Machine learning-assisted crystal engineering of a zeolite.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Faujasite (FAU) zeolites are crucial as catalysts and adsorbents.
• Controlling synthesis conditions is key to tailoring zeolite microstructures and properties.
• Machine learning (ML) offers a powerful tool for understanding complex material synthesis relationships.

Purpose of the Study:

• To demonstrate the utility of ML algorithms, specifically Geometric Harmonics, in predicting zeolite properties.
• To identify optimal synthesis conditions for enhancing the Si/Al ratio in FAU zeolites.
• To compare the performance of ML approaches against traditional methods like Neural Networks and Gaussian Process Regression.

Main Methods:

• Application of Machine Learning algorithms (Geometric Harmonics, Neural Networks, Gaussian Process Regression) to model zeolite synthesis.
• Analysis of input parameters (composition, conditions) and output characteristics (microstructure).
• Experimental synthesis and characterization of FAU zeolites with varying Si/Al ratios.

Main Results:

• ML successfully captured the relationship between synthesis inputs and FAU zeolite microstructural outputs.
• Identified synthesis conditions to achieve a record Si/Al ratio of 3.5 in FAU zeolite via direct, organic-free synthesis.
• Reduced Na2O content was identified as critical for high Si/Al ratios in FAU materials.

Conclusions:

• ML provides valuable insights into zeolite synthesis, facilitating the design of materials with targeted properties.
• The synthesized high Si/Al ratio FAU zeolite (Si/Al = 3.5) yields a superior acid catalyst for propane cracking and dehydrogenation.
• This work establishes a new benchmark for FAU zeolite synthesis and catalytic performance.