Prediction of biphasic separation in CO₂ absorption using a molecular surface information-based machine learning...

Area of Science:

• Chemical Engineering
• Environmental Science
• Computational Chemistry

Background:

• Global warming necessitates advanced carbon dioxide (CO2) capture technologies.
• Biphasic absorbents offer a promising avenue for energy-efficient CO2 capture processes.
• These systems, typically mixed solvents of alkanolamine and organic solvents, exhibit complex phase behavior requiring extensive experimental screening.

Purpose of the Study:

• To develop a predictive method for determining the phase states of mixed-solvent biphasic absorbents.
• To assess the efficacy of quantum calculations and machine learning models in predicting phase behavior.
• To identify key molecular descriptors influencing phase transitions during CO2 absorption.

Main Methods:

• Utilized quantum calculations and machine learning algorithms (random forest, logistic regression, support vector machine).
• Trained models on a dataset of 61 mixed-solvent absorbents (alkanolamine/glycol ether or alcohol).
• Evaluated model performance in predicting phase states before and after CO2 absorption.

Main Results:

• Machine learning models achieved over 90% accuracy in predicting the phase states of mixed-solvent absorbents.
• The models successfully predicted phase states both before and after CO2 absorption.
• Analysis revealed that the molecular surface charge of amine species is a critical factor in determining phase behavior during CO2 absorption.

Conclusions:

• A robust predictive method for biphasic absorbent phase states was successfully developed using machine learning.
• This approach significantly reduces the need for extensive experimental screening in developing CO2 capture solvents.
• Understanding the role of amine molecular surface charge can guide the design of more efficient CO2 capture systems.