On grid-independency of CFD-DEM simulations of cluster-induced turbulence

Abstract

We present a comparative study on the influence of three different particle (data) mapping methods on various one-point and two-point cluster-induced turbulence (CIT) statistics in an unbounded fluidization system. These methods include the particle centroid method (PCM), the divided particle volume method (DPVM), and a newly proposed implementation of Gaussian kernel method referred to as GaussFace method. In PCM method, the entire particle data is allocated to the cell, in which the particle centroid is located. However, in DPVM, the particle data is subdivided into smaller volumes using a satellite point method, allowing the distribution of particle properties among the cells associated with the satellite points. We performed simulations of dilute and dense systems at grid sizes between 2.22$d_p-8.88d_p$ and two smoothing characteristic sizes. Our results reveal that solely using GaussFace method, which is based on distributing particle data to surrounding cells employing a Gaussian kernel, leads to a smoother particle field compared to the PCM and DPVM. Furthermore, we find that grid-independent results can only be obtained by GaussFace method together with subsequent smoothing. We also observe that the grid-independence criterion identified for the dilute system is also valid for the dense system. In addition, comparing two different methods of separating the spatially correlated and uncorrelated particle velocity components reveals that the commonly adopted fixed filter approach grossly underestimates the granular temperature in the dilute regions. The findings of our study can serve as a reference for obtaining high-resolution CFD-DEM results, which, in turn, can be used to develop coarse-grid models.