A Bearing Fault Diagnosis Method Based on MS-TCN

To address the shortcomings of traditional fault diagnosis algorithms, including inadequate fault feature extraction and low identification accuracy, an end-to-end rolling bearing fault diagnosis method is proposed, utilizing a multi-stage temporal convolutional network (MS-TCN) that incorporates attention mechanisms and soft thresholding to enhance feature extraction and diagnostic accuracy. First, in the initial stage, a hybrid attention module (Convolutional Block Attention Module, CBAM) is employed to extract sensitive information related to fault characteristics, enhancing feature representation. In the refinement stage, a soft thresholding function is applied to retain critical features while eliminating noise-related ones. Subsequently, a multi-head attention mechanism is introduced to compensate for global relationship modeling. Experimental validation on the Case Western Reserve University dataset demonstrates that the proposed method achieves an average fault recognition rate of 99.8%, outperforming other models in classification accuracy. This approach holds significant value for bearing fault diagnosis.