Scholarlink

<p>Strength and biological adhesion are critical performance indicators for the sustainable application of eco-friendly marine mortar blocks. However, the effects of mix proportion variations on strength and adhesion differ markedly. This discrepancy leads to low efficiency and high cost in traditional design methods. Alkalinity, a key parameter influencing bio-adhesion, serves as an indirect indicator of the material's environmental compatibility. In this work, eco-friendly marine mortar specimens with various mix proportions were prepared, and their compressive strength and alkalinity were systematically measured. Using a collected dataset of 464 mix formulations from the literature, seven machine learning models were compared, and a hybrid ensemble framework based on a back-propagation neural network (BPNN) was developed. The proposed framework demonstrated superior prediction accuracy (R<sup>2</sup> &gt; 0.98), excellent generalization capability, and interpretability via SHAP analysis. Compared to single-algorithm models, the ensemble model improved the prediction accuracy by 67.8%. Furthermore, a performance-driven inverse design methodology was implemented. The optimal mix proportions generated by this method were experimentally validated, with all prediction errors below 3%. This study provides an effective strategy to reduce the trial-and-error cost in optimizing marine mortar formulations, thereby supporting advancements in sustainable coastal engineering.</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

Machine learning-driven optimization framework for strength and bio-adhesion performance in marine engineering ecological mortar blocks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月18日

查看全文
  • Wangshun Li,
  • Penghui Yan,
  • Bin Tian,
  • Xiaochun Lu,
  • Ya Ban,
  • La Zhang,
  • Xiaogang Zhang,
  • Qiang Liu

摘要

Strength and biological adhesion are critical performance indicators for the sustainable application of eco-friendly marine mortar blocks. However, the effects of mix proportion variations on strength and adhesion differ markedly. This discrepancy leads to low efficiency and high cost in traditional design methods. Alkalinity, a key parameter influencing bio-adhesion, serves as an indirect indicator of the material's environmental compatibility. In this work, eco-friendly marine mortar specimens with various mix proportions were prepared, and their compressive strength and alkalinity were systematically measured. Using a collected dataset of 464 mix formulations from the literature, seven machine learning models were compared, and a hybrid ensemble framework based on a back-propagation neural network (BPNN) was developed. The proposed framework demonstrated superior prediction accuracy (R2 > 0.98), excellent generalization capability, and interpretability via SHAP analysis. Compared to single-algorithm models, the ensemble model improved the prediction accuracy by 67.8%. Furthermore, a performance-driven inverse design methodology was implemented. The optimal mix proportions generated by this method were experimentally validated, with all prediction errors below 3%. This study provides an effective strategy to reduce the trial-and-error cost in optimizing marine mortar formulations, thereby supporting advancements in sustainable coastal engineering.