An Adaptive Modeling Method for a Robot Belt Grinding Process

Abstract

A robot belt grinding system has a good prospect for releasing hand grinders from their dirty and noisy work environment. However, as a kind of manufacturing system with a flexible grinder, it is a challenge to model its processes and control grinding removal precisely for free-formed surfaces. In the belt grinding process, material removal is related to a variety of factors, such as workpiece shape, contact force, and robot velocity. Some factors of the grinding process, such as belt wear, are time variant. In order to control material removal in the robot grinding process, an effective approach is to build a grinding process model that can track changes in the working condition and predict material removal precisely. In this paper, an adaptive modeling method based on statistic machine learning is proposed. The major idea is to build an initial model based on support vector regression using historical grinding data serving as training samples. Afterward, the trained model is modified according to in situ measurement data. Robot control parameters can then be calculated using the grinding process model. The results of the blade grinding experiments demonstrate that this approach is workable and effective.