Model Identification using a Statistical Cluster LPC approach with Application to Motion of a Brushless Motor

This paper presents a new statistical method based on Cluster Last Principal Component (CLPC) algorithm to identify nonlinear, time-varying, dynamical models from input-output data clusters of black boxes. Each of data clusters is on a time window. For every data cluster an appraiser updates the parameters of a Gaussian time-varying model via an optimality design criterion that maximises the Likelihood function and the estimated steady-state parameters of this model are quasi-constant values. An application to identify the nonlinear model of a control system of a brushless motor is developed. By applying of CLPC algorithm to this system, the actual angular positions of the brushless motor and the control torque have been estimated. Numerical tests of simulation in Matlab envinronment demonstrate the effectiveness of the proposed algorithm.