CIVL 6960 Structural

Reliability
Lecture 16
03/18/24
CIVL 6960 Structural Reliability

10. SYSTEM RELIABILITY ANALYSIS

10.1 First-order Bounds on System Reliability
10.2 Failure Mode Approach

2
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

What Is System Reliability?

Up until now we have considered a single limit state function for describing the failure of our
structure:

For example:

Failure occurs when the external moment, 𝑀, due to 𝑄 exceeds the moment capacity, 𝑀! , of the
beam given by:

𝐺 𝑀! , 𝑀 = 𝑀! − 𝑀

Question: what about shear?

3
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

What Is System Reliability?

Even if we can describe the reliability of a member with a single limit state function, most
structures are composed of many members:

System reliability is concerned with the estimation of safety while considering

multiple limit sates/failure modes

4
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Component Reliability
We can imagine describing the reliability of a single component, e.g. , a truss element in tension,
through a single limit state function (single failure scenario)

Component
reliability

For assessing component reliability, we

have introduced a number of possible
approaches:

• Monte Carlo simulation

• Importance Sampling
• Second Moment methods
• First-Oder Reliability Methods (FORM)

5
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Question: if in most cases we are interested in the reliability of the “system”, why
focus on component reliability?

• Even for a single component, the calculation of its reliability is complex! Just imagine
what is going to happen if I now want to consider a system with several hundred
components!!!

• The reliability of a system is in general a function of the reliability of its components.

6
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 1: Series System

7
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 1: Series System

Failure of this system (𝐹 is a random load) will occur if
any one of its components fail (series system)!

Therefore calling:
𝐹"#" = system failure event
𝐹$ = component failure event

we have:

𝑃 𝐹"#" = 𝑃(𝐹% ∪ 𝐹& ∪ ⋯ ∪ 𝐹' ) (1)

Observation: the failure events, 𝐹$ , are generally

statistically dependent (complicates the evaluation of
Eq. (1))

For example, for this structure the axial stress in these

two members will be negatively correlated

8
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 2: Parallel System

9
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 2: Parallel System

Failure of this system will occur only if all of its components
fail (parallel system)!

Therefore calling:
𝐹"#" = system failure event
𝐹$ = component failure event

we have:

𝑃 𝐹"#" = 𝑃(𝐹% ∩ 𝐹& ) (2)

We can again observe the statistical dependence of the

failure events due to the mechanics of the problem

Observation: In general, our system will have components both in parallel as well as in series

10
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Summary
Conceptually system failure involves unions and intersections of component failure events.

We observed that in general the component failure events will be statistically dependent:

This means that in general we need more than the component failure
probabilities to assess the system failure probability

Example:
Consider a two-component series system with 𝐹% and 𝐹& the component
failure events. The probability of the system failure event 𝐹"#" is given by:

𝑃 𝐹"#" = 𝑃 𝐹% ∪ 𝐹& = 𝑃 𝐹% + 𝑃 𝐹& − 𝑃 𝐹% ∩ 𝐹&

Cannot in general be evaluated

from component probabilities

Question: is there any way to estimate the system level failure

probability knowing only the component failure probabilities?

11
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

10. SYSTEM RELIABILITY ANALYSIS

10.1 First-order Bounds on System Reliability
10.2 The Failure Mode Approach

12
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

First-order Bounds on System Reliability

Series System (Weakest Link Model)
3 failure mode example
𝐹"#" = system failure event 𝑆
𝐹$ = component failure event 𝐹#
we have:
𝐹!
𝐹"
𝑃 𝐹"#" = 𝑃(𝐹% ∪ 𝐹& ∪ ⋯ ∪ 𝐹( )

where 𝑚 is the total number of components (i.e.

limit states)

For series systems a first order lower bound is +

given by: 𝐹$ ⊆ 6 𝐹) = 𝐹"#"
)*%

𝑚
max 𝑃 𝐹) ≤ 𝑃(𝐹"#" )
)*%

3
𝑃 𝐹"#" = 𝑃 𝐹% ∪ 𝐹& ∪ 𝐹+ ≥ max 𝑃 𝐹)
)*%

13
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

First-order Bounds on System Reliability

Series System (Weakest Link Model)
3 failure mode example
For series systems a first order upper bound is given by:

