节点文献

采用GPU加速复杂结构角系数计算的方法研究

An Investigation on the Method to Speed Up View Factor Computation With GPU of a Complex Configuration

  • 推荐 CAJ下载
  • PDF下载
  • 不支持迅雷等下载工具,请取消加速工具后下载。

【作者】 胡金龙吉洪湖黄伟斯仁

【Author】 HU Jin-Long;JI Hong-Hu;HUANG Wei;SI Ren;College of Energy and Power,Nanjing University of Aeronautics and Astronautics;

【机构】 南京航空航天大学能源与动力学院

【摘要】 采用离散传递法计算目标的红外辐射特征首先要计算角系数,对于复杂目标,角系数计算会耗费大量时间。针对角系数计算中遮挡判断问题,提出了一种不需要求解微元面中心点连线与遮挡面交点的矢量法。开发了基于矢量法的GPU并行程序,并使用一个有理论解的模型验证了程序的正确性,最后针对航空发动机排气系统复杂模型,在一台配置有448个核心的GPU处理器和主频为3.4GHz CPU处理器的工作站上,评估了基于矢量法的GPU程序相对单核CPU程序的加速效果。结果表明;在微元面数量为13670时,矢量法可获得约73倍的加速比,是采用线面求交法获得的加速比的1.4倍。使用离散传递法计算航空发动机排气系统的红外辐射特征时,仅将角系数模块使用GPU并行计算,可减少45%的计算时间。

【Abstract】 To calculate the IR signature by utilizing of Discrete Transfer Method,the view factor should be calculated in advance,which is time consuming.A vector method was proposed to detect if there is an obstruction between two surface elements without the need to finding the intersection point of the line between the centroids of the two surface elements and the obstruction.A computing code based on the vector method was developed by using CUDA.The vector method and the code was validated with a theoretical model.The speed-up ratio of the vector method was study with an aero-engine exhaust system model which was divided into 13670 triangle surface elements.A workstation with a 448 cores GPU and 3.4GHz CPU was used in the study.The result show that the speed-up ratio of vector method is 73 times,comparing with single CPU code,which is 1.4 times of the speed-up ratio by using Line-Plane Intersection Method.The calculating time can be reduced by 45% if only the view factor calculation was implemented with GPU for computing the IR signature of the exhaust system by using Discrete Transfer Method.

【关键词】 角系数GPU遮挡判断并行计算
【Key words】 view factorGPUobstruction detectionparallel computing
【基金】 航空科学基金(No.20132752039)
  • 【文献出处】 工程热物理学报 ,Journal of Engineering Thermophysics , 编辑部邮箱 ,2017年06期
  • 【分类号】O434.31;TP338.6
  • 【被引频次】1
  • 【下载频次】90
节点文献中: 

本文链接的文献网络图示:

本文的引文网络