Fault reconstruction approach for saturated dynamic systems using adaptive estimation and optimization

Modern industrial systems are becoming more complex and expensive, which have higher requirements on safety and reliability, and less tolerance on system performance degradation due to anomalies and faults. As a result, there is a strong motivation to address real-time monitoring and diagnosis techniques to detect a fault at an early stage and assess the severity of the faulty component. In this paper, real-time input and output data are used to monitor real-time and reconstruct faults for saturated dynamic systems using adaptive estimation and optimization techniques. Specifically, the saturation signals are partitioned into the conventional inputs without saturation and the beyond-saturation input signals. An unknown input decoupling approach is used to decouple unknown input uncertainty partially, and adaptive techniques are addressed to simultaneously reconstruct the additive faults and the beyond-saturation input signals. The estimator gains are obtained via the optimization for solving strict linear matrix inequalities to attenuate the effect from un-decoupled uncertainty to the estimation error dynamics. It is worth highlighting that the saturated input signals can also be reconstructed, which is particularly important when the saturated inputs cannot be measured directly, or it is costly to install extra sensors for the measurement. The proposed approach is to design offline, but implement using real-time input and output data, implying excellent real-time performance. The effectiveness of the proposed algorithm is demonstrated by an aircraft system and a single-link flexible joint robotic system.