An Estimation Technique Applied to Failure Detection

Abstract

To control a dynamic system in presence of sensor failures, the usual approach introduces three sensors for each important output and a voting procedure to select the correct information. The authors propose to replace the third sensor by estimations of the output measurements that express the natural redundancy of the outputs of the system.Obviously, these estimations must not be disturbed by any failure. Among several possibilities, the authors have chosen a method that rejects the incoherent information detected by the two sets of sensors, until the correct sensor is selected. Such a solution is a nonstationary one because of the modifications of the observation equation. In order to reduce the computation time, a bank of stationary filters adapted to each particular configuration of the data acquisition system was used.In the practical example solved, automatic landing of the AIRBUS aircraft, the authors have to deal with new difficulties : a part of the system itself is nonstationary. So a separation into a large stationary system and the remaining one, nonlinear and nonstationary, is done.Good results are obtained with a computation time not too large for an onboard computer.