Kinematic Calibration Of High Speed Parallel Robot Based On Monocular Vision

According to the requirements of the domestic industrial automation production line for the precise picking and placing of light and small materials,aiming at a classical Delta high speed parallel robot with an active joint driven by an external rotating pair and a slave joint containing a parallelogram,the kinematic calibration method based on monocular vision is developed.The theories and methods of geometric error modeling,visual error measurement,error parameter identification and compensation are analyzed.The purpose of this dissertation is to direct the parallel robot to realize fast kinematic calibration in the field.The main contents of the study are as follows.Geometric error modeling.Based on the closed vector chain and the first order perturbation technique,a general and systematic method of geometric error modeling for high speed parallel manipulators is proposed.By using this method,the explicit form of the full parameter geometric error model of a single branch chain is established,and according to the principle and characteristics of the parallel mechanism,the geometric error sources that affect the controllable and uncontrollable pose errors at the end of the moving platform are effectively separated.Thus a simplified model of compensated geometric error source is constructed,and the influence coefficient of each error source on the end position error is further analyzed by the forward kinematic,which provides a theoretical basis for further research.Error measurement method.A vision measurement system for measuring the end position of Delta robot is designed by analyzing the characteristics of the mechanism and comparing many visual measurement methods.A vision measurement scheme of terminal pose can be worked out to ensure high measuring accuracy.Then a method based on Zhang Zhengyou calibration method is proposed to measure the end position of parallel robot.This method has high calibration efficiency,high calibration precision,high flexibility and low cost.And further analysis and detection of visual positioning accuracy provide a guarantee of measurement accuracy for the subsequent kinematic calibration research.Error parameter identification.According to the invariance of distance error and the influence of random error,a parameter identification model based on distance error of vision measurement is constructed.The model shows the mapping relation between error parameters and distance errors of end.A distance measurement method based on monocular vision is proposed.The distance between the camera center is calculated by means of the rigid body transformation and the equivalent transformation principle of the vision measurement.Then the distance error of the manipulator end is obtained by using the parallelogram property.This method solves the problem that the distance information in the whole workspace can not bemeasured because the camera has a small field of vision.An algorithm for selecting the optimal pose set of the measuring points is presented,which takes into account the number of measured points and the optimal pose set.This algorithm can not only guarantee higher compensation accuracy of the end effector but also shorten the calibration time.The regression model of ordinary least square method is constructed.In order to overcome the problem of multiple collinearity,the identification algorithm is improved by reducing the dimension or introducing the regularization parameter.Thus regression models of truncated singular value decomposition method,principal component regression method and adaptive ridge regression algorithm are established.It provides a theoretical basis and a strong guarantee for the reliability and stability of the error parameter identification results.Error compensation method.A real-time error compensation method is proposed by modifying the input value of active joint.The computer simulation of kinematics calibration is accomplished by using the hierarchical error compensation method.The simulation results of rough calibration and fine calibration show that the identification model is correct and the identification results of the selected measuring points are more accurate and stable.The reliability and stability of the identification results of principal component regression algorithm and adaptive ridge regression algorithm are better,and the identification robustness of adaptive ridge regression algorithm is slightly better than principal component regression algorithm.The effectiveness of the proposed kinematics calibration method is verified by the calibration experiments on the actual prototype.The position error of the actual Delta robot is less than 0.387 mm and the distance error is less than 0.251 mm after the fine compensation.