Bearing Domain Adversarial Fault Diagnosis Method Based on Wasserstein Distance Driven by Digital Twin

When solving the problems of data scarcity and cross-domain adaptability in bearing fault diagnosis, traditional methods often find it difficult to obtain a large amount of fault data, which limits the application and development of data-driven methods. Therefore, this paper proposes a novel domain adversarial fault diagnosis method driven by digital twins. First, by constructing a dynamic simulation model of bearing faults, the vibration signals of bearing faults are successfully simulated, thereby obtaining rich simulated fault data. Secondly, by using the SAE network and Wasserstein distance transfer strategy, the fault diagnosis knowledge is automatically learned from the simulated time domain signals, and this knowledge is successfully transfer to the diagnosis problems under different working conditions. Finally, by introducing the domain adversarial mechanism and gradient reversal layer (GRL), the transfer of target domain data features to source domain data features is realized. At the same time, the accurate classification of fault features and the adaptation of domain distribution are realized in the feature extraction process, which effectively solves the domain offset problem in cross-domain fault diagnosis.