Data analysis

Reliability and Risk Analysis

Master in Nuclear Science and Technology

Joint Master in Safe and Reliable Nuclear Applications (SARENA)
March 2023

D. Cuervo

RRA. 2022-23 Data analysis 1

 Data analysis

Outline

 Introduction and objectives

 Reliability functions
 Failure models
 Maintainability and availability
 Data bases
 Common cause failures

RRA. 2022-23 Data analysis 2

 Data analysis

Introduction

 Once the Fault Tree for the system/function has been

develop, probability of occurrence of the Basic Events is
needed
 This probability is established after the analysis of data
obtained when component reliability tests are performed
 The failure mode of the component can be modeled
statistically by different probability distributions
depending on the its characteristics
 Extended approach must be taken when maintenance is
included in the analysis

RRA. 2022-23 Data analysis 3

 Data analysis

Main objectives
 Know the reliability functions used to determine
probability of failure
 Develop the models for the data required by the Fault
Trees
 Study the different statistical models that may be
applied depending on the failure mode
 Understand the concept of repair and maintenance and
the relationship with availability of a component to
perform the work
 Know the models applicable to this type of components
 Understand the concept of Common Caused Failures and
know the main approaches to treat them

RRA. 2022-23 Data analysis 4

 Reliability functions

RRA. 2022-23 Data analysis 5

 Reliability functions

Probability of basic events

 To quantify the fault tree, it is necessary to know the
probability that a component does not perform its
function
 This is related to the availability of the component that
depends on:
 Failure mode of the component
 Discrete
– On demand
 Continuous
– In operation
– On stand-by
 Unavailability due to preventive maintenance and testing
 Unavailability due to corrective maintenance
 Common cause failures

RRA. 2022-23 Data analysis 6

 Reliability functions

Random variable Time to Failure

 The way to analyze the probability that a component is

not available at a given time is to use the random
variable Time to Failure
 This is:
• In the non repairable components: survival time of the
component
• In the repairable components: time elapsed from the temporal
origin to the first failure or from last repair to next failure

T
1
0
F t
RRA. 2022-23 Data analysis 7
 Reliability functions

Models
 The behavior of the random variable T may be adjusted
by statistical models that allow to predict the probability
that a component will not be available when needed
 Concepts of reliability, maintainability and availability
are basic to know the failure probability depending on
the operation and failure mode of the component
 There exist a wide variety of probability distributions

Some of them represent adequately the

reliability, maintainability and availability of
different types of components and failure modes

RRA. 2022-23 Data analysis 8

 Reliability functions

What is reliability?
 Reliability of a system (component, product, structure or
service) is the probability to perform its function without
failure, in conditions and time specified
 This definition includes this basic concepts
1. Satisfactory operation
2. In specified conditions
3. For a given period of time
4. Being expressed as a probability
 It may be applicable to:
• Repairable components : it may be more than one failure
• Unrepairable components : they are discarded after failure

RRA. 2022-23 Data analysis 9

 Reliability functions

Reliability functions

 To study the component reliability, several functions

may be used that relate to each other
 These are:
• Failure probability function or unreliability function
• Reliability function
• Failure density function
• Failure rate function or hazard function

RRA. 2022-23 Data analysis 10

 Reliability functions

Reliability function R(t)

 Probability of correct operation of a component in

stated conditions during a time t, considering that it
was at operating conditions at the beginning of the
process
R(t) = Pr(T > t) R(0 ) = 1 R(∞ ) = 0 T
1
0
1 F t

R(t)
0 t
RRA. 2022-23 Data analysis 11
 Reliability functions

Failure probability function or unreliability

function F(t)
 Cumulative distribution function for the random
variable T
 Probability of failure of a component in stated
conditions during a time t, considering that it was at
operating conditions at the beginning of the process
F(t) = Pr(T ≤ t) F (0) = 0 F (∞ ) = 1 R(t) + F(t) = 1

T
F(t) 1
0
F t
t
RRA. 2022-23 Data analysis 12
 Reliability functions

Failure density f(t)

 Probability density function (pdf) of the random

variable T
dF(t) dR(t) Pr(t < T ≤ t + ∆t )
f(t) = = − = lim
dt dt ∆t →0 ∆t

RRA. 2022-23 Data analysis 13

 Reliability functions

Failure density f(t)

 f(t)∆t is the unconditional probability that the

component will fail in the interval (t,t+ ∆t]
f(t) ∆t = Pr(t < T ≤ t + ∆t)
 It admits two interpretations:
 Probability per unit time that the component or system
experiences its first failure at time t, given that the
component or system was operating at time zero
 Expected ratio of failures in a small interval ∆t after t for a
population initially in operation

