Scholarlink

<p>Polyethylene fiber-based engineered cementitious composites (PE-ECC) offer exceptional compressive strength (CS), tensile strength (TS), ductility, crack control, and durability properties. Their mechanical performance depends on highly nonlinear interactions among the matrix composition, fiber geometry, and curing conditions. Accurately capturing these interactions using conventional experimental approaches is time-consuming and costly. This study developed a comprehensive and interpretable machine learning (ML) framework to predict the CS and TS of PE-ECC. An experimental database was constructed from published experimental studies incorporating the mixture proportions, fiber characteristics, and curing parameters. Six ML models (e.g., DT, RF, SVR, XGBoost, LightGBM, and CatBoost) were trained and optimized using systematic data preprocessing, outlier removal, feature standardization, and grid-search-based hyperparameter tuning. The model performance was rigorously evaluated using (e.g., R<sup>2</sup>, Adj. R<sup>2</sup>, MAE, RMSE, MSE, and SMAPE) supported by fluctuation, distribution, residual, and radar-based multi-metric analyses. Among the evaluated models, XGBoost demonstrated superior predictive accuracy and stability, achieving testing R<sup>2</sup> values of 0.93 and 0.90 for CS and TS, respectively. XGBoost proved to be the most effective method for this task, minimizing the prediction errors while capturing the nonlinear micromechanical interactions of PE-ECC more effectively than other ensemble and individual learners. SHapley Additive exPlanations (SHAP) was employed to ensure model interpretability and identify governing parameters. The results indicate that CS is primarily controlled by matrix-related variables, particularly the water-to-binder ratio (W/B), whereas TS is dominated by fiber-related parameters such as fiber length (L<sub>f</sub>) and the GGBS-to-binder ratio (GS/B). To facilitate practical implementation, an interactive graphical user interface (GUI) was developed, enabling the rapid and reliable prediction of the mechanical properties of PE-ECC without coding or extensive laboratory testing. The proposed framework provides a robust, interpretable, and application-oriented tool for accelerating PE-ECC mix design, reducing experimental effort, and supporting the development of durable and sustainable cementitious composites.</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

Data-driven modeling of mechanical properties in polyethylene fiber-based engineered cementitious composites (PE-ECC)

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Md Mehedi Hasan,
  • Md Nasir Uddin,
  • Md Sagar Akon,
  • Abba Bashir,
  • Borhan Uddin Rabbani

摘要

Polyethylene fiber-based engineered cementitious composites (PE-ECC) offer exceptional compressive strength (CS), tensile strength (TS), ductility, crack control, and durability properties. Their mechanical performance depends on highly nonlinear interactions among the matrix composition, fiber geometry, and curing conditions. Accurately capturing these interactions using conventional experimental approaches is time-consuming and costly. This study developed a comprehensive and interpretable machine learning (ML) framework to predict the CS and TS of PE-ECC. An experimental database was constructed from published experimental studies incorporating the mixture proportions, fiber characteristics, and curing parameters. Six ML models (e.g., DT, RF, SVR, XGBoost, LightGBM, and CatBoost) were trained and optimized using systematic data preprocessing, outlier removal, feature standardization, and grid-search-based hyperparameter tuning. The model performance was rigorously evaluated using (e.g., R2, Adj. R2, MAE, RMSE, MSE, and SMAPE) supported by fluctuation, distribution, residual, and radar-based multi-metric analyses. Among the evaluated models, XGBoost demonstrated superior predictive accuracy and stability, achieving testing R2 values of 0.93 and 0.90 for CS and TS, respectively. XGBoost proved to be the most effective method for this task, minimizing the prediction errors while capturing the nonlinear micromechanical interactions of PE-ECC more effectively than other ensemble and individual learners. SHapley Additive exPlanations (SHAP) was employed to ensure model interpretability and identify governing parameters. The results indicate that CS is primarily controlled by matrix-related variables, particularly the water-to-binder ratio (W/B), whereas TS is dominated by fiber-related parameters such as fiber length (Lf) and the GGBS-to-binder ratio (GS/B). To facilitate practical implementation, an interactive graphical user interface (GUI) was developed, enabling the rapid and reliable prediction of the mechanical properties of PE-ECC without coding or extensive laboratory testing. The proposed framework provides a robust, interpretable, and application-oriented tool for accelerating PE-ECC mix design, reducing experimental effort, and supporting the development of durable and sustainable cementitious composites.