ayz2/drivaerml_processed

DrivAerML (processed)

This is a processed, downsampled derivative of DrivAerML, not the original dataset. Fields were converted to a common frame and non-dimensionalization, rows were randomly subsampled and some columns were dropped (details below). For the original data, see the upstream source.

A preprocessed, downsampled release of DrivAerML. Total size is 0.75 TB. Every sample is a float32 point cloud [n_points, n_cols] stored as .npy, in a common frame and non-dimensionalization shared across the campaigns of this release.

Layout

collated/                 surface
  manifest.json           [{stem, n_points, n_points_source, ...}]
  splits/{train,val}.json
  norm_stats*.npz         per-column normalization statistics
  samples/<stem>.npy
volume_collated/          volume
  manifest.json
  norm_stats_volume*.npz
  samples/<stem>.npy
RELEASE.json              downsampling factors, columns and row totals

Columns

Surface (collated, 7 columns):

col	name	meaning
0	x	position
1	y	position
2	z	position
3	p	static pressure (raw solver units; Cp = (p - p_inf) / q)
4	tau_x	wall shear stress (raw)
5	tau_y	wall shear stress (raw)
6	tau_z	wall shear stress (raw)

Dropped from this release: Cp, pPrime2.

Volume (volume_collated, 7 columns):

col	name	meaning
0	x	position
1	y	position
2	z	position
3	ux	velocity
4	uy	velocity
5	uz	velocity
6	p	static pressure (raw solver units; Cp = (p - p_inf) / q)

Downsampling

tree	samples	row factor	rows (release)	rows (collated source)
collated	483	1/3	1.384e+09	4.151e+09
volume_collated	483	1/3	2.545e+10	7.635e+10

Rows within each file are in random order: they were permuted when the files were written. The release keeps the first n // factor rows of each collated sample, which is a uniform random subsample of the mesh points. For the same reason, any contiguous window arr[s:s+k] is a uniform random crop, so np.load(..., mmap_mode='r') plus a slice is the intended access pattern. The shipped norm_stats*.npz were computed on the full collated trees and remain valid for the subsample.

Splits

The volume tree uses the surface splits (keyed by stem).

• train: 386 samples
• val: 97 samples

Source dataset and preprocessing

DrivAerML

What it is. 500 morphs of the DrivAer notchback body, released as time-averaged hybrid-RANS/LES statistics — the highest-fidelity car campaign here.

Generation. SA-based σ-DDES: a Spalart-Allmaras RANS branch with the Nicoud σ-model as the SGS operator, plus Enhanced Protection shielding (after Deck & Renard's ZDES) to keep LES content out of attached boundary layers; Spalding all-y⁺ wall functions. Run on a custom pimpleFoam derivative (Upstream CFD, OpenFOAM v2212) with modified Rhie-Chow interpolation and second-order implicit Euler time integration. ~160 M cells, first boundary-layer height 0.75 mm, 7 layers to 12 mm total. Averaging continues until the drag estimate settles to ±1.5 drag counts, generally 40–60 convective time units (CTU based on the 2.786 m wheelbase).

Parameters. U∞ = 38.889 m/s (140 km/h), Re = 7.2e6 on the wheelbase (1.2e7 if length-based), air ν = 1.507e-5 m²/s, ρ = 1.2041 kg/m³, zero incidence.

Cost. ~40 h on 1536 cores (AWS hpc6a.48xlarge) per case ≈ 6.1e4 core-hours, so ≈ 3.1e7 core-hours for the campaign — the most expensive dataset in the set by a wide margin.

Processing. Surface is cell-based (boundary_<N>.vtp, cell data already aligned with centroids, so no cross-file matching): [x, y, z, p, τx, τy, τz, Cp, pPrime2] (9 cols, kinematic p, CF_SIGN = −1). 483 of the 500 runs are present (runs 8, 9, 80–99 and 167 are missing upstream). The volume is delivered as fixed-size 25 GiB split parts that must be concatenated byte-for-byte into one 46 GB polyhedral VTU (166 M points / 147 M cells); the builder reconstructs one file at a time on local NVMe and deletes it after reading, since holding all 483 at once would be ~8 TB. Fields are read from point data to avoid cell_centers() on 147 M polyhedra. Axes match DrivAerNet++, except that the floor is at z ≈ −0.318, not 0 — so the ground-plane guard used for the other car campaigns is deliberately disabled here. The raw domain is a full (not half) wind tunnel, x ∈ [−40, 80], y ∈ [−22, 22], z ∈ [−0.318, 19.68]; the crop keeps ~92% of points. Splits: random 80/20 (386/97).

Citation

TODO: this work, and the upstream dataset's citation.