Scholarlink

<p>This research aims to experimentally investigate the mechanical behavior of natural fiber-reinforced geopolymer concrete (FRGPC) using Himalayan Giant Nettle Fiber (HGN) and to confirm the results using machine learning (ML) models. Ground granulated blast furnace slag and fly ash were employed as binders in 60:40, 70:30, and 80:20 ratios, along with fiber contents varying between 0% and 2% by binder weight. The optimum mix of 70% GGBFS, 30% FA, and 1% HGN fiber exhibited the maximum mechanical strengths, like compressive strength (CS) of 72.6 MPa, flexural strength (FS) of 13.5 MPa, and split tensile strength (STS) of 9.4 MPa at 28 days. Strengths decreased for fiber concentrations exceeding 1% due to improper distribution and excessive porosity, as the slump reduced to 35 mm. Hybrid ML models (Random Forest (RF) and CatBoost (CATB)) are trained with optimal hyperparameters obtained from Grid Search and Krill Herd (KH) algorithms to predict the mechanical properties. The most accurate CATB-KH model performed well for CS (R<sup>2</sup> = 0.9852, RMSE = 0.0484), FS (R<sup>2</sup> = 0.9882, RMSE = 0.0377), and STS (R<sup>2</sup> = 0.9916, RMSE = 0.0342). Interpretability by SHapley Additive exPlanations (SHAP) analysis revealed that curing age was the most significant parameter for CS, while fiber content was the most significant for FS and STS. Local Interpretable Model Agnostic Explanation (LIME) examination also verified model-specific behavior predictions at the individual level. The good agreement between the experimental results and the projections of ML enhances the confidence in the reproducibility of the finding with higher accuracy and demonstrates the practical potential of HGN fiber for sustainable structural applications.</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

Experimental investigation with machine learning validation on the mechanical strength of natural fiber reinforced geopolymer concrete

  • 期刊论文

  • 发表日期：

    2026年4月2日

查看全文
  • Suebha Khatoon,
  • Kaliluthin A K,
  • Sanjog Chhetri Sapkota

摘要

This research aims to experimentally investigate the mechanical behavior of natural fiber-reinforced geopolymer concrete (FRGPC) using Himalayan Giant Nettle Fiber (HGN) and to confirm the results using machine learning (ML) models. Ground granulated blast furnace slag and fly ash were employed as binders in 60:40, 70:30, and 80:20 ratios, along with fiber contents varying between 0% and 2% by binder weight. The optimum mix of 70% GGBFS, 30% FA, and 1% HGN fiber exhibited the maximum mechanical strengths, like compressive strength (CS) of 72.6 MPa, flexural strength (FS) of 13.5 MPa, and split tensile strength (STS) of 9.4 MPa at 28 days. Strengths decreased for fiber concentrations exceeding 1% due to improper distribution and excessive porosity, as the slump reduced to 35 mm. Hybrid ML models (Random Forest (RF) and CatBoost (CATB)) are trained with optimal hyperparameters obtained from Grid Search and Krill Herd (KH) algorithms to predict the mechanical properties. The most accurate CATB-KH model performed well for CS (R2 = 0.9852, RMSE = 0.0484), FS (R2 = 0.9882, RMSE = 0.0377), and STS (R2 = 0.9916, RMSE = 0.0342). Interpretability by SHapley Additive exPlanations (SHAP) analysis revealed that curing age was the most significant parameter for CS, while fiber content was the most significant for FS and STS. Local Interpretable Model Agnostic Explanation (LIME) examination also verified model-specific behavior predictions at the individual level. The good agreement between the experimental results and the projections of ML enhances the confidence in the reproducibility of the finding with higher accuracy and demonstrates the practical potential of HGN fiber for sustainable structural applications.