Development of a Learning-Based Control System for Robot-Assisted Grinding of Rubber Roller

Grinding rubber rollers to a specific geometry presents significant challenges due to the material's non-linear behavior. Currently, this process has not been successfully automated, primarily because the control concepts implemented thus far do not achieve the desired accuracy for economical operation. Consequently, these rollers are manually ground to the required geometry on a grinding wheel. Previous work has led to the development of a system that automates the grinding of rollers using a robot. The objective of this paper is to demonstrate, through modeling, how a learning-based control system can be developed for this process to reduce production time and facilitate an economical operation. To this end, we propose two control concepts: one that directly predicts the required control variable using a learning-based controller, and another that employs an AI-corrected control mechanism, which utilizes the control variable provided by a proportional controller to correct the error between the control variable and the control difference. Finally, we conduct a comparative simulation of the performance of these concepts against a conventional controller.