INTRODUCTION TO RELIABILITY

CALCULATION METHODS

CHANAN SINGH
DEPARTMENT OF ELECTRICAL & COMPUTER
ENGINEERING
TEXAS A&M UNIVERSITY
COLLEGE STATION, TX

May, 2006

1
 INTRODUCTION TO QUANTITATIVE
RELIABILITY
 RELIABILITY RELATES TO THE ABILITY OF A
SYSTEM TO PERFORM ITS INTENDED FUNCTION
 IN A QUALITATIVE SENSE, PLANNERS AND
DESIGNERS ARE ALWAYS CONCERNED WITH
RELIABILITY
 WHEN QUANTITATIVELY DEFINED, RELIABILITY
BECOMES A PARAMETER THAT CAN BE TRADED
OFF WITH OTHER PARAMETERS LIKE COST
 NECESSITY OF QUANTITATIVE RELIABILITY:
-EVER INCREASING COMPLEXITY OF SYSTEMS
-EVALUATION OF ALTERNATE DESIGNS
-COST COMPETITVENESS AND COST-BENEFIT
TRADE OFF

2
 MEASURES OF RELIABILITY

BASIC INDICES

 PROBAILITY OF FAILURE – LONG RUN

FRACTION OF TIME SYSTEM IS FAILED
 FREQUENCY OF FAILURE – EXPECTED OR
AVERAGE NUMBER OF TIMES PER UNIT TIME
 MEAN DURATION OF FAILURE – MEAN
DURATION OF A SINGLE FAILURE

OTHER INDICES CAN BE OBTAINED AS

FUNCTIONS OF THESE BASIC INDICES

3
COST VS RELIABILITY

Total Cost

Investment Cost

Failure Cost
To Customer

ROPT
4 Optimum
 SEQUENCE OF PRESENTATION

 DESCRIBE AN EXAMPLE THAT WILL BE USED TO

ILLUSTRATE CONCEPTS
 INTRODUCE ANALYTICAL METHODS OF
RELIABILITY EVALUATION
 INTRODUCE MONTE CARLO SIMULATION
METHODS OF RELIABILITY EVALUATION

5
 EXAMPLE SYSTEM

Generators Transmission Load

Figure 1: System diagram

Data__________________________________________________________________

Generators:
Each generator either has full capacity of 50 MW or 0 MW when failed.
Failure rate of each generator is 0.1/day and mean-repair-time is 12 hours.

Transmission Lines:
The failure rate of each transmission line is assumed to be 10 f/y during the
normal weather and 100 f/y during the adverse weather. The mean down
time is 8 hours. Capacity of each line is 100 MW.

Weather:
The weather fluctuates between normal and adverse state with mean duration
of normal state 200 hours and that of adverse state 6 hours.

Breakers:
Breakers are assumed perfectly reliable except that the pair B1&B2 or
B3&B4 may not open on fault on the transmission line with probability 0.1.

Load:
Load fluctuates between two states, 140 MW and 50 MW with mean
duration in each state of 8hr and 16hr respectively.

FOR THE DESCRIBED SYSTEM,

HOW CAN YOU CALCULATE THE FOLLOWING BASIC

RELIABILITY INDICES ?

1. Loss of load probability

2. Frequency of loss of load

6
3. Mean duration of loss of load

7
 METHODS OF QUANTITATIVE RELIABILITY
ANALYSIS

 ANALYTICAL METHODS

- STATE SPACE USING MARKOV PROCESSES

- MIN CUT SETS
- NETWORK REDUCTION

 MONTE CARLO SIMULATION

- RANDOM SAMPLING
- TIME SEQUENTIAL

8
 Markov Processes

Using

Frequency Balance Approach

9
 Concept of Transition Rate

Consider two system states i and j.

Transition rate from state i to j is the mean number of transitions from

state i to j per unit of time in state i.

If the system is observed for T hours and Ti hours are spent in state i,
then the transition rate from state i to j is given by

ij = nij / Ti

where

nij = number of transitions from state i to j during the

period of observation


1 2

 

 

3 4


Fig. 1 Four state model

10
Example 1 : A 2- state component

 
1 2

UP DN

Fig. 2 A 2-state model

Let UP state be #1 and Down state be #2.

