Parameter influence analysis in a 3D TBM model via sensitivity analysis and Gaussian process metamodels

Urban tunnel excavation with Tunnel Boring Machines (TBMs) can induce surface settlements that may affect nearby structures. Despite improvements in TBM technology, deformations persist due to factors such as over-excavation, shield geometry, grout behavior, and soil properties. Three-dimensional finite element models (FEM) are widely used to simulate these effects but are computationally expensive, limiting their use in parametric studies. This work presents a 3D FEM simulator of mechanized tunneling and investigates two key aspects: (i) the influence of modeling parameters-domain geometry and mesh density-on numerical accuracy and computational cost, and (ii) the relative importance of uncertain input parameters on model outputs. A global sensitivity analysis was performed using Sobol' indices to quantify parameter influence. To reduce the computational cost, Gaussian process (GP) metamodels were built and validated using leave-one-out cross-validation. The resulting framework enables efficient screening of influential inputs and supports reduced-order modeling in design or calibration tasks.