Evaluation of density, viscosity, surface tension and CO2 solubility for single, binary and ternary aqueous solutions of MDEA, PZ and 12 common ILs by using artificial neural network (ANN) technique

Abstract

In this study, density, viscosity, surface tension and CO2 solubility for single, binary and ternary aqueous solutions of N-methyldiethanolamine (MDEA), piperazine (PZ) and 12 common ionic liquids (ILs) were predicted by applying artificial neural network (ANN) technique. The input data included operating temperature (293.15–373.15K) and pressure (0.177–3938.400kPa) in addition to the weight fractions of aqueous solutions of MDEA, PZ and ILs which were respectively in the range of (0.0–1.0), (0.0–0.09) and (0.0–1.0) as well as the molecular weight of ILs (148.18–505.00g/mol) and their acentric factors (0.325–1.261). More than 2600 experimental data points for density, viscosity, surface tension and CO2 solubility were collected from literature. By using the Levenberg-Marquardt back-propagation and tan-sigmoid as learning algorithm and transfer function, respectively, four ANN models were examined to treat these data. It was found that the best ANN architectures for predicting these properties were respectively (7:3:0:1), (7:6:0:1), (7:4:0:1) and (7:10:0:1). The calculated properties were compared with the corresponding experimental data which indicated a negligible error.