Physical-informed machine learning prediction of properties of oxide glasses
Résumé
Glass is a material for the future with an infinity of compositions. To speed up the finding of new compositions with specific values of properties, the deep learning is an efficient method. Here, this method is used to predict physical properties as density, Young's modulus or dynamical viscosity. To do that, physical inspired models are developed. The general method is first presented with the databases used in this contribution. To predict density, the fitting is operated on the molar volume. For the Young's modulus, the fitting is achieved on the packing factor. The Poisson coefficient is determined according to the Makishima-Mackenzie's model. For each prediction, a comparison with experimental data is provided. Finally, predictions are used to see which oxide is the more relevant to enhance the Young's modulus, for instance.
Origine | Fichiers produits par l'(les) auteur(s) |
---|---|
Licence |