Robust Identification and Fault Diagnosis using Set-membership Approaches

Author: BLESA IZQUIERDO, JOAQUIN

Date of  defense: 20-06-2011

Abstract

This thesis is devoted to the robust identification and fault diagnosis using set-membership approaches. These methods use mathematical models whose parameters are unknown but bounded by known uncertain sets with the purpose of finding possible inconsistencies between system measurements and the modelled behavior. In order to fit better the behavior of real complex systems that can be non-linear and/or distributed parameter systems, multiple input multiple output (MIMO) linear parameter varying (LPV) models in regressor form and complex shapes (zonotopes and polytopes) for bounding the parameter uncertainty have been choosen. In particular, two set-membership approaches have been studied: the interval predictor and bounded error approaches that consider unknown parameters bounded by a set. The difference of these two methods are the consideration of the variation of the parameters in the bounded set: while interval predictor approach considers that parameters can vary at will within the bounded uncertain set, the bounded error approach considers that the parameters are unknown within the bounded uncertain set but it is known that they will not vary. Different identification methods and robust fault detection algorithms have been developed considering the two set-membership approaches for multiple output LPV systems with uncertain parameters bounded by a zonotope. A new method that is an intermediate approach to the studied set-membership approaches has been proposed. This method is able to handle systems with parametric uncertainty and bounded parameter variation between samples. A similar method was proposed in Ingimundarson et al., (2009) where the uncertainty in parameters was bounded by a zonotope, whereas in the method proposed in this thesis, the uncertainty is considered to be bounded by a polytope. It is shown that consistency checks indicating faults can be performed in a natural manner with a polytope description of the feasible parameter set. On the other hand, the fault sensitivity analysis for the different considered faults, computed with the mathematical models, has been used in order to determine the fault detection behavior and for fault isolation and estimation purposes. Along this disertation, six cases studies have been used to illustrate the effectiveness of the proposed methods and algorithms.