Nonlinear Friction Identification of A Linear Voice Coil DC Motor

Abstract

Friction is a nonlinear phenomenon that has a major effect on the performance of servomechanisms and positioning systems such as CNC machines, robot manipulators, printers and disk drives. For this reason, accurate knowledge of the system parameters and friction characteristics should be included in the system model in order to design high performance control systems. This research aims to develop an identification procedure for nonlinear friction in a linear positioning system. Three methods are used. The first method is based on the free-fall of an object released from rest and it is used to find the mass, Coulomb friction and viscous friction coefficient. The second method is based on the steady state response (SSR) of the velocity control loop using a PI controller and it is used to find the friction force characteristics by measuring the current at constant velocity. The friction-velocity relation is plotted at different velocity values and the Coulomb friction, viscous friction, and stiction friction coefficients are extracted from the graph itself. The third method is based on artificial neural networks (ANNs). ANNs represent a powerful data-driven, self-adaptive, and flexible computational tool, which has the ability to capture the nonlinear and complex characteristics of any physical process with a high degree of accuracy. In this research work, an adaptive linear neuron (ADALINE) network based on three different network structures is used to estimate the parameters of the system including the Coulomb friction and viscous friction coefficients. The training of the network is carried out in off-line and on-line modes. Finally, a comparison between these three methods is made to evaluate the advantages and disadvantages of each technique. An experimental setup with a linear voice coil DC motor and a high resolution position sensor is developed to test the different identification and control algorithms. The setup is interfaced with a dSPACE 1103 Controller Board. The experimental results obtained show that the ADALINE network is able to estimate the system parameters with a percentage error of 4.71% for the mass, 24.63% for the viscous, and 62.29% and 14.86% for the Coulomb friction in the positive and negative directions of motion, respectively.