Professional Documents
Culture Documents
Reliability L D Arya PDF
Reliability L D Arya PDF
RELIABILITY
ENGINEERING
HAPPY BIRTHDAY!!!!!!!!!!!!!
Place: Indore.
From:
• Chinmay Sharma
• Mughashe A. Assumi
• Shoab Ahmed Khan
• Shobhit Sharma
CONTENTS
1. Reliability Engineering Introduction.
Duration.
Chapter 1
1.1 General
Now a days it is the responsibility of an engineer to ascertain adequate operation of a
system/component/device for a required duration with provided environment. It is expected that
failure should be as small as possible. Qualitatively the word reliability has been used since ages,
but for professional engineers some quantitative measure is required and than its physical
interpretation is also required to draw some meaningful conclusions. Safety also is related to
reliability. Safety of working person is important while system fails or performs inadequate
operation. The term quality control which is associated usually with a manufacturing process is
in a same short term reliability evaluation. Sometimes in system operation and planning ‘quality’
is included in the word adequate performance of the system.
The classical definition of the reliability is stated as follows.
“Reliability is the probability of a device performing its function adequately for an intended
period under the operating conditions encountered.”
The definition has four basic parts i.e.
1. Probability - This provides a numerical measure to the reliability whose sec is between [0, 1].
Probability is a mathematical concept. It magnitude alone is not any importance until some
system description is associated with it.
2. Adequate performance – The numerical measure has to be evaluated with respect to adequate
performance of the system. A specific professional engineer may decide adequacy criteria
about a system. This is an engineering judgment and may require significant computational
effort for a complex system e.g. aerospace, power network etc. A power supply point one
will say adequate if voltage and frequency are in desired range.
3. Time period – Reliability of a component i.e. probability of adequately performing in general
a function of time. Obviously reliability decreases as time period of study increases. It
decreases from value 1 to value 0 as t approaches infinity. R(t) i.e. reliability function is a
decreasing function of time.
4. Operating condition/ Environment - Reliability of a component will depend on the
environment in which it is functioning. A personal computer (PC) put in a laboratory of the
institute under air-conditioner will have higher reliability than a PC in a hostel room without
air-conditioner.
Initial application of reliability studies in military applications. Then reliability study was
introduced to electronics nuclear science and now a days it is one of the most important
component for the all engineering disciplines. The beginning of the development of reliability
theory can be traced back to the days of 2 world war.
The first reliability evaluation was reported to occur when an explanation was sought for the
poor performance of German V1 and V2 missiles. These missiles were constructed of large
number of components. Each component considered highly reliable. It was revealed in the report
that though each component was highly reliable but the missile system as whole was quite
unreliable. This is due the simple fact that the reliability of system was product of reliabilities of
individual components (assuming independent failure). Today this solution is well known, but
then it was a revelation at that time. Reliability studies get significant attention after 1965.
Earlier importance was given to generating system reliability evaluation.
For different system there may be many more indices defined. Once the reliability indices are
obtained the planner may us these values for modification of the system by selecting a threshold
value of these indices.
Fig. 1.1
(2)Low level Redundancy (LLR): In a low level redundancy, redundancy is added at component
level as shown in fig. below
Fig. 1.2
Chapter – 2
2.1 Introduction
This chapter deals with some important formulas of basic probability theory which are used in
reliability engineering applications.
( ) = 1
0 ≤ ( ) ≤1
3. Frequency interpretation
An experiment is performed N times independently and let us assume number of times favorable
outcome occurs is ni, which is for even then if N is ‘large’ then ( ) is approximately
written as
!
( ) ≃
"
(a) If events #, %, ... ' are mutually exclusive (non-compatible), then following relation is
written
(b) If events #, %, ... ' are not mutually exclusive (compatible), then following relation is
observed
(∑ ) = ( ) = 1
Where ( %⁄ # ) is the probability of occurrence of event % under the assumption that # has
occurred.
Fig. 2.1
Event ‘A’ is defined on the sample space as shown in Fig-2.1. Now the event ‘A’ can be written
as follows
= ) .0
( ) = ) ( .0 )
( ) = ) ( ⁄0 ) (0 )
Ex. 2.1: A factory shop transformer feeds four motors with input powers as 10, 20, 20, 50 kVA
respectively. The operation of these motors is associated with the production process in such a
manner that when the first motor is running, the probability of switching on of the second motor
is 0.7, when the first and second motors are running, the probability of switching on of the third
motor is 0.6; when the first three motors are running; the probability of switching on of the
fourth motor is 0.3. Assume the probability of switching on of the first motor is 0.8. Determine
the probability of a load of 100 kVA on the factory shop of transformer.
Solution:
One is required to calculate the probability of all four motors running simultaneous i.e.
( #. %. 2. 3) = ( 3⁄ #. %. 2) ( 2⁄ #. %) ( %⁄ #) ( #)
Given
( #) = 0.8
( %⁄ # ) = 0.7
( 2⁄ # . %) = 0.6
( 3⁄ #. %. 2) = 0.3
Probability of load being 100 kVA = All the four motors are on = 0.3 × 0.6 × 0.7 × 0.8
= 0.1008
Fig. 2.2
Solution:
A. probability of success requires all four components must in successful operation i.e.
p? = (@. #. %. 2)
= (@ ) ( # ) ( % ) ( 2 )
= 0.848
B. Alternative approach
q? = (@ + #
+ % + 2)
p? = 1 − q?
Ex. 2.3: Consider transmission system with three line and three node, generator connected at
node one only. Capacity of each line is 100 MW. Loads at node 2 and 3 are (a) 75 and 125 MW
respectively. (b) 50 MW each. Obtain reliability of transmission system assuming reliability of
each line 0.96.
Solution:
Case 1:
Fig. 2.3
It is obvious from the above diagram that to meet this load at nodes 2 and 3 all the three lines
must be in ‘up’ state then only load at two nodes can be met. Hence probability of success is
written as
= 0.962
= 0.8847
Case 2:
Fig. 2.4
= FE# E% E2 , E# E% E2 , E# E% E2 , E# E% E2 , E# E% E2 , E# E% E2 , E# E% E2 , E# E% E2 G
# = E# E% E2
A% = E# E% E2
A2 = E# E% E2
A3 = E# E% E2
( )= ( #) + ( %) + ( 2) + ( 3)
P(A# ) = 0.962
( %) = 0.96% × 0.04
( 2) = 0.96% × 0.04
( 3) = 0.96% × 0.04
Note: At least two transmission lines are required in up state for system success.
Output of an experiment is events which may not be numbers. In application studies one requires
the output of experiments in a special order and in the form of numbers. Hence according to a
‘rule’ or ‘function’ each event of sample space is mapped into a number space. Mapping is not
one to one. More than one events may be mapped into one number. The rule or function J(K )
which maps sample space into number space (also known as range space) is defined as random
variable, which is illustrated with the help of following Fig. 2.5.
Fig. 2.5
Where K# , K% , K2 , … K' are the elements of sample space, and L# , L% , L2 , … L1 are elements in
number space. According to the definition
J(K ) = L ; ≤ !
If there are finite or infinite countable events in sample space then random variable is discrete
random variable. If elements in sample space are uncountable by infinite then random variable is
continuous.
= NOO , O, O , P
It is obvious sample space has four elements and they are not numbers. ‘H’ denotes head and ‘T’
denotes tail. One may define random variable as a ‘rule’ as follows ‘X’
Q = N0, 1, 2P
Obviously elements in number space are three. The two elements of sample space i.e. HT, TH
are assigned number ‘1’ according to rule ‘X’. This is an example of discrete random variable.
X: Number of trials / tossing required to get first head. In view of this the sample space can be
directly written as
= N1, 2, 3, 4 … P
= Q
Since number of trials theoretically may range from one to infinity. This is again an example of
discrete random variable.
Ex.: An equipment is put to service and one measures its life length. The ‘life length of the
component’ is a rule and is a random variable ‘X’. This is one of the most important random
variable in reliability studies. This is an example of continuous random variable.
A Probability is associated with each element of sample space. Hence an element of number
space will also be assigned a probability value. Hence one can write
RJ = L S = ( L )
Then following relations follows from the basic axioms of probability theory
0 ≤ (L )≤1
) (L )=1
TUV WXX
The discrete sequence satisfying above conditions is called probability distribution function
(pdf).
'
RJ ≤ L' S = Y( L' ) = ) (L )
Z-[
Y (−∞) = 0, Y (∞) = 1
] = ^ RJS = ) L ( L )
Z-[
_ = ^ R(L − ]
% )% S = ) (L − ])% ( L )
Z-[
Large value of variance signifies large variation around mean value and signifies large
uncertainty.From relation (2.5) it follows
^ RJ % S = _ % + ]%
Since ^ R]% S = ]%
Ex. 2.4: A coin is tossed twice. Obtain probability distribution function of random variable as
number of tails. It is assumed an unbiased coins and P(H) = P(T) = 0.5 i.e. probability of
occurrence of ‘H’ and ‘T’ equals 0.5.
Solution:
(OO) = 1⁄4
( O) = (O ) = 1⁄4
( ) = 1⁄4
According
] = )L (L )
= 1⁄2 + 1⁄2 = 1
= 1a + 0 + 1a = 0.5
4 4
Ex.: A coin is tossed till first tail occurs. Number of trials is as random variables. It is assumed
that point value of random variable is r. This means in (b − 1) trials ‘H’ occurs and in b cd trial
‘T’ occurs. Trials are assumed to be independent. In general further it is assumed that probability
of occurrence of ‘H’ is < and that of ‘T’ is e .Then a sequence for r trials is written as
(b) = <.
fg <. < …g<.
gghg gi e .
V-#
) (b) = ) e. <V-#
V
1
= e. =1
1−<
Binomial Distribution
Assume one perform ‘!’ independent trials and in each trial probability of occurrence of an event
(success) and non-occurrence (failure) are ‘<’ and ‘e’ respectively. Then probability of
occurrence of ‘b’ success in ‘!’ trials is written as follows
<. <gghg
fg … …g…
gi< . fg
e. egghg
… …g…
gie = <V e'-V
V '-V
But these ′b′ success can occur 'kV ways hence total probability expression is written as follows
'
kV <V e'-V
Ex. 2.5: A biased coin is tossed thrice. Probability of getting head is 3a4 and that of tail is 1a4.
Obtain probability of getting two heads in three trials.
Solution:
%
= 2k% m3a4n m1a4n
9 1
= 3× ×
16 4
27
=
64
Ex. 2.6: A power plant has ten generators each having probability of success as 0.94. Assume
independent failures. Determine probability of operating at least four generators.
Solution:
! = 10
< = 0.94
e = 0.06
#p
Poisson Distribution
This is another important distribution which is applicable when (i) Total trials are not known. (ii)
Probability of occurrence of the event is very small.
]
-q V
(b) = b = 0, 1, 2, … ∞
b!
] is parameter of distribution.
Assume
! → ∞ (large)
!< = ]
!(! − 1)(! − 2) … (! − b − 1) V
(b) = < (1 − <)'-V
b!
(!<)V (1 − <)'
=
b!
]V (1 − <)'
=
b!
]
-q V
=
b!
Note:
b = (1 − <)'
≅ −!< = −]
Ex. 2.7: Probability of failing a landing gear of an aircraft while landing is 10-y . Determine the
probability of surviving the landing gear the designed landing 10000.
Solution:
! = 103
< = 10-y
b = 0 (none failure)
]
-q V
(b) =
b!
(0) = -p.#
= 0.905
≅ 0.905
Solution:
]
-q p
RJ = 0S = 0.1 = = -q
0!
−] = ! 0.1
] = − ! 0.1
[
]
-q V
RJ ≥ 3S = )
b!
VZ2
-q p
] -q #
] ]
-q %
=1− − −
0! 1! 2!
RL ≤ J ≤ L + ∆LS
(L) = lim
∆|→p ∆L
According to basic axims of probability theory any function will qualify to be pdf if it satisfies
following conditions
} (L )~L = 1
-[
This means function always has non-negative values and total area under the curve is unity.
