Data-Driven Passivity Analysis and Fault Detection Using Reinforcement Learning

This paper presents a novel approach for passivity analysis and hierarchical fault detection of passive systems employing model-free reinforcement learning (RL). The proposed method can analyze the passivity of a system without knowing or identifying the system model and furthermore construct a fault detection logic grounded on energy indicators from the analysis result. Initially, the data-driven Bellman optimality equation is formulated, which is equivalent to the system's passivity condition. Subsequently, the RL algorithm is delineated, and its time efficient advantage is elucidated in terms of both convergence and computational complexity. Simultaneously, the Bellman optimality equation in RL is clarified to be equivalent to the energy conservation constraint in the system. Based on this revelation, a hierarchical detection method based on the energy performance indicator is introduced. This approach can effectively detect faults within passive systems online and assess the severity of their consequences. The effectiveness of the proposed method is validated through simulation.