(147b) A Coarse-Grain CFD-DEM Approach Based on Particle Filtering for Simulating Fluidized Polydisperse Particles

Highly resolved simulations of circulating fluidized bed (CFB) risers at industrial scale are prohibitively expensive. A common approach to reduce the computational cost is to lump particles into parcels, each containing many particles. However, existing approaches are based on ad-hoc corrections and fail to converge to the underlying deterministic equations in the limit the parcel approaches a single particle. In this work, a rigorous formulation of the filtered Computational Fluid Dynamics-Discrete Element Method (CFD-DEM) equations are presented. While exact, the equations result in unclosed terms, notably a sub-filtered drag force, that require new models. The unclosed terms are informed by highly resolved CFD-DEM simulations of moderately dense gas-solid flows. We consider simulations of monodisperse and polydisperse Geldart-A particles in unbounded flows. Parcels are identified from the highly resolved simulations efficiently, in a consistent manner with the underlying formulation, using a KDTree algorithm. Collisions between the parcels are handled using soft-sphere approach that modifies the coefficient of restitution based on the number of particles per parcel. Variation in particle size and velocity within each parcel is quantified. The relative contribution of the sub-filtered drag is found to increase with number of particles per parcel, and depend on the local filtered volume fraction and Reynolds number. Symbolic regression is employed to arrive at close-formed algebraic models. The accuracy and speed-up of the proposed filtered CFD-DEM approach compared to the highly resolved simulations will be presented.