Federated Learning for Efficient Condition Monitoring and Anomaly Detection in Industrial Cyber-Physical Systems

The proliferation of Internet of Things (IoT) devices and To address these challenges, we propose an approach leveraging autonomous systems within cyber-physical systems (CPS) has federated learning (FL) for both anomaly detection and heightened the importance of anomaly detection and localization localization in CPS. While FL offers a foundation for distributed in industrial component health monitoring [1]. Modern model training [1], [3], [5], basic FL implementations CPS, encompassing smart grids and industrial control systems, face several limitations in CPS environments, including (1) generate vast amounts of data from numerous sensors and inability to handle varying sensor reliability and data quality, actuators [2]. With millions of machine failures occurring (2) vulnerability to node failures and subsequent data loss, (3) globally each year, the financial losses due to downtime and inefficient resource utilization across heterogeneous nodes, and repairs are substantial. Early detection and precise localization (4) limited adaptation to dynamic operational conditions. of sensor anomalies are crucial to minimize these losses and Our framework enhances traditional FL approaches through prevent cascading failures.