Explainable machine learning (XML) framework for seismic assessment of structures using Extreme Gradient Boosting (XGBoost)

초록

In the field of structural engineering, optimizing retrofitting strategies for bolstering seismic resilience stands as a pressing challenge. Existing methods are often limited by the time-intensive nature of nonlinear time history (NLTH) analysis and the lack of transparency in machine learning (ML) techniques. This study presents an innovative framework for optimizing retrofitting strategies in structural engineering to enhance seismic resilience. The framework integrates eXtreme Gradient Boosting (XGBoost) and the Spring-Rotational Friction Damper (SRFD) retrofit system, known for its ability to dissipate seismic energy and incorporate self-centering mechanisms. The approach improves transparency in machine learning processes and streamlines design optimization. It uses eXplainable Artificial Intelligence (XAI) methods, such as SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME), to provide insights into model predictions and ensure clear decision-making processes. The framework uses data-driven optimization to tailor design parameters to specific seismic hazards, enhancing seismic resilience. Its accuracy was validated through a comprehensive analysis, showing low residual errors, favorable learning curves, and a mean squared error (MSE) of 0.00142. The research evaluates the framework using 2D and 3D case study structures, comparing metrics like maximum displacement, residual drift, maximum inter-story drift (MIDR), and energy dissipation. The seismic performance evaluation confirmed the effectiveness of the design procedure for determining optimal retrofit system parameters estimated by the eXplainable Machine Learning (XML) framework. This represents a significant advancement in seismic assessment methodologies, enabling engineers to make informed decisions about building safety and promoting the adoption of ML-based approaches in earthquake engineering. © 2025 The Authors

키워드