Stability of rectangular tunnels in cohesive-frictional soil under surcharge loading using isogeometric analysis and Bayesian neural networks

This study evaluates the stability of rectangular tunnels in cohesive-frictional soils under surcharge loading using a combination of IsoGeometric Analysis and artificial neural networks. A dataset of 12,946 samples was generated automatically to analyze a wide range of soil profiles and tunnel geometries. Stability solutions were derived using IsoGeometric Analysis coupled with second-order cone programming, enabling precise and efficient assessments of ultimate surcharge loading. A key contribution of this study is the development of a closed-form solution through a Bayesian regularized neural network, which significantly improves accuracy compared to existing methods. Advanced data visualization techniques, including two- and three-dimensional partial dependency plots, were used to reveal complex relationships among design parameters. Sensitivity analyses provided valuable insights for optimizing tunnel designs, enhancing decision-making processes in geotechnical engineering. This study aims to equip engineers with practical tools for designing rectangular tunnels in real-world applications.