Machine Learning Prediction of Solvent-Assisted Depolymerization in Epoxy Covalent Adaptable Networks

Area of Science:

• Polymer Science and Engineering
• Materials Chemistry
• Computational Materials Science

Background:

• Thermosetting polymers pose recycling challenges due to their cross-linked structures.
• Sustainable waste management necessitates predictive models for thermoset depolymerization.
• Solvent-assisted depolymerization is a key strategy for breaking down thermoset networks.

Purpose of the Study:

• To develop a machine learning framework for modeling thermoset depolymerization.
• To predict the depolymerization behavior of covalent adaptable networks (CANs).
• To guide solvent selection and process optimization for thermoset recycling.

Main Methods:

• Compiled a curated dataset from published literature on CAN depolymerization.
• Utilized material descriptors and processing parameters as model features.
• Employed tree-based machine learning models, including XGBoost, with hyperparameter optimization and cross-validation.
• Applied Shapley additive explanations (SHAP) for model interpretation.

Main Results:

• XGBoost demonstrated the highest predictive accuracy in modeling depolymerization behavior.
• SHAP analysis quantified the influence of material descriptors and processing parameters on depolymerization time.
• Independent validation confirmed reasonable agreement between model predictions and experimental data.
• Identified systematic statistical relationships between material properties, processing conditions, and depolymerization kinetics.

Conclusions:

• The developed data-driven framework provides a quantitative tool for thermoset recycling.
• Machine learning enables prediction of solvent-assisted depolymerization kinetics.
• This approach facilitates informed solvent selection and process optimization for sustainable thermoset waste management.