Reliability Engineering
Uploaded by
Boomiga RaghuReliability Engineering
Uploaded by
Boomiga RaghuReliability
Engineering
Therory and Practice
Hamid Bazargan_Harandi
January 2023
2
To my parents
The late Mohammad Ali B1azargan (1905-1967)
&
The late Robabeh Eslampanah (1921-1999)
3 Reliabilty Engineering
C H A P T E R T I T L E S
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
About the author …………………………………………….476
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
FO REWORD
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
Nomenclature Nomenclature
Symbols Description
1)sample size 2)the number
A,B,C n 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
A 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, OR Operational readiness
GPD) 2) Bartlet's test
C statistic
1)A parameter of some
distributions (Weibull ,
P,Q
GEV, Failure probability of a
GPD) p device
CDF Cumulative
distribution Probability density function
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
f pdf of system's lifetime Pi (t ) The probability that the
system is in state i at time t
f pdf of random variable
X Fraction defective in a
population
CDF of strength Failure (unreliability)
F Q
probability of a device
pdf of stress(load)
f S (s ) Failure probability of ith
component
CDF of strength
F (s ) R
pdf of strength
f ( ) ∗( ) The optimum reliability of
the system at time t
F ( )
pdf of sample min.
The required reliability of
F ∗F convolution of FX,FY the system
∗ The reliability allocated to
critical value of F ith unit
F,mn, distribution for n &m The given reliability for ith
degrees of freedom at unit
given sig. lev. α n The reliability of n-
component system(a active)
R (t)
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
standby ( 1 active 1 standby)
deg. of f. at lev. α/2 2 The reliability of 2-
R
H,L,N,O active
component system( both
active)
ℎ( ) Instantaneous failure rate The reliability function of
function Rs(t) the switch
L Lower Specification Level System reliability function
Rsys(t)
13 Reliabilty Engineering
Nomenclature Nomenclature
( ) 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
Stress, Load
S under half load
Maximum of stress(load), In load sharing:The failure
Smax upper extreme value of S rate for the component
under full load
Average stress value
Smean The failure rate of the
Minimum of stress(load),
Smin 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 upper specification The standard devaition of
U level of a dimention of a apopulation
product Standard devation of load
The difference of U al L sistribution
( − ) 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
V Coefficient of variation of measurement in a part
safety factor(SF)
In AGREE Method: the standard devaition of a
The importance factor measurement
of the ith subsystem
sum 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
composition operator
α, β, … or acting over a function
(in UGF Technique)
The probability of Type I
error End of Eexample
The probability of Type II End of theorem ; End of
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 TˆBF (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:
∞
R(t) = Pr(X > ) = Pr(X ≥ t) = ∫ f(x)dx − 1 − F(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 = MTBF , R(t = MTBF) = Pr(X > μ) =
Y is exponentially distributed, then:
( )= = , ( )= ≅ 0.37
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.
Y = Min(X, T) ⟹ E(Y) = E[Min(X, T)]
Min (X,T)+Max(X,T)=X+T⟹
E[Min (X,T)] =E(X+T)-E[ Max(X,T)]
E(X + T) = E(X) + T
E{Max(X, T)} = E(Max(X, T)|X > )Pr(X > )
+ E(Max(X, T)|X ≤ T)Pr(X ≤ T)
ﺍE(Max(X, T)|X > ) = E(X|X > ) ,
X is exponentially distributed and therefore is memoryless; then:
E(X|X > ) = T + λ and E(Max(X, T)|X ≤ T) = E(T) = T.
then
λ λ λ
E{Max(X, T)}=( T + λ) × e +T 1−e = T+ λe
Chap 1 Introduction and Basic Concepts 26
E(Y) = E{Min(X, T)} = E(X + T) − E[ Max(X, T)]
= + 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 t
( ) ( ) ( ) ( )
( ≤ ≤ | > )= =
( ) ( )
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( ) = lim∆ → (∆ )× ( )
= ( )
− = ( )
, (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 ‘chance’ of failure (though
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 h(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 P (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) = ∑∞ Pr(X = j) =∑∞ P (j),
The function h(k) given below, is known as the rate
function of "item with discrete lifetime distribution"(Xie et al,2002):
h(k) = Pr(X = k | K ≥ k) = Pr(X = k)/ Pr(X ≥ k) ⟹
P (k) P (k)
h(k) = = . (14 − 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 R(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)
f (x )dx
a
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 ( = 0, , = 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 t 2 t
1
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(
=1
=2
= 0.2
= 0.3
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( , ),…, 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
19000 − 20000
R(19000) = Pr( > 19000) = Pr Z > =
2000
Pr(Z>0.5)=1-Pr(Z<-0.5)=0.69146=69.15%
35 Reliabilty Engineering
1 - t-200002
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;0,∞)
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 = 3 = 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 Engineering
( )= −1 (1-22-3)
3)
median= (1-22-4)
4)
Figure 1.3 shows the pdf of 2 sample lognormal
distribution
distributions.
Fig. 1.3 The pdf of 2 lognormal distribution
1-8-4-1
1 Calculation of the parameters(μ,σ)
parameters( of lognormal
distribution from the mean and variance
Given E((T) and Var(T) as the mean and variance of a
lognormal random variable, the parameters of the distribution
is calculated from:
( )
= ln ( )
+1 , (1-22
22-5)
= ln ( ) − . (1-22
22-6)
1-8-4-2
2 The relationship between lognormal (μ,σ)
( ) and normal
(μ,σ)) distributions
Chap 1 Introduction and Basic Concepts 38
If T is distributed log-normally :T~lognormal(μ, σ), then
X=ln(T)is a normal random variable: lnT~N(μ, σ).
If X is distributed normally :X~N(μ, σ), then T=e is a
lognormal random variable: e ~logN(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):
( ) = lognormalcdf( , μ, )
( ) = 1 − lognormalcdf( , 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
(150) = > = 0.496
( )= ( = 150) ≅ 0.0027
√ ,
41 Reliabilty Engineering
( ) /
ℎ(150) = ( )
= = 0.0053 1 =
.
53 failures 10000 .
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 function 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 tA
C t A
B
f t e tA (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(1+ )-B2 Γ(1+ ) (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(1 + ) ≅ A + 0.9B , = Γ(1 + 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
, ,1 , or , ,2 . 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
E (X )
(1-24-2)
V ar ( X )
(1-24-3)
2
M GF (t ) (
)
(1-24-4)
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
1-8-8-1 Erlang 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( + )
+ >0
(1-25)
Γ( )
Figure 1-10 shows three functions related to 7 different
distributions.
49 Reliabilty Engineering
Fig. 1-10
1 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 Engineering
Figures 1-11--1 to 1-11-5 show some other
er variations of bathtub
curve which happen in practice,
Fig 1-11-2
2 Some variations of bathtub curve(Kuo&Zuo,2003)
cu ve(Kuo&Zuo,2003)
Fig 1-11-3
3 A variation of bathtub curve for some mechanical devices
(Ireson,1995 page18-2)
Chap 1 Introduction and Basic Concepts 52
Fig 1-11
11-4 Another variation of Bathtub failure rate function
(Nahmias, 2004, Fig.12-4)
Figure 1-11-5
5 shows different forms of bathtub failure curve due to
different levels of stress on some mechanical devices.
Fig1-11--5
5 Effects of stress levels on mechanical failure rates
(Ireson,199
n,1995 Fig18-1, Stamatis,2010 Fig. 6-3
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 = 0ﻭ > 0 , 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
The hazard function might be of the following power
function:
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 h(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 − 30 − 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:
F (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:
h(x) = p (1 − 30 − 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 x n n 1
n n x n 1 x n x x 1 n x
E ( X ) x p x 1 p n p 1 p np p 1 p
x 0 x x 1 x 1 x 1 x 1
n 1 n 1
y n 1 y
x 1 y E (X ) np p 1 p
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…) as well as time 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
(0.0256×515.8)0e-0.0256×515.8
Pr(X=0)= =1.8×10-6
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
x ∈ {0, 1 , … , n}
n n n x
Binomial
( − )
(1 − p) ( − )
∑ ( − )
∈ { , ,…, }
Poisson
x k x
x ∈ {0, 1 , … } e x x 1 k
e
x! k 1 k ! x !(1 )
k 0 k!
Geometric
( − ) −( − ) p
x ∈ {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 (a1 b1 ) (a2 b2 ) (ai bi ) (an bn ) sum
... ..
.
Frequ. o
.. ..
and let
N 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 t = b is estimated as:
N (bi )
Rˆ (bi ) (1-34)
N0
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 number of items failed during (t )i
f(t) = = (1-36)
N 0 × (b a ) ( )× (t)i
i i
The cumulative distribution function(CDF) estimate
The CDF at t = b is estimated from:
1
N (bi )
( )=1− ( )= (1-37)
N0
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 0 = 46
Solution
The above relationships were used to estimate ℎ( ), ( ),
( ), ( ). The results are shown in the following table:
b ( ) ℎ( )
i ( ) multiply multiply
N(a i ) N(bi )
ai N(b ) by: by:
i i
ai ( t ) i (b )
10 10
=
N 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 =
( 1)
→ (20000) =
(20000)
=
46 − 19 27
46
=
46
= 0.587
N0 N0
N ( a i ) − N ( bi )
ℎ( ) = →
N ( ai ) bi ai
hˆ ( t ) = 46 − 27
= 0.207 × 10 0 t 20000
(46)(20000)
ai − bi
( )= →
N0 × (b a )
i i
20000<t 40000 ( )=
( )( )
= 0.12 × 10
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 = 1. Therefore according to Eqs.
1-34 through 1-37:
Chap 1 Introduction and Basic Concepts 68
N (b i )
( )= , b1 = , ℎ( ) = .
N 0 × (b a ) N0 N ( a i ) (b a )
i i i i
The following table and figures shows the results of the
calculations based on these equations.
fˆ t = N 0 (b1i ai ) Rˆ (b i )
hˆ t = N(a )1(b
i ai )
a i t bi for ai t bi
i
N (bi )
(× 0.01) N0
b
(× 0.01)
9
1 0 5 1/(10*5)= 2 1/(10*5)= 2
10
8
2 5+ 10 1/(10*5)=2 1/(9*5)= 2.22
10
7
3 10+ 17.5 1/(10*7.5)= 1.33 1/(8*7.5)= 1.67
10
6
4 17.5+ 30 1/(10*12.5)=0. 1/(7*12.5)= 1.14
10
5 30+ 40 1/(10*10)= 1 5
1/(6*10)= 1.67
10
4
6 40+ 55 1/(10*15)=0. 67 1/(5*15)= 1.33
10
3
7 55+ 67.5 1/(10*12.5)=0. 8 1/(4*12.5)= 2
10
2
8 67.5+ 82.5 1/(10*15)=0. 67 1/(3*15)= 2.22
10
1
9 82.5+ 100 1/(10*17.5)=0. 1/(2*17.5)= 2.86
10
10 100+ 117.5 1/(10*17.5)=0. 0 1/(1*17.5)= 5.7
The following figures shows the functions of Example 1-
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
fˆ t =
f
hˆ t =
i , Rˆ (bi ) NN(b0i ) , i
N (b a ) N (a )(b a )
0 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
Rˆ (bi )
fˆ t = N 0 (bf ii ai ) N (b i ) hˆ t = N (ai )(f ibi ai )
no. of N0
a i t bi a i t bi
i ai bi
failures
1 0 3 185 185/(800×3)=0.0771 615 185/(800×3)=0.0771
800
2 3 6 42 42/(800×3)=0.0175 573 42/(615×3)=0.0227
800
3 6 9 36 36/(800×3)=0.015 537 36/(573×3)=0.0209
800
4 9 12 30 30/(800×3)=0.0125 507 30/(537×3)=0.0175
800
490
5 12 15 17 17/(800×3)=0.0071 17/(507×3)=0.0112
800
482
6 15 18 8 8/(800×3)=0.0033 8/(490×3)=0.0054
800
468
7 18 21 14 14/(800×3)=0.0058 14/(482×3)=0.0097
800
8 21 24 9 9/(800×3)=0.00375 459 9/(468×3)=0.0064
800
9 24 27 6 6/(800×3)=0.0025 453 6/(459×3)=0.0044
800
10 27 30 3 3/(800×3)=0.0013 450 3/(453×3)=0.0022
800
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
h(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
18 16 14 12 11 10 9 8 7 6
Failures in
200
180 162
0.1
the interval 146 119 107 96 86 77 69 62
0.1 0.1 0.098
h(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
18 16 14 12 11 10 9 8 7 6
Failures in
200
180 162
0.1
the interval 146 119 107 96 86 77 69 62
0.1 0.1 0.098
h(t) ≅ 0.1
20 18 16 14 13 12 11 10 9 8 7 6
= 2 × 10
D D D D D D D D D D D D
f̂(t)
t in hour
f̂(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
F̂(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
ˆ ˆ
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) R(t) f(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 h(t), this interval was combined with the previous
interval and h(t=300) was estimated as follows:
2
h = = 0.0170
(325 − 300)(8 − 4 + 0.7)
With empirical data this kind of smoothing must frequently be done
Some of the calculations are shown below:
.
h = =0.0024 R =1− = R (195)=0.9167
( )( . ) .
