Scholarlink

<p>Crack propagation in reinforced concrete structures poses significant challenges to structural integrity, safety, and long-term durability. Traditional fracture mechanics approaches, such as Paris’ Law and linear elastic fracture mechanics (LEFM), often fail to capture the complex interactions of structural, material, and environmental variables observed in real-world conditions. This study presents a comparative analysis of classical and ensemble machine learning (ML) models for predicting crack propagation rate (CPR) in concrete elements. A dataset comprising 1200 samples collected from laboratory tests and field monitoring was used to train and validate seven supervised learning algorithms: multiple linear regression (MLR), decision tree regression (DTR), artificial neural networks (ANN), random forest regression (RFR), stochastic gradient boosting (SGB), CatBoost, and XGBoost. Feature selection methods, including recursive feature elimination and SHAP (SHapley Additive exPlanations), identified initial crack width, load cycles, and load magnitude as the most influential predictors. XGBoost outperformed all other models, achieving the highest R<sup>2</sup> (0.91), accuracy (92.1%), and robustness against noise and small data sizes. SHAP analysis also enhanced model interpretability, a key factor for engineering adoption. The proposed ML framework offers a scalable, accurate, and explainable solution for real-time structural health monitoring, supporting proactive maintenance and infrastructure 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

Predicting crack propagation rate in reinforced concrete structures using classical and ensemble machine learning with SHAP-based interpretability

  • 期刊论文

  • 发表日期：

    2025年11月3日

查看全文
  • Md Ebad,
  • Sarvesh Vyas

摘要

Crack propagation in reinforced concrete structures poses significant challenges to structural integrity, safety, and long-term durability. Traditional fracture mechanics approaches, such as Paris’ Law and linear elastic fracture mechanics (LEFM), often fail to capture the complex interactions of structural, material, and environmental variables observed in real-world conditions. This study presents a comparative analysis of classical and ensemble machine learning (ML) models for predicting crack propagation rate (CPR) in concrete elements. A dataset comprising 1200 samples collected from laboratory tests and field monitoring was used to train and validate seven supervised learning algorithms: multiple linear regression (MLR), decision tree regression (DTR), artificial neural networks (ANN), random forest regression (RFR), stochastic gradient boosting (SGB), CatBoost, and XGBoost. Feature selection methods, including recursive feature elimination and SHAP (SHapley Additive exPlanations), identified initial crack width, load cycles, and load magnitude as the most influential predictors. XGBoost outperformed all other models, achieving the highest R2 (0.91), accuracy (92.1%), and robustness against noise and small data sizes. SHAP analysis also enhanced model interpretability, a key factor for engineering adoption. The proposed ML framework offers a scalable, accurate, and explainable solution for real-time structural health monitoring, supporting proactive maintenance and infrastructure resilience.