Intelligent Fault Diagnosis of Rolling Bearings Based on Attention Mechanism and Joint Correlation Alignment

Due to the data distributions in the source domain (SD) and the target domain (TD) are significantly different, many techniques have been developed to adapt to different domains. However, in the existing domain adaptation (DA) methods, the influence of domain samples on network weight has not been considered. Moreover, most of these methods only focused on marginal distribution alignment (MDA), while the class conditional distribution alignment (CDA) has been ignored. In view of the above shortcomings, an adaptive network based on attention mechanism and joint correlation alignment (AM-JCA) is proposed in this paper. The attention mechanism is utilized to extract crucial fault feature information. The joint correlation alignment method is used to effectively reduce the difference of marginal distribution between two domains, and further minimize the difference of conditional distribution between different categories. In addition, in order to learn more separable fault features, a new N-Softmax loss is proposed, which has a stronger classification ability than the traditional softmax. Finally, experimental results show that the proposed method has significant effects and superiority.