Accelerating amine-based CO₂ capture with machine learning: From molecular screening to process optimization.

Area of Science:

• Chemical Engineering
• Materials Science
• Artificial Intelligence

Background:

• Amine-based CO2 capture is a mature technology for industrial carbon reduction.
• Vast chemical spaces and complex interactions challenge traditional experimental methods.
• Machine learning offers novel strategies to overcome these limitations.

Purpose of the Study:

• To explore machine learning applications for optimizing amine-based CO2 capture systems.
• To enhance precision, reduce regeneration energy, and improve material design.
• To demonstrate industrial applicability and cost-effectiveness.

Main Methods:

• Ensemble learning algorithms for liquid amine systems.
• Interpretable models to identify key molecular descriptors.
• Differential descriptor methods for solid amine systems.
• Virtual screening of large chemical databases.
• Mechanistic analysis of adsorption properties.

Main Results:

• Precision improved to <0.93% in liquid systems; 34% regeneration energy reduction achieved.
• Solid amine model performance improved from R²=0.5102 to 0.79.
• Identified 11% of screened candidates with superior CO2 binding.
• 2642 synthesizable high-performance molecules identified.
• Physical properties identified as dominant factors in adsorption.

Conclusions:

• Machine learning revolutionizes amine-based CO2 capture design, moving towards intelligent paradigms.
• Significant cost reductions (35.76%) and profit improvements (15-25%) demonstrated in industrial applications.
• Physics-constrained algorithms and unified frameworks are needed for further development and translation.