摘要
在分布式并行计算机环境下进行了有限元并行算法的研究,建立了结构动力分析的两种隐式积分方法(Newmark方法和Wilson-θ方法)的并行化方法与算法步骤,设计了变带宽一维存储时有效刚度矩阵的三角分解并行算法;基于Transputer的分布式MIMD并行计算机上,采用3L并行FORTRAN编写了计算程序,并将其移植到有限元串并行混合分析软件PFEM中。以平面问题和空间板弯问题作为实例进行了数值计算。结果表明计算方法具有较高的并行效率。当自由度为7579,最大带宽为726时,2个和3个处理器工作的并行效率分别为0.70和0.55。
The parallel algorithms for FEM in the distributed memory environment are studied. Two implicit parallel time integration algorithms are presented, which are, respectively, the paralleled Newmark method and Wilsion θ method. Emphasis is placed on the parallel LDL T decomposition of the equivalent stiffness matrices in the two methods. The algorithms are carried out in the MIMD environment on Transputer, and are also integrated into the available PFEM package which is a serial parallel mixed FEM softward. As verifying examples, a plane problem in elasticity and a 3D plate problem are presented. The result shows that the parallel efficiencies in present method are satisfactory as well, which are 0.70 and 0.55 respectively when the degree of freedom (DOF) is 7579, and the maximum bandwidth is 726 under two processors and three processors.
出处
《清华大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
1998年第11期77-80,共4页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金
关键词
有限元法
隐式积分方法
并行算法
结构动力分析
finite element methods
implicit integration method
parallel algorithms
distributed memory MIMD computer