EUVIMEDEUVIMED entdecken
Crossref

Learning a family of rheological constitutive models using neural operators

Deepak Mangal, Milad Saadat, Safa Jamali

Journal of Rheology · 01.03.2025 · 69 (2) · 55-67

DOI: 10.1122/8.0000908 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Journal of Rheology, online erschienen am 21.01.2025. Verfasst von Deepak Mangal und 2 weiteren Autor:innen. Die Arbeit wurde bisher 7-mal zitiert (Crossref) und nennt 44 Literaturangaben. Gefördert von: National Science Foundation, U.S. Department of Defense.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Data-driven frameworks have emerged as powerful tools for predictions of the rheological properties of a complex fluid by combining phenomenological solutions and observed experimental data. These physics-informed machine learning algorithms usually tackle an instance-wise problem, meaning that the stored solution (or parameters) cannot be generalized to other geometries or flow protocols. Neural operators, on the other hand, offer exceptional capabilities in learning families of constitutive relations, i.e., the operator that maps the input function space to the output space. In this study, we explore the application and accuracy of neural operators in learning constitutive models under diverse flow conditions and fluids of choice. Specifically, we employ four neural operator architectures—Physics-Informed Deep Neural Operators and Fourier Neural Operators, along with data-driven DeepONets and FNOs—to learn and predict two classes of complex fluids: thixotropic elastoviscoplastic (TEVP) and tensorial 2D Giesekus, under several canonical and arbitrary flow protocols. For TEVP fluid, the outputs include shear stress and structure parameter, while for the tensorial 2D Giesekus model, the outputs are the shear and normal stresses. Our results clearly show that neural operators can effectively learn the mapping between the input and output spaces for both scalar and tensorial constitutive models. Furthermore, they can replicate arbitrary flow protocols when trained with sufficient random input functions. Overall, our findings highlight the potential of neural operators to serve as digital twins for rheometers well beyond the common flow kinematics applied in conventional rheometry.

Förderung

  • National Science Foundation
  • U.S. Department of Defense

Zitieren

APA

Mangal, D., Saadat, M., & Jamali, S. (2025). Learning a family of rheological constitutive models using neural operators. Journal of Rheology, 69(2), 55-67. https://doi.org/10.1122/8.0000908

Vancouver

Mangal D, Saadat M, Jamali S. Learning a family of rheological constitutive models using neural operators. Journal of Rheology. 2025;69(2):55-67. doi:10.1122/8.0000908

BibTeX
@article{Mangal2025,
  author = {Mangal, Deepak and Saadat, Milad and Jamali, Safa},
  title = {Learning a family of rheological constitutive models using neural operators},
  journal = {Journal of Rheology},
  year = {2025},
  volume = {69},
  number = {2},
  pages = {55--67},
  doi = {10.1122/8.0000908}
}
Kostenlos starten →