A bearing fault diagnosis method based on an incremental broad federated learning system

Abstract To address the challenges of long global model training time and model forgetting in deep network-based federated learning methods, a novel federated learning method based on an incremental broad federated learning system is proposed. This method leverages random mapping as a bridge between broad learning and federated learning, constructing federated learning framework based on broad learning systems. This framework ensures local user data privacy while enabling random feature extraction from networks. A multi-user federated sparse feature extraction method combining random feature mapping with sparse autoencoders is proposed, and a broad global model is quickly constructed. An incremental learning method of broad federated learning is proposed, which can dynamically update the model parameters according to the new data of users. Experiments conducted on three datasets have demonstrated that the proposed method achieves outstanding classification performance across different scenarios.