Fault-tolerant sensor reconciliation schemes based on unknown input observers

This paper proposes two fault-tolerant sensor reconciliation design methods for over-sensed plants. The aim of the reconciliator is to detect the presence of faults on the existing physical sensors and hide the corrupted measurements in the generation of a virtual output, which one would like to be generated in a reliable way in spite of fault occurrences and hence trustfully usable for control purposes. The sensor faults here considered are limited to variations of both sensor gain and offset values. The proposed approach envisages the use of an unknown input observers coupled with an 'ad-hoc' parameters estimator used to estimate online the sensor effectiveness matrix at each time instant. In the paper, two design methodologies are described, based one on the linear parameter varying polytopic formulation and the other on the linear fractional transformation paradigm. All main properties of the considered schemes are investigated and rigorously proved.