Self-Organizing Neural Network Application to Drill Wear Classification

Abstract

The article describes an application of a simulated neural network to drill wear classification from cutting force signals generated by the drilling process. As the input to the neural network, a multicomponent vector composed of a sensory part and a descriptive part is used. The components of the sensory part represent characteristic features of the cutting momentum and the feed force power spectra, while the descriptive part encodes the corresponding drill wear class. During adaptation, the self-organizing neural network is used to form a set of prototype vectors representing an empirical model of the observed drilling process. The model is used in the analysis mode of the system for an on-line classification of the drill wear from the cutting forces. The performance of the developed information processing system is experimentally demonstrated by classification of drill wear during machining on a steel workpiece.