Fault diagnosis of non-linear dynamical systems using analytical and soft computing methods

The paper deals with the problems of robust fault detection using analytical methods (observers and unknown input observers) and soft computing techniques (neural networks, neuro-fuzzy networks and genetic programming). The model-based approach to Fault Detection and Isolation (FDI) is considered. In particular, observers for non-linear Lipschitz systems and extended unknown input observers are discussed. In the case of soft computing techniques, the main objective is to show how to employ the bounded-error approach to determine the uncertainty of the GMDH and neuro-fuzzy networks. It is shown that based on soft computing models uncertainty defined as a confidence range for the model output, adaptive thresholds can be defined. The final part of the paper presents two illustrative examples that confirm the effectiveness of the unknown input observers and the neuro-fuzzy networks approaches.