Solving dynamic fault trees using a new Hybrid Bayesian Network inference algorithm

Abstract

We present a hybrid Bayesian network (HBN) framework to analyse dynamic fault trees. By incorporating a new approximate inference algorithm for HBNs involving dynamically discretising the domain of all continuous variables, accurate approximations for the failure distribution of both static and dynamic fault tree constructs are obtained. Unlike in other approaches no numerical integration techniques or simulation methods are required. Moreover, no exact expression for the posterior marginal is needed and no conditional probability tables need to be completed. Sensitivity analysis, uncertainty, diagnosis, common cause failure analysis, can all be easily performed within this framework. Posterior estimates of parameterised marginal failure distributions can also be obtained using available raw failure data together with prior information from expert judgement.