An implementation of a novel vision‐based robotic tracking system

Purpose The purpose of the paper is to present an approach to detect and isolate the sensor failures, using a bank of extended Kalman filters (EKFs) using an innovative initialization of covariance matrix using system dynamics. Design/methodology/approach The EKF is developed for nonlinear flight dynamic estimation of a spacecraft and the effects of the sensor failures using a bank of Kalman filters in investigated. The approach is to develop fast convergence Kalman filter algorithm based on covariance matrix computation for rapid sensor fault detection. The proposed nonlinear filter has been tested and compared with the classical Kalman filter schemes via simulations performed on the model of a space vehicle; this simulation activity has shown the benefits of the novel approach. Findings In the simulations, the rotational dynamics of a spacecraft dynamic model are considered, and the sensor failures are detected and isolated. Research limitations/implications A novel fast convergence Kalman filter for detection and isolation of faulty sensors applied to the three axis spacecraft attitude control problem is examined and an effective approach to isolate the faulty sensor measurements is proposed. Advantages of using innovative initialization of covariance matrix are presented in the paper. The proposed scheme enhances the improvement in estimation accuracy. The proposed method takes advantage of both the fast convergence capability and the robustness of numerical stability. Quaternion‐based initialization of the covariance matrix is not considered in this paper. Originality/value A new fast converging Kalman filter for sensor fault detection and isolation by innovative initialization of covariance matrix applied to a nonlinear spacecraft dynamic model is examined and an effective approach to isolate the measurements from failed sensors is proposed. An EKF has been developed for the nonlinear dynamic estimation of an orbiting spacecraft. The proposed methodology detects and decides if and where a sensor fault has occurred, isolates the faulty sensor, and outputs the corresponding healthy sensor measurement.