Fault-Tolerant Sensor Reconciliation Schemes via LFT Unknown Input Observers

Abstract

This paper presents a fault-tolerant sensor reconciliation design approach for over-sensed plants (see Fig. 1). The reconciliator is in charge of detecting, at each time instant, the possibly faulty physical sensors y and generating a virtual output z (with dimy ≥ dimz) where the corrupted measures coming from the pool of redundant physical output are removed. In this way the virtual output is always healthy and usable for control purposes without requiring the reconfiguration of the nominal control law. The approach is based on the use of Unknown Input Observers (UIO) with Linear Fractional Transformation (LFT) parameter dependency and works together with an "ad-hoc" parameters estimator that is designed to estimate on-line at each time instant the sensor effectiveness matrix. The sensor faults here considered are limited to variation of sensors’ gain and offset values. All main properties of the scheme are investigated and rigorously proved. A final simulation example is included to show the effectiveness of the proposed scheme.