广东工业大学学报 ›› 2015, Vol. 32 ›› Issue (04): 138-144.doi: 10.3969/j.issn.1007-7162.2015.04.025

• 综合研究 • 上一篇    下一篇

有限元求解器Calculix预处理并行优化方法

林欣达1,林穗1,姜文超1,李东明2,王多强3   

  1. 1.广东工业大学 计算机学院,广东 广州 510006;2.广州船舶及海洋工程设计研究院,广东 广州 510250;3.华中科技大学 计算机科学与技术学院,湖北 武汉 430074
  • 收稿日期:2014-04-11 出版日期:2015-12-04 发布日期:2015-12-04
  • 作者简介:林欣达(1989-),男,硕士研究生,主要研究方向为并行与分布式计算.
  • 基金资助:

    广东省自然科学基金资助项目(S2012040006729);广州市科技计划项目(2012Y200040)

The Parallel Optimization of Preconditioning for the Finite Element Solution Calculix

Lin Xin-da1, Lin Sui1, Jiang Wen-chao1, Li Dong-ming2, Wang Duo-qiang3   

  1. 1.School of Computers, Guangdong University of Technology, Guangzhou 510006, China;2.Guangzhou Marine Engineering Design & Research Institute, Guangzhou 510250,China;3.School of Computer Science & Technology, Huazhong University of Science and Technology, Wuhan 430074,China
  • Received:2014-04-11 Online:2015-12-04 Published:2015-12-04

摘要: 针对船舶疲劳强度分析中大规模数值计算分析的效率问题,在广州超算中心先导系统环境下对开源有限元求解器Calculix中求解线性方程组迭代法的预处理方法进行并行优化,提出基于列主元的多行双门槛的不完全LU分解预处理方法,证明该方法下的不完全LU分解可以进行下去,并在广州超算中心先导系统环境下开发原型系统,运用实际测试算例验证了该方法在相同的条件下,有效缩短了船舶疲劳强度分析中数值分析部分的计算时间.

关键词: 有限元求解器;稀疏线性方程组;不完全预条件方法;船舶疲劳强度分析;超级计算机

Abstract: Aiming at the problems of large scale numerical computing and analysis in ship strength fatigue analysis, this paper presents an improved mechanism with optimized pretreatment of linear equation group iteration at open-source finite element solver Calculix in National Supercomputing Center in Guangzhou. It can resolve the multi-row double threshold incomplete decomposition based on column pivoting and the incomplete LU decomposition proves to proceed. Hence the prototype system has been developed in the center. And the practical experiments and testing results show that the computing time of ship strength fatigue analysis can be reduced effectively.

Key words: finite element solver; sparse linear equations; incomplete decomposition preconditioning; ship fatigue strength analysis; supercomputer

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!