Manipulator Control System Model Estimation Unsing a Real Time Finite State Machine based on Statistical LPC Analysis

Abstract

This paper presents a new statistical method based on a real-time Last Principal Component (LPC) algorithm to estimate single-input-single-output (SISO) and multiple-input-multiple-output (MIMO) varying time dynamical models from input output data clusters of non stationary black boxes. Each of data clusters is on a time window. A real time estimation algorithm via statistical LPC algorithm and an appraiser called "finite state machine" is then described. For every data cluster the finite state machine updates the parameters of a Gaussian varying time model via an optimality design criterion that maximises the Likelihood function. Using the LPC algorithm and the finite state machine, the estimated steady-state parameters of the model are constant values. An application to estimate the control system model of an industrial manipulator is developed. In fact many mathematical models of physical systems are non stationary such as industrial manipulator model. By applying of the real time LPC algorithm to control system of a two links planar industrial manipulator, the actual positions of the joints and the control torque have been estimated and numerical tests of simulation in Matlab 6.5 envinronment demonstrate the effectiveness of this algorithm.