Fault diagnosis of nonlinear systems using a hybrid approach (Q1022039)

An overview of the fault diagnosis literature in general and fault detection methodologies in particular is given. The fault diagnosis problem in nonlinear systems is defined. A comprehensive literature review and analysis of different approaches to fault detection, isolation and identification of both linear and nonlinear systems are presented. Both model-based and computational intelligence-based approaches to fault diagnosis have been extensively reviewed and analyzed and a number of well-known methodologies within each framework are further demonstrated and their respective pros and cons are cited. Both the series-parallel and the robust parallel structures of the hybrid nonlinear fault detection isolation and identification methodology under full-state measurement assumption, which is the core contribution of this monograph are demonstrated. The specific formulation of the problem of fault detection isolation and identification in nonlinear parameter estimation problem using the notion parameterized fault models is introduced. A short survey of various model-based and computational intelligence-based nonlinear parameter estimation techniques is also performed. The theory of state estimation or filtering has been comprehensively reviewed in order to design and develop a fault tolerant state estimator that enables fault detection, isolation and identification under partial-state measurement conditions. A specific adaptive neural estimator is then designed and its integration with the proposed hybrid fault detection, isolation and identification is described. A spacecraft attitude control system and reaction wheel actuators to which the proposed fault diagnosis algorithms are applied is explained. Simulation results demonstrating the effectiveness and validating the properties, such as robustness, of the proposed fault detection, isolation and identification algorithms have also been proposed.