Application of Kohonen Neural Network to Tool Monitoring in Blanking

Abstract

A dynamic growth of artificial neural network applications in engineering is mainly confined to multi-layered feed-forward back-propagation neural networks which are based on supervised learning. In this research an alternative approach based on unsupervised learning was investigated. It was implemented in the form of a modified self-organizing Kohonen network and used for the classification of tool wear in blanking. The modification, based on Ahalt's implementation of a conscience mechanism, enabled very close tool states to be distinguished. The results achieved suggest that the method could be successfully used in monitoring and diagnostic tasks in all kinds of industrial processes.