Recent advances in machine learning (ML) have the potential to substantially accelerate aircraft wing design, a process that has traditionally required extensive high-fidelity simulation and design iteration. Existing public datasets for ML-based wing design have focused primarily on aerodynamic performance, with limited consideration of the multidisciplinary interactions that govern practical aircraft design. In particular, efficient wing design requires simultaneous consideration of aerodynamic performance and structural integrity, yet no publicly available datasets capture this coupled aerostructural behavior at high-fidelity. To address this need, a high-fidelity multidisciplinary dataset for static aerostructural analysis of transonic wings is introduced. The dataset includes 2471 samples, along with three out-of-distribution samples. The wing planform geometry is parameterized by span, sweep, dihedral, tip chord, and tip twist, while the airfoil cross-sections are parameterized using the class-shape transformation method. The analysis is further parameterized by Mach number, angle of attack, altitude, and structural load factor. The utility of the dataset is demonstrated by training state-of-the-art ML models to predict aerodynamic surface quantities, integrated aerodynamic performance quantities, and the failure distribution of the wingbox. ML models achieve at most 0.0309 normalized root mean squared error (NRMSE) in the prediction of aerodynamic quantities and at most 0.1189 NRMSE in the prediction of failure distribution of the wingbox. By providing coupled aerodynamic and structural responses for a diverse set of transonic wings, this dataset establishes a benchmark for developing, evaluating, and comparing multi-physics ML models for wing design. Broadly, it is intended to encourage creation of similar high-fidelity multi-physics benchmark datasets across engineering domains.
Figures & tables
Figure 1: Different views of the outer wing model and internal wingbox structure depicting the parameters used in this dataset.
Type
Parameter
Lower bound
Upper bound
Flow
Mach number, M∞
0.75
0.90
Angle of attack, α
2∘
5∘
Altitude, h
10,668 m (35,000 ft)
12,192 m (40,000 ft)
Wing planform
Span, b
13 m
18 m
Leading edge sweep, ΛLE
25∘
32∘
Dihedral, Γ
2∘
6∘
Table 2: Parameters used in this work, along with their respective lower and upper bound.
Figure 2: Samples from the dataset: (a) wing planform view, and (b) wing airfoil shapes along with upper and lower bound.
Figure 3: The out-of-distribution wing planform inspired from the Onera M6 wing: (a) planform view, and (b) cross-section airfoil.
Parameter
First sample
Second sample
Third sample
M∞
0.84
0.84
0.73
α
3∘
0∘
0∘
h
11,430 m (37,500 ft)
9,754 m (32,000 ft)
9,754 m (32,000 ft)
b
15.5 m
15.5 m
17.69 m
ΛLE
30∘
30∘
25.75∘
Γ
0∘
0∘
5.24∘
Table 3: Parameters for the out-of-distribution samples provided in this work. The parameter which is out-of-distribution is indicated in red.
Figure 4: An XDSM diagram illustrating the aerostructural framework used for evaluating samples in the design of experiment (DoE) and generating the dataset. The stacked diagonal blocks denote that the corresponding process is executed in parallel.
Figure 5: Thickness distribution in the baseline wingbox mesh: (a) skin, and (b) spar. The rib thickness is held constant and is equal to 3 mm.
Property
Values
Density
2500 kg/m 3
Young’s modulus
70 ×109 Pa
Poisson’s ratio
0.3
Yield stress
350 ×106 Pa
Table 4: Material properties for the wingbox model.
Figure 6: Results of static aerostructural analysis for selected samples in the dataset.
File name
Description
parameters.json
Parameters defining the geometry and flow conditions
Wing surface ( Cp , Cf , etc.) data stored as VTK file
surface_solution.hdf5
Wing surface ( Cp , Cf , etc.) data stored as point cloud
structural_solution.vtk
Displacement, stress, strain, and failure fields in VTK format
wing_model.stl
Wing geometry as triangulated surface mesh
Table 5: Details about various files for each sample in the dataset. Here, Cp refers to coefficient of pressure, Cf is skin friction coefficient, and M is Mach number. The CL , CD , and CM denotes lift, drag, and pitching moment coefficient.
Figure 7: Visualization of the prediction of Cp and failure distribution using ML models for a wing shape at a Mach number of 0.869, angle of attack of 4.82∘ , and altitude of 11462 m.
Figure 8: Prediction of Cp distributions at 25%, 50%, and 75% spanwise sections of the wing shape and operating conditions corresponding to Figure 7 .
Model
Cp
Cf
CD
CL
NRMSE
R2
NRMSE
R2
NRMSE
R2
NRMSE
R2
MHA-Net
0.0213
0.9984
0.0307
0.9967
0.0107
0.9975
0.0087
0.9975
UQAeroNet
0.0212
0.9984
0.0309
0.9968
0.0109
0.9974
0.0084
0.9977
Table 6: ML model evaluation for aerodynamics discipline of aeroelastic analysis.
Figure 9: Visualization of the predicted Cp and failure distribution using ML models for the first OOD sample in Table 3 .
Figure 10: Prediction of Cp distributions at 25%, 50%, and 75% spanwise sections for the first OOD sample in Table 3 .
Appendix figures & tables14 assets
Supplementary material from the paper’s appendix.
Appendix
Figure 11: Computational meshes used for analysis in this work: (a) aerodynamics (A2 level) and (b) structural (S2 level).
