Calibration of a serial robot using a laser tracker

Abstract

The positioning performance of an industrial robot ABB IRB 1600-6/1.45 has been studied with a laser tracker. Performing some axis-by-axis analyses, we found that axes 2, 3 and 6 have a non-geometrical behavior. A 34-parameter model was used to represent the real robot. This error model takes into account the geometrical errors due to fabrication as well as four error parameters related to stiffness (in axes 2 and 3) and four other error parameters used to fit a second-order Fourier series to the non-linear behavior of axis 6. The Nelder-Mead non linear optimization technique was used to find the error parameters that best fit the measures acquired with the laser tracker.
\n
\nAn algebraic solution for the inverse kinematics is not possible for the 34-parameter model. We therefore propose a numerical and iterative inverse algorithm to recalculate the robot targets into so-called fake targets. No more than three iterations are needed to accurately obtain the joint angles corresponding to a given pose of the end-effector.
\n
\nSimilar tests to the ones proposed by the ISO 9283 norm are performed to compare the accuracy of the nominal and improved robot models. The validation of the accuracy is done with a large number of measures. For the 34-parameter model the mean / maximum position errors are reduced from 0.979 mm / 2.326 mm to 0.329 mm / 0.916 mm (verification performed with around 1000 measurements), at a 6 kg payload, for eight points on the endeffector and for the complete robot workspace (or almost complete, since we had to avoid some obstacles).
\n
\nAnalyses were performed with the expected errors. They allow to “pre-validate” the models without having to take extra measurements. It was found that this pre-validation is very close to the real validation.