Excavation-induced fault activation has become a major barrier in deep rock excavation. Multiple lines of evidence indicate that shear stress perturbation is one of the main causes of fault activation. However, prediction and control of shear stress perturbation during deep rock excavation remain challenging. Here we combine numerical modelling and machine learning to forecast the evolution of shear stress along pre-existing faults adjacent to the deep tunnel excavation. We simulate the shear stress distributions along the faults with various dip angles and distances to the tunnel under different in-situ stress conditions. We then train three machine learning models with the key factors collected from the numerical models and test the models to show a general trend of shear stress distribution along the faults. Our study demonstrates a better understanding of shear stress perturbation during deep rock excavation and suggests practical methods to control and mitigate the excavation-induced seismic events.

Underground excavation,fault slip,induced seismicity,numerical simulation,machine learning