Scholarlink

<p>A hybrid finite element–machine learning framework was developed to predict the bearing capacity of piled raft foundations in soft clay. A high-quality dataset was generated through three-dimensional finite element simulations of 45 piled raft configurations in which the number of piles, pile diameter, length-to-diameter ratio, spacing-to-diameter ratio, and pile–soil stiffness ratio were systematically varied. The dataset was used to train and evaluate several regression algorithms, including support vector machines and Gaussian process regression models. Two predictive structures were examined. The first model incorporated all geometric and stiffness parameters, while the second model used only geometric variables to quantify the influence of stiffness. Model performance was assessed using mean absolute error, root mean square error, and the coefficient of determination based on an independent test dataset. The optimized Gaussian process regression model achieved the highest predictive accuracy, reaching a test coefficient of determination of 0.991. The geometry-based model maintained strong predictive capability with a coefficient of determination of 0.960. The developed surrogate model enabled instantaneous prediction compared with computationally intensive numerical simulations, supporting rapid design exploration and efficient parametric analysis for piled raft foundation systems in soft clay.</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

A Hybrid FEM-Machine Learning Framework for Prediction of Piled Raft Bearing Capacity in Soft Clay

  • 期刊论文

  • 发表日期：

    2026年4月23日

查看全文
  • Danish Ahmed,
  • Muhammad Ajmal,
  • Siti Noor Linda bt Taib,
  • Nuha Jamal Alzayani,
  • Tahar Ayadat,
  • Sani I. Abba

摘要

A hybrid finite element–machine learning framework was developed to predict the bearing capacity of piled raft foundations in soft clay. A high-quality dataset was generated through three-dimensional finite element simulations of 45 piled raft configurations in which the number of piles, pile diameter, length-to-diameter ratio, spacing-to-diameter ratio, and pile–soil stiffness ratio were systematically varied. The dataset was used to train and evaluate several regression algorithms, including support vector machines and Gaussian process regression models. Two predictive structures were examined. The first model incorporated all geometric and stiffness parameters, while the second model used only geometric variables to quantify the influence of stiffness. Model performance was assessed using mean absolute error, root mean square error, and the coefficient of determination based on an independent test dataset. The optimized Gaussian process regression model achieved the highest predictive accuracy, reaching a test coefficient of determination of 0.991. The geometry-based model maintained strong predictive capability with a coefficient of determination of 0.960. The developed surrogate model enabled instantaneous prediction compared with computationally intensive numerical simulations, supporting rapid design exploration and efficient parametric analysis for piled raft foundation systems in soft clay.