Structural and mechanical properties of alkali silicate glasses: Insights from molecular dynamics simulations and artificial intelligence

MS Idrissi and A El Hamdaoui and T Chafiq and S Ouaskit, JOURNAL OF NON- CRYSTALLINE SOLIDS, 666, 123706 (2025).

DOI: 10.1016/j.jnoncrysol.2025.123706

The various properties of alkali silicate glasses could be predicted by applying different techniques of artificial intelligence (AI), including machine learning (ML) and deep learning (DL). The training process requires huge datasets; however, the experimental values for different properties may not always be available. In that case, the datasets can be generated by molecular dynamics simulations (MD). In this paper, ML and DL algorithms have been performed to predict the structural and mechanical properties from chemical composition for three alkali silicate glasses (lithium, sodium, and potassium silicate) based on a dataset prepared using MD simulations. Notably, we demonstrate that even with a highly limited coarse-resolution dataset, fine-tuned models particularly artificial neural networks (ANN) can achieve predictive performance comparable to that of models trained on the full high- resolution MD dataset. This clearly illustrates the efficiency and practicality of ML approaches in reducing the computational cost of glass design by achieving accurate predictions with significantly fewer data points. The performance of different models developed within is compared on the basis of the accuracy and model complexity. Polynomial regression (PR), and artificial neural networks (ANN) gave, among others, the best prediction performance. These results serve as a stepping stone toward the machine learning guided design of novel glasses with specific properties.

Return to Publications page