Augmenting Milling Process Data for Shape Error Prediction

Abstract

New integrated sensors and connected machine tools generate a tremendous amount of in-depth process data. The continuous transformation of the obtained data into deployable machining knowledge allows for faster ramp-ups, more reliable process outcome and higher profitability. A system for recording data from various sources - including a simultaneous material removal simulation - is implemented to aggregate and store process data. In addition to the simulation results, process data from the machine control, cutting forces and shape error samples are collected. A series of slot milling processes are carried out with varying cutting speed, feed per tooth and width of cut in a full factional design. In order to continuously evaluate process data, automatized methods are required. This is achieved using the simulation results to determine all relevant cutting conditions. Dependencies between cutting parameters, sensor signals and cutting result are identified and quantified. However, a one-dimensional model does not predict the shape error accurately. As an alternative model, a multidimensional model based on a Support Vector Machine is trained, using process forces and simulation data. The obtained prediction accuracy is significantly higher compared to the one-dimensional model and can be used to design highly reliable cutting processes.