Scholarlink

This study investigates the prediction of bored pile load–displacement behaviour using machine learning (ML) models. The models were trained on a dataset derived from ground investigations and 28 full-scale uplift pile tests and validated against nine independent field tests. Key input features included pile geometries, cone tip resistance (qc), sleeve friction (fs) and pile head displacement (δ). Where direct CPT data was unavailable, parameters were derived from standard penetration test (SPT) data using established correlations. Five models were developed and compared: random forest (RF), gradient boosting (GB), polynomial regression (PR), multilayer perceptron (MLP) and linear regression (LR). Model performance was evaluated using the coefficient of determination (R2), mean squared error (MSE), root mean squared error (RMSE) and mean absolute error (MAE). Ensemble models (RF and GB) achieved the highest predictive accuracy (R2 ≈ 0.999), with validation results supporting their generalisation to unseen field data. Feature importance analysis confirmed that pile geometry and CPT parameters significantly influence uplift capacity. The findings demonstrate that machine learning models may supplement traditional bored pile design practices, especially when traditional data is limited.

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

Bored Pile Capacity Prediction Based on CPT Data Using Machine Learning Models

  • 会议论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年

查看全文
  • Abdulhakim Mawas,
  • Omar Hamza

摘要

This study investigates the prediction of bored pile load–displacement behaviour using machine learning (ML) models. The models were trained on a dataset derived from ground investigations and 28 full-scale uplift pile tests and validated against nine independent field tests. Key input features included pile geometries, cone tip resistance (qc), sleeve friction (fs) and pile head displacement (δ). Where direct CPT data was unavailable, parameters were derived from standard penetration test (SPT) data using established correlations. Five models were developed and compared: random forest (RF), gradient boosting (GB), polynomial regression (PR), multilayer perceptron (MLP) and linear regression (LR). Model performance was evaluated using the coefficient of determination (R2), mean squared error (MSE), root mean squared error (RMSE) and mean absolute error (MAE). Ensemble models (RF and GB) achieved the highest predictive accuracy (R2 ≈ 0.999), with validation results supporting their generalisation to unseen field data. Feature importance analysis confirmed that pile geometry and CPT parameters significantly influence uplift capacity. The findings demonstrate that machine learning models may supplement traditional bored pile design practices, especially when traditional data is limited.