EE5712 Power System Reliability

:: Reliability Theory
Panida Jirutitijaroen
Addition rule
Multiplication rule
Conditional probability rule
Complementation rule
Failure distribution
Cumulative failure distribution
Survival function
Hazard function

REVIEW: PROBABILITY THEORY

 Addition Rule
A method of finding a probability of union of two
events.
P(E1 U E2) = P(E1) + P(E2) – P(E1 ∩ E2)

• If E1 and E2 are mutually exclusive, then

P(E1 U E2) = P(E1) + P(E2).

E1 E2 E1 E2
 Multiplication Rule
A method of finding probability of intersection of
two events.

P(E1 ∩ E2) = P(E1)×P(E2 | E1)

• If E1 and E2 are independent, then

P(E1 ∩ E2) = P(E1)×P(E2).
 Conditional Probability Rule
If an event E depends on a number of mutually
exclusive events Bj, then
P(E) = Σj [ P(E | Bj)×P(Bj) ]

B3
B1
E B5
B2 B4
 Complementation Rule
Probability of the set of outcomes that are not
included in an event.
P(Ē) = 1 – P(E)

• Example
Probability of success = 1 – Probability of failure
 X: Time to Failure

A generator start working at

time x = 0 G

Outcome: time to It can fail at any time, x ≥ 0

failure

Random Variable, X x≥0

 Failure Distribution

Probability that a component fails at time x.

Pr x  X  x  x 
f x   lim
x 0 x
Exponential distribution function

f x  0.5e0.5 x , x  0
0.5

Pr 2  X  3   0.5e 0.5 x dx
3

1 2 3 4 5
 Cumulative Failure Distribution
A probability that a component fails before
or at time x.
F  x   Pr  X  x    f t dt
x



Exponential distribution function

1 2 3 4 5
 Survival Function
A function that gives probability of a
component survival beyond time x.

R x   Pr  X  x   1  Pr  X  x   1  F  x 
• Time to failure of a component is a random variable,
X.
• Commonly used in reliability theory


R4  Pr  X  4   0.5e 0.5 x dx
0.5
4

1 2 3 4 5
 Hazard Function
A function that gives a rate at time x, at which
a component fails ( i.e. failure rate), given that
it has survived for time x.

• Denoted by Ф(x),
Probability of a component fails between time x and x+Δx given
that it has survived for time x

Pr  x  X  x  x | X  x 
 x   lim
x 0 x
 An Important Relationship
• From hazard function as Δx → 0,
  x x  Pr x  X  x  x | X  x 
• This gives probability of failure of a
component in interval (x, x+Δx).
 x 

Pr  x  X  x  x | X  x 

x x+ Δx
 Important Note
• Although , for simplicity, survival function and
hazard rate function have been described with
respect to a component failure, they apply to
any random variable.
• For example if the random variable is time to
repair, then Ф(x) represents the repair rate
 Exponential Distribution Function
• Non-negative continuous random variable
• Commonly used to represent up time of a
component
• Probability that a component will fail at time x
is,
f x  ex , x  0
• Mean time to failure (MTTF) is E X  
1

 Exponential Distribution Example

x1 x2 x3 x4 x5

x1  x2    xn
• Mean time to failure is 
n
• n = number of failures
1 n
 
MTTF Total observatio n time
• λ is called “Failure rate”.
 Constant Hazard Function
• If we assume that the up time of a component
is exponentially distributed, the failure rate is
constant.
f x  e  x
,x0 Rx  ex , x  0
f x 
 x    , x  0
R x 

• This is also called “memoryless property”

 Memoryless Property
• Observe a component up to time t. and a
component has not failed yet.
• Interested to find residual lifetime, Y = X-t.
FY  x   Pr Y  x | X  t   Pr  X  t  x | X  t 
• Distribution of a residual lifetime is independent
of the time that the component has been in
operations.
• This component does not age, only random
failure
Characteristics of Exponential Dist.

f x   e  x
Rx  ex  x   
 Triangle Relationship
Probability density function
f x 
f x 
dR x  

dx 
x
  t dt
  f t dt
 f t dt  x e 0 x

Rx  1 dRx   x 

Survival function Rx  dx Hazard function

x
  t dt

e 0
Reliability Theory
Reliability Measure
Simple reliability evaluation methods

OUTLINE
Reliability evaluation
Failure distribution
Survival function
Hazard rate function

RELIABILITY THEORY
 Reliability Evaluation
• Concern with time of a component/system to
fail.
• At time = 0, probability of failure = 0.
This looks like
cumulative
distribution
function!
t=0

• As time → ∞, probability of failure → 1.

 Cumulative Failure Distribution
• Cumulative failure distribution

t=0

• Measure of probability of failure as a function

of time
• Denote as Q(t).
 Survival Function
• More interested to know what is the survival
probability.
• Denoted by R(t), or equivalently, S(t)
• Probability of a component/system surviving
• Complement of probability of failure
R(t) = 1 – Q(t)
 Failure Probability Density Function
• Denoted by f(t).
• Recall: pdf= derivative (cumulative function)
dQ t 
f t  
dt f(t)

d 1  Rt  dR t 
f t   
t
dt dt
Qt    f t dt
0 t 
Q(t) R(t) = Survival Function
Rt   1   f t dt   f t dt 0 t time
0 t
 Hazard Rate Function
• Equivalently
– transition rate function, failure rate function,
repair rate function, force of mortality
• Definition:

Measure of the rate at which failure occur

• Compute as per unit to the number of

components
 Hazard Rate Calculation
• Denote by φ(t).
• Depends on
– number of failure in given time
– number of components exposed to failure

Number of failures per unit time

φ(t) =
Number of components exposed to failure
 Example
• N0 = number of component tested
• Ns(t) = number of component survived at t
• Nf(t) = number of failure at t

R(t) = Ns(t) Q(t) = Nf(t)

N0 N0
Number of failures per unit time
φ(t) =
Number of components exposed to failure

φ(t) = dNf(t)/dt
Ns(t)
 φ(t) VS. R(t)
• From f(t) = dQ(t)/dt,
f(t) = dNf(t)/dt
N0
φ(t) = dNf(t)/dt = N0 dNf(t)/dt = f(t)
Ns(t) Ns(t) N0 R(t)

• Hazard rate function is a conditional function

of failure density function.
 φ(t) VS. f(t)
• Failure density function
– Probability of failure in any period of time
• Hazard rate function
– Probability of failure in next period of time, given
that it has survived up to time t
• φ(t) equivalent to f(t) but covers only time up
to point of interest.
• Need to normalize back to unity for times up
to “t”.
 Example: Hazard Rate Calculation
• An equipment initially contain 1000 identical components has the
following history data in 100 hours.

Taken from “Reliability Evaluation of Engineering Systems: concepts and techniques” by Roy billinton and Ron Allan
Hazard Rate Function
 A Bath Tub Curve
• Typical hazard function of a component.
 t 

Burn-in period Useful life Wear-out period

• It is fairly common to assume constant transition

rates in reliability modeling.
 A Two-State Component
• Consider a two-state component
• λ(t) is a hazard function of the
up time, called failure rate.
• μ(t) is a hazard function of the λ(t)

down time, called repair rate. UP DOWN

• Generally, hazard function is μ(t)
called ‘transition rate’ in
reliability work.
Mean time to failure
Mean time to repair
Mean time between failure
System availability
Frequency of failure

RELIABILITY MEASURE
 Mean Time to Failure (MTTF)

Average time that the system is in good

working condition, also called mean up
time.

• If system spend Tsuccess hours in success states,

the mean up time is given by
MTTF = Tsuccess / nsuccess-to-failure
nsuccess-to-failure = number of transitions from success states
to failure states during T hours.
 A Two-State Example

UP

nUD nUD nUD

DOWN

Tup
Tdown

Mean time to failure (MTTF) = ∑ Tup

∑ nUD
 Mean Time to Repair (MTTR)

Average time that the system is in repair

condition, also called mean down time.

• If system spend Tfailure hours in failure states,

the mean down time is given by
MTTR = Tfailure / nfailure-to-success
nfailure-to-success = number of transitions from failure states to
success states during T hours.
 A Two-State Example

UP

nUD nUD nUD

DOWN

Tup
Tdown

Mean time to repair (MTTR) = ∑ Tdown

∑ nUD
 Mean Cycle Time

Average time between failures, also called

Mean time Between Failure (MTBF)

• If system has mean up time of MTTF, and

mean down time of MTTR, the mean cycle
time is
MTBF = MTTF + MTTR
 A Two-State Example

UP

nUD nUD nUD

DOWN

Tup
Tdown

Mean cycle time (MTBF) = ∑ Tup + ∑ Tdown

∑ nUD
 System Availability

Probability of being found in the success

states.

• If system has mean up time of MTTF, and

mean down time of MTTR, the mean cycle
time is MTTF + MTTR, system availability is
found from
A = MTTF/ (MTTF + MTTR)
 System Unavailability

Probability of being found in the failure

states.

• If system has mean up time of MTTF, and

mean down time of MTTR, the mean cycle
time is MTTF + MTTR, system unavailability is
found from
U = MTTR/ (MTTF + MTTR)
 Example
• A two-state system has the following history over
10 years,
– 20 transitions from up to down state
– 9 years spent in up state
• Find the followings
– Mean up time in days
– Mean down time in days
– Mean cycle time in days
– System availability
– System unavailability
 Other Measures
• Frequency of failure
• Equivalent failure rate
Introduction
Probability convolution
Failure mode and effects analysis
Series-Parallel network
Conditional Probability Approach
Cut-set or tie-set method

SIMPLE RELIABILITY EVALUATION

METHODS
 Introduction
• Simple system reliability analysis
• Use only probability rules
• Assumptions
– Known failure probability of each component
– independent failure
– Simple system configurations, series/parallel
• Solution: Failure probability (equivalently,
system unavailability)
 Probability Convolution Method
• Two independent random variables, X and Y
with probability density function
• Interest to find the distribution of a new
random variable Z = X + Y
• Continuous case:

z t   x  y t    x yt  d
• Discrete case: 


z n   x  y n   xm yn  m
m  
 Excess Generation at Load Point
• X: generating capacity distribution
• Y: load distribution
• Interest to find Z: Excess generation
Z=X–Y
• System loss is when Z < 0
• System unavailability is Pr{Z < 0}
• Need to find the distribution of Z
 Probability Convolution for Z
• Z = X – Y = X + Y’, where Y’ = -Y
• Continuous case:
 
zt   x  yt    x yt  d   x y  t d
 
• Discrete case:

z n   x  yn   xmym  n
m  
Excess Generation Distribution Example

• Interest to find excess capacity (E), given that the

generation capacity (G) and load (L) has the
following distribution

Capacity (MW) Probability Load (MW) Probability

0 0.000001 0 0.00
50 0.000297 50 0.20
100 0.029403 100 0.75
150 0.970299 150 0.05
 Convolution Example
n G: Capacity (MW) Probability L: Load (MW) Probability
0 0 0.000001 0 0.00
1 50 0.000297 50 0.20
2 100 0.029403 100 0.75
3 150 0.970299 150 0.05

• From z*n+ = ∑(all m) x[m]y[m-n]

• E[-150] = G[0]L[150]
• E[-100] = G[0]L[100] + G[50]L[150]
• E[-50] = G[0]L[50] + G[50]L[100] + G[100]L[150]
• E[0] = G[0]L[0] + G[50]L[50] + G[100]L[100] + G[150]L[150]
• E[50] = G[50]L[0] + G[100]L[150] + G[150]L[100]
• E[100] = G[100]L[0] + G[150]L[50]
• E[150] = G[150]L[0]
 Failure Mode and Effects Analysis
• Model the system
• Categorized into subsystems
• Define function of each system and its requirement
• Block diagram
• Examine all failure modes of a component
• For example, A circuit breaker
– ground faults
– failure to open
– undesired tripping
• Study the effects according to the failure
 You are the weakest link!!

"A chain is only as strong as its weakest link”

Does this mean that system reliability is
determined by the least reliable component in
the system?
 Series/Parallel Network
• Define by logical relationship between failure
of the network and failure of the components
• Usually consistent with system structure
• Not necessarily refer to physical connections
• Series network: the success of all components
causes the system to success
• Parallel network: the failure of all components
causes the system to fail
 Series System

A B

• Let PsA and PsB be the success probability of

component A and B
• System availability is
A = PsA × PsB
 Parallel System

• Let PfA and PfB be the failure probability of

component A and B
• System unavailability is
U = PfA × PfB
 Series/Parallel Example
2

3
1 5 8
4

6 7

• If all component has failure probability of 0.01,

evaluate system reliability in terms of availability
and unavailability
 Applications
Probability Convolution Series/Parallel Network
• Develop excess generating • Distribution systems
capacity distribution • Simple substation
• Single area analysis configuration
• Injected power modeling at • Components
each bus – Circuit breakers
– Transformers
– Transmission lines
 Limitations

2 3

 
1 3
1 4
5 5 6
2 4 7 10
8 9
Series or Parallel??
 Conditional Probability Approach
• Decompose a complex system into simpler
subsystems
• Each subsystem is disjoint event
• Use conditional probability rule to calculate
system failure probability
• Denote key component, X, the probability of
system failure is calculated from.
Pf = Pr{system fails | X fails} × Pr {X fails} +
Pr{system fails | X works} × Pr {X works}
 Conditional Probability Example
• Assume that each component has failure probability of 0.01
and component 5 is key component, calculate system
failure probability
1 3
5
5 is working 2 4 5 fails

1 3 1 3

2 4 2 4
subsystem A subsystem B
 Failure Probability Calculation
Pf = Pr{system fails | 5 works} × Pr {5 works} +
Pr{system fails | 5 fails} × Pr {5 fails}
• Pf = Pr{A fails} × 0.99 + Pr{B fails} × 0.01
 Limitations