Mutual Exclusive
𝑆
𝐹#
(

𝑃 𝐹"#" ≤ min 1, ; 𝑃(𝐹) ) 𝐹! 𝐹"

)*%

+ +

6 𝐹) = ; 𝐹)
)*% )*%

𝑃 𝐹"#" = 𝑃 𝐹% + 𝑃 𝐹& + 𝑃(𝐹+ )

14
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

First-order Bounds on System Reliability

Series System (Weakest Link Model)

The combination of the lower and upper bounds are known as the Frechet inequality:

(
𝑚
max 𝑃 𝐹) ≤ 𝑃(𝐹"#" ) ≤ min 1, ; 𝑃(𝐹) )
)*%
)*%

Observation 1:

It can be demonstrated that these are the narrowest possible bounds when only the
component probabilities are known

These limits can, at times, be too wide to be useful.

However, through physical considerations on the system they can often be narrowed.

15
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 1: Series System

The splicing of large reinforcing bars often is done using sleeves, as shown Figure 1.

Suppose each bar has a mean yield strength of 60 ksi with a COV of 0.15, and the mean capacity
of the sleeve is 140 kips with a COV of 0.20. The applied tension (𝑇) has a mean of 60 kips and a
standard deviation of 15 kips. The net bar area is 2.0 in& .

For the system described above, calculate, using second moment methods, the first-order bounds
on the system failure probability.

Assume Normal distributions and statistically independent resistances.

1 2

Figure 1

Note: the axial force (load effect due to the applied tension
𝑇) of each component is 𝑇

16
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 1: Series System

The system fails is bar 1 fails or sleeve fails or bar 2 fails. So, this system is a series system.

Let:
𝐹"#" = system failure
𝐹% = bar 1 failure
𝐹& = sleeve failure
𝐹+ = bar 2 failure

So, 𝑃 𝐹"#" = 𝑃(𝐹% ∪ 𝐹& ∪ 𝐹+ )

We are able to find the component failure probabilities with the second moment information of the
random variables related to each component’s load and resistance. However, we are not given
the statistical dependence information of the component failure events, so we cannot calculate the
exact failure probability of the system. That is why here we want to calculate the first-order
bounds on the system failure probability instead.

To calculate the first-order bounds on the system failure probability, we have to calculate the
failure probabilities of each component, i.e., 𝑃 𝐹% , 𝑃 𝐹& and 𝑃(𝐹+ )

17
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 1: Series System

Each component failure corresponds to a specific limit state function. Let’s calculate one-by-one:

For bar 1:
The component failure is defined as 𝐹%

The limit state function is defined as: 𝐺% = 𝑅% − 𝑆% = 𝑓# 𝐴 − 𝑇, where 𝑓# is the yield strength of the
bar and 𝐴 is the net bar area which is 2.0 in& . So, the second-moment information is:
𝜇,! = 60, 𝜎,! = 𝜇,! COV,! = 60×0.15 = 9
𝜇 - = 60, 𝜎- = 15

Since all random variables in this example are normal and 𝐺% is a linear limit state function, we
can find that 𝐺% is also normal with mean and standard deviation calculated by
𝜇." = 𝜇,! 𝐴 − 𝜇 - = 60×2 − 60 = 60

𝜎." = 𝜎,&! 𝐴& + 𝜎-& = 9& ×2& + 15& = 23.43

Using the second moment method, we can obtain the component failure probability of bar 1:
𝜇. 60
𝑃 𝐹% = 𝑃 𝐺% ≤ 0 = Φ − " = Φ − = 0.005223
𝜎." 23.43

Since the resistance and load effect of bar 1 and bar 2 are identical, the component failure
probability of bar 2 is also 0.005223, i.e., 𝑃 𝐹+ = 𝑃 𝐹% = 0.005223

18
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 1: Series System

Each component failure corresponds to a specific limit state function. Let’s calculate one-by-one:

For sleeve:
The component failure is defined as 𝐹&

The limit state function is defined as: 𝐺& = 𝑅& − 𝑆& = 𝐹# − 𝑇, where 𝐹# is the capacity of the
sleeve. So, the second-moment information is:
𝜇/! = 140, 𝜎/! = 𝜇/! COV/! = 140×0.2 = 28
𝜇 - = 60, 𝜎- = 15

