Cross-scale analysis for the thermo-hydro-mechanical (THM) effects on the mechanical behaviors of fractured rock: Integrating mesostructure-based DEM modeling and machine learning

Abstract

The excavation process induces the emergence of damage zones in the on-site engineering rock masses, significantly affecting the parameter evaluation of surrounding rock masses. Therefore, an in-depth understanding of the thermo-hydro-mechanical (THM) effects on the cross-scale failure process of fractured rock becomes progressively crucial in natural energy exploitation projects. In this study, a coupled thermo-hydro-mesostructure-based DEM (T-H-MSBM) model was developed to reconstruct rock microstructures and distinguish the THM responses of varying mineral grains, micro-defects and cracks in fractured granite. Five sets of numerical fractured granite in terms of different damage degrees were generated by restoring the form of excavation damage on site, and a series of compression simulations were conducted on the T-H-MSBMs under coupled temperature (25–250 °C), pore pressure (0–15 MPa) and confining pressure (15 MPa) conditions. Based on the cross-scale failure analysis on the fractured granite during the THM-compression loading process, the interplay between THM treatment and damage degree on the mechanical properties of fractured granite was revealed, and the main mechanisms affecting the varied macro mechanical properties were further discussed. Results indicate that both temperature and pore pressure exert the amplified deteriorating effect on the macro mechanical properties of fractured granite with increasing damage degree, while the temperature dependence becomes significantly more pronounced in the fractured granite with low damage degree. With an increase in the damage degree of fractured rock, the abundance of intra-mineral tensile cracks accelerates the initiation of pore pressure-induced cracks, and the coupled increase in pore pressure and temperature enhances the role of quartz-feldspar cracks in promoting the cracks in quartz. Finally, utilizing the optimized kernel ridge regression (KRR) model employed the grid search method, the correlation information database of rock macro-mechanical strength and THM factors were generated. The entire training time was less than 3 s, with the relative average errors of 0.7 MPa, which demonstrated the robust capability of machine learning method in predicting the coupled THM phenomena spanning multiple scales in geological materials and systems.