Further one writes
R ≤ J ≤ ~ S = } (L)~L
This implies probability that random variable lies between R , ~S is area under the curve between
R , ~S as shown below in fig. 2.6
Fig 2.6
It is stressed here that point probabilities are zero in case of continuous random variable
RJ = €S = } (L)~L = 0
•
Y (L) = RJ ≤ LS = } (L )~L
-[
Fig 2.7
~Y(L)
(L ) =
~L
] = ^ RJS = } L. (L )~L
-[
Median Value:
Mode (Š‹Œ•Ž ):
•L
L1U•ƒ = (L)
L1U•ƒ
Variance (•‘ ):
It shows dispersion around mean value. Larger variance shows larger uncertainty in the random
variable. It is defined as follows
_ = } (L − ])% . (L )~L
%
-[
^ RL % S = _ % + ]%
As earlier derived in discrete random variable case, above relation holds irrespective of
distribution function.
Note: Assume X is any random variable and with pdf (L) and ”(L) is any function of X then
expected value of ”(L) is
(L ) = •. L 0<L<1
=0 otherwise
Solution:
•>0
#
•
} •L ~L = 1 = =1
2
p
•=2
(L) = 2L 0<L<1
Mean value:
# %
Mean value =–p 2L % ~L = = 0.67
2
#
4 4
= } —L % + − L˜ 2L ~L
3 3
p
#
8 8 17
} —2L 2 + L − L % ˜ ~L =
3 3 18
p
L1ĥ W' :
|„…†‡ˆ‰
} 2L~L = 0.5
p
1
L1ĥ W' = = 0.707
√2
•L
L1U•ƒ = (2L) = 1
1
Fig 2.8
Let us assume X is a continuous random variable. One has large observed value of this random
variable i.e. L# , L% , … Lš . Probability distribution function can be approximated as follows from
frequency interpretation
ℎ(L, L + ∆L)
(L ) ≅ ›(L ) =
". ∆L
Where ›(L) is approximate value of pdf at ‘L’. ℎ(L, L + ∆L) denotes number of observation
between RL, L + ∆LS. By varying ‘L’ and selecting ∆L, a curve ›(L) may be plotted. Curve
fitting technique may be used for distribution identification. Alternatively after noting the shape
of the plot a standard pdf may be selected e.g. exponential, Normal etc. and then parameter may
be estimated using estimation theory. A typical plot is shown in Fig 2.9 below
Fig 2.9
1
(L) = •<L<D
D−•
Fig 2.10
|
1
Y (L ) = RJ ≤ L S = } ~L
D−•
W
L−•
Y(L) =
D−•
•
L •+D
]=} ~L =
D−• 2
W
•
•+D % 1
’•bRJ S = _ = } žL −
%
Ÿ ~L
2 D−•
W
(D − •)%
=
12
(L) = -¡|
, L≥0
is a parameter of distribution.
Y (L ) = } -¡|
~L = 1 − -¡|
Fig 2.11
[
1
] = } L. -¡|
~L =
p
[
1 1
_ % = } (L − )% . -¡|
~L = %
p
|„…†‡ˆ‰
} -¡|
~L = 0.5
p
1− -¡|„…†‡ˆ‰
= 0.5
-¡|„…†‡ˆ‰
= 0.5
ln 0.5
L1ƒ• W' = −
•L
L1U•ƒ = R -¡|
S=0
0<L<∞
1 |-q ¦
(L) = -p.y£
¤
¥
, −∞≤L ≤∞
√2¢_
It is bell shaped curve with two parameters. The parameters themselves are mean (]) and
variance(_ % ). A typical sketch of this function is shown in Fig 2.12
Fig. 2.12
Where
1 |-q ¦
"(], _ % ) = -p.y£
¤
¥
√2¢_
Analytical evaluation of above integral does not exist. Hence cumulative probabilities are
evaluated numerically. Define a new random variable
L−]
§=
_
Y (L ) = ¨(§) = } "(0,1)~§
-[
Hence cumulative probabilities RJ ≤ LS is area under the standard "(0,1) curve from −∞ to §
as shown in Fig 2.13
Fig 2.13
¨(∞) = 1
¨(−∞) = 0
¨(0) = 0.5
Tabulated values of ¨(§) for various values of § in the range -4 to 4 are available. Since area
under the curve "(0,1) outside above range is negligible. Further following key points may be
remembered
R] − _ ≤ J ≤ ] + _S = 0.683
R] − 2_ ≤ J ≤ ] + 2_S = 0.945
R] − 3_ ≤ J ≤ ] + 3_S = 0.997
R] − 4_ ≤ J ≤ ] + 4_S = 0.999
R• ≤ J ≤ DS = } "(], _ % )~L
W
©¦
= } "(0,1)~§
©ª
= ¨(§% ) − ¨(§# )
Ex. 2.10: Load at substation is normal distribution with mean 1000 MW and standard deviation
100 MW. Obtain following
(i) RE ≥ 1100S
(iii) RE ≤ 800S
Solution:
1 X-q ¦
()= -p.y£
¤
¥
√2¢_X
1 -p.y£
X-#ppp ¦
¥
= #pp
√2¢. 100
= "(1000, 100% )
(i)
= } "(0,1) ~§
#
= ¨(∞) − ¨(1)
= 1 − 0.84 = 0.16
(ii)
##pp
= } "(0,1) ~§
-#
(iii)
¬pp
-%
= } "(0,1) ~§
-[
(L) −∞≤L ≤∞
If it is required to obtain truncated distribution between rangeR•, DS. Then truncated distribution
∗(
L) •≤L≤D
∗(
Area under the curve L ) in the range R•, DS must be one. Hence constant € is defined as
€ = } (L)~L
W
∗(
(L)
L) = •≤L≤D
€
For example a Normal distribution truncated left to ′•′ and right to ‘D’ is shown in Fig 2.14
Fig 2.14
Similarly a truncated exponential distribution truncated to left of ′•′ is shown in Fig 2.15
Fig 2.15
Ex. 2.11: A random variable ‘J’ is normally distributed with mean value 1000 and standard
deviation 100. Obtain truncated distribution left to zero. Assuming that random variable has only
non-negative values.
Solution:
q
Note that in this example ratio of = 10.
¤
= } "(0,1) ~§
-#p
=1
Hence
∗(
L) = (L)
q
No truncation is required. In fact it is obvious from the ratio of . If this ratio is more than 3
¤
then truncation is not required and area under the curve left to zero will be practically zero in
such case.
Ex. 2.12: A random variable ′J′ is normally distributed with mean value 1000 and standard
deviation 500. Obtain truncated distribution left to zero.
Solution:
(L ) = "(1000, 500% )
= } "(0,1) ~§
-%
≅ 0.9772
Hence
∗(
"(1000, 500% )
L) =
0.9772
Exercise #1
Ex 1: A power plant contains 5 units of each of 100 MW and probability of successful working
of each unit is 0.94. Load on the plant is constant as 250 MW. Obtain probability of load
exceeding the capacity and calculate expected value of load not supplied.
Ex 2: A power plant contains 3 units of capacities 50, 100, 150 MW with probabilities of
successful operation as 0.90, 0.92, and 0.96 respectively. Load on the plant is Normally
distributed with mean 100 MW and standard deviation 20 MW. Obtain probability of load
exceeding the capacity and expected loss of load.
Ex 5: Assume X has Poisson distribution with parameter′]′. Find the value of ‘r’ for which
RJ = bS is maximum.
Ex 6: The number of particles emitted from a radioactive source during a period is a random
variable with Poisson distribution. If the probability of no emissions equals 0.80. What is the
probability that at least three emissions will occur.
Ex 7: X is a random variable with exponential distribution with parameter 0.1. Obtain from
physical consideration the value of a random variable lies between 0, 1. Obtain modified
exponential distribution.
Ex 8: Assume X has distribution "R], _ % S. Obtain the value of constant ‘k’ such that
RJ ≥ €S = 0.33 RJ ≤ €S
K 2 + 3K % + 4K + 10 + € = 0
Ex 11: The diameter of a conductor ‘X’ is a random variable with probability density function as
Ex 12: An anti aircraft gun system has probability of hitting an aircraft is 0.0095 for each shot.
Determine the number of shots required to be fired so that probability of hitting with two or more
shots is at least 0.96.
Solution: (Hint)
Let us assume number of shots required is ! and since < = 0.0095 given. Parameters of
poisson distribution is ] = !<
RJ ≥ 2S = 1 − RJ = 0S − RJ = 1S = 0.96
Solve for n?
-q V
]
(b) =
b!
1− -q
− -q
] = 0.96
-q (
1 + ]) = 0.04
Exercise # 2
(i) ± RJ + S = ±(J)
(ii) ±R JS = %
±RJS
Where c is a constant.
® = •J + D
1
(L ) = •≤L≤D
D−•
Obtain pdf of
L−•
=
D−•
Obtain pdf of
® = 4 − J%
(i) ® = 2J + 3
(ii) = Q
(i) ® = sin J
(ii) ® = cos J
Ex 7: A random variable ‘X’ is uniformly distributed between R−4,4S. Obtain pdf of
® = |J|
Ex 8: A random variable ‘X’ is uniformly distributed between R−2¢, 2¢S. Obtain pdf of
(i) ® = sin J
(ii) ® = cos J
Ex 9: Assume J# , J% , … J' are random variables with pdf as # (L), % (L), … ' (L). Obtain pdf of
Chapter 3
3.1 Introduction
Fig. 3.1
® = O(J)
Then for any probability associated with range space ¶ is equal to an equivalent space in range
space |.
For example if
® = •J % •>0
Then
RJ > 1S = R® > •S
Hence event
(L ) = -|
L≥0
® = 3J + 2
It is obvious domain of ® is ≥2
Fig. 3.2
Suppose one define the event N® ≥ 8P which yields from the function NJ ≥ 2P
Hence R® ≥ 8S = RJ ≥ 2S
=} -|
~L = -%
Assume J is a random variable having probability function (L). Define ® as another random
variable
® = O(J)
Fig. 3.3
R® ≤ S = RO(J) ≤ S
= RJ ≤ O -# ( )S
Since R® ≤ S = @ ( ) =cdf of ®
@ ( ) = Y(L)
Hence
~@ ( ) ~Y (L )
g(y) = =
~ ~L
~L
g(y) = f(x)
~
Where ”( ) is pdf of ®.
L = O -# ( )
Again the random variable Y is defined as decreasing function of ′J′ as shown in Fig. 3.4
® = O(J)
Fig. 3.4
@ ( ) = R® < S = RJ > LS
= 1 − RJ ≤ LS
= 1 − Y(L)
~@ ( ) ~L
g(y) = = − (L )
~ ~
•|
Since ® is a decreasing function will be negative and hence above relation is written as
•¶
Assume a random variable J has probability density function (L) and ® = O(J) has multiple
solutions when it is inverted for J i.e.
L = O -# ( )
Fig. 3.5
@ ( ) = R® ≤ S = RJ ≤ L# S + RL% ≤ J ≤ L2 S + RJ ≥ L3 S + …
~@ ( )
”( ) =
~
•|¦ •|½
Since •¶
and •¶
are negative above relation is written as
~L
”( ) = ) (L ) ¾ ¾
~
® = Y(L)
J = Y -# (®)
Fig. 3.6
@ ( ) = R® ≤ S
= RJ ≤ Y -# ( )S
= YmY -# ( )n =
”( ) = 1 , 0< <1
This result is useful in generating random variates from distribution (L) if random variates are
known from uniform distribution R0 , 1S.
(L ) = 3L % , 0<L<1
Obtain distribution of ®
® = 3J + 1
Solution:
®−1
J=
3
~L 1
=
~ 3
3 1 1
”( ) = ( − 1)% = ( − 1)% 1≤ ≤4
9 3 9
Solution:
<= %
#a
=< %
1
()= 8 < < 10
2
~
”(<) = ( ) ¾ ¾
~<
1 1
= . <-#/%
2 2
1
= 64 < < < 100
4“<
Ex. 3.4 : A random variable J is uniformly distributed R−1 , 1S . Obtain probability density
function of ® = J % .
