Scholarlink

<p>Elevated reinforced concrete (RC) water tanks are critical infrastructure but remain seismically vulnerable due to fluid–structure interaction (FSI) complexities. Traditional finite element modeling (FEM) is computationally intensive, limiting rapid design iterations. This study proposes a novel machine learning (ML) framework to predict seismic responses of RC elevated tanks, explicitly capturing FSI dynamics. The main contribution of this work is developing and validating a computationally efficient ML framework that accurately predicts the complex seismic responses of elevated RC tanks, including base shear, overturning moment, and maximum displacement. This approach significantly outperforms traditional methods in speed while maintaining high accuracy. Using Housner’s Two-Mass Model integrated with SAP2000 simulations, a comprehensive dataset of 5,000 cases was generated, reflecting variability in geometry, staging, and ground motion. Eight ML models, including gradient-boosting algorithms and neural networks, are rigorously evaluated. Support Vector Regression (SVR) achieves exceptional accuracy (R<sup>2</sup> = 0.991 for base shear, R<sup>2</sup> = 0.988 for maximum displacement), reducing computation time from hours to seconds. Additionally, CatBoost emerges as an efficient alternative (R<sup>2</sup> = 0.978 for base shear, training time ≈ 28&#xa0;s), enabling real-time optimization. SHapley Additive exPlanations (SHAP) analysis identified tank slenderness (H/L) and Arias Intensity as key seismic risk drivers, offering actionable design insights. This framework enables rapid design iterations, quick vulnerability assessments, and potential real-time monitoring, while promoting sustainability through optimized material use and enhanced community resilience. </p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A machine learning framework for predicting seismic behavior in elevated reinforced concrete tanks

  • 期刊论文

  • 发表日期：

    2025年5月13日

查看全文
  • A. Aziz Al-Ayoubi,
  • Varatharajan Thirumurugan,
  • K. S. Satyanarayanan

摘要

Elevated reinforced concrete (RC) water tanks are critical infrastructure but remain seismically vulnerable due to fluid–structure interaction (FSI) complexities. Traditional finite element modeling (FEM) is computationally intensive, limiting rapid design iterations. This study proposes a novel machine learning (ML) framework to predict seismic responses of RC elevated tanks, explicitly capturing FSI dynamics. The main contribution of this work is developing and validating a computationally efficient ML framework that accurately predicts the complex seismic responses of elevated RC tanks, including base shear, overturning moment, and maximum displacement. This approach significantly outperforms traditional methods in speed while maintaining high accuracy. Using Housner’s Two-Mass Model integrated with SAP2000 simulations, a comprehensive dataset of 5,000 cases was generated, reflecting variability in geometry, staging, and ground motion. Eight ML models, including gradient-boosting algorithms and neural networks, are rigorously evaluated. Support Vector Regression (SVR) achieves exceptional accuracy (R2 = 0.991 for base shear, R2 = 0.988 for maximum displacement), reducing computation time from hours to seconds. Additionally, CatBoost emerges as an efficient alternative (R2 = 0.978 for base shear, training time ≈ 28 s), enabling real-time optimization. SHapley Additive exPlanations (SHAP) analysis identified tank slenderness (H/L) and Arias Intensity as key seismic risk drivers, offering actionable design insights. This framework enables rapid design iterations, quick vulnerability assessments, and potential real-time monitoring, while promoting sustainability through optimized material use and enhanced community resilience.