1

Introduction to
Reliability Engineering

Unit 7a

Spring 2024
References
2
• Ebeling, C.E., Introduction to Reliability and Maintainability
Engineering, 2nd ed, Waveland Press, 2019, Chapter 1, 2
(Ebeling, 2019)
• Modarres, M., M. Kaminskiy, V. Krivtsov, Reliability Engineering and Risk
Analysis, 2nd ed, Taylor & Francis, 2010 (Modarres, RERA)
• Jordaan, Ian, Decisions Under Uncertainty– Probabilistic Analysis for
Engineering Decisions, Cambridge University Press, 2005 (Jordaan,
2005)
• Modarres, M., Risk Analysis in Engineering, Taylor&Francis, 2006
(Modarres, RAE)
• Modarres, M., Reliability Engineering and Risk Analysis in Engineering,
Marcel Dekker, 1999 (Modarres, RE)
• O’Connor, P.D.T., Practical Reliability Engineering, 4th ed, Wiley, 2002
(O’Connor, PRE)
• Rausand, M. and Hoyland, A., System Reliability Theory, 2nd edition,
Wiley, 2004
Things Fail! Consider Causes
3
• 1978 - Ford Pinto was recalled for modifications to the fuel tank to
reduce fuel leakage and fires resulting from rear-end collisions.
Numerous reported deaths, lawsuits, and the negative publicity
eventually contributed to discontinued production of the Pinto.
– Faulty design
• 1979 - The left engine of a DC-10 broke away from the aircraft
during take-off killing 271 people.
– Poor maintenance procedures, engine removal procedures introduced
unacceptable stresses on the pylons.
• 1979 - The Three Mile Island disaster resulted in a partial meltdown
of a nuclear reactor
– Cooling water flow to reactor failed. Backup system was down for
routine maintenance. Warning lights were hidden by maintenance tags.
Emergency relief valve failed to close causing additional water to be lost
from the cooling system. Operators were reading gauges that were not
working properly. Result of both mechanical and human error and
organizational factors.
More Things Fail!
4
• 1986 - Explosion of the space shuttle Challenger
– The below freezing temperatures prior to launch contributed to the failure
by making the rubber O-rings brittle, which failed to seal four sections of
booster rocket. Also, failure to investigate and learn from Near Misses.

• 2003 - Space Shuttle Columbia disintegrated during re-entry into

Earth's atmosphere, loss of all 7 crew members
– Loss was a result of damage sustained during launch when a piece of
foam insulation the size of a small briefcase broke off the Space Shuttle
external tank under the aerodynamic forces of launch. Also, failure to
investigate and learn from Near Misses.
• 2007 - The entire span of the Interstate 35W bridge collapsed where
the freeway crosses the river in Minneapolis.
– Failure of undersized, steel gusset plates was reason for collapse. Engineers who
designed the bridge in the 1960s either failed to properly calculate the thickness
needed for the plates that were to hold the bridge together. Faulty Design, failure
to inspect and repair bridge
Incidents still keep happening!
5

• 2009- Macondo Disaster

– Blowout preventer failed to shutoff fuel kick

• 2019- Multiple fire and explosion in fuel storage tank

farm at TPC Group, Port Neches
– 3-day mandatory evacuation, 3 injuries due to pipeline rupture
Why Study Reliability?
6

• Reliability Engineering has grown in importance due to its

recognized direct connection with risk:
o Profit considerations resulting from the high cost of risk due
to failures, their repairs, and warranty programs
o Public awareness and insistence on product quality
o New laws and regulations concerning product liability
o Government requirements to meet Reliability and
Maintainability performance specifications
o Increased complexity and sophistication of systems requires
greater analysis
Risk, Performance, Cost
7
Identifying components critical to performance are
required for risk analysis. Understanding their performance
involves estimating reliability, availability, maintainability.

Engineering decisions involve risk-based synergistic

effects and balance among risk, performance, and cost.
Reliability Estimation
8

• Probabilistic
• Challenge>Capacity?
• Reliability R(t) vs. Maintainability H(t) vs. Availability A(t)
 Predicting Reliability: Example
9
• For a new TV unit produced by the XYZ Company, the
following distribution of the fraction of units that failed
was obtained from a reliability testing program.

