Search Results

A simulation framework for vehicle aerodynamics using up to 10 billion fully unstructured cells has been developed on a world-fastest class supercomputer, called the K computer, in Kobe, Japan. The simulation software FrontFlow/red-Aero was fully optimized on the K computer to utilize up to 10,000 processors with tens of thousands of cores. A hybrid parallelization method using MPI among processors and OpenMP among cores inside each processor was adopted. The code was specially tuned for unsteady aerodynamic simulation including large-eddy simulation, and low Mach number approximation was adopted to avoid excessive iterations usually required for the fully incompressible algorithm. The automated mesh refining system was developed to generate unstructured meshes of up to 10 billion cells. In the system, users only generate unstructured meshes in the order of tens of millions of cells directly using commercial preprocessing software.

In open-grille vehicle aerodynamics simulation using computational fluid dynamics, in addition to basic flow characteristics, such as turbulent flow with a Reynolds number of several million on the bluff body, it is important to accurately estimate the cooling air flow introduced from the front opening. It is therefore necessary to reproduce the detailed geometry of the entire vehicle including the engine bay as precisely as possible. However, there is a problem of generating a good-quality calculation grid with a small workload. It usually takes several days to a week for the pretreatment process to make the geometry data ‘clean’ or ‘watertight’. The authors proposed a computational method for complex geometries with a hierarchical Cartesian grid and a topology-independent immersed boundary method with dummy cells that discretize the geometry on a cell-by-cell basis and can set an imaginary point arbitrarily.