Since all random variables in this example are normal and 𝐺& is a linear limit state function, we
can find that 𝐺& is also normal with mean and standard deviation calculated by
𝜇.# = 𝜇/! − 𝜇 - = 140 − 60 = 80

𝜎.# = 𝜎/&! + 𝜎-& = 28& + 15& = 31.76

Using the second moment method, we can obtain the component failure probability of bar 1:
𝜇. 80
𝑃 𝐹& = 𝑃 𝐺& ≤ 0 = Φ − # = Φ − = 0.005893
𝜎.# 31.76

19
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

Example 1: Series System

For series systems, the first-order bounds on the system failure probability is determined by

(
𝑚
max 𝑃 𝐹) ≤ 𝑃(𝐹"#" ) ≤ min 1, ; 𝑃(𝐹) )
)*%
)*%

Lower bound:
max 𝑃 𝐹) = max(0.005223, 0.005893, 0.005223) = 0.005893
)*% 12 (

Upper bound:
(

min 1, ; 𝑃(𝐹) ) = min(1, 0.005223 + 0.005893 + 0.005223) = 0.016339

)*%

0.005893 ≤ 𝑃(𝐹"#" ) ≤ 0.016339

20
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

First-order Bounds on System Reliability

Parallel System (Redundant Systems)
3 failure mode example
𝐹"#" = system failure event 𝑆
𝐹$ = component failure event 𝐹#
we have:
𝐹!
𝐹"
𝑃 𝐹"#" = 𝑃(𝐹% ∩ 𝐹& ∩ ⋯ ∩ 𝐹( )

where 𝑚 is the total number of components (i.e.,

limit states) 𝐹% ∩ 𝐹& ∩ 𝐹+

For parallel systems a first order upper bound is +

given by:
] 𝐹) ⊆ 𝐹$ for 𝑖 = 1, … , 𝑚
)*%
𝑚
𝑃 𝐹"#" ≤ min 𝑃 𝐹)
)*%

3
𝑃 𝐹"#" = 𝑃 𝐹% ∩ 𝐹& ∩ 𝐹+ ≤ min 𝑃 𝐹)
)*%

21
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

First-order Bounds on System Reliability

Parallel System (Redundant Systems)
To find a lower bound consider the probability of survival of the system 𝑃(𝑆"#" ):
1 − 𝑃 𝐹"#" = 𝑃 𝑆"#" = 𝑃(𝑆% ∪ 𝑆& ∪ ⋯ ∪ 𝑆( ) De Morgan’s law

𝐸% ⋂𝐸& … ⋂𝐸( = 𝐸% ∪ 𝐸& ∪ ⋯ ∪ 𝐸(

(
upper bound for a series system
Survival events 𝑃 𝑆"#" ≤ ; 𝑃(𝑆) )
(ignoring intersection)
$*%

Failure events 1 − 𝑃 𝐹"#" ≤ ; 1 − 𝑃(𝐹) )

$*%
(

𝑃 𝐹"#" ≥ max 0,1 − ; 1 − 𝑃(𝐹) )

)*%
(

𝑃 𝐹"#" ≥ 1 − ; 1 − 𝑃(𝐹) )
$*%

22
Dr. Min Li 3/18/24
CIVL 6960 Structural Reliability

First-order Bounds on System Reliability

Parallel System (Redundant Systems)
The combination of the lower and upper bounds provides the following bounds for parallel systems:

(
𝑚
max 0,1 − 𝑚 + ; 𝑃(𝐹) ) ≤ 𝑃 𝐹"#" ≤ min 𝑃 𝐹)
)*%
)*%

Observation 1: It can be demonstrated that these are the narrowest possible bounds when only the
component probabilities are known.

Observation 2: the upper bound indicates how the probability of failure of a redundant system is
smaller that the minimum component failure probability.

Observation 3: the lower bound is invariably equal to 0.

(
For positively correlated component
failure events the following lower bound c 𝑃 𝐹) ≤ 𝑃(𝐹"#" )
can be used instead: )*%

23
Dr. Min Li 3/18/24