A Neural Network Method for Characterizing Particle Clusters in Gas–Solid Fluidized Beds

Cluster quantification in fluidized beds is notoriously difficult and heavily depends on empirical correlations with poor generalization ability. In this study, a simple and general approach was presented to characterize local clusters using machine learning. First the adaptive Otsu’s threshold was used to identify clusters from the flow field generated by CFD-DEM simulation, then the information regarding cluster characteristics and local hydrodynamics were extracted to generate a data set for machine learning. The Random Forest results indicate that pressure drop is the most dominant influence on cluster size, while solids loading ratio and gas velocity influence cluster velocity significantly. The effects of other local hydrodynamic parameters such as solid concentration and Reynolds number can not be ignored either. Accordingly, a Deep Neural Network was trained to predict cluster characteristics based on the accessible mesoscale and macroscopic flow parameters. Good predictive capability was achieved without a mechanistic understanding of cluster dynamics.