Computational & Technology Resources
an online resource for computational,
engineering & technology publications
PROCEEDINGS OF THE SIXTH INTERNATIONAL CONFERENCE ON PARALLEL, DISTRIBUTED, GPU AND CLOUD COMPUTING FOR ENGINEERING
Edited by: P. Iványi and B.H.V. Topping
Aerodynamic characteristics study on high speed train based on an efficient parallel CFD solver
S. Ding1, D. Chen1 and G. Zheng2
1CRRC Qingdao Sifang Co., Ltd.
S. Ding, D. Chen, G. Zheng, "Aerodynamic characteristics study on high speed train based on an efficient parallel CFD solver", in P. Iványi, B.H.V. Topping, (Editors), "Proceedings of the Sixth International Conference on Parallel, Distributed, GPU and Cloud Computing for Engineering", Civil-Comp Press, Stirlingshire, UK, Paper 8, 2019. doi:10.4203/ccp.112.8
Keywords: CFD solver, parallel computing, high speed train, dyeing layered communication technology.
In the computation of aerodynamic problems, both precision and efficiency should meet higher requirements because of the large aerodynamic mesh, which often reaches the range of ten million. For this issue, a parallel computing solver of Euler and Navier-Stokes equations based on hybrid mesh is developed. Hybrid mesh is easier to be generated on complex shape when compared with structured meshes and also relatively suitable for the simulation of boundary layer flow in contrast with unstructured meshes, therefore being more fit for engineering applications. For parallel computing, the Pmetis code from the University of Minnesota is used for mesh division and dyeing layered communication technology based on MPI is used for information delivery, so as to improve communication efficiency and reduce communication congestion. In addition, implicit time-marching is optimized for better parallel computing. This above-mentioned solver is applied for steady and unsteady aerodynamic problems for standard EMU in the conditions of open line and cross wind. Specifically, nodecentered second-order discrete scheme and k-! turbulence model are used in the computation, while moving wall boundary condition is considered to simulate ground effect. Compared with experiments, the results of commercial software and CFD solver match well with the measured values. Furthermore, CFD solver behaves better on efficiency while the precision is not worse. The parallel implementation efficiency almost reach over 90% for CPUs from 2 to 256.
purchase the full-text of this paper (price £22)