Transfer Learning for Polymer Mechanics: A Fusion Approach to Bridge Molecular Dynamics Simulations and Experiments...

Area of Science:

• Polymer Science
• Materials Science
• Computational Chemistry

Background:

• The stress-strain curve is crucial for understanding polymeric material mechanics, particularly for solution-polymerized styrene-butadiene rubber (SSBR).
• Molecular dynamics (MD) simulations offer microscale insights but often yield inaccurate stress values due to high strain rates compared to experimental data.

Purpose of the Study:

• To develop a robust method for accurately predicting SSBR stress-strain curves by bridging the gap between MD simulations and experimental observations.
• To enhance the performance optimization of SSBR through accurate mechanical behavior prediction.

Main Methods:

• A weighted fusion framework integrating transfer learning with a hybrid Long Short-Term Memory-Multilayer Perceptron (LSTM-MLP) and eXtreme Gradient Boosting (XGBoost) model was employed.
• A dataset comprising 100 simulated SSBR stress-strain curves and 5 experimental curves was utilized for model pretraining and fine-tuning.

Main Results:

• The proposed model achieved stress-strain predictions consistent with experimental data, outperforming alternative machine learning baselines in accuracy.
• Correlation analysis identified the influence of SSBR's four structural units (styrene, 1,2-butadiene, cis-1,4-butadiene, trans-1,4-butadiene) on mechanical properties.

Conclusions:

• The developed transfer learning framework effectively reconciles discrepancies between simulated and experimental SSBR mechanical data.
• The findings provide theoretical insights for targeted enhancement of SSBR performance by understanding structure-property relationships.