Transition rate from up to down state = failure rate

= n 12 / T1

= 1 / ( T 1 / n12 )

= 1/ MUT

where

MUT= mean up time

Also

Transition rate from down to up state = repair rate

= n 21 / T2

= 1/( T 2 / n21 )
=1 / MDT
where
MDT = mean down time of the component

11
 Concept of Frequency

Frequency of encountering state j from state i is the expected (mean)

number of transitions from state i to state j per unit time.

Fr(ij) = steady state or average frequency of transition from state i to

j

= nij / T

= ( Ti / T) (nij / Ti )

= pi ij
where

pi = long term fraction of time spent in state i

= steady state probability of system state i

12
 State transition diagram
 
10MW 0 MW

UP DN
2 State Model for a 10MW Generator

1u  
 1 2u
30 MW 3u
 
  
1 D 1u 1 u
2 2u 3 2D 4 2u
20 MW 3 U 20 MW 3u 20 MW 3 D
    

    
< 10MW

1 D 1u 1 D
5 2D 7 2D 6 2u
10 MW 3 U 10 MW 3 D 10 MW 3 D

  

1 D
8 2D
  0 MW 3 D 

Fig 5 Model for three 2-state units

Concept of Frequency Balance

In steady state or average behavior,

Frequency of encountering a state (or a subset of states) equals the

frequency of exiting from the state (or the subset of states).
Example 2:

13
Consider the state transition diagram for three 2-state units.

Equation for state 2 can be written as,

p1  1 +p 5  2 + p 6 3 = p 2 ( 1 + 2 +  3 )
Calculation of state probabilities:

If components are independent, system state probabilities can be found

by the product of unit state probabilities.
If components are not independent then

- write an equation for each of n system states using frequency

balance.

- any n-1 equations together with

 p
n
i 1
1
i

can be solved to find state probabilities.

EQUATIONS ARRANGED IN MATRIX FORM :

THE STATE PROBABILITIES CAN BE OBTAINED BY

SOLVING
 
BP = C
 
WHERE
B : MATRIX OBTAINED FROM THE TRANSPOSE OF
TRANSITION RATE MATRIX R BY REPLACING THE
ELEMENTS OF AN ARBITRARILY SELECTED ROW k
BY 1s
R : MATRIX OF TRANSITION RATES SUCH THAT ITS
ELEMENT ri j = λi j
λij : CONSTANT TRANSITION RATE FROM STATE i TO j

14
P : COLUMN VECTOR WHOSE iTH TERM pi IS THE
STEADYSTATE PROBABILITY OF THE SYSTEM BEING
IN STATE i
C COLUMN VECTOR WITH kTH ELEMENT EQUAL
TO ONE AND OTHER ELEMENTS SET TO ZERO.

15
Other indices:

Mean cycle time of an event (MCT) = 1 / frequency of the event

Mean duration of the event = MCT x Prob. of the event

Thus

Mean cycle time between failures = 1/ freq of failure

Mean down time (MDT) = prob of failure / freq of failure

Mean up time = MCT - MDT

= prob of system up / freq of failure

16
 Frequency of a Set of States

X Y
j
i
S

j
i S

Y
X Disjoint Subsets

Fig 4 Concept of set frequency

Frequency of encountering subset Y from subset X,

Fr(X  Y) =  pi  ij
iX jY

Therefore freq of encountering subset X,

Fr(X) =  pi  ij
i(S-X) jX

To find the frequency of a subset of states:

1. Draw boundary around the subset of states.

17
2. Find the expected transition rate into the boundary or out of the
boundary.

Example:

For the case of three 2-state units,

Fr(capacity  10) = p2 ( 2 + 3 ) + p3 ( 1 + 3 ) + p4 (1 + 2 )

=p5(  1 +  2 ) +p7 ( 2 +3) + p6 (1 + 3)

This frequency is typically called cumulative frequency.

