PART V: Product and Process Design
CHAPT ER
| 11)
Reliability
Chapter Outline
11-1 Introduction
11-2. Reliability
11-3 Life-Cycle Curve and Probability Distributions in Modeling Reliability
11-4 System Reliability
11-5 Operating Characteristic Curves
11-6 Reliability and Life Testing Plans
11-7. Summary
Case Study
Key Terms
Exercises
Referencés
jan ra
aoe R11 Symbols
nile 2 Constant failure rate : : | vt
Ft Probability density function for time to failure ad
{{) Probability distribution function for time to failure
(i) Reliability function at time r |
rt) Fairare-rate function
Probability of lot acceptance | on
6 Meantlife of product | ef
n Sample size At
r Rejection number
T Predetermined test time
a Producer's risk |
@ Mean life associated with producer's risk inet
B
‘Consumer's risk give
@ Mean life associated with consumer's risk
a
} | rob
l |
11-1 INTRODUCTION |i
| pod
In earlier chapters we dealt with quality assurance and control and explored the nin | wal
of quality during the manufacture of the product or the rendering of a service, Our pone
cussion focused on a specific instant in time. In this chapter we examine the conceptol
reliability.
Reliability is a measure of the quality of the product over the long run. Inhee the
the concept of reliability is an extended time period over which the expected operatin
of the product is considered: that is, we expect the product will function acrordng»
certain expectations over a stipulated period of time.
To ensure customer satisfaction in the performance phase, we address mesus |
to improve reliability in the design phase. The complex nature of products rat | Ty
many components in their construction; thus, we need to be able to calculate syS'€™ |
-reliability. With the customer and warranty costs in mind, we must know the chance" | a
successful operation of the product for at least a certain stipulated period of time Such ty
information helps the manufacturer to select the parameters of a warranty. policy.
|
|
| 11-2 RELIABILITY ~
Reliability is the probability of a product performing its intended function forasitl
Period of time under certain specified conditions, Four aspects of reliability 7%
waluear this definition. First, reliability is a probability-related concepts Ne ts
alue of this probability is between 0 and 1. Second, the functional performance
a eaiee as 7 meet certain stipulations. Product design will usually ensue or
he Meh luct that meets or exceeds the stipulated requirements. For eat wn
hase, i ig strength of a nylon cord is expected to be 1000 kg. then in its OP yi
Plies sabe cord must be able to bear weights of 1000 kg or more. "Third, rlab
plies successful operation over a certain period of time. Although 10 Pte ert
Feast a inal ney the time requirement ensures satisfactory performates b
ed period (say, 2 years), Fourth, operating or envitOO™ME 4 spt
t ;
ions under which product use takes place are specified. Thus, if the nylon’ forma ®
itions indoors, then satisfactory ee the UA
ions. In the context of these four sper sul pf
scribed as having a probability of sue
000 kg for 2 years under dry conditions
ified to be used under di
= 4
expected for use under those. seca
bility of the nylon cord might be de:
mance of .92 in bearing loads of 1|
. LIFE-CYCLE CURVE AND PRO!
14 ot MODELING RELIABILITY BABILITY DISTRIBUTIONS
Most products gO through three distinct phases from product i
fue 2] SHOWS atypical life-cycle curve for which the failure rare hs cloner os
Figen of time. This curve is often referred to asthe bathtub curve; it con wa ot the
tetiing phase, the chance-failure phase, and the wear-out phase (Besicrfeld 199,
paisa and Sheldon 1968; Halpem 1978). Phase (Besterfield 1990;
‘The debugging phase, also known as the infants
in ine failure rate as initial problems identified during eres Assinar tea
‘he chance-failure phase, between times f, and f, is then encountered failures occur
‘Tndomly and independently. This phase, in which the failure rate is constant, ily
resents the useful life of the product. Following this is the wear-out phase, inwhich
anincrease in the failure rate is observed. Here, at the end of their useful life, "parts i
and wear out.
