Automatic mesh generation based on skilled knowledge by genetic algorithm

Abstract

This paper describes an automatic quadrilateral finite-element mesh generation method for 2 dimensional models based on knowledge of skilled analysts and by using a genetic algorithm. The analysts subdivide a given geometrical model into regions step-by-step, based an their knowledge, and mesh them finely. We derived subdivision rules and a fitness function from the knowledge. The operation order of the rules is not unique so that one model could be subdivided into different sets of regions by the analysts. We regard the problem of subdivision for a geometrical model as a combinatorial optimization problem of the rules, because there must be many choices of the rules at each subdivision step. In addition to using a genetic algorithm, we have proposed a strategy like simulated annealing, which is carried out mesh generation as generative differentiation similar to living organism. We have applied the method to two-dimensional models, and confirmed the effectiveness of the new algorithm.