ayz2/drivaernet_processed

DrivAerNet++ (processed)

This is a processed, downsampled derivative of DrivAerNet++, 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 DrivAerNet++ (upstream: https://dataverse.harvard.edu). Total size is 0.76 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/{test_old,train,train_old,val,val_old}.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)

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	8,128	1/1	3.658e+09	3.658e+09
volume_collated	8,128	1/5	2.356e+10	1.178e+11

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).

• test_old: 1,385 samples
• train: 6,502 samples
• train_old: 5,394 samples
• val: 1,626 samples
• val_old: 1,349 samples

Source dataset and preprocessing

DrivAerNet++

What it is. 8000 parametrically morphed DrivAer passenger-car bodies (fastback, estateback and notchback, with open/closed and smooth/detailed wheels, and detailed ICE or smooth EV underbodies) under 26 geometric design parameters. It models external automotive aerodynamics — drag, base-wake structure and underbody flow — as a shape-to-field regression problem.

Generation. Steady incompressible RANS with Menter's k-ω SST closure, simpleFoam (OpenFOAM v11), SIMPLE coupling, wall-modelled via nutUSpaldingWallFunction. ~24 M cells per case (500–750 k on the car surface). 7000 iterations per case, with forces averaged over the last 1000. This is the only two-equation closure in the ten.

Parameters. U∞ = 30 m/s, Re 8.37e6–1.01e7 (varying with car length), air at ν = 1.56e-5 m²/s, ρ = 1.184 kg/m³. Zero incidence and zero yaw.

Cost. 256 cores per case on a 60-node / 2880-core cluster; 3.0e6 CPU-hours in total (≈ 375 CPU-h per case), 39 TB released across 834 332 files.

Processing. The raw release stores pressure and wall-shear-stress in separate legacy VTK files that contain the same point set in a different order; pressure is realigned to the shear point order by exact coordinate matching before stacking. Surface output is [x, y, z, p, τx, τy, τz] (7 cols, kinematic p in m²/s², converted to Cp/Cf by the loader with q = ½·30²). Native axes are x streamwise (wake at +x), y lateral (symmetric about 0), z vertical with the floor at z ≈ 0; no reorientation is needed, but CF_SIGN = −1 applies. The volume release is inconsistent about domain extent — some runs are half-domain about y = 0, others full — so half-domain runs are mirrored across y = 0 with u_y negated, giving a full-body cloud for every sample (recorded per sample in the manifest). Volume crop margins are [0.5, 0.75, 0.0] L below and [2.0, 0.75, 0.5] L above the paired surface bbox (stream/width/vertical), i.e. the ground plane is never extended downward. Volume output is [x, y, z, ux, uy, uz, p]. Splits: random 80/20 (6502/1626).

Citation

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