计算物理 ›› 2009, Vol. 26 ›› Issue (5): 699-702.

• 研究论文 • 上一篇    下一篇

基于Tahoe框架的某夹具并行计算

范宣华1,2, 吴瑞安1, 郝志明1, 何颖波1   

  1. 1. 中国工程物理研究院总体工程研究所, 四川 绵阳 621900;
    2. 北京大学力学与空天技术系, 北京 100871
  • 收稿日期:2008-05-04 修回日期:2008-07-26 出版日期:2009-09-25 发布日期:2009-09-25
  • 作者简介:范宣华(1981-),男,工程师,博士生,从事振动理论及并行计算方面的研究,四川绵阳919信箱402分箱621900.
  • 基金资助:
    中国工程物理研究院十一五重大预研专项课题资助项目

Parallel Computing of Clamp Structure in Tahoe Frame

FAN Xuanhua1,2, WU Ruian1, HAO Zhiming1, HE Yingbo1   

  1. 1. Institute of Systems Engineering, China Academy of Engineering Physics, Mianyang 621900, China;
    2. Department of Mechanics and Aerospace Engineering, Peking University, Beijing 100871, China
  • Received:2008-05-04 Revised:2008-07-26 Online:2009-09-25 Published:2009-09-25

摘要: 在开源软件Tahoe框架基础上,结合有限元前后处理程序MSC.Patran及Tecplot,对某复杂夹具进行建模.通过区域分解、编制接口和采用PHG中提供的PCG(preconditioned conjugate gradient,预处理共轭梯度法)迭代解法成功实现262×104自由度模型的串、并行计算.结果表明,并行计算收敛速度更快,4进程并行计算时间不到串行计算时间的1/4.通过与商用程序MSC.Nastran比较,验证计算结果的正确性.利用大型并行计算机对该模型并行计算性能进行研究,获得最高32进程的并行计算加速比.研究表明,改进后的Tahoe计算框架对于开展大规模自由度下的结构并行计算分析研究是可行的,并且随计算节点增加,并行计算过程基本呈线性加速.

关键词: 并行计算, 有限元, 区域分解, 预处理共轭梯度法, 加速比

Abstract: In Tahoe frame (an open source software) and finite element (FE) pre-processing/post-processing soft wares (MSC.Patran and Tecplote),an FE model of clamp structure was created.Serial and parallel computations of a model with 2.62 million-freedom degree are achieved successfully with domain decomposition,compiling of interfaces and iterative algorithm solvers provided in PHG solvers(preconditioned conjugate gradient methods).It indicates that convergence of parallel computation is faster than serial one's,and computation time of a four-procedure parallel is less than one forth of that of serial one.Accuracy of computing is validated by comparing with result of MSC.Nastran.Parallel computing performance of the frame is tested with a large parallel computer.The maximum testing number is 32 nodes.It shows that an updated Tahoe frame is feasible in parallel computing of large-scale freedom degree models.Parallel computing time is nearly linearly reduced with increasing of computing nodes.

Key words: parallel computing, finite element, domain decomposition, preconditioned conjugate gradient, speedup

中图分类号: