Figure 11. Random Forest
out-of-bag permuted predictor importance estimates for local segregation
data in terms of percentage mass of the more massive species for binary
mixtures and broad PSDs. The fast fluidized bed had 324 datasets14, while the turbulent fluidized bed had 190 datasets21.
While the mixture compositions were fixed in both the binary mixtures
and broad PSDs for the fast fluidized bed, the percentage mass of the
binary mixtures (namely, 25%, 50% and 75% of fine, which refers to
the narrow PSD glass) and also the widths of the broad PSDs
(σ /dave = 10%, 25%, 40%, and 65%) were
varied in the turbulent bed. Figure 12 shows the random forest analysis
of the relative influence of the variables for the binary mixtures and
broad PSDs. Although r/R was the most influential in segregation
in the fast fluidized bed (Figure 11) 25, it had
little influence on both non-monodisperse particle systems (i.e., binary
mixtures and broad PSDs) in the turbulent bed (Figure 12). Figure 12
shows clearly that the relative influences for the two types of
non-monodisperse mixtures were different, which agrees with earlier
studies that also indicated different behaviors between binary mixtures
and broad PSDs 14,16,20,21,37-39. For binary mixtures,
the composition of the constituents was approximately as important as
the overall mass flux (Gs ) in influencing
segregation. On the other hand, for the broad PSDs, h/H was the
most influential, followed by σ /dave andGs , which were equally influential.
Interestingly, in both cases, the extent of polydispersity
(specifically, represented by the percent fines for the binary mixtures
and σ /dave for the broad PSDs) andGs had similar influences on segregation.