Machine learning–based seismic resilience optimization and rapid assessment of reinforced cement concrete bridge

Abstract

After an earthquake, resilient transportation planners must estimate reinforced cement concrete bridge structural safety, recovery time, economic losses, and long-term sustainability under tight time and data restrictions. Traditional seismic assessment frameworks use nonlinear time-history analysis, simplified fragility curves, and unconnected loss or life-cycle models, which increase computational cost, fragment uncertainty treatment, resilience, cost, and environmental performance optimization. Current machine learning methods are faster but disregard system-level interconnections, recovery dynamics, and sustainability implications, restricting their application for policy-level decision-making. To address these limits, this study provides a unified machine learning–based prediction and optimization method for rapid RCC bridge seismic resistance, post-earthquake losses, and sustainability metrics assessment. The six-block analytical process ensures data continuity. First, a SeisBridge Diffusion Surrogate (SB-DS) imitates nonlinear seismic response utilizing graph-based structural representations and conditional diffusion models to reliably and uncertainty-awarely estimate engineering demand parameters at low computational Second, topology-aware message transmission utilizing a Graph Fragility and Damage Inference Network (G-FraDIN) translates these demands into component- and system-level damage state probabilities, surpassing standalone component fragility models. Third, for risk assessment, a Recovery–Loss Multi-Task Emulator with Conformal Uncertainty (ReLo-ConfMT) predicts repair time, direct and indirect losses, and functionality trajectories with statistically calibrated prediction interval Fourth, the hybrid ML–LCA HyLiSE-Bridge life-cycle carbon and energy estimates include seismic damage and repair trajectories. Fifth, a Resilience–Loss–Sustainability Multi-Objective Bayesian Optimizer (ReLoS-MOBO) discovers Pareto-optimal bridge designs and retrofits that balance resilience, economic risk, and environmental performance The Scenario-Consistent Policy Evaluator with Explainable Learning (SCOPE-XL) finds robust design solutions under stakeholder requirements and openly interprets key design drivers. The proposed paradigm decreases computational effort by orders of magnitude, improves prediction accuracy and uncertainty consistency, and offers resilience–sustainability trade-offs. Its integrated and interpretable nature aids emergency response planning, infrastructure investment, and sustainable bridge earthquake designs.