Research on Fault Diagnosis of Metro Traction Motor Bearings Based on Information Fusion and Optimized DRSN-GRU

Abstract Aiming at the current limitations of insufficient feature extraction and inadequate diagnostic accuracy in fault detection of subway traction motor bearings subjected to complex operating conditions, An intelligent fault identification approach integrating multi-source information fusion and optimized DRSN-GRU is developed in this study. The proposed method first converts vibration signals and acoustic emission (AE) signals into time-frequency graph via continuous wavelet transform (CWT), and achieves effective fusion of multi-modal signals using an attention-based feature fusion (AFF) mechanism. Subsequently, a hybrid model integrating a particle swarm optimization (PSO)-enhanced deep residual shrinkage network (DRSN) and gated recurrent unit (GRU) is constructed. Specifically, The DRSN framework incorporates a convolutional block attention module (CBAM), substantially enhancing its ability to extract crucial fault-related features. Finally, the collected bearing fault signals of metro traction motors were then fed into the developed diagnostic network for model training. The final results indicate that the diagnostic accuracy of fused signals is higher than that of single signals, the method proposed in this paper can effectively extract fault features under complex operating conditions and attains high accuracy in compound fault diagnosis.