.
h = =0.0050 R = 1− = R(300)=0.5595
( )( . ) .
.
h = =0.0082 R = 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 = t( ) i − 0.3
1 = 1
(t − t ( ) )(n + 0.4) n + 0.4 1− F(t) (t ( ) − t ( ) )(n − i + 0.7)
1 − 0.3
1 1
10 + 0.4
1 5 (10 − 5)(10 + 0.4) 0.9327 (10 − 5)(10 − 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
117.5 Cannot be computed 0.9327 0.0673 Cannot be computed
10
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 , X , … , 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 X( ) denote the smallest of the n
observations. If X , X , … , X are independent then:
Pr X( ) > = Pr[(X > ), (X > ), … , (X > )] = Pr(X > ) ⇒
Pr 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 (x) 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:
lim → ( )= ( )=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 y = F , in Eq. 1-42 i.e
F ( )
(y) = 1 − [1 − F (y)]
( )
=1− 1−
Since F[F (x)] = x then the CDF of is:
( )= 1− 1− 0≤ ≤
We know from mathematics that
lim → { 1− }=e u ≥ 0 therefore:
lim ( )= 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 to1 −
( )=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
1−e :
( ) = lim ( )=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:
F ( )
(y) = Pr(limiting ( ) < )
From this relationship calculate F ( )
in terms of
Pr(U< ⋯)= .
Step 5 Calculate F ( )
(y), considering step 4 and
Eq. 1-45 i. e. Pr ( ≤ ) = 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
( ) = ln = − ln(1 − )
1−
Taylor expansion of f(x) about x=a is:
( )= ( )+ ( )( − ) + ( − ) ( ) +…
! !
This expansion for ln(1 + x) , −1 < ≤ 1 is:
ln(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
F ( )
(y) = Pr( ( ) < ) = ≤ = ( ≤ )
⟹ ( )
( )= ( )
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 others
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,…N
t 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 (D − ),
then
t = k(D − d ), where k>0 is the constant of proportionality.
The cumulative distribution of t 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
F (x) = , 0 ≤ 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− ( > )=1− ( > ,..., > )
Assuming ,..., are independent and similar, we could
write:
( ) = 1− ( > )… ( > )= 1− 1− ( ) … 1− ( )
( )= ( < ) = 1− 1− ( )
In mathematics it is shown that for 0 < < 1 , [1 − ]
approaches as ⟶ ∞ , then
Chap 1 Introduction and Basic Concepts 90
Since 0 ≤ ( ) ≤ 1 and there are a lot of pits in the pipe,
× ( )
therefore ( ) ≅ 1 − .
We saw earlier F (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, = 10 ,
=10 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 = 128. However here the pdf is
( )= and to find the value of λ the following
equation has to be solved:
1
1 = in
( )= ( ) = 16
128 1
( )= in
128
1 ∫ 1
( )= = ⟹ =
1− 128 1− 128
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 = 128.
Substituting the followings in ( ) =
= 10 , = 10 , = , = 128 , ( ) = 0.9
Yields t ≅ 242 hr.
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…. Interested readers 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
1
Extreme value 1=Fisher Tippet 1
93 Reliabilty Engineering
states that sum of the sample elements(∑ ) tends towards a
normal distribution with meannμ 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’ test described below:
Take a random sample of size at least 20 :t , … , t ≥ 20,
where t 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(1 − , 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 =
10%. 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 Ln -
20 20
B 19.34
20+1
1+
6 20
χ ,
= chi2inv(0.05,19) = 10.1170
χ ,
= chi2inv(0.95,19) = 30.1415
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:
∑ . .
= = ≅ 41.82, ( )= × = .
.
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 (1) , ..., 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):
(1 4 8 )
i a
F (i )
n b
0.27 0.23
where a 0.20 b 0.20
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 FX [ xˆ ( i ) ] F (i ) for each F(i), i=1,..,n calculate
xˆ(i ) , i=1,..,n from where F X
is the cumulative
distribution function of the distribution under study.
Plot the pairs ( x(i)& xˆ ( 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) & xˆ ( i ) to 1; but the vice
99 Reliabilty Engineering
versa is not necessarily true i.e. if the correlation
coefficient of x(i) & xˆ ( 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 ) xˆ ( i ) x ( i ) xˆ ( i )
R (1-49)
2 2 2
n x (i ) x (i ) n xˆ ( i ) 2 xˆ ( i )
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
ˆ X 54.81 , ˆ 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 xˆ ( i ) is calculated by equating the F(i)to the
Chap 1 Introduction and Basic Concepts 100
normal standard cumulative distribution, and calculating xˆ ( i )
from these equations.
xˆ(i ) ˆ s
Pr( X xˆ(i ) ) Pr(Z ) F (i) ˆ X , ˆ .
ˆ c4
sample calculation follows:
For i 1 :
xˆ (i ) ˆ xˆ (1) 54.81
Pr( Z ) F (i ) , Pr( Z ) 0.0294 xˆ (1) 32.37
ˆ 11.7287/0.9876
or ( ) = (0.0294,54.81,11.8751) = 32.3698.
The following table contains all the results
Rank(i) x (i) F(i) xˆ ( i ) Rank(i) x (i) F(i) xˆ ( 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 xˆ ( 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'…);qqplot(X,pd)
The correlation coefficient(r) between xˆ ( 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 F (x) and F (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):
( ) = Pr( + ≤ )=∫ Pr( + ≤ | = ) ( ) =
∫ Pr( + ≤ | = ) ( ) =∫ 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)dy = f (x)f (a − x)dx (1 − 51)
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) = (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 0 ≤ ≤ 1, ( ) is nonzero and
105 Reliabilty Engineering
(a − y) ≠ 0 0≤ − ≤ 1 or a − 1 ≤ ≤a
Then the limits of the integral are the intersection of the
intervals 0 ≤ ≤ 1 and a − 1 ≤ ≤ a . To calculate the
resulting interval which depend on the value, we divide the
range of 0 ≤ a ≤ 2 into 0 ≤ a ≤ 1 and 1 ≤ a ≤ 2:
If 0 ≤ a ≤ 1, as the following figure shows, the limits of integral
would be 0 ≤ y ≤ a:
If 1 ≤ a ≤ 2, using a similar figure it could be shown that the
limits of integral would be a − 1 ≤ ≤ 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) = (a − y) ( )
Considering the range of X and Y, we could say that a ≥ 0.
To find the limits of the above integral notice that:
f (y) ≠ 0 for ≥ 0 and
(a − y) ≠ 0 for a − y ≥ 0 ≤ .
Therefore the range of integral to become nonzero is 0 ≤ y ≤ a:
( )
f (a) = ∫ f (a − y)f (y)dy = ∫ e e dy = ae , 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)21
( )= ae a a0
(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
f (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 a − y = t, the range of t would be is a − 1 ≤ t ≤a
1
(a) = (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( )∗ = F ∗. (1-53)
and also
F ∗ F= F ∗ (1-54)
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 (at )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
1-16 The pdf of the difference of 2 independent
and nonnegative random variables
Suppose X1, X2 are 2 independent nonnegative random
variables with density functions f X and f X and let Y=X2-X1. Y is
1 2
sometimes called interference random variable. The density
function of Y is calculated from (extracted from K&L page
125):
fX1 ( y x2) fX2 (x2)dx2 y 0
X 0
fY ( y) fX1 ( y x2) fX2 (x2 )dx2 2 1-55
y 0
x2
fX1 ( y x2) fX2 (x2 )dx2
x2 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:Pr(a ≤ Y ≤ b) = ∫ f(y)dy .
For discrete random variable Pr(a ≤ Y ≤ b) = ∑ p(y).
where f(y) and 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
Pr( − < < + )>1− , ≥1 (1-57-2)
Example 1-24
The strength of a kind of component is a random variable
with μ̂=X=40 . 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
2) the distribution is unknown but σ=2
3) can the strength in part a be exponentially distribute?
4) the distribution is unknown but has one mode and σ=2
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<X<X+kσ >1-
2
k
X-kσ=34, X+kσ=46 k= 3
1 1 8
Pr(40-3×2< X < 40+3×2) > 1- 2
=1- 2 = =%88.9
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 2.25×9
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.
Definition of Likelihood function
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:
= Pr( = ,…, = ) = Pr( = ) × … × Pr( = )
(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
1
distribution with f x 0 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 ( ) ˆ
d xi X
b) ( )= = =∑ = =1
. . . .
Lemma1
1
Based on page 290 Bowker & Liberman (1972).
117 Reliabilty Engineering
If θ is the maximum likelihood estimator of θ and T(θ) is
the function of θ possessing a single inverse(i.e. its derivative is
always positive or negative) , then T 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:
MLE[θ(λ)] = θ(MLE(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 1 xi
k
n n n x k n x k
L p x i q n x i p x 1q n x 1 ... p q
i 1 x i x1 xk
k k
k
n xi
i 1
( n x i )
i 1
L p q
i 1 x i
k k
k
n xi (n x i )
ln(L ) ln( ) ln( p i 1 ) ln(q i 1 )
i 1 x i
k
n k k
ln(L ) ln x i ln p (n x i ) ln(1 p ),
i 1 x i i 1 i 1
k k k k
x i (n x i ) 1 p (n x i ) nk x i
0 0
ln(L ) i 1 i 1 i 1 i 1
1 p
k
k
p p p
x
i 1
i x
i 1
i
1 xi
1 k 1 MLE ( p )
nk i 1
p kn
xi i 1
End of Example
k 1 MLE ( p )
x
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 (x) = ;
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 = 0.062 which is the mean of the distribution.
Thus λ = = 16.13, 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 = 15.9. 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
k − 5, n = 100, MLE(p) = = = ⟹
kn n 100
MLE(p) = 0.016
Using MATLAB:
Given = 2, : = 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
k
x i
i 1
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, ….,k
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
1 X i n
E (X ) M 1 ˆ
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
Xi
E X M 1
n
2 2
E
X M 2 2 2 X i
n
X i
ˆ X
2
n
2 ^
2
Xi
2
2 X i 2 X
n
n
Chap 1 Introduction and Basic Concepts 124
In statistics theory it is proved that instead of the above
n
2
x
i 1
i
2
estimate for variance i.e. X sample variance i.e.
n
n n
(x i X )2 x i
2
nX 2
S2 i 1 i 1
is a good estimate for the
n 1 n 1
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 f(x) = e ,
θ
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.
C-1 Kolmogorov-Smironov(KS) test
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 = 0.05?
[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 = 0.05 .
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=[….data];
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 (X ≤ 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(X ≤ a) = 0.3935
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
F-1: Finding Roots of Polynomial Equation
To find the roots of x + ⋯.+ x + ⋯+ − 0 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
= 100√10. 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.
h(t) = 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 ) 105 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 (0 ) 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
_
b2
Rayleigh distribution with CDF F(x)=1-e ?
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 = Pr(E ) and the system
reliability( ) equals:
Chap. 2 Static Reliability Models 138
= ( ∩ …∩ ).