RRA. 2022-23 Data analysis 14

 Reliability functions

Failure rate or hazard function h(t)

 h(t)∆t is the conditional probability that the

component will fail in the interval (t,t+ ∆t], given that
it has survived until time t
 It admits two interpretations:
• Probability per unit time that the component experiences
its first failure at time t, given that the component or
system was operating at time zero and time t
• Expected ratio of failures in a small interval ∆t after t for a
population of components in operation initially and at time
t

RRA. 2022-23 Data analysis 15

 Reliability functions

Failure rate or hazard function h(t)

 h(t)∆t is the conditional probability that the

component will fail in the interval (t,t+ ∆t], given that
it has survived until time t
h(t) ∆t = Pr(t < T ≤ t + ∆t | T ≥ t)

Pr(t < T ≤ t + ∆t) f(t)

h(t) = lim =
∆t →0 Pr(T > t) ⋅ ∆t R(t)

f (t )
h(t ) =
R (t )

RRA. 2022-23 Data analysis 16

 Reliability functions

Cumulative failure rate or cumulative hazard

H(t)
 Number of expected failures in the temporal interval
(0,t)

t dH (t )
H (t ) = ∫ h(t ) dt ⇒ h(t ) =
0 dt

RRA. 2022-23 Data analysis 17

 Reliability functions

Relationship between reliability functions and

the failure rate

dR (t )
f (t ) d h(t) parameter to determine
h(t ) == − dt = ln [ R (t ) ]
R (t ) R (t ) dt at reliabilty studies
 t 
R (t ) = exp − ∫ h(t ) dt 
 0 
 t 
1 − exp  − ∫ h(t )dt 
F (t ) =
 0 
 t 
f (t ) = h(t )·R (t ) = h(t )·exp − ∫ h(t ) dt 
 0 
RRA. 2022-23 Data analysis 18
 Reliability functions

Example

 Calculate the values of the failure probability

functions at 600 hours of a batch of 2000
components knowing that 521 have failed up to this
moment and 18 will fail in the following 24 hours.
 Solution

RRA. 2022-23 Data analysis 19

 Reliability functions

Example

 Calculate the values of the failure probability

functions at 600 hours of a batch of 2000
components knowing that 521 have failed up to this
moment and 18 will fail in the following 24 hours.
 Solution
• Operating components at 600h: 1479
• R(600) = 1479/2000 = 0.74
• F(600) = 521/2000 = 0.26 (= 1-R(600))
• f(600) = 18/(2000*24) = 0.0004
• h(600) = 18/(1479*24)= 0.0005

RRA. 2022-23 Data analysis 20

 Reliability functions

Example

 Calculate the values of the failure probability

functions at 2400 hours of a batch of 2000
components knowing that 1402 have failed up to this
moment and 7 will fail in the following 24 hours.
 Solution

RRA. 2022-23 Data analysis 21

 Reliability functions

Example

 Calculate the values of the failure probability

functions at 2400 hours of a batch of 2000
components knowing that 1402 have failed up to this
moment and 7 will fail in the following 24 hours.
 Solution
• Operating components at 2400h: 598
• R(2400) = 598/2000 = 0.30
• F(2400) = 1402/2000 = 0.70
• f(2400) = 7/(2000*24) = 0.0002
• h(2400) = 7/(598*24)= 0.0005

RRA. 2022-23 Data analysis 22

 Reliability functions

Mean time to failure MTTF

 First moment of the failure probability density f(t) is

the expected time to failure of a component or
system
∞ ∞
MTTF = − t ⋅ R(t) ∞
0
+ ∫ R(t) dt
0
⇒ MTTF = ∫ R(t) dt
0

 It is a useful indication of the average life for a device

RRA. 2022-23 Data analysis 23

 Reliability functions

Bathtub curve

 Failure rates of many components along life follows

the classic “bathtub curve”

RRA. 2022-23 Data analysis 24

 Reliability functions

Bathtub curve

 Failure rates of many components along life follows

the classic “bathtub curve”

RRA. 2022-23 Data analysis 25

 Reliability functions

Bathtub curve description

 Infant mortality or debugging period

• Failures arising because of poor quality control practices
• Failure rate is decreasing
 Useful life period
• Failures occur at a rather uniform rate corresponding to
random failures
 Wear out period
• Failures arising from degradation of the component
• Failure rate is increasing

RRA. 2022-23 Data analysis 26

 Failure models

RRA. 2022-23 Data analysis 27

 Failure models

Reliability distributions

 Discrete distributions or time-independent

• Bernoulli trials
• Binomial distribution
 Life distributions or time dependent
• Exponential distribution
• Gamma distribution
• Weibull distribution
• Other

