Machine learning-based framework for rapid assessment of seismic resilience and sustainability metrics for regional RC bridges

Abstract

The seismic resilience and environmental impact of reinforced concrete (RC) bridges during earthquakes are crucial for maintaining the functionality and sustainability of transportation networks in earthquake-prone regions. This study presents a machine learning (ML)-based framework that integrates performance-based earthquake engineering (PBEE) principles and ML models to rapidly assess seismic resilience and post-earthquake losses, including repair time, repair costs, and sustainability metrics quantified by carbon footprint, for regional RC bridges. Based on twelve bridge key attributes, including column height and diameter, 1000 finite element (FE) bridge models are systematically developed through the Latin Hypercube Sampling (LHS) method and subjected to 100 ground motions to compute probabilistic seismic demand models, system-level fragility, seismic resilience, and post-earthquake losses. Through hyperparameter tuning and k-fold cross-validation, six ML models are optimized with the artificial neural network (ANN) achieving superior accuracy in predicting seismic resilience. Subsequently, the developed ANN framework is applied to representative regional RC bridges, facilitating rapid and reliable predictions of seismic resilience and post-earthquake losses across varying bridge attributes. Overall, the developed framework serves as an efficient and practical tool for decision-makers, providing valuable insights to enhance seismic resilience and sustainability metrics while optimizing post-earthquake recovery strategies for critical infrastructure.