Paper Number
ML11                         My Program 

Session
AI and ML in Rheology

Title
Suspension-balance neural networks for modeling concentrated suspension rheology in confined flows

Presentation Date and Time
October 22, 2025 (Wednesday) 2:50

Track / Room
Track 6 / Sweeney Ballroom C

Authors (Click on name to view author profile)

  1. Davis, Michael (Florida State University)
  2. Castillo Sanchez, Hugo A. (FAMU-FSU College of Engineering, Chemical and Biomedical Engineering)
  3. Rao, Rekha R. (Sandia National Laboratories)
  4. Liu, Leo (FAMU-FSU College of Engineering, Chemical and Biomedical Engineering)

Author and Affiliation Lines (in printed abstract book)
Michael Davis1, Hugo A. Castillo Sanchez2, Rekha R. Rao3 and Leo Liu2
1Florida State University, Tallahassee, FL 32311; 2Chemical and Biomedical Engineering, FAMU-FSU College of Engineering, Tallahassee, FL 32311; 3Sandia National Laboratories, Albuquerque, NM 87123

Speaker / Presenter 
Castillo Sanchez, Hugo A.

Keywords
theoretical methods; computational methods; artificial intelligence; bio-fluid dynamics; biorheology; methods; machine learning; networks; non-Newtonian fluids; particualte systems; rheometry; suspensions; techniques

Text of Abstract
Understanding the complex rheology of concentrated particle suspensions in confined geometries is critical for applications ranging from biomedical microfluidics to industrial coatings. However, traditional continuum models often fail to capture the intricate coupling between hydrodynamics, particle interactions, and confinement effects, especially at high volume fractions. In this work, we introduce Suspension-Balance Neural Networks (SB-NNs), a physics-informed machine learning framework that integrates particle-scale stress balance principles with neural network architectures to model the flow behavior of concentrated suspensions in confined geometries. Our approach learns spatially resolved stress and velocity fields directly from sparse datasets from direct numerical simulations and microfluidic experiments, while embedding conservation laws, particle migration tendencies, and non-Newtonian constitutive behavior into the model structure. We validate the SB-NNs across a range of particle concentrations and confinement ratios, demonstrating superior accuracy in predicting shear-induced particle migration and plug-flow development compared to traditional continuum or empirical models. Furthermore, the framework captures emergent features such as micro-structural heterogeneity that arise uniquely under confinement. These results offer new insight into the mesoscale physics of suspension flows and highlight the potential of hybrid physics-ML tools to serve as surrogate models for complex rheological systems. The implications of this work extend to improving process design in suspension-based technologies and enhancing diagnostic capabilities in biomedical suspension flows such as blood.