Performance Optimization and Crack Control of UHPC-Ordinary Concrete Composite Tunnel Lining Structures Under High Stress and Seismic Loading

The increasing construction of tunnels in seismically active regions demands lining systems capable of withstanding combined static overburden pressures and dynamic ground motions without compromising waterproofing integrity through excessive cracking. While Ultra-High Performance Concrete (UHPC) offers exceptional mechanical properties—compressive strengths of 150–250 MPa and strain-hardening tensile behavior—its application as a composite lining material remains hindered by the absence of integrated optimization frameworks that balance crack control, structural capacity, and material economy. This paper presents a hybrid framework integrating Physics-Informed Neural Networks (PINNs) with Non-dominated Sorting Genetic Algorithm II (NSGA-II) for the multi-objective optimization of UHPC-ordinary concrete composite tunnel linings. A constitutive model-constrained PINN surrogate is developed to predict nonlinear seismic responses, achieving R² values of 0.987, 0.992, and 0.984 for crack width, displacement, and stress, respectively, with per-evaluation speedup exceeding 10⁶ relative to finite element simulations. The PINN replaces the expensive numerical solver within an NSGA-II optimization loop, reducing total computational cost by 95.7% while identifying Pareto-optimal designs. The novel functionally graded composite lining architecture features a UHPC secondary layer with engineered interfacial bonding to an ordinary concrete substrate. Optimization across 2,400 design configurations yields an optimal design at UHPC thickness 120 mm, fiber volume fraction 2.5%, and groove depth 15 mm, achieving 42.3% crack width reduction, 38.6% seismic capacity increase, and 42.0% cost reduction relative to a uniform UHPC alternative while maintaining a practical cost index of 1.42 relative to conventional C30 lining. The Pareto-optimal design outperforms eight baseline methods—including direct FEM-NSGA-II, GPR-NSGA-II, and XGBoost-NSGA-II—across all competing objectives. Crack widths remain below 0.10 mm at peak ground accelerations up to 0.50g, ensuring waterproofing integrity under severe seismic events. These results establish a computationally efficient pathway toward data-driven design of resilient underground infrastructure.