Assuming the operation of the subsystems are independent of
each other, we have:
= ( )… ( )= ×…× ⇒
=∏ . (2-1)
n
Since 0 < <1 therefore ≤ min { } ; that is the reliability of
i1
a series system with independent components is not greater than
the least reliable component(K&L page 56). It is worth
mentioning that in an n-
n component series system, if the
reliability of all components is equal to = 1 − , = 1, …,n
then: = (1 − ) . Now notice since according to binomial
expansion:
Therefore
( − 1)
(1 − ) = 1 + (− ) + (− ) +. . . +(− )
2
Ignoring higher order terms, could be approximated 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 1 q i q i 1
i 1 i 1
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 − ) 0.95 = (1 − ) ⇒ = 0.0051 ⇒
=1− = 0.9949
Using approximation:
≅1− ⇒ 0.95 = 1 − (10)( ) ⇒ = 0.005
=1− = .995
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:
= Pr(E |E , … , E , E )Pr(E |E , … , E ) … Pr(E |E )Pr(E )
=Pr(nth component is on|other components are on)× … × Pr(1st component on) (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 Rsys is equal to:
R = The probabilty that either 1 or 2 works= Pr(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 − Pr(both components fail ) = 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 ≥ 0.80 ⇒ ≥ 1 − (1 − 0.8)6 =
02353 = 23.53%
b- = 1 − (0.5) = 0.999 ⇒ = 10
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 − 0.98) = .999 → =
(.001)/ (.02) ≅ 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 + 0.15) = 0.75
The reliability of the system is
: R = 1 − (1 − 0.75) = 0.98
: R = (0.75) = 0.42
Select parallel configuration.
b)
The false signal probability is 0.1, then the probability of false
signal in a parallel arrangement is:
1 − [1 − (1 − 0.9) ] = (0.1) = 0.001.
Since the probability of without-error operation of the series
system is (0.9) ; then the probability of a false signal for series
configuration is 1 − (0.9) = 0.271
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.
2-4-1-1 Two-component system with 1 active and 1 standby-
Perfect switching
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 Engineering
=
Pr(system works)
works = Pr( A works) + Pr ( A does not work ) × Pr( B works)
works
= + ((1 − )= + −
Therefore the reliability of a two-component
component standby system is
the same as to that of a 2-component
component active system as given below:
2
R 1 (1 R 1 )(1 R 2 ) Ps=1 (2-3-1
1)
sys
2-4-1-2 n-component
component system with 1 active and n-1
1 standby-
standby
Perfect switching
Consider an n-component
component standby system with one normally
operating subsystem and n-1 in standby status(Fig 2-3--1). The
system is functional until n failures occur.. The reliability of this
system is:
n
standby R 1 (1 R 1 ).....(1 R n ) (2--3-2)
sys
Fig. 2-3-1
2 RDB of n-component
component standby system
(https://egyankosh.ac.in/bitstream/123456789/35169/1/Unit 15.pdf)
(https://egyankosh.ac.in/bitstream/123456789/35169/1/Unit-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 ….standby systems, the result would be Eq. 2-3-2.
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
2-4-1-3 Two-component system with 1 active and 1 standby-
Imperfect switching
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 0 < < 1 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)(1-0.92×0.96)=0.9883
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 × 0 .8 5 )(1 -0 .7 5 × 0 .8 )= 0 .9 2 3 0
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 − )(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,…,n. Then each subsystem reliability is equal
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
2-6 k-out-of-n configuration
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
Rsys = ∑n−k
x=0 x
(1 − R)x (R)n−x (2-5)
159 Reliabilty Engineering
ℎ :
Rsys = binocdf(n − k, n, 1 − R) (2-5-1)
The system failure probabilty or system unreliabilty is calculated
from:
1− =1− ( − , ,1 − )
Alternatively,
we could say that system work as far as n-k 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 k 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:
Rsys = 1 − binocdf(k − 1, 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 3 ∫0 3−1 (
1 − )5−3 =6R5-15R4+10R3
R=0.9 ⟹ =0.99144
Using Σ in Eq. 2-6:
161 Reliabilty Engineering
k−1 3−1
n 5
Rsys = 1 − 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) =
0.9914
Eq. 2-6-1:
=1− (2,5,0.9) = 1 − 0.00856=0.99144
End of Example
Example 2-13
The system shown in the following figure has only 4
components A , A , A and 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,
If 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 1,2 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 1 &2 work,
If 1 &3 work,
If 2 &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
2-6-1-1 Upper bound for k-out-of-n reliability
The reliability of k/n configuration is given by Eq. 5-2 :
n
R = (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 − 0.001 = 0.999
= 1− ( − 1, , )
1− (2,6,0.999) = 0.99999999998502
The probability of failure(unreliability probability):
1− ≅ 1.4976 10 .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.
2-7 Complex System Analysis
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 Reliability Evaluation
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.
2-7-2-1 Transforming a delta configuration into an
equivalent star configuration (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(1r3 r1)(1r2) 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
R1 (2-10-1)
B
[1 (1 r3 )(1 r2r1 )][1 (1 r1 )(1 r2r3 )][1 (1 r2 )(1 r3 r1 )]
1 R1
1 (1 r1 )(1 r2r3 )
ABC
2 R 2 (2-10-2)
C
ABC
3 R 3 (2-10-3)
A
A 1 (1 r r )(1 r ) B 1 (1 r r )(1 r )
1 2 3 2 3 1
C 1 (1 r3 r1 )(1 r2 )
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 r1, r2,r3 have to be found from the 3 equations
of Sec. 2-7-2-3 in terms of R1,R2,R3. 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 r3 r2 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:
r3 r2 r RY2 0 (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 Static Reliability Models 176
Solution No.1: star-delta approach
The vortices 1,2 and 3 in the given RBD constitute a delta
configuration. The
Th equivalent star configuration has 3
components with the following reliabilities:
ABC ABC ABC
R1 R2 R3
B C A
where
A = B = C = l-[l-(0.9)(0.9)](l-0.9)=0.981
l
3
R =R =R = 0.981 =0.9904
1 2 3 0.981
Replacing the delta configuration with this star configuration
would yield the following equivalent RBD for the system
177 Reliabilty Engineering
The reliability of the system based on the new RBD is calculated
as follows:
RSys = 0.9904 ×(1-(1-0.9904
×(1 ×0.9) (1-0.9904 ×0.9))= 0.9787.
Solution No.2 Conditional Probability Approach
The Element No 3 is chosen as the key element. If this element is
not functional, the following RBD with a reliability denoted by R SP
would be obtained.:
RSP =1-(1-0.9×0.9)(1-0.9×0.9)= 0.9639
Chap. 2 Static Reliability Models 178
Suppose Element No. 3 in the original RBD works all the time.
Replacing it with a line would result the following RBD with a
reliability dented by RPS .
R PS =[1-(1-0.9)(1-0.9)][1-(1-0.9)(1-0.9)]=0.9801
=[1-(1-0.9)(1-0.9)][1-(1-0.9)(1-0.9)]=0
Now Rsys ( the reliability of the system)
system is calculated using
sing Bayes
Rule (Eq. 2-8)
8) as follows:
Rsys =RSP×(1-r3)+RPS×r3=0.9639×0.1 +0.9801 ×0.9= 0.9
0.9785
2-8
8 Calculation of upper and lower bounds for
complex system using cut and tie set
set analysis
In this section a procedure is introduced for calculating an
upper and a lower reliability bound for complex systems. This
procedure is based on the so-called
so called tie and cut sets. The RBS of
some complex systems are shown below.
179 Reliabilty Engineering
Fig. 2-9 Three complex systems
Before performing the calculations, some definitions are
introduced:
Definition of minimal cut set
This set is a minimal set of components which by failing
guarantee the failure of the system(Grosh,1989 page125); in
other words the failure of its components cause interruption of
all paths from the input to the output of the system.
Definition of minimal path
A minimal path is a minimal set of components by functioning
ensures the system operation(Grosh,1989 page125).
Chap. 2 Static Reliability Models 180
Notice that
-Each
Each complex system has usually several minimal path and cut
sets. A set could be both cut and path.
- In the foregoing discussion, by mentioning cut set, minimal cut
set is meant and by mentioning path set, minimal path set is meant.
Example 2-21
2 (Grosh,1989 page 125)
Find the minimal path and cut sets for the following RBD.
Solution
The minimal cut sets are:
Cut set no. components
1 C1C2
2 C4C5
3 C1C3C5
4 C2C3C4
The minimal path are:
181 Reliabilty Engineering
Cut set no. components
1 C1C4
2 C2C5
3 C1C3C5
4 C2C3C4
End of Example
xample
Example 2-22
2 (O'connor,2003 page175)
The RBD of a system is given below. Show the minimal cut
and path sets on the RBD.
Solution
The minimal cut sets are (1-3 2-3 and 4 )as
as shown below:
The minimal path sets are (1-2-4 3-4 ) shown below:
Chap. 2 Static Reliability Models 182
End of example
Example 2-23
2 (Shooman,2002 page285)
What are the minimal cut and path sets between source a and
target b given below?
Solution
If all components fail but Component 1,the connection
between a and b does not interrupts then the set {1} is a tie set
If all components fail but Components 2& 5, the connection
between a and b does not interrupts then the set {2,5} is a tie
set
…
Functioning of the set with minimum elements 6,4 ensures
ensure the
system operation , 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 Static Reliability Models 184
After calculating R N 1 , the reliability of Network N1and that
of network N2 ( R N 2 ),we have the following inequality for Rsys ,
the reliability of the original system (based on Grosh, 1989 p 125 -6)
RN 2 Rsys RN 1 (2-13)
For some reasons such as dependence of the subsystems of
Network N1, The calculated upper bound( R N 1 ) is usually over-
over
estimated by this method. If R N 1 exceeds 1 set R N 1 1
Example 2-24
2
Draw the auxiliary networks for calculating the upper and
lower bounds for RBD of Example 2-21.
2
Solution a)Auxiliary network N1
Network N1 which is a parallel configuration of all the
minimal path elements in series is shown below:
The reliability of this RBD given by the following relationship
is the upper bound for the original system reliability of Example
2-21:
185 Reliabilty Engineering
RN 1 1 (1 r1r4 )(1 r2r5 )(1 r1r3r5 )(1 r2r3r4 )
b)Auxiliary network N2
Network N2 which is the series configuration of all the minimal
cut elements in parallel is shown below:
Network N2
The reliability 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 the auxiliary networks for calculating the upper and
lower bounds for the following RBD. The number in the box is
the reliability 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+…+ product of
unreliabilities of the components of Ac )
T
Ri
K 1 i B K
Product of reliabilities of 1st path components +…+ Product of reliabilities
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
Network N1 Network N2
T=2, B1={1, 2, 4}, B2={ 3, 4} C=3, A1={1, 3}, A2={2, 3}, A3={4}
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
2-9 Applications of Bays reliability in Design
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 i,j=1,2,…,k; i≠j ), 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(B1) 08
.
Chap. 2 Static Reliability Models 192
B2 = the event that the mean life is 30,000 kilometers, Pr(B2 ) 0.2
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(X³40000,μ=50000)=Pr Z>
(50000)(0.10)
=Pr(Z³-2)=0.9772
40000-30000
Pr(A B2 )=Pr(Z> )=0.00045
30000×0.10
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 μ=50000 km 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
1
From K&L chapter 3
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 Engineering
of losing a wheel bolt are 0.00001,
0.00001, what is the reliability of this
bolt system?
6- -(Problemn 7 Page 70 K&L) The system diagram given below
describes the circuitry for a neutral start switch on a manual
automobile transmission. According to the service manual, in
order to start this vehicle the clutch pedal must be fully
depressed and the ignition switch must be in the start position.
(a) Define reliability as it relates to this system
(b) Draw an appropriate reliability block diagram(RBD)
diagram
(c) Assuming that each functional block in your RBD has a
0.0001 chance of failing, calculate the system reliability.
7- A manufacturer wishes to know the reliability of a skid
protection system Io be used on military tractor trailers. The
system consists of:
(a) Two battery or generator powered sensors per wheel.
(b) One logic unit per sensor to predict wheel skid.
skid
(c) A command unit, which operates an electric or an engine
vacuum solenoid.
Chap. 2 Static Reliability Models 196
(d) The solenoids in (c) operate an actuator that controls the
pressure to the brake.
. The system diagram (not reliability diagram) is shown below.
8) A DC battery has a time to failure that is normally
normally distributed
with a mean of 30 hours and a standard deviation of 30 hours,
(a) What is the 25-hour
25 reliability?
