Combined Kalman and sliding innovation filtering: An adaptive estimation strategy

Abstract

This paper proposes a new adaptive estimation strategy for a nonlinear system with modeling uncertainties. The extended Kalman filter (EKF) and unscented Kalman filter (UKF) are optimal estimators which have been used extensively for state estimation in literature and industry. While the EKF uses a first order Taylor series expansion to approximate nonlinearities, the UKF uses sigma points from the projected probability distribution of states. The sliding innovation filter (SIF) is a suboptimal, yet robust estimation strategy which has recently been proposed. For nonlinear systems, the extended SIF (ESIF) is formulated by using a first order Taylor series expansion like the EKF. This work proposes a novel adaptive estimation strategy which combines and balances the optimality of the EKF and UKF with the robustness of the ESIF. These new methods are referred to as the EKF-ESIF and UKF-ESIF, respectively. A time-varying sliding boundary layer is used as a means of detecting the presence of faults or uncertainties and as a criterion for switching between the EKF or UKF and the ESIF. In normal operating conditions the algorithm computes estimates using an optimal KF-based gain, and an SIF-based gain when a fault is detected. The system examined in this study consists of a magnetorheological (MR) damper with a constant current. Faults or uncertainties are introduced as unwanted behavior in the power supply in the form of undercurrent and overcurrent. The robustness of the EKF-ESIF and UKF-ESIF was validated for force estimation exerted by the MR damper and the results were compared with the standard EKF and UKF.