Effect of Lagrangian Time Scales and Reynolds Number on the Prediction of Droplet Penetration through Vertically Oriented Turbulent Pipe Flows

ZACHARY MILANI, Edgar Matida, Leonid Nichman, R. Timothy Patterson, Carleton University

     Abstract Number: 208
     Working Group: Aerosol Physics

Abstract
Aerosol deposition is a known bias in many common two-phase flow scenarios for example when pollen or dust accumulates in ventilation ducts, pharmaceutical aerosols are lost in the upper respiratory tract, and during the deposition of solid-phase combustion emissions inside of sampling systems. In these cases, numerical simulations can help to accurately predict aerosol deposition and losses. Using an eddy interaction model (EIM) to simulate aerosol deposition is attractive when compared to the time-consuming and computationally expensive direct numerical simulation (DNS) and large eddy simulation (LES) methods. There are, however, still challenges associated with using EIMs e.g., turbulence anisotropy treatment near pipe walls and the choice of proper Lagrangian time scales. To investigate these, a fundamental experiment studying droplet penetration in a vertically oriented turbulent pipe flow was simulated using a Lagrangian standard random-walk EIM code that was developed in house. Two Reynolds numbers (ReD = 5,300 and 19,000) were simulated using flow statistics (i.e., mean velocities, root mean square fluctuation velocities, and turbulence dissipation rate) that were obtained from DNS data found in the literature. These one-way coupled simulations involved the release of thousands of monodisperse droplets ranging from 1 to 27 µm into the computational domain. Improvements to droplet penetration estimates were achieved in previous work (flow simulations of ReD = 11,700 and 37,700) by introducing new eddy time scales based on the ratio of Lagrangian and moving Eulerian time scales, and the dissipation and velocity fluctuations normal to the walls. These time scales are further developed and systematically tested across a wider range of scenarios in this work. Additional random-walk models are implemented and compared to the standard model used in the original work.