Deep generative model‐based generation method of stochastic structural planes of rock masses in tunnels

Abstract

Tunnels stand as indispensable pillars of transportation infrastructure, assuming a central and transformative role in fostering the sustainable evolution of urban. The excavation process of tunnels presents a spectrum of geological challenges, encompassing the potential for instability and collapse. Ensuring the stability of the tunnel is a top priority in tunnel construction. The destabilization leading to collapse in certain tunnels is intricately connected to the structural planes of the rock mass. Accurately obtaining the distribution of structural planes within the rock mass is the necessary basis for maintaining the stability of the tunnel. The conventional Monte Carlo method generates each parameter of stochastic structural planes separately without considering the correlations between the parameters. To address this limitation, we propose a stochastic structural plane generation method based on deep generative model (DGM). The model takes the measured factual structural plane data as input, and the neural network realizes the generation of structural plane data with automatic learning of the distribution law of structural planes and the correlations between each parameters without assuming the probability distribution of stochastic structural planes in advance. This method has been used for stochastic structural plane generation of the rock mass in the Yuelongmen tunnel located in Mianyang City, Sichuan Province. The validation results show that the proposed DGM‐based method automatically captures the correlation between structural plane parameters while ensuring the greater accuracy of the generated structural planes.