FINITE ELEMENT MESHING USING KOHONEN'S SELF-ORGANIZING MAPS

Abstract

This paper presents an application of self-organizing maps for finite element meshing. The main idea is to give a geometrical interpretation to the neural distribution in the synaptic space. A preliminary partitioning of the geometry into convex subdomains is necessary. For each subdomain, a network is chosen from the node distribution on the contour, a distribution which is already known. The weights of the peripheral neurons are computed to coincide with this distribution. The meshing is obtained after the network has organized itself.