CO2 absorption behavior in dual-amine blends of primary and tertiary amine: machine learning, NMR analysis, and performance evaluation

Accurately predicting CO2 equilibrium solubility is essential for optimizing amine-based sorbents in carbon capture processes. However, traditional experimental approaches require extensive testing, making data collection time-consuming and often incomplete. In this study, we address this challenge by developing six machine learning models to predict CO2 solubility in three dual-amine blends (MEA + MDEA, MEA + DEEA, and MEA + DMEA), significantly reducing the reliance on labor-intensive experiments. The PSO-BPNN model achieved the highest accuracy, with errors below 2.2 %, demonstrating the potential of machine learning to provide reliable solubility data without exhaustive experimental measurements. Additionally, we experimentally evaluated the CO2 absorption and desorption performance of MEA-based blends, complemented by 13C NMR spectroscopy to elucidate reaction mechanisms. Among the tested systems, the 2 M MEA + 3 M DEEA blend exhibited the highest CO2 loading, fastest desorption rate, and lowest heat duty compared to 5 M MEA, underscoring its potential as a superior sorbent for energy-efficient CO2 capture.