Wind tunnel

The results above are from the key Slipstream solve (10 October, stock model, 20 mm cells). The flow lines, the pressure skin and the notes under “Reading the picture” are from the earlier rev B solve at 30 mm, which also modelled open ducts and turning tyres; Slipstream does not yet, so those two sections below describe rev B.

The solver

This is a lattice-Boltzmann solve, the method most of the road-car industry now uses for exterior aero. Rather than solving the Navier–Stokes equations directly, it tracks packets of air moving between the cells of a fixed grid and colliding — and the correct fluid behaviour falls out of the bookkeeping. It handles messy geometry well, which matters here, because the geometry is the car.

The model from the garage page was voxelised: every triangle rasterised onto a grid (20 mm in the Slipstream solves, 30 mm in rev B), the shell closed, and everything the outside air could not reach marked as solid. Air enters at the nose end of the domain at a fixed speed, leaves freely at the tail end, and the floor runs backwards at the same speed as the air — a rolling road, as a real tunnel would use.

Air through the car, not just round it (rev B)

The donor model has its grilles painted on rather than cut out, so the openings were authored to the real car's layout and carved out of the voxel body: the two kidneys, the lower intake and a corner duct each side. On the page the bodywork itself is left whole — cutting the same boxes out of the display mesh tore the grille surround — so the lines that vanish into the nose are going in where the real openings are. They feed a shrouded plenum, so the air cannot simply slip round the edge of the core, then pass through a radiator modelled as a porous slab — partial bounce-back, with the resistance set from a duct calibration to behave like a real core and condenser stack. Behind that is a firewall, so nothing vents into the cabin, and the only way out is a 30 mm slot across the trailing edge of the bonnet. That is a vent slot, deliberately, not the factory panel gap; a 5 mm shutline is far below anything this grid could resolve.

Tyres that turn (rev B)

The wheels do not rotate as geometry — they do not need to. In lattice Boltzmann the bounce-back off a solid cell can carry a wall velocity, so every cell on each tyre is given its own surface velocity from ω × r about the axle, at the speed that rolls without slip on the moving floor. The air leaves the surface with the right tangential momentum, which is how production solves treat a tyre. Checked first on a spinning cylinder: the air one cell off the surface comes out at 0.85 of the surface speed.

Where tyre meets road is the awkward part. A circular section touching a moving floor is a singular corner and, left alone, squirts a spurious jet sideways. Each tyre is therefore cut to a flat 150 mm contact patch with vertical sides, which is both the standard fix and closer to what a loaded tyre actually does.

Turbulence

Nothing resolves every eddy on a car at road speed; that would need more grid than exists in this room. The small scales are handled by a Smagorinsky sub-grid model, which adds extra viscosity exactly where the strain rate is high. The large structures — the separation off the tail, the pair of trailing vortices, the wake's length — are resolved, and those are what set drag.

Where it stops being true

The grid is still coarse. In the Slipstream solves one cell is 20 mm and the car is 206 cells long; in rev B it was 30 mm and 138, with the gap under the floor three to four cells and each wheel 22 across. That is enough for the shapes of things and not enough for a boundary layer, so surface friction drag is understated while the staircase surface catches pressure a smooth panel would not.

The radiator's resistance is tuned, not measured — no one has put a pressure drop across this car's core. The engine bay is empty behind the core, where a real one is full of engine, and the only outlet is the bonnet vent, where a real car dumps most of its cooling air under the floor.

The donor model carries a splitter, side skirts and aftermarket wheels the real car does not, and the tunnel is narrow enough (10% blockage in the Slipstream solves, 19% in rev B) that the walls are closer to the car than any real facility would allow.

The Reynolds number of the solve is far below the road value. For a bluff body whose separation points are set by sharp edges — the trailing edge of the hatch, the arch lips — the wake's shape is largely insensitive to that, which is why the picture is worth looking at. The absolute numbers are not.