Fault Diagnosis of Subway Traction Motor Bearings Under Variable Conditions Based on BA-VMD and SA-CNN Information Fusion

Traditional approaches for identifying bearing defects in metro traction systems often suffer from low diagnostic efficiency and accuracy. To address this, we propose an information fusion approach using the Bat Algorithm-Optimized Variational Mode Decomposition (BA-VMD) and the Self-Attention Convolutional Neural Network (SA-CNN). Vibration and acoustic emission signals are denoised via BA-VMD to optimize decomposition, followed by a diagnosis model utilizing attention-based fusion and SA-CNN to enhance key feature extraction. Experiments on subway traction motor bearings under varying operating conditions demonstrate the method’s efficacy. Results indicate that BA-VMD achieves a signal-to-noise ratio of 6.791, which is 1.595 higher than that of EMD (5.196). Furthermore, the SA-CNN model achieves an average diagnostic accuracy of 98.6%, significantly outperforming MLP (93.57%) and SVM (90.90%). These findings confirm that the proposed framework ensures accurate and stable bearing fault detection in highly variable operating conditions.