Search for: All records

Machine learning (ML) is emerging as a powerful tool to predict the properties of materials, including glasses. Informing ML models with knowledge of how glass composition affects short-range atomic structure has the potential to enhance the ability of composition-property models to extrapolate accurately outside of their training sets. Here, we introduce an approach wherein statistical mechanics informs a ML model that can predict the non-linear composition-structure relations in oxide glasses. This combined model offers an improved prediction compared to models relying solely on statistical physics or machine learning individually. Specifically, we show that the combined model accurately both interpolates and extrapolates the structure of Na₂O–SiO₂glasses. Importantly, the model is able to extrapolate predictions outside its training set, which is evidenced by the fact that it is able to predict the structure of a glass series that was kept fully hidden from the model during its training.

 

Nanoindentation is a widely used method to probe the mechanical properties of glasses. However, interpreting glasses’ response to nanoindentation can be challenging due to the complex nature of the stress field under the indenter tip and the lack ofin situcharacterization techniques. Here, we present a numerical model describing the nanoindentation of an archetypical soda‐lime silicate window glass by means of peridynamic simulations. We show that, although it does not capture shear flow and permanent densification, peridynamics exhibits a good agreement with experimental nanoindentation data and offers a direct access to the stress field forming under the indenter tip.

 

Topological constraint theory (TCT) has enabled the prediction of various properties of oxide glasses as a function of their composition and structure. However, the robust application of TCT relies on accurate knowledge of the network structure and topology. Here, based on classical molecular dynamics simulations, we derive a fully analytical model describing the topology of the calcium aluminosilicate [(CaO)_x(Al₂O₃)_y(SiO₂)_1−x−y, CAS] ternary system. This model yields the state of rigidity (flexible, isostatic, or stressed‐rigid) of CAS systems as a function of composition and temperature. These results reveal the existence of correlations between network topology and glass‐forming ability. This study suggests that glass‐forming ability is encoded in the network topology of the liquid state rather than that of the glassy state.

 

Lithium aluminoborate glasses have recently been found to feature high resistance to crack initiation during indentation, but suffer from relatively low hardness and chemical durability. To further understand the mechanical properties of this glass family and their correlation with the network structure, we here study the effect of adding SiO₂to a 25Li₂O–20Al₂O₃–55B₂O₃glass on the structure and mechanical properties. Addition of silica increases the average network rigidity, but meanwhile its open tetrahedral structure decreases the atomic packing density. Consequently, we only observe a minor increase in hardness and glass transition temperature, and a decrease in Poisson's ratio. The addition of SiO₂, and thus removal of Al₂O₃and/or B₂O₃, also makes the network less structurally adaptive to applied stress, since Al and B easily increase their coordination number under pressure, while this is not the case for Si under modest pressures. As such, although the silica‐containing networks have more free volume, they cannot densify more during indentation, which in turn leads to an overall decrease in crack resistance upon SiO₂addition. Our work shows that, although pure silica glass has very high glass transition temperature and relatively high hardness, its addition in oxide glasses does not necessarily lead to significant increase in these properties due to the complex structural interactions in mixed network former glasses and the competitive effects of free volume and network rigidity.

 

Although peridynamics is widely used to investigate mechanical responses in materials, the ability of peridynamics to capture the main features of realistic stress states remains unknown. Here, we present a procedure that combines analytic investigation and numerical simulation to capture the elastic field in the mixed boundary condition. By using the displacement potential function, the mixed boundary condition elasticity problem is reduced to a single partial differential equation which can be analytically solved through Fourier analysis. To validate the peridynamic model, we conduct a numerical uniaxial tensile test using peridynamics, which is further compared with the analytic solution through a convergence study. We find that, when the parameters are carefully calibrated, the numerical predicted stress distribution agrees very well with the one obtained from the theoretical calculation.