the assembly

The minimum of a sample

� ) the CDF of standard
The ith value of an ordered � normal distribution
�) sample
The ith value of a sample;
moment gegerating
The occurrence time of the
function
ith Failure; the ith failure
of randdom varaible X
time of the device
�The critival value of chi-
r
Note :  X (i )  X i
r
2,n � squared distribution wth
i 1 i 1
degre. free=n and sig .
level = α

α, β, …  ��  composition operator
acting over a function
(in UGF Technique)
 � The probability of Type I
End��� Eexample
error
End of theorem ; End of
 The probability of Type II ������� proof
error
�����
The strength of the ith link
δ� (component)

� �
 14

� �

�
15 Reliabilty Engineering

Chapter 1
Introduction
&
Basic
Concepts
Chap 1 Introduction and Basic Concepts 16

1
Introduction and Basic Concepts

Aims of the chapter

This chapter is concerned with definitions and basic concepts

needed in a reliability course such as MTTF,MTTR, reliability
function, hazard function and their estimation. Bathtub curve,
cumulative distribution function of extreme values of samples
are also discussed and a goodness of test for exponential
distribution is explained.

1.1 Introduction

In general, the reason one is concerned with the reliability of

components of electrical and mechanical systems is to ensure
that the systems will be reasonably free from failure(Grant
&Leavenworth, 1988 page 606). Failure of products could
incur a great loss or lead to personal injury, or severe physical
or environmental damage and even lead to death, This proves
the importance of product reliability in various fields including
air-space. According to Bazovsky(2004) "reliability has added a
new dimension to quality control work without subtracting
anything from traditional quality control work and methods."

�
17 Reliabilty Engineering

Here the quality characteristic is life. Gathering of lifetime data

is often expensive and its statistical analysis is an important
topic in reliability engineering. It is reminded that control
charts such as �chart and p-chart could be constructed using
lifetime data but these charts, despite their effectiveness, do not
answer such questions as what percent of the products live more
than 1000 hours with 90% of probability. Before defining the
term reliability let us define the term failure.

1-2 Failure

American National Standard defines failure as "The

termination of the ability of any item to perform its required
function (IIE Terminology page 8-9).

1-3 Reliability

In general reliability is the ability of a device, a system or a

unit to perform a function or some required functions without
any failure under some conditions for a stated amount of
operation. The amount of operation could be expressed in time,
kilometers, working cycles, number of times it operates��

When defining this characteristic by such terms as "assessed

reliability" and "predicted reliability" the following is useful
Chap 1 Introduction and Basic Concepts 18

IIE terminology defines this term as " the probability that an

item will perform a required function under stated conditions
for a stated period of time" (IIE Terminology page 8-22).

The reliability is sometimes expressed as a success ratio.

The objectives of Reliability Engineering

The Reliability Engineer must1

 Apply engineering knowledge and techniques to reduce

the occurrence of failures
 Determine the cause of each failure and make necessary
adjustments to correct the issue or completely address
the root cause
 Identify different ways to address failures should the root
cause prove uncorrectable
 Do reliability estimations for new designs and
continually analyze reliability data

Moreover reliability engineers check new installations to

ensure they adhere to functional specifications. They guide users
to ensure the reliability and maintainability of equipment,
processes, utilities, facilities, controls, and safety/security
�����������������������������������������������������������

1
From (https://www.techslang.com/definition/what-is-reliability-engineering/ )

�
19 Reliabilty Engineering

systems. That includes helping them come up with asset

maintenance and risk management plans.

Reliability engineers develop solutions to repetitive failures

and all other problems that adversely affect the users��
operations. They work with production teams to analyze assets��
performance.

Overall, reliability engineering can minimize failures,

enhance effectiveness, reduce repair times, streamline
maintenance processes, and offer protection against injuries and
death.( End of quotation from https://www.techslang.com/definition/what-is-reliability-engineering/)

Also reliability engineers must

 Be able to enhance and optimize systems' reliability

Reliability Importance: The Reasons

Some of the reasons why product reliability is important are:

-Greater safety for industries such a space industry

-Greater product reliability causes more reputation

-Customers' request and consent

-Although greater reliability incur a higher cost, but the

overall cost including that of maintenance and repair is less.
Chap 1 Introduction and Basic Concepts 20

It is worth mentioning that reliability theory has application

to many fields including air-space industry, home appliances,
transportation, buildings, and electronic industry.

1-3-1 abbreviations

TTF Time To Failure,

TTR Time To Repair, repair time
TBF time between 2 successive failures( for repairable devices)

MTTF Mean Time to failure

MTTR Mean Time to Repair
MTBM Mean time between maintenance, The average length
of time between one maintenance action and another
for a component
MTFF Mean time to first failure
MTBF Mean time between failures( for repairable devices)

Note that TBF is equal to the sum of TTR and TTF (Fig.1.1).

TBF=TTR+TTF (1-1-1)

Fig. 1.1 Graphical representation of Eq. 1.1.

�
21 Reliabilty Engineering

Taking the average of both sides of Eq. 1-1-1 yields:

MTBF=MTTR+MTTF (1-1-2)

If the probability distribution function or the cumulative distribution

function of the time between failures is not known, mean time
between failures(MTBF)could be estimated from (Tersine, 1985,
p202):

WT
M T�BF  (1-2)
N

Where N is the number of failures during the time WT.

Mean time to failure (MTTF) and mean time to repair(MTTR)

could be estimated in a similar manner.

In the continuation of this section, some terms used for

measuring reliability such as reliability function, mean life time,
hazard function or failure rate function are described. It is worth
remembering that since we accept that in a population of a
product, the products fail in different times, even if they work
under the same conditions, it is concluded that the failure
phenomena has to be treated statistically. That is why the
definition of reliability basics concepts is based on Profanity
theory.
Chap 1 Introduction and Basic Concepts 22

1-4 System reliability function: R(t)

Reliability function for devices with continuous lifetime

distribution is defined as:

∞
���� = ���X > ) = ���X ≥ t) = ∫ f(x)�� − � − �(x) (1-3)

where

( ) is the probability density function(pdf) of time to failure

(X=TTF),

( ) is the cumulative distribution function of lifetime(X),

Reliability function for devices with discrete lifetime

distribution is defined as:

� �� ( > ) , k = 1, 2, . . .. (1-4-1)
( ) = ∑∞ ( ), (1-4-2)

Where ( ) is the failure probability at time j.

For example if the failure probability at any time is p and the

distribution is geometric, then the reliability of time j is:

( ) = (1 − ) , k = 1, 2, . . .. (1-5)

If k is largish and p is small, then:

�
23 Reliabilty Engineering

� )
( )≅ . (1-6)

1-5 Calculation of average lifetime

If the lifetime is a continuous random number with

probability density function f(x) of cumulative distribution F(x),
its average is calculated from:

∞
( )=∫ ∞
( ) . (1-7)�

It is proved that for a continuous distribution:

∞
( ) = ∫ [1 − ( )] −∫ ∞
( ) (1-8)

And since lifetime(X) does not accept negative values, then

lifetime average could be calculated from:

∞
( ) = ∫ [1 − ( )] (1-9)

or from :

( )=∫ ( ) . (1-10)

It worth noting that 2 systems with equal lifetime average

might have different reliability.
Chap 1 Introduction and Basic Concepts 24

Example 1-1

If X with normal distribution ~ ( = , ) and Y

with exponential distribution ~ = =

represent the lifetimes of 2 products, find the average lifetime
and the reliability of each product for a mission time equal to
= .

Solution

X is normally distributed; then

E(X) = μ = ���� , R(t = ����) = ��(X > μ) =

Y is exponentially distributed, then:

( )= = , ( )= ≅ 0.��

Example 1-2

The life time of a critical component is exponentially

distributed with parameter λ� If the component fails or if its
lifetime reaches T it is replace with a ne one. How much time
on the average is needed to replace the component?

�
25 Reliabilty Engineering

Solution

Let

X=the life time of the component and

Y= the replacement time.

� = ������ �) ⟹ E(Y) = E[������ �)]

Min (X,T)+Max(X,T)=X+T⟹

E[Min (X,T)] =E(X+T)-E[ Max(X,T)]

�(X + T) = E(X) + T

E{����X, T)} = E(����X, T)|X > )���X > )

+ E(����X, T)|X ≤ T)���X ≤ T)

‫������ﺍ‬X, T)|X > ) = E(X|X > )�,

X is exponentially distributed and therefore is memoryless; then:

��X|X > ) = T + λ� and �(����X, T)|X ≤ T) = E(T) = T.

then

λ λ λ
�{����X, T)}=(�� + λ) × e +T 1−e = T+ λe
Chap 1 Introduction and Basic Concepts 26

�(Y) = E{������ �)} = E(X + T) − E[ ����X, T)]

= � � − (T + e λ � ⇒ �E(Y) = (1 − e λ �
λ λ λ

1-6 Failure rate

The probability of failure of a system in a given interval � , ]

is(K&L page 12):

Pr(t1  X  t2 )  � ( )− ( )= ( ) − ( ).

conditioning on the event the item is working at time �

( ) ( ) ( ) ( )
( ≤ ≤ | > )= =
( ) ( )

If this conditional probability is averaged over � , ] an

average rate of failure is obtained from the following (Ravindran,
2016 p17-12)

( )− ( )
( < < | > ) ( ) ( )− ( )
= =
− − ( − ) ( )

This is called the failure rate during interval � , ] (K&L p12).

1-6-1 Instantaneous rate function (hazard function)

for continuous life time distributions

In the above expression, let � , ] = [ , + ∆ ] then the

average rate of failure would be
� �− ( +∆ )
.
∆ × ( )

�
27 Reliabilty Engineering

When  t approaches zero( t  0 ) in the above fraction, a

function called instantaneous rate function or hazard function is
obtained:

� � ( ∆ ) ( ) ( )
ℎ( ) = ���∆ → (∆ )× ( )
= ( )
− = ( )
, (1-11)

Then for devices with continuous lifetime having pdf f(t) and
reliability(survivor) function R(t), the hazard function is defined
as the ratio of the probability density function to the survivor
function.

� � ′( )
ℎ( ) = � �
= ( )
. (1-12)

The ratio is a function of �t. In practice, t could be time,

number of cycles or revolutions ,km, events,��

h(t) represents the conditional probability density that an item of

age t will fail(Ross, 1985 page 194). However, we can see from the
definition the hazard function is the ����������������������������
it is a normalized probability, not a probability) at time t, given
that the individual has survived until time (https://web.stat.tamu.edu/~
suhasini/teaching613/chapter6.pdf).
Chap 1 Introduction and Basic Concepts 28

It is worth mentioning that it is probable as much as ℎ( ) ×

that a component with lifetime t will fail during the small
interval dt, since

The importance of the hazard function is that it indicates the

change in the failure rate over the life of a population of
devices. For example, two designs may provide the same
reliability at a specific time; however the failure rates up to this
point in time can differ(K&L page12).,

The hazard function of a device is not necessarily the same

in different lifetime intervals.

1-6-2Necessary condition for being a hazard function

All hazard functions must satisfy two conditions(Ravindran,

2016 page17-12). They cannot be negative

ℎ( ) ≥ 0 ≥0 (1-13-1)

and it could be proved that if a function �(t) is a hazard

function then:

∫ ℎ� � = ∞. (1-13-2)

�
29 Reliabilty Engineering

1-6-3 Hazard function and reliability function for

discrete life time distributions

If a product life time(X) has a discrete distribution and � (k)

is the probability that the product fails at time k, according to
Eqs. 1-4 the reliability would be:

R(k) = Pr(X≥k) = ∑∞ ���X = j) =∑∞ � �j),

���The function h(k) given below, is known as the rate

function of "item with discrete lifetime distribution"(Xie et al,2002):

��k) = ���X = k | K ≥ k) = ���X = k)/ ���X ≥ k) ⟹

P (k) � �k)
h(k) = � . (�� − 1)
∑∞ P (k) R(k)

For example if the lifetime of an item has a Poisson

distribution with parameter λ, then:

� � ( )
h(k)= ( )
=∑ ( )
= .
! ∑∞
!

( , )
in MATLAB: h(k)= ( , )
.
Chap 1 Introduction and Basic Concepts 30

1-6-4 Calculation of ( ), ( ), ( ), ( ) given any of

them

If any of the four functions ℎ( ), ( ), ( ), ( ) are

known, the other three are uniquely obtainable from it as
described below(Grosh,1989 page16):

h(t) is known:

f(t) R'(t) t R'(x) t

h(t)=  dt=-h(t)dt   dx=- h(x)dx
R(t) R(t) 0 R(x) 0

Assuming �(0) = 1, integrating yields ,then

∫ � )
( )= (1-14-1)
( ) = ℎ( ) ( ) (1-14-2)

( )=1− ( ) (1-14-3)

f(t) is known:

(1-15-1) (1-15-2) (1-15-3)

�
31 Reliabilty Engineering

F(t) is known:

( )
( )= ( ) ( )=1− ( ) ℎ( ) = ( )

(1-16-1) (1-16-2) (1-16-3)

R(t) known

-R'(t)
h(t)= ( ) = ℎ( ) ( ) ( )=1− ( )
R(t)

(1-17-1) (1-17-2) (1-17-3)

1-7 The pdf of a part of a distribution

If f (x ) , the probability density function of a random

variable is known, the density function of part (a b) of the
random variable is:

1
f a b (x )  b
 f (x ) (1-18)

a
f (x )dx
Chap 1 Introduction and Basic Concepts 32

1-8 Some continuous distributions used in

reliability theory1

Below some useful probability distributions related to

lifetime and reliability subject are reminded.

1-8-1 Exponential distribution

Exponential distribution is a distribution whose density function is

�
� �� , ≥ 0, >0 = > 0 (1-19-1)

The reliability function related to this distribution is as follows:

∞
� )= ( > )=∫ � � ≥ 0 (1-19-2)

The hazard function is :

f (t ) 1
h (t )    (1-19-3)
R (t ) 

It is clear that the rate function of an exponential distribution

is constant and independent of time. Conversely if we know the

�����������������������������������������������������������
1
some softwares such as ARENA could determine the best distributions that
fit a data set(e.g in �ARENA tools-input analyzer- new-file data file- use
existing- fit all)

�
33 Reliabilty Engineering

failure rate of a random variable is time independent (constant),

it s exponentially distributed.

It is worth mentioning that

-A gamma distribution with parameters�( = 1, β� is an

exponential distribution

-A Weibull distribution with parameters�� � �, , = 1) is an

exponential distribution

-The minimum of n independent exponential distributions

with parameters ,…, follows an exponential
distribution with parameters ∑ .
-According to Eq.1-18, the density function of a section of
an exponential random variable say section (0 D) is:

�� ��� 0< < .�

1-8-2 Normal(Gaussian) distribution

The pdf of a normal distribution which is sometimes called
Gaussian distribution is


 x   2
f (x ) 
1
e 2 2 ,   x   (1-20-1)
 2
The rate function is:
1
1  t    2    t   
h (t )  exp    1   Z     (1-20-2)
 2  2 2    
Chap 1 Introduction and Basic Concepts 34

where
Z is the cumulative distribution function(CDF) of standard normal.

ℎ�
��
�� � �

t=TTF
Fig. 1.1 shows the rate function of 2 normal distributions.

It is worth mentioning that normal distribution has the

additive property i.e. if n independent normal distributions
N( � �,����� � � are added to give another random
variable Y, Y also follows a normal distribution,

Example 1.3

Suppose the time to failure of a device is normally distributed

with mean of 20000 cycles and standard deviation of 2000
cycles. Find the value of reliability(survivor) function and
hazard function at 19000 cycles.

Solution Using Table D at the end of the book

����� − �����
�������� = ��� > ������ = �� � > �
����

Pr(Z>0.5)=1-Pr(Z<-0.5)=0.69146=69.15%

�
35 Reliabilty Engineering

- t-200002
2

1
f(t)= e 2(2000 )  f(19000)=0.000176
2000 2π

f(19000) 0.000176
h(19000)= = =0.000245 failures/cycle
R(19000) 0.69146

i.e. 245 failures per 1 million cycles.

1-8-3 Truncated normal distribution(μ�σ���∞)

If we truncate the values of a normal distribution from

below zero, the density function of the remaining values within
the interval [0; ∞] would be derived from Eq. 1-18 as follows:

μ
f (t ) = e σ , t ≥ 0 , σ > 0 . < μ < ∞ (1-21-1)
σ√ π

where

∞
1 − −
� = > =1−Φ � )
√2

Note that
-The probabilities of this distribution is not calculated in the
same manner which is done in classical normal distributions.
- The mean and variance of this distribution does not equal
 and  2 . The mean is
Chap 1 Introduction and Basic Concepts 36

2
1 μ 
 -  
E  X     e 2 σ  (1-21-2)
a
This truncated distribution has an increasing rate function.
Figure 1-2 shows the function for typical one plotted with the
following MATLAB commands:
mu=3;sigma=0.1;t=0:.01:10;f=(normpdf(t,mu,sigma))/(normcdf(mu/si
gma))./(1-normcdf(t,mu,sigma));plot(t,f);

Fig. 1-2 Instantaneous rate function of a truncated

with � ��� = 0.1 ≥0

1-8-4 Log- normal distribution(μ�σ�

The probability density function of a log-normal distribution is:

� �= ≥0, > 0 −∞ < < ∞�� (1-22-1)

√

The mean, variance and median of the distribution is as follows:

( )= (1-22-2)

�
37 Reliabilty Engin
neering

( )= −1 (1-22-3
3)
median= (1-22-4
4)
Figure 1.3 shows
s the pdf of 2 sample lognormal
distribution
ns.

Fiig. 1.3 The pdf of 2 lognormal distribution

1-8-4-1 Calcculation of the parameters(μ,σ) of lognormal

disstribution from the mean and variance

Given ���T)� ���� ����T)� as the mean and variancce of a

lognormal raandom variable, the parameters of the distr
tribution
is calculated
d from:

( )
= �� ( )
� � ��������������� (1-22
2-5)

= �� ( ) − . (1-22
2-6)

1-8-4-2 Thee relationship between lognormal (μ,σ) and norrmal

(μ,σ) distribu
utions
Chap 1 Introduction and Basic Concepts 38

If T � is distributed log-normally :T�����������μ, σ), then

X=ln(T)is a normal random variable: lnT�N(μ, σ).

If X � is distributed normally :X��(μ, σ), then T=� is a

lognormal random variable: � ������μ, σ).

Fig. 1-4 compares the 2 distributions:

�
Fig 1.4 Normal and lognormal distributions�
To calculate the probabilities in this distribution proceed as
follows:
FT (t )  Pr(T  t ) 
lnT   ln t   ln t  
Pr(lnT  ln t )  Pr(  )  Pr(Z  )
  
ln t  
Then FT (t )   Z ( ) (1-22-7)

where �is the CDF of standard normal.
The reliability function or survivor function is given by :

�
39 Reliabilty Engineering

� )= > =1− (1-22-8)

The command lognormalcdf from Table H at the end of the

book might be used to calculate ( ) and R(T):
� � = ������������� � μ� �
� ) = 1 − ������������� � μ� �

The instantaneous failure rate function is:

  ln t    2 
exp  
1  2 2 
h (t )  (1-22-9)
  t 2   ln t    
1  Z    
  

In MATLAB, h(t) could be calculated by dividing commands

lognormalpdf to 1-lognormalcdf :e.g.:
y=(lognpdf(x,1,2.5))./(1-logncdf(x,1,2.5));plot(x,y)�
Fig 1-5 shows the rate functions for 4 lognormal distributions.
This distribution seems to have little except its mathematical
tactability to recommend itself as a failure distribution ; it does
seem to give good fit to repair time distributions(Barlow and
Proschan,1995 page17).
Chap 1 Introduction and Basic Concepts 40

Fig 1-5 The hazard function for some lognormal variable with�� μ = 1
(the horizontal axis is the time and the vertical is the failure function)

Example 1-4

If time to failure of a device is lognormally distributed with

� = 5, = 1). Find the values of the reliability and hazard
functions for t-150 units of time.

Solution

(���) = > = 0.���

( )= ( = ���) ≅ 0.����
√ ,

�
41 Reliabilty Engineering

( ) /
ℎ(���� � ( )
= = 0.���� 1 =
.

�� �������� ����� .

1-8-5 Uniform distribution

If the probability density function of a random variable is as

follows:

( )= ≤ ≤
(1-23-1)
0 .

The variable is said to be uniformly distributed over � ]�

For example the density function of the uniform distribution in

1
 0  t  0
the interval ( 0  0 ) is: f (t )  0
0 other


In this distribution:

( )= (1-23-2)

( )= (1-23-3)

� �
ℎ( ) = � )
= ≤ ≤ (1-23-4)
Chap 1 Introduction and Basic Concepts 42

Figures 1-6 and 1-7 shows the density��������� f��x) and the
hazard function ℎ( ).

Fig 1.7 The hazard function of Fig 1.7 The density function of
a uniform distribution a uniform distribution

1-8-6 Weibull distribution

If the probability density function of a random variable is as

follows:

C
C 1  tA
C  t  A  
f t     e  B 
tA (1-24-1)
B B 

This continuous distribution is called Weibull after Swedish

mathematician and engineer Waloddi Weibull.

The CDF, the reliability (survivor) function, the rate function

are:

( )=1− (1-24-2)

�
43 Reliabilty Engineering

( )= (1-24-3)
C 1
� � C t  A 
ℎ( ) = � )
=   (1-24-4)
B B 

and the mean, variance and median are:

1
E (T )  A  B  (1  ) (1-24-5)
C

2  1 
Var(T)  B2 Γ��� ���2  Γ��� � (1-24-6)
C  C 

1
median  A  B (ln 2) C (1-24-7)

A is called the location parameter, B is the scale parameter and

C is the shape parameter.

An interpretation of location parameter in reliability theory :

A minimum life time of A is guaranteed.

Figure 1-8 shows some Weibull distribution pdf 's.

In weibull distribution:

-If A=0 , the distribution is called 2-paramter distribution.

-If A=0 C=1 , the distribution is exponential distribution

whose hazard function is constant.
Chap 1 Introduction and Basic Concepts 44

-If A=0, C=2 the distribution is called Rayleigh distribution

whose hazard function is linear. This distribution is frequently
used as the statistical distribution of sea wave height and to
model the behavior of some communication channels.

Fig. 1-8 Plot of the Weibull distribution for scale parameter B=1 and
five values of shape parameter (extracted from Grant&Leavenworth ,1988
page605)

Given a random sample …., , the following relations

could be used to estimate the parameters B and C of a Weibull
distribution with A=0. These relations are related to maximum
likelihood estimation(MLE) method in statistics theory.

∑ ∑
= ∑
− (1-23-8)

∑
= (1-23-9)

�
45 Reliabilty Engineering

- In a exceptional case where the shape parameter of Weibul

(A,B,C) is C=3.44, the distribution could be approximated with
a normal distribution with parameters (Carter,1986 as refrenced
by O'Connor, 2003page122 )

� + Γ�� � ) ≅ A + 0.9��, � Γ�� � Γ( )) ≅ 0.3 �

������
Example 1-5
Write a MALAB code to estimate the parameters B and C
of a Weibull distribution from which the following random
sample is at hand:
X=[113.0634 49.5432 53.4872 93.7147 74.0594
114.3216 97.1033 61.5069 74.7216 52.8807];
Furthermore estimate the parameters with wblfit MATLAB
command.
Solution
%Sample X=[X(1)......X(n)]
X=[113.0634 49.5432 53.4872 93.7147 74.0594
114.3216 97.1033 61.5069 74.7216 52.8807];
for C=.01:0.001:40
for I=1:length(X)
LNX(I)=log(X(I));
XIC(I)=X(I)^C;XICLNX(I)=XIC(I)*LNX(I);
end
A= C-(sum(XICLNX)/sum(XIC)-sum(LNX)/length(X))^(-1);
if ( abs(A)<= 0.001 ) C1=C; disp(sprintf('C= %6.4f ', C1))
end
end
B=(sum(X.^C1)/(length(X)))^(1/C1);
disp(sprintf('B= %6.4f ', B))
with MATLAB command wblfit:
Chap 1 Introduction and Basic Concepts 46

>>wblfit(X)
ans = 87.1543 3.7149

1-8-7 GEV distribution

Fisher and Tippet presented three statistical distributions�

�� , ,� , ��or� , ,� . Jerkinson (1955) showed
that these 3 are special cases of one distribution which was
called later generalized extreme value (GEV) distribution. The
characteristics of this distribution as well as Weibull distribution
and general Pareto distribution are given in Table F at the end of
the book.

1-8-8 Gamma distribution

The gamma distribution is another continuous distribution used

in reliability work to fit failure data. It is sufficiently flexible. A
random variable X which follows a Gamma distribution has 2
positive parameter:   0   0 with the following pdf:

 
 x  1e x x 0
f  x       (1-24-1)

0 x 0

Furthermore:
�����������������������������������������������������������

1 Extreme Value
2
Fisher-Tippet I or Gumbel distribution

�
47 Reliabilty Engineering

 (1-24-2)
E (X ) 

 (1-24-3)
V ar ( X ) 
2
  (1-24-4)
M GF (t )  ( )
 t

is called shape parameter and is the scale parameter.

The failure rate function of the gamma distribution does not

exist in a simple closed form. Figure 1.9 shows the function.
For α > 1 it is increasing , for α = 1 it is constant and for
α < 1it is decreasing

Fig 1-9 The hazard function of gamma distribution

������������������������������

If is some positive integer n, the distribution is called Erlang

and in the pdf we could replace Γ(n) with Γ(n) = (n − 1)!.

If α = 1 the distribution is exponential;

Chap 1 Introduction and Basic Concepts 48

If α = 2 `the distribution is Rayleigh.

An application of Erlang distribution is to calculate the

probabilities related to the nth occurrence in a Poisson process.

In Erlang distribution with parameters n and the kth moment

about the origin is:

Γ( + ) (1-25)
� + > 0�
Γ( )

Figure 1-10 shows three functions related to 7 different

distributions.

�
49 Reliabilty Engin
neering

Fig
g. 1-10 pdf, h(t),R(t) of some distributions
Chap 1 Introduction and Basic Concepts 50

1-9 Bathtub curve hazard function

The hazard function varies with time. A well-known pattern

is called bathtub curve whose ideal form is shown in Fig.1-11-1.

Fig 1-11-1 Ideal Bathtub curve

It comprises of three parts: 1)Infancy(early-life period ),

2)useful life, 3)aging. The first part represents the failure rate of
early life period which is decreasing. The second part has a
constant rate of failure, The last part is the wear-out period and
has an increasing failure rate.

A distribution which could be used for each of the 3 parts of

Fig.1-11-1 is Weibull distribution with different shape
parameter C as described below:

0< C <1 : 0< t < t1


C=1 : t1 < t <t 2

C >1 : t  t2

�
51 Reliabilty Engin
neering

Figures 1-11--1 to 1-11-5 show some other variations of batht

htub
curve which happen
h in practice,

�
Fig 1-11-2
2 Some variations of bathtub curve(Kuo&Zuo,2
2003)

Fig 1-11-3 A variation of bathtub curve for some mechanicall devices

(Ireson,1995 page18-2)
Chap 1 Introdu
uction and Basic Concepts 52

Fig 1-11
1-4 Another variation of Bathtub failure rate functtion
(Nahmias, 2004, Fig.12-4)

Figure 1-11-5
5 shows different forms of bathtub failure curve due
d to
different leveels of stress on some mechanical devices.

Fig1-11--5 Effects of stress levels on mechanical failure raates

(Ireson
n,1995 Fig18-1, Stamatis,2010 Fig. 6-3 page163 )

�
53 Reliabilty Engineering

1-9-1 Some forms of hazard functions

For each part of the bath tub curve different hazard function
is appropriate. So next some kinds of failure rate function are
considered(K&L page 28) .
a- Constant hazard( failure rate) function
If the failure rate function is ℎ( ) = i.e. is constant and
does not depend on time ,according to the relationship
between the hazard function and the density function of
lifetime(TTF);

∫ � � ∫
( ) = ℎ( ) = = >0

( )
( )= = =
ℎ( )

Therefore if the hazard function is constant , the lifetime is

exponentially distributed.

The concept of being constant is illustrated in the histogram

and table of Example 1-11.

b- Linear hazard function�

If the failure rate function is ℎ( ) = + , ≥ 0 which
represent a line, then

∫ � )
( )= = (1-26)
Chap 1 Introduction and Basic Concepts 54

For the case = 0‫ > ��ﻭ‬0 , the hazard function is linearly

increasing:

ℎ( ) = ≥ 0, The density function is

∫ ( )
( ) = ℎ( ) ⟹

( )= ∫
= ≥0 (1-27-1)

Which corresponds a Rayleigh distribution or a Weibull distribution

with parameters A=0, B= , C=2 whose reliability function is

( )= (1-27-2)
For the case that the filature rate of a device is like a bathtub except
the first and last part are linear, the hazard function is as follows(K&L
page 29):

− + > 0, 0< <

⎧
ℎ( ) = > 0, < < (1-28)
⎨
⎩ ( − )+ > 0, >

This hazrd function linearly decreases to at time , remains

costant until time , and then linearly increases.

c- Power function�Model
����������������������������������������������������
�����������
Ct C t
h (t ) = =
B B B
Then the density function and the reliability functions would be

�
55 Reliabilty Engineering

∫ ( ) C t
f (t ) = h (t )e = e C, B > 0
B B
R(t) = e
Which corresponds to a 2-parameter Weibull distribution

d- Hazard function of form � ) = + �

If the failure rate function is of the form �(t) = λ + Ct

where �C , k are constants then:

∫ � )
( )= =

e- Hazard function of form ( )= �

If the failure rate function is ℎ( ) = γe �where λ, γ are

constants , the function increase or decreases sharply and

(K&L page 30):

( )= (1-29-1)

The distribution is a kind of GEV distribution with the

reliability function:

( )= (1-29-2)
Chap 1 Introduction and Basic Concepts 56

1-10 Some discrete distributions

Below some discrete distributions are reminded.

1-10-1 Geometric distribution

Consider running an experiment(trial) which has two outcomes

(failure or success). Let p = the probability of success in each trial.
Now notice the two distributions described below:

a) Geometric distribution for failures

We perform the above experiment until a success occurs. Let

X = the number of failures before the first success,

then the probability function of random variable X is given by:

P (x) = p(1 − p) . 0 < <1,x = 0,1,2 … (1 − �� − 1)

The following figure shows the function for � .

�
57 Reliabilty Engineering

Fig. 1.12 The probability function of a Geometric distribution (p=0.1)

The CDF, the mean and the variance are:

� (x) = 1 − p(1 − p) (1-30-2)

1-p 1-p
E(X)= 1-30-3 Var(X)= 1-30-4 
p p2
In this distribution the hazard function is constant:
�(x) = p ��1 − �� − 5)

b)Geometric distribution for success

If We perform the above experiment until the first success

occurs and define a random variable

X = the number of trials until the first success, and

Chap 1 Introduction and Basic Concepts 58

then the probability function of random variable X is given by:

P (x) = p(1 − p) . 0< <1, x = 1,2 … 1-31-1

and
1 1-p
E(X)= 1-31-2  Var(X)= 1-31-3
p p2

Geometric distribution is the only discrete distribution which is

memory-less.

1-10-2 Binomial distribution

If the probability function of a random variable x with

binomial distribution which has 2 parameters positive integer n,
0<p<1 is as follows:

� )= (1 − ) , = 0,1, … (1-32-1)

As proved in Example 1-6 the mean of binomial distribution is

E(X)=np. (1-32-2)

The variance is

Var(X)=np(1-p). (1-32-3)

Example 1-6
Find the mean of a binomial distribution .

�
59 Reliabilty Engineering

Solution

E(X)=∑ (1 − )

Since the value of the first term( i.e. for x=0) is zero and
k   k  1  then
i  k  
i   i 1 

n
n  n
 n 1  x x n n  1
  x 1
E ( X )   x   p x 1  p   n   p 1  p   np    p 1  p 
n x n x n x

x 0 x  x 1  x  1  x 1  x  1 

n 1 n  1
  y
x  1  y  E (X )  np    p 1  p 
n 1 y

y o  y 
n
n 
(a  b ) n     a i b n i  E ( X )  np ( p  1  p ) n 1  np
i 1  i 

For The binomial distribution with parameters ~ ( , ) ���������

( )= .

End of Example

1-10-2 Poisson distribution

The Poisson distribution gives the probability of a given

number of events happening in a specified time period.

If we let X = the number of events in a unit time, then the

probability function is given by:

� �� >0 = 0,1,2, … (1-33-1)

�
Chap 1 Introduction and Basic Concepts 60

and

Var(X)=E(X)= (1-33-2)

Therefore it could be concluded that if the mean and variance of

a random variable are not equal, its distribution is not Poisson.

Now let X = the number of events in a time interval and

λ = the average number of events occurring in a unit time

interval then the probability function is given by:

� )
� ( )= = 0,1,2, … (1-33-3)
!
and
Var(X)=E(X)= (1-33-4)

In fact t could be expressed in other units (length unit , space

unit��������������������������Ireson et al, 1996).

Example 1-7 (Ireson et al, 1996page 11-26).

The failure of an electricity transfer line has roughly a

Poission distribution with annual mean of 0.0256 failure per
1000 feet(nearly 26 failures per one million feet). Find the
probability that no failures occurs along 515.8 feet of the line.

Solution

 =0.0256

�
61 Reliabilty Engineering

Let X= the number failures occurring along t feet.

For t=515.8 feet, the desired probability is

(0.0256�������0�-0.0256������
Pr(X=0)= =1.8���-6
0!

The following g table shows the probability function, the CDF,

the hazard function of some discrete probability distributions.

Distribution h(x )
( = ) ( ≤ )

Discrete uniform 1 x 1 x
� ∈ ��, 1 , … , n}
n n n x
Binomial
( − )
(1 − p) ( − )
∑ ( − )
∈ � � �, … , }
Poisson
x
� ∈ {0, 1 , … }  e
x  x  k
e   x 1  k
x! k 1 k ! x !(1  )
k 0 k!
Geometric
( − ) −� − ) �� p
� ∈ {0, 1 , … }
Hyper-geometric
−

max 0, m+ n -N  −
− −
− −
 x ∑ ( , ) −
−
min  n, m
Chap 1 Introduction and Basic Concepts 62

1-11 On accelerated life testing (ALT), parametric

and non-parametric reliability analysis

In reliability theory, to speed out obtaining life data for a kind

of device, system or component there are some tests called
accelerated life testing and to estimate the values related to
reliability of a device, system or component there are 2
methods: parametric and nonparametric. These concepts are
briefly described below.

1-11-1 Accelerated life testing(ALT)

To perform reliability analysis for a device ,system or a

component the analyst needs life data. In conventional life
testing, to obtain life data some devices are set on a test under
normal condition until they fail. Obtaining life data in this way
is very time consuming and sometimes impossible. Accelerated
life testing (ALT) is the process of testing a product by
subjecting it to conditions (stress, strain, temperatures, voltage,
vibration rate, pressure etc.) in excess of its normal service. The
life test data is extrapolated to obtain the estimates of normal
time to failure. ALT produces the required data in a short
amount of time. Tobias & Trindad(2019) and Cabarbaye(2019)
are 2 references among many others which deal with ALT.

�
63 Reliabilty Engineering

1-11-2 Parametric reliability analysis

In parametric models of reliability analysis, a statistical

distribution such as exponential, Weibull, lognormal and normal
is used to fit the life data or failure rate to estimate the values
such as reliability, failure rate of components.

1-11-3 Non-parametric reliability analysis

Nonparametric analysis allows the analyst to characterize life

data without assuming an underlying distribution. There are
some methods in this kind of analysis including Kaplan�Meier
method, simple actuarial method and standard actuarial method.
Below nonparametric estimation of functions � ), ( ),
( ), ( ) from grouped observations and from ordered sample
is described.

1-11-3-1 Non-parametric Estimation of � ), ( ), ( ), ( )from

Grouped Observations

Below it is described how the functions ℎ( ), ( ), ( ),

( ) for a product could be estimated from a frequency
distribution table of grouped lifetimes or from a random
sample of lifetime

���
Chap 1 Introduction and Basic Concepts 64

1-11-3-2 Non-parametric Estimation of

( ), ( ), ( ), ( )from frequency table

Suppose the following frequency distribution table has been

prepared from a random sample of size N put on test for the
lifetime of an item :

.
interval (a� b� ) (a� b� ) (ai bi ) (an bn ) sum
... ..
.
Frequ. o
.. ..

and let

� The size of the�initial sample put on lifetime test at t=0�

(t)� The number of survivor components (or the number still

working adequately)at time t�

Given the above frequency table, the four functions could be

estimated as follows:

The reliability function estimate

The reliability function at � = b �is estimated as:

N (bi )
R� (bi )  (1-34)
N�

�
65 Reliabilty Engineering

The hazard function estimate

The hazard function related to ith interval i.e. ai bi �could be
estimated as follows(K&L page 13 Grosh ,1989 page 3):
N ( a i ) N ( bi ) (t )i ��
ℎ( ) = � (1-35-1)
N ( a i )  (b i  a i ) N (a i )  (  t ) i

If during (t )i  bi  ai one item fails then

ℎ( ) = (1-35-2)
N (a i )  (  t ) i

The density function estimate

The density function for � < is estimated as follows:

ai bi ���������� ������������������� (t )i

�(t) = = (1-36)
N � � (b a ) � )× (t)i �
i i

The cumulative distribution function(CDF) estimate

The CDF at � = b �is estimated from:

N (bi )
� )=1− ( )= � (1-37)
N�

Example 1-8

46 components were placed on a life test. The system is

observed every 20000 hours and number survivors are written
down (see the following frequency distribution table). Estimate
ℎ( ), ( ), ( ), ( ).
Chap 1 Introduction and Basic Concepts 66

Cum. N (t ) =the number of

Time intervals(hr) failures in interval
frequ. components working at
Time t�
0-20000 19 19 27
20000+-40000 11 30 16
40000+-60000 7 37 9
60000+-80000 5 42 4
80000+-190000 4 46 0
>100000 0 46 0
Sum � � ��

Solution

The above relationships were used to estimate ℎ( ), ( ),

� �, ( ). The results are shown in the following table:

b  ( ) ℎ� �
i � � multiply multiply
��� � � ���� �
i ai ai  ( t ) i ��� �
�
�b ) by:
��
by:
��
=
� ai  t  b i

1 0 20000 46 27 0.587 0.413 0.207 0.207

2 20000 40000 27 16 0.348 0.652 0.120 0.204

3 40000 60000 16 9 0.196 0.804 0.076 0.219

4 60000 80000 9 4 0.087 0.913 0.055 0.278

5 80000 100000 4 0 0.0 1.000 0.044 0.500

To see how the estimates were calculated, Sample calculations

are shown below:

�
67 Reliabilty Engineering

� �� (������ �� − �� ��
b1 �� → � ������� = �
��
�
��
� 0.���
N� N�

N ( a i ) − N ( bi )
ℎ( ) = →
N ( ai ) bi  ai

h� ( t ) � �� − ��
= 0.��� � �� � 0  t  20000
����(������

ai − bi
� �� � →��
N� � (b  a )
i i

20000<t  40000 � ( )=
( )( )
= 0.�� � �� �

Example 1-9

10 components were placed on life test. If one failure has

occurred in each of the time intervals given in the following
table estimate and plot the density function, the reliability
function and hazard function for the time intervals. (Grosh.,1989
Example 1.1 )

Solution

Note that for all intervals � �. Therefore according to Eqs.

1-34 through 1-37:

Chap 1 Introdu
uction and Basic Concepts 68

N (b i )
� )= �� b1 �� ,�ℎ� � � �
N � × (b a ) N� N ( a i ) (b  a )
i i
i i

The followiing table and figures shows the results of the

calculations based on these equations.

��  �  � N � (b�i ai ) � � R� (b i )
��  �  � ��� �
�
� �  (b i ai )

�

����� �  �  �� N (bi ) for ��  �  ��

�× 0.���
� N�
�
�× 0.���
�
1 0 5 1/(10*5)= 2 1/(10*5)=
=2
��
�
2 5+ 1
10 1/(10*5)=2 1/(9*5)= 2.22
2
��
�
3 10+ 1
17.5 1/(10*7.5)= 1.33 1/(8*7.5)=
= 1.67
��
�
4 17.5+ 3
30 1/(10*12.5)=0. 1/(7*12.5))= 1.14
��
�
5 30+ 4
40 1/(10*10)= 1 1/(6*10)=
= 1.67
��
�
6 40+ 5
55 1/(10*15)=0. 67 1/(5*15)=
= 1.33
��
�
7 55+ 6
67.5 1/(10*12.5)=0. 8 1/(4*12.5))= 2
��
�
8 67.5+ 8
82.5 1/(10*15)=0. 67 1/(3*15)=
= 2.22
��
�
9 82.5+ 1
100 1/(10*17.5)=0. 1/(2*17.5))= 2.86
��
10 100+ 1
117.5 1/(10*17.5)=0. 0 1/(1*17.5))= 5.7

The followin
ng figures shows the functions of Example 1--9:

�
69 Reliabilty Engineering

Histogram of failure rate Polygon of failure rate

Histogram of density function Polygon of density function

Cumulative density function and Reliability function

Example 1-10

800 units of a product were placed on the life test and every 3
hours the number of failures were recorded (see table below).
Estimate and plot the density function, the reliability function &
hazard function for the time intervals.(Example. 1.2 Grosh,1989 )

Solution

The appropriate equations are:

f f
f�  �  � i �� , R� (bi )  NN(b�i ) , i
h�  �  � ��
N (b  a ) N (a )(b  a )
� i i i i i
Chap 1 Introduction and Basic Concepts 70

The following table and figures shows the results of the

Calculations:
Calculations of Example 1-10

Density function Failure rate

R� (bi )
f�  �  � N � (bf ii ai ) � �  N (b i ) h�  �  � N (ai )(f ibi ai ) � ���
ai bi no. of N�

i ����� �  �  �� ��� �  �  ��
failures
���
1 0 3 185 185/(800�3)=0.0771 185/(800�3)=0.0771
���
���
2 3 6 42 42/(800���������� 42/(615����������
���
���
3 6 9 36 36/(800��������� 36/(573����������
���
���
4 9 12 30 30/(800���������� 30/(537����������
���
���
5 12 15 17 17/(800���������� 17/(507����������
���
���
6 15 18 8 8/(800���������� 8/(490����������
���
���
7 18 21 14 14/(800���������� 14/(482����������
���
���
8 21 24 9 9/(800����������� 9/(468����������
���
���
9 24 27 6 6/(800���������� 6/(459����������
���
���
10 27 30 3 3/(800���������� 3/(453����������
���

�
71 Reliabilty Engineering

Histogram of failure rate Polygon of failure rate

Histogram of density Polygon of density function

function

Cumulative Distr .
function and
Reliability function�

The functions of Example 1-10 (Grosh,1989 Example 1.2)

Example 1-11

Estimate the density function, the reliability function and the

rate function related to a product whose life test results for 200
units are shown on the following histogram.
Chap 1 Introduction and Basic Concepts 72

Fig. 1-13 Histogram of 200 Switch lifetimes (Feigenbaum,1990)

Solution

��(t)٬ f(t) , R(t) were calculated using Eqs. 1-35-1, 1-36-1 and
1-37. The following table shows the results:

Estimation of � ), ( ), ( ), ( ) for the lifetime of the switches in

Histogram of Fig 1-12
1000-hr
1 2 3 4 5 6 7 8 9 10 11 12
interval
Items volume
working at
200 180 162 146 132 119 107 96 86 77 69 62
the beginning
of the interval
Failure
20 18 16 14 13 12 11 10 9 8 7 6
frequency
No. of
20
Failures in  18

16

14 12 11 10 9 8 7 6
200
180 162
0.1
the interval 146 119 107 96 86 77 69 62
0.1 0.1 0.098
�(t) ≅ 0.1

�
73 Reliabilty Engineering

Estimation of � ), ( ), ( ), ( ) for the lifetime of the switches in

Histogram of Fig 1-12
1000-hr
1 2 3 4 5 6 7 8 9 10 11 12
interval
Items volume
working at
200 180 162 146 132 119 107 96 86 77 69 62
the beginning
of the interval
Failure
20 18 16 14 13 12 11 10 9 8 7 6
frequency
No. of
20
Failures in  18

16

14 12 11 10 9 8 7 6
200
180 162
0.1
the interval 146 119 107 96 86 77 69 62
0.1 0.1 0.098
�(t) ≅ 0.1

20 18 16 14 13 12 11 10 9 8 7 6
= 2 × ��
D D D D D D D D D D D D
�
f(t)
t �in hour
�
f(t) 20 18 16 14 13 12 11 10 9 8 7 6
t �in 1000
200 200 200 200 200 200 200 200 200 200 200 200
hours

�
F(t) t �in 20 38 54 68 81 93 104 114 123 131 138 144
1000 hours
200 200 200 200 200 200 200 200 200 200 200 200

� �
R(t)=1-F(t) 0.8
0.90 0.73 0.66 0.6 .54 .48 0.43 0.39 .36 .48 0.28
1

End of Example��
Chap 1 Introduction and Basic Concepts 74

1-11-2 Estimation of � �, ( ), ( ), ( ) from ordered

random sample

Suppose a random sample of n units of a product were

placed on life test and the n failure time are :

� � � () <⋯< ( ).

h(t), R(t), F(t), f(t) could be estimated similar to Example 1-8 by

forming subintervals with frequency 1. However at the ith ordered
time i.e. � � �they could be estimated from the following relations as
well( K&L p 32):

�
�) = �
(1-38)
�
�) =1− �
(1-39)
() − ( )
ℎ () = ⟹
( )− () ()

ℎ () = ( . )
(1-40)
( ) ()

( )− ( )
�) = ⟹
( )− ()

() = (
(1-41)
( ) () . )

Example 1-12

8 units of a kind of spring were placed on the life test. The

spring failed at the following kilo cycles:

190٬ 245٬ 265٬ 300٬ 320٬ 325٬ 370‫ ﻭ‬400

�
75 Reliabilty Engineering

Estimate F (t), R (t), f (t) and h (t)at the above points in

time.

Solution

The computations are shown in the following table.

Failure � �) F(t) �(t) �(t) h(t)

No.(i)
1 190 0.083 0.917 0.0022 0.0024
2 245 0.202 0.798 0.0060 0.0075
3 265 0.321 0.679 0.0034 0.0050
4 300 0.440 0.560 0.0059 0.0170*
5 320 0.560 0.440 0.0248
6 325 0.679 0.321 0.0025 0.0082
7 370 0.798 0.202 0.0040 0.0198
n=8 400 0.917 0.083 _ _
Because the short interval of time between failures 5 and 6 produced
a large increase in ��t), this interval was combined with the previous
interval and h(t=300) was estimated as follows:
2
� = = 0.�����
(��� − ����(8 − 4 + 0.7)
With empirical data this kind of smoothing must frequently be done

�Some of the calculations are shown below:

.
� � =0.0024 � =1− = R (195)=0.9167
( )( . ) .

.
� � =0.0050 � = 1− = R(300)=0.5595
( )( . ) .

.
� = =0.0082 � = 1− = R(325)=0.3214
( )( . ) .
Chap 1 Introduction and Basic Concepts 76

Example 1-13

In Example 1-9, 10 units of a kind of components were placed on

life test. The failures occurred at the following times:

5 10 17.5 30 40 55 67.5 82.5 100 117.5

Use Eqs. 1-38 through 1-41 to estimate F (t), R (t), f (t) and h (t)

Solution

The computations are shown in the following table1.

f(t)= F(t) R(t) =

h(t)=
i � � t( ) i − 0.3
1 = 1
(t − t ( ) )(n + 0.4) n + 0.4 1− F(t) (t ( ) − t ( ) )(n − i + 0.7)

� − 0.3
1 �
�� � 0.4
1 5 (�� − 5)(�� � 0.4) 0.9327 ��� − 5)(�� − 1 + 0.7)
=
= 0.0192 = 0.0206
0.0673
2 10 0.0128 0.1635 0.8365 0.0153
3 17. 5 0.0077 0.2596 0.7404 0.0104
4 30 0.0096 0.3558 0.6442 0.0149
5 40 0.0064 0.4519 0.5481 0.0117
6 55 0.0077 0.5481 0.4519 0.0170
7 67.5 0.0064 0.6442 0.3558 0.0180
8 82. 5 0.0055 0.7404 0.2596 0.0212
9 100 0.0055 0.8365 0.1635 0.0336
10 117.5 Cannot be computed 0.9327 0.0673 Cannot be computed

End of Example

�����������������������������������������������������������
1
Prepared by:Mr M Morrdi former student of Kerman University

�
77 Reliabilty Engineering

In summary, if the lifetime distribution is known or could be

specified, the relationships related to the distribution have to be
used for reliability computations; if the distribution is unknown
but sufficient data regarding the lifetime is available, the
relationships related to grouped data should be used, otherwise
prepare an ordered sample of the lifetimes and perform the
calculations using the ordered sample.

1-12 The density function &Cumulative distribution function of

sample minimum

Suppose random sample of size n (� � � � … , X �) is taken

from a d\statistical distribution having CDF � ) −∞< <
∞. Either the smallest or the largest of the n observations is
referred to as an extreme value statistic. Practical applications
of extreme value statistics are many; e.g a chain is not stronger
than its weakest link. Let �� ( ) denote the smallest of the n
observations. If � , X , … , X are independent then:

�� �( ) > = ���(X > ), (X > ), … , (X > )] = ���X > ) ⇒

�� X � ) > = [1 − F (y)] or

� )
( ) = 1 − [1 − ( )] −∞ < < ∞ (1-42)

If the distribution of is continuous then the density function

of � ) is:
Chap 1 Introduction and Basic Concepts 78

� )
( )= ( )
( ) −∞< <∞ (1-43)

Example 1-14

Random samples of size n are taken from a population whose

pdf and CDF are:

� )= ≥ 0, ( )= 1− ≥0

Find the pdf and CDF of the smallest extreme value.

Solution

The CDF of the smallest value of the n observation is given by

Eq. 1-42

1− ≥0
( )= 1− 1−1+ =
� �
0 <0

The pdf is given by Eq. 1-43� � )

( )= , ≥ 0;

therefore:

The minimum of the samples of size n taken from an

exponential distribution with parameter has an
exponential distribution with parameter .
End of example

�
79 Reliabilty Engineering

1-12-1The CDF of the minimum of samples of

largish size taken from a population with known
( )

If the CDF (�)�of the population from which the samples

are taken is known, as the sample size(n) becomes large1 the
following approximate approach is helpful in the study of the
sample minimum distribution (from K&L page 42).
To derive � �
( ), when the sampling size(n) from the

distribution with CDF ( ) is large, let random variable Un be

defined as: U = � ) which has a value between 0 &1.
This variable is used in determining the limiting distribution of
� ). Below it is shown that ( ) i.e. the CDF of U is as
follows:

( )=1− 1− (1-44)

And as n approaches infinity we have:

��� → ( )= ( )=1− (1-45)

Proof:

� )= ( ≤ )= ( ) ≤ ⇒

( )= ( ) ≤ = ( ) ≤ ⇒

�����������������������������������������������������������
1
This assumption in the fracture of structures is logical because the number
of their defects are large.
Chap 1 Introduction and Basic Concepts 80

� )= ( )
0≤ ≤

Substituting �� � � , in Eq. 1-42 i.e�

�� ( )
(y) = 1 − [1 − F (y)] �

� �
�1− 1−

Since ��� �x)] = x�� then the CDF of is:

� )= 1− 1− 0≤ ≤ �

We know from mathematics that

��� → � �− }=e u ≥ 0 �therefore:�

� ��� � �� 1− ≥0
→

Here it is reasoned that(K&L p 42, Mann et al,1974p 102):that

since the sequence of the following CDF's converges to� −

( )=1− 1− ,…, ( )=1− 1−

1

Therefore the sequence of random variable �.

=1 ( ) , =2 ( ) ,… , = ( )

�
81 Reliabilty Engineering

Converges in distribution to a random variable U with CDF

�−e :

( ) = ��� � ��1−
→

The pdf of U is:

( )
� )= = ( )= ≥ 0.End of proof

Now notice that

since � � � then� ( ) = and the sequence

of random variables � ( ) converges in distribution to a random

variable, say Y, where

� = = ( )

Thus for large sample size( n) the limiting distribution of the

smallest extreme value( � �) is given by the distribution of Y as
described in the following steps;

Derivation of ( )
i.e.�the CDF of sample minimum

or � ) when →∞

Step1 Given � � substitute x= � ��

Chap 1 Introduction and Basic Concepts 82

Step 2 Let �= ( ) then calculate its inverse i.e.

