Granular filters are traditionally used for water filtration and recently they are being extensively used in several chemical engineering applications. Computational fluid dynamics (CFD) simulations are a cost-effective tool for the design and development of

granular filters in applications such as fast pyrolysis of biomass for bio-oil production. The predictive capability of CFD simulations of granular filtration strongly depends on the equations governing the concentration of particulates and the model for the filtration rate. The primary objective of this work is to understand and investigate the filtration of inertial particulates in a granular filter, and develop high fidelity models using direct numerical simulations. Particle-resolved direct numerical simulation (PR-DNS) is a first-principles approach to develop accurate models for interphase momentum, energy, heat transfer in gas-solid flow and can be developed to study granular filtration. Another objective is to test these developed models in the CFD code ANSYS-FLUENT to simulate a full-scale moving bed granular filter.

A direct numerical simulation-Lagrangian particle tracking (DNS-LPT) approach has been developed to simulate moving-bed granular filtration. It is established that DNS-LPT simulations give numerically converged results. The penetration and single-collector efficiency obtained from DNS-LPT gives good match with published results.

The DNS-LPT results show that for inertial particles in a granular filter there is a significant nonzero mean slip between particles and fluid. A modified effective Stokes number that gives a good collapse of single-collector efficiency is obtained from DNS-LPT data. Using DNS-LPT simulations we developed a model for filter coefficient in terms of the modified effective Stokes number that can be used in CFD simulations.

An analytical framework for calculating filter efficiency of polydisperse particles in a granular bed is developed for cases where inertial impaction and interception are the principal filtration mechanisms. The developed framework can be used for both the Stokes flow and moderate Reynolds number. The results obtained from the analytical

framework give a good match with the DNS-LPT results.

The DNS-LPT approach has been used to study bouncing of particles from granule surface by implementing hard-sphere collision between particles and granules. The DNS-LPT results of filter efficiency is compared with the results obtained using laser-based experiments performed by collaborators. The DNS-LPT simulations for bouncing particles are used to develop a model for adhesion probability of inertial particles in a granular filter. In addition to the model development, the developed models are implemented and tested in the CFD code ANSYS-FLUENT to simulate a full-scale moving-bed granular

filter.