Study on BFGS-MLM Algorithm in Dynamics Parameter Identification of Industrial Robots

Abstract

A parameter identification algorithm is proposed for solving the problem in dynamics parameter identification of robots. Firstly, use the nonlinear Deami-Heimann empirical friction model to describe the frictional characteristics between the joints and establish a dynamics identification model. Secondly, in order to overcome the slow convergence speed, the BFGS-MLM (Modified Levenberg-Marquardt) algorithm based on NMLM (New Modified Levenberg-Marquardt) algorithm is proposed for the parameter identification process. This method converts the nonlinear dynamics parameter identification problem into a nonlinear least squares problem, and the parameters to be identified are obtained by iteratively solving the optimal value. In the identification process, the line search strategy is used to solve the optimal iterative step size of the BFGS-MLM algorithm. The quasi-Newton method combined with the BFGS (Broyden, Fletcher, Goldforb and Shanno) correction formula is used to solve the approximate Hesse inverse matrix of the LM (Levenberg-Marquardt) step, which makes the algorithm have higher convergence speed. Finally, it is verified by experiments that this parameter identification method is feasible. The proposed BFGS-MLM algorithm can effectively improve the identification accuracy of dynamics models and the iterative convergence speed. It can provide a new solution for more complex nonlinear dynamics parameter identification problems.