Generation of robust residuals based on unknown input observers with optimal approximate decoupling

This paper proposes a method of robust residual generation to linear system influenced by the unknown inputs. The effect of the disturbances and uncertainties can generate difficulties on the level of the stage of decision to diagnosis (false alarm), it is then shown that a robust observer with the unknown inputs guarantees robustness system, based on unknown inputs observer by optimal approximate decoupling in the diagnosis. This last leaned on the minimal and the maximal sensibility at the unknown inputs and the faults respectively, who guarantees the obtaining of robust residuals checked by the constraints of robustness/sensitivity in the form of Linear Matrix Inequalities (LMIs). Finally, a simple application is presented to study the performances of this observer.