An algorithm for predicting the hydrodynamic and mass transfer parameters in agitated reactors

A large number of experimental data points (7374) obtained in our laboratory as well as from the literature, covering wide ranges of reactor geometry (reactor diameter and type, impeller diameter and gas distribution scheme), physicochemical properties (liquid and gas density and molecular weight, liquid viscosity and surface tension, diffusivity) and operating variables (superficial gas velocity, temperature, pressure, mixing speed, liquid height and mixtures) were used to develop empirical as well as back-propagation neural network (BPNN) correlations in order to predict the hydrodynamic and mass transfer parameters in gas-liquid agitated reactors (ARs). The empirical and BPNN correlations developed were incorporated in a calculation algorithm for predicting the gas holdup (epsilon(G)), volumetric mass transfer coefficients (kappa(L)a), Sauter mean bubble diameter (d(s)), gas-liquid interfacial area (a) and liquid-side mass transfer coefficient (kappa(L)) in ARs, operating in surface-aeration, gas-inducing and gas-sparging modes. The algorithm was used to predict the effects of liquid viscosity and hydrogen mole fraction in the feed gas (H-2 + N-2) on the hydrodynamic and mass transfer parameters for the soybean oil hydrogenation process conducted in a large-scale gas-sparging agitated reactor (7000 kg soybean oil capacity). The predictions showed that increasing the liquid-phase viscosity, mimicking the evolution of the hydrogenation of soybean oil in a batch reactor, decreased epsilon(G) and increased d(s), resulting in a decrease of a. The decrease of the gas holdup with increasing the liquid-phase viscosity was related to the increase of gas bubble coalescence in the reactor. Increasing liquid-phase viscosity, however, decreased kappa(L) as well as kappa(L)a values for both H-2 and N-2 within the range H-2 mole fraction (0-1) used. This kappa(L) behavior indicated that the effect of viscosity on kappa(L) is more significant than that of d(s), since kappa(L) was reported to be proportional to d(s). The predictions also showed that increasing the H-2 mole fraction in the feed to the reactor decreased epsilon(G) and increased d(s), resulting in a decrease of a and an increase of kappa(L) as well as kappa(L)a for both H-2 and N-2 within the range of liquid-phase viscosity used (0.0023-0.0047 Pa s). The decrease of the gas holdup with increasing the H-2 mole fraction in the feed gas was attributed to the decrease of the density (momentum) of the gas mixture. The increase of kappa(L) values with increasing the H-2 mole fraction in the feed gas was related to the increase of d(s). The predicted kappa(L)a values indicated that the mass transfer behavior in the large-scale gas-sparging reactor proposed for soybean oil hydrogenation was controlled by the mass transfer coefficient, kappa(L). Also, under similar conditions, kappa(L)a values for H-2 in soybean oil when using the gaseous mixture (H-2 + N-2) were lower than those obtained for H-2 (as a single-component); and kappa(L) values for H-2 were consistently greater than those of N-2 within the ranges of the operating conditions used in the simulation. (c) 2005 Elsevier B.V. All rights reserved.