UAV-AID-100

UAV-AID-100 is the full-CFD public release of the UAV Aerodynamic Information Dataset (UAV-AID).

It contains 100 three-dimensional fixed-wing UAV lifting-surface configurations, selected to provide broad coverage of the UAV-AID geometric and operating-condition design space.

For each wing, the dataset provides the geometry, computational mesh, CFD setup and solution files, solver histories, aerodynamic metadata, and surface distributions of pressure and wall shear stress over the available angles of attack.

The dataset is intended for research in:

• aerodynamic surrogate modelling,
• geometric deep learning (graph neural networks / neural fields),
• surface-field prediction,
• CFD reduced-order modelling,
• aerodynamic design-space exploration,
• aerodynamic coefficient prediction,
• CFD post-processing and benchmarking.

Dataset Overview

UAV-AID was generated using an automated three-dimensional CFD workflow for fixed-wing UAV lifting surfaces. ("An Automated Framework for Streamlined CFD-Based Design and Optimization of Fixed-Wing UAV Wings")

The complete UAV-AID design-space registry contains 10,000 wing configurations spanning conventional, flying-wing and blended-wing-body (BWB) geometries.

CFD production was initiated for 5,572 configurations, of which 5,448 produced at least one aerodynamic solution, resulting in 64,422 wing–angle-of-attack CFD records.

Two nested public releases are provided:

Release	Contents
UAV-AID-1000	1,000-wing aerodynamic-polar dataset
UAV-AID-100	100-wing full-CFD dataset, nested within UAV-AID-1000

This repository contains UAV-AID-100.

The 100 wings were selected deterministically from the 1,000-wing public subset using a space-filling farthest-point (maximin) procedure. The selection was performed using the planform parameters, B-GAN airfoil latent variables, velocity, and Reynolds number without using aerodynamic responses as selection variables.

Dataset Structure

The repository is organized by wing:

UAV-AID-100/
│
├── wing_1/
│   ├── wing_1.xml
│   ├── wing_1.pkl
│   ├── wing_1.vsp3
│   ├── wing_1.step
│   ├── wing_1.msh.h5
│   │
│   ├── wing_1_airfoil_1.dat
│   ├── wing_1_airfoil_2.dat
│   ├── wing_1_airfoil_3.dat
│   │
│   ├── <AoA-specific Fluent case files>
│   ├── <AoA-specific Fluent solution files>
│   ├── <AoA-specific force/moment histories>
│   └── <AoA-specific surface-field files>
│
├── wing_2/
│   └── ...
│
└── wing_...

The exact files available for a wing depend on the simulated angle-of-attack range and retained CFD solutions.

Data Products

Geometry

Each wing is represented parametrically using two trapezoidal planform sections defined by three spanwise chord stations.

The planform is described by 11 parameters:

Variable	Description	Units
$b_1$	Root-to-kink semi-span distance	m
$b_2$	Semi-span	m
$c_1$	Centerline chord	m
$c_2$	Kink chord	m
$c_3$	Tip chord	m
$Λ_{1_{c/4}}$	Inboard quarter-chord sweep	deg
$Λ_{2_{c/4}}$	Outboard quarter-chord sweep	deg
$i_{1_{c/4}}$	Kink twist	deg
$i_{2_{c/4}}$	Tip twist	deg
$γ_1$	Inboard dihedral	deg
$γ_2$	Outboard dihedral	deg

The resulting parameterization includes conventional wings, flying-wing configurations, and BWB-type lifting surfaces.

Airfoil Representation

Three airfoils define each wing:

• centerline,
• kink,
• tip.

Each airfoil is represented by an eight-dimensional latent vector from a Bézier Generative Adversarial Network (B-GAN).

Therefore, each wing contains 24 airfoil latent variables in addition to the 11 planform variables, resulting in a 35-dimensional geometric sampling space.

