Computational fluid dynamics (CFD) analysis consists of three phases: 1) pre-processing, e.g., grid generation; 2) fluid analysis; and 3) post-processing, e.g., visualization. With the recent improvement in computer performance, pre-processing takes up a higher proportion of the total number of processes in extensive and complex CFD analyses. In this study, we developed a novel method for automatic generation of robust structured grids during the initial stage of fluid analysis to reduce the computational cost of grid generation. By implementing this method, we successfully reduced the computational cost of grid generation, and approximately thirty million grids were generated on the K computer within thirty minutes. By using these generated grids, over eighty percent weak scaling up to 4000 nodes on the K computer is obtained.
|Translated title of the contribution||J011011 Development of large-scale grid generation method for high-performance parallel fluid computation|
|Publication status||Published - Sept 8 2013|