Consistent Multi-Threaded Mesh Refinement with Adaptive Length Scale Estimation for Moving Boundary Problems

Abstract

Computational fluid dynamic simulations with moving boundaries are widely involved in high performance computing applications. For problems with large-displacement or large-deformation boundaries, mesh cells near the boundaries are often excessively stretched or compressed, thus it's hard to maintain a high-quality mesh. To deal with the distorted cells, this paper adopts the mesh refinement method based on the open source software OpenFOAM. In order to achieve the desired effects of localization and adaptation, we propose an adaptive length scale estimation algorithm based on the specified growth factor and current edge lengths. Considering the inconsistency problems for the original implementation of parallelization, an optimized multi-threaded master/worker model is developed for the process of edge checking. Experiments show that our adaptive length scale estimation algorithm works well for moving boundary problems. Compared to the original mesh deformation, using the adaptive mesh refinement could greatly improve the mesh quality. In parallel testing, all the results are consistent and a maximum speedup of 3.8 is achieved on a computing node of 24 cores.