Enhancing seismic performance of highway bridges group with laminated rubber bearings via artificial neural networks and multi-objective genetic algorithm

Maximizing the effectiveness of seismic isolation devices is crucial for enhancing bridge seismic performance at the regional level. This study proposes an intensive optimization framework for bridge seismic performance that accounts for structural size variations and ground motion uncertainties. Using Latin Hypercube Sampling (LHS), a representative group of 1000 bridges is generated for parametric finite element analysis. A high-precision, interpretable Artificial Neural Network (ANN) model is developed to predict the seismic response of bridges with laminated rubber bearings (LRBs) and identify crucial seismic isolation and geometric design parameters. The optimization process employs a Nondominated Sorting Genetic Algorithm II (NSGA-II) to provide multi-objective solutions tailored to various design strategies and earthquake intensities. The framework's effectiveness is validated through a bridge case study. The results indicate that the ANN prediction model achieves high accuracy, with R² values exceeding 0.84. SHapley Additive exPlanations (SHAP) analysis highlights that the key geometric features affecting seismic response are pier diameter (D p), pier height (H p), and reinforcement strength (f s). Critical seismic isolation parameters include the shear key gap (ΔGap), yield strength (f sk), and friction coefficient (μ b) of LRBs. It is recommended that seismic designs for small-to-medium span highway bridges incorporate an optimized shear key gap and reduce the friction coefficient of LRBs to enhance seismic isolation and ensure structural integrity under varying ground motion intensities. To facilitate practical application, the optimization framework is implemented into a graphical user interface (GUI), enabling engineers to select optimal design strategies and earthquake intensities based on specific project requirements.