OpenFOAM for Car Aero: How to Find Models and Get a Simulation Running
Getting a car into OpenFOAM is mostly a geometry problem
The CFD solver is the easy part. OpenFOAM ships with everything you need to run steady-state external aerodynamics — simpleFoam for the solver, snappyHexMesh for meshing, ParaView for visualization. What slows most people down is finding a usable car geometry in the first place, and then getting it into a state where the mesher will actually accept it.
Where to find car models
GrabCAD is the most practical starting point. Search your specific model by chassis code rather than a generic name and you’ll usually turn up several community-uploaded bodies. Quality varies enormously — some are detailed enough to be genuinely useful, others barely more than a silhouette — but it’s free and the library is large.
FetchCFD is worth bookmarking alongside it. The site hosts complete CFD setups alongside downloadable geometries, so you can see how someone else approached the same or a similar shape before you commit hours to your own configuration. The automotive section has race car and road car models, mostly in formats that import cleanly into FreeCAD or Blender for prep work.
For validation before you trust results on your own geometry, the DrivAer model is the standard open-source reference for road car aerodynamics. It’s a modular design — three rear-end configurations, two underbody variants — built specifically so researchers can compare results against one another and against a published wind tunnel dataset. Run the DrivAer first and check your drag coefficient against the published data. If the numbers are badly off, something in your setup is wrong and you want to know that before you spend time on your actual car.
The Ahmed body as a sanity check
Even simpler than the DrivAer is the Ahmed body — essentially a box with a slanted rear window. It’s been a CFD benchmark since 1984, the flow physics are well-documented, and it runs fast. It’s not a car, but it’s a reliable first test. If your Ahmed body drag coefficient is wildly off the published values, something is wrong with your mesh or your boundary conditions, and that problem will follow you into every subsequent simulation.
Cleaning the geometry — the part nobody warns you about
This is where most people lose a week. Community STL files are rarely watertight. snappyHexMesh needs a closed, manifold surface: no gaps, no reversed normals, no overlapping faces. A model that looks fine in a CAD viewer can still have dozens of these problems hiding underneath the surface.
FreeCAD has basic repair tools built in. For more intensive cleanup, Blender’s 3D Print Toolbox add-on identifies non-manifold edges and lets you fix them in the same session. If you’re spending more than an hour on manual repairs, Meshmixer is faster for patching holes and rebuilding small sections. The goal is a single closed solid that doesn’t leave snappyHexMesh guessing about inside versus outside.
One specific thing to watch: wheel wells. Car models from community libraries often have wheels sitting inside the arches as separate bodies, creating surface intersections at every junction. Either remove the wheels entirely and simulate the body only, or properly merge them into one watertight solid before you try to mesh. Leaving them as floating separate bodies causes snappyHexMesh to produce a corrupted mesh that simpleFoam will reject at the first time step.
Setting up the OpenFOAM simulation
simpleFoam with the k-omega SST turbulence model is the standard starting point for external car aero. It’s a steady-state RANS solver — not the most physically accurate approach for separated flow, but fast enough to give you drag numbers and qualitative pressure distributions on a desktop machine in a reasonable time frame.
The domain size matters more than most tutorials emphasize. A common starting point is roughly 5 car lengths upstream, 10 downstream, and 3–4 car widths on each side and above. Too small and your boundary conditions start interfering with the flow field around the car. Too large and you’re wasting mesh cells on regions that don’t matter.
The louisgag repository on GitHub is a complete working external car flow example — snappyHexMeshDict and all boundary conditions already configured, built for a published research paper so the settings are sensible. Cloning it and swapping in your own geometry is a much faster route than building a case from scratch. Adjust the surface refinement regions around the areas you actually care about: spoilers, splitters, diffusers.
What to realistically expect from the results
RANS CFD on a desktop gives you a useful qualitative picture. Pressure distributions, broad flow separation zones, whether a wing is producing the attached flow you think it is — all of that is readable. Absolute drag numbers are roughly in the right ballpark but should not be treated as ground truth.
Fine detail around mirrors, vents, and complex geometry transitions is where RANS struggles. Transient effects are out of scope entirely. Accurate absolute predictions need higher-fidelity methods — LES or detached eddy simulation — on meshes that demand serious compute.
Where RANS genuinely earns its keep for enthusiast aero work is in relative comparisons: wing versus no wing, two spoiler profiles against each other, different splitter lengths. The absolute numbers may drift, but the delta between configurations tends to stay reliable. That’s the question worth asking the simulation.