Machine-learning-accelerated multimodal characterization and multiobjective design optimization of natural porous...

Area of Science:

• Materials Science
• Computational Chemistry
• Green Chemistry

Background:

• Natural porous materials, including nanoporous clays, are cost-effective adsorbents and catalysts.
• Material performance hinges on pore morphology and surface activity, characterized by properties like surface area, pore volume, and pH.
• Experimental characterization is often costly and time-consuming, especially when optimizing materials for specific applications.

Purpose of the Study:

• To apply tree-based machine learning methods for accelerated characterization of natural porous materials.
• To develop predictive models for material properties based on experimental data.
• To identify key factors influencing material properties and optimize processing conditions.

Main Methods:

• Training tree-based machine learning models on experimental datasets of natural porous materials.
• Utilizing feature importance analysis to identify critical material properties and processing parameters.
• Applying models for high-throughput exploration of processing parameter-property correlations and multiobjective optimization.

Main Results:

• Developed models accurately predict experimental characterization outcomes (R² from 0.78 to 0.99).
• Identified key factors influencing material properties through feature importance analysis.
• Pinpointed optimal processing conditions for clays in acid catalysis, leading to a synthesized material with improved acid character.

Conclusions:

• Machine learning offers a powerful, accelerated approach to characterizing and optimizing natural porous materials.
• The developed models facilitate efficient exploration of material processing-property relationships.
• Optimized clay materials demonstrate enhanced performance in acid-catalyzed degradation, achieving 79% removal of chlorophyll-a.