A novel transfer learning method for bearing fault diagnosis under different working conditions

Transfer learning has attracted great attention in intelligent fault diagnosis of bearings under different working conditions. However, existing studies have the following limitation. (1) The metric of feature distribution discrepancy between different working conditions is not sufficiently domain adaptive. (2) The decision boundaries among different classes are not sufficiently clear in the target domain. To overcome the aforementioned limitations: (1) A fault transfer diagnosis model based on deep convolution Wasserstein adversarial networks (DCWANs) is proposed to handle the first limitation; (2) A variance constraint is developed for the DCWAN-based model to increase the aggregation of extracted features, which enlarges the margins among features of different classes in the source domain and also helps in feature extraction by adaptively aligning features by classes under different working conditions, thus, overcoming the second limitation. Experimental results showed that the proposed model achieves a higher fault diagnosis accuracy than existing models.

Automated summary

A fault transfer diagnosis model based on DCWANs was proposed to address limitations in existing studies, with a variance constraint developed to increase feature extraction aggregation and align features by classes adaptively. Experimental results demonstrated higher fault diagnosis accuracy compared to existing models.