18
 Equivalent Transition Rate

The equivalent transition rate from X to Y in Fig 4 is given by

XY = Fr(XY) / Prob(X)

Solving Example Problem using Markov Approach1

1
Calculations provided by Maja Knezev

19
1. System State Description & Equivalents

The first task is to obtain probabilities for the generators, transmission lines and
loads, which are independent parts of the system.
1. 1. Generators

Up State  Down
50 MW  State
0 MW
Figure 2: Each Generator has two possible states

1
  730 / year
  0.1 / day  36.5 / year 12 (1)
8760

Figure 3: State Transition Diagram – Generator System

20
Merging Identical Capacity States

State 2 State 3
State 1 12  23  34 State 4
Two UP, one DN One UP, two DN  43
All UP -150MW  21  32 All DN - 0MW
100MW 50MW

Figure 4: Equivalent State Transition Diagram – Generator System

Equivalent transition rates:

P1  P1  P1

12G   3
P1
P2   P3   P4 
 21G  
P2  P3  P4
2 P2   2 P3   2 P4 
 23G   2
P2  P3  P4
(2)
2 P   2 P6   2 P7 
 32G  5  2
P5  P6  P7
P5   P6   P7 
34G  
P5  P6  P7
P8   P8   P8 
 43G   3
P8

Transition rate matrix is:

 3 3 0 0 
   (  2) 2 0 
RG   (3)
 0 2  ( 2   )  
 
 0 0 3  3 

If we substitute values for  and  obtained in (1) into the matrix (3), transition rate matrix
for the generator system is:

 109.5 109.5 0 0 
 730  803 73 0 
RG  (4)
 0 1460  1496.5 36.5 
 
 0 0 2190  2190

1. 2. Transmission Lines

21
 Up State  Down
100 MW  State
0 MW

Figure 5: Each Transmission line has two possible states

During the normal weather   10 / year ,

During the adverse weather    100 / year
1
   1095 / year
8 (5)
8760

If all the breakers are perfectly reliable, for the two-transmission-line system, there will be
4 states.

 State 2 
1D 2U
 100 MW 
State 1 State 4
1U 2U 1D 2D
200 MW 0 MW

State 3 
1U 2D
 100 MW 

Figure 6: Four State Transition Diagram – Transmission System

If breakers may not open on command:

22
 State 5 0.1λ 0.9λ State 2 
1D 2U   1D 2U
0 MW   100 MW 
State 1 State 4
1U 2U 1D 2D
200 MW 0 MW

State 6  State 3 
1U 2D 1U 2D
0 MW 0.1λ 0.9λ 100 MW

 
Figure 7: Six State Transition Diagram – Transmission System

Merging of states:

State 1 State 2 State 3 State 4

12  23  34
one UP one DN  21 two UP  32
one UP one DN  43 two DN
0MW 200MW 100MW 0MW

Figure 8: Equivalent Four State Transition Diagram – Transmission System

23
Equivalent transition rates:

P5   P6 
 12  
P5  P6

P1 (0.1  0.1 )
 21   0.2
P1

P1 (0.9  0.9 )
 23   1.8
P1

P2   P3 
 32  
P2  P3

P2   P3 
 34  
P2  P3

P4 (  )
 43   2 (6)
P4

1.2.1 Weather

State 4  State 1  State 2 2μ State 3

1 UP, 1 2 UP 1 UP, 1  2
DOWN 0.2λ 200 1.8λ DOWN  DOWN
0 MW  MW  100 MW 0 MW
S S S S

N N N N
State 8  State 5  State 6 2μ State 7
1 UP, 1 2 UP 1 UP, 1  2
DOWN 0.2λ́ 200 1.8λ́ DOWN ’ DOWN
0 MW ’’” MW ’ 100 MW 0 MW
’’
’’ Eight State Transition Diagram – Transmission System
Figure 9: Equivalent

24
 1
N  43.8 / year
Transition rate from normal weather to adverse weather is: 200
8760

1
S  1460 / year
Transition rate from adverse weather to normal weather is: 6
8760

Transition rate matrix of transmission system is:

 - ( 2  N
 

 0

 
R 
T  S

 0
 0


 0

(7)

1.3 Load

State 1 1 State 2
140 MW 21 50 MW
2

25
 Figure 10: State Transition Diagram – Load
1
12   1095 / year
8
8760

1
 21   547.5 / year
16
8760

Transition rate matrix:

  1095 1095 
RL  
 547.5 (8)
 547.5 

26
2 Steady State Probabilities, Frequency and Mean Duration of Loss of Load

2. 1. Generation System

In order to get the steady probability of each state, we can write:

 P1G  0
  0
T  P2 G 
4
RG  
 P 3G  0 P
i 1
iG 1 (9)
   
 P4 G  0

Using the RG we obtained in equation (4), solving equations (9), we get the steady state
probability of each state.

