sahni-lab/multi-res-wedge-duct-cfd-database

Wedge Duct Multi-Resolution Adaptive CFD Dataset

Supersonic/hypersonic inviscid flow over a wedge in a duct, solved with Ansys Fluent on a sequence of anisotropically adapted meshes. Each sample is one set of inflow conditions drawn from a 1750-point sampling plan. For every sample, the dataset provides the volume solution, the derived fields that drive the adaptation (gradients, Hessians, shock flags, size field) and the boundary-surface solution, for each of four adaptation cycles.

Contents

Samples	228 (metadata/samples.parquet)
Cycles per sample	4 (cycle0 … cycle3)
Mesh	3D slab, one layer of wedge (prism) cells through the thickness (z)

Layout

metadata/
  samples.parquet        one row per sample: inflow conditions
  cycles.parquet         one row per (sample, cycle): mesh size and file paths
  fields.json            field names, original solver names, units, descriptions
  reflag_summary.parquet shock-flag recomputation statistics
  lambda_values.parquet  Hessian eigenvalue percentile statistics
geometry/                Parasolid model, Simmetrix model and starter mesh
data/NNNN/cycleK/
  volume.vtu             flow solution on the cycle-K mesh
  supp.vtu               adaptation fields on the same mesh
  surface.vtm            boundary surfaces (multiblock), blocks in surface/<name>.vtu

All .vtu files are VTK XML unstructured grids (binary, zlib), readable with ParaView, VTK, pyvista or meshio. Fields are point data.

Samples (metadata/samples.parquet)

column	units	description
sample_id		four-digit ID, matches data/NNNN/
original_id		ID in the original 1750-point sampling plan
T0	K	total temperature
P0	Pa	total pressure
Ts	K	static temperature
Ps	Pa	static pressure
M		freestream Mach number
altitude_km	km	altitude used for the static conditions

Adaptation cycles

cycleK holds the solution on the mesh after K adaptations; cycle0 is the starter mesh. Minimum mesh size in adaptation cycle N is h_min = 0.00651125 / (4 · 2^(N-1)), maximum size h_max = 0.05209448, and maximum requested aspect ratio 32. The adaptation is driven by the static temperature Hessian, modified at shocks flagged from the temperature and pressure fields.

Fields

volume.vtu and surface/<block>.vtu: Static_Pressure, Mach_Number, Velocity_Magnitude, Sound_Speed, Static_Temperature, Density, Dynamic_Pressure, Enthalpy, Specific_Heat_Ratio__gamma_, Gas_Constant__R_, Velocity (3).

supp.vtu: temperature_gradient (3), temperature_hessian (9), temperature_flag_direction, temperature_seeds, temperature_flag, pressure_gradient (3), pressure_hessian (9), pressure_flag_direction, pressure_seeds, pressure_flag, size_field (6).

size_field is the size field prescribed for the next adaptation, stored as two orthogonal 3-component vectors (components 0–2 and 3–5). Each vector gives a direction, and its magnitude is the requested mesh size in that direction. The cases are 2D, so the third (through-thickness) direction does not apply and is omitted.

Surface blocks: top, outlet, bottom, wedge, bottom-2, inlet, front, back.

Loading

import meshio, pandas as pd
from huggingface_hub import snapshot_download

root = snapshot_download("sahni-lab/multi-res-wedge-duct-cfd-database", repo_type="dataset",
                         allow_patterns=["metadata/*", "data/0000/*"])
samples = pd.read_parquet(f"{root}/metadata/samples.parquet")
vol = meshio.read(f"{root}/data/0000/cycle0/volume.vtu")
print(vol.point_data["Mach_Number"].shape)

Citation

A paper describing this dataset is forthcoming at AIAA SciTech 2027. The citation will be added here once it is published.