On parametric solution of fault detection problems

Abstract

We discuss the solution of the fault detection problem in presence of parametric uncertainties. The basic approach is an extension of the nullspace method for constant systems to the case of linear parameter varying (LPV) models. In a general setting, we consider the case when part of the unknown parameters are non-measurable and part of them are measurable. The resulting LPV-gain scheduled fault detection filter provides robustness with respect to both types of parametric uncertainties. Symbolic and numerical computational procedures which underlie the proposed synthesis approach are discussed.