Abstract: The post-fracture load-bearing properties of laminated glass is crucial for the safety of architectural glass structures under long-duration wind disasters or explosions. Crack morphology significantly impacts the tension-stiffening effect and the triggering of local failure in laminated glass. Fragmentation tests and post-fracture uniaxial tensile tests were conducted on laminated tempered glass with SG interlayer to obtain the testing datasets of crack morphological images and associated mechanical properties. The refined finite element (FE) models were then calibrated with experimental outcomes to determine key parameters such as fragment density, minimum nearest neighbour distance, and effective bonding coefficients corresponding to the crack morphology. Subsequently, the batch generation of FE models for fractured laminated glass by adjusting surface compressive stress was established, which utilized Voronoi tessellation to approximate the actual morphology to expand the simulation dataset. The fast evaluation method based on convolutional neural network (CNN) with high accuracy for the equivalent tensile properties of fractured laminated glass was developed via image recognition training on crack morphology.

Key words: laminated glass, SG interlayer, post-fracture property, crack morphology, fast evaluation method, convolutional neural network