Scholarlink

<p>For practical large-scale applications of computational fluid dynamics in the aerospace industry, implicit flow solvers are necessitated for efficient simulations. This paper presents the implementation of a solver that employs an unstructured finite volume approach and a Multi-Colored Gauss-Seidel (MCGS) method for steady-state compressible flow simulations on a server equipped with multiple Graphics Processing Units (GPUs). The mesh partition process is completed with PyMetis, and Message Passing Interface (MPI) is utilized for communications between mesh partitions. A parallel coloring algorithm is employed in the pre-processing module. The code is developed using a hybrid programming approach, with the main framework written in Python and the GPU kernel source codes written in C. The transonic turbulent flows over the CHN-T1 transport aircraft model are simulated on unstructured hybrid meshes. The numerical results are compared with experimental data, and the performance of the developed flow simulation framework is analysed.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A multi-colored Gauss-Seidel solver for aerodynamic simulations of a transport aircraft model on graphics processing units

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年6月23日

查看全文
  • Liu Yang,
  • Jian Yang

摘要

For practical large-scale applications of computational fluid dynamics in the aerospace industry, implicit flow solvers are necessitated for efficient simulations. This paper presents the implementation of a solver that employs an unstructured finite volume approach and a Multi-Colored Gauss-Seidel (MCGS) method for steady-state compressible flow simulations on a server equipped with multiple Graphics Processing Units (GPUs). The mesh partition process is completed with PyMetis, and Message Passing Interface (MPI) is utilized for communications between mesh partitions. A parallel coloring algorithm is employed in the pre-processing module. The code is developed using a hybrid programming approach, with the main framework written in Python and the GPU kernel source codes written in C. The transonic turbulent flows over the CHN-T1 transport aircraft model are simulated on unstructured hybrid meshes. The numerical results are compared with experimental data, and the performance of the developed flow simulation framework is analysed.