Self-Supervised Time-Frequency Alignment with Convolutional Block Attention Module

In various complex industrial scenarios, the collection of bearing fault data is expensive and time-consuming, making intelligent fault diagnosis a highly challenging task in scenarios with few labeled fault samples. To address this issue, we propose a self-supervised time-frequency alignment with convolutional block attention module (SSTFA-CBAM). This framework combines the characteristics of time-domain and frequency-domain signals, and achieves efficient feature extraction and transfer learning through self-supervised learning. Firstly, an overlapping segmentation strategy is adopted to increase the diversity of the original samples. Then, through fast Fourier transform and Gaussian noise enhancement, the data is input into frequency domain and time domain encoders respectively to extract deep features of unlabeled data. During this process, we utilized the convolutional block attention module (CBAM) to adaptively weight the features, enhancing the network's attention to key features. Finally, a small number of labeled samples are used to fine tune the classifier after the time-domain encoder, achieving accurate judgment of fault data. The SSTFA-CBAM framework significantly enhances the reliability of bearing fault diagnosis through precise feature extraction and self-supervised learning, providing robust data support for Predictive Health Management (PHM) systems and reinforcing the overall stability and safety of industrial systems.