Air data sensor fault detection and diagnosis with application to real flight data

Abstract

Air Data Sensor (ADS) Fault Detection and Diagnosis (FDD) is important for the safety of the aircraft. Adaptive Fading Unscented Kalman Filter (AFUKF) is able to tackle the ADS FDD using the kinematic model of the aircraft. This paper proposes a Robust Three-Step Extended Kalman Filter (RTSEKF) for the estimation of the ADS faults. The RTSEKF is extended to Robust Three-Step Unscented Kalman Filter (RTSUKF) to reduce the influence of linearization errors when coping with nonlinear systems. However, it is found that the RTSUKF is still sensitive to initial conditions. The problem is analyzed and a Modified Robust Three-Step Unscented Kalman Filter (MRTSUKF) is proposed to guarantee its performance. The performance of the AFUKF and the MRTSUKF is tested using the simulated data of a Cessna Citation II aircraft. The similarities and differences between them are also presented. Furthermore, the performance of these two approaches is validated using the real flight test data of the aircraft. The results demonstrate their ability to be applied in practice.