Fault detection for uncertain sampled-data systems via deterministic learning

In this paper, an approach for rapid fault detection for a class of nonlinear sampled-data systems is proposed. Firstly, a learning estimator is constructed to capture the unknown system dynamics effects in sampled-data systems. The key issue in the learning process is that partial neural weights will converge into their optimal values based on the deterministic learning theory. Then a knowledge bank can be established, which stores the knowledge of various system dynamics effects, such as the Euler approximation modeling error, effect of the unstructured modeling uncertainty and different faults dynamics. Secondly, by utilizing knowledge bank, a set of estimators are constructed. The learned knowledge can quickly be recalled to compensate the unknown system dynamics effect. As a result, the occurrence of a fault can be rapidly detected. Finally, a rigorous analysis for characterizing the detection capability of the proposed scheme is given. Simulation study is included to demonstrate the effectiveness of the approach.