Beyond Fragility: Physics-Driven Neural Surrogates for Seismic Resilience Prediction of Bridges

Jacob Atkins¹Donya Hajializadeh²Waqas Iqbal¹Farahnaz Soleimani^1*

• ¹Oregon State University, Corvallis, United States
• ²University of Surrey, Guildford, United Kingdom

Traditional fragility-based methods are rigorous but can be computationally intensive and difficult to scale across large bridge inventories, particularly when resilience assessments require propagating fragility outputs through functionality and recovery models for time-dependent decision support. This study introduces a physics-driven neural surrogate framework that complements fragility-informed workflows by directly predicting the seismic resilience index of bridges as a continuous, system-level metric. Using pre-1971 concrete box-girder bridges as a case study to illustrate the proposed approach, the framework integrates high-fidelity nonlinear time-history simulation performed in OpenSees with a multilayer perceptron (MLP) neural network. The simulation-informed dataset captures complex structural and seismic interactions across 1,600 bridge–ground motion scenarios. A systematic hyperparameter tuning process, spanning loss functions, optimizers, network depth, and regularization, is employed to optimize model performance. The final MLP architecture achieves over 97% prediction accuracy and significantly outperforms ensemble learning (EL) models. By learning directly from physics-based simulations, the surrogate model enables rapid, scalable, and interpretable resilience estimation, supporting efficient retrofit prioritization, emergency planning, and resilience-informed design in seismically active regions.