An improved multi-channel and multi-scale domain adversarial neural network for fault diagnosis of the rolling bearing

Abstract

To improve the fault diagnosis accuracy of rolling bearings under diverse working conditions, an improved domain adversarial neural network is proposed, the feature extraction module is reconstructed by multi-channel and multi-scale CNN-LSTM-ECA (MMCLE) in the proposed network. The MMCLE module consists of several key components. Firstly, the multi-channel multi-scale Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) are established to extract spatial features and temporal dependencies of the input data. Then, the Efficient Channel Attention (ECA) module is introduced to weight the effective feature channels. Finally, the domain adversarial training is employed to extract common features from both the source and target domains. By minimizing the domain offset between these domains, the faults of rolling bearing under diverse working conditions can be accurately diagnosed. The simulation results show that, based on the proposed MMCLE model, the domain offset issue can be effectively addressed, and the fault diagnosis accuracy can be improved for samples in the target domain under diverse working conditions. The accuracy and feasibility of the proposed method can be effectively verified.