A Domain Generalization Network via Meta-Learning for Motor Fault Diagnosis Under Unseen Working Conditions

Intelligent motor fault diagnosis has emerged as an effective tool to ensure reliability and safety of industrial equipment. However, as motors always work under varying conditions, intelligent fault diagnosis methods may fail when facing an unseen working condition. To tackle this problem and improve the generalization ability of model, a domain generalization network based on meta-learning is constructed for motor fault diagnosis. A synthetic testing domain is split from source domains, and meta-train and meta-test are performed on the training and testing domains, respectively. The meta-learning process aims to ensure that the model has good generalization on unseen domains. Experiments on multiple cross-condition motor fault diagnosis tasks verify the effectiveness of the proposed method.