Application of neural networks to automatically classify rotational parts into part families

Abstract

This thesis details the procedure for automatic classification of rotational parts into part families using an artificial neural network. The classification is based on geometric features and tolerances. The neural network paradigm employed belongs to a class of Adaptive Resonance Theory Models. The training of the network was done on a commercially available software package. The major conclusions that can be drawn from this research are: 1) The ART2 paradigm in neural networks is capable of automatically grouping parts into part families. 2) The number of groups increased with an increase in vigilance of the system. The number of groups also decreased with a decrease in vigilance. 3) The order in which the parts are input to the system has no effect on the system performance. Thus the overall conclusion of this thesis is that neural networks are indeed an appropriate tool to automatically group rotational parts into part families.