Research on Time-Frequency Fusion-Based Swintransformer and SE-CNN Dual-Channel Bearing Fault Diagnosis Method

Bearings are one of the important components with a high failure rate in rotating machinery, and their working state directly affects the performance of the entire equipment. Therefore, it is of great significance to diagnose bearing faults. However, since the bearing fault features are difficult to extract effectively, the diagnostic performance is not high. To address this problem, this study proposes a dual-channel bearing fault diagnosis method based on Swin-transformer and SE-CNN based on time-frequency fusion. This method constructs time-frequency fusion features by combining fast Fourier transform and variational mode decomposition, and uses it as the input of the dual-channel model. The Swin-transformer model is used to extract local spatial features, and the SE-CNN model is used to strengthen the key feature channel weights through the channel attention mechanism to form a complementary feature enhancement mechanism. Experiments on the CWRU dataset demonstrate that this method's model accuracy reaches 99.06%, which is 4.2%-12.7% greater than the Convolutional Neural Network, Transformer and other models. Its improved fault diagnosis and feature extraction capabilities support the advancement of intelligent bearing fault diagnosis research and practical implementation.