RRA. 2022-23 Data analysis 28

 Failure models

Bernoulli trials

 Used for a single device that operates on demand

 Probability of occurrence for an event (failure) with
two states (failure/operation) is p
 Unreliability
1 si x < 0

F(x
= ; p)  p si 0 ≤ x < 1
0 si x ≥ 1


RRA. 2022-23 Data analysis 29

 Failure models

Bernoulli trials

 Failure experience for selected MOV type over 10-yr

period

 Operating experience from plant records:

• 1430 demands of MOV to open
• 3 failures
 P = 3/1430 = 2·10-3

RRA. 2022-23 Data analysis 30

 Failure models

Binomial distribution
 Used for a single device that operates on demand and
can be repaired to an "as good as new" state
immediately after it fails
 It is based on Bernoulli trials with probability of
occurrence p constant from trial to trial
 Models the probability of failure on demand after n
demands
 Unreliability (probability of k
failures in n demands)
k
F(k ; n, p)= Pr( K ≤ k=
) ∑ p
i =0
i
(1 − p ) n −i

RRA. 2022-23 Data analysis 31

 Failure models

Exponential distribution
 Used in most of the reliability analysis to describe useful
life period
 Describes the time to failure in a Poisson process that is
the one in which the events happen in a continuous and
independent manner with a constant mean rate
• If a unit has survived to time t, its probability of survival that is,
its reliability, over an additional time period is the same
regardless of the present age t
 It is considered that the component is kept new up to
failure
 It is used for components that are not repaired or
maintained
RRA. 2022-23 Data analysis 32
 Failure models

Exponential distribution

 A constant failure rate defines an exponential

distribution, where t represents the time
h(t ) = λ λ ≥ 0, t ≥ 0
 Cumulative failure rate in period (0,t)
t
H (t ) = ∫ h(t )dt =λt
 Unreliability and reliability
0

= − H (t )
R(t ) e= e − λt t >0 1 − e − λt
F (t ) = t >0
 Failure density
dR(t)
f(t) = − λ, t ≥ 0 f(t) = λ e − λ t λ, t ≥ 0
dt
RRA. 2022-23 Data analysis 33
 Failure models

Exponential distribution

 Mean Time To Failure

∞
1
MTTF = ∫ R(t )dt = λ
o

 Rare event approximation (if λt << 1)

F (t ) ≈ λt

RRA. 2022-23 Data analysis 34

 Failure models

Exponential distribution

Failure rate Failure density

F ( t , 0.5) 0.8

F ( t , 1) 0.6

F ( t , 1.5)

F ( t , 3)
0.4 Unreliability
0.2

0
0 1 2 3 4 5
t

RRA. 2022-23 Data analysis 35

 Failure models

Gamma distribution
 It is especially appropriate for systems subjected to
repetitive, random shocks generated according to
the Poisson distribution
 The failure probability then depends on the number
of shocks the device has undergone, i.e., its age
 It is used to analyze wear-out period in components
life
• Each of the shocks is a random variable therefore subject
to an exponential distribution
• It represents the probability that a component suffers k
shocks and after the next one it fails

RRA. 2022-23 Data analysis 36

 Failure models

Gamma distribution

 Failure density of the Gamma distribution with

parameters λ and k with k ∈N+ (Erlang distribution)

λ k t k −1e − λt
=f (t ) λ, k, t ≥ 0
( k − 1)!
 k: shape parameter (number of shocks suffered)
 λ: scale parameter (failure rate of the exponential
distribution for the shocks)

RRA. 2022-23 Data analysis 37

 Failure models

Gamma distribution

 Unreliability
k - 1 (λ ⋅ t)i
F (t ) = 1 − e-λ⋅t ⋅ ∑
1 − R(t) = λ, t > 0
i = 0 i!
 Failure rate f(t) λ k t k −1
h(t) = = λ, t ≥ 0
R(t) k −1
λi ⋅ t i
Γ(k)∑
i =0 i!
 MTTF (up to k failures) E(T) = MTTF =
k
λ

RRA. 2022-23 Data analysis 38

 Failure models

Gamma distribution

Failure rate Failure density

0.8
F ( t , 3 , 1)

F ( t , 3 , 2) 0.6

Unreliability
F ( t , 3 , 4)
0.4
F ( t , 3 , 7)
0.2

0
0
0 1 2 3 4 5
0 t 5

RRA. 2022-23 Data analysis 39

 Failure models

Example
 The adjustment time T for mechanical device follows
an exponential distribution, with an average time of
150 hours. According to the manufacturer standards,
some components must be replaced after 3
consecutive adjustments. It is assumed that the
substitution time T’ of the components is a gamma
distribution
 It must be calculated
• Gamma distribution parameters
• Average time to substitute the components
• Probability of needing spare parts in 400 hours

