Reliability Theory

 What is reliability?
 Why it is important to study?
 Use of reliability in day to day life.

Reliability may be defined in several ways:

The idea that an item is fit for a purpose with

respect to time.
 In the most discrete and practical sense: "Items
that do not fail in use are reliable" and "Items that
do fail in use are not reliable”.
Reliability of a device is denoted as R (t ) and
defined as
R(t )  P T  t  , t  0.

Where “T” is a continuous random variable

representing the life length of the device.
Some property of the reliability function R (t ) :


(i) R(t )   f (u)du
t
Where “f ” is probability density function (p.d.f) of “T”.

We assume
f (t )  0 for t  0

(ii ) R(t )   f (t )
(iii ) R(t ) is a nonincreasing function of t

(iv) R(0)  1 , R()  0

(v) R(t )  1  F (t ) where F (t ) is the cdf of T

(v) R(t )  1  F (t ) where F (t ) is the cdf of T

R(t )  P T  t 

or , R(t ) 1  P T  t 

or , R(t ) 1  F (t )

Hence the results

(iv) R(0)  1 , R()  0

(v)  (iv)
(iii ) R(t ) is a nonincreasing function of t
(iv)  (iii)

(ii ) R(t )   f (t )
(v)  (ii )

(v)  R(t )  1  F (t )
 R(t )   F (t )   f (t )

Hence the results

Conditional failure (hazard) rate function:

It is denoted by Z (t ) or h(t )

P  t  T  t  t T  t 
Z (t )  lim
t  0 t

For small t
Z (t )t  P t  T  t  t T  t 
 For small t
Z (t )t  P t  T  t  t T  t 

For small t , Z (t )t approximates the conditional probability

that the device will fail during the interval  t , t  t  given
that it has survived at the time t

In particular, if Z (t )  

 Hazard rate is constant

In that case we can say that the device is as good as

new at any time of its life.
 f (t )
Exercise: Show that Z (t )  .
R (t )

P  t  T  t  t T  t 
Z (t )  lim
t 0 t

P  t  T  t  t   T  t  
Z (t )  lim
t 0 t.P T  t 

P  t  T  t  t  
Z (t )  lim
t 0 t.P T  t 
 t t

 f (u )du
t
Z (t )  lim
t 0 t.R(t )
F (t  t )  F (t )
Z (t )  lim
t 0 t.R(t )

F (t ) f (t )
Z (t )   .
R(t ) R(t )
 Exercise :

Find the failure rate function for Weibull Distribution:


  t 
F (t )  1  e
f (t ) F (t )
 Z (t )   Z (t ) 
R(t ) 1  F (t )
  1
  t 
F (t )  e   t  

  t   1
e   t    1
 Z (t )      t 
 t 
e
Failure Patterns
 Theorem:

If T is the time to failure of a device with pdf f(t)

and cdf F, and Z(t) is the failure rate function of the
device, then :

t t


 Z ( s ) ds 
 Z ( s ) ds
R(t )  e 0
, f (t )  Z (t )e 0
, for t  0.
 f ( s) F ( s ) R( s )
Z (s)   
R( s) R( s) R( s)
Integrating, Note that R (0)  1
t t
R( s )
0 Z ( s ) ds   o R(s) ds
t
t

 Z ( s ) ds
 ln  R(t )    Z ( s)ds  R (t )  e 0

o
 f (t )
Z (t ) 
R(t )

f (t )  Z (t ) R(t )


 Z ( s ) ds
f (t )  Z (t )e 0
 Theorem 6.3:

Expected system life time: Theorem 6.3 shows how

mean/average lifetime of a system can be
determined from a knowledge of the reliability
function R(t)


E T    R( s )ds
0

Where T is the time to failure

 To prove that, it is sufficient to show that

E T    P T  t  dt
0

 
R.H .S     f (u )du  dt
0 t 

u 
    f (u )dt  du
0 0  
  uf (u )du  L.H .S
0
Example:

F (t )  t /10 0  t  10
1 t  10
Find E (T )

Clealy :
R (t )  1  F (t )
 1  t /10 0  t  10
0 t 0
 
E T    R( s )ds
0
10
 s 
E T    1   ds
0
10 

10
2
 s 
E T    s    10  5  5
 20  0
 Reliability of a system
• Series System:- A system whose
components(say n components) are arranged in
such a way that the system fails whenever any
of its components fail is called a series system.

