Kalman-Filter Based Control and Performance Monitoring Systems

Abstract

This paper addresses the problem to design control and performance monitoring systems for robotic manipulators. Basically three control schemes are proposed. The first one is a modification of the computed torque method where the input torques necessary to drive the robotic manipulator are on-line computed as functions of the optimal – unbiased, minimum variance – estimates of the joint coordinates and velocities, and the corrected accelerations. The optimal state estimator based on a stochastic discrete-time linear dynamic model and previous observations is generated the discrete Kalman filter. The idea of the second control system is to use a better reference trajectory for linearization in each working cycle. Two ways are given to perform this relinearization process: the relinearization is performed around the estimated trajectory, and the extended Kalman filter. The third novel control method addresses the problem to design a nonlinear feedback controller in case of uncertain dynamical parameters and other disturbances. Monitoring schemes for real-time detection and isolation of failures in robotic manipulators are outlined in the second part of the paper. The problem of fault detection is formulated as a problem in Hypothesis Testing by regarding the normal operation of the robot manipulator as the null hypothesis.