� ) = ⋯ in terms of

Step 3 Calculate the limiting � ) in terms of U(=limiting �

from step 2

Step 4 Calculate the following:

� ( )
(y) = ���limiting � � � )

From this relationship calculate � ( )

in terms of

���U< ⋯)=�� .

Step 5 Calculate � ( )
(y), considering step 4 and

Eq. 1-45 i. e. ���( ≤ ) = 1 − .

Examples 1-15 and 1-16 illustrates the derivation.

Example 1-15

Random samples of size n are taken from a uniform

distribution on [0 1]. What is the CDF and pdf of the smallest
extreme value when → ∞�
Solution

The density function of the uniform distribution is:

�
83 Reliabilty Engineering

1
� )= 0≤ ≤
0 .

To derive the CDF of the sample minimum as� → ∞ the above

5 steps are followed:

Step 1

−0 ( )
� ( )= ⇒ ( ) =

Step 2

( )
� ( ) ⇒ = ⇒ ( ) =

Step 3

Let U = the limiting value of � , then:

� � → = ( )

Step 4

( )
( )= ( ) ≤ = ≤ = ≤

⟹ ( )
( )= ≤ =
Chap 1 Introduction and Basic Concepts 84

Step 5

According to Eq. 1-45 the CDF of U is

( )=1− > 0; then

( )
( )= =1− , ≥0 ⟹

( )
( )=1− , ≥0

Taking derivative yields the pdf as follows:

( )= ≥ 0 .End of Example
( )
0 ℎ

Example 1-16

Random samples of size n are taken from an exponential

distribution with parameter . What is the CDF and pdf of the
smallest extreme value when → ∞.
Solution

The density function of the uniform distribution is:

� �� � ≥0
0 .

�
85 Reliabilty Engineering

To derive the CDF of the sample minimum as� → ∞ the above

5 steps are followed:

Step 1

� )=1− � � ) =1− ( )

Step 2
� ( ) ⇒ = 1− ( ) ⟹
1 � 1
( ) = �� = − ���1 − )
1−

Taylor expansion of f(x) about x=a is:

� )= ( )+ ( )( − ) + ( − ) ( ) +�
! !

This expansion for ���1 + x) , −1 < ≤ 1 is:

���1 + ) = 0 + − + − +�

Let �− ; then

1 1 1
� ) =− − − − −⋯
2 3

Step 3

Let as → ∞ U = the limiting value of � . Ignoring the terms

of order 2 and higher we could say that the distribution of � )

when → ∞ approaches the distribution of ∶

( ) → = × .
Chap 1 Introduction and Basic Concepts 86

Step 4

� ( )
(y) = ��� ( ) < ) = ≤ = ( ≤ )

⟹ ( )
( )= ( )

Step 5

According to Eq. 1-45 the CDF of U is

( )=1− > 0; then

( )
( )= ( )=1− , ≥0 ⟹

( )
( )=1− , ≥0

Taking derivative yields the pdf as follows:

nλy
(y) = e y≥0
f ( )
0 ������

End of Example

The following example (extracted from K&L p 45) shows an

application of GEV distribution to reliability.

�
87 Reliabilty Engineering

Example 1-17

An applications of the extreme value distribution is to the study the failures of car exhauhs caused by
corrosion.

Consider a kind of automotive exhaust pipe that has various

pits when new. The exhaust gases and other corrosives increase
the depth of these pits and, ultimately, a failure occurs when the
exhaust gases can escape through one pit that has penetrated the
thickness of the pipe and has become a hole. If we assume that
the time of penetration is proportional to the difference between
the pipe thickness(D) and the initial pit depth� � and has a

truncated exponential probability distribution between (0 D),

show that the time to failure of the exhaust pipes is a GEV
distribution and find the reliability function.

Solution

Symbols

D Exhaust pipe thickness

Initial pit depth of ith pit i=1,2,��
� Failure time of ith pit
N Number of pits.
T Failure time of the exhaust pipe

The distribution of is a truncated exponential with the

following density function:
Chap 1 Introduction and Basic Concepts 88

� )= = 0≤ ≤
(0≤ ≤ ) 1−

Since the failure time of ith pit (t )is proportional to (� − ),

then

� = k(D − d ), where k>0 is the constant of proportionality.

The cumulative distribution of � is as follows

� )= ( ≤ )= ( − )≤ = ( − ≤ ) ⟹

( )= − ≤ .

Since the maximum of is D, then:

� )= − ≤ ≤ = ( )− −

where

D is the thickness of the pipe and

di, the initial depth of the ith pit, i=l,2,,..,N.

N is the number of pits.

The di's constitute a random sample from a truncated

exponential distribution defined on the interval (0 D) :

�
89 Reliabilty Engineering

� (x) = , � ≤ x ≤ D, i = 1,2, . . . , N

Therefore

( )
� )= ( )− − = - =

.
= ⇒

Since − ≥ 0� �and − ≤ �then �0 ≤ ≤ ��and

−1
� )= 0≤ ≤ , i = 1,2, . . . , N
−1

Let T =the failure time of the entire exhaust pipe, then

N
= m in ( ) and its CDF is:
i �

� )= ( < )=1− ( > )=1− ( > ,..., > )

Assuming ,..., are independent and similar, we could

write:

� ) = 1− ( > )… ( > )= 1− 1− ( ) … 1− ( )

( )= ( < ) = 1− 1− ( )

In mathematics it is shown that for � < < 1 , [1 − ]

approaches as ⟶ ∞ , then
Chap 1 Introduction and Basic Concepts 90

Since �� ≤ � ) ≤ 1 and there are a lot of pits in the pipe,

× ( )
therefore � ) ≅ 1 − .

We saw earlier � (t)���� then:

� )=1− ⇒ ( )=1− ( )=

This is an extreme value or a GEV distribution(K&L page 46) .

Example 1-18

In the previous example, suppose � ℎ, = �� �

=�� � ℎ / and the average depth of pits is .
Find the life time that will give a reliability of 90% .

Solution

If the pdf of the initial depth of a pit were λe , the average

depth would be �
= and λ = ���. However here the pdf is

� �= and to find the value of λ the following

equation has to be solved:

1
1 = � ���
� )= ( ) = ��
��� 1
( )= �����
���
1 ∫ 1
( )= = ⟹ = �
1− ��� 1− ���

�
91 Reliabilty Engineering

∫
Solving the equation = in MATLAB:

>>syms landa x; landa =solve((int(landa*x*exp(-

landa*x),x,0,1/16))./(1-exp(-landa/16))==1/128)

landa=127.64972

Notice that ignoring from the denominator yields�λ = ����

Substituting the followings in � � �

= �� � , = �� � , = , = �����, ( ) = 0.9�

Yields � ≅ ������.

Fisher -Tippet and central limit Theorems

Fisher -Tippet Theorem �

If X 1 ,..., X n are independent and identically distributed(iid)

random variables then as n increases
the distribution of the maximum of these variables approaches a
GEV distribution and
the distribution of the minimum of these variables ] approaches
another GEV distribution.

It is worth mentioning that:

1) in the original theorem by Fisher and Tippet, states that
the limiting distribution is one of three extreme value
Chap 1 Introduction and Basic Concepts 92

distributions (EV11=FT1, EV2=FT2 and EV3=FT3) but

Jerkinson(1955) showed that the aforementioned three
limiting distributions can be unified into a single
expression known as the generalized extreme value
(GEV) distribution.
2) This theorem is used in extreme value analysis such
finding the possible maximum of wind speed, wave
length, etc����������������aders could refer to references
such as Coles (2001).
3) For using this theorem, some random samples of largish
size from the desired population are needed. Extract
their minimum or maximum and prepare a vector (titled
say Data) of the minima or the maxima. Then use
gevfit(Data) command to estimate the fitting GEV.

Central Limit Theorem �

According to the central limit theorem the mean of

random samples X 1 ,..., X n of sufficiently large size n from a

population with mean�μ and finite variance�σ2 , tends towards a

2
normal distribution with mean�μ and finite variance ; even if the
√

original distribution is not normally distributed. The theorem also

�����������������������������������������������������������

�
���������������������������������

�
93 Reliabilty Engineering

states that sum of the sample elements(∑ ) tends towards a

normal distribution with mean�μ and finite variance σ2 .

1-13 Bartlett's goodness of fit(GOF) test for

exponential distribution

To deal with the following hypotheses using a GOF test

H0: The distribution is exponential

H1: The distribution is not exponential

we could use the Bartlett�������� described below:

Take a random sample of size at least 20 :� , … , t ≥ ��,

where � is the time of the ith event; calculate the statistic B
given by(K&L p239):

  1  r   1  r 
2r ln    i 1t i     i 1 ln t i 
 r   r 
B  
(1-46)
r 1
1
6r

which has a chi-squared distribution with r-1 degrees of freedom

under the null hypothesis H0. If B is outside the
interval�[ χ ,
χ ,
], reject H0; α is the level of

significance of the test.

Chap 1 Introduction and Basic Concepts 94

χ ,
is read from Table E or calculated in MATLAB from

chi2cdf(� − , r − 1)

χ ,
is read from Table E or calculated in MATLAB

from chi2cdf( , r − 1).

�

Example 1-19

The following sample of size r=20 is available from the

lifetime of a kind of electric bulb. Does an exponential
distribution fit the lifetime data? Use Bartlett's test with��α �
���. If the answer is yes, give the mean and the pdf of the
distribution?

50.1 20.9 31.1 96.5 36.3 99.1 42.6 84.9 6.2 32.0
30.4 87.7 14.2 4.6 2.5 1.8 11.5 84.6 88.6 10.7

Solution

H0: The lifetime distribution is exponential

H1: The distribution is not exponential

Let t i , i  1,..., r  20 be the sample values. We use Bartlett's

test:

�
95 Reliabilty Engineering

  1  r   1  r 
2r ln    i 1t i     i 1 ln t i 
 r   r 
B  
r 1
1
6r


20
i 1 i
t  50.1  20.9  ...  88.6  10.7  836.3


20
i 1
Ln (t i )  Ln 50.1  Ln 20.9  ...  Ln 88.6  Ln10.7 63.9385
  836.3  63.94 
2��� ��  � 
  20  20 
B  19.34
20+1
1+
6  20

χ ,
= ��������0.������ = �������
��

χ ,
= ��������0.������ = �������
�

H0 is not rejected because, the value of statistic B does not

fall outside χ ,
χ ,
. Therefore the
�

distribution of the bulbs are fitted to an exponential

distribution with the mean and pdf:

∑ . .
� = ≅ ������� ( )= × = .
.

It is worth mentioning that using Kolomogrov-Smironov

test in MATLAB does not reject H0
Chap 1 Introduction and Basic Concepts 96

>>Data=[�
50.1
�
10.7];
>> H=kstest(Data, [Data expcdf(Data,mean(Data))] , 0.1)
H=0

This means that H0 is not rejected at the significance level

of 10%.

1-14 Q-Q plot

Quantile-Quantile(Q-Q) plot is a graphical device to observe

whether a particular distribution fits a dataset or not. In this
graph the observed data and the corresponding data obtained
from the distribution are plotted versus each other in an X-Y
coordinate plane. The better the population follows the
distribution, the closer the points to the angle bisector of the first
quarter of the X-Y plane .The procedure for preparing a Q-Q
plot is as follows:

 Sort The sample of data from minimum to maximum,

giving rank 1 through n: X (�) , ..., X ( n )

 Allocate a number F(i), called plotting position calculated

from one the following formulae to each x ( i ) . In fact

�
97 Reliabilty Engineering

F(i) is a number near to relative frequency and an

estimate for the cumulative distribution function at x (i ) .

There are many formulae for plotting position including the

followings:

A)Gumbel Plotting position

One of the first formulae for plotting position was given by

Gumbel:

i
F (i )  , i  1,...,n
n 1 (1-47)

B) Plotting position for normal distribution

There are some formulae for the normal case including

(Besterfield,1990 page52):

i  0.5
F (i )  (1-47-1)
n
or (Goda,2000 page 287):

i  0.375
F (i )  . (1-47-2)
n  0.25
Chap 1 Introduction and Basic Concepts 98

C) Plotting position for Weibull distribution with

parameters A,B,C

The Plotting position for Weibull distribution with parameters

A,B,C is (Goda,2000 page 287):

i a
F (i )  (�  � � )
n b

�.�� �.��
where a  �.��  b  �.�� 
C C

D) Plotting position for Exponential Distribution

Since Exponential distribution could be considered a Weibull

with C=1 then:

i  0.47
F (i )  (1-48-1)
n  0.43

 From F X [ x� ( i ) ]  F (i ) for each F(i), i=1,..,n calculate

x�(i ) , i=1,..,n from where F X

is the cumulative

distribution function of the distribution under study.

 Plot the pairs (� x(i)& x� ( i ) � �� in an X-Y coordinate plane,

and fit a line to the points. The closer this line to the
angle bisector of the first quarter of the plane, the better
fits the distribution to the dataset. It is worth knowing
that the better the distribution fits the data set the closer
the correlation coefficient of x(i) & x� ( i ) to 1; but the vice

�
99 Reliabilty Engineering

versa is not necessarily true i.e. if the correlation

coefficient of x(i) & x� ( i ) is close to 1,necessarily the

distribution does not fit the dataset well.

The correlation coefficient is calculated by the following

formula:

n  x ( i ) x� ( i )   x ( i )  x� ( i )
R  (1-49)
n  x (i )    x (i )  n  x� ( i ) 2    x� ( i ) 
2 2 2

Example 1-20

The following table shows a sorted random sample, x(i)� 's,

from a population. Is the sample a representative of normal
distribution?

Solution
To answer, a Q-Q plot is drawn. The mean and variance of
the distribution is estimated as follows:
s
�  X  54. , �   11.7287
c4

F(i), i=1,..,n was computed using F (i )  i  0.375 as the

n  0.25
plotting position, and inserted in the table. Then the
corresponding x� ( i ) is calculated by equating the F(i)to the
Chap 1 Introduction and Basic Concepts 100

normal standard cumulative distribution, and calculating x� ( i )

from these equations.

x�(i )  � s
Pr(X  x�(i ) )  Pr(Z  )  F (i) �  X , �  .
� c4
sample calculation follows:
For i  1 :
x� (i )  � x� (1)  54.81
Pr( Z  )  F (i ) , Pr( Z  )  0.0294  x� (1)  32.37
� 11.7287/0.9876

or � ) = (0.������������������� = ��������

The following table contains all the results

Rank(i) x (i) F(i) x� ( i ) Rank(i) x (i) F(i) x� ( i )

1 32 0.0294 32.37 12 59 0.5471 56.21

2 34 0.0765 37.84 13 59 0.5941 57.64
3 39 0.1235 41.06 14 60 0.6412 59.10
4 44 0.1706 43.51 15 61 0.6882 60.64
5 46 0.2176 45.55 16 64 0.7353 62.28
6 47 0.2647 47.34 17 67 0.7824 64.08
7 50 0.3118 48.98 18 68 0.8294 66.11
8 51 0.3588 50.52 19 70 0.8765 68.56
9 51 0.4059 51.98 20 70 0.9235 71.78
10 52 0.4529 53.41 21 71 0.9706 77.25
11 56 0.5000 54.81

Fig.1.14 2 shows x ( i ) 's versus x� ( i ) 's and a line fitted to them.

�
101 Reliabilty Engineering

QQplot
80

pp=(i-0.375)/(n+0.25)
70

60
Predicted

50

40

30
30 40 50 60 70 80
Observed

Fig. 1.14-1 Q-Q plot with Fig. 1.14-2 Q-Q plot

F(i )=
i-0.375 .
with F(i )= i-0.5 .
n+0.25 n

Since the points are near to the fitted line and the line is close to
the angle bisector of the first quarter of the X-Y coordinate
plane, it is concluded that the normal distribution fits the dataset.

In MATLAB, the command qqplot could be utilized to make a

Q-Q plot from the normally distributed dataset X. The Q-Q plot
of Fig. 1.13-1 was made by this command. The difference of
the two plots is not significant. It is worth mentioning that if the
dataset X is not normally distributed, the following MATLAB
command could be used to plot the Q-Q plot:

�X=[data]; pd=makedist('Distribution name'���������������

The correlation coefficient(r) between x� ( i ) , x ( i ) is calculated by

r=corrcoef(X,Xhat);r=R(1,2)
Chap 1 Introduction and Basic Concepts 102

where
X is the vector containing x(i), i,=1,2,3..

Xhat is the vector containing � ), = 1,2, .. which gives

0.9826. This value, being near to 1, together with the Q-Q plot
of Fig. 1-13=1 or Fig. 1-13-2 indicate that normal distribution is
a good fit for the data best fit.

1-15 Convolution

Since the concept of convolution of functions in statistics and

probability is used to find the distribution of the sum of
independent continuous random variables X and Y having the
density function (pdf), �f (x) ( ) and the cumulative
distribution function � ��� and � (y); this concept is reviewed
below.

1-15-1 CDF and pdf of sum of independent variables

X and Y

Let X and Y be 2 independent random variables and Z=X+Y.

a)The CDF of Z i.e. � � is derived from the following

relationship which is called the convolution of �� �and is
denoted by ∗ :

( )= ∗ ( )=∫ F (a − y) ( ) (1-50)

�
103 Reliabilty Engineering

Proof (from Ross,1983 page 17):

� ) = ��� + ≤ )=∫ ��� + ≤ | = ) ( ) =

∫ ��� + ≤ | = ) ( ) =∫ F (a − y) ( ) .

End of proof of section a

b) The pdf of Z i.e. � � is derived from the following

relationship which is called the convolution of �� and is

denoted by ∗ :

f (a) = f ∗ f (a) = f (a − y)f (y)�� � � ��)f (a − x)�� ����(1 − ����

where

� ) is the pdf of random variable X

� ) is the pdf of random variable Y

Proof (from Ross,1985 page 54):

� )= F (a − y) ( ) = F (a − y) ( ) ⇒

( )= (a − y) ( ) .

End of proof of section b

Chap 1 Introduction and Basic Concepts 104

Notice that

- When calculating the convolution integral, usually it is easier

to set the simpler function as the second function(Gordon,
1993).

-If the distribution of X+Y is known for us, there is no need for
the above integrations

-The concept of convolution has been extended for more than 2

functions.

Example 1-20 (Ross, 1985page 54)

X and Y are independent uniformly distributed random
variables on the interval (0 1). Find the density function of
X+Y.
Solution

��) = (a − y) ( )

Considering the uniform distribution of X and Y on (0 1) ,it is

evident that a value of X+Y, say a, lies in the interval (0 2);
mathematically 0 ≤ a ≤ 2 .

To find the limits of the above integral notice that:

For � ≤ ≤ 1, � � is nonzero and

�
105 Reliabilty Engineering

�a − y) ≠ 0 0≤ − ≤ 1 ���a − 1 ≤ ≤a

Then the limits of the integral are the intersection of the

intervals � ≤ ≤ 1 ����a − 1 ≤ ≤ a . To calculate the
resulting interval which depend on the value, we divide the
range of �� ≤ a ≤ 2 into � ≤ a ≤ 1 and � ≤ a ≤ 2:

If � ≤ a ≤ 1, as the following figure shows, the limits of integral

would be � ≤ y ≤ a:

If � ≤ a ≤ 2, using a similar figure it could be shown that the

limits of integral would be � − � ≤ ≤ 1.

Therefore

�a) = (a − y) ( )
∞

⎧
1×1 0≤a≤1
⎪
⎪
� ⇒
⎨
⎪
⎪ 1×1 1≤a≤2
⎩
Chap 1 Introduction and Basic Concepts 106

0≤a≤1
�a) =
2− 1≤a≤2

Example 1-21

X and Y are 2 independent random variables with density

functions � �� � ≥ 0 and � )= , ≥ 0. Find the
density function of Z=X+Y.

Solution No.1

��) = (a − y) ( )

Considering the range of X and Y, we could say that a ≥ ��

To find the limits of the above integral notice that:

� �y) ≠ 0 for � ≥ 0 and

�a − y) ≠ 0 for � − y ≥ 0 ≤ .

Therefore the range of integral to become nonzero is � ≤ y ≤ a:

� )
f (a) = ∫ f (a − y)f (y)�� � ∫ � e �� � �� � a ≥ 0

Solution No.2

( )=∫ (a − x) ( ) .

�
107 Reliabilty Engineering

with a similar reasoning for the range of integral:

� )
( )= = a≥0

Solution No.3

As it is well known that the sum of 2 independent variable with

the same parameter has a ( = 2, )� distribution, then:

e a (a)n 1 e a (a)21

� )=   ae a a0
(n  1)! (2  1)!

Example 1-22

X and Y are independent random variables with normal( � �

and Uniform(0,1) distribution respectively. Find the density
function of X+Y.

Solution

� �= (a − y) ( )

The range of a=x+y is −∞ < < ∞��

To find the limits of the above integral notice that:

Chap 1 Introduction and Basic Concepts 108

� (y) ≠ 0 for 0 ≤ ≤ 1 and

�a − y) ≠ 0 for −∞ ≤ a − y ≤ ∞�� −∞≤y≤∞.

The intersection of these 2 interval is0 ≤ y ≤ 1.Therefore the

integral is not zero between 0 to 1:

�a) = (a − y) × 1

If we let � − y = t, the range of t would be is a − 1 ≤ t ≤a

1
��) = (t)(− ) = →
√2

(a) = (a) − (a − 1) −∞ < <∞

where is the CDF of standard normal distribution.

1-15-2 n-fold convolution of f with itself

Suppose we have n independent random variables with the

same density function f(x) and we want to derive the density
function of their sum i.e. � � � ∑ ( ). � � which called n-
fold convolution of f with itself and denoted by � � )] ∗ id
defined as follows:
� ) = ( ) ∗ [ ( )]( )∗ . (1-52)

�
109 Reliabilty Engineering

Furthermore, denoting ∗ ��� ∗, ∗, the n-fold

convolution of F(cumulative distribution function) with itself is
the distribution of the sum of n independent random variables
each having distribution F(Ross,1983, page17) is denoted by:
� ∗ F( )∗ = F ∗� (1-53)

and also

� ∗ F��� ∗ ������������������������������������������

Example 1-23
Find the probability density function of the sum of the
exponentially distributed lifetimes of 3 independent
components with parameter �
Solution
Using convolution:
� ) = [ ( )]( )∗ = ( ) ∗ [ ( )]( )∗

= ∗ ( )∗

[f(t)]( )∗ = [ f (t )] ∗ �

[f (a)]2*   f (a  t )f (t )dt a 0


Since the lifetimes are exponentially distributed

t  0 and a  t  0(  t  a ), then < < and:
Chap 1 Introduction and Basic Concepts 110

a a
[f (a)]2*   e  (at )e  (t )dt   2e a  dt   2ae a
0 0

2 t
Continue with convolving  te and λe to reach the

solution which is: (λa) e .

The solution was, because the sum of 3 independent

exponential distributions with the same parameter λ has a
Gamma distribution with parameters (n=3, λ).
End of Example

�������������������������������������������������
���������������������������������

Suppose X1, X2 are 2 independent nonnegative random

variables with density functions f X and f X and let Y=X2-X1. Y is
� �

sometimes called interference random variable. The density

function of Y is calculated from (extracted from K&L page
125):

  fX� (y  x�) fX� (x�)dx� y �
X �
fY (y)   fX� (y  x�) fX� (x� )dx�   � ���� �

  fX� (y  x�) fX� (x� )dx�
x�
y �
x� y
Needless to say that if the distribution of X2-X1 is known
There is no need for using The above equation.

�
111 Reliabilty Engineering

1-17 Percentage of a distribution being outside

limits

To calculate the proportion of a random variable which fell

inside and outside any specified limits, lets distinguish 2 cases:

a)If the distribution of Y is completely known

The calculation is as simple as follows:

For continuous random variable:���a ≤ Y ≤ b) = ∫ f(y)����

For discrete random variable����a ≤ Y ≤ b) = ∑ p(y).

where �(y)����� p(y) are density or probability function of Y.
b) If the distribution of Y is unknown
In this case Tchebycheff inequality could be used which
states that in all statistical distributions the fraction falling
outside�� � � > 1 is at most :

− ≥ ≤ , >1 (1-56-1)

or
− ≤ ≤ + >1− , > 1 (1-56-2)

where
and are the mean an standard deviation of Y.
�c) A table containing some intervals and frequencies
In this case if the frequency distribution shows the
distribution of Y has only one mode and the mode is the same as
the arithmetic mean and the frequencies decline continuously on
Chap 1 Introduction and Basic Concepts 112

both side of the mode,(Grant &Leavenworth,1988),according to

an adaptation of the above inequality by Camp and Meidel:
− ≥ ≤ = .
, ≥1 (1-57-1)

Or
��� − < < + )> 1− , ≥1 (1-57-2)

Example 1-24

The strength of a kind of component is a random variable

with ����� . For each of the following cases determine, what
percentage of the components fall within the specification limits
( 34, 46).
1)The strength is normally distributed with σ��
2) the distribution is unknown but σ��
3) can the strength in part a be exponentially distribute?
4) the distribution is unknown but has one mode and σ��

Solution
Normal distribution
a b 
Pr  a  X  b   Pr( Z  )
 
34-40 46-40
Pr( <Z< )=Pr  -3<Z<3
2 2
From Table C:
Pr  -3<Z<3 =0.99865 - 0.00135=0.9973

�
113 Reliabilty Engineering

With MATLAB:

Pr  34 < X < 46 = normcdf(46,40,2)-normcdf(34,40,2) = 0.9973

That is 99.73% of the product fall within (34 46)
2)Using Tchebychef Inequality:
1
Pr  X-kσ������σ  ���
2
k
X-kσ���� ���σ���  �� �
1 1 8
Pr(40-3��� � � �������� ������ 2
��� 2 � ������
k 3 9
In this case more than 88.99% of the product fall within 34
and 46.
3)The distribution cannot be exponential because the mean
and standard deviation of exponential distribution are equal.
4)Assume the conditions for applying Camp-Meidell
inequality holds, therefore :

More than 1- 1 4
=1- 2 of the product fall within (34 46) ;
2.25k 2 9k
or
1 1
Pr  34<X<46  >1- =1- =%95.06
2 2.25��
2.25k

Appendix 1 Parameter Estimation Techniques

There many techniques for estimating the parameters of a

statistical distribution including the following:
1)Maximum Likelihood Estimation(MLE)
2)Method of Moments(MOM)
Chap 1 Introduction and Basic Concepts 114

3)Least Squares Method(LSE)

4)Bayes Method
5) Geometric Mean Slope1
6) Pickands Estimators
7)Heuristic Algorithms
8) Minimum chi-squared Estimation
9)Using Inverse Probability theory
10)Bootrap Estimation Method
From the above methods, MLE and MOM are described briefly below.

1-Maximum Likelihood Estimation Method

To use maximum likelihood method, we first need to define

likelihood function. Likelihood is a concept that works with
joint distributions.

����������������������������������

Suppose a random sample� ��� has been taken from a

continuous or discrete distribution. The following joint
probability function L is called likelihood function:
For continuous distribution:

L  f ( x1  xn )  f x1 ( x1 ) f xn ( xn )
(1�58�1)
�����������������������������������������������������������
1
Refer to page 25.33 Handbook of Reliability by Irenson et al,1996

�
115 Reliabilty Engineering

For discrete distribution:

� ��� = ,…, = ) = ��� = ) × … × ��� = )���
(1-58-2)�� ��
e.g.:
for exponential distribution:

� L  ( e   x 1 )... ( e   x n )   n e    x i
(1-58-3)��
for binomial distribution:�������������������������������
� �∏ ∑
× (1 − )∑ ( )
� (1�58�4)

Notice that the calculation of this method is based on the

assumption that �…, are independent and identically
distributed(iid).

Steps of Maximum Likelihood Estimation(MLE) Method

To estimate the parameters � 1, 2, ... , k of a
distribution, MLE method could be used through a 3-step
process.

1. Find the likelihood function L for the given random

variables (X1, X2, ... , Xn),
2. Maximize the likelihood function by taking the
derivatives of L with respect to θ�

Notice that log(x) is a monotone1-increasing function of x,

maximizing logarithm of a function is equivalent to maximizing
the function(based on Barlow and Proshan,1996 p166).

�����������������������������������������������������������
1
A function is monotonic if its first derivative is always positive or negative
Chap 1 Introduction and Basic Concepts 116

Therefore it is often simpler to maximize the logarithm of

function L rather than L itself (Bowker and Lieberman, 1972
page287)

3. Estimate the value of θ1, θ2, ... , θk by setting the

derivatives obtained in Step 2 equal to zero.

Notice that if ��, is random sample from a uniform

�
distribution with f  x   �  x   ٬it is proved that:

n
MEL    max  xi 
i 1

Example 1-25
a)Given a random sample ��� from an exponential
distribution, use MLE method to estimate �
b)If the sample is (1.1, 0.9, 1.21, 0.8)calculate the value of �
Solution
a)Since the sample is random, 's are independent.
L   n e   x i  lnL  ln   n e   x i   nln     x i 
d  lnL  n 1
 0  MLE ( )  �  
d  xi X

b) ( )= = =∑ = =1
. . . .

Lemma1

�����������������������������������������������������������

�
��������� �����������������������������������

�
117 Reliabilty Engineering

If θ is the maximum likelihood estimator of θ and �(θ) is

the function of θ possessing a single inverse(i.e. its derivative is
always positive or negative) , then � θ is the maximum
likelihood estimator of � ��
M L E [ g ( )]  g [ M L E ( )]

End of lemma
Example 1-26
If �…, is a random sample of size n, taken from an

exponentially distributed lifetime, estimate θ � .

Solution
Since θ �λ� = <0 therefore θ has a unique inverse and

according to the above lemma:

����θ(λ)] = θ(����λ)) = = .

Example 1-27
Given a random sample …., from a Weibull
distribution with location parameter A=0, use MLE method to
derive the relations for calculating the scale and shape
parameters B and C.

Answer

∑ ∑
� ∑
− (1-59-1)

∑
� � (1-59-2)
 Chap 1 Introduction and Basic Concepts 118

It is worth mentioning that the MATLAB command wbfit

estimates the parameters of a Weibull distribution.
Example 1-28
a)Write a MATLAB code to return the estimates of a 2-p
Weibull distribution from which the following sample is
available.
b)Also use wblfit function to estimate the parameters.
Solution
a)
%Sample X=[X(1)......X(n)]
X=[113.0634 49.5432 53.4872 93.7147 74.0594 114.3216 97.1033
61.5069 74.7216 52.8807];
for C=.01:0.001:40
for I=1:length(X)
LNX(I)=log(X(I));XIC(I)=X(I)^C;XICLNX(I)=XIC(I)* LNX(I);
end
A=C-(sum(XICLNX)/sum(XIC)-sum(LNX)/length(X))^(-1);
if abs(A)<= 0.001 )C1=C;disp(sprintf('C= %6.4f ', C1)) end
end
B=(sum(X.^C1)/(length(X)))^(1/C1);
disp(sprintf('B= %6.4f ', B))
��b)
>> wblfit(X)
ans = 87.1543 3.7149

Example 1-29

Let ,i=1,..,k be the number of successes in a sample of size n

from a binomial distribution with parameter p, Find MLE(p).

�
119 Reliabilty Engineering

Solution
k
n 
L   Px i Px i    p x i q n  x i
i � xi 
k
n  n   n  x k n x k
L     p x i q n  x i    p x �q n  x �  ...   p q 
i �  x i  x� xk 
k k
k
n  xi  ( n x i )
L     p i �
 q i �

i �  x i 
k k
k
n  xi (n x i )
ln(L )  ln(  )  ln( p i � )  ln(q i � )
i �  x i 

k
n  k k
ln(L )   ln     x i ln p   (n  x i ) ln(�  p ),
i �  x i  i � i �

k k k k

 ln(L ) x i  (n  x i ) � p  (n  x i ) nk   x i
� i �
 i �
 �  i �
 i �

p p � p p k k

x
i �
i x
i �
i

� nk  xi
�  k  �  MLE ( p )  i �

p kn
xi i �
End of Example

x
k  �  MLE ( p ) 
n

MATLAB commands for estimating Distributions

parameters
The MATLAB commands for estimating the parameters of
some statistical distributions are given in Table H.
For example, given a sample the MATLAB command
Chap 1 Introduction and Basic Concepts 120

expfit returns parameter θ for ��) = ;�

poissfit returns parameter of a Poisson distribution

binofit returns parameter p of a binomial distribution.

Example 1-30
The following sample shows the lifetime(in year) of some
units randomly taken from a batch of a device having an
exponentially distributed lifetime. Estimate the parameter of the
distribution.
0.04 0.15 0.04 0.09 0.03 0.01 0.04 0.06 0.01 0.15

Solution
X=[ 0.04 0.15 0.04 0.09 0.03 0.01 0.04 0.06 0.01 0.15 ] ; expfit(X)

This yields θ � 0.��� which is the mean of the distribution.

Thus λ = � � �����, which the average number of annual
θ

failures.
Example 1-30
The annual number of failures of a device has a Poisson
distribution; given the following sample estimate the distribution
parameter.
17 13 19 8 17 17 12 19 18 19

Solution

>> X=[17 13 19 8 17 17 12 19 18 19 ];poissfit(X)

This returns λ = ����. End of Example

�
121 Reliabilty Engineering

Example 1-30

To estimate the failure probability(p) of the cables used in the

construction of a kind of bridge, 5 samples of size n from this
kind of cable was set to life test. The number of failures in these
5 samples are as follows:
� 2, : = 1, : = 4, : = 0, : = 1. What is the
maximum likelihood estimate of p?
Solution
2+1+4+0+1
∑ x X 5
� − 5, n = ���� ����p) = � = ⟹
�� n ���
����p) = 0.���
Using MATLAB:
Given � �� � = 1, : = 4, : = 0, : = 1 As seen
below, binofit function returns 5 estimates with mean
of� ̂ =0.016.

>>x=[2 1 4 0 1]; P=binofit(x,100); phat =mean(P)

phat=0.016.

2- Methods of Moments(MOM)

A widely used technique in estimation is method of

moments .
Before describing the method, it is reminded that:

the kth moment of random variable X about zero(0) is

Chap 1 Introduction and Basic Concepts 122

E( ) = expected value of and

the kth moment about 0 of sample ��, is
n


i �
x k
i
Mk 
n .
Methods of moments for estimating the parameters of the
distribution of random sample X is based on the fact the g the
kth moment of X could be estimated by the kth sample moment
i.e. � �= , k=1, 2��

Steps of MOM
To estimate parameters ��, of the statistical
distribution of random variable X ,
i-Compute � �, = , … , in terms of �…, .
Notice that always the first moment of a distribution is its
mean and the second moment is equal to the variance of the
distribution plus the squared mean.
ii-Form the k equations
� �= , j=1, ����
iii-Solve the equations for the parameters�� ,…, .
The resulting values are called method of moments
estimators for the parameters.

�
123 Reliabilty Engineering

Example 1-32
Let �…, be a random sample taken from an
exponential distribution with parameter , Estimate the
parameter by the method of moments.
Solution
� X i n
E (X )  M �    � 
 n X i
End of example
Example 1-33
Let �…, be a random sample taken from an normal
distribution with parameters �� , estimate the
parameters by the method of moments.

Solution
 X i
 
E  X   M �  n
 
 X
E  �   M �  �
 �   �   X i
 n
 X i
    �  X
n 
 � ^ �
X i
 �   X i   �  
�
X
�

  n
n
Chap 1 Introduction and Basic Concepts 124

In statistics theory it is proved that instead of the above

n

x i
�

estimate for variance i.e. i �

X �
sample variance i.e.
n
n n

 (x i  X )� x i
�
 nX �

S�  i �
 i �
is a good estimate for the
n � n �
variance of X because of being unbiased i.e. � )= .

Appendix 2: Application of MATLAB in Reliability theory

Softwares have provided calculations easy. Here some MATLAB

functions which might be used in reliability subject is described.

A-Plotting the frequency distribution of lifetime

Given a sample of the lifetime of a kind of product, the

frequency distribution which consists of classes and their
corresponding frequencies could be plotted using the following
command;
>>hist(Data,K)
where
Data is a vector consisting the life times of a sample selected at
random from a lot of the product,

K is the number of classes into which the range of life

time is desired for partitioning.

�
125 Reliabilty Engineering

B-Parameter estimation

Once the histogram in the above section was prepared one

might guess the statistical distribution to which the life tomes
belong. The parameters of this distribution could be estimated
using some commands given in Table H at the end of the book
e.g.:
x
1 −θ
To estimate θ in an exponential distribution with �(x) = θ
e ,
thetahat=expfit(Data)

To estimate the parameters of a normal distribution:

[muhat,sigmahat] = normfit(Data)

To estimate the parameters B and C of a Weibull distribution

� )
with reliability function �and A=0:

BhatandChat= wblfit(data)

where Data is a vector containing a sample of the life times of

the product.

C-Goodness of fit (GoF)

Goodness of fit implies a comparison of the observed data

with the data expected under the model using some fit. It
describes how well a statistical model fits a set of observations.
Chap 1 Introduction and Basic Concepts 126

To accomplish this, one could use several GoF tests as well as a

kind of graph known as Q-Q plot.

There are a number GoF tests including Pearson chi-squared

test for continuous ad discrete distributions and
Kolmogorov_Smironov test for continuous distributions. In this
section the latter test is introduced.

���������������������������������

The Kolmogorov-Smirnov test is used to examine whether a

sample comes from a population with a specific CDF F(x) or not:

The distribution with CDF F(x) fits the data

CDF F(x) does not fit the data

The MALAB function related to this test is as follows:

H = kstest(Data,CDF, )�

where

Data A column vector containing data

Level of significance e.g . 0.05,0.10
CDF hypothesized, continuous cumulative distribution function F(x)

Examples for the format of specifying the desired CDF are

given below:
[Data wblcdf(Data,A, B)]�,

�
127 Reliabilty Engineering

[Data expcdf (Data, )],

����[Data, [Data normcdf(X, μ‫ﻭ‬σ )].

if omitted or

If the CDF is unspecified (i.e., set to an empty matrix []), the

hypothetical distribution is assumed to be a standard normal:
N(0,1).

H indicates the result of the test:

H = 0 ⟹ Do not reject the null hypothesis at significance level .

H = 1 ⟹ Reject the null hypothesis at significance level .

Example 1-33

Could it be said that the following sample comes from an

exponential distribution with significance level � ����?

[110, 520, 645, 680, 330, 75, 95, 480, 360, 575, 1065, 170,
415, 15, 20,1275, 270, 90, 1500, 1923, 715, 1523,427, 730,
1120, 390, 240, 40, 220, 673, 2397, 1032, 315]

Solution
Entering the data as a column vector:
>>Data=[�
110
520
�
1032
315];
Giving the command:
Chap 1 Introduction and Basic Concepts 128

>>H=kstest(Data, [Data expcdf(Data, mean(Data))], 0.05)

The answer for H is 0; i.e. it is not rejected that the data belongs
to an exponential distribution with significance level � ���� .

C-2 Q-Q plot

Quantile-quantile ( Q-Q) plot examines the conformity

between the empirical distribution and the given theoretical
distribution through command qqplot( X , pd) which displays
the quantiles of the sample data X versus the theoretical
quantiles of the distribution specified by the probability
distribution object pd:

>>�X=[���������
pd=makedist(distribution name ); e.g.
pd=makedist ('exponential')
pd=makedist ('Gamma')
qqplot(X,pd)

Figure1-15 shows a sample Q-Q plot.

Fig. 1-15 A sample Q-Q plot

�
129 Reliabilty Engineering

The more the points near to the line and the line near to the
bisector of the first quarter, the better the distribution fits the
data.

D Calculation of Reliability

Once the distribution of the lifetime of a device is determined

to be one of the well-known distributions, the reliability could
be easily calculated using the CDF commands in Table H; e.g.
for exponential :
R=1-expcdf(x,θ) where x is mission time and θ is the
distribution mean,
for normal :
R = 1-normcdf(x, μ, σ) where x is the mission time and μ, σ
are the distribution mean and standard deviation.

for 2-parameter Weibull( location parameter A=0)

R = 1-wblcdf(x, B,C) where where x is the mission time and

B,C are the scale and shape parameters of the distribution.

E Calculation of the inverse of cumulative distribution

function (CDF )and Reliability function

The inverse of a CDF gives a value say a associated with

random variable X such that the probability of the variable being
less than or equal (� ≤ a)to a is equal to the given cumulative
probability.
Chap 1 Introduction and Basic Concepts 130

Table H shows the MATLB command for this purpose. The

commands have the suffix .inv
Example 1-34

Suppose the life time(X) of a device is exponentially

distributed with mean of 100 hours,

a)Find a in Pr(� ≤ a) = 0.����

answer :

>>x=expinv(0.3935,100)

x=50

b) Find the reliability for lifetime equal to 50.

answer

>> p=1-expcdf(50,100)

p = 0.6065

c)Find the lifetime value for which the reliability of the device
is 0.6065.

>>x= expinv(1-0.6065,100)

x = 50.0051

�
131 Reliabilty Engineering

F Finding polynomial roots and Solution of Equations in

MATLAB

During the calculations of problems the following MATLAB

commands might be helpful

������������������������������������������

To find the roots of x + ⋯.+ x + ⋯+ − � the

following command in MATLAB is used:
>>roots(�a … a a ]).

F-2: Finding solution of algebraic equations

Solve function in MATLAB finds the solution of an equation

of asimultaneous equations. For example to find the value of

∫
which satisfies the equation = , the following
( )

commands could be used:

>>syms landa x; landa =solve((int(landa*x*exp(-landa*x),

x,0,1/16))./(1-exp(-landa/16))==1/128)

landa=127.65

One who brag ,

will not become savant
Chap 1 Introduction and Basic Concepts 132

Exercises

1-(Problem 3 Page49 K&L) Two designs for a critical

component are being studied for adoption. From extensive
testing on prototypes it was found that the time to failure(TTF)
is Weibull distributed with a minimum life of zero. Design I
costs $1,200 to build and has Weibull parameters of C= 2 and
� ���√��. Design II costs $1500 to build and has Weibull
parameters of C= 3 and B= 100 hours,

(a) The component has a 10 hour guaranteed life. Which design

should the manufacturer produce and why?

(b) For a 15 hour guaranteed life what should the choice be?

2(Problem 6 Page49 K&L) Consider the piecewise linear

bathtub hazard function defined over three regions of interest
given below.

���� = b − c t, 0≤t≤t �

h(t) = b − c t − c (t − t ), t < ≤t �

h(t) = b − c t − c (t − t ) + c (t − t ), t < < ∞�

The constants b and c in the above expressions are determined

so that they satisfy the normal requirements for h(t) to be a
hazard function. Find the reliability function based on the above
hazard function.

�
133 Reliabilty Engineering

3(Problem 11 Page49 K&L) If h(t) is a hazard function prove

that∫ ℎ( ) ⟶ ∞.

4- (Problem 7 Page50 K&L) Which of the following functions

can serve as hazard function
-at at +5 -3 e2t
a) e b) e c)ct d)et e) 3
t
Develop the density and Reliability functions for those which
are hazard function.

5- (Problem 9 Page50 K&L)

50 automobile components are placed on test with a hazard

[unction as below:

h(t ) ��� where t is in kilometers

Compute the expected number of failures after 10,000

kilometres.

Hint: Let X= the life time; use binomial distribution with

n  50, p  Pr( X  104 )

6- Repeat the previous Problem for hazard functions

- 2 t
i)h(t)=10-6t ii)h(t)=10-4 e 100 t  km

7- (Problem 15 Page50 K&L)Given the population distribution

is uniform on (0 ) , find the CDF and the pdf for the smallest
Chap 1 Introduction and Basic Concepts 134

extreme value in a random sample of size n. use the exact theory

not the approximation for large n.

8-(Problem 17 Page50 K&L)Rework the previous problem,

assuming n large and asymptotic distribution of extreme values.

9-If a population is uniformly distributed on (�  ) find the

expected value and the variance of the minimum of the samples
taken from this distribution.

10-What is the maximum likelihood estimate for parameter b in

x2
_
Rayleigh distribution with CDF F(x)=1-e b2
?

11- Write the required relations for estimating the parameters of

Bernoulli and normal distributions using MLE method.

12-Assuming f(x) is a unimodal pdf with modal value , prove

that h '( x )  h 2 ( x ) .

( ) ( ) ( ) ( ) ( )
Solution:ℎ� ) = ( )
⟹ℎ( )= ( )

( ) ( ) ( ) ( ) × ( ) ( ) ( ) ( ) ( )
⟹ℎ( )= = ⟹ℎ( )= =
( ) ( ) ( )

 h '( x )  h 2 ( x )
( ) ( )
×
( ) ( )

�
135 Reliabilty Engineering

Chapter 2
Static Models
in
Reliability
Chap. 2 Static Reliability Models 136

2
Static Reliability Models

Aims of the chapter

This chapter is concerned with modeling the reliability of the
systems in which time coordinate is not presented in the
reliability of their subsystems or components. In this regard the
chapter after introducing a diagram called reliability block
diagram (RBD); deals with some reliability configurations such
as series, parallel, k-out-of n configurations. Furthermore
calculation of upper and lower reliability bound for the complex
systems is described.

2-1 Definition of static reliability models

Here, the word static means that the time coordinate is not
presented in the calculations. In modeling a system from a
reliability stand point using static models, the component or
subsystem reliabilities are considered to be constants; thus some
base time period is implied(K&L page 55). Before dealing with
some conventional component configurations, a graph is
introduced below.

�
137 Reliabilty Engineering

2-2 Reliability Block Diagram

A reliability block diagram(RBD) is a graphical model of the

elements of a system permitting the calculation of system
reliability given the reliability of the elements . Figures 2-1 and
2-2 are RBB examples. Each component or subsystem of the
system is presented by a block or box in the RBD. This useful
and important graph is used in the calculation of systems'
reliability. Now some conventional configurations and their
RBDs are addressed below.

2-3 series configuration

A series system is one that requires all of its subsystems to

function in order for the system itself to function; in other
words, it has a configuration such that if any one of the sub-
systems fails, the entire system fails. Figure 2-1 shows the RBD
of a system with series configuration(or simply series system).�

1 2 3 n

Fig. 2.1 RBD of a series configuration

Let Ei=event that subsystem i operates successfully, then the

reliability of ith subsystem is R = ��(E ) and the system
reliability( ) equals:
Chap. 2 Staticc Reliability Models 13
38

= ( ∩ …∩ ).

Assuming
g the operation of the subsystems are indepen
ndent of
each other, we
w have:

= ( )… ( )= ×…× ⇒

=∏ . (2-1)

n
Since � � �� therefore ≤ min � � ; that is the reliab
bility of
i�

a series systeem with independent components is not greatter than

the least relia
iable component(K&L page 56). It is worth
mentioning that
t in an n- component series system, if the
reliability off all components is equal to = 1 − , = 1, …,n
then: = (1 − ) . Now notice since according to biinomial
�
expansion:

Therefore

( − 1)
�1 − ) = � � (− ) + �− � �� � � ��− �
2

Ignoring hig
gher order terms, could be approximated
d as
follows if < 1:

�
139 Reliabilty Engineering

: = (1 − ) ≅ 1 − , < 1.

If the components do not have the same reliability and ith

element failure probability is denoted by =1− , an
approximation for the series system reliability is given by(K&L
page 58):

n n
R sys  �   q i q i �
i � i �

Example 2-1

A ten-component system with 95% reliability is to be

designed. The system is to be designed in such a way that if any
component fails the system would fail. What should be the
reliability(R) of each component?

Solution

The system configuration is series and its reliability is : �

= (1 − ) ���� � � (1 − ) ⇒ = 0.���� ⇒

=1− = 0.9949

Using approximation:

≅1− ⇒ 0.�� � � − (���( ) ⇒ = 0.���

�1− = .���
Chap. 2 Static Reliability Models 140

Notice that if the components are not independent Eq. 2-1

could not be used for computing the reliability of a series
system. In this case, the chain rule for factorization, as
described below, might be useful:

Chain Rule for Factorization

Let Ei=event that subsystem i operates successfully; then the

system reliability( ) equals:

= ( ∩ ∩ …∩ ∩ ),

Using the chain rule for factorization, this joint probability can
be rewritten as follows:

� ���E |E , … , E , E )���E |E , … , E ) … ���E |E )���E )

������������������������������������������������� � × �������������������) (2-2)

2-4 Parallel configuration

A system is said to have a parallel configuration if any of the

elements in its structure permit the system to function; in other
words parallel system is a configuration that works as long as
not all of the system components fail.
Assuming the components work independently of one another,

= 1 − (1 − )(1 − ) … (1 − )=1− (1 − ) ( − )

�
141 Reliabilty Engineering

It could be easily verified that the reliability of a parallel

system is more than any of its components reliability.

Fig. 2-2 Reliability block diagram of a

parallel system-all components working

Proof of Eq. 3-2

Suppose in a 2-component system, components 1and 2 are

connected in parallel, and let E1 and E2 denote the events that
1and 2 operate successfully. Since the systems works if either 1
or 2 works ,then the system reliability ���� is equal to:

� = �������������� that either 1 or 2 works� ���E + E )

( )= ( )=
R = ( 1 + 2) = ( 1) + ( 2) − ( 1 2) =
R + R − R R = R + R (1 − R ) =
1 − (1 − R ) + R (1 − R ) = 1 − (1 − R )(1 − R )
Or
R = 1 − ������������������������) = 1 − (1 − R )(1 − R ).
Chap. 2 Static Reliability Models 142

And in general if n components of a system are connected in

parallel independent of each other ,then the reliability of n-
component parallel system is:
�
= 1 − (1 − )(1 − ) … (1 − )=1− (1 − ). (2 = 3 − 2)

If the unreliability of component i is dented by �1− then:

=1− … . (2 = 3 − 3)

End of proof

Example 2-2

a)A 6-componet parallel system with a reliability of 80% is

to be designed, determine the reliability of each component.

b)How many components in a parallel system leads in a

reliability of 99.9% system reliability?

Solution
1
a- = 1 − (1 − )6 ≥ 0.��� ⇒ ≥ 1 − (1 − 0.8)6 =
������� = �������
b- = 1 − (0.5) = 0.���� ⇒ = ��

Example 2-3 (K&L page 71)

In an electrical distribution system, electronically operated

circuit breakers(CBs) can be activated to interrupt the current. If
the current exceeds 105% of the rated line current it is required
that the circuit breakers open, thereby disconnecting the supply.

�
143 Reliabilty Engineering

The probability that a circuit breaker functions correctly is 0,98,

and each breaker has its own line voltage sensor. If the
reliability associated with interrupting the circuit is to be at least
0.999, how many circuit breakers in series are necessary to
achieve the desired reliability?

Solution

98% of the times a CB could disconnect the un-allowed

current load. To increase the reliability associated with
interrupting the circuit, series configuration cannot be used
because a series configuration of number of the� � CBs would
result in a reliability of less than 98%; instead a parallel
configuration is used, the necessary number of CBs in parallel
is calculated as follows: � 1 − (1 − 0.��� = .��� → =
(.����/ (.��� ≅ 2 �
Example 2-3 (from K&L)

A detection system for the CO level in a test cell is under

cosideration. Specifically, there is a sensor available that will
close a circuit and thereby signal the personnel if it detects a
particular level of CO concentration. However, this sensor can
fail in the following ways:
Chap. 2 Static Reliability Models 144

Failure Statuts Probability

Signal high CO level when none is present 0.10
Not detect high CO level when it is present 0.15
Obviously, the sensor is not too reliable and it is ecided to use 3
of them ia DC circuit.
(a) Arrange the sensors such that the probability of
delecting a high emission level if it is present, is maximized.
(b) Calculate the probability of a false signal for each
arrangement considered in (a)
Solution
a) The reliability of the sensor is:
� =1− = 1 − (0.1 + 0.��� � 0.��
The reliability of the system is
� : R = 1 − (1 − 0.��� � 0.��
: R = (0.��� � 0.��
Select parallel configuration.
b)
The false signal probability is 0.1, then the probability of false
signal in a parallel arrangement is:
�� − �1 − (1 − 0.9) ] = (0.1) = 0.����
Since the probability of without-error operation of the series
system is ����) ; then the probability of a false signal for series
configuration is � − (0.9) = 0.���

�
145 Reliabilty Engineering

2-4-1 Types of parallel configurations

In a parallel system the components might be arranged in the

following ways;

1-Active redundant

In this parallel configuration more than one components are

active and failure of one component still makes the system keep
working(Fig 2-2). Eqs. 2-3-1 through 2-3-3 are related to this
type of parallel systems.