The generated airfoils are also provided explicitly as coordinate files in .dat format, so use of the B-GAN model is not required to reconstruct the released geometries.

Each airfoil coordinate file contains 192 normalized coordinate points.

Design Space

The principal geometric constraints of UAV-AID are:

Quantity	Range / Constraint
$b_2$	$1–10 m$
$b_1$	$0.2 b_2$ – $0.7 b_2$
$c_1$	$0.2–10 m$
$c_2$	$0.3 c_1 – c_1$
$c_3$	$0.3 c_2 – c_2$
$Λ_{1_{c/4}}$	$0–60°$
$Λ_{2_{c/4}}$	$0–60°$
$i_{1_{c/4}}$	$−5–5°$
$i_{2_{c/4}}$	$−7–7°$
$γ_1$	$−10–10°$
$γ_2$	$−10–10°$
Aspect ratio	$3–20$

The aspect-ratio population was constrained to approximately reproduce the distribution observed in published fixed-wing UAV data.

UAV Scale and Operating Conditions

The design space covers multiple UAV scales.

UAV-AID class	Full wingspan	Assigned velocity
Class I	$2 ≤ b < 8 m$	$6–42 m/s$
Class II	$8 ≤ b < 18 m$	$20–42 m/s$
Class III	$18 ≤ b ≤ 20 m$	$27–103 m/s$

Each wing is assigned a single operating velocity, which remains fixed throughout its angle-of-attack sweep.

All production conditions are defined at sea level.

CFD Methodology

The aerodynamic solutions were generated using an automated CFD pipeline.

Setting	Value
Flow model	3D steady RANS
Solver	ANSYS Fluent
Solver implementation	Native GPU solver
Turbulence model	k-ω SST
Near-wall design target	mean y+ ≤ 1
Reference length	Mean Aerodynamic Chord
Reference area	Half-wing planform area
Maximum iterations	2200 / 2500 depending on AoA

The nominal angle-of-attack sequence is:

-4°, -2°, 2°, 4°, 6°, 8°, 10°, 12°, 14°, 16°, 18°, 19°, 20°, 21°

Individual polars may terminate earlier because of detected stall, convergence behaviour, or workflow limitations.

Aerodynamic Quantities

The CFD outputs include integrated aerodynamic quantities such as:

Name	Symbol
Lift Coefficient	$C_L$
Drag Coefficient	$C_D$
Pitching Moment Coefficient at y=0	$C_{M,0}$
Pitching Moment Coefficient at y=c/4	$C_{M,c/4}$
Lift	$L$
Drag	$D$
Force along x-axis	$F_x$
Force along z-axis	$F_z$
Pitching Moment at y=0	$M_0$
Pitching Moment at y=c/4	$M_{c/4}$

Numerical metadata include quantities such as:

convergence status
iteration count
solution time
mean y+
maximum y+
operating conditions
reference quantities
stall information

Surface Field Data

For the available CFD operating points, surface-resolved data are provided on the wing surface.

Typical exported fields include:

Node_ID
x
y
z
pressure
x-wall-shear-stress
y-wall-shear-stress
z-wall-shear-stress

File Formats

Depending on the data product, the repository contains:

File	Description
.dat	Airfoil coordinates
.pkl	Parametric wing definition
.vsp3	OpenVSP geometry
.step	CAD representation
.msh.h5	ANSYS Fluent computational mesh
.ansa.gz	BETA CAE ANSA CFD pre-processing file
.cas.h5	Fluent case/setup
.dat.h5	Fluent CFD solution
.out	Force and moment iteration histories
.csv / .parquet	Exported surface or compiled data
.xml	Wing and CFD summary metadata

Some formats, particularly .cas.h5, .dat.h5, and .msh.h5, require ANSYS Fluent or compatible post-processing software for direct inspection. .ansa.gz files require active ANSA license.

The exported tabular surface-field data can be used independently of ANSYS Fluent.

