Scholarlink

<p>This paper discusses an advanced machine-learning model to predict the compressive strength (MPa) of Magnesium chloride-modified concrete. In this experiment, a constant 50% proportion of ordinary Portland cement was maintained, and the remaining 50% was replaced with magnesium chloride (MgCl<sub>2</sub>) and ground granulated blast furnace slag (GGBS) at various proportions. A dataset was designed with 10 input parameters, including quantity (kg/m3) of cement, GGBS, MgCl<sub>2</sub>, natural sand (N-sand), manufactured sand (M-sand), 10&#xa0;mm and 20&#xa0;mm coarse aggregate, water, age, and the desired output of compressive strength (MPa). The data was divided into train (70%), validation (15%) and test (15%). The four advanced machine learning algorithms are applied to predict the compressive strength of concrete: XGBoost (Extreme gradient boosting), NGBoost (Natural gradient boosting), GBDT (Gradient boosting decision tree), and Tab transformer. The performance of the model was assessed using the coefficient of determination (R<sup>2</sup>), Normalised mean squared error (NMSE) and Root mean squared error (RMSE). Visual comparison of model performance was presented using Taylor diagrams, and further interpretability was enhanced with Shapley additive explanations (SHAP) plots. The findings support the hypothesis that the proposed models help model the nonlinear relationships between input variables and optimise the design of sustainable concrete mixes. The results provide a sound, fact-based approach to developing sustainable SCM-based concrete, as they show that boosting-based models can describe nonlinear relationships among mix ingredients and curing age.</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

Deep tabular transformer and boosting models for accurate strength prediction of sustainable high-performance concrete using high dosage supplementary cementitious materials

  • 期刊论文

  • 发表日期：

    2026年1月8日

查看全文
  • B. Narendra Kumar,
  • P. Sai vineela,
  • Bhupesh Deka

摘要

This paper discusses an advanced machine-learning model to predict the compressive strength (MPa) of Magnesium chloride-modified concrete. In this experiment, a constant 50% proportion of ordinary Portland cement was maintained, and the remaining 50% was replaced with magnesium chloride (MgCl2) and ground granulated blast furnace slag (GGBS) at various proportions. A dataset was designed with 10 input parameters, including quantity (kg/m3) of cement, GGBS, MgCl2, natural sand (N-sand), manufactured sand (M-sand), 10 mm and 20 mm coarse aggregate, water, age, and the desired output of compressive strength (MPa). The data was divided into train (70%), validation (15%) and test (15%). The four advanced machine learning algorithms are applied to predict the compressive strength of concrete: XGBoost (Extreme gradient boosting), NGBoost (Natural gradient boosting), GBDT (Gradient boosting decision tree), and Tab transformer. The performance of the model was assessed using the coefficient of determination (R2), Normalised mean squared error (NMSE) and Root mean squared error (RMSE). Visual comparison of model performance was presented using Taylor diagrams, and further interpretability was enhanced with Shapley additive explanations (SHAP) plots. The findings support the hypothesis that the proposed models help model the nonlinear relationships between input variables and optimise the design of sustainable concrete mixes. The results provide a sound, fact-based approach to developing sustainable SCM-based concrete, as they show that boosting-based models can describe nonlinear relationships among mix ingredients and curing age.