Clusters have a significant impact on gas–solid hydrodynamics, thus have garnered sustained research efforts for several decades. Recently, the image processing technology has been popularly used in characterizing clusters due to its non-invasive nature and high spatiotemporal resolution. This study introduced an innovative method for cluster identification by utilizing non-local means filtering and the maximum between-cluster variance. An exhaustive evaluation of image processing-based identification techniques was conducted using the flow data from CFD-DEM simulations. The approach can accurately reproduce the cluster dynamics throughout the entire bed without the need of manual adjustments, whereas traditional global threshold methods often fall short in reliably predicting clusters with low concentration or larger size. At lower gas velocities, there is a clear correlation between cluster characteristic parameters, while this type of relationship diminishes at higher gas velocities. Furthermore, clusters at different horizontal positions exhibit distinct morphologies and movement patterns.
Read full abstract