Rolling bearing fault diagnosis method based on dual attention multi-scale convolutional network

To address the significant decline in the accuracy of rolling bearing fault diagnosis under variable working conditions and strong noise, which is caused by feature distribution shift and fault feature submersion, this paper proposes a bearing fault diagnosis method based on the Dual Attention Multi-Scale Convolutional Neural Network (DAMSCNN). First, a parallel multi-scale convolution module with multiple kernel sizes extracts fault features under different receptive fields, overcoming the limitation of fixed receptive fields in single-scale convolution. Second, a cascaded channel-spatial dual attention mechanism is designed following a progressive weighting strategy. A residual connection is embedded in the channel attention branch to highlight fault-sensitive channels and suppress redundant ones. The spatial attention branch fuses global average pooling and max pooling to strengthen fault impulse regions and suppress interference from working condition fluctuations and background noise. Finally, the weighted features are classified by a Softmax classifier. Experiments on the CWRU bearing dataset demonstrate that DAMSCNN achieves a peak diagnostic accuracy of 99.69% in cross-condition tasks and maintains accuracy above 92% under −5 dB SNR, outperforming BiTCN-Transformer, MTFC-T, and CTCA in accuracy, recall, and F1-score, while also offering lower computational cost and faster inference speed.