(b) When should a battery be replaced to ensure, that there is no
more than a 10% chance of failure prior to replacement?
(c) Two batteries are connected in parallel to power a light.
Assuming, that the light does not fail, what is the 35-
35 hour
reliability for the power source?
(d) A particular battery has been in continuous use for 30 hours.
What
hat is the probability that this battery will last another 4
hours?
9)Calculate the reliability of the following two systems, where
each component 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(R ) is given by R = R .
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:
R = 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 R=component reliability
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
a) n = 3 m = 4, R = 0.7
b) n = 3 m = 2, R = 0.9
Solution
a)
From Eq. 3.1
R = [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) for n = 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 Pr(X = x ) = p and ∑ p = 1. the
mapping x ⟶ p is usually called probability mass function.
The mean of and MGF of this rv is:
( )=∑
k
t x i
( ) or ( )= ( )= p e
i 0
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
∑
( )= ( ).
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 x = 1 , x = 2 and
x = 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
∑
( )= ( ) (3 − 6)
i 1
4. If in the definition of the MGF e 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= 1, … , p= 1, … ,
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) =
Pr(X = 1, X = 0.5) + Pr(X = 1, X = 1) + Pr(X = 1, X = 2) ⟹
( = 1) = 0.06 + 0.36 + 0.18 = 0.6
Therefore according to the calculations in the table;
= { = 1 , = 2 , = 4 , = 16 } ,
= (0.6, 0.04, 0.24, 0.12).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 (z) (z)
(Livitin, 2010page 8). According to this notation for 2 variables:
( ) =⊗ ( ), ( )] = [ ( )⊗ ( ) (3 − 9)
for n variables(Livitin, 2010page 8):
(z) ( ) =⊗ ( ( ), … , ( )) (3 − 10)
U (z) = .
n
f ( x i j1 ,..., x n j n )
⊗ P Z =
pij i z
… (3 − 11)
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 U(z) 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:
( ) = 0.06 + 0.04 + 0.36 + 0.24 + 0.18 + 0.12
Merging the like forms results in:
( ) = 0.6 + 0.04 + 0.24 + 0.12
As you may have noticed, this function represents the
probability function for Y as follows:
Y = (1, 2, 4, 16) , q = (0.6, 0.04, 0.24, 0.12)
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 Y=X 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 ) 0.6×0.1 Z +0.6×0.6Z +0.6×0.3Z +
(40.5 ) (41) (42 )
0.4×0.1Z +0.4×0.6Z +0.4×0.3Z
1 2 4 16
( ) = 0.6 + 0.04 + 0.24 + 0.12
From this function the probability function of Y is obtained:
= (1, 2, 4, 16)
probabilities = = (0.6, 0. 04, 0.24, 0.12)
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
2 × 3 × 3 × 2 × 2 = 72; 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
( , )
0.42 ( , )
+ 0.21 +
. × . . × .
( , ) ( , )
0.07 018
. × . . × .
( , ) ( , )
+0.09 + 0.03 ⟹
( ) = 0.63 + 0.27 + 0.1
( )= ( )⊗ ( )
= (0.1 + 0.5 + 0.4 ) ⊗ (0.6 + 0.4 )=
( , ) ( , ) ( , [) ( , )
0.06 +0.04 +0.3 0.2
( , ) ( , )
+0.24 + 0.16 ⟹
( ) = 0.64 + 0.36
( )= ( )⊗ ( )=
= (0.63 + 0.27 + 0.1 ) ⊗ (0.64 + 0.36 )=
= 0.4032 + 0.2268 + 0.1728 + +0.0972 + 0.064 + 0.036 =
( ) = 0.4032 + 0.3996 + 0.01612 + 0.036
( )= ( ) ⊗× ( )
= (0.4032 + 0.3996 + 0.01612
.
+ 0.036 ) ⊗× (0.5 + 0.5 )
After necessary calculations and simplification, the final answer
for U (z) is:
( ) = 0.2016 + 0.1998 + 0.2822 + 0.018
+ 0.1998 + 0.0806 + 0.018
From U (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 + 0.15 + 0.4 + 0.35 ,
Find 20-hour reliability of the system.
Solution
= ( > 20) = 0.40 + 0.35 = 0.75.
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 X is the state of jth subsystem(with 2 values :either
x =1=working or x =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 R 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 = 0 or 1.
Let the static reliability of jth subsystem = , then
= = 1− = 0 or = 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
value of
, , Pr( X x ) p X (x)
= 0 down, = 1 working
0,0,0 0 (1 − )(1 − )(1
0,0,1 0 (1R )(1R )R )
1
−
2 3
0,1,0 0 (1 − ) (1 − )
0,1,1 1 (1 − )
1,0,0 0 (1 − )(1 − )
1,0,1 1 (1 − )
1,1,0 0 (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) (1 ≤ ≤ ) 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
U (Z) = 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.
Exercises
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’s sight,
so keep looking
how you are
acting
229 Reliabilty Engineering
Chapter 4
Structural
reliability
Analysis
Chap. 4 Structural Reliability Analysis 230
Structural Reliability Analysis
Aims of the chapter
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 1.52.
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:
2S
1 2
E ( )
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 S ≥ δ.
4-3 System reliability -Load & Strength variable
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)
s f ( y s )f (s )ds y 0
s
y
and the system reliability(R):
R Pr(Y 0) fY (y ) (4-6)
y 0
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
¥ 25 1 s-15
R= f s (s) 1-Fδ (s) ds= 1- ds
s=0 s=20 25-20 25-15
1 25
R= (25-s)ds=0.25
50 s=20
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
1
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(1 + ) ≅ A + 0.9B , = Γ(1 + Γ( )) ≅ 0.3
237 Reliabilty Engineering
Fig 4-2-1Normally distributed load and strength: interference
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 2
S 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
.9999999
.999968
.999999
.97725
.99865
.99977
.5987
.6915
.7734
.8413
.8944
.9332
.9599
.9937
R=
.50
ϕ ( )
239 Reliabilty Engineering
Fig. 4-3 Plot of Reliability versus Safety Margin( K&L page 80 Redrawn)
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 λ Exp(λδ ) Exp(λ )
(4-9)
R=
λ λδ
Ref:K&L page 157
2 σ λ Normal Exp( λ)
(4-10)
R = 1 − exp (−μ × λ + ) (μ, σ)
Ref:K&L page 157
3 SM =
μδ μ (4-11) N(σδﻭμδ ) N(σ ﻭμ )
σδ σ K&L page 126
R = ϕ (SM)=normcdf(SM)
Calculable in MATLAB
4 = (z)=normcdf(z) logN(σδ ﻭμδ ) logN(σ ﻭμ )
z=
(4-12)
Ref:K&L page 130
ﻭ ﻭ are the parameters not
the mean and standard deviation
s
5 x
s Gamma Gamma
x a 1 (1 x )b 1dx ,
(a b )
R (a, , δ ) (b, , )
(a )(b )
(4-13)
x 0
Ref:K&L page 141
betacdf ( s s , a, b ) : MATLAB
Calculable in MATLAB
6 − Weibull Normal
R = Pr( > ) = + × (A,B,C) (μ, σ)
√2
Ref:K&L page
(4-14)
∞
× ∫ exp [− − 0.5 + ] ,
=
142& Appendix III
Calculable in Matlab , Maple…
7 ∞
Bδ Aδ − A Weib( , Weib( ,
R=1− e exp [−( y δ + ) ]dy , ) , )
B Bδ
= 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 − Pr(Z < ),
μ −μ δ−μ −
SM = = =
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 SM = 8, the reliability and the failure probability
is:
= (1 − ) = 1 × 10−15
R=Pr(Z < = 8)= normcdf(8) = 0.999999999999999
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
−
=( )
= Pr( > ) = 1 − e exp [−( ) ]
=1− e exp [− ( ) ]
[ )]
=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 60000 Kpa and
standard deviation of 20000 Kpa is applied to a structure which
Chap. 4 Structural Reliability Analysis 246
has an lognormal distributed strength with a mean of 10
Kpa and standard deviation of 10 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 :
( ) 20000
= ln + 1 = ln ( + 1) =
( ) 60000
(20000^2/60000^2 + 1) = 0.1054
= ln ( ) − = log(60000)-0.5*.1054= 10.9494
( ) 10000 10000
= ln + 1 = ln ( + 1) = + 1 = 0.01
( ) 100000 100000
= ln ( ) − =log(100000)-0.5*.01= 11.5079
b) Using MALAB software and Eq. 4-12:
= (z)=normcdf(z) z=
. .
z= = = 1.6441
√ . .
247 Reliabilty Engineering
c)According to Eqs. 1-22-2 &3
0.1054
( )= = 10.9494 + = 60000
2
( )= −1 .
( )−1 = (2 ∗ 10.9494 + 0.1054) ∗
( (0.1054) − 1) = 20004 .. 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, = 0.018) fits the
strength. Since the exponential distribution could be considered
Gamma(b=1, = 0.01), therefore according to Eq. 4-13 the
reliability of the structure is:
λs
x= λ
s +λ δ
Γ(a+b)
R= x a-1 (1-x) b-1dx=
Γ(a)Γ(b) x=0
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
Γ(2+1)
Γ(2)Γ(1)
x=0
1×1 0
x=0
or by MATLAB
0.01
R betacdf ( 0.01+0.018 ,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.
1
ARENAtools-input analyzer- new-file data file- use existing- fit all
249 Reliabilty Engineering
Solution
From Eq. 4-9:
R= = = 0.60
From Eq. 4-15:
Since exponential distribution is a special case of Weibull
distribution, then
C =C =1 = =0
=1− e exp [−( ) ]
= = , = ⟹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
(4-16)
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 (4.5) =
0.999997. 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(5000N, 400N). The load it has to withstand is also
normally distributed S~N(3500N, 400N). 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)
5000 − 3500
= ≤ = ⎛ ≤ ⎞ = 0/99598
400 + 400
⎝ ⎠
b)
R=Pr δ>L =0.99598 Pr(Z< L'-5000 )=0.00402
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.
4-4 Calculation of structures' reliability : Load
or Strength deterministic
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 F ( ) then the reliability of the system ( R)
is the shaded area in the figure calculated as follows:
δ
= Pr <δ =F δ = ( ) (4 − 18)
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) .
= 0.911
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 S if the probability density function of strength
(capacity) is ( ) and the CDF is Fδ then the reliability of the
system (R) is the shaded area in the figure calculated as follows:
= Pr S = ( ) = 1 − Fδ (s) (4 − 19)
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 2 (X ) 1]
using MATLAB
σ=sqrt(log(4002 /50002 +1))=0.1
2
μ=lnE(X)- σ =ln(5000)- 0.01 =8.5122
2 2
R=Pr δ>S=4000 =
Pr(lnδ>ln4000=8.294)=Pr(Z> 8.294-8.5122
0.1 =-2.182)=0.9854
or using MATLAB:
R= 1 − Pr( < 4000) = 1 − ogncdf(4000,8.5122,0.1) = 0.9854
End of Example
259 Reliabilty Engineering
4-5 Interrelation between reliability(R) and
safety factor(SF): Strength( ) and Load(S)
independent and normally distributed
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 SF 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):
n≥ (4-20)
×
R≥1− ( )
(4-21)
where
R The system reliability
n The mean of SF
V = 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:
S2
1 2
n (4-22-1 )
S S
2
2 2 4
2 S S
SF
S (4-22-2 )
S S
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 n = 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 V = 16% 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 0.8357
= = = 0.18
μ 4.6426
= 0.95.
. ∗ .
≥1− ( . )
. ∗ .
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⟹ ≥ . = 4.6426
. ∗
.