Probability Distributions to Model Failure Rate -
Exponential Distribution
The life-cycle curve of Figure 11-1 shows the variation of the failure rate as a
function of time. For the chance-failure phase, which represents the useful life of the
product, the failure rate is constant. As a result, the exponential distribution can be
ried to describe the time to failure of the product for this phase. In Chapter 4, the ex-
ponental distribution was shown to have a probability density function given by
f(g=aew, 120 (ua)
rate, Figure 4-24 shows this density function
where 2 denotes the failure
ure (MTTF) for the exponential distribution was given in
The mean time to fail
eq, (4.40) as
MTTF=1/A 1.2)
Thus, if the failure rate is constant, the mean time to failure is the reciprocal of the fail-
ure rate. For repairable equipment, this is also equal to the mann time between fi
ures (MTBF). There will be a difference between MTBF and MTTF only if there is a
significant repair or replacement time upon failure of the product
Debugging ‘Wear-out
Chance-failure phase phase
phase
i a
Time (t)
FIGURE 11-1 A typical life-cycle curve.
515 @
ReliabilityPCD ah Peer eres eee EG na eo
a The reliability at time , R(s), is the probability of the product lasting up to a,
CHAPTER 11 least time 1. It is given by
R()=1-FO
1
a1 [eva- 7 (3)
where F(t) represents the cumulative distribution function at time ¢ ‘The cumulative
distribution function for the exponential distribution is shown in Figure 4-25, Figure
11-2 shows the reliability function, R(0, for the exponential failure distribution. At
time 0, the reliability is 1, as it should be. Reliability decreases exponentially with time.
Tn general, the failure-rate function r(‘) is given by the ratio ofthe time-tofallue
probability density function to the reliability function. We have
£0
rO=R@ (4)
For the exponential failure distribution
de®
O= Sr
implying a constant failure rate, as mentioned earlier.
Example 11-1: An amplifier has an exponential time-to-failure distribution with 2 failure
rate of 8% per 1000 h. What is the reliability of the amplifier at 5000 h? Find the mean timeto
failure.
Solution. The constant failure rate 4 is obtained as
The reliability at 5000 h is
=
=
3
FIGURE 11-2 Reliability function for the exponential
time-to-failure distribution.
Time (t)me to failure is
snl
MTTF = 1/A = 1/0.00008 = 12,500 h
|-2: What is the highest fail
ple 11 2: lure rate for a
hat is, successful operation) of 95% at
ie (oral distribution. ‘at 4000 h?
Product if itis t
Assume that the (ey a Probability of
"
" ie time to failure follows
pero
The reliability at 4000 h is .
Solution. ry is 95. If the constant failure rate is given by 2, we have
R()=e*
’
195 = e-A000
Wehave
40004 = 0.05129
A= 0.0000128/h = 12.8/10°h
‘us the highest failure rate is 12.8/10° h for a reliability of .95 at 4000 h.
Weibull Distribution
The Weibull distribution is used to model the time to failure of products that
we a varying failure rate. It is therefore a candidate to model the debugging phase
(tilure rate decreases with time), or the wear-out phase (failure rate increases with
tine) (Henley and Kumamoto 1981). The Weibull distribution was introduced in
Chapter 4. It is a three-parameter distribution whose density function is given in
¢4 (443) as
no=A(! NP eal (3? y Ry (1s)
a\ a a
(a0 < 7<), a scale parameter, a (a> 0),
Ne
The parameters are a location parameter, ° i
anda shape parameter B (B> 0). The probability density functions for 7= 0,a=1 and
‘eeral values of B (B= 0.5, 1,2. and 4) are shown in Figure 426. ie yag
“Appropriate parameters are used to model a wide variety oF sitar Wye
aad pT pe Werbll dieseibution reduces to the exponential distribution. Por relabi
‘iy modeling, the location parameter 7will be zero. If a= 1 and B=0.5, for am
theaiure ee descents wath time and can therefore be used to mode) comPArEN
debugging phase. Likewise, if a= 1 and B= 355, the failure fate increases with ine
on so can be used to model products in the wear-out phase. In this case, the
istibution approximates the normal distribution. |
The reliability Tanction for the Weibull distribution is given by
cy (11.6)
R()=exP| a
The mean time to failure, as given by €4- (444). is
1 (11.7)
MTTF = ar 3!
me-to-failure probability istibulion is
Thefailure-rate function r(1) for the Weibull
fo. _ Be (11.8)
ripe san” B
RO, re density
ws the shape of a nction for the Weib\ | failus
; i ce fu for the Wei
3 shor he failure a }
ig
"etion. for values of the parame!9518
CHAPTER 11
Failure cate (rt)
FIGURE 11-3 Failure-rate functions for the Weibull time-
to-failure distribution for f = 0.5, 1, and 3. Time (1)
Example 11-3: Capacitors in an electrical circuit have a time-to-failure distribution that can
be modeled by the Weibull distribution with a scale parameter of 400 h and a shape parameter
of 0.2, What is the reliability of the capacitor after 600 h of operation? Find the mean time to
failure. Is the failure rate increasing or decreasing with time?
