Sensor Fault Detection for UAVs using a Nonlinear Dynamic Model and the IMM-UKF Algorithm

This paper presents a method of Fault Detection, Identification and Accommodation for inertial sensors in Unmanned Aerial Vehicles. A nonlinear model of the aircraft's dynamics replace the traditional inertial navigation equations and is used in conjunction with the Interacting Multiple Model and the Unscented Kalman Filter for improving state estimation in presence of inertial sensor faults. Performance comparisons are made between filters using the inertial navigation equations and the dynamic model for the fault-free conditions. It is shown that a matched UKF will result in adequate state estimation regardless of the failure mode and that the IMM-UKF algorithm is a step closer to achieving the same performance. The IMM-UKF is shown capable of maintaining stable state estimates in the presence of all single inertial sensor faults.