Efficient belief rule-based network for planetary gearbox wear state characterization using multi-channel lubricant debris information

Real-time gearbox wear state characterization through online oil analysis technology is crucial in improving system reliability and safety. However, the information conflict in multi-channel data of ferromagnetic wear particles and the low computational efficiency of existing methods hinder its practical implementation in safety–critical industries such as nuclear power plants. To this end, this paper proposes an Efficient Belief Rule-Based Network (EBRBNet). The structure of EBRBNet is established based on the belief rule-based theory, suited to efficiently fusing multi-channel abrasive information and dealing with the problem of information conflict in an interpretable way. Moreover, the parameter softmax normalization is developed to prevent gradient explosion, and a gradient-based optimization framework is developed to improve computational efficiency in EBRBNet. The performance of EBRBNet is evaluated using an online gearbox oil monitoring dataset and publicly available datasets. The results indicate that the method achieves interpretable information fusion and higher accuracy in wear state characterization.