Fault Diagnosis in Nonlinear Systems Using Learning and Sliding Mode Approaches with Applications for Satellite Control Systems

In this thesis, model based fault detection, isolation, and estimation problem in several classes of nonlinear systems is studied using sliding mode and learning approaches. First, a fault diagnosis scheme using a bank of repetitive learning observers is presented. The diagnostic observers are established in a generalized observer scheme, and the observer inputs are repetitively updated using the output estimation error in a proportional-integral structure. Next, a framework for robust fault diagnosis using sliding mode and learning approaches is proposed to deal with various types of faults in a class of nonlinear systems with triangular input form. In the designed diagnostic observers, first order and second order sliding modes are used respectively, to achieve robust state estimation in the presence of uncertainties, and additional online estimators are established to characterize the faults. In order to guarantee that the sliding mode is able to distinguish the system uncertainties from the faults, two iterative adaptive laws are used to update the sliding mode switching gains. Moreover, different online fault estimators are developed using neural state space models, iterative learning algorithms, and wavelet networks. Another class of nonlinear systems where an unmeasurable part of state can be described as a nonlinear function of the output and its derivatives is considered next. Accordingly, a class of fault diagnosis schemes using high order sliding mode differentiators (HOSMDs) and online estimators are proposed, where neural adaptive estimators and iterative neuron PID estimators are designed. Additionally, a fault diagnosis scheme using HOSMDs and neural networks based uncertainty observers is designed in order to achieve a better performance in robust fault detection. If the