INVERSE OF KEY PARAMETERS OF NONLINEAR FRICTION MODEL OF ROBOT JOINTS

An accurate description of nonlinear friction of robot joints has important theoretical and scientific significance for improving trajectory accuracy, positioning accuracy and reliability of robot. However, the robot joints usually contain the motors, reducers, actuators and sensors, which are complex electromechanical coupling system. With the change of service time and working conditions, the friction parameters of robot joints also have significant time-varying effect, which is difficult to accurately describe, resulting in the decrease of trajectory accuracy and great difficulty for robot precision maintenance in the later stage. Therefore, this paper quantitatively evaluates the influence of friction parameters on the output torque of the robot, and proposes an inverse method for nonlinear friction parameters of the robot joints considering time-varying effects. Firstly, a general nonlinear friction model of the robot joint is established. The robot joint constant speed tracking experiment was designed, and the data collected by the experiment were processed by the Kalman filter. Then the relationship between the joint velocity and the driving motor current was established, and the general nonlinear friction model of the joint was established. Secondly, the key parameters of nonlinear friction model are selected. The dynamics model of robot containing nonlinear friction was established. The joint torques were calculated based on the excitation trajectory, and the friction parameters with high sensitivity to joint torques were selected for sensitivity analysis. Thirdly, a data set corresponding to the joint output torque and friction parameters is established. The friction parameter value space was constructed based on the actual working conditions, and the optimal Latin hypercube method was used to sample the friction parameters, which were substituted into the robot dynamics model to calculate the corresponding torques, and the one-to-one data set corresponding to the joint output torques and friction parameters were obtained. Finally, the inverse problem neural network is established and trained, and the key parameters of the nonlinear friction model are reversed and verified. The results illustrate that the accurate description of nonlinear joint friction reduces the influence of friction moment mutation on the trajectory of the robot, and significantly improves the trajectory accuracy.