A decentralized fault detection and prediction scheme for nonlinear interconnected continuous-time systems

Complex nonlinear systems such as an aircraft, trains, automobiles, power plants and chemical plants are represented as nonlinear interconnected subsystems. Therefore, in this paper a novel decentralized fault diagnosis and prognosis (FDP) methodology is proposed for such large-scale systems. Current FDP approaches require the knowledge of the entire state or its estimated vector. But the main goal in this work is to design a local fault detector (LFD) or observer for each subsystem based on the measured local states of the subsystem alone. A local residual signal is generated via the measured states of the local subsystem and the estimated states provided by the LFD. A fault is detected when this local residual exceeds a predefined threshold. The adaptive online approximator in each LFD is activated upon detection to compensate the fault dynamics due to local and non-local faults. A novel update law for tuning the parameters of the online approximator is derived. Upon detection, faults local to the subsystem and to other subsystems are isolated. In addition, the proposed scheme provides the time to failure (or remaining useful life) information by using local measurements and the parameter update law of the LFD. Simulation results verify the effectiveness of the proposed decentralized FDP scheme.