Solution:
Fig. 3.7
1
(L ) = , −1 <L <1
2
L# = −“ , L%Z “
~L# ~L%
”( ) = (L# ) ¾ ¾ + (L% ) ¾ ¾
~ ~
1 1 1
= À + Á
2 2“ 2“
1
=
2“
Chapter 4
4.1 Introduction
A component is put to service. It’s time to failure or life length ‘T’ is a random variable. Then
reliability of the component is defined as
( ) = R > S
Hence reliability is the probability of survival of the component at time ‘t’ i.e. it fails at time
greater than ‘t’ .Assume f (t) is probability density function of ‘T’. It is popularly known as
failure density function or simply failure law. Then reliability function can be written as follows
( ) =} ( ) ~
c
R ≤ S = 1− ( ) =} ( ) ~
p
( )=1−} ( ) ~
p
Fig 4.1
Area under the curve right to the line at t represents reliability ( ) and that of left to its
represents failure probability ( ). As it increases ( ) decreases and Y( ) increases. Ultimately
(∞) becomes zero and (∞) becomes one.
In fact assume "p components are put to service and "Â ( ) are survivals at time‘t’.
Ä ( ) = šÅ (c) … (1)
R š Æ
›( ) = ÇšÈ
… (2)
šÉ Ç c
# •šÈ
f (t) = … (3)
šÉ •c
Ë"T is number of components failed in the interval (t, t+∆T). A curve can be plotted for different
interval (t, t+∆T) as shown in the figure 4.2.
Fig 4.2
Plotting it on a graph a standard failure law is identified and parameter can be identified e\with
estimation theory. Similarly the graph for R (t) may be plotted using (1).
≤ ≤ +Ë
ℎ ( ) = lim — ˜ … (4)
Çc→p >
It is failure rate denoted by probability of failing the component in the intervalR ≤ ≤ + Δt]
given that component is working at ‘t’ divided by time interval and taking limit as Δt approaches
zero. Relation (4) for ℎ ( ) is written as
R ≤ ≤ +Ë S
ℎ( ) =
∆c→p Ë R < S
( ). Ë
=
Ë ( )
( )
= … (5)
( )
Note: From equation (4) it is clear that ℎ(0) = (0) i.e. hazard function and failure density
function both are equal at = 0. Precisely ( ) is also a rate but with reference to original
sample at = 0.
Ë"T
ℎ( ) =
Çc→p "Â Ë
1 ~"T
ℎ( ) =
"Â Ë ~
Above relation is obtained even using relation (5) and putting in it for ( ) and ( ).
( )= R ≤® ≤ +Ë S
Çc→p
Ë"T
= lim
∆c→p "p Ë
1 ~"T
=
"p ~
Since ( ) = 1 − Y( )
~ ( )
= − ( ) … (6)
~
1 ~ ( )
ℎ( ) =−
( ) ~
~
ln ( ) = −ℎ( )
~
! ( ) = − } ℎ( ) ~ = −O ( )
p
( )= - (c)
… (7)
Meantime to failure (MTTF) is expected value of life length of the component ‘T’ and is given
as
[
Y= } . ( )~
p
[
~
= } ž− Ÿ~
p ~
[
= R− . ( ) S [
p + } ( )~
p
[
Hence Y = –p ( )
Instantaneous failure rate i.e. hazard function may be (i) Decreasing (ii) Constant or (iii)
Increasing. Usually in the life of a component/device/system all three types of failure rates are
predominant in three stages of the system. Initial period where failure rate is decreasing is known
as debugging period or infant mortality period. Failure rate is decreasing in this period because
time passes it’s design/manufacturing defects are removed. Failure in this period is due to
indogenous reasons. That is internal regions are mainly responsible for the failure rate of the
system. Second stage is the useful period of the device where failure rate is fairly constant.
Failures in this region are also known as random failures. Here in this useful period failures are
by chance and reasons are exogenous. Reasons for failures are external e.g. overstressing,
accident etc. In the third region which is known as wear out region, ageing effect dominates and
increasing failure rate is observed due o wear out. These three failure rate are shown in fig. 4.3.
This curve also shows the combination of three known as λ-curve/bath-tub curve.
Fig 4.3
Curve (a),(b) and (c) represents endogenous failure rates, exogenous failure rate and wear out
failure rates respectively. Curve (d) represents total failure rate. Even in the initial period wear
out failure, will be present but may be very small hence in curve (c) is shown increasing becomes
dominant in the region III. Similarly even in debugging period random failures may be present.
A typical bath tub curve which accounts all three types of hazard rate is given as follows
ℎ( ) = kp + k# 'ª
+ k% '¦
Constants kp , k# & k% and exponents !# & !% are to be selected properly according to their
significance in respective regions.
Fig 4.4
Hence the exponential failure law has constant failure rate. Hence if a component has constant
failure law it implies that it has exponential failure density function. All other distributions have
time dependent failure law. An important property of exponential failure law is that reliability of
the component for a period ‘t’ , given that component is working at the beginning of the period is
independent of the beginning of the interval. Hence in a sense this law is memory less. Hence the
component failure probability in any interval is same if it is working at the starting of the
interval. This is illustrated as follows:
R p ≤ ≤ p+Ë S
R p ≤ ≤ p +Ë / > pS =
R > pS
cÉ Îc
–c -¡c
~
= É
-¡c
=1 − -¡c
Hence the probability of failure of the component is independent of ‘t0’ i.e. the starting of the
interval. Reliability for a period ‘t’ given that component is working at p is independent of t0
and is given as
( / p) = R > + p / > pS = -¡c
Ex. 4.1: A component having constant failure rate has reliability 0.96. Obtain (i) failure rate (ii)
MTTF (iii) Reliability for a period of two years (iv) Median time (v) mode.
Solution:
( ) = -¡c
= 0.96 = -¡
= − ! 0.96 < b •b
(2) = .% X' p.«Ï
t median :
e-Ñ: = 0.5
− 1ƒ• W' = ! 0.5
! 0.5
1ĥ W' =
! 0.96
1U•ƒ = •L -¡c
= 0 0< <∞
€−
( ) = 1 − Y( ) = 1 − /€ =
€
( ) 1
ℎ( ) = =
( ) €−
#
The hazard rate function is increasing which increases from Ò at t=0 to ∞ at t=k.
•
€−
Y = } ~ = €a2
p €
Fig 4.5
Ex.4.2: A component has uniform failure law between [0,10].Determine its reliability for 1 Hr
period provided it is working at (i) the end of first hour (ii) the end of second hour.
Solution:
1
( ) = ;0 < < 10
10
£a ¥= R > p + ⁄ > pS
p
#p
1
} ~
cÉ Îc
10 10 − p −
= =
#p 1 10 − p
–c 10 ~
É
m1a1n = = 8a9
#p-#-#
#p-#
9
m1a0n = = 0.9
10
m1a2n = = 7a8
#p-%-#
#p-%
1 c-q ¦
( )= -p.y £
¤
¥
√2¢ _
Y= ]
c
Y( ) = R < S = } "(], _ % ) ~
-[
−]
= Ô (§) = Ô ž Ÿ
_
−]
( ) = 1− Ôž Ÿ
_
− ] -#
h(t) = "(], _ % ). —1 − Ô ž Ÿ˜
_
Fig 4.6
This distribution function is applicable when two random variables comes one after the other in
stages and each of them is having exponential distribution. Specifically this is a failure law of a
standby system when main component is in operation and other is in stand by mode having
failure rate λ1 & λ2 .This law is written as
λ# λ%
¯ -Ѫ :
– -Ѧ :
°
λ% − λ#
λ% -Ѫ :
– λ# -Ѧ :
Y( ) = 1 −
λ% − λ#
λ% -Ѫ :
– λ# -Ѧ :
( ) = 1 − Y( ) =
λ% − λ#
[
λ% – λ#
-Ѫ : -Ѧ :
Y= } ~
p λ% − λ#
1 1
= +
λ# λ%
λ# λ%
ℎ( ) = ¯ -Ѫ :
– -Ѧ :
°
λ% -Ѫ : – λ
#
-Ѧ :
Fig 4.7
λ# λ%
( )= lim ¯ -Ѫ :
– -Ѧ :
°
Ñ% → Ñ# λ% − λ#
λ# λ% -Ѫ:
= lim ¯1 – ( Ѧ -Ѫ ) :
°
Ñ% → Ñ# λ% − λ#
λ# λ% -Ѫ: (λ% − λ# )% t %
= lim À1 – 1 + (λ% − λ# ) − + …..Á
Ñ% ,Ñ# → Ñ λ% − λ# 2!
( )= % -Ñ :
[
( )= } % -Ñ :
~
c
= (1 + ) -Ñ :
[
2
Y = } (1 + λt) -Ñ:
=
p
2 %
( )= -Ñ :
, >0
2!
[ 2 % 2
2!
} -Ñ :
~ = =1
p 2! 2! 2
3 2 y 3 V V-#
-Ñ :
, ,… -Ñ :
3! 4! (b − 1)!
V V-#
( )= -Ñ :
, ≥0
(b − 1)!
( )% ( )V-#
( ) = Ø1 + + +⋯ Ù -Ñ :
2! (b − 1)!
[ ( )% ( )V-#
Y = } Ø1 + + +⋯ Ù -Ñ :
~
p 2! (b − 1)!
b
=
V V-#
-Ñ :
(b − 1)!
ℎ( ) =
( )% ( )V-#
ž1 + + 2! + ⋯ Ÿ -Ñ :
(b − 1)!
V V-#
(b − 1)!
=
( )% ( )V-#
ž1 + + +⋯ Ÿ
2! (b − 1)!
Fig 4.8
It is a mixture failure law it represents the failure distribution of a group of devices which comes
from more than one brand. Each brand has different constant failure rates. It is represented as
( ) = Ú# #
-Ѫ :
+ Ú% %
-Ѧ :
+ Ú2 2
-ÑÛ :
+ ⋯ +ÚV V
-ÑÜ :
Ú ≥ 0 & )Ú = 1
[
( ) = } )Ú -ÑÝ :
~
c
[
=) } Ú -ÑÝ :
~
c
[
= À) −Ú
-ÑÝ :
Á
c
( ) = )Ú -ÑÝ :
∑ Ú -ÑÝ :
ℎ( ) =
∑Ú -ÑÝ :
The hazard rate is decreasing and its plot is shown in fig 4.9
Fig 4.9
Wear in period is the time period spent by the system before it is put to service. At times it is
desirable to put or operate the component at reduced stress and level so that its manufacturing
defects may be identified and removed and may give increased reliability for a specified period.
It is expected that it is desirable to put the component in wear in period under debugging period
when the hazard rate is decreasing. It is of no use putting a component in wear in period if failure
is constant or increasing. Let us evaluate the reliability for a period ‘t’, given that component is
working at ‘ p ’ i.e.
£a ¥= R > p + ⁄ > pS
p
R > p+ S ( p + )
= =
R > pS p
Hence
Þ ßÞ
- –É É d(c) •c
£a ¥= Þ
p - –É É d(c) •c
ÞÉ ß Þ
Þ ßÞ
-À} d(c)•c - –É É d(c)•c Á
= É
Þ
- –Þ ÉßÞ d(c)•c
= É
:É Î :
ln R £tat ¥ = − } h(t)dt
p
:É
1 ~ £ at ¥
p
= Rℎ( p ) − ℎ( + )S
£ at ¥ ~ p
p
p
~ £ at ¥
p
= Rℎ( p ) − ℎ( + )S £ at ¥
~ p p p
•á£ca: ¥
•cÆ
É
> 0 if failure rate is decreasing
•á£ca: ¥
É
< 0 if failure rate is increasing
•cÆ
•á£ca: ¥
É
= 0 if failure rate is constant
•cÆ
Hence components should be if failure rate is high and decreasing this only happens under
debugging period.
Exercise # 1
( )= -âc ¦
, ≥ 0 & α is a positive constant.
Obtain
(iii) MTTF.
Obtain
(i) MTTF
(ii) Failure law.
(iii)Hazard Rate.
Ex:4 A components have constant failure rate 0.4/year. Determine the warranty period to be set
by the manufacturer so that no more than 4% components are returned for service.
Ex:5 A computer manufacturers provide warranty for one year period . Determine the maximum
tolerable failure rate. Assume constant failure rate and the manufacturer does not expect more
than 5% sold computers to be returned for service during warranty period.
= k# , > p
Ex :8 two components have same MTTF. One of them has constant failure and other have
linearly increasing failure rate 2 Ú .Obatin (i) Ú in terms of . (ii) If for each component the
design life reliability is 0.9.Determine the ratio of design lives of two components.
(i) Probability of failing it in third year provided that it survives for two years.
(ii) If upon failure the device is immediately replaced. Determine the probability that (a) No
failure will occur in two years (b) More than one failure will occur in two years.
