Scholarlink

<p>Accurate characterization of the residual strengths of mortar exposed to enhanced thermal conditions is crucial for optimizing the design, performance, and structural integrity of reinforced concrete structures during service. This study employs data-driven predictive approaches to estimate the strengths of pozzolanic mortar, where cement content is partially replaced with wood ash and fly ash, thermally treated from 0 to 800&#xa0;°C, and cured for up to 365&#xa0;days. The experimental dataset consisted of 324 data points for tensile and compressive strength measurements, with seven predictor variables for model development. Three machine learning (ML) models, including the Levenberg–Marquardt trained artificial neural network (ANN-LM), adaptive neuro-fuzzy inference systems (ANFIS), and multivariable regression analysis (MVRA), were developed for predictions. Extensive hyperparameter tuning of the ANN-LM models was conducted, evaluating 20 different architectures. The ANN-LM 7-9-1 and ANN-LM 7-12-1 architecture emerged as the best-performing models for estimating pozzolanic mortar’s tensile and compressive strengths. Performance evaluation metrics revealed that ANN-LM 7-9-1 achieved correlation coefficient (R), variance accounted for (VAF), and relative root mean square error (RRMSE) values of 0.9626, 92.68%, and 0.1852 for compressive strength. In contrast, ANN-LM 7-12-1 achieved 0.9364, 87.69%, and 0.5216 for tensile strength, demonstrating superior performance compared to the other models. The top-performing ANN-LMs were transformed into easy-to-use computational formulas or expressions to predict pozzolanic mortar strength and facilitate real-world application. Sensitivity analyses indicated that all seven features were significant, with mortar weight and cement content being the most influential. The two top-performing ANN-LM models, with the derived equations, are recommended for accurately forecasting pozzolanic mortar strengths within the experimental dataset's range for practitioners. This study provides practitioners with precise predictive tools, thereby minimizing reliance on time-consuming and costly experimental studies.</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

Comparative Modeling of Compressive and Tensile Strengths in Thermally Exposed Pozzolanic Mortar Using ANN-LM, ANFIS, and MVRA with Closed-Form Equations

  • 期刊论文

  • 发表日期：

    2025年7月25日

查看全文
  • Nafiu Olanrewaju Ogunsola,
  • Ajibola Ibrahim Quadri,
  • AbdulBasit Olamide Bankole

摘要

Accurate characterization of the residual strengths of mortar exposed to enhanced thermal conditions is crucial for optimizing the design, performance, and structural integrity of reinforced concrete structures during service. This study employs data-driven predictive approaches to estimate the strengths of pozzolanic mortar, where cement content is partially replaced with wood ash and fly ash, thermally treated from 0 to 800 °C, and cured for up to 365 days. The experimental dataset consisted of 324 data points for tensile and compressive strength measurements, with seven predictor variables for model development. Three machine learning (ML) models, including the Levenberg–Marquardt trained artificial neural network (ANN-LM), adaptive neuro-fuzzy inference systems (ANFIS), and multivariable regression analysis (MVRA), were developed for predictions. Extensive hyperparameter tuning of the ANN-LM models was conducted, evaluating 20 different architectures. The ANN-LM 7-9-1 and ANN-LM 7-12-1 architecture emerged as the best-performing models for estimating pozzolanic mortar’s tensile and compressive strengths. Performance evaluation metrics revealed that ANN-LM 7-9-1 achieved correlation coefficient (R), variance accounted for (VAF), and relative root mean square error (RRMSE) values of 0.9626, 92.68%, and 0.1852 for compressive strength. In contrast, ANN-LM 7-12-1 achieved 0.9364, 87.69%, and 0.5216 for tensile strength, demonstrating superior performance compared to the other models. The top-performing ANN-LMs were transformed into easy-to-use computational formulas or expressions to predict pozzolanic mortar strength and facilitate real-world application. Sensitivity analyses indicated that all seven features were significant, with mortar weight and cement content being the most influential. The two top-performing ANN-LM models, with the derived equations, are recommended for accurately forecasting pozzolanic mortar strengths within the experimental dataset's range for practitioners. This study provides practitioners with precise predictive tools, thereby minimizing reliance on time-consuming and costly experimental studies.