Scholarlink

<p>Sustainable concrete design, with increased use of cementitious materials and microfines, requires reliable predictions of mechanical properties. The three fundamental mechanical characteristics of concrete, including compressive strength, split tensile strength, and flexural strength, can be estimated simultaneously with the aid of the information-supported approach suggested in the current research. The experimental database, based on seven controlled mix designs, incorporated eight key input variables: cement, ground granulated blast-furnace slag (GGBS), microfines, manufactured sand (M-sand), coarse aggregates (10&#xa0;mm and 20&#xa0;mm), water content, and curing age. Model performance was evaluated using a mix-wise cross-validation strategy to prevent data leakage and ensure unbiased generalization. The two advanced machine learning algorithms, TabTransformer and Natural Gradient Boosting (NGBoost), enable the model results to be easily interpreted in terms of material behaviour and evaluated using the coefficient of determination (R<sup>2</sup>), normalised mean squared error (NMSE), and root mean squared error (RMSE). The interpretability of the model was measured with Shapley Additive Explanations (SHAP) that quantify the contribution of the individual attributes. Whereas NGBoost achieved superior predictive performance on the test data, with R<sup>2</sup> values of 0.995 ± 0.003 for compressive strength, 0.903 for split tensile strength, and 0.924 for flexural strength, it showed more stable convergence across all strength parameters than Tabtransformer, which also exhibited strong predictive performance. Moreover, the interpretability analysis revealed that the two most significant variables for determining strength gain were curing age and water-to-binder ratio. To minimize experimental burden and enable the design of sustainable concrete mixtures informed by the data, the proposed framework provides a reliable and interpretable approach to multi-property strength prediction.</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

Tripartite strength prediction of sustainable micro fine based concrete using tabular deep learning and probabilistic boosting frameworks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月28日

查看全文
  • B. Narendra Kumar,
  • Sai Vineela,
  • Sathvik Sharath Chandra,
  • Somula Ramasubbareddy

摘要

Sustainable concrete design, with increased use of cementitious materials and microfines, requires reliable predictions of mechanical properties. The three fundamental mechanical characteristics of concrete, including compressive strength, split tensile strength, and flexural strength, can be estimated simultaneously with the aid of the information-supported approach suggested in the current research. The experimental database, based on seven controlled mix designs, incorporated eight key input variables: cement, ground granulated blast-furnace slag (GGBS), microfines, manufactured sand (M-sand), coarse aggregates (10 mm and 20 mm), water content, and curing age. Model performance was evaluated using a mix-wise cross-validation strategy to prevent data leakage and ensure unbiased generalization. The two advanced machine learning algorithms, TabTransformer and Natural Gradient Boosting (NGBoost), enable the model results to be easily interpreted in terms of material behaviour and evaluated using the coefficient of determination (R2), normalised mean squared error (NMSE), and root mean squared error (RMSE). The interpretability of the model was measured with Shapley Additive Explanations (SHAP) that quantify the contribution of the individual attributes. Whereas NGBoost achieved superior predictive performance on the test data, with R2 values of 0.995 ± 0.003 for compressive strength, 0.903 for split tensile strength, and 0.924 for flexural strength, it showed more stable convergence across all strength parameters than Tabtransformer, which also exhibited strong predictive performance. Moreover, the interpretability analysis revealed that the two most significant variables for determining strength gain were curing age and water-to-binder ratio. To minimize experimental burden and enable the design of sustainable concrete mixtures informed by the data, the proposed framework provides a reliable and interpretable approach to multi-property strength prediction.