About this Abstract |
Meeting |
2020 TMS Annual Meeting & Exhibition
|
Symposium
|
Computational Discovery and Design of Emerging Materials
|
Presentation Title |
Predicting Physical Properties of SiO2-based Glasses by Machine Learning |
Author(s) |
Yong-Jie Hu, Ge Zhao, Bo Liu, Yang Chen, Kerby Shedden, Liang Qi |
On-Site Speaker (Planned) |
Liang Qi |
Abstract Scope |
SiO2-based glasses have diverse applications as both structural and functional materials. It is difficult to efficiently predict and optimize their physical properties according to the chemical composition due to their non-crystalline structures. We have developed machine learning (ML) techniques to predict their physical properties across a complex compositional space. For densities and elastic moduli of glasses, our approach relies on a training set generated by high-throughput molecular dynamic simulations and descriptors based on fundamental physics of interatomic bonding. For properties that are difficult to be calculated by atomistic simulations, such as liquidus temperatures and dielectric constants, the training sets are directly obtained from available glass property databases such as SciGlass and descriptors based on crystalline oxide properties from first-principles calculations. We also applied ML models to generate a compositional-property database that allows for a fruitful overview of the physical properties of the general multi-component glass systems. |
Proceedings Inclusion? |
Planned: Supplemental Proceedings volume |