New methods for calculating the force distribution within belt grinding processes

Abstract

In order to simulate belt grinding processes (e.g. for process planning or path planning) one usually needs information about the contact zone and contact forces. Typically, an unacceptable computational effort is required for good simulation results, since these contact problems are usually of a nonlinear nature. In this paper, the application of support vector machines (SVM) is presented. The SVM is a learning machine that aims at finding a function that optimally fits given observations. The main advantage of SVM is its fast evaluation during simulation. However, a single training phase with an extensive amount of observation data has to be done once before the simulation can take place. From a practical point of view, it is very often not feasible to sample these observation data by experiments. At this point special Finite element methods for contact problems can be applied very efficiently. In order to obtain as accurate as possible training data, an adaptive finite element method for contact problems has been developed.