Failure characteristics of tunnels neighbouring karst fissures: Insights from laboratory observations and machine learning-interpreted simulations

Abstract

This study primarily investigates the mechanical responses and failure mechanisms of tunnels adjacent to karst fissures through detailed laboratory model experiments, supplemented by machine learning-interpreted simulations. Utilizing a novel tunnel-karst fissure model test system, we meticulously analysed stress and displacement evolution within the surrounding rock during tunnel excavation, elucidating unevenness and complexity of the rock displacement-stress field induced by the fissure’s low stiffness. The experimental findings delineate the critical phases of rock instability, showcasing how hydraulic pressures interact with karst features to induce failure. Key experimental results indicate a mechanism for the formation of water inrush channels, which are crucial for understanding tunnel vulnerability near karst fissures. Complementing the physical experiments, a deep residual network (ResNet), interpreted using SHapley Additive exPlanations (SHAP), was employed to quantify the impacts of fissure positions, lateral pressure coefficients, and hydraulic pressures on the critical safety distances of fissures, according to simulation results. This integrated approach enhances the precision of risk assessments and supports the development of targeted mitigation strategies in karst-affected tunnelling projects, thereby improving the safety and sustainability of underground constructions.