CHAPTER

Reliability & Risk Analysis

for Systems

15
Probability, Statistics, and Reliability for
Engineers and Scientists, Third Edition
Bilal M. Ayyub and Richard H. McCuen
CHAPMAN
HALL/CRC
 Reliability of Systems
System analysis requires the recognition and
modeling of some system characteristics that
include:
1. post-failure behavior of a component
2. contribution of a component failure or
failure-mode occurrence to the system's
failure (Functional modeling)
3. statistical correlation among failure
modes and components' failure
4. definition of failure at the system level

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Systems in Series

1 2 3 ... n-1 n

n components

4 7

8 9 11
1 12
3 5

2 6 10 13

1 2 3 ... 12 13

13 components

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Reliability Computations for Systems in Series

Non-correlated and correlated performance functions for the

components.

Correlated Non-correlated

< Pf < 1-

n failure probability of the (1

= the ith component,
-P )( 1-P for)i( =1-1,
P 2, ...,
)( n.1-P )
max (P ) f f f fn
i=1 fi 1 2 n-1

P
f
i

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Example: Truss Reliability

Failure probabilities for components

0.01 0.01

0.005
0.03
0.01 0.01
0.01 0.03
0.01
0.02 0.02 0.02 0.02

System in series
max (0.01, 0.02, 0.03, 0.005) < Pf
1 - (1-0.01)6 (1-0.02)4 (1-0.03)2 (1-0.005) > Pf
0.03 < Pf < 0.1870

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Systems in Parallel

3
n components
.
.
.
n

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
Reliability Computations for Systems in Parallel

n < P < n
 Pf f min
(P )
i 1 i i=1 fi

P = the failure probability of the i component, for i =

th

1, 2,f..., n.
i

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
Example: Reliability of a Piping System in Parallel
Pipe 1, Failure Probability = 0.01

Pipe 2, Failure Probability = 0.03

Pipe 3, Failure Probability = 0.001

Pipe 4, Failure Probability = 0.02

A. Flow Input B. Flow Output

Pf = (0.01)(0.03)(0.001)(0.02) = 6x10-9 (non-correlated failures)

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Mixed Systems in Series and Parallel

In general, many engineering problems can be modeled using systems with components that are mixed in
parallel and series.

Example: Reliability of a Mixed Piping System

(non-correlated failures)

Pipe 1, Failure Probability = 0.01 Pipe 6, Failure Probability = 0.01

Pipe 2, Failure Probability = 0.03 Pipe 7, Failure Probability = 0.03
Pipe 3, Failure Probability = 0.001 Pipe 8, Failure Probability = 0.001

Pipe 4, Failure Probability = 0.02 Pipe 9, Failure Probability = 0.02

Pipe 5, Failure
Probability
A. Flow Input = 0.00001 B. Flow Output

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Pipes 1, 2, 3, and 4 Pipes 6, 7, 8, and 9
Failure Probability Failure Probability
= 6E-9 = 6E-9

Pipe 5, Failure
Probability
A. Flow Input = 0.00001 B. Flow Output

Pf = 1 - (1 - 6x10-9)(1 - 6x10-9)(1 - 0.00001)

= 0.0000100

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
Fault Tree Analysis
OR Gate

The reliability evaluation of

complex engineering AND Gate
systems requires analytical
methods that are systematic Event to be Decomposed Further
and methodical.
Basic Event

Fault tree analysis (FTA) is a

method that can be used for Undeveloped Event

this purpose.
Switch or House Event

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 System Definition
for a Dam
Functions Breakdown

Dam Dam System

Serviceability Safety Upstream and Downstream and

Requirements Requirements Dam Facility
Water Inflow Water Outflow

Dam Dam
Strength Flood Control
Water Drainage Structure Foundation Downstream
Release Precipitation Flood Plain
Basin Dams

Pool Water Structural/ Stability

Level Geotechnical Seismic Spillway Turbines
Integrity
Upstream Property
Flood Plain Downstream Dams History and Environment
Dams Faults
Flood Outlet Gates Population
Warning
River, Soil Moisture Equipment
Tributaries, Wildlife Vegetation
and other
Water Bodies Reservoir
Snow and Ice Soil

Earth Vegetation Initial Level Capacity

Adjacent to Adjacent to
Water Bodies Water Bodies

Slides Debris

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
Example: Reliability of a Mixed Piping System