End of Example
4-6 Determining the structural reliability
bounds using nonlinear programming(NLP)
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
f , (δ, S) Joint pdf of the strength and stress(load)
263 Reliabilty Engineering
fS 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 , S max ]. 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) Fδ (u)f s (u)du [1 FS (u )]f (u )du (4-25)
δ min min
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
Max/Min 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( 129×10 -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 =30,σS =3) :
32-30
p1 Pr(30 s 32) Pr(0 Z ) 0.2475
3
269 Reliabilty Engineering
p 2 Pr(32 s 34) normcdf(34,30,3)- normcdf(32,30,3) 0.1613
p3=normcdf(36,30,3)- normcdf(34,30,3)= 0.0685
p4= normcdf(38,30,3)- normcdf(36,30,3) = 0.0189
p5=normcdf(40,30,3)- normcdf(38,30,3) = 0.0034
p6= normcdf(42,30,3)- normcdf(40,30,3) = 3.9739 10-4
p7=normcdf(44,30,3)- normcdf(42,30,3)= 3.014110-5
p8= normcdf(46,30,3)- normcdf(44,30,3) =1.4824 10-6
p 9 Pr(46 s 48)
normcdf(48,30,3)- normcdf(46,30,3) 4.7226 10-8
p 1 0 P r( 4 8 s 5 0 )
n o rm c d f(5 0 ,3 0 ,3 )- n o rm c d f(4 8 ,3 0 ,3 ) 9 .7 3 5 0 1 0 - 1 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
-10 -10
Up10= 9.7350 10 +(0.02)* 9.7350 10 ;
-10 -10
Lp10= 9.7350 10 - (0.02)* 9.7350 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 ﻭq
i i and their limits
n
30-30
p
i 1
i Pr(s min ) Pr( Z
3
) 0.5
n
q i Pr( s max ) 1 exp[ (
50-30 2
) ]=0.105.
i 1 60
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 1 2
n
0.02 0.02
0.105-( ) 0.105 q i 0.105+( ) 0.105
2 i 1 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 :
Lp =0.2475- 0.02 × 0.2475 p Up =0.2475+ 0.02 × 0.2475
1 1 1
0.1613- 0.02 × 0.1613 p 0.1613+ 0.02 × 0.1613
2
0.0685- 0.02 × 0.0685 p 0.0685+ 0.02 ×0.0685
3
0.0189- 0.02 × 0.0189 p 0.0189+ 0.02 ×0.0189
4
0.0034- 0.02 × 0.0034 p £0.0034+ 0.02 ×0.0034
5
-4 -4 -4 -4
3.9739×10 - 0.02 × 3.9739×10 p 3.9739×10 + 0.02 ×3.9739×10
6
-5 -5 -5 -5
3.0141×10 - 0.02 × 3.0141×10 p 3.0141×10 + 0.02 ×3.0141×10
7
-6 -6 -6 -6
1.4824×10 - 0.02 × 1.4824×10 £p £1.4824×10 + 0.02 ×1.4824×10
8
-8 -8 -8 -8
4.7226×10 - 0.02 × 4.7226×10 p 4.7226×10 + 0.02 ×4.7226×10
9
-10 -10 -10 -10
9.7350×10 - 0.02 ×9.7350×10 p 9.7350×10 + 0.02 ×9.7350×10
10
The constraints of Type Lq,i qi Uq,i i 1,...,10 :
Chap. 4 Structural Reliability Analysis 274
Lq =0.0011- 0.02 * 0.0011 q Uq =0.0011+ 0.02 * 0.0011
1 1 1
Lq2=0.0033- 0.02 * 0.0033 q Uq2=0.0033+ 0.02 * 0.0033
2
Lq3=0.0055- 0.02 * 0.0055 q Uq3=0.0055+ 0.02 * 0.0055
3
Lq4=0.0077- 0.02 * 0.0077 q Uq4=0.0077+ 0.02 * 0.0077
4
Lq5=0.0098- 0.02 * 0.0098£ q Uq5=0.0098+ 0.02 * 0.0098
5
Lq6=0.0118- 0.02 * 0.0118 q Uq6=0.0118+ 0.02 * 0.0118
6
Lq7=0.0138- 0.02 * 0.0138 q Uq7=0.0138+ 0.02 * 0.0138
7
Lq8=0.0157- 0.02 * 0.0157 q Uq=0.0157+ 0.02 * 0.0157
8
Lq9=0.0174- 0.02 * 0.0174 q Uq9=0.0174+ 0.02 * 0.0174
9
Lq10=0.0191- 0.02 * 0.0191 q Uq10=0.0191+ 0.02 * 0.0191
10
Constraints related to p i & q i :
n
ap pi b p 0 .5-(0.02/2)*0.5 p1 ... p10 0.5+(0.02/2)*0.5
i 1
n
aq q i b q 0.1055-(0.02/2)*0.105 q1 ... q10 0.105+(0.02/2)*0.105
i 1
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
Appendix : Other definition of safety
margin(SM) and its relationship to safety
factor(SM)
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).
SM = − 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
SM = −1 (p-2) or
×
SF = ( p-3)
( + )×
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 4
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« strength of
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 = 0.2
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
normcdf(-35/5)+exp((.001^2*5^2)/35-2*0.001)*(1-normcdf((-35+ (.001^2*5^2)/5 -4))
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.
5.1 Introduction
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 S = 12
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.
5.2 Certain properties of a function of some
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 1 , 2 then
ℎ( ) = × ( )+ × ( ) (5-3)
In general, if the inverse function has n roots 1 , … , 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 2 2 and = . Find the pdf of Y.
f (x ) e
2
Solution
= ( )= ,
pdf of Y = ℎ( ) = × [ ( )] , ( )= ( )
1
= ⟹ = = ( ) =
2
ln y
1 2
f (k ( y )) e 2
2
2
1 1
ln y
ℎ( ) = × [ ( )]= × e
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 2 2 . Find the
f (x ) e
2
probability density function of the cross section i.e. = .
Solution
4A 4A
= ⟹ D1 ,D2 ;
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
dA
| 2 A 2 | |
dA A
2 2
4A 4A
1 1
h (A ) [exp( 2
) exp(
2
)] .
A 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 pdf ( , ), then
End of Proof
5-2-3 Variance of sum and difference of 2 random
variables
If and 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
5-2-3-1 Variance of sum of 2 independent random variables
If and 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
From Eq. 5-8:
( . )
( )≅ ( )+ "( ) = (2 ) + (2 ) = 4.01
From Eq. 5-9:
( )≅[ ′( )] ⟹ ( ) ≅ (2π×2×0.1)2 = 0.16
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 X , … , 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):
( ) ≅ ( ,…, )+ ∑ ⃒ ( ) (5-10-1)
Or in vector form:
|
⎛ . ⎞
( ) ≅ ( ,…, )+ ( )…. ( ) ⎜ . ⎟ (5-10-2)
⎜ . ⎟
⎝ | ⎠
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
X ,…, having standard deviations ,…, then:
⎧ ⎫
⎪ ( ,…, ) ⎪
( )≅∑ × ⃒ (5-11)
⎨ ⎬
⎪ ⋮ ⎪
⎩ ⎭
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 20000
=− , = . ⃒ = = 160
5
1 10000
= , =0. ( , )= ⟹
5
2 0
10000 1(0.4 0)
160 ⟹
( )≅ +
5 2
( ) ≅ 2000 = 20000 = 20
According to Eq. 5-11:
Chap. 5 Combinations of random variables + tolerance 292
( )≅ ( ⃒ ) + ( ⃒ )
1 −10000
= (1000) + (0.4)
5 5
⟹ ( ) ≅ 65600 ⟹ ≅ 256.1 = 2561
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 :
2
X X Y
E 1 (5-12-1)
Y Y Y
From Eq. 5-12-1 it is concluded that:
2
1 1 Y (5-12-2)
E 1
Y Y Y
Furthermore the variance of is approximated by:
4
2 2 2
X X
x Y Y (5-12-3)
V ar
Y Y X Y Y
293 Reliabilty Engineering
Example 5.5
Repeat Example 5.4 using Eqs. 5-12-1 & 5-12-3.
Solution
2
P A
1
10000 0.4
E( ) = ≅ = ∗ 1+ = 2012.8 N / cm 2
5 5
A A
⟹ E( ) = 20128
4
2 2 2
P P
P A A
( ) = Var
A A P A A
10000 1000 0.4 0.4
= ∗ + − = 65436
5 10000 5 5
≅ 255.8 N / cm 2 = 2558 . End of Example
5.3 Statistical Tolerance
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 X = + ،. . . + . The mean and variance
of X denoted by sum and σ S2 u m equals:
sum=n part (5-13-1)
2
Sum 2
nPart (5-13-2)
295 Reliabilty Engineering
part
2
sum (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 × standrd deviation, then substituting =
( )
and = in Eq. 5-13-3 results in:
(5-14)
2
U L sum
(U L ) part .
n
( ) ( )
Let ∆= and = then:
∆
= (5-15-1)
√
∆= √ (5-15-2)
Then the tolerance specification of a product assembled from n
similar parts with specification ± t would be × ± ∆. If the
n parts have different specifications a ± t … a ± t … a ± t
( − )
then if = , the specification of the assembled part
2
would be ∑ a ± ∆′ where:
∆′ = 2
. (5 − 16)
=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 is10 × 2.000 ± ∆
where ∆= √ = 0.012√10=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 = = = 0.012 . 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
(0.012) + (0.015)
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’s 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 Combinatio
ombinations of random variables + tolerance 298
Distribution nominal Value ±
E1 normal E1* =40 0.5 40 ±0.5
1 1 1
N normal N* = 1% ±0.01×
2 2 2
K normal K =3
*
2% 3±0.02×3
The mean of the distributions are equal to their nominal values.
Solution
Taylor series 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
(5-17)
For the linearization of = , is expanded into a
Taylor series around the nominal values E1* , N * , K * with help of
the above relationship as follows:
299 Reliabilty Engineering
1
E 0 E 1*N *K * E 1 E 1* N *K * N N * E 1* K * K K * E 1* N *
1!
*
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 E1* , N * , K * , therefore the
distribution of has the mean E 0 calculated from:
E E N *K *E E *K *N E *N *K 2N *E *K *
0 1 1 1 1
Since the mean of the distributions of , , equal E1* , N * , K *
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 =E 1* N * K *
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,
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 ± b the largest value of the standard
deviation of the dimension denoted by σ is given by(see
K&L pages93-96):
σ = (5-19)
where z is the critical value related to normal distribution
given in Table D or by a software such as MATLAB; e.g. for
=0.27% Z z0.00135 norminv(1-0.000135) 3 and
2
therefore:
0.5 1
×0.01 3×0.02
E = , N = 2
= 0.0017 , k = =0.02
1
3 3 3
Substituting numerical values in the right hand side of Eq. 5-18
gives E 0.512 . * * *
Eo E1 N K 40× 12 ×3=60 is the nominal value
0
of the output voltage. To write the specification of output
voltage E0 as 60 ± Δ note that the distribution of E0 can be
approximated by a normal with mean equal to the nominal
value(60); therefore for α = 0.27% from Eq.5-19 :
301 Reliabilty Engineering
Δ Δ Δ Δ
0.512 = = = = ⟹ Δ = 1.536
. 0.00135 3
1.536
=2.56% E 0 : 60±2.56% .
60
: 60 ± 1.536 or since
Therefore the output voltage E is 60 volts±2.56%.
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 4000 lb 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 100×103 psi , 5×103 psi , S 4×103lb , S 100lb ,
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±1.5% 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.
Exercises1
In the following problems assume all dimensions are normally distributed and the tolerance
range is 6 sigma(±3σ)
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 ±0.002. The tolerance for
the distance L between the centers of the holes is ±0.004. Find
the tolerance for the dimension h.
305 Reliabilty Engineering
3.A rectangular solid bar has the following dimensions:
X : 2±0.002m ,Y : 1±0.001m , Z : 4±0.008m . 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 ±20' D 0.800±0.002 in d 0.400 ± 0.001 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
1
random variable with parameters n= 17 and 20kPa .
307 Reliabilty Engineering
Determine the value which the random variable smean will exceed
only 1.3% of the time. Also determine the value which the
random variable samp will not exceed 90% of the time.
Hint: For largish n a gamma distribution (n, ) could be
√
approximated with a normal distribution ( = , = );
becaue it be consisidered the sum of n inpenendent exponential
distribution with parameter ( ),
7. The analysis of the loading of a component revealed a load
diagram as shown in the Following figure..
The four forces F1, F2, F3, and F4 are random variables, the
distributions of which are given in the following table
Chap. 5 Combinations of random variables + tolerance 308
Force Distribution Parameters
F1 Exponen. 1
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 RT.
309 Reliabilty Engineering
Answer is given in K&L page 105.
