Increasing Interpretability and Feasibility of Data Driven-Based Unknown Fault Diagnosis in Permanent Magnet Synchronous Motors

Data-driven methods often rely on historical data to establish diagnostic models, and model performance depends on the quantity and quality of data. However, historical data is difficult to access and of low quality in practical engineering applications. To solve the above problems, a device modeling-based method for unknown fault interpretability and feasibility enhancement of permanent magnet synchronous motors (PMSM) is proposed. First, a nonlinear fault contact force model coupled with time-varying contact deformation and time-varying stiffness is constructed. The variation rule of fault vibration response characteristics under different fault sizes is obtained. The method satisfies the requirement of reinforcement learning (RL) on the amount of data and improves the diagnosis accuracy under incomplete data. Second, a rotary evaporation strategy is proposed to eliminate the impact of knowledge memory and accurate modeling on diagnostic efficiency. The method better accepts the constraints between evaporation loss and bearing state category loss and improves the real-time performance of fault diagnosis. Finally, its performance is compared with existing advanced methods on a self-built experimental platform that includes multiple working conditions. The experimental results demonstrate the advantages of the method in terms of motor diagnostic performance and interpretability.