A Comparative Study of DPM AI-Generated and Traditional Airfoils for Lift Performance in an Experimental Wind Tunnel Setup
Katelyn Ozeki
American School in Japan, Tokyo, Japan
Publication date: November 20, 2025
American School in Japan, Tokyo, Japan
Publication date: November 20, 2025
DOI: http://doi.org/10.34614/JIYRC2025II34
ABSTRACT
Airfoil design has conventionally relied on computational methods; yet recent advances in generative AI have introduced new possibilities for generating novel geometries. Traditionally, artificial intelligence in aerodynamics has concentrated on performance prediction and analysis, but its application has begun to expand towards airfoil design. This study investigates diffusion probabilistic models (DPMs) as an alternative generative approach for airfoil design, comparing aerodynamic characteristics of two distinct DPM-generated datasets (Wagenaar+ ‘24, Graves+ ‘24), along with assessing their performance against conventional airfoils. An experimental wind tunnel test was conducted on all airfoils, produced by 3D printer, to evaluate lift at set angles corresponding to expected lift coefficients, maximum lift and optimal angles of attack. Findings indicated that one of the two DPM-generated airfoil datasets outperformed the traditional airfoils in lift performance, whereas the other achieved less favorable results, emphasizing the importance of training datasets, parameters and model objectives on AI outcomes. Differences in the traditional airfoils and generated airfoils were also examined in detail with varying results. Overall, this study highlights the potential of DPM generative approaches to advance airfoil design, while suggesting next steps that include further research into other aerodynamic principles and the effects of varying algorithm inputs.
Airfoil design has conventionally relied on computational methods; yet recent advances in generative AI have introduced new possibilities for generating novel geometries. Traditionally, artificial intelligence in aerodynamics has concentrated on performance prediction and analysis, but its application has begun to expand towards airfoil design. This study investigates diffusion probabilistic models (DPMs) as an alternative generative approach for airfoil design, comparing aerodynamic characteristics of two distinct DPM-generated datasets (Wagenaar+ ‘24, Graves+ ‘24), along with assessing their performance against conventional airfoils. An experimental wind tunnel test was conducted on all airfoils, produced by 3D printer, to evaluate lift at set angles corresponding to expected lift coefficients, maximum lift and optimal angles of attack. Findings indicated that one of the two DPM-generated airfoil datasets outperformed the traditional airfoils in lift performance, whereas the other achieved less favorable results, emphasizing the importance of training datasets, parameters and model objectives on AI outcomes. Differences in the traditional airfoils and generated airfoils were also examined in detail with varying results. Overall, this study highlights the potential of DPM generative approaches to advance airfoil design, while suggesting next steps that include further research into other aerodynamic principles and the effects of varying algorithm inputs.
