MNE4109 Reliability Engineering and Risk Analysis

Topic 1: Random variable and probability distribution

Wei WANG
wwang326@cityu.edu.hk
Topic
o Basic notions of probability theory
 Basic Definitions
 Boolean Logic
 Definitions of probability
 Probability laws
 Random variables
 Probability distributions for reliability, safety and risk

different failure times

X time
Probability distribution to
X time represent the failure time Random
variable
fT(t) P(t)

time
2
Random Variable
Random variable
o Experiment: ε
o Sample space: Ω X(ω) random variable in ℛ
o Outcome: 𝜔 quantifies outcomes of a random
occurrence or an experiment;
can take on many values;
measurable;
real value.
Univocal mapping

Ω  Specific value;
 Function;

𝜔−

𝜔+

ℛ
𝑋(𝜔− ) 𝑋(𝜔+ )

4
Random variable - example
o Experiment: ε = die toss
o Sample space: Ω = {1, 2, 3, 4, 5, 6}; X(ω) in ℛ
o Outcome: 𝜔

Univocal mapping

1 2 3
4 5 6

ℛ
1 2 3 4 5 6

5
Random variable - represent events
o Experiment: ε = die toss
o Sample space: Ω = {1, 2, 3, 4, 5, 6}; X(ω) in ℛ
o Event: 𝐸1 = 1,2,3,4 𝐸1 = 𝑋 < 4.236
𝐸2 = ∅ 𝐸2 = 𝑋 < 0
𝐸3 = Ω 𝐸3 = 𝑋 < +∞
Univocal mapping

1 2 3
4 5 6

ℛ
1 2 3 4 5 6

6
Random variable
o Experiment: ε
o Sample space: Ω X(ω) random variable in ℛ
o Outcome: 𝜔

General mathematical models of random behavior

Applied to different physical phenomena

which behave similarly
o Discrete;
o Continuous.

7
Probability distributions for reliability, safety and risk analysis
Probability functions (I)
gives a probability that X can take a value
o Cumulative Distribution Function (CDF) no larger than an independent value x.

𝐹𝑋 𝑥 = 𝑃 𝑋 ≤ 𝑥 Discrete: x
o Properties: Continuous: x
 lim 𝐹𝑋 𝑥 = 0
𝑥→−∞

 lim 𝐹𝑋 𝑥 = 1
𝑥→+∞

 𝐹𝑋 𝑥 is a non-decreasing function of x;
 The probability that X takes on a value in the interval [a,b] is:
𝑃 𝑎 < 𝑋 ≤ 𝑏 = 𝐹𝑋 𝑏 − 𝐹𝑋 𝑎 FX(x)

x
9
Probability functions (II, discrete random variable)
o 𝑋 is random variable takes discrete values 𝑥𝑖 , 𝑖 = 1, … , 𝑛.
o Probability Mass Function (PMF):
𝑓𝑋 𝑥𝑖 = 𝑃 𝑋 = 𝑥𝑖 = 𝑝𝑖 𝑓𝑋 (𝑥𝑖 ) 0.512

𝑛 0.384
෍ 𝑓𝑋 (𝑥𝑖 ) = 1
𝑖=1
0.096
0.008
PMF gives each discrete value 𝑥𝑖 with a
probability assignment 𝑝𝑖 , indicating the x
probability that X can take the value 𝑥𝑖 . 0 1 2 3

10
Probability functions (II, discrete random variable)
o 𝑋 is random variable takes discrete values 𝑥𝑖 , 𝑖 = 1, … , 𝑛.
o Probability Mass Function (PMF):
𝑓𝑋 𝑥𝑖 = 𝑃 𝑋 = 𝑥𝑖 = 𝑝𝑖 𝑓𝑋 (𝑥𝑖 ) 0.512

𝑛 0.384
෍ 𝑓𝑋 (𝑥𝑖 ) = 1 = 𝐹𝑋 𝑥 , 𝑥 ≥ 𝑥𝑛
𝑖=1
0.096
0.008
o Cumulative Distribution Function (CDF)
x
0 1 2 3
𝐹𝑋 𝑥 = 𝑃 𝑋 ≤ 𝑥 = ෍ 𝑓(𝑥𝑖 ) FX(t) 1.000
𝑥𝑖 ≤𝑥
stepwise
0.616
increase

