Paper Number
ML8                         My Program 

Session
AI and ML in Rheology

Title
Augmenting machine learning of universal viscoelastic constitutive relationships through curriculum learning using first normal stress difference measurements in LAOS

Presentation Date and Time
October 22, 2025 (Wednesday) 1:50

Track / Room
Track 6 / Sweeney Ballroom C

Authors (Click on name to view author profile)

  1. King, Nicholas (MIT)
  2. Pashkovski, Eugene (The Lubrizol corporation)
  3. Patterson, Reid (The Lubrizol corporation)
  4. Rockwell, Paige (The Lubrizol corporation)
  5. McKinley, Gareth H. (Massachusetts Institute of Technology, Mechanical Engineering)

Author and Affiliation Lines (in printed abstract book)
Nicholas King1, Eugene Pashkovski2, Reid Patterson2, Paige Rockwell2 and Gareth H. McKinley1
1Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139; 2The Lubrizol corporation, Wickliffe, OH 44092

Speaker / Presenter 
King, Nicholas

Keywords
computational methods; artificial intelligence; industrial applications; machine learning; polymer melts

Text of Abstract
Large amplitude oscillatory shear (LAOS) is a key technique for characterizing nonlinear viscoelasticity, from the response of polymer melts to foods and other soft materials. For highly elastic materials such as polymer melts, the first normal stress difference (N1) can become much larger than the shear stress at sufficiently large strains, serving as a sensitive probe of the material’s nonlinear rheological characteristics. The LAOS shear stress and N1 responses thus together provide an extended rheological fingerprint of a complex fluid. We use the Fourier-Tschebyshev framework recently introduced by King et al (2025) to determine the N1 material functions for a silicone oil (PDMS) and a thermoplastic polyurethane (TPU) melt. Experiments are performed using the Gaborheometry strain sweep technique, enabling rapid and quantitative determination of experimental N1 data from small to large strain amplitudes. The measured material response for PDMS is well-described by a conventional multimode differential constitutive equation. However, for the TPU we observe a local ‘band gap’ at which there is a finite but non-oscillatory N1 response at a specific angular frequency and strain amplitude. This cannot be predicted by the widely used Giesekus model for polymer melts, and motivates the search for new constitutive equations that can describe such nonlinear fluid behavior. We use machine learning techniques to learn a tensorial constitutive equation (a Rheological Universal Differential Equation (RUDE)) that can describe the fluid’s extended rheological fingerprint, taking a step towards a more accurate digital fluid twin that enables the prediction of the fluid response under different processing conditions. The framework we outline for analyzing N1 is complementary to the established framework for analyzing the nonlinear shear stresses in LAOS, and is helpful for augmented feature representation in machine learning, hence more fully quantifying the nonlinear viscoelastic properties of a wide range of soft materials.