anon-ml-2026/Wing-MD

Wing-MD: A Multidisciplinary Dataset for Machine Learning-Driven Aerostructural Wing Design

Wing-MD is a dataset that comprises of aerostructural analysis of multiple wing geometries in different operating conditions. This aerostructural wing dataset can be used to train ML models for performing aerostructural analysis of 3D transonic wing geometries. It can also be used to benchmark new ML-based approaches for multidisciplinary analysis and optimization (MDAO).

NOTE: If you do not want to download the full dataset, then you can use the train and test parquet files which contains input parameter values and corresponding scalar output values. You can also download desired number of individual sample folders. Refer below for more details about what is there in individual folders.

Dataset details:

This repository consists of two separate datasets - one for training (2471 sample points) and other for testing (492 sample points). Both training or testing dataset consists of multiple sub-folders, each containing solution files generated during aerostructural analysis. Specifically, each sub-folder (either in testing or training dataset folder) consists of following files:

1. parameters.json: A JSON file that contains the value of parameters used for the analysis. The code snippet below shows an example from one of the sub-folders. There are 21 parameters in this dataset.

  {
    # operating conditions
    "mach": 0.8946880746322314,
    "alpha": 3.0906308137746494,
    "altitude": 11059.583794251872,
    # airfoil upper surface CST coefficients
    "upper_cst_0": 0.16031407680528947,
    "upper_cst_1": 0.18262630639228722,
    "upper_cst_2": 0.15150491612696584,
    "upper_cst_3": 0.1742494674503922,
    "upper_cst_4": 0.17949621814086716,
    "upper_cst_5": 0.2194988873820222,
    # airfoil lower surface CST coefficients
    "lower_cst_0": -0.1968537762484503,
    "lower_cst_1": -0.1142864070908371,
    "lower_cst_2": -0.20078643470620053,
    "lower_cst_3": -0.14899202308429582,
    "lower_cst_4": -0.11176538864315147,
    "lower_cst_5": 0.06286741918698549,
    # wing planform variables
    "span": 13.093900077141413,
    "sweep": 25.317460425398412,
    "dihedral": 5.021048856644773,
    "twist": -3.3509339884995013,
    "tip_chord": 2.023372652014176,
    # structural variables
    "load_factor": 0.8938751865414288
  }

2. scalar_outputs.json: A JSON file that contains the value of scalar outputs from the aerostructural analysis. The code snippet below shows an example from one of the sub-folders. There are 9 scalar outputs in this dataset.

  {
    "cd": 0.050106008124448494, # drag coefficient
    "cdp": 0.04662458623241084, # pressure component of drag coefficient
    "cdv": 0.003481421892037658, # skin friction component drag coefficient
    "cl": 0.14791799066412906, # lift coefficient
    "clp": 0.14796187098001148, # pressure component of lift coefficient
    "clv": -4.3880315882470754e-05, # skin friction component lift coefficient
    "cmx": 0.13776796962006113, # moment coefficient along x axis
    "cmy": -0.14908249833037057, # moment coefficient along y axis
    "cmz": -0.06786819421734241 # moment coefficient along z axis
  }

3. surface_solution.vtu: A VTU file that contains wing surface solution from aerodynamics solver. This file contains following quantitites for each surface mesh node:

• Deflected wing surface mesh
• cp: wing surface pressure coefficient
• cf: wing surface skin friciton (magnitude) coefficient
• cfx: wing surface skin friciton (x-direction) coefficient
• cfy: wing surface skin friciton (y-direction) coefficient
• cfz: wing surface skin friciton (z-direction) coefficient
• Temperature, density, pressure, and surface $y^+$ values

You can easily read this file using pyvista library in Python. This file will be useful for graph/mesh-based ML models.

4. surface_outputs.hdf5: A HDF5 file that contains same quantities as the surface_solution.vtu file but in a point-cloud format. Note that mesh coordinates are available in this file as a point-cloud (connectivity information is not available). You can read this file using h5py Python library. This file will be useful for point-cloud based ML models.

5. structural_solution.vtk: A VTU file that contains wingbox solution from structural solver. This file contains following quantities (for each element in wingbox) from the fnitie element analysis:

• u, v, w: translational displacements in x, y, and z direction
• rotx, roty, rotz: rotational displacements about x, y, and z direction
• ex0, ey0, exy0: membrane strains
• ex1, ey1, exy1: bending strains
• eyz0, exz0: transverse shear strains
• erot: drilling strain
• sx0, sy0, sxy0: membrane stress
• sx1, sy1, sxy1: bending stress
• syz0, sxz0: transverse shear stress
• srot: drilling stress
• failure: failure criteria
• fx, fy, fz: applied force in x, y, and z direction
• mx, my, mz: applied moment about x, y, and z direction
• dv1: shell thickness for each element

You can easily read this file using pyvista library in Python.

6. wing_model.stl: A STL file representing the undeflected wing geometry model. This file can be used as an input to ML models, if required.

NOTE: Please contact us if you want the raw CGNS surface and/or volume solution files generated by the CFD solver or if you want the raw structural solution from the FEM solver. To limit the size of the dataset, it is not included in this repository.