Paper Number
AM15 

Session
AI and ML Based Rheological Characterization

Title
Data-driven constitutive modeling in fluidic four-roll mill flows via small-angle X-ray scattering

Presentation Date and Time
October 10, 2022 (Monday) 4:45

Track / Room
Track 6 / Mayfair

Authors (Click on name to view author profile)

  1. Young, Charles D. (University of Wisconsin-Madison, Chemical and Biological Engineering)
  2. Corona, Patrick T. (University of California, Santa Barbara, Chemical Engineering)
  3. Datta, Anukta (University of California, Santa Barbara, Chemical Engineering)
  4. Helgeson, Matthew E. (University of California, Santa Barbara, Chemical Engineering)
  5. Graham, Michael D. (University of Wisconsin - Madison, Chemical and Biological Engineering)

Author and Affiliation Lines (in printed abstract book)
Charles D. Young1, Patrick T. Corona2, Anukta Datta2, Matthew E. Helgeson2 and Michael D. Graham1
1Chemical and Biological Engineering, University of Wisconsin - Madison, Madison, WI 53715; 2Chemical Engineering, University of California, Santa Barbara, Santa Barbara, CA 93106

Speaker / Presenter 
Young, Charles D.

Keywords
theoretical methods; computational methods; AI based; ML based

Text of Abstract
Two challenges in learning rheological models from data are the availability of training data sets and the measurement of sample deformations under mixed rotational and extensional flows commonly encountered in processing applications. To address this challenge, we use data from scanning small-angle X-ray scattering (sSAXS) measurements in a fluidic four-roll mill (FFoRM). The FFoRM-sSAXS approach provides a large data set of nanostructural measurements along diverse 2D Lagrangian deformation trajectories. We propose a machine learning framework in which FFoRM-sSAXS data is used to train a model which can predict the nanostructural evolution of the fluid for an arbitrary deformation input. We first use autoencoders to learn a reduced order model from scattering data. We then learn the time evolution in the reduced state using a neural network approximation to the governing differential equation. Finally, we learn a transformation from the state data embedded in the scattering intensity to the stress exerted on the fluid. The framework is tested on a rigid rod suspension and compared to theoretical constitutive models and bulk rheological data. We consider an ideal synthetic data set generated from numerical simulations of dilute suspensions, then extend the framework to experimental measurements of semidilute suspensions.