Fault diagnosis of unknown device based on dynamic model and domain generalization

The data-driven fault diagnosis of rolling bearings is widely used, but the methods often rely on a large amount of available data. However, in actual industrial processes, the operating conditions of bearings are complex and variable, and it is difficult to obtain the working condition data that needs to be diagnosed, which increases the difficulty of cross working device fault diagnosis. This article proposes a fault diagnosis method that combines simulation models with domain generalization, and establishes a new domain generalization network. Train the model using simulation models to generate bearing vibration data under target operating device, and generalize it to other actual operating devices. Combining knowledge distillation and adversarial methods in domain generalization networks, embedding the student network of knowledge distillation into the inter domain invariant feature extractor to extract intra domain and inter domain invariant features. The method has been experimentally validated to have high accuracy even under unknown device.