Improving the quality of finite volume meshes through genetic optimisation

Abstract

Mesh quality issues can have a substantial impact on the solution process in Computational Fluid Dynamics, leading to poor quality solutions, hindering convergence and in some cases, causing the solution to diverge. In many areas of application, there is an interest in automated generation of finite volume meshes, where a meshing algorithm controlled by pre-specified parameters is applied to a pre-existing CAD geometry. In such cases, the user is typically confronted with a large number of controllable parameters, and adjusting these takes time and perseverance. The process can, however, be regarded as a multi-input and possibly multi-objective optimisation process which can be optimised by application of Genetic Algorithm techniques. We have developed a GA optimisation code in the language Python, including an implementation of the NGSA-II multi-objective optimisation algorithm, and applied to control the mesh generation process using the snappyHexMesh automated mesher in OpenFOAM. We demonstrate the results on three selected cases, demonstrating significant improvement in mesh quality in all cases.