Paper Number
AM9 

Session
AI and ML Based Rheological Characterization

Title
Viscoelastic free surface flows: From models to experiments and somewhere in between

Presentation Date and Time
October 10, 2022 (Monday) 2:10

Track / Room
Track 6 / Mayfair

Authors (Click on name to view author profile)

  1. Bolintineanu, Dan (Sandia National Laboratories)
  2. Hamersky, Mark (The Procter & Gamble Co)
  3. Hartt, William H. (The Procter & Gamble Co)
  4. Lindberg, Seth (The Procter & Gamble Co)
  5. Ortiz, Weston (University of New Mexico)
  6. Rao, Rekha (Sandia National Labs)

Author and Affiliation Lines (in printed abstract book)
Dan Bolintineanu1, Mark Hamersky2, William H. Hartt2, Seth Lindberg2, Weston Ortiz3 and Rekha Rao1
1Sandia National Laboratories, Albuquerque, NM 94551; 2The Procter & Gamble Co, Cincinnati, OH 45232; 3University of New Mexico, Albuquerque, OH 94551

Speaker / Presenter 
Hartt, William H.

Keywords
experimental methods; theoretical methods; computational methods; AI based; ML based; polymer solutions

Text of Abstract

Fiber formation of polymeric materials is an industrial relevant process that involves history dependent shear, extensional, and elastic material response. The polymer solution is compressed and sheared in a die and then exhibits die swell, a predominantly elastic response, as the fibers are extruded. Computational fluid dynamics (CFD) models are needed to describe these flows, to inform process design and perform process optimization. However, CFD of viscoelastic flows has both increased complexity and increased computational costs compared to Newtonian flows, since a polymer stress constitutive equation is necessary to describe the polymer rheology.

In this paper, we investigate die swell and CaBER flows of polymeric solutions where there is also a free surface. Multimode linear and exponential included Phan-Thien-Tanner (PTT) constitutive equations are investigated for the viscoelastic response. An arbitrary-Lagrangian-Eulerian (ALE) implementation with pseudo-solid mesh motion is used to solve for the location of the free surface. A log-conformation tensor formulation of the stress equations is used to improve stability of the method as the fluid elasticity is increased. Die swell data is available for a number of model fluids, with varying viscosities, elastic behavior, and shear and extensional thinning. The epsilon parameter in the PTT model is investigated to improve the model predictions to the data, implying that extensional rheology can also impact swell ratios.

Some preliminary work investigating incorporating AI/ML into our rheological measurements will also be discussed. Physics-informed neural networks (PINNs) are a recently developed numerical technique that augments traditional approaches to fluid mechanics modeling with high-dimensional and heterogeneous data. We will apply a PINNs framework to utilize data from a CaBER experiment.