Developing Reduced-Order Models for Predicting Aerodynamic Performance of Novel Wing

Session Number

2

Advisor(s)

Katherine J. Asztalos, Argonne National Laboratory

Location

A115

Discipline

Engineering

Start Date

15-4-2026 11:10 AM

End Date

15-4-2026 11:55 AM

Abstract

Traditional aerodynamic optimization has relied heavily on the use of computationally expensive methods, requiring hundreds to thousands of Computational Fluid Dynamics simulations and data to produce optimal designs, often taking days to weeks to complete. This study presents a Generative Adversarial Network (GAN) to optimize this process by generating candidate airfoils that are compared with real airfoil data (gathered from UIUC airfoil database consisting of 200 airfoil profiles, 100 normalized coordinates) to optimizing the creation of realistic airfoils that can be continually refined to meet target metrics. The GAN utilizes custom multi-objective loss functions for optimization purposes, drawing on geometric loss (loss characterized by physical properties), aerodynamic loss (error loss calculated through predicted and actual aerodynamic coefficients), and smoothness loss (characterized by surface coordinates). Additionally, batch analyses are performed in XFLR5 through the integration of XFLRpy python programs. The batch analyses display physical characteristics alongside crucial airfoil coefficients and environmental conditions like Coefficients of Lift, drag, angles of attack, Reynold’s numbers, and many graphs relating these important quantities. This suggested model significantly cuts down on computation cost & time, allows for the easy adaptation and creation of reliable airfoil designs, and refines new and pre-existing airfoil geometry

Share

COinS
 
Apr 15th, 11:10 AM Apr 15th, 11:55 AM

Developing Reduced-Order Models for Predicting Aerodynamic Performance of Novel Wing

A115

Traditional aerodynamic optimization has relied heavily on the use of computationally expensive methods, requiring hundreds to thousands of Computational Fluid Dynamics simulations and data to produce optimal designs, often taking days to weeks to complete. This study presents a Generative Adversarial Network (GAN) to optimize this process by generating candidate airfoils that are compared with real airfoil data (gathered from UIUC airfoil database consisting of 200 airfoil profiles, 100 normalized coordinates) to optimizing the creation of realistic airfoils that can be continually refined to meet target metrics. The GAN utilizes custom multi-objective loss functions for optimization purposes, drawing on geometric loss (loss characterized by physical properties), aerodynamic loss (error loss calculated through predicted and actual aerodynamic coefficients), and smoothness loss (characterized by surface coordinates). Additionally, batch analyses are performed in XFLR5 through the integration of XFLRpy python programs. The batch analyses display physical characteristics alongside crucial airfoil coefficients and environmental conditions like Coefficients of Lift, drag, angles of attack, Reynold’s numbers, and many graphs relating these important quantities. This suggested model significantly cuts down on computation cost & time, allows for the easy adaptation and creation of reliable airfoil designs, and refines new and pre-existing airfoil geometry