Explicit and implicit Bayesian Network-based methods for the risk assessment of systems subject to probabilistic common-cause failures

Abstract

In an engineering system, multiple components may fail simultaneously due to a shared cause or common cause (CC). This kind of failure is referred to as a common-cause failure (CCF), and it contributes greatly to system risk. However, the propagation of a cause condition to the affected components is not always certain (deterministic). The propagation may be probabilistic due to many reasons, such as system specific defences against cause condition propagation. This kind of CCF is also called the probabilistic CCF (PCCF) which is defined as any condition or event that causes multiple components to fail or malfunction simultaneously with different occurrence probabilities. A system subject to PCCFs is sometimes affected by multiple CCs. This paper proposes explicit and implicit Bayesian Network (BN)-based methods to model systems subject to PCCFs considering different relationships among multiple CCs within a single model. The originality of this work lies in the use of BN to model the probabilistic dependencies among multiple CCs under different relationships and their affected components. Both methods are suitable for systems in which multiple CCs affect different components with different dependencies. Finally, the proposed methods are applied to model the risk of gas explosion in coal mines to evaluate the occurrence probability of accidents.