Intelligent diagnosis method for machine faults based on federated transfer learning

Intelligent fault diagnosis model based on federated learning can effectively solve the problem of fault data privacy and sharing, and ignores the difference of fault data distribution. Transfer learning can avoid the difference of data distribution. Combining the advantages of transfer learning and federated learning, a fault diagnosis model with data privacy based on federated transfer learning is proposed to achieve cross-domain fault diagnosis without sharing the data. In the constructed model, the local models on the client are firstly established based on deep convolution neural network to extract the feature of the source and target domains. The alignment loss is introduced to minimize the similar feature distribution differences among different source domains and target domain. The parameters of local models are fused and updated to generate a global model, which can not only identify the fault types in the target domain, but also retain the ability to recognize the fault types in the source domain. The two experiments, including different bearings with same feature distribution and label and the bearing and planetary gear in the same transmission system with similar feature distribution, are used to verify the effectiveness of the proposed model. The experiments suggest that the fault diagnosis model based on federated transfer learning can reduce the difference of the newly added fault type data distribution, and can accurately recognize the fault data of the source domain and target domain. Compared with the traditional diagnosis model based on deep learning, transfer learning and federated learning, the proposed model can effective perform the cross-domain fault diagnosis with data privacy.