Mohamed Elrefaie

Building foundation physics models

Surrogate modeling of the aerodynamic performance for airfoils in transonic regime


Conference paper


Mohamed Elrefaie, Tarek Ayman, Mayar Elrefaie, Eman Sayed, Mahmoud Ayyad, Mohamed M AbdelRahman
AIAA SCITECH 2024 Forum, 2024


View PDF
Cite

Cite

APA   Click to copy
Elrefaie, M., Ayman, T., Elrefaie, M., Sayed, E., Ayyad, M., & AbdelRahman, M. M. (2024). Surrogate modeling of the aerodynamic performance for airfoils in transonic regime. In AIAA SCITECH 2024 Forum. https://doi.org/10.2514/6.2024-2220


Chicago/Turabian   Click to copy
Elrefaie, Mohamed, Tarek Ayman, Mayar Elrefaie, Eman Sayed, Mahmoud Ayyad, and Mohamed M AbdelRahman. “Surrogate Modeling of the Aerodynamic Performance for Airfoils in Transonic Regime.” In AIAA SCITECH 2024 Forum, 2024.


MLA   Click to copy
Elrefaie, Mohamed, et al. “Surrogate Modeling of the Aerodynamic Performance for Airfoils in Transonic Regime.” AIAA SCITECH 2024 Forum, 2024, doi:10.2514/6.2024-2220.


BibTeX   Click to copy

@inproceedings{mohamed2024a,
  title = {Surrogate modeling of the aerodynamic performance for airfoils in transonic regime},
  year = {2024},
  doi = {10.2514/6.2024-2220},
  author = {Elrefaie, Mohamed and Ayman, Tarek and Elrefaie, Mayar and Sayed, Eman and Ayyad, Mahmoud and AbdelRahman, Mohamed M},
  booktitle = {AIAA SCITECH 2024 Forum}
}

Advancements in generative AI models have notably enhanced the automation of 3D shape generation, presenting transformative possibilities in the design of wings for aerospace applications. The optimization of such shapes relies on a large number of numerical simulations, which pose a computational challenge in the preliminary design stages. In this paper, we compare different machine learning models for surrogate modeling of the aerodynamic performance of airfoils for the transonic regime. We propose a new representation of the airfoils by combining geometric and aerodynamic features to comprehensively characterize the airfoil and its operating flight conditions. A training dataset that includes eight different transonic airfoils was generated, where we examined each airfoil under various operational flight conditions, encompassing a wide range of Angle of Attack (AoA) and freestream Mach numbers. This resulted in a dataset comprising 1,362 data points. The surrogate models employed in our study are primarily ensemble learning methods, including Random Forest, Gradient Boosting, and Support Vector Machines, complemented by deep learning techniques. We conduct a comparative analysis of these models to evaluate their efficacy in predicting aerodynamic coefficients. Our experiments show that different surrogate models can accurately and efficiently predict aerodynamic coefficients with an R2 of 99.6% for unseen flight conditions. The dataset and code used in our study are accessible to the public at: https://github.com/Mohamedelrefaie/TransonicSurrogate
We conducted a grid convergence study to ensure that the simulations are accurate and high-fidelity. Notable differences appeared between grids with 28,458 and 56,916 cells, so we selected a 20,196-cell grid to maintain computational efficiency while accurately capturing key physics, including shock wave formation.
Grid Convergence Study for the inviscid case about RAE2822 airfoil at 𝑀∞ = 0.73 and AoA= 3.19◦.

Share



Follow this website


You need to create an Owlstown account to follow this website.


Sign up

Already an Owlstown member?

Log in