9. A kind of beam has a rectangular-shaped
rectangular cross-section.
section. The
dimensions of the cross-section,
cross section, denoted by a & b, are random
variables with mean 20 mm, 60 mm and standard
a b
deviation 0.02 mm, 0.1 mm. A stress is applied to the
a b
beam. The bending stress is a random variable with mean
2000Nm and standard deviation
deviation 10Nm. Find an approximate
value for the maximum bending stress(S). The dimensions are
assumed independent.
Anyone pursuing his goals honestly
does not slip up
and if he does, he can seek a way out
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 310
Chapter 6
Estimation of
Mean Life &
Reliability,
Related
Experiments
&Tests
311 Reliabilty Engineering
6
Estimation of Mean Lifetime &Reliability and
Related Experiments &Tests
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 ،. . . ،x is obtained. The following equation gives an
unbiased estimate for the mean lifetime ( )of the product:
( 6-1)
∑
θ= =
This equation could be used for any product with any lifetime
distribution.
6.3 Tests for 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 “suspended,” “censored,” or “truncated”
.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=[….data]; pd=makedist('exponential', 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). Estimate the mean life
parameter.
Solution
M TˆB F = = = 12.25 ℎ End of Example
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.
6-4-1-1Type I censoring life test 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 *
ˆ (6-3)
r r
∗
Where r ≥ 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
ˆ (6-4)
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
No
Cycles at 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 replacement.
Solution
= n = 9 × 20000 = 180000
Ten failures occurred during the test duration i.e. r=10,
therefore :
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 318
T 180000
M TˆB F = θ = ˆ = = 18000
r 10
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 24×5000=120,000 hr
*
ˆ T nt 120000 =8571hr .
MTTF
r r 14 End of Example
6-4-1-2Type I censoring life test without replacement
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
ˆ i 1 (6-5)
r r
319 Reliabilty Engineering
where
r number of items failed
T total operational time for the n units tested
∗ duration of life test
n 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
t* 30days , T ( n r )t * xi T (20-9)×30+171.3 501.3
l 1
ˆ 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.
6-4-2-1Type II censoring life test 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 ≤ n), the test is terminated. The estimate for
MTTF is(K&L page252, Lewis,1994 page239):
( ) ( ) ∑
= = (6-6)
xi ith value in the sample containing failure times
( ) The time of the rth failure
T 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)×41.2+27.4+...+14.4+41.2=624.5
i 1
Then according to Eq. 6-6
ˆ ˆ T 624.5 =62.45 days . End of Example
MTBF
r 10
6-4-2-2Type II censoring life test with replacement
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 ) (6-8)
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 ) ; ≥ 1 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 (r )
fails times and the mean time to failure is ˆ . Hence we
r
n
have the following estimate for MTBF:
T nx ( r )
MTˆBF ˆ (6-9)
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 ( ) = 840
= 6 × 840 = 5040 hr ˆ ˆ T 5040 630 hr
MTTF
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.
ˆ
r r
Time-terminated With 6-3 r
(Type-I Replacement ( n r )t * x ( i )
T i 1
censoring) ˆ
r r
Without 6-5 ( − ) +∑
( )
Replacement =
Failure- Without 6-6 nx
terminated Replacement (r )
(Type-II r
censoring) With 6-8
Replacement
T nt *
ˆ
r r
6-5 On the Accelerated Life Testing(ALT)
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.
6-6 Confidence interval for mean lifetime-
Exponential distribution case
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<G2 ) =1-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, 1 − 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 estimated from ˆ .
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
∑ ( )
= = = = 83665
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 chi2inv(.025,16) 6.91,
2
χ16,0/025 chi2inv(0.975,16) 28.85
2T 2T 2×669320 2×669320
2
θ 2 θ
χ α χ α 28.85 6.91
2r, 2r,1-
2 2
θ L =46400 θ 193725 cycles=θ U
c)CI for reliability function , if the lifetime is
exponentially distributed:
t t t t
L U 46419 193736
e R (t ) e or e R (t ) e
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 328
for t=4000 cycles, CI is:
4000 4000
- -
46400
e 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
was r=20 failures and mean of MTTF
ˆ 5000 . Find a 95% CI
for MTTF. Assume the lifetime is exponentially distributed.
Solution
According to Eq. 6-11:
2T 2T
Pr 2 θ 2 =1-α ,T= r×θ=20×5000, 1-α=0.9
χ α χ α
2r, 2 2r1 ,(1- )
2
χ 240,0.05 =55.76, χ 240,0.95 =26.51
From Table E:
Substituting the numerical values yields:
Pr(1793.4 θ 3772.2)=0.9 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 1
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
1
from ttps://web.cortland.edu/matresearch/CensorDatSTART.pdf
Chap 6 Estimation of Lifetime &Reliability Experiments &Tests 330
1-α=0.95 α=0.05,
MATLAB
χ 22,0.975 =chi2inv(0.025,2)=0.0506,
χ 22,0/975 = chi2inv(1-0.025,2)=7.3778
2×5×15.5 2×5×15.5
n=5,x (1) =15.5 θ 21 θ 3063
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 1 − 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…denoted by
up to which the fraction of the products fail. , in other words is
such that = 1− where R(.) is the reliability function.
331 Reliabilty Engineering
Suppose the lifetime distribution of the product is exponential
with mean θ then = 1− or = × ln and if the
estimate for is denoted by then we have the following
estimate for : = × ln
If we have a 1- CI such as < < for the mean
lifetime, then the following interval would be 1- CI for :
× ln < < × ln ( ) (6-13-1)
× ln( ) < < × ln ( ) (6-13-2)
where
= is when (the time or the kilometer or the temperature
or…) that fraction p 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.
6-7 Reliability Acceptance Sampling Plans
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
1
H108., Quality Control and Reliability Handbook (Interim) Sampling
Procedures and Tables for Life and Reliability Testing (Based on Exponential
Distribution),, in (Supply and Logistics)
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 − 14)
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 = 0.10), and to reject 95% of the
lots with mean life θ = 2000hr (β = 0.05). 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 1 1 1 1
335 3 5 10 20 3Reliabilty
5 Engineering
10 20
/
r n n n n n n n n
1 0 r
= 0.01 = ﻭ0.01 = 05 = ﻭ0.0.01
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 = 0.01ﻭβ = 0.05 α = 0.05ﻭβ = 0.05
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
= 0.01ﻭ = 0.1 = 0.05 = ﻭ0.1
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
= 0.01ﻭ = 0.25 = 0.05 = ﻭ0.25
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
= 0.1 = ﻭ0.01 = 0.25 = ﻭ0.01
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 = 0.1 = ﻭ0.05 α = 0.25 = ﻭ0.05
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 = 0.1 = ﻭ0.1 α = 0.25 = ﻭ0.1
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 = ﻭ0.25 α = 0.25 = ﻭ0.25
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 = 0.10 ,
& β = 0.05. 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.
6-7-3-1 Operating Characteristic curve for single sampling
plans (Pa versus Reliability)
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
= Pr( ≤ ) = ∑ (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 Estimation of Lifetime &Reliability Experiments &Tests 338
Fig. 6.1 OC curve for plan n=10, c=2
b)the lifetime
time is exponentially distributed then for T=100⟹
T=100
= = 0. From the curve or the table Pa=0.93.
6-7-3-2
2 Operating Characteristic curve for single sampling
plans (Pa versus mean lifetime)
The following example illustrates how to
t plot the OC curve
an
n acceptance sampling plan for inspecting a lot. If the sample
meets a special criterion, the
t lot will be accepted.
Example 6-12
6 (OC curve:Pa plotted versus mean lifetime)
(based on Grant&Leevenworth,1988 page 585)
Consider the following acceptance sampling plan 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 = ) or the product reliability. Table 6-3 shows the
λ
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
Pr( ≤ 2)
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 Engineering
To know how Pa is calculated , suppose 0.0003 ; since
n=11000 then np ' 11000 3.3 . The approximate value for
Pa from Poisson CDF table : Pa Pr( X 2) 0.359 0.360 .
The exact value of Pa is calculable from MATLAB :
= Pr((≤ 2) = binocdf(2,11000, 0.0003) = 0.3594
3594.
The following MATLAB commands plots the OC curve.
p=1/17000:.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 example
6-8
8 statistical hypothesis testing on mean and
minimum lifetime
In life testing, situations frequently arise where it is important
to determine if a new system meets a design goal 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, H0 : 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) )
i 1
n x (1)
ˆ ' ;calculate F0 ,
r 1 ˆ '
Using Table A at the end of the book or MATLAB command
finv (1 , 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)
ˆ '
Note that if Ho is rejected, the mean lifetime in this case is
estimated by ˆ ' . The minimum life is estimated from
t ˆ
ˆ '
x(1) and the relabity is estimared from e ˆ ' (K&L page258-9).
ˆ
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) )
i=1 (15-10)(2099199-190437)
θ= = +
r-1 10-1
=1908762
(190437-190437)+(245595-190437)+....+( 2099199-190437+ )
θ=2767421
10-1
nx (1) 15×190437
F0 = = =1.03
ˆ
θ' 2767421
The critical value of F is not in Table A, Using MATLAB:
Fα,2,2r-2 =F0.05,2,18 =finv(0.95,2,18)=3.5546 F0 <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
2 2r
6.Calculate 0 ,
0
7.Using Table E at the end of the book or
MATLAB command chi2inv(α, 2r), find 21 ,2r i.e. the critical
value of chi-squared distribution for the given .
8. Reject Ho if
2 2r
0 2,2r (6-18 )
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 =10 6 cycles
H1 :θ>θ 0
Solution
2 2r
0 r 10, n 15, 5%
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 chi2inv(0.95,20) yields:
2
%5,20 31.41 . 02 %5,20
2
H 0 : 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 = ( ), ( ), … , ( )
,
S = ( ), ( ), … , ( )
. For simplicity S 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, 2 − 2 deg of free.
Obtainable from Table A or MATLAB command
(1 − , 2 − 2, 2 − 2)
r2
C ( 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 0 if 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 0 if 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
Exercises1
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 = 0 .1 ).
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
10. Plot1 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
Dynamic Models+ Availability, Application of
Markov Chain
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
-R sys (t)
h 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 1
j
R sys (t) j i
h sys (t)=
m
R sys (t)
R i (t )
t 1
m
f 1 (t ) f m (t ) 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 ) , h (t ) ln R (t )
0
dt
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
m
d
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: h (t) = λ + C t , i = 1,2, . . . , m
where ٬ ﻭk constant values.
Then using Eq. 1-14-1
∫ ( )
( )= =
And using Eq. 7-1
( ) = exp − ∑ − ∑ (7-2-1)
Let ∗ = ∑ , ∗
=∑ , = ∑ = ∗
then (K&L p39)
∗ ( ∗ )
( ∗ ∗× × )
( )=e ∗
.
For large m it could be shown that :
∑ ∗
lim →∞ ( )= = = (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 (t) = + , is exponentially distributed with parameter
∑ .
Note that if ℎ (t) = Eq. 7-2-2 holds regardless of the value
of m. By substituting = 0 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 1...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 = constant
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 n
n
(7-4-2)
1 i=1 i
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 )dds ( 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 Fx(a), if
X(1) denotes the minimum we have:
n (7-8)
FX (1) ( y ) 1 1 FX ( y )
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:
n
G ( y ) F (1) ( y ) 1 1 F ( y )
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
n
1-FC (y)=1- 1-1-Fδ (y)
we could conclude that the reliability
of the n-link chain which equals RC=Pr( >s) is obtainable from:
n
RC f S (s ) 1 F (s ) ds (7-9)
0
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 °C. The system is a 10-
link series chain. The strength of each link is exponentially
distributed with mean 600 °C. 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
s s 10
1 500 600
R sys e e ds
0 500
56 S
1 3000 1 3000 6
R sys e ds
500 s 0 500 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.
