Asynchronous Federated Broad Learning System for Intelligent Fault Diagnosis of Wind Turbine

Purpose: Intelligent fault diagnosis (IFD) is crucial for ensuring the operation safety of wind turbines. However, the problem of data scarcity in individual wind farm and the concern of data privacy across multiple wind farms have posed significant challenges for the engineering application. Federated learning offers an effective method by aggregating multiple client-side datasets for global fault diagnosis modeling while ensuring data privacy. Current federated learning based IFD methods have the two following shortcomings: (1) The time-consuming synchronous aggregation of client updates. (2) The issue of global performance degradation during client aggregation. To overcome the above-mentioned shortcomings, this paper proposes an asynchronous federated broad learning system (AFBLS) model for IFD of wind turbines. Methods: Firstly, independent IFD models on client sides are established based on broad learning systems by using the data of each client. Then, an asynchronous federated aggregation method for client models is developed to achieve rapid online incremental updates of the global diagnosis model without data leakage. Finally, an Adaptive Aggregation Selection strategy is developed to mitigate the precision degradation caused by the outdated aggregations problem, so as to strengthen the global diagnosis model continuously. Results: The feasibility and effectiveness of the proposed AFBLS model are validated by two cases of blade icing detection and main bearing wear diagnosis. The experiments prove that the proposed AFBLS can improve the results of the aggregation model. Conclusion: The research show that the proposed AFBLS has the capacities of integrating new client diagnosis models asynchronously and improving the diagnosis precision of the global model over time.