Representing parameterised fault trees using Bayesian networks

Fault trees are used to model how failures lead to hazards and so to estimate the frequencies of the identified hazards of a system. Large systems, such as a rail network, do not give rise to endless different hazards. Rather, similar hazards arise repeatedly but with different frequency depending on factors such as location. Several authors have identified the need to build models to estimate both system-wide average hazard frequencies and hazard frequencies in specific situations. Fault trees can be used for this but they grow as additional factors are considered. In this paper, we describe a compact model using Bayesian networks. The fault tree notation is retained; with base events parameterised by variables in the Bayesian net to represent a mixture of related fault trees compactly. We use a simple example to describe the model structure and report on ongoing work on a model of train derailment.