7-1-3-1 Reliability in active redundancy
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 X denote the lifetime of the system, and , i=1,..n
denote the lifetime of the subsystems. Then in an active
redundant system:
1
From: , Lewis(1994) page 263, Grosh(1989 )page169,Li(2016)and
https://www.weibull.com/hotwire/issue21/
Chap 7 Dynamic Models+Availability, Markov Chain 366
( ≤ )= [ ( ,…, )≤ ]= ( ≤ ,…, ≤ )
Assuming , i=1,..n are independent:
( ≤ ,…, ≤ )= ( ≤ )… ( ≤ )
( ≤ )= ( ≤ )… ( ≤ )
⇒
1− ( ) = [1 − ( )][1 − ( )] … [1 − ( )] ⇒
( ) = 1 − [1 − ( )][1 − ( )] … [1 − ( )]
7-1-3-2 Reliability and MTBF in active system with exponentially-
distributed-lifetime components
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 ) eti , 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− =
e 1t e 2t et ( 1 2 ) t 0 (7-12-1)
If , are equal (λ1 =λ2 ) then:
1 2
R sys (t)=2e -λt -e -2λt =1-(1-e -λt ) 2 (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
0
1 1 1 1 1 1 1 (7 − 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 Mn .
k 1
Eq. 7-15 could be verified for n=2 from Eq. 7-13:
1 1 1 3 3
n=2 MTBFsys = + - = = θ
λ λ 2λ 2λ 2
and for n=3 from Eq. 7 − 14 − 2:
3 1 1 11 11
n=3 MTBFsys = -3× + = = θ .
λ 2λ 3λ 6λ 6
369 Reliability Engineering
The reliability of the above n-components active parallel system
is:
R s y s ( t ) = 1 -( 1 -e -λ t ) n (7-16)
Table 7-1Values of =∑ for
n=1,...,32 (Grosh,1989 p136)
n M n M
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 )
7-1-3-3 Reliability of standby parallel system
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.
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 t , R 2 (t)=e -λ t . 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-λ1t + λ1e-λ2 t (e(λ 2 -λ1 )t -1)
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) et (1t) 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
5
Then for t 10 , 100
2 2
R 91%, R 84.5%
standby active
2 2
5
for t 100 , 100 ,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
R (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:
2
3
t
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: = 20 , = , λ =
= , = 0.99
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
(20) = 1 − (1 − e 25 )(1 − e 10 )
= 1 − (1 − 0.4493)(1 − 0.13532) = 0.5238
The 20-year reliability of the second subsystem
(disinfection unit) is calculated from Eq. 7-21:
377 Reliability Engineering
2 202
t 20 20 2
R (t ) e t 1 t R (20) e 25 (1 25 25 ) 0.932
2 2
The entire system reliability is currently:
= 0.5238 ∗ .932 ∗ .99 = 0.4833
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 i
n ( t )
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
Failure rate and mean lifetime of active & standby
parallel systems having similar components
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)
0
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, λ=0.001, according to Eq. 7-19
2
R (t)=e -λt (1+λt) =0.99532
standby
Notice that if both units were active the according to Eq. 7-10:
2
R (100)=1-(1-e -λt ) 2 = 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.
Life pdf & reliability for Perfect switch: 2-component
standby system
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)
Special case: Constant failure rate & perfect switch( = )
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):
2 1
1e t 2e t
R sys (t) (7-30)
1 2 2 1
Chap 7 Dynamic Models+Availability, Markov Chain 384
Proof:
According to Eq. 7-18:
t t
Rsys (t)=R 1 (t)+ R 2 (t-t ) f1 (t ) dt =e 1t e 2 (t t ') 1e 1t ' dt
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 )
Life pdf & reliability for Perfect switch: 3-component
standby system
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
2
3 λt .
Eq. 7-21 i.e. R (t)=e -λt 1+λt+
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 j i
i 1 j 1
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 gives
i 1 l 1 i
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
Pˆs (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
Imperfect switch ,2-component standby system:
Reliability function and life time pdf
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 P = 1 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 1 p 1 .
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]et et (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):
2 t
R (t ) e t 1 1 e s t 0 (7-43)
sys s
The mean time to failure of is calculated as follows:
2
MTTFsys R (t )dt e t 1 1e s t dt
sys
0 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 Engineering
2
-Substituting
Substituting P s 1 in the above relationships yields R (t ) and
sys
MTTFsys for perfect switching which we saw earlier.
-The
The reliability function of a two-component
two component system having similar
active and standby units with the same failure rate λ is :
2
Calculated from Eq. 7-43
7
i.e. R (t ) e t 1 st given
sys
1 e
s
λs as the switch failure rate
2
Or is calculated from Eq. 7-39
7 i.e. R (t ) e t (1 Ps t ) if the
sys
reliability of the switch is given as a constant value Ps .
7 1
Example 7-6
A system is composed of 1 active unit A, 1 redundant
standby unit B and a switch S. The lifetime of the units is
exponentially distributed with 1 failure/1000hr. The key has a
constant rate of failure per 10000 hours. Find 500-hr
500 hr reliability
of the system and
a its mean lifetime.
1
From Dr Eshargh's pamphlet, Faculty member of Sharif University of Tech
, Tehran. Figure from K & L p 59.
Chap 7 Dynamic Models+Availability, Markov Chain 392
Solution:
Eq. 7-43 gives the reliability function:
103 failure / hr 110001, s 1100001 104,t 500
2
t s t
R (t ) e 1 1e 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 P = 0.97:
2
(t) = e−λt (1 + PS × λt)
Rsys
393 Reliability Engineering
2
(1000) = e−0.5 (1 + 0.97 × 500 × 10−6 × 1000)=0.9007
Rsys
For P = 1 from Eq. 7-19:
2 λ .
(1000) = e (1 + λt) = e (1 + 0.5)=0..9098
R
λt = 500 × 10 × 1000 = 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 t 1 (1 p )t e t (1 Ps t ) t 0
sys
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
1e 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
0
dt t
1 e s
switch average reliability Ps (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 1 (1 p ) (1 e ) dt
0 0
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):
2 t
R (t ) e t 1 1 e s t0
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 2
f x f a x a f a x a f a ...
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
1e s 1e s e dt
0
Ps
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
1e s or 1 e s or λs ≅ 61 ∗ 10−6 the 2 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
2 t
R e t 1 1 e s 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 + =0.10λ .
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 + = 0 .1 0 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
1
K&L page 222, Lewis(11996)page 260&chapter11
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.
Special case:
2-component shared loading: constant failure rates
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 t -λ t
when one unit fails. substituting g(t)=λ ge g , Rg (t)=e g and
-λ t
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 t 2 g t 2g 2 g t
R f (t)=e R sys (t ) e (e f t e ) (7-51)
2g 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× 10 per hour under partial load(
half load here) and 4× 10 under full load. Calculate 1000-hr
reliability of the system.
Solution
λ = 0.001 ℎ , λ = 4 × 10 ℎ
According to Eq.7-51:
2 g t 2g 2 g t
R sys (t ) e (e f t e )
2g f
R (1000) =
× . ×
2 × 0.001 . × × . ×
e + (e −e )
2 × 0.001 − 0.004
Using MATLAB
=exp(-2*0.001*1000)+2*0.001*(exp(-0.004*1000)-exp(-2*0.001*1000))/(2*.001-0.004)
⇒R (1000) = 25.24% End of Example
Chap 7 Dynamic Models+Availability, Markov Chain 402
7-2 System Effectiveness Measures
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’g 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 (7-52-2)
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 e - γ t , 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)
Mean time to repair(MTTR)
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 1
i 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.
Operational readiness(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. (7-54)
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
7-3 Availability
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…) and uses only operating time and actual repair time .
Therefore is computed from(K&L page 227):
operating time
= (7-55)
operating time + a. 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 =
. .
a.r.t.= total down time- administrative& logistic times=6-2=4.
8
A = = 0.69 = 66%
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 availability function for repairable systems –time to
failure (TTF)and repair time :exponential
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
= lim → ( )= (7-58-1)
μ
λ
or = (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 10 working hours. The average of its
actual repair time is 89.3 hr. The other subsystem failure rate is
4.78 failures per 10 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 × 10−6 ⟹
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
= 4.78 × 10
θ MTBF2 =1/4.78*10^-6= 209205 hr
(26280)= exp(-26280*4.78*(10^-6))= 0.8819
2 209205
= = 209205 =0.9958
2+ . . . 2 +890.3
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
A * (T ) A (t )dt (7-60)
T 0
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 ) R (t )dt (7-61)
T 0
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 ) R (t )dt TTF ~ exp( ) R (t ) e s t
T 0
1 T 1 e sT
Ps e s t dt
T 0 sT
Since the Taylor expansion of f(x) about a is;
1 1 2
f x f a x a f a x a f a ...
1! 2!
T
Then the expansion of e about a=0 :
- T T-0 (T-0) 2
e s =e-λs (0) + (-λ s )e-λs T |T=0 + [-λ s (-λ s )e-λsT ]|T=0 +...
1! 2!
1 1
e- sT 1-λ s T+ (λ s T)2 + (λ s T)3 +....
2 6
1 -λ T
Then for sT 1 , approximately e s 1-λsT+ (λsT) and:
2
2
417 Reliability Engineering
1
1-1+λ s T- (λ s T) 2
1-e-λs T 2 1
Ps = 1- λ s T A * (T )
λsT λs T 2
∗(
b) ) = 0.95 then
1 1
0.95 1- λsT T T 0.1 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 is the
μ
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
A =
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 (1e ) (7-65)
421 Reliability Engineering
7-4Application of Markov Chains to System
Reliability Analysis
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 Dynamic Models+Availability, Markov Chain 422
Fig 7-8
8 shows a diagram of the state space of the critical
system.. The possible states are denoted by symbols such as
(Computer A is active) , (Computer B is in standby mode)
Fig 7-8
7 The 9 possible states of a two-unit
unit system
(Barlow and Proschan , 1996p120)
The state space has 9 states labeled 0 through 8. For example
state o indicates that computer A is used actively while
computer B is operating and standby. If no failures occur in a
time interval of length t 0 , measured from the moment the
system enter State 0, preventive maintenance
ce is performed on
computer A and State 1 begins. If no failure occurs, the state
423 Reliability Engineering
passes around the perimeter of the square…. States 4,6 & 8 are
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.
Exercises
1. A system composing 20 independent components, each
having failure rate 0.0001 + 0.0005 , 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 t -λ t
g(t)=λ ge g under shared load and pdf f(t)=λ f e f under full
2 g t 2g 2 g t
load, .R sys (t ) e (e f t e ) would be the
2g 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
Enhancement, Optimization & Allocation of
Reliability
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.
8-1 Enhancement(Improvement) of system
reliability
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=0.8 =0.512.
b) The enhancement requires that each component has the
reliability of √0.95 . 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.
1
https://reliabilityanalyticstoolkit.appspot.com/standby_redundancy_integrate
Chap 8 Enhancement, Optimization & Allocation of Reliability 430
8-2 Reliability Optimization
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,…, xk) be the decision
variables; the model could be written as follows:
/ Max ( )
. . ( )≤0 = 1,2, … ,
ℎ( )=0 = 1,2, … ,
≥0
could be a cost function, the reliability function in the system
(series, parallel, structural load-strength systems…) or the
average lifetime.
431 Reliability Engineering
As an illustration consider the following series--parallel
system shown in Fig 8-5.
8
Fig.
Fig.8-5 A series-parallel system(Faghih,1996 p102)
k
Total number of components in the system is n i
. If we
i 1
suppose the components in Subsystem are identical, each
having reliability R i , then
n
The reliability if 1st subsystem= 1 (1 R1) 1 ,
n
The reliability if 1st Subsystem i 1 (1 R1 ) 1 .
The reliability of the entire system is:
k
R sys 1 (1 R i ) ni (8-1)
i 1
where
k number of subsystems
Ri The reliability of each component in Subsystem i
ni Number of components in Subsystem i
Now let
k = Maximum budget available
Ci = The cost of each component in Subsystem i
Then :
Chap 8 Enhancement, Optimization & Allocation of Reliability 432
n C
i 1
i i C ; (8-2)
And given specified values of Ri, i=1,2,…,k, the problem
would be determine ni, i=1,2,…,k in such a way that the
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
n
M ax R 1 (1 R ) i
i 1
i
s .t .
k
a ij ni b j i 1, 2,..., k j 1, 2,.., m
i 1
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
ni Number of components in ith 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
s.t.
