Scholarlink

<p>This study delves into predicting the residual gamma-ray linear attenuation coefficient (Rµ)&#xa0;values of concrete incorporating nano-additives, specifically nanocarbon tubes (NCTs) and nano-alumina (NAl), under elevated temperatures using various artificial intelligence (AI) models. Four AI-based prediction models of artificial neural networks (ANN), fuzzy logic models (FLM), water cycle algorithm (WCA), and genetic algorithm (GA) were trained using available literature data, which includes experimental results of 104 post-heating <i>µ</i> values varying by temperature degree, temperature exposure period, nanomaterial type, and nanomaterial replacement ratio. Results showed that ANN and FLM models demonstrated strong potential for predicting Rµ values, achieving coefficient of determination (R<sup>2</sup>) values of 0.989 and 0.999, respectively, for the training datasets. However, their practical application is limited by the challenge of formulating concise&#xa0;and direct prediction equations. Conversely, metaheuristic algorithms such as WCA and GA yield highly accurate predictions and enable the derivation of robust predictive equations. The developed equations using WCA and GA demonstrated excellent performance, achieving high R<sup>2</sup> values of 0.959 and 0.907, respectively, for the training datasets. Moreover, these models exhibited superior validation for residual Rµ values after elevated temperatures exposure, with mean absolute errors (MAEs) of 0.0322 and 0.0501 for training, 0.049 and 0.054 for validation, and 0.0499 and 0.0575 for testing datasets, respectively. Furthermore, sensitivity analysis using Shapley Additive Explanation (SHAP) was conducted to elucidate the impact and relationship between the input variables and the outputs of Rµ values. The SHAP results indicated that temperature degree exerted the most significant influence on Rµ values, followed by %NCTs, %NAl, and finally, time of exposure. The average absolute SHAP values for these variables were 2.1, 1.7, 1.6, and 1, respectively. This study’s findings emphatically underscore the effectiveness of AI-based models in predicting concrete radiation shielding behavior. Crucially, it provides valuable insights into the intricate, nonlinear relationships among the various variables that govern this behavior.</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

Enhancing predictive accuracy of nano-additive concrete gamma ray attenuation at high temperatures using AI-based models

  • 期刊论文

  • 发表日期：

    2025年8月2日

查看全文
  • Alaa A. Mahmoud,
  • Alaa A. El-Sayed,
  • Ayman M. Aboraya,
  • Islam N.Fathy,
  • Islam M. Nabil

摘要

This study delves into predicting the residual gamma-ray linear attenuation coefficient (Rµ) values of concrete incorporating nano-additives, specifically nanocarbon tubes (NCTs) and nano-alumina (NAl), under elevated temperatures using various artificial intelligence (AI) models. Four AI-based prediction models of artificial neural networks (ANN), fuzzy logic models (FLM), water cycle algorithm (WCA), and genetic algorithm (GA) were trained using available literature data, which includes experimental results of 104 post-heating µ values varying by temperature degree, temperature exposure period, nanomaterial type, and nanomaterial replacement ratio. Results showed that ANN and FLM models demonstrated strong potential for predicting Rµ values, achieving coefficient of determination (R2) values of 0.989 and 0.999, respectively, for the training datasets. However, their practical application is limited by the challenge of formulating concise and direct prediction equations. Conversely, metaheuristic algorithms such as WCA and GA yield highly accurate predictions and enable the derivation of robust predictive equations. The developed equations using WCA and GA demonstrated excellent performance, achieving high R2 values of 0.959 and 0.907, respectively, for the training datasets. Moreover, these models exhibited superior validation for residual Rµ values after elevated temperatures exposure, with mean absolute errors (MAEs) of 0.0322 and 0.0501 for training, 0.049 and 0.054 for validation, and 0.0499 and 0.0575 for testing datasets, respectively. Furthermore, sensitivity analysis using Shapley Additive Explanation (SHAP) was conducted to elucidate the impact and relationship between the input variables and the outputs of Rµ values. The SHAP results indicated that temperature degree exerted the most significant influence on Rµ values, followed by %NCTs, %NAl, and finally, time of exposure. The average absolute SHAP values for these variables were 2.1, 1.7, 1.6, and 1, respectively. This study’s findings emphatically underscore the effectiveness of AI-based models in predicting concrete radiation shielding behavior. Crucially, it provides valuable insights into the intricate, nonlinear relationships among the various variables that govern this behavior.