2-Standby redundant

In this parallel configuration one component works and some

similar components are waiting to replace the on-line
component when it fails. he system is functional until all
components fail. It is worth mentioning that waiting (standby)
units might be subject to failure when waiting for eplacement.

3-Shared parallel configuration

In the shared parallel system, the failure rate of surviving

components increases as failures occur. An automobile wheel
assembly is an example of the shared parallel arrangement; if a
lug nut comes loose the remaining nuts must support an
increased load, and hence the failure rate is increased with each
successive failure. Thus, the shared parallel is not truly a static
model (K&L page99).
Chap. 2 Static Reliability Models 146

Definition of Perfect and Imperfect switching

In standby redundant configurations by "perfect switching" it

is meant that when the active component fails and a standby
component is to replace it by the help of a switch, the switch
does not fail during or before replacement operation,
mathematically the switch is 100% reliable (Ps=1).
"Imperfect switching" refers to the situation in which the
switch has a probability of failing to change over from active
component A to component B when A fails in a standby
redundant configuration.

���������������������������������������������������������
�������������������

Consider a system composed of an original and a backup

component shown in Fig. 2-3. When the original component
fails, a perfect switch(i.e.100% reliable) turns on the standby
backup component and the system continues to operate. Let
R1,R2 denote the reliability of the components.

Fig. 2-3 RDB of a 2-component standby system

The reliability of this system is equal to:

�
147 Reliabilty Engiineering

=
�����������
������ � ��� A ������ � ���( A ��������������� × ��� B ������
�
�� + (1 − )= + −
Therefore th
he reliability of a two-component standby sy
ystem is
the same as to
t that of a 2-component active system as given below
b :
2
R  1  (1  R 1 )(1  R 2 ) Ps�1 (2-3-1
1)
syys

�������������
������������������������������������������
������
��������������������

Consider an n-component standby system with one normally

n
operating su
ubsystem and n-1 in standby status(Fig 2-3--1). The
system is fun
nctional until n failures occur. The reliability
y of this
system is:
n
stan
ndby R  1  (1  R 1 ).....(1  R n ) (2--3-2)
sys

Fig
g. 2-3-1 RDB of n-component standby system
m
(https:///egyankosh.ac.in/bitstream/123456789/35169/1/Unit-1
15.pdf)
Chap. 2 Static Reliability Models 148

Proof of Eq. 2-3-2

Suppose a system includes 1 active subsystem and 2 redundant

subsystems in standby status with a perfect switch. The
3
reliability of this system( R )is:
sys

Stand by 3
 Reliabilty1-active 1 standby 
R  1  1  [1  (1  R1 )(1  R 2 )]1  R 3 
sys  
Standby 3
R  (1  R1 )(1  R 2 )(1  R 3 ) Ps  1
sys

This was proof for a 3-component standby system; if the

calculations continue in a similar manner for 4-component, 5-
component ��������������������������������������������-3-2.
End of proof .

Example 2-5

A parallel system has an active device with 90% reliability.

When this active fails, a perfect switch replaces it by a standby
backup with 80% reliability. Calculate the system reliability.

Solution

2
R =1-(1-R1 )(1-R 2 ) =1-(1-0.9)(1-0.8)=0.98
sys

�
149 Reliabilty Engineering

���������������������������������������������������������
��������������������

Consider a system composed of an original and a backup

component as shown in Fig. 2-3. When the original component
fails, an imperfect switch with reliability � � � � turns on
the standby backup and the system continues to operate. Let
R1,R2 denote the reliability of the components in the system
R1,R2 and Ps are constants not functions. The redundant
component do not share any of the load and is not probable to be
in a failure mode before turning on. In this case the system
reliability ( R2 ) is given by the following relationship (based on
sys

Billinton&Roy,1992 page77 Eq.4-12):

2
R =R 1 +Ps R 2 (1-R 1 )=1-(1-R 1 )(1-Ps R 2 ), Ps  1 (2-3-3)
sys

Example 2-6

Evaluate the reliability of the system in Fig. 2-3 if A has the

reliability of 0.9, B has a reliability given A has failed of 0.96
and switch has a reliability of 0.98.
Solution
2
R =1-(1-R1 )(1-Ps R 2 )=1-(1-0.9)(1-0.92��������.9883
sys
Chap. 2 Static Reliability Models 150

2-5 Combination of Series and Parallel Configurations

Some systems have a series-parallel configuration as the ones

shown in the following example.

Example 2-7 : Consider the series-parallel configurations

A&B given below; which one is more reliable? A or B?
A
1 3

2 4
�

1 2

��B
3 4
Solution

Assuming the components are independent from each other,

� [1 − (1 − )(1 − )][1 − (1 − )(1 − )]

= 1 − (1 − )(1 − )

− = (1 − )(1 − )+ (1 − )(1 − ) > 0�

Then A is more reliable than B.

�
151 Reliabilty Engineering

Example 2-8
In example 2-7 let R 1  0.95 , R 2  0.85 , R3  0.75, R 4  0.8 and

the components work independent of each other. Calculate the

reliability of the configuration.
Solution
R A = [1 -(1 -0 .9 5 ) (1 -0 .8 5 )] [ 1 -(1 -0 .7 5 )(1 -0 .8 ) ] = 0 .9 4 2 9

R B = 1 -(1 -0 .9 5 � � �� � ��� �� �� � � � �� �� � �� � � �

2-5-1 Redundancy Level

Redundancy is the duplication of critical
components or functions of a system. It is a common
approach to improve the reliability and availability a
system. In this chapter you were introduced with active
and standby redundancies.
One of the most fundamental determinants of component
configuration concerns the level at which redundancy is to be
provided. Lewis,1994, Chap. 9). In this regard the following
redundancies are introduced:
 High-level redundancy(HL) or the system level redundancy
 Low-level(LL) redundancy or the component level redundancy
High-level redundancy involves the duplication of the entire
system while low-level redundancy is limited to the duplication
of components or subsystems.
Chap. 2 Static Reliability Models 152

High-level redundancy(HL) or the system level redundancy

Suppose we have n types of component and using k
components from each type, a series configuration is formed. If
the n subsystems are connected as a parallel configuration then
high-level(HL) redundancy or the system level redundancy has
been formed(Fig 2-4-1).

a b c

a b c

Fig. 2-4-1 Example of High-level(HL) redundancy (Lewis, 1994 p272)

The reliability of the above RBD is;

= 1 − [(1 − )]

Low-level(LL) redundancy or the component level redundancy

Suppose we have n types of component and using k
components from each type, a parallel configuration is formed.
If the n subsystems are connected as a series configuration,
then low-level(LL) redundancy or the component level
redundancy has been formed(Fig 2-4-2).

�
153 Reliabilty Engineering

Fig. 2-4-2 Low-level(LL) redundancy (Lewis, 1994 p272)

Remember that if 2 similar components with reliability R are

paralleled then the resulted system would have the following
reliability:

��� − (1 − )(1 − ) = 2 −

Therefore the reliability of the above RBD is equal to:

= (2 − ) 2 − (2 − )

If � = then

− =6 (1 − ) > 0 ⇒ > .

Regardless of how many components the original system has

in series, and regardless of whether two or more components are
put in parallel, low-level redundancy yields higher reliability,
but only if a very important condition is met. The failures must
be truly independent in both configurations (Lewis,1994 page 273)
Chap. 2 Static Reliability Models 154

Example 2-9

Find the reliability of a system having the RBD shown in Fig.

2-5-1 in which m series subsystem of n components are
parallel.

1 2 3 n

1 2 3 n

1 2 3 n

Fig. 2-5-1 The RBD of part a of Example 2-9

Solution Let r denote the reliability of ith component

for i=1,2,����������������������������������������������
to �∏ r , and the system reliability is given by:
= 1 − (1 − ∏n=1 ) (2-4-1)
where
m number of subsystems
n number of components in each subsystem
the reliability of ith component i=1,2,...,n

b) Find the reliability of the following system

�
155 Reliabilty Engineering

1 1 1

2 2 2

3 3 3

n n n �
Fig.2-5-2 The RBD of part b of Example 2-9

Solution
The figure shows that the system has m parallel subsystems each
having n components. The reliability of this LL redundancy
system is given by:

= [1 − ∏ (1 − )] (2-4-2 )

Example 2-10( Lewis, 1994 page271)

Find the reliability of the RBD given in the following figure:

1 2
4
3
Chap. 2 Static Reliability Models 156

Solution: = [1 − (1 − )(1 − )]( )

Example 2-11
In the following 2-component system, the reliability of the
components are respectively R � R . If the reliability of this
system is not enough for us, and we have the following two
options to use, which one has more reliability? A or B?

Solution

� � + − )

=[ + −( )( )]( )

− =( )( − )

Therefore if � , configuration B would be more reliable than

configuration A ��

Not surprisingly, this expression indicates that the greatest

reliability is achieved in the redundant configuration if we
duplicate the component that is least reliable; if R2 > R1 then
system B is preferable, and conversely. This rule of thumb can
be generalized to systems with any number of non-redundant
components; the largest gains are to be achieved by making the
least reliable components redundant(Lewis, 1994 page271).

�
157 Reliabilty Engineering

������������������������������

A k-out-of-n system has n identical independent components

or subsystems, of which only k need to be functioning for
system success(Fig. 2-6). It is supposed that Each component
works, independently of all the other components ..

Fig 2-6 A k-out-of-n system

xamples of real world applications

Applications of k-out-of-n systems can be found in many
areas such as communication, electric and electronic, safety
monitoring systems and human organizations.
In a cable-supported bridge having n supporting cables, at
least k cables must be working
A committee with n members who must decide to accept or
reject innovation-oriented projects and the committee will
accept a project when k or more member (Nordmann and Pham 1999)
Chap. 2 Static Reliability Models 158

2-6-1 Reliability of k-out-n configuration

In k-out-of-n configuration, if k=1 the configuration is active

parallel configuration with the following reliability:
� 1 − (1 − ) … . (1 − )
If k=n, the configuration would be:�
= × …×
If k=n-1,the system does not fail with the failure of one
component, but it fails with the failure of 2 components.
In the simplest form, let the reliability of all components be
the same and equal to R. To compute the reliability of this k-
out-of system notice that(Lewis, 1994 page 269):
For identical components, the reliability of an k-n system
may be determined by the binomial distribution. Suppose that p
is the probability of failure over some period of time for one
unit. That is, p = 1 - R, where R is the component reliability.
From the binomial distribution the probability that n units
will fail is just

( = )= (1 − )

The n-k system will function if there are no more than n-k
failures. Thus the reliability is as follows:

( ≤ − )= (1 − )

Substituting 1-R for p yields;

�
���� � ∑n−k
x=0 �
(1 − R)x (R)n−x (2-5)

�
159 Reliabilty Engineering

ℎ :
���� � ��������� − �, n, 1 − R) (2-5-1)
The system failure probabilty or system unreliabilty is calculated

from:

�− =1− ( − , ,1 − )

Alternatively,
we could say that system work as far as ��� components out
of n component fail. Then the probability of system
failure is given by:

( > − )= (1 − )

Then the system reliability ( ) is;

−
=1− (1 − )
= − +1

Or we could say that system work as far as � components

out of n component work. Then the reliability of system
is given by:

(1 − ) −
=
=

where R is component reliability.

This relationship and the following integral are equal(Barlow &

proshan,1998 page218):
Chap. 2 Static Reliability Models 160

= ∫ (1 − ) .
In summary, if the reliability of components in a k-n system is
denoted by R then:

=∑ (1 − ) = ∫ (1 − ) . (2-6)

where
R component reliability
system reliability
n Total number of components in the system
i no. of components that work
n-i no. of components that fail

using MATLAB:
���� � � − ��������� − �, n, R) (2-6-1)

Notice that Eqs. 2-5 and 2-6 are equal.

Example 2-12
A system has a 3 out of 5 active redundancy configuration.
The reliability of each component is R=0.9. Calculate the
reliability of the system.
Solution
Using integral of Eq. 2-6:
5 3−1 (
=3 3 ∫0 1 − )5−3 ���������������

R=0.9 ⟹ ���������

Using Σ in Eq. 2-6:

�
161 Reliabilty Engineering

k−1 3−1
� 5
���� � � − Ri (1 − R)n−i = 1 − 0.9i (1 − 0.9)5−i �
i i
i=0 i=0

With Matlab:
Eq. 6-2
3*nchoosek(5,3)*int(R^2*(1-R)^2)= 6R5-15R4+10R3
Eq. 2-5-1
( − , , 1 − ), = (5 − 3,5,1 − .9) =
0.����
Eq. 2-6-1:
=1− (2,5,0.9) = 1 − 0.�����=0.99144
End of Example
Example 2-13
The system shown in the following figure has only 4
components ��A , A , A ���� A . Each component works,
independently of all the other components .Determine the
configuration of the system.
Chap. 2 Static Reliability Models 162

Solution
It is evident from the figure that the system works in the
following conditions:
If �� A , A & work,
If �� A , A & A work,
If �� A , A & work,
If �� A , A & work,
���� A , A , A & work.
Therefore the system works if at least 3 components work; i.e.
the system has a k=3/n=4 configuration. End of Example
Example 2-14
The system shown in the following figure has only 3
components ��2 3 . Each component works, independently
of the two others .Determine the configuration of the system.

1 2

2 3

1 3

Solution
It is evident from the figure that the system works in the
following conditions:

If � &2 work,
If �� &3 work,
If �� &3 work,

�
163 Reliabilty Engineering

If all 3 components work.

That is at least 2 components out of 3 must work in order to
have a working system. Therefore the system has a 2-out-of-3
configuration.
Example 2-15
Verify that a k-out-of-n configuration converts to a series
configuration for k=n and reduces to parallel configuration for
k=1.
Solution
According to Eq. 6-2, the reliability of a k/n system is:

= (1 − )

substituting k=n:

= (1 − ) = (1 − ) =

k=n⟹ =

substituting k=1:
= (1 − )

= (1 − ) − (1 − ) ⟹
0

= (1 − ) − (1 − )

According to Newton binomial expansion:

Chap. 2 Static Reliability Models 164

� + ) = ( )

then

�1 − ) = (1 − + ) =1 =1

therefore

= (1 − ) − (1 − ) = 1 − (1 − )

i.e. for k=1 k/n configuration reduces to parallel configuration.

End of Example

�����������������������������������������������

The reliability of k/n configuration is given by Eq. 5-2 :

n
� = (1 − R) (R)
x

Then

=1− (1 − ) (R)

(1 − ) (R) ( )
=1−
− +1
(1 − ) (R) ( )
− −⋯
− +2
Then

≤1− (1 − ) (2 − 7)
− +1

This is an upper bound for the reliability of a k-out-of n

configuration.

�
165 Reliabilty Engineering

Example 2-16
A pressure vessel is equipped with 6 relief values that work
independent of each other. Three values are enough for the safe
operation of the vessel. The failure probability of each valve is
0.1%. Calculate the probability of safe working of the vessel.
Solution
The reliability of this system, which has a 3-out-of-6
reliability configuration, is given by Eq. 2-6-1:

� 1 − 0.��� � 0.���

= 1− ( − 1, , )

1− (2,6,0.���� = 0.��������������

The probability of failure(unreliability probability):

�− ≅ 1.����  �� .End of Example

Notice that:
In Reliability literature related to k-out-of-n configuration
sometimes the binary variables and are defined as:

1
� �= �
0 ℎ

i.e. The structure of the ith component is described by this binary

variable.
Chap. 2 Static Reliability Models 166

⎧1 ≥
⎪
� �
⎨
⎪0 < .
⎩

The binary variable indicates the state of the k/n system.

These binary variables might be used in the optimization

problems related to k/n systems.

����������������������������

Not all designs can easily be tackled for reliability

computations. Certain designs such as those shown below are
so complex that pure parallel or series are not appropriate for the
calculation of their reliability. Such systems are known as
complex systems. There are some methods for calculating the
reliability of complex system including:
a)Enumeration method
b)�Path Tracing
c) Conditional Probability Method or Application of Bayes
theorem or Conditioning on a key element
d)Delta-Star Transformation Approach for �����������������������
e) Method based on Markov chain
f) Cut and tie set analysis �
Methods c, d and f are described below.

�
167 Reliabilty Engineering

2-7-1 Conditional Probability Method

To evaluate the reliability of a complex system using

conditional probability approach or the decomposition method ,
follow the steps explained below:
1.Choose a component, say K, which appears to bind
together the reliability of the system as a keystone component.
A poor choice may increase the number of steps in the
calculations.

2. Decompose the original system first by considering the

keystone component to be working all the time, which means is
100% reliable. Secondly, consider it as not working (which
means that it is not reliable at all or has failed). Prepare 2 new
RBDs as reduced subsystems:

in the first one, replace the working K by a line in the

reliability block diagram of the original RBD. This means that
the information can flow in either direction with no interruption.

For preparing the other subsystem remove component K

from the RBD of the original system. This is because if
component K does not work, it means that the path(s) of
information which goes/go through component K is/are
interrupted. Hence, information cannot pass through component
K
Chap. 2 Static Reliability Models 168

3)Calculate the reliabilities of the reduced subsystems and

then calculate the reliability of the system (Rsys) from:
Rsys =
Pr(system success| component K is working) +
Pr(system success | component K does not work at all)

It is worth explaining that this method has been extended to

choosing more than one key element. For more details refer to
references such as Wang & Jiang(2004).

Example 2-17(Lewis,2014Page 281)

Calculate the reliability of a system with the following RBD:

Solution
Component 2a is chosen as the key element and system
operation is conditioned on:
i) 2a works all the time
ii) 2a does not work at all
With the following symbols :
Y The event that the original system works successfully
x The event that Component 2a fails

�
169 Reliabilty Engineering

The event that Component 2a works

R1 The reliability of Components 1a&1b
R2 The reliability of Components 2a&2b�
R1 The reliability of Components 3a&3b�

The sample space(SS) includes 2 events:� = { , }.

Applying Bayes theorem:

{ }= ( | ) ( )+ ( | ) ( ) (2-8)

Let � { }= { }=
Then � { }= 1− and and Eq. 2-8 is could be written as:

� ( )= (1 − )+ (2-9)

Now, we must evaluate the conditional reliabilities and

.
For in which 2a has failed, all paths leading through 2a in
the original RBD are disconnected. The resulting RBD is as
follows:

This reduced system forms a series configuration of components

1b, 2b and 3b then:

� ( | )=
Chap. 2 Static Reliability Models 170

Conversely, for in which 2a is operational a line is drawn

instead of 2a. This action bypasses component 2b.

Therefore in this case the resulting RBD appears as follows:

which has the following reliability

� ( | ) = [1 − (1 − ) ][1 − (1 − ) ��

Substituting the expressions for ��and into Eq. 2.8, the

reliability of the original system is as follows:
� �1 − ) + [1 − (1 − ) ][1 − (1 − ) ]( ) (10-2)

2-7-2 Delta-Star Transformation Approach for

Reliability Evaluation

In this section, the reader is presented with the so-called delta

configuration and star configuration, delta-Star conversion and
the use of this transformation to simplify complex reliability
block diagrams.

�
171 Reliabilty Engineering

Figure 2-7 shows the star and delta configurations. Here we

use the capital letter R for the reliability of each component of
star connections and lower case r for that of delta configuration.
Capital letter V stands for vortices.

Fig. 2.7 Star and delta configurations

Delta- Star transformations help us to transform some reliability

networks into series parallel networks. This is dealt below.

����������������������������������������������������
����������������������������� (Grosh,1989 Page137)�����
Suppose a delta configuration is completely known and given
in Fig 2-7 and we would like to find a star configuration which
has the same reliability. The reliability equivalence of the 2
configurations in Fig. 2-7 requires that:
1)The reliability of the section between vortices V1 and V 2 in
star configuration ( R1 in series with R3 ) must be equivalent to
the reliability of same section in delta configuration
(subsystem� r in parallel with" in series with "); i.e.

R3R1 1(1r3 r1)(1r2)  C r2  rr

1 3  rr
1 2r3
Chap. 2 Static Reliability Models 172

2)With a similar argument for vortices 1 & 2,we could write :

R1R 2  1  (1  r1 r2 )(1  r3 )  A  r3  r2 r1  r1r2 r3
3)Similarly:

R R  1  (1  r r )(1  r )  B  r  r r  r r r
2 3 23 1 1 23 123

The equivalent star configuration:

Solving the above 3 equations simultaneously for R1 , R 2 , R 3 would

result in:

 ABC
R�  (2-10-1)
 B
 [�  (�  r� )(�  r�r� )][�  (�  r� )(�  r�r� )][�  (�  r� )(�  r� r� )]
1 R� 
 �  (�  r� )(�  r�r� )

ABC
2 R �  (2-10-2)
 C
 ABC
3 R �  (2-10-3)
 A
A  �  (�  r r )(�  r ) B  �  (�  r r )(�  r )
 � � � � � �

C  �  (�  r� r� )(�  r� )

Example 2-18

Suppose the components of the delta configuration in Fig 2.7

has the following reliabilities r1=0.7 , r2=0.8 , r3=0.9. Find the
equivalent star configuration.

�
173 Reliabilty Engineering

Solution

ABC ABC ABC

R1  R2  R3 
B C A

A=1-(1-r1r2 )(1-r3 )=1-(1-r3 -r1r2 +r1r2 r3 )=0.956

B=1-(1-r2 r3 )(1-r1 )=0.916 �
C=1-(1-r3 r1 )(1-r2 )=1-(1-r2 -r1r3 +r1r2 r3 )=0.926

0.900497
ABC=0.900497 R1 = =0.9831
0.916
0.900497 0.900497
R2 = = 0.9725 R 3 = =.9419
0.926 0.956

2-7-2-3 Transforming a star configuration into an

equivalent delta configuration
In this section ��������� have to be found from the 3 equations
of Sec. 2-7-2-3 in terms of ��������. The following MATLAB
code could be used to do this:
% Star2Delta.m
clc;
clear;
close all;
% Parameters input
R1=input('Please Insert R1 value in Y config.: ');
R2=input('Please Insert R2 value in Y config.: ');
R3=input('Please Insert R3 value in Y config.: ');
global A B C;
A=R1*R2;
B=R2*R3;
C=R1*R3;
Chap. 2 Static Reliability Models 174

%% calculations
options=optimset('Display','Off');
Eq=fsolve(@Delta,[0.5,0.5,0.5],options);

Eq1=Eq(1);
Eq2=Eq(2);
Eq3=Eq(3);
display(['r1= ' num2str(Eq1), ' r2= '
num2str(Eq2) ,' r3= ' num2str(Eq3)])
The sub-code Delta.m used above is as follows:

function W=Delta(r)
global A B C;
W=[1-(1-r(1)*r(2))*(1-r(3))-A;
1-(1-r(3)*r(2))*(1-r(1))-B;
1-(1-r(1)*r(3))*(1-r(2))-C];
Performing star2Delta using the data in Example 2-11
>>star2Delta
Please Insert R1 value: 0.9831
Please Insert R2 value: 0.9725
Please Insert R3 value: 0.9419
Results: r1= 0.69991 �� r2= 0.79993 �� r3= 0.90017
Special case: identical components
If a delta configuration consists of 3 identical component
with a reliability of r ; the equivalent star configuration must
have the following components:

R1  R2  R3  r3  r2  r  RY (2-11)

Conversely a star configuration with identical components RY

has an equivalent delta configuration in which all its 3
components has the reliability r obtained from the following:

r�  r�  r  RY� � (2-12)

�
175 Reliabilty Engineering

Example 2-19

In a delta configuration the reliability of the 3 subsystems is

rΔ =0.9 . Find the equivalent star configuration.

Solution

R Y = -rΔ3 +rΔ2 +rΔ =0.99045 i.e. R1 =R 2 =R 3 =R Y =0.99045

Conversely if Eq. 2-12 is solved with R Y =0.99045 ,

-rΔ3 +rΔ2 +rΔ -0.990452 =0 would give 3 answers(-0.995 ,1.0952 and

0.9) for rΔ , the acceptable answer is rΔ =0.9 .

End of Example

The following example illustrates the delta-star

transformation approach for reliability evaluation of complex
systems.

Example 2-19

A system has the following RBD. The components work

independent of each and the reliability of %90. Find the
reliability of the system.
Chap. 2 Staticc Reliability Models 17
76

Solution No.1
N : star-delta approach

The vortice
ces 1,2 and 3 in the given RBD constitute a deelta
configuration
n. The equivalent star configuration has 3
components with the following reliabilities:

ABCC ABC ABC

R1  R2  R3 
B C A

where

A = B = C = l-[l-(0.9)(0.9)](l-0.9)=0.981��������

3
R =R =R = 0.981 =0.9904
1 2 3 0.981

Replacing th
he delta configuration with this star configurat
ation

would yield the following equivalent RBD for the system

m

�
177 Reliabilty Engiineering

The reliabilit
ity of the system based on the new RBD is cal
alculated
as follows:

RSys = 0.9904
4 ���-(1-0.9904 ��������-0.9904 ������= 0.9787.

Solution No.2
N Conditional Probability Approach

The Elementt No 3 is chosen as the key element. If this elemeent is

b R SP
not functionaal, the following RBD with a reliability denoted by
would be obttained.:

RSP =1-(1-0.9
9������������������������
Chap. 2 Staticc Reliability Models 17
78

Suppose Eleement No. 3 in the original RBD works all thee time.
Replacing it with a line would result the following RBD with
w a

reliability deented by RPS .

R PS =[1-(1-0
0.9)(1-0.9)][1-(1-0.9)(1-0.9)]=0.9801

Now Rsys � thee reliability of the system) is calculated using

g Bayes

Rule (Eq. 2-8) as follows:

Rsys =RSP������3���PS��3�����������������������������785 �

���������
���������������������������������������
������������������������������������������������

In this seection a procedure is introduced for calculaating an

upper and a lower reliability bound for complex system
ms. This
procedure iss based on the so-called tie and cut sets. The RBS of
some compleex systems are shown below.

�
179 Reliabilty Engiineering

Fig. 2-9 Three complex systems

Before perfo
orming the calculations, some definitions are
introduced:

Definition of
o minimal cut set

This set is
i a minimal set of components which by faiiling
guarantee th
he failure of the system(Grosh,1989 page125);; in
other words the failure of its components cause interruptiion of
all paths from
m the input to the output of the system.

Definition of
o minimal path

A minimal path
p is a minimal set of components by functiioning
ensures the system
s operation(Grosh,1989 page125).
Chap. 2 Staticc Reliability Models 18
80

Notice that

-Each compllex system has usually several minimal path and

a cut
sets. A set co
ould be both cut and path.

- In the foreggoing discussion, by mentioning cut set, minimaal cut

a by mentioning path set, minimal path set iss meant.
set is meant and

Example 2-21
2 (Grosh,1989 page 125)

Find the miinimal path and cut sets for the following RB
BD.

Solution

The minimal
al cut sets are:

Cut set no. components

1 C1C2

2 C4C5
3 C1C3C5
4 C2C3C4

The minimal
al path are:

�
181 Reliabilty Engiineering

Cut set no. components

1 C1C4
2 C2C5
3 C1C3C5
4 C2C3C4

mple
End of Exam

Example 2-22
2 (O'connor,2003 page175)

The RBD
D of a system is given below. Show the miniimal cut

and path setss on the RBD.

Solution

The minim
mal cut sets are (1-3 2-3 and 4 )as shown below:

The minim
mal path sets are (1-2-4 3-4 ) shown below::
Chap. 2 Staticc Reliability Models 18
82

End of exam
mple

Example 2-23
2 (����������������������

What are thee minimal cut and path sets between source a and
target b giveen below?

Solution
If all co
omponents fail but Component 1,the con
nnection
nd b does not interrupts then the set �1} is a tie set
between a an
If all compon
nents fail but Components 2& 5, the connectiion
nd b does not interrupts then the set �2,5} is a tie
between a an
set
�
Functioning of the set with minimum elements 6,4 ensur
ures the
system operaation , then Set {6,4} is a minimal path.
���

�
183 Reliabilty Engineering

The other cut and tie sets are determined with similar reasoning.
The following table shows the minimal cut sets and minimal
paths.

The minimal cut sets and minimal paths of Example 2-23

Path no. Components Tie set no. Components
1 5,4,1 1 1
2 2,6,1 2 2,5
3 3,6,5,1 3 6,4
4 4,3,2,1 4 2,3,4
5 6,3,5

End of Example

2-8-1 Calculation of reliability upper& lower bounds

for complex systems using auxiliary networks

To compute upper and lower reliability bounds for a complex

system, based on the minimal cut sets and minimal paths, two
auxiliary systems are constructed(Grosh, 1989, page 125).
1)Auxiliary network N1 is composed of parallel
configuration of all the minimal path elements in series. This
network is based on this fact that as far as one minimal path
work the system works.
2)Auxiliary network N2 is composed of the series
configuration of all the minimal cut elements in
parallel(Grosh,1989 p 127).
Chap. 2 Staticc Reliability Models 18
84

After caalculating R N 1 , the reliability of Network N1and

a that

or Rsys ,
of network N2 ( R N 2 ),we have the following inequality fo

the reliability
ty of the original system (based on Grosh, 1989 p 125 -6)
RN 2  Rsys  RN 1 (2-13)
For somee reasons such as dependence of the subsysstems of
Network N1, The calculated upper bound( R N 1 ) is usuallly over-

estimated by
y this method. If R N 1 exceeds 1 set R N 1  1

Example 2-24
2
Draw th
he auxiliary networks for calculating the up
pper and
lower bound
ds for RBD of Example 2-21.
Solution a)Auxiliary network N1
Network N1 which is a parallel configuration of all the
h elements in series is shown below:
minimal path

The reliabiliity of this RBD given by the following relattionship

is the upper bound for the original system reliability of Example
E
2-21:

�
185 Reliabilty Engiineering

RN 1  1 (1 r1r4 )(1 r2r5 )(1 r1r3r5 )(1 r2r3r4 )

b)Auxiliary network
n N2
Network N2
N which is the series configuration of all the minimal
cut elements in
i parallel is shown below:

Network N2
The reliabilit
ity of the above RBD given by the following
relationship is the lower bound for the original system
reliability:
RN2 =
1-(1-r1)(1-r2 ) 1-(1-r4 )(1-r5 ) 1-(1-r1)(1-r3)(1-r5 ) 1-(1-r2 )(1-r3)(1-r4 )

Then RN 2  Rsys  RN 1 � �

Example 2-25
2
Draw thee auxiliary networks for calculating the upp
per and
lower bound
ds for the following RBD. The number in thee box is
the reliability
ty of the component.
Chap. 2 Static Reliability Models 186

Solution
The minimal paths are:
1-3-6-7 1-3-5-7 1-2-4-7 1-2-5-7 �
The minimal cut set are:�
1 7 2-3 4-5-6 2-5-6 3-4-5

The auxiliary networks N1 and N2 are shown below.

Network N1

�
187 Reliabilty Engineering

������������������Network�N2

After calculating the reliability of Network N1( R N 1 ) and that

of Network N2( R N 2 ) we could write RN 2  Rsys  RN 1

2-8-2 An approximate formula for the upper and

Lower reliability bounds of complex systems

Approximate bounds on system reliability from minimal cut

sets and minimal path are given below (O'Connor& Kleyner 2012, p153)

Cut sets Paths

C T
(2-14)
1    (1  Ri )  Rsys    Ri
j 1 i A j k 1 i B k

c The number of minimal cut sets

N The number of ct set

T The number minimal paths(tie sets)

Aj The components of jth cut set, j 1,2,...,c

The components of kth path, k 1,2,...,T

Chap. 2 Static Reliability Models 188

C
1    (1  R i ) =
j 1 i A j

1-( Product of unreliabilities of the components of A1+��� �roduct of

unreliabilities of the components of Ac )
T


K 1 i  B K
Ri 

Product of reliabilities of 1st path components +����������������������������

of last path components

In the relationship 2-14, if the calculated upper bound is

greater than 1, let the bound equal to 1; if the calculated lower
bound is negative let the lower bound equal to zero.
Example 2-26
Calculate the upper and lower bound for the reliability of the
following system. Each component has the reliability of 90%.

Solution
The minimal cut sets and the minimal paths are given in the
following table.

�
189 Reliabilty Engineering
�
�
Cut Set No. j� components�
Path No. k� components�
j=1� 1-3
k=1� 1-2-4
j=2� 2-3
k=2� 3-4
� j=3� 4

Method 1: Using auxiliary networks

����������� ������������

��������������������������������� �����������������������������������

�
�
The reliability of the system( Rsys )lies between:

R N1 =1-(1-R1R 2R 4 )(1-R 3R 4 )=0.948

and

R N =[1-(1-R1)(1-R3 )][1-(1-R2 )(1-R3 )])(R4 )=0.8821

2

Then 0.8821<Rsys <0.948 .

Chap. 2 Static Reliability Models 190

Method 2: Using Relationship 2-14

C T
1    (1  Ri )  Rsys    Ri 
j 1 i A j k 1 i B k

1-  (1-R1 )(1-R3 )+(1-R 2 )(1-R3 )+(1-R 4 ) <Rsys <R1R 2R 4 +R3R 4

1-  (1-R1 )(1-R 3 )+(1-R 2 )(1-R 3 )+(1-R 4 ) =0.88

R1R2R4 +R3R4 =0.729+0.81=1.539 1

Then 0.88<Rsys <1.

It is obvious that this was an illustration for the methods.

There was no need to apply these 2 method to this simple RBD;
because its exact value of reliability is simply calculated as
follows: [1-(1-R1R2 )(1-R3 )](R4 )=0.8829 �

�����������������������������������������������

According Bays' theory, if the sample space of an experiment is

SS  {H 1  ...  H K } and H i 's are mutually exclusive( i.e.�

Hi  H j = for ��������������≠� ), then:

Pr( Hi ) Pr( B / Hi )
Pr( Hi | B)  k (2-15)
 Pr(H ) Pr(B H )
i 1
i i

where B is an event defined on Sample space SS.

�
191 Reliabilty Engineering

Example 2-27(K&L page 392)

Suppose a design engineer has developed a new mechanical

system that has never been built or tested before. The engineer
believes, based on his previous experience and intuition, that if
the system has been designed properly to meet the performance
criteria, the time to failure, is normally distributed with a mean
of 50,000 kilometers. If the system is improperly designed, the
mean life may be 30,000 kilometers. Based on his experience,
the engineer has good confidence in his design. A priori, he says
that the probability that the design has a mean life of 50,000
kilometers is 0.80, and hence 0.20 is the probability that the
design has a mean life of 30,000 kilometers.

A single prototype is built and tested in a simulated environment

that duplicates as nearly as possible the actual environment. The
system is tested, but economic considerations dictate that (he
testing be stopped at 40,000 kilometers. The engineer also says
that, based on past experience, it is known that the standard
deviation for the life of the system is 10% of the mean life.
The objective is to predict the reliability of the design .
Solution
Let
A = the event that the system has been tested and operated
successfully for 40,000 kilometers.
B1 =the event that the mean life is 50,000 kilometers, Pr(B�) ��
.
Chap. 2 Static Reliability Models 192

B2 = the event that the mean life is 30,000 kilometers, Pr(B� ) �.�
SS=B1 UB2
Pr(Bi )Pr(A Bi )
Pr(Bi A)= i=1,2
Pr(B1 )Pr(A B1 )+Pr(B2 )Pr(A B2 )
Pr(B1 )=0.8
 40000-50000 
Pr(A B1 )=Pr(X�������μ����������  �� 
 (50000)(0.10) 
=Pr(Z�����������

40000-30000
Pr(A B2 )=Pr(Z> )=0.00045 
30000�����

Pr(B1 )Pr(A B1 )
Pr(B1|A)= =0.9999
Pr(B1 )Pr(A B1 )+Pr(B2 )Pr(A B2 )
Pr(B2 |A)=1-Pr(B1|A)=1-0.9999
Suppose later, after production stage, one of these devices
was selected for a mission of 35000-km- experiment. Assuming
a normally distributed life time with mean 50000 km and
=5000 km the reliability is given by
35000-50000
R=Pr(X>35000)=Pr(Z> =Pr(Z>-3)=0.99865
5000
We are 99.99% confident about the mean used for this
calculation because Pr(B1 | A)  0.9999. Therefore we are 99.99%

confident that the reliability of the device is 0.99865 .

However before performing the test on the prototype which

lead to a life of 40000 km we were able to state with 80%
certainty(obtained from the engineers original beliefs) that the

�
193 Reliabilty Engineering

35000-km reliability of the device is 0.99865 . Because we

were 80% confident about the mean used for the calculation.
After the testing which was terminated after 40000 km, the
probability of the event ����� �� conditioned on Event A
was obtained 99.99%. Therefore the confidence for the
predicted reliability shows increase using the Bays' theorem.
End of Example

Exercises1

1-(Problem 1 Page 68 K&L) Calculate the reliability of each of the

following RBDs, where each component has the indicated
reliability

�����������������������������������������������������������
�
���������������������
Chap. 2 Static Reliability Models 194

2. A system consists of 100 parts connected functionally in

series. Each part has a 1000- hour reliability of 0.9999.
Calculate the reliability of the system.
3 A system is comprised of four major subsystems in parallel.
Each subsystem has a reliability of 0.900. At least two of the
four subsystems must operate if the system is to perform
properly. What is the reliability of this system?
4 A system is comprised of 10 subsystems connected
functionally in series, [f a system reliability of 0.999 is
desired, what is the minimum subsystem reliability that is
needed?
5- -(Problem 5 Page 69 K&L) Assume that 4 wheel bolts are
adequate from a design standpoint. However, the wheel
attachment under consideration has 5 bolts. If the chances

�
195 Reliabilty Engiineering

of losing a wheel
w bolt are 0.00001, what is the reliability
y of this
bolt system??
6- -(Problemnn 7 Page 70 K&L) The system diagram given beelow
describes thee circuitry for a neutral start switch on a manu
ual
automobile transmission.
t According to the service manuaal, in
order to startt this vehicle the clutch pedal must be fully
depressed an
nd the ignition switch must be in the start posi
sition.
(a) Define reliability
r as it relates to this system
(b) Draw an
n appropriate reliability block diagram(RBD)
(c) Assumin
ng that each functional block in your RBD has a
0.0001 chanc
nce of failing, calculate the system reliability.

7- A manufa
facturer wishes to know the reliability of a skiid
protection sy
ystem Io be used on military tractor trailers. The
T
system consiists of:
(a) Two batt
ttery or generator powered sensors per wheel..
(b) One logiic unit per sensor to predict wheel skid.
(c) A comm
mand unit, which operates an electric or an en
ngine
vacuum soleenoid.
Chap. 2 Staticc Reliability Models 19
96

(d) The soleenoids in (c) operate an actuator that controls the

pressure to the
th brake.
. The system
m diagram (not reliability diagram) is shown below.
b

8) A DC bat
attery has a time to failure that is normally disstributed
with a mean
n of 30 hours and a standard deviation of 30 ho
ours,
(a) What is the 25-hour reliability?
(b) When sh
hould a battery be replaced to ensure, that theere is no
more than a 10% chance of failure prior to replacement?
(c) Two baatteries are connected in parallel to power a light.
Assuming, that
t the light does not fail, what is the 35- hour
reliability fo
or the power source?
(d) A particcular battery has been in continuous use for 30
0 hours.
What is thee probability that this battery will last an
nother 4
hours?
9)Calculate the reliability of the following two systemss, where
each compon
nent has the indicated reliability

�
197 Reliabilty Engineering

10)A customer of a bank uses 2 different cards for Automatic

Teller Machines. One of them is connected to 2 accounts with
reliabilities RI and RII and the other is connected to the account
with reliability RI. The reliability of the cards is R. Which
card do you prefer?

Ans. RI  RII .
11)To have a 6-component series system of at least 95%
reliability, how many components do you suggest?
12) A system consists of several components with 94%
reliability To have a system with 95% reliability, what
configuration do you suggest and many components?
14) In the following RBD, each component has the indicated
reliability. B2 is a standby component which replaces B1 by a
switch. The failure probability of the switch and the standby
component when they are needed is negligible. Calculate the
reliability the system with the given RBD and compare it with
the case as if there is no redundant standby component.
Chap. 2 Static Reliability Models 198

It is best to start every thing with

trustfulness and
end it with
faithfulness

�
199 Reliabilty Engineering

Chapter 3
Reliability
Considerations
in Design
+UGF
Technique
Chap. 3 Reliability in Design + UGF Technique� 200

3
Reliability Considerations in Design + UGF Technique

Aims of the chapter

This chapter is divided into two sections. First section deals

with reliability considerations in design. The second section
introduces one of the methods used to conduct reliability
analysis; i.e. the universal generating function(UGF) method
which is a method of modern discrete mathematics.

3.1 Reliability Considerations in Design

The design process dictates the system configuration and the
configuration chosen influences the reliability level as well as
the cost of achieving this level. Thus, a preliminary reliability
analysis as well as the many other design factors should be
considered during the design phase. .(K&L page 62)
Since the designer is the system architect he or she should be
familiar with the basic reliability analysis concepts that can be
used to evaluate the design. Only after the design is completed can an
independent reliability group analyse and test the product. So it is
important that the designer evaluate the reliability levels and costs of
various designs before the final choice is made.(K&L page 63)

A frequently used measure of complexity is the number of

components in a system. It is a fundamental tenet of reliability

�
201 Reliabilty Engineering

engineering that as he complexity of a system increases, the

reliability will decrease, unless compensatory measures are
taken(Lewis, 1994pace).

This section will emphasize some trade-offs between

reliability and the number of components. This might be helpful
to a designer in developing alternatives(K&L page 63).

3.1-1Reliability considerations in series configurations

Consider the series system shown in Fig 3.1

1 2 3 n

Fig 3-1 A series configuration

If the reliability of each component is equal to R i.e.

� R = R = ⋯ = R = R then according to Eq.2-1 the
system reliability(� ) is given by �� = R .
� depends on R and n(reliability and number of
components). This relationship is shown in Fig 3.2.
Chap. 3 Reliability in Design + UGF Technique� 202

Fig 3.2 The relationship between An n-component system's

reliability, and the number components for 3 values of R

Some considerations on the design of a series configuration

follows. According to Fig 3.2:

1)For a given component reliability(R), the reliability of a series

system can be improved by decreasing the number of
components in series. Conversely the system reliability id
decreased as the number of components increases.

2)for a given number of components , the reliability of a series

system will improve if components of greater reliability are
used.

3)When the number of components is increased, the system

reliability will not change if components of appropriate greater
reliability are used.

The following figure also conveys some concepts similar to

those that does Fig. 3.2.

�
203 Reliabilty Engineering

Fig.3.3 Series system reliability as a function of number and

reliability of components (Lewis,1994 page 253)

3.1-2 Reliability considerations in parallel configurations

The paralleling of components is usually mentioned as a

means to improve system reliability. However the gains are not
always realizable(K&L page 64). Consider the RBD of m-

component system in which all components are actively

parallel. If the reliability of components are the same and equal
to R, the reliability of the system shown in Fig. 3-4 is:
� = 1 − (1 − R) . Figures 3.5 and 3.6 plot this relationship.
Chap. 3 Reliability in Design + UGF Technique� 204

Fig 3-4 A Parallel-active system

Fig. 3.5 Parallel system reliability as a function of reliability of

components for 4 values of m (K&L page 64)

�
205 Reliabilty Engineering

1
R=0.9
0.95
R=0.6
0.9 R=0.7

R=component reliability in
������������������������
System Reliability

0.85
series configuration

0.8

0.75

0.7

0.65

0.6
1 2 3 4 5 6 7 8 9 10
no. of components
.

Fig. 3.6 Parallel-active system reliability as function of no. of

components for 3 values of R.

Some considerations regarding this system follows:

1- For a given component reliability the more the number of

components (m) the more the system reliability; however for
m>4 the increase slows down(see Fig. 3.5 and 3.6 )

2- To use cheaper and less reliable components and at the same

time to keep the system reliability fixed, the number of
component has to be increased, as it is evident from Fig. 6.3.

3- For a given number of components, the more the component

reliability (R) the more the system reliability.
Chap. 3 Reliability in Design + UGF Technique� 206

For example if m =2 Fig. 3-6 gives the system reliability0.84

for R=o.6 and 0.91 for R=o.7.

It is worth knowing that

designing a parallel system for a mechanical device is usually
extremely difficult. Some forms of parallel arrangement such as
providing spare parts (a standby parallel arrangement) or using a
load-sharing design (a shared parallel arrangement) are probably
more representative of the true situation.(K&L).

3.1-3 Reliability considerations in series-parallel

configurations

Remember that given an n-component series system, we can

either provide redundant components, which give a system
design diagram as shown in Fig.3.7, or provide a total redundant
system as shown in Fig.3.9.
As you know the former redundancy is known as low-level
redundancy whereas the latter(system level redundancy) is also
called high-level redundancy.

�
207 Reliabilty Engineering

Fig. 3-7 Component redundancy or low-level (LL) redundancy

Now we would like to make some comparisons between LL
and HL redundancies. Assume all components are independent
of each other and have the same reliability of R.

Reliability of LL redundancy
The reliability of LL configuration in Fig 3.7 is given by:

= [1 − (1 − ) ] � (3-1)

This equation is plotted in Fig 3.8

Chap. 3 Reliability in Design + UGF Technique� 208

Fig. 3.8 LL configuration's reliability in terms of no. of subsystems (n)

and number of components(K&L p66)

Reliability of HL redundancy
The reliability of HL configuration in Fig 3.9 is:

� = 1 − [1 − ] ، (3-2)

where

m is the number of subsystem and

n is the number of components in each subsystem

�
209 Reliabilty Engineering

Fig. 3.9 High-level or system redundancy

Eq. 2-3 is plotted in Fig. 3.10 in terms of n for 3 values of m and

2 values of R.

Fig. 3.10 HL configuration's reliability in terms of no. of

components of subsystems (n) (K&L p67)

Now we would like to compare HL and LL redundancies.

Chap. 3 Reliability in Design + UGF Technique� 210

By comparing the graphs in Figs. 3.8 and 3.10 it is evident

that the low-level redundancy gives a higher system reliability in
all cases. However, the difference is not as pronounced if
components have high reliabilities. Basically the 2 Figures
indicate that providing spare components will result in better
overall reliability than providing a spare system. Of course, this
can be applied at different levels to subsystems, depending on
the possible system breakdown, for, in some instances, design or
system peculiarities make it impossible to apply all of these
rules. Also the total system operation must be considered. For
instance, if your automotive brake system fails at 80 km/h in
heavy traffic it would not do you any good to have a complete
set of components in your glove compartment. So the rules must
be used as guides and applied with discretion( K&L page67-68).
Example 3.1 Find the reliability of the LL redundancy given
in Fig. 3.7 and that of the HL redundancy given in Fig 3-9 for
���� � � m = 4, R = 0.7
���� � � m = 2, R = 0.9�
Solution
a)
From Eq. 3.1
� � = [1 − (1 − R) ] , n = 3 m = 4, R = 0.7���
( R sys ) LL  0.9759
�
From Eq. 3.2
� R = 1 − [1 − R ] , =3 = 4, = 0.7
( R sys ) H L  0.81368
b) ����� � 3 m = 2, R = 0.9From Eqs. 3.1 and 3.2⟹
(Rsys )HL =0.93 (Rsys )LL =0.97
End of example

�
211 Reliabilty Engineering

3-2Universal Generating Function(UGF) analysis of

Reliability Systems
Before dealing with Universal Generating Function.; It is
worth reminding that in the reliability analysis of system 2
different systems are identified: binary -state system and multi-
state system (MSS).
a binary -state system assumes only two possible states for a
system and its components: either perfect functional or
completely down
and multi-state systems reliability models allow both the
system and its components to assume more than two levels i.e.
different performance levels and several failure modes with
various effects on the entire system performance . In other
words, in multi-state systems the system and its components
have multiple possible states: some intermediate states as well
as complete failure and perfect functioning.
Different methods, such as Monte Carlo simulations,
extension Boolean models, stochastic processes and the
universal generating function (UGF) method have been
proposed to conduct the reliability analysis of MSSs1.
Although the UGF method which is a method of modern
discrete mathematics has a high computing speed in the
reliability assessment of multi-state systems (MSSs), it can be
�����������������������������������������������������������
1
Jinhua, Mi, et al, 2015
Belief Universal Generating Function Analysis of Multi-State Systems
Under Epistemic Uncertainty and Common Cause Failures IEEE
Transactions on Reliability Vol. 64 No.4
�
Chap. 3 Reliability in Design + UGF Technique� 212

used for the analysis of binary-state systems. we proceed now

with the use of UGF in the latter case.

Universal generating function(UGF) is an extension of

moment generating function(MGF) and probability generating
function(PGF). UGF could be used in determining the prob-
abilistic distribution of complicated functions of some discrete
random variables. Before proceeding with UGF, some
definitions are reminded.

3.2-1 Moment generating function of discrete random variables

Consider a discrete random variable(rv) X which can take

values x 0 ,..., x k such that ���X = x ) = p � and ∑ p = 1. the
mapping � ⟶ p is usually called probability mass function.
The mean of and MGF of this rv is:
� )=∑ �
k
( ) ��� ( )= ( )= p e
i 0
i
t x i

For example if X has a binomial distribution� � , ),

then� ( ) = ( +1− ) .

Some properties of moments generating functions (MGFs) are

as follows:

1. The MGF of a random variable(rv) is unique i.e.

if the MGF exists for an rv, then there one and only one
distribution associated with that MGF uniquely defines the
distribution of the rv.

2. The MGF of the sum of some independent rv's is

equal to the product of the MGFs of the rv's:

�
213 Reliabilty Engineering

k
� ∑
( )=  � �.
i 1
(k )
d
3. E (X k )  � )|
dt k
i.e. the kth derivative of � �with respect to t gives the
th
value of k moment of the distribution about the origin,
for t=0.
4. The additive property of some distributions such as
Poisson and normal distributions could be verified by
MGF.
3.2-2 Z-transform or probability generating function of
discrete random variables
The probability generating function of a discrete random
variable is defined as follows:

� �= ( )= (3 − 3)

This function is also called Z-transform if variable X.

Example 3.2
If a variable X takes on the values � = 1 , � � 2 and
� � 5 with probabilities 0.3 ,0.5 and 0.10. Find the Z-transform
of the variable.
Solution
� �= = 0.3 × + 0.6 × + 0.1 ×

End of Example

Some properties of z-transform

A useful property of z-transform is its ability to solve
difference equations. Some other common properties are:

1.The probability function of a random variable whose z-

transform is � ) is derived from the following relationship:
( )
� × ( )| (3-4)
!
 Chap. 3 Reliability in Design + UGF Technique� 214

2. The first derivative of � ) with respect to z gives the value

of the distribution mean, for z=0.

� )= ( )= ( )= (3 − 5)

=1 ⟹ ( )=∑ = ( ) ��
3.The z-transform of the sum of some independent rv's is equal
to the product of the z-transforms of the rv's:
k

∑
� )=  � ������������������������������� − ��
i 1
4. If in the definition of the MGF � is replaced by z, then one
gets z-transform of the random variable.
Example 3.3

Suppose k independent trials each having 2 outcomes:

success with probability and failure with probability 1-� . are
performed independently. For jth trial let defined below:
=1 = =1
� ‫�ﻳﺎ‬ ( )=
=0 =1− 1− =0
Therefore the z-transform of is:� ( )= + (1 − )
k
If X   X j then X represents the number of successes that
j 1

occurs in the k independent trials and the z-transform of X is:

ω (z) = [πz + (1 − π)] × … × [πz + (1 − π)] = [πz + (1 − π)]

This z-transform is that of a binomial random variable with

parameters (k, ) whose probability function is given by:
� (1 − ) 0≤ ≤

End of Example

�
215 Reliabilty Engineering

3-2-3 The Universal Generating Function(UGF)

Consider independent random variables ��, ,…, with

mapping � . If we want to find the probability function of
� � ), we have to obtain a vector Y composing all
possible values of f and the probability of the occurrence of the
values.
Each possible value of f corresponds to a combination of the
values of its arguments �…, . Let the probability function
of taking values be represented by:

x= �, … , p= �, … ,

Then the total number of possible combinations constituting

k
the range of � … ) is �  � � , where is the number of
i 1

possible values that takes.

