On Identifiability of Bias-Type Actuator-Sensor Faults in Multiple-Model-Based Fault Detection and Identification

Abstract This paper explores a class of multiple-model-based fault detection and identi cation(FDI) methods for bias-type faults in actuators and sensors. These methods employbanks of Kalman-Bucy lters to detect the faults, determine the fault pattern, andestimate the fault values, wherein each Kalman-Bucy lter is tuned to a di erentfailure pattern. Necessary and sucient conditions are presented for identi abilityof actuator faults, sensor faults, and simultaneous actuator and sensor faults. It isshown that FDI of simultaneous actuator and sensor faults is not possible using thesemethods when all sensors have biases. 1 Introduction Failures in control e ectors and sensors can cause poor performance or instabilityin dynamical systems. In particular, faults in ight control systems for aircraft orspacecraft can lead to loss of control and serious incidents. Therefore, rapid faultdetection and identi cation (FDI) in actuators and sensors is important for enhancingight safety. One approach to actuator and sensor FDI is based on multiple-modelmethods [1], [2], which have been extended to detect faults and identify the faultpattern as well as the fault values [3], [4]. Such methods typically use banks ofKalman-Bucy lters (or Extended Kalman lters) and multiple hypothesis testing,and have been reported to be e ective for bias-type faults such as aircraft controlsurfaces getting stuck at unknown values, or sensors (e.g., rate gyros) that developunknown constant or slowly-varying biases. The underlying requirement for thesemethods is that the faults should be identi able. Identi ability of bias-type faultswas considered in [4] and preliminary results were presented. This paper focuses ingreater detail on the identi ability of actuator faults, sensor faults, and simultaneousactuator and sensor faults.