Scholarlink

<p>Accurately predicting the failure of reinforced concrete (RC) structures under impact loads is critical for ensuring structural safety and resilience. This study addresses this challenge by developing a robust machine learning framework to classify impact damage in RC panels into four primary modes: no damage, penetration, scabbing, and perforation, using a diverse experimental dataset of 254 tests. To solve this classification problem, a suite of ensemble models was developed and evaluated, primarily a Separate Stacking Model (SSM) architecture utilizing Deep Artificial Neural Networks (DANNs) as base learners. The framework’s performance was benchmarked against standard algorithms, and model interpretability was achieved using the SHAP (SHapley Additive exPlanations) framework. Using a rank-sum aggregation across different metrics, the SSM-AdaBoost and SSM-GaussianNB models emerged as top-performing. Furthermore, SHAP analysis identified panel geometry and impactor characteristics as the most significant physical drivers of failure. In conclusion, this work presents a highly accurate and interpretable predictive tool. The integration of optimized ensemble methods with explainability provides a reliable solution that not only enhances predictive capabilities but also offers actionable insights, contributing to the safer and more efficient design of impact-resistant structures.</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

Ensemble deep neural network model for failure mode prediction of reinforced concrete panels under impact loads

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月21日

查看全文
  • Mohammad Sadegh Barkhordari,
  • Duc-Kien Thai,
  • Shekufe Khoshnazar,
  • Quoc Hoan Doan

摘要

Accurately predicting the failure of reinforced concrete (RC) structures under impact loads is critical for ensuring structural safety and resilience. This study addresses this challenge by developing a robust machine learning framework to classify impact damage in RC panels into four primary modes: no damage, penetration, scabbing, and perforation, using a diverse experimental dataset of 254 tests. To solve this classification problem, a suite of ensemble models was developed and evaluated, primarily a Separate Stacking Model (SSM) architecture utilizing Deep Artificial Neural Networks (DANNs) as base learners. The framework’s performance was benchmarked against standard algorithms, and model interpretability was achieved using the SHAP (SHapley Additive exPlanations) framework. Using a rank-sum aggregation across different metrics, the SSM-AdaBoost and SSM-GaussianNB models emerged as top-performing. Furthermore, SHAP analysis identified panel geometry and impactor characteristics as the most significant physical drivers of failure. In conclusion, this work presents a highly accurate and interpretable predictive tool. The integration of optimized ensemble methods with explainability provides a reliable solution that not only enhances predictive capabilities but also offers actionable insights, contributing to the safer and more efficient design of impact-resistant structures.