Fault detection based on an improved zonotopic Kalman filter with application to a wind turbine drivetrain

Abstract

This paper proposes a sensor fault detection method based on an improved zonotopic Kalman filter (ZKF) for discrete-time systems with parameter uncertainty. In the residual generation step, an improved ZKF is designed to generate robust residuals. The improved ZKF is designed by directly optimizing the estimated interval widths, which provides a clear geometric interpretation and yields tighter uncertainty bounds compared to the commonly used Frobenius norm optimization method. Moreover, the gain matrix of the improved ZKF is computed by the linear programming method, which is numerically efficient. In the residual evaluation step, the improved ZKF is used to obtain guaranteed adaptive thresholds. Then, to illustrate the superiority of the proposed fault detection method, a comparison study with application to a wind turbine drivetrain is proposed, which illustrates that the proposed method can achieve more accurate fault detection results compared with the commonly-used Frobenius norm optimization approach.