Prediction of CO2 Mass Transfer Flux in Aqueous Amine Solutions Using Artificial Neural Networks

In the present research, neural networks were applied to predict mass transfer flux of CO2 in aqueous amine solutions. Buckingham π theorem was used to determine the effective dimensionless parameters on CO2 mass transfer flux in reactive separation processes. The dimensionless parameters including CO2 loading, the ratio of CO2 diffusion coefficient of gas to a liquid, the ratio of the CO2 partial pressure to the total pressure, the ratio of film thickness of gas to liquid and film parameter as input variables and mass transfer flux of CO2 as output variables were in the modeling. A multilayer perceptron network was used in the prediction of CO2 mass transfer flux. As a case study, experimental data of CO2 absorption into Piperazine solutions were used in the learning, testing, and evaluating steps of the multilayer perceptron. The optimal structure of the multilayer perceptron contains 21 and 17 neurons in two hidden layers. The predicting results of the network indicated that the mean square error for mass transfer flux was 4.48%. In addition, the results of the multilayer perceptron were compared with the predictions of other researchers’ results. The findings revealed that the artificial neural network computes the mass transfer flux of CO2 more accurately and more quickly.