RRA. 2022-23 Data analysis 40

 Failure models

Example

 Solution
• Gamma distribution parameters
 Device adjustment failure rate λ=1/150  Scale parameter
 Shape parameter: k=3
• Average time to substitute the components
 MTTF(T’)=k/ λ = 450 h
• Probability of needing spare parts in 400 hours
k -1
λi ⋅ t i
F(t , λ , k ) = 1 − e − λ⋅t
∑
i=0 i!
400
−
150  4000 4001 4002 
F( 400,1/150,3) = 1 − e 1500 ⋅ 0! + 1501 ⋅1! + 1502 ⋅ 2!  = 0, 498
 
RRA. 2022-23 Data analysis 41
 Failure models

Weibull distribution

 Generic distribution for general purpose

 It is defined from the cumulative failure rate
β
t 
=H (t )   α, β ≥ 0
α 
• α: scale parameter
• β: shape parameter or “characteristic life”
 Failure rate
β −1
dH (t ) β  t 
h(t ) =
= α, β ≥ 0
dt α  α 
RRA. 2022-23 Data analysis 42
 Failure models

Weibull distribution
 Hazard rate satisfies a power law β-1 as a function of
time
 Therefore, if
• β<1 means that failure rate decreases with time
 This happens if there exists a significant infant mortality
• β=1 means that failure rate is constant in time
 This means that random events are the cause of failure and it is
representing useful life for the component
• β>1 means that failure rate increases with time
 This happens if there exists “wear-out” of the component
 This allows to build the full
bathtub curve

RRA. 2022-23 Data analysis 43

 Failure models

Weibull distribution

 Unreliability β
t 
− 
1 − e− H (t ) =
F (t ) = 1− e α 

 Failure density
β
β −1 t 
βt  − 
f (t ) =   e α 
α α 
 MTTF
 1
E (t ) = MTTF = α ·Γ 1 + 
 β

RRA. 2022-23 Data analysis 44

 Failure models

Weibull distribution

Failure rate Failure density

1
0.8
F ( t , 1 , 0.5)
F ( t , 1 , 1) 0.6

F ( t , 1 , 1.5)

F ( t , 1 , 3)
0.4 Unreliability
0.2
0
0
0 1 2 3 4 5
0 t 5
RRA. 2022-23 Data analysis 45
 Failure models

Estimation of values of reliability functions

 Obtain directly the values of the probability functions
from test or operating data of the systems or
components using estimators
 It may be applied to:
• n components that are not repaired
• 1 component that is repaired n times
 The application of these methods requires that the repaired system
(component) be as new

The data of n failures of a repairable system are

equivalent to the failure of n identical systems

 Large samples (n> 20) are considered to be n the number

of elements in the sample. If the sample is small (n≤20)
the estimators must be corrected
RRA. 2022-23 Data analysis 46
 Failure models

Estimation of values of reliability functions

 Reliability ˆ ni
R(ti ) =
n0

 Unreliability ˆ n0 − ni
F(ti ) =
n0
 Failure density
ˆf(t ) = Δni
i
n0 Δti
 Failure rate
ˆ Δ ni
h(ti) =
ni ∆ti

RRA. 2022-23 Data analysis 47

 Failure models

Adjustment to a Weibull distribution

 Determination of values for R(t)

n
Rˆ (ti ) = i
n0

ti

RRA. 2022-23 Data analysis 48

 Failure models

Adjustment to a Weibull distribution

 Determination of α and β parameters

β
t 
− 

• Inverting terms t 
β
R(t ) = e α 
1  
α 
=e
R(t )
• Taking algorithms
1
ln ln ( ) = β ln t − β ln α 1
R(t) Yˆ = ln ln( )
ˆ
R(t)
1
ln ln( = ) β ln ti − β ln α
ˆ
R(ti )

Yˆ βX i − β ln α
=
Minimum square root method
from experimental data
X i = ln ti
RRA. 2022-23 Data analysis 49
 Maintainability and availability

RRA. 2022-23 Data analysis 50

 Maintenance and availability

Unavailability due to maintenance

 Maintenance is necessary for the components to
operate as planned
 Maintenance activities
• Preventive: Periodic, scheduled  Considered in PRA
• Corrective: Responds to degraded conditions or failures 
Usually not considered in PRA (the failed component is not
repaired)
 The component may not be able to perform its
function during the maintenance time
• Power disconnected, parts removed, sections of pipe
isolated, signals deactivated

