Control Change Cause Analysis-Based Bayesian Network Modeling for System Risk Assessment

The key problems in Bayesian network (BN) modeling are how to consider expert knowledge and select a structure learning algorithm for system risk assessment, because of the uncertainty and inaccuracy in the expert knowledge and high time-consumption in the structure learning for real systems. A control change cause analysis (3CA)-based BN modeling approach is proposed for system risk assessments in this paper. 3CA is used to identify the substances and causal relationships among different nodes in a new four-layer BN model, including cause, change, ineffective control, and risk layers. The four-layer BN model can be built by connecting the four-layer nodes with causal relationships, thus avoiding the use of a complicated structure learning algorithm. We use the proposed approach to perform risk assessment of a vehicle engine fuel supply system and obtain the assessment results by calculating the posterior probability of causes that lead to different operational risks. It can be used in various complex systems to perform their risk assessment conveniently and successfully.