Since � are independent the probability of the jth
combination of the realization of f variates is equal to:
probability of jth variate of f= =∐
where(Levitin,2010page)
is the probability of the realization of the arguments
composing the combination.
And the corresponding value of f can be obtained as:
� ,…,X .
Some combinations might have the same values. Since all
combinations are mutually exclusive, therefore the probability
Chap. 3 Reliability in Design + UGF Technique� 216

that the function f takes on some value is equal to the sum of the
combination producing this value(Levitin,2010 page 6). Let
be a set of combinations producing the value . If the total
number of different realization of the function � � � is H,
then the probability function of f is:
Y  f h :1  h  H  ,

 
n (3 − 7)
q    pij i :1  h  H 
 ( x 1 j1 ,..., x 1 j n )A h i 1` 

Example 3-4

Consider independent random variables X 1, X 2 with the

following probability functions:
0.1 = 0.5
� � = 0.6 =1 ( ) = 0.6 =1
0.4 =4
0.3 =2
Find the probability function of � � � �= .

Solution
All possible combinations of X 1 , X 2 and the probability
function of Y is given in the following table:�

X1
� X2 � �
1 1 1 0.5
0.6  0.1=0.06
2 4 0.5 2 0.4  0.1=0.04
3 1 1 1 0.6  0.6=0.36
4 4 1 4 0.4  0.6=0.24
5 1 2 1 0.6  0.3=0.18
6 4 2 16 0.4  0.3=0.12
As the table shows some combinations have the same value .
Since all combinations are mutually independent then the

�
217 Reliabilty Engineering

probability of the occurrence of this same value is the some of

the probabilities of the combinations producing the value. e.g.:
� � 1) =
���X = 1, X = 0.5) + ���X = 1, X = 1) + ���X = 1, X = 2) ⟹
( = 1) = 0.�� � 0.�� � 0.�� � 0.6
Therefore according to the calculations in the table;
� � � = 1 , = 2 , = 4 , = ���} , �
= (0.6, 0.��� 0.��� 0.���.End of Example
For solving problems such as the one given in Example 3-4,
another approach called the UGF technique could be used. The
technique, based on using z-transform and a composition
operator (denoted by ⊗ ), is described below.

Let takes on � ,…, with probabilities ,…, ,

the corresponding z-transform is the following polynomial:

� )= (3 − 8)

As you know the z-transform of the sum of independent random

variables X 1 ,..., X n is the product of their z-transforms:

The probability function of a combination of several

independent random variable such as Y given in Example 3-4
could be obtained from its z-transform. Therefore if one could
find the z-transform of a combination, it will be easy to obtain
the probability function of the combination.
Chap. 3 Reliability in Design + UGF Technique� 218

UGF Technique

In the so-called UGF technique, To calculate the z-transform,

U(z) , of every arbitrary combination(function) of independent
random variables X1,...,Xn , replace the product operator ( ) on
the z-transforms in Eq. 3-6 with an appropriate operator denoted
by ⊗ . Here the z-transform of random variable of

independent variables X 1,..., X i ,..., X n is denoted by � ).

The z- transform of � ,…, ) is denoted by ����� (��

(Livitin, 2010page 8). According to this notation for 2 variables:

� ) =⊗ ( ), ( )] = [ ( )⊗ ( ) (3 − 9)

for n variables(Livitin, 2010page 8):

�z) ( ) =⊗ ( ( ), … , ( )� (3 − ����

U (z) =� .
 f ( x i j1 ,..., x n j n ) 
n
⊗ P Z =
  pij i z
…  ��� (3 − ���
 i 1 
The technique based on using z-transform and composition
operators�⊗ is named universal z-transform or universal
(moment) generating function (UGF) technique (Livitin, 2010
page 8 ). UGF technique has applications such as finding the
probability function of an arbitrary function of several
independent random variables and finding the reliability of
complicated systems. For other applications refer to reverences
such as chapter 2& 3 in Levitin(2010).

�
219 Reliabilty Engineering

Notice that(Livitin,2010 page 8):

1-Although � � resembles a polynomial, � ) is not
necessarily a polynomial.
2-When the � ) represents the probability function of a
random function � ,…, ), the expected value of this
function can be obtained as the first derivative of ���� at z=1.

Example 3-5(Levitin, 2010, page9)

Consider the probability function of Y from the table in
Example 3-4. The z-transform of Y takes the form:
� ) = 0.�� � 0.�� � 0.�� � 0.�� � 0.�� � 0.��
Merging the like forms results in:

� ) = 0.6 + 0.�� � 0.�� � 0.��

As you may have noticed, this function represents the

probability function for Y as follows:

� = (1, 2, 4, ��� , q = (0.6, 0.��� 0.��� 0.���

which is the same as what was obtained in Example3-4.

The described technique of determining the probability

functions is based on an enumerative approach, which is
extremely time consuming. Fortunately, many functions used in
reliability engineering produce the same values for different
combinations of the values of their arguments ( ′ ). The
combination of recursive determination of the functions with
simplification techniques based on the like terms collection
Chap. 3 Reliability in Design + UGF Technique� 220

allows one to reduce considerably the computations needed to

obtain the probability function of complicated functions.
The following procedure is easier for solving this example.
Based on the data in Example 3-4 the u-function of and is
as follows:

U1 (z )  0.6z 1  0.4z 4 , U 2 (z )  0.1z 0.5  0.6z 1  0.3z 2

Let u-function of ��� be denoted by � �, then

according to Eq, 3.9

U Y ( z )  U 1 ( z )  power U 2 ( z ) 
U Y ( z )  (0.6 z 1  0.4z 4 )  power (0.1z 0.5  0.6z 1  0.3z 2 )

(10.5 ) (11) (12 )

U Y (z )  0.6���� Z +0.6�0.6��Z +0.6�0.3��Z +
(40.5 ) (41) (42 )
0.4�����Z +0.4�0.6��Z +0.4�0.3��Z 

1 � � ��
� � = 0.6 + 0.�� + 0.�� + 0.��

From this function the probability function of Y is obtained:

� (1, 2, 4, ���
�������������� = = (0.6, 0. ��� 0.��� 0.���
End of example

Example 3-6( based on Livitin, 2010 page 9)

Random variables �…, are independent and the data for

their probability functions are given in the following table:

�
221 Reliabilty Engineering

( ) ( )
x 1,0 =8 p1,0 =0.7 5 0.6 0.1 0 0.6 1 0.5
x 1,1 =10 p1,1 = 0.3 8 0.3 0.5 2 0.4 1.5 0.5
12 0.1 0.4

Find the probability function of

� � � �=[ ( , )+ ( , )]( ).

Solution
The total number of term multiplication procedures that one has
to perform using enumerative approach is
� × 3 × 3 × 2 × 2 = ��; however applying UGF technique as
performed below reduces this amount to only 26 (Livitin,2010
page11).

The u-function of the variables are:

� �� � , +
,
, = 0.7 + 0.3
( )= , , +
,
, +
,
, = 0.6 + 0.3 + 0.1
( )= , , + , , +
,
, = 0.1 + 0.5 + 0.4
( )= , , +
,
, = 0.6 + 0.4
( )= , , +
,
, = 0.5 + 0.5 . .

Let us introduce the following 3 auxiliary random variables:

� ( , ) = ( , ) = +

Therefore � . Using composition operators on pairs of u-

functions, the probability function of Y is obtained as follows:
��
� �= ( )⊗ ( )=
= (0.7 + 0.3 ) ⊗ (0.6 + 0.3 + 0.1 )=
Chap. 3 Reliability in Design + UGF Technique� 222

( , ) ( , )
���� + 0.�� +
� � . � � .
( , ) ( , )
0.�� ���
� × . � � .
( , ) ( , )
+0.�� + 0.�� ⟹
( ) = 0.�� � �.�� + 0.1

( )= ( )⊗ ( )
= (0.1 + 0.5 + 0.4 ) ⊗ (0.6 + 0.4 )=

( , ) ( , ) ( , [) ( , )
0.�� +0.�� +0.3 0.2
( , ) ( , )
+0.�� + 0.�� ⟹
( ) = 0.�� � 0.��

� )= ( )⊗ ( )=

= (0.�� � �.�� + 0.1 ) ⊗ (0.�� � 0.�� �=

= 0.���� � 0.���� � �.���� + +0.���� + 0.��� + 0.��� =

�������
� � = 0.���� � 0.���� + 0.����� + 0.���
��
� � = ( ) ⊗× ( )
= (0.���� � 0.���� + 0.�����
.
+ 0.��� ) ⊗× (0.5 + 0.5 )
After necessary calculations and simplification, the final answer
for � �z) is:

� ) = 0.���� � 0.���� + 0.���� + 0.���

+ 0.���� + 0.���� + 0.���
From � (z) the probability function of Y is obtained as follows:

Y=( 8, 10, 12, 14, 15, 18, 21)

q= (0.02016, 0.1998, 0.2822, 0.018, 0.1998, 0.0806, 0.018)

End of Example

3-2-4 derivation of the reliability using UGF�

Given the UGF of a system, its reliability could be

estimated> This is illustrated in the following Example.

�
223 Reliabilty Engineering

Example 3-7
The universal generating function of a system is:
�
� ) = 0.1 + 0.�� + 0.4 + 0.�� ,
Find 20-hour reliability of the system.
Solution
= ( > ��� = 0.�� � 0.�� � �.���
End of Example

3-2-4 Reliability Analysis of Binary -State Systems using UGF�

Symbols
System Reliability
The reliability of jth subsystem
� The state of jth subsystem(either 1=working or 0=down)
X The system structure function
The UGF method is very effective for the reliability analysis of
multistate systems; however it could be used for binary-state
systems, though that effective as compared to conventional
methods (see Kuo &Zuo,2003).

This section focus on the allocation of UGF technique to

reliability systems whose components and the system itself have
only 2 states: either working or not.

ّFig 3�10 The RBD of a ‫ﺱ‬eries-parallel system

(Livitin, 2010page 30)
Chap. 3 Reliability in Design + UGF Technique� 224

Consider the RBD of a system given in Fig. 3-10 with the

system structure function(Livitin, 2010page 32)

� [ , ( , )]
where � is the state of jth subsystem(with 2 values :either
� =1=working or � �0=down).
Let
Rj=Pj be the reliability of jth subsystem for a fixed mission
time, the probability that is on working conditions during the
mission time
and 1-Pj be the probability that the jth subsystem is down.
Then the expected value of is:
�0 1− +1 = =
where � is jth subsystem reliability.
Therefore for a fixed mission time the system reliability
equals the expected value of .
Similarly the reliability of the system for a fixed mission
equals the expected value of the system structure function X:
= ( ), (3-12)
Where
� ( , ,…, )
The state of jth subsystem(either 1=working or
0=failed)

Therefore for a fixed mission time the system reliability equals

the expected value of X.

�
225 Reliabilty Engineering

Usually the element reliability vector is known and we would

like to obtain the system reliability as a function of ′ . In
systems with independent elements, such functions are available
and depend on the system structure.

Example 3-8(Livitin, 2010Page 31)

Consider the following RBD, Xj denotes the state variable

of jth subsystem taking values �� � � ���1.

Let the static reliability of jth subsystem = �,�then

= = 1− = 0 ��� = 1 = 1,2,3

If the subsystems are independent then:

( = ∩ = ∩ = )=
(1 − ) (1 − ) (1 − )

Suppose the system structure function or system state variable is

� [ , ( , )], then the probability function of X is
as the following table shows:
Chap. 3 Reliability in Design + UGF Technique� 226

�
v�������
, , Pr( X  x )  p X (x)
� 0 d���� = 1 working

0,0,0 0 �� − )(1 − )(1

0,0,1 0 (1R1)(1R2)R3
0,1,00 �1 − ) (1 − )
0,1,11 �1 − )
1,0,00 �1 − )(1 − )
1,0,11 �1 − )
1,1,00 �1 − )
1,1,1 1
According to Eq. 3-12 for a binary system :
= ( )= ( ) = [(1 − )(1 − )(1 − ) × 0] + ⋯ + [ × 1]

� =( + − ) = [1 − (1 − )(1 − )].�

If in this binary system R1=0.95; R2=0.9; R3=0.85�, the system

reliability would be:

R3*(1-(1-R1)*(1-R2))= 0.8458

Having the reliability functions of independent system

elements Rj (t) (� ≤ ≤ ) one ean obtain the system reliability
function Rsys(t) by substituting Rj with Rj (t) (Livitin 2010,p32)�
Example 3-9
Consider the system form previous example whose RBD is

and assume that the reliability functions of the system element

are:

�
227 Reliabilty Engineering

� )= ( )= ( )=
Then this series-parallel system reliability is:
( ) = { ( )] = 3( )[ 1( )+ 2( )− 1( ) 2( ))]��
( )
+ − , �
( )= ( )[1 − (1- ( ))(1 − ( ))]= [1 − (1- )(1 − )]

End of Example

At the end, it is worth mentioning that having the u-functions

of the elements of an n-element binary system of the form

� ��) = 1 − R Z + R Z 0≤j≤n (3-13)

and the system structure function � ∅( , ,…, )

The system reliability measure can now be obtained

as(Livitin,2010, page34):

( )
E(X)� �1) = =�|z=1������������������������(3-14)

where U (z )   [U (z ),...,U n (z )] .
 1

The application of UGF technique to n-element binary system is

discussed in references such as Livitin(2010) pages 32-41. Moreover
Wei-Chang(2009) is a reference on UGF.

����������

1.Consider the system given below, composed of 4 like elements

having discrete life time of 20٬10 and 30 days with probabilities 0.2 ٬
Chap. 3 Reliability in Design + UGF Technique� 228

0.3and 0.5 . Calculate the UGF or U(z) of this system and the 10-daay
reliability of the system.

2.Repeat the previous example for the following RBD:

Keep in mind
that you are
never absent
from

God����������
so keep looking
how you are
acting

�
229 Reliabilty Engineering

Chapter 4
Structural
reliability
Analysis
�

�
Chap. 4 Structural Reliability Analysis 230

���������������������������������

��������������������

This chapter focuses on the reliability of the networks and

structures whose strength (capacity) and/or loads are
probabilistic. Reliability expressions for various statistical
distributions of strength and load namely normal, exponential,
lognormal, gamma and Weibull are presented.

4.1 Introduction
Designers of systems such as structures take many factors
into considerations including the reliability. Strength(capacity)
and load(stress) are 2 variables that affect the reliability of
structures(dams ,bridges; communication networks and
antenna). To be reliable, structures require to withstand ultimate
loads without failure.
There are 2 approaches for this purpose: deterministic and
probabilistic.

�
231 Reliabilty Engineering

The deterministic approach seeks out a worst case and

specifies a factor of safety for the extreme case to use in the
design. The probabilistic approaches utilize the statistical
distribution of input variables (here mainly load and strength) to
calculate reliability.
It should be added that in both approaches the amount of data
influences the results.
��Load-strength Interference Analysis
While the deterministic approach adopts the safety factor as
stability index, the probabilistic methods adopt as the probability
of failure (Queiroz, 2016)

Structural failure occurs when load(stress) exceeds capacity

(strength).Figure 1.4 shows such a case.

Fig 4-1The interference of the time-dependent load and strength

(Rausand & Hsyland, 2004 Fig 1.2)
Chap. 4 Structural Reliability Analysis 232

4-2-1Deterministic approach: Application of Safety Factor

Safety factor is defined as1

SF= (4-1)

where �is the strength and s is the load.

SF<1 results in failure. An acceptable SF is traditionally 1.5 2.
To cover unknowns and ensure safety, the deterministic
approach introduces conservatism by specifying a largish factor
of safety(SF). Calculating such a conservative SF requires a
high experience. On the other hand, this approach practically
forgets the randomness nature of design variables and
parameters (load , strength��. Of course the specialists of this
approach may notice the randomness of them but in
computations, the specialists act as if they are not probabilistic.

4-2-2Probabilistic Design Approach

The probabilistic approach incorporate the variability of

input parameters and variables and utilizes their statistical
characterization and attempts to provide a desired reliability in
the design. Probabilistic approach uses different methods. In

�����������������������������������������������������������
1
When the strength and load are independent random
variables, the average SF,E(SF), is approximately:
E ( )   �S 
E  SF   �  �  �
2
E ( s )  E (s ) 
MIL-HDBK-17-3E, Working Draft page 6-7
https://www.gla.ac.uk/external/asranet/Resources/milhdbk.pdf
�

�
233 Reliabilty Engineering

its simplest form, the measure of reliability is made by

comparing a component's stress to its strength(MIL-HDBK-17-3E).
The system does not fail as far as load(s) is less than its
strength(δ) and fails when � ≥ δ.

��������������������������������������������������

When the strength ( ) and/or the load(s) are random variables.

The reliability( R) of the system is given by

R  P r( S   ) (4-2-1)
R  P r( S    0 ) (4-2-2)
 (4-2-3)
R  Pr(  1)
S
R  P r( S F  1) (4-2-4)

Let � − �then
� ( > 0) (4-2-5)���

If the distribution of Y is not known�, the following

relationship might be helpful:

R  Pr   S    f ,S  , S  d  dS , (4  3)
 S
where f  ,S  , S  is the joint probability density function of

strength and load.

If S and� are independent, then:
Chap. 4 Structural Reliability Analysis 234


  
 
R   f  ( )   f s (s )ds  d    f s (s )   f  ( )d   ds 
 0 s 0  s 0  s 

R   f s (s ) 1  F (s )ds (4  4)
s 0

where

fs The pdf load

f The pdf strength

F The joint pdf of strength

Moreover, if s and are independent, the pdf of

= − might be calculated from(K&L page 125):


  f  ( y  s )f s (s )ds y  0
0
f Y ( y )   f  ( y  s )f s (s )ds    (4-5)
 f ( y  s )f (s )ds y  0
 
s
s
 y

and the system reliability(R):

  
(4-6)
R  Pr(Y  0)  
y 0
fY (y )    f  ( y  s )f
y 0 s  0
s (s )dsdy �

Example 4-1
The stress and the strength distributions for a component are
uniform over the interval :
Strength: [15 25]
Stress : [20 25

�
235 Reliabilty Engineering

How many percent of this kind of component break in a single

application of the load?

Solution

1  s-15 
R=  f s (s) 1-Fδ (s)  ds= 
� ��
1- ds 
s=0 s=20 25-20  25-15 
1 25
50 s=20
R= (25-s)ds=0.25

100(1-R)=75% break.

End of Example �

4-3-1 Definition of safety margin(SM)

Safety margin is an index related to the subject of reliability

defined as follows:
= (4-7) ��������� �

where
� are the means of the strength and load,
� are the variances of the strength and load.
In a structure, if � , the more s far from the less failure
probability and the more reliability. Then the more the
denominator the more the reliability; the less variation of the
load and strength (or the less the denominator), the more we are
confident. Therefore the greater SM>0, the more reliable the
 Chap. 4 Structural Reliability Analysis 236

μ
structure. It is worth noticing that actually SM equals where

Y=δ-S. or it equals the reciprocal of the coefficient of variation of

Y.

An application of SM is in the calculation of structures'

reliability when the strength and the load are independent and
normally1 distributed (See Figs. 4-2-1 & 2)

Fig 4-2-1 Normally distributed load and strength: Non- interference

�����������������������������������������������������������

�In a exceptional case where the distribution is Weibul(A,B,C)

�

with shape parameter C=3.44, the distribution could be

approximated with a normal distribution with parameters
(Carter,1986 as refrence by O'Connor, 2003 )
� + Γ�� � ) ≅ A + 0.9��, � Γ�� � Γ( )) ≅ 0.3 �
������ �

�
237 Reliabilty Engineering

�
Fig 4-2-1Normally distributed load and strength: interfere���

Suppose in a structure
the load is normally distributes with parameters and .
the strength is normal with parameters and
the strength and the load are independent.
Then:

� − ~ (   S , 2 s2 )

� ( > 0)

  S   S
R  Pr( Z   )  Pr( Z  )
 
2
S
2
  S2   2

Since � then

R   (SM ) (4-8)
Z

where ϕ is the CDF of standard normal distribution.

Chap. 4 Structural Reliability Analysis 238

Therefore when the load and the strength are independent and
normally distributed, the reliability(R) is calculated from Eq. 4-
8. The more SM the greater R. More specifically on the average
the more the difference ((strength-load) or the less the variances
of load and strength the more R. Moreover

.Negative safety margin indicates that on the average load is

greater than strength of the structure which is dangerous;

.|SM|=∞ indicated that load and strength are deterministic.

It is worth mentioning that if a random sample of normally
distributed strength and a sample of normally distributed load is
available, the estimates of the mean and variance of ��
could be used when using Eqs. 4-7 and 4-8.
Example 4-1a
The strength and the load related to a structure are normally
distributed. Calculate the reliability for 0,25<SM< 6.
Solution
The following table shows the reliability calculated from Eq.
4-8 for several SM . Figure 4-3 shows the related plot.
SM
0

0.5

1.5
0.25

0.75

1.25

1.75

2.75

3.5
3

4
5
6
1

.999968�

.9999999
.999999
.97725

.99865
.99977
.5987
.6915
.7734
.8413
.8944
.9332
.9599

.9937

R=
.50

ϕ ( )

�
239 Reliabilty Engineering

�
��������� ���������������������������������������������������������������
0
log(1-R)(log of failure probability per application of load)

-2

-4

-6

-8

-10

-12

-14

-16
0 2 4 6 8 10
safety margin �
Fig 4-4 Plot of Logarithm of unreliability versus safety margin(SM)

Fig 4-4 shows the logarithm of unreliability ( failure

probability) per application of load versus SM. The figure has
been plotted using the following MATLAB commands:
Chap. 4 Structural Reliability Analysis 240

=0:.25:10;R=normcdf(SM); F=log10(1-R);plot(SM , F)
Table 4.1 gives the reader an idea about the variability in R
related to different magnitudes of variability in normal y
distributed strength and stress random variables(K&L page 79)

Table 4-1 Effects of different cases of normally distributed

load and stress on reliability
Case No.

�Strength Stress SF ��R=

SM
  S S  / S
�� )

1 50000 2000 20000 2500 2.5 9.37 1.0

2 50000 8000 20000 3000 2.5 3.51 0.9997
3 50000 10000 20000 3000 2.5 2.87 0.9979
4 50000 8000 20000 7500 2.5 2.73 0.9969
5 50000 12000 20000 6,000 2.5 2.236 0.987
6 25000 2000 10000 2500 2.5 4.69 0.9(5)86
�7 25000 1000 10000 1500 2.5 8.32 0.9(16)6
8 50000 20000 10000 5000 5.0 1.8\94 0.9738
9 50000 2000 40000 2500 1.25 3.123 0.99909
10 50000 5000 10000 5000 5.0 5.65 0.9(8)2

End of Example �
Figure 5.4 is a sample plot of log(1-R) versus SM per
application of load. The figure shows if the SM of a design
lies in the third region (i.e. if SM is greater than a threshold), the
logarithm of failure probability is very small and the failure
probability becomes infinitesimal and the design is said to be
intrinsically reliable .

�
 241 Reliabilty Engineering

Table 4-1 Expressions for different independent Distributions of Load (S)

& Strength�� )
Case No.

Eq No.
Dist .of Dist .of
Eq. for reliability Strength(�δ) Load(s)

1 R=
λ Exp(λδ ) Exp(λ )

(4-9)
λ λδ
Ref:K&L page 157
2 σ λ Normal Exp( λ)
R = 1 − ��� (−μ × λ + ) (μ� σ)

(4-10)
������������������������������������������ Ref:K&L page 157
3 �� =
μδ μ
�
(4-11) N(σδ‫ﻭ‬μδ ) �(σ ‫ﻭ‬μ )
σδ σ K&L page 126
� = ϕ (SM)=normcdf(SM)
Calculable in MATLAB
4 � (z)=normcdf(z)���������������������������� �����σδ ‫ﻭ‬μδ ) �����σ ‫ﻭ‬μ )

�����������z= Ref:K&L page 130

(4-12)

�
‫ � �ﻭ ��� �ﻭ‬are the parameters not
the mean and standard deviation
5
s

(a  b )
x 
s  Gamma Gamma
R
(a )(b )  x a � (�  x )b �dx , (a,�, δ � � (b,�, )
(4-13)


x �
Ref:K&L page 141
 betacdf ( s s , a, b ) : MATLAB

Calculable in MATLAB
6 � = ��� > ) =
−
+ ×
Weibull
(A,B,C)
Normal
(μ� σ)
√2
Ref:K&L page
(4-14)

∞
× ∫ �����− − 0.5 + ] ,
=
142& Appendix III
Calculable in Matlab , Maple���
7 ��1−
∞
Bδ
e ���� �−( y
B
δ �
�δ − �
�δ
) ]��
����� ,
, )
����� ,
, )

= Ref:K&L page 146

(4-15)

In K&L is seen in
C
= d the expression which
seems to be a typo
Chap. 4 Structural Reliability Analysis 242

Fig. 4.5 Characteristic regions of a typical log(1-R) Vs SM curve

(O'Connor, 2003 page119)
Example 4-2
A normally distributed load with parameters ( , ) was
applied once to a structure with constant strength ,
a) Plot the failure probability(F=1-R) versus SM and also plot
log(1-R) versus SM.
b) Calculate a fixed value(δ) for the strength in terms of μ
,σ � such that the structure lies in the intrinsically reliable
region.

�
243 Reliabilty Engineering

Solution
F = 1 − R = 1 − ��(Z < ),
μ −μ δ−μ −
�� = = =
σ +σ σ +0

�The following figure shows failure probability( F )versus SM

plotted by the following MATLAB commands:

SM=.001:.01:8.5 ;F=(1-normcdf(SM)); plot(SM,F)

The following figure shows logarithm of F versus SM plotted

using: SM=.001:.01:8 ;LF=log(1-normcdf(SM)); plot(SM,LF).

This figure shows that for SM>8 the structure is intrinsically

reliable:
−
= >8 ⟹ > +8

Note that for �� � 8, the reliability and the failure probability

is:
� (1 − ) = 1 × ��−�� �

R=���Z < = 8)=� ���������) = 0.���������������

Chap. 4 Structural Reliability Analysis 244

4-3-2 Reliability Computation for Probabilistic

independent load and strength
The reference K&L has done a lot of computations for
deriving the reliability of systems having various independent
distributions of load and strength. Table 4-1 shows the results.
It is worth mentioning that
1. exponential distribution could be considered a special form
of gamma and Weibull distribution. Therefore if , for example,
our structure has a Weibull-distributed load independent from
the exponentially distributed strength, then we could use case
no. 7 of Table 4-1 to calculate the reliability.
2. If we have a structure with Weibull-distributed load and
strength having the same shape parameter C and zero location
parameter then the reliability of the structure ( R) is given by:

( )
�( ) ( )
����� �����������(4-15-1)

Proof: Form Eq. 4.15

−
�� )

= ��� > ) = 1 − e ���� �−� ) ]

=1− e ���� �− � ) ]

[ )]
=1− e ⟹

�
245 Reliabilty Engineering

c
�1− e , = d ⟹

c [ ]( )
=1− e d

1 c
=1− 1+ e d ⟹
1+

∞ C
− 1+ − 1+ (0)C
e −e ( )C
=1− =1−
( )C + ( )C
1+

( )
⟹R=( ) ( )
End of proof

Notice that

-if C=1 i.e. the load and strength are both exponentially
distributed with parameters λ � �and λδ = respectively,

then Eq.4.9 is obtained

-if the load and strength are both Rayleigh distributed with
scale parameters �and � respectively, then the reliability is
calculated from Eq. 4-15-1 for C=1:

�( )
�( ) ( )
. ( 4-15-2 )

Example 4-3-1
A lognormal distributed load with mean of ����� Kpa and
standard deviation of������ Kpa is applied to a structure which
Chap. 4 Structural Reliability Analysis 246

has an lognormal distributed strength with a mean of ��

Kpa and standard deviation of �� K pa .
a) Find the parameters of th4e distributions:
b) Estimate the system's reliability
c) Using Eqs. 1-22-1 & 1-22-2 verify the parameters
obtained in part a for load distribution.
Solution

a)Using MALAB software and Eqs. 1-22-5 & 1-22-6 :

( ) �����
� �� + 1 = ���( � 1) =
( ) �����

(������2/������2 + 1) = 0.�����

�� = �� � ) − = log(60000)-0.5*.1054= 10.9494

( ) ����� �����
� �� + 1 = ���( � 1) = � 1 = 0.��
( ) ������ ������

� �� � ) − =log(100000)-0.5*.01= 11.5079

b) Using MALAB software and Eq. 4-12:

� (z)=normcdf(z) z�
�

. .
�� = = 1.�����
√ . .

�
247 Reliabilty Engineering

c)According to Eqs. 1-22-2 &3

0.����
� �= = ������� � = ������
�

( )= −1 .

( )−1 = (2 ∗ ������� � 0.����� ∗

( (0.����� − 1) = ����� .. The difference is due to

approximations.

Example 4-3-2
The following random sample is from the load random
variable applied to a structure. The values are in KPa.
284.9188 104.1661 20.6819 461.9137 197.4067
159.5707 161.5850 50.4525 130.1263 161.5384
418.1608 29.1977 80.9464 7.0621 76.3582
87.0721 16.6974 64.4067 159.6288 39.7292

The strength is also a random variable , of which is as follows:

115.3541 26.0195 153.8555 264.6725 51.3116

168.7690 214.7956 18.8720 22.9266 139.8943
66.7714 151.4207 164.2746 153.6125 219.3838
149.3005 206.9543 30.3550 91.1805 81.8012
Assuming the 2 random variables are independent, calculate the
reliability.
Chap. 4 Structural Reliability Analysis 248

Solution
Using softwares such as ARENA1 or goodness-of-fit tests or
Q-Q plot help us to consider the load is exponentially distributed
with mean 100KPa and know that a Rayleigh distribution with
mean 111 or equivalently a Gamma(a=2, = 0.���) fits the
strength. Since the exponential distribution could be considered
Gamma(b=1, = 0.��), therefore according to Eq. 4-13 the
reliability of the structure is:
λs
x= λ
s �λ δ
�����
R=
Γ���Γ��� 
x=0
x a-1 (1-x) b-1dx=

λ
betacdf( λs �λs δ ,a,b) MATLAB

x= 0.01 0.01
0.01
x= 0.01+0.018 0.028
�����
 x 2-1 (1-x)1-1dx= 2  x(1-x) 0dx=x 2 ] 0.028 =0.1276
������ ��� 0
x=0
x=0

or by MATLAB

R  betacdf ( 0.01+0.018
0.01
,2,1) =0.1276 �

Example 4-3-3
The strength( ) of a component and the stress (S) applied to
it are exponentially distributed with means 150 and 100 psi
respectively. Find the reliability( R) of the component using
Eqs. 4.9 &4-15.

�����������������������������������������������������������
�
����������������������������������������������������������������������

�
249 Reliabilty Engineering

Solution
From Eq. 4-9:

�= = = 0.��� ������������

From Eq. 4-15:

Since exponential distribution is a special case of Weibull
distribution, then
� �� �1 � �0

=1− e ���� �−� ) ]

= = , = ⟹d = d ⟹

1
=1− e

Let = , = . Using the following MATLAB

instructions results in R= �

>>bd=150;bs=100;syms x;W=[(1/bd)*exp(-(x/bd)-x/bs)]

>>R=1- int(W,x,0,inf)

4-3-3 Definition of Loading Roughness

Load roughness (LR) is a factor in load-strength interference

which combines the load information and strength information
Chap. 4 Structural Reliability Analysis 250

(Wu & Xi, 2010)1. Quantifies LR is defined as follows

(O'Connor &Keleyner,2012 page 121) :

S
����������� LR 
2 2
�������������������
  S

where �

S is the standard deviation of load(stress) random variable

 2   2 is the standard deviation of the difference   S .

 S

An application of LR is in the calculation of the failure and

reliability of components and systems subject to multiple
application of loads. The most reliable situations are those with
low LR and high SM; and the least reliable situations are those
with high LR and low SM(Reuben,1994 page 209-210).
SM and LR allow, in theory, to analyze the way in which
load and strength distributions interfere and so generate a
probability of failure(O'Connor &Keleyner,2012 page 121).

�����������������������������������������������������������
1
Wu,Y. Xi,L. 2010 Load-roughness impact on reliability considering
dependent failure ,Proceeding 16th ISSAT conference on Reliability and
Quality in Design

�
251 Reliabilty Engineering

Moreover a value of LR allows to know about the variations of

S and such that:
For a largish SM, if the variance of load is small and that of
strength is large then LR will be small(e.g. 0.3); and for a fixed
SM, as the load spread becomes wider than that of the strength,
LR increases. Therefore for a largish SM if the variance of load
is large and that of strength is small then LR will be large (e.g.
0.9).
Figure 4-6 shows four cases in which the distributions of
strength and load are normal and have no considerable overlap.
The LR for each case is indicated on the figure; SM=4.5 and
single application of load results in a reliability of Φ ���5) =
0.������. This figure also shows that:
a)For SM=4.5 if we know that LR is small, it is concluded
the variation of strength and load is large and small respectively.
a)For SM=4.5 if we know that LR is large, it is concluded the
variation of strength and load is small and large respectively.
Chap. 4 Structural Reliability Analysis 252

Fig. 4.6 Four cases of normally distributed load and strength

with SM=4.5 and 4 different LRs(King.1990 page 348)

To study more about the effects of loading roughness and

safety margin, refer to O'Connor &Keleyner(2012) Fig5.2.
Example 4.4(O'Connor,2003Example 4-1)
The strength of a component is normally distributed
� δ~N(�����, ����). The load it has to withstand is also
normally distributed ��N(������ ����). Assume the strength is
independent of the laod.

�
253 Reliabilty Engineering

a)What is the component reliability per application of load?

b) Find the extreme load L such that Pr   L  equals the

answer in part a.
Solution
a)

���� − ����
� � ≤ = ⎛� ≤ ⎞ = 0/������
��� � ���
⎝ ⎠
b)

R=Pr  δ��  ��������  ����� L'-5000 ��������� 

400 �
L'-5000 =-2/65  L=3940 N
400
or by MATLAB

L=norminv(1-0.99598,5000,400)=�3939.85N

4.3-4 Effect of Safety Margin and Loading Roughness

on Reliability (Multiple Load Applications)

The reliability for multiple load application is calculated

from(O'Connor, Kleyner, 2014 page 124):
n

 
R   f  ( )   f s (s )ds  d  (4-17)
 0 s 0 
where
n is loading times , the number of load applications which are independent

f s (s ) is the load pdf and

f  ( ) is the strength pdf independent from the load.

Chap. 4 Structural Reliability Analysis 254

Fig 4-8 Failure probability versus SM for large n as well as n=1

(O'Connor, Kleyner, 2014 page 125)
Figure 4.8 shows the effects of different values of LR and
SM on failure probability per load application for large values of
n as well for single load application(n=1).
For details refer to O'Connor& Kleyner( 2014) and Wenxue
Qian et al ( 2014). Wenxue studies the reliability of a com-

�
255 Reliabilty Engineering

ponent in relation to some parameters including LR, SM and

loading times(n). O'Connor gives 2 examples which illustrate
the application of load-strength analysis to design of an
electronic device and a mechanical one.

���������������������������������������������������
���������������������������

In this section those systems are considered in which the

capacity has known deterministic value and the load is a
random variable or vice versa the load is known and the
capacity is a random variable.

4-4-1 Calculation of structures' reliability when

strength is deterministic

Consider a system with a known capacity δ and a distribution

of possible loads as plotted in Fig. 4.11.

Fig . 4-11 Interpretation of reliability -Load :variable, strength: fixed

Chap. 4 Structural Reliability Analysis 256

For a fixed δ if the probability density function of load is

� ) and the CDF is � ( ) then the reliability of the system ( R)
is the shaded area in the figure calculated as follows:
δ
� �� �δ �� δ � � � �� (4 − ���

Example 4-6 (Lewis, 1994 p181)

Suppose the bending moment on a match stick during
striking has an exponential distribution. The match stick have
the given strength δ and break 20% of the time. The
manufacturer increases the strength by 50%. What fraction of
the strengthened matches are expected to break as they are
struck?
Solution
Bending moment � ( ),

=F δ � � ) = =1− �

0.8 = 1 − ⟹ = 0.2�

Now the strength is multiplied by 1.5 i.e. δ =1.5 δ

then
. . . .
New � ∫ =1− =1− =1−
(0.2) .
= 0.����
The fraction of the strengthened matches expected to
break is 1-0.911= 8.9%

�
257 Reliabilty Engineering

4-4-2 Calculation of structures' reliability when load

is deterministic
Consider a system with a known load S and a distribution of
the strength as plotted in Fig. 4.12.

Fig . 4-12 Interpretation of reliability - strength: variable, load: fixed

For a fixed � if the probability density function of strength

(capacity) is ( ) and the CDF is �δ then the reliability of the
system (R) is the shaded area in the figure calculated as follows:

� Pr   S  � ( ) = 1 − Fδ ��) (4 − ���

Example 4-7
A component is subject to the fixed load of 4000N. The
strength is log-normally distributed with mean of 5000N and
Chap. 4 Structural Reliability Analysis 258

standard deviation of 400N. Calculate the reliability for single

load application.

Solution

To find the answer we have to calculate the parameters μ�σ

from the mean and standard deviation using Eq.1-22:

  ln[var(X ) / E � (X )  1] 
using MATLAB
����log(4002 /50002 +1))����
2
μ�������� σ ����5000�� ���� �������
2 2

R=Pr  δ�������  �

Pr(ln������������������� 8.294-8.5122
0.1 =-2.182��������

or using MATLAB:

�� 1 − ��� < ����� = 1 − ������������8.������.1) = 0.�����

End of Example

�
259 Reliabilty Engineering

���������������������������������������������
������������������������������ ��������������
�������������������������������������

Two different methods were pointed out in this chapter to

cope with load-strength interference: the reliability-based and
safety factor (SF) methods. Are the safety factor and the
reliability concepts contradictory or they are interrelated? To

answer this question note = therefore if the load(S) and

the strength( ) are random variables then �� would be a random

variable. The following 2 inequalities has been developed to
show the interrelation of random variable SF(having a mean of
n and coefficient of variation V ) and the system reliability(R).
If and R are independent and normally distributed, then (Dao-
Thein&Massoud,1974):
�≥ (4-20)

×
�≥1− ( )
(4-21)

where

R The system reliability

� The mean of SF
� = V = ������
n
 Chap. 4 Structural Reliability Analysis 260

If S and δ are independent with mean and standard deviation

of (  , ) & (  ,   ), Based on Eqs. 5-12-1 ,5-12-3the mean and
S S

variance of� = could be approximated from: �

   S� 
������������������ n  �  �  ��������������(4-22-1 )
S  S 
�����

  
�
 � � �
��  �
       S
 
  S
 
 
 (4-22-2 )
 S     S   S  
SF
   

Inequalities 4-20&4-21give a lower bound for and R

repectively. The relationship between the reliability(R), Vn and
the lower bound of is shown in Fig. 4-13.

Vn(%)�
Fig 4.13 Plot of � � versus Vn for six levels of R

(Dao-Thein & Massoud,1974)

�
261 Reliabilty Engineering

According to this figure the reliabilities of 90% , 95%, 97%

with � � ��� corresponds to the average safety factor of at
least 2,3 and 10 respectively.
It is worth mentioning that if the load and strength have
normal distributions S ~ N (S ,S ) and  ~ N ( , ) then
(Handbook Sharpe, 2008 page 271):

  n S
�Φ � � (4-23)
 
2

2
s

Φ is the standard normal distribution that could read from

Table C or calculated usin a softwares such as MATLAB.
Needless to say n  1 results in Eq. 4.8. references such as
Lewis(1994)page 182 has more on SF and R.
Example 4-8-1
The safety factor (SF)of a structure is a random variable with
a standard deviation of 0.8357. The mean of SF is at least
4.4626. Find the reliability of the structure.

Solution
From Fig. 4-21 the minimum of the reliability is 0.95.
using Inequality 4-21:

σ 0.����
� = � 0.��
μ ������

. ∗ .
≥1− ( . )
= 0.��.
. ∗ .
End of Example
Chap. 4 Structural Reliability Analysis 262

Example 4-8-1
The reliability of a structure should be at least 0.95 . the
coefficient of variation is 18%. How much is the safety factor
on average?

Solution

Eq. 4-20⟹ ≥ . = 4.����

. ∗
.

End of Example

��������������������������������������������
����������������������������������������

In the real world, it might be difficult to know the true

distributions over the complete range of the stress and the
strength random variables(K&L page 88); Therefore the reliability
cannot be calculated using Eq. 4-3 i.e.

R  Pr   S    f  ,S  , S  d dS
 S
or equivalently, in the case of independence of stress and
strength, the failure probability cannot be computed from:

 
R  Pr(S   )   F (s )f s (s )ds   [1 Fs ( )]f  ( )d  (4-24)�
 

Where

� � (δ, S)� Joint pdf of the strength and stress(load)

�
263 Reliabilty Engineering

�� f S pdf of the stress(load)

FS CDF of the stress(load)

F CDF of the strength

f pdf of the strength

Fig. 4-14 Load- strength interference (K&L page123)

A procedure has been developed for these cases which
calculates a minimum and a maximum for the reliability. Note
that the reliability(R) depends on the interference of the two
random variables(stress and strength); hence only the local
information in the interference range is needed to compute R
(K&L page 88). In this procedure smax and min are determined

as the upper limit for S and the lower limit for respectively
forming the interference interval [  ��� � � ��� ].  min and S max are

either known from the pdfs of  and S respectively or their

Chap. 4 Structural Reliability Analysis 264

values are estimated according to the accuracy desired(K&L

page 88); then a lower and upper bound is calculated for the
reliability according to the following algorithm which uses
nonlinear programming(NLP).

4-6-1 The algorithm for Reliability Lower & Upper

Bounds using NLP

Step 1

Determine the load-strength interference interval [min smax ]

in such a manner that the probability beyond the interval is
ignorable.
Now the system failure probability( R ��could be calculated

from:
s max s max
R=Pr(s>δ�  δ min
�δ ���� � �����  
 min
��  FS �u ��f  �u �du (4-25)

Step 2

Divide the interval [min s max ] into n equal subintervals:

Step 3

Let the probabilities ( p1,...pn ) and (q1,...qn ) be defined as:

�
265 Reliabilty Engineering

p i  P r(a i 1  S  a i ) i  1, . .. , n (4-25)
q i  P r (a i 1    a i ) i  1, . . . , n (4-26)
Now it can be shown that Eq. 4-25 could be approximated
by(K&L page89):
n
  i  n   n 
R    pi   q k      qi   pk  (4-27)
i 1   k 1   i 1   k i 

Step 4

pi and qi are the probability of occurring the load and strength

in an interval. To add more uncertainty, lower and upper limits
could be considered for them:
Lp ,i  pi U p,i i 1,..., n (4-28-1)

Lq ,i  qi Uq ,i i 1,..., n (4-28-2)

Therefore in this step determine a lower bound and an upper

bound for each of ( p1,...pn ) & (q1,...qn ) . Some designers state

these bounds as pi pi and qi qi where  is a fraction

between zero and 1.

Step 5

Since(K&L page 89):

n
Pr(  s max )   q i , � (4-29-1)
i 1
Chap. 4 Structural Reliability Analysis 266

n
Pr(s   min )   p i ( 4-29-2)
i 1

and p i � & q i � are 2 probability values, to add more

uncertainty, consider upper and lower bounds for them:

n n
ap   pi  b p , aq   q i bq �
i 1 i 1

Therefore in this step determine bounds for p i and p i

One advice is to locate the bounds 2 from p i or p i .

Step 6

Now we would like to use the above bounds in order to

determine the upper and lower bounds for �and for R=1- � To
accomplish this the following 2 nonlinear models have to be
developed and solved:

�
267 Reliabilty Engineering

n
  i  n   n 
������� R    p i   q k      q i   p k  
i 1   k 1   i 1   k  i 
s .t .
L p ,i  p i  U p ,i i  1,..., n
L q ,i  q i  U q ,i i  1,..., n
n
ap   p i  b p ,
i 1
n
aq   q i bq
i 1

pi  0 qi  0

Use the optimal values of the 2 objective functions as the lower

and upper bounds for the failure probability of the system( R ).

End of algorithm.

It should be pointed out that some researchers have used linear

programming to calculate bounds for reliability e.g. see Song��
Kiureghian(2003)

Example 4-9(K&L page 89)

The load on a structure is normally distributed with mean

30MPa and standard deviation 3MPa(coefficient of variation
equal to 0.1). The strength has a Weibull distribution with CDF

( )=1− and parameters:

Minimum strength A  3 0 M P a , B  60M Pa , C=2or3or4

Chap. 4 Structural Reliability Analysis 268

Find the lower and upper bound for the reliability of this
structure.

Solution

Step 1 [min Smax ] =?

 min is set equal to A=30. If we set S max  50 based on

S  6 S  48 ,the probability that the load reaches an amount
greater than it is infinitesimal( 129��� -9 ).
Step 2
The interval [min  30,Smax  50] is divided into ten subinterval

With length 50-30  2 and :

10

a 0 =30 a1 =32 ..... a 9 =48 a10 =50

Step 3 Determining pi's & qi's

pi's based on the load distribution i.e. N(μS ����σS ��� :

32-30
p�  Pr(30  s  32)  Pr(0  Z  )  0.2475 �
3

�
269 Reliabilty Engineering

p 2  Pr(32  s  34)  ����������������������������������  0.1613

���������������������������������������������
�����������������������������������������������
�����������������������������������������������
����������������������������������������������� ����
��������������������������������������������� ����
���������������������������������������������� ����
p 9  Pr(46  s  48)
 ����������������������������������  �������  ����
p 1 0  P r( 4 8  s  5 0 ) 
� � �� � � ��� � �� � �� �� �� � �� � � ��� � �� � �� �  �� �� � � �  � � � � � �

Calculation of qi's based on the strength distribution i.e. Weibull

with parameters A=30,B=60,C=2

q1  Pr(30    32)  Fweibul (32)  Fweibul (30) 

 32  30 2   30  30 2 
  1  exp( )    1  exp( )   0.0011
 60   60 
or
wblcdf (32  30, 60, 2)  wblcdf (30  30, 60, 2)  0.0011
q 2  Pr(32    34) 
exp(-((32-30)/60)^2)-exp(-((34-30)/60)^2)  0.0033
Chap. 4 Structural Reliability Analysis 270

q 3  Pr(34    36)  0.0055

q 4  Pr(36    38)  0.0077
q 5  Pr(38    40)  0.0098
q 6  Pr(40    42)  0.0118
q 7  Pr(42    44)  0.0138
q 8  Pr(44    46)  0.0157
q 9  Pr(46    48)  0.0174
q10  Pr(48    50)  Fweibul (50)  Fweibul (48)
 50  30 2   48  30 2 
  1  exp( )    1  exp( )   0.0191
 60   60 

Step 4

Suppose an uncertainty of    2 % (K&L page90 )was

present in ( p1,...pn ) and (q1,...qn ) calculated above; therefore

each of ( p1,...pn ) lies in the interval [Lpi Upi] where

Upi=pi+(0.02)*pi ; Lpi=pi-(0.02)*pi

and qi lies in the interval [Lqi Uqi] where

Uqi=qi+(0.02)*qi ; Lqi=qi-(0.02)*qi

As sample calculations the bounds for ( p1, p2 , p3, p10 ) are:

Up1=0.2475+(0.02)* 0.2475 ; Lp1=0.2475-(0.02)* 0.2475 ;

Up2=0.1613+(0.02)* 0.1613 ; Lp2=0.1613-(0.02)* 0.1613 ;
Up3=0.0685+(0.02)* 0.0685; Lp3=0.0685-(0.02)* 0.0685;

�
271 Reliabilty Engineering

Up10= 9.7350 10 +(0.02)* 9.7350 10

-10 -10
;

Lp10= 9.7350 10 - (0.02)* 9.7350 10

-10 -10
;

The bounds for (q1,...q10) with  (   2 %) are calculated from

Uqi=qi+(0.02)*qi ; Lqi=qi-(0.02)*qi i=1,2,..,10

As sample calculation:

q1=0.0011;Uq1=0.0011+(0.02)*0.0011;Lq1=0.0011-(0.02)* 0.0011

Step 5 Calculation of  p ‫ �ﻭ‬ q

i i and their limits

n
30-30
p
i �
i  Pr(s   min )  Pr( Z 
3
)  0.5
n
50-30 2

i �
q i  Pr(  s max )  �  exp[ (
60
) ]=0.105.

Suppose an uncertainty of 1% was present in q i and p i

calculated above(K&L page90); therefore they lie in the

following intervals (K&L page 90):

n
0.02 0.02
0.5-( )  0.5   p i  0.5+( )  0.5
2 i � 2
n
0.02 0.02
0.105-( )  0.105   q i  0.105+( )  0.105
2 i � 2
Chap. 4 Structural Reliability Analysis 272

The following MATLAB code performs steps 3, 4 and 5:

format long
%input
r=[30 50]; %input 1
n=10; %input 2
alpha=0.02; %input 3
%algorithm
a=alpha;% specify an appropriate value for alpha
x=r(1):(r(2)-r(1))/n:r(2);
for i=1:n
%Load
p(i)=normcdf(x(i+1),30,3)-normcdf(x(i),30,3);
%Strength
q(i)=wblcdf(x(i+1)-30,60,2)-wblcdf(x(i)-30,60,2);
end
p=p';
q=q';
Up=p+p*a
Lp=p-p*a
Uq=q+q*a
Lq=q-q*a
Sigmap=normcdf(min(x),30,3)
Sigmaq=wblcdf(max(x)-30,60,2)
Usimgap=Sigmap+a*0.5*Sigmap
Lsigmap=Sigmap-a*0.5*Sigmap
Usigmaq=Sigmaq+a*0.5*Sigmaq
Lsigmaq=Sigmaq-a*0.5*Sigmaq

�
273 Reliabilty Engineering

Step 6 Lower and upper bound for failure probability ( R ):

Step 6-1 The following model has to be solved once for
maximization and once again for immunization to find the upper limit
for failure probability( R ).
The objective function is:
10
  k i  
Max R    pi   q k  
i 1   k 1  
or
Max R  p1q1  p 2 (q1  q 2 )  p3 (q1  q 2  q 3 )  ...  p10 (q1  q 2  ...  q10 )

s .t .

The constraints of type L p ,i  pi  U p ,i i  1,...,10 :

�   � �  
�� �������� ���� ��������  �  �� �������� ���� ��������

�������  ����  ��������  �  �������  ����  ��������

�
�������  ����  ��������  �  �������  ����  �������
�
�������  ����  ��������  �  �������  ����  �������
�
�������  ����  ��������  �  ��������  ����  �������
�
��������� �  ����  �����������  �  ���������� �  ����  ����������
�� �� �� ��
�
��������� �  ����  �����������  �  ��������� �  ����  ����������
�� �� �� ��
�
��������� �  ����  �����������  ��  ���������� �  ����  ����������
�� �� �� ��
�
��������� �  ����  �����������  �  ��������� �  ����  ����������
�� �� �� ��
�
��  ����  ���������� ��  ����  ����������
��� ��� ��� ���
��������� �  �  ��������� �
��

The constraints of Type Lq,i  qi  Uq,i i  1,...,10 :

Chap. 4 Structural Reliability Analysis 274

�   � �  
�� �������� ���� ��������  �  �� �������� ���� ��������

�����������  ����  ��������  �  �����������  ����  ��������

�
�����������  ����  ��������  �  �����������  ����  ��������
�
�����������  ����  ��������  �  �����������  ����  ��������
�
�����������  ����  ���������  �  �����������  ����  ��������
�
�����������  ����  ��������  �  �����������  ����  ��������
�
�����������  ����  ��������  �  �����������  ����  ��������
�
�����������  ����  ��������  �  ����������  ����  ��������
�
�����������  ����  ��������  �  �����������  ����  ��������
�
������������  ����  ��������  �  ������������  ����  ��������
��

Constraints related to p i &� q i :�

n
ap   pi  b p  � .5-(0.02/2)*0.5  p1  ...  p10  0.5+(0.02/2)*0.5
i �
n
aq   q i b q  0.1055-(0.02/2)*0.105  q�  ...  q��  0.105+(0.02/2)*0.105
i �

0  pi  1 0  pi  1
Softwares such as Lingo or GAMS give the following results for the
maximization and minimization :
The objective function has the optimal values:

If minimized Rmin =0.00704= thelower limit

If maximized Rmax =0.0076= the upper limit

Therefore for = 2% and the Weibull shape parameter C= 2,

the failure probability lies in [ 0.00704 0.0076] and the
reliability lies in:[ 1-0. 0076 1-0.00704 ] =[0.9924 0.99295 ].
The following table shows the unreliability limits for other
cases(from K&L page 91):

�
275 Reliabilty Engineering

Uncertainty Shape Parameter

( )%
C=2 C=3 C=4
R m in R max R m in R max R m in R max

2 0.00704 0.0076 0.00123 0.00133 0.00024 0.00026

4 0.00677 0.00789 0.00118 0.00138 0.00023 0.00027
6 0.00650 0.00818 0.00113 0.00143 0.00022 0.00029
8 0.00624 0.00848 0.00108 0.00148 0.00021 0.00030
10 0.00598 0.00878 0.00104 0.00154 0.00020 0.00031

End of Example

��������������������������������������
������������������������������������������
�����������

As well as the definition given in Eq. 4-7 for safety

margin(SM), SM is usually expressed as the allowable working
stress ( ) divided by the applied stress f minus 1(Ireson et
al,1996 page 18-13).