RRA. 2022-23 Data analysis 51

 Maintenance and availability

Random variable Time to Repair

 The way to analyze the probability that a component,

not initially available, will be available at a given
moment is to use the random variable Time to
Repair T
 This can only occur on repairable systems and is the
time taken to keep the product looking like new, after
failure, and includes waiting, administrative, repair,
testing, and inspection times
1 T

0
R t
RRA. 2022-23 Data analysis 52
 Maintenance and availability

Maintainability function

 Cumulative distribution function for the random

variable T
 Probability of repair of a component in stated
conditions not before time t, considering that it was
was failed at the beginning of the process
M(t) = Pr(T ≤ t) M (0 ) = 0 M (∞ ) = 1

M(t)

RRA. 2022-23 Data analysis 53

 Maintenance and availability

Repair density m(t)

 Probability density function (pdf) of the random

variable T
dM(t) Pr (t <T ≤ t+∆t)
m(t)= = lim
dt Δt →0 Δt
 m(t)∆t is the unconditional probability that the
component will be repaired in the interval (t,t+ ∆t]

m(t)Δt = Pr(t < T ≤ t + Δt)

RRA. 2022-23 Data analysis 54

 Maintenance and availability

Repair rate g(t)

 g(t)∆t is the conditional probability that the

component will be repaired in the interval (t,t+ ∆t],
given that it was failed until time t

g(t)Δt = Pr( t < T ≤ t + Δt | T > t )

Pr(t < T ≤ t + Δt) m(t)

g(t) = lim =
Δt →0 Δt ⋅ Pr(T > t) 1 − M(t)

RRA. 2022-23 Data analysis 55

 Maintenance and availability

Relationship between reliability functions and

the repair rate
 Number of expected repairs in the temporal interval
(0,t)
t
dG (t )
G (t ) = ∫ g(t)dt ⇒ g (t ) =
0
dt

 Maintainability and repair density as a function of

repair rate
t
M(t) = 1- exp (- ∫ g(t)dt ) = 1- exp (- G(t))
0t
m(t) = g(t) exp (- ∫ g(t)dt ) = g(t) ⋅ exp (- G(t))
0

RRA. 2022-23 Data analysis 56

 Maintenance and availability

Maintainability distribution

 In this case, exponential distribution is habitually

used to model the behavior

Constant repair rate

RRA. 2022-23 Data analysis 57

 Maintenance and availability

Constant repair rate

 A constant repair rate μ gives an exponential repair
distribution where t represents time
g(t) = μ , μ, t ≥ 0
• Number of expected repairs in the temporal interval (0,t)
t
G(t) = ∫ g(t)dt =μ ⋅ t
• Maintainability 0

M(t) = 1 − e − G(t) M(t) = 1 − e − μ⋅t

• Repair density dM(t)
m(t) = − m(t) = μe − μ t
dt
• Mean time to repair (MTTR) ∞
1
MTTR = ∫ ( 1 − M(t)) ⋅ dt =
o
μ

RRA. 2022-23 Data analysis 58

 Maintenance and availability

Repairable system
 Repairable system: not discarded when it fails, but
corrected and reinstalled
• Repair may include replacement of components deemed non-
repairable
 Cycle of a repairable system:
• Time to failure followed by repair time
 Availability of a repairable system: probability of being
able to perform its function, when required
• Availability
• Stationary or asymptotic availability
• Average or interval availability

RRA. 2022-23 Data analysis 59

 Maintenance and availability

Availability A(t)
 Probability that a repairable product is operational at
time t, given that it was in operational condition at the
initial time
 Comparing with reliability
• The reliability of a product requires its operational state during
the period of time [0, t] A(t) ≥ R(t)
• If the product is not repaired A(t) = R(t)
 Unavailability Q(t): probability that a repairable
product is not operational at time t, having it working at
the initial time
 Then we have
A(t) + Q(t) = 1

RRA. 2022-23 Data analysis 60

 Maintenance and availability

Stationary or asymptotic availability A

 Availability limit
A =lim A(t )
t →∞

 Stationary unavailability
Q =lim Q(t )
t →∞
• For most repairable products: Instant availability and
unavailability quickly tends to asymptotic values
1 A(t)
A
Q Q(t)
0
t

RRA. 2022-23 Data analysis 61

 Maintenance and availability

Average availability A

 Proportion of time that a repairable product is

operational in the time interval [0, τ]
1τ
A (τ ) = ∫ A(t)dt
aver τ 0

 Average unavailability: Proportion of time that a

repairable product is not operational in the time
interval [0, τ] τ 1
Q (τ ) = ∫ Q(t)dt
aver τ 0

