Scholarlink

<p>Accurate prediction of the compressive strength of high-performance concrete (UHPC) is crucially important for mix optimization, structural safety, and low-cost in the experiments. Traditional approaches adopted in regression-based methods may not be sufficiently adequate to quantify the nonlinear and multivariate interactions of the concrete constituents, and therefore, there is a need for robust predictive frameworks. This work investigates the use of three state-of-the-art machine learning models, Gradient Boosting Machines (GBM), Random Forests (RF), and Deep Neural Networks (DNN), to predict compressive strength of UHPC based on a data set of 1030 standardized samples that include cement, blast furnace slag, fly ash, water, superplasticizer, fine and coarse aggregates and curing-age as input variables. Data preprocessing was performed with min-max normalization, which was followed by systematic training and test phases. Model performance was evaluated using several statistical indicators, i.e., R<sup>2</sup>, RMSE, MAE, Willmott’s index, and weighted mean absolute percentage error, and also graphical tools such as Regression Error Characteristic (REC) curves, Taylor diagrams, and ranking analysis. The results demonstrated that GBM achieved the highest accuracy during training (R² = 0.999, RMSE = 0.649) and maintained superior generalization in testing (R² = 0.910). RF provided competitive performance, while DNN, though comparatively weaker, effectively captured nonlinear patterns. Interpretability analyses using SHAP, LIME, and Partial Dependence Plots confirmed the influence of admixtures, fly ash, and curing age as dominant factors governing strength. Overall, the study establishes RF as a dependable tool for UHPC strength prediction while underscoring the potential of ensemble learning to enhance data-driven decision-making in modern concrete 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

Data-driven prediction of ultra-high-performance concrete compressive strength using ensemble and neural network models

  • 期刊论文

  • 发表日期：

    2025年12月5日

查看全文
  • Sushila Sharma,
  • Avijit Burman,
  • Pijush Samui

摘要

Accurate prediction of the compressive strength of high-performance concrete (UHPC) is crucially important for mix optimization, structural safety, and low-cost in the experiments. Traditional approaches adopted in regression-based methods may not be sufficiently adequate to quantify the nonlinear and multivariate interactions of the concrete constituents, and therefore, there is a need for robust predictive frameworks. This work investigates the use of three state-of-the-art machine learning models, Gradient Boosting Machines (GBM), Random Forests (RF), and Deep Neural Networks (DNN), to predict compressive strength of UHPC based on a data set of 1030 standardized samples that include cement, blast furnace slag, fly ash, water, superplasticizer, fine and coarse aggregates and curing-age as input variables. Data preprocessing was performed with min-max normalization, which was followed by systematic training and test phases. Model performance was evaluated using several statistical indicators, i.e., R2, RMSE, MAE, Willmott’s index, and weighted mean absolute percentage error, and also graphical tools such as Regression Error Characteristic (REC) curves, Taylor diagrams, and ranking analysis. The results demonstrated that GBM achieved the highest accuracy during training (R² = 0.999, RMSE = 0.649) and maintained superior generalization in testing (R² = 0.910). RF provided competitive performance, while DNN, though comparatively weaker, effectively captured nonlinear patterns. Interpretability analyses using SHAP, LIME, and Partial Dependence Plots confirmed the influence of admixtures, fly ash, and curing age as dominant factors governing strength. Overall, the study establishes RF as a dependable tool for UHPC strength prediction while underscoring the potential of ensemble learning to enhance data-driven decision-making in modern concrete engineering.