Machine Learning Based Reconstruction of Process Forces

During milling, process forces are acting on the cutting tool, causing tool deflection and subsequently a shape deviation of the workpiece. To compensate these effects, knowledge of the process forces is required. In this work, machine learning (ML) methods are applied to reconstruct process forces from the drive signals of two different milling centers. The results of a linear regression, bagged trees and a stacked LSTM are presented. The approaches show different results depending on the milling center. Only for the LSTM an error lower than 30 N is achieved for both machine tools. Independent of the ML approach, the results strongly depend on the selection of milling processes used for training.