A Bayesian network development methodology for fault analysis; case study of the automotive aftertreatment system

Abstract

This paper proposes a structured methodology for generating a Bayesian network (BN) structure for an engineered system and investigates the impact of integrating engineering analysis with a data-driven methodology for fault analysis. The approach differs from the state of the art by using different initial information to build the BN structure. This method identifies the cause-and-effect relationships in a system by Causal Loop Diagram (CLD) and based on that, builds the Bayesian Network structure for the system. One of the challenges in identifying the root cause for a fault is to determine the way in which the related variable causes the fault. To deal with this challenge, the proposed methodology exploits Dynamic Fault Tree Analysis (DFTA), CLD and the correlation between variables. To demonstrate and evaluate the effectiveness of the presented method, it is implemented on the data-driven methodology applied to the automotive Selective Catalytic Reduction (SCR) system and the obtained results have been compared and discussed. The proposed methodology offers a comprehensive approach to build a BN structure for an engineered system, which can enhance the system's reliability analysis.