STFT-CBAM Hybrid Domain Adaptation for Bearing Fault Diagnosis Under Varying Conditions

Aiming at the distribution difference caused by the change of operating conditions in bearing fault diagnosis, a domain adaptive method based on STFT spectrum feature extraction and CBAM attention mechanism (STFT-CBAM-DAMS) is proposed in this paper. Firstly, short-time Fourier transform (STFT) was used to extract the spectral features, and power spectral density (PSD) was calculated for normalization to enhance the ability of feature expression. Then, data enhancement and signal sharpening techniques are used to optimize data quality. In the feature extraction stage, the residual network (ResNet) of CBAM attention mechanism is introduced, and the joint maximum average difference (JMMD) and conditional adduction domain adaptive (CDAN) + entropy weighting (E) strategies are combined to dynamically adjust the distribution difference between source domain and target domain, which significantly improves the cross-domain diagnostic capability of the model. The experimental findings demonstrate that this method exhibits outstanding fault diagnosis capabilities even under complex operating conditions. It offers a high-precision and highly robust solution for the intelligent operation and maintenance of industrial equipment.