Scholarlink

<p>Traditional approaches for evaluating the mechanical properties of concrete incorporating untreated foundry sand (UFS) remain constrained by labor-intensive testing protocols and empirical approximations. This study proposes a hybrid framework that integrates systematic experimental validation with interpretable machine learning (ML) to predict the compressive strength (CS), split tensile strength (STS), and flexural strength (FS) of UFS-modified concrete, thereby enabling rapid and sustainable mix optimization. Experimental investigations were carried out on M20, M40, and M70 concrete grades, with fine aggregates replaced by UFS at 0–50%. A comprehensive dataset of 422 data points was generated from standardized tests on 54 cubes, 18 cylinders, and 18 beams after 28-day curing. Six ML algorithms—Ridge Regression, LASSO, Stochastic-M5P, Random Tree, Bagging, and Stochastic-Random Forest—were trained and benchmarked. The results demonstrate that 30% UFS replacement enhances CS (37.11&#xa0;MPa, + 34%), STS, and FS in M20 concrete, while M40 achieves peak strength at 20% UFS. Among the models, Stochastic-Random Forest attained near-perfect training accuracy for CS (<i>R</i> = 0.999, MAE = 0.067&#xa0;MPa). In contrast, Stochastic-Random Tree provided superior generalizability for STS (testing <i>R</i> = 0.95, MAE = 0.277&#xa0;MPa) and FS (testing <i>R</i> = 0.996, MAE = 0.544&#xa0;MPa). Sensitivity analysis revealed cement content, water–cement ratio, and curing duration as critical predictors, with UFS% exhibiting nonlinear effects on FS. By uniting experimental data with interpretable ML, this work advances the predictive design of eco-efficient concrete, mitigates the environmental burden of UFS disposal, and reduces dependence on natural aggregates. Furthermore, a graphical user interface (GUI) has been developed to translate the proposed framework into a practical decision-support tool for sustainable construction engineering.</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

Analyzing Compressive, Flexural, and Tensile Strength of Concrete Incorporating Used Foundry Sand: Experimental and Machine Learning Insights

  • 期刊论文

  • 发表日期：

    2025年9月28日

查看全文
  • Ganesh Shiva Sagar,
  • Shivanjali Mukthi,
  • Vikas Mehta

摘要

Traditional approaches for evaluating the mechanical properties of concrete incorporating untreated foundry sand (UFS) remain constrained by labor-intensive testing protocols and empirical approximations. This study proposes a hybrid framework that integrates systematic experimental validation with interpretable machine learning (ML) to predict the compressive strength (CS), split tensile strength (STS), and flexural strength (FS) of UFS-modified concrete, thereby enabling rapid and sustainable mix optimization. Experimental investigations were carried out on M20, M40, and M70 concrete grades, with fine aggregates replaced by UFS at 0–50%. A comprehensive dataset of 422 data points was generated from standardized tests on 54 cubes, 18 cylinders, and 18 beams after 28-day curing. Six ML algorithms—Ridge Regression, LASSO, Stochastic-M5P, Random Tree, Bagging, and Stochastic-Random Forest—were trained and benchmarked. The results demonstrate that 30% UFS replacement enhances CS (37.11 MPa, + 34%), STS, and FS in M20 concrete, while M40 achieves peak strength at 20% UFS. Among the models, Stochastic-Random Forest attained near-perfect training accuracy for CS (R = 0.999, MAE = 0.067 MPa). In contrast, Stochastic-Random Tree provided superior generalizability for STS (testing R = 0.95, MAE = 0.277 MPa) and FS (testing R = 0.996, MAE = 0.544 MPa). Sensitivity analysis revealed cement content, water–cement ratio, and curing duration as critical predictors, with UFS% exhibiting nonlinear effects on FS. By uniting experimental data with interpretable ML, this work advances the predictive design of eco-efficient concrete, mitigates the environmental burden of UFS disposal, and reduces dependence on natural aggregates. Furthermore, a graphical user interface (GUI) has been developed to translate the proposed framework into a practical decision-support tool for sustainable construction engineering.