n
R s  
i 1
Ri

w here R s  r e li a b i li t y o f s e r i e s s y s t e m
R i  R e lia b ility o f th e ith c o m p o n e n t
 Parallel System

• A system whose components are arranged in

such a way that the system fails only if all its
components fail is called a parallel system
n
R s (t )  1   (1  Ri (t ))
i 1

where R s  reliability of parallel system

R i  Reliability of ith component of the system

Example 1: Consider a series system of three independent
components and the life length of each component follows
uniformly distribution over (0,10).
(i) Find the reliability of the system.
(ii) Find the expected system life.
3 3
Solution:
Rs (t )   Ri (t )   1  Fi (t ) 
i 1 i 1

Fi (t )  t /10 0  t  10
1 t  10
3
Rs (t )  1  Fi (t ) 
 1  t /10  3 0  t  10
0 t  10
 
E T    R( s )ds
0

10 3
 s 
E T    1   ds
0
10 

5
E T  
2
Example 2:
Consider a system consisting of eight components .
The system consists of five assemblies in series, where
assembly 1 consists of component 1 with reliability
0.99, assembly II consists component 2 and 3 in
parallel with reliabilities 0.95 and 0.95, assembly III
consists component 4,5,6 in parallel with reliabilities
0.96, 0.92, 0.85, assembly IV and V consist of
component 7 and 8 with reliabilities 0.95, 0.82
respectively. calculate system reliability.
Exercise VI.7 :
Let Y be the life of an independent series system with n
component. Find the pdf and cdf of Y. Also prove that the hazard
rate function of the system is the sum of the hazard functions of its
components.

Solution:
n
Rs (t )   Ri (t )
i 1
n
FY (t )  1  Rs (t )  1   Ri (t )
i 1
 d n 
fY (t )   Rs (t )     Ri (t ) 
dt  i 1 

n
 n

or , fY (t )    Ri (t ) R j (t ) 

i 1 

j 1 
 j i 
 n
 n

  Ri (t ) R j (t ) 

i 1 

fY (t )
j 1 
Z s (t )    j i 
n
Rs (t )
 Ri (t )
i 1

R1(t ) R2 (t ) Rn (t )

Z s (t )      
R1 (t ) R2 (t ) Rn (t )
 f1 (t ) f 2 (t ) f n (t )
Z s (t )     
R1 (t ) R2 (t ) Rn (t )

Z s (t )  Z1 (t )  Z 2 (t )    Z n (t )

n
Z s (t )   Z i (t )
i 1
Exercise VI.8.
Consider an n-component independent series system.
Given that these components have constant failure rates
1 ,  2 ,  ,  n respectively. Find the
(i) reliability of the system.
(ii) hazard rate of the system.
(iii) pdf of the life-length of the system.
Also express the mean life of the system in terms of
mean life of its components.
Solution:
n n
Rs (t )   Ri (t )   1  Fi (t ) 
i 1 i 1
  n 
n
 i t

  i  t

Rs (t )   e e  i 1 

i 1

Z (t )  Z1 (t)  Z2 (t )  Zn (t)

 1  2    n
 n 
d    n  i t
 
fs (t)   (Rs (t))   i  e  i1 
dt  i1 
The mean life length of the system is

1
E (T )   s 
1   2     n

1

1 1 1
  
1 2 n
where i is the mean life-length of th
i component.
Exercise VI.9.
Let T be the life length of an independent parallel
system with n-components.
(i) Find the pdf and cdf of T
(ii) Show that t

t


 Z ( s ) ds n 
 Z i ( s ) ds
e 0
 1   1  e 0 

i 1 

Solution:  
Hint: Use properties of reliability, f(t) and their
definitions.
 Example:
Probability that a device can survive after
0,1,2,3,4 or more shocks are 1,0.8,0.4,0.2, 0
respectively. If the arrival of shocks follow
Poisson distribution with λ=0.15, find R(10).
Solution:

2 3
 t  ( t ) ( t ) 
R(t )  e 1  0.8(t )  0.4  0.2 
 2 6 
Put λ=0.15 and t=10 in the above.
 CERTAIN LIFE MODELS

Exponential: In reliability, this distribution plays a

vital role. This is the distribution during phase II.

“Useful life phases” of the device, when failure

occurs due to an external causes, which may be
called “Shock/shocks”
 Suppose that a device fails due to shocks, which occur in a Poisson process
with a rate  . Assume that the device fails as soon as the first shock arrives.
Then:

F (t )  1  e   t , t  0

or ,
f (t )   e   t , t  0
 0 else where

Thus the failure-time distribution is exponential distribution.

