| Co-Investigator(Kenkyū-buntansha) |
OKUDA Hiroshi The Univeristy of Tokyo, Graduate School of Enginnering, Associate Professor, 大学院・工学系研究科, 助教授 (90224154)
YOSHIMURA Shinobu The Univeristy of Tokyo, Environmental Studies, Professor, 大学院・新領域創成科学研究科, 教授 (90201053)
|
|---|
| Research Abstract |
The requirements for analyzing various kind of phenomena in nuclear power plants are the capability to analyze the complex phenomena in nuclear reactor, the efficiency and speed of pre-processing for analyzing them accurately, to make the main analysis process seemless and using massively parallel computers for reducing time and storage costs.
In this study, therefore, we developed the following key technologies ; (1) introducing a new analysis strategy, which does not need finite element meshes and are suitable for massively parallel environment, (2) using such a meshless method, developing the massively parallel large scale 3-dimensional analysis codes, which includes free surface and oscillate solid, (3) finally, developing the non-linear and dynamic stress analysis system for the massively parallel large scale 3-dimensional structures considering the effect of thermal-fluid coupled analysis.
In this study, we used Free Mesh Method (FMM) as a basic technology, which was a kind of meshless methods. Using this method, we developed fluid-structure coupled analysis system. Concretely, (1) developing the domain decomposition module for analyzing in parallel from nodal based data, (2) developing parallel iterative linear solver with the preconditioning for structure and fluid analysis, (3) developing prototype system for evaluate the system, (4) verifying the efficiency of developed system on the PC clusters, (5) verifying the efficiency of developed system on Hitachi SR2201 and SR8000, (6) integrating the developed system with user interface functions.
Because of the high applicability of Free Mesh Method to the parallel processing, we could develop the 3-dimensional large, scale coupled analysis system, which showed high parallel efficiency on both PC clusters, and parallel computers.
|
