Figure 2. Box plots of local cluster characteristics. The left column corresponds to 1188 datasets from a fast fluidized bed 16,17, while the right column corresponds to 378 datasets from a turbulent fluidized bed20.
To understand the relative influence of the various variables, Figure 3 displays the random forest analysis results. The variables include particle properties (namely, dave andρp ), riser position (namely, radial position normalized with respect to the riser radius (r/R ), and axial position normalized with respect to the riser height (h/H )) and operating condition (namely, Ug andGs ). Regarding cluster probability, the similarity between the fast and turbulent fluidized beds lies inh/H being the most dominant and Ug being least dominant in both fluidization regimes. For the cluster duration and frequency, the relative influences were different for the two fluidized beds. For the fast fluidized bed, four of the six variables (namely, dave , ρp ,r/R, and Gs ) were approximately similar in influencing cluster duration, while dave was the most influential on cluster frequency. The lack of a more distinctively dominant variable among the six evaluated here is because the single most influential factor has been reported earlier to be whether the particle system was monodisperse or not (Figure 4) 26.
As for cluster duration and frequency in the turbulent fluidized bed,h/H was by far the most dominant, which is tied to the steep particle concentration drop with height typical of this regime2, while r/R was the least important, which is consistent with the relatively flat radial profiles of chaotic parameters reported earlier 11. The different variables governing the cluster duration and frequency suggest that the underlying mechanisms leading to cluster formation were different in the different regimes.