Experimental measurement and thermodynamic/intelligent modeling of yellow 2 G solubility in supercritical carbon dioxide with/without co-solvent

Abstract

In this study, the solubility of an azo dye, Yellow 2 G, was investigated in supercritical CO2, both with and without co-solvents (methanol and dimethyl sulfoxide (DMSO)), spanning temperatures from 308 to 338 K and pressures from 10 to 34 MPa. The supercritical solubility of Yellow 2 G ranged from 0.052×10−6 to 0.559×10−6 mole fractions, while in the presence of methanol and DMSO, it varied from 0.5×10−6 to 5.92×10−6 and 0.46×10−6 to 4.6×10−6, respectively. Co-solvents significantly increased supercritical solubility, with over a 9-fold increase with methanol and more than a 7-fold increase with DMSO. Semi-empirical models, Peng-Robinson (PR), Soave Redlich Kwong (SRK), Perturbed-Chain Statistical Associating Fluid (PCP-SAFT), and quadrupolar cubic plus association theory (qCPA) equations of state, and a multilayer perceptron (MLP) neural network were used to assess the solubility data. Among the semi-empirical models, the Khansary model for the binary system and the Jouyban and Soltani-Mazloumi models for the ternary system demonstrated the highest conformity. Among models based on equations of state, the PCP-SAFT model demonstrates the highest accuracy in correlating solubility data for both binary and ternary systems. Notably, the accuracy of the SRK model in correlating solubility data for the binary system matches that of PCP-SAFT. The proposed MLP neural network provided highly accurate estimations of Yellow 2 G solubility in both binary and ternary systems, achieving an accuracy of over 99 %.