0.104

0.008
0.000
x
0 1 2 3
11
Summary measures (discrete random variable)
o Mean value (expected value):
𝑛

𝜇𝑋 = 𝐸 𝑥 = ෍ 𝑥𝑖 𝑝𝑖
𝑖=1

A point at which the probability mass is concentrated on average

o Variance:
𝑛

𝑉𝑎𝑟 𝑋 = 𝜎𝑋2 = ෍ 𝑥𝑖 − 𝜇𝑋 2 𝑝𝑖
𝑖=1

It is a measure of the dispersion of the values around the mean

12
Probability functions (III, continuous random variable)
o Let X be a random variable which takes continuous values in ℝ.
o Cumulative Distribution Function (CDF): FX(x)
𝑥
𝐹𝑋 𝑥 = 𝑃 𝑋 ≤ 𝑥 = න 𝑓𝑋 𝜏 𝑑𝜏
−∞

x
x x+dx
o Consider a small interval 𝑥, 𝑥 + 𝑑𝑥 , the probability is:
𝑃 𝑥 ≤ 𝑋 < 𝑥 + 𝑑𝑥 = 𝐹𝑋 𝑥 + 𝑑𝑥 − 𝐹𝑋 (𝑥) fX(x)
𝑓𝑋 𝑥 𝑑𝑥
o Define: Probability Density Function (PDF):
𝐹𝑋 𝑥 + 𝑑𝑥 − 𝐹𝑋 𝑥 𝑑𝐹𝑋
𝑓𝑋 𝑥 = lim =
𝑑𝑥→0 𝑑𝑥 𝑑𝑥 x
x x+dx
 𝑓𝑋 (𝑥) is not probability but probability per unit x (i.e., probability density);
 𝑓𝑋 (𝑥) ≥ 0;
+∞
 ‫׬‬−∞ 𝑓𝑋 𝑥 𝑑𝑥 = 1.
13
Summary measures (continuous random variable)
o Distribution Percentiles (𝑥𝛼 ):
𝛼 FX(x)
𝐹𝑋 𝑥𝛼 =
100
0.90
o Median (𝑥50 ):
𝐹𝑋 𝑥50 = 0.50 0.10
𝑥10 𝑥90 x
The probability to be below or above is equal.

o Mean value (expected value):

𝑛
+∞
𝜇𝑋 = 𝐸 𝑥 = න 𝑥𝑓𝑋 𝑥 𝑑𝑥 continuous = ෍ 𝑥𝑖 𝑝𝑖 discrete
−∞ 𝑖=1
The probability mass is concentrated on average.
o Variance:
𝑛
+∞
𝑉𝑎𝑟 𝑋 = 𝜎𝑋2 = න 𝑥 − 𝜇𝑋 2 𝑓𝑋 𝑥 𝑑𝑥 continuous = ෍ 𝑥𝑖 − 𝜇𝑋 2 𝑝𝑖 discrete
−∞ 𝑖=1
It is a measure of the dispersion of the values around the mean.
14
Reliability
o T = Time to failure of a component (random variable);
o Probability Density Function (PDF) at time t: 𝑓𝑇(𝑡);
o Cumulative Distribution Function (CDF) at time t: 𝐹𝑇(𝑡) = 𝑃(𝑇 < 𝑡)
(probability of having failure before t);
o Reliability at time t: 𝑅 𝑡 = Probability that the component does not fail up to t:
NO failure
𝑅 𝑡 = 𝑃 𝑇 > 𝑡 = 1 − 𝐹𝑇 𝑡 X time
0 𝑡

fT(t) 𝑃 𝑇<𝑡 FT(t)

𝑃 𝑡1 ≤ 𝑇 < 𝑡2 1 𝐹𝑇 𝑡
𝑹 𝒕

t 0 t
t1 t2 t

18
PDF: interpretation
o We start out a new item at time t = 0, when we ask a question:
«What is the probability that the item will fail in an interval [𝑡, 𝑡 + 𝑑𝑡]?»

