FEM Simulation Data-driven Fault Diagnosis Method for Bearing based on Improved Dynamic Adversarial Adaptive Network with Attention

In practical engineering, data has the characteristic of no label. Existing researches on bearing fault diagnosis without label mostly use test bench data to construct source domain, which is expensive. Therefore, FEM simulation data-driven fault diagnosis method for bearing based on improved dynamic adversarial adaptive network with attention is proposed in this paper. Firstly, finite element simulation methods are used to approximate operating conditions of target domain to obtain different health state simulation data, which are used as source domain data to reduce hardware resource dependence on fault simulation of the experimental platform. Then, dynamic adversarial adaptive networks based on efficient channel attention are constructed, and the attention mechanism and dynamic adversarial adaptive networks are used to achieve enhanced feature extraction ability of the model and sub-domain alignment, global alignment and fault identification of cross-domain samples, respectively. Finally, a large number of labeled finite element simulation samples and unlabeled data from the target domain are used to train the model for cross-domain diagnosis. The method validation was conducted on the bearing experimental dataset, and the results showed that the finite element simulation data can effectively replace the test bench data to achieve high-precision fault diagnosis in the target domain without labels. The cross-domain diagnosis ability of the proposed attention dynamic adversarial adaptive network is significantly better than that of the comparative method.