Quantitative characterization of crack coalescence mechanisms in granite at the mesoscopic scale using deep learning

This study quantitatively characterized the crack coalescence mechanisms in granite at the mesoscopic scale using deep learning and scanning electron microscopy (SEM). The meso-morphology of the crack surface created by crack coalescence was acquired by SEM. The fractal dimension of the morphology image was computed. Mesoscopic tensile and shear morphologies on the crack surface were quantitatively characterized by deep learning. The results showed: (1) The fractal dimension of the crack surface morphology caused by shear coalescence (1.927) was the highest, followed by that induced by tensile-shear coalescence (1.901), while the lowest was caused by tensile coalescence (1.871). (2) Shear morphology proportions were 14.3, 29.1, and 55.8% for crack surfaces from tensile coalescence, tensile-shear coalescence, and shear coalescence, respectively. (3) The crack coalescence behavior on the specimen surface was linked to that inside the specimen. A high ratio of shear to tensile strain on the sample surface was associated with a large proportion of shear morphology inside the specimen. (4) The meso-mechanisms of crack coalescence and crack propagation were compared, and crack coalescence exhibited more pronounced shear characteristics than crack propagation. The findings of this study offered new insights into the mesoscopic mechanisms of crack coalescence.