Deep Learning-Assisted Elucidation of Structure-Performance Relationships in MOF-NH₃ Adsorption Refrigeration Working...

Area of Science:

• Materials Science
• Chemical Engineering
• Sustainable Energy

Background:

• Adsorption refrigeration systems utilizing metal-organic frameworks (MOFs) and ammonia offer a carbon-neutral solution for converting low-grade heat into cooling.
• The precise adsorption mechanisms of MOFs under saturated ammonia conditions are not fully understood due to limited data.

Purpose of the Study:

• To establish a comprehensive database of MOF-ammonia adsorption properties.
• To develop predictive models for MOF performance in adsorption refrigeration.
• To identify key MOF structural descriptors influencing adsorption capacity.

Main Methods:

• A large-scale database of 9835 MOF structures was created using high-throughput grand canonical Monte Carlo simulations.
• Machine learning and deep learning models, including Convolutional Neural Networks (CNNs), were trained using 42 MOF parameters.
• SHAP (SHapley Additive exPlanations) values were used to analyze feature importance and quantify parameter effects.

Main Results:

• The CNN model achieved the highest prediction accuracy (R² = 0.880 ± 0.030) for cyclic adsorption capacity.
• Volume-related parameters, surface area, and pore size were identified as critical for MOF performance.
• Specific pore volume was found to be a key descriptor, capturing steric and adsorption potential effects.

Conclusions:

• This data-driven approach enhances the understanding of MOF-ammonia adsorption mechanisms.
• The study provides quantitative insights into optimal MOF structural parameters for adsorption refrigeration.
• Findings guide the rational design of advanced materials for next-generation carbon-neutral cooling technologies.