Research Review on Federated Learning Technology for Fault Diagnosis

Federated learning (FL) is a distributed machine learning (ML) method. This technology only needs to exchange model parameters without sharing private data and plays an important role in industrial fault diagnosis. This paper focuses on analysing four mainstream technical schemes: fault diagnosis methods based on traditional federated learning, fault diagnosis methods based on federated deep learning, optimized federated fault diagnosis methods for nonindependent and identically distributed (non-IID) data, and lightweight federated fault diagnosis methods for edge device deployment. This paper systematically sorts out the federated learning technologies for fault diagnosis, summarizes the practical challenges existing in these technical schemes, and proposes corresponding future research prospects, providing specific references for the innovation and implementation of federated learning technologies for fault diagnosis.