Contribution à l’amélioration de la précision absolue des robots parallèles

Abstract

The purpose of the present study is to improve the absolute accuracy of parallel robots, using geometric calibration methods. These methods identify the values of the robot’s geometric parameters, to improve the correspondence between the real robot and the mathematical model used in its controller. In addition to the compensation of geometric errors, the proposed calibration approaches allow to identify accurately the base frame of each of the studied robots. The developed methods are applied using two parallel robots with less than six degrees of freedom (DOFs): a precision positioning table with three DOFs (PreXYT) and a planar fivebar robot (DexTAR) with two DOFs. For the first robot, the calibration is performed by first using a direct identification method. The second work to improve the absolute accuracy of PreXYT is based on the direct kinematic calibration method. The accuracy of the five-bar robot is improved by using a self-calibration approach that exploits the working modes and assembly modes, to reduce the number of calibration positions. Therefore, all possible robot configurations for each calibration position are retained. This approach is particularly attractive in its simplicity: at each calibration position, a precision ball is permanently installed as a target for measurements. The positions of these balls, placed on a removable platform, is measured only once, using a coordinate measuring machine (CMM). After reinstalling the platform of the robot, the calibration can be done anytime using only the information from the actuator encoders. Calibration and validation data are collected using two measuring devices. The first device is an articulated arm coordinate measurement, from FARO Technologies, the second is a Mitutoyo CMM. The measurement uncertainties of these machines are respectively ±18 μm and ±2.7 μm. Knowing that the quality of calibration is inversely proportional to the measurements uncertainties, using accurate instruments with near-complete geometric models allowed us to achieve these results: the maximum errors in position and orientation were reduced respectively to 0.044 mm and 0.009° for the PreXYT, within a circle of 170 mm in diameter. For the robot DexTAR, the maximum error of position was reduced to 0.077 mm throughout its workspace, i.e. 600 mm × 600 mm. Improve the accuracy of robots beyond these values, using only kinematic approaches, may be unlikely. In this sense, the addition of modeling and compensation of non-kinematic errors would be useful to obtain better results.

Date	1 Aug 2012
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Ilian Bonev (Supervisor)