Physics-informed neural networks for understanding shear migration of particles in viscous flow

Abstract

We harness the physics-informed neural network (PINN) approach to extend the utility of phenomenological models for particle migration in shear flow. Specifically, we propose to constrain the neural network training via a model for the physics of shear-induced particle migration in suspensions. Then, we train the PINN against experimental data from the literature, showing that this approach provides both better fidelity to the experiments, and a novel understanding of the relative roles of the hypothesized migration fluxes. We first verify the PINN approach for solving the inverse problem of radial particle migration in a non-Brownian suspension in an annular Couette flow. In this classical case, the PINN yields the same value (as reported in the literature) for the ratio of the two parameters of the empirical model. Next, we apply the PINN approach to analyze experiments on particle migration in both non-Brownian and Brownian suspensions in Poiseuille slot flow, for which a definitive calibration of the phenomenological migration model has been lacking. Using the PINN approach, we identify the unknown/empirical parameters in the physical model through the inverse solver capability of PINNs. Specifically, the values are significantly different from those for the Couette cell, highlighting an inconsistency in the literature that uses the latter value for Poiseuille flow. Importantly, the PINN results also show that the inferred values of the empirical model's parameters vary with the shear Péclet number and the particle bulk volume fraction of the suspension, instead of being constant as assumed in some previous literature.

arXiv comment19 pages, 9 figures; v2: minor updates & provided data accessibility statement; v3: accepted for publication in International Journal of Multiphase Flow

References in corpus (9)

• SciPy 1.0--Fundamental Algorithms for Scientific Computing in Python
• Machine Learning for Fluid Mechanics
• NSFnets (Navier-Stokes Flow nets): Physics-informed neural networks for the incompressible Navier-Stokes equations
• A composite neural network that learns from multi-fidelity data: Application to function approximation and inverse PDE problems
• A Physics-Guided Neural Network Framework for Elastic Plates: Comparison of Governing Equations-Based and Energy-Based Approaches
• Physics informed neural networks for continuum micromechanics
• Physics-Informed Neural Network Method for Forward and Backward Advection-Dispersion Equations
• Learning Unknown Physics of non-Newtonian Fluids
• A two-fluid model for numerical simulation of shear-dominated suspension flows