It is assumed that the component is having a constant failure rate. Once the component fails it is
replaced by a similar component immediately having same failure rate. The following analysis is
applicable even if the component is repairable and repair time is negligible. Such that as the
component fails, repaired and put back to service.
p( +~ ) = p( ) (1 − ~ )
Pp (t + dt) − Pp (t)
= −λ Pp (t)
dt
Similarly
P# (t + dt) − P# (t)
= λ Pp (t) − λ# (t)
dt
(K + ) # (K) =
K+
λ
P# (s) =
(s + λ)%
P# (t) = λ t e-Ñ:
P (t + dt) − P (t)
=λP -# (t) − λ P (t)
dt
λ P -# (s)
' (K) =
(K + )
Now for ! = 2
P# (s)
P% (s) =
(K + )
.λ λ% t % -Ñ:
P% (s) = & P (t ) = e
(K + )% %
2!
Proceeding in similar manner general solution of the general differential equation is written as
λ t -Ñ:
P (t) = e
n!
Above solution gives probability of n failures in time ‘ ’.This solution describes Poisson process.
If for a specific time one defines
]=
] e-æ
P =
n!
It is Poisson distribution.
Now if one has N spare components to be replaced than reliability of the system is written as
š
λ t -Ñ: ( )% ( )š
=) e = Ø1 + + +⋯ Ù -Ñ :
n! 2! "!
'Zp
V
λ t -Ñ:
RJ ≤ bS = ) e
n!
'Zp
& RJ > bS = 1 − RJ ≤ bS
Ex : A component has failure rate 0.5 per year. Calculate number of spare components required
to get reliability 0.99 for two years of operations.
Solutions;
Let us assume the spare components required are ‘n’. The reliability with n spare is given by
( )% ( )š
Ø1 + + +⋯ Ù -Ñ :
= 0.99
2! "!
= 0.5 × 2 = 1
!=0
1
-.#p
= = 0.3678
!=1
2
(1 + ) -.#p
= = 0.7357
!=2
2.5
(1 + 1 + 0.5) -.#p
= = 0.9195
!=3
1
ž2.5 + Ÿ -#.p
= 0.9820
6
!=4
1
ž2.67 + Ÿ -#.p
= 0.9970
24
1( )= ( ) 0< < p
1( )= ( p ). ( − p) p < <2 p
1 (2 p ) = ( p )%
1( )= ( p ). ( − 2 p ) 2 p < <3 p
It is expected in general that with increasing failure rate reliability will decrease with a lower rate
than what it would have been without maintenance as shown in figure below.
Fig 4.10
Y= } 1( )~
p
c %cÉ ('Î#)cÉ
= } ( )~ + } ( p ). ( − 2 p )~ + … … … } ( p )' . ( − ! p ) ~ + ….
p cÉ 'cÉ
[ ('Î#)cÉ
= ) } ( p )' . ( − ! p ) ~
'Zp 'cÉ
[ cÉ
= ) } ( p )' . ( é ) ~ é
'Zp p
[ cÉ
= ) ( p )' . } ( é ) ~ é
'Zp p
cÉ
–p ( é) ~ é
=
1 − ( p)
[
1
) ( p )' =
1 − ( p)
'Zp
Effect of preventive maintenance has desired and quite definite effect when failure rate is
increasing due to ageing or wear causes. It is of no use when hazard rate is constant or
decreasing. In such situation even reliability may be reduced due to imperfect maintenance.
Let us assume general expression for failure rate as described by weibull failure law
ℎ( ) = Ú ã ä-#
( )= -âc ê
1 (! p )
-â'cÉ ê
= = â ä
!(!ä-# − 1)
(! p ) -â('cÉ )ê p
This implies that β >1.This implies increasing failure rate. If β =1, there is no improvement in
reliability . this implies constant failure rate .if β < 1 (decreasing failure rate) then reliability
with maintenance will decrease. These aspects have been shown in fig 4.11
Fig 4.11
In most devices encountered in electrical or electronics having moving or rotating devices and
may have slightly increasing failure rate. An excellent example of this is alternator due to wear it
may have increasing failure rate but due to proper preventive maintenance over an adequate
interval the hazard rate can be assumed constant. Assume the failure rate of the component due
to random failures and wear as follows
ℎ( ) = Úp + Úã ä-#
,ã > 1
( )= -(âÉ cÎ â c ê )
If it assumed that preventive maintenance is carried out at an interval p, then reliability with
maintenance is given as follows
1( )= . . … . . ìℎ b ! < < (! + 1)
-'âÉ cÆ -'âcÆ ê -âÉ (c-'cÆ ) -â(c-'cÉ ) ê
p p
1 ~ 1
ℎ1 ( ) = −
1( ) ~
íb ℎ1 ( ) = Úp + Úã ( − ! U )ä-#
than the first term than overall hazard rate can be approximated as a constant by averaging over
U this is shown in figure below.
Fig 4.12
Úã
λîïð = Úp + ä-#
2 p
If the maintenance is not ideal i.e. imperfect. Assume p is the probability of unsatisfactory
maintenance. This means per repair reliability reduces by Factor (1 − <). Then in ‘n’ repair the
factor by which this reliability with preventive maintenance reduces is (1 − p) . Hence
reliability
1 ( p) <
= 'âcÆ ñ
À!ä-# − 1 − Á
(! p ) Ú p
ä
<
! ä-# − 1 − >0
Ú p
ä
Ú p
ä
! ä-# − Ú p
ä
− <>0
<< Ú p
ä
(!ä-# − 1)
ℎ( ) = 0.2 , K ! •bK
Solution:
O ( ) = } 0.2 ~ = 0.1 %
Since p << 1
(1 − p) ≅ e- ò
( )= -âc ê
O ! 1( )= . .
-'âcÆ ê -'u -â(c-'cÉ ) ê
1( )• =! p
1 (! p ) = -(âcÆ ê Î 'u)
The ratio of reliability with imperfect maintenance and without maintenance is given as
1 (! p ) − ('âcÆ ê Î 'u)
= = Râ'ê cÆ ê – ' âcÆ ê -'uS
(! p ) -(â'cÆ ê)
( p) = -p.# cÉ ¦
= -p.3 ×p.%y
= -p.p%y
= 0.9753
1( )= = = .7788
-p.p%y ×#p -p.%y
(b)
′1 (5) = (1 − .01)#p × -p.%y
= 0.7043
(c) Reliability function with maintenance is written as
1 ~ 1
ℎ1 ( ) = − = 0.2 ( − ! ), ! < < (! + 1)
1( ) ~
p p p
ℎ1 (! p ) = 0
ℎ1 (! p ) + ℎ1 ((!+1) p )
= = 0.05
2
1 = -p.y ×y
= -p.%y
= .7788
Fig 4.13
Chapter 5
5.1 General
This chapter presents reliability evaluation methodologies which are represented by reliability
network or logic diagram. The reliability network may be a series or parallel or series-parallel or
a complex configuration. It is to be stressed here that the reliability networks in no way
represents the connection of components in a physical system. For example two transmission
lines operating in electrically parallel will be considered in series configuration for reliability
calculation if successful operation of both is required to meet the demand. The relation used is
derived will be applicable for reliability evaluation without time factor or with time dependents
function under the assumption that all the component are put in service simultaneously(This will
not be the case in standby system). Generally it is assumed that components in system fail
independently.
A series network is simplest and block diagram representation of it is shown in fig 5.1.
Fig. 5.1
Physically a system configuration will be a series reliability network if system fails even if a
single component fails or system survives if all the components are working successfully. This is
also known as Non-redundant system.
'
 = õ b … (5.1)
Z#
Sometimes this is also known as static reliability. Now assume T# , T% , T2 , … T are life lengths of
these components and f# (t), f% (t), f2 (t), … f (t) are the failure density function of respective
components. Life length ‘T’ of the series system will be minimum of individual life lengths i.e.
= minR T# , T% , T2 , … T S … (5.2)
The above relation implies that if ( ≥ ) then all T# , T% , T2 , … T must be greater than‘t’ (if ‘T’ is
minimum). Above relation can be written in terms of probability as follows
'
 = õ b ( ) … (5.5)
Z#
Failure density function of series system may also obtained using (5.4)
'
1 − F(t) = õ(1 − Y ( ))
Z#
'
F(t) = 1 − õ(1 − Y ( ))
Z#
' '
~ ~
( ) = ÷1 − õ(1 − Y ( ))ø = − ùõm1 − Y ( )nú
~ ~
Z# Z#
~ ~
( )= õ ( )=− '
( )
~ ~
=! '-#
( )φ(t)
'
Â( )=õ -¡‡ c
Z#
Â( )= -¡Å c
 =)
( )= Â
-¡Å c
Hence the series system has also exponential failure law. Failure rate is sum of individual failure
rates. Further MTTF of system
1 1
Y= =
 ∑
'
þ Â
ý = = õb
•
þb•
Z#
•
It is obvious for series system the reliability importance is highest of ‘weakest’ component.
Hence the component which has lowest reliability may be improved in design / quality /
strengthing if at all overall system reliability need to be improved.
Fig. 5.2
System fails if all components fail. System performs its function even if a single component is
working. Since all components are connected in the system thus this arrangement is also known
as active parallel. Probability of failure of the system i.e. un-reliability is given by (if there
reliability or independent time factor)
'
u = õ(1 − b )
Z#
'
u = 1 − õ(1 − b )
Z#
u = ) b − ) b b+ + ) b b+ b• − … + (−1)'-# b# b% … b'
u = 1 − (1 − b)'
Now assume each component has f# (t), f% (t) … f (t) as failure density function and
corresponding r# (t), r% (t) … r (t) as reliability function. Life length of system is given as
= max R T# , T% , T2 , … T S
Following is written
( ≤ ) = (T# ≤ t) (T% ≤ t) … (T ≤ )
@ ( ) = Y# ( )Y% ( ) … Y' ( )
@( ) = õ Y ( )
( ) = 1 − õ(1 − b ( ))
u( ) = 1 − õ(1 − )
-¡‡ c
~
”( ) = õ(1 − -¡‡ c
)
~
~
∏(1 − -¡‡ c
)
ℎ( ) = ~
1 − ∏ (1 − -¡‡ c )
It is noted that each component has constant failure rate but the failure rate is time dependent.
The plot of the hazard rate is given in fig 5.3 as increasing.
Fig. 5.3
• = ! N λ# , λ% , … λ P
Y = } ù1 − õm1 − -¡‡ c
nú ~
∞
Y=} ) -¡‡ c
−) -(¡‡ Ρ )c
+ ) -(¡‡ Ρ Ρ )c
+ … + (−1)' - ∑ ¡‡ c
~
p + • +
1 1 1 1
=) −) + ) + … + (−1)'
( + +) ( + + + •) ∑
+ • +
For example ! = 2
1 1 1
Y% = + −
λ# λ% λ# + λ%
For ! = 3
1 1 1 1 1 1
Y2 = + − − − +
λ# λ% λ# + λ% λ% + λ2 λ2 + λ# λ# + λ% + λ2
If n units are identical having each as constant failure rate λ than expression for MTTF is given
as
'
'
• (−1)
•Î# ∑ ' • (−1)•-#
Y=) =
€ €
•Z#
!=2
1.5
Y% =
It is true that reliability of parallel system increases as more number of components are
connected. But common failure may not permit to achieve ideal results. The cause of common
failure includes common operational environment, external catastrophic events, and operational
errors and common external power source etc.
The common cause failure effects is modeled by connecting a hypothetical unit in series with
parallel combination representing the reliability owing common cause failure as shown in fig 5.4
below.
Fig. 5.4
λ# , λ% are failure rate of each unit independently and λ is failure rate due to common cause. The
reliability of the system is written as
=R -¡ª c
+ -¡¦ c
+ (-¡ª Ρ¦ )c
S -¡ c
λ# = λ# + λ , λ% = λ% + λ
=R -(Ѫ -Ñ )c
+ -(¡¦ -Ñ )c
+ (-¡ª Ρ¦ -%¡ )c
S -¡ c
=R -¡ª c
+ -¡¦ c
+ (-¡ª Ρ¦ -¡ )c
S
λ# = λ% = λ
λ# = λ% = λ
=R2 -¡ c
− -(%¡ -¡ )c
S
At times a factor ã is defined as the ratio of common cause failure to total failure rate as
λ
ã=
λ
ã- factor may be substituted in above relation and reliability expression is written as follows
= R2 -¡ c
− -(%¡ -ä¡ )c
S
= ¯2 -¡ c
− -(%-ä)¡ c
°
=R 2− -(#-ä)¡ c
S -¡ c
At times this technique is known ã- factor technique. General expression for ‘n’ component in
parallel accounting common cause failure is written as
( ) = R 1 − (1 − -¡c '
) S -¡ c
Ex. A system constitutes of three components each having constant failure rate λ. For system
success at least two units must be available. Obtain (a) Reliability (b) Hazard rate (c) MTTF of
the system.
