Paper Number
ML16                         My Program 

Session
AI and ML in Rheology

Title
Obtaining rheological constitutive equations for geopolymer systems from scarce data

Presentation Date and Time
October 23, 2025 (Thursday) 8:45

Track / Room
Track 6 / Sweeney Ballroom C

Authors (Click on name to view author profile)

  1. Dabiri, Donya (Northeastern University, Department of Mechanical and Industrial Engineering)
  2. Egnaczyk, Ted (University of Delaware, Department of Chemical and Biomolecular Engineering)
  3. Wagner, Norman (University of Delaware)
  4. Jamali, Safa (Northeastern University, Mechanical and Industrial Engineering)

Author and Affiliation Lines (in printed abstract book)
Donya Dabiri1, Ted Egnaczyk2, Norman Wagner2 and Safa Jamali1
1Mechanical and Industrial Engineering, Northeastern University, Boston, MA 02115; 2University of Delaware, Newark, DE

Speaker / Presenter 
Dabiri, Donya

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

Text of Abstract
Understanding the behavior of complex fluids through their constitutive dynamics has long been a subject of considerable interest. Traditionally, this involved phenomenological or empirical formulations of material functions. In recent years, the Sparse Identification of Nonlinear Dynamical Systems (SINDy) technique has demonstrated strong potential in extracting compact yet accurate models from noisy datasets. Recently, Rheo-SINDy has been developed with specific applications of model discovery in rheology as well. In this work, we use a hybrid framework that combines Rheology-informed Neural Networks (RhINNs) with SINDy to uncover constitutive relations in geopolymer systems using experimental data, where classical rheological models fall short or are absent. Our dataset includes time sweeps of the viscoelastic moduli. This integrated approach leverages machine learning and experimental data—capitalizing on SINDy’s ability to infer governing equations with limited information, and RhINNs’ strength in learning and solving such equations. We validate our method by constructing simplified yet expressive models that accurately capture the behavior of geopolymers across different flow protocols.