EUVIMEDEUVIMED entdecken
Crossref

Hybrid physics-data-driven deep learning for pore-scale transport in microfluidic system

Yajuan Zhuang, Qiang Ye, Nanlin Liu, Xiaoting Xie, Hengnian Yan, Lingzao Zeng

Physics of Fluids · 01.07.2025 · 37 (7)

DOI: 10.1063/5.0271043 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Physics of Fluids, online erschienen am 23.07.2025. Verfasst von Yajuan Zhuang und 5 weiteren Autor:innen. Die Arbeit wurde bisher 7-mal zitiert (Crossref) und nennt 47 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Microfluidic platforms serve as powerful tools for pore-scale transport studies, generating datasets ideal for data-driven modeling. While conventional physics-driven models dominate flow characterization, their simplified two-dimensional (2D) implementations introduce substantial bias despite computational efficiency. Conversely, pure data-driven approaches face generalization limitations. To address both challenges, we developed a hybrid framework embedding physical transport mechanisms into neural networks. Through microfluidic experiments comparing four modeling approaches (pure data-driven, hybrid, 2D physics-based, and 2D depth-averaged models), we demonstrated that the hybrid model achieved consistent predictive accuracy across temporal scales. It outperformed traditional 2D physics-based models, which exhibited significant prediction bias. The model generalized well at different flow rates, as evidenced by the accurate reconstruction of the concentration field. Both hybrid and data-driven models performed comparably in short-term forecasting with sufficient data, but the hybrid framework uniquely maintained long-term prediction reliability. Quantitative analysis revealed monotonic accuracy improvements with increased experimental data in hybrid model bias correction. This work establishes the first experimental validation of hybrid modeling for pore-scale solute transport in microfluidic systems, bridging microscale observations with deep learning for enhanced porous media transport prediction.

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Zhuang, Y., Ye, Q., Liu, N., Xie, X., Yan, H., & Zeng, L. (2025). Hybrid physics-data-driven deep learning for pore-scale transport in microfluidic system. Physics of Fluids, 37(7). https://doi.org/10.1063/5.0271043

Vancouver

Zhuang Y, Ye Q, Liu N, Xie X, Yan H, Zeng L. Hybrid physics-data-driven deep learning for pore-scale transport in microfluidic system. Physics of Fluids. 2025;37(7). doi:10.1063/5.0271043

BibTeX
@article{Zhuang2025,
  author = {Zhuang, Yajuan and Ye, Qiang and Liu, Nanlin and Xie, Xiaoting and Yan, Hengnian and Zeng, Lingzao},
  title = {Hybrid physics-data-driven deep learning for pore-scale transport in microfluidic system},
  journal = {Physics of Fluids},
  year = {2025},
  volume = {37},
  number = {7},
  doi = {10.1063/5.0271043}
}
Kostenlos starten →