A multi-source unsupervised fault diagnosis network with residual enhancement attention module for rotating machinery cross-operating conditions

Abstract

Domain adaptation methods based on average statistical metrics or single-source domains may encounter performance deficiencies of rotating machinery fault diagnosis. To this end, this paper proposes a multi-source domain adaptive network with the residual enhancement attention module (MDAN-REAM). Firstly, extracting feature information was performed for each combination of source and target domains by common feature extractor with the REAM. Secondly, domain-specific features were extracted by a domain adaptation method based on mean square statistics discrepancy (MSSD). Finally, fault diagnosis on the target domain was performed using all source domain classifiers. And the multi-classifier metric was applied to align the prediction discrepancies among all classifiers to improving fault diagnosis accuracy. Two experimental cases were designed to evaluate the proposed method. Experimental results demonstrate that the proposed method exhibits superior performance compared to many popular methods.