Scholarlink

<p>This study aims to predict the vertical displacement (<i>UZ</i>) at the tip of bored piles using the XGBoost machine learning algorithm. The training data were generated from three-dimensional finite element simulations conducted with PLAXIS software. These simulations included various scenarios by altering key physical parameters of Clayey Sand soil, such as load, void ratio, elastic modulus, and other mechanical properties. The XGBoost model was trained on this dataset to accurately forecast pile settlement. Model performance was evaluated by comparing its predictions with field data obtained from static load tests. The results showed excellent agreement, with a coefficient of determination (<i>R</i><sup>2</sup>) close to 0.99 and a root mean square error (RMSE) under 21&#xa0;mm. These findings confirm the high predictive power of the model. More importantly, integrating FEM simulations with machine learning significantly reduces the need for repeated, time-consuming numerical analyses. This hybrid approach offers a fast, reliable, and cost-effective tool for designing bored pile foundations. Furthermore, the research demonstrates the potential of applying machine learning to solve complex geotechnical problems. It contributes to improving the efficiency and accuracy of foundation design in practical engineering.</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

XGBoost-Based Prediction of Bored Pile Settlement on Clayey Sand Using FEM-Based Data

  • 期刊论文

  • 发表日期：

    2025年7月5日

查看全文
  • Tuan Nguyen Anh,
  • Luan Vo Nhat,
  • Hoa Tran Vu Van

摘要

This study aims to predict the vertical displacement (UZ) at the tip of bored piles using the XGBoost machine learning algorithm. The training data were generated from three-dimensional finite element simulations conducted with PLAXIS software. These simulations included various scenarios by altering key physical parameters of Clayey Sand soil, such as load, void ratio, elastic modulus, and other mechanical properties. The XGBoost model was trained on this dataset to accurately forecast pile settlement. Model performance was evaluated by comparing its predictions with field data obtained from static load tests. The results showed excellent agreement, with a coefficient of determination (R2) close to 0.99 and a root mean square error (RMSE) under 21 mm. These findings confirm the high predictive power of the model. More importantly, integrating FEM simulations with machine learning significantly reduces the need for repeated, time-consuming numerical analyses. This hybrid approach offers a fast, reliable, and cost-effective tool for designing bored pile foundations. Furthermore, the research demonstrates the potential of applying machine learning to solve complex geotechnical problems. It contributes to improving the efficiency and accuracy of foundation design in practical engineering.