RRA. 2022-23 Data analysis 62

 Maintenance and availability

Availability
 We must study the complete failure-repair process:
the one in which the repair time is not small
compared with the time in which the system is
operating
 We use the new concepts:
 Failure intensity w(t)
 Repair intensity v(t)
 And the known concepts
• H (t) expected number of failures in the interval (0, t)
• G (t) expected number of repairs in the interval (0, t)

RRA. 2022-23 Data analysis 63

 Maintenance and availability

Availability

 Through a complex theoretical process

Initially Result
Expected
Rates Densities Intensities Unavailability
numbers
λ(t), µ(t) f(t), m(t) w(t), v(t) Q(t)
 H(0,t), G(0,t)

 It can be demonstrated that

• Unavailability
Q(t) = H(0,t) - G(0,t)
• Then Availability is:
A(t) = 1 − Q(t)
RRA. 2022-23 Data analysis 64
 Maintenance and availability

Components in operation with repair

 In the case of components in operation, we can

consider the model of constant failure rate λ and
repair µ f (t )= λ ⋅ e − λt g (t )= µ ⋅ e µ − t

 Applying the theory, we obtain:

• Instant unavailability λ
Q(t) = H(0,t) - G(0,t) Q(t) =
λ+µ
(
1 − e −(λ − µ )t )

• Stationary unavailability (asymptotic)

λ λ MTTR MTTR
Q∞ = lim
t →∞ λ + µ
(
1 − e −(λ − µ )t ) Q∞ =
= =
λ + µ MTTR + MTTF MTBF

RRA. 2022-23 Data analysis 65

 Maintenance and availability

Components in operation with repair

• Instant availability
µ λ
A(t ) =
1 − Q(t ) = + ⋅ e-(λ + µ ) t
λ+µ λ+µ
• Stationary availability (asymptotic)

µMTTF MTTF
A∞ lim=
= A(t ) = =
t →∞ µ + λ MTTR + MTTF MTBF

 If MTTF >> MTTR ó λ >> µ

Q( t ) = λ ⋅ t A( t ) = 1 − λ ⋅ t

RRA. 2022-23 Data analysis 66

 Maintenance and availability

Components in stand-by with surveillance

 Failure model to account for repairable failures of stand-by
components that will only be detected during:
• Actual demands
• Test demands during surveillance
 The model assumes negligible downtime for test and repair,
negligible test inefficiencies, and that, at the end of a test, the
component unavailability is zero
 We use the unreliability obtained averaging F(t) over the interval
between surveillances TS:
T T
1
= F
1 S
∫
TS 0
F=(t )dt
1 S
∫
TS 0
− λt
(1 − e )dt F =1 −
λTS
1 − (
e − λTS
)

F(t)
F

TS TS TS t
RRA. 2022-23 Data analysis 67
 Maintenance and availability

Components in stand-by with surveillance

 If λTS << 1 (λTS < 0.1)

 1 − λTS  1  (λTS ) 2  λTS
F =lim 1 − ( 1− e ) = 1−  −λTS + =
λTS →0
 λTS  λTS  2  2
− λTS (λTS ) 2
because e ≈ 1 − λTS + when λTS → 0
2
λTS
F=
2
F(t)

TS TS TS t

RRA. 2022-23 Data analysis 68

 In Saphire

RRA. 2022-23 Data analysis 69

 In SAPHIRE

Demand probabilities

RRA. 2022-23 Data analysis 70

 In SAPHIRE

Exponential distribution

RRA. 2022-23 Data analysis 71

 In SAPHIRE

Exponential distribution approximation

RRA. 2022-23 Data analysis 72

 In SAPHIRE

Unavailability

RRA. 2022-23 Data analysis 73

 In SAPHIRE

Stationary unavailability with MTTF >> MTTR

RRA. 2022-23 Data analysis 74

 In SAPHIRE

Stand-by component with surveillance

RRA. 2022-23 Data analysis 75

 In SAPHIRE

Stand-by component with surveillance

approximation

RRA. 2022-23 Data analysis 76

 Data bases

RRA. 2022-23 Data analysis 77

 Data

Required data
 To be able to quantify both the event tree and the fault
trees, certain data are needed, mostly as a result of
statistical analysis
• Initiating event frequencies
• Basic independent failure events
 Demand failure probabilities
 Operating and standby failure rates
• Unavailability
 Frequencies and duration of tests and preventive maintenance
 Unavailability time of each component due to corrective maintenance
• Special event probabilities
• Common cause failures: probabilities or rates
• Probabilities of human error
RRA. 2022-23 Data analysis 78
 Data

