CFD simulation pipeline using the OpenFOAM framework to generate a 2D database for machine learning purposes using NACA symmetrical, 4- and 5-digit airfoils.

The pipeline can generate results of subsonic simulations for Mach numbers M < 0.6 and any angle of attack. However, it is advised to stay within alpha = [-20, 20] deg as any lower or higher angle will only predict separation.

The detailed explanation about the simulation pipeline as well as the performed verification and validation can be found here (LINK TO PAPER).

To reference this algorithm, please cite the repository as:

@misc{widmann2023nacaFOAM,
  title={Automated Aerodynamic 2D Data Generation for Machine Learning},
  author={Widmann, Sebastian and Schlichter, Philipp and Reck, Michaela and Indinger, Thomas},
  year={2023},
  publisher={GitHub},
  howpublished={\url{https://github.com/sebastianwidmann/nacaFOAM}},
}

The subsets of the airfoilMNIST datasets are subsequently linked here once the data has been published.

• airfoilMNIST-raw: https://web.tresorit.com/l/caxBx#uO5w7Q1BS6g1SvDAsrdyVA (currently a snippet (10% of the total dataset), containing all airfoils and angles of attack for Mach = 0.2 - more will be added asap)
• airfoilMNIST: Not uploaded yet
• airfoilMNIST-incompressible: https://syncandshare.lrz.de/getlink/fiHGg7CzJ1tajtpDxwR65D/

• OpenFOAM v2206 or newer
• Python packages as set in requirements.txt

• To execute the main script nacaFOAM.py, one has to enter the OpenFOAM environment or source toward the bashrc script
  • Ubuntu: source /lib/openfoam/openfoam2206/etc/bashrc

• Typically, Python is installed in the usr/bin/python3.XX directory on local Linux machines. However, this is often not true for HPC systems.
  • By default, nacaFOAM.py will look for the executable of Python in the following directory -> usr/bin/python3
  • If a virtual environment is used, the directory to the Python executable must be specified in the shebang line at the beginning of nacaFOAM.py

• To execute the main script, 3 parameters must be provided in the form of a parameter list (see line 335 in nacaFOAM.py)
  • Angles of attack
  • Mach numbers
  • Airfoil codes
• The simulations will be saved into the database directory which will be generated once the main script is executed. To minimise the storage capacity, it is advised to delete the raw data folders and only keep the VTK files in the database folder.

Optional If rhoSimpleFOAM should run on a different core count than 4, respecify the following arguments

• nacaFOAM.py: the argument nProc in line 336,
• nacaFOAM.py: the argument nr in line 262,
• nacaFOAM-template/system/decomposeParDict: modify the argument numberOfSubdomains in line 17

• Specify range of angles of attack as list or array of strings in line 325

  • angles = np.arange(-5, 21, 1)
    

• Specify range of Mach numbers as list or array in line 326

  • angles = np.arange(0.05, 0.65, 0.05)
    

• Specify range of NACA airfoils through function generateNacaAirfoils()

  • nacas = generateNacaAirfoils()
    

    • In the function the different digits can be modified
      • NACA 4-series --> see lines 42-44
      • NACA 5-series --> see lines 53-56
• Execute nacaFOAM.py script

• Easiest way to run a single simulation is by replacing lines 325-327 with the following lines

  • angles = np.array([0])
    

  • machs = np.array([0.15])
    

  • nacas = ['0012']
    

• Execute nacaFOAM.py script

• To generate the textfile with the lift, drag and moment coefficients for all simulations, simply execute the postProcessing.py script after the nacaFOAM.py script has been executed sucessfully. This script will generate a new .csv file within the database directory.