Fault Diagnosis for a Class of Sampled-data Systems via Deterministic Learning

摘要

In this paper, an approach for rapid detection of small oscillation faults for a class of sampled-data nonlinear system is proposed based on deterministic learning theory. Firstly, based on the Euler approximate discrete time model of the continuous-time system, a training estimator is constructed to learn the normal mode and the fault modes. By using the deterministic learning theory and stability results of linear discrete time-varying systems, the modeling uncertainty and the fault functions are locally-accurately approximated. The obtained knowledge are stored in constant RBF networks. Secondly, by utilizing learned knowledge, a set of estimators are constructed. One estimator represents the normal mode, whereas the others represent the fault modes. The average L₁ norms of the residuals are taken as the measure of the differences of dynamics between the monitored system and the estimators. The occurrence of a oscillation fault can be rapidly detected according to the smallest residual principle. Simulation study is included to demonstrate the effectiveness of the approach.