Physics-Embedded Neural Network: A Novel Approach to Design Polymeric Materials.

Area of Science:

• Materials Science
• Computational Materials Science
• Polymer Science

Background:

• Machine learning is trending in polymeric materials design.
• Conventional neural networks lack physical constraints, interpretability, and struggle with limited data.

Purpose of the Study:

• To propose a Physics-Embedded Neural Network (PENN) for improved polymer material design.
• To embed physical laws into machine learning models for enhanced accuracy and interpretability.

Main Methods:

• Incorporated the Yeoh hyperelastic constitutive model into the neural network architecture.
• Pre-trained the model on molecular dynamics (MD) simulation data.
• Fine-tuned the model using a transfer learning strategy with limited experimental data and uncertainty quantification.

Main Results:

• The PENN model demonstrated improved physical plausibility and interpretability.
• Accurate predictions were achieved even with limited experimental data.
• The model enabled performance-driven inverse design for guiding polymer development.

Conclusions:

• PENN effectively bridges the gap between simulation and experiment in polymer science.
• The physics-embedded approach transforms predictive modeling into a practical tool for materials discovery.
• This work advances the design of high-performance polymeric materials.