ABSTRACT
Model based fault diagnosis relies on process models and differs in terms of adopted model as well as algorithms used. The system model expressed in descriptor form is a natural starting point for modeling of complex industrial processes (Nikoukhah 1992). Recent studies (Darouach 1992, Chisci 1992, Mandela 2010, Puranik 2012) demonstrate state estimation of descriptor systems by Kalman filters and its variants. However, the application of such filters to fault detection of descriptor systems has not been assessed. This paper aims to analyse fault detection (FD) of nonlinear descriptor system (NLDS) by implementation of such filters. An Ensemble Kalman filter (EnKF) is evaluated for state estimation and filter performance is examined. Further, the residuals are generated and analysed by various residual evaluation functions. The approach for state estimation, input fault detection and performance analysis of residual evaluation functions is demonstrated on a slow electrochemical descriptor system.
