Intelligent fault diagnosis of rolling bearing based on a deep transfer learning network

Abstract

Rolling bearing of rotating machinery's key component will inevitably fail due to the complex and changeable operating environment such as variable speed, large disturbance, high and low temperature. It is quite challenging to obtain abundant labeled bearing fault samples because the rotating machinery is typically in a healthy and operational state. For addressing the issue, an intelligent fault diagnosis method based on a deep transfer learning network is proposed. First, a bidirectional gated recurrent unit (Bi-GRU) network is utilized to mine the latent relationship between labeled source domain samples and few labeled target domain samples, the parameters of Bi-GRU are trained to obtain the instance transfer bidirectional gated recurrent unit model (ITBi-GRU), and auxiliary samples are generated based on the ITBi-GRU. Second, as a feature transfer learning method, joint distribution adaptation is used to simultaneously decrease the distribution discrepancies between the generated auxiliary samples and the unlabeled target domain samples. Finally, extensive experiments are employed to evaluate the effectiveness of the proposed method in the case of scarce labeled samples.