| 引用本文: | 徐权,刘田田,冷珏琳,等.面向复杂几何模型的多级并行四面体网格生成算法.[J].国防科技大学学报,2021,43(2):33-39.[点击复制] |
| XU Quan,LIU Tiantian,LENG Juelin,et al.Multilevel parallel tetrahedral mesh generation for complex geometric models[J].Journal of National University of Defense Technology,2021,43(2):33-39[点击复制] |
|
| |
|
|
| 本文已被：浏览 11615次 下载 5825次 |
| 面向复杂几何模型的多级并行四面体网格生成算法 |
| 徐权1,2,刘田田1,2,冷珏琳1,2,杨洋1,2,郑澎3 |
|
(1. 中物院高性能数值模拟软件中心, 北京 100088;2. 北京应用物理与计算数学研究所, 北京 100094;3. 中国工程物理研究院计算机应用研究所, 四川 绵阳 621900)
|
| 摘要: |
| 高性能计算机的快速发展为数值模拟提供了必需的硬件环境,数值模拟领域对网格的需求已高达数十亿到数百亿量级,而网格生成作为数值模拟的重要一环,其发展则相对滞后,很难满足并行数值模拟求解器对大规模网格的需求。本文面向复杂几何模型提出一种多级并行四面体网格生成算法,该算法首先基于模型的几何特征建立网格的尺寸场,并基于尺寸场和几何实体间的邻接关系对几何实体进行分组,将分组后的几何实体分配到不同的计算节点,在计算节点间采用前沿推进法实现三角形面网格的并行生成,然后在计算节点内对三角形面网格进行二级区域分解,将分解后的子网格分配到各进程中,最后在进程内采用多线程并行方法实现四面体网格的并行生成。通过实际应用三峡大坝模型进行验证,该算法具有良好的并行效率和可扩展性,可以在数千处理器核上实现十亿规模高质量四面体网格的并行生成。 |
| 关键词: 网格生成 区域分解 四面体网格 并行计算 |
| DOI：10.11887/j.cn.202102005 |
| 投稿日期：2019-09-30 |
| 基金项目:科技部国家重点研发计划资助项目(2017YFB0202203)；国家自然科学基金资助项目(11801037) |
|
| Multilevel parallel tetrahedral mesh generation for complex geometric models |
| XU Quan1,2, LIU Tiantian1,2, LENG Juelin1,2, YANG Yang1,2, ZHENG Peng3 |
|
(1. CAEP Software Center for High Performance Numerical Simulation, Beijing 100088, China;2. Institute of Applied Physics and Computational Mathematics, Beijing 100094, China;3. Institute of Computer Application CAEP, Mianyang 621900, China)
|
| Abstract: |
| The rapid development of high-performance computer provides the hardware environment for high performance numerical simulations. Many numerical simulations have the ability of handling meshes from billions of elements to tens of billions. However, the development of parallel mesh generation, which is a critical step in numerical simulation, is relatively lagging behind. Thus, a multilevel parallel unstructured tetrahedral mesh generation algorithm for complex geometric models was presented. Firstly, a sizing-function of the mesh was established on the basis of the geometric features of the model, and the geometric entities were grouped on the basis of the sizing-function and the adjacency relationship between the geometric entities. The grouped geometric entities were distributed to different computing nodes. Then the triangular meshes were generated by the advancing front method among the nodes. And the triangular meshes were decomposed into sub-meshes which were distributed to each process. Finally, a multi-threaded parallel algorithm was used to generate tetrahedral meshes in the process. The proposed parallel mesh generation algorithm was validated on the Three Gorges Dam model, and the results show that the presented algorithm has good parallel efficiency and scalability, and can generate billions of high quality tetrahedral meshes for modern supercomputers with thousands of processor cores. |
| Keywords: mesh generation domain decomposition tetrahedral mesh parallel computing |
|
|
|
|
|
