Binary mixtures of biomass and inert components in fluidized beds: Experimental and neural network exploration

[Display omitted] •Hydrodynamic of multicomponent particle beds involving biomass is investigated.•Conventional pressure drop experimental methods and ANN techniques are adopted.•ANNs demonstrated good performances in prediction of both outputs.•Biomass volume fraction significantly affects the complete fluidization velocity.•Minimum fluidization velocity is determined by inert density and biomass sphericity. In light of the little understanding of the hydrodynamics of multicomponent particle beds involving biomass, a detailed investigation has been performed, which combines well-known experimental and theoretical approaches, relying, respectively, on conventional pressure drop methods and artificial neural network (ANN) techniques. Specific research tasks related to this research includes: i. to experimentally investigate by means of visual observation the mixing and segregation behavior of selected binary mixtures by varying the biomass size and shape as well as the properties (size and density) of the granular solids in cold flow experiments; ii. to carry out a systematic experimental investigation on the effect of the biomass weight and volume fractions on the characteristic velocities (i.e., complete fluidization velocities and minimum slugging velocity) of the investigated binary mixtures in order to select the critical weight fraction of biomass in the mixtures beyond which the fluidization properties deteriorate (e.g., channelling, segregation, slugging); iii. to analyze the results obtained in about 80 cold flow experiments by means of ANN techniques in order to scrutinize the key factors that influence the behavior and the characteristic properties of binary mixtures. Experimental results suggest that the bed components’ density difference prevails over the size difference in determining the mixing/segregation behavior of binary fluidized bed, whereas the velocities of minimum and complete fluidization increased with the increase of the biomass weight fraction in the bed. The training of ANNs demonstrated good performances for both outputs (Umf and Ucf), in particular, best predictions have been obtained for Umf with a MAPE