Modelling failure behavior to predict Reliability:

Larger fraction of failures initially:
0.12

0.1
Fraction Failed

0.08

F(t) 0.06

0.04

0.02

0
1000 3000 5000 7000 9000

Operating Hours

Ebeling, 2010
 Predicting Reliability: Example
10
• From these data, fraction failed, F(t), was derived as,

where F(t) is the cumulative probability of a TV failure occurring by

time t (in operating hours):

• Assuming the typical consumer will use the TV an average of 3

hours a day, 1095 operating hours (3 x 365) will be observed for
the first year. Therefore, the cumulative probability of a unit
failing by t = 1095 hr is

• With over ten percent of the units sold expected to fail during the
first year, the company decided to initiate a Reliability Growth
program to improve product reliability, reduce warranty costs,
and increase customer satisfaction.
 The Reliability Function
11

• Let random variable T be the time to failure of a

component.
• Reliability at time T = t is the probability of working up to
and at time t.
• R(t) is also is the probability of failing AFTER time t
designated by R(t) = P(T > t)
• Simply stated, Reliability R(t) is the probability that a
component or system will survive at least until time t.
 Graph of a Reliability CDF
12
Three characteristics of an acceptable Reliability distribution, R(t):
1.2

Probability of surviving
R(t) 1
1. R(T = 0) ~ 0, set = 1, because it is
0.8 tested and observed to be working
0.6
2. Monotonically decreasing
0.4

0.2
3. R(T ∞) approaches 0
t
0
0 20 40 60 80 100 120

Based on this model, R(t), a cumulative distribution

function (cdf), is a monotonically decreasing function
of time, because the influences that reduce reliability
Ebeling, 2010
accumulate with time.
The Failure CDF
13

• Failure distribution: F(t) = 1 – R(t)

• F(t) = P(T< t), designated the failure cdf, is the probability
of failure by time t.
• Because R(t) and F(t) are complementary functions, R(t) +
F(t) = 1 for each value of t.

Logic expression:
 Graph of a Failure CDF
14

Three characteristics of an acceptable Failure distribution, F(t):

1.2

F(t) The probability of a failure

1

3. approaches 1
0.8

0.6

2. monotonically increasing
0.4

0.2
1. F(T = 0) ~ 0, set = 0, because
0 tested and observed to be t
0
working
20 40 60 80 100 120

Just as R(t) is monotonic decreasing with time, F(t) is a

monotonic increasing function of time. As a unit operates
in time, the probability of failure increases.
Ebeling,. 2010
Probability Density Function (PDF)
15

• Related to R(t) and F(t) is the failure probability density

function (pdf), which is designated by f(t). (Recall pdf and
pmf)
• Note that f(t) is the slope of F(t) and the slope of -R(t). So
f(t) can be derived by taking the first derivative of F(t) or
R(t).
 Graph of a Pr Density Function (PDF)
16
f(t)
0.02

0.018

0.016

0.014

0.012

0.01

0.008

0.006

0.004

0.002

0
0 20 40 60 80 100 120

t
The area under a portion of a pdf between t1= 0 and t2 is the
probability of failure occurring within [t1= 0, t2]. This corresponds
to a point on the F(t) cdf. The area under the entire pdf equals 1.
Ebeling, 2010
Review: Relationship of PDF and CDF
17
• The cumulative failure and reliability distributions can be
derived by integrating the failure pdf.

T = time of failure

F(t) is the probability of a failure up to time t.

T = time of failure

R(t) is the cumulative probability of working at time t

and the cumulative probability of failing after time t.
 Exercise
18

• Some passive components of a distribution system

for natural gas have the following reliability
function:

Find: a. R(3 yrs)

b. The CDF, F(t)
c. Pr failure in (1<T<3) = P{1<T<3}
d. The probability density function, f(t)

Ebeling, 2010
19
 Mean Time to Failure (MTTF)
Critical for System Management!
20

• The MTTF is the mean of the pdf failure probability

distribution, f(t). MTTF is predicted by either of the following
equivalent methods (use the easier method!)
– Calculation of the mean time to failure T or expected
value of T, E(T), using f(t), the pdf failure distribution, as
the weight function for t.
– Integration (by parts) of the reliability function R(t) over all
T.
 Derivation of MTTF
21

Integration by parts:

because

Goes to 0 faster than t ∞

 Exercise: MTTF
22

• For the distribution system, find the MTTF using f(t).

Given, f(t) = t/50.

• Method 1:
Exercise - MTTF Revisited
23

• Method 2:
For the distribution system, find the MTTF using R(t).
Begin by finding F(t) from f(t):
 Median Time to Failure
24
• MTTF is a weighted average that is an acceptable point-
value representative of time to failure when the distribution
is symmetric, such as the Normal distribution, or a fairly
symmetric distribution.
• Median time to failure is the time for which 50% of the failures
occur before the median and 50% occur after the median
time to failure.
• Median time to failure is an acceptable point value
representative of failure times when the distribution is
skewed, where the median value is between the mode and
the mean.

Pr of failure for T ≥ median

Most Likely Time to Failure (Mode)
25
• The Mode or most likely time to failure is the value of t
that maximizes f(t), providing f(t) pdf has a single peak
• An unimodal f(t) (single peak), for any unit interval of
time, is centered at the mode, the probability of a
failure within that interval is greater than in any other
unit time interval.

