Dynamic Parameter Identification Of Industrial Robots

Industrial robots have become an indispensable means to increase productivity. The ever increasing quality standards and new applications impose higher requirements on accuracy, reliability and performance of a robot. At present, most domestic robots still use PID motion controllers, which have not considered the complex nonlinear dynamics of a robot, resulting in the limitations of the accuracy and performance of a robot. A main reason why the dynamic characteristics of a robot are not included in these controllers is that accurate dynamic parameters are difficult to obtain. The experimental parameter identification of a robot is the only effective method to obtain accurate dynamic parameters. Robot identification is an experimental technique to estimate realistic dynamic models from motion data and actuator toques measured during well-designed experiments.Since the large difference between the values of the inertial parameters of the wrist with respect to those of the base, a step-by-step identificaiton procedure is proposed in this paper, the inertial parameters of the wrist and the base are estimated seperately. The proposed method simplifys the identification procedure, and can obtain the dynamic parameters of six-dof serial robots simply and effectively. The identification model of robots dynamic parameters becomes very complicated with the increase of joints, which makes the identification equations of the redundant-degree-of-freedom robots are difficult to obtain, a new sequential identification procedure is proposed. In each identification experiment, only three axes are moved, and the others are locked. The proposed method simplifys the identification procedure, the computational efficiency is significantly improved, and can obtain the dynamic parameters of the redundant-degree-of- freedom robots effeciently.Due to the growing importance of the robot payload, it is necessary to extend the robot identification method to estimate the inertial parameters of the robot payload. In the identification of payload parameters, not only the actuator torque prediction accuracy is important, but special attention is paid to the accuracy of the individual parameter estimates. This paper presents a payload identification approach which does not require a full identification of the manipulator, but compensates for robot dynamic models to obtain the ineritial parameters of payload. Robot payload identification imposes high requirements on experiments, and requires more prior information and calibration to achieve satisfying accuracy requirements.The obtained dynamic robot model is used to improve the path tracking accuracy of robots. In order to achieve the requirement of real-time calculation, the dynamic model is simplified under the condition that the accuracy of predicted torques are maintained. The simplified dynamics model is taken as a torque feedforward loop to estimate the joint torques of robots. The executed trajectory of the feedforward controller is much closer to the desired one, and the output torques are smoother.Since the dynamics and structures of parallel robots are extremely complex, and the corresponding observation matrix is difficult to solve and simplify by the geometric structures of robots directly, a procedure of dynamic parameter identification based on the principle of virtual work and singular value decomposition are adopted. Combining the characteristics of the symmetry of parallel mechanisms and repetitive of the branches, parts of small inertial parameters are eliminated, so that the identification models can be simplified, and the identification accuracy of part parameters are improved. Taking a Stewart platform as an example in this paper, the corresponding validation is completed.