LECTURE NOTE : SYSTEM RELIABILITY

The functional behaviour of most systems can be characterized by a network diagram whose
nodes denote the subsystems of the system and the branches of the network represent the
functional relationship between these subsystems.
In general, a network with n nodes in which the operation or nonoperation of a node is
independent of the operation or nonoperation of the other nodes will be called structure of
order n and the nodes of the network will be referred to as the component of the structure.

Series System
A series system with n independent components is a system whose network representation is a
series structure of order n.

Figure 1 : Series Structure

Let XS be a random variable denoting lifetime of the system and Xi be the lifetime of its ith
component (I = 1,2,…,n) . Let FS(x) be the life distribution (failure time distribution function)
of the system and Fi(x) denote the life distribution (failure time distribution function) of the ith
component (i = 1,2,…,n). Further, let Fi (t) be the reliability of the ith component and FS (t) be
the reliability of the system.
The series system fails the moment a component of the system fails. Therefore, the lifetime of
the series system is the lifetime of its weakest component, i.e.,
Xs = min (X1, X2,…,Xn) = X(1) , which is the smallest order statistic.

Series System- Life Distribution Function

FS (t) = P  XS  x 

= P  X (1)  x 

= 1 - P  X (1)  x  = 1 - P  X1  x, X 2  x , ..., X n  x 

n
=1-  P X
i =1
i  x

n
=1-  1 - P  X
i =1
i  x 

n
=1-  1 - F (x) .
i =1
i

1
Series System- Reliability Function

FS (x) = 1 - FS (x)
n
=  1 - F (x)
i =1
i

n
=  F (x) .
i =1
i

Series System- Failure Density Function

dFS (x)
fS (x) =
dx
n −1

 (1 - Fj (x) ) + f 2 (x)  (1 - Fj (x) ) +...+ f n (x)  (1 - F (x) )

n n
= f1 (x) j
j=2 j=1 j=1
j2

 (1 - F (x) ) , where f (x) =

n n
dFi (x)
=  fi (x)
i =1 ji
j i
dx
, 1  i  n.

Series System- Mean Time to Failure


E  XS  =  FS (x)dx
0

 n
=   F (x) dx .
0 i =1
i

Theorem 1: For a series system , the failure rate of the syatem is equa to the sum of the failure
rates of its components.
Proof: Let hi(.) be the failure rate function of ith component of the series system , i=1,2,…,n.
n
Since FS (x) =  F (x)
i =1
i

x
n 
− h i (u ) du
= e
i =1
0

n x
−   hi (u ) du
=e i = 10

2
 x n
−  i= 1 hi (u ) du
=e 0

n
Let  hi (u ) =h(u) , then reliability function of the system is:
i =1


− h (u ) du
FS (x) = e 0
.

 h(.) is the failure rate of the series system.

PARALLEL SYSTEM
A parallel system with n independent components is a system whose network representation is
a parallel structure of order n.

Figure 2: Parallel Structure

The parallel system fails only when all its components fail, therefore the lifetime of the parallel
system is the life time of its strongest component., i.e.,
XP = max(X1, X2,…,Xn) = X(1) , which is the largest order statistic.

Parallel System- Life Distribution Function

FP (t) = P  X P  x 

= P  X (n )  x 

= P  X1  x, X 2  x , ..., X n  x 
n
=  P X
i =1
i  x

n
=  F (x)
i =1
i

3
 n
=  (1 − F (x) ) .
i =1
i

Parallel System- Reliability Function

n
FP (x) = 1 -  (1 − Fi (x) ) .
i =1

Parallel System- Failure Density Function

dFP (x)
f P (x) =
dx
n n n −1
= f1 (x)  Fj (x) + f 2 (x)
j=2
 Fj (x) +...+ f n (x)
j=1
 F (x)
j=1
j

j2

n n
dFi (x)
=  fi (x)
i =1
 F (x) , where f (x) =
ji
j i
dx
, 1  i  n.

Parallel System- Failure Rate Function

f P (x)
h(x) =
FP (x)
n n

 fi (x)
i =1
 F (x)
ji
j

= n
1 -  (1 − Fi (x) )
i =1

which is clearly not equal to the sum of the component failure rates unless n = 1.

