EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

GeoDiff: A Geometric Conditional Diffusion Model for Interpretable Geometric Control and Intelligent Optimization in Airfoil Design

Mingdong Li, Yufei Zhang, Yingyuan Liu, Kang An, Leqin Wang

Journal of Computational Design and Engineering · 12.11.2026

DOI: 10.1093/jcde/qwag097 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Computational Design and Engineering, online erschienen am 12.11.2026. Verfasst von Mingdong Li und 4 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

To address the issues of opaque control interfaces and the inability to design airfoils based on physical parameters in data‑driven airfoil design, this paper proposes GeoDiff—a geometric conditional diffusion model and explores its integration with reinforcement learning (RL) and genetic algorithms for multi-objective optimization under multi-constraint scenarios. Serving as the core generative module, GeoDiff has been experimentally validated for its generation accuracy, demonstrating its ability to generate airfoils whose key geometric features closely resemble those of a given reference airfoil based on its conditional information. Monotonic response tests further indicate that GeoDiff yields monotonically consistent geometric and aerodynamic performance responses to variations in geometric inputs, thereby confirming its conditional controllability. Moreover, this paper proposes a fixed‑seed 40-step DDIM sampling strategy combined with geometric condition perturbations. Building on this, the study further investigates the effectiveness of incorporating GeoDiff into optimization loops using proximal policy optimization (PPO) and NSGA-II. Results for the representative airfoil NACA4412 indicate that, under fair comparison conditions, the GeoDiff‑NSGA-II and GeoDiff‑PPO frameworks achieve superior optimization performance compared to the conventional CST‑NSGA-II approach, albeit at the cost of some computational overhead.

Zitieren

APA

Li, M., Zhang, Y., Liu, Y., An, K., & Wang, L. (2026). GeoDiff: A Geometric Conditional Diffusion Model for Interpretable Geometric Control and Intelligent Optimization in Airfoil Design. Journal of Computational Design and Engineering. https://doi.org/10.1093/jcde/qwag097

Vancouver

Li M, Zhang Y, Liu Y, An K, Wang L. GeoDiff: A Geometric Conditional Diffusion Model for Interpretable Geometric Control and Intelligent Optimization in Airfoil Design. Journal of Computational Design and Engineering. 2026. doi:10.1093/jcde/qwag097

BibTeX
@article{Li2026,
  author = {Li, Mingdong and Zhang, Yufei and Liu, Yingyuan and An, Kang and Wang, Leqin},
  title = {GeoDiff: A Geometric Conditional Diffusion Model for Interpretable Geometric Control and Intelligent Optimization in Airfoil Design},
  journal = {Journal of Computational Design and Engineering},
  year = {2026},
  doi = {10.1093/jcde/qwag097}
}
Kostenlos starten →