Solution. The parameters of the Weibull distribution are a = 400 h and f= 02. The
location parameter 7is 0 for such reliability problems. The reliability after 600 h of operation §
given by R= «l-(2)
---(&) |
3381
‘The mean time to failure is
The failure-rate function is
0.
10= oop
=0.0603r°*
This function decreases with time. It would model components in the debugging Phas*
-YSTEM RELIABILITY. oe
Most products are made up of a number of components. The reliabil
onent and the configuration of the system consisting of these components det ‘desis
Fhe system reliability (that is, the reliability of the product). Although product
11-4 S$)
com
ity of cath inesagetureand maintenance influence reliability, improving reliability is ager the
| Mn of design. One common approach for increasing the reliability of the system is
| ai gundance in design, which is usually achieved by placing components in par-
| emiGTong as one component operates, the system operates, In this section we
how to compute system reliability for systems that have components in
‘
trate
ceo arallel, or both. .
cis, in paral
| systems with Components in Series “
gaa 11-4 shows a system with three components (A, B, and C) in eres. For the sys-
fio opefate, each component must operate. It is assumed that the components op-
wie independently of each other (that is, the failure of one component has no
{gyence onthe failure of any other component). In general if there aren components
| duis, where the reliability of the ith component is denoted by Ri, the system relia-
vityis
Ry=RyXR2X+X Ry (119)
~ | Thesystem reliability decreases as the number of components in series increases.
though overdesign in each component improves reliability, its impact would be
tibet by thé number of components in series. Moreover, manufacturing capabilities
and resource limitations restrict the maximum reliability of any given component.
1x | Product redesign that reduces the number of components in series is a viable
et allemative.
0
Example 11-4; A module of a satellite monitoring system has 500 components in series. ‘The
{ relablty ofeach component is 999. Find the reliability ofthe ‘module. If the number of compo-
|
¢ | nents in series is reduced (0 200, what is the reliability of the module?
Solution, The system reliability for the module is
4 Ry= (.999)5 = .606
| Note that even with a high reliability of .999 for each component, the system reliability is only
‘95%. When the number of components in series is reduced to 200, the reliability of the module is
' R,= (.999)=.819
Use of the Exponential Model
Ifthe system components can be assumed to have a time to failure given by the
exponential distribution and each component has a constant failure rate, we can com
Put the system reliability, failure rate, and mean time to failure, As noted earlier,
then the components are in the chance-failure phase, the assumption of a constant
, lure rate could be justified.
steg SUPPOSE the system has n components in series, each with exponentially distrib
ie ime-to-failure with failure rates Ay, 22, « Aw The system reliability is found as
Product of the component reliabilities:
\
Rose aK KM
=enp| - (S| (11.10)
Fiation (11.10) implies that the time to failure of the system is exponentially
ted with an equivalent failure rate of S“”., ,. Thus, if each component that
{+}
distri
519 w
Reliability
i FIGURE 11-4 System with components in series.ae
= 520 fails is replaced immediately
CHAPTER 11 to failure for the system is given by
1
MTIF=—
Da
i
When all components in series have an id
system (a special case of eq. (11-11)] is given by
1
ME Sad
Example 11-5: The automatic focus unit
Each component has an exponential time-to-
{05 per 4000 h, What is the reliability of each
liability of the automatic focus unit for
Solution. The failure rate for each component is
A =.05/4000 h
=125x10%h
“The reliability of each component after 2000 h of operation is
xpf- (12.5 x 10°*)2000]
975
The reliability of the automatic focus unit after 2000 h of operation is
R, = exp[-(10 x 12.5 x 10°°)2000]
=.719
‘The mean time to failure of the automatic focus unit is
MTTF = 1(10 x 12.5 x 10)
= 8000 h
Example 11
with another that has the same failure rate the meaning
jentical failure rate, say A, the MTTF for the
of a television camera has 10 components in series,
“allure distribution with a constant failure rate gt 44°
‘component after 2000 h of operation? Find th
‘2000 h of operation. What is its mean time-to-failure?
