UCLA

Understanding the compositional dependence of properties of silicate glass is critical to design novel glasses for various technology applications. With the development in molecular dynamics simulations and machine learning techniques, a combined and fully computational approach, which is able to reveal the relationship between glass composition and its mechanical properties, can be developed and served as a guide prior to experiments and manufacturing. On one hand, machine learning is a powerful tool to predict the properties based on the existing database. On the other hand, molecular dynamics simulation cannot only produce sufficient data points for machine learning models but also provide a detailed picture of the atomic structure of glasses. This atomic-scale knowledge from molecular dynamics simulation contains an intrinsic relationship between glass compositions and their mechanical properties.

Here, we first use molecular dynamics simulation to generate the dataset for calcium aluminosilicate glasses and apply different machine learning models to predict their Young’s modulus using glass compositions in Chapter 1. Next, we apply topological constraint theory to quantify the atomic structures of simulated glasses and use this knowledge to predict Young’s modulus for calcium aluminosilicate glass family in Chapter 2. Last, in Chapter 3, we propose a fully analytical model to link the network topology with glass compositions.