Detecting faults in a hydraulic system using neural network and observer approaches

A robust fault detection observer (RFDO), which includes a robust residual, is designed for a general nonlinear system with polynomial-type nonlinearities. Two main theoretical results are given for the existence, stability and the detectability of the RFDO. The approach is applied to a three-tank hydraulic system where two fault scenarios are considered. Three different degree observers, with corresponding residuals, are assessed. This observer approach is then compared to the performance of a neural network approach under the same fault conditions.