Top Event:
No Flow from A to B

Piping Section I Piping Piping Section III

Section
II: Pipe 5

Pipe Pipe Pipe Pipe Pipe Pipe Pipe Pipe

1 2 3 4 6 7 8 9

P =P  P1P2 P3 P4P5P6 P7 P8 P9

f

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 P(S1) = P  P1P2 P3 P4 
 
= P(P1) P(P2) P(P3) P(P4)
= (0.01)(0.03)(0.001)(0.02)
= 6x10-9

Similarly, the failure probabilities of sections II

and III, respectively, are given by
P(S2)= 0.00001
P(S3)= P(S1) = 6x10-9

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Now, the occurrence probability of the top event can be
computed as:
Pf = P   S1  S2  S3 
= P(S1)  + P(S2)   + P(S3)  

- P(S1)P(S2) - P(S1)P(S3) - P(S2)P(S3)

+ P(S1)P(S2)P(S3)

Pf = 6x10-9 + 0.00001 + 6x10-9

- (6x10-9)(0.00001) - (6x10-9)(6x10-9)
- (0.00001)(6x10-9)
+ (6x10-9)(0.00001)(6x10-9)
= 0.0000100

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Event Tree Analysis

1. Event tree analysis is a method to study the effect of

an initiating event on a system.
2. The initiating event can be the failure of a component
in the system or the top event as was defined in the fault
tree analysis.
3. The effect of an initiating event on a system depends
on what might happen next and the sequence of
occurrence.
4. As a result, several possible scenarios can be
developed with possibly very severe impact on the
system.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Example: Nuclear Regulatory Commission, WASH 1400 study

Emergency
Electric Core Fission Containment
Pipe Break Power Cooling Product Integrity Tree Branch Probability Consequences
System Removal

Survival, 1-P(E4) P(E)(1-P(E1))(1-P(E2))(1-P(E3))(1-P(E4)) Very Small Radioactive Release

Survival
1-P(E3) Failure, P(E4) P(E)(1-P(E1))(1-P(E2))(1-P(E3))P(E4) Small Radioactive Release
Survival
1-P(E2) Survival, 1-P(E4) P(E)(1-P(E1))(1-P(E2))P(E3)(1-P(E4)) Small Radioactive Release
Failure
P(E3) Failure, P(E4) P(E)(1-P(E1))(1-P(E2))P(E3)P(E4) Medium Radioactive Release
Survival
Survival, 1-P(E4) Large Radioactive Release
1-P(E1) Survival P(E)(1-P(E1))P(E2)(1-P(E3))(1-P(E4))
1-P(E3) Failure, P(E4) P(E)(1-P(E1))P(E2)(1-P(E3))P(E4) Large Radioactive Release
Failure
Survival, 1-P(E4) Large Radioactive Release
P(E2) P(E)(1-P(E1))P(E2)P(E3)(1-P(E4))
Failure
P(E3) Failure, P(E4) Very Large Radioactive Release
Initiating Event P(E)(1-P(E1))P(E2)P(E3)P(E4)
P(E) Survival, 1-P(E4) Very Large Radioactive Release
Survival P(E)P(E1)(1-P(E2))(1-P(E3))(1-P(E4))
1-P(E3) Failure, P(E4) Very Large Radioactive Release
Survival P(E)P(E1)(1-P(E2))(1-P(E3))P(E4)
1-P(E2) Survival, 1-P(E4) P(E)P(E1)(1-P(E2))P(E3)(1-P(E4)) Very Large Radioactive Release
Failure
P(E3) Failure, P(E4) Very Large Radioactive Release
Failure P(E)P(E1)(1-P(E2))P(E3)P(E4)
P(E1) Survival, 1-P(E4) Very Large Radioactive Release
Survival P(E)P(E1)P(E2)(1-P(E3))(1-P(E4))
1-P(E3) Failure, P(E4) Very Large Radioactive Release
Failure P(E)P(E1)P(E2)(1-P(E3))P(E4)
P(E2) Survival, 1-P(E4) Very Large Radioactive Release
Failure P(E)P(E1)P(E2)P(E3)(1-P(E4))
P(E3) Failure, P(E4) P(E)P(E1)P(E2)P(E3)P(E4) Very Large Radioactive Release

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Assume
P(E) = 1E-04
P(E1) = 1E-07
P(E2) = 1E-05
P(E3) = 1E-06
P(E4) = 1E-05

P(Ei)(1-P(Ej))  P(Ei)
where P(Ei) and P(Ej) are small numbers.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
Computation of Branch Probabilities in Event Tree Analysis

