Statistical analysis on nanostructure–mechanical property relations for xSiO₂–(1-x)Al₂O₃ aluminosilicate glass with voids and inclusions

The complex constituent elements and nanoscale structures of a silicate glass material significantly affect its general properties, which challenges the prediction of macroscopic structure-property relations. In this study, a statistical analysis procedure is proposed to establish nanoscale structure-property relations of xSiO₂–(1-x)Al₂O₃ aluminosilicate glass. The nanostructural complexities in terms of Al₂O₃ content as well as type, size and distribution of internal defects are considered by using multiple nanostructure descriptors. Based on principal component analysis (PCA) and polynomial regression, we then establish high dimensional predictive models with simple formulas and high accuracies. Using the developed model, massive data is generated (over 30,000 data points) for a multivariate global sensitivity analysis (GSA). Finally, nanostructural variables influential to the mutually correlated mechanical properties are identified; their individual, interactional and total effects are quantitatively evaluated. The findings of this work not only reveal the complex nanostructure-property relations for the aluminosilicate glass, but also offer a feasible solution to model establishment and GSA performance for a high dimensional system, both of which usually require massive training data and are computationally too expensive.