Nonlinear set membership filtering using ellipsoids and its application in fault diagnosis

A set membership filtering algorithm using ellipsoidal sets for nonlinear systems with unknown but bounded noises is proposed and applied to guaranteed fault detection and isolation(FDI).After the nonlinear state and observation equations are expended in a Taylor series,a tight box outerbounding the region in which the linearization remainder potentially lies is found using interval analysis.Assuming that boxes are employed to bound the process and observation noises,the time and observation updates in the algorithm require computing a subminimal-volume ellipsoid containing the vector sum of an ellipsoid and segments and the intersection of an ellipsoid and strips respectively.Based on the filtering algorithm,sensor FDI approaches are developed.Since the set membership filtering is a guaranteed state estimation,the FDI approaches are guaranteed FDI approaches,which means that if a fault is detected,there is truly a fault in the system.The proposed approaches are applied to a two-state nonlinear system to gain insight into their inner-workings.