大规模非结构化四面体网格高性能解耦并行生成算法

面向大规模科学计算，研究非结构化网格并行生成方法，提出高质量、高效率的四面体解耦并行网格生成算法，解决大规模科学计算计算中的网格生成与再生成、网格光顺与移动等难点问题。重点研究分区算法，提出基于AFT-Delaunay算法的虚拟界面诱导分区界墙生成方法，实现子区域间的完全解耦；研究基于节点移动的网格并行优化算法，提出解耦的并行点球弹簧修匀法，实现网格光顺与移动一体化；研究高时空效率的网格数据结构，提出基于点-单元结构的拓扑信息存储结构，减少数据冗余，提高网格拓扑信息的查找、添加、删除和更新操作效率；研究非流形约束的AFT-Delaunay网格生成算法，提出基于八叉树背景网格的高效布点算法，提高算法在分区内的生成速度。研发出并行程序，实现十亿至百亿个单元量级的三维非结构化四面体网格的并行生成。建立三维并行网格生成通用平台，为大规模科学计算提供核心支撑技术。

For large-scale numerical simulations, we investigate parallel mesh generation method and propose a high quality and effectiveness decoupled parallel tetrahedral mesh generation algorithm, and establish a general platform for mesh parallel generation, to resolve mesh generation, mesh smoothing, remeshing and reposition problems in large-scale numerical simulation. We study domain decomposition approaches, and propose a virtually inducting decomposition method based on AFT-Delaunay mesh generation to enhance element quality around sub-domain boundary. We research a reposition-based mesh optimization method, and propose a decoupled ball-vertex method to reduce the degree of coupling and increase the speed of optimization in classic spring analogy. We study high-effective mesh data structure, and propose node-element based topology data structure to reduce data redundancy, and enhance data inquiring, adding, deleting and updating operations for unstructured mesh data. We study un-manifold constrained AFT-Delaunay mesh generation method, and propose octree-based inner nodes generation method to improve the speed of mesh generation in sub-domain. Also, we utilize the above studies to achieve the parallel mesh generation in 1-10 billions elements magnitude, and establish a parallel mesh generation software platform which provide modeling technical support for large-scale numerical simulation.