DETECTION OF TOOL BREAKAGE IN MILLING OPERATIONS .2. THE NEURAL-NETWORK APPROACH

In this study, supervised and unsupervised neural network systems were used to detect tool breakage in milling operations. The restricted Coulomb energy (RCE)-type neural network was used for supervised learning. The effectiveness of the encoding method was tested using the RCE network on both simulated and experimental cutting force signals. The RCE network correctly categorized more than 98% of the presented data sets after training, which included simulated and experimental cutting force data. An unsupervised neural network was prepared based on the adaptive resonance theory (ART2). The ART2 network was initially trained using simulated signals. The ART2 network was then used to monitor tool breakage while continuously updating learned recognition codes and establishing new categories as needed. The accuracy of ART2 networks on tool breakage detection and proper vigilance factor selection for minimum node assignment issues are discussed. The ART2 network correctly categorized 97.2% of the presented experimental data sets after initial learning on simulated cases.