Bearing Fault Detection Based on Convolutional Self-Attention Mechanism

In view of the deficiency in traditional bearing fault diagnosis which depended on manual diagnosis and mechanical theory, the deep learning fault detection method based on bearing fault data-driven was studied. By analyzing the advantages and disadvantages of convolutional neural networks (CNNs), recurrent neural networks (RNNs) and self-attention mechanism, a rolling bearing fault detection method based on CSAM was proposed, which effectively integrated the powerful feature processing ability of CNNs and the local feature processing ability of self-attention mechanism. Bearing fault detection models based on CNNs, RNNs and CSAM are respectively constructed. Typical rolling bearing fault experimental data sets were selected to compare and analyze the effect of the CSAM method and the existing algorithm for the classification and identification of bearing faults. The results shown that the fault identification rate of CSAM method was 29.2% and 9.5% higher than that of CNNs and RNNs respectively in the test dataset on which the bearing fault was determined, which verified that the method can significantly improve the accuracy of bearing fault detection.