Machine learning models for seismic analysis of buckling-restrained braced frames

• Published in: Journal of Building Engineering 2024-12, Vol.98, p.111398, Article 111398
• Main Authors: Anand, T.P., Pandikkadavath, Muhamed Safeer, Mangalathu, Sujith, Sahoo, Dipti Ranjan
• Format: Article
• Language: English
• Subjects: Accumulated local effect; Buckling restrained braces; Inter-storey drift ratio; Machine learning; Nonlinear responses history analysis; Residual drift ratio; Shapley additive explanations
• ISSN: 2352-7102; 2352-7102
• DOI: 10.1016/j.jobe.2024.111398

The current research addresses the need for efficient analysis, design, and post-earthquake damage assessment of buckling-restrained braced frames by developing machine learning (ML) models to predict multiple engineering demand parameters (EDPs) under various seismic loading conditions. To this end, a database of 16,694 records is formulated using numerical simulations. Then, nine supervised ML algorithms are optimised through hyperparameter tuning and validated to identify the most effective prediction model. The extreme gradient boosting (XGBoost) model demonstrated superior performance in estimating the inter-storey drift ratio (IDR), residual drift ratio (RDR), maximum ductility demand (μmax), and cumulative ductility demand (μcum). Consequently, a user-friendly graphical user interface is devised for its seamless implementation. Finally, interpretable ML techniques, such as Shapley additive explanations (SHAP) and accumulated local effects, are applied to the XGBoost model to discern key input parameters and prediction trends. Pseudo-spectral acceleration at 2.0 s is identified as the most influential variable for predicting IDR, RDR, and μmax, while Arias intensity is the most significant for predicting μcum. The top-ranking earthquake parameters identified through SHAP aid structural designers in assessing optimal intensity measures for fragility analysis. •Creation of an open-source database that links structural and seismic features of BRBFs with engineering demand parameters (EDPs).•Demonstration of the effectiveness of XGBoost model in predicting multiple EDPs.•Utilisation of SHAP to identify potential intensity measures in fragility analysis for all four EDPs.•Application of ALE to identify key input parameter prediction trends and threshold values.