�� � − 1. (p-1)

Any negative SM value indicate that the structure will fail

because the applied stress of the allowable material strength.
This is only for unidirectional stresses ; biaxial and tri-axial
stresses require further analysis(Ireson et al,1996 page 18-13).
It is reminded that the safety factor(SF) which is a strength
design factor is defined by the ratio of a critical design strength
parameter (tensile, yield, etc) to the anticipated operating stress
under normal operating conditions((Ireson et al,1996 page 18-
12). For example let
denote the material strength and
Chap. 4 Structural Reliability Analysis 276

denote the allowable working stress

Then the factor of safety becomes:

�� SF= .
fmw


Therefore f mw = and:
SF

�� = − 1� (����� or
×

�� = ���������������������������
� + )×

where
SM The safety margin or margin of safety
SF The safety factor
The strength of the material
The maximum allowable working stress
The stress applied to the structure.

Have a good opinion of God,

for whoever has a good opinion of God
He will treat him in the same way

�
277 Reliabilty Engineering

Exercises

In the following problems the stress and the load(stress) are independent.

1-(Problem 5 Page159 K&L) The strength and the stress S for the
design of a component are logomrally distributed with the
following infonnation on and S:

E( )=750.00 Mpa � �50.00 MPa

E(S)=500.00 Mpa �80.00 Mpa

2-(Problem 3Page159 K&L)A component is to be designed for a

specified reliability of 0.990. The stress and the strength
random variables are known to be lognormally distributed for
this component with the following information

E( )=1100.00 MPa, E(S) = 850.00 , �100.00 MPa

Determine the maximum allowable standard deviation of the

stress that can be be appliedto the component which will give
us the desired reliability.

3 -(Problem 9 Page160 K&L)The strength of a component has a

gamma distribution with parameters a=4,�, δ 1. The failure
inducing stress also is gamma distributed with b=2,� 1.
Compute the reliability of the component.
Chap. 4 Structural Reliability Analysis 278

4-(Problem 10 Page160 K&L) In Exercise 3, assume that δ �

and � 2.5 . Compute the reliability of the component for this
case.

5-(Problem 11 Page161 K&L) A leaf spring for a truck is to be

designed for a reliability 0.9995 based on the fatigue failure of
the leaf spring. The fatigue strength of the material out of
which this spring is made is Weibull distributed with the
following parameters:

A= 500.00 M Pa B= 500.00 M Pa C=3.0

The random loading of the spring induces stresses that are

assumed to be normally distributed with a coefficient of
variation of 0.08. Compute the permissible normal stress
parameters that would yield the specified reliability.

6-(Problem 13 Page160 K&L)The strength of a component is

lognormally distributed with a mean of 800.00MPa and standard
deviation of 150.00 MPa. The failure governing stresses have
normal distribution with a mean of 600.00 MPa and a standard
deviation of 110.00 MPa. Compute the reliability of the
component.

7-(Problem 15 Page161 K&L)The stress acting on a component

is uniformly distributed over an interval{ 10,40]. Th��������������

�
279 Reliabilty Engineering

the component follows normal distribution (35 , 5)Derive an

expression for the reliability of the component. Find R

8-(Problem 16 Page162 K&L)The stress acting on a component

is uniformly distributed over [ 10 , 30] The strength of the
component has a three-parameter Weibull distribution with
parameters A=20, B=30 and C=3. Derive an expression for the
reliability of the component and caculate its numerical value.

9-(Problem 17 Page162 K&L)The stress acting on a component

is uniformly distributed over an interval [Smin Smax]. The
strength of the component has gamma distribution with parar
ters n and . Derive an expression for the reliability of the
component. Let

Smin = 10 Smax=30 n=5 � �.2

Find R.

10-(Problem 8 Page160 K&L) The strength of a component is

lognormally distributed with a mean value of 400 MPa and a
standard deviation of 50 MPa. The stress acting on the
component is normally distributed with a mean value of 250
MPa and a standard deviation of 50 MPa. Compute the bounds
on reliability for equal to 5% .
Chap. 4 Structural Reliability Analysis 280

11-(Problem 9 Page160 K&L)

The stress and the strength distributions for a component are
Weibull with the following parameters:
Strength: A=300 MPa, B=400 MPa, C=3
Stress : A=150 MPa, B=300 MPa, C= 4
Compute the bounds on reliability for =0.05

12- Find the a component reliability with exponentially

distributed strength with parameter   0.001 and normally

distributed stress N (  s  35 K pa ,  s  5 K pa ) using the

following equation (Eq. 6-31K&L p139)

s 2 s2 s  2 s2
R  Z ( )  exp(  s  )  [1  Z ( )]
s 2 s

������������������������������������������������������������������������������))�

13- Is it possible to derive Eq.4.9 from Eq.4-13 or 4-15 of this

chapter?

�
281 Reliabilty Engineering

Chapter 5
on the
Combination of
Random Variables
in Design
& a Glance at the
Tolerance Concept
Chap. 5 Combinations of random variables + tolerance 282

5
On the combinations of random variables in design ;
A glance at the tolerance concept

Aims of the chapter

This chapter is concerned with finding some properties of a

function of several random variables. The chapter also reviews
the concept of tolerance in designs quickly.

������������������

The reliability of an engineering design is often a function of

several quantities. Variability is inherent in most of these
quantities; i.e. most of them are random variables(RVs). As an
example consider the design of a beam. Stresses in beams due
to bending is very important for an engineer; therefore he is
usually interested in computing the bending stresses. The
following formula gives the maximum stress in a beam due to
bending(K&L page 95:

M c
S  (5-1-1)
I

�
283 Reliabilty Engineering

where
S = the maximum stress at the farthest surface from the neutral
axis (it can be at top or at bottom), kPa

c = the maximum distance from the neutral axis to the extreme

fiber (again, this can be to the top or bottom of the shape), m

I = the moment of inertia of the beam cross section about the

centroidal axis, m4

M = the bending moment along the length of the beam where

the stress is calculated, N.m

if the maximum bending stress is required then M is the

maximum bending moment acting on the beam

The moment of inertia of a beam having circular cross

section with radius r meters and thickness of t meters is
×
� . Thus according to Eq. 5-1-1, ‫= � ﺑﺮﺍﺑﺮ‬ gives the

maximum fiber stress in such a beam. If the beam has a

rectangular cross-section of height a meters and width b

meters, then = in and the corresponding stress is

�
calculated from � = ��
Chap. 5 Combinations of random variables + tolerance 284

Fig.5.1 Bending moment on a beam- An illustration

It is worth mentioning that if yield stress(� ) replaces S,

the external bending moment( ) causing the beam to reach the
yield point is calculated as follows:
� (5-1-2)

�
285 Reliabilty Engineering

Figure 5-1 illustrates this equation.

In real world, quantities such as M,c,b & a are random

variables. Hence to compute any property of a function of
these random variables such as M c , we need to know how
S
I

to combine as M ،c ،b ‫ ﻭ‬a. Now for this reason we focus on how

to find certain properties of a function of random variables.

����������������������������������������������
�����������������

In this section, given random variables ,…, we are to

show how to determine certain properties ( ,…, ) i.e. a

function of them which is in turn a random variable(K&L p 96).

5.2.1 The pdf of a function of one random variable

Suppose we are given Y=g(X) where X is a random variable

with known density function(pdf) f(x) and we would like to find
the pdf h(y)for the random variable Y. h(y) is given bythe
following relationship(K&L p97):

ℎ( ) = × [ ( )] (5-2)

where

k(y) is the inverse function of g i.e. � �� ( ) = ,�

Chap. 5 Combinations of random variables + tolerance 286

represents the absolute value of the derivative of � )

with respect to y.

If � ( ) has 2 answers � , 2 then

ℎ( ) = × ( )+ × ( ) (5-3)

In general, if the inverse function has n roots � , … , 2 , Eq. 5.3

will have n items, one term for each root(K&L page97).

Example 5.1
Random variable X is normally distributed with density
2
 x  
function(pdf) 1 and � . Find the pdf of Y.
f (x )  e 2 2
 2

Solution
� � )= ,�

�������Y = ℎ( ) = × [ ( )] , ( )= ( )�

1
= ⟹ = = ( )�������� = ��

2
 ln y  
1 2 �
f (k ( y ))  e 2
 2

2
ℎ( ) = × [ ( )]=
1
× e
1
  ln y   
;
 2 2
2

This the pdf of lognormal distribution. This is well known in

statistics that if random variable X is normally distributed,
has a lognormal distribution.End of Example

�
287 Reliabilty Engineering

Example 5.2 (K&L p97)

The diameter(D) of the circular cross section of a kind of rod
2
 x  
has a normal distributionwith 1 . Find the
f (x )  e 2 2
 2
probability density function of the cross section i.e.� = .

Solution

4A 4A
= ⟹ �1   ��2   ;
4  

According to Eq. 5.3:

ℎ( ) = × ( )+ × ( )

Here �y=A , k=D����� � ������� � ��thenℎ( ),the pdf of

the cross section is calculated from:

4A 4A
ℎ( ) = ×  � ×  �

 

1 4 4 1 1 1 dD 1
D  A 2 | dD |  2  A 2 | | �

 dA
 dA A
2 2
 4A   4A 
     

[exp(    )  exp(   
 
)] ���
1 1
h (A )  
A  2 2 2
2 2

End of Example
Chap. 5 Combinations of random variables + tolerance 288

5-2-2 Mean of 2 random variables

If X , Y are random variables( no matter whether continuous

or discrete, independent or not) with mean � then:

E[X+Y]=E[X]+E[Y] or �
= 1
+ 2
(5-4)

Proof for continuous case(Ross, 1985 p46):

If X and Y are continuous random variables with marginal

density functions � ) ،� � ) and joint ����� ( , ), then

End of Proof

5-2-3 Variance of sum and difference of 2 random

variables

If � ���� are 2 random variables with variances �

and coefficient correlation ρ then for � + :
= + +2 ( )= + +2 (5-5)
and for� = − :
� − (5-6-1)

� � −2 ( )= + −2 . (5-6-2)

�
289 Reliabilty Engineering

����������������� �����������������������������������������

If � ���� are 2 independent random variables with

variances � then for � + :

� � . (5-7)

5-2-4 Approximating mean and variance of a

function of a random variable

Suppose X is a random variable with mean μ and variance

σ . Let Y be a function of X, then (K&L p101):
� � = [ ( )] ≅ ( )+ "( ), (5-8)

and also(K&L p102):

( ) ≅ [ ′( )] . (5-9)

Example 5.3(K&L page102)

The radius of a kid of a bar is a random variable with mean
�2 & standard deviation � . Find the values
of the mean and the standard deviation of the bar cross section.

Solution
The cross section of the is calculated from:
� ( )= ′( ) = 2 "( ) = 2
��������������
( . )
� )≅ ( )+ "( ) = (2 ) + (2 ) =������ �
From Eq. 5-9:

( )≅[ ′( )] ⟹ ( ) ≅ (2π�������2 ������ �

End of Example
Chap. 5 Combinations of random variables + tolerance 290

5-2-5 Approximating the mean of a function of some

independent random variables

If � , … , are independent random variables with means

�…, and variances ( ),�� ( ), an approximation
of the mean of a function of these variables � �X , … , ) is
given by(K&L page 103):

� � ≅ � ,…, )+ ∑ ⃒ ( )����������������������

Or in vector form:
|
⎛ . ⎞
� ) ≅ ( ,…, )+ ( )…. ( ) ⎜ . ⎟����������������
⎜ . ⎟

⎝ | ⎠

by � it is meant to replace ��'s with� �'s.

5-2-6 Approximating the variance of a function of

some independent random variables

Let � (X , … , ) be a function of independent variables

� ,…, having standard deviations �…, then:

⎧ ⎫
⎪ ( ,…, ) ⎪
( )≅∑ × ⃒ �����������������
⎨ ⎬
⎪ ⋮ ⎪
⎩ ⎭

Example 5.4(K&L page104)

�
291 Reliabilty Engineering

The load f acting on a bar in tension has a mean value �

10,000 N and a standard deviation � 1,000 N. The mean
value or the cross-section area A is �5.0 cm1, and the
standard deviation of A is � 0.4 cm. Find the mean and
standard deviation of the tensile stress S on the bar.

Solution

= = ( , ),

According to Eq. 5-10-2:

⃒
1 ⎛ ⎞
� )≅ ( , )+ ( )⎜ ⎟�
2
⃒
⎝ ⎠
2 2
2 �����
=− , = . ⃒ = = ����
�

1 �����
= , =0. ( , )= ⟹�
�

2 0
����� 1(0.4 0)
��� ⟹ �
( )≅ +
� �

( ) ≅ ����� = ������ = �� �

�����������������������
Chap. 5 Combinations of random variables + tolerance 292

( )≅ ( ⃒ ) + ( ⃒ )
1 −�����
= (����� + (0.4) �
5 �

⟹ ( ) ≅ ������� ⟹ ≅ ����� = ���� �

Therefore a stress with mean 20 MPa and standard deviation

2.56 MPa is acting on the bar.

End of Example

5-2-7Approximating the mean and variance of

Let X and Y be 2 independent random variables with mean

‫��ﻭ‬ and variance and . The mean of the quotient is

approximately :

 X  X   Y  
�

E   �     �����������������
 Y  Y   Y  

From Eq. 5-12-1 it is concluded that:

1  Y  ������������������������
2
��  1  
E   1    
 Y  Y   Y  

Furthermore the variance of is approximated by:

�
  �
 X   X 
� �

��V ar        x     Y    Y   �����������

 Y   Y    X    Y    Y  
 

�
293 Reliabilty Engineering

Example 5.5

Repeat Example 5.4 using Eqs. 5-12-1 & 5-12-3.

Solution

P    A  
�
����� 0.4
�� ) = ≅ �    � ∗ 1+ = �����8 N / cm 2
A   A   � 5
 
⟹ E( ) = ����� �

�
  �
 P   P 
� �

( ) = Var       P     A    A  
 A    A    P    A    A  
  
����� ���� 0.4 0.4
� ∗ � − = ������
� ����� 5 5

� ≅ ����� N / cm 2 � ���� � End of Example

�����������������������������

Since it is impossible to make everything to an exact size, the

specification for design dimensions and variables is usually
given as a nominal value plus minus a number. For example in
2.500  0.003, 2.500 is the nominal value and  0.003 is the
tolerance. Tolerance is the total amount a dimension may vary
Chap. 5 Combinations of random variables + tolerance 294

and is the difference between the upper and lower (minimum)

limits of the specification.
It is worth mentioning that tolerance is sometimes written as
a percent; e.g. 2�500  0.12%. By this notation it is meant that

tolerance is  0.12
100  2.500=  0.003.

Next, after reminding the calculation of tolerance of linearly

and nonlinearly assembled parts, an example shows the
calculation of reliability when tolerance is given. Thought the
chapter it is assumed that the tolerance of a component or
assembly is ± where is the standard deviation of the
component or the assembly. k is a constant which is usually
equal 3.

5-3-1 Relationship of assembly tolerance parts

tolerance

Consider a product composed of n similar parts. Let the

dimension of each part be denoted by � i = 1, … , n with mean
2
and variance  part and  part . The dimension of the assembled

product is therefore � = + ،. . . + . The mean and variance

of X denoted by sum and σ S2 u m equals:

����� ���� (5-13-1)

Sum
2
 nPart
2
(5-13-2)

�
295 Reliabilty Engineering

 part 
 sum
2
(5-13-3)
n

Let us denote the specification of each part by a  t and that

of the assembled part by b   and tolerance limits by U and L.
� ) ( )
Assuming = , = i.e. the tolerance
( )
equals � ������������������ then substituting = �
( )
and = in Eq. 5-13-3 results in:
(5-14)
U  L sum
2

(U  L ) part  .
n
�

( ) ( )
Let��∆� �and � then:

∆
���������� � �������������������������
√

��������∆� √ ������������������������������������������

Then the tolerance specification of a product assembled from n

similar parts with specification � � would be � � ∆. If the
n parts have different specifications � � t … a ± t … a ± t
� − )
then if 2
= �, the specification of the assembled part

would be ∑ a ± ∆′ where:

∆′ = 2
. (5 − ����
=1
�
Chap. 5 Combinations of random variables + tolerance 296

Example 5-6

Suppose 10 similar parts with specification 2.000 � 0.012

are assembled in a series configuration. What is the
specification of the assembled product?

Solution

The specification of the assembled part is��� × 2.��� � ∆

where ∆= √ = 0.���√��=0.0379 or: 20.000 � 0.0379.
End of Example

Example 5-7

We would like to produce a product with specification 20.000

� 0.0379 as an assembly of 10 similar parts. What should be
the tolerance of each part?

Solution

.
The specification�of each of the 10 parts must be ±
∆ .
where � = = 0.��� . End of Example
√ √

Example 5-8

�
297 Reliabilty Engineering

A part with specification 2.000 � 0.012 is assembled with

another part have the specification of 3.000 � 0.016. What is
the specification of the assembly?

Solution

The specification of the assembled part is 2.000+3.000�∆′

where ∆′ is given by Eq. 5-16 as follows:
∆′ = ∑ = =0.020. Therefore the
(0.���� � (0.����
specification is: 3.000�0.020 End of Example

5-3-2 Tolerance in complex systems

In the previous section, tolerances for series configurations

were considered. In this section an example illustrates how to
calculate the tolerance of an assembly in which a nonlinear
function of the components exists. This is usually accomplished
by linearization of the function by Taylor�� expansion up to the
first order around the nominal dimensions.
Example 5-9( Extracted from Bowker-Lieberman,1972 p94)
In the electrical circuit shown below, find the tolerance of the
output voltage � . The components have the following
specifications:
Chap. 5 Com
mbinations of random variables + tolerance 298

������������� ������� ����� �

�
�� ������ �� ���� �������� ��� �� ����
� � �
� ������ �� � �� ������
� � �
�
� ������ � �� �� ��������

The mean off the distributions are equal to their nominal values.
v

Solution
Taylor seeries of a 3-variable function up to the first order is
as follows:

1   
f  x , y , z   f a,b,c    x a  y b   z c  f  x , y , z  ...
1! x y z  x  a, y b, z c

������

For the liinearization of � , is expanded

d into a
Taylor seriees around the nominal values ��* , � * , � * with help of
the above rel
elationship as follows:

�
299 Reliabilty Engineering

1
E 0  E 1*N *K * 
1! 
    
E 1  E 1* N *K *  N  N * E 1* K *  K  K * E 1* N *   
*
E 0  N *K *E 1  E 1 K * N  E 1*N *K  2E 1*N * K *

This approximation is a linear combination of � , . Since random

variables � , have the means ��* , � * , � * , therefore the

distribution of has the mean  E 0 calculated from:

0 
E  E N *K *E  E *K *N  E *N *K  2N *E *K *
1 1 1 1 
Since the mean of the distributions of � � equal ��* , � * , � *

therefore the mean of �is:

* *
 E  N * K *  EE  E 1* K * NN  E 1* K * N *  2N * K * E 1*
0
1
1

or  Eo �� �*  � *  � *

The variance of ,denoted by  E2 , is calculated from the linear

0

approximation of indepenvent variables , , i.e:

E  N *K *E  E *K *N  E *N *K  2N *E *K * 
0 1 1 1 1

     
2 2 2
2 * * 2 * * 2
E  N K  E  E1 K   E1*N *  K2 (5-18)
0 1 N

No notice that under the following 3 assumptions

Chap. 5 Combinations of random variables + tolerance 300

a)During production, the dimension of each component can be

centered at its nominal values i.e. the magnitude of the variance
of each component is such that the natural tolerance limits
coincides with the specifications limits � �,
b)The distributions are normally distributed with means equal to
nominal values
c)For each component all but 100α� of the values will fall
within the corresponding � � the largest value of the standard
deviation of the dimension denoted by σ is given by(see
K&L pages93-96):

σ = � � ���������������������������

where z is the critical value related to normal distribution

given in Table D or by a software such as MATLAB; e.g. for
 ������  Z  z�������  �������(����������)  � and
�

therefore:
�
��� ����� ������
E �
�
� N � �
� ������ � k � �����
� � �
Substituting numerical values in the right hand side of Eq. 5-18
* * *
gives  E  ����� . Eo  E� N K  40� 12 ����� is the nominal value
�

of the output voltage. To write the specification of output

voltage E� as �� � Δ note that the distribution of E� can be
approximated by a normal with mean equal to the nominal
value(60); therefore for α = 0.��� from Eq.5-19 :

�
301 Reliabilty Engineering

Δ Δ Δ Δ
�.��� � = = � ⟹ Δ = 1.���
. 0.����� 3

�����
� �� : �� � 1.����� ���or�since�� ������  E 0 � �������� ��
��

Therefore the output voltage � is 60 volts�2.���.

The folowing example shows how tolerance affects reliability. �

Example 5-10 (Based on K&L��page 165)

A circular bar is subjected to a tension load S, shown below.

Due to the nature of manufacturing, the diameter d of the bar is

a random variable and due to various raw materials used the
ultimate tensile strength of the rod is also a random variable

with mean   10000 psi and standard deviation 5000psi.

The random variable S has S ����� �� and S  100 lb. If
the load and strength are normally distributed , the following
Chap. 5 Combinations of random variables + tolerance 302

equation1 is used to calculate the reliability of these kind of

bars:

R  Pr(S   )  Z (z ) (5-20)�
where �

= and
( . )
�

Φ ( ) is the CDF of normal standard distribution.

The diameter of the bar has a mean ��of r  0.12635 inches and its
specification is r  p % . The load and the strength are normally
distributed. To know how the variations in the rod diameter affect the
rod reliability, conduct a sensitivity analysis of the rod reliability
with respect to the rod radius.

Solution

The following table shows the reliability of the rod computed

using MATLAB from Eq. 5-20 for seven values of p in r  p%

and   100���3 psi �  ����3 psi � S  ����3lb � S  ���lb ,  r  0.12635

�����������������������������������������������������������
1
For proof see K&L page 165

�
303 Reliabilty Engineering

p z R   Z (z )  normcdf  z 

0 3.760 0.999915
0.5 3.756 0.999914
1.0 3.74 0.999908
1.5 3.72 0.999900
3.0 3.61 0.999847
5.0 3.37 0.999624
7.0 3.10 0.999032
If for example the specification of the radius of the rod is
0/12635����� i.e. p=1.5, z turns to be z=3.72 and the rod
reliability would be 99.99% as calculated below:
p=1.5;
z=(10^5-
(4000/(pi*(0.12635)^2)))/sqrt(5000^2+((100^2+((4*(0.01*p)^2)
/9)*4000^2))/(pi^2*0.12635^4))
R= normcdf(z)

End of Example.� �

������������

In the following problems assume all dimensions are normally distributed and the tolerance
range is 6 sigma(��σ�

1. The parts of a contact assembly for a relay are shown in the

following figure . The dimension x represents the amount of

�����������������������������������������������������������
1
Problems 1 through 7 are from Chap. 5 page 113 problems(1,3,5,7,9,11,13)
Chap. 5 Combinations of random variables + tolerance 304

intentional overtravel (called "wipe") of the upper contact that

would occur if the upper contact was clamped to the part at left.
Find the nominal dimension x and its tolerance.�

2. A partially finished connecting rod is shown in the following

figure. Each radius has a tolerance of ��������������������������
the distance L between the centers of the holes is �������� �����
the tolerance for the dimension h.

�
305 Reliabilty Engineering

3.A rectangular solid bar has the following dimensions:

X : 2������m �Y � �������m � Z � �������m . Find the

specification of V=XYZ.

4.The head of a screw is shown in the following figure. The

various dimensions are formed in such a manner that there is no
association between them; that is, they are mutually
independent. Determine the tolerance for H, the depth of the
screw head. The dimensions 9, D, and d and their tolerances are:

 :90o ���� D  ����������� in d  ����� � ����� in

Chap. 5 Combinations of random variables + tolerance 306

5. Let random variable � , where� is a constant and the

3 variables are independent and have the following properties.

Variable
mean 4 2 1
��stand. devia 0.4 0.2 0.1

Approximate the expected value and standard deviation of Y.

6. An automotive component is subjected to a fluctuating stress

with mean and the amplitude = , =

, as shown in the following figure. The maximum

value of the stress ( smax ) is a normally distributed random

variable with mean= 600 kPa and standard deviation=40kPa.
The minimum value of the stress( smin) is a gamma distributed
�
random variable with parameters n= 17 and  20kPa .


�
307 Reliabilty Engiineering

Determine th
he value which the random variable smean willl exceed
only 1.3% off the time. Also determine the value which th
he
random variaable samp will not exceed 90% of the time.
Hint: For lar
argish n a gamma distribution (n, ) could be
√
approximateed with a normal distribution ( � � � �;

becaue it be consisidered the sum of n inpenendent expon

nential
distribution with
w parameter ( ),

7. The analyysis of the loading of a component revealed a load

diagram as shown
s in the Following figure..

The four forc

rces F1, F2, F3, and F4 are random variables, the
th
distributionss of which are given in the following table
Chap. 5 Combinations of random variables + tolerance 308

Force Distribution Parameters

F1 Exponen. �� �  2.6kN


F2 Normal ��   41.6kN ,   11.03kN

F3 Normal   37.5kN ,   3.9kN

F4 Normal   39kN ,   10.07kN

Calculate the mean and the variance of the magnitude of the

horizontal resultant load.

Hint: The gamma distribution (n, ) may be approximated by a

√
normal distribution with � and � .

8.(Example5-6 page 104K&L)

An electrical circuit has two resistances R1,�R2� in parallel as

shown below. The value of s each resistance is a random
variable. We know that

 R 1  100  ,  R 1  10  ,  R 2  200  ,  R 2  15  ,

Determine the mean and standard deviation of R T.

�
309 Reliabilty Engiineering

Answer is giiven in K&L page 105.

9. A kind off beam has a rectangular-shaped cross-sectio

ion. The
dimensions of the cross-section, denoted by a & b, are random
variables wiith mean   20 mm,   60 mm and standard
s
a b
deviation   0.02 mm,   0.1 mm. A stress is applieed to the
a b
beam. The bending stress is a random variable with
h mean
2000Nm and
nd standard deviation 10Nm. Find an appro
oximate
value for thee maximum bending stress(S). The dimensiions are
assumed ind
dependent.

Anyo
one pursuing his goals honestlly
does not slip up
and iff he does, he can seek a way out
o
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 310

Chapter 6
Estimation of
Mean Life &
Reliability,
Related
Experiments
&Tests

�
311 Reliabilty Engineering

6
����������������������������������������������
����������������������������

Aims of the chapter

This chapter deals with the estimation of 2 statistical

measures related to a product i.e. reliability and expected value
of products lifetime. Some standard experiments and statistical
tests of hypothesis are also mentioned, and some acceptance
sampling plans based on the lifetime are introduced. The
emphasis is on the products whose lifetimes are exponentially
distributed.

6.1 Introduction
The problem of estimation of the lifetime and the reliability
of products is a common problem in the control of products
quality. When we have the lifetimes of a random sample of the
product, one obvious way to estimate the mean lifetime is
calculating the sample mean. Another way is performing special
life tests on the sample and then calculating the mean. A third
way of life testing is called accelerated life testing which
involves the acceleration of failures to quantify the life
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 312

characteristics of the product at normal use conditions; in other

words it involves capturing product life data under accelerated
stress. These 3 ways are pointed out in this chapter. The present
chapter also mentions some statistical tests hypothesis and some
acceptance sampling plans related to lifetime.
6.2 Estimation of product mean life given a lifetime
sample of size n
Suppose a random sample is taken from a product and the
products in the sample are tested until all of them fail and the
lifetimes � ،� � � ،x is obtained. The following equation gives an

unbiased estimate for the mean lifetime ( )of the product:

∑
θ� = (�6-1)

This equation could be used for any product with any lifetime
distribution.
6.3 ��������� Estimating Mean Life �
Consider a life testing where n items are simultaneously
placed on test. The purpose of the life tests here is to calculate a
point estimate and sometimes interval estimates for the product
mean life. It often occurs that we need to discontinue the life test
before all the elements in the sample fail. In such cases, we say
that the test has been ������������� ������������ ��� ������������
.Censoring schemes employed during the life test make the
inspection as a cost effective one. Time censoring (Type I),
failure censoring (Type II) are 2 common types of the censoring

�
313 Reliabilty Engineering

schemes employed in life tests. Each of these 2 types might be

performed in 2 ways:
i)Censoring schemes with replacement.

Replacement during a life testing means that once observing

a failure item, it is replaced by a new or repaired one. In other
words, the total number of inspected items during the test
remains constant n.

ii)Censoring schemes without replacement

6.3.1 Time censoring (Type-I)

In a time censoring scheme, n items are simultaneously

placed on the test and the test terminates at some specified time.

6.3.2 Failure censoring (Type-II)

In a failure censoring scheme, n items are simultaneously

placed on the test and the test continues until particular number
of failures , say r ,occurs.
To summarize the above discussion: it often occurs that we
need to truncate our life test before all the elements in the
sample experience the failure. Two common types of truncation
or censoring are :
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 314

ℎ
⎧ :
⎪ ℎ

⎨ ℎ
⎪ :
⎩ ℎ

6.4 Estimation of mean life ��

List of Symbols
Number of units of product placed on life test
Number of failures
∗ Predetermined amount of time for a life test
�
The total operation time of all test items
Time to failure for product no. i
� �� Time to failure for ith failure

� )� Time to failure for failure no. r

Mean lifetime of the product

Estimate of

The mean life � ) or MTBF of products is estimated from the

following general formula��

θ� (6-2)

where
T = the total operation time of all items placed on test including
those failed,
r = total number of failures occurred during the life test

�
315 Reliabilty Engineering

Needless to say if the lifetime of a product is exponentially

�
distributed with pdf ( )= , Eq. 6.2 estimates the

parameter of this diminution.

Note that;
To verify "that an exponential distribution fits the life data"
a test of hypothesis such as Bartlett's test (see K&L page 239) or
Q_Q plot with following MATLAB command could be used:

X=[�����������������������xponential', mean(X));qqplot(X,pd).

Example 6-1(K&L page 251)

A truck was shaken on a simulator for a total time of'245 hours.

During this life test 20 failures occurred. The time between
failures can be well approximated by an exponential
distribution(see page 240 K&L). �����������������������
�����������
Solution

M T�B F = = = ������ℎ ��������������� �

Calculation of T for = = =

To calculate T for estimating , let us distinguish the following

cases for discontinuing our tests:

-Type I tests(time truncation) with replacement

Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 316

- Type I tests(time truncation) without replacement

-Type II tests(failure truncation) with replacement

-Type II tests(failure truncation) without replacement

6-4-1 Type I censoring life test

In type-I censoring at a predetermined time, say�� ∗ , the life

test is terminated. The test could be performed without or with
replacement.

���������������������������������������������������

In time �truncated tests with replacement, in fact all n items

∗
work until the predetermined time �� and the total operation
time(T) of all items placed on test including those failed
∗
is:� = × ; therefore:

T nt *
�   ������������������
r r

∗
Where � ≥ 1 is the number of failures dung the test time .
An application of this equation is , for example, when we have
where we have n test stands, and we cycle each
test stand for  cycles. As items fail they are replaced. Where a
truncation time is specified this is called Type I censoring. Here
we have(K&L page 252)

�
317 Reliabilty Engineering

n
�  ���������������������
r ��

Example 6-2(Example 10-12 K&L page 252 based on Example 10-3 page 241)

Nine stands are used for testing the life of a kind of switch.
As items fail they are replaced. Each stand was cycled 20,000
times, and a counter recorded the cycle number at which failures
occurred. The following table contains the data. Estimate MTBF.

Stand no. 1 2 3 4 5 6 7 8 9
4100, 18100, 18950

Cycles at

No
No failure
3100, 8100

which
6700

4600

5400

2600

4700

failure
failure
occurred

Counters were not reset when a new switch was placed on

test. Thus counts are continuous from zero. Total test is 20,000
cycles per stand with repla��������

Solution

� n � � � ����� � �������

Ten failures occurred during the test duration i.e. r=10,

therefore :
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 318

T ������
M T�B F � θ = �  � � ������
r ��

This means that on the average, each switch could be

cycled 20,000 times before failure.

Example 6-3(Lewis,1994 page240)

A chemical plant has 24 process control circuits. During 5000

hr of plant operation. The circuits experience 14 failures. After
each failure the unit is immediately replaced. What is the MTTF
for the control circuits?

Solution

T  24������������� ������

 *
�    T  nt  120000 =8571hr �
MTTF
r r 14 End of Example

������������������������������������������������������

In type I censoring the test is terminated at some

predetermined time; in nonreplacement case the number of
failures 1  r  n and MTTF is estimated from (Lewis, 1994page
239,Mann,1974 page 173):
r
( n  r )t *   x ( i )
T
�   i 1 ����������������(6-5)
r r ��

�
319 Reliabilty Engineering

where

� number of items failed

� total operational time for the n units tested
∗ duration of life test
�
�� number of items placed on life test

x (i ) time of the ith failure x (1)  x (2)  ...  x (i )  ...

Example 6-4:

20 units of a kind of gyroscope were placed on a 30-day life

test without replacement. 9 units failed at the following
times(indays) 14.4, 5.1, 27.7, 29.1, 23.6, 20.00, 10.5, 13.5, 27.4.
Estimate MTTF .

Solution

x1 �14.4����� x9 � 27.4 x 171.3

i
r
��  30days , T  ( n  r )t *   xi T  (20-9)���������  �����
l 1

�    T � 501.3 =55.7 days .End of Example
MTTF
r 9
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 320

6-4-2 Type -II censoring life test

In Type II or failure censoring, the test is discontinued after
occurring a predetermined number of failures ( r ). This type
might be performed with or without replacement.

�������������������������������������������������������

In this kind of experiment n units of a product are placed

simultaneously on life test and failed units are not replaced.
When the number of the failure reaches the predetermined
number r (1≤�r ≤�����������st is terminated. The estimate for
MTTF is(K&L page252, Lewis,1994 page239):

� ) ( ) ∑
= = �����
��
�� ith value in the sample containing failure times
� ( ) The time of the rth failure

�� Total operation time of all items placed on life test

r r
Total operation time of failed items
 x i   x (i ) �
i 1 i 1

� − ) () operation time of the functional items at the end of the test

References such as Mann(1974) page 164 provide some
descriptions on the proof of Eq. 6.6 .

It is obvious that:

�
321 Reliabilty Engineering

1. If the time of the life test is such that all n items fail then
∑
r=n and Eq. 6.1 i.e. θ = is obtained.

2. If x (r ) coincides the test time �� ∗

the equations of Sec

6-4-1-2 &6-4-2-1give the same result.

Example 6-5

The director of a laboratory, place 20 units of a kind of

gyroscope on life test and decides to stop the test whenever the
tenth failure occurs. At time41.2 the tenth failure occurs and the
experiment terminates. The time of the other failures are:
14.4 ٬5.1 ٬ 27.7 ٬29.1 ٬�23.6 ٬�20.0 ٬10.5 ٬�13.5 ٬�14.4. Find
the MTTF of the gyroscope.

Solution
r
T  (n  r ) x( r )   xi  (20-10)�����������������������������5
i �
Then according to Eq. 6-6

�  �  T  624.5 =62.45 days . End of Example

MTBF
r 10

����������������������������������������������������

In this kind of experiment n units of a product are placed

simultaneously on life test; when a unit fails it is replaced. The
test is terminated when the number of failures reaches a
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 322

predetermined number . in other words the test terminates at

time x(r). The estimate for MTTF is(Lewis,1994 page239):

 nx
  ( r ) �������������������������������������
r

n The number of items placed initially on the life test

x(r) The time when the rth failure occurs

Description of Eq. 6-8:

The experiment ends at time x ( r ) ; ≥ � and r could be less

or greater than n. during the experiment time totally r failures

occurs for all the n units, therefore each unit on the average
x
fails ��times and the mean time to failure is �  ( r ) . Hence we
r
n
have the following estimate for MTBF:

T nx ( r )
MT�BF  �   ��������������������
r r

Example 6-6(lewis,1994 page241)

Six units of a new high-precision pressure monitor are placed

on an industrial furnace. After each failure the monitor is
immediately replaced. However, the eighth failure occurs after
only 840 hours of service. It is decided that the high-temperature

�
323 Reliabilty Engineering

environment is too severe for the instruments to function

reliably, and the furnace is shut down to replace the pressure
monitors with a more reliable, and expensive, design. Assuming
that the failures are random, estimate the MTTF of the monitors.
Solution
� ( ) =6 =8 ( ) = ����
�  �  T 5040
= 6 �������������������� MTTF   630 ���
r 8
Tests Summary
Table 6.1 summarizes the relationships related to the above life
tests:
Table 6.1 � Equations for estimating MTTF or MTBF�= )
Type of Eq. T nt *
Experiment
Replacement
No.
�   �
r r
Time-terminated With 6-3 r

(Type-I Replacement ( n  r )t *   x ( i )
T
censoring) �   i 1 �
r r
Without 6-5 ( − ) ( ) +∑
Replacement �
Failure- Without 6-6  nx
terminated Replacement   (r ) �
(Type-II r
censoring) With 6-8 T nt *
Replacement �   �
r r
�
�����������������������������������������������
At the end of the subject of experiments, it is worth
mentioning that there are some life tests called accelerated life
tests for quick obtaining of lifetime data.
Conventional way for preparing the lifetime data of a product
is to place some units of it on life test under normal use

conditions, until all fail. This procedure for obtaining TTFs is

Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 324

difficult and sometimes impossible. Some experiment methods

called Accelerated Life Tests have been developed to expedite
the test and save time and cost during the design and
development validation phase in many industries. ALT perform
1
the life test at a high level of a parameter or a variable. The
results obtained at high levels of the accelerating variables are
then extrapolated to provide information about the product life
under normal use conditions.
In accelerated life tests, place a sample of the product on life
test under elevated working conditions (temperature, voltage,
pressure, rate, vibration, humidity and so on...) in order to
accelerate the failure mechanisms. The results are then used to
extrapolate to usual operating conditions.
Many references including Cabarbaye(2019), Tobias&
Trindade (2012) discuss ALT. Softwares such as Minitab
perform data analysis for ALT.

�������������������������������������������
������������������������������

Whenever a sample is taken from a population, different

estimate for a parameter of the population is obtained. To
modify this difficulty on could construct a confidence interval
for the parameter. Suppose ��، is a random sample from a
�����������������������������������������������������������

1Pascual et al,2006 Accelerated Life Test Models and Data Analysis, In

book: Springer Handbook of Engineering Statistics (pp.397-426)

�
325 Reliabilty Engineering

distribution with unknown parameter θ, the interval [G1, G2]

where G1, G2 are function of �…، is called a 1-� confidence

interval (CI) for θ� if :

�� Pr (G1 <θ��2 � ���α (6-10)

Next the confidence interval for the mean of exponentially

distributed lifetimes. To do this consider a life test of a sample
taken from exponentially-distributed-life product ,terminated
after time T during which the number of failed items has
reached r. assuming zero-minimum life, � − α confidence
interval for the mean of an exponentially-distributed �life
 
product is(K&L page 253):  2T 2T .
 2 2 
 2r , 2 r ,1
 
 2 2 

This confidence interval (CI)which could be written as:

 
 2T 2T  (6-11)
Pr  2    2   1 
  
 2 r ,2 2 r ,1
2


Is based on the assumption the lifetime of the product is

T
exponentially distribute d with mean θ���������� from �  .
r
2T
The proof relies on the fact that random variable has a Chi-

Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 326

squared distribution with 2r degrees of freedom(Appendix 10.D

K&L page 281).

This CI for example, is suitable when we have n test stands

where items are replaced as they fail and the test is discontinue
d at a predetermined time. Or when we might drive vehicles
over e.g. a 40,000 km test schedule and elect to count failures
rather than failure intervals(K&L page 254).

Example 6.7

8 leaf springs were tested to failure . The results, in cycles,

follow:
X�1) X�2)� X�3)� X�4)� X�5)� X�6)� X�7)�� X�8)�
8712 21915 39400 54613 79000 110200 151208 204312
a)Estimate the mean lifetime
b) Suppose the lifetime is exponentially distributed, find a
95% confidence interval for
c) find a 95% confidence interval for the spring reliability if
vibrated 4000 cycles.

Solution

a)The point estimate for the mean life( ) is : �

�
327 Reliabilty Engineering

r 8
T=(n-r)x (r) + x i =(8-8)x (8) + x i 
i=1 i=1
8 8
T   x i =  x (i) =669320
i=1 i=1

∑ � �
= = = = �����

b)If the lifetime of the spring is exponentially

distributed, the CI for  is given from Eq. 6-11:

2T 2T
2
θ 2 .
χ α χ α
2r, 2r,1-
2 2

r  n  8,  =5% . From Table E or MATLAB:

2
χ16,0.975  ����������������  �����
2
χ16,0/025  �����������������  �����

2T 2T 2������� ��������
2
θ 2  θ 
χ α χ α ����� ����
2r, 2r,1-
2 2

θ L ������  θ  ������ �������θ U

c)CI for reliability function , if the lifetime is

exponentially distributed:
t t t t
   
L U
e  R (t )  e or e 46419
 R (t )  e 193736
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 328

for t=4000 cycles, CI is:

4000 4000
- -
e 46400
 R(4000)  e 193725 
0.9174=91.74%  R(4000)  0.9796=97.96%

Example 6.8:

The elements of a random sample a of a kind of electronic

circuit were placed on life test without replacement. The result

�    5000 . Find a 95% CI
was r=20 failures and mean of MTTF
for MTTF. Assume the lifetime is exponentially distributed.
Solution
According to Eq. 6-11:

 
 2T 2T  
Pr  2  θ  2  ���α ��� ��θ��������� ��α����
χ α χ
 2r, 2 2r1 ,(1- ) 
α
2 

χ 240,0.05 ������� χ 240,0.95 ������

From Table E:
Substituting the numerical values yields:
Pr(1793.4  θ  ����������� therefore a 95% CI for the

mean life is: (1793.4 , 3772.2). End of Example

�
329 Reliabilty Engineering

Example 6.91 for r=1:

Suppose the life time of a kind of product is exponentially

distributed . Due the fact that the product is

1) expensive and its life test is destructive or

2) expensive and a small quantity of it is available

The failure on only one unit of it for lie testing is affordable.

n=5 units of the product were placed simultaneously on life
test and when the first failure occurred at 15. 5 hr the test was
terminated. Find a 95% confidence interval for MTTF.

Solution
The test is of Type II without replacement, therefore:

r
T  (n  r ) x( r )   xi , r =1,x (1) =15.5 ,
i �
r
T=(n-r)x (r) + x i =(n-1)x (1) +x1 =(n-1)x (1) +x (1) =nx (1)
i=1

From Eq. 6-11:

2T 2T 2nx (1) 2nx (1)

2
θ 2  2 θ 2
χ α χ α χ α χ α
2r, 2r,1- 2, 2,1-
2 2 2 2

�����������������������������������������������������������
�
�������������������������������������������������������������
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 330

1-α�����  α������
MATLAB 
χ 22,0.975 �������������������������
χ 22,0/975 ���������������������������

2������� 2���15.5
n=5,x (1) =15.5  θ  ��  θ  ����
7.3778 0.0506

End of Example

6-6-1 Lower-bound confidence interval(CI) for mean

of exponential distribution
After a life test, the lower-bound � − α�CI for the mean of
an exponentially distributed lifetime could be calculated from
(K&L page257):
≥ = (6-12)
,

Where T is total test schedule and r the number of failures

occurred during the test

6-6-2 The confidence interval for the time during which

fraction p of exponentially-distributed-life products fail

Sometimes we would like construct a confidence interval for

the time or the kilometer or the temperature or������������
up to which the fraction of the products fail. � in other words is
such that � � �− where R(.) is the reliability function.

�
331 Reliabilty Engineering

Suppose the lifetime distribution of the product is exponential

�
with mean θ then = 1− or� = × �� and if the

estimate for is denoted by then we have the following

estimate for : � × ��

If we have a 1- CI such as � � � for the mean

lifetime, then the following interval would be 1- CI for �

× �� � < × ���( ) (6-13-1)

� ��� ) < < × ���( ) (6-13-2)

where
� is when (the time or the kilometer or the temperature
or������������������ of the working product fail or fraction R=1-p
of them do not.
Last part of problem 6 of this chapter exercises uses Sec.
6.6.2.
������������������������������������������
During the past years several researches have been done on
the subject of sampling from a lot of products to accept or reject
it based on product lifetime. Single, double and multiple
sampling plans have been developed in this regard which are
called Reliability Acceptance Sampling Plans (RASP) ,utilized
to inspect the quality of a lot for acceptance.
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 332

Among the first researches for life testing based on samples

taken from populations whose life follow exponential
distribution is the reference H108 Handbook1
This 88-page handbook is primarily concerned with three
different types of life test sampling plans and includes a number
of accompanying tables. These plans are
1) test terminated upon the occurrence of pre-assigned
number of failures,
2) test terminated at a predesigned time, and
3) sequential life testing plans.
Also provided are a set of 90 operating characteristic (oc) curves
applicable for the above three test plans. Some descriptions of
the sampling plans of this handbook is given in the author's
quality control book2.
Table 6-2 of this chapter is a sample of the handbook tables.
This table helps to determine single sampling plans for
inspecting lots.
Here single sampling plans are schemes in which a decision
to accept or reject an inspection lot is based on the inspection of
a single sample. A single sampling plan consists of a single
sample of size n placed on life test for a time T hours, with

�����������������������������������������������������������
�
�������������������������������������������������������������������
�����������������������������������������������������������������������������
�������������������������������������������
2
https://opentextbc.ca/oerdiscipline/chapter/industrial-engineering or
https://archive.org/details/statistical-methods-august-2020-bazargan or
https://opentextbc.ca/oerdiscipline/chapter/statistics/

�
333 Reliabilty Engineering

associated acceptance and rejection number(r). Note that during

the test time failed items are replaced with new ones unless the
number of items has exceeded r-1.

6-7-1 Type I&II errors of Sampling plans

The Inspection of a lot for accepting or rejecting using

sampling plans have might encounter errors. These inspection
errors are classified into two categories: Type I and Type II.
Type I error results in the rejection of a conforming lot, while
Type II error causes the inspector to accept a defective lot. Here
(in a single sampling with parameters n and c) The type I &II
errors could be described as follows:
When the lot posses a conforming reliability such as R1,and
the plan rejects it Type I error ( rejecting conforming lots)
occurs with the following probability.

�1− (1 − ) (6 − ����

When the lot posses a nonconforming reliability such as R2,

and the plan accepts it, Type II error ( accepting nonconforming
lots) occurs with the following probability:

� �1 − ) (6 − 15)�

6-7-2 Design of single plans using Table 6-2

Given the test time T, type-I error probability( ) for the

desired mean�θ and type-II error probability( ) for the
undesired mean�θ ,calculate � . Read the plan indices
( sample size n and rejection number r ) from Table 6-2.
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 334

Example 6.10
Design a plan whose test is of type "life tests terminated at
pre-assigned time" in such a way the test time does not exceed
T=500 hr . The plan is wanted to accept 90% the lots having
mean life θ =10000 hr (α = 0.��), and to reject 95% of the
lots with mean life θ � ������� (β = 0.���� The life is
assumed to be exponentially distributed.

Solution

= = , = =

�
 Table 6-2 Acceptance Sampling Plans for some � � & with a test
terminated at pre-assigned time with replacement (Table 2C-4��in H108 Handbook)�
� � T / 0 � T / 0
� � 1 1 � � 1 1 �
� � � � � �
335 3 5 �� �� 3Reliabilty
5 Engineering
�� ��
 /
r �n n ��n �n n n n ��n
1 0 r
� �.��‫� � ﻭ‬.�� � ��‫� � ﻭ‬.0.��
2thirds� 136� 3 551 1103 220 9 23 397 795 1591
1half� 46 9 158 317 634 3 72 120 241 483
1third 19 3 51 103 206 1 25 38 76 153
1 fifth 9 1 17 35 70 7 ��9 16 32 65�
1tenth 5 4 6 12 25� 4 ��4 6� 13 27�
α = 0.��‫ﻭ‬β � 0.�� α � 0.��‫ﻭ‬β � 0.��
2thirds 101� 2 395 790 158 6 16 270 541 1082
1half 35 6 113� 227 454 2 47 78 157 314
1third 15 2 37 74 149 1 16 27 54 108
1 fifth 8 8 14 29 58 5 6 10 19 39�
1tenth 4 3 4 8 16 3 ��3 4 8 16�
� 0.��‫ﻭ‬ � �.1� = 0.��‫� � ﻭ‬.1
2thirds 83 1 316 632 126 5 13 216 433 867
1half 30 5 93 187 374 1 37 62 124 248
1third 13 1 30 60 121 8 11 19 39 79
1 fifth 7 7 11 23 46 4 ��4 7 13 27�
1tenth 4 2 4 8 16� 3 ��3 4 8 16�
� 0.��‫ﻭ‬ � �.��� � �.��‫� � ﻭ‬.��
2thirds 60 1 217 434 869 3 77 129 258 517
1half 22 3 62 125 251 1 23 38 76 153
1third 10 1 20 41 82 6 7 13 26 52
1 fifth 5 4 7 13 25 3 ��3 4 8 16�
1tenth 3 2 2 4 8� 2 ��1 2 3 7�
� �.1‫ = ﻭ‬0.��� � �.��‫� � ﻭ‬.��
2thirds 77 1 329 659 131 5 14 234 469 939�
1half 26 5 98 197 394 1 42 70 140 281�
1third 11 2 35 70 140� 7 15 25 50 101�
1 fifth 5 7 12 24 48 3 5 8 17 34�
1tenth 3 3 5� 11 22 2 ��2 4 9 19�
α � ��1‫ = ﻭ‬0.�� α � 0.��‫� � ﻭ‬.��
2thirds 52 1 214 429 859 3 84 140 280 560
1half 18 3 64 128 256 1 25 43 86 172�
1third 8� 1 23 46 93 5 10 16 33 67�
1 fifth 4 5 8 17 34 2 ��3 5 10 19�
1tenth 2� 2 3 5 10 2 ��2 4 9 19�
α � ��1‫ = ﻭ‬0.�� α = 0.��‫� � ﻭ‬.1
2thirds 41 9 165 330 660 2 58 98 196 392
1half 15 3 51 102 205 8 17 29 59 119
1third 6 9 15 31 63 4 7 12 25 50
1 fifth 3 4 6 11 22 2 ��3 4 9 19�
1tenth 2 2 2 5 10 1 ��1 2 3 5�
α � 0.1‫ = ﻭ‬0.��� α � 0.��‫� � ﻭ‬.��
2thirds 25 5 94 188 376 1 28 47 95 190
1half 9 1 27 54 108 5 10 16 33 67
1third 4 5 8 17 34 2 2 4 9 19
1 fifth 3 3 5 11 22� 1 1 2 3 6
1tenth 2 1 2 5 10 1 1 1� 2 5
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 336

Table 6-2 gives n=34 & r=4 from with� = , = ,α = 0.���,

& β = 0.��. That is a random sample of size 34 is taken from the

lot and its items are put on life tested simultaneously; unless the
number of failures before 500 hr is equal to r=4 if before the end
of 500-hr test, a failure occurs it is replaced by a new one, and
the total number of failures is updated. End of Example

6-7-3 The operating characteristic (OC) curve of

single sampling plans

Remember that in quality control the so called OC curves for

an acceptance sampling plan plots the probability of accepting a
lot MTBF or the reliability of the products. An application of
the OC curve which is plotted for single, double and multiple
sampling plans is providing easy comparison ( ) versus a
parameter related to the lot such as MTTF or of plans. What
follows next is plotting the OC curve for single sampling plans.

