High-Throughput Screening and Prediction of High Modulus of Resilience Polymers Using Explainable Machine Learning

Area of Science:

• Materials Science
• Computational Chemistry
• Polymer Science

Background:

• The modulus of resilience (R) quantifies a material's ability to store and release elastic strain energy, crucial for mechanical systems.
• Improving R requires a high yield strength (σy) and low Young's modulus (E), a challenging combination as these properties often correlate.
• Discovering polymers with optimal R is essential for developing advanced materials.

Purpose of the Study:

• To develop a computational method for rapidly identifying polymers with a high modulus of resilience.
• To leverage machine learning (ML) and molecular dynamics (MD) simulations for accelerated materials discovery.
• To identify key polymer substructures influencing mechanical properties for targeted material design.

Main Methods:

• Trained single-task, multitask, and Evidential Deep Learning models to predict polymer mechanical properties (E and σy) using experimental data.
• Employed explainable ML models to identify critical substructures affecting polymer mechanical properties.
• Validated ML predictions using high-fidelity molecular dynamics (MD) simulations.

Main Results:

• Identified 10 new real polymers and 10 hypothetical polyimides with exceptional modulus of resilience.
• ML models accurately predicted properties for 12,854 real polymers and 8 million hypothetical polyimides.
• MD simulations confirmed the enhanced modulus of resilience in the newly discovered polymers.

Conclusions:

• The proposed ML-driven approach significantly accelerates the discovery of high-performance polymers.
• This method effectively combines ML predictions with MD validation for efficient material design.
• The approach is applicable to broader polymer material discovery challenges, including membranes and dielectrics.