Three-Dimensional Crack Recognition by Unsupervised Machine Learning

Abstract

Many macrocracks are usually generated during the fracturing of rocks. Elucidating the spatial distribution of cracks provides the basis for understanding crack nucleation and fracture formation in rock mechanics. Considering either a single microcrack or all the microcracks provides a limited interpretation of rock mass failure that is often induced by different macrocracks. Here we recognize macrocracks based on a three-dimensional (3D) crack model, implemented using an unsupervised machine learning algorithm and microcrack coordinates. This approach recognized microcracks that coalesce to form a macrocrack in three dimensions. Rock fracturing was performed using a triaxial loading test, and the coordinate data were obtained via the acoustic emission (AE) technique. The results show that the main macrocracks are distributed throughout the whole granite specimen, and smaller macrocracks form near the unloading surface. The AE-recognized crack pattern was found to be consistent with the actual cracks. The adaptability of the proposed method and the potential research and applications were discussed. This approach provides a means to understand the formation and distribution of rock fractures.