EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Physics-informed Kolmogorov–Arnold network for consolidation of high-water-content soft soil

Delian Ma, Liyou Chen, Lijian Ni, Kaiyue Chen, Shaoxiang Zeng, Mingyuan Wang

AIP Advances · 01.11.2026 · 16 (11)

DOI: 10.1063/5.0332414 ↗

Kurz gefasst

Fachartikel in der Zeitschrift AIP Advances, online erschienen am 19.11.2026. Verfasst von Delian Ma und 5 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Major Science and Technology Program of the Zhejiang Provincial Department of Water Resources, "Pioneer" and "Leading Goose" R&D Program of Zhejiang.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Vacuum preloading with geotextile confinement can enhance the consolidation efficiency of high-water-content soft soil. An accurate prediction of this consolidation behavior is crucial for engineering decision-making. This study proposes a physics-informed Kolmogorov–Arnold network (PIKAN) for consolidation prediction of geotextile-confined soft soil under drainage enhancement. First, the nonlinear time-dependent consolidation governing equation, along with the initial and mixed boundary conditions, is embedded into a unified loss function to guide physics-constrained learning and enforce physically consistent predictions. Next, a KAN backbone with learnable B-spline edge functions replaces the conventional fully connected architecture, improving the ability of the model to capture localized solution features in physics-informed training. To improve boundary compliance and training stability under mixed boundary conditions, a curriculum-based training strategy is introduced, incorporating stratified sampling, staged learning-rate adjustment, and progressive reinforcement of Dirichlet constraints. The proposed model is validated through one-dimensional and two-dimensional consolidation benchmark problems, as well as a large-strain consolidation case. Statistical results from 20 independent runs demonstrate that the proposed PIKAN achieves superior boundary compliance and predictive stability compared with multilayer perceptron-based physics-informed neural networks. The model can be extended to other coupled consolidation scenarios with strong nonlinearity and mixed boundary conditions, offering a general route toward robust and physically consistent spatiotemporal prediction in soft ground improvement applications.

Förderung

  • Major Science and Technology Program of the Zhejiang Provincial Department of Water Resources
  • "Pioneer" and "Leading Goose" R&D Program of Zhejiang

Zitieren

APA

Ma, D., Chen, L., Ni, L., Chen, K., Zeng, S., & Wang, M. (2026). Physics-informed Kolmogorov–Arnold network for consolidation of high-water-content soft soil. AIP Advances, 16(11). https://doi.org/10.1063/5.0332414

Vancouver

Ma D, Chen L, Ni L, Chen K, Zeng S, Wang M. Physics-informed Kolmogorov–Arnold network for consolidation of high-water-content soft soil. AIP Advances. 2026;16(11). doi:10.1063/5.0332414

BibTeX
@article{Ma2026,
  author = {Ma, Delian and Chen, Liyou and Ni, Lijian and Chen, Kaiyue and Zeng, Shaoxiang and Wang, Mingyuan},
  title = {Physics-informed Kolmogorov–Arnold network for consolidation of high-water-content soft soil},
  journal = {AIP Advances},
  year = {2026},
  volume = {16},
  number = {11},
  doi = {10.1063/5.0332414}
}
Kostenlos starten →