Scholarlink

<p>Predicting the settlement (UZ) of bored piles in clayey sand is a highly nonlinear geotechnical problem that depends on numerous soil parameters and complex loading conditions. In particular, the mechanical behaviour of clayey sand, being sensitive to variations in moisture content, density, and particle structure, poses significant challenges for traditional analysis methods such as the Finite Element Method (FEM). To address these limitations, this study utilizes output data from 1,200 numerical simulations to train and evaluate the performance of three widely used machine learning algorithms: Artificial Neural Network (ANN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) in modelling the relationship between applied load and pile settlement. Evaluation on an independent test set shows that the ANN model achieved the highest accuracy, with a coefficient of determination R<sup>2</sup> = 0.9946 and RMSE = 15.14&#xa0;mm, outperforming both XGBoost (R<sup>2</sup> = 0.9870) and Random Forest (R<sup>2</sup> = 0.9738). Moreover, ANN predictions closely matched the results from field load tests conducted at 11 different loading levels, further validating its reliability. These findings confirm that machine learning, particularly ANN, can serve as an effective support tool in foundation design, helping to reduce both computation time and cost compared to traditional FEM analyses while maintaining the accuracy needed in geotechnical environments characterized by uncertainty.</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

Hybrid Machine Learning for Predicting Bored Pile Settlement in Clayey Sand: A Comparative Study Using FEM Data

  • 期刊论文

  • 发表日期：

    2025年9月9日

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

摘要

Predicting the settlement (UZ) of bored piles in clayey sand is a highly nonlinear geotechnical problem that depends on numerous soil parameters and complex loading conditions. In particular, the mechanical behaviour of clayey sand, being sensitive to variations in moisture content, density, and particle structure, poses significant challenges for traditional analysis methods such as the Finite Element Method (FEM). To address these limitations, this study utilizes output data from 1,200 numerical simulations to train and evaluate the performance of three widely used machine learning algorithms: Artificial Neural Network (ANN), Random Forest (RF), and Extreme Gradient Boosting (XGBoost) in modelling the relationship between applied load and pile settlement. Evaluation on an independent test set shows that the ANN model achieved the highest accuracy, with a coefficient of determination R2 = 0.9946 and RMSE = 15.14 mm, outperforming both XGBoost (R2 = 0.9870) and Random Forest (R2 = 0.9738). Moreover, ANN predictions closely matched the results from field load tests conducted at 11 different loading levels, further validating its reliability. These findings confirm that machine learning, particularly ANN, can serve as an effective support tool in foundation design, helping to reduce both computation time and cost compared to traditional FEM analyses while maintaining the accuracy needed in geotechnical environments characterized by uncertainty.