An Adaptive Accuracy Threshold Aggregation Strategy Based on Federated Learning

Fault diagnosis can be used to identify the type and severity of fault accurately and automatically. If the data is transmitted to the cloud, it will increase the risk of data security and computing. Therefore, the demand for artificial intelligence operations on edge network devices is increasing day by day. Because federated learning mechanism does not need to transmit information data, it is more suitable for edge network machine learning scenarios with limited data. Therefore, this paper proposes a bearing fault diagnosis method based on Federated learning. In the model aggregation, according to the accuracy threshold adaptive algorithm, high-quality local models are selected to participate in the aggregation, so as to reduce the number of communication. The experimental results show that when the communication times are reduced to 77.6%, the accuracy is only reduced by 2.8%.