Towards manufacturing robotics accuracy degradation assessment: A vision-based data-driven implementation

In this manuscript we report on a vision-based data-driven methodology for industrial robot health assessment. We provide an experimental evidence of the usefulness of our methodology on a system comprised of a 6-axis industrial robot, two monocular cameras and five binary squared fiducial markers. The fiducial marker system permits to accurately track the deviation of the end-effector along a fixed non-trivial trajectory. Moreover, we monitor the trajectory deflection using three gradually increasing weights attached to the end-effector. When the robot is loaded with the maximum allowed payload, a deviation of 0.77mm is identified in the Z-coordinate of the end-effector. Tracing trajectory information, we train five supervised learning regression models. Such models are afterwards used to predict the deviation of the end-effector, using the pose estimation provided by the visual tracking system. As a result of this study, we show that this procedure is a stable, robust, rigorous and reliable tool for robot trajectory deviation estimation and it even allows to identify the mechanical element producing non-kinematic errors.