Dynamic Parameter Identification for Reconfigurable Robot Using Adaline Neural Network

Abstract

This paper discusses the issues of the dynamic parameter identification for the reconfigurable robot and presents an identification approach based on the Adaline neural network. The process of the dynamic parameter identification is mainly divided into three steps. The first is to establish the dynamic model of the reconfigurable robot and compensate the joint friction with Coulomb viscous friction model. Moreover, the finite Fourier series is designed by the excitation trajectory and the genetic algorithm is used to optimize the trajectory. The second is to build an Adaline neural network model and to train the weight of the network by the motion and torques information of the reconfigurable robot. The third is to simulate the three-degree-of-freedom reconfigurable robot and compare the predicate value by the Adaline neural network with the one by the linear least squares. Finally, the accuracy of the dynamic parameters is verified by using the simulation results of the torque errors. And the simulation results show that the proposed Adaline neural network can improve the identification precision of the dynamic parameters.