𝑃 𝑡 ≤ 𝑇 < 𝑡 + 𝑑𝑡 = 𝐹𝑇 𝑡 + 𝑑𝑡 − 𝐹𝑇 (𝑡) ≈ 𝑓𝑇 𝑡 𝑑𝑡
PDF
measures the
X time failure probability
0 t t+dt per unit time.

19
Hazard function
o We start out a new item at time t = 0, when we ask a question:
«What is the probability that the item will fail in an interval [𝑡, 𝑡 + 𝑑𝑡]?»

𝑃 𝑡 ≤ 𝑇 < 𝑡 + 𝑑𝑡 = 𝐹𝑇 𝑡 + 𝑑𝑡 − 𝐹𝑇 (𝑡) ≈ 𝑓𝑇 𝑡 𝑑𝑡
PDF
measures the
X time failure probability
0 t t+dt per unit time.

o When the item has survived until time t (age), we ask again:
«What is the probability that the item will fail in the next interval [𝑡, 𝑡 + 𝑑𝑡]?»

Hazard function
𝑃 𝑡 ≤ 𝑇 < 𝑡 + 𝑑𝑡 𝑇 > 𝑡 ≈ 𝒉𝑻 𝒕 𝑑𝑡 (also known as the
failure rate, hazard
NO failure rate) measures the
propensity to fail,
X time given the survival age.
0 t t+dt
20
Hazard function & reliability
𝑃 𝑡 ≤ 𝑇 < 𝑡 + 𝑑𝑡 𝑓𝑇 𝑡 𝑑𝑡
𝒉𝑻 𝒕 𝑑𝑡 = 𝑃 𝑡 ≤ 𝑇 < 𝑡 + 𝑑𝑡 𝑇 > 𝑡 = =
𝑃 𝑇>𝑡 𝑅 𝑡

𝑑𝐹𝑇 𝑡 𝑑(1 − 𝑅(𝑡)) 𝑑𝑅(𝑡)

𝑓𝑇 𝑡 = = =−
𝑑𝑡 𝑑𝑡 𝑑𝑡

𝑑𝑅(𝑡)
ℎ 𝑇 𝑡 𝑑𝑡 = −
𝑅(𝑡)

Integral

𝑡
න ℎ 𝑇 𝜏 𝑑𝜏 = −ln 𝑅(𝑡)
0

𝑡
𝑅 𝑡 = 𝑒 − ‫׬‬0 ℎ𝑇 𝜏 𝑑𝜏
21
Hazard function: the Bath-Tub Curve
A hazard function commonly shows three distinct phases:
Items are tested at the factory before distributed
o Decreasing - infant mortality or burn in: to users  infant mortality is removed.
Failures due to defective pieces of equipment not manufactured or constructed
properly (missing parts, substandard material batches, damage in shipping, ...)
o Constant - useful life:
Random failures due to unavoidable loads coming from (earthquakes, power
surges, vibration, temperature fluctuations,...)
o Increasing - ageing:
Aging failures due to cumulative effects such as corrosion, embrittlement,
fatigue, cracking, …

(i) (ii) (iii)

Congenital Human accidents Loss of bone mass  femur fracture

defects Arterial hardening,… 22
Univariate probability distributions
Univariate discrete probability distributions
o Binomial distribution
o Geometric distribution
o Poisson distribution
Binomial distribution
o Y = discrete random variable with only two possible outcomes:
1, success 𝑃 𝑌=1 =𝑝
𝑌=ቊ Bernoulli process
0, failure 𝑃 𝑌 =0 =1−𝑝
o We perform n different trials of the experiment (independent), 𝑌1 , … , 𝑌𝑛 .

o X = discrete random variable counting the number of success out of the n trial
(independently from the sequence with which successes appear):
𝑛

𝑋 = ෍ 𝑌𝑖
𝑖=1
Ω = 0,1,2, … , 𝑛

𝑛 𝑥 𝑛−𝑥 𝑛 𝑛!
o PMF of X: 𝑏 𝑥; 𝑛, 𝑝 = 𝑝 1−𝑝 𝑥 = 0,1,2, ⋯ , 𝑛 𝑥
=
𝑥 𝑛 − 𝑥 ! 𝑥!

