Leveraging Geospatial Technologies and AI for Synthesizing Earthquake-Tsunami Fragility Functions in Infrastructure Digital Twins

Abstract

This study presents a novel machine learning (ML)-based framework for synthesizing three-dimensional (3D) fragility surfaces for earthquake-tsunami multi-hazard analysis. Traditional fragility functions rely on computationally intensive finite element (FE) simulations or empirical models, limiting scalability for multi-hazard applications. The proposed ML model integrates independent two-dimensional (2D) fragility curves for earthquake and tsunami hazards using Random Forest and Extreme Gradient Boosting algorithms, efficiently generating synthesized fragility surfaces with high predictive accuracy. By incorporating geospatial technologies and artificial intelligence (AI), the framework enables real-time hazard and damage assessment, supporting resilience planning and mitigation strategies. The study further explores applications in structural retrofitting, where ML-generated fragility surfaces inform quantitative evaluations of retrofitting strategies for multi-hazard resilience. The approach is validated using the Pseudo Seaside testbed, demonstrating its capability in community-scale risk assessments.