�����������������������������������������������������������
������������������������������

In a single acceptance sampling plan, we take a sample of

size n from our lot and place all the n products on life test for a
period of time T as prescribed by the plan. If the reliability of
the products for the time T is R, the failure probability is p=1-R.
The probability that a large lot is accepted in a single sampling
plan of size n and acceptance number c is calculated from:

�
337 Reliabilty Engineering

= ��� ≤ ) = ∑ (1 − ) (6-16)

This equation, which gives the probability of the failure of at

most c products during the test, could be used to to plot OC

curve (Pa versus R).

Example 6-11(plotted versus R) OC curve:Pa

A plan with n=10, acceptance number c=2 and T=100 is used

to accept or reject a large lot.
a)Plot the OC curve of the plan in such a way that Pa be
plotted versus the product Reliability.
b) If the lifetime of the products are exponentially distributed
with mean 950 hours. Find the probability of accepting a large
lot of this product.
Solution
a) The following table shows the values of Pa for 11
values of R calculated using Eq. 6-16 or MATLAB
command binocdf.
R 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Pa 0 0 0.001 0.0016 0.012 0.055 0.17 0.39 0.68 0.93 1

and the following MATLAB commands plot the OC curve

(see Fig. 6.1) R=0:0.1:1; Pa= binocdf(2,10,1-R); plot(R,Pa).
Chap 6 Estimat
ation of Lifetime &Reliability Experiments &Tests 338

Fig. 6.1 OC curve for plan n=10, c=2

b)the lifetim
me is exponentially distributed then for T=100
0⟹

� � ���From the curve or the table Pa=0.93.

�����������
��������������������������������������������
������
������������
���������������������

The folllowing example illustrates how to plot the OC

C curve
an acceptancce sampling plan for inspecting a lot. If thee sample
meets a speccial criterion, the lot will be accepted.

Example 6-12
6 (OC curve:Pa plotted versus mean lifetime)

(based on Gran
ant&Leevenworth,1988 page 585)
Consideer the following acceptance sampling plan
n to be
taken from a largish lot and plot its OC curves using various
MTTF. The plan is as follows:

�
339 Reliabilty Engineering

Take a random sample of 22 units from the lot of product,

and apply life test. Whenever an item fails, replace it with
another item selected at random from the lot. If the test
continues for 500 hr with no more than 2 failures, accept the lot.
If 3 failures [ or more] occur before 500 hr of testing. Terminate
the test and reject the lot.
Solution
We suppose simultaneous testing of 22 units for 500 hours
or 110 units for 100 hours or 11000 units for 1 hour give the
same results. To plot the OC curve, note that totally we have 22 �

500=11000 item-hours with acceptance number c= 2. Assume

the failure probability is the same for all 11000 unit-hours.
Define the problem as a single acceptance sampling plan with
n=11000,and c=2,let:
X� = number failures in 11000 unit-hours during 1 hour test,
then Pa  Pr( X  2 given The probability of failure of one unit in an hour,
or
the failure rate of one unit per hour or the failure probability of one unit-hour
or
the proportion of binomial distribution p')

 Pr( X  2 | given mean of number of failures in a sample of 11000  11000  np ' )

Here �has a value less than 1 and is interpreted as the
probability of the failure of one unit in an hour.
The failure rate takes the place of fraction nonconformities
(Grant & Leavenworth,1988,page 586). Then the probability
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 340

that a unit fails in an hour is � To plot the OC curve, the exact

value of lot acceptance probability corresponding to a particular
could be calculated from: Pa=binocdf(2,n=11000,p'=λ).
In MATLAB, the approximate probability for various may
be calculated from Pa≅poisscdf(2,np').
As stated earlier, in chapter 10, Pa were plotted versus p'.
But in this chapter the horizontal axis is either the product mean
life ( θ � λ � � � or the product reliability. Table 6-3 shows the

probabilities for some values of θ. Figure 6-2 shows the

corresponding OC curve.
Table 6-3 Acceptance probability of in Example 14.15 for
various mean lifetimes (Grant Leavenworth,1988,Page586)

Calculation of OC curve for acceptance sampling plan requiring 11,000 item

hours of life testing with an acceptance number of 2. Calculation assumes
that the failure rate  is independent of the age of the item tested

Failure rate Mean life Expected average number of Probability of acceptance

per hour, failures in 11,000 test hours (probability of 2 or less

� hours�
λ� � ( np '  11000 ) failures) from Pois. Dist.
0.00002 50000 0.22 0.999
0.00005 20000 0.55 0.982
0.00006� 16667 0.66 0.971
0.00008 12500 0.88 0.939�
0.00010� 10000 1.1 0.900
0.000125 8000 1.375 0.839
0.00015 6667 1.65 0.770�
0.00020� 5000 2.2 0.623
0.00025 4000 2.75 0.480�
0.00030 3333� 3.3 0.360
0.00040 2500 4.4 0.185
0.00050 2000 5.5 0.088
0.00060 1667 6.6 0.040
0.00080 1250 8.8 0.007

�
341 Reliabilty Engiineering

To know
w how Pa is calculated , suppose   0.0003 ; since
n=11000 theen np '  11000  3.3 . The approximate value
va for
Pa from Poiisson CDF table : Pa  Pr( X  2)  0.359  0.360 � .
The exact vaalue of Pa is calculable from MATLAB :
= ����≤ 2) = ���������������� �.����� = 0.���
���
The followin
ng MATLAB commands plots the OC curve.
p=1/17000:.0
00001:1/1000;Pa=binocdf(2,11000,p);plot(1../p,Pa)

�
Fig. 6-2 OC curve for Example 14.15
(acceptance probabilityversus mean life).

End of exam
mple

���������������������������������������������
����
�������
�����������

In life tessting, situations frequently arise where it is im

mportant
to determin
ne if a new system meets a design goall or an
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 342

established standard. This leads to the area of statistical

inference called hypothesis testing(K&L page 263). Chapter 1
introduced Bartlett's goodness of fit test for the assessment of
the hypothesis that the distributional form was exponential. Here
2 tests on the mean and the minimum lifetime are presented.

6-8-1 Test of hypothesis on minimum life of an

exponentially distributed lifetime(K&L page263)

To deal with the following hypotheses on minimum life(  )

of an exponentially distributed lifetime, H 0 :   0 H1 :   0
given significance level(Type I error probability ) of ,
Take a random sample of size n, and
Place all of the n products simultaneously on life test, without
replacement,
Continue the test until the time that rth failure occurs( ≤ is a
predetermined number).
Prepare an ordered sample of the failure times: x (1), x (2),�� x (r) .�
Calculate the mean lifetime from
r
(n  r )(x ( r )  x (1) )   (x ( i )  x (1) )
n  x (1)
� '  i 1
;calculate F0  ,
r 1 � '
Using Table A at the end of the book or MATLAB command
finv (�   , 2,2r-2) , find F ,2,2 r  2 , the critical value of F

distribution for the given .

�
343 Reliabilty Engineering

n  x(1)
Ho is rejected if F0   F ,2,2r 2 (6 17) �����
� '

Note that if �Ho is rejected, the mean lifetime in this case is

estimated by � ' . The minimum life is estimated from

t �
� � ' 
  x(1)  and the relabity is estimared from e � ' (K&L page258-9).
n

��Example 6-13 (K&L page 263)

The data in the following table represents cycles to failure

for throttle return springs. Twenty springs were tested under
conditions similar to those encountered in actual use. The test
was truncated at the time of the tenth failure. Can we conclude
with 95% confidence that the minimum life(  ) is greater than
zero?
failure
1 2 3 4 5 6 7 8 9 10
no. (i)

Cycles to
190437

245593

277761

432298

530100

626300

1043307

1055528

1221393

2099199

failure
x(i)�

Solution
H0 :   0

H1 :   0
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 344

10
(n-r)(x (r) -x (1) )+  (x (i) -x (1) )
 (15-10)(2099199-190437)
θ� i=1
� �
r-1 10-1
=1908762
(190437-190437)+(245595-190437)+....+( 2099199-190437+ ) 
 θ��������
10-1
nx (1) 15�������
F0 = = =1.03
�
θ� 2767421

The critical value of F is not in Table A, Using MATLAB:

F������ =F��������� =finv(0.95,2,18)=3.5546 F� <F0.05,2,18

H0 is not rejected i.e. it is not concluded minimum life(  ) is

other than zero. End of example �

6-8-2 Test of hypothesis on mean lifetime and failure

rate concerning exponential distribution(K&L page263)

To test the following statistical test of hypothesis on mean

lifetime( ) of an exponentially distributed product with
significance level of

H 0 :   0
H1 :   0

Perform the following steps:

�
345 Reliabilty Engineering

1.Take a random sample of size n, and

2.Place all of the n products simultaneously on life test, without
replacement
3.Continue the test until the time that rth failure occurs( ≤ is
a predetermined number).
4.Prepare the ordered sample of the failure times: x (1), x (2),��x (r).�

5.Estimate the mean lifetime from   ( n  r ) x ( r )   x (i )


r

2r 
6.Calculate  0 
2
,
0
7.Using Table E at the end of the book or
MATLAB command���������α, 2r), find  21 ,2r i.e. the critical

value of chi-squared distribution for the given .

8. Reject Ho if


2r
������������������  
2
  2,2r �������������������������������(6-18���������������
0
0

The criteria of the above test is applicable for performing the

test on the failure rate(  ) of exponentially distributed products
i.e. H 0 :   0 H 1 :   0 .
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 346

Example 6-14

A random sample of certain exponentially-distributed-life

product was placed on a life test without replacement. The test
was terminated when the tenth failure occurred. Other data are ;

n  15,  x(i) =18217915,x(10)  2099199 .

Perform the following test of hypothesis with   %5 on the
mean lifetime( ):
H 0 :θ  θ 0 ��� 6 ������
H1 :θ�θ 0

Solution

2r
 
2
r  10, n  15,   5%
0
0
The test is of non-replacement failure-terminated type, therefore
according to Eq. 6.6

 (n  r )x ( r )   x i (2)(10)(2871391)
  2871391,  02   57
r 106

Table E or MATLAB command ��������0.������ yields:

%5,20
2
 31.41 . 02  %5,20
2
 H � :   θ0 is strongly rejected, &

H1 :θ�θ0 is not rejected �

�
347 Reliabilty Engineering

6-8-3 Comparison of two designs

An engineer would like to compare 2 designs of a product.

From Design 1 a random sample of size n1 and from Design 2 a
random sample of size n2 is taken and placed on life test. The
tests are terminated when �failure occurs in the sample of
Design 1 and �failure occurs in the sample of Design 2.
≤ and ≤ are predetermined truncation points.

Suppose the failure times are � � � �, ( ), … , ( )

,

S = ( ), ( ), … , ( )
. For simplicity � will be

assigned such that � ) ≤ ( ). If the lifetimes are

exponentially distributed, to compare their mean lifetimes
1 ,2 i.e. performing the following test:
H 0 : 1  2
1  2
H1 : 
1  2
The statistic under the null hypothesis is (K&L p265):
� × �������������������������������� (6� 19)��������

where

n1 Sample size taken from Design 1

n2 Sample size taken from Design 2
r1 Truncation point of the test of Sample 1 ≤
Terminate the lifetest when �failure occurs
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 348

r2 Truncation point of the test of Sample 1 ≤

Terminate the lifetest when �failure occurs
�� �
, x
x
1(r ) 2(r ) Occurrence time of r1th & r2 th failure in Design 1 & 2 respectively
1 2

x ,x
2(1) 1(1)
The occurrence time of first failure in Design 1 & 2 respectively

x 1( j ) , x 2 ( j ) Elements of the samples

F ,2 r2  2,2 r1  2 α -significance-level critical value of an F distributions

2
2
with � − �,�2 − 2 deg of free.
Obtainable from Table A or MATLAB command
(1 − , 2 − 2, 2 − 2)

C r2
( n2  r2 )( x2( r2 )  x2(1) )   ( x2( j )  x2(1) )
j 1
C r1
( n1  r1 )( x1( r1 )  x1(1) )   ( x1( j )  x1(1) )
i 1

Rejection Criteria:
Reject � ��� is outside
For H1 :1 2
 F α F ,2 r2  2,2 r1  2 
 1 2 ,2r2 -2,2r1 -2 2  �

For H1 :1 2 Reject � �� F0 >F ,2r1 -2,2r2 -2

To test if the failure rates of the 2 designs are significantly

H 0 : 1  2
different or not i.e. , the above test could be used.
H 1 : 1  2

It is our duty to act in such a manner

that can be universalized i.e.
we would want everyone else to act
in a similar manner
(Kant)

�
349 Reliabilty Engineering

�����������

In the problems, if the type of distribution is needed and not specified,

exponential distribution is assumed
1. The weather radar system on a commercial aircraft has
an MTBF of 1,140 hours. Assume an exponential time to
failure distribution and answer the following questions:
(a) What is the probability of failure in a 4-hour flight?
(b) What is the maximum length of flight such that the
reliability will not be less than 0.99? (Assume that the
system is in continuous operation during flight.)
2. The MTBF of a kind of tank is 810 kilometers.
Assuming an exponential distribution:
a) What is the maximum mission length such that there
will be a 0.98 chance of the tank returning?
b) What is the probability of the tank returning from a
160 kilometer mission?
c) How many tanks should be sent out
on the 160 kilometer mission to obtain a probability of
0.99 .that at least five tanks will arrive at the target area
(assume 80 kilometers to target).
3. Ten engines of a new design were each driven the
equivalent of 50,000 kilometers. Odometer readings
were recorded whenever an unscheduled maintenance

�����������������������������������������������������������
1
Exercises 1 through 8 are from Chap. 9 K&L page 269 problems
1,2,3,4,6,8,9,10
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 350

action occurred. The odometer readings for each vehicle follow:

Motor
Odometer readings
no.
1 220 11970 21397 27766
2 45270 48836
3 25695 25989 30980 32769 47459
4 4200 14672 21831 29187 31964 36535 44094
5 3900 29147 31613 37524 43601 45208
6 1275 21183 23649 33348 40907
7 3730 6300 11840
8 22565 22710 28301 31628 45784 47213
9 12759 14548 19539 41108 44550
10 12212 18727 41854 42169 47996

a) Can the data be represented by the exponential distribution? ( 0.1)

b) If answer to part (a) is yes, estimate the MTBF.

4. The following data represent kilometers to failure:

43000 27200 10600 12400

27000 4100 200000 18200
68000 40500 109000 14200
46000 2600 2400 24500

a) Assess the feasibility of using the exponential

distribution to model this
situation. Assuming that the exponential is applicable:
b) estimate the MTBF;
c) set a 90% lower confidence limit on the 10%failure
kilometer.
d) With 90% confidence, quote the 2,400 km reliability.

�
351 Reliabilty Engineering

5. An automobile was driven over a 120,000 kilometer test

course. The following represent odometer readings at
which a particular type of failure
4123 27720 63582
4497 28496 66057
10506 40887 100763
12317 48323

Assuming an exponential distribution as representative, is there

any evidence that the failure rate in the first 40,000 km is different
than in the last 80,000 km? ( α � � �� ).
6. For a test vehicle, major electrical failures occurred at
the following kilometers
63 17393 23128
114 18707 24145
14820 19179 33832
16105 22642 34345

The vehicle was driven a total of 36,000 kilometers.

(a) Estimate the MTBF.
(b)Determine the 90% two-sided confidence interval for
the MTBF.
(c) Estimate the reliability function.
(d) Determine the 95% lower confidence limit for the
1,200 kilometer reliability.
(e) With 90% confidence estimate the kilometer at which
10% of the population will fail.
7. In 600,000 test kilometers accumulated on 6 vehicles, a
total of 69 failure occurred. Assuming an exponential
failure distribution:
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 352

a) Estimate the MTB

b) Find the 90% lower confidence limit on the MTBF.
c) Find the 90% lower confidence limit on the reliability function.
8. A transmission valve operated for 9,276 cycles before
the test was discontinued. The test was stopped because
an oil pump failure caused the valve to burn out.
a) Set a 90% lower confidence limit on the MTBF.
b) A second valve of a different design failed at 19,460 cycles.
Management would like your recommendation as to which
valve is best. What would be you answer?
9. (K&L��pp 239-240�) A device was placed on 245-hr lifetest had
20 failures occurred on the following times:
Failure times during 245-hr lifetest
21.2 74.7 108.6 157.4
49.9 76.8 112.9 164.7
59.2 84.3 127.0 196.8
62 91 143.9 214.4
74.6 93.3 151.6 218.9

Ignoring the repair times, could we say that Time Between

Failure(TBF) follows an exponential distribution?
Hint: At first compute the TBFs which are the following values
Time between failures
21.2 0.1 15.3 5.8

47.9-21.2=26.7 2.1 4.3 7.3

11.3 7.5 14.1 32.1

2.8 6.7 16.9 17.6

12.6 3.2 7.7 4.5

�
353 Reliabilty Engineering

1
10. Plot the OC curve (Pa versus MTBF) for the following
sampling plan:
Twenty units are randomly taken from a largish sample, and
simultaneously placed on lifetest. Whenever an item fails, it
is replaced with another item selected at random from the
lot. If the test continues for 500 hr with not more than 2
failures accept the lot. If 3 failures occur before the 500 hr of
testing, reject the lot and terminate the test.

�����������������������������������������������������������
1
Solution on Page 586 Grant &Leavnworth(1988)
Chap 7 Dynamic Models+Availability, Markov Chain 354

Chapter 7
Dynamic
Models,
Availability,
Application of
Markov Chain to
Reliability

�
355 Reliability Engineering

7
����������������������������������������������
�������������

Aims of the chapter

This chapter deals with time related reliability models or

dynamic models. Series systems and two but types of parallel
redundancy, namely active redundancy and standby redundancy
are introduced. Some system attributes such as maintainability,
serviceability as well as availability are defined. The chapter
also points out the application of Markov chains to reliability
analysis.

7.1 Dynamic Models in Reliability

Dynamic models (time dependent) are important, realistic,
and more appropriate than static models which were covered in
chapter 2. Incorporating time into static models , this chapter
deals with series, weakest link, active parallel, standby
parallel(perfect switching and imperfect switching).

7-1-1 Series Systems

Series systems are those in which all components are

required to be in a state of functioning for the system to be
Chap 7 Dynamic Models+Availability, Markov Chain 356

functioning. If a system is assembled of m components in series,

its reliability( ) is:

( )=∏ ( ) (7-1)

where ( ), = 1, … , are the reliabilities of the

components.

Proof No.1
' m m

h sys (t)=
-R sys (t)
R sys (t )   R i (t )  [1  Fi (t )] �
R sys (t) t 1 t 1

 m m
 m
 m

R (t)= f i (t ) R j (t )   f i (t ) R j (t ) 
'   
sys 
i 1 
  
j 1  i 1  j 1 
 j i   j i 

Therefore

 m m

 f (t ) R (t ) 
  i 
i 1 
j 
'
R sys (t) j 1 
h sys (t)=    j  i 
m �
R sys (t)
 R i (t )
t 1

f 1 (t ) f m (t ) m
f i (t ) m
hsys (t)= +...+ =   hi (t )
R1 (t ) R m (t ) i 1 R i (t ) i 1

�
357 Reliability Engineering

Proof No.2



 h ( ) d  
d
R (t )  e 0
  h ( )d    ln R (t ) ,
0
h (t )  
dt
ln R (t )

R sys (t )  R 1 (t )  ...  R n (t )  ln R sys (t )   ln R i (t ) 

d d m
ln R sys (t )   ln R i (t )   hi (t )
dt dt i 1
d m
hsys (t )  ln R sys (t )      h i (t )   hsys (t )   hi (t )
dt i 1

End of proof

Now Suppose the failure rate functions of m independent

components of a system are:�� (t) = λ � � � , i = 1,2, . . . , m
where������ ٬ ‫ ﻭ‬k constant values.
Then using Eq. 1-14-1

∫ ( )
� )= =

And using Eq. 7-1

( ) = ��� − ∑ − ∑ (7-2-1)

Let ∗ = ∑ , ∗
=∑ , = ∑ = ∗

then (K&L p39)

∗ ( ∗ )
( ∗ ∗× × )
( )=e ∗
.

For large m it could be shown that :

∑ ∗
��� →∞ ( )= = = (7-2-2)
Chap 7 Dynamic Models+Availability, Markov Chain 358

That is the time between failure of a series system, with largish

number of components with failure rate function

ℎ ��� = + , is exponentially distributed with parameter

∑ .

Note that if ℎ �t) = Eq. 7-2-2 holds regardless of the value

of m. By substituting � � in Eq. 7-2-1 this fact could easily
be verified.

Example 7-1

The failure rates of an n-component series system are

constant values λ i�i ����n . The components are independent.

Find the failure rate function, the reliability function and MTBF
of the system.
Solution
The failure rates of the components are constant i.e. the
lifetime distributions are exponential. The lifetime of a series
configuration equals the lifetime of the component that has the
minimum life among the components. On the other hand the
minimum if some exponentially distributed random variables
has an exponential distribution. Therefore the distribution of
this system is exponential with the following functions:
According to Eq. 7-2 ,the system failure rate function ( h sys ) is

n n
h sys (t)=  h i (t)=  λ i � ��������
i=1 i=1

�
359 Reliability Engineering

Then the reliability function is:

-t   λ
������������������������ R sys (t)=e i ��������������������(7-4-1)

The mean lifetime of this series system is

1
1 1  n 1
M T BF      (7-4-2)
 i=1  i 
n n
1
λ
i=1
i 
i=1 M T BFi

1
where MTBFi   i  . End of Example
λi

Mean lifetime and reliability function of a series

system of identical components with failure rate

Consider a series system having n independent components

whose lifetimes are exponentially distributed with mean
1
MTBFpart  , then according to Eq. 7=4-1&2 mean time

between failures of the system and the reliability function the
system are:

-n t
R sys (t)=e (7-4-3)

1 MTBFpart
MTBF   ��������������������� (7-4-4)
n n

7-1-2 Series Chain Model or Weakest Link Model

A chain-model system works like a chain. A chain is not

stronger than its weakest link.
Chap 7 Dynamic Models+Availability, Markov Chain 360

The series chain model is a series system in that if anyone

component fails the system will fail; however, the concept of
how a component fails is different. As an example of this
concept of failure, consider a circuit composed of n identical
components, and this circuit is subjected to thermal stresses. Let
us assume for simplicity that the thermal stresses are the main
cause of failure. In this situation the one component having the
least resistance to the thermal stresses will be the first to fail.
Then, in this case, the system reliability will be(K&L page 214):

= ( ,1, … , ) ( 7-5) �
where Rj is the reliability of the ith component and describes the
component's resistance to failure from thermal stresses.

Calculation of a component reliability (Rj)

If the strength of a component and the stress applied to the

system are random variables denoted by  ,S the chain will
break if the applied stress exceeds the strength of anyone link.
Hence to compute the component reliability, their joint density
function should be integrated over   s (see Fig 7.1):

�� Ri  Pr(  s )   f ( , s )dds ������7-6)�

  s ‫ﻭ‬s

With the assumption that  and s are independent:

�
361 Reliability Engineering


  
Ri  s  0 s  s f  ( )d   ds 
f ( s )  f s (s ) 1  F (s ) ds � (7-7��
0

Fig 7-1 The region where the strength is more than stress

Let C be the random variable representing the strength of the

n-link chain. This will be:

n
 C  min  i
i 0

where i is the strength of the ith link.

According to Eq. 1-42 in Sec. 1-12 on the minimum of a
random sample of size n from a continuous distribution �����, if
X(1) �denotes�the minimum we have:

n
FX (1) ( y )  1  1  FX ( y )  (7-8)
Chap 7 Dynamic Models+Availability, Markov Chain 362

In the above chain model if the cumulative distribution function

(CDF) of the strength of each the n links is F (y ) , then the CDF

of the chain strength, G ( y ) , equals:

G ( y )  F (1) ( y )  1  1  F ( y ) �
n

Where

F (y ) is the CDF of each link of the chain

� F (1) ( y ) is the CDF of the strength of the weakest link.

On the other hand according to the concept of Eq. 7-7 i.e.


R i   f s (s ) 1  F (s )ds and the equation
0


1-FC (y)=1- 1-1-Fδ (y)
n
 we could conclude that the reliability
of the n-link chain which equals ������  ��� is obtainable from:


RC   f S (s ) 1  F (s ) ds
n

0 (7-9)

Where
RC The reliability of an n-link chain

F (y ) The CDF of the strength of an individual component.

f S (s ) The pdf of the stress acting on the system

Note the similarity of Eq. 7-7 with Eq. 7-9 for n=1.

�
363 Reliability Engineering

Example 7-2

A system is exposed to heat stress whose value is

exponentially distribute with mean 500 ��. The system is a 10-
link series chain. The strength of each link is exponentially
distributed with mean 600 ��. Find the reliability of the chain.

Solution

The probability density function(pdf) of the strength of each

link:

t  t
1  600
f  part (t )  e , F part (t )  1  e 600 ,
600

s
1  500
The pdf of the stress applied to the system is f  e
s 500
According to Eq. 7-9:

 n
R sys   f (s ) 1  F part (s ) ds n  10
0 s

10
1  500   600 
s s

R sys   e e  ds 
0 500  
56 S
1   3000 1 3000 6
R sys 
500 s  0
e ds  
500 56

56

End of Example
Chap 7 Dynamic Models+Availability, Markov Chain 364

It is worth mentioning that in this chapter the subsystems

(components ) of a system are assumed to be independent of
each other unless specified something else.

7-1-3 Parallel systems

A parallel system is one that will fail only if all of its

subsystems fail. Three models related to this kind is studied
here: active parallel redundant, standby parallel redundant and
shared load parallel.
Definition of Redundancy
In reliability engineering, redundancy may be defined as the
duplication of the components of a system with the intention of
increasing reliability of the system and an alternative to failing
condition. Two types of commonly applied redundancy are
active redundancy and standby redundancy.
Active((Hot) Redundancy Definition
Active redundancy does not require the external components
or devices to perform the function of detection, decision and
switching when an element or path in the redundant structure
fails(Based on Li,2016). The redundant elements are always in
operation to share the load of the system. This redundancy is
also called hot redundancy(Shooma,2002 page 336)
Standby Redundancy Definition
Standby redundancy is defined as the redundancy that
requires the external elements or devices to detect, make a
decision and switch to another element or path as a replacement

�
365 Reliability Engineering

for a failed element or path(Li,2016). In this type extra units are

not brought into use until the main unit fails. It is the primary
consideration in determining whether cold, warm or hot standby
is to be used(Lewis, 1994, p263).
Definition of cold & warm standby redundancy1:
In cold standby the secondary unit is under no stress.
In warm standby the secondary unit is under a stress less that
of the main unit and more than that of cold standby.
It is worth mentioning that if nothing is said about the
coldness or warmness of the standby components in this book,
they are assumed cold.

�����������������������������������������

In a parallel system with active redundant, all subsystems are

working and if the components are independent, its reliability is
derived from:

� Rsys (t )  1 1 R1(t )1 R2 (t )...1 Rn (t ) (7-10)

�

Proof

Let � denote the lifetime of the system, and , i=1,..n

denote the lifetime of the subsystems. Then in an active
redundant system:
�����������������������������������������������������������
�
���������������������������������������������������������������
������������������������������������������
Chap 7 Dynamic Models+Availability, Markov Chain 366

( ≤ )= [ ( ,…, )≤ ]= ( ≤ ,…, ≤ )

Assuming , i=1,..n are independent:

( ≤ ,…, ≤ )= ( ≤ )… ( ≤ )�
( ≤ )= ( ≤ )… ( ≤ )
⇒
�− ( ) = [1 − ( )][1 − ( )] … [1 − ( )] ⇒

( ) = 1 − [1 − ( )][1 − ( )] … [1 − ( )]

��������������������������������������������������������������������
���������������������������������

In an active parallel system whose components failure rates

 , i  1,..., n are constant(or the lifetimes are exponentially
i
distributed), the reliability of each component is calculated from
Ri (t )  eti , t  0 , and the system reliability is calculated from

n
R sys (t )  1   (1  e  t i ) t  0 ������������������(7-11)�
i 1

The Proof is similar to that of Eq. 3-2.

Note that Eq. 7.11 implicitly shows that the lifetime of active
systems in not exponentially distributed.
Two- component active parallel system
For n=2 from Eq. 7-10 we could write:

�
 367 Reliability Engineering

( ) = 1 − (1 − ( ))(1 − ( )) ��

Rsys (t )  R1(t )  R2 (t )  R1(t )R2 (t ) �

If the failure rate of the components are constants  ,  then from

1 2
Eq. 7-11:

( )=1− 1− 1− =
 1t
� e  e 2t  et ( 1 2 ) t 0 ��������������

If  ,  are equal (λ1 �λ2  � then:

1 2

R sys (t)=2e -λ� -e -�λ� =1-(1-e -λ� ) � (7-12-2)

The system mean life is:

 1 1 1
(MTBF ) sys   Rsys (t )dt    (7-13)�
0 1 2 1  2

Three-component active parallel system

For n=3 from from Eq. 7-10

( ) = 1 − (1 − ( ))(1 − ( ))(1 − ( ))

If the failure rate of the components are constants  ,  , 

1 2 3
then
Chap 7 Dynamic Models+Availability, Markov Chain 368

( ) = 1− 1− 1− 1− (7 − 14 − 1)

and the system mean life is:


MTBFSys   R sys (t )dt 
�

1 1 1 1 1 1 1 ��� − 14 − 2)
+ + - - - +
λ1 λ 2 λ 3 λ1 �λ 2 λ1 �λ 3 λ 2 �λ 3 λ1 �λ 2 �λ 3

Mean lifetime and reliability for active parallel systems

having components with constant failure rate 
If all n components of an active parallel system are
independent and their lifetimes are exponentially distributed
with parameter  and mean lifetime  then(based on
Garosh,1989 page135):

n
1
���� MTBFSysActive    (7-15)�
k 1 k

n
1
Table 7-1 gives the values of k
k 1
 Mn .

Eq. 7-15 could be verified for n=2 from Eq. 7-13:

1 1 1 3 3
n=2 MTBFsys = + - = = θ
λ λ �λ �λ �
and for n=3 from Eq. � − 14 − 2:

3 1 1 11 11
n=3 MTBFsys = -3� � � � θ .
λ �λ �λ �λ �

�
369 Reliability Engineering

The reliability of the above n-components active parallel system

is:

R s y s ( t ) = 1 -( 1 -e -λ � ) � ����������������(7-16)�

Table 7-1Values of �∑ for

n=1,...,32 (Grosh,1989 p136)
n� M � n� �
1 1 17 3.4295
2 1.5 18 3.4951
3 1.8333 19 3.5477
4 2.0833 20 3.5877
5 2.2833 21 3.6453
6 2.4500 22 3.7343
7 2.5928 23 3.7343
8 2.7178 24 3.7759
9 2.9289 25 3.8159
10 2.9289 26 3.8544
11 3.0198 27 3.8914
12 3.1032 28 3.9271
13 3.1801 29 3.9616
14 3.2515 30 3.9950
15 3.3182 31 4.0272
16 3.3807 32 4.0585

It is seen from Table 7-1 that "for a system to have double the
mean life of a single component, it must consist of 4
components. To triple the mean life the system must have 11
components. Theoretically, there is no limit to how much the
system mean life can be extended but the cost of extending life
through mere redundancy is usually prohibitive. Redesign
should be performed to excessive redundancy"(Grosh,1989 p 135).
Given the reliability function of an active parallel system,
R active (t ) , the failure rate function is obtained from:
Chap 7 Dynamic Models+Availability, Markov Chain 370

d
Ractive (t )
�� hactive (t )  
dt ����������(7-17)�
Ractive (t )

������������������������������������������������

Consider the standby parallel system shown in Fig 7.2. In the

n-unit system only one unit works and when it fails the one of
the n-1 standby units replaces it by the help of a switch. Upon
failure, this active unit is replaced with another standby unit.
The process continues until no more standby redundant units is
available.

1�
�

� 2

� Switch
n

Fig. 7,2 A standby parallel system

The switch could be an operator or a device such as a
hydraulic valve or electric relay or a contractor. Let the
probability of successful operation of the switch for replacing
the unit be denoted by Ps. If Ps is 1 the case is called perfect
switching; if less than one, the case is called imperfect
switching.

Case 1: Perfect switching

�
371 Reliability Engineering

"Perfect switching means perfectness of a detection and

switching mechanism used to detect failure of a component and
to activate a redundant component"1. In other word
Perfect switching is a situation where the reliability of the
switch when needed to perform its function is 100% (Ps=1) i.e.
no failure is assumed for the switch when needed to perform its
task. Below the reliability of systems having one active unit and
n-1 standby redundant (backup) units are analyzed.

Perfect switch :Two-component parallel system

with an active unit and a standby redundant unit

Consider a system with one active subsystem(unit) and one

standby (backup) unit which replaces the active upon failure by
the help of a switch of 100% reliability (Ps=1). The reliability of
this system is(K&L page 219):

2 t d 
R (t)=R 1 (t)- R 2 (t-t )  R 1 (t ) dt =
standby 0
 dt   (7-18)�
t
=R1 (t)+  R 2 (t-t )f1 (t )dt 
0

Where

2 Reliability function of a system having one

R (t)
standby active unit and n-1 standby redundant units.
�����������������������������������������������������������
1
Pradip Kundu & Asok Nanda� Redundancy Allocation in a System: A Brief Review
Chap 7 Dynamic Models+Availability, Markov Chain 372

R1(t) Reliability function of the main unit

R2(t) Reliability function of the redundant unit

f1 (t) Pdf of the main unit

For proof, interested readers could refer to Lewis(1994) page

255 or K&L page 219.

Example 7-3-1

A system has one active unit with failure rate � and

one standby redundant unit with failure rate � . When the

active unit fails a perfect switch replaces it with the other unit.
Calculate the reliability of the system.

Solution

The constant failure implies that the life time is exponentially

distributed ; then R 1 (t)=e -λ � , R 2 (t)=e �λ � . According to Eq. 7-18:
1 2

2 t d 
R (t)=R1 (t)- R 2 (t-t )  R 1 (t ) dt =
standby 0
 dt  
t   (t-t )  d ' 
e  1t   e 2  e  1t dt ' 
0
 dt  
t   (t-t ) t
=e  1t   e 2 1e  1t ' dt '  e  1t  1e  2t  e( 2 1 )t 'dt ' 
0 0
2
1
R (t) = e-λ1� + λ1��λ2 � ���λ 2 �λ1 �� ���
standby λ 2 � λ1

�
373 Reliability Engineering

1  251 2  101
2
R (20)=exp(-20/10)+(1/10)*exp(-20/25)*(exp(20/(1/25-1/10))-1)/(1/25-1/10)  0.8842
standby

End of Example

Perfect Switch : Two-component parallel system

with identical active and standby units(failure rate= )

Assume that the main and standby units are identical, each
with a constant failure rate and the switch is perfect. Then:
2
R Sys (t )  e  t (1  t ) t 0 (7-19)�

Proof
Since R1(t)= R2(t)= e   t then according to Eq. 7-18:

2 t d 
R (t)=R1 (t)-  R 2 (t-t )  R 1 (t ) dt =
standby 0
 dt  
t
e   t   e   (t t ') -e   t '  dt =e   t  te   t  e   t (1  t )
0

End of Proof .
Example 7-3-2
Calculate and compare the reliability of two configurations
of a two-unit system (active parallel and standby parallel). The
failure rate of the unit is 5%.

Solution
Chap 7 Dynamic Models+Availability, Markov Chain 374

Since the components failure rate is constant, the lifetime is

exponentially distributed. The reliability for standby
configuration is calculated from Eq.7-19.

2
R (t)  et (1t) t  0
standby

The reliability for active configuration is calculated from

Eq.7-19.

2
R sys (t )  1   (1  e  t i ) t  0 1  2  
i 1

Then for t  10 ,   100

5
�

2 2
R  91%, R  84.5%
standby active

2 2
for t  100 ,   100
5
,R  4%, R  1.3% �
standby active

Perfect Switch: Three-unit standby system (1 active

&2 standby)
For the reliability of a three-unit standby system the
following relationship holds (K&L p 220)
3 2 t t-t1
R (t)= R (t)+ f1(t1) f2 (t2 )R3(t-t1-t2 )dt2dt1 (7-20)
standby standby 0 0

where

�
 375 Reliability Engineering

3
�� R The reliability of 3-unit standby system
standby

2
�� R The reliability of 2-unit standby system
standby

The density function of active unit

� The density function of standby unit No. 1
� (t)� The density function of standby unit No. 2
If the 3 units are identical and their lifetime is exponentially
distributed with parameter λ then:

  t  
2
3
 t
R (t )  e 1  t   t 0 (7-21)
sys
 2 

Example 7-4
The reliability of the water supply system of a city is a
concern of the city council. The council would like to ensure
the inhabitants that the system will work for 20 years with a
reliability of 95%. At the time being the water is supplied by a
reservoir and a river ( with mean lifetime of 25 and 10 years
respectively) in parallel. The water is then disinfected in a
building that has an active unit for disinfection and 2 standby
backup units. Each of the disinfection units is designed for a
useful life of 25 years. After disinfection the water goes to
distribution subsystem, which is 99% reliable. The council is to
decide whether to allow the municipality to add a deep well to
Chap 7 Dynamic Models+Availability, Markov Chain 376

the water supply subsystem or not? What do you suggest?.

Assume all life times are exponentially distributed.

Solution

The RBD of the system of the Proble3m is as follows:

Fig 7.3 The RBD of the wateter supply system of Example 7-4

Data: � �� , = , λ = �

= , = 0.�� ���

The reliability of the first subsystem (reservoir+ river) is

calculated from Eq.7-10, assuming the reservoir and the river
are both active:
20 20
���� = 1 − (1 − e 25 ��1 − e 10 �

= 1 − (1 − 0.�����(1 − 0.������ � 0.����

The 20-year reliability of the second subsystem

(disinfection unit) is calculated from Eq. 7-21:

�
377 Reliability Engineering

 t  
202
 2

R (t )  e t 1  t 
20 2
R (20)  e (1  25  25 )  0.932
20
 25
 2  2
The entire system reliability is currently:

= 0.���� ∗ ���� ∗ ��� � 0.����

Suppose the permission for a well of reliability of R' is issued by
the council; then the system reliability would be:
=[1-(1-.4493)(1-0.13532)(1-R')]*0.932*.99
It is evident the if even if R' has its greatest value i.e. 1 the
system will have a reliability of [1-(0)] *0.932*.99= 0.9227.
The council cannot issue the permission because is interested
in 95% reliability. End of Example
Perfect Switch: n-unit standby system (1active &n-1 standby)
Consider an n-component system has 1 active unit and n-1
standby units whose lifetimes are exponentially distributed with
parameter λ . If the switch is 100% perfect then(O'Connor 2003
page 167,K&L page 221):

 n 1 
  ( t )
i
n 
R (t )  e  t  i 0  �������������������(7-22)�
standby
 i! 
 

n
- R standby '(t)
h standby (t)= n
��������������������������(7-23)�
R standby (t)

�
Chap 7 Dynamic Models+Availability, Markov Chain 378

Comparison of Two 2-component parallel

configurations (active and standby)

Figure 7.4 shows "both the reliability and the failure rate for
[2-component] active and standby parallel systems, along with
the results for a system consisting of a single unit with the
assumption that the lifetimes are exponentially distributed. At
intermediate times the failure rate for the standby system is
smaller than for the active parallel system. This is reflected in a
larger reliability for the standby system(Lewis,1994 page 256).

Fig 7-4 Properties of two-unit parallel systems(Lewis, 1994 page 256):

a) instantaneous failure rate b) Reliability function

���������������������������������������������������
�������������������������������������������

In this section active and standby configurations of a parallel

system having similar subsystems are considered and their
mean lifetime and reliability functions are mentioned.

�
379 Reliability Engineering

2-&3-component active and perfect switch standby

systems

Failure rate function of 2-component systems:

a)The instantaneous failure rate function of an active parallel
whose two active components have exponentially-distributed
life is given by:

1 dR active (t )
hactive (t )    �����������
R active (t ) dt

 1  e  t 
hactive (t )     t 
(7-24)
 1  0.5e 

b)The instantaneous failure rate function of a parallel system

with 1 active and 1 standby component and a perfect switch
whose 2 components have exponentially-distributed lifetime is:
1 d
hstandby (t )   Rstand (t ) 
Rstand (t ) dt

 t 
hstandby (t )     (7-25)
 1  t 

Mean lifetime of 2-component system

The mean time to failure(MTTF) of a system could be

calculated from integrating its reliability function over(0 ∞):
Chap 7 Dynamic Models+Availability, Markov Chain 380


MTTF  �
R (t )dt . (7-26)

If the mean lifetime of a component( M T T F part ) is given.

The mean lifetime of an active parallel 2-component system is:

MTTFactive  3 MTTFpart (7-27-1)

2

And that of a standby parallel 2-component system is:

MTTFstamdby  2MTTFpart (7-27-2)

Mean lifetime of 3-component standby system

Given the lifetime of each component , M T T F part , in a 3-

component standby system whose switch is perfect i.e. PS =1

and has 1 active and 2 standby independent components, the
system mean lifetime is given by:

MTTFstamdby  3MTTFpart (7-27-3)�

Mean lifetime of n-component standby system-perfect switch

Given the lifetime of each component , M T T F part , in an n-

component standby system whose switch is perfect i.e. PS =1

and has 1 active and n-1 standby independent exponential-
lifetime components, the system mean lifetime is given by:

n
MTBFstandby = =n  MTBFpart (7-27-4)


�
381 Reliability Engineering

Where  is the parameter of the exponential distribution of the

lifetime of each component .

Proof

The lifetime(X) of a system having 1 active unit,n-1

redundant standby components and a perfect switch is the sum
of the lifetimes( i=1,..,n) of its components:
n
=∑ ⟶ E (X standby )  E ( X i )   E (X i )   MTBFi
i 1

n
E (X standby )  E ( X i )   E (X i )   MTBFi
i 1

If Xi is exponentially distributed with parameter i then:

1
MTBFsys   (7-27-5)�
i

n
If 1  ...  n   then MTBFstandby = =n  MTBFpart and the


proof is complete.

Example 7-5

A system has 1 active unit , 1 cold standby unit and a

perfect switch. The failure rate of both units is 1 failures per
1000 hours. Calculate 100-hr reliability of the system.
Solution
Chap 7 Dynamic Models+Availability, Markov Chain 382

t=100, �.���, according to Eq. 7-19

2
R (t)=e -λ� (1+λ�� ��������
standby

Notice that if both units were active the according to Eq. 7-10:
2
R (100)=1-(1-e -λ� ) � = 0.9909 .
active

Reliability function and lifetime pdf of n-

component standby system-Perfect switch

This section deals with the probability density function of the

life time and the reliability function of a system having 1 active
unit and n-1 cold standby units.

�������������������������������������������������������
���������������

A single unit is put in service. A perfect switch replaces it by

a cold backup unit as soon as a failure occurs. The density
function of the lifetime of this system is derived from the
following convolution(Grosh, 1989 p164):

t
f sys (t )   f 1 (z )f 2 (t  z )dz ����������������������(7-28).
z 0

where

�
383 Reliability Engineering

f sys The pdf of system lifetime(T)

f1 The pdf of the main device lifetime(T1)�
f2 The pdf of the backup lifetime(T2)�

����������������������������������������������������� � ��

If both devices have constant failure rates λ 1�λ2 and � 1, the

lifetime pdf of the system is as follows:

( )= ( − ) (7-29)

Proof(Grosh, 1989, page166)

t
f sys (t )   f 1 (z )f 2 (t  z )dz  �
z 0

( ) ( )( )
( )= =

( ) ⟹
= 1− = +

( )= ( − )

The reliability function in this case is(Grosh, 1989, p166):

�e   t �e   t
� �

R sys (t)   (7-30)

�  � �  �
Chap 7 Dynamic Models+Availability, Markov Chain 384

Proof:

According to Eq. 7-18:

Rsys (t)=R 1 (t)+  R 2 (t-t )  f1 (t ) dt =e  1t   e  2 (t t ') 1e  1t '  dt  
t t

0 0
 2 t  1t
1e e
Rsys (t)   2
1  2 2  1
Notice that for � � use Eq. 7-19 i.e. R sys (t )  e  t (1  t )

�������������������������������������������������������
���������������

A single unit is put in service. A perfect switch replaces it by

a cold backup unit as soon as a failure occurs. When the backup
fails, it is replaced by the other back. The density function of
the lifetime of this system is derived from (Grosh, 1989 p165):
t z (7-31)
f sys (t )    f 1 (w )f 2 (z w )f 3 (t  z )dwdz
z 0 w 0

The mean lifetime of this system and its reliability function is

derived from :
 (7-32)
Rsys (t )   f sys (x )dx
x t

  (7-33)
MTTFsys   xf sys (x )dx   Rsys (t )dt
x 0 0

where

f sys The pdf of the system lifetime(T)

�
385 Reliability Engineering

f1 The pdf of the main device lifetime(T1)�

f2 The pdf of the 1st backup lifetime(T2)�
f3 The pdf of the 2nd backup lifetime(T2)
In the above discussion it is assumed that the backup units
are cold i.e. they are under no stress when they are in the
standby mode.
In this three-components system if all units have the same
constant failure rate λ , the reliability function is calculated from
  λ�   .
2
3
Eq. 7-21 i.e. R (t)=e -λ� 1+λ�� 
sys
 2 

Reliability of n-component standby system :Perfect Switch

Consider a case where 1 unit is active, n-1 components are

cold and in standby mode and the switch is 100% reliable. If the

units have constant failure rates 1,..., n , the reliability

function of the system would be(Garosh,1989page169):
n n n j
R (t )   e  i t  (7-34)�
sys
i 1 j 1  j  i
j i

If 1  ...  n   then(Grosh,1989 p167)��

n (t ) n 1e   t
R (t )  e   t  te   t  ....  (7-35)�
sys ( n  1)!
Chap 7 Dynamic Models+Availability, Markov Chain 386

n n
1
Remember as proved earlier MTTFsys   MTTFi  
i 1 l 1 i
gives

the system's mean time to failure and for 1  ...  n  

n
MTTFsys  .


In the above discussions it is assumed that the backup units

are cold i.e. they are under no stress when they are in the
standby mode.

Case 2: Imperfect switching

In this case the switch has a reliability of less than 1; i.e.

failure of the detection and switching mechanism is probable
and therefore the standby unit cannot replace the failed unit. Let
Ps denotes the probability of failure of the detection and
switching mechanism. Ps could be estimated as follows
(Billinton &Allan,1992)
A
P�s  (7-36)�
B

where

A = the number of times the switch works when required

B = the total number of times the switch is required to perform

its function

�
387 Reliability Engineering

����������������������������������������������
���������������������������������������

n
Let R (t ) denote the reliability function of a system having 1
sys

active unit and n-1 redundant standby units with an imperfect

2
switch(Ps<1). R (t ) is obtained from(K&L page 221):
sys

2 t
R (t)=R1(t)+Ps  f1(t')R2 (t-t)dt (7-37)
sys 0
where
2
R (t ) Reliability function of 2-component standby system
sys

R1(t) Reliability function of active unit

R2 (t) Reliability function of standby unit
Notice that
- Although it is probable the standby unit does not work when
required to replace the active unit, this probability has not been
taken into account.
- Substituting � � �� into Eq. 7-37 yields Eq. 7-18.
2-component standby , constant failure rate
Imperfect switch with reliability
Consider a system which has 1 active unit with constant
failure rate λ , 1 redundant standby unit with constant failure
rate λ and an imperfect switch having reliability P  �  p  � .
s
2
R (t ) is obtained from(Lewis,1994 page 340,341):
sys
Chap 7 Dynamic Models+Availability, Markov Chain 388

2 (1  p)1  1t  2t

R (t)=e 1t  (e  e ) (7-38)
sys 2  1

Let the units are similar i.e. 1  2   then:

2
R (t)=[1+(1-p)t]et  et (1 Ps t) (7-39)
sys

where

p = the failure probability of the switch when required

to perform its task

Ps 1 p =Switch reliability

2-component standby , constant failure rate

Imperfect switch with reliability function ( )
Consider a system which has 1 active unit with constant
failure rate λ , 1 redundant standby unit with constant failure
rate λ and an imperfect switch having reliability function
2
R S (t) . In this case R (t ) is obtained from(K&L page222):
sys

2 t
n  2 R (t)  R1(t)   f1(t)  Rs (t ') R2 (t t)dt (7-40)
sys 0

where

�
389 Reliability Engineering

2
R (t ) Reliability function of 2-component standby system
sys

R1(t) Reliability function of active unit

R2 (t) Reliability function of standby unit
R S (t) Reliability function of switch
f1 pdf of main unit lifetime(T1)

Special case: Constant failure rates

Let failure rates of the active unit, the cold redundant standby
unit and the switch is constant and equal 2 , 1 &  then:
s
t 2 1 1 1
MTTFsys   R (t )dt   
0 sys 1 (1  2   )(1   ) (1  2   )2
s s s
or
1 1 1 1
MTTFsys   (  ) (7-41)
1 1  2   1   2
s s

Furthermore if 1  2   then :

1  1 1
MTTFsys-standby   (  )
     s   s 

For a system with 1 active , 1 standby unit and imperfect switch

1 1 
MT TFsys-standby    (7-42)
  s (   )
s s

where

 = the failure rate of both units

Chap 7 Dynamic Models+Availability, Markov Chain 390

 =the failure rate of the switch

s

Eq. 7-40 for this case yields(K&L p222):

    t 
(t )  e  t 1 
2
R  1  e s   t 0 (7-����
sys  s   

The mean time to failure of is calculated as follows:

 2 
  1e s t dt 
MTTFsys   R (t )dt  e t 1  
0
sys
0  s  

1 1 
MTTFsys    which is Eq.7-42.
   (   )
s s s

Using Eq. 7.39 for the reliability function of a 2-component system

whose active and standby units have the same mean time to failure
1
MTTFpart  and the switch is imperfect with reliability P s :


 2
1+P s
M TTFsys   R sys (t )dt   (1+P s ) M T T F part (7-45)
0


Notice:

�
391 Reliability Eng
gineering

2
-Substituting P s  1 in the above relationships yields R (t ) and
sys

MTTFsys forr perfect switching which we saw earlier.

-The reliabilitty function of a two-component system having siimilar

active and staandby units with the same failure rate λ is :

  given
om Eq. 7-43 i.e. R (t )  e t 1   st
2
Calculated fro 1 e 
sys
 s  
λs as the sw
witch failure rate

2
ated from Eq. 7-39 i.e. R (t )  e t (1  Ps t ) if thee
Or is calculat
sys

reliability of the
t switch is given as a constant value Ps .

7 1
Example 7-6

A system
m is composed of 1 active unit A, 1 red
dundant
standby unit
it B and a switch S. The lifetime of the units is
exponentiallly distributed with 1 failure/1000hr. The keey has a
constant ratee of failure per 10000 hours. Find 500-hr reeliability
of the system
m and its mean lifetime.

�����������������������������������������������������������
1
From Dr Esh hargh's pamphlet, Faculty member of Sharif University
y of Tech
, Tehran. Figurre from K & L p 59.
Chap 7 Dynamic Models+Availability, Markov Chain 392

Solution:

Eq. 7-43 gives the reliability function:

 103 failure / hr 110001, s  1100001 104,t  500

t    s t 
2
R (t )  e 1  1e  90.2%
sys  s  

Eq. 7-44 gives the reliability function:

1 1 
MTTFsys     1909.9 hr End of Example�
 s s (s   )

Example 7-7 (Ebrahimi, 1992, p 267)

The failure rate of a device is constant and equal to 500

failures per 1 million hours (  6 ). To enhance the
 50 0  1 0

reliability another unit is used as standby redundant which

replaces the device upon failure by switch having 97%
reliability. Find 1000-hr reliability of this system. Solve the
problem again for a perfect switch.

