Digital twins enable domain adaptive rolling bearing fault diagnosis

Abstract Aiming at the problems of large differences in signal feature distribution, serious noise interference and scarcity of fault samples leading to low diagnostic accuracy and poor generalization ability of the model for the same type of rolling bearings under different load conditions and structural parameters in industrial production, a digital twin-enabled domain adaptive fault diagnosis method based on MCNN-BiLSTM-Attention is proposed. Firstly, a rolling bearing digital twin model is established to generate rich twin data based on different bearing structures and fault parameters. Secondly, a transfer learning approach is used to solve the problem of inconsistent data feature distribution between the twin and the real fault data as well as the fault data of bearings with different structural parameters, and transfer fault features in different domains are extracted using a Multi-scale Convolutional Neural Network (MCNN) followed by a Bidirectional Long Short-Term Memory (BiLSTM) network with an attention mechanism,and the CORAL loss function, which takes into account the difference in the statistical distribution of the data features between the source domain and the target domain, is used for the adaptive operation in different domains during the network training process, and finally, the proposed methodology is evaluated on the rolling bearing fault data sets from Case Western Reserve University (CWRU) and Jiangnan University. The results show that, compared with other advanced models, the model has the highest accuracy in the task of transfer fault diagnosis of rolling bearings with different loading conditions and different structural parameters under different signal-to-noise ratio noise, which proves the effectiveness and feasibility of the model.