Predictive modelling of carbon dioxide solubility in aqueous polyamine blended sodium glycinate blends using thermodynamic and machine learning approach

Abstract

This study investigates the enhancement of CO ₂ solubility in aqueous sodium glycinate (SG) solutions blended with two polyamines: piperazine (PZ) and dipropylenetriamine (DPTA). Vapour–liquid equilibrium (VLE) data were experimentally measured using an equilibrium stirred cell at 313–333 K and CO ₂ partial pressures of 2–120 kPa. Six blend compositions were investigated, containing 5, 10, or 15 mass% PZ or DPTA combined with 25, 20, or 15 mass% SG, respectively, at a constant total amine concentration of 30 mass%. CO ₂ loading (mol CO ₂ /mol total amine) increased substantially with higher polyamine content, with the 15 mass% DPTA +15 mass% SG blend showing the highest capacity. Solvent density was correlated using excess molar volume models, yielding mean absolute errors of 0.0265% (PZ–SG) and 0.0258% (DPTA–SG). The VLE data were successfully fitted using modified Kent–Eisenberg models, with mean absolute errors (MAE) of 0.75% and 0.80%, respectively. Four machine learning models (XGBoost, ANN, random forest, SVR) were evaluated using a combined dataset of 660 experimental and literature data points. XGBoost showed the strongest predictive capability ( R ²  = 0.996), outperforming random forest ( R ²  = 0.99), ANN ( R ²  = 0.972) and SVR ( R ²  = 0.92). The isosteric heat of CO ₂ absorption, estimated via the Clausius–Clapeyron equation, ranged from 25 to 55 kJ/mol CO ₂ —lower than 30 wt.% MEA (~85 kJ/mol). These results demonstrate that polyamine‐activated SG solvents significantly enhance CO ₂ capture and that data‐driven models effectively predict CO ₂ solubility for solvent screening and process design.