Scholarlink

<p>This research develops a hybrid approach of experiment with the calculation of strength development in GGBS (Ground Granulated Blast Furnace Slag) and Hydrated lime modified concrete. A designed experimental programme was carried out under a separate laboratory set up, where the proportioning, casting, curing and mechanical characterising of 105 concrete datapoints were conducted to produce high fidelity data on compressive, split tensile and flexural strengths. The dataset consists of 10 input features like cement, GGBS, Hydrated lime, M-sand, N-sand, 20&#xa0;mm aggregate, 10&#xa0;mm aggregate, water, age and output as target strength. These datasets were then modelled with the help of Random Forest (RF) and Natural Gradient Boosting (NG Boost) in order to solve the unknown nonlinear and multi-parameter interactions that determined the development of strengths in blended binder systems. The data is divided into 60% for training, 20% for validation and 20% for testing. RF yielded high predictive fidelity of compressive strength with R<sup>2</sup> of 0.9901 (training), 0.9811 (validation), 0.9700 (testing), with RMSE of 1.57–3.47&#xa0;MPa, and NG Boost demonstrated the same accuracy. Split tensile strength proved to be highly predictive with RF yielding R<sup>2</sup> = 0.9912−0.9903 and NG Boost yielding R<sup>2</sup> = 0.9933–0.9944 (RMSE = 0.158–0.248&#xa0;MPa). Although more susceptible to material variability, flexural strength was well modelled within an R<sup>2</sup> range of 0.785–0.936 and RMSE of 0.155–0.347&#xa0;MPa. Additionally, 5-Fold cross-validation is performed to avoid the overfitting of data. A brief SHAP analysis showed the age of curing and the binder content to be the most significant predictors, but radar-metric analysis confirmed the consistency of generalisation among all measures of strength. The findings confirm that the integration of obtained lab data and sophisticated ensemble as well as probabilistic models gives a clear and dependable avenue towards prediction and optimisation of the performance of low-carbon cementitious structures.</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

Predictive modelling of nonlinear strength development in GGBS-hydrated lime concrete using ensemble and probabilistic learning frameworks

  • 期刊论文

  • 发表日期：

    2026年2月17日

查看全文
  • B. Narendra Kumar,
  • C. Nanda Kishore

摘要

This research develops a hybrid approach of experiment with the calculation of strength development in GGBS (Ground Granulated Blast Furnace Slag) and Hydrated lime modified concrete. A designed experimental programme was carried out under a separate laboratory set up, where the proportioning, casting, curing and mechanical characterising of 105 concrete datapoints were conducted to produce high fidelity data on compressive, split tensile and flexural strengths. The dataset consists of 10 input features like cement, GGBS, Hydrated lime, M-sand, N-sand, 20 mm aggregate, 10 mm aggregate, water, age and output as target strength. These datasets were then modelled with the help of Random Forest (RF) and Natural Gradient Boosting (NG Boost) in order to solve the unknown nonlinear and multi-parameter interactions that determined the development of strengths in blended binder systems. The data is divided into 60% for training, 20% for validation and 20% for testing. RF yielded high predictive fidelity of compressive strength with R2 of 0.9901 (training), 0.9811 (validation), 0.9700 (testing), with RMSE of 1.57–3.47 MPa, and NG Boost demonstrated the same accuracy. Split tensile strength proved to be highly predictive with RF yielding R2 = 0.9912−0.9903 and NG Boost yielding R2 = 0.9933–0.9944 (RMSE = 0.158–0.248 MPa). Although more susceptible to material variability, flexural strength was well modelled within an R2 range of 0.785–0.936 and RMSE of 0.155–0.347 MPa. Additionally, 5-Fold cross-validation is performed to avoid the overfitting of data. A brief SHAP analysis showed the age of curing and the binder content to be the most significant predictors, but radar-metric analysis confirmed the consistency of generalisation among all measures of strength. The findings confirm that the integration of obtained lab data and sophisticated ensemble as well as probabilistic models gives a clear and dependable avenue towards prediction and optimisation of the performance of low-carbon cementitious structures.