Research on Collapse Risk Assessment of Karst Tunnels Based on BN Self-Learning

Abstract

The high risk of collapse is a key issue affecting the construction safety of karst tunnels. A risk assessment method for karst tunnel collapse based on data-driven Bayesian Network (BN) self-learning is proposed in this study. The finite element calculation is used to analyze the distribution law of the plastic zone of the tunnel and the karst cave surrounding rock under different combinations of parameters, and a four-factor three-level data case database is established. Through the self-learning of the BN database, a Bayesian Network model of karst tunnel collapse risk assessment with nodes of four types of karst cave parameters is established. The specific probability distribution state and sensitivity of the parameters of different types of karst caves under the condition of whether the tunnel and the karst cave plastic zone are connected or not are studied. The research results show that the distance and angle of the karst cave are the main influencing parameters of the tunnel collapse probability, and the diameter and number of the karst cave are the secondary influencing parameters. Among them, the distance, diameter, and number of karst caves are proportional to the probability of tunnel collapse, and the most unfavorable orientation of karst caves is 45° above the tunnel. When the tunnel passes through the karst area, it should avoid the radial intersection with the karst cave at the arch waist while staying away from the karst cave. The results of this work can provide a reference for the construction safety of karst tunnels under similar conditions.