Mbrb: Micro-Belief Rule Base Model Based on Cautious Conjunctive Rule for Interpretable Fault Diagnosis

Abstract

In aerospace engineering, medical diagnosis and other fields, the implementation of interpretable fault diagnosis (IFD) technology is critical for improving the credibility of diagnosis results and ensuring the reliability of complex electromechanical systems. As a typical interpretable modeling method, the belief rule base (BRB) approach still has two problems that must be solved for application to IFD: 1) complex rule structures increase the cost of model interpretability, and 2) over-optimization of parameters diminishes the interpretability. To solve these problems, this paper proposes a modified micro-belief rule structure and develops a new micro-belief rule base (MBRB) model based on this structure. In addition, considering the independence of rules and the correlations between multiple BRBs, a cautious conjunctive rule-based reasoning process is established as the inference engine of MBRB. Moreover, an interpretable optimization method based on projection covariance matrix adaptive evolutionary strategy (PCMAES) is proposed for the MBRB model to balance model interpretability and diagnostic accuracy. Finally, the availability and effectiveness of the proposed MBRB is verified through an electromagnetic relay experiment.

Keywords: Belief rule base, interpretability, fault diagnosis, cautious conjunctive rule, projection covariance matrix adaptive evolutionary strategy