Let us assume, “p” is probability that even after the
shock, no failure occurs.

R(t )  P  No failure occurs in the time interval (0,t ]


R(t )  P n shocks arrive in time (0,t]  device survives all n shocks
n0

Using:

 A
P( A  B)  P( A)P B


R(t )  P Xt  n P The device will survives all n shocks 
n0
 Xt  n
 
R(t )  P Xt  n pn
n0

Where Xt is the number of shocks which arrives in the time-interval

(0,t]. Under the assumption that the shocks arrive in (0,t ] is a Poisson
process, then Xt has a Poisson distribution with parameter t

et (t)n n
R(t)   p , t 0
n0 n!
n

 pt   (1 p)t
R(t)  et
 n!
 R(t)  e ...........(*)
n0

Differentiate (*) with respect to t we get

  (1 p)t
 R(t)  e ...........(*)
 (1 p)t
R(t)   f (t)  (1 p)e
 (1 p)t
 f (t)  (1 p)e , t 0

Thus T has exponential distribution with parameter  (1  p )

Put p=0, we get

t
 f (t)  e
 Hazard rate of exponential distribution

t
f (t)  e
or ,
F (t )  1  e  t , t  0

f (t ) f (t )
Z (t )   
R(t ) 1  F (t )

This is the phase-II in failure pattern graph

 Gamma Distribution

Suppose that the shocks occur in a Poisson Process. And

the device fails as soon as the r-th shock arrives

R (t )  P (T  t )

 P  There will be no failure during the interval (0,t]

 P  There will be less than r -shocks in the interval (0,t]

Let :
X t be the random varible "Number of Shocks that arrive in (0,t]"

Then Clearly distribution of X follows poisson with

parameter  t

R (t )  P ( X t  r  1)
k
r 1
 e t  t 
k 0 k!

Differentiating with respect to t we get

 r 1
  t 
R(t )   e  t , t  0
( r  1)!

r 1
  t 
f (t )  e  t , t  0
(r )

Which is a Gamma distribution

Failure rate of Gamma distribution

r 1
  t  t
f (t )  e , t  0
 (r )
r 1   t
f (t ) t e
Z (t )   
R (t ) r 1   x
x
t
e dx
  2
 r 1   x

L et D    x e d x  .
 t 

r  2  t
 r 1   x r  t

t e (r  1  t )  x e dx  t e 
Z (t )   t 
D

r 1   x r  t
Now let  (t)=( r  1   t )  x e dx  t e
t

  (t)=  ( r  1)  x r  2 e   x dx, (using by parts)
t
C ase I : If 0  r  1.
Then  ( t )  0 for all t  0.
  ( t ) is an increasing function and
increases from a negative value to zero.
  ( t )  0 for all t  0.
 Z ( t )  for all t  0.
 Z ( t ) is a decresing function of t.
Thus Gamma distributio n for (0  r  1)
can be used as a life- model for any device
when the device in phase-I.
 C ase II : If r  1.
r 1 r 1
let t  ,  (t )  0  Z (t )  0 for t 
 
r 1
le t t  , th e n   (t)< 0 a s r> 1 .

  ( t ) is d e c re a sin g fu n c tio n a n d
d e c re a se s fro m a p o sitiv e v a lu e to z e ro .
  ( t )  0 fo r a ll t  0 .
 Z  ( t )  0 fo r a ll t  0 .
 Z ( t ) is a n in c re sin g fu n c tio n .
T h u s G a m m a d istrib u tio n fo r ( r  1)
c a n b e u se d a s a life - m o d e l fo r a n y d e v ic e
w h e n th e d e v ic e in p h a s e -III.
 C ase III : If r  1.
Then the Gamma distribution is changed
to exponetial distribution with parameter .
 Z (t )   (constant function).
Thus Gamma distribution for (r  1)
can be used as a life- model for any device
when the device in phase-II.
 Failure rate of Normal Distribution

Let T be the life length of any device which follows

2
normal distribution with mean  and variance  .
T 
 is follows standard normal distribution.

Assume that, the mean is so large.

 P(T  0) is neglegible, so that P (T  0)  1.

 2
 1 1  t  
  t   
  
 1  F (t )
2  
1  t  
 
2
 e 
2  
e

 2    
Z (t )  2  
 2 (1  F (t ))2

2
1  t  
1    t 
 1  F (t )
2  
Now let  (t)= e 
2   
C ase I : If 0  t   .
t
Then  0, and  ( t )  0 .