Solution:

Eq. 7-39 gives the reliability function. For � � 0.��:

�
��� � e−λt (1 + PS × λt)�
R���

�
393 Reliability Engineering

�
������ = e−0.5 (1 + 0.�� � ��� � ��−6 × �������������
R���

For � � 1 from Eq. 7-19:

� λ .
(����� = e (1 + � � e (1 + 0.5)=0..9098
R

λ� = ��� � �� � ���� � 0.5 is the mean of the failures of

each device per 1000 hours. End of Example� �
Comparison of reliability function of active and
standby systems-imperfect switching
Eqs. 7-43 &7-45 give the following 2 apparent different
relationships for 2-component standby system-imperfect switch:

2    e s t 
1)R (t )  e t 1  1   t 0
sys  s  
�
2
2) R (t )  e
sys
 t
1  (1  p )t   e  t
(1  Ps t ) t 0

if the failure rate of the switch( s ) is given the former relationship is

used, if the reliability of the switch( P s ) is given the latter is used.

Question: Are these 2 relationship equivalent?

Answer: Suppose they are equal i.e.

Chap 7 Dynamic Models+Availability, Markov Chain 394

   t    e  t 1  (1  p )t  e  t (1  P t ) 
e  t 1  1  e s   
 s 
s

 t
1e s
Ps 
s t

This the switch average reliability. That is they would be

equivalent if is calculated as the average of its reliability

function ( e s t )of constant- failure- rate switch over period(0 t):

t
s t
e dt  t
1 e s
switch average reliability  Ps  0
 (7-46)
t 0 s t

2-component standby system with similar units,

imperfect switch and warm redundant unit

Consider a system that has an active unit and a standby unit.

both the main and the secondary units have the same failure rate
λ but the secondary unit has   as failure rate while is standby
mode. The switch is imperfect with the failure probability p=1-Ps.
The expression for the reliability function of this system is
(Lewis,1994 p262):

 
      
R (t )  e t 1  PS   (1  e t ) (7-47)
sys   
 

The system mean time to failures is calculated as follows:

�
395 Reliability Engineering

 
 
t t
t  t  
MTTFsys   Rsys (t )dt   e
0 0 1  (1  p )  (1  e

) dt  �
 
 

Ps 1 p
1  1 
1   1  
MTTFsys   �(7-48)�
 

Where

p� Switch failure probability

Ps  1  p � Switch reliability
 � Failure rate of each unit while in active mode
 � Failure rate of secondary unit while in standby mode

Example 7-8

Consider a 2-component standby system in which the switch

for replacing the units is imperfect and its lifetime is
exponentially distributed The main unit and the secondary units
when becomes active fail at the constant rate of  . We know
that:
Given the failure rate of the switch( �,the reliability of the
system is calculated from Eq. 7-43 i.e. (from: K&L p222):

   t 
R (t )  e   t 1   1  e s  
2
t0
sys
 s  
Chap 7 Dynamic Models+Availability, Markov Chain 396

If the failure rate of the secondary unit while in standby

mode is   then given the value for the reliability of the
switch, the reliability of the system is calculated from Eq. 7-47
i.e. (from: Lewis p262):
 
2
    
R (t )  e t 1  Ps  (1  e t ) .
sys   
 
a)Specify a condition under which these two expressions
identical if     .

b)If in Example 7-7   6 under which condition do

 50 0  10

the 2 relationship give the same result?

Answer :a)

Substituting     in second relationship and equating the 2

expression we would have the following result:

 t
1 e s 
Ps  
s 1  e t

We the Taylor expansion of f(x) about a is:

1 1
f  x   f a   x  a  f   a    x  a  f  a   ...
2

1! 2!

Therefore e
t is expanded as follows:

�
397 Reliability Engineering

×
=1+ !
(− )+ ⋯�

There fore ≅1− � or

1 e t  t (7-49)

Therefore we must have

 t  t t  s t
Ps 
1e s

 1e s
 
0
e dt
s t s t t

That is under the condition that the value given for the switch
reliability( Ps ) equals the mean of reliability function of
exponential- distributed �lifetime switch over period(0 t) the 2
expression give the same results.

Answer: b)
Since the specified   equals  in Example 7-7, therefore if
 t  1000
1e s ��� 1 e s �����λs ≅ �� ∗ ��−6 �������relationship
Ps  0.97 
st s 1000

give the same result.

End of Example�
Therefore the following expressions for the reliability of a two-
component standby system ( imperfect switch warm standby )

 
2
      
R (t )  e t 1  (1  p )  (1  e t )
sys   
 
Chap 7 Dynamic Models+Availability, Markov Chain 398

   t 
 e  t 1   1  e s
2
R  t 0
sys
 s  
2
R (t )  e t 1 (1 p )t   e t (1 Ps t ) t 0
sys

Give the same results if the failure rate of the redundant unit
while in standby mode equals the failure rate of the active unit
(   ) and the given reliability value for the switch(Ps) equals
the value obtained from Eq. 7-46.
Example 7-9( Lewis,1994 page 262)
An engineer designs a standby system with two identical
units to have an idealized MTTF of 1000 days. To be
conservative, she then assumes a switching failure probability of
10%( p=0.10 or PS =0.9 ) and the failure rate of the unit in

standby of 10% of the unit in operation i.e. λ + �����λ .

Assuming constant failure rates, estimate the reduced MTTF of
the system with switching and standby failures included.

Solution

The system has n=2 components (1 active 1 standby). Since

the failure rate of each component is constant; therefore the
distribution of the lifetime of the units are exponentially
distributed .
Let  = the comstant failure rate of each unit in operation i.e. when is

active. If the standby unit were cold, according to Eq. 7-27-4 for the
idealized case:

�
399 Reliability Engineering

MTTFsys=2  1 �⟹ 1000  2  1    0.002 per day .

 
In imperfect switching according to Eq. 7-48 with p  0.1 &

λ + � � �� � λ :

1 p 1  0.1
1 1

1  1  0.1  909days
MTTFsys  
 0.002

End of Example�
The relationships for 3-component standby systems could be
studied in K&L pages 221-222. The interested readers in the
reliability function and the MTTF of the general case n=n, could
refer to Niaki & Yaghoubi(2020).

7-1-4 Shared load parallel configuration1

Up to now when analyzing redundancy, independence was

assumed among the units within system. In other words, it was
assumed that the failure of a unit does not affect the failure rates.
In this section , the load-sharing systems are considered, where
the assumption of independence is no longer valid. When units
in a system fail one by one, the total load of the system is
redistributed among the surviving units, resulting in an increased
load shared by each surviving unit. For an example of load
sharing consider a section of a machine which has several
�����������������������������������������������������������
�
��������������������������������������������������
Chap 7 Dynamic Models+Availability, Markov Chain 400

screws When a screw break, the total load is redistributed

among the surviving screws.
Here analysis will be limited to the case of two units. In this
load sharing case the reliability function of the system is(K&L
p224):
2 t
Rsys (t )  R g (t )  2 g (t1 )R g (t1 )Rf (t  t1 )dt1 (7-50)
0

Where

g(t) The pdf for TTF under half load

f(t) The pdf for TTF under full load
R g (t ) 
 t g ( )d
R f (t ) 
  t f ( )d 

For the proof refer to K&L page223.

For many probability density functions, calculation of Eq. 7-

50 is difficult. As an easy example constant-rate-failure is
considered.
������� ������
�����������������������������������������������������
Consider a load sharing system that has 2 identical
components. The failure rate of each is constant λ g when both
units work. The failure rate increases to another constant λ f
-λ � -λ �
when one unit fails. substituting g(t)=λ g� g , Rg (t)=e g and
-λ �
R f (t)=e f in Eq. 7-50 gives the system reliability function
for this case as follows(K&L page 224)

�
401 Reliability Engineering

-λ f � 2 g t 2g 2 g t
R f (t)=e R sys (t )  e  (e  f t  e ) (7-51)
2g  f
�

Where

Half load failure load

Full load failure load�

Example 7-10

Consider a 2- component load sharing parallel configuration

in which the failure rate is 1� �� �per hour under partial load(
half load here) and 4� �� under full load. Calculate 1000-hr
reliability of the system.
Solution

λ � 0.���� ℎ �����λ �= 4 × �� � ℎ �

According to Eq.7-51:

2 g t 2g 2 g t
R sys (t )  e  (e  f t  e )
2g  f
� (����� =

× . ×
2 × 0.��� . × × . ×
e + �e −e )
� × 0.��� − 0.���
Using MATLAB�
�exp(-2*0.001*1000)+2*0.001*(exp(-0.004*1000)-exp(-2*0.001*1000))/(2*.001-0.004)

⇒� (����� = ������ End of Example

Chap 7 Dynamic Models+Availability, Markov Chain 402

�����������������������������������

Reliability is not the only index used to characterize the

performance of an engineering system. Some other features are
serviceability, maintainability, operational readiness and
availability described below.

Serviceability�

Serviceability�is the measure of the features that support the

ease and speed of which corrective maintenance and preventive
maintenance can be conducted on a system. In a simple
statement we could say serviceability is used to present the
degree of the difficulty with which equipment can be repaired.
When it is said equipment 1 is more serviceable than equipment
2 , it is meant that the better Serviceability the shorter the active
repair time.

This index is difficult to measure on a ratio scale; however it

can easily be measured on an is usually expressed as ranking.
Serviceability is difficult to measure on a ratio scale;
however, it can easily be measured m ordinal scale by a
specifically developed rating and/or ranking procedure, which
requires that systems be compared and ranked according to the
ease of serviceability (Handbook of industrial Eng�� edited by Gavriel Salvendy ).

�
403 Reliability Engineering

Maintainability

In MIL-STD-721C, maintainability is defined as follows:

The measure of the ability of an item to be retained in or
restored to specified condition when maintenance is performed
by personnel having specified skill levels, using prescribed
procedures and resources, at each prescribed level of
maintenance and repair(Ireson, et al, 1996 page 15-3)
While the reliability engineer is concerned with many
physical characteristics that affect system components, such as
temperature, humidity, shock, and vibration, the maintainability
engineer will be concerned with the physical partitioning of a
system into repairable items; the accessibility, weight, and
volume of these items; the skills and training of maintenance
crew; and the availability of the appropriate tools and equipment
for conducting maintenance activities(Ireson, et al, 1996 page 15-3).
The maintainability index of a machine is the probability that
it restores to working status within a specified period. Notice by
the term "down time" used sometimes here it is meant all the
time period the machine is out of service. This time period
includes the time necessary to detect the failure, the repair time,
administrative and logistic times.

Maintainability function

Maintainability function for a device, denoted by M(t), is the

probability that the maintenance task considered will be
successfully completed before a specified time t:
Chap 7 Dynamic Models+Availability, Markov Chain 404

�� M (t )  Pr(T repair  t ) (7-52-1)

where random variable is

= The time required for completing the service,
maintenance, replacing new units�
If m(t) denotes the probability density function of then :

t
M (t )   m (x )dx ����������������������
0

Therefore the maintainability function for a device,

represents the probability that the device restores( gets out of
down state) successfully within a specified time. It is worth
noting that exponential, log-normal distribution Weibull are 3
distributions frequently used for service times.
Example 7-11

The total service and maintenance time of a dive has

the pdf m ( t ) = γ � - γ � , find the maintainability function
of the device.

Solution
t

M (t )  1 e  t  1 e 
End of Example��� ����

It is worth mentioning the term dependability has been

introduced to cover all important aspects of a device to function
satisfactorily including reliability, availability, maintainability,
quality and safety. Interested readers could refer to references

�
405 Reliability Engineering

such as Standards IEC 60050-192 & IEC 60300 �, Eusgeld et al 2008)

and Martha et al (2022)

��������������������������

A widely used maintainability parameter is MTTR . For an n-

component system it is calculated from(Ireson, et al. 1996 page
15-6):
n

 (MTTR )  
i i n 
MTTR  i 1
  i
 (MTTRi ) ��������������������7-53 )�������
 i
 
 i 1 i 

�����
where
i Failure rate of the ith repairable component
MTTR i Mean time to repair ith repairable unit
n number components in the system
i A fraction of failures per unit time related to
 i

ith unit
It might be useful for some readers to know that some
references such as the manual of MIL-HDBK-472 standard
deal with MTTR in details. This manual is comprehensive
design tool for maintainability prediction analysis including
calculating MTTR.
����������������������(OR)
�
"The term operational readiness(O.R.) is defined as the
probability that either a system is operating or can operate
Chap 7 Dynamic Models+Availability, Markov Chain 406

satisfactorily when the system is used under stated conditions.

Operational readiness is more encompassing than the term
availability" (K&L page 225).
3 time periods are used in the calculation of OR(K&L p 226):
operating time  idle time
O.R.  ������������������
operating time  idle time  down time

Example 7-12(K&L p226)

The following figure shows the status of a machine over a

time horizon graphically. Suppose the total operating time of
the machine is 8 time units, the idle time and the downtime is 6
time units each. Calculate the operational readiness of the
machine

Fig 7-5 A Machine status over a time horizon(K&L p226)

Solution

operating time  idle time 8+6

OR    0.70
operating time  idle time  down time 8+6+6

�
407 Reliability Engineering

This mean that the machine is ready to perform its function 70%
of the time. End of Example��� ����

�����������������

Availability as a measure of system effectiveness is defined

"as the probability that an item will be available when required,
or the proportion of total time the item is available for use"
(O'connor,2003 page 300). At first availability is studied for
the case where different times(operating, down..) are fixed and
not variable. Here availability is denoted by A . Availability
which excludes free(idle) time would be estimated from(K&L
page 227)

operating time
A (7-54-1)
operating time  down time

7-3-1 Intrinsic Availability

Intrinsic availability index( ) of a machine does not include

administrative time and logistic time in the down time of the
machine. In other words , it ignores administrative delays( such
as the time it takes to find a repairman, spare components,
tools��� ���� ����� ����� ��������g time and actual repair time .
Therefore is computed from(K&L page 227):
���������� ����
= (7-55)�
������������ ����� � �� r. t
where a.r.t. is the actual repair time shown in Fig 7-5.
Chap 7 Dynamic Models+Availability, Markov Chain 408

Example 7-12:
The down time in one complete cycle of a machine is 6 time
units and its operating time is 8 units(See Fig 7-5). Find the
machine availability (A).
Solution
operating time 8
A A  =0.57  57%
operating time  down time 6+8
End of Example���

Example 7-14 In Example 7-13 if the administrative time is

one time unit , and the logistic time is also one time unit. Find
the intrinsic availability.

Solution
� � . .
a.r.t.= total down time- administrative& logistic times=6-2=4.
8
� = = 0.�� � ���
8+4
Examples 7-13&14 show that by eliminating the administrative
and logistics time in the repair cycle, the current availability of
0.57 can be increased in the limit to the intrinsic availability of
0.66.� There is a potential for a 9% improvement in availability��

End of Example���

7-3-2 Availability function

The availability , like the reliability ,is time dependent. The

above relationships for availability give fixed values
independent of time.

�
409 Reliability Engineering

The reliability function for repairable systems ,A(t), is

defined as the probability that the system operates at time t
irrespective of its past history of breakdown and repair(Grosh,
1989 page 268).
Assuming specific models for both the failure and downtime
(repair time)distributions, the maintainability and availability
functions could be derived. Considering the simplest possible
case (using the exponential distribution with parameters �and
respectively for time to failure and repair time) yields a
differential equation for availability function of this simple case
dA (t )
(K&L page228):  (  )  A (t )  
dt

����������������������������������������������������������
�������������������������������������������

The following solution of the above differential equation is

the availability function for a system with exponentially
distributed TTF and downtime (repair time) having parameters
� respectively(Grosh, 1989 p 270 , K&L 280)

  (   )t
A (t )   e ������� ��������������(7-56)
   
Chap 7 Dynamic Models+Availability, Markov Chain 410

7-3-3 availability function for nonrepairable systems:

exponential lifetime

For a system which is not repairable   0 . If the lifetime of

the system is exponentially distributed, from Eq. 7-56 it is
concluded that its availability equals its reliability at time t:

nonrepairable system A(t)=R(t) (7-57)

7-3-4 Steady state availability

A fraction of the total time that the device is ready to perform

its duty in the long range is called steady state availability
dented by A, If the operating time (TTF) of the machine is

exponentially distributed with mean MTTF= 1 and the down

λ

time has the mean MTTR= 1 as an index of maintainability,



then(K&L page 228):

μ
�� = ��� → � �� μ
(7-58-1)�

or � = λ
�� (7-58-2)
�
μ λ

Needles to say that A has a value between 0 and 1.Notice

that:
-If only the life time is in exponential form and the down time is
not exponentially distributed, the relationship for the steady
state availability is the same as Eq.7.58. and more generally

�
411 Reliability Engineering

according to Ross(985) page 402 : If the on & off distributions

are arbitrary continuous distributions with respective mean

& then it follows from the theory of alternating renewal

μ

process (see page 287 of Ross,1985) that A(t) in the long range

approaches to = .
�
μ

Therefore we could say if the operating time and the repair

time have continuous probability distributions with mean MTBF
and MTTR, the steady-state availability would be:
MTBF
A= (7-59-1)
MTBF+MTTR
Fig.6.7 shows a nomogram (abaque) for this relationship
between availability and MTBF(a measure of reliability) and
MTTR(a measure of maintainability). That is draw a line which
connects the given MTBF on the MTBF scale and the given
MTTR on the MTTR scale. The intersection of this line and A
scale is value equal to what is resulted from Eq. 7-59.1.
It is reminded that
- if the life time and the repair time are not random variables
use Eq. 7-54&55 for calculating availability.
-A device can have low availability, high reliability and vice
versa.
- Index A could be used for comparing two types of a device
that have the same reliability.
Chap 7 Dynamic Models+Availability, Markov Chain 412

Fig. 7.6 a nomogram form of Eq. 7-59-1 (Ebrahimi,1992 p331)

Example 7-15
It is desired that a machine which has an exponentially
distributed lifetime with mean 3000 hours to possess a steady-

�
413 Reliability Engineering

state availability of 99.95%. What should be the mean time to

repair(MTTR)?
Solution
Using Eq. 7-59-1:

MTBF 3000
A  0.9995   MTTR  1.6 hr
MTBF  MTTR 3000+MTTR

Using the nomograph of Fig 7-6

The line connecting 3000 on the MTBF scale to b0.995 on
Scale A , gives MTTR=.6 on MTTR scale.

7-3-5 Intrinsic availability in long rage

The intrinsic availability in steady state can be calculated

from (K&L page 228,Stapl, 2009 p 344):
MTBF
AI (steady )  (7-59-2) �����
MTBF  m.a.r .t .
��
Where m.a.r.t is the mean of actual repair time.
Notice that
-Actual repair time does not include logistic and administrative
times as has been shown in Fig 7.5 for a deterministic case.
-The nomogram of Fig. 7-6 could be used for this relationship.
Chap 7 Dynamic Models+Availability, Markov Chain 414

Example 7-16
A series system has 2 subsystems. One the is a compressor
with 80.37 failures per �� working hours. The average of its
actual repair time is 89.3 hr. The other subsystem failure rate is
4.78 failures per �� working hours and the actual repair time
on the average is 890.3. Calculate 26280-hr reliability of the
each subsystem , their steady state intrinsic availability
Solution
Constant failure rate λ implies that the life time distribution
1
is exponential with mean θ � and reliability function .
λ

therefore
For compressor:

λ =80.37 � ��−6 ⟹
θ MTBF1=1/80.37*10^-6=12442.4
� )=exp(-(t/ ))
(26280)=exp (-26280*80.37*(10^-6))=0.1210
.
� . . .
= . .
= 0.9929

For other subsystem

� �.�� � ��
θ � MTBF2 =1/4.78*10^-6= 209205 hr
�26280)= exp(-26280*4.78*(10^-6))= 0.8819
2 �������
= 2+ . . . 2
= �������+����3 =0.9958

�
415 Reliability Engineering

7-3-6 Mission Availability

The average of reliability function, A(t), over time period T

in some references is denoted by A*(T) and named mission
availability or interval availability. Given the availability
function over a period (0 T), A*(T) is calculated as follows:

1 T
T 0
A * (T )  A (t )dt (7-60)

Fig 7.7 The average of a typical availability function

Figure 7-7 shows this integration graphically.

Since for a non repairable system its availability equals its
reliability at time t :A(t)=R(t) therefore(Lewis,1994 p301):

1 T
A * (T ) 
T 
0
R (t )dt (7-61)

That is the mission availability and the average of the reliability

function related to the same time period are equal.
Chap 7 Dynamic Models+Availability, Markov Chain 416

Example 7-17
The lifetime of an non-repairable switch with known MTTF
is exponentially distributed with parameter λ �
a)Calculate the parametric average of the reliability function of
the switch.
b)(Lewis ,1994 p301)
The system mission availability must be 0.95. Find the
maximum design life that can be tolerated in terms of the
MTTF.
Solution
a)For this switch which is non-repairable, A * i.e. the mission
availability equals the average of the reliability function(Ps) .
1 T
Ps  A * (T ) 
T  0
R (t )dt TTF ~ exp( )  R (t )  e  s t

1 T 1  e  sT
Ps 
T 
0
e  s t dt 
sT
Since the Taylor expansion of f(x) about a is;
1 1
f  x   f a   x  a  f   a    x  a  f  a   ...
2

1! 2!
 T
Then the expansion of e about a=0 :
- T
T-0 (T-0) 2
e s =e-λs ��� + (-λ s ���λs � �T=0 � ��λ s ��λ s ���λs� ��T=0 ���� 
1! 2!
1 1
e- sT  1-λ s �� �λ s ��2 � �λ s ��3 �����
2 6
1 -λ �
Then for sT  1 , approximately e s  1-λs�� �λs�� and:
2

�
417 Reliability Engineering

1
1-1+λ s �� �λ s �� 2
1-e-λs � 2 1
Ps =   1- λ s �  A * �T �
λs� λs � �
∗� � = 0.�� then
b)
1 �
0.95  1- λs�  T   T  ���  MTTF
2 10
s
End of Example

7-3-7 System Availability in terms of components'

Availability

As stated earlier according to Ross(985) page 402 :

"If the on & off distributions for component i are arbitrary

continuous distributions with respective mean & �

μ

1,2, … , then it follows from the theory of alternating renewal

process (see page 287 of Ross,1985) that ( )in the long

range approaches to
1
λi
� )⟶ = 1 1 " (7-62)
+
λi μi

Where �is the mean lifetime of component I and μ

is the

mean of its downtime.

Consider a system composes of n independent components

with reliabilities ��� . Let f ( R1 ,..., R n ) denoted the system

reliability function. The steady state availability of the system
,A, is calculated from( according to examples on page 402 Ross1985)::
Chap 7 Dynamic Models+Availability, Markov Chain 418

������ � � , … , ) �������������������(7-63)

where

The reliability function of n-component system

� f
such as Eq. 2-1 or Eq.2-3.
1 The steady-state availability of component i which
λ
� �
1
+
1 replaces in � ,…, )
λ μ
1 The average lifetime of component i
�
λ
1 The average downtime of component i
� �
μ

Example 7-18
The lifetime and downtimes of n independent components,

on the average, are �� and ,i=1,..,n and their reliability are

�…� . Calculate the parametric expression for the steady

state availability of the system in both series and parallel
configuration.
Solution
For series configuration: = ( ,…, )= ×…×
Therefore according to Eq. 7-63 the steady-state availability
of the series configuration of the n components is:

= ( ,…, )= ×. . .×

For parallel configuration:

�
419 Reliability Engineering

= 2
( 1, … , ) = 1 − (1 − 1) ×. . .× (1 − )�

= ( ,…, ) =1 − (1 − ). . . (1 − )

Example 7-19

A 2-unit system fails when either of its units fail. The units
have the steady state availability of 0.9958 and 0.9929.
Calculate the steady state availability of the system.

Solution

The configuration is series.

= ( 1, 2) = 1 × 2 ���������������

� � , )= × ������������������A=0.9929*0.9958=0.9887

7-3-8 The steady-state availability in Preventive

Maintenance

From O'connor (2003 )page 402:

"Maintainability affects availability directly. The time taken
to repair failures and to carry out routine preventive
maintenance removes the system from the available state. There
is thus a close relationship between reliability and
maintainability, one affecting the other and both affecting
Chap 7 Dynamic Models+Availability, Markov Chain 420

availability and costs. In the steady state, i.e. after any transient
behavior has settled down and assuming that maintenance
actions occur at a constant rate" (O'connor,2003 page 402):
C
= × � (7-64 )
+ C+T
where
C= Preventive maintenance cycle [ e.g. every 1000 hr]
T=Total time required to perform preventive maintenance tasks

7-3-9 Definition of Unavailability Function

The unavailability is the event that at a point of time a system

or a device does not perform its duty under specific conditions.
If the value of steady state availability is A the unavailability in
steady state would be 1-A.

Unavailability Function: Lifetime Exponential

In a special case where the lifetime and downtime are

exponentially distributed, the Instantaneous unavailability
function would be (O'Connor, 2003 page168):
  ()t  (  )t
U(t)  1  e  (1e ) �(7-65)�
      

�
421 Reliability Engineering

������������������������������������������
���������������������

In this section we would like to have a glance at the

application of Markov chains to the reliability analysis of
repairable or nonrepairble systems whose lifetime follow
exponential distribution. For other distributions Monte Carlo
simulation is applicable. See references such as Chap. 8 of
Smith(1993).
If each component has approximately an exponential failure
law, the complete system can be described approximately by a
Markov process and to predict the future state of the system,
knowledge of the history of such systems contains no predictive
value( extracted from Barlow & Proschan,1996 page119).
Figure 7-8 shows the state space of a 2-unit critical system. In
this system two identical computers A & B are connected in
parallel in such a way that both are operating although only one
is in actual service. At a time of a computer failure, repair is
done readily. Preventive maintenance for a specified computer is

scheduled after t 0 hours if one computer is active and the other

is on an operating standby basis. If the first computer fails and
the second fails during the downtime the first one the
consequences could be catastrophic(Barlow & Proschan,1996 p120).
Chap 7 Dynam
mic Models+Availability, Markov Chain 422

Fig 7-8 shows

s a diagram of the state space of the critical
system. Thee possible states are denoted by symbols succh as
(Computer A is active) , (Computer B is in standby mo
ode)

F 7-8 The 9 possible states of a two-unit systeem

Fig
(Barlow and Proschan , 1996p120)

The state space has 9 states labeled 0 through 8. For example

e
state o ind
dicates that computer A is used actively
y while
computer B is operating and standby. If no failures occcur in a

time intervaal of length t 0 , measured from the mom

ment the
system enterr State 0, preventive maintenance is perform
rmed on
computer A and State 1 begins. If no failure occurs, the
th state

�
423 Reliability Engineering

passes around the perimeter of the square����������������������

unfavorable. (Barlow and Proschan , 1996page121).
The user of this system may be interested in such information
as the mean system down time during a specified time interval,
the probability that the system is down more than x minutes at
any one time. Under certain reasonable assumptions on time to
failure(TTF) , the time to perform repair(TTR), etc. The
operation of the system can be described by a semi-Markov
process to get the desired information. Chapter 5 of Barlow&
Proschan(1996) deal with this system in detail.
As another example consider a system having 3 components
or units a, b, c. To use a Markov chain the states of this system
are defined as combinations of operating and failed components.
As the following table shows the system, depending on the
operation or failure of the components(o=operating X= failed),
has 8 states(Lewis,1994page 326):
unit State�
1 2 3 4 5 6 7 8
a o� X o� o� X X o X
b o o X o X o X X�
c o o o X o X X X
o=operating X= failed

Chapter 11 of Lewis(1994) deals with Markov analysts of 3

configurations related to this case and it is worth mentioning
that many references deal with the application of Markov
Chap 7 Dynamic Models+Availability, Markov Chain 424

process to reliability theory; e.g. Dhillion(2006) pages 49-54.

Barlow & Proschan(1996) page 120 and pages 192-197.

������������

1. A system composing 20 independent components, each

having failure rate �.���� � 0.���� , fails when one its
components fail. Calculate the system failure rate and its
reliability for t=10.
Hint: Use Eq. 7-2 for calculating the failure rate.
2. A standby parallel system has identical active unit and
redundant standby unit with constant failure rate λ. Show
that the instantaneous failure rate function for the system
 2t
is: hstandby (t )  ,
1  t
3.Derive Eq. 7-51 by the help of 7-50 i.e. prove that in a
two-component load sharing parallel system with pdf
-λ � -λ �
g(t)=λ g� g under shared load and pdf f(t)=λ f � f under full
� g t �g � g t
load, .R sys (t )  e  (e  f t  e ) would be the
�g  f
system reliability.

Helping one person might not change

the whole world,but it could change
the world for one person

�
425 Reliability Engineering

Chapter 8
Enhancement,
Optimization
&Allocation
of Reliability
Chap 8 Enhancement, Optimization & Allocation of Reliability 426

8
�������������������������������������������
������������

Aims of the chapter

Due to the importance of the design phase in setting the

reliability of products, this chapter deals with how to enhance ,
optimize reliability and to allocate reliability to each component
in the system to have a more reliable design.

���������������������������������������
������������

There are two conventional approaches to improve the

reliability of a system(based on Shooman,2002 page 335):
1)Enhancing the reliability of the system components
2)Active(or hot) redundancy and standby redundancy
These two approaches are described below.

8-1-1 Improving Component reliability

An approach for enhancing the reliability of a system is

improving the reliability of the basic elements, Ri, by allocating
some or all of the cost budget to fund redesign for higher

�
427 Reliability Engineering

reliability.( Shooman,2002 page 335) Figure 3-3 and 305 could

help to clarify this approach.

Example 8.1
The series system shown in Fig. 8.1 is composed of k=3
identical components with a reliability of 0.80 each.

Fig. 8-1 A k-component series system.

a) Calculate the current reliability of the

system.
b) What do you suggest for the reliability of
each component in order to enhance the
system reliability to 0.95?

Solution

a) R=�.8 =0.512.

b) The enhancement requires that each component has the

reliability of √0.�� . End of Example

8-1-2 Active(Hot) and standby redundancy

Another approach to enhance systems reliability is to place

redundant components in parallel with the operating components
either in active(hot) or standby status.
Chap 8 Enhancement, Optimization & Allocation of Reliability 428

8-1-2 Active redundancy

In this way of enhancement components are placed in parallel

with the subsystems that operate continuously (see Fig. 8.2) This
is ordinary parallel redundancy(hot redundancy).

Fig. 8-2 The k-component system with active redundancy.

(Shooman,2002 p336)

8-1-2 Standby redundancy

This form of redundancy places components in standby

parallel with k subsystems and switch them in when an on-line
failure is detected(Shooman,2002 page336). Figure 8.3 shows
this case. The redundant components of the system shown in the
figure are cold. On the figure denotes the failure rate of the
operating unit and dentoes the failure rate of the redundant
unite in the standby mode.

�
429 Reliability Engineering

Fig. 8.3 A standby redundant parallel system

A combination of active and standby redundancy is shown in Fig. 8-4

Fig 8-4 Combination of active and standby redundancy1.

�����������������������������������������������������������
�
����������������������������������������������������������������������������� �
Chap 8 Enhancement, Optimization & Allocation of Reliability 430

������������������������������

Optimization play a distinguished role in system design, Its

objective in reliability subject is to help developing a more
effective and safe design that works within existing constraints.

Two conventional reliability optimization problems are:

maximizing system reliability with cost constraints or
minimizing system cost subject to the constraint that the
reliability be greater than a given minimum. As an application
you know that adding redundant components in parallel to a
system improves the system reliability. However this approach
enhances the cost, weight and volume of the system. Therefore
an optimization problem has to be presented and solved in such
a way that the optimum design considers the constraints as well
as maximizing the reliability.
To write a general model, let x= (x1,����k) be the decision
variables; the model could be written as follows:

/ ���� ( ) ��

. . ( )≤0 = 1,2, … ,
ℎ( )=0 = 1,2, … ,
����������� ≥ ��

could be a cost function, the reliability function in the system

(series, parallel, structural load-strength systems��� ��� ����
average lifetime.

�
431 Reliability Eng
gineering

As an illustration
i consider the following series--parallel
system show
wn in Fig 8-5.

Fiig.8-5 A series-parallel system(Faghih,1996 p102)

k
Total num
mber of components in the system is n
i 1
i
. If we

suppose thee components in Subsystem ��are identicaal, each

bility R i , then
having reliab
bility if 1st subsystem= 1 (1 R1) 1 ,
n
The reliab
bility if 1st Subsystem i  1 (1 R1) 1 .
n
The reliab
The reliab
bility of the entire system is:

k
R sys   1  (1  R i ) ni  �����������������������
i 1

where
k numbber of subsystems
Ri The reliability
r of each component in Subsystem i
ni Numb ber of components in Subsystem i
Now let
k = Maxiimum budget available
Ci = The cost
c of each component in Subsystem i
Then :
Chap 8 Enhancement, Optimization & Allocation of Reliability 432

k
�������� n C
i 1
i i  C ����������������� (8-2)�

And given specified values of Ri, i=1,2,����� ���� ��������

would be determine ni, i=1,2,���� ��� ����� �� ���� ����� ����
reliability of the system is maximized and the constraints be
satisfied. The model could be written as:

k
M ax R sys  f ( n 1 ,..., n k )   1  (1  R i ) n i 
i 1

s .t .
k

n C
i 1
i i C

0  R i  1 i  1,..., k
n 0 n i =Integer
i

Another sample model could be the following:

k
M ax R   1  (1  R ) i 
n
i 1 
i 
s .t .
k

a
i 1
ij ni  b j i  1, 2,..., k j  1, 2,.., �

0  Ri  1
n 0 n i =Integer
i

where

aij Amount of jth material used for components of ith

subsystem
bj Amount of jth material available
k Number of subsystems

�
433 Reliability Engineering

� Number of materials
th
ni � Number of components in i subsystem

Cost function and MTTF or MTBF could be the objective

function .The following model which determines the optimum

values of ni , i  1,..., n such that the cost is minimized and

ensures that the system reliability will not be less than R 0 :

k
Min Z   ni C i �
i 1

�����

k
R sys   1  (1  R i )ni   R 0
i 1

0  R i  1 i  1,..., k
n 0
i
n = Integer
i

To maximize the system mean lifetime of the system in Fig

8.5 and ensuring that the system reliability exceeds , the
following model could be used:
Chap 8 Enhancement, Optimization & Allocation of Reliability 434

MaxZ  (MTTF )sys  Min{MTTFi , i  1, 2,..., k }

s .t .
k

 1  (1  Ri )
ni 
i 1
  R 0
0<R <1 i=1,2,...,k
i
n 0
i
n = Integer
i

Remember that if an active parallel configuration has ni

exponentially-distributed-lifetime components with identical
parameters � =⋯=θ = θ , then according to Eq. 7-15
ni
MTTFi    h1 .
h 1

To know more about reliability optimization, the reader could

read references such as Chap 6 of Barlow&Proschan(1996).

8-2-1 Methods for the solution of The above problems

There are several methods and softwares for solving

reliability optimization problems including(Yi-Chic,2002):
1.Exact methods (such as Brach and bound algorithm,
dynamic programming, Cutting plane algorithm, Surrogate
constraint method). This methods are time consuming for large-
scale problems. Kuo and Prasad (2000) provides a good
overview of the methods that have been developed since 1977
for solving various reliability optimization problems.

�
435 Reliability Engineering

2.Heuristic methods, Artificial intelligence(Genetic algorithms,

Simulated annealing, Artificial neural networks , Tabu search,..)
and other methods such as Lagrange multiplier technique,
geometric programming, Random search; some of these
methods give approximate solution.
Notice that:
-For application of optimization to structural reliability see
references such as K&L page 423and Xie, Zhai(2021)�
-If we have simultaneous objective functions such as
�f ، … ،f ،f ) which are to be maximized and ��′ ، � ،�′ ،�′ �

subject which are to be minimized subject to constraints

�� ، … g ) ≥, ≤, = 0; multi-criteria decision making (MCDM)
techniques could be used. A mathematical model of an MCDM
problem could be written as follows:
����f ( ), f ( ), . . . , f (x)}
����� � �, f ( ), . . . , f ( )}
s.t.
� � ){≤, =, ≥}0 i = 1,2, . . . , m
where x is a vector including the decision variables

��������������������������������

In the subject of reliability there is a problem called

reliability allocation in which it is discussed how much the
reliabilities of all or some of the components or subsystems

�����������������������������������������������������������

1
The refrence of this chapter is mainly K&L, Chap 14.
Chap 8 Enhancement, Optimization & Allocation of Reliability 436

( � � … ) of a given system should be to achieve a specified

overall system reliability ( ). This process requires solving the
following inequality(K&L page 404):
� � ,….., )≥ (8-3)

Where

The unknown reliability of component i

The required reliability for the system
f The functional relationship between the components and
the system

Time and cost could be included in the problem , i.e. ′ be

time-dependent and total cost be minimized.
The solution procedure is not difficult for series, parallel and
k-out- n configuration; however the solution for complex
configuration is not mathematically easy.

Most of the basic reliability allocation models are based on

the assumption that component failures are independent, the
failure of any component results in system failure (i.e., the
system is composed of units in series), and that the failure rates
of the components are constant. The independence assumption
leads to the following equation

� � ,…, )= ( )….. ( )≥ ( ) (8-3-1)

�
437 Reliability Engineering

Let = constant failure rate of the ith component. The system

has a series configuration. Therefore the lifetime of the system is
the minimum of the component lifetimes which are of
exponential form; therefore the lifetime of the system is
exponentially distributed. As a special case of Eq. 8-3-1, if the
goal value of the failure rate of the system is ,then Eq. 8-3-1
becomes (K&L p 407):

�..e ≥ (8-4)
or
� λ +. . . . +λ ≤ (8-5)
Theoretically, the above equation has an infinite number of
solutions, assuming no restrictions on the allocation. The
problem is to establish a procedure that yields a unique or
limited number of solutions by which consistent and reasonable
reliabilities may be allocated1. Some of these procedures are:

1. Equal Apportionment Technique

2. ARINC Apportionment Technique
3. AGREE Apportionment Technique
4. Feasibility of Objective apportionment
5. Repairable System Apportionment
6. Minimum Effort Algorithm
7. Growth apportionment

�����������������������������������������������������������
1
From: http://reliabilityanalytics.com/blog/2011/10/09/reliability-allocation/
Chap 8 Enhancement, Optimization & Allocation of Reliability 438

8. Dynamic Programming

The first 3 methods are described below

8-3-1 Equal Apportionment Technique(ETA)

This technique allocates equal reliability to the n components

of the system to achieve the system reliability requirement.
When no information on the system is available, other than the
fact that n components are to be used in series or parallel , equal
apportionment( ) to each subsystem would seem reasonable.
Furthermore allocation of the same reliability( ) to all
components of a k-out-of-n system is usual.
For series configuration, nth root of the system reliability
requirement( ) would be apportioned to each subsystem:

� � × …× ⟹ =

= = i = 1, … , (8-5-1)

For active parallel configuration, according to Eq.�(2 − 3 − 1):

= 1-(1-R1) (1-R2) �����-Rn)

= R�� i = 1, … , n ⟹

= 1 − (1 − R) = 1- ⟹1-R= �1 − )⟹

=1− (1 − )�������������������(8-5-2)

�
 439 Reliability Engineering

And for k-out-of-n configuration, according to Eq. 2-5 (Chap 2):

n
� � (1 − R) (R)
x

Given� � k and n, R has to be determined in such a way that

��∑ �1 − R) (R) − R )����

�������������������������������������

Example 8-2
The reliability requirement for a 3-component series system
is = 0.8573. Find the reliability of each component using
equal apportionment technique.

Solution

According to Eq. 8-5-1:

� �� = 0.8573

��The reliability of each component is;

�� = √0.���� � �.���� ����� ���������������

Example 8-3

The reliability of a 3-component active parallel system is

required to be = 0.8573. Use ETA to determine the
reliability of each component.
Chap 8 Enhancement, Optimization & Allocation of Reliability 440

Solution

According to Eq. 8-5-2:

�1− (1 − ) n=3 � 0.�����⟹�

�1− (1 − 0.����� � 0.���� �

Example 8-4

The reliability of a 3-out-of -5 configuration is required to

have the reliability of = 0.99144. Calculate R, the reliability
of each of the five components.

Solution

According to Eq,2-5 in Chap. 2, R is derived from:

� (1 − ) ( ) �

then: �

�
(1 − ) ( ) − 0.����� � 0

The following MATAB commands yields = 0.9.

f=inline ('binocdf(5-3,5,1-R)-0.99144');R=fzero ( f,0.5).�

�
441 Reliability Engineering

8-3-2 The ARINC apportionment technique for series system

with independent subsystems having exponential lifetimes

The ARINC allocation method was developed by a research

center associated with Aeronautical Radio Incorporation. This
method is applicable to a series system whose subsystems have
constant failure rate( ) and their mission times equal the
system mission time. Let
∗
The allocated failure rate to the subsystem
The desired failure rate given for the entire system

∗
The ARINC method tries to choose such that(K&L p 407):

∗
+. . . . + ∗
≤ �

Steps of ARINC apportionment technique are(K&L page408):

I. Determine the subsystem failure rates ( � i = 1,2, …) from
the past data, observed or estimated.

II. Assign a weighting factor ( ) to each subsystem

according to the failure rates determined in step I, where is
given by
� ���������� (8-6)�
n


i 1
i

III. Allocate new subsystem failure rates (λ∗ ′ ) calculated from

the following relationship( assuming λ∗ +. . . . +λ∗ = λ );
Chap 8 Enhancement, Optimization & Allocation of Reliability 442

∗
= ��������������(8-7)�

where λ is the desired failure rate for the entire system.

Example 8-5(based on Example 14.2 K&L page408)

Consider a series system composed of three subsystems with

constant failure rates. The mean lifetimes are 200, 333, and 1000
hours respectively. The system has a mission time of t=20 hours.
A system reliability of 0.95 is required( � �.��). Use ARINC
method to find the reliability requirements for the subsystems.

Solution

Since the failure rates of the subsystems are constant, their

lifetimes are exponentially distributed and the lifetime
distribution of this series system is also exponential. Therefore
the ARINC method could be used:
1 1
� = � 0.���� = 0.���� , = 0.����
���
n
∗
= ,�w i  i /  i
i 1

�.���
�� � 0.����� w = 0.����� �
�.��� � 0.��� � 0.���
w = 0.����
To find the required failure rate() for this exponentially-
distributed-lifetime system we could write:

���� = e ⟹ 0.�� � � ⟹ = 0.������ ℎ �

�
443 Reliability Engineering

The required failure rate for each subsystem is calculated

∗
from Eq. 8-7 i.e. � as follows:

∗
= (0.�����0.������ = 0.�����

∗
= (0.�����0.������ = 0.������

∗
= (0.�����0.������ = 0.�������

Since the lifetime of each subsystem is exponentially

distributed, the allocated reliability for them to ensure a 20-hour
operation of the system are:

∗ (��� ∗ ( . )
=e =e = 0.��������������

∗ ���� ∗
=e = 0.���, R∗ (��� = 0.������ �

8-3-3 The AGREE allocation method for pseudo-series

system with independent exponential-lifetime subsystems

A method of apportionment is outlined by the Advisory

Group on the Reliability of Electronic Equipment (AGREE)
takes into consideration both the complexity and importance of
each subsystem. In this method for each subsystem a factor
called importace index is introduced to express the degree of
impotance between the system failure and the subsystem. It is
assumed that the subsystems have constant failure rates.
The method applies to any unit that can be decoposed into a
series ofindependent subsystems(Grosh,1989 p149); some of
Chap 8 Enhancement, Optimization & Allocation of Reliability 444

which are taken out of the system before the end of the mission
time considered for the system. Notice that the total system
under consideration is not truly a series system unless all of the
importance indices (wi's) equal unity and the mission time of all
subsystems ((ti's) are equal(Grosh,1989 p150).
To reach a target MTBF for the system, this method uses
Equation 8-8 which calculates an approximate value for the
MTBF of each subsystem(Grosh,1989 p150).
( )( )( )
���� � [ ( )]
= 1,2, … (8-8)

This is equivalent to
the following failure rate for the subsystem(K&L p409):
[ ∗( )]
� = 1,2, … (8-9-1)

or
the following reliability for the subsystem
∗
� )= (8-9-2)
where
Number of components of in the ith subsystem
N Total number of components in the system: N= ∑
� failure rate of ith subsystem
� � The required system reliability for a mission time t
∗
The reliability allocated to ith subsystem
System mission time���
The mission time for ith subsystem; the time period
required for the ith subsystem to operate from the

�
445 Reliability Engineering

beginning of the operation of the system and the

subsystem is not needed any more(� � ≤ )
Importance index for the ith subsystem; probability that
the system mission fails if ith subsystem fails.
is the following quotient(Grosh, 1989 p149)
����������������
�
���������� �����������������

�� states that for the successful operation of the

system, the ith subsystem must work successfully.
The more wi's closer to 1 the better the results. Small
wi's causes poor results by the AGREE formula .
The reliability allocated to the ith subsystem is calculated from:
∗
� (8-10-1)

or

∗ [ ( )]
� 1− (8-10-2)

Example 8-6(K&L page 410)

A system consisting of four subsystems is required to

demonstrate a reliability level of 0.95 for 10 hours of continuous
operation. Subsystems I and 3 are essential for the successful
operation of the system. Subsystem 2 has to function for only 9
hours for the operation of the system, and its importance factor
is 0.95. Subsystem 4 has an importance factor of 0.90 and must
function for 8 hours for the system to function. Solve the
Chap 8 Enhancement, Optimization & Allocation of Reliability 446

reliability allocation problem by AGREE method using the data

Given below:

1 10 1 15
2 9 0.95 25
3 10 1 100
4 8 0.9 70
sum N=210

Solution

Data: t=10 N = ∑ = 210 , (10)= 0.95

According to Eq. 8.8 the required mean lifetime for each
subsystem is:

= �
[− ( )]

The equivalent failure rate is � .

��� ∗ � ∗ ��
= � �����4 �
�����− 0.���

1 1
= = = 0.�������� � ����� � ��
�����4

��� ∗ �.�� ∗ �
= � �����2 = 0.�������
(����− 0.���

��� ∗ � ∗ ��
= � �������� = 0.������
����− 0.���

�
447 Reliability Engineering

��� ∗ �.9 ∗ 8
= � ����� = 0.������
(���(− 0.���

Eq. 8-10-1&2 allocates the required reliability to each subsystem:

∗ 1−[ ( )]
=1− �

∗
1 − (0.���
=1− = 0.������
1

∗ ∗ ( . )( )
��� = e =e = 0.�����

similarly

∗
1 − (0.���
��− � �.������
0.��

∗ ∗ ( . )( )
or � e =e = 0.�����
∗ (t)
� = 0.������,
∗ (t)
�� = 0.�����

These four reliabilities result in a reliability of

∗ ∗ ∗ ∗
� × × = 0.94788=94.79 % for the system which is
slightly less than the system reliability requirement 0.95. This is
a result of the approximate nature of the AGREE formula and
that ����� are less than unity.End of Example

The readers interested in AGREE method for parallel

configurations could refer to references such as Grosh (1989).
 Chap 8 Enhancement, Optimization & Allocation of Reliability 448

����������

1. We would like to design an active parallel system of n

identical subsystems. The lifetime of each subsystem is
exponentially distributed with average lifetime of 100 hours.
Prepare a mathematical model to determine n in such a way that
the system reliability is greater than 0.95 and the average system
life time in maximized. For calculating the reliabilities of
subsystems use mission time of 100 hours.

2. The monthly failure rates of the subsystems of a series system

are constant and their estimates are �50 × �� � 18 × �� � 2.3 ×
�� � 5.6 × �� �failure per month. Use ARINC technique to
assign reliabilities to the subsystems such that the system
reliability would be 0.98 for 36- month mission time.

3. (From K&L page433) A system consists of five subsystems

in series. The system reliability goal is 0.990 for 10 hours of
operation. The necessary information for the subsystem is given below

Number of � Operating
Subsystem No.� �� subsystems time
ni wi ti
1 25 1.00 10
2 80 0.97 9
3 45 1.00 10
4 60 0.93 7
5 70 1.00 10

�
449 Reliability Engineering

4. (K&L page 436) Consider the design reliability problem

when both the stress(s) and the strength ( ) are normally
distributed. The reliability goal for the component is 0.99. The
cost functions and the constrains for the 4 parameters are:

parameter� The terms of the cost function Constraints(units in MPa)�

.
μ � ������ � (μ ) � ����0 ≤ μ ≤ ������
.
σ � ��� � (σ ) � ���� ≤ σ ≤ ������
.
���� � ( ) � ����� ≤ ≤ �����
.
��� � ( ) � ��� ≤ ≤ ����
Formulate the model of this problem to determine the values
of ‫ �ﻭ‬, , in such a way that the sum of the terms in
second column is minimized and the constraints given in the
table are satisfied.

5. Solve Example 8-1 of this chapter assuming k=3 active

parallel components.

6. A system consists of four subsystems having constant failure

rates. The system will fail if a subsystem fails. The current
lifetimes of the subsystems are estimated to be
250, 142.75, 12 and 20 hours. Assign required reliability to
ensure the system will have a reliability of 0.99 for a mission
time of 50 hours.

You can never satisfy people by your

property. So, you can attract
their satisfaction by your behaviour
 450

�����������
American National Standard ,1989
Bazargan, Hamid,
Statistical methods in Quality control
Downloadable from:
https://opentextbc.ca/oerdiscipline/chapter/industrial-engineering or
https://archive.org/details/statistical-methods-august-2020-bazargan or
https://opentextbc.ca/oerdiscipline/chapter/statistics/�����
Bazovsky, Igor, 2004
Reliability Theory and Practice
Dover Publication Inc
Barlow,E.B.& Proschan F,L1996 �������
Math. Theory of Reliability,
S.I.A.M
Billinton, R, Allan, R.,1992
Reliability Evaluation of Engineering Systems: Concepts and Techniques
Plenum Publishers
Bowker, A., H. and. Leiberman, G. J.,�����
Engineering Statistics
Prentice Hall, Inc., Englewood Cliffs, N.J
Cabarbaye,A.2019
������� Implementation of accelerated life testing
Cab Innovation, France(www.cabinovation.com)
Carter,A.D.S., 1986
Mechanical Reliability
Macmillan
Chen, Qiming, 2004
"The probabibity, identification, and prevention of rare events in
power systems "
Iowa State univ. Retrospective Theses andDissertations.Paper 1149.
Coles , Stuart, 2001
An Introduction to Statistical Modeling of Extreme Values �����������
Springer Verlag, London.
Dhillon, B.S.2006
Maintainability, Maintenance, and Reliability for Engineers
CRC Press
Taylor & Francis Group
Dao-Thein, M.&Massoud,M.1974
On the Relation Between the Factor of Safety and Reliability
Jr of Engineering for Industry , 96(3): 853�857.
doi: https://doi.org/10.1115/1.3438452
https://citeseerx.ist.psu.edu/viewdoc/download?doi=
10.1.1.1058.9057&rep=rep1&type=pdf
Ebeling, Charles E.1997
Introduction to Reliability and Maintainability Engineering
McGraw-Hill

�
451 Reliability Engineering

Ebrahimi, Ghanbar, 1992,

Quality Control: "The assurance Sciences" (in Persian)
Tehran University Publications No. 2137, Tehran
Epstein,B,1960
Elements of the Theory of Extreme values
Technometrics vol2 #1 pp27-41
Faghih, N. 1996
Maintenave Engineering(Persian)
Navid Publicationm Shiraz, Iran
Eusgeld, F.C. Freiling, and R. Reussner (Eds.): 2008
Dependability Metrics,
LNCS 4909, pp. 59�103, 2008. c Springer-Verlag Berlin Heidelberg
Fegenbaum, A.V.,1991
Total Quality Control; McGraw-Hill
Goda, Y., 2000
Random Seas and Design of Maritime Structures.
2nd edition, World Scientific, Singapore.
Gordon, E. C.,1993
Signal and linear system analysis
Allied pulishers
Grant,E.G. , Leavenworth, E.,S., 1988
Statistical quality control
McGraw Hill,New York
Grosh, D.L ,1989
A Primer of Reliability Theory
Wiley
Guiqin Chen , a, Xiuling Wei, b, Gang Cui, c, Xinge Wang, d Lei Li, e,2013
Application of MATLAB in the Analysis of Reliability Data
Advanced Material Research vols 760-762 pp1004-1007
Hsyland,A.,Rausand,M., 2004
System Relaibility Theory: Models, Statistical Methods & Applications
Wiley
IEC 60050-192:2015
International Electrotechnical Vocabulary (IEV) - Part 192:
Dependability
IEC 60300-1:2014
Dendability management - Part 1: Guidance for management and application
IIE Terminology
Industrial Engineering Terminology
Institute if Industrial Engineers,McGraw-Hill
Ireson, W.G.,Coombs, C.F.,Moss, R.Y.1996
Handbook of Reliability Engineering and Management
McGraw Hill���
Jerkinson, A. F., 1955.
The Frequently Distribution of The Annual Maximum (or Minimum)
Values of ��Meteorological Elements.
Quarterly Jr. of The Royal Met. Soc. Vol. 81.
 452

Kapur, K.C.& Lamberson,L.R.

