Federated Domain Generalization: A Secure and Robust Framework for Intelligent Fault Diagnosis

The maturation of sensor network technologies has promoted the emergence of the Industrial Internet of Things, which has been collecting an increasing volume of monitoring data. Transforming these data into actionable intelligence for equipment fault diagnosis can reduce unscheduled downtime and performance degradation. In conventional artificial intelligence paradigms, abundant individual data distributed across clients' devices needs to be delivered to a central storage for data analysis and knowledge extraction, which may violate data privacy requirements and neglect distribution discrepancy across different clients. To tackle the issue of privacy disclosure, an edge-cloud integrated federated learning framework is developed. Then, a two-stage training mechanism is designed to establish a domain-agnostic fault diagnosis model that can achieve satisfactory diagnostic performance on unseen target domains. Comprehensive simulated experiments on two rotating machines indicate that the proposed method possesses good generalization ability and can meet the requirement of privacy protection.