Predictive Control of a Robotic Grinding System

Abstract

This paper describes a modified predictive controller that has been interfaced with the robotic weld bead grinding system currently under development at The Charles Stark Draper Laboratory. The objective of the predictive control in the grinding system is to invert the previously modeled grinding dynamics. This is accomplished via a link between the controller and a grinding simulator that has been developed, based on an exchange of geometric and force data used by both algorithms. The interface utilizes vision data obtained from previous grinding passes to generate a force trajectory used by the robot on the subsequent pass. This trajectory is precomputed by the predictive controller to yield a specific weld bead shape. Due to system limitations, measurement noise, and modelling errors, the beads are not ground perfectly flat. However, their cross sectional area (or height) variations are reduced significantly. Thus, the bead is flattened although not completely.