A machine learning prediction framework for damping performance of silicone rubber based on molecular dynamics and VAE-OK data augmentation

Henan Tang, Xuefeng Yu, Qingxian Liu, Yunlong Li, Jingyao Dong, Shuaida Wang, Bin Yang,
A machine learning prediction framework for damping performance of silicone rubber based on molecular dynamics and VAE-OK data augmentation,
Polymer,
Volume 340,
2025,
129269,
ISSN 0032-3861,
https://doi.org/10.1016/j.polymer.2025.129269.
(https://www.sciencedirect.com/science/article/pii/S0032386125012558)
Abstract: The remarkable damping characteristics of the silicone rubber (SR) have piqued the curiosity of scientists. Molecular dynamics (MD) simulations are helpful for studying the impact of crucial structural factors on damping performance at the molecular level. There is a great deal of potential to enhance the efficiency and accuracy of MD simulations by combining machine learning method. In this study, a novel data augmentation-machine learning technique trained by small-sample MD results were proposed to predict the damping performance of SR. The methodology encompassed: (1) employing a variational autoencoder for rapid scaling of MD computational data; (2) utilizing Ordinary Kriging for labeling virtual samples; (3) employing the Gradient Boosting Algorithm Model for damping performance prediction. Crucial structural factors such as the hydrogen bond number, free volume, and volume modulus of SR were focused. The findings demonstrated the effectiveness on the data augmentation and GBR prediction model based on augmented MD samples. The updated regression models showed significant improvements in the accuracy and reliability. With this approach, the issue of inadequate data for machine learning applications in materials science could be effectively tackled. Subsequently, the model's robustness and predictive capability were validated, demonstrating strong internal and external prediction performance.
Keywords: Silicone rubber; Loss factor; GBR gradient regression; Molecular dynamics simulation; Data enhancement; Ordinary Kriging