Fault Diagnosis in Satellite System: A Federated Learning Approach with Communication Constraints

Satellites operate in harsh space environments for extended periods, making fault diagnosis a critical strategy for extending their lifespan. The federated learning fault diagnosis method based on telemetry data is a novel approach in the field of satellite fault diagnosis. This method can both address the data privacy issues across different TT&C centers and improve the poor fault diagnosis performance caused by the low quality of the training data set of a single TT&C center. However, challenges arise in the deployment and application of federated learning methods due to "Communication-limited" and "Extreme Heterogeneity" at some TT&C centers. To solve such a problem, this paper proposes a streaming federated learning framework for satellite fault diagnosis-FedSAT, with efficient communication and privacy protection. Under this framework, the TT&C center with communication constraints only needs to participate in a single round of federated learning for the system to obtain a well-performing and stable model. Simulations verify the effectiveness of the proposed framework.