𝐸 𝑋 = 𝑛𝑝
𝑉𝑎𝑟 𝑋 = 𝑛𝑝 1 − 𝑝 25
Geometric distribution
o Y = discrete random variable with only two possible outcomes:
1, success 𝑃 𝑌=1 =𝑝
𝑌=ቊ Bernoulli process
0, failure 𝑃 𝑌 =0 =1−𝑝
o We perform n different trials of the experiment (independent), 𝑌1 , … , 𝑌𝑛 .

o T = trial of the first success (or number of trials between two successive
occurrences of success);

Ω = 0,1,2, … , 𝑛

o PMF of T: 𝑔 𝑡; 𝑝 = 1 − 𝑝 𝑡−1 𝑝 𝑡 = 1,2, ⋯

∞
𝑡−1 𝑝 2
1
𝐸 𝑋 = ෍𝑡 1 − 𝑝 =𝑝 1+2 1−𝑝 +3 1−𝑝 +⋯ =
𝑝
𝑖=1
The number of trials that wait, on average,
1−𝑝 for a success after a previous success.
𝑉𝑎𝑟 𝑋 =
𝑝2 26
Poisson distribution
o Stochastic events that occur in a continuum period, with a constant rate of
occurrence, 𝜆:

o K = discrete random variable, indicating the number of events in the period of

observation (0,𝑡):

𝜆𝑡 𝑘 −𝜆𝑡
o PMF of K: 𝑝 𝑘; 0, 𝑡 , 𝜆 = 𝑒 𝑘 = 0,1,2, ⋯
𝑘!
𝐸 𝑋 = 𝜆𝑡
𝑉𝑎𝑟 𝑋 = 𝜆𝑡

27
Univariate continuous probability distributions
o Exponential distribution
o Weibull distribution
o Normal distribution
o Uniform distribution
Exponential distribution
o When the item has survived until time t:

NO failure

time
0 t
𝑡
𝑅 𝑡 = 𝑒 − ‫׬‬0 ℎ𝑇 𝜏 𝑑𝜏

Hazard function ℎ 𝑇 𝑡 = 𝜆 (constant)

𝜆𝑡 𝑘 −𝜆𝑡 λ𝑡 0 −λ𝑡
𝑅 𝑡 = 𝑃 𝑇 > 𝑡 = 𝑃 no failure in 0, 𝑡 = 𝑒 = 𝑒 = 𝑒 −λ𝑡
𝑘! 0!
Poisson process where no failure
event occurs in the period (0,𝑡).

𝐹𝑇 𝑡 = 1 − 𝑅 𝑡 = 1 − 𝑃 𝑇 > 𝑡 = 1 − 𝑒 −𝜆𝑡 Note: It is the only

distribution characterized
𝑓𝑇 (𝑡) = 𝜆𝑒 −𝜆𝑡 by a constant failure rate.
29
Hazard function: the Bath-Tub Curve
A hazard function commonly shows three distinct phases:
Items are tested at the factory before distributed
o Decreasing - infant mortality or burn in: to users  infant mortality is removed.
Failures due to defective pieces of equipment not manufactured or constructed
properly (missing parts, substandard material batches, damage in shipping, ...)
o Constant - useful life:
Random failures due to unavoidable loads coming from without (earthquakes,
power surges, vibration, temperature fluctuations,...)
o Increasing - ageing:
Aging failures due to cumulative effects such as corrosion, embrittlement,
fatigue, cracking, …
(i) (ii) (iii)

constant
failure
rate

30
Exponential distribution
o Mean:
+∞ +∞
1 the expected length of time that an
𝐸 𝑇 =න 𝑡𝑓 𝑡 𝑑𝑡 = න 𝑡𝜆𝑒 −𝜆𝑡 𝑑𝑡 = = 𝑀𝑇𝑇𝐹
0 0 𝜆 item survives in its operation.
Integration by parts

o Variance:
1
𝑉𝑎𝑟 𝑋 =
𝜆2

31
Exponential distribution: momerylessness
o A component with constant failure rate, 𝜆, is found still operational at a given
time 𝑡1 (i.e., age of the component).
«What is the probability that it will fail in the next time length 𝜏?»