Parallel System- Mean Time to Failure


E  X P  =  FP (x)dx
0



(1 − Fi (x) )dx .
n
= 0  
1 -
i =1 

4
COMPARISON OF A PARALLEL SYSTEM WITH n INDEPENDENT COMPONENTS
WITH A SERIES SYSTEM CONTAINING THE SAME n INDEPENDENT
COMPONENTS

(i) Series system never has a larger MTTF:

Since X(1)  X(n) , therefore, E[X(1)]  E[X(n)] ,i.e., E[XS]  E[XP].

(ii) If 0 < Fi (x)  1 , i = 1,2,…,n, then

n n
1 -  (1 − Fi (x) ) > Fj (x) >  Fi (x) :
i =1 i =1

Since XP > Xj > Xs  E[XP] > E[Xj] > E[Xs]

  

(1 − Fi (x) ) dx >  Fj (x) dx >
n n

0 1 - 
i =1  0
  F (x) dx
0 i =1
i

n n
Or 1 -  (1 − Fi (x) ) > Fj (x) >  Fi (x) .
i =1 i =1

In other words parallel systems have higher reliability than any of their components, and each
component in a series system has higher reliability than the system. The former fact is important
since it tells the system designer that one way to increase the reliability of the system is to
duplicate components by placing additional ones in parallel. The duplication of components,
in particular, placing additional components in parallel, is an example of design technique
called redundancy.

k-out-of-n (k/n) SYSTEM

Such a system functions iff at least k of the n components are functioning. Series system is n-
out-of-n system and parallel system is 1-out-of-n system.

5
 Figure 3: 2-out-of-3 Structure

Let it be assumed that the n components in a k-out-of-n system have the same failure time
distribution, G(.) .

k-out-of-n system : Reliability Function

Fk /n (x) = P  X k /n  x 
= P[ atleast k out of n components function beyond time x]
= P[ {exactly k out of n components function beyond time x} OR
{exactly k+1 out of n components function beyond time x} OR
{exactly k+2 out of n components function beyond time x} OR
… OR
{exactly n out of n components function beyond time x}]
= P[ {exactly k out of n components function beyond time x}]
+ P [{exactly k+1 out of n components function beyond time x}]
+P[exactly k+2 out of n components function beyond time x}]
+… + P[{exactly n out of n components function beyond time x}] (By axiom of finite
additivity)
n
n
  i  ( P X > x ) (1 − P  Xi > x )
i n −k
= i
i=k  

n
  i  ( G(x) ) (1 − G(x) )
n
i n −i
= .
i=k  

k-out-of-n system : Failure Time Distribution

6
Fk /n (x) = 1 − Fk /n (x)

n
  i  ( G(x) ) (1 − G(x) )
n
i n −i
=1-
i=k  
k −1
n
  i  ( G(x) ) (1 − G(x) )
i n −i
=
i =0  
k −1
 n 
  n − i  ( G(x) ) (1 − G(x) )
i n −i
=
i =0  
Let n-i = j

when i = 0 then j= n, and when j = k-1 then j = n-k+1.
Thus,
n
n
   ( G(x) ) (1 − G(x) ) .
j n− j
Fk /n (x) =
j= n − k +1  j 

k-out-of-n system : Failure Density Function

dFk /n (x)
f k /n (x) =
dx

d  n n n− j 
    ( G(x) ) (1 − G(x) ) 
j
=
dx  j=n −k +1  j  
n
n n
n
   j ( G(x) ) g(x) (1 − G(x) ) -    ( G(x) ) (n − j) (1 − G(x) )
j−1 n− j j n − j−1
=
j= n − k +1  j  j= n − k +1  j 

n n −1
n! n!
 ( G(x) ) g(x) (1 − G(x) ) -  ( G(x) ) (1 − G(x) )
j−1 n− j j n − j−1
=
j= n − k +1 ( j − 1)!(n − j)! j= n − k +1 j!(n − j − 1)!

n!
( G(x) ) g(x) (1 − G(x) ) .
n −k k −1
=
(n − k)!1!(k − 1)!

k-out-of-n system : Mean Time To Failure

We derive recurrence relationship for finding out MTTF(k/n)
Consider MTTF(k-1/n) – MTTF(k/n)

