Intelligent crack classification and precursory characteristics in the granite shear fracturing process

Rock mass instability and fracture monitoring are essential for deep mining and geotechnical safety, yet traditional acoustic emission (AE) b-value analysis lacks explicit physical mechanisms and quantitative warning thresholds, while the RA–AF method for tensile-shear crack classification suffers from subjective boundary definitions and poor robustness. In this study, shear-induced AE tests on granite were processed via a hybrid GMM–SVM model to intelligently classify RA–AF data. The results yield average shear and tensile crack proportions of 66.29% and 33.71%, respectively, and reveal a distinct three-stage shear-tensile-shear evolutionary pattern. During the b-value decline phase, the shear crack ratio reaches 94.33%, demonstrating that the drop is driven by large-scale shear failure. We therefore propose a dual-criterion early-warning approach: simultaneous sharp decreases in b-value and a shear ratio of 94.33% can serve as quantitative precursors to instability. This framework integrates crack size and fracture mode, overcoming the qualitative limitations of conventional b-value analysis and offering physically meaningful, multi-dimensional criteria for hazard warning in rock engineering.