Scholarlink

<p>In composite structures using steel reinforced concrete (SRC), it is essential to consider the ultimate bond stress (τu) between profiled steel and concrete. The present techniques for measuring the τu of profiled steel–concrete may lack accuracy when used to larger-scale environments. This results from the development of these models with a restricted quantity of test data sets prior to their implementation. This study evaluates the τu of profiled steel–concrete utilizing various machine learning approaches that need data use. According to a computer database including 269 test results from previous studies, eight input parameters influence the ultimate bond stress of profiled steel–concrete. During the process of developing and evaluating the framework, seventy percent of the data was used as a learning set, fifteen percent as a validating set, and fifteen percent as an assessing set. To calculate, Adaptive boosting (ADA) was employed. Its reliability is greatly impacted by the ADA hyperparameters, which need to be chosen using metaheuristic optimization techniques. This is accomplished using the Horned Lizard Optimizer (HLO) and the Graylag Goose Optimizer (GGO). This technique utilizes a more extensive, curated dataset and non-linear learning capabilities to capture intricate interactions among various input variables, in contrast to typical empirical or semi-empirical models that rely on restricted datasets and linear assumptions. It surpasses current methodologies by providing enhanced generalization, robustness across data subsets, and increased prediction accuracy, especially via the ADA-GGO model, which exhibits more stable and dependable performance, rendering it more appropriate for practical, large-scale engineering applications. The data indicates that both ADA-GG and ADA-HL have a high probability of correctly estimating the value of τu. The ADA-GG indicated that the training, validation, and assessment phases had respective U (95%) index values of 0.1806, 0.2373, and 0.2019. Comparable U (95%) values for the ADA-HL were 0.2636, 0.2915, and 0.2670.</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

Adaptive boosting estimation algorithms on bond of steel–concrete in steel reinforced composite structures

  • 期刊论文

  • 发表日期：

    2025年9月10日

查看全文
  • Dayuan Jiang,
  • Yingjing Lan,
  • Guozhe Yue

摘要

In composite structures using steel reinforced concrete (SRC), it is essential to consider the ultimate bond stress (τu) between profiled steel and concrete. The present techniques for measuring the τu of profiled steel–concrete may lack accuracy when used to larger-scale environments. This results from the development of these models with a restricted quantity of test data sets prior to their implementation. This study evaluates the τu of profiled steel–concrete utilizing various machine learning approaches that need data use. According to a computer database including 269 test results from previous studies, eight input parameters influence the ultimate bond stress of profiled steel–concrete. During the process of developing and evaluating the framework, seventy percent of the data was used as a learning set, fifteen percent as a validating set, and fifteen percent as an assessing set. To calculate, Adaptive boosting (ADA) was employed. Its reliability is greatly impacted by the ADA hyperparameters, which need to be chosen using metaheuristic optimization techniques. This is accomplished using the Horned Lizard Optimizer (HLO) and the Graylag Goose Optimizer (GGO). This technique utilizes a more extensive, curated dataset and non-linear learning capabilities to capture intricate interactions among various input variables, in contrast to typical empirical or semi-empirical models that rely on restricted datasets and linear assumptions. It surpasses current methodologies by providing enhanced generalization, robustness across data subsets, and increased prediction accuracy, especially via the ADA-GGO model, which exhibits more stable and dependable performance, rendering it more appropriate for practical, large-scale engineering applications. The data indicates that both ADA-GG and ADA-HL have a high probability of correctly estimating the value of τu. The ADA-GG indicated that the training, validation, and assessment phases had respective U (95%) index values of 0.1806, 0.2373, and 0.2019. Comparable U (95%) values for the ADA-HL were 0.2636, 0.2915, and 0.2670.