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

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