Reliability in Engineering Design,
John Wiley
Kalaiselvan,C., L. Bhaskara Rao 2016
Comparison of reliability techniques of parametric and non-parametric method
Engineering Science and Technology,Vol 19, Issue 2, pp 691-699
Kaushik,A,Singh,R,2021
An Introduction to Probability and Statistics
K.K publication
King, R,1990
Safety in Process Industries
Butterworth-Heinemann,
Kuo, W. and Prasad, V, R.2000
An annotated overview of system-reliability optimization,
IEEE Trans on Reliability 49 176-187.
Kuo, Way; Lin, Hsin-Hui; Xu, Zhongkai; Zhang, Weixing; 2012
Reliability Optimization with the Lagrange- Multiplier
and Branch- and-Bound Technique
Reliability, IEEE Transactions , Volume: R-36 Issue: 5
Kuo, Way; Zuo,M.J, 2003,
Optimal Relaibility Modelling � Principles and applications
John Wiley
Lewis,E.E.,1994
Introduction to Reliability Eng'g,
Wiley
Li, James, 2016
Reliability Comparative Evaluation of Active Redundancy vs.
Standby Redundancy
Int. Jr of Math,, Eng;g.and Manag.Sciences Vol. 1, No. 3, 122�129
Mann, N.R., Schafer ,R.E.,Singpurwala. 1974
Methods for Statistical. Analysis of Reliability &Life Data
Wiley
Mathworks�������������������
Martha de Souza, G. F., Netto,A.C., Melani,R.F., Michalski,M.A. ... Renan da Silva,2022
Reliability Analysis and Asset Management of Engineering Systems,
Elsevier Inc.
Nordmann, L., Pham, H., 1999
Weighted voting systems
IEEE Transactions on Reliability, 1999
Niaki S, T., Yaghoubi, A.,2020
Exact Equations for The Reliability and Mean Time to Failure Of
1-Out-Of-N Cold Standby System with Imperfect Switching
Journal of Optimization in Industrial Engineering Vol.14, Issue 2
O'Connor,P.D.,T.2003
Practical Reliability Engineering,
Wiley

�
453 Reliability Engineering

O'Connor, Patrick P, Kleyner, Andre 2012

Practical Reliability Engineering,
Wiley
Queiroz, I. M., 2016
Comparison between Deterministic and Probabilistic
Stability Analysis, Featuring Consequent Risk Assessment
International Jr of Geotechnical and Geological Eng'g Vol:10, No:6,
Ravinran, A. Ravi,(Editor),2016
Operations Research and Management Science Handbook
CRC Prress
Reuben,R,1994
Materials in Marine Technology
Springer-Verlag
Ross,S.M.1985
Introduction to probability models,
Academic Press Edition
Salvendy, Gavriel( Editor),2001
Handbook of Industrial Eng'g: Technology and Operations Management
John Wiley
Sharpe, William, 2008
Handbook of Experimental Solid Mechanics
Springer Science & Business Media
Smith, D.J.1993
Reliability. Maintainability & Risk
Butter worth &Heinemann
Shooman, M.L ,2002
Reliability of Computer systems and networks
John Wiley
Song, J. , Kiureghian ,A.D. 2003,
Bounds on System Reliability by Linear Programming,
J. Eng. Mech. 129:627-636.
Stamatis,D.H. 2010
The OEE Primer: Understanding Overall Equipment
Effectiveness, Reliability, and Maintainability
CRC press
Stapelberg, R.F.,2009
Handbook of Reliabi., Availab., Maintain. and Safety in Eng'g Design
Springer Science & Business Media
Tersine, R., J., 1985
Production and Operations Management
Elsevier Science publishing Co., Amsterdam
Thompson, W.A1988. �����
Point Process Models with Applications to Safety and Reliability
Chapman and Hall,
Tobias, P,A, Trindade ,D,C, 2012
Applied Reliability
CRC Press
 454

Wang Wendai, Jiang, Mingxiao, 2004

Generalized decomposition method for complex systems
Publisher: IEEE,conference : 26-29 Jan. 2004 Los Angeles, CA, USA
DOI: 10.1109/RAMS.2004.1285416
Wei-Chang,Leh,2009
A simple universal generating function method for estimating the
reliability of general multi-state node networks
IIE Transactions Volume 41, 2008 - Issue 1:Special Issue Honoring
Richard Barlow
Wenxue Qian, Xiaowei Yin, Liyang Xie, 2014
Reliability Modeling and Assessment of Component with Multiple Weak
Sites under Complex Loading
Mathematical Problems in Engineering Volume 2014
Xie, M, Gaudoin, O.,BracQuemond, C, 2002
Redefinning, failure rate function for discrete distributions
International Journal of Reliability, Quality and Safety Engineering
Vol. 9, No. 3 pp 275�285
Xie, Xin, Zhai, Hao,2021
Structural reliability optimization model with �
Evolutionary Intelligence , springer Published on line March 2021
https://doi.org/10.1007/s12065-021-00588-9
Yi-Chic Hsieh. 2002
A two phase LP for redundancy allocation problem
Yogoslav Jr of O.R.

�����������������������������������
�

�
455 Reliability Engineering

�
�����������������
�

�
�������������������������������������������������������( Mood et al,1974 ) Example � � � � . ���������������������������������

 1� 2 3 4 5 6 7 8 9 10 12 15 20 30� 60 120 
.10 39.9 49.5 53.6 55.8 57.2 58.2 58.9 59.4 59.9 60.2 60.7 61.2 61.7 62.3 62.8 63.1 63.3
.05 161 200 216 225 230 234 237 239 241 242 244 246 248 250 252 253 254�
.02 5 648 800 864 900 922 937 948 957 963 969 977 985 993 1000 1010 1010 1020
.01� 1 4050 5000 5400 5620 5760 5860 5930 5980 6020 6060 6110 6160 6210 6260 6310 6340 6370
.00 5 16200 20000 21600 22500 23100 23400 23700 23900 24100 24200 24400 24600 24800 25000 25200 25400 25500�
.10� 8.53 9.00 9.16 9.24 9.29 9.33 9.35 9.37 9.38 9.39 9.41 9.42� 9.44 9.46 9.47 9.48 9.49
.05 18.5 19.0 19.2 19.2 19.3 19.3 19.4 19.4 19.4 19.4 19.4 19.4 19.5 19.5 19.5 19.5 19.5
.02 5 38.5 39.0 39.2 39.2 39.3 39.3 39.4 39.4 39.4 39.4 39.4 39.4 39.4 39.5 39.5 39.5 39.5
2
.01� 98.5 99.0 99.2 99.2 99.3 99.3 99.4 99.4 99.4 99.4 99.4 99.4 99.4 99.5 99.5 99.5 99.5
.00 5 199 199 199 199 199 199 199 199 199 199 199 199 199 199 199 199 199
.10 5.54 5.46 5.39 5.34 5.31 5.28 5.27 5.25 5.24 5.23� 5.22 5.20 5.18 5.17 5.15 5.14 5.13
.05 10.1 9.55 9.28 9.12 9.01� 8.94 8.89 8.85 8.81 8.79 8.74 8.70 8.66 8.62 8.57 8.55 8.53
.02 5 17.4 16.0 15.4 15.1 14.9 14.7 14.6 14.5 14.5 14.4 14.3 14.3 14.2 14.1 14.0 13.9 13.9
3
.01� 34 .1 30.8 29.5 28.7 28.2 27.9 27.7 27.5 27.3 27.2 27.1 26.9 26.7 26.5 26.3 26.2 26.1
.00 5 55.6 49.8 47.5 46.2 45.4 44.8 44.4 44.1 43.9 43.7 43.4 43.1 42.8 42.5 42.1 42.0 41.8
.10� 4.54 4.32 4.19 4.11 4.05 4.01 3.98 3.95 3.93 3.92 3.90 3.87 3.84 3.82 3.79 3.78 3.76
.05 7.71 6.94 6.59 6.39 6.26 6.16 6.09 6.04 6.00 5.96 5.91 5.86 5.80 5.75 5.69 5.66 5.63
.02 5 12.2 10.6 9.98 9.60 9.36 9.20 9.07 8.98 8.90 8.84 8.75 8.66 8.56 8.46 8.3& 8.31 8.26
.01� 4 21.2 18.0 16.7 16.0 15.5 15.2 15.0 14.8 14.7 14.5 14.4 14.2 14.0 13.8 13.7 13.6 13.5

.00 5 31.3 26.3 24.3 23.2 22.5 22.0 21.6 21.4 21.1 21.0 20.7 20.4 20.2 19.9 19.6 19.5 19.3
457 ���Reliability Engineering
Table A� -continued
�
�
.10 4.06� 3.78 3.62 3.52 3.45 3.40 3.37 3.34 3.32 3.30 3.27 3.24 3.21 3.17 3.14 3.12 3.11
.05 6.61 5.79 5.41 5.19 5.05 4.95 4.88 4.82 4.77 4.74 4.68 4.62 4.56 4.50 4.43 4.40 4.37
.02 5 10.0 8.43 7.76 7.39 7.15 6.98 6.85 6.76 6.68 6.62 6.52 6.43 6.33 6.23 6.12 6.07 6.02
5
.01� 16.3 13.3 12.1 11.4 11.0 10.7 10.5 10.3 10.2 10.1 9.89 9.72 9.55 9.38 9.20 9.11 9.02
.00 5 22.8 18.3 16.5 15.6 14.9 14.5 14.2 14.0 13.8 13.6 13.4 13.1 12.9 12.7 12.4 12.3 12.1
.10 3.78 3.46 3.29 3.18 3.11 3.05 3.01 2.98 2.96 2.94 2.90 2.87 2.84 2.80 2.76 2.74 2.72
.05 5.99 5.14 4.76 4.53� 4.39 4.28 4.21 4.15 4.10 4.06 4.00 3.94 3.87 3.81 3.74 3.70 3.67
.02 5 8.81 7.26 6.60 6.23 5.99 5.82 5.70 5.60 5.52 5.46 5.37 5.27 5.17 5.07 4.96 4.90 4.85
6
.01� 13.7 10.9 9.78 9.15 8.75 8.47 8.26 8.10 7.98 7.87 7.72 7.56 7.40 7.23 7.06 6.97 6.88
.00 5 18.6 14.5 12.9 12.0 11.5 11.1 10.8 10.6 10.4 10.2 10.0 9.81 9.59 9.36 9.12 9.00 8.88
.10 3.59 3.26 3.07 2.96 2.88 2.83 2.78 2.75 2.72 2.70 2.67 2 . 63 2.59 2.56 2.51 2.49 2.47�
.05 5.59 4.74 4.35 4.12 3.97 3.87 3.79 3.73 3.68 3.64 3.57 3.51 3.44 3.38 3.30 3.27 3.23
.02 5 8.07 6.54 5.89 5.52 5.29 5.12 4.99 4.90 4.82 4.76 4.67 4.57 4.47 4.36 4.25 4.20 4.14
.01� 7 12.2� 9.55 8.45 7.85 7.46 7.19 6.99 6.84 6.72 6.62 6.47 6.31 6.16 5.99� 5.82 5.74 5.65
.00 5 16.2 12.4 10.9 10.1 9.52 9.16 8.89 8.68 8.51 8.38 8.18 7.97 7.75 7.53 7.31 7.19 7.08
.10 3.46 3.11 2.92 2.81 2.73 2.67 2.62 2.59 2.56 2.54 2.50 2.46 2.42 2.38 2.34 2.31 2.29
.05 5.32 4.46 4.07 3.84 3.69 3.58 3.50 3.44 3.39 3.35 3.28 3.22 3.15 3.08 3.01 2.97 2.93
.02 5 7.57 6.06 5.42 5.05 4.82 4.65 4.53 4.43 4.36 4.30 4.20 4.10 4.00 3.89 3.78 3.73 3.67
.01� 8� 11.3 8.65 7.59 7.01 6.63 6.37 6.18 6.03 5.91 5.81 5.67 5.52 5.36 5.20 5.03 4.95 4.86
.00 5� 14.7 11.0 9.60 8.81 8.30 7.95 7.69 7.50 7.34 7.21 7.01 6.81� 6.61 6.40 6.18 6.06� 5.95�

�
 458

Table A -continued Fm,n������e.g. � � , , = 7.13�

 1� 2 3 4 5 6 7 8� 9 10 12 15 20� 30 60 120 ∞
.10 3.36 3.01 2.81 2.69 2.61 2.55 2.51 2.47 2.44 2.42 2.38 2.34 2.30 2.25 2.21 2.18 2.16
.05 5.12 4.26 3.86 3.63 3.48 3.37 3.29 3.23 3.18 3.14 3.07 3.01 2.94 2.86 2.79 2.75 2.71
.02 5 7.21 5.71 5.08 4.72 4.48 4.32 4.20 4.10� 4.03 3.96 3.87 3.77 3.67 3.56 3.45 3.39 3.33
.01 9 10 .6 8.02 6.99 6.42 6.06 5.80 5.61 5.47 5.35 5.26 5.11 4.96 4.81 4.65 4.48 4.4O 4.31
.005 13.6 10.1 8.72 7.96 7.47 7.13 6.88 6.69 6.54 6.42 6.23 6.03 5.83 5.62 5.41 5.30 5.19�
.10 3.29 2.92 2.73 2.61 2.52 2.46 2.41 2.38 2.35 2.32 2.28 2.24 2.20 2.15 2.11 2.08 2.06
.05 4.96 4.10 3.71 3.48 3.33 3.22 3.14 3.07 3.02 2.98 2.91 2.84 2.77 2.70 2.62 2.58 2.54
.025 10
6.94 5.46 4.83 4.47 4.24 4.07 3.95 3.85 3.78 3.72 3.62 3.52 3.42 3.31 3.20 3.14 3.08
.01� 10.0 7.56 6.55 5.99 5.64 5.39 5.20 5.06 4.94 4.85 4.71 4.56 4.41 4.25 4.08 4.00 3.91
.005 12.8 9.43 8.08 7.34 6.87 6.54 6.30 6.12 5.97 5.85 5.66 5.47 5.27 5.07 4.86 4.75 4.64
.10 3.18 2 81 2.61 2.48 2 39 2.33 2.28 2.24 2.21 2.19 2.15 2.10 2.06 2 .01 1.96 1.93 1.90
.05 4.75 3.89 3.49 3.26 3.11 3.00 2.91 2.85 2.80 2.75 2.69 2.62 2.54 2.47 2.38 2.34 2.30
.025 12
6.55 5.10 4.47 4.12 3.89 3.73 3.61 3.51 3.44 3.37 3.28 3.18 3.07 2.96 2.85 2.79 2.72
.01 9.33 6.93 5.95 5.41 5.06 4.82 4.64 4.50 4.39 4.30 4.16 4.01 3.86 3.70 3.54 3.45 3.36
.005 11.8 8.51 7.23 6.52 6.07 5.76 5.52 5.35 5.20 5.09 4.91 4.72 4.53 4.33 4.12 4.01 3.90
.10 3.07 2.70 2.49 2.36 2.27 2.21 2:16 2.12 2.09 2.06 2.02 1.97 1.92 1.87 1.82 1.79 1.76
.05 4.54 3.68 3.29 3.06 2.90 2.79 2.71 2.64 2.59 2.54 2.48 2.40 2.33 2.25 2.16 2.11 2.07
.025 6.20 4.77 4.15 3.80 3.58 3.41 3.29 3.20 3.12 3.06 2.96 2.86 2.76 2.64 2.52 2.46 2.40
.01� 15 8.68 6.36 5.42 4.89 4.56 4.32 4.14 4.00 3.89 3.80 3.67 3.52 3.37 3.21 3.05 2.96 2.87

.005 10.8 7.70 6.48 5.80 5.37 5.07 4.85 4.67 4.54 4.42 4.25 4.07 3.88 3.69 3.48 3.37 3.26

�
459 ���Reliability Engineering
�
Table A -continued �

.10 2.97 2.59 2.38 2.25 2.16 2.09 2.04 2.00 1.96 1.94 1.89 1.84 1.79 1.74 1.68 1.64 1.61
.05 4.35 3.49 3.10 2.87 2.71 2.60 2.51 2.45 2.39 2.35 2.28 2.20 2.12 2.04 1.95 1.90 1.84
.025 5.87 4.46 3.86 3.51 3.29 3.13 3.01 2.91 2.84 2.77 2.68 2.57 2.46 2.35 2.22 2.16 2.09
�
.01� 20� 8.10 5.85 4.94 4.43 4.10 3.87 3.70 3.56 3.46 3.37 3.23 3.09 2.94 2.78 2.61 2.52 2.42
.005 9.94 6.99 5.82 5.17 4.76 4.47 4.26 4.09 3.96 3.85 3.68 3.50 3.32 3.12 2.92 2.81 2.69
.10 2.88 2.49 2.28 2.14 2.05 1.98 1.93 1.88 1.85 1.82 1.77 1.72 1.67 1.61 1.54 1.50 1.46
.05 4.17 3.32 2.92 2.69 2.53 2.42 2.33 2.27 2.21 2.16 2.09 2.01 1.93 1.84 1.74 1.68 1.62
.025
� 5.57 4.18 3.59 3.25 3.03 2.87 2.75 2.65 2.57 2.51 2.41 2.31 2.20 2.07 1.94 1.87 1.79
.01� 30 7.56 5.39 4.51 4.02 3.70 3.47 3.30 3.17 3.07 2.98 2.84 2.70 2.55 2.39 2.21 2.11 2.01
.005 9.18 6.35 5.24� 4.62 4.23 3.95 3.74 3.58 3.45 3.34 3.18 3.01 2.82 2.63 2.42 2.30 2.18
.10 2.79 2.39 2.18 2.04 1.95 1.87 1.82 1.77 1.74 1.71 1.66 1.60 1.54 1.48 1.40 1.35 1.29
.05 4.00 3.15 2.76 2.53 2.37 2.25 2.17 2.10 2.04 1.99 1.92 1.84 1.75 1.65 1.53 1.47 1.39
.025 5.29 3.93 3.34 3.01 2.79 2.63 2.51 2.41 2.33 2.27 2.17 2.06 1.94 1.82 1.67 1.58 1.48
.01� 60 7.08 4.98 4.13 3.65 3.34 3.12 2.95 2.82 2.72 2.63 2.50 2.35 2.20 2.03 1.84 1.73 1.60
.005 8.49 5.80 4.73 4.14 3.76 3.49 3.29 3.13 3.01 2.90 2.74 2.57 2.39 2.19 1.96 1.83 1.69
.10� 2.75 2.35 2.13 1.99 1.90 1.82 1.77 1.72 1.68 1.65 1.60 1.54 1.48 1.41 1.32 1.26 1.19
.05 3.92 3.07 2.68 2.45 2.29 2.18 2.09 2.02 1.96 1.91 1.83 1.75 1.66 1.55 1.43 1.35 1.25
.025 5.15 3.80 3.23 2.89 2.67 2.52 2.39 2.30 2.22 2.16 2.05 1.94 1.82 1.69 1.63 1.43 1.31
.01 120 6.85 4.79 3.95 3.48 3.17 2.96 2.79 2.66 2.56 2.47 2.34 2.19 2.03 1.86 1.66 1.53 1.38
.005 8.18 5.54 4.50 3.92 3.55 3.28 3.09 2.93 2.81 2.71 2.54 2.37 2.19 1.98 1.75 1.61 1.43
.10 2.71 2.30 2.08 1.94 1.85 1.77 1.72 1.67 1.63 1.60 1.55 1.49 1.42 1.34 1.24 1.17 1.00
.05� 3.84 3.00 2. 60 2.37 2. 21 2.10 2.01 1.94 1.88 1.83 1.75 1.67 1.57 1.46 1.32 1.22 1.00
.025� 5.02 3.69 3.12 2.79 2.57 2.41 2.29 2.19 2.11 2.05 1.94 1.83 1.71 1.57 1.39 1.27 1.00
.01�  6.63 4.61 3.78 3.32 3.02 2.80 2.64 2.51 2.41 2.32 2.18 2.04 1.88 1.70 1.47 1.32 1.00
.005� 7.88 5.30 4.28� 3.72 3.35 3.09 2.90 2.74 2.62 2.52 2.36 2.19 2.00 1.79 1.53 1.36 1.00
 460
λ ������������������������������������������������������� Pr(X  k ) �
or k
np 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
0.01 0.990 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000

0.10 0.905 0.995 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000

0.20 0.819 0.982 0.999 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000

0.30 0.741 0.963 0.996 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
0.40 0.670 0.938 0.992 0.999 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
0.80 0.449 0.809 0.953 0.991 0.999 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000
1.00 0.368 0.736 0.920 0.981 0.996 0.999 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000 1.000

2.00 0.135 0.406 0.677 0.857 0.947 0.983 0.995 0.999 1.000 1.000 1.000 1.000 1.000 1.000 1.000

3.00 0.050 0.199 0.423 0.647 0.815 0.916 0.966 0.988 0.996 0.999 1.000 1.000 1.000 1.000 1.000

4.00 0.018 0.092 0.238 0.433 0.629 0.785 0.889 0.949 0.979 0.992 0.997 0.999 1.000 1.000 1.000

5.00 0.007 0.040 0.125 0.265 0.440 0.616 0.762 0.867 0.932 0.968 0.990 0.995 0.998 0.999 1.000

6.00 0.002 0.017 0.062 0.151 0.285 0.446 0.606 0.744 0.847 0.916 0.960 0.980 0.991 0.996 0.999

6.20 0.002 0.015 0.054 0.134 0.259 0.414 0.574 0.716 0.826 0.902 0.950 0.975 0.989 0.995 0.998
66 40 0 002 0 012 0 046 0 119 0 235 0 384 0 542 0 687 0 803 0 886 0 940 0 969 0 986 0 994 0 997
60 0 001 0 010 0 040 0 105 0 213 0 355 0 511 0 658 0 780 0 869 0 930 0 963 0 982 0 992 0 997

�
461 ���Reliability Engineering
λ ������������������������������������������������������� Pr(X  k ) �
or k
np 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
7.00 0.001 0.007 0.030 0.082 0.173 0.301 0.450 0.599 0.729 0.830 0.900 0.947 0.973 0.987 0.994

8.00 0.000 0.003 0.014 0.042 0.100 0.191 0.313 0.453 0.593 0.717 0.820 0.888 0.936 0.966 0.983

9.00 0.000 0.001 0.006 0.021 0.055 0.116 0.207 0.324 0.456 0.587 0.710 0.803 0.876 0.926 0.959

10.00 0.000 0.000 0.003 0.010 0.029 0.067 0.130 0.220 0.333 0.458 0.580 0.697 0.792 0.864 0.917

11.00 0.000 0.000 0.001 0.005 0.015 0.038 0.079 0.143 0.232 0.341 0.460 0.579 0.689 0.781 0.854
12.00 0.000 0.000 0.001 0.002 0.008 0.020 0.046 0.090 0.155 0.242 0.350 0.462 0.576 0.682 0.772
12.50 0.000 0.000 0.000 0.002 0.005 0.015 0.035 0.070 0.125 0.201 0.300 0.406 0.519 0.628 0.725
13.00 0.000 0.000 0.000 0.001 0.004 0.011 0.026 0.054 0.100 0.166 0.250 0.353 0.463 0.573 0.675
13.50 0.000 0.000 0.000 0.001 0.003 0.008 0.019 0.041 0.079 0.135 0.210 0.304 0.409 0.518 0.623
14.00 0.000 0.000 0.000 0.000 0.002 0.006 0.014 0.032 0.062 0.109 0.180 0.260 0.358 0.464 0.570
14.50 0.000 0.000 0.000 0.000 0.001 0.004 0.010 0.024 0.048 0.088 0.140 0.220 0.311 0.413 0.518
15.00 0.000 0.000 0.000 0.000 0.001 0.003 0.008 0.018 0.037 0.070 0.120 0.185 0.268 0.363 0.466
15.50 0.000 0.000 0.000 0.000 0.001 0.002 0.006 0.013 0.029 0.055 0.100 0.154 0.228 0.317 0.415
16.00 0.000 0.000 0.000 0.000 0.000 0.001 0.004 0.010 0.022 0.043 0.080 0.127 0.193 0.275 0.368
16.50 0.000 0.000 0.000 0.000 0.000 0.001 0.003 0.007 0.017 0.034 0.060 0.104 0.162 0.236 0.323
17.00 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.005 0.013 0.026 0.050 0.085 0.135 0.201 0.281
17.50 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.004 0.009 0.020 0.040 0.068 0.112 0.170 0.243

18.00 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.003 0.007 0.015 0.030 0.055 0.092 0.143 0.208
 462
λ ������������������������������������������������������� Pr(X  k ) �
or k
np 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
18.50 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.005 0.012 0.020 0.044 0.075 0.119 0.177
19.00 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.004 0.009 0.020 0.035 0.061 0.098 0.150
19.50 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.003 0.007 0.010 0.027 0.049 0.081 0.126
20.00 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.005 0.010 0.021 0.039 0.066 0.105

20.50 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.004 0.010 0.017 0.031 0.054 0.087
21.00 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.003 0.010 0.013 0.025 0.043 0.072
21.50 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.005 0.010 0.019 0.035 0.059
22.00 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.004 0.008 0.015 0.028 0.048
22.50 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.003 0.006 0.012 0.022 0.039

23.00 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.004 0.009 0.017 0.031
23.50 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.003 0.007 0.014 0.025
24.00 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.003 0.005 0.011 0.020
24.50 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.001 0.002 0.004 0.008 0.016

�
463 ���Reliability Engineering
Table C Area under standard normal curve����Z < )) e.g. ��� < −3.��� = 0.������ �
x  �
z 
 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0.00
-3.5 0.00017 0.00017 0.00018 0.00019 0.00019 0.0002 0.00021 0.00022 0.00022 0.00023
-3.4 0.00024 0.00025 0.00026 0.00027 0.00028 0.00029 0.0003 0.00031 0.00032 0.00034
-3.3 0.00035 0.00036 0.00038 0.00039 0.0004 0.00042 0.00043 0.00045 0.00047 0.00048
-3.2 0.0005 0.00052 0.00054 0.00056 0.00058 0.0006 0.00062 0.00064 0.00066 0.00069
-3.1 0.00071 0.00074 0.00076 0.00079 0.00082 0.00084 0.00087 0.0009 0.00094 0.00097
-3 0.001 0.00104 0.00107 0.00111 0.00114 0.00118 0.00122 0.00126 0.00131 0.00135
-2.9 0.00139 0.00144 0.00149 0.00154 0.00159 0.00164 0.00169 0.00175 0.00181 0.00187
-2.8 0.00193 0.00199 0.00205 0.00212 0.00219 0.00226 0.00233 0.0024 0.00248 0.00256
-2.7 0.00264 0.00272 0.0028 0.00289 0.00298 0.00307 0.00317 0.00326 0.00336 0.00347
-2.6 0.00357 0.00368 0.00379 0.00391 0.00402 0.00415 0.00427 0.0044 0.00453 0.00466
-2.5 0.0048 0.00494 0.00508 0.00523 0.00539 0.00554 0.0057 0.00587 0.00604 0.00621
-2.4 0.00639 0.00657 0.00676 0.00695 0.00714 0.00734 0.00755 0.00776 0.00798 0.0082
-2.3 0.00842 0.00866 0.00889 0.00914 0.00939 0.00964 0.0099 0.01017 0.01044 0.01072
-2.2 0.01101 0.01130 0.0116 0.01191 0.01222 0.01255 0.01287 0.01321 0.01355 0.01390
-2.1 0.01426 0.01463 0.015 0.01539 0.01578 0.01618 0.01659 0.01700 0.01743 0.01786
-2 0.01831 0.01876 0.01923 0.0197 0.02018 0.02068 0.02118 0.02169 0.02222 0.02275
-1.9 0.0233 0.02385 0.02442 0.025 0.02559 0.02619 0.0268 0.02743 0.02807 0.02872
-1.8 0.02938 0.03005 0.03074 0.03144 0.03216 0.03288 0.03362 0.03438 0.03515 0.03593
-1.7 0.03673 0.03754 0.03836 0.0392 0.04006 0.04093 0.04182 0.04272 0.04363 0.04457
 464

Table C continued e.g. :���Z < −1.��� = 0.������� ��

Z 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.02 0.01 0.00
-1.6 0.04551 0.04648 0.04746 0.04846 0.04947 0.0505 0.05155 0.05262 0.0537 0.0548
-1.5 0.05592 0.05705 0.05821 0.05938 0.06057 0.06178 0.06301 0.06426 0.06552 0.06681
-1.4 0.06811 0.06944 0.07078 0.07215 0.07353 0.07493 0.07636 0.0778 0.07927 0.08076
-1.3 0.08226 0.08379 0.08534 0.08691 0.08851 0.09012 0.09176 0.09342 0.0951 0.0968
-1.2 0.09853 0.10027 0.10204 0.10383 0.10565 0.10749 0.10935 0.11123 0.11314 0.11507
-1.1 0.11702 0.119 0.121 0.12302 0.12507 0.12714 0.12924 0.13136 0.1335 0.13567
-1� 0.13786 0.14007 0.14231 0.14457 0.14686 0.14917 0.15151 0.15386 0.15625 0.15866
-0.9 0.16109 0.16354 0.16602 0.16853 0.17106 0.17361 0.17619 0.17879 0.18141 0.18406
-0.8 0.18673 0.18943 0.19215 0.19489 0.19766 0.20045 0.20327 0.20611 0.20897 0.21186
-0.7 0.21476 0.2177 0.22065 0.22363 0.22663 0.22965 0.2327 0.23576 0.23885 0.24196
-0.6 0.2451 0.24825 0.25143 0.25463 0.25785 0.26109 0.26435 0.26763 0.27093 0.27425
-0.5 0.2776 0.28096 0.28434 0.28774 0.29116 0.2946 0.29806 0.30153 0.30503 0.30854
-0.4 0.31207 0.31561 0.31918 0.32276 0.32636 0.32997 0.3336 0.33724 0.3409 0.34458
-0.3 0.34827 0.35197 0.35569 0.35942 0.36317 0.36693 0.3707 0.37448 0.37828 0.38209
-0.2 0.38591 0.38974 0.39358 0.39743 0.40129 0.40517 0.40905 0.41294 0.41683 0.42074
-0.1 0.42465 0.42858 0.43251 0.43644 0.44038 0.44433 0.44828 0.45224 0.4562 0.46017
0 0.46414 0.46812 0.4721 0.47608 0.48006 0.48405 0.48803 0.49202 0.49601 0.5

�
 Table C continued e.g.������ < 1.��� = 0.������
465 ���Reliability Engineering
� 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
�
0 0.5 0.50399 0.50798 0.51197 0.51595 0.51994 0.52392 0.5279 0.53188 0.53586
0.1 0.53983 0.5438 0.54776 0.55172 0.55567 0.55962 0.56356 0.56749 0.57142 0.57535
0.2 0.57926 0.58317 0.58706 0.59095 0.59483 0.59871 0.60257 0.60642 0.61026 0.61409
0.3 0.61791 0.62172 0.62552 0.6293 0.63307 0.63683 0.64058 0.64431 0.64803 0.65173
0.4 0.65542 0.6591 0.66276 0.6664 0.67003 0.67364 0.67724 0.68082 0.68439 0.68793
0.5 0.69146 0.69497 0.69847 0.70194 0.7054 0.70884 0.71226 0.71566 0.71904 0.7224
0.6 0.72575 0.72907 0.73237 0.73565 0.73891 0.74215 0.74537 0.74857 0.75175 0.7549
0.7 0.75804 0.76115 0.76424 0.7673 0.77035 0.77337 0.77637 0.77935 0.7823 0.78524
0.8 0.78814 0.79103 0.79389 0.79673 0.79955 0.80234 0.80511 0.80785 0.81057 0.81327
0.9 0.81594 0.81859 0.82121 0.82381 0.82639 0.82894 0.83147 0.83398 0.83646 0.83891
1 0.84134 0.84375 0.84614 0.84849 0.85083 0.85314 0.85543 0.85769 0.85993 0.86214
1.1 0.86433 0.8665 0.86864 0.87076 0.87286 0.87493 0.87698 0.879 0.881 0.88298
1.2 0.88493 0.88686 0.88877 0.89065 0.89251 0.89435 0.89617 0.89796 0.89973 0.90147
1.3 0.9032 0.9049 0.90658 0.90824 0.90988 0.91149 0.91309 0.91466 0.91621 0.91774
1.4 0.91924 0.92073 0.9222 0.92364 0.92507 0.92647 0.92785 0.92922 0.93056 0.93189
1.5 0.93319 0.93448 0.93574 0.93699 0.93822 0.93943 0.94062 0.94179 0.94295 0.94408
1.6 0.9452 0.9463 0.94738 0.94845 0.9495 0.95053 0.95154 0.95254 0.95352 0.95449
1.7 0.95543 0.95637 0.95728 0.95818 0.95907 0.95994 0.9608 0.96164 0.96246 0.96327
1.8 0.96407 0.96485 0.96562 0.96638 0.96712 0.96784 0.96856 0.96926 0.96995 0.97062
 466

Table C continued e.g. ���Z < 3.��� = 0.������

Z 0.00 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
1.9 0.97128 0.97193 0.97257 0.9732 0.97381 0.97441 0.975 0.97558 0.97615 0.9767
2 0.97725 0.97778 0.97831 0.97882 0.97932 0.97982 0.9803 0.98077 0.98124 0.98169
2.1 0.98214 0.98257 0.983 0.98341 0.98382 0.98422 0.98461 0.985 0.98537 0.98574
2.2 0.9861 0.98645 0.98679 0.98713 0.98745 0.98778 0.98809 0.9884 0.9887 0.98899
2.3 0.98928 0.98956 0.98983 0.9901 0.99036 0.99061 0.99086 0.99111 0.99134 0.99158
2.4 0.9918 0.99202 0.99224 0.99245 0.99266 0.99286 0.99305 0.99324 0.99343 0.99361
2.5 0.99379 0.99396 0.99413 0.9943 0.99446 0.99461 0.99477 0.99492 0.99506 0.9952
2.6 0.99534 0.99547 0.9956 0.99573 0.99585 0.99598 0.99609 0.99621 0.99632 0.99643
2.7 0.99653 0.99664 0.99674 0.99683 0.99693 0.99702 0.99711 0.9972 0.99728 0.99736
2.8 0.99744 0.99752 0.9976 0.99767 0.99774 0.99781 0.99788 0.99795 0.99801 0.99807
2.9 0.99813 0.99819 0.99825 0.99831 0.99836 0.99841 0.99846 0.99851 0.99856 0.99861
3 0.99865 0.99869 0.99874 0.99878 0.99882 0.99886 0.99889 0.99893 0.99896 0.999
3.1 0.99903 0.99906 0.9991 0.99913 0.99916 0.99918 0.99921 0.99924 0.99926 0.99929
3.2 0.99931 0.99934 0.99936 0.99938 0.9994 0.99942 0.99944 0.99946 0.99948 0.9995
3.3 0.99952 0.99953 0.99955 0.99957 0.99958 0.9996 0.99961 0.99962 0.99964 0.99965
3.4 0.99966 0.99968 0.99969 0.9997 0.99971 0.99972 0.99973 0.99974 0.99975 0.99976
3.5 0.99977 0.99978 0.99978 0.99979 0.9998 0.99981 0.99981 0.99982 0.99983 0.99983

�
467 ���Reliability Engineering

������������������������������������������������ � e.g.� � = 1.��

Z 0 0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09
0 0.5 0.49601 0.49202 0.48803 0.48405 0.48006 0.47608 0.4721 0.46812 0.46414
0.1 0.46017 0.4562 0.45224 0.44828 0.44433 0.44038 0.43644 0.43251 0.42858 0.42465
0.2 0.42074 0.41683 0.41294 0.40905 0.40517 0.40129 0.39743 0.39358 0.38974 0.38591
0.3 0.38209 0.37828 0.37448 0.3707 0.36693 0.36317 0.35942 0.35569 0.35197 0.34827
0.4 0.34458 0.3409 0.33724 0.3336 0.32997 0.32636 0.32276 0.31918 0.31561 0.31207
0.5 0.30854 0.30503 0.30153 0.29806 0.2946 0.29116 0.28774 0.28434 0.28096 0.2776
0.6 0.27425 0.27093 0.26763 0.26435 0.26109 0.25785 0.25463 0.25143 0.24825 0.2451�
0.7 0.24196 0.23885 0.23576 0.2327 0.22965 0.22663 0.22363 0.22065 0.2177 0.21476
0.8 0.21186 0.20897 0.20611 0.20327 0.20045 0.19766 0.19489 0.19215 0.18943 0.18673
0.9 0.18406 0.18141 0.17879 0.17619 0.17361 0.17106 0.16853 0.16602 0.16354 0.16109
1 0.15866 0.15625 0.15386 0.15151 0.14917 0.14686 0.14457 0.14231 0.14007 0.13786
1.1 0.13567 0.1335 0.13136 0.12924 0.12714 0.12507 0.12302 0.121 0.119 0.11702
1.2 0.11507 0.11314 0.11123 0.10935 0.10749 0.10565 0.10383 0.10204 0.10027 0.09853
1.3 0.0968 0.0951 0.09342 0.09176 0.09012 0.08851 0.08691 0.08534 0.08379 0.08226
1.4 0.08076 0.07927 0.0778 0.07636 0.07493 0.07353 0.07215 0.07078 0.06944 0.06811
1.5 0.06681 0.06552 0.06426 0.06301 0.06178 0.06057 0.05938 0.05821 0.05705 0.05592
1.6 0.0548 0.0537 0.05262 0.05155 0.0505 0.04947 0.04846 0.04746 0.04648 0.04551
�
 468
Table D Crtical values of standard normal continued e.g for � �.�� � � = 1.96 �

Z 0 0.01 0.02 0.03 0.04 0.05� 0.06 0.07 0.08 0.09

1.8 0.03593 0.03515 0.03438 0.03362 0.03288 0.03216 0.03144 0.03074 0.03005 0.02938
1.9 0.02872 0.02807 0.02743 0.0268 0.02619 0.02559 0.025 0.02442 0.02385 0.0233
2 0.02275 0.02222 0.02169 0.02118 0.02068 0.02018 0.0197 0.01923 0.01876 0.01831
2.1 0.01786 0.01743 0.017 0.01659 0.01618 0.01578 0.01539 0.015 0.01463 0.01426
2.2 0.0139 0.01355 0.01321 0.01287 0.01255 0.01222 0.01191 0.0116 0.0113� 0.01101 �

2.3 0.01072 0.01044 0.01017 0.0099 0.00964 0.00939� 0.00914 0.00889 0.00866 0.00842
2.4 0.0082 0.00798 0.00776 0.00755 0.00734 0.00714 0.00695 0.00676 0.00657 0.00639
2.5 0.00621 0.00604 0.00587 0.0057 0.00554 0.00539 0.00523 0.00508 0.00494 0.0048
2.6 0.00466 0.00453 0.0044 0.00427 0.00415 0.00402 0.00391 0.00379� 0.00368 0.00357
2.7 0.00347 0.00336 0.00326 0.00317 0.00307 0.00298 0.00289 0.0028 0.00272 0.00264
2.8 0.00256 0.00248 0.0024 0.00233 0.00226 0.00219 0.00212 0.00205 0.00199 0.00193
2.9 0.00187 0.00181 0.00175 0.00169 0.00164 0.00159 0.00154� 0.00149 0.00144 0.00139�
3 0.00135 0.00131 0.00126 0.00122 0.00118 0.00114 0.00111 0.00107 0.00104 0.001

�
469 ���Reliability Engineering
�
�

�������������������������������� ����������������� � �


� .001 .005 .010 .025 .050 .100 .900 .950 .975 .990 .995 .999
1 1.83� 7.88� 6.63� 5.02� 3.84� 2.71� 0.02� 0.00� 0.00� 0.00� 0.00� 0.00�
2� 13.82� 10.60� 9.21� 7.38� 5.99� 4.61� 0.21 0.10� 0.05� 0.02� 0.01� 0.00
3 16.27� 12.84� 11.34� 9.35� 7.81� 6.25� 0.58 0.35� 0.22� 0.11� 0.07� 0.02�
4� 18.47� 14.86� 13.28� 11.14� 9.49� 7.78� 1.06� 0.71� 0.48� 0.30� 0.21� 0.09
5� 20.52� 16.75� 15.09� 12.83� 11.07� 9.24� 1.61� 1.15 0.83� 0.55� 0.41� 0.21
6 22.46� 18.55� 16.81� 14.45 12.59� 10.64 2.20� 1.64� 1.24� 0.87� 0.68� 0.38�
7 24.32� 20.28� 18.48� 16.01� 14.07� 12.02� 2.83� 2.17� 1.69� 1.24� 0.99� 0.60�
8� 26.13� 21.95� 20.09� 17.53� 15.51� 13.36� 3.49� 2.73� 2.18 1.65� 1.34� 0.86
9 27.88� 23.59� 21.67� 19.02� 16.92� 14.68� 4.17� 3.33� 2.70 2.09� 1.73� 1.15�
10� 29.59� 25.19� 23.21� 20.48� 18.31� 15.99 4.87� 3.94� 3.25� 2.56� 2.16� 1.48
11� 31.26� 26.76� 24.72� 21.92� 19.68� 17.28� 5.58� 4.57 3.82� 3.05� 2.60� 1.83�
12� 32.91� 28.30� 26.22� 23.34 21.03� 18.55� 6.30� 5.23� 4.40� 3.57� 3.07 2.21�
13 34.53� 29.82� 27.69 24.74� 22.36� 19.81� 7.04� 5.89� 5.01� 4.11� 3.57� 2.62
14 36.12� 31.32� 29.14� 26.12� 23.68� 21.06� 7.79 6.57� 5.63� 4.66� 4.07� 3.04�
15 37.70� 32.80� 30.58� 27.49� 25.00� 22.31� 8.55� 7.26� 6.26� 5.23� 4.60� 3.48
16 39.25� 34.27� 32.00� 28.85� 26.30� 23.54 9.31 7.96� 6.91� 5.81 5.14 3.94
 470
Table E continud
α .001� .005� .010� .025� .050� .100 .900 .950� .975 .990� .995� .999

17 4.79� 35.72� 33.41 30.19� 27.59� 24.77� 10.09� 8.67� 7.56� 6.41� 5.70� 4.42
18 42.31� 37.16� 34.81� 31.53� 28.87� 25.99 10.86� 9.39� 8.23� 7.01� 6.26� 4.91�
19� 43.82� 38.58� 36.19� 32.85� 30.14� 27.20� 11.65� 10.12� 8.91� 7.63� 6.84� 5.41�
20 45.32� 40.00� 37.57� 34.17� 31.41� 28.41� 2.44� 10.85� 9.59� 8.26 7.43� 5.92
21� 46.80� 41.40� 38.93� 35.48� 32.67� 29.62� 13.24� 11.59� 10.28� 8.90� 8.03� 6.45
22 48.27� 42.80� 40.29� 36.78� 33.92� 30.81� 14.04� 12.34� 10.98 9.54� 8.64� 6.98�
23 49.73� 44.18� 41.64� 38.08� 35.17� 32.01� 14.85 13.09� 11.69� 10.20� 9.26� 7.53�
24� 51.18� 45.56� 42.98� 39.36� 36.42� 33.20� 15.66 13.85� 12.40� 1.86� 9.89� 8.09�
25 52.62� 46.93� 44.31� 40.65� 37.65� 34.38� 16.47� 14.61� 13.12� 11.52� 10.52� 8.65
26 54.05� 48.29� 45.64� 41.92� 38.89� 35.56� 17.29� 15.38� 13.84� 12.20� 11.16� 9.22�
27� 55.47� 49.64� 46.96� 43.19� 40.11� 36.74� 18.11� 16.15� 14.57� 12.88� 11.81� 9.80�
28 56.89� 50.99� 48.28� 44.46� 41.34� 37.92� 18.94� 16.93� 15.31� 13.56� 12.46� 10.39�
29 58.30� 52.34� 49.59� 45.72� 42.56� 39.09� 19.77� 17.71� 16.05� 14.26� 13.12� 10.99�
30 59.70 53.67� 50.89� 46.98� 43.77� 40.26� 20.60� 18.49� 16.79 14.95� 13.79 11.59
40 73.40� 66.77� 63.69 59.34� 55.76� 51.81� 29.05� 26.51� 24.43� 22.16� 20.71� 17.92�
50 86.67� 79.49� 76.15� 71.42� 67.50� 63.17� 37.69� 34.76� 32.36� 29.71� 27.99� 24.67�
60 99.61� 91.95� 88.38� 83.30� 79.08� 74.40� 46.46� 43.19� 40.48� 37.48� 35.53� 31.74
70 112.32� 104.21� 100.43� 95.02� 90.53� 85.53� 55.33� 51.74� 48.76� 45.44� 43.28� 39.04
80 124.84� 116.32� 112.33� 106.63� 101.88� 96.58� 64.28� 60.39� 57.15� 53.54� 51.17� 46.52
90 137.20 128.30� 124.12� 118.14� 113.15� 107.57� 73.29� 69.13� 65.65� 61.75� 59.20� 54.16
100 149.45 140.17 135.81 129.56 124.34 118.50 82.36 77.93 74.22 70.06 67.33 61.92

�
471 ���Reliability Engineering

��������������������������������������������������������������
pdf MGF Mean Variance
1 ,a  x  b  (t )  e  e
tb ta
(a  b ) (b  a ) 2
Uniform (a,b)�
b a t (b  a ) 2 12 �
Exponential with  1 1
Parmeter 
  x , x  0 �  t  2
�

Gammaa(n, λ� e x (x )n 1   

n
n n
n integer ,   0� ,x  0  
(n 1)!   t   2
�
2 n
 x  
1   2t 2  
Normal((μ��σ� e 2 2
 x  exp  t   � 2�
 2  2 

 ln t   
2
 
2

1 2

exp  
 2μ�σ2 σ2
Lognormal (μ��σ)� e 2 t 0 e  (e -1)
 t 2  2  �
�

Beta with parameters� (a  b ) a a b

cx a1(1  x )b 1,0x 1 c 
a>0 ،b>0 (a)(b ) a b (a  b ) 2 (a  b  1)
 472

�
473 ���Reliability Engineering

Table F Chracteristic of some continuous distributions continued �

A= location parameter B= scale parameter C= shape parameter�
Variance� mean� CDF� pdf� �

C x  A C  1  ( x  A )C
A  B  (1  1 )
x A C
 (1  ) 
2 ( )
B (1  )  B
2 2 2 1 ( ) B B e B Weibull�
C C
C � 1 e B x  A� x A
1
xA  C
1 (1  C x-A ) -1- C1 valid on
1  (1  C  ) valid on
B B B
� A B
A  x   w hen C  0 or A  x   when C  0 or GPD
1 C � A  x  A  B w hen C  0 A  x  A  B when C  0
C � C
 B
0 x  A  C & C0 
0 x  A B & C0
 
C
1
 1  x
1
 A C 1 xA
 x  A (1  C )  e(1  C B ) C
  (1 C  ) C  B
� � e B 
for 
x  A  B &C  0 GEV�
 B  C
B
x  A  &C  0
1 
x  A & C0 
 C  C

 1 x  A  B &C  0
 C
 �
�

 2B2 A  γB  x A  xA

6 �   0.05772� e e B
�
1
B
exp(  x  A  e
B
B )
�
FT1

A Weibull distribution with A= 0 & C=1 is an exponential distribution. A Weibull distribution with A= 0 & C=2 is Rayleigh distribution
A GPD distribution with A= 0 & C=0 is an exponential distribution
 474

������������������������������������������������������������
Probability function p(x) MGF� Z-transform mean Variance�
n  x n x
Binomial (n,p),0≤�≤�   p (1  p ) x  0,1,  , n
[ pe t  (1 p )]n [ pz  (1  p )]n np n p (1  p ) �
x 
�

Poisson with x
parameter e 
x  0,1,2,... exp[(et 1)] e  ( z 1)  
x! �
Geometric for success
x 1 pe t pz 1 1 p
with parameter p(1 p) x  1,2, 1  (1  p ) z p
0  p  1� 1  (1  p )e t p2 �
p
Geometric for failure , p 1 p 1 p
with parameter p(1  p) x
x  0,1, 2,� 1  (1  p )e t
0  p  1� t   ln(1  p )
1  (1  p )z p � p2 �
�

�
�����������������1����������������������������������
��
�� Parameter Density/probabality
estimator
Random numbers Inverse of CDF CDF(F(x)) ��function
Distribution
Beta betafit(X) betarnd(A,B,m,n,o,...) betainv(P,A,B) betacdf(x,A,B) betapdf(x,A,B)
Poisson poissfit(X) poissrnd(λ,m,n) poissinv(P,�λ) poisscdf(x,�λ) poisspdf(x,�λ)
Binomial binofit(X,n binornd(N,P,m,n) binoinv(Y,N,P) binocdf(x,N,P) binopdf(x,N,P)
Neg Bino. nbinfit(X) nbinrnd(R,P,m,n) nbininv(Y,R,P) nbincdf(x,R,P) nbinpdf(x,R,P)
Hyp. Geo. hygernd(M,K,N,m,n) hygeinv(P,M,K,N) hygecdf(x,M,K,N) hygepdf(x,M,K,N
Gamma gamfit(X) gamrnd(n,�λ,m,n) gaminv(P, n,�λ) gamcdf(x, n,�λ) gampdf(x,n,�λ)
Lognormal lognfit(X) lognrnd(μ��σ����� logninv(P,μ��σ� logncdf(x,μ��σ� lognpdf(x,μ��σ�
Chi-Squa.. chi2rnd(V,m,n) chi2inv(P,V) chi2cdf(x,V) chi2pdf(x,V)
Normal normfit(X) normrnd(μ��σ����� norminv(P,μ��σ� normcdf(x μ��σ� normpdf(x, μ��σ�
Exponential expfit(X) exprnd(mu,m,n) expinv(P,mu) expcdf(x, mu) exppdf(x, mu)
Geometry geornd(P,m,n) geoinv(Y,P) geocdf(x,P) geopdf(x,P)
Weibull wblfit(X) wblrnd(B,C,m,n) wblinv(P, B,C) wblcdf(x, B,C) wblpdf(x-A, B,C)
Uniform unifit(X) unifrnd(A,B,m,n) unifinv(P,A,B) unifcdf(x,A,B) unifpdf(x,A,B)
F frnd(V1,V2,m,n) finv(P, V1,V2) fcdf(x, V1,V2) fpdf(x, V1,V2)
GEV gevfit(X) gevrnd(C,B,A) gevinv (P,C,B,A) gevcdf(x,C,B,A) gevpdf(C,B,A)
GPD gpfit gprnd gpinv gpcdf gppdf
Rayleigh raylfit(X) raylrnd(B,m,n) raylinv(P,B) raylcdf(x,B) raylpdf(x,B)
t trnd(V,m,n) tinv(P,V) tcdf(x,V) tpdf(x,V)

�����������������������������������������������������������

1
Prepared by by: Mr Mohsen Abyar : A graduate of Shahid Bahonar University of Kerman, Iran
 ���������������������
���� ������� ��������� � ���� ����� ��� ����������� ������������ ����� ����� � ��
��������������������������������������������������������������
��� ��� ����� ����������� �������������������� ���� ��� ������ � ��� ����
������������������������������������������������������������
�����������������������������������������������������������������
���������� ��� ����� ���� ���� ������ �������� ��� ��������� ����
��������������������������������������������������������������
��������������������������������

���������������������������������������������������������������������
������������ ���� ��������� ���� ����������� �������� ���������
�������� ��� ���� �������� ��� ����� ����� ���� ���� ��� �� ��������
������� ��� �� ����������� ��� ���� ��������� �������� �������
��������� ��� ��� ���� ��� ������������ ���� ������ ����
����������������� ��� ���� �������� ��� ������������ ��� ��������
�������������������������������������

�����������������������������������������
������������������������������������������
����������������
������������������������������������� �