Effect of grid type and refinement method on CFD-DEM solution trend with grid size

Abstract

Identification and quantification of numerical error is complicated when Computational Fluid Dynamics (CFD) is coupled with Discrete Element Method (DEM) to model particle-fluid interactions. The presence of solid particles in the model disrupts the typical monotonic convergence of CFD with grid size, due to the changing distribution of calculated void fraction. Quantification of numerical error is crucial to the development of CFD-DEM as an analysis and design tool, but investigations of this error have been limited. Guidelines are needed for model design to permit the quantification of numerical error, and subsequent grid refinement to minimize error. The present study investigates the solution trend and relative numerical error for different computational grid types over a range of grid sizes. Simulations of flow through a stationary particle bed are conducted using computational grids for columns with circular and square cross-sections. A hybrid rectangular and cylindrical hex mesh is used for the circular cross-section column, while an initially cubic hex mesh is applied to the square cross-section column. Non-uniform, uniform, and one-dimensional refinements are individually performed on grids for both types of columns. The grid refinement study procedure by Celik et al. (2008) is applied to the solution trend of simulations using each uniform refinement method, and the resulting average grid convergence index (GCI) is shown in several cases to be a good predictor of the relative numerical error.