MSTD: a framework for rolling bearing fault diagnosis based on multi-scale and soft-threshold denoising

In real industrial environments, the data collected is often subject to noise interference, which leads to challenges for traditional deep learning fault diagnosis models, including insufficient autonomous learning capabilities, complex feature extraction processes, and sensitivity to noise. To address these issues, this paper proposes a bearing fault diagnosis convolutional neural network based on multi-scale and soft-threshold denoising (MSTD). In this model, Poly Kernel Feature Fusion (PKFF) is utilised to extract high-quality features from raw vibration data, enhancing the model's ability to recognise complex fault patterns; Residual anchor attention (RAA) reduces redundancy and irrelevant information between channels, optimising inter-channel information to ensure the model's effectiveness when handling multi-channel data; Soft threshold CG (ST-CG) utilises a soft-threshold denoising method to suppress noise after feature extraction, thereby more clearly revealing fault characteristics. Soft-threshold processing not only enhances the model's noise resistance but also preserves critical fault information, enabling the model to maintain high diagnostic accuracy even under low signal-to-noise ratio conditions. The proposed model demonstrates outstanding performance on the CWRU and MFS-MG datasets, exhibiting superior feature extraction capabilities and robustness in high-noise environments compared to other models, significantly improving diagnostic accuracy and noise resistance.