Prediction of lining response for twin tunnels constructed in anisotropic clay using machine learning techniques

Abstract

Excessive structural forces generated inside tunnel linings could affect the safety and serviceability of tunnels, emphasizing the need to accurately predict the forces acting on tunnel linings during the preliminary design phase. In this study, an anisotropic soil model developed by Norwegian Geotechnical Institute (NGI) based on the Active-Direct shear-Passive concept (NGI-ADP model) was adopted to conduct finite element (FE) analyses. A total of 682 cases were modeled to analyze the effects of five key parameters on twin-tunnel structural forces; these parameters included twin-tunnel arrangements and subsurface soil properties: burial depth H, tunnel center-to-center distance D, soil strength suA, stiffness ratio Gu/suA, and degree of anisotropy suP/suA. The significant factors contributing to the bending moment and thrust force of the linings were the tunnel distance and overlying soil depth, respectively. The degree of anisotropy of the surrounding soil was found to be extremely important in simulating the twin-tunnel construction, and severe design errors could be made if the soil anisotropy is ignored. A cutting-edge application of machine learning in the construction of twin tunnels is presented; multivariate adaptive regression splines and decision tree regressor methods are developed to predict the maximum bending moment within the first tunnel’s linings based on the constructed FE cases. The developed prediction model can enable engineers to estimate the structural response of twin tunnels more accurately in order to meet the specific target reliability indices of projects.