7
 
MTTF(k-1/n) – MTTF(k/n) =   F
0
k −1/n (x) − Fk /n (x)  dx

 n n

n 
=      ( G(x) ) (1 − G(x) ) −   i  ( G(x) ) (1 − G(x) )
n
i n −i i n −i
 dx
0  i = k −1     
i i =k


 n 
 ( G(x) ) (1 − G(x) )
k −1 n − k +1
=  dx .
0
k − 1

Let G(x) = e−x , x  0,   0 . In other words suppose that each component of k/n system
has constant hazard rate  .Then,


 n  −x k −1
 ( e ) (1 − e )
−x n − k +1
MTTF(k-1/n) – MTTF(k/n) =   dx
0
k − 1


 n  −x k −1 n −k +1  n − k + 1
=   (e )    ( −e ) dx
−x j

0
k − 1 j= 0  j 

 n  n −k +1  n − k + 1
  ( −e )
−x k + j−1
=  dx
 k − 1 j=0  j 0

1  n  n −k +1 j  n − k + 1 1
=    (−1)   .
  k − 1 j=0  j  (k + j − 1)

Using the identity

N
 N 1 N!(a − 1)!
 (−1)  j
 = , a  1 , we have
j= 0  j  (a + j) (N + a)!

1
MTTF(k-1/n) – MTTF(k/n) = .
 (k − 1)

Using the above recurrence relationship one can find MTTF(1/n) system when each
component has constant hazard rate  as follows:

1
MTTF(1/n) = + MTTF(2/n)


1 1
= + + MTTF(3/n)
 2

8
 1 1 1
= + + + MTTF(4/n)
 2 3

1 1 1 1
= + + + + MTTF(5/n)
 2 3 4

1 1 1 1 1
= + + + + ... + + MTTF(n/n).
 2 3 4 (n − 1)

Now using Theorem 1.1 , hazard rate of a series system (n/n system) is sum total of hazard rate
of its components. Thus when each component has constant hazard rate  , the hazard rate of
the series system is n 

1
 MTTF(n/n) = .
n

1 1 1 1 1 1 1 n 1
Thus, MTTF(1/n) = + + + + ... + + =  .
 2 3 4 (n − 1) n  i =1 i

We can also find MTTF(k/n) system when each component has constant hazard rate 
as follows:

1
Since MTTF(k-1/n) – MTTF(k/n) =
 (k − 1)

1
 MTTF(k/n) = MTTF(k-1/n) -
 (k − 1)

1 1
= MTTF(k-2/n) - − −
(k − 2) (k − 1)

1 1 1
= MTTF(k-3/n) − − −
(k − 3) (k − 2) (k − 1)

1 1 1 1
= MTTF(k-4/n) − − − −
(k − 4) (k − 3) (k − 2) (k − 1)

9
 1 1 1 1 1
= MTTF(1/n) − − ... − − − −
 (k − 4) (k − 3) (k − 2) (k − 1)

1 k −1 1
= MTTF(1/n) − 
 i =1 i

1 n 1 1 k −1 1
=  − 
 i =1 i  i =1 i

1 n 1
=  .
 i=k i

SERIES-PARALLEL SYSTEM

A series-parallel system of order (m,n) is a system consisting of m identical parallel systems

each of order n arranged in series. Figure 4 depicts such a system and reduces to a series-
parallel system of order (m,n) when n=m and mi = m ,i=1,2,…,n

Figure 4: A Series-Parallel structure

For a series-parallel system of order (m.n) the reliability metrics are as under:

If the components are independent and Fi (x) equals the reliability of the ith type component for
i=1,2,…,n, then the system reliability function is:

m
 n

FS−P (x) = 1 −  (1 − Fi (x) )  .
 i =1 

We shall assume that all components have the same reliability function: Fi (x) = G(x) .Then,

10
Series-Parallel System- Reliability Function is:

(
FS− P (x) = 1 − (1 − G(x) ) ).
n m

Series-Parallel System- Life Time Distribution is:

(
FS− P (x) = 1- 1 − (1 − G(x) )
n m
).
Series-Parallel System- Failure Density is:

(
fS−P (x) = mn 1 − (1 − G(x) ) )
n m −1
(1 − G(x) )
n −1
g(x) .

