Federated learning architecture for bearing fault diagnosis

Federated learning (FL) is a distributed machine learning and it can obtain the participants, such as many companies or personal mobiles. The key point of federated learning is that it does not require the participants to send their data to the others, and FL only upload the gradients or weights of model trained by local data of each participant. Therefore, in this study, we combine the C fraction and gradient aggregation to implement the FL architecture for diagnosis of bearing fault. Finally, in our experiments, even if only a small number of clients participant in training, the testing accuracy can reach to 99 %. Furthermore, we use the number of turns to evaluate the impact of C fraction in testing.