Work

Predicting stiffness and toughness of aluminosilicate glasses using an interpretable machine learning model

Year: 2025

Type: article

Abstract: • High-throughput MD simulations of calcium aluminosilicate glasses were carried out. • Neural networks models can well capture both the elastic and fracture behaviors. • The prediction models were interpreted by the Shapely additive explanations algorithms. • Oxygen tricluster fraction is the key factor for controlling mechanical properties. The increasing demand for lighter and more durable glass materials relies on the development of stiffer, stronger, and tougher glasses. However, the design of new glasses with targeted properties is largely impeded due to the lack of composition-structure–property models. Here, we combine machine learning with high-throughput molecular dynamics simulations to predict the mechanical properties of 231 calcium aluminosilicate (CAS) glass compositions under varying preparation conditions. We demonstrate that prediction models based on neural networks can well capture both the elastic and fracture behaviors of CAS glasses. By interpretating the prediction model, we demonstrate that the Al 2 O 3 content is the primary factor determining mechanical properties. Specifically, an increase in Al 2 O 3 content leads to higher modulus, tensile strength, and toughness. The roles of preparation pressure and cooling rate are positively correlated with modulus and tensile strength, respectively. Structure analyses reveal that the fraction of oxygen triclusters is the key factor for controlling both the elastic and fracture behavior of the CAS glasses. Based on these findings, our work facilitates the rational design of new oxide glasses with targeted properties.

Source: Engineering Fracture Mechanics

DOI: 10.1016/j.engfracmech.2025.110961

Authors: Tao Du, Zhimin Chen, Sidsel M. Johansen, Qiangqiang Zhang, Yuanzheng Yue, Morten M. Smedskjaer

Institutions: Hong Kong Polytechnic University, Aalborg University, Lanzhou University

Language: English

Topic: Machine Learning in Materials Science

Subfield: Materials Chemistry

Field: Materials Science

Domain: Physical Sciences

Open Access status: hybrid

Estimated APC paid: $4,120