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
ni
1 (1 Ri )
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 (f′ ، … ،f′ ،f′ )
subject which are to be minimized subject to constraints
(g ، … g ) ≥, ≤, = 0; multi-criteria decision making (MCDM)
techniques could be used. A mathematical model of an MCDM
problem could be written as follows:
Max{f ( ), f ( ), . . . , f (x)}
Min{f ( ), f ( ), . . . , f ( )}
s.t.
g ( ){≤, =, ≥}0 i = 1,2, . . . , m
where x is a vector including the decision variables
8-3 Reliability Allocation 1
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) …..(1-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
R = (1 − R) (R)
x
GivenR , k and n, R has to be determined in such a way that
(∑ (1 − R) (R) − R )=0.
This could be done using a software.
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:
= =3 = 0.8573
The reliability of each component is;
= √0.8573 = 0.95
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 = 0.8573 ⟹
=1− (1 − 0.8573) = 0.4774
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:
5
(1 − ) ( ) − 0.99144 = 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):
∗ ∗
+. . . . + ≤ 0
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( = 0.95). 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
= = = 0.005, = 0.003 , = 0.001
200
n
∗
= , w i i / i
i 1
0.005
= = 0.555, w = 0.333 ,
0.005 + 0.003 + 0.001
w = 0.111
To find the required failure rate() for this exponentially-
distributed-lifetime system we could write:
(20) = e ⟹ 0.95 = e ⟹ = 0.00256 ℎ
443 Reliability Engineering
The required failure rate for each subsystem is calculated
∗
from Eq. 8-7 i.e. = as follows:
∗
= (0.555)(0.00256) = 0.00142
∗
= (0.333)(0.00256) = 0.000582
∗
= (0.111)(0.00256) = 0.000284
Since the lifetime of each subsystem is exponentially
distributed, the allocated reliability for them to ensure a 20-hour
operation of the system are:
∗ (20) ∗ ( . )
=e =e = 0.97
∗ (20) ∗
=e = 0.98 , R∗ (20) = 0.99
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).
( )( )( )
MTBF = [ ( )]
= 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(0 < ≤ )
Importance index for the ith subsystem; probability that
the system mission fails if ith subsystem fails.
is the following quotient(Grosh, 1989 p149)
subsystem fails
=
number of subsystem failure
=1 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 = .
210 ∗ 1 ∗ 10
= = 2729.4
(15)(− 0.95)
1 1
= = = 0.00036638 = 36638 × 10
2729.4
210 ∗ 0.95 ∗ 9
= = 1400.2 = 0.0007142
(25)(− 0.95)
210 ∗ 1 ∗ 10
= = 409.41 = 0.002442
100(− 0.95)
447 Reliability Engineering
210 ∗ 0.9 ∗ 8
= = 421.1 = 0.002374
(70)(− 0.95)
Eq. 8-10-1&2 allocates the required reliability to each subsystem:
∗ 1−[ ( )]
=1−
∗
1 − (0.95)
=1− = 0.99634
1
∗ ∗ ( . )( )
or = e =e = 0.99634
similarly
∗
1 − (0.95)
=1− = 0.99359
0.95
∗ ∗ ( . )( )
or = e =e = 0.99359
∗ (t)
= 0.975870,
∗ (t)
= 0.98116
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 and 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
Exercises
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 150 × 10 , 18 × 10 , 2.3 ×
10 , 5.6 × 10 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 0.0002 × (μ ) 30000 ≤ μ ≤ 75000
.
σ 800 × (σ ) 1000 ≤ σ ≤ 10000
.
8997 × ( ) 10000 ≤ ≤ 68000
.
366 × ( ) 500 ≤ ≤ 7500
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
References
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., 1972
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·Using MATLAB. 2006
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.
Every thing comes to him who waits
455 Reliability Engineering
T A B L E S
Table A Crtical values of F distribution Fm,n ( 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 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 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 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 12 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 Table B Some values of CDF of Poisson Distribution 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
6.40 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
6.60 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 Table B Some values of CDF of Poisson Distribution 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 Table B Some values of CDF of Poisson Distribution 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(Pr(Z < )) e.g. Pr( < −3.00) = 0.0013
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. :Pr(Z < −1.56) = 0.05938
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
465
Table C continued e.g. Pr( < 1.04) = 0.14917 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. Pr(Z < 3.44) = 0.99971
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
Table D Critical values of standard normal ( ) e.g. . = 1.64
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 = 0.05 = . = 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
Table E Critical values of Chi Distribution( , )
.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
Table F Chracteristic of some continuous distributions
pdf MGF Mean Variance
1 ,a x b tb
(t ) e e
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
n
Gammaa(n, λ) e x (x )n 1 n n
n integer , 0 ,x 0
(n 1)! t 2
2 n
1
x
2 2
2t 2
Normal((μ, σ) e x exp t 2
2 2
2 2
ln t
1 2 exp 2μ+σ2 σ2
Lognormal (μ, σ) e 2 t 0 2
e (e -1)
t 2
Beta with parameters (a b ) a a b
a>0 ،b>0 cx a1(1 x )b 1,0x 1 c
(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 )
2 x A C ( )
B 2 (1 C2 ) B 2 (1 C1 ) (
B
) B B e B Weibull
C 1 e x A x A
1
xA C
1 (1 C x-A ) -1- C1 valid on
1 (1 C ) valid on
B B B B
A 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 & C0
0
x A B & C0
C
1
1 x
1
A C 1 xA
x A (1 C ) e(1 C B ) C
(1 C ) C B
e B
for
x A B &C 0 GEV
C
B B
1 x A & C0 x A &C 0
C
C
x A B &C 0
1
C
2B2 A γB x A xA
B 1
exp( x A e FT1
6 0.05772 e e B B
B )
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
Table G Chracteristic of some discrete distributions
Probability function p(x) MGF Z-transform mean Variance
n x n x
Binomial (n,p),0≤p≤1 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,
0 p 1 1 (1 p )e t 1 (1 p ) z p p2
Geometric for failure p
x
, p 1 p 1 p
with parameter p(1 p) x 0,1, 2, 1 (1 p )e t
0 p 1 t ln(1 p )
1 (1 p )z p p2
Table H MATLAB1 Commands related to Distributions
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,m,n) 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,m,n) 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
ABOUT THE AUTHOR
The author received his B.S. in Industrial Engineering (IE) from a
University of Technology in Tehran, in 1976 and his MS degree
in IE from University of Pittsburgh(Pitt) ,PA in 1978. He was
employed as a faculty member in Kerman, Iran in 1979. He
started to continue his studies for PhD at Pitt in 1985; after 2
semesters he left USA for home; however he received PhD
from Brunel University of London in July 2006. He has taught
some courses for over 30 years
The author has published some textbooks in Persian; some articles in
conferences and journals and supervised several graduate
theses. He was retired in 2015 from his job as a faculty
member at a university in his hometown Kerman, Iran.
Chairman of IE and ME departments are among his
responsibilities at the College of Engineering of Shahid
Bahonar University of Kerman , Iran.
If youth but knew, If old age but could,
Si jeunesse savait, Si vieillesse pouvait
(French Proverb)
You might also like
- Analysis of The Micro-Environment of VinfastNo ratings yetAnalysis of The Micro-Environment of Vinfast1 page
- Engineering Reliability - Richard E. Barlow100% (1)Engineering Reliability - Richard E. Barlow221 pages
- Semester & Branch: I Me - : Course Plan 2012-2013No ratings yetSemester & Branch: I Me - : Course Plan 2012-20136 pages
- Reliability Engineering ECE-419E: EWP New DelhiNo ratings yetReliability Engineering ECE-419E: EWP New Delhi3 pages
- An introdution to reliability and maintainability engineering 2nd ed Edition Ebeling - Quickly download the ebook in PDF format for unlimited reading100% (1)An introdution to reliability and maintainability engineering 2nd ed Edition Ebeling - Quickly download the ebook in PDF format for unlimited reading49 pages
- Reliability Engineering (Lecture Note On Topic 1 and 2)No ratings yetReliability Engineering (Lecture Note On Topic 1 and 2)23 pages
- An introdution to reliability and maintainability engineering 2nd ed Edition Ebeling - Read the ebook online or download it to own the full contentNo ratings yetAn introdution to reliability and maintainability engineering 2nd ed Edition Ebeling - Read the ebook online or download it to own the full content80 pages
- Download ebooks file (Ebook) An introdution to reliability and maintainability engineering by Ebeling, Charles E ISBN 9781577666257, 1577666259 all chapters100% (1)Download ebooks file (Ebook) An introdution to reliability and maintainability engineering by Ebeling, Charles E ISBN 9781577666257, 1577666259 all chapters82 pages
- RE Student Note-Lecture-1 New 2020 05 22 13 59 25 259No ratings yetRE Student Note-Lecture-1 New 2020 05 22 13 59 25 2594 pages
- Instant Access to An introdution to reliability and maintainability engineering 2nd ed Edition Ebeling ebook Full Chapters100% (3)Instant Access to An introdution to reliability and maintainability engineering 2nd ed Edition Ebeling ebook Full Chapters71 pages
- Introduction To Reliability in Mechanical DesignNo ratings yetIntroduction To Reliability in Mechanical Design38 pages
- SY1920 2ndsem EE181 ACR-Herrera-Juan-PaoloNo ratings yetSY1920 2ndsem EE181 ACR-Herrera-Juan-Paolo14 pages
- Practical Reliability Engineering: Third RevisedNo ratings yetPractical Reliability Engineering: Third Revised7 pages
- INDE 6336 Reliability Engineering: Instructor: Dr. Qianmei (May) Feng E217-D3, (713) 743-2870 Qmfeng@uh - EduNo ratings yetINDE 6336 Reliability Engineering: Instructor: Dr. Qianmei (May) Feng E217-D3, (713) 743-2870 Qmfeng@uh - Edu27 pages
- Probabilistic Reliability Engineering (PDFDrive)No ratings yetProbabilistic Reliability Engineering (PDFDrive)536 pages
- Dimitri Kececioglu - Reliability Engineering Handbook Vol. 2 PDFNo ratings yetDimitri Kececioglu - Reliability Engineering Handbook Vol. 2 PDF586 pages
- Reliability Engineering Handbook (Volume 2) by Dimitri KececiogluNo ratings yetReliability Engineering Handbook (Volume 2) by Dimitri Kececioglu586 pages
- Risk Analysis For Information and Systems Engineering: INSE 6320 - Week 6100% (1)Risk Analysis For Information and Systems Engineering: INSE 6320 - Week 69 pages
- Dimitri B.) Reliability Life Testing Vol-INo ratings yetDimitri B.) Reliability Life Testing Vol-I798 pages
- IE 443 Chapter 3 Reliability Analysis 2023No ratings yetIE 443 Chapter 3 Reliability Analysis 202355 pages
- Advanced Multiplayer Game Development with Ureal Engine 5: A Comprehensive Guide to C++ ScriptingFrom EverandAdvanced Multiplayer Game Development with Ureal Engine 5: A Comprehensive Guide to C++ ScriptingNo ratings yet
- Instant download Human Variation Races Types and Ethnic Groups 6° Edition Stephen Molnar pdf all chapter100% (1)Instant download Human Variation Races Types and Ethnic Groups 6° Edition Stephen Molnar pdf all chapter67 pages
- Liquid-Propellant Rocket Engine Injector Dynamics: Vladimir G. Bazarov and Vigor YangNo ratings yetLiquid-Propellant Rocket Engine Injector Dynamics: Vladimir G. Bazarov and Vigor Yang10 pages
- Geoforum: Marien González-Hidalgo, Christos ZografosNo ratings yetGeoforum: Marien González-Hidalgo, Christos Zografos13 pages
- A Novel Study On Machine Learning Algorithm BasedNo ratings yetA Novel Study On Machine Learning Algorithm Based10 pages
- Electrical Estimation & Costing - 18EE822 2022-23: Mrs - Tarakeshwari V, Asst. Professor Department of EEE, SJBITNo ratings yetElectrical Estimation & Costing - 18EE822 2022-23: Mrs - Tarakeshwari V, Asst. Professor Department of EEE, SJBIT27 pages
- Module 2: Lecture 5 Fourier Series Decomposition and Its ApplicationsNo ratings yetModule 2: Lecture 5 Fourier Series Decomposition and Its Applications4 pages
- Risks and Risk Management in Agriculture: Ludwig TheuvsenNo ratings yetRisks and Risk Management in Agriculture: Ludwig Theuvsen13 pages
