System identification of the nonlinear residual errors of an industrial robot using massive measurements

Abstract

Abstract In this paper, an industrial robot is calibrated by identifying the nonlinear residual errors from massively measured tool positions using a laser tracker. A fully automatic measuring system is developed to collect more than 10,000 robot configurations in an efficient manner, which communicates with the robot controller and the laser tracker in real time. The robot configurations are programmed using a laser traceable path planning algorithm, which ensures the tool positions always locate in the expected envelope and the mirror ball keeps aligning against the laser tracker without occlusion. Afterwards, the robot motion control program is automatically generated using a robot simulation and off-line programming system. In order to obtain relatively higher positioning accuracy and less computation time, the off-line calibration procedure is subdivided into two steps. The first step solves the significant structural errors using the model-based parametric calibration. The second step further identifies the nonlinear residual errors using a deep neural network (DNN), so that the deviations related to different robot configurations can be reduced and the overall workspace accuracy can be improved to a much higher level. After calibration, the mean/maximum positioning errors in the measured envelope are reduced from 1.81 mm/1.96 mm, respectively, to 0.10 mm/0.22 mm. The calibrated model is eventually integrated into the off-line programming system, and effectively compensate a robot welding trajectory.