Scholarlink

<p>This study introduces a machine learning (ML) approach to predict the required undrained shear strength of Deep Cement Mixed (DCM) (C<sub>uc</sub>) columns used in pavement foundations. It presents a data-driven design method as an alternative to traditional techniques for estimating DCM column strength, marking a shift in geotechnical design practices. For this, two experimental datasets from laboratory investigations of soft soil improvement were combined to create a training database. To overcome limited experimental data, SMOTE-like interpolation methods were used to expand the database while maintaining geotechnical accuracy. Using the Pycaret library, a thorough analysis of different regression algorithms identified four top models: <i>Gradient Boosting Regression</i> (gbr), <i>AdaBoost</i> (ada), <i>Random Forest</i> (rf), