Stack Autoencoder Transfer Learning Algorithm for Bearing Fault Diagnosis Based on Class Separation and Domain Fusion

Abstract

In intelligent fault diagnosis, transfer learning can reduce the requirement of sufficient labeled data and the same data distribution. However, for the diagnosis of a new machine, there are still some limitations, such as low accuracy or the demand for some labeled data with fault information in the new machine. In this article, we propose a stack autoencoder transfer learning algorithm based on the class separation and domain fusion (SAE-CSDF) to solve these problems. According to the characteristics of bearing faults, the proposed weighted domain fusion strategy can ensure the direction and balance in the transfer process. The proposed class separation degree can improve the accuracy of the target domain indirectly by extending the differences between the classes in the source domain. The effectiveness of the SAE-CSDF is verified via the mutual transfer of two public datasets and one laboratory dataset. The results show that the accuracy of the algorithm can reach 97% in the transfer between different machines, even if there is no labeled fault data in the new machine.