Research · written up April 2026
Ahmed body: y⁺ pass on the baseline grid
An Ahmed body is a smoothed block used as a standard car-like shape in wind tunnels and in CFD. y⁺ is how far the first grid point sits from the surface, measured in the flow's own units; a wall function assumes that point lands inside a particular band.
The question
Wall-resolved or wall-modelled: how much of the drag disagreement is the wall treatment and how much is the geometry?
yplus_body_timeseries.dat in aeroAUTO
@ 3a3a889. Log axis: the series spans three decades. Nothing is smoothed and
no sample is dropped.
The case runs a steady RANS configuration with k-ω SST and a wall-function wall treatment, which is only valid where the first cell centre lands in the log layer. The study's own README puts that target at y⁺ between 30 and 300, and the body at 80 to 120.
It does not land there. Over the second half of the run the surface average settles at 79, comfortably inside the band, while the maximum settles near 1040, three times above it. The minimum is below the band on every one of the 347 samples, and the maximum is above it on 346 of them, 99.7 per cent. So the mesh satisfies the wall function on average and violates it everywhere at once: part of the body is being resolved and part of it is being modelled, and which part is decided by the local flow rather than by anything chosen.
The drag gap, and why it stays unattributed
Drag over the settled window is Cd = 0.2253 with a standard deviation of 0.0008 across the window's samples, against a published value of 0.285 for the 25-degree body. The gap is about 21 per cent, and until the wall treatment is uniform across the surface there is no way to say how much of that is the wall model and how much is everything else.
What it does not settle
The meshed geometry on this case is not the standard Ahmed body: the slant came out at 21.67 degrees rather than 25, so the published value above is a marker on the axis, not a comparison. See aeroAUTO.
A y⁺ series is a property of the mesh and the flow together. It says the wall treatment is inconsistent; it does not say what the drag would be if it were consistent, and nothing here is a grid-independence result.
- Where it stands
- Over the second half of the run the surface average settles near 79, inside the band a wall function assumes, and the maximum near 1040, three times above it.
- Next
- Re-run with y⁺ < 1 on the body; compare drag convergence and the y⁺ map against the baseline case.
- Tools
- OpenFOAM · Python