Three-dimensional undrained stability analysis of circular tunnel heading in anisotropic and heterogeneous clay: FELA, ANN, MARS, and XGBoost

The tunnel face stability in undrained anisotropic clay that has an increasing shear strength with depth is investigated by using three-dimensional (3D) finite element limit analysis (FELA). Four dimensionless input parameters are considered in this study to develop the dataset of the tunnel stability factors, which include the overburden stress ratio, cover-depth ratio, anisotropic ratio, and strength gradient ratio. Based on the created FELA dataset, three machine learning models, namely Artificial Neural Network (ANN), Multivariate Adaptive Regression Splines (MARS), and Extreme Gradient Boosting (XGBoost), are carried out and optimized for their structures, and their performances are ranked by regression parameters and quantity methods. The evaluation result shows that the MARS model has the highest effectiveness, and its optimized structure is implemented to propose an explicit predictive function for the tunnel stability problem. Additionally, the sensitivity analysis is followed up to quantify the impact of relative design parameters on the stability performance, indicating that the strength gradient ratio is the most important variable, followed by the anisotropic ratio, cover-depth ratio, and overburden stress ratio, respectively. Furthermore, the closed-form equation derived using the MARS method provides a reliable tool for tunnel design practitioners.