Branch Branch Probability Expression Approximate Expression Branch

No. for Branch Probability Probability
1. P(E)(1-P(E1))(1-P(E2))(1-P(E3))(1-P(E4)) P(E) 1E-04
2. P(E)(1-P(E1))(1-P(E2))(1-P(E3))P(E4) P(E)P(E4) 1E-09
3. P(E)(1-P(E1))(1-P(E2))P(E3)(1-P(E4)) P(E)P(E3) 1E-10
4. P(E)(1-P(E1))(1-P(E2))P(E3)P(E4) P(E)P(E3)P(E4) 1E-15
5. P(E)(1-P(E1))P(E2)(1-P(E3))(1-P(E4)) P(E)P(E2) 1E-09
6. P(E)(1-P(E1))P(E2)(1-P(E3))P(E4) P(E)P(E2)P(E4) 1E-14
7. P(E)(1-P(E1))P(E2)P(E3)(1-P(E4)) P(E)P(E2)P(E3) 1E-15
8. P(E)(1-P(E1))P(E2)P(E3)P(E4) P(E)P(E2)P(E3)P(E4) 1E-20
9. P(E)P(E1)(1-P(E2))(1-P(E3))(1-P(E4)) P(E)P(E1) 1E-11
10. P(E)P(E1)(1-P(E2))(1-P(E3))P(E4) P(E)P(E1)P(E4) 1E-16
11. P(E)P(E1)(1-P(E2))P(E3)(1-P(E4)) P(E)P(E1)P(E3) 1E-17
12. P(E)P(E1)(1-P(E2))P(E3)P(E4) P(E)P(E1)P(E3)P(E4) 1E-22
13. P(E)P(E1)P(E2)(1-P(E3))(1-P(E4)) P(E)P(E1)P(E2) 1E-16
14. P(E)P(E1)P(E2)(1-P(E3))P(E4) P(E)P(E1)P(E2)P(E4) 1E-21
15. P(E)P(E1)P(E2)P(E3)(1-P(E4)) P(E)P(E1)P(E2)P(E3) 1E-22
16. P(E)P(E1)P(E2)P(E3)P(E4) None 1E-27

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
Observations
1. Branch 1 has the largest occurrence probability, but it has a
small consequence
2. Branches 8 to 16 have the highest level of consequence
3. Branch 9 has the highest occurrence probability
4. Branch 9 can be considered as the highest risk branch
5. The occurrence probability of Branch 9 can be reduced by
improving the failure probability of the pipe breakage (P(E)), or
improving the electric power failure probability (P(E1)).
6. Options of system improvement can be identified that affect
these two events.
7. Cost-benefit analysis using decision trees can be utilized to
select the best option.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Risk-Based Decision Analysis
 The risk of an event can be defined as a combination
of both its occurrence probability and its occurrence
consequence.
 The combination can be in the form of their product.
 Alternatively, it can be considered to be an ordered
pair of occurrence probability and its occurrence
consequence, i.e., (probability, consequences).
 For several events of interest, risk plots can be
produced using these ordered pairs for the events and
a coordinate system of occurrence probability and
consequence.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Lines of
Constant Risk

Increasing Risk
Likelihood

Consequence

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Risk Methods

Risk Methods

Risk
Risk Management
Communication

Risk Risk Control:

Decision Making Engineering
Assessment Monitoring Community

Media and
Risk Analysis: Risk Evaluation: Public
Hazard Identification Risk Acceptance
Risk Estimation Option Analysis

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 A decision model is a systematic framework
for decision making in risk analysis. In order to
construct a decision model, the following
elements of the decision model need to be
defined:

1. objectives of decision analysis,

2. decision variables,
3. decision outcomes, and
4. associated probabilities and consequences.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Objectives of Decision Analysis

 Engineering decision problems can be classified

into single- and multiple-objective problems.
 Example objectives are minimizing the total
expected cost, maximizing safety, maximizing
the total expected utility value, and maximizing
the total expected profit.
 Decision analysis requires the definition of
these objectives.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
Decision Variables
 The decision variables are the feasible options or
alternatives available to the decision maker at any stage
of the decision-making process.
 Ranges of values that can be taken by the decision
variables should be defined.
 Example decision variables can include, for example,
what and when to inspect components or equipment,
which inspection methods to use, assessing the
significance of detected damage, and repair/replace
decisions.
 Assigning a value to a decision variable means making a
decision at that point of a decision-making process.
These points within the decision-making process are
called decision nodes( ).
Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Decision Outcomes
 The decision outcomes are the events that can happen as a
result of a decision.
 They are random in nature, and their occurrence cannot be
fully controlled by the decision maker.
 Example decision outcomes can include, for example, the
outcomes of an inspection (detection or non-detection of a
damage), and the outcomes of a repair (satisfactory or non-
satisfactory repair).
 Therefore, the decision outcomes with the associated
occurrence probabilities need to be defined. The decision
outcomes can occur after making a decision at points within
the decision-making process called chance nodes ( ).