Solution:
%a = b 2 + 3b % (1 − b)
2
= -2¡c
+3 -%¡c
−3 -2¡c
=3 -%¡c
−2 -2¡c
= 3b % − 2b 2
−~ %/2
( )= =6 -%¡c
−6 -2¡c
~
6 -%¡c
−6 -2¡c
ℎ( ) =
3 -%¡c − 2 -2¡c
Fig.5.5
3 2 5
= − =
2 3 3
This arrangement is also known as k-out of n configuration. In this ‘n’ units are operating in
parallel the success requires at least ‘k’ unit in operation or in up states, where k is given as
1< k < n.
Fig.5.6
Assuming each component is identical and having reliability ‘r’ for a specified duration. Using
Binomial distribution probability success of ‘m’ success out of ‘n’ unit is written as follows
Since at least ‘k’ units are required for system success. Reliability expression is written as
'
•-#
= 1 − ) 'k1 b 1 (1 − b)'-1
1Z#
If each component has failure rate λ, then the reliability •/' is given as
'
•/' = ) '
k1 -¡1c
(1 − )
-¡c '-1
1Z•
Using above failure density function, hazard rate and MTTF of partially redundant system may
be calculated.
If the units are not identical than Binomial distribution is not applicable. In this situation a
product in the following manner is formed and states where at least ‘k’ units are available are
selected. There probabilities are calculated and added to give required results.
= <# <% <2 … <' + e# <% <2 … <' + <# e% <2 … <' + ⋯ + e# e% e2 … e'
Ex. A system of three components each having failure rate as #, %, 2. System success requires
at least two components in operation. Obtain
Solution:
%a = b# b% b2 + e# b% b2 + b# e% b2 + b# b% e2
2
= -(¡ª Ρ¦ )c
+ -(¡¦ Î¡Û )c
+ -(¡ª Î¡Û )c
−2 -(¡ª Ρ¦ Î¡Û )c
= b# b% + b% b2 + b# b2 − 2b# b% b2
(2) MTTF
1 1 1 2
= + + −
#+ % %+ 2 #+ 2 # + % + 2
This section deals with reliability evaluation of a non series parallel network. Following four
methods are discussed
1) Event space
2) Decomposition
3) Tie-set and
4) Cut- set
This is one of the simplest techniques for evaluating reliability of a network. The method is
implemented in the following steps.
Step –1. Based on the total number of components, enumerate all possible states of the network.
Assuming each component has two states (up and down), total no. of states will 2 . n denotes
total no. of components. These events may be identified using a event-tree as shown in fig. 5.7
for three component namely a,b,c. The event tree will grow in size as the number of component
increases.
Step –2. Looking at the network, partition the total number of states in two sets i.e. ‘w’ – success
set of event and ‘f’ – fail set of event.
Step –3. Evaluate the probabilities these events and obtain reliability of the system as follows
= ) (^ )
∈
=1− = ) (^+ )
+∈T
"ç : Simple addition in above relation is due to the event enumerated being mutually
exclusive.
^L. Evaluate the reliability of following network using event space method.
Fig. 5.8
Solution:
^# = •D ~ ^« = •D ~
^% = •D ̅~ ^#p = •D ̅~
^2 = •D ~̅ ^## = •D ~̅
^y = •D ~ ^#2 = •D ~
^Ï = •D ̅~ ^#3 = • D ̅~
^ = •D ~̅ ^#y = •D ~̅
Matching the event with the network following events are in ‘w’ sets
W : [^# , ^% , ^2 , ^3 , ^y , ^ , ^« , ^#2 ]
= (•D ~) + (•D ̅~) + (•D ~̅ ) + (•D ̅~̅ ) + (•D ~) + (•D ~̅ ) + (•D ~) + (•D ~)
= b 3 + 4b 2 (1 − b) + 3b % (1 − b)%
= 3b % − 2b 2
Further assume each component has constant failure rate λ i.e. b = -¡c
. The above reliability
expression reduces to
=3 -%¡c
−2 -2¡c
Failure law
( )=6 -%¡c
−6 -2¡c
6 ( -%¡c − -2¡c
)
ℎ( ) =
3 -%¡c − 2 -2¡c
3 2 9−4 5
Y= − = =
2 3 6 6
The method is simple but as number of component increasing states also increases and thus
computational burden increases. Total number of states are 2 , if n denotes total no. of
component, but total number of states can be reduced by considering the failure of a small
number of components. Since failure of large number of components is unlikely i.e. probability
of such events/ states with failure of more than a specific number of component is negligible.
In this technique a key-stone component is selected as ′J′ and based on this component’s state
system success is written as follows
= .J + .J
( ) = ( . J) + ( . J)
Decomposition is based on that system success will occur either in intersection with J working
or with intersection J has failed. From above relation reliability is written as follows
Where
Fig. 5.9
As shown in above Figure the network is decomposed into two networks ′ ′ and ′0′. In ′ ′
system success is represented when ′J′ is working which is shown as ‘short-circuit’ and where as
in ′0′ system success is evaluated while ′J′ has failed whivh is shown as ‘open-circuit’
condition. The methodology is explained by evaluating the reliability of the following system
Fig. 5.10
(k ) = (J) = b
mkn = mJn = (1 − b )
Fig. 5.11
It is easier to evaluate the ( ⁄J) from above network. Close observation of above network
reveals that reliability of above network is simply ′D′ and ′~′ in parallel combination i.e.
( ⁄J) = b• + b• − b• b•
Fig. 5.12
m ⁄Jn = bW b•
( ) = (b• + b• − b• b• )b + bW b• (1 − b )
( ) = b• b + b b• − b• b b• + bW b• − bW b• b
þ ( )
= b• − b• b = b• (1 − b )
þbW
þ ( )
= b − b b• + bW − bW b = b (1 − b• ) + bW (1 − b )
þb•
þ ( )
= b• + b• − b• b• − bW b•
þb
þ ( )
= b − b• b = b (1 − b• )
þb•
þ ( ) þ ( )
= = b − b%
þbW þb•
þ ( ) þ ( )
= = 2(b − b % )
þb• þb
Hence it is obvious if each component has reliability ′D′ and ′ ′ has twice reliability importance
as compared to ′•′ and ′~′. Hence if system reliability is to be improved then it should be
improved by investing money in components ′D′ and ′ ′ both or ′D′or ′ ′.
This method is also known as path tracing method. A tie-set is a set of components which
provide path from input to output while all of the components in the set are in up state. Minimal
tie-set is a set of minimum number of components which while all of them in up state provides a
path (success). Even if a single component is removed from this ‘minimal tie-set’ the remaining
set of component no longer provides a path (success) even if all of them are in up state. The
methodology is explained in following steps
R # , %, 2 … … … V S
Step-2: For system success at least one tie-set must be available. Thus equivalent network is
drawn as follows
Fig. 5.13
= )
Z#
= ) ( )−) ( +) + ) ( + •) + ⋯ + (−1)V-# ( # % … … … V)
+ • +
Consider the complex network used in previous sub-section. Looking at the network Tie-sets are
obtained as follows
# = •. D , % = .~ , 2 = D.
= ( #) + ( % ) + ( 2 ) − ( # %) − ( # 2) − ( % 2) + ( # % 2)
= bW b• + b b• + b• b − bW b• b − b• b b•
If bW = b• = b = b• = b we obatin
= 3b % − 2b 2
A cut-set is defined as a set of component which fail system fails. A minimal cut-set is a set of
component which fail system fails; even if a single component is removed from minimal cut-set
the remaining components do not form a cut-set. The method is explained in following steps
Step-2: For system failure at least one cut-set should occur i.e. system will fail even if all the
elements in any one of the minimal ‘cut-set’ fails. The equivalent diagram is shown in following
Fig. 5.14.
Fig. 5.14
= ) ( k ) − ) ( k k+ ) + ) ( k k+ k• ) … . . (−1)V-# (k# k% … … … kV )
+ • +
Fig. 5.15
k# = • , k% = D ~, k2 = D , k3 = • ~
(i) The method can be programmed on a digital computer. No doubt tie-set method also can be
programmed.
(ii) While one identifies cut-set, various failure modes are directly available. This enables one to
strengthen the weak components or devices.
(iii) In cut-set method approximate evaluation of reliability is possible. This may be achieved in
one or two following ways
(a) Limiting the order of cut-sets i.e. neglecting the higher order cut-sets which contain more
than a specified number of elements. Since probabilities of occurrence of such ‘cuts’ may be
negligible.
(b) By truncating the unreliability relation ′ ′ by retaining only first order terms as follows
≅ ) (k )
≅1−
Failure probability thus obtained will be on higher side and reliability will be on lower side.
Usually in engineering application each component is highly reliable and above approximations
are justified in practice
Method for obtaining tie-sets which are required in Tie-Set method of reliability evaluation.
Assuming a reliability network is given, form a connection matrix "; defined as follows
" = 1,
=0 Otherwise
Where ′€′ is the node being eliminated. Ultimately after eliminating all intermediate nodes a
2 × 2 matric will result. All minimal tie-set can be obtained from this matrix from " 'u c×U cu c
element. In this process at times a non-minimal tie-set may result which should be ignored and
identified
Ex.: Obtain tie-sets of the following network using Node elimination method
Fig. 5.16
1 • • 0
0 1 1 ~
"=
0 0 1 D
0 0 0 1
Node € = 2 to be eliminated
" é## = 1
" é 2# = 0
" é 22 = 1
" é 3# = 0
" é 32 = 0
" é 33 = 1
1 (• + ) ~
" é = ù0 1 Dú
0 0 1
€=3
" éé## = 1
= ~ + (• + )D
= ~ + •D + D
" éé 3# = 0
" éé 33 = 1
1 •D + D + ~
" éé =
0 1
# = •D, % =D , 2 = ~
Once tie-set are obtained are obtained a matrix in the following format is prepared which exhibit
in a row (i) elements present in k :; tie-set (k :; column) i.e. A Ò represent an element present in
i:; path and, k :; column is considered for this element.
This means A Ò = 1, if i:; tie-set contains k :; element and 0 otherwise. Then if a specific column
contains all ones then the corresponding elements form a first order cut. Now examine sum of
two column’s combinations (excluding the first order cut). Combinations giving one at every
place will form second order cut. Thus continuing the increased numbers of combinations higher
order cuts are obtained. It is stressed here that while looking at combinations the combinations of
lower order which forms cut should not be considered. The procedure is illustrated with the
following example shown in following fig. 5.17.
Fig. 5.17
T# = ab , T% = ~ , T2 = D
Elements
a b c d e
Tie-sets
T1 1 1 0 0 0
T2 0 0 1 1 0
T3 0 1 1 0 1
It is obvious from the above table; there is no column which contains all ones. Hence there is no
first order cut-set.
k# = • , k% D , k2 D~
k3 •. . ~
It has been considered that the component has two states i.e. ‘up’ and ‘down’. In many situations
the component may have more than two states. A generator may have ‘up’, ‘down’ and a
‘derated’ states. A diode may function properly; fail by open circuit or short circuit. Thus diode
has two failure modes and one state of normal operation. Reliability of such system may be
evaluated either by event-space
space or decomposition method.
Consider the example of two diodes operating in parallel for rectification purpose as shown in
fig. 5.18.
Fig. 5.18
Reliability of above configuration is required for rectification purpose. Assume p # , p!# , p?#
are the probabilities of normal, open circuit and short circuit for diode one and p % , p!% , p?% are
the same probabilities of diode number two. Events may be obtained by product rule or by tree.