If generators are independent probabilities can be calculated by product rule also.

Probabilities calculated in either way are the same.


Pu   0.95238


Pd   0.047619


P1G= Pu * Pu * Pu =0.8638377 (10)

P1G= 3*Pu * Pu * Pd =0.1295725
P1G= 3*Pu * Pd * Pd =0.00647876
P1G= Pd * Pd * Pd =0.000107979

2. 2. Transmission System

We have the following equations:

 P1  0
P   
 2  0 
 P 3  0 
 P  0  8
T 4
RT  
 P5  0 P
i 1
i 1 (11)
 P  0 
 6  
 P7  0
  0 
 P8   

27
Using the RT we obtained in equation (7), solving equations (11), we get the steady state
probability of each state:

RtT =

1.0e+003 *

-0.0638 1.0950 0 1.0950 1.4600 0 0 0

0.0180 -1.1488 2.1900 0 0 1.4600 0 0
0 0.0100 -2.2338 0 0 0 1.4600 0
0.0020 0 0 -1.1388 0 0 0 1.4600
0.0438 0 0 0 -1.6600 1.0950 0 1.0950
0 0.0438 0 0 0.1800 -2.6550 2.1900 0
0 0 0.0438 0 0 0.1000 -3.6500 0
0 0 0 0.0438 0.0200 0 0 -2.5550

P1=0.9507726 (12)
P2=0.01787034
P3=0.0001196528
P4=0.0019820304
P5=0.02678843
P6=0.002162378
P7=0.000060686
P8=0.00024383

We can also reduce the eight-state transmission transition diagram given in Fig 11 to a
three-state diagram with respect to the capacities of the states:

State 1 21T 32T State 3

State 2
200MW 12T 100MW 23T 0MW

31T 13T

Figure 11: Equivalent Three State Transition Diagram – Transmission System

For the reduced model, the following results apply:

P1T  P1'  P5'  0.9507726  0.02678843=0.97756103

P2T  P2'  P6'  0.01787034  0.002162378=0.020032718
P3T  P3'  P4'  P7'  P8'  0.0001196528  0.0019820304+0.000060686+0.00024383
P3T  P3'  P4'  P7'  P8'  0.0024061992

28
 P1'  1.8  P5'  1.8' 0.9507726  18  0.02678843  180
12T    22.439339
P1'  P5' 0.97756103

P2'   P6' 
 21T     10
P2'  P6'

P2'   P6' ' 0.01787034 10  0.002162378 100

23T    19.714808
P2'  P6' 0.020032718

P3'  2   P7'  2
 32T   1095
P3'  P7'  P4'  P8'

P1'  0.2  P5'  0.2' 0.9507726 2  0.02678843 20

13T    2.493259
P1'  P5' 0.97756103

P4'   P8' 
 31T  '  2090 (13)
P3  P7'  P4'  P8'

2. 3. Load

The following equations apply:

 P  0  2
R TL  1L    
P2 L  0
P i 1
iL 1 (14)

Using the RL we obtained in equation (8), solving equations (14), we get the steady state
probability in each state:

 P1L  0.3333333
 P   0.6666667
 2L   

29
2. 4. Solution for the System

Steady state probability, frequency and mean time of loss of load could be found using the
following table:

P1G = 0.8638377 P1T = 0.97756103 P 1L = 0.3333333

P2G = 0.1295725 P2T = 0.020032718 P2L = 0.6666667
P3G = 0.00647876 P3T = 0.0024061992
P4G = 0.000107979

System Generation, Probability of Transition to the states Loss of

State transmission, load system state with loss of load load
system state
1 111 0.281485 2,3,4 No
( 12T 13T ,  12 G )
 , 
2 121 Yes
3 131 Yes
4 211 Yes
5 221 Yes
6 231 Yes
7 311 Yes
8 321 Yes
9 331 Yes
10 411 Yes
11 421 Yes
12 431 Yes
13 112 0.562969 15(  13T ) No
14 122 0.0115367 2,15 No
( 21L ,  23T )

15 132 Yes
16 212 0.0844453 4,18 No
(  21L , 13T )
17 222 0.0017305 5,18 No
( 21L ,  23T )

18 232 Yes
19 312 0.00422226 7,21,22, No
(  21L , 13T , 34G )
20 322 0.000086525 8,21,23 No
(  21L ,  23T , 34G )
21 332 Yes
22 412 Yes
23 422 Yes
24 432 Yes

We can calculate the probability of states having no load loss. Those probabilities are
obtained for the generators, transmission lines and loads as independent.

