Pointwise   The Connector newsletter Pointwise Facebook Pointwise GitHub Pointwise LinkedIn Another Fine Mesh blog Pointwise Twitter Pointwise YouTube Y+ Calculator



Not for the Faint of Heart: Swift and Pointwise Team Up for CFD on One Billion Cells

Swift Engineering, Inc. of San Clemente, California, specializes in the design, engineering and manufacture of composite structure vehicles. The company was founded in 1983 to design and manufacture race cars for motorsports and they have had many successes there, including designs for Indy Cars, Formula Atlantic, and Formula Nippon racers. In 2000, Swift diversified into the aerospace market with a design for an unmanned air vehicle, and aerospace has since grown into the majority of their business. The common threads in their work have been rapid turnaround and innovative designs that exceed their customers' expectations.

Another thread running through Swift's history is an emphasis on fluid dynamics. In 1993, they developed North America's most advanced rolling-road wind tunnel for measuring drag, downforce, side force and wheel torque on road vehicles. They have also been innovative in applying computational fluid dynamics (CFD) to their designs and were recognized at the Supercomputing 2011 conference for their accomplishments, receiving the International Data Corporation's HPC Innovation Excellence Award for their continued development of next-generation virtual simulation engineering and analysis for aerospace and motorsports vehicles.

In 2011, Swift engineers performed a study of the requirements for highly-detailed CFD analysis on very large (one-billion-cell) meshes. They used the Pointwise preprocessing software from Pointwise, Inc. for the computational meshes and CFD++ from Metacomp Technologies, Inc. for flow solutions. On the hardware side, they used a Cray CX1 supercomputer including a visualization and preprocessing blade with 192GB of Netlist HyperCloud memory.

For the initial study, Swift was assisted by Pointwise, Inc. engineers, who developed an automated grid refinement technique using Pointwise's Glyph scripting language. The script automatically generates triangular surface and hybrid volume grids of a desired refinement level, so that grids of increasing sizes could be generated automatically. For the volume grid, Pointwise's T-Rex, anisotropic tetrahedral extrusion technique was used to generate high-aspect ratio tetrahedra in the boundary layer that were then combined into prism cells to reduce cell count and produce a more flow-aligned grid, as shown in Figure 1.

cut through a prism-tet mesh

Figure 1 - Cut through volume grid showing prisms in boundary layer region with surrounding tetrahedral cells. +

In addition, a block of hexahedral cells was placed in the wake behind the vehicle, as shown in Figure 2, to resolve the wake without unnecessarily increasing grid cell count. The script was initially applied to a simplified sedan geometry to make sure it was working properly and to obtain initial timing and memory usage. Running on single processor of the CX1 the CPU time and memory required for mesh generation were found to scale linearly with mesh size, as shown in Figure 3. Even the largest mesh, with more than 1 billion cells, could be generated in a reasonable time on a single CPU workstation. Generation of the mesh containing 1.197 billion cells used approximately 85 gigabytes of memory and just over 19 hours of CPU time. And since the CPU time and memory vary linearly with the number of cells it is easy to predict the resources needed for a given mesh size.

hex mesh used to resolve wake

Figure 2 - Surface grid on Indy Car and hexahedral cell wake block. +

After applying the script to generate meshes with up to 1.2 Billion cells on the simplified sedan, Swift engineers then applied it to a proposed Indy Car design, shown in Figures 1 and 2, with similar results. Their conclusion from these studies is that Pointwise is memory efficient and capable of generating large grids on a single workstation, and with the use of Netlist's HyperCloud memory can generate meshes with well more than a billion cells. The Glyph scripting language built into Pointwise can fully automate the meshing procedure and makes generating grids of varying sizes simple. Finally, they found the anisotropic tetrahedral cells with prism recombination generated by Pointwise T-Rex methodology to be very effective at resolving flow features, even on complex geometries. Further studies will be conducted to examine application of these techniques to other geometries.

memory and CPU usage for mesh generation

Figure 3 - Memory and CPU time to generate grids up to 1.2 billion cells. +

Better, Faster Meshes Start with a Free Trial

Get your heart racing with a free trial of Pointwise meshing. Whether you're using a billion cells or not quite so many, Pointwise can speed you through your mesh generation.

Free Trial