On-line metal cutting tool condition monitoring. II: tool-state classification using multi-layer perceptron neural networks

This paper outlines a neural networks based modular tool condition monitoring system for cutting tool-state classification. Test cuts were conducted on EN24 alloy steel using P15 and P25 coated cemented carbide inserts and on-line cutting forces and vibration data acquired. Simultaneously the wear lengths on the cutting edges were measured, and these together with the processed data were fed to a neural network trained to distinguish tool-state. The first part of the investigation concentrated on tool-state classification using a single wear indicator and progressing to two wear indicators. The developed system was tested for a variety of cutting conditions and its ability to distinguish changes in tooling material and cutting conditions from those arising from tool wear was assessed. The system was found to be capable of accurate tool state classification in excess of 90% accuracy but deteriorated when the cutting conditions were significantly changed. (C) 2000 Elsevier Science Ltd. All rights reserved.