Type of data
 Generic data: provide information on different plants
• Large data volume vs. specific data
• Data from other plants: they may be similar or different to
those of the plant under study
• The distribution function of the generic data constitutes the “a
priori” function in the Bayesian estimation
 Specific data: allows to introduce the experience of the
plant into the analysis
• They represent a lower volume of data than generic ones
• They are directly applicable to the modeled basic events
• This information is specific to the plant in question and is
considered as a function of evidence in Bayes, it is applied to a
direct estimation
RRA. 2022-23 Data analysis 79
 Data

Generic sources of data

 NUREG/CR 4550 Vol.1 Analysis of Core Damage Frequency From Internal
Events: Methodology Guidelines September 1987
 NUREG/CR-2728 Interim Reliability Evaluation Program Procedure Guide,
Sandia National Laboratories, January 1983
 IEEE Standard 500, IEEE Guide to the Collection and Presentation of
Electrical, Electronic, Sensing Component, and Mechanical Equipment
Reliability Data for Nuclear-Power Generating Stations, Appendix D,
Reliability Data for Nuclear-Power Generating Stations, IEEE 1984
 NUREGs based on LERs (Licensee Event Report)
• Power switches
• Instrumentation channels and transmitters
 And a lot more…

Compilation: IAEA-TECDOC-478

RRA. 2022-23 Data analysis 80

 Data

Generic sources of data

 IAEA-TECDOC-478

RRA. 2022-23 Data analysis 81

 Data

Generic sources of data

 IAEA-TECDOC-478

RRA. 2022-23 Data analysis 82

 Common Cause Failures

RRA. 2022-23 Data analysis 83

 Common Cause Failures

Types of dependent failures

 There are three general types of dependent faults
• Certain initiating events (eg, fires, floods, earthquakes, loss of service
water)
• Dependencies between systems including:
 Functional dependencies (eg, AC current dependence)
 Shared Equipment Dependencies (eg, High Pressure Injection (HPCI) and
Insulated Refrigeration System (RCIC) share common Condensate Tank Suction
Valve (CST)
• Intercomponent dependencies (eg, existing design flaws in multiple
similar valves)
 Modeling
• The first two types are explicitly modeled in event and fault trees,
• The third type is known as common cause failure (CCF) (i.e. residual
dependencies are not explicitly modeled) and is dealt with
parametrically
RRA. 2022-23 Data analysis 84
 Common Cause Failures

Types of dependent failures

 Common Cause Failures (CCFs) are those faults that

simultaneously affect more than one redundancy of
the components (same function and system)
 Break the hypothesis of independence
• Failure of one component increases the probability of
failure of others
 They need a specific treatment
 They constitute a common cause component group
(CCCG)

RRA. 2022-23 Data analysis 85

 Common Cause Failures

Examples of CCF factors

 Similar design
 Spare parts from the same supplier
 Physically close, subject to the same environmental
conditions
• Radioactivity
• Temperature
• Corrosion
 Maintenance / calibration performed by the same
personnel or the same shift
 Using similar procedures
 Etc.

RRA. 2022-23 Data analysis 86

 Common Cause Failures

Examples of components with possibility of CCF

 Pumps
 Valves: Motorized, Safety / Relief, Pneumatic, Control
/ Air Operated, Hydraulic
 Fans, Fan-coils, cooling units
 Compressors
 Switches (medium voltage)
 Batteries and chargers
 Diesel generators

RRA. 2022-23 Data analysis 87

 Common Cause Failures

Basic Parameter Model

 Expands the failure probability (or failure rate) of a

component in a common-cause component group
(CCCG) into terms involving
• independent failure of the component and
• combinations of CCFs with the other components in the
CCCG

𝐴𝐴𝑡𝑡 = 𝐴𝐴𝐼𝐼 + 𝐶𝐶𝐴𝐴𝐴𝐴 + 𝐶𝐶𝐴𝐴𝐴𝐴 + 𝐶𝐶𝐴𝐴𝐴𝐴𝐴𝐴

𝐵𝐵𝑡𝑡 = 𝐵𝐵𝐼𝐼 + 𝐶𝐶𝐴𝐴𝐴𝐴 + 𝐶𝐶𝐴𝐴𝐴𝐴 + 𝐶𝐶𝐴𝐴𝐴𝐴𝐴𝐴
𝐶𝐶𝑡𝑡 = 𝐶𝐶𝐼𝐼 + 𝐶𝐶𝐴𝐴𝐴𝐴 + 𝐶𝐶𝐴𝐴𝐴𝐴 + 𝐶𝐶𝐴𝐴𝐴𝐴𝐴𝐴

RRA. 2022-23 Data analysis 88

 Common Cause Failures

Basic Parameter Model

 Symmetry assumption: the probabilities of similar

events involving similar components (i.e., events in
the same CCCG) are the same
Pr 𝐴𝐴𝐼𝐼 = Pr 𝐵𝐵𝐼𝐼 = Pr 𝐶𝐶𝐼𝐼 = 𝑄𝑄1
Pr 𝐶𝐶𝐴𝐴𝐴𝐴 = Pr 𝐶𝐶𝐴𝐴𝐴𝐴 = Pr 𝐶𝐶𝐵𝐵𝐵𝐵 = 𝑄𝑄2
Pr 𝐶𝐶𝐴𝐴𝐴𝐴𝐴𝐴 = 𝑄𝑄3

