Innovative approaches for predicting seismic stability of circular and rectangular tunnels in cohesive-frictional soils using machine learning and finite element limit analysis

This paper investigates the stability solutions for plane-strain circular and rectangular tunnels in cohesive-frictional soils using the Mohr-Coulomb failure criteria. The study examines the impact of pseudo-static seismic body forces on tunnel failure behavior during earthquakes. Stability analysis is conducted using two-dimensional finite element limit analysis (2D FELA), ensuring reliable results. A comprehensive range of dimensionless input parameters is systematically examined, including tunnel cover depth ratio (H/D), normalized tunnel shape ratio (B/D), coefficient of horizontal earthquake acceleration (k_h), normalized soil strength (γD/c), and soil friction angle (φ), to analyze the stability load coefficient (σ_i/c) or the required load at the tunnel circumference. The collapse mechanisms of circular and rectangular tunnels in Mohr-Coulomb soils are thoroughly discussed. The findings offer innovative solutions for improving tunnel design practices in cohesive-frictional soils under pseudo-static seismic forces. Additionally, this study introduces a machine learning model integrating a random forest (RF) technique with the dragonfly optimization algorithm (DOA) to develop surrogate models for predicting the seismic stability load factor of tunnels. The proposed RF-DOA hybrid model is validated, showing strong agreement with numerical FELA results. Shapley analysis reveals normalized soil strength (γD/c) as the most influential factor. These findings provide a reliable solution and an effective tool for enhancing tunnel design in cohesive-frictional soils under earthquake conditions.