An Interpretable Bearing Fault Diagnosis Method Based on Belief Rule Base With Interval Structured

Rolling bearings are crucial components in industrial applications, forming the core of modern rotating machinery. The belief rule base with interval structure (IBRB-r) is a rule-based expert system designed to describe the causality of bearing faults. It is mainly applicable in situations where data within the intervals are relatively concentrated. However, in practical industrial applications, bearing operation data is often dispersed. This leads to uneven data distribution within the reference interval, affecting the model’s accuracy. Additionally, the interpretability of the parameters decreases after the IBRB-r model is optimized. To address these issues, an adaptive rule matching interval structured belief rule (IBRB-Di) is proposed for bearing fault diagnosis. First, a rule matching method that can be dynamically computed within an reference interval is introduced. Also, a new computation method for the weights of rules activation within an interval is presented. These methods adapt to the dispersion of data and simplify the number of rules. Second, the projected covariance matrix adaptive evolution strategy with expert constraints (P-CMA-ES) is adopted to optimize the model. This aims to balance accuracy and interpretability. Finally, the method is experimentally validated using the Case Western Reserve University bearing dataset. This verification confirms the effectiveness of the proposed method.