Machine Learning Guided Synthesis of Flash Graphene.

Area of Science:

• Nanoscience and Materials Science
• Computational Chemistry
• Chemical Engineering

Background:

• Flash Joule heating enables synthesis of nanomaterials like graphene from waste materials.
• Understanding the complex variables governing nanocrystal formation in this process is crucial but poorly understood.

Purpose of the Study:

• To explore factors driving amorphous carbon to graphene nanocrystal transformation using machine learning.
• To identify key parameters for optimizing graphene synthesis via flash Joule heating.

Main Methods:

• Construction of machine learning (ML) models, including XGBoost regression, to analyze flash Joule heating synthesis.
• Utilizing feature importance assays, decision trees, and partial dependence analyses to interpret ML model outputs.
• Application of Bayesian meta-learning algorithms for automated improvement of bulk crystallinity.

Main Results:

• XGBoost model achieved a high accuracy (r² = 0.8051 ± 0.054) in predicting crystallinity.
• Identified starting material properties and stochastic current fluctuations as critical factors.
• Charge and current density emerged as key predictors, suggesting a shift in reaction kinetics.

Conclusions:

• Machine learning provides powerful insights into complex nanomanufacturing processes like flash Joule heating.
• This study demonstrates ML's utility in optimizing the synthesis of 2D crystals with desired properties.
• The findings pave the way for more efficient and controlled production of graphene and other nanomaterials.