Actuator Fault Detection and Estimation for Hydrofoil Attitude Control Systems: A Gaussian Mixture Model-Aided UKF Approach

Hydrofoils, as the actuators in hydrofoil attitude control systems (HACSs), are prone to faults and failures resulting in fatal accidents. In this article, a fault detection and estimation (FDE) strategy for such failures is presented based on a developed baseline model. The actuator fault is first parameterized by rewriting the dynamic model of the HACS. To address non-Gaussian factors in the joint estimation of the unscented Kalman filter (UKF), the Gaussian mixture model (GMM) is then integrated with the conventional UKF to approximate the complex distributions. Considering the problem of singular or ill-conditioned covariance matrices in the GMM, a covariance regularization approach based on the posterior estimation with the inverse-Wishart distribution is explored. Actuator anomalies can be detected via a scale metric utilizing the covariance matrices, together with the peak signal-to-noise ratio (PSNR) index, and accurate fault estimation is implemented by isolating fault information from the augmented state even in the presence of non-Gaussian disturbances. The effectiveness of the proposed strategy is validated by the experiment carried out on a hardware-in-loop (HIL) platform. The RMSE results for fault estimation obtained through the method proposed in this article exhibit the reductions of 19.23% and 31.77% for the constant-value fault scenario, 5.03% and 16.31% for the time-varying fault mode, when compared with the results obtained using the joint UKF and the Bayesian-based adaptive Kalman filter with Gaussian-inverse-Wishart mixture (BAKF-GIWM) techniques. These findings underscore the superior efficacy and performance of the proposed method.