Scholarlink

<p>This research was aimed at systematically comparing the advanced machine learning models to predict the compressive strength (MPa) of concrete mixes on a dataset that involved 11 relevant factors including cement (kg/m<sup>3</sup>), ground granulated blast-furnace slag [GGBS] (kg/m<sup>3</sup>), bentonite (kg/m<sup>3</sup>), calcite (kg/m<sup>3</sup>), manufactured sand [M-sand] (kg/m<sup>3</sup>), natural sand [N-sand] (kg/m<sup>3</sup>), coarse aggregate (20&#xa0;mm &amp; 10&#xa0;mm), water, age, and compressive strength. In this experiment, 50% of the cement was replaced with Supplementary Cementitious Materials (SCMs). Within this replacement, Ground Granulated Blast Furnace Slag (GGBS) was used as the major constituent with replacement levels of up to 50%, while bentonite was varied up to 20% and calcite up to 7.5%. These proportions were systematically adjusted across the 144 mixes to study their individual and combined effects on compressive strength and were produced in a continuous laboratory program, giving the experimental data consistency and reliability. The data was divided between test (20%), training (60%) and validation (20%). The results obtained using four models, namely Deep Neural Network (DNN), Recurrent Neural Network (RNN), Liquid Neural Network (LNN) and Extreme Gradient Boosting (XGBoost), were discussed in order to compare their performance. Model performance was determined by the coefficient of determination (R<sup>2</sup>), normalised mean squared error (NMSE) and the mean absolute percentage error (MAPE). XGBoost has been remarkable in its training prediction with an R<sup>2</sup> value of 0.977 and an MAPE value of 2.47%. Meanwhile, RNN appeared the most reliable and the most transferable in terms of all data partitions’ performance, having validation and testing R<sup>2</sup> values of 0.921 and 0.884, respectively, as well as the lowest testing MAPE of 6.22%.</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

Comparative analysis of advanced machine learning models for predicting compressive strength of Supplementary Cementitious Materials (SCM) based concrete

  • 期刊论文

  • 发表日期：

    2025年9月2日

查看全文
  • B. Narendra Kumar,
  • A. Meghanadha Reddy,
  • Sayanti Chatterjee,
  • Gokulan Ravindiran

摘要

This research was aimed at systematically comparing the advanced machine learning models to predict the compressive strength (MPa) of concrete mixes on a dataset that involved 11 relevant factors including cement (kg/m3), ground granulated blast-furnace slag [GGBS] (kg/m3), bentonite (kg/m3), calcite (kg/m3), manufactured sand [M-sand] (kg/m3), natural sand [N-sand] (kg/m3), coarse aggregate (20 mm & 10 mm), water, age, and compressive strength. In this experiment, 50% of the cement was replaced with Supplementary Cementitious Materials (SCMs). Within this replacement, Ground Granulated Blast Furnace Slag (GGBS) was used as the major constituent with replacement levels of up to 50%, while bentonite was varied up to 20% and calcite up to 7.5%. These proportions were systematically adjusted across the 144 mixes to study their individual and combined effects on compressive strength and were produced in a continuous laboratory program, giving the experimental data consistency and reliability. The data was divided between test (20%), training (60%) and validation (20%). The results obtained using four models, namely Deep Neural Network (DNN), Recurrent Neural Network (RNN), Liquid Neural Network (LNN) and Extreme Gradient Boosting (XGBoost), were discussed in order to compare their performance. Model performance was determined by the coefficient of determination (R2), normalised mean squared error (NMSE) and the mean absolute percentage error (MAPE). XGBoost has been remarkable in its training prediction with an R2 value of 0.977 and an MAPE value of 2.47%. Meanwhile, RNN appeared the most reliable and the most transferable in terms of all data partitions’ performance, having validation and testing R2 values of 0.921 and 0.884, respectively, as well as the lowest testing MAPE of 6.22%.