SBIR Phase II: Implicit Massively Parallel Navier-Stokes Solver for Hybrid Structured-Unstructured Grids

[*] 9530230 这个小型企业创新研究第二阶段项目将开发一个高效的计算流体动力学（CFD）软件包，该软件包使用混合结构-非结构网格来最大限度地减少网格生成工作，并使用大规模并行处理（MPP）计算机来最大限度地减少运行时间。最近出现的高性能MPP计算机理论上减少了使用CFD设计和分析涉及流体流动的系统所需的计算机时间。然而，在实践中，今天MPP中使用的算法是为顺序计算机开发的，因此不能充分利用大规模并行处理的全部功能。此外，MPP计算机还没有简化结构化网格生成过程，这可能需要数周的复杂配置。在第一阶段，一个有效的隐式有限体积法求解二维Navier-Stokes方程的MPP计算机上使用混合结构-非结构网格和区域分解。结构化网格用于求解近壁边界层，非结构化网格用于离散流场的其余部分。在第二阶段，该算法将被扩展到3D，域分解将被扩展和改进，并将增加准动态负载平衡。此外，潜在的市场将扩大增加不可压缩流的能力，在网络上的工程工作站并行运行的能力，和一个互动的MOTIF用户界面，这将使并行几乎透明的用户。 所提出的求解算法有可能大大减少进行CFD分析所需的时间。混合网格的使用将简化实际（复杂）几何形状的网格生成过程，MPP计算机的使用将减少运行时间。由此产生的分析包将在航空航天工业和其他工业中销售。 ***

*** Imlay 9530230 This Small Business Innovation Research Phase II project will develop an efficient Computational Fluid Dynamics (CFD) software package which uses hybrid structured-unstructured grids to minimize the grid generation effort, and massively parallel processing (MPP) computers to minimize run time. The recent emergence of high performance MPP computers has theoretically reduced the computer time required for the design and analysis of systems involving fluid flow using CFD. However, in practice, algorithms used in MPP today were developed for sequential computers and therefore do not harness the full power of massively parallel processing. Also, MPP computers have not simplified the structured grid generation process, which may take weeks for complex configurations. During phase I, an efficient implicit finite-volume method was developed for solving the 2D Navier-Stokes equations on MPP computers using hybrid structured-unstructured grids and domain decompostion. The structured grids are used to resolve boundary layers near walls and unstructured grids are used to discretize the remainder of the flow field. During phase II, the algorithm will be extended to 3D, the domain decompostion will be expanded and improved, and quasi-dynamic load balancing will be added. Also, the potential market will be broadened by adding incompressible flow capability, the ability to run in parallel on networks of engineering workstations, and an interactive MOTIF user interface which will make the parallelism nearly transparent to the users. The proposed solution algorithm has the potential to dramatically reduce the time required to perform a CFD analysis. The use of hybrid grids will simplify the grid generation process for practical (complex) geometries and the use of MPP computers will reduce the run time. The resulting analysis package will be marketable to the aerospace industry and other industries. ***