• The probability of occurrence of any basic event within a

given CCCG is assumed to depend only on the number and
not on the specific components in that basic event.

RRA. 2022-23 Data analysis 89

 Common Cause Failures

Basic Parameter Model (BPM)

 For a CCCG of size m,𝑚𝑚the BPM defines
𝑚𝑚 − 1 (𝑚𝑚)
𝑄𝑄𝑡𝑡 = � 𝑄𝑄
𝑘𝑘 − 1 𝑘𝑘
𝑘𝑘=1
• where Q(m)k = probability of a CC basic event involving k specific
components in a CCCG of size m, ( 1 ≤ k ≤ m)

 BPM estimates the basic event probabilities directly

 Basic parametric model may be reformulated to calculate
the basic event probabilities from a set of parameters
• Beta factor model
• Multiple Greek letters
• Alpha factor model

RRA. 2022-23 Data analysis 90

 Common Cause Failures

Beta factor model

 A single parameter model, that is, it uses one parameter (β)
in addition to the total component failure probability to
calculate the common cause failure probabilities.
 It was the first model to be applied to common cause events
in risk and reliability studies.
 The model assumes that a constant fraction (β) of the
component failure probability can be associated with
common cause events shared by other components in that
group
 Another assumption is that whenever a common cause event
occurs, all components within the common cause component
group fail
• Therefore, for a group of m components, all Qk(m)'s are zero except
Q1(m) and Qm(m).

RRA. 2022-23 Data analysis 91

 Common Cause Failures

Beta factor model

 Taking into account all this failure probabilities are

calculated as follows
Q= Q (m) (m)
+Q(m) Q1( m )= (1 − β )Qt( m )
t 1 m

(m)
Q = βQ (m) Qm( m )
m t β = (m)
Q1 − Qm( m )
(m) β
Q = Q1( m )
 Therefore m
1− β
(m) 1 + β (m)
Q t = Q1
1− β

RRA. 2022-23 Data analysis 92

 Common Cause Failures

Beta factor model

RRA. 2022-23 Data analysis 93

 Common Cause Failures

Multiple Greek Letters model

 Is the most general of a number of recent extensions

of the beta-factor model
 In this model, other parameters in addition to the
beta factor are introduced to account more explicitly
for higher order redundancies and to allow for
different probabilities of failures of subgroups of the
common cause component group

RRA. 2022-23 Data analysis 94

 Common Cause Failures

Alpha factor model

 For several practical reasons, it is often more
convenient to rewrite the Q k(m) ’s of the BPM in
terms of other more easily quantifiable parameters
 The alpha-factor model develops CCF probabilities
from a set of failure ratios and the total component
failure probability (Qt).
• Qt = total failure probability of each component due to all
independent and CCF events
• αk = fraction of the total number of failure events that
occur in the CCCG that involve the failure of k components
due to a common cause.
 Obtained from testing
RRA. 2022-23 Data analysis 95
 Common Cause Failures

Alpha factor model

 Using these parameters, depending on the

assumption regarding the way components in the
CCCG are tested, the probability of a CCBE involving
failure of k components in a CCCG of m components
is given by the following:

(𝑚𝑚) 1
𝑄𝑄𝑘𝑘 = 𝛼𝛼 𝑄𝑄
𝑚𝑚 − 1 𝑘𝑘 𝑡𝑡
𝑘𝑘 − 1

RRA. 2022-23 Data analysis 96

 Summary

RRA. 2022-23 Data analysis 97

 Summary

 Unreliability and unavailability must be modeled to be

introduced in the Reduced Boolean Equation
 There are different reliability functions depending on the
failure mode being studied
 Maintainability must be adequately taken into account
when conducting a PSA to derive the availability
 Common cause failures (CCF) are of great importance
when quantifying the PSA since they increase the
unreliability of the component and therefore the system

RRA. 2022-23 Data analysis 98

 Reliability and Risk Analysis

Master in Nuclear Science and Technology

Joint Master in Safe and Reliable Nuclear
Applications (SARENA)
RRA. 2022-23 Data analysis 99