Recommended Uses

UAV-AID-100 is primarily intended for research requiring access to distributed or solver-level CFD information.

Examples include:

Surface-field surrogate modelling

Wing surface mesh + operating condition
      ↓
Graph / Point Cloud representation
      ↓
GNN / Neural Field
      ↓
Local aerodynamic quantities (pressure / wall-shear distribution)

Integrated aerodynamic prediction

Parametric Wing geometry + Operating Condition
    ↓
ML surrogate
    ↓
CL, CD, CM

Relationship to UAV-AID-1000

UAV-AID-100 is a nested subset of UAV-AID-1000.

Every wing included in this repository is therefore also represented in the coefficient-level public release.

Users interested only in aerodynamic coefficients and geometry parameters should generally use UAV-AID-1000, since it provides substantially greater geometric coverage with a much smaller storage requirement.

UAV-AID-100 should be used when CFD meshes, complete numerical solutions, solver histories, or surface fields are required.

Data Quality and Curation

The parent UAV-AID database underwent a hierarchical quality audit covering:

production status
polar completeness
CFD convergence
near-wall resolution
coefficient plausibility
statistical coefficient outliers

The public datasets are selected from the curated aerodynamic population.

Users can nevertheless apply additional numerical-quality criteria depending on their application.

For example, the supplied mean and maximum y+ values can be used to define stricter near-wall-resolution requirements.

Important Limitations

UAV-AID represents isolated lifting surfaces, not complete aircraft.

Unless explicitly represented by a particular geometry, the CFD data do not include:

fuselage effects
empennage effects
propulsion-system effects
propeller slipstream
engine installation effects
complete-aircraft interference

All simulations use a fully turbulent steady RANS k-ω SST model.

The data should therefore not be interpreted as reference solutions for laminar-transition-dominated flows or intrinsically unsteady post-stall aerodynamics.

This is particularly relevant for the lowest-Reynolds-number Class I configurations.

High-angle-of-attack data should also be interpreted with care because CFD convergence becomes more difficult as separation and stall are approached.

Download

The complete repository can be downloaded with:

from huggingface_hub import snapshot_download

snapshot_download(
    repo_id="george-erfem/UAV-AID-100",
    repo_type="dataset",
    local_dir="UAV-AID-100",
)

Individual files or wing directories can also be retrieved directly through the Hugging Face Hub.

Related Dataset

The coefficient-level companion dataset is:

UAV-AID-1000 — 1,000-Wing Aerodynamic Polar Dataset

UAV-AID GitHub Repository

Paper

The complete methodology, design-space definition, dataset audit, aerodynamic characterization, and machine-learning benchmarks are described in:

G. Efrem, C. Pliakos, D. Terzis, and P. Panagiotou,
"UAV-AID: UAV Aerodynamic Information Dataset",
[Journal / conference when available], [year].

Add DOI / publication URL after publication.

If you use UAV-AID in academic work, please cite the dataset paper.

@article{efrem_uavaid,
  title   = {UAV-AID: UAV Aerodynamic Information Dataset},
  author  = {Efrem, Giorgos and Pliakos, Chris and Terzis, Demetris and Panagiotou, Pericles},
  journal = {...},
  year    = {...},
  doi     = {...}
}

License

This dataset is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0).

Software distributed alongside the dataset may use a separate software license.

Acknowledgments

This research was supported by the Hellenic Foundation for Research and Innovation (HFRI) under the Basic Research Financing (Horizontal support for all Sciences), National Recovery and Resilience Plan, Project Number 016429, Project Acronym INDIANA.

Computational results were produced using the Aristotle University of Thessaloniki High Performance Computing Infrastructure and Resources.

The authors also acknowledge the National Technical University of Athens for supporting the wind-tunnel experimental validation campaign.

Contact

For questions regarding UAV-AID:

UAV-iRC / Aristotle University of Thessaloniki

egiorgosm@meng.auth.gr