Nonlinear Estimation of Sensor Faults With Unknown Dynamics for a Fixed Wing Unmanned Aerial Vehicle

In this paper, the estimation of additive inertial navigation sensor faults with unknown dynamics is considered with application to the longitudinal navigation and control of a fixed wing unmanned aerial vehicle. The faulty measurement is on the pitch angle. A jump Markov regularized particle filter is proposed for fault and state estimation of the nonlinear aircraft dynamics, with a Markovian jump strategy to manage the probabilistic transitions between the fault free and faulty modes. The jump strategy uses a small number of sentinel particles to continue testing the alternate hypothesis under both fault free and faulty modes. The proposed filter is shown to outperform the regularized particle filter for this application in terms of fault estimation accuracy and convergence time for scenarios involving both abrupt and incipient faults, without prior knowledge of the fault models. The state estimation is also more accurate and robust to faults using the proposed approach. The root-mean-square error for the altitude is reduced by 77% using the jump Markov regularized particle filter under a pitch sensor fault amplitude of up to 10 degrees. Performance enhancement compared to the regularized particle filter was found to be more pronounced when fault amplitudes increase.