Enhanced residual convolutional domain adaptation network with CBAM for RUL prediction of cross-machine rolling bearing

Abstract

Remaining useful life (RUL) prediction of rolling bearing is one of the important measures to ensure the reliable operation of mechanical equipment. Most of the existing methods are domain adaptation (DA) based RUL prediction on the same machine with different conditions, but few on cross-machine. DA can cope with the data distribution discrepancy (domain shift) under different machines or other conditions, but the potential negative transfer will affect the effect of DA and prediction performance. Therefore, an enhanced residual convolutional domain adaptation network (ERCDAN) is designed for cross-machine rolling bearing RUL prediction. Firstly, the enhanced residual convolutional module (ERCM) is designed for degradation feature extraction from limited data, and with the convolutional block attention module (CBAM) to enhance the extracted features. Secondly, the DA module with a collaborative full connection structure and attenuation multi-kernel maximum mean discrepancy is designed for mitigating negative transfer to effective domain-invariant feature extraction. Finally, the experimental analysis of cross-machine rolling bearing RUL prediction is conducted on the PHM2012, XJTU-SY, and EBFL datasets. The results show that the proposed method can not only effectively achieve cross-machine RUL prediction, but also has good cross-bearing prediction performance with different conditions on the same machine, reflecting good generalization performance.