An Adaptive Mesh Generation With Parameterized Learning

Abstract

This paper presents an adaptive mesh generation with parameterized learning. The present method does not need to perform iterative processes of field analysis in contrast with the conventional adaptive meshing methods. The present method evaluates mesh qualities for each element by means of evaluation function, which is weighted linear combination of shape and area of elements, distance from material corners and so on. The element which has the worst value of the evaluation function is divided into a few elements according to its shape, and this procedure is repeated until the prescribed number of elements is obtained. By using the simple Genetic Algorithm (sGA), the weighting coefficients are optimized through learning with example models such that the resultant mesh has the lowest numerical error.
The good mesh can be obtained without time-consuming computation, since the weight values for the mesh features are learned by the sGA. The present method would allow us to realize effective design and development of electromagnetic machine and devices.