Investigating the structure and fracture toughness of spodumene glass- ceramics using machine learning

DM Sun and MH Shi and X Ge and FM Cao and LR Jensen and JW Ding and JA Zhao and T Du and MM Smedskjaer, CERAMICS INTERNATIONAL, 51, 47563-47576 (2025).

DOI: 10.1016/j.ceramint.2025.07.259

Crystals in glass-ceramics can impede crack propagation and cause crack deflection to prevent crack extension, improving the fracture toughness of the glass. However, detailed atomic-scale understanding of crack initiation and propagation in glass-ceramics remains elusive. In this work, we first experimentally investigate the fracture patterns of Li2O-Al2O3-SiO2 (LAS) glass-ceramics, finding enhanced crack deflection for higher crystallinity. To investigate the influence of different crystal morphologies on the fracture mode and understand the structure- mechanics relationships, we then simulate the fracture behavior of LAS glass-ceramics using molecular dynamics simulations. We find that a high frequency of Al coordination number changes (bond switching events) occur during the fracture process, consuming the fracture energy. We then use a classification-based machine learning model to calculate 'softness' based on the static structure. Subsequent simulations show that softness can predict Al bond switching and thus capture the long- term dynamic behavior of Al atoms. Notably, although we did not use crystallinity as the training dataset, the machine learning model can learn the structure features related to bond switching probability and differentiate the fracture behavior at different stages. As such, softness can effectively predict the fracture behavior, i.e., predict the microscopic failure mode of glass-ceramics based on their structure. This work thus reveals the influence of the initial atomic-scale structure on different fracture modes in glass-ceramics, which is beneficial for guiding the design of novel glass-ceramic compositions.

Return to Publications page