Scholarlink

<p>The rapid usage of recycled aggregate concrete and nano-modified binders necessitates prediction frameworks that can handle tightly correlated mechanical, durability, and functional properties with limited experimental data samples. Traditional empirical formulations and single-output learning models cannot represent nonlinear, multiscale recycled aggregate, nano-admixture, curing history, and microstructure interactions. These limits limit material optimization reliability and prevent the design of durable and intelligent concrete systems for sustainable infrastructure sets. Next generation artificial intelligence frameworks using graph neural networks, capsule networks, neural ordinary differential equations, and neural architecture search predict nano-modified recycled aggregate concrete’s compressive, tensile, flexural, freeze–thaw, chloride penetration, and self-sensing electrical behavior. Microstructural interaction models and physical limitations ensure material believability across varied compositions and curing regimes. A quantitative analysis of over 100 experimental mix configurations indicates significant accuracy gains over multi-output baselines. EvoConcreteNet predicted flexural strength with a R² of 0.95, while GraphSenseNet achieved a coefficient of determination of 0.96 for compressive strength with mean absolute errors &lt; 2.5&#xa0;MPa. CapsuleRACNet achieved a R² above 0.97 for electrical resistance estimate, while ContinuousConcreteODE reduced freeze-thaw cycle prediction errors by over 40% compared to traditional regressors. Integrated design reduced forecast uncertainty, maintained physically consistent trends, and allowed multi-property assessment in one modeling step. Using physics-aware, multi-architecture deep learning, paper can logically create durable, self-sensing, and environmentally robust recycled aggregate concretes for sustainable construction systems.</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

Multi-property performance prediction of nano-material-enhanced recycled aggregate sustainable concrete: application of next-generation artificial intelligence techniques

  • 期刊论文

  • 发表日期：

    2026年3月17日

查看全文
  • Vaishali Mendhe,
  • Chetan D Karekar,
  • Shradhesh Marve,
  • Lowlesh N. Yadav,
  • Hirkani Padwad,
  • Nischal Puri,
  • Nilesh Shelke,
  • Aseel Smerat,
  • Vikrant S. Vairagade

摘要

The rapid usage of recycled aggregate concrete and nano-modified binders necessitates prediction frameworks that can handle tightly correlated mechanical, durability, and functional properties with limited experimental data samples. Traditional empirical formulations and single-output learning models cannot represent nonlinear, multiscale recycled aggregate, nano-admixture, curing history, and microstructure interactions. These limits limit material optimization reliability and prevent the design of durable and intelligent concrete systems for sustainable infrastructure sets. Next generation artificial intelligence frameworks using graph neural networks, capsule networks, neural ordinary differential equations, and neural architecture search predict nano-modified recycled aggregate concrete’s compressive, tensile, flexural, freeze–thaw, chloride penetration, and self-sensing electrical behavior. Microstructural interaction models and physical limitations ensure material believability across varied compositions and curing regimes. A quantitative analysis of over 100 experimental mix configurations indicates significant accuracy gains over multi-output baselines. EvoConcreteNet predicted flexural strength with a R² of 0.95, while GraphSenseNet achieved a coefficient of determination of 0.96 for compressive strength with mean absolute errors < 2.5 MPa. CapsuleRACNet achieved a R² above 0.97 for electrical resistance estimate, while ContinuousConcreteODE reduced freeze-thaw cycle prediction errors by over 40% compared to traditional regressors. Integrated design reduced forecast uncertainty, maintained physically consistent trends, and allowed multi-property assessment in one modeling step. Using physics-aware, multi-architecture deep learning, paper can logically create durable, self-sensing, and environmentally robust recycled aggregate concretes for sustainable construction systems.