Research on Dynamic Parameter Identification of Large Inertia Industrial Robot Based on RBFNNs

Abstract

Aiming at the accuracy of the large inertia industrial robot dynamic model, a radial basis function neural networks (RBFNNs) weighted least square (WLS) identification scheme is proposed to further improve the accuracy of the dynamic model. Based on the dynamic linearization model of a large inertia industrial robot, the open-source toolbox Sympybotics is introduced to assist in obtaining the minimum inertia parameter set and observation matrix. The finite-term Fourier series is selected as the excitation trajectory while the condition number of the observation matrix is applied as the performance index for optimization. Its purpose is to ensure that the impact of external disturbances on the identification data is minimized while fully exciting the robot dynamics. Based on the actual operating data, the weighted least squares method is used to identify the kinetic parameters to obtain a rough solution of the kinetic parameters. Further, the accurate solution is obtained by nonlinear constraint function optimization and RBFNNs optimization. The experimental results show that the proposed method could guarantee the accuracy of the dynamic model of the large inertia industrial robot effectively, which provides an important technical support for its high-performance motion control.