EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Physics-Informed Neural Network Simulation of Proppant Transport in Low-Viscosity Fracturing Fluid

Yuyang Liu, Kai Shen, Guanglong Sheng, Hui Zhao

Processes · 21.07.2026 · 14 (14) · 2352

DOI: 10.3390/pr14142352 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Processes, online erschienen am 21.07.2026. Verfasst von Yuyang Liu und 3 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: National Natural Science Foundation of China, National Major Science and Technology Projects, “Tianshan Talents” Training Program-Science and Technology Innovation Team Project of Xinjiang Uygur Autonomous Region.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Proppant transport and deposition directly control fracture conductivity during hydraulic fracturing, especially under low-viscosity fracturing fluid conditions where strong convection, settling, deposition, and erosion coexist. To improve the computational efficiency of proppant transport simulation, this study develops a dual-network physics-informed neural network (dual-PINN) framework based on a previously developed fifth-order WENO proppant transport model. In the proposed framework, PINNc is used to predict the proppant concentration field, while PINNhp is used to predict the deposited proppant bed height. The governing physical constraints, including convection, effective diffusion, settling, deposition, and erosion, are embedded into the loss function to achieve coupled concentration–deposition prediction. The model is validated using a benchmark case with a fracture length of 100 m, fracture height of 30 m, fracture width of 0.002 m, injection rate of 0.015 m3/s, and inlet proppant concentration of 0.1. Compared with the fifth-order WENO reference solution, the RMSE values of the predicted proppant concentration at 100 s, 400 s, and 1500 s are 6.36 × 10−3, 7.38 × 10−3, and 8.28 × 10−3, respectively. The offline training time of the dual-PINN model is approximately 15 min, and the trained model requires only about 3 ms for one forward prediction. In comparison, the fifth-order WENO method requires 0.43–22.3 s for a single simulation under the tested mesh resolutions, corresponding to an online speed-up of approximately 143 to 7430. Sensitivity analyses further show that the proposed model maintains stable and physically consistent responses for inlet proppant concentrations of 0.01–0.25 and injection rates of 0.01–0.06 m3/s. These results demonstrate that the proposed dual-PINN framework can provide an efficient mesh-free surrogate for rapid proppant transport prediction and fracturing parameter optimization.

Förderung

  • National Natural Science Foundation of China
  • National Major Science and Technology Projects
  • “Tianshan Talents” Training Program-Science and Technology Innovation Team Project of Xinjiang Uygur Autonomous Region

Zitieren

APA

Liu, Y., Shen, K., Sheng, G., & Zhao, H. (2026). Physics-Informed Neural Network Simulation of Proppant Transport in Low-Viscosity Fracturing Fluid. Processes, 14(14), 2352. https://doi.org/10.3390/pr14142352

Vancouver

Liu Y, Shen K, Sheng G, Zhao H. Physics-Informed Neural Network Simulation of Proppant Transport in Low-Viscosity Fracturing Fluid. Processes. 2026;14(14):2352. doi:10.3390/pr14142352

BibTeX
@article{Liu2026,
  author = {Liu, Yuyang and Shen, Kai and Sheng, Guanglong and Zhao, Hui},
  title = {Physics-Informed Neural Network Simulation of Proppant Transport in Low-Viscosity Fracturing Fluid},
  journal = {Processes},
  year = {2026},
  volume = {14},
  number = {14},
  pages = {2352},
  doi = {10.3390/pr14142352}
}
Kostenlos starten →