A neural network generator for tetrahedral meshes

An automatic mesh generator for tetrahedral finite elements, based on an artificial neural network, is described. Starting from an initial moderately coarse mesh, the generator grows the mesh up to the user-specified number of nodes. The growth is driven by a node probability density function, which is piecewise constant on the initial tetrahedra and expresses the user's expertise on the problem. The generator takes the mesh quality into account in several parts, so that high-quality meshes are obtained.