Scholarlink

<p>The traditional concrete mix design method has the bottleneck of insufficient collaborative prediction of multi-objective performance and lack of reverse mix ratio derivation, which restricts the accurate regulation of concrete performance. In this paper, a concrete database was established based on 207 sets of test results. Through six machine learning algorithms and Bayesian hyperparameter optimization, the 7d and 28d compressive strength, slump, expansion degree, initial setting time and final setting time were collaboratively predicted and reversely designed. Through the TOPSIS decision theory, it is found that the DNN algorithm has good prediction accuracy and generalization performance. The importance ranking, threshold effect and interrelationship of each parameter are obtained by combining SHAP and PDP analysis. In addition, the mix ratio parameters under different properties are predicted by reverse design, and the intelligent inversion from performance requirements to material mix ratio is realized.</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-objective design method of concrete mix ratio based on machine learning: forward prediction and reverse design

  • 期刊论文

  • 发表日期：

    2026年2月3日

查看全文
  • Bin Yang,
  • Yue Li,
  • Hui Lin,
  • Xiwang Chen,
  • Kun Ni,
  • Wei Li,
  • Jiale Shen,
  • Tao Liang

摘要

The traditional concrete mix design method has the bottleneck of insufficient collaborative prediction of multi-objective performance and lack of reverse mix ratio derivation, which restricts the accurate regulation of concrete performance. In this paper, a concrete database was established based on 207 sets of test results. Through six machine learning algorithms and Bayesian hyperparameter optimization, the 7d and 28d compressive strength, slump, expansion degree, initial setting time and final setting time were collaboratively predicted and reversely designed. Through the TOPSIS decision theory, it is found that the DNN algorithm has good prediction accuracy and generalization performance. The importance ranking, threshold effect and interrelationship of each parameter are obtained by combining SHAP and PDP analysis. In addition, the mix ratio parameters under different properties are predicted by reverse design, and the intelligent inversion from performance requirements to material mix ratio is realized.