| Abstract
| - Interfacial areas (aw) and volumetric mass-transfer coefficients (kLaw, KLaw, kGaw, and KGaw)required for randomly packed tower design were gathered from the literature to generate aworking database including over 3780 measurements. A set of artificial neural networkcorrelations for the gas−liquid interfacial area and the pure local mass-transfer coefficients wasproposed. Thus, the gas−liquid interfacial area and the pure local mass-transfer coefficients(kγ, where γ = G or L) were extracted using a reconciliation procedure which combined actuallymeasured interfacial areas with pseudo interfacial areas inferred from the actually measuredvolumetric mass-transfer coefficients. The neural network weights of the two aw and kγcorrelations were adjusted using a least-squares composite criterion simultaneously over thefive mass-transfer parameters. The first correlation representing the gas−liquid interfacial area[aw/aT = f(ReL,FrL,EoL,χ,K)] yielded an average absolute relative error (AARE) of 22.5% for the325 measurements available. The second one, representing either kG or kL, was also implementedusing the following structure: Shγ = f(Reγ,Frγ,Scγ,χ). The combination of both correlationpredictions (i.e., kγaw) yielded an AARE of 24.4% for the local and global volumetric mass-transfercoefficients (3455 data).
|
