Abstract

 The Pseudo-Direct Numerical Simulation (P-DNS) method is a numerical approach to solve multiscale fluid dynamics phenomena overcoming the computational burden associated with direct numerical resolution of the governing equations. In P-DNS, both the coarse scale and the fine scale are solved numerically. The most expensive part of the computations, the solutions of the fine scale, are parametrized to be performed offline and store their results in dimensionless databases. This latter allows us to construct synthetic models to emulate the fine scale behavior when coarse scale solutions are computed. In this work, a multiphase representative volume element (RVE) is designed for the specific case of a dispersed flow of heavy particles in air. Making use of high performance computing facilities, a set of numerical experiments for a wide range of volume fractions, particle distribution sizes, and inertial forces introduced as shear loads are carried out. Quantitative results of the statistically stationary turbulent state are obtained. This brings insights into the turbulence modulation phenomenon, i.e. the change of the kinetic energy of the carrier flow due to the presence of the particles. The database of RVE results is stored in an artificial neural network to facilitate its use in general flow solvers via a fast interpolation technique. Global case studies of particulate flows are analyzed, highlighting the differences and similarities found using the fine-scale model developed vis-a-vis a standard turbulence modeling.