Paper Number
ML5                         My Program 

Session
AI and ML in Rheology

Title
Inferring viscoelastic model parameters from complex flows using physics informed neural networks

Presentation Date and Time
October 22, 2025 (Wednesday) 11:10

Track / Room
Track 6 / Sweeney Ballroom C

Authors (Click on name to view author profile)

  1. Bolintineanu, Dan S. (Sanda National Laboratories, Energetics, Multiphase, and Soft Matter Science)
  2. Sankaran, Shyam (University of Pennsylvania, Mechanical Engineering and Applied Mechanics)
  3. Trask, Nathaniel A. (University of Pennsylvania, Mechanical Engineering and Applied Mechanics)
  4. Perdikaris, Paris G. (University of Pennsylvania, Mechanical Engineering and Applied Mechanics)
  5. Rao, Rekha R. (Sandia National Laboratories)
  6. Ortiz, Weston (University of New Mexico)

Author and Affiliation Lines (in printed abstract book)
Dan S. Bolintineanu1, Shyam Sankaran2, Nathaniel A. Trask2, Paris G. Perdikaris2, Rekha R. Rao1 and Weston Ortiz3
1Energetics, Multiphase, and Soft Matter Science, Sanda National Laboratories, Albuquerque, NM 87111; 2Mechanical Engineering and Applied Mechanics, University of Pennsylvania, Philadelphia, PA 19104; 3University of New Mexico, Albuquerque, NM 87111

Speaker / Presenter 
Bolintineanu, Dan S.

Keywords
computational methods; artificial intelligence; methods; machine learning; non-Newtonian fluids

Text of Abstract
Viscoelastic constitutive models often entail many parameters, and are difficult to calibrate using standard rheological measurements. We present a deep learning inverse calibration framework based on physics-informed neural networks (PINNs) that combines full-field data from multiple data sources to infer constitutive parameters. We demonstrate our approach using simulation data from the extensional and static phases of a capillary break extensional rheometry (CaBER) flow simulation. The use of full-field velocity and stress data yields adequate inference of rheological parameters for both a Newtonian and linear Phan-Thien-Tanner (LPTT) fluid, with improvements in the latter case achieved by combining data from the static and extensional phases. We explore the robustness of this approach to the amount of available data, as well as to Gaussian noise in the data, in order to assess the potential for use with experimental datasets. Using velocity and stress data only on the boundary leads to incomplete parameter identification for the single-mode LPTT model. However, by augmenting the dataset with measurements from a simple Poiseuille flow experiment, we obtain accurate parameter estimates even with limited data. The ease of using multiple data sources in a PINNs framework demonstrates the usefulness of this approach for augmenting traditional model calibrations, and potentially combining full-field measurements with standard rheological measurements for improved model parameter inference. This method provides a crucial step towards efficient and accurate characterization of complex fluids, circumventing many limitations of conventional methods that rely on idealized flow conditions and simplified geometries. *Sandia National Laboratories is a multi mission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC., a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525