Scholarlink

<p>The study explores the use of banana leaf ash (BLA), an agricultural waste, as a sustainable partial cement substitute in M30-grade concrete. Specimens incorporating 5%, 10%, and 15% BLA were tested for compressive, flexural, and split strengths at 7, 14, and 28&#xa0;days. Results indicated that BLA improved mechanical performance relative to the control mix, with 10% replacement achieving optimum strength and 15% providing slightly lower strength but enhanced sustainability. Advanced machine learning models, Deep Neural Networks (DNN), Natural Gradient Boosting (NGBoost), Bayesian Neural Networks (BNN), and TabNet, were developed for multi-output strength prediction. NGBoost achieved the highest accuracy, DNN delivered robust and efficient predictions, and BNN provided effective uncertainty quantification. SHAP and permutation importance analyses identified BLA content, water–cement ratio, and coarse aggregate as the most influential features, while regression plots, REC curves, violin plots, and Taylor diagrams confirmed model reliability. Carbon footprint analysis showed up to 15% reduction in embodied CO₂ emissions with BLA, and a Sustainability Index (SI) highlighted 10% as the optimal balance of strength and environmental benefit. These findings demonstrate the potential of combining BLA with AI-driven modelling for eco-efficient concrete design, supporting the United Nations Sustainable Development Goals.</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

Sustainable concrete incorporating banana leaf ash: strength prediction and optimisation using advanced machine learning algorithms

  • 期刊论文

  • 发表日期：

    2025年9月4日

查看全文
  • Shahaji Patil,
  • Gayathri Niveda,
  • B. J. Phanindra Babu,
  • G. M. Jayashree,
  • Anigowda,
  • Sathvik Sharath Chandra

摘要

The study explores the use of banana leaf ash (BLA), an agricultural waste, as a sustainable partial cement substitute in M30-grade concrete. Specimens incorporating 5%, 10%, and 15% BLA were tested for compressive, flexural, and split strengths at 7, 14, and 28 days. Results indicated that BLA improved mechanical performance relative to the control mix, with 10% replacement achieving optimum strength and 15% providing slightly lower strength but enhanced sustainability. Advanced machine learning models, Deep Neural Networks (DNN), Natural Gradient Boosting (NGBoost), Bayesian Neural Networks (BNN), and TabNet, were developed for multi-output strength prediction. NGBoost achieved the highest accuracy, DNN delivered robust and efficient predictions, and BNN provided effective uncertainty quantification. SHAP and permutation importance analyses identified BLA content, water–cement ratio, and coarse aggregate as the most influential features, while regression plots, REC curves, violin plots, and Taylor diagrams confirmed model reliability. Carbon footprint analysis showed up to 15% reduction in embodied CO₂ emissions with BLA, and a Sustainability Index (SI) highlighted 10% as the optimal balance of strength and environmental benefit. These findings demonstrate the potential of combining BLA with AI-driven modelling for eco-efficient concrete design, supporting the United Nations Sustainable Development Goals.