Deep Learning-based Identification of Crack Pattern Evolution Leading to Catastrophic Rupture in Sandstones

Understanding the failure mechanisms of rocks is crucial for various engineering applications and natural hazard mitigation. However, the mechanical response associated with crack pattern evolution and effective indicators for predicting catastrophic rupture remain unexplored. This study presents a novel precursory predictor for catastrophic rupture and elucidates the relationship between stress softening and crack evolution in sandstone. A deep learning-based segmentation model is developed to identify evolving crack patterns. Results demonstrate that crack propagation follows an acceleration–deceleration trend, with catastrophic rupture occurring after a time lag from the peak crack rate. A substantial number of cracks form during the post-peak phase, indicating that the peak crack rate predicts impending rupture. Furthermore, rapid stress softening near the rupture point follows a negative logarithmic function relationship with the cumulative crack area or length. The findings indicate that catastrophic rupture is induced by the formation of new localized zones and their interaction with existing ones rather than the growth of pre-existing damage zones. This study comprehensively characterizes unstable crack propagation and offers new insights into brittle failure mechanisms, with implications for predicting rupture in geological and engineering contexts.