A data-driven method in predicting crack coalescence based on pre-existing flaw geometry

Chunjiang Zou, Boyuan Chen, Huakun Yu, Sihan Yan, Ruoge Wang, Ruoxi Zhu, Yunshun Li, Daokun Zhang

Abstract: The coalescence of the discontinuities often leads to rock mass failure. This study develops a novel data-driven approach to predict crack coalescence in Carrara marble with double pre-existing flaws, focusing on flaw geometric parameters: inclination angle (β), bridging angle (α), and ligament length (L). Using 252 groups of numerical simulations validated against physical Carrara marble experiments, a Gaussian Process classifier was trained to predict coalescence types (no coalescence, direct, indirect) with high accuracy. Results reveal direct coalescence dominates (78 %), with β exerting the strongest influence (40 % weight). Non-linear relationships between flaw geometry and coalescence were quantified, enabling probabilistic predictions for untested configurations. This method eliminates the need for resource-intensive simulations or experiments, offering an efficient tool for failure forecasting in the future. Findings can enhance rock mass stability assessments in tunneling, mining, and rock slope, and will advance predictive modeling in geohazard mitigation.

External IDs:doi:10.1016/j.enggeo.2025.108362