Development of direct numerical simulations for multiphase fluid flow benchmarking

This study investigates the dynamics of multiphase turbulent flows and various coupling methods between the solid particles and the liquid at different particle concentrations and Stokes numbers. Simulations are performed altering the treatment of the particle phase to assess how the coupling method used affects model predictions. Direct numerical simulation (DNS) of a single-phase turbulent channel flow at a shear Reynolds number of 300 is performed and coupled to a Lagrangian particle tracker (LPT) to simulate the particle phase. These techniques are first validated against previous DNS-based results with good agreement obtained. A series of deterministic simulations using one-, two- and four-way coupling between the fluid and the particles is then performed at two particle concentrations and two Stokes numbers. Differences in the velocity, turbulence and drag forces on both the fluid and the particles, caused by changes in the latter variables, are assessed and differences explained in terms of the effects of forces acting on the particles, particle collisions and turbophoresis. To reduce model run times, a stochastic collision metric is added to the LPT and tested for its effectiveness. Both fictional particle and direct simulation Monte Carlo techniques are investigated, with the latter producing superior results after some modification, and in all but one case is found to reduce model run times, albeit with some loss in accuracy. Finally, the agglomeration of particles is considered using both deterministic and stochastic approaches. The stochastic technique requires further development where agglomeration is implemented, and although successful the rate of particle collision remains high, with agglomerates being formed faster than turbophoretic effects can disperse them. Overall, the study improves our understanding of particle-laden flows, provides benchmark solutions against which more pragmatic predictive approaches can be assessed, and represents a significant step in the development of more computationally efficient stochastic techniques.