Investigating vorticity in granular flows through particle image velocimetry and a neural particle method

Abstract

Understanding the physics behind dense granular flows poses challenging open problems due to their complex dynamics. This study investigates the vorticity generation in granular flows through a combined analysis of Neural Particle Method (NPM) simulations and Particle Image Velocimetry (PIV) measurements obtained from laboratory column collapse experiments. The adaptive refinement neural particle method (arNPM), a Lagrangian physics-informed neural network approach, implemented with the μ(I) rheology, provides a comprehensive analysis for granular flows through particle tracking. Numerical results reveal the high vorticity in the fluid-like zone, indicating strong local shear and rotational motion during spreading. To verify and corroborate these findings, PIV analysis is performed to reconstruct the velocity field and compute corresponding vorticity distributions. Different post-processing techniques on the PIV raw data, including short-time temporal averaging, are applied to reduce measurement noise and improve consistency of the velocity and vorticity fields. The comparison between simulations and experiments shows reasonably good agreement, demonstrating that vorticity provides a physically meaningful indicator of localized shear and of the transition between quasi-static and fluid-like granular regimes. Specifically, the short-time temporal averaging approach successfully isolates the macroscopic rotational motion from particle-scale noise. Finally, we highlight the potential for integrating physics-informed simulations with experimental measurements to enhance their accuracy and reliability and, hence, to improve the comprehension of complex granular flow phenomena.