In all there are nine states. States corresponding to system success are added and reliability is
evaluated as
pé = p # . p % + p!# . p % + p!% . p #
pé ! = p!# . p!%
Fig. 5.19
The equivalent diode has pé , pé ! , pé ? as probabilities of normal, open and short circuit states.
p # =p % =p ,
p!# = p!% = p! ,
p?# = p?% = p?
pé p %
+ 2p p!
Fig. 5.20
In this case total numbers of states are same i.e. nine as earlier but the equivalent three states are
identified as follows
péé = p %
+ 2p p?
péé ? = p? %
It is obvious from expressions of pé and péé that for reliability enhancement two diodes should
be connected in series if probability of failure of diode by short circuit is more than that of
probability of failure by open circuit, other wise they should be connected in parallel.
Assume the two diodes have been connected in parallel as shown below
Fig. 5.21
. J" + . Jš + . J#$
= p # . p!% + p # . p % + p!# . p %
Fig. 5.22
Parallel combination of diode may be assumed as a single component ′J′. Then reliability
expression is written as
p?2 + p 2 p #. p % + p!# . p % + p!% . p # +p 2 p?# . p?% + p!# . p?% + p!% . p?# + p # . p?%
+ p % . p?# + 0. (p!# . p!% )
Exercise # 1
1. Determine (i) Reliability (ii) Hazard rate (iii) MTTF (iv) Failure Law for the following
network
Fig. 5.23
Fig. 5.24
4. A non-redundant system is designed with 100 component and has design life reliability as
0.92. The system is redesigned with 80 component. Determine new design-life reliability.
Assuming same design-life. Assume each component has same constant failure rate.
5. Thermocouples are used in a system each having failure rate 0.009/hr. How many
thermocouples must be placed in parallel if the system is to run for 100 hrs. with a system failure
probability not more than 0.05. Assume all failures are independent.
6. A disk drive has constant failure rate and an MTTF of 5000 hr. Determine
(i) Probability of failure for one year of operation if two of the drives are placed in parallel and
(ii) Again determine the probability of failure for one year of operation if common mode
failures are characterized by ã = 0.2.
7. A system consists of two sub-systems in parallel. Each sub-system has reliability function as
-(ca%)¦
( )=
& is a constant.
8. A component has two failure modes and represented by a three state model. Failure rates to
mode one and two are # 0.1/ yr. , % = 0.05/ yr. Obtain reliability of the component for one
year. Probability of other two failed states.
(i)
(ii)
Chapter 6
6.1 Introduction
This section discusses reliability evaluation of a standby system under the assumptions
1) Failure rate of switch is zero
2) Probability of failing the switch on demand is zero
3) Failure rate of idle component is zero while in standby mode
A standby system with ‘r’ standby components is shown in fig. 6.1 below
Fig. 6.1
Further assume that failure rates of each component as λ1, λ2, λ3……. . λr+1. A state transition
diagram of the system is shown below.
Fig 6.2
State 2 – Second standby is in operation [main & first standby has failed]
State n – nth standby is in operation [main & (n-1) standby component have failed]
State 0 equation
pp t + dt = p! (t)(1 − #~
= (Probability of system in ‘0th’ state at‘t’) (probability the component does not fail)
p# t + dt − p# t
= λ# pp − λ% p#
dt
p% t + dt = p# t (λ2dt) + p% t 1- λ3dt)
p% t + dt − p% t
= λ% p# − λ2 p%
dt
<å% = λ% p# − λ2 p% … (6.3)
<'å = λ p -# −λ Î# p … (6.4)
<(å = p A …(6.5)
−λ# λ# 0 0 0 …
0 −λ% λ% 0 0 …
0 0 0 −λ2 λ2 …
⋮ ⋮ ⋮ ⋮ ⋮ ⋱
[1 0 0 ⋯ 0]
In sequence above equation are solved and if there are ‘r’ standby then reliability of the system is
given by
)<
Zp
1− ()
Ex. Assume the system consists of one main component and one standby having failure rate as
λ# and λ% respectively.
pp t + p# t
pp t e-Ѫ:
p# t is obtained by solving
<å# = λ# pp − λ% p#
K
λ#
#
K + λ# K + λ%
aλ − λ aλ − λ
λ# λ#
% # % #
K −
#
K + λ# K + λ%
λ# λ#
p# t e-Ѫ: − e-Ѧ:
λ% − λ# λ% − λ#
λ# λ#
e-Ѫ: + e-Ѫ: − e-Ѧ:
λ% − λ# λ% − λ#
~
−
λ# λ%
= [e-λª : − e-λ¦ : ]
~ λ% − λ#
Further if the two failure rates are equal λ# λ% λ then R(t), f(t) expression reduces to
= (1 + )e-Ñ:
( ) = λ % te-Ñ:
In general for ‘r’ standby system where each component has ‘λ’ failure rate
(λt) -Ñ:
p = e
n!
V
(λt) -Ñ:
( )=) e
n!
'Zp
Assume that system consists of one main and one standby component each having same failure
rate. Further assume that p? is the probability of successful change-over. The state transition
diagram is shown in figure below
Fig. 6.3
e = 1 − <Â
State 1 – Main component has failed and change-over has been successful and standby is
operating
−λ p? λ q? λ
[p!å p#å p%å ] [pp p# p% ] ù 0 −λ λ ú
0 0 0
p!å = −λpp
pp = e-Ñ:
p? λ
K+ p#
K+
p? λ
p# s
K+ %
p# t p? λte-Ñ:
Reliability is given as
pp + p# 1 + p? λt) e-Ñ:
1 + p?
Y=
λ
~
( )= −
~
(b)Assume switching is imperfect and switch has a failure rate *+ (imperfect switching)
In addition to unsuccessful change-over the switch while in connection having a failure rate λ? .
The state diagram is drawn as
Fig. 6.4
State 0 – represents the main component is in operation and standby is in standing mode
State 1 – main component has failed and standby is in operation (with successful) change-over
I. By unsuccessful change-over
II. By failure of switch while main component is in operation
III. Failure of switch while standby is in operation
IV. Failure of standby component while in operation
− λ + λ? p? λ q? λ + λ?
[p!å p#å p%å ] [pp p# p% ] ù 0 −(λ + λ? ) (λ + λ? ) ú
0 0 0
pp = e-(ÑÎÑ,):
p#å p? λpp − λ + λ? p#
p? λ
K + λ + λ? p#
K + λ + λ?
p? λ
p# s
K + λ + λ? %
p# t p? λte- ÑÎÑ, :
Reliability is given as
The result is expected. Previous result multiplied by reliability of switch (e-Ñ,: ). This is
equivalent to imperfect switch in series.
(c)Assume that the standby components may fail during idle mode and failure rate is *- .
Fig. 6.5
− λ + λ? + λ p? λ λ q? λ + λ?
[p!å p#å p%å p2å ] [pp p# p% p2 ] 0 −(λ + λ? ) 0 −(λ + λ? )
0 0 −λ? λ?
0 0 0 0
State 0 – represents the main component is in operation and standby is in idle mode
State 1 – main component has failed, successful change-over has taken place and standby
component is in operation
State 2 – main component is in operation but the standby component has failed in idle mode
State 3 – is absorbing state which is reached via either unsuccessful change over, switch failure
or by standby component failure
Reliability is given by
pp + p# + p%
p!å − λ + λ? + λ pp
pp e- ÑÎÑ, ÎÑÝ :
p#å p? λpp − λ + λ? p#
p? λ
K + λ + λ? p#
K + λ + λ? + λ
p? λ
p# s
K + λ + λ? K + λ + λ? + λ
p? λ - ÑÎÑ :
p# t ¯e , − e- ÑÎÑ, ÎÑÝ :
°
λ
p%å λ pp − λ? p%
λ
P% s
K + λ? K + λ + λ? + λ
λ λ
a λ+λ a λ+λ
P% s −
K + λ? K + λ + λ? + λ
Prof. L.D. Arya
Reliability Engineering 122
λ
p% t [e-Ñ,: − e- ÑÎÑ,ÎÑÝ : ]
λ+λ
Reliability is given as
p? λ - ÑÎÑ : λ
e- ÑÎÑ,ÎÑÝ : + ¯e , − e- ÑÎÑ, ÎÑÝ : ° + ¯e-Ñ,: − e- ÑÎÑ,ÎÑÝ : °
λ λ+λ
Assume two systems are operating in active parallel and sharing the total load/demand. One of
the system fails, the other can meet the demand but operating under stressed condition and hence
its failure rate will increase. The reliability of such load sharing system is obtained using state
transition diagram. In earlier analysis of parallel system the two components were operating in
parallel and sharing the demand, but when one of them fails the other meets the demand but
failure rate remains same. The state transition diagram of load sharing system is given as
follows. Assume λ# , λ% are the failure rates of each component while both are in operation.
λ# Îand λ% Î are failure rates of these components while only one of them is working. The four
state model of the load-sharing system is given as
Fig. 6.6
− λ# + λ% λ# λ% 0
0 3
[p!å p#å p%å p2å ] [pp p# p% p2 ] / 0 −λ% Î 0 λ% Î 2
/ 0 0 −λ# Î λ# Î 2
. 0 0 0 0 1
p!å − λ# + λ% pp
pp e- ѪÎѦ :
p#å λ# pp − λ% Î p#
λ#
mK + λ% Î np#
K + λ# + λ%
λ#
p# s
K + λ# + λ% mK + λ% Î n
λ# λ#
4 4
λ% Î − λ# − λ% λ% Î − λ# − λ%
P# s −
K + λ# + λ% mK + λ% Î n
λ#
p# t [e-Ѧ − e- ѪÎѦ : ]
ß
:
λ# + λ% − λ% Î
p%å λ% pp − λ# Î p%
λ%
P% s
mK + λ# n K + λ# + λ%
Î
λ#
p% t [e-Ѫ − e- ѪÎѦ : ]
ß
:
λ# + λ% − λ# Î
Reliability ( ) = pp + p# + p%
λ# λ%
e- + e-Ѧ − e- ѪÎѦ [e-Ѫ
ß ß
Ѫ ÎѦ : : : :
+
λ# + λ% − λ% Î
(λ# + λ% − λ# Î
− e- Ѫ ÎѦ :
]
Note: if λ# Î λ# , λ% Î = λ%
λ# = λ% = λ
λ# Î = λ% Î = λÎ
2λ
( ) = e-%Ñ: + ¯e-Ñ : − e-%Ñ: °
ß
(2λ − λ )
Î
1
( )= ¯2λe-Ñ : − λÎ e-%Ñ: °
ß
(2λ − λÎ )
If λÎ = 2λ
Then show
( ) = (1 + 2 ) e-%Ñ:
Exercise # 1
Ex. 1. A standby system consists of one main component and one standby having failure rate λ#
= 0.4, λ% = 0.2. Probability of successful change over is 0.95. Failure rate of switch is 0.01 per
year. Also the standby component’s failure rate during idle period is 0.05 per year. Obtain
reliability for one year.
Ex. 2 A non-redundant system consists of 50 components each having failure rate 0.01 per year.
Obtain
Ex. 3 Three components are operating in parallel and sharing a load. Each having same failure
rate λ = 0.2 per year. If one of the components fails the remaining two component can meet the
demand but then each one has failure rate 0.3. If again one component fails the remaining one
component can meet the demand but then failure rate become 0.4 per year. Obtain reliability for
one year.
Chapter – 7
7.1 General
Concept of reliability function discussed in previous chapters is of great significance for non-
repairable or mission oriented systems. A repairable system once fails repaired and then again is
ready for service. In such situation one is interested in knowing the percentage of time the
system is in upstate and failure frequency of the system. Usually constant failure rate is assumed
in such analysis. Since repair time of the system is also a random variable. Repair rate of the
system is also treated as a constant i.e. repair density function is also assumed as exponential.
Usually it is expected that large number of failures in a system occurs which requires small
repair time whereas few failure will occur which requires long repair time. Hence plot of
estimate of repair density function will look as shown in fig 7.1.
