Risk Assessment and Decision Analysis with Bayesian Networks

377

12

Systems Reliability Modeling

12.1 Introduction

Traditionally, assessing the reliability of a complex system has been based

on considering physical failures of hardware components. Although this

remains an important aspect of reliability assessment, it is becoming

increasingly important to also consider the impact of design faults. Design

faults are deviations from expected or intended requirements, and they

have the potential to trigger failures in operation and can dominate overall

reliability.

In this chapter we outline ways of assessing the reliability of complex

systems in terms of both their component failure behavior and also in

terms of the extent to which they contain design defects. We show how

Bayesian networks (BNs) can be used to model systems’ reliability, from

failure data collected during test or operation, to predict future systems’

reliability or to diagnose faults in such systems. Moreover, we also show

how we can take account of the structure of the system in the analysis,

especially the fault tolerance, redundancy, and other reliability enhanc-

ing methods used in the design.

We distinguish two broadly different scenarios when assessing a sys-

tem’s reliability:

1. Discrete (probability of failure on demand)—In this scenario

we are interested in reliability associated with a nite set of

uses of the system. For example, if the system is a military air-

craft carrying out a number of missions during a conict, then

we would be interested in the probability of failure for a given

mission. This scenario is discussed in Section 12.2.

2. Continuous (time to failure)—In this scenario we are interested

in reliability associated with a system operating in continuous

time. For example, for a military aircraft we would be inter-

ested in the number of miles it could y before a failure occurs.

This scenario is discussed in Section 12.3.

Section 12.4 gives an example of system fault monitoring (in real time

or near real time) using a dynamic Bayesian network (DBN) to monitor,

diagnose failures in, or control a working system.

The boundaries between hard-

ware and software are becoming

increasingly blurred in the complex

mechatronic (mechanical, elec-

tronic, electrical, and software) and

embedded systems used today.

Visit www.bayesianrisk.com for your free Bayesian network software and models in

this chapter

The literature on reliability mod-

eling is too large to do it justice

here, so we focus on examples of

using BNs for reliability modeling

for a limited number of commonly

encountered scenarios, including

dynamic fault trees (DFTs), a more

recent innovation.