The fracture behaviour of glass has been studied since quite some time mainly with the focus on deriving strength or fracture mechanical properties. While such quantities are critical for the design assessment in facade engineering and structural glazing, this research investigates the fracture pattern from a geometrical point of view for different levels of thermal pre-stress, different glass thickness and support conditions, in order to calibrate statistical predictive models for e.g. fragment count, fragment shape parameters, homogeneity of the fracture pattern etc. As this research is inherently data-driven, we first present the conduction of experiments and the collection of corresponding data. In a subsequent step, we employ computer vision and Bayesian quantification algorithms to identify and model several key quantities of the fracture geometry.
The results indicate our models to significantly capture main statistical properties and the method serves as a template framework for future data-driven research in glass fracture modelling.