Experimental and predictive modeling of density of AMP–DA2MP and MDEA–DA2MP bi-solvent amine blends for CO2 capture

The density of CO₂-unloaded and CO₂-loaded aqueous solutions of 2-amino-2-methyl-1-propanol (AMP) + 1,5-diamino-2-methylpentane (DA2MP) and methyldiethanolamine (MDEA) + DA2MP bi-solvent blends were experimentally investigated under varying amine concentrations (2–5 M), temperatures (25–50 °C), and CO₂ loadings (0–1.0 mol CO₂/mol amine). To predict the experimental results, 4 modeling approaches were applied: artificial neural networks (ANN), response surface methodology (RSM), Nwaoha’s nonlinear model, and Nwaoha et al.’s modified nonlinear model. Results showed that higher CO₂ loading increased density, while higher temperature and amine blend concentration reduced density. Predictive modeling demonstrated that the ANN exhibited the highest accuracy, achieving a maximum absolute average deviation (%AAD) of 0.38%, a high coefficient of determination (R²) of 0.9999, and the highest mean squared error (MSE) of 1.6E-5. Meanwhile, the modified nonlinear model had a maximum %AAD of 0.78%, the highest R² of 0.9697, and the highest MSE value of 1.05E-4. The nonlinear model achieved maximum %AAD of 0.9155%, highest R² of 0.9309, and highest MSE value of 1.15E-4. The RSM showed maximum %AAD of 2.1401%, highest R² of 0.9782, and MSE of 4.97E-4. The predictive performance of the models follows this trend: ANN > modified nonlinear model > nonlinear model > RSM. Notably, the modified nonlinear model, despite requiring significantly less computational effort and expertise compared to ANN and RSM, proved to be highly effective for accurately predicting density.