Robot Error Calibration Based on Improved CSO Algorithm Kinematics and Improved CSO-Elman Neural Network Non-kinematics

Aiming at the positioning error calibration problem of industrial robots, we combine the kinematics and non-kinematics aspects to calibrate the positioning error of robots. Aiming at the kinematics of a robot, a kinematics error model is developed and an improved chicken swarm optimication (CSO) algorithm is proposed to identify the geometric parameter error of the robot. The effect of the proposed algorithm is verified by comparing the Levenberg-Marquardt iterative algorithm and the particle swarm optimization algorithm. The IRB1200 robot is taken as the experimental object, and error data are collected using an APIT3 laser tracker. A robot error calibration experiment platform is built to conduct experiments. The experimental measurement shows that the average positioning error of the robot end is decreased from 2.76 mm to 1.45 mm, which is increased by 47.5%. Furthermore, for the non-kinematic aspect of the robot, the improved CSO algorithm proposed in the kinematic error calibration is used to optimize the initial threshold and weight of the Elman neural network, and the Elman neural network optimized with initial parameters is used to establish the mapping relationship between the robot end position error and the robot joint angle to predict the robot position error in a trained robot cube space. The prediction effect of the ordinary Elman neural network was compared. The experimental measurement shows that the average positioning of the robot end is improved by 34.9% compared with that before calibration, which verifies the fitting prediction effect of the neural network proposed in this study.