Scholarlink

<p>This study proposes an innovative method for assessing liquefaction potential by incorporating field data from the dynamic cone penetration test (DCPT) with advanced machine learning algorithms. The integration of these techniques aims to enhance the accuracy and reliability of liquefaction evaluations. The proposed framework addresses the limitations of the traditional methods by combining cost-efficient and expeditious DCPT with state-of-the-art computational tools. DCPT data, including the penetration rate (<i>e</i>) and dynamic resistance (<i>q</i><sub>d</sub>), were collected from various sites and used to calculate the critical <i>e</i>/<i>q</i><sub>d</sub> ratio for determining the liquefaction susceptibility. A threshold criterion was established, classifying soils with <i>e</i>/<i>q</i><sub>d</sub> ≤ 4 as non-liquefiable and those with <i>e</i>/<i>q</i><sub>d</sub> &gt; 4 as liquefiable. The prediction accuracies for liquefied and non-liquefied cases were 82.7% and 86.5%, respectively. Machine-learning models, including SVM-PSO, SVM-GWO, SVM-GA, and SVM-FF, were employed to confirm the evaluation procedure. Hyperparameter tuning was performed to optimize the model performance, with SVM-PSO achieving the highest <i>R</i><sup>2</sup> (0.999) and the lowest RMSE (0.220). The results underscore the efficacy of the <i>e/q</i><sub>d</sub> ratio within the DCPT framework, offering a robust tool for assessing the liquefaction susceptibility. This innovative approach enhances the accuracy, efficiency, and adaptability of liquefaction assessments across diverse soil conditions.</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

Applications of Dynamic Cone Penetration Test for Estimating Liquefaction Susceptibility Using Machine Learning Paradigms

  • 期刊论文

  • 发表日期：

    2025年1月2日

查看全文
  • Shubhendu Vikram Singh,
  • Sufyan Ghani

摘要

This study proposes an innovative method for assessing liquefaction potential by incorporating field data from the dynamic cone penetration test (DCPT) with advanced machine learning algorithms. The integration of these techniques aims to enhance the accuracy and reliability of liquefaction evaluations. The proposed framework addresses the limitations of the traditional methods by combining cost-efficient and expeditious DCPT with state-of-the-art computational tools. DCPT data, including the penetration rate (e) and dynamic resistance (qd), were collected from various sites and used to calculate the critical e/qd ratio for determining the liquefaction susceptibility. A threshold criterion was established, classifying soils with e/qd ≤ 4 as non-liquefiable and those with e/qd > 4 as liquefiable. The prediction accuracies for liquefied and non-liquefied cases were 82.7% and 86.5%, respectively. Machine-learning models, including SVM-PSO, SVM-GWO, SVM-GA, and SVM-FF, were employed to confirm the evaluation procedure. Hyperparameter tuning was performed to optimize the model performance, with SVM-PSO achieving the highest R2 (0.999) and the lowest RMSE (0.220). The results underscore the efficacy of the e/qd ratio within the DCPT framework, offering a robust tool for assessing the liquefaction susceptibility. This innovative approach enhances the accuracy, efficiency, and adaptability of liquefaction assessments across diverse soil conditions.