Modeling contaminant transport in industrial ventilation applications is difficult because the flow fields encountered are time-dependent, three-dimensional, and turbulent. For contaminants comprised of dense particles or droplets, previous studies have shown that interactions between particles and turbulent eddies results in locally large concentrations, as much as 20-40 times the global mean value. These interactions are often established by the presence of a worker near a local exhaust device. Consequently, suspended particulates may pose significantly greater health risk to workers than has been previously thought. Computational models present a powerful tool for assessing the risk posed by suspended particles. Unfortunately, current models are unable to accurately account for the complex interactions between particles and turbulence and, therefore, new approaches are needed. Because of the development of dynamic subgrid-scale turbulence models, large eddy simulation (LES) has emerged as a powerful approach for predicting complex flows, but without requiring ad hoc calibration of model constants. Consistent with our long-term objective in developing of reliable predictive methods for modeling transport in industrial ventilation applications, the purpose of this investigation was to develop and validate new computational models based on LES for predicting particulate transport. The computational technique was based on LES of the filtered Navier Stokes equations. The equations governing conservation of mass and momentum for an imcompressible fluid were solved using the fractional step method on a staggered grid. Second-order central differences were used to discretize spatial derivatives, the discrete system was time-advanced using mixed explicit/implicit scheme (second-order Adams-Bashforth for the convective terms and Crank-Nicholson for the linear terms). The subgrid-scale stress in the momentum equations was closed using the Smagorinsky eddy viscosity model. The coefficient in the Smagorinsky model was calculated using the dynamic procedures outlined in Germano et al. (1991) and Meneveau et al. (1996). Following development of the solution method for the carrier flow, particles tracking capabilities were incorporated into the computer code. The equation of motion used in this work included nonlinear drag and gravitational drift and is appropriate to describing the motion of a spherical particle in a nonuniform, gas-phase turbulent carrier flow. A series of interpolatation schemes were tested, and it was found that fourth-order accurate Lagrange polynomials accurately recovered fluid velocities at particle positions. LES predictions of several particle-laden flows were evaluated using both experimental measurements and results from direct numerical simulation. A wide array of test problems were considered, including prediction of particle deposition onto the wall of a turbulent boundary layer, particle transport in fully-developed channel flow, and prediction of a high Reynolds number particle-laden mixing layer. Reynolds numbers in the carrier phase varied from 3,300 in the channel (based on centerline velocity and channel half-width) to over 20,000 in the mixing layer (based on the velocity difference across the layer and vorticity thickness), a range covering the Reynolds numbers encountered in ventilation applications. Aerodynamic diameters considered in the simulations were in the inspirable range. The actual values were chosen the same as in the available experimental measurements, ranging for 30 to 130 ~m. LES predictions compare favorably with experiments, predicting the enhanced particle fluctuating velocities with increasing response time in boundary layers and the increased lateral dispersion in the mixing layer. Significantly, LES also predicts the complex structural features of particle laden turbulence, i.e. preferential concentration of particles into regions of high strain rate. Both qualitative and quantitative measures of preferential concentration are in good agreement with experimental measurements and results for direct numerical simulation.
Department of Mechanical Engineering, University of Vermont, Burlington, Vermont 05405