Finite Element Mesh Generation using Growing Cell Structures Networks

A new method for the automated generation of eecient meshings for the nite element method (FEM) is presented and applied to a general problem from the eld of computer graphics | radiosity. Radiosity designates a class of algorithms which calculate the global ow of light in a virtual three-dimensional environment. The representation of radiosity computed by using an FEM approach can be regarded as a mesh of reference transfer rays. This representation must be as compact as possible due to the huge geometries demanded in todays applications. To obtain this compactness, in this work, the generation of the meshing is accomplished using the incremental supervised growing cell structures approach (ISGCS). It creates a neural network with a very eecient internal representation which is interpreted and utilized directly as radiosity FEM meshing.