Fig 7.1
To determine availability function Markov process is utilized. Precisely a repair density function
is defined as
[ ≤ ≤ + ∆ V]
lim
V V V
V V
∆c5 →p ∆V
!V ( V , ∆ V )
≈
"V ∆ V
J( + ∆ ) = 7
4 ()
J = , J( − ∆ ) = €, J( − 2∆ ) = …
X(t + ∆ ) = j
4()
=P X t =i …(7.1)
= p (t, ∆t)
It is obvious that in a Markov process ,the process (probabilistic) a state ‘j’ at (t + ∆t) depend on
the state occupied by the process at t i.e. ‘i’ not on the earlier states. In this sense a Markov
process is memory less process. The conditional probabilities in (1) i.e. p (t, ∆t) are known as
transitional probabilities. Further if transitional probabilities are independent of ‘t’ then the
Markov process is known as homogeneous/stationary process. In view of the above eqn.(7.1) is
written as
J( + ∆ ) = 7
4 ) = e +∆
J( = …(7.2)
e + may be defined as transition rate from state i to j. In terms of reliability as shall be made clear
this may be either a failure or repair rate provided both are having exponential probability
density functions. Consider the general state transition diagram shown below
Fig 7.2
Let us assume p# t , p% (t), . . . p (t) are state probabilities at time ‘t’. General expression for
p (t + ∆t) [probability of system in i:; state at (t+∆t)] is given as
Defining
p t + ∆t − p t
) qÒ pÒ t
∆t
ÒZ#
Taking limit as ∆ → 0
på; = ) qÒ pÒ t … (7.4)
ÒZ#
på (t) = p A …(7.5)
Off- diagonal
=e+
This represents a discrete Markov process or chain and represented in matrix form as
p (r+1) = p r P …(7.6)
) pÒ = 1 … (7.8)
Ex 1. Consider following two state continuous Markov process model. Obtain steady state
probabilities.
Fig 7.3
−0.2 p# + 2 p% = 0
0.2 p# − 2 p% = 0
These two equations are not independent. Hence one of them is considered.
0.2 p# − 2 p% = 0
p# + p% 1
2 10
p# =
2.2 11
0.2 1
p% = =
2.2 11
Put b = 2
2
p (3) = p (2) A′ = p (0) A′
The general solution is written as
C
p (r) = p 0 A′
C. Steady state solution of discrete Markov process
p (r+1) = p r P
p (r+1) = p r P = <(
p =p P
p [ I –P ] = 0
D. Complete solution of discrete Markov chain is given as
p (r) = p (0) PC
Fig 7.4
Matrix equation
K ç −0.1 0.1
[K ý − ] = −
0 K 10 −10
s + 0.1 −0.1
=
−10 s + 10
s + 0.1 −0.1 -#
[K ý − ]-#
−10 s + 10
1 K + 10 .1
K(K + 10.1) 10 K + .1
1 K + 10 .1
[<# K <% K ] [<# 0 <% 0 ]
K(K + 10.1) 10 K + .1
K + 10
<# (K) =
K(K + 10.1)
10 0.1
<# ( ) = + -#p.#c
10.1 10.1
0.1
<% (K) =
K(K + 10.1)
0.1 0.1
<% ( ) = − -#p.#c
10.1 10.1
Consider a single component having •!~ ] as failure and repair rate respectively. The state
transition diagram of the system is shown in fig below
Fig. 7.5
The state -1 represents ‘up’ and state -2 represents ‘down’ state. The state space equations are
given in matrix form as follows
K ç −λ λ
RK ý − S = −— ˜
0 K μ −μ
s + λ −λ
=— ˜
−μ s + μ
s+λ −λ -#
RK ý − S-# = — ˜
−μ s+μ
1 s+μ λ
= — ˜
K(K + ] + ) μ s+λ
1 s+μ λ
R<# (K) <% (K)S = R1 0S — ˜
K(K + ] + ) μ s+λ
K+]
<# (K) =
K(K + ] + )
<% (K) =
K(K + ] + )
K+]
<# ( ) = E-# R S
K(K + ] + )
]
= + -(¡Îq)c
+] +]
<% ( ) = E-# — ˜
K(K + ] + )
= − -(¡Îq)c
+] +]
The state probability <# ( ) is known as availability and <% ( ) is known as unavailability. Fig. 7.6
shows the plot of these functions.
Fig. 7.6
It is observed the both availability and unavailability approaches to steady state values as given
]
<#ÂÂ , <%ÂÂ =
+] +]
The same figure also shows the plot of reliability which is exponentially decreasing which does
not consider repair. No path of return from absorbing states.
Assume that two components ‘a’ and ‘b’ has failure and repair rate as λî , λ and μî , μ .
Fig 7.7
State probabilities may be easily evaluated by using the method of discretization. It requires a
suitable choice of ‘∆ ’. If one does not have experience on the system initially any value of ∆
may be selected and state probabilities trajectories are obtained. Then a smaller value of ∆ is
selected and then again state probabilities are obtained. The difference in two calculated state
probabilities showed be less than a tolerance value specified. Otherwise further reduce ∆ and
repeat calculation of state probabilities till convergence.
State probabilities in long run are evaluated by solving following algebraic equations.
− λî + λ λî λ 0
μî −(μî + λ ) 0 λ
[p# p% p2 p3 ]
μ −(λî + μ )
= [0 0 0 0]
0 λî
0 μ μî −(μî + μ )
p# + p% + p2 + p3 1
μî μ λî μ μî λ λî λ
D D D D
p# , p% = , p2 = , p3 =
= (λî + μî )(λ + μ )
W = • = , ]W = ]• = ]
The state transition diagram will reduce to as shown in fig 7.6 below.
Fig 7.8
−2λ 2λ 0
[p#å p%å p2å ] [p# p% p2 ] ù μ − λ + μ)
( λ ú
0 2μ −2μ
−2λ 2λ 0
[p# p% p2 ] ù μ −(λ + μ) λ ú = [0 0 0]
0 2μ −2μ
p# + p% + p2 1
Solution gives
μ% 2λμ λ%
p# , p% = , p2 =
λ+μ % (λ + μ)% (λ + μ)%
System availability is evaluated by applying system adequacy criterion. For example at least one
of the units is required for system success then system availability is given by
¶ = p# + p%
μ% + 2λμ
=
(λ + μ)%
Prof. L.D. Arya
Reliability Engineering 139
Unavailability is given as
λ%
=
¶Â
(λ + μ)%
Fig 7.9
If the units are identical then above state transition diagram reduces to
Fig 7.10
State equations are written and solutions for eight states (different components case) is written
(long run).
μî μ μ λî μ μ μî λ μ μî μ λ
p#
D D D D
, p% = , p2 = , p3 =
λî λ μ λî μ λ μî λ λ λî λ λ
D D D D
py = , pÏ = , p = , p¬ =
If the components are identical then states are identical then states are reduced from eight to four
as shown in fig 7.8.
μ2 3μ% λ 3λ% μ λ2
(λ + μ)2 (λ + μ)2 (λ + μ)2 (λ + μ)2
p# = , p% = , p2 = , p3 =
If at least two units are required from adequacy view point the system availability is given as
μ2 + 3μ% λ
(λ + μ)2
¶ = p# + p% =
3λ% μ + λ2
(λ + μ)2
¶ = p2 + p3 =
Ex. A power plant has three units each 100 MW. The failure rate and repair rate of each unit is
given as 0.4 and 9.6 per year respectively. Obtain capacity distribution of the plant.
Solution:
«.Ï
Availability of each unit = «.ÏÎp.3
= 0.96
p.3
Unavailability of each unit = = 0.04
«.ÏÎp.3
Fig 7.11
Assume a component has failure rate and repair rate as λ and µ respectively. MTTF and MTTR
# #
respectively. Life length (TH ) and time to repair (TC) are random variables
¡ q
are given as and
and each having exponential distribution function. Then cycle time i.e. sums of ‘TH’ and ‘TC’ is
also a random variable when the process of failure and repair comes in stages. More over there
two random variables are independent. Hence as in the case of standby system the cycle time
(T=TH + TC ) has Hypo-exponential distribution function and given as
]
”( ) = R -¡c
− -qc
S
]−
1 1
= +
]
Fig 7.12
Cycle time is also known as MTBF i.e. mean time between failures. Average failure frequency or
frequency is given as
1 ]
= =
1 1 +]
+]
]
=
+]
]
=ž Ÿ =ž Ÿ]
+] +]
Now consider general state transition diagram with state probabilities p# , p% , p2 , … p . further
assume t# , t % , t 2 , … t are time spent in these states respectively
e = )e +
+
Then frequency of i:; state i.e. state ‘i’ one side and remaining states on other side is given as
number of transitions per unit time from i:; state.
qt
T!
t
= ž Ÿq
T!
=p q
Individual frequency of a state is not of importance when large numbers of components are
involved in the systems. In such situation transitions per unit time (frequency) from a group of
states e.g. set of events which corresponds to system success is of significance. This frequency
corresponds to failure frequency. Consider a general state transition diagram as shown in figure
7.10. The total states are partitioned in two groups i.e. ‘w’ and ‘f’. The states in ‘w’ and ‘f’
correspond to working (up) and failed (down) state respectively. If system is in a state of ‘w’ set
the system is ‘up’ otherwise if system encountered a state in ‘f’ the system is in ‘down’ state.
Fig 7.13
In such situations transition outside the group are considered. Transitions among states in a
group are not considered. Hence frequency of transitions from ‘w’ to ‘f’ is written as
)< )e +
∈ +∈T
Frequency will be same as calculated as transitions from ‘f’ to ‘w’ and given
) <+ ) e+
+∈T ∈
Ex. A system consists of two components having μ# 9.6, μ% = 9.4, λ# = 0.4 and λ% = 0.6.
Solution:
Fig 7.14
μ# μ%
(λ# + μ# )(λ% +μ% )
p# = = 0.94 × 0.96
λ# μ%
(λ# + μ# )(λ% +μ% )
p% = = 0.04 × 0.94
μ# λ%
p2 0.96 × 0.06
λ# + μ# λ% +μ%
λ# λ%
p3 = = 0.04 × 0.06
(λ# + μ# )(λ% +μ% )
= 0.94 × 0.96
f% = (0.04 × 0.94)(μ# + λ% )
= 0.04 × 0.06 × 19
#% = p# λ% + p% λ% = λ% (p# + p% )
23 = p2 μ% + p3 μ% = μ% (p2 + p3 )
Note: #% = 23
A methodology has been summarized in following steps for calculating failure frequency, mean
up time (MUT), mean down time (MDT).
Step 1 – It is assumed that failure and repair rate of each component present in the system is
provided. Based on this rates draw state transition diagram.
<( = 0
)< 1
Step 3 – apply system adequacy criterion and partition the state space in two groups i.e. ‘w’ and
‘f’. Obtain steady state availability and unavailability as follows
= )< )e +
∈ +∈T
Step 5 – The general state transition diagram is represented as two states, one component
equivalent as follows
Fig 7.15
Where λI and μI is equivalent failure and repair rate. From above equivalent diagram frequency
is written as
< u ƒ = <•' ]ƒ
ƒ = a< u
, ]ƒ = a<•'
1 < u
= = = < u
1 <•'
= = = <•'
]
Ex. A system consists of three identical components having failure and repair rates as λ and µ
respectively. The system success requires at least two components in up state. Obtain (1)
(2) MUT (3) MDT.
Solution:
μ2
(λ + μ)2
p# = = 0.982
3μ% λ
(λ + μ)2
p% = = 3 × 0.98% × 0.02
3λ% μ
(λ + μ)2
p2 = = 3 × 0.98 × 0.02%
λ2
(λ + μ)2
p3 = = 0.022
μ% (μ + 3λ)
(λ + μ)2
< u = p# + p% =
λ% (3μ + λ)
(λ + μ)2
<•' = p2 + p3 =
# = <# × 0 + <% × 2
(λ + μ)2 1000
= =
6μ% λ% 6 × 9.8% × 0.2%
µ% (µ + 3λ)
=
6μ% λ%
λ% (3µ + λ)
=
6μ% λ%
<# = <% ]
]
<# <%
<# + <% 1
]
<% + <% = 1
<% =
+]
<# 1 − <% = 1 −
+]
]
=
+]
Ex. Consider another example one component which has three states. λ ,µ and r are failure, repair
and reinstallation rate. The state transition diagram is shown in fig 7.16.
Fig 7.16
State – 1 is up state. State – 2 is the state where the component has failed. State - 3 is an
additional state where the component has been repaired but not put to service. Reinstallation
crew is a different term and this time is also a random variable assumed as exponential
distribution.