30
From Table 1, we can get the steady state probability of the loss of load as follows.

P=1-
(0.281485+0.562969+0.0115367+0.0844453+0.0017305+0.00422226+0.000086525)
P=0.053524715 (15)

The frequency of loss of load is:

F   P  
i ij  P1 ( 12T   13T   12 G )  P13  13T  P14 ( 21L   23T )
i X j X

(16)
 P16 (21L  13T )  P17 (21L  23T )  P19 (21L  13T  34G )  P20 (21L  23T  34G )

F = 95.742635 /year

Values needed for F that are calculated previously:

12T  22.439339  23T  19.714808 13T  2.493259
12 G  3  109.5
 23G  2  73  21L  547.5 / year
34 G    36.5

The mean time of loss of load is:

P 0.053524715
MD    4.89726hours (17)
F 95.742635 / year

31
 -Cut Set Method
 A cut set is a set of components or conditions that cause system
failure.

 A min cut set is a cut set that does not contain any cut set as
a subset.

 In this presentation a cut set implies a min cut set.

 The term component will be used to indicate both a physical

component as well as a condition.

 Components in a given cut set are in parallel, as they all need

to fail to cause system failure.

 Cut sets are in series as any cut set can cause system failure.

32
Frequency & Duration Equations For Cut Sets

First Order Cut Set: One component involved

 csk  i

rcsk  ri

where

i , ri  Failure rate and mean duration of component i

csk ,r csk  Failure rate and mean duration of cut set k that contains

component i

Second Order Cut Set k:Two components involved

i  j (ri  rj )
csk 
1  i ri   j rj

ri rj
rcsk 
ri  rj

where
i ,  j  Failure rates of components i and j comprising cut set k
ri , rj  Mean failure durations of components i and j comprising
cut set k
Second Order Cut Set with Components subject to
Normal and Adverse Weather.
i , i  Failure
rate of component i in the normal and adverse
weather
N , S  Mean duration of normal and adverse weather
csk  a  b  c  d
N   Nr   Nr
a  ( i j i  j i j)
N  S N  ri N  rj

33
 N r Sri r Srj
b  (i i  j   j j i )
NS N S  ri N S  rj
S Nri Nrj
c  (i j   j i )
N S N  ri N  rj
S Sri Srj
d  (i j   j i )
NS S  ri S  rj
where
  Component due to both failures occurring during normal
a

weather
  Initial failure in normal weather, second failure in adverse
b

weather
  Initial failure in adverse weather, second failure in normal
c

weather
  Both failures during adverse weather.
d

Combining n Cut Sets

T  cs1  cs 2    csn

rT  (cs1rcs1  cs 2 rcs 2    csn rcsn ) / T

34
 APPLICATION OF CUT SET METHOD TO
EXAMPLE SYSTEM
Cut set 1: One line failure and breaker stuck.
N   S
av   12.621
N S
cs1  2  av  0.1  2.524 f / y
rcs1  8hr

Cut set 2: One generator failure and load changes from

50 to 140
g  0.1 / day  36.5 / year
8760
load   547.5 / year
16
rg  12hr  .00137 yr
rload  8hr  .000913 yr
 g load (rg  rload )
cs 2   88.306 / yr
(1   g rg  load rload )

35
 rg rload
rcs 2   4.8hr
rg  rload

Cut set 3:One line failure (breaker not stuck) and load
changes from 50 to 140
l  av  0.9
rl  8hr

l load (rl  rload )

cs 3   15.03 / yr
(1  l rl  load rload )
rl rload
rcs 3   4hr
rl  rload
Cut set 4: Two lines fail(breaker not stuck)

For each line

  10  .9  9 / yr
   100  .9  90 / yr
r  8hr  .000913 yr
N  200hr  .022831yr
S  6hr  .000685 yr

Applying the equation for second order cut set exposed

to fluctuating environment,
cs 4  .3888 / yr
rcs 4  4hr.

