| Industrial robot,as a flexible intelligent machine,has been widely used in all walks of life in national production.But its processing,installation,assembly and geometric error due to factors such as transmission or component wear,deformation of bar,end load change and the influence of the geometric error caused by the factors such as temperature,lead to actual model parameters and the nominal size,easy to a drop in accuracy,stability,reduce and control model is not accurate.Through investigation,it is found that the calibration method of measuring the end-effector of the robot and identifying the error vector of joint parameters combined with inverse kinematics is the necessary way to improve the accuracy and stability.However,the following problems are prone to occur: First,the current calibration methods are limited by the working range of the measuring instruments,and the spatial coordinates of the end-effector are often difficult to determine,which only ensures the accuracy of the measured pose,but fails to calibrate the full pose of the robot in the feasible domain.Secondly,for the geometric calibration method using joint axis vector to identify joint parameters,the analysis of a single joint needs to lock the other joints at the same time,which seriously affects the real-time and authenticity of measurement and identification.Third,the end-effector pose measurement requires continuous iteration to obtain joint parameters through inverse kinematics,but the efficiency of identification and calibration is affected in the process of continuous iteration.Fourthly,the identified results are error variables of each parameter under the global least squares assumption,rather than real kinematic parameters or time-varying parameter time-history curves,which can only provide data support for offline calibration effectively,but are not suitable for real-time online compensation and calibration methods.Finally,limited by the current experimental system that only measures the end effector of the robot and the theoretical method of identification combined with inverse kinematics,this calibration method with the sole purpose of improving the end accuracy rarely systematically analyzes and explores the source and location of errors.Therefore,how to make up for the limitations of the existing robot calibration methods,and explore the experimental system based on error system analysis,is the only way to improve the core competence of industrial robots.In this paper,a new experimental system of "four-dimensional measurement" based on the "joint motion space" of the 6 axis series rotating robot is constructed based on the multi-camera vision measurement system and the geometric calibration method of the overall measurement of the robot and the separate analysis of the joint.A number of measuring points are arranged for all the joints and rods of the robot,and the control points of each joint are identified and tracked by the multi-camera vision measurement system under the premise that the robot moves in a wide range in any feasible domain.On the basis of obtaining the coordinate sets of the same control points in two states(or time),the optimal absolute rotation matrix and translation vector under the definition of least squares are derived,and the relative rotation matrix and translation vector are derived based on the properties of implicated motion.The joint axis vectors and DH parameters of the robot were obtained by combining Rodrigues transform,and the time history curve of joint Angle,the only joint variable,was analyzed.Then,on the basis of identification,the influencing factors of joint axis vector estimation are systematically analyzed,which lays a foundation for experimental modeling.Finally,the robot is calibrated and compensated by ELM neural network.MATLAB robot toolbox is used as a simulation tool,combined with spline interpolation and other methods to simulate the measurement method in the experiment,and white Gaussian noise is added to simulate the measurement noise.On the basis of verifying the effectiveness of identification and calibration by simulation,the identification and calibration of Eston industrial robot ER20-30-1880 and Guangzhou numerical control industrial robot RB03 are verified respectively.It not only verifies the effectiveness of the method,but also improves the absolute positioning accuracy of the robot.The position error increased from 7.440 mm to0.159 mm,and the attitude increased from 3.073 degrees to 0.077 degrees.At the same time,the absolute positioning accuracy of the robot is improved,and the basic picture of the error analysis of the industrial robot based on the multi-camera vision measurement system is preliminarily constructed.This method improves the traditional method of measuring only the end-effector in robot calibration,and measures and identifies all control points of each joint.The measurement space is expanded,and the robot can be measured in any feasible area.Moreover,the motion mode to be measured is a large space complex motion with six joints moving simultaneously,which not only improves the measurement efficiency,but also realizes the practicability and authenticity of the robot measurement.In terms of identification,the pose information of the end-effector can be obtained by using the unique forward kinematics while determining the identification parameters of each joint under the technical route of robot whole machine measurement and joint analysis alone.It avoids the problem of low iteration efficiency caused by the inverse kinematics calculation based on the end information in the traditional identification method,and greatly improves the identification efficiency.Under the identification system of this method,the real joint axis vector and DH parameters are reconstructed according to the time series.It breaks through the traditional identification method which can only determine the error variable of each parameter under the global optimal condition,and lays a foundation for the research of error source and location analysis,online health monitoring,online pose calibration and so on.Finally,according to the parameter identification results of the ELM neural network and the proposed method,the whole pose calibration of the robot is achieved,which can calibrate both geometric and non-geometric errors at the same time,which not only makes up for the limitation of the traditional method to improve the accuracy only in the calibration area,but also takes into account the non-geometric errors affecting the end accuracy. |
