Microscopic cracking behaviors of rocks under uniaxial compression with microscopic multiphase heterogeneity by deep learning

Abstract

Cracking behaviors of rocks significantly affect the safety and stability of the explorations of underground space and deep resources. To understand deeply the microscopic cracking process and mechanical property of rocks, X-ray micro-computed tomography (X-μCT) is applied to capture the rock microstructures. The digital color difference UNet (DCD-UNet)-based deep learning algorithm with 3D reconstruction is proposed to reconstruct the multiphase heterogeneity microstructure models of rocks. The microscopic cracking and mechanical properties are studied based on the proposed microstructure-based peridynamic model. Results show that the DCD-UNet algorithm is more effective to recognize and to represent the microscopic multiphase heterogeneity of rocks. As damage characteristic index of multiphase rocks increases, transgranular cracks in the same grain phase, transgranular and intergranular cracks of pore-grain phase, intergranular and secondary transgranular cracks and transgranular crack between different grains propagate. The ultimate microscopic failure modes of rocks are mainly controlled by the transgranular cracks-based T1-shear, T3-shear, T1-tension, T2-tension and T3-tension failures, and the intergranular cracks-based T1-tension, T1-shear and T3-shear failures under uniaxial compression.