EUVIMEDEUVIMED entdecken
Crossref

A fast deep-learning surrogate with autoencoded boundary feature extraction for supersonic combustion flow field prediction

Tong Zhao, Yvpeng Liu, Chong Wu, Bing Liu, Shaohua Zhu, Jian An, Junqiang Bai

Physics of Fluids · 01.11.2026 · 38 (11)

DOI: 10.1063/5.0348875 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Physics of Fluids, online erschienen am 05.11.2026. Verfasst von Tong Zhao und 6 weiteren Autor:innen. Die Arbeit nennt 52 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Basic Research Program of Shaanxi Province.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Machine learning surrogates provide an efficient approach to accelerating computational fluid dynamics simulations. However, their application in complex system such as supersonic reacting flows remains challenging because of the strong coupling among heterogeneous boundary information, mesh topology, and thermochemical flow structures. This study proposes an encoder–decoder surrogate, named supersonic combustion flow prediction (SCFP), which reformulates flow field prediction as a boundary conditioned solution operator approximation. Instead of directly regressing the flow field from the original inputs, SCFP encodes the physical boundary conditions and geometric descriptors into a compact latent vector on a low-dimensional feature manifold. A decoder then combines the latent vector with the coordinates of each cell to reconstruct velocity, pressure, temperature, and species fields at each cell. SCFP is evaluated on datasets of three-dimensional supersonic cavity combustors. The results show that encoding multimodal boundary information as a 17-dimensional vector enables a continuous mapping of input parameters from physical space to the feature manifold. Compared with seven benchmark models, SCFP achieved lower relative L2 error and higher correlation coefficients. For each case containing approximately 600 000 cells, the complete three-dimensional flow fields are inferred in less than 10 s. This work demonstrates that organizing multimodal boundary information on a continuous latent manifold improves the prediction accuracy of reacting flows and design space exploration of high-speed propulsion systems.

Förderung

  • National Natural Science Foundation of China
  • Natural Science Basic Research Program of Shaanxi Province

Zitieren

APA

Zhao, T., Liu, Y., Wu, C., Liu, B., Zhu, S., An, J., & Bai, J. (2026). A fast deep-learning surrogate with autoencoded boundary feature extraction for supersonic combustion flow field prediction. Physics of Fluids, 38(11). https://doi.org/10.1063/5.0348875

Vancouver

Zhao T, Liu Y, Wu C, Liu B, Zhu S, An J, et al. A fast deep-learning surrogate with autoencoded boundary feature extraction for supersonic combustion flow field prediction. Physics of Fluids. 2026;38(11). doi:10.1063/5.0348875

BibTeX
@article{Zhao2026,
  author = {Zhao, Tong and Liu, Yvpeng and Wu, Chong and Liu, Bing and Zhu, Shaohua and An, Jian and Bai, Junqiang},
  title = {A fast deep-learning surrogate with autoencoded boundary feature extraction for supersonic combustion flow field prediction},
  journal = {Physics of Fluids},
  year = {2026},
  volume = {38},
  number = {11},
  doi = {10.1063/5.0348875}
}
Kostenlos starten →