An Integrated Method for Identification and Predictive Modeling of Rock Discontinuities in Tunnels: Methodology and Experimental Validation

Characterizing discontinuities ahead of the tunnel face is critical for geological prediction and risk control. However, this task is difficult because the rock mass is exposed only during excavation, while hidden structures usually lack direct validation. This study proposes a closed-loop framework that combines discontinuity identification, geometric connectivity analysis, and ahead-of-face prediction. First, MSECNet extracts discontinuities from exposed tunnel faces by using surface-normal information to guide the representation of geometric edges. The extracted traces are then analyzed under coplanarity constraints, so that their spatial connectivity can be inferred. Based on these connected traces, a three-dimensional discontinuity network is reconstructed by integrating deterministic geometry with stochastic variability. Excavation mileage is further introduced as a sequential variable to predict discontinuity distributions in the unexcavated rock mass. To support quantitative validation, a physical model testbed is developed to reproduce the progressive exposure of tunnel faces. Point clouds acquired by three-dimensional scanning are used to update the network and verify the predicted structures at successive excavation stages. In this way, discontinuity prediction is transformed from static interpretation of a single exposed face into a spatiotemporal inference process constrained by geometry and supported by physical evidence. The proposed framework provides an experimental and algorithmic basis for reliable discontinuity forecasting in tunnel engineering and offers a transferable route for integrating digital mapping, structural analysis, and geological prediction.