Back to Search Start Over

Parallel 3D Mesh Generation using Geometry Decomposition.

Authors :
Jurczyk, Tomasz
Głut, Barbara
Breitkopf, Piotr
Source :
AIP Conference Proceedings. 2007, Vol. 908 Issue 1, p1579-1584. 6p. 1 Diagram, 1 Chart, 1 Graph.
Publication Year :
2007

Abstract

The paper refers to the problem of parallel generation of 3D meshes for complex objects. Simulation of processes using finite element method (FEM) consists essentially of two phases: generation of finite element mesh and solving the appropriate set of algebraic equations. The parallel solving of such problems has been often reduced to sequential generation of a mesh and than decomposing of this mesh. The latter operation was usually performed sequentially as well. In such approach the only part being parallelized is the solver. In the recent years much attention has been directed to the task of parallelization of the mesh generation process. The need for considering this problem results mainly from the fact that the simulations are currently being run on meshes with very large number of elements. In such cases the sequential generation of meshes poses significant problems regarding both the amount of the required memory and the discretization time. In this work there is described the decomposition strategy of the initial surface 3D mesh for subsequent generation of the volume mesh. Surface meshes are the input data which are being decomposed into sub-domains by cutting them with separators. Next, the surface meshes are constructed for the separator, which are closing the new sub-domains. There is ensured the compatibility of the surface meshes at the interface. This procedure is being continued until the prescribed number of sub-domains is reached. Generation of volume meshes can then be performed for each closed sub-domain on the parallel computer. As a result, the whole volume mesh needn’t be stored in the memory of the single computational node. During the simulation phase there is interchanged only the information about the interface meshes, which stay compatible. Moreover, for single-processor unit, the method can give additional benefits. It’s possible to partition the domain depending on the amount of the available memory and then generate the volume meshes sequentially for each sub-domain. This approach allows to use the sequential mesh generator without any modifications. © 2007 American Institute of Physics [ABSTRACT FROM AUTHOR]

Details

Language :
English
ISSN :
0094243X
Volume :
908
Issue :
1
Database :
Academic Search Index
Journal :
AIP Conference Proceedings
Publication Type :
Conference
Accession number :
25209748
Full Text :
https://doi.org/10.1063/1.2741034