Refer to Example 11-5 concernin; i
g the automatic focus unit of at
camera, which has 10 similar components in series. It is desired for the focus unit
ae
(Ul,
Tere!
elevision |
to have arel
bility of | |
bt of $8 afer 2000» of operation, What would be the mean time to fire of the india |
Solution. Let A, be the failure rate of the focus unit. Then, A, = 102, where Aree y
is found from :
an
the failure rate of each component. To achieve the reliability of .95 after 2000 h, the valve {ft
|
|
95 = exp[-A,(2000)}
or
0.0000256/h
‘The failure rate for each component is
A=aslo
= 0.00000256/h
=256/10%h
The mean lime to failure for cach component
nt would be
/ MITF=1/A
a Raebraking mechanism is a critical system in the
Sessothatifone fails, the brakes still work. (ne SUOmebIle. Dual subsystems thus
Suppose we have n components in Parallel, with the reliabilit: of the
l. th
satdenoted by Ri 1.2... .m. Assuming tha the components opertc mean,
wlindependently of each other, the probability of failure of each component maven
bfi=1~R,, Now, the system fails only if all the components fail. Thus the probabil-
iy of system failure is ;
Fy=(1~ Ry)(1~ Ra) += (1 Ry)
=Tla-r)
‘ot
Thereliability of the system is the complement of F, and is given by
R=1-F,
| =! fla R) (11.12)
i=
| Use of the Exponential Model
Ifthe time to failure of each component can be modeled by the exponential dis
‘rbution, each with a constant failure rate 4,,¢= 1, .....n, the system reliability, as
Suming independence of component operation, is given by
| R=) fla-e
et
E -Tla-e*) (11.13)
it
| The time-to-fal t exponentially distributed.
-t0-failure distribution of the system is Not Expo! ly
.__ In the special case where all components have the same failure rate A, the system
| ‘lability is given by
R=1-(1-6°"
S218
Reliability
in parallel.
c FIGURE 11-5 System with component?a5 " For this situation, the mean time to failure for the system with 7 identical
in parallel, assuming that each failed component is immediately replaced by an iden,
cal component, is given by
Af sade vie
MITF=3\1+5+3 a dig
Example 11-7: Find the reliability of the system shown in Figure 11-5 with three
nents (A.B, and C) in parallel. The reliabilities of A, B, and C are 95, 92, and 90, respect,
Solution. The system reliability is
R,=1-(1—95)(1 = 92)(1 = 90) #
= 1.0004 = 9996
Note that the system reliability is much higher than that of the individual components. Design.
ers can increase system reliability by placing more components in parallel, but the cost of head.
ditional components necessitates a trade-off between the two objectives.
Example 11-8: For the system shown in Figure 11-5, determine the system reliability for
2000 h of operation, and find the mean time to failure. Assume that all three components have
‘an identical time-to-failure distribution that is exponential, with a constant failure rate
$f 0.0005/h, What is the mean time to failure of each component? If it is desired for the system
to have a mean time to failure of 4000 h, what should the mean time to failure be for each
component? 7
Solution. The failure rate of each component is 2 = 0.0005/h. For 2000 h of operation,
the system reliability is
R,=1~ {1 — expf-(0.0005)2000}}?
1~ (.63212)° |
=.7474
The mean time to failure for the system is
«
ti
By placing three identical components in parallel. the system MTTF has been increased ®
___ Fora desired system MTTF of 4000 h, of elo
vidual oT ested sistem 9 |; We now calculate the required MTTF i
so00= (14342
7273
where 2s the failure rate for each component. Solving for A, we get
1.8333,
az =0.
00046/h
‘Thus, the MTTF for each component would have to be
\
1 i
MTTF 2
A oooows =717591Rams with Components in Series ang inP,
‘aralle| ~
systems often consist of co,
ool e m|
miity calculations are based on then nn that are both in seri
iously disc series
“ussed
nents operate independently, and in parallel,
ce
‘Oncepts, assuming that
eco
Find the reliability of the
114
gample
ts are in seric eight-cor
Bereoponen ies and some are jenecomponent
Talons: Ras = 92, Raa = 90, Ras an Baal. The velabitiee a in Figure 11.6;
Raa = 96, Ra, = 95, Reon she components
98, Ro, = 90, Ry, = 92. and
33 = 92, and
Solution. We first find the reliabil
sash Ao, As,and A, the reliabil vig Saeh subsystem. For the subays
stem with compo-
R
1-(1- Ra, Rag\(1-R,
= [1 (92)(90)I~ Cane
ray | OPK SOE - (88). 96)
Saal, the reliability of the subsystem with components B;, B.
1» Bz, and By is
Rp=1-(1~Rp,)(1- Rey)
1 (1 Res)
=1-(1-.95)(1 -.90)(1- ie
rt )(1 ~ 90)(1 - 92)
Tesjstem reliability is given by
Ry x Rox Ro,
(.9733)(.9996)(.93)
= 9048
Use of the Exponential Model
fae time to failure for each compone
moa system reliability and mean time to failure can
tions using previously discussed procedures.
ture rate and the mean time to failure for the eight
“The failure rates (number of units per hour) for the
= 0.0016, 4, = 0.0060.
‘nt can be assumed to be exponentially dis-
be calculated under cer-
Exar
me 11-10: Find the system fail
eee system shown in Figure 11-6.
ents are as follows: Aa, = 0.0006, 2a; = 0.0045, Aas = 0.0035; Aas
#y,=00060, 4p, = 0.0060, Acy Ps
0050.
«i, SMlution. First we compute failure rates for the subsystems. We have 2 failure rate
ini + a3) = 0.0051 for the Ai/A2 re rate is
Oye, ‘subsystem: for the As/As subsystem. the fail
»* 4a.) =00051. The mean time to
failure for the Ay/A2/As/Aa subsystem 5
1 1
arr, = te ( 1+ 5) = 20808"
Mi star i)
The
ce
an ime to failure forthe subsystem consisting of components Rasa
1+5tg
1 1 1) sosss5m
verre sa 23
By+ 524
CHAPTER 11
The system failure rate is b
+ 0.005 = 0.01167
294.118 — 305.;
The mean time ‘o failure for the system is c
MTTF, = 85.69 h
Systems with Standby Components ~
y
In a standby configuration, one or more parallel components wait to take over open.
tion upon failure of the currently operating component. Here, it is assumed tha ™
only one component in the parallel configuration is operating at any given time. Be. (s
cause of this, the system reliability is higher than for comparable systems with compo.
nents in parallel. In parallel systems discussed previously, all components are assumed:
to operate simultaneously. Figure 11-7 shows a standby system with a basic component
and two standby components in parallel. Typically, a failure-sensing mechanism
triggers the operation of a standby component when the currently operating com-
ponent fails.
r
Use of the Exponential Model
If the time to failure of the components is assumed to be exponential with failure
rate A, the number of failures in a certain time 1 adheres to a Poisson distribution with
Parameter /. Using the Poisson distribution introduced in Chapter 4, the probabiliy
Of x failures in time 1 is given by
aragy® re
See (115)
P(x)
x!
For a system that has a basic component in parallel with one standby component, the
system will be operational at time 1, as long as there is no more than one failure. Thert-
fore, the system reliability would be | a
Rees ean,
For a standby system with a basic component and two standby components (shown
Figure 11-7), the system will be operational if th Jess than of
equal to 2. The system reliability is is wumber of fellures jp es
Ree era edt ae
Basic
component
Standby *
component
Standby \
FIGURE 11-7 Standby system, component 2 \
‘:
| eral, if there are n components on standl
i iby along with the basi
| jot (+1) components inthe system] the system scale scomponent fora its
| liability
| Ro=e*
! (11.16)
| spemean time to failure for such a system is
MTTF="+ 1
{ 2 (1417)
mple 11-11: Find the reliability of the Standby system shown in Figure 11-7,
| jysic component and two standby components, each having an exponential time
iipaion, The failure rate for each component is 0.004/h and the period of operation is 306 5,
+ Whatis the mean time to failure?
Ifthe three components are in parallel (not in a standby mode),
| thesystem? What is the mean time to failure in this situation?
Solution. The failure rate A for each component is 0,004/r. For 300 h of operation, the
stem reliability is
! nner ( 14414 2)
jomeamr|
2
=e cmc + 0.004(300) +. I
121 41.2+0.72}
879
what is the reliability of
n+l_ 3
MTTF —_ =750h
A 0.004
Ifthe system has all three components in parallel, the probability of failure of each com-
+ ponent is
ew
— ¢-21004(300) = 6988
The system reliability is found as
(6988)
= 659
The mean time to failure for this parallel system is
1 west
sie 3(ss3)
= 458.333 h
fi at the syst ‘ability and MTTF of the standby and pars lel systems differ sig:
pte thy the standby and paral
y
licantly tem rel
1c CURVES”
15 OPERATING CHARACTERIST!