A Novel Bilateral Fuzzy Adaptive Unscented Kalman Filter and its Implementation to Nonlinear Systems with Additive Noise

Abstract

This paper introduces a novel approach, called a bilateral fuzzy adaptive unscented Kalman filter (BFAUKF), for fault detection in nonlinear systems. This algorithm uses a Mamdani fuzzy logic in the determination of the measurement noise covariance which is needed in the implementation of the unscented Kalman filter (UKF). By doing so, we can achieve better accuracy and shorter computation time in the detection of fault when it is compared with the stand alone UKF estimation method. To show the effectiveness of this algorithm, a fault detection design based on the proposed approach is developed for an inverted pendulum system. The simulation results show a significant improvement of 65% on fault detection accuracy and 30% on computation time in comparison with the conventional UKF algorithm.