Fault detection and isolation in an inertial navigation system using a bank of unscented H∞ filters

In order to ensure safe operation and meet reliability standards at a safety-critical level, instrument failures have to be robustly handled through effective fault diagnosis. A popular approach to fault detection for non-linear systems is the extended Kalman filter (EKF). It has, however, been shown to lack robustness in the face of non-Gaussian noise disturbances and modelling errors. An alternative to the EKF, the extended H_∞ (EHF) filter is capable of robust estimation even in the presence of coloured noise; though, modelled on the EKF, it does inherit certain of its shortcomings. A recent addition to the H_∞ family of filters, the unscented H_∞ filter (UHF) promises both robustness and excellent estimation performance in non-linear, non-Gaussian settings. This paper presents arguably the first application of the UHF to an FDI task: sensor fault detection and isolation (FDI) in a strap-down inertial navigation system (INS) of the type commonly mounted on smaller unmanned aircraft. We apply the UHF in a bank of dedicated observers within an analytical redundancy framework. Results are comparable to the EKF under a Gaussianity assumption.