A health management modeling method for complex electromechanical systems based on a Huffman tree-structured belief rule base

Abstract The traditional belief rule base (BRB) expert system often suffers from combinatorial rule explosion in complex electromechanical system health management, where the number of rules grows exponentially with feature dimensionality. While hierarchical reduction methods have been applied to mitigate this issue, improper feature partitioning often compromises inference accuracy and model interpretability. To address these limitations, this study introduces a multilayer tree-structured BRB expert system that integrates the concept of a Huffman tree for self-organizing hierarchy construction. The framework jointly considers feature importance and correlation, assigning high-weight, low-redundancy features to upper layers close to the final output, while relegating low-weight, high-redundancy features to lower layers. This design effectively compresses the rule scale and enhances decision-making traceability. Validation on the National Aeronautics and Space Administration milling dataset demonstrates that the proposed method significantly mitigates combinatorial explosion while improving prediction accuracy and interpretability. Generalizability is further confirmed through experiments on the Commercial Modular Aero-Propulsion System Simulation turbofan engine dataset, underscoring its applicability to health management tasks for complex electromechanical systems.