𝑃 𝑡1 < 𝑇 ≤ 𝑡1 + 𝜏 𝑇 > 𝑡1

𝑃(𝑡1 < 𝑇 ≤ 𝑡1 + 𝜏) 𝐹 𝑡1 + 𝜏 − 𝐹 𝑡1
= =
𝑃(𝑇 > 𝑡1 ) 𝑅(𝑡1 )

1 − 𝑒 −𝜆(𝑡1+𝜏) − 1 − 𝑒 −𝜆𝑡1 𝑒 −𝜆𝑡1 − 𝑒 −𝜆(𝑡1+𝜏)

= =
𝑒 −𝜆𝑡1 𝑒 −𝜆𝑡1

= 1 − 𝑒 −𝜆𝜏 =𝐹(𝜏)

o Still exponential with failure rate 𝝀!

o The probability that it will fail in a period of time length 𝝉 does not depend
from the component age 𝒕𝟏 (when the component is as good as new).
o The age does not influence the failure process.
32
Weibull distribution
o The age of a component influences its failure process so that the hazard rate
does not remain constant throughout the lifetime:

X time
0 t

Hazard function ℎ 𝑇 𝑡 ≠ constant

𝒉 𝒕 = 𝝀𝜶𝒕𝜶−𝟏 𝒕 > 𝟎
𝑡 𝑡
− ‫׬‬0 𝜆𝛼𝜏𝛼−1 𝑑𝜏 𝛼
𝐹𝑇 𝑡 = 𝑃 𝑇 < 𝑡 = 1 − 𝑒 − ‫׬‬0 ℎ𝑇 𝜏 𝑑𝜏 =1−𝑒 = 1 − 𝑒 −𝜆𝑡

𝑑𝐹𝑇 𝑡 𝛼
𝑓𝑇 (𝑡) = = 𝜆𝛼𝑡 𝛼−1 𝑒 −𝜆𝑡
𝑑𝑡

33
Hazard function: the Bath-Tub Curve
A hazard function commonly shows three distinct phases:
Items are tested at the factory before distributed
o Decreasing - infant mortality or burn in: to users  infant mortality is removed.
Failures due to defective pieces of equipment not manufactured or constructed
properly (missing parts, substandard material batches, damage in shipping, ...)
o Constant - useful life:
Random failures due to unavoidable loads coming from without (earthquakes,
power surges, vibration, temperature fluctuations,...)
o Increasing - ageing:
Aging failures due to cumulative effects such as corrosion, embrittlement,
fatigue, cracking, …
𝜶<1 𝜶=1 𝜶>1

𝒉 𝒕 = 𝝀𝜶𝒕𝜶−𝟏

34
Normal (or Gaussian) distribution
o PDF of X: 𝑋~𝑁 𝜇𝑋 , 𝜎𝑋

1 1 𝑥−𝜇𝑋 2
−
𝑓𝑋 𝑥; 𝜇𝑋 , 𝜎𝑋 = 𝑒 2 𝜎𝑋 −∞ < 𝑥, 𝜇𝑋 < ∞; 𝜎𝑋 > 0
2𝜋𝜎𝑋
𝐸 𝑇 = 𝜇𝑋
𝑉𝑎𝑟 𝑋 = 𝜎𝑋2 o Symmetric about the mean;
o Data near the mean are more frequent in
fX(x) occurrence than data far from the mean.

fX(x1)dx It is the only distribution

with a symmetric bell shape!

fX(x2)dx
x
t1 t2
t3 t4

35
Uniform distribution
o PDF of X: 𝑈 𝑎, 𝑏
1 o Equal probability over a given range;
𝑓𝑋 𝑥; 𝑎, 𝑏 = ቐ𝑏 − 𝑎 , 𝑎 < 𝑥 < 𝑏 o Used for generation of random numbers
0, otherwise e.g. in Monte Carlo sampling.

o CDF:
0, 𝑥<𝑎
𝑥−𝑎
𝐹𝑋 𝑥 = , 𝑎≤𝑥≤𝑏
𝑏−𝑎
1, 𝑥>𝑏

𝑎+𝑏
𝐸𝑇 =
2

𝑏−𝑎 2
𝑉𝑎𝑟 𝑋 =
12

38