AI-Driven Insight into Polycarbonate Synthesis from CO₂: Database Construction and Beyond.

Area of Science:

• Materials Science
• Polymer Chemistry
• Computational Chemistry

Background:

• Developing new polymers via epoxide and CO2 copolymerization is challenging due to unpredictable outcomes.
• Current trial-and-error methods are resource-intensive, leading to significant financial and time losses.
• Artificial Intelligence (AI) offers potential solutions, but high-quality data for polymers is scarce.

Purpose of the Study:

• To create the first dataset linking epoxy comonomer structure, catalysts, and conditions to polymerization success.
• To develop and validate a machine learning (ML) analytical pipeline for predicting polymer properties.
• To mitigate the challenges and reduce the costs associated with novel polymer development.

Main Methods:

• Compilation of a novel dataset detailing epoxy comonomer structure, catalyst, and polymerization conditions.
• Development of an ML-based analytical pipeline, incorporating AutoML for hyperparameter tuning.
• Addressing the dimensionality problem inherent in complex material datasets.

Main Results:

• The ML pipeline successfully predicted molecular weight (R2=0.79), polydispersity index (R2=0.86), and conversion rate (R2=0.93).
• Initial results highlight the significance of managing data dimensionality for accurate predictions.
• The automated pipeline demonstrates scalability and potential for future research.

Conclusions:

• The developed ML pipeline provides accurate predictions for key polymer characteristics, reducing experimental uncertainty.
• This approach offers a foundation for data-driven polymer design, minimizing resource expenditure.
• The study paves the way for more efficient and predictable synthesis of novel polymeric materials.