Following two frequency balance equations are written with respect to state -1 and state – 2
λ
f# p# λ = p2 r or p2 p
r #
λ
f% = p% μ = p# λ or p% = p
μ #
p# + p% + p2 1
λ λ
p# + p# + p# = 1
μ r
μr
p#
μr + λr + λμ
λr
p%
μr + λr + λμ
μλ
p2
μr + λr + λμ
Note: as r → ∞
]
p# →
+]
p% →
+]
And p2 → 0
Exercise # 1
−3 3 0 0
10 −12 2 0
<( å <( , =
0 20 −21 1
0 0 30 −30
rate as follows λ# = 0.1, μ# = 9.8, λ% = 0.1, μ% = 9.9, λ2 = 0.3, μ2 = 9.7. Calculate failure
Ex 2. A power plant contains 3 units of 100, 200 and 50 MW. Each having failure and repair
frequency, MUT, MDT. If load on the plant is 150 MW. All rates are in per year.
Chapter 8
8.1 General
One important aspect of reliability evaluation concerns about meeting the load by a supply
system while the system is in operation. Depending on the type of system, such type of reliability
evaluation requires probability of capacity of the system more than the load or probability of
load exceeding the capacity. Important ingredients of such studies are suitable load model and
capacity model. Then probability theory is applied to merge these two models to evaluate
reliability or failure probability. Various loads may be encountered in practice. For electric
equipment voltage may be a load (stress). Depending on the magnitude of voltage applied the
probability of failure may vary. Depending upon the available generation capacity there is
associated risk of not meeting the load. In such case depending on the nature of study a suitable
model of load is required. Whereas generation capacity may be modeled using Markov analysis.
In many mechanical systems load may vary periodically. In many such situations reliability is
viewed as static and not as a function of time. In general load may be electrical, thermal,
mechanical or chemical. Load may be measured in terms of volts, power, degrees, kilograms,
operations per hour, Km per hour or any other unit. Let us assume that L and C are the random
variables representing Load and Capacity. Then depending upon the system and modeling aspect
both of them may be continuous random variable. Load and Capacity both may be discrete
random variable. One of them may be continuous and the other discrete random variable. It is
interesting many times to evaluate the probability density function of safety factors or surplus
capacity, both of them are functions of load and capacity.
= a
= −
Assume fl(l) as probability distribution function which is met by a constant capacity “c” as
shown in fig. 8.1 below:
Fig 8.1
From fig 8.1 reliability P[C>L] or P [L<C] is calculated as shaded area i.e. area left to vertical
line at C.
Ex:1 Assume load is uniformly distributed between [0.8, 1.2] and capacity is fixed at 1.0. Obtain
reliability.
Solution:
Fig 8.2
1
X = , 0.8 < < 1.2
0.4
Reliability is given as
#
1 1 − 0.8
=} = = 0.5
p.¬ 0.4 0.4
Assume fc(c) as probability density function of capacity, load as fixed L. The situation is
presented as in fig. 8.3
Fig 8.3
The reliability in this case is the area under fc(c) right to vertical line at ‘L’
[
P[C > E] = = } f (c)dc
Ex:2 The capacity of distribution substation is having Normal distribution N (100,100) and load
is fixed in an hourly operation 50 MW. Obtain reliability of the system.
Solution:
1 -#pp ¦
f (c) = e
-p.y
#p
√2π σ
σ = 10
[
1 -#pp ¦
R=} e dc
-p.y
#p
yp √2π 10
1
= Φ (∞) − Φ (−5) ≅ 1
[ -N¦
} e % dz
-y √2π
Assume fl(l) and fc(c) as probability density functions of load and capacity and arbitrary as
shown in fig. 8.4 as follows
Fig 8.4
Now if one considers an arbitrary value of capacity ‘c’. Then reliability as a function of ‘c’ is
written as
b } ~ … (8.2)
0
Now since ‘c’ itself is a random variable and having a probability density function fc(c). The
overall reliability may be calculated as expected value of r(c).
[
= } b( ) ( ) ~ … (8.3)
p
\Putting the value of r(c) in (3) from (2), the value of ‘R’ is obtained as
[
=} {} X( ~ } ( )~
p p
[
=} } X( ( )~ ~ … (8.4)
p p
One very important case observed in reliability is when load and capacity both are having
Normal distributions.
1 J-J̅
¦
fJ (l) = Nml,̅ _X n = RS
-p.y— ˜
%
e
√2π σJ
1
f (c) = N(c, _
- ¦
R
%) -p.y— ˜
= e
√2π σ
J-J̅
[
1 1
¦
- ¦
RS R
-p.y— ˜ -p.y— ˜
=}} e e dl dc
p √2π σJ √2π σ
p
1 J-J̅
¦ ¦
Î[ -
RS e R
-p.y— ˜ -p.y— ˜
2π σ σJ -[ -[
} } e dl dc
− ̅ − ̅
L= , =
_ _X
íb _X + ̅ ≤ _ L + ̅
¤ -U̅
Or ≤ ¤T
L+ ¤T
X ZS̅
Î[ X Yß— X ˜
1 S S
= } } e-p.y(V ÎW ) dy dx
¦ ¦
2π -[ Z[
Fig 8.5
-X ̅
The region of integration is right of plane of the line £
¤
¤T
L+ ¤T
¥.It is observed that x and
¤
y both varies. If one rotates x and y axis anticlockwise by an angle Ө = •! -#
¤T
. Then the
limit of integration will become constant (β) independent of ‘x’. Then evaluation of integral
becomes simple. This means following substitution is made
L′ cos Ө sin Ө L
— ˜ =
′ − sin Ө cos Ө
L cos Ө −sin Ө L′
— ˜
− sin Ө cos Ө ′
=
~L ~ = ~L’ ~ ’
x % + y % = x′% + y′%
_ − ̅
≤ L+À Á
_X _X
R -J̅
y′ cos Ө + x′sin Ө ≤ [x′ cos Ө − y′sin Ө] +
RS RS
After simplification
’≤ β
c − l̅
β =
“σJ % + σ %
1 Î[ ê
= } } -p.y(|é¦ Î¶é¦ )
~ ′ ~L′ … (8.5)
2¢ -[ Z[
1 Î[
1 ê
= } À } -p.y¶é¦
~ ′Á -p.y|é¦
~L′ … (5)
√2¢ -[ √2¢ Z[
Now the inner integral is independent of x’ and can be evaluated from the standard tables of
Normal distribution. This is recognized as Φ (β).
R= Φ (β)
It is obvious from above expression that reliability depends on (i) Mean value of load and
capacity (ii) Standard deviation of load and capacity. The positive difference between
̅ & l ̅ should be high for adequate reliability. In addition to this variances of load and capacity
should be small.
Ex: Load and capacity at a distribution substation are having Normal distribution:
fJ (l) = N (100,100)
f (c)= N (150,144)
Solution:
l =100 MW, σJ = 10
̅ = 150 MW, σ = 12
150 − 100 50
ã= = = 3.003
√144 + 100 √244
Let us define
±= = Safety factor.
J̅
σJ
^X = a ̅ = Co-efficient of variation of load.
l
σ
^ = ac = Co-efficient of variation of capacity.
c − l̅ ̅ ± − 1) (± − 1)
ã= = =
“σJ % + σ %
_ ^X% +
% “^X% + ^% ± %
^
% %
Assume l0,l1,l2……lk are the discrete load levels with probability distribution
p(l0),p(l1),p(l2)……p(lk). Capacity has discrete levels c0,c1,c2….cm with probability distribution
p(c0),p(c1),p(c2)….p(cm).Further assuming load and capacity states as independent. Following
combined state diagram may be obtained.
Total number of states are (k+1)(m+1). The probabilities of each state, say pij is calculated
p = p(c ) . p(l )
The states are partitioned in two groups one Positive Margin (PM) group and other Negative
Margin (NM) group. In PM group (ci ≥ lj) and in NM group each state has (ci < lj).The reliability
is obtained as
` = ) abc
b,c∈de
` = ) abc
b,c∈fe
Ex: A system load and capacity are modified as discrete random variable with pdf as follows
lj 50 100 150
P(lj) 0.4 0.3 0.3
cj 0 150 200
P(cj) 0.01 0.04 0.95
Solution:
NM GROUP
______________________________________________________________________________
PM GROUP
8.6 Capacity as continuous random variable and load as discrete random variable.
Assume that f c is the probability density function of capacity and p(li) for l0,l1,l2……lk is the
distribution of discrete load model. Then reliability for load level li is given as
< / ) = g
∞
( ) ~ for i=0,1,2,3….k.
. # + . % + ⋯….+ . •
)<£ a ¥.<
• [
) À} f c Á <
X‡
Ex: The capacity of a distribution sub-station is uniformly distributed [0.5,1]. The load levels and
its pdf is given as follows:
Solution;
#
1
< /0.6) = } ~ = 0.4/0.5 = 0.8
p.Ï 0.5
#
1 .2
<( /0.8) = } ~ = = 0.4
p.¬ 0.5 .5
<( / ) = } ~
0
Then reliability expression is written using total probability theorem as in previous section.
Ex: A power plant has 3 generators each of 100 MW each having availability 0.95. The load is
having Normal distribution with N(100,100).Obtain availability of the system.
Solution:
ci p(ci) p(ci)
0 0.053 0.000125
2pp J-#pp ¦
1 -p.y— ˜
%p
1 N¦
<m a300n = } e #p dl } e-% dz = 1 = Φ(20)
-[ √2π 10 -[ √2π 10
-#p i¦
<m a0n } e-¦ dz = Φ −10) = 0
#
√%h #p
-[
= <m a300n <(300) + <m a200n< 200 + <m a100n<(100) + <m a0n< 0
` = dNn > EP = dNnk > ok P. dNn‘ > o‘ P. dNnl > ol P … dNnm > om P
` = pk . p‘ . pl … pm
` = õ pb
b
Ex: The load on a system is exponentially distributed with parameters α.The capacity of the
system is ‘k’ a constant. The load is repeated periodically ‘n’ times obtain reliability expression.
Solution:
(-âX) •
= ¯− °p = 1− (-â•)
(-â•) '
= ¯1 − °
Ex: The capacity of a plant is uniformly distributed between [0.8,1.0S. Load on the plant is
constant as 0.9 pu. Obtain reliability for the 24 repeated operations .
Solution:
#
#
p.%
r=} dc = 0.5
p.«
= (0.5)24
Assume that number of occurrences of a load in a specified time ‘t’ is not known but given by
Poisson process as
( )' . (-¡c)
'( ) =
!!
Where Pn(t) denotes probability of occurrence n times load in time ‘t’. λ denotes average
frequency of load occurrences. Further reliability for n-occurrences of load for identical load and
capacity distributions is rn. Then overall reliability for n-load occurrences is written as
b' '( ) = (reliability with ‘n’ random levels) ×( probability of occurrence of ‘n’ loads).
= ) r P (t)
[
(Ñ:)‰ .ƒ (ZsÞ)
'Zp r
=∑[
'!
(b )'
= (-¡c)
)
!!
= (-¡c)
. (V¡c)
(-¡c)(#-V)
=
Exercise # 1
Ex 1: An electrical design engineer observes that fifty percent of lightning load (in terms of
volts) on a surge protective device are of magnitude more than 1000 V. Pdf of such load has
been estimated as
± Ú -ât
v≥0
Obtain:
(i) Ú
(iii) For what voltage the system should be designed if failure probability should be less
than or equal to 0.05.
-(Xaypp)Û
( 1−
Where l is in kg. Determine the capacity for which the cable must be designed if the
probability failure should not exceed 0.005.
X( Ú 0<Ú< ∞
-âX
1
( )= € ≤ ≤ 2€
€
=0 K èℎ b
Ex 5: A building is designed to withstand winds of speed up to 180 Kmph. Hurricane winds are
normally distributed with a mean 100 Kmph and standard deviation of 10 Kmph .occurrence of
Obtain:
Ex 6: A generating system has four units each of 100 MW. Each having failure and repair rate as
0.1 per year and 20 per year respectively. The load is modeled as discrete random variable as
References
1. Reliability, Quality and Safety for Engineers – B.S. Dillon, CRC press.
2. Introduction to Reliability Engineering – E.E. Lewis, John Wiley and Sons.
3. Reliability Evaluation of Engineering Systems: Concepts and Techniques – R. Billiton
and R. Allam
4. Probabilistic Reliability: An Engineering Approach – M.L. Shooman, McGraw Hill.