Adaptive hybrid Unscented Kalman filter for aircraft sensor fault Detection, Isolation and Reconstruction

Abstract

In this paper, a new approach for input and output Fault Detection, Isolation and Re- construction (FDIR) is proposed. Robust Adaptive Kalman Filters do not consider the reconstruction of faults and they require testing the innovation covariance in order to distinguish between an input fault and an output fault. Sliding Mode Observers requires a secondary observer to reconstruct output faults. The approach proposed in this paper consists of three filters: an Adaptive Fading Unscented Kalman Filter (AFUKF), an Augmented Unscented Kalman Filter (AUKF) and an Adaptive Hybrid Unscented Kalman Filter (AHUKF). The AFUKF is derived and proven to be able to reconstruct output faults in an unbiased sense. Input faults are detected and reconstructed by the AUKF. Finally, these two filters are integrated by the AHUKF through an adaptive integration scheme, which can distinguish between input faults and output faults without testing the innovation covariance. Additionally, the kinematic equations are used for aircraft sensor FDIR rather than the dynamic equations, which makes this approach robust to model uncertainties. The proposed approach is tested with the Cessna Citation II CE-550 model with the objective of the Inertial Measurement Unit (IMU) sensor and Air Data Sensors (ADS) FDIR. The simulation results demonstrate the effectiveness and efficiency of the proposed approach, showing its feasibility for input and output FDIR.