Skip to content

drivaer

English | Japanese

« Home · « External flow and wind engineering

Summary: the wake of the DrivAer fastback (a realistic car shape) on the D3Q27 cumulant block-forest AMR-LBM. The inflow is a section cut from a steady OpenFOAM solution (simpleFoam, k-omega SST). The velocity field at y = 0 and the wake fluctuations are compared with the FVM. This is the case of the OpenRheo part 2 talk at the OpenCAE Symposium 2026 (section 7.4).

Setup

The LBM domain (8.3 x 8.2 x 5.1 m) holds the rear of the car and sits inside the FVM domain (half model, 40 x 5 x 8 m). The LBM inflow is the FVM section at x = 1.6 m, 2.2 m ahead of the rear: its mean velocity plus digital-filter fluctuations built from k.

DrivAer setup: (a) FVM domain and boundary conditions, (b) FVM mesh and prism layers at y = 0, (c) LBM domain, refinement levels and boundary conditions, and the mean inflow vectors at y = 0 (every 0.2 m)

  • (a) FVM and LBM domains.
  • (b) FVM mesh at y = 0 (3.43M cells, snappyHexMesh).
  • (c) LBM domain and refinement boxes (level 0/1/2 = 4.4/2.2/1.1 cm), with the mean inflow velocity vectors at y = 0 (every 0.2 m). The figure is the one of the OpenCAE Symposium 2026 talk.

Results

drivaer y=0

  • (a) FVM mean u.
  • (b) LBM instantaneous u at t = 1.65 s.
  • (c) LBM u against time at the four points of (b).

drivaer wake y=0 animation

|u| of the LBM on the plane y = 0 (0 to 22 m/s, x from the rear end). Same set-up as the table below (wall function evaluated with the velocity relative to the moving floor and wheels), rerun to write the plane every 100 steps (0.014 s); the last 2.1 s (150 frames) play in 10 s. The shear layer from the roof becomes turbulent as it moves downstream, and vortices are exchanged in the recirculation behind the rear end.

FVM (simpleFoam, k-omega SST) LBM (D3Q27 cumulant)
reverse-flow length behind the rear at y = 0 0.72 m 0.93 m (1.02 m with a 0.55 cm underbody; 1.08 m with 0.55 cm also around the rear end)
underbody u at the rear 12.8 m/s 12.9 m/s (0.55 cm underbody; 12.0 m/s with 0.55 cm also around the rear end)
Cd 0.2571 (mean of iterations 501-1000) not evaluated

The LBM numbers are from the rerun after the wall function was changed to use the velocity relative to the moving floor and wheels (2026-10-29/30, Miyabi-G GH200, 56,000 steps, averaged from step 28,000). The change against the earlier runs is within 2.5 % in Cd and 0.06 m in the reverse-flow length. The run with 0.55 cm cells also around the rear end (x -0.29 to 0.82 m, |y| < 0.84 m, z < 1.02 m) was added on 2026-10-30.

Wake probes u (mean / std, m/s):

point mean std
1 -2.27 1.67
2 11.47 1.16
3 12.23 1.88
4 2.93 3.46

The spectral peaks lie at 3.75-5.0 Hz (St = 0.33-0.44 with the car height 1.418 m and 16 m/s; frequency resolution 0.63 Hz). The peak moves between the first and second 0.8 s of the same run, so the frequency is not settled at this averaging length.

Numbers of the OpenCAE Symposium 2026 talk

From the OpenCAE Symposium 2026 talk; the whole talk is summarised in OpenCAE Symposium 2026 talk.

  • FVM: simpleFoam, k-omega SST, 3.43 M cells (snappyHexMesh, 6 prism layers on body, wheels and ground), rotating wheels; Cd 0.257 at 16 m/s (experiment [Collin] 0.252).
  • LBM inflow: the FVM velocity plus digital-filter fluctuations from k.
  • LBM: D3Q27 cumulant, finest 1.1 cm (19.82 M cells), about 45 minutes on one Miyabi-G GH200; Spalding wall function, moving floor, rotating rear wheels.
FVM LBM LBM, underbody cells halved
Rear recirculation length 0.72 m 0.92 m 1.08 m
Speed 0.1 m above the floor at the rear 12.8 m/s 8.8 m/s 12.6 m/s

Wake standard deviation 1.2–3.4 m/s; spectral peaks of the shear layer and wake 3.75–5.0 Hz (St = 0.33–0.44 on the car height). The figures and numbers of this table (including the 0.55 cm underbody column) are the abstract's runs; the rerun with the wall function relative to the moving floor and wheels gives 0.93 m and 9.4 m/s (Cd within 2 %). The current values are in the table under Results.

Settings

Item Value
Geometry DrivAer fastback with wheels
Speed 16 m/s, floor moving at 16 m/s, rear wheels rotating
Lattice D3Q27, cumulant collision, wall model (Spalding)
Refinement 3 levels, finest cell 1.1 cm, 19,818,112 cells
Steps 16,000, statistics over steps 8,000-16,000. The instantaneous field and the probe records come from a 56,000-step run
Machine Miyabi-G (GH200)
FVM the DrivAer fastback case of the OpenFOAM training course: 3.43M cells, 64 Grace ranks, 335 s

Geometry

The DrivAer geometry is under the licence of TUM (Technical University of Munich), so this repository and wiki contain no geometry files (STL, meshes); users obtain the geometry themselves. The training-course case fetches the surfaces distributed with the OpenFOAM-13 tutorial (incompressibleFluid/drivaerFastback) with run/getGeometry.

FVM case of the training course

The FVM solution is the DrivAer fastback case of the OpenFOAM training course (OpenCAE Society).

Content Link
Case files Flow2/drivaerFastback
Course document (Markdown, Japanese) Supercomputer-OpenFOAM-Training-drivaerFastback-Flow2.md
Common settings and mesh run/base
CPU run (simpleFoam) run/CPU
GPU runs (AmgX, openfoam-ecse) run/AmgX, run/eCSE
Geometry download run/getGeometry

Data

File Content
README.md Data provenance, run ids, the source of the numbers on this page
domain.json FVM domain, LBM domain and refinement boxes, inlet section
probe_stats.py Mean, std and spectral peak at the four wake points (its input probes_y0.csv is too large for the wiki)

Validation

Tier T3 — demo (no quantitative gate)
QoI reverse-flow length at y = 0, wake u
Reference OpenFOAM simpleFoam k-omega SST of the training course (no published reference)
Gate — no gate

Source and runs

The case ran on a development branch (drivaer) that is not yet in the public main. The data provenance and the numbers come from the README.md of the symposium paper data: run ids run_nh32_ml2.3548845, run_nh32_ml2.3553357, and 3553360 -> 3556610.