A Model Ensemble Approach Enables Data-Driven Property Prediction for Chemically Deconstructable Thermosets in the...

Area of Science:

• Materials Science
• Polymer Chemistry
• Computational Chemistry

Background:

• Thermosets pose sustainability challenges due to their complex composition and lack of predictive property knowledge.
• Developing deconstructable thermosets with tunable properties is crucial for environmental sustainability.
• Computational methods struggle with amorphous, multicomponent thermoset systems.

Purpose of the Study:

• To develop a predictive model for thermoset properties using machine learning.
• To accelerate the discovery of sustainable thermosets with controlled deconstructability.
• To enable property prediction based on molecular building blocks and formulation.

Main Methods:

• A closed-loop strategy combining experimentation and machine learning (ML) was employed.
• A dataset of 101 polydicyclopentadiene (pDCPD) thermoset examples was utilized.
• Molecular features and formulation variables were used as ML model inputs, with uncertainty quantification via ensembling.

Main Results:

• Accurate predictions of glass transition temperature (Tg) for pDCPD thermosets with cleavable bifunctional silyl ether (BSE) comonomers/cross-linkers.
• Predictions achieved an accuracy within <15 °C for diverse BSE compositions.
• The ML model effectively handled multicomponent, amorphous thermoset systems.

Conclusions:

• This data-driven approach facilitates the prediction of thermoset properties from molecular components.
• The strategy accelerates the discovery of advanced plastics, rubbers, and composites.
• Enables the design of thermosets with improved functionality and controlled deconstructability for sustainability.