Robot Kinematics Parameter Identification And Error Compensation Based On RPSO-DFNN

The absolute positioning accuracy of a robot refers to the deviation between the theoretical and actual values of the robot.In actual production and processing.Industrial robots are affected by various factors such as long working hours and excessive loads,resulting in problems such as deformation of the connecting rods..The deformation of the connecting rod will lead to the change of kinematics parameters and kinematics errors.To improve the absolute positioning accuracy of robots,this article focuses on the design of robot measurement devices,joint compensation methods for robot absolute positioning errors,derivation and simulation of error compensation algorithms,and grinding experiments.The main research content is as follows:(1)Firstly,design a robot processing platform.Secondly,a robot end detection device was constructed using tools such as wire sensors,data acquisition cards,and robot flanges.Use industrial profiles to build a detection device installation frame,install a wire sensor on the installation frame,and fix the wire end of the sensor on the robot flange.Using a laser interferometer as a calibration tool,the calibration method,modeling,and uncertainty of the coordinate detection device were analyzed.Finally,the DH method is used to analyze the robot’s kinematics model and error model.(2)To improve the absolute positioning accuracy of robots,a joint compensation algorithm for robot kinematics parameters is proposed.Firstly,using DH model and error model,the joint compensation theory of robot kinematics parameters and robot joint variables is proposed.Secondly,the ring particle swarm optimization(RPSO)algorithm is proposed to improve the iteration mode of particles to compensate the kinematics parameters of the robot.Then,using full connection neural network,Relu activation function,loss function and other methods,a dual channel feedforward neural network(DFNN)is designed to predict robot joint variables.Finally,the ring particle swarm optimization algorithm and the dual channel feedforward neural network are used to simultaneously compensate the kinematics parameters and joint variables of the robot.(3)The robot kinematics error compensation theory is simulated and verified.On the one hand,four different algorithms,namely particle swarm optimization algorithm,ring particle swarm optimization algorithm,dual channel feedforward neural network and RPSO-DFNN,were used for simulation,and the accuracy after simulation was improved by 76.9%,79.35%,83.72% and 84.75%respectively.On the other hand,by comparing the Mayfly algorithm with the Sparrow algorithm,the compensated accuracy was improved by 75.76% and 75.12%,respectively.The accuracy of the joint compensation method can also be improved by 8.99% and 9.63%,respectively.(4)Verify the trajectory accuracy and machining accuracy of the robot.Use a laser interferometer to measure the trajectory error of the robot in the x,y,and z directions.The detection platform was built,and Tianyuan 3D scanner was used to detect the robot machined workpiece before and after calibration.The point cloud data was collected for error analysis and flatness analysis.After optimization,the average trajectory error on the x,y,and z axes was reduced by 0.185 mm,0.397 mm,and 0.264 mm,respectively,and the workpiece over cutting phenomenon was reduced by 23.975%.