Deep transfer attention network for intelligent fault diagnosis of rolling bearings

Abstract

Deep learning based on fault diagnosis methods of rolling bearings has attracted much attention with the rapid development of artificial intelligence. However, these methods trained by bearing datasets from one equipment cannot be well applied to correctly identify the fault information of the bearing datasets from another equipment. The main reason for this problem is that the datasets from the different equipment have different probability distributions. In order to solve the above problem, a deep transfer attention network is proposed for intelligent fault diagnosis of the rolling bearings. The steps of this method are listed as follows. Firstly, the feature extraction network is used to extract the features from different bearing datasets. Secondly, the different transfer attention is given to different areas of the sample data by the attention mechanism and distribution differences among the features is reduced by domain adaptation. Finally, the fault recognition network is trained to identify fault status for different bearings. The proposed method is verified by using fault datasets from different bearings equipment. The results show that the proposed method can achieve the fault identification of different bearings equipment.