Mesh level
Number of cells
Number of constant layers
y+ max
A0
14,535,680
4
∼0.2
A1
1,816,960
2
∼0.4
A2
227,120
1
∼0.9
Appendix
Table 7: Different computational fluid dynamics mesh levels used in this work.
Mesh level
Number of nodes
Number of elements
S0
56658
57600
S1
13938
14400
S2
3378
3600
Appendix
Table 8: Different structural mesh levels used in this work.
Mesh level
CD (d.c.)
CL
Cm
Max. vertical displacement (m)
Max failure
A0-S0
241.85
0.507
-0.797
0.978
0.652
A1-S1
242.74
0.494
-0.774
0.942
0.557
A2-S2
244.50
0.498
-0.782
0.961
0.497
Appendix
Table 9: Grid convergence study for multidisciplinary analysis using different mesh levels.
Figure 12: Distribution of various operating conditions in the training dataset.
Figure 13: Distribution of various wing planform parameters in the training dataset.
Figure 14: Distribution of the load factor parameter in the training dataset.
Figure 15: Distribution of airfoil lower surface CST coefficients in the training dataset.
Figure 16: Distribution of airfoil upper surface CST coefficients in the training dataset.
Figure 17: Pairwise plot illustrating relationship between important scalar quantities of interest (QoI) in the dataset. The diagonal plots show the distribution of an individual QoI, while the off-diagonal plots depict the correlation between the quantities.
Figure 18: Visualization of the prediction of Cf and failure distribution using ML models for a wing shape at a Mach number of 0.831, angle of attack of 4.95∘ , and altitude of 11352 m.
Figure 19: Prediction of Cf distribution at 25%, 50%, and 75% spanwise sections of the wing shape and operating conditions in Figure 18 .
Figure 20: Visualization of the predicted Cf and failure distribution using ML models for the third OOD sample in Table 3 .
Figure 21: Prediction of Cf distributions at 25%, 50%, and 75% spanwise sections for the third OOD sample in Table 3 .
Accurate machine-learning models for aerodynamic prediction are essential for accelerating shape optimization, yet remain challenging to develop for complex three-dimensional configurations due to the high cost of generating training data. This work introduces a methodology for efficiently constructing accurate surrogate models for design purposes by first pre-training a large-scale model on diverse geometries and then fine-tuning it with a few more detailed task-specific samples. A Transformer-based architecture, AeroTransformer, is developed and tailored for large-scale training to learn aerodynamics. The methodology is evaluated on transonic wings, where the model is pre-trained on SuperWing, a dataset of nearly 30000 samples with broad geometric diversity, and subsequently fine-tuned to handle specific wing shapes perturbed from the Common Research Model. Results show that, with 450 task-specific samples, the proposed methodology achieves 0.36% error on surface-flow prediction, reducing 84.2% compared to training from scratch. The influence of model configurations and training strategies is also systematically studied to provide guidance on effectively training and deploying such models under limited data and computational budgets. To facilitate reuse, we release the datasets and the pre-trained models at https://github.com/tum-pbs/AeroTransformer. An interactive design tool is also built on the pre-trained model and is available online at https://webwing.pbs.cit.tum.de.
Yunjia Yang, Babak Gholami, Caglar Gurbuz +2
Technical University of Munich, Garching, D-85748, Germany · BMW Group, Bremer Str. 6, Munich, D-80788, Germany
This paper describes the first-ever open-source high-fidelity CFD dataset of a high-lift aircraft for the purpose of AI surrogate model development. The dataset is composed of 1800 samples, arising from 180 geometry variants and 10 angles of attack for the high-lift NASA Common Research Model (CRM) geometry, used within the AIAA High-Lift Prediction Workshop series. One of the novelties of this dataset is the use of a GPU-accelerated high-fidelity explicit, wall-modeled LES approach for each simulation, using solution-adapted grids between 300M and 500M cells. This ensures the greatest possible accuracy given known challenges in steady-state RANS approaches for these portions of the flight envelope. The entire dataset (geometries, time-averaged volume and surface variables and integral forces) are available, free of charge with a permissive open-source license (CC-BY-4.0). By making this data publicly available, we aim to accelerate the research and development of AI surrogate modeling within the aerospace industry.
Neil Ashton, Adam Clark, Liam Heidt +11
NVIDIA, Santa Clara, US · The Boeing Company, Seattle, US · Cadence Design Systems, Santa Clara, US
Experimental investigation and modeling of flapping-wing aerial vehicles are limited by the scarcity of large-scale records that temporally align wing kinematics, aerodynamic responses, and flight states. Existing datasets are often limited in scale and affected by measurement noise and temporal misalignment between rapidly varying wing motion and the associated dynamic response, particularly during high-frequency flapping. We introduce FlapKAD, an episode-structured simulation dataset comprising 2,000 rigid-wing flight episodes and 720,152 valid time steps, with bilateral wing kinematics, aerodynamic responses, and flight states recorded synchronously within a common clock. FlapKAD supports a unified bidirectional sequence-prediction benchmark constructed from the same temporally aligned episodes. The forward task predicts future vertical force coefficients and body vertical velocity from histories of realized flap and twist angles, whereas the inverse task reconstructs future flap- and twist-angle trajectories from the corresponding response histories. A benchmark of eight representative time-series architectures across multiple prediction horizons reveals direction- and horizon-dependent model behavior, systematically higher reconstruction errors for twist angle than for flap angle, and no consistent advantage from increased architectural complexity. FlapKAD provides a reproducible dataset and benchmark for studying coupled wing-kinematic, aerodynamic, and flight-state dynamics in flapping-wing aerial vehicles.