Research And Implementation Of Error Compensation Technology On Six-joint Robot

With the application of industrial robot becoming more and more popular, people put forward more requirements on the robot’s motion precision. Since the various error factors, there is a certain error between the theory pose and the actual pose, which is a great obstacle on the robot’s promotion. Calibration technology is a common method to improve the motion precision. Error compensation is an important part in calibration. Based on the industrial robot calibration and error compensation, this paper taking Yaskawa MOTOMAN-MH6 as the prototype robot, mainly does the following work:Based on DH and MDH method, the kinematics of MOTOMAN-MH6 was deduced. Error model was established upon the analysis on error factors. The kinematics simulation in MATLAB laid a theoretical foundation for the following algorithm design.During the measurement, end position information is easy to get but the gesture is not easy. Based on end position constraint, a calibration method with the goal of building linear equations was proposed. The analysis of degradation on A provided a theoretical basis for the selection of the parameters. A cross bar calibration method was designed. The effectiveness of algorithm was verified in MATLAB. However, only position information couldn’t get the whole parameters identification.In order to achieve the full recognition, using the all pose information,a genetic tabu-search algorithm(GTSA) was proposed. GTSA put parameter identification as a multi-parameter optimization problem. According to the simulation results in MATLAB, the algorithm is effective. End error compensation could be fulfilled by updating kinematics model. Nevertheless, the update could result in the robot kinematics model no longer met the Pieper criterion and algebraic solution very complicated or inexistent.When kinematics couldn’t update, an end pose prediction and error compensation algorithm based neural network was proposed. Neural network has strong ability to nonlinear approximation. If BP network conducted the mapping of target pose and actual pose, the problem of pose prediction was solved. If BP network conducted the mapping of target pose and fixed pose, the problem of error compensation was solved. Simulation shows that, after add compensation component, the mean error distance has decreased by 61.54%, which illustrates the algorithm is effective.