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Associated Probabilities and Consequences

 The decision variables take values that can have

associated costs. These costs can be considered as the
direct consequences of making these decisions.
 The decision outcomes have both consequences and
occurrence probabilities. The probabilities are needed
due to the random (chance) nature of these outcomes.
 The consequences can include, for example, the cost of
failure due to damage that was not detected by an
inspection method.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Decision Trees

 The elements of a decision model need to be

considered in a systematic form in order to make
decisions that meet the objectives of the
decision-making process.
 Decision trees are commonly used to examine
the available information for the purpose of
decision making.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Example: Decision Analysis in Inspection Planning

Branch Cost
01:Detection
P(O 1):.25
A1:Visual Inspection C(A 1)+P(O 1)*C(O 1)+P(O 2)*C(O 2)
C(O 1 ):$10/FT
= $40.5
C(A 1):$0.5/FT 02:Non-Detection
P(O 2):.75
C(O 2 ):$50/FT

03:Detection
P(O 3):.4
A2:Dye Penetrant Test C(O 3):$10/FT C(A 2)+P(O 3)*C(O 3)+P(O 4)*C(O 4)
C(A 2):$1.0/FT = $35.0
04:Non-Detection
P(O 4):.6
Test/Inspect C(O 4):$50/FT
Ships Butt
Welds
05:Detection
P(O 5):.6
A3:Magnetic Particle Test C(O 5):$10/FT C(A 3)+P(O 5)*C(O 5)+P(O 6)* C(O 6)
C(A 3):$4.0/FT = $30.0
06:Non-Detection
P(O 6):.4
Key: C(O 6):$50/FT

= Decision Node
= Chance Node
P( ) = Probability 07:Detection
C( ) = Cost of ( ) P(O 7):.7
TC( ) = Total Cost A4:Ultrasonic Test C(O 7):$10/FT C(A 4)+P(O 7)*C(O 7)+P(O 8)*C(O 8)
Ai = Alternativei
C(A 4):$15.0/FT = $37.0
Oj = Outcomej
08:Non-Detection
P(O 8):.3
C(O 8):$50/FT

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 Problems
For the following performance function, determine
the safety index () using (a) the first-order second-
moment method, and (b) the advanced second-
moment method:
Z =X - X - X
1 2 3
The non-correlated random variables are assumed
to have the following probabilistic characteristics:

Random Mean Coefficient of Distribution

variable value variation Type

X1 1 0.25 Normal
X2 5 0.05 Normal
X3 4 0.2 Normal

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 For the following performance function, determine the
safety index () using (a) the first-order second-moment
method, and (b) the advanced second-moment method:

Z =X - X - X
The non-correlated random
1 2 variables
3 are assumed to
have the following probabilistic characteristics:

Random Mean Coefficient of Distribution

variable value variation Type

X1 1 0.25 Lognormal
X2 5 0.05 Lognormal
X3 4 0.2 Lognormal

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 The change in the length of a rod due to axial force P is
given by: PL
L 
AE

where L = length of rod, P = applied axial force, A = cross sectional

area of rod, and E = modulus of elasticity. Assume that the change
in length of the rod must not exceed 0.15 in. Using 2,000 simulation
cycles in (a) the direct Monte Carlo simulation method, and (b) the
conditional expectation method, determine the failure probability of
the rod in meeting this limit. Assume the non-correlated random
variables to have the following probabilistic characteristics:

Random Mean value Coefficient Distribution

variable of variation Type

P 100 kips 0.35 Lognormal

L 20 in 0.05 Normal
E 30,000 ksi 0.10 Lognormal
A 1 in2 0.05 Normal

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 For the rod in the previous problem, study the
effect of increasing the number of simulation
cycles on the estimated failure probability. Use
the following numbers of simulation cycles: 20,
100, 500, 1000, 2000, 10,000, and 100,000.
Provide your results in the form of plots of
estimated failure probability, and its coefficient of
variation as a function of the number of simulation
cycles.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen
 References
1. Ayyub, B.M., and McCuen, R., Probability, Statistics and
Reliability for Engineers, CRC Press, FL, 1997.
2. Hart, "Uncertainty Analysis, Loads, and Safety in Structural
Engineering," Prentice Hall, NY, 1982.
3. Ang, and Tang, "Probability Concepts in Engineering
Planning and Design," Volume II, John Wiley and Sons,
NY, 1982.
4. Melchers, "Structural Reliability Analysis and Prediction,"
Ellis Horwood Limited, UK, 1987.
5. Madsen, Krenk, and Lind, "Methods of Structural Safety,"
Prentice Hall, Englewood Cliffs, NJ, 1986
6. Thoft-Christensen, and Baker, "Structural Reliability Theory
and Its Applications,“ Springer-Verlag, New York, 1982.
7. Thoft-Christensen, and Murotsu,
"Applications of Structural Systems
Reliability Theory," Springer-Verlag,
Berlin, 1986.

Probability, Statistics and Reliability for Engineers and Scientists, Third Edition, Bilal M. Ayyub and Richard H. McCuen