LAUSR

Abstract During the manufacturing of machined workpieces with very narrow tolerances in serial production, slight quality deviations can cause high scrap rates and a waste of resources. In-process quality surveillance makes it possible to take measurements during the machining process as soon as a deviation occurs. Within this context, the quality of machined workpieces has to be predicted simultaneous to the machining process without affecting it. In this paper, the machine learning method of random forests (RF) is employed to predict geometrical and dimensional quality characteristics of reamed bores based on process data. Process data were collected during the serial production of high-precision hydraulic valves by a milling-machine. It is found that the values of the mean absolute error were on average very low for all predicted quality characteristics. The results have shown that RF is able to predict quality characteristics on basis of process data accurately which can lead to a more sustainable manufacturing of drilled and reamed bores.