A SELF-ORGANIZING NEURAL NETWORK APPROACH FOR AUTOMATIC MESH GENERATION

Abstract

A new automatic mesh generator, SOFT (Self-Organizing Finite-element Tessellation), based on self-organizing neural networks has been demonstrated. With user-supplied mesh density function and boundary mesh this mesh generator provides a graded mesh, of which asymptotic characteristics are quite similar to weighted Dirichlet tessellation and dual Delaunay triangulation. Local mesh restrictions such as fixed boundary and/or internal meshes are easily incorporated in this new mesh generator. Although the algorithm is applicable to general n-dimensional meshes 2-dimensional rectangular and triangular meshes are presented for simplicity.