2 3
1 3

 
1 4
5 5 6
2 4 7 10
8 9
Series or Parallel??
 Cut-Set and Tie-Set Method
• Evaluate reliability of a block diagram network
• Assume independent failure
• Definition
– Cut set
– Minimal cut set
– Tie set
– Minimal tie set
• Use probability rules to calculate system
availability
 Cut-Set

A set of components whose failure causes

system failure

1 3
5
2 4

Example (1,2), (1,2,3), (1,2,4), (1,2,5), (1,3,5) , (1,2,3,4) ,

(1,2,3,4,5) , …
 Minimal Cut-Set

A smallest set of components whose failure

causes system failure

1 3
5
2 4

Minimal cut-set = { (1,2), (3,4), (1,4,5) , (2,3,5) }

 Cut-Set Method
• Enumerate all minimal cut-set in the system.
• Failure of all components in a minimal cut-set
causes system failure.
• This implies parallel connections among these
components.
• Each minimal cut set causes system failure.
• This implies series connections among the
minimal cut sets.
• Draw equivalent system and use series/parallel
method to compute for system availability.
 Example
1 3
5
2 4
Minimal cut-set = { (1,2), (3,4), (1,4,5) , (2,3,5) }

1 3 1 2
4 3
2 4 5 5
 System Unavailability Calculation
• If the system has C1, …, Cn minimal cut sets,
system failure probability is found from
Pf = Pr,Ĉ1 U Ĉ2 U … U Ĉn},
Ĉ is an event that the cut-set fails
• These minimal cut sets are not disjoint,
Pf = Pr,Ĉ1 U Ĉ2 U … U Ĉn}
= ∑i Pr{Ĉi} - ∑i≠j Pr{Ĉi ∩ Ĉj- + ∑i≠j≠k Pr{Ĉi ∩ Ĉj ∩ Ĉk}
- … + (-1)ⁿ⁻1Pr,Ĉ1 ∩ … ∩ Ĉn}
 Probability Approximation
• Total number of terms in previous equation is
2ⁿ-1.
• Use Boole’s inequality,
Pf = Pr,Ĉ1 U Ĉ2 U … U Ĉn- ≤ ∑i Pr{Ĉi}
• Upper bound on failure probability
• Lower bound on success probability
 Tie-Set

A set of components whose function

causes system success

1 3
5
2 4

Example (1,3), (1,3,4), (1,3,5), (1,4,5), (2,4) , (2,3,4) ,

(1,2,3,4,5) , …
 Minimal Tie-Set

A smallest set of components whose

function causes system success

1 3
5
2 4

Minimal tie-set = { (1,3), (2,4), (1,4,5) , (2,3,5) }

 Tie-Set Method
• Enumerate all minimal tie-set in the system.
• Success of all components in a minimal tie-set
causes system success.
• This implies series connections among these
components.
• Each minimal tie set causes system success.
• This implies parallel connections among the
minimal tie sets.
• Draw equivalent system and use series/parallel
method to compute for system availability.
 Example
1 3
5
2 4
Minimal tie-set = { (1,3), (2,4), (1,4,5) , (2,3,5) }

1 3
2 4
1 4 5
2 3 5
 System Availability Calculation
• If the system has T1, …, Tn minimal tie-sets,
system success probability is found from
Ps = Pr{T1 U T2 U … U Tn},
T is an event that the tie-set is success
• These minimal tie sets are not disjoint,
Ps = Pr{T1 U T2 U … U Tn}
= ∑i Pr{Ti} - ∑i≠j Pr{Ti ∩ Tj- + ∑i≠j≠k Pr{Ti ∩ Tj ∩ Tk}
- … + (-1)ⁿ⁻1 Pr,T1 ∩ … ∩ Tn}
 Probability Approximation
• Total number of terms in previous equation is
2ⁿ-1.
• Use Boole’s inequality,
Ps = Pr{T1 U T2 U … U Tn- ≤ ∑i Pr{Ti}
• Upper bound on success probability
• Lower bound on failure probability
 Bounds on Probability Approximation

Cut-set Method Tie-set Method

• Compute for failure • Compute for success
probability probability
• Upper bound on failure • Upper bound on success
probability probability
• Lower bound on success • Lower bound on failure
probability probability
 Limitations

2 3
1 3

 
1 4
5 5 6
2 4 7 10
8 9
Series or Parallel??
 Summary
• Reliability theory
– Failure distribution
– Survival function
– Hazard rate function
• Reliability Measure
• Simple reliability analysis
– Probability convolution method
– Series/parallel system
– Conditional probability approach
– Cut-set/Tie-set method
• Input: failure probability of each component
• Output: failure probability of a system
 About Next Lecture
• Interest to know how often the system fails
– “Frequency of failure” is another reliability
measure.
• Need to study stochastic process
– Model stochastic behavior of a system
– Transition rate from one state to others
– From success to failure state called “Failure rate”
– From failure to success state called “Repair rate”