Reconstruction-based Set-valued Observer - A New Perspective for Fault Detection within Uncertain Systems

This paper presents an extension of a particular type of observer called the Set-valued Observer; this kind of observer is very well suited for uncertain fault detection. But some limitations restrict its use. Indeed, all the sensors are needed to observe the state and as a consequence this method does not allow fault detection when some sensor information is not available. Other work has focused on the well-known Luenberger Observer applied to uncertain systems; but once again, this option is limited. Indeed, it is difficult to converge the algorithm because of the wrapping effect induced by recursivity. Here a new approach is proposed combining the power of the two algorithms. The Luenberger Observer coupled with the SetValued Observer allows us to reconstruct the states without divergence. This combination is a substantial contribution for fault detection within uncertain systems.