Scholarlink

<p>A major challenge in external beam-column joint (BCJ) design is reinforcement congestion in the joint area, which hinders concrete pouring and weakens the joint, reducing its capacity. This study investigates the use of engineered cementitious composites (ECCs) combined with two reinforcement stirrups as an optimized solution to mitigate congestion while maintaining structural integrity. The selection of two stirrups was based on achieving an optimal balance between structural performance, reinforcement congestion mitigation, and BCJ capacity, which could not be effectively achieved with either full or no stirrups. To evaluate BCJ performance, finite element modeling (FEM) and machine learning (ML) algorithms were utilized. FEM simulations examined various ECC-stirrup configurations, revealing that introducing ECC in the joint and extending it 457.2 mm into the beam effectively relocated plastic hinges, promoted ductile failure, and prevented catastrophic structural collapse. Notably, the FEM-9 model demonstrated superior ductility, confirming that targeted ECC placement significantly enhances BCJ capacity. The application of ECC increased peak load capacity by up to 29.5%, demonstrating its effectiveness in BCJ strengthening. Furthermore, this study systematically evaluates seven ML models using a dataset of 220 data points generated from a parametric study, using the optimized FEM-9 configuration as the baseline model. Models’ performance was assessed using the coefficient of determination (R<sup>2</sup>) and other statistical measures. To enhance interpretability, Shapley Additive Explanations (SHAP) and Partial Dependence Plots (PDPs) were employed, revealing key input–output relationships. Results indicate that AdaBoost outperforms other models, achieving R<sup>2</sup> values of 0.998 for training and 0.976 for testing, making it the most effective predictive model for BCJ performance.</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

Strengthening and predictive design of ECC-reinforced beam–column joints using FEM and explainable machine learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月31日

查看全文
  • Fayiz Amin,
  • Mehran Sahil,
  • Abdulmoez Al Ismaeel,
  • Muhammad Faisal Niaz,
  • Hafiz Ahmed Waqas,
  • Khan Abdul Majid,
  • Saed A. M. Nassar

摘要

A major challenge in external beam-column joint (BCJ) design is reinforcement congestion in the joint area, which hinders concrete pouring and weakens the joint, reducing its capacity. This study investigates the use of engineered cementitious composites (ECCs) combined with two reinforcement stirrups as an optimized solution to mitigate congestion while maintaining structural integrity. The selection of two stirrups was based on achieving an optimal balance between structural performance, reinforcement congestion mitigation, and BCJ capacity, which could not be effectively achieved with either full or no stirrups. To evaluate BCJ performance, finite element modeling (FEM) and machine learning (ML) algorithms were utilized. FEM simulations examined various ECC-stirrup configurations, revealing that introducing ECC in the joint and extending it 457.2 mm into the beam effectively relocated plastic hinges, promoted ductile failure, and prevented catastrophic structural collapse. Notably, the FEM-9 model demonstrated superior ductility, confirming that targeted ECC placement significantly enhances BCJ capacity. The application of ECC increased peak load capacity by up to 29.5%, demonstrating its effectiveness in BCJ strengthening. Furthermore, this study systematically evaluates seven ML models using a dataset of 220 data points generated from a parametric study, using the optimized FEM-9 configuration as the baseline model. Models’ performance was assessed using the coefficient of determination (R2) and other statistical measures. To enhance interpretability, Shapley Additive Explanations (SHAP) and Partial Dependence Plots (PDPs) were employed, revealing key input–output relationships. Results indicate that AdaBoost outperforms other models, achieving R2 values of 0.998 for training and 0.976 for testing, making it the most effective predictive model for BCJ performance.