Étalonnage de robots industriels à l'aide d'un système portable de photogrammétrie

Abstract

The present work intends to analyze the use of a portable photogrammetric system for industrial robot calibration. The device used in this study is the MaxSHOT 3D system provided by Creaform. Robot calibration consists in identifying a geometric or non-geometric mathematical model of the robot in order to improve the control of the robot and increase its positioning accuracy. The following thesis presents a non-geometric calibration of a relatively small robot, the FANUC LR Mate 200iC. For comparative purposes, the calibration procedure is also used with a second device, the FARO Laser Tracker ION. This laser tracker is also used to acquire a validation dataset of approximately 1,000 measurements. First, the robot mathematical model is elaborated. Forward and inverse kinematic models are presented following the modified Denavit-Hartenberg method (DHM). In addition, joint compliance is modeled using the Newton-Euler algorithm that considers the joints as torsion springs. This yields a category 3 model in order to proceed to a non-geometric calibration. Next, the portable photogrammetry system and the laser tracker are introduced. Being that these two devices are diametrically opposite in their function, the typical usage is compared. The portable photogrammetry system is affordable (approximately CA$ 27,000), easy to use, and allows the observation of a scene (hundreds of 3D point). However, it requires that the measured objects be stationary, it calls for a large number of manipulations and is nearly impossible to use in an automated manner. In contrast, the laser tracker can measure continuous trajectories and allows full automation of the measurement process. Yet, it is very expensive (more than CA$ 100,000), allows the measurement of only one 3D point at a time, and is very sensitive to environmental conditions (e.g. temperature, vibrations, air currents, etc.). Taking the limitations of both measurement devices into consideration, a pseudo-random algorithm was used to generate a 1 000 robot poses. This algorithm ensures that for each proposed configuration, the measurement targets on the robot end-effector are visible by both measuring devices. Subsequently, an observability analysis is performed to select the poses most appropriate for calibration in order to identify the robot’s parameters. A set of only 34 configurations are used for the identification process, leaving the other 966 configurations for the validation phase. The final section of the thesis presents the least squares robot parameter identification method. The identified models are presented and their performance is analyzed. Calibration using data from the MaxSHOT 3D gives a robot’s positioning accuracy of 0.469 mm, while 0.365 mm is obtained by using data from the laser tracker. Admittedly, as the validation measurements were acquired with the laser tracker, bias favors this device. Due to this bias and because there is a similarity between the accuracy obtained with both devices, relative to absolute precision, their calibration performance is concluded to be similar. Nevertheless, it is suggested to consider all other features, as they these are still two very different devices.

Date	9 Sept 2015
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Ilian Bonev (Supervisor)