• In each case of mean, median, or mode for a

particular distribution, the most representative point
value time to failure point value can be selected based
on the Skewness and Peakedness (Kurtosis) shape
of the distribution and the application.
Exercise: Central Tendency for Skewed Distribution
26
R(t) = 1 - t2 / 100 = 0.5 at the median

Solve for t at the median: F(t) = t2 /100 = 0.5

t2 = 50 or, tmedian = 7.07 yr

tmean = 6.67 yr (MTTF calculated before)

f(t)
by inspection
tmode = 10 yr
10 t
So, tmean < tmedian < tmode

This is an example for a distribution tail skewed to the left, where

tmean < tmedian < tmode and tmean weighted by left-skewed tail.
 Central Tendency for Skewed Distribution
27

Central measures for a Central measures for a

distribution tail skewed to the distribution tail skewed to the
right with tmode < tmedian < right with tmode < tmedian < tmean
tmean and tmean , which is and tmean biased by long tail to
biased by tail to the right. the right.

Ebeling, 2010
 Design Life
28
• The design life is defined as the time tR that corresponds
to a specified reliability value R that we may require for a
critical component, where R(tR) = R, where as t increases
to time tR, the reliability drops to R(tR).

• Example: Find the time, t0.99 such that R(t0.99) = 0.99

• t0.99 is the 0.99 probability or 99 percent design life.
• One percent of units are expected to fail by time t0.99

Use probabilistic language for an uncertain event, and do not state “will fail by”
as if the event time is known exactly.
 Variance and Standard Deviation
29
• Variance of t is the weighted average squared difference of a
time of failure, t, from the mean time of failure, MTTF, as shown
below:

• Equivalently by expansion of the above expression, variance is

the expected squared time of failure minus the square of the
expected time of failure, E(t2)-[E(t)]2, as shown in the
computational form:

= mean of square – square of mean

• The larger the variance, the less representative the mean,

median, or mode central measures, are of the failure time
distribution for a component
 Exercise - Variance for f(t) = t/50
30

Coefficient of Variation: unitless Dispersion relative to the Mean

cov = σ/μ = 2.36/6.66 ~ 0.354

Standard deviation, the square root of variance, is useful as a metric of

variability in addition to variance, because it has the same units as the
random variable.
Failure Rate, λ(t)
31

• Also called the hazard rate

• λ(t) is a conditional probability of failure rate for T > t
given the probability that the component is operating at
time t.
• λ is a function of t, λ(t). So it may increase, decrease or
stay constant with time.

𝑓𝑓(𝑡𝑡)
𝜆𝜆 𝑡𝑡 =
𝑅𝑅(𝑡𝑡)
 Failure Rate Function
(‘Hazard Rate Function’)
32

From
Obtain
P(T>t)

So

Form λ(t) by dividing by ∆t and taking the limit to dt:

Failure Rate Function λ(t) from f(t) = t/50
33

λ (t)
Example of a
IFR continuously increasing
failure rate (IFR) as the
component ages
t
10 yr
Note the rapid acceleration of failure rate for t approaching 10 yr.
Relation of Failure Rate and Reliability
34
• Begin with λ(t) and derive R(t)

𝒕𝒕
− ∫𝟎𝟎 𝝀𝝀 𝒕𝒕 𝒅𝒅𝒅𝒅
Integrate: 𝑹𝑹 𝒕𝒕 = 𝒆𝒆

Example: If
Review 1.2
R(t)
1 35
• R(t)=Pr(T>t) 0.8
0.6

• F(t) = Pr(T<t) = 1-R(t) 0.4

0.2
0

• 0 20 40 60 80 100 120

∝ ∝ ∝ 1.2
𝑑𝑑𝑑𝑑(𝑡𝑡) F(t)
• 𝑀𝑀𝑀𝑀𝑀𝑀𝑀𝑀 = � 𝑡𝑡𝑡𝑡 𝑡𝑡 𝑑𝑑𝑑𝑑 = � −𝑡𝑡
0 0 𝑑𝑑𝑑𝑑
𝑑𝑑𝑑𝑑 = � 𝑅𝑅 𝑡𝑡 𝑑𝑑𝑑𝑑
0
1
0.8

•
0.6
0.4
0.2

• 0
0 50 100 150

• 0.02 f(t)
0.015

• 0.01

0.005

0
0 50 100 150
 Probability Density Function
(PDF)
36

Pr of failure between a and b:

𝑏𝑏
P{a≤ 𝑇𝑇 ≤ 𝑏𝑏} = 𝐹𝐹 𝑏𝑏 − 𝐹𝐹 𝑎𝑎 = 𝑅𝑅(𝑎𝑎) − 𝑅𝑅(𝑏𝑏) = ∫𝑎𝑎 𝑓𝑓 𝑥𝑥 𝑑𝑑𝑑𝑑