Series-Parallel System- Failure Rate Function is:

mn (1 − G(x) )
n −1
g(x)
h S−P (x) = .
1 − (1 − G(x)) n

Series-Parallel System- MTTF is:

( )
 
1 − (1 − G(x) )
n m
MTTF(S-P) =  FS−P (x) dx =
0 0
dx .

CALCULATION OF MTTF (S-P) WHEN EACH COMPONENT HAS CONSTANT HAZARD RATE 

( )

1 − (1 − e −x )
n m
MTTF(S-P) = dx
0

dz dz
Let 1 − e−x = z , then e−x dx = dz  dx = −x
or dx =
e (1 − z)

 (1 − z )
n m dz
 MTTF(S-P) =
0
(1 − z)

1 (1 − z )
1 n

(1 − z n ) dz
m −1
= 
 0 (1 − z)

Now

11
 (1 − z ) = 1 + z + z
n
2
n −1
+ ... + z n-1 =  z i
(1 − z) i =0

1 n −1 i
1
z (1 − z n ) dz .
m −1
 MTTF(S-P) =  
 i =0 0

dy dy
Let z n = y  nz n-1dz = dy  dz = n-1
or dz = (n-1)/n
nz ny

1 n −1 i/n
1
dy
 y (1 − y )
m −1
 MTTF(S-P) = 
 i =0 0 ny(n −1)/n

1 n −1 i +n1 - 1
1

 (1 − y ) dy
m −1
= 
n i = 0 0
y

1 n −1  i + 1 
=  
n i = 0  n
, m


 i +1 

n −1  m
1  n 
= 
n i = 0  i + 1 
 + m
 n 

 i + 1  
(m − 1)!  n  − 1 !
n −1
  
= 
n i = 0  i + 1  
,
 n  + m − 1 !
  

where [x] is the greatest integer not exceeding x.

PARALLEL-SERIES SYSTEM

A parallel- series system of order (m,n) is a system consisting of m identical series system each
of order n arranged in parallel. Figure 5 depicts such a system and reduces to a parallel-series
system of order (m,n) when ni = n ,i=1,2,…,n

12
 Figure 5: A parallel-series structure

For a parallel-series system of order (m,n) the reliability metrics are as under

If the components are independent and Fi (x) equals the reliability of the ith type component for

i=1,2,…,n, then the system reliability function is:

m
 n

FP −S (x) = 1- 1 −  Fi (x)  .
 i =1 

n
Since the reliability function of a series system of order n is  F (x) and we have m such
i =1
i

systems connected in parallel we have the above result.

We shall assume that all components have the same reliability function: Fi (x) = G(x) .Then,

Parallel – Series System- Reliability Function is:

(
FP −S (x) = 1- 1 − ( G(x) ) )
n m
.

Parallel – Series System- Life Time Distribution is:

(
FP −S (x) = 1 − ( G(x) ) )
n m
.

Parallel – Series System- Failure Density is:

13
 (
f P −S (x) = mn g(x) 1 − ( G(x) ) )
n m −1
( G(x) )
n −1
.

Parallel – Series System- Failure Rate Function is:

h P −S (x) =
(
mn g(x) 1 − ( G(x) ) )
n m −1
( G(x) )
n −1

.
(
1- 1 − ( G(x) ) )
n m

Parallel – Series System- MTTF is:

( )
 
 
0 FP−S (x) dx = 0 1- 1 − ( G(x) )
n m
MTTF(S-P) =  dx .


CALCULATION OF MTTF (P-S) WHEN EACH COMPONENT HAS CONSTANT HAZARD RATE 

 (1 − (1 − e ) ) dx

−nx m
MTTF(P-S) =
0

dz dz
Let 1 − e−nx = z , then ne−nx dx = dz  dx = −nx
or dx =
ne n(1 − z)

 (1 − z ) n(1 − z)
dz
 MTTF(P-S) = m

1 (1 − z )
1 m

n 0 (1 − z)
= dz

Now

(1 − z ) = 1 + z + z
n
2
+ ... + z n-1 =  z i
n −1

(1 − z) i =0

1 n −1 i
1

 MTTF(P-S) =  z dz
n 0 i =0

1 n −1 i
1
 MTTF(P-S) =  z dz
n i =0 0

1 n −1 1
= 
n i =0 (i + 1)

14
Let i+1= j ,

1 m 1
then MTTF(P-S) =  .
n j=1 j

15