Automatic Tool Wear Inspection by Cascading Sensor and Image Data

Robbert Verbeke, Lars De Pauw, Fabian Fingerhut, Tom Jacobs, Toon Goedemé, Elena Tsiporkova

Abstract: This paper explores the possibility to apply data-driven methods to improve the replacement strategy of worn cutting tools used in milling industry. While the data was generated in a controlled environment, the conditions under which the data were generated varied to make them more realistic. Both indirect (sensor data) and direct (images) data was captured and individually modelled. We propose a cascading approach, combining both modalities in a sequential way, and show that this methodology leads to very accurate tool replacement strategy while keeping production efficiency high, as long as the cutting speed does not change drastically.

External IDs:dblp:conf/pkdd/VerbekePFJGT23