 Z (t )  0 for all 0  t   .
 Z (t ) is an incresing function.
 C ase II : If t   .
1
T h en   ( t )   (1  F ( t )).

   ( t )  0 fo r all t   .
  ( t ) is a d ecreasin g fu n ctio n an d
d ecreases fro m a p o sitive valu e to zero .
  ( t )  0 fo r all t   .
 Z  ( t )  0 fo r all t   .
 Z ( t ) is a in creasin g fu n ctio n o f t  0 .
T h u s N o rm al d istrib u tio n fo r (  larg e )
can b e u sed as a life- m o d el fo r an y d ev ice
w h en th e d evice in p h ase-III.
Statistical Methods of Determining Failure Time
Distribution

ˆ i  0.5
F (ti )  , using correction for continuity.
n
n(ti )  n  t  ti 
f d (t )  , ti  t  t i   t i
n ti

n ( ti )  n  t   t i 
Z d (t )  , ti  t  ti   ti
n(ti ) ti
Data Analysis to check suitability of
exponential distribution
Ex. V11:
100 components were put on life-test. The test
was terminated as soon as the 11th components
failed. The life-length of those components which
failed were as follows:
8, 16.1, 24.3, 32.55, 40.9, 57.8, 66.4, 75.1, 83.9,
92.85. Guess which distribution fits the data.
Estimate the parameters.
Failure No. 1 2 3 4 5 6
Time to 8 16.1 24.3 32.55 40.9 49.3
failure
Failure No. 7 8 9 10 11

Time to 57.8 66.4 75.1 83.9 92.85

Failure
 Case-I:
Time Time
fd (t) fd (t)
interval interval
0-8 0.00125 49.3-57.8 0.00117
8-16.1 0.00123 57.8-66.4 0.00116
16.1-24.3 0.00121 66.4-75.1 0.00114
24.3-32.55 0.00121 75.1-83.9 0.00113
32.55-40.9 0.00119 83.9-92.85 0.00111
40.9-49.3 0.00119
Case-II:

ti  ln(1 Fˆ (ti )) ti  ln(1 Fˆ (ti ))

8 0.00501 57.8 0.0672

16.1 0.0151 66.4 0.0780
24.3 0.0253 75.1 0.0888
32.55 0.0356 83.9 0.0998
40.9 0.0460 92.85 0.1109
49.3 0.0566
 Case-III:
Time Time
Zd (t) Zd (t)
interval interval
0-8 0.00125 49.3-57.8 0.00125
8-16.1 0.00125 57.8-66.4 0.00125
16.1-24.3 0.00124 66.4-75.1 0.00125
24.3-32.55 0.00126 75.1-83.9 0.00125
32.55-40.9 0.00124 83.9-92.85 0.00124
40.9-49.3 0.00125
Since Zd (t) is almost constant, we can assume
exponential distribution fits the data. Now using
least square method we can determine the
parameter  .

Calculation for .
11

 Z (ti )
ˆ  i1
 0 .0 0 1 2 5 .
11
 Data Analysis to check suitability of
weibull distribution with α=0.
Weibull distribution with parameter (, ,).
A random variable T is said to have Weibull
distribution if its density function is given by

   1  t  

  t    
 


 e ,t  
f (t )      

 0, t  
Special case of Weibull distribution when 0.

 1 t 
 t   
 
f (t )    e , t  0.
  
It can be proved that the hazard rate and reliability
of the above Weibull distribution are

 1 t 
 t   
 
Z (t )    , R (t )  e , t  0.
  
Consider the reliability R(t ) .

t 
 
 
R (t )  e , t  0.
 

t  t 
ln( R (t ))       ln( R (t ))   
   
  1 
 ln  ln      ln(t )   ln 
  R (t )  
   1 
 ln  ln      ln( t )   ln 
  1  F (t )  
  1 
 ln  ln      ln( t )   ln 
 1  Fˆ (t )   i
  i 

   1   
Now plot  ln ( ti ) , ln  ln    
  ˆ  
   1  F (ti )   
    1   
If the points  l n ( t i ) , l n  ln    
  ˆ  
   1  F (ti )   

lies almost on a straight line, we can assume

Weibull distribution with α=0 fits the data.

How to calculate the parameter  and .

Case-I (using graphical method)
ˆ  slope of the above line
and ˆ can be obtained by putting any value for t i

in the equation given below:

1   1 
ˆ  ln(ti )  ln  ln   
ˆ   1  Fˆ (ti )  
Case-ii (using least square method )

Least square method

Suppose ( xi , yi ), i  1, 2, , n given. We want to fit one
straight line bx  a to the above data. Least square

method estimates a and b by solving

n 2

min  yi  bxi  a  . (1)

( a ,b )
i 1
Using derivative concept, we can easily say that
equation (1) gives minimum when

n  xi y i   x i  y i
bˆ  2
n  xi    xi 
2

 yi  bˆ  xi
aˆ 
n
In our problem,

xi  ln(t i )
  1 
y i  ln  ln  
 
 1  ˆ
F ( t ) 
  i 

b
a    ln( )
Using least square method, we can estimate

ˆ
  bˆ
aˆ

ˆ
ˆ  e
Consider the reliability Z (t ) .
 1
 t 
Z (t )    , t  0.
  

ln(Z (t ))  ln    ( 1)ln(t )  ( 1)ln
 
 ln(Z (t ))  ( 1)ln(t )  ln    ln
ln( Z d (ti ))  (   1) ln(ti )  ln    ln 

Now plot  ln ( t ), ln  Z
i d (ti ) 
If the points  ln ( t i ), ln  Z d ( t i )  

lies almost on a straight line, we can assume

Weibull distribution with α=0 fits the data.

How to calculate the parameter  and .

Case-I (using graphical method)
ˆ  1  slope of the above line
ˆ  1+slope of the above line
and ˆ can be obtained by putting any value for t i

in the equation given below:

ln ˆ  ( ˆ  1) ln(ti )  ln  Z d (ti ) 
ln ˆ 
ˆ
Case-ii (using least square method )

Least square method

Suppose ( xi , yi ), i  1, 2, , n given. We want to fit one
straight line bx  a to the above data. Least square

method estimates a and b by solving

n 2

min  yi  bxi  a  . (1)

( a ,b )
i 1
Using derivative concept, we can easily say that
equation (1) gives minimum when

n  xi y i   x i  y i
bˆ  2
n  xi    xi 
2

 yi  bˆ  xi
aˆ 
n
Using least square method, we can estimate

ˆ ˆ
  b  1
l n ˆ  aˆ
ˆ
ˆ  e
In our problem,

xi  ln(ti )
yi  ln  Z d (ti ) 
b   1
a  ln    ln 
Ex. V12:
100 components were put on life-test. The test
was terminated as soon as the 7th components
failed. The life-length of those components which
failed were 2, 3.3, 4.7, 5.9, 7, 8, 9.
(a) Using hazard rate, check whether Weibull distribution fits
the data or not ? Estimate the parameters using
(i) graphical method.
(ii) least square method.
(b) Using cdf/reliability, check whether Weibull distribution fits
the data or not ? Estimate the parameters using
(i) graphical method.
(ii) least square method.
 Failure No. 1 2 3 4 5 6 7
Time to failure 2 3.3 4.7 5.9 7 8 9

Time interval Zd (t)

0–2 0.005
2 - 3.3 0.0077
3.3 – 4.7 0.0072
4.7 – 5.9 0.0085
5.9 – 7 0.0094
7-8 0.0105
8-9 0.0106
Solution (a):

ti ln(ti ) ln(Zd (ti ))

2 0.69 -5.29
3.3 1.19 -4.85
4.7 1.54 -4.87
5.9 1.77 -4.76
7 1.94 -4.66
8 2.07 -4.55
9 2.19 -4.54
since the points  ln(ti ),ln  Zd (ti )  lies almost on
a straight line, we can assume Weibull
distribution fits the data.

a(i). Calculation for parameters.

ˆ  1  slope  1  0.01  1.01

ln ˆ  ( ˆ  1 ) ln ( t i )  ln  Z d ( t i ) 
ˆ
ˆ  e
ln 1 .0 1  ( 0 .0 1 )  0 .6 9  5 .2 9
1 .0 1
 e  1 8 8 .7 9
a(ii). Calculation for parameters

x i  ln ( t i ) , y i  ln  Z d ( t i )  , i  1, 2 ,  , 7 .
7 7

i 1
x i  1 1 .3 9 ,  y i   3 3 .5 2
i 1
7 7
2
i 1
x i  2 0 .2 4 ,  x i y i   5 3 .7 4
i 1

 bˆ  0 .4 4 , aˆ   7 .1 3
ln (1 .4 4 )  7 .1 3
 ˆ  1 .4 4 , ˆ  e 1 .4 4
 1 8 2 .3 2 .