Abstract:Particle collisions driven by turbulent fluctuations play a key role in such diverse problems as cloud formation, aerosol powder manufacturing and inhalation drug therapy to name a few. In all of these examples (and many others) turbulent fluctuations increase the rate of collisions relative to the background collision rate driven by Brownian motion. Furthermore, turbulence can spontaneously generate very large fluctuations in the particle concentration field. This “clustering” is caused by the inertial mismatch between the heavy particles and the lighter surrounding gas; vortices in the flow “centrifuge” the heavier particles out of vortex cores and into the straining regions that lie in between the vortices. Because collision is a binary process, concentration fluctuations further enhance the turbulent coagulation rate by as much as two orders of magnitude. An effect of this size must be accounted for in a rational model of turbulent coagulation. Sundaram & Collins (J. Fluid Mech. 1997) showed that the radial distribution function (RDF) of the particle population, evaluated at contact, precisely corrects the collision kernel for clustering. Subsequent work has explored the dependence of the RDF on the system parameters (e.g., particle size, concentration, response time and Reynolds number) using direct numerical simulations. These results have improved our understanding and ability to predict the effect of the first three parameters; however, owing to the limited range of Reynolds number that can be reached in a numerical simulation, questions remain over the scaling of the RDF with Reynolds number. This is a critical issue for high-Reynolds-number applications such as cloud physics, where values of the Reynolds number can be 1–2 orders of magnitude greater than can be simulated. We will present our highest Reynolds number simulations to date and show our attempts to resolve this issue. Recently, the ability to measure three-dimensional particle positions using holography has been realized (e.g., Meng & Pu, J. Opt. Soc. Am. 2003). With holography, the optical image that is produced contains fringes that, upon inverting the laser, reproduce the original image in three dimensions. The hologram can then be scanned using a digital camera to obtain the particle positions. An important consideration with this study is the need to differentiate individual particles. We developed a search algorithm that locates particle centers, even in the presence of optical aberations and speckle noise. The algorithm has been used to obtain the first experimental RDF measurements to date. Thus far we see good agreement between the experimentally obtained RDF and the simulations. Besides validating the simulations, experiments can span a much broader range of Reynolds numbers, providing critical data that may help resolve the open questions associated with this parameter.

Investigating the influence of particle distribution on force and torque statistics using hierarchical machine learning

A New Particle Generation Method for Arbitrary 2D Geometries in SPH Modeling

Turbulent Coagulation of Aerosol Particles: New Insights from Direct Numerical Simulations and Holographic Imaging Experiments

Flow past a random array of statistically homogeneously distributed stationary Platonic polyhedrons: Data analysis, Probability maps and Deep Learning models

Particle clustering in turbulence: Prediction of spatial and statistical properties with deep learning

Microstructure-based prediction of hydrodynamic forces in stationary particle assemblies

Physics-informed neural network for modelling force and torque fluctuations in a random array of bidisperse spheres

Physics informed machine learning with Smoothed Particle Hydrodynamics: Hierarchy of reduced Lagrangian models of turbulence

Learning locally dominant force balances in active particle systems

Machine learning for physics-informed generation of dispersed multiphase flow using generative adversarial networks

Adaptive In-Focus Particle Detection and Segmentation in Holographic 3d Image with Mechanism-Guided Machine Learning

High-resolution fluid–particle interactions: a machine learning approach

Prediction of force chains for dense granular flows using machine learning approach

Hierarchical Self-Organization in Fixed-Magnetization Particle Systems

Physics-guided Design and Learning of Neural Networks for Predicting Drag Force on Particle Suspensions in Moving Fluids

Physics-Guided Deep Learning for Drag Force Prediction in Dense Fluid-Particulate Systems

Subspace Graph Physics: Real-Time Rigid Body-Driven Granular Flow Simulation

A systematic approach for numerical research of realistic shaped particle-fluid interactions

Inferring Latent States and Refining Force Estimates via Hierarchical Dirichlet Process Modeling in Single Particle Tracking Experiments

Prospects of federated machine learning in fluid dynamics

Neural SPH: Improved Neural Modeling of Lagrangian Fluid Dynamics