Research on Parameter Identification Method for Robotic Manipulators Joint Friction Model Based on PINN

Abstract

In the research and application of robotic manipulator systems, the friction phenomenon poses challenges to system stability and control precision. To further improve the parameter identification accuracy in traditional friction modeling for robotic manipulators, this paper proposes a friction model parameter identification method based on the Physics Informed Neural Network (PINN). The proposed method takes the relative velocity and normal pressure in the motion of the robotic manipulators as the information input items, with the friction and model parameters as outputs. The network parameters and identification parameters are updated according to the Adam method, achieving a more precise identification of friction parameters. It comprehensively considers friction mechanism information and data information to jointly construct the objective optimization function. Through simulation comparisons with noisy/noise-free data, the PINN method is validated to have higher identification accuracy than Genetic Algorithm (GA) and Particle Swarm Optimization (PSO), with an average reduction of $30 \%$ and $50 \%$ in the identification error rates for noise-free and noisy data, respectively.