ライブラリ登録: Guest
Begell Digital Portal Begellデジタルライブラリー 電子書籍 ジャーナル 参考文献と会報 リサーチ集
International Journal for Multiscale Computational Engineering
インパクトファクター: 1.016 5年インパクトファクター: 1.194 SJR: 0.554 SNIP: 0.68 CiteScore™: 1.18

ISSN 印刷: 1543-1649
ISSN オンライン: 1940-4352

International Journal for Multiscale Computational Engineering

DOI: 10.1615/IntJMultCompEng.v2.i3.50
23 pages

Multilevel Parallel Programming for Multiscale Modeling of Composite Materials

Paul M. Eder
Department of Mechanical Engineering, Ohio State University, Columbus, Ohio 43210
James E. Giuliani
Science and Technology Support Group, The Ohio Supercomputer Center 1224 Kinnear Rd., Columbus, Ohio, 43210
Somnath Ghosh
Department of Civil Engineering, Johns Hopkins University, Baltimore, MD 21218

要約

This paper presents work aimed at implementing an efficient multilevel parallel model for an adaptive multiscale finite element model. The FEM model combines a traditional displacement based finite element model with a microstructural Voronoi cell finite element method (VCFEM) for multiscale modeling of heterogeneous microstructures with nonuniform microstructural heterogeneities. Three levels of hierarchy are used in the model, including (a) level-0 of pure macroscopic analysis; (b) level-1 of macro-micro coupling, used for signaling the switch over from macroscopic analyses to pure microscopic analyses; and (c) level-2 regions of pure microscopic modeling. A distributed/shared memory (DSM) cluster system is used for code development and execution, where multiprocessor nodes offer shared memory on the node and distributed memory between the nodes. The approach uses multiple parallel models to efficiently distribute the level-1 and level-2 workloads across multiple workstations based on computational requirements. The Message Passing Interface (MPI) library is used for distributed memory decomposition between nodes, and multithreading using the OpenMP (OMP) library is used for shared memory decomposition on each node. An efficient iterative multigrid solver is also integrated. The details of these implementations are discussed and numerical results, which demonstrate the ability of the parallel model to solve problems in a fast and efficient manner, are provided.