Multi-Objective Optimization of Steel-Reinforced Concrete Hybrid Segment Support for TBM Tunnels Based on Surrogate Model

TBM tunnels crossing fault fracture zones require support systems for the control of deformation; ideally, these systems should be economically and computationally feasible to design. Conventional numerical optimization involves repeatedly evaluating high-fidelity models, and the associated computational burden increases rapidly as structural variables and competing objectives are introduced. This study develops and applies a surrogate-assisted multi-objective optimization framework for a steel-reinforced concrete hybrid segment support system in a tunnel located in a fault fracture zone. The framework combines Optimal Latin Hypercube Sampling (OLHS) for design-space coverage, Kriging surrogate modeling for rapid response prediction, SHapley Additive exPlanations (SHAP) for parameter importance interpretation, and the Non-dominated Sorting Genetic Algorithm II (NSGA-II) to conduct a Pareto search. Two engineering objectives—tunnel crown displacement and haunch displacement—are predicted using the surrogate model, while a material-based support cost index is evaluated independently from the design variables. The selected compromise solution reported in the original analysis reduces the crown displacement, haunch displacement, and support cost index by 7.02%, 9.28%, and 12.84%, respectively, relative to the reported original scheme. Direct numerical analysis of the selected continuous optimum yields crown and haunch displacements of approximately 2.25 and 1.27 mm, respectively, consistent with the surrogate-based results. Although available field monitoring observations provide an additional engineering consistency check, the short monitoring duration and incomplete documentation of the underlying project data limit claims of full-scale causal validation. The main contribution of this study is therefore the integration of expensive numerical analysis, interpretable surrogate modeling, multi-objective optimization, and engineering-level verification for the investigated hybrid segment system.