基于多块网格对低层建筑风场分区并行计算

发布时间：2018-03-15 11:03

本文选题：低层建筑风场　切入点：多块网格　出处：《中南林业科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：建筑风场之紊流的数值仿真研究,特别是大规模并行计算研究是当前国内外研究的热点问题,是特别重要的研究领域。准确预测建筑风场可满足工程实践中对建筑规划及设计的需要。工程应用中的各类建筑风场往往网格系统生成较复杂、数值计算量大等特点。如何提高各类建筑风场之紊流数值仿真效率,是近年来的研究热点。本文在湖南省自然科学基金项目“基于多块网格及大涡模拟的三维建筑风场并行分区计算研究(编号:14JJ2103)”的资助下,基于开源程序包Caffa3d.MBRi和工程流变学湖南省重点实验室高性能计算集群对低层建筑风场开展分区并行计算研究。本文主要结论如下:(1)首先,对Caffa3d.MBRi开源程序包进行了调试,基于Caffa3d.MBRi程序包建立数值模拟多块网格系统,然后开展分区并行计算,最后将数值模拟结果与实验结果进行比较,验证了开源程序包Caffa3d.MBRi数值模拟结果的可靠性。(2)基于中南林业科技大学的Dell高性能计算集群平台和Caffa3d.MBRi程序包,在linux系统下搭建了风场数值仿真平台并开展了相关算例多工况分区并行数值模拟,研究了不同屋顶坡度(单坡0°、10°、30°、45°、60°,双坡10°、30°、45°、60°)的单体低层建筑风场,结果表明,在相同的边界条件下,双坡建筑屋面坡度在30度左右和单坡建筑屋面坡度在10度到30度之间,屋面受到的风压相对较小。(3)基于中南林业科技大学的DELL集群平台和Caffa3d.MBRi程序包对双体低层建筑风场开展了多工况干扰特性的数值模拟研究,通过对数值模拟结果的流场分析和表面风压分析发现,双体建筑串列布置时,上风建筑对下风建筑风压影响较大;双体建筑并列布置时,两幢建筑相互之间的影响较小。(4)利用所搭建的风场数值仿真平台,基于自定义分区方法,对研究对象之低层建筑风场开展大规模分区并行数值仿真,研究发现通过使用适当的核数和均衡配置每个核的计算任务实现负载均衡,能充分发挥集群的峰值性能,可提高并行计算效率。本文的创新点主要是搭建了基于多块网格系统及Caffa3d.MBRi程序包的风场数值仿真平台;利用所构建的风场数值仿真平台开展大规模分区并行计算,探讨了负载均衡的分区方式对集群效率的影响,以挖掘集群的峰值性能。
[Abstract]:Numerical simulation of turbulence in building wind field, especially large-scale parallel computation, is a hot issue at home and abroad. Accurate prediction of building wind field can meet the needs of building planning and design in engineering practice. How to improve the efficiency of turbulent numerical simulation of various building wind fields, It is a hot research topic in recent years. This paper is supported by the Hunan Natural Science Foundation project, "the study of 3D wind field parallel partition computing based on multi-grid and large eddy simulation (No.: 14JJ2103)". Based on open source package (Caffa3d.MBRi) and high performance computing cluster of key laboratory of engineering rheology in Hunan province, the paper studies the wind field of low-rise buildings. The main conclusions of this paper are as follows: (1) first of all, the open source package of Caffa3d.MBRi is debugged. The numerical simulation multi-grid system is built based on the Caffa3d.MBRi package, and then the partitioned parallel computing is carried out. Finally, the numerical simulation results are compared with the experimental results. Verify the reliability of Caffa3d.MBRi numerical simulation results of open source package. 2) based on the Dell high performance computing cluster platform and Caffa3d.MBRi package of Central South University of Forestry Science and Technology, A numerical simulation platform of wind field is set up in linux system and parallel numerical simulation is carried out for multiple working conditions. The wind fields of single roof slope (single slope 0 掳10 掳10 掳30 掳30 掳30 掳45 掳60 掳, double slope 10 掳30 掳30 掳45 掳60 掳) are studied. The results show that under the same boundary conditions, The slope of double slope building roof is about 30 degrees and that of single slope building roof is between 10 degrees and 30 degrees. The wind pressure on the roof is relatively small.) based on the DELL cluster platform and Caffa3d.MBRi package of the Central South Forestry Science and Technology University, the numerical simulation of the wind field of the low-rise catamaran building is carried out. Through the flow field analysis and surface wind pressure analysis of the numerical simulation results, it is found that the upper wind building has a great influence on the wind pressure of the downwind building when the catamaric building is arranged in tandem, and when the double body building is arranged in parallel, the wind pressure of the downwind building is greatly affected. The influence of two buildings on each other is small. (4) using the wind field numerical simulation platform, based on the custom partition method, the large-scale parallel numerical simulation of the wind field of the low-rise buildings is carried out. It is found that load balancing can be realized by using proper number of cores and balancing the computing tasks of each core, which can give full play to the peak performance of the cluster. The main innovation of this paper is to build a wind field numerical simulation platform based on multi-grid system and Caffa3d.MBRi package, and to use the wind field numerical simulation platform to carry out large-scale partitioned parallel computing. This paper discusses the effect of load-balanced partition on cluster efficiency in order to tap the peak performance of cluster.
【学位授予单位】：中南林业科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TU312.1

【相似文献】