For the system

T  cs1  cs 2  cs 3  cs 4  106.25 / yr
(cs1rcs1  cs 2 rcs 2  cs 3 rcs 3  cs 4 rcs 4 )
rT   4.76hr
T
1
T 
rT
T
Frequency of failure = T  100.45 / yr
T  T

36
 T
Probability of failure = T   T = .0546

37
Monte Carlo Simulation

38
 Introduction

 The Monte Carlo method mimics the failure and repair history of
components and the system by using the probability distributions
of component states.
 Statistics are collected and indices estimated by statistical
inference.
 Two main approaches: random sampling , sequential simulation.

Random number generator:

Each number should have equal probability of taking on any one of the
possible values and it must be statistically independent of the other
numbers in the sequence. Random numbers in a given range follow a
uniform probability density function.

Multiplicative congruential method:

Random number Rn+1, can be obtained from Rn :

Rn+1 = (a Rn ) (modulo m)

where
a,m= positive integers, a < m.

Rn+1 is the remainder when (aRn) is divided by m.

suggested values
a=455 470 314
m=231 -1 = 2147 483 647
R0 = seed
= any integer between 1 and 2147 483 64

Range can be limited by truncation. For example if rn between 0 and

999 are required, the last three digits of the random number generated
can be picked up. If rn between 0 and 1 is required, put a decimal before
the rn generated.

39
 Random sampling
Sampling a component state:

Consider a component that has probability distribution:

State number Probability

(random variable)
1 .1
2 .2
3 .4
4 .2
5 .1

Let us assume that the random numbers lie in the range 0 to 1. We can
assign the random numbers proportional to their probability as follows:

Random State sampled

number drawn
0 to.1 1
.1+ to .3 2
.3+ to .7 3
.7+ to .9 4
.9+ to 1. 5

So if a rn is .56 ,then we say that state number 3 is sampled.

This procedure can be more simply carried out by using a cum prob or
probability distribution function. The prob mass function and
corresponding probability distribution function for this component are
shown below:

40
 1.0 1.0
0.9 0.9
0.8 0.8
0.7 0.7

P(X= x ) 0.6 0.6

0.5 0.5 F(x)=P(X < x)
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
Random Observation
0 0
1 2 3 4 5 1 2 3 4 5
x x
Now you can place the rn on the vertical axis and read the value of the
state sampled on the horizontal axis. It can be seen that this is equivalent
to proportional sampling.

Sampling a system state:

If a system consists of n independent components, then to sample a
system state, n random numbers will be needed to sample the state of
each component. For example for a system of two components with the
pdfs shown above, sampling may proceed as follows. Random numbers
used are found by a computer program and are shown on the next page.
RN for component 1 RN for comp 2 System state
.946 .601 (5,3)
.655 .671 (3,3)
.791 .333 (4,3)
.345 .532 (3,3)
.438 .087 (3,1)
.311 .693 (3,3)
.333 .918 (3,5)
.998 .209 (5,2)
.923 .883 (5,4)
.851 .135 (4,2)
.651 .034 (3,1)
.316 .525 (3,3)
.965 .427 (5,3)
.839 .434 (4,3)

41
Now if you want to estimate the probability of say state (3,3)
-
P(3,3) = n/N

where

n = number of times state sampled

N = total number of samples.

From the table

-
P(3,3) = 4/14

=2/7

=.286

The actual prob of (3,3) is

.4x.4 = .16

If this sampling and estimation are continued the estimated value will
come close to .16.

You can appreciate on of the problems of Monte Carlo that the indices
obtained are estimates. So one must have some criterion to decide
whether the indices have converged or not.

A Sequence of Random Numbers Generated

42
.946 .308
.601 .116
.655 .864
.671 .917
.791 .126
.332 .157
.345 .837
.532 .029
.438 .865
.087 .603
.311 .203
.693 .823
.333 .969
.918 .509
.998 .407
.209 .355
.923 .885
.883 .931
.851 .896
.135 .680
.651 .057
.034 .922
.316 .113
.525 .852
.965 .931
.427 .538
.839 .504
.434 .925

43
 Sequential Simulation
Sequential simulation can be performed either by advancing time in
fixed steps or by advancing to the next event.

Fixed time interval method:

This method is useful when using Markov chains,i.e., when transition

probabilities over a time step are defined. This is simulated using
basically the proportionate allocation technique described under
sampling. This is indeed sampling conditioned on a given state. This can
be illustrated by taking an example. Assume a two state system and the
probability of transiting from state to state over a single step given by
the following matrix:

Initial State Final State

0 1
0 0.3 0.7
1 0.4 0.6

Starting in state 0, select a rn and the next state is determined as follows:

digit event
0 to 0.3 stay in state 0
0.3+ to 1.0 transit to 1

Similarly starting in state 1,

digit event
0 to 0.4 transit to 0
0.4+ to 1.0 stay in 1

44
Construction of a realization for 10 steps:

Step RN State
1 .947 0
2 .601 1
3 .655 1
4 .671 1
5 .791 1
6 .333 1
7 .345 0
8 .531 1
9 .478 1
10 .087 1

Next event method:

This method is useful when the times in system states are defined using
continuous variables with pdfs. Keeping in mind that discrete rvs are
only a special case of continuous rvs ,the previously described method of
reading a variable from the pdf using a rn can be used. Consider for
example a rv X representing the up time of a component and
distribution

F(x) = P(X x)

as shown in the following figure. Now if a rn between 0 and 1. is drawn,

then the time to failure can be read as shown:

RN
F(x) = P(X < x)

Value of RV

RV X

45
Continuous distributions are approximated by discrete distributions
whose irregularly spaced points have equal probabilities. The accuracy
can be increased by increasing the number of intervals into which (0,1)
is divided. This requires additional data in the form of tables. Although
the method is quite general, its disadvantages are the great amount of
work required to develop tables and possible computer storage
problems. The following analytic inversion approach is simpler.

Let Z be a random number in the range 0 to 1 with a uniform

probability density function, i.e., a triangular distribution function:
0 Z  0


f ( z)  1 0  Z 1
0 Z  0



Similarly

0 Z  0


F ( z)  z 0  Z 1
1 Z  1



Let F(x) be the distribution from which the random observations are to
be generated. Let

z = F(x)

Solving the equation for x gives a random observation of X. That the

observations so generated do have F(x) as the probability distribution
can be shown as follows.

Let  be the inverse of F; then

x = (z)

Now x is the random observation generated. We determine its

probability distribution as follows

P(x  X) = P(F(x)  F(X))

= P(z F(X))
= F(X)

46
Therefore the distribution function of x is F(X), as required. In the case
of several important distributions, special techniques have been
developed for efficient random sampling.

In most studies, the distributions assumed for up and down times are
exponential. The exponential distribution has the following probability
distribution

P( X  x)  1  ex

where 1/ is the mean of the random variable X. Setting this function
equal to a random decimal number between 0 and 1,

z  1  e  x

Since the complement of such a random number is also a random

number, the above equation can as well be written as

z  ex

Taking the natural logarithm of both sides and simplifying, we get

ln( z)
x


which is the desired random observation from the exponential

distribution having 1/ as the mean.

This method is used to determine the time to the next transition for
every component, using  or  for , depending on whether the
component is UP or DOWN. The smallest of these times indicates the
most imminent event, and the corresponding component is assigned a
change of state. If this event also results in a change of status, (e.g.,
failure or restoration) of the system, then the corresponding system
indices are updated.

47
The procedure can be illustrated using an example of two components
with data as given below.
Component  (f/hr)  (rep/hr)
1 .01 .1
2 .005 .1

Time Random Number Time to change Compon

for Component ent state
1 2 1 2 1 2

0 .946 .601 5 101 U U

5 .655 - 0/4  96 D U
9 .670 - 0/40  92 U U
49 .790 - 0/2  52 D U
51 .332 - 0/110 50  U U
101 - .345 60 0/11  U D
112 - .531 49  0/127 U U
161 .437 - 0/8  78 D U
169 .087 - 0/244 70  U U
239 - .311 174 0/12  U D
251 - .693 162 0/73  U U
324 - .333 89 0/ U D


48
 indicates time causing change and is added to obtain the total time.

49
 REFERENCES

1. C. Singh, “Course Notes: Electrical Power System

Reliability”, http://www.ece.tamu.edu/People/bios/singh/
2. C. Singh, R. Billinton, “System Reliability Modelling and
Evaluation”, Hutchinson Educational, 1978, London
3. R. Billinton, R. Allen, “Reliability Evaluation of Power
Systems”, Plenum Press, 1984

50