Fault detection techniques based on set-membership state estimation for uncertain systems

This thesis proposes a new Fault Detection approach for linear systems with interval uncertainties, bounded perturbations and bounded measurement noises. In this context, the Fault Detection is based on a set-membership state estimation of the system. The main contributions of this thesis are divided into three parts:- The first part proposes an improved method which combines the good accuracy of the zonotopic set-membership state estimation and the reduced complexity of the ellipsoidal set-membership estimation.- In the second part, a new ellipsoidal state estimation approach based on the minimization of the ellipsoidal radius is developed, leading to Linear Matrix Inequality optimization problems. In this context, both multivariable linear time-invariant systems and linear time-variant systems are considered. An extension of these approaches to systems with interval uncertainties is also proposed. - In the continuity of the previous approaches, two Fault Detection techniques have been proposed in the third part based on these set-membership estimation techniques. The first technique allows to detect sensor faults by checking the consistency between the model and the measurements. The second technique is based on Multiple Models. It deals with actuator/component/sensor faults in the same time. A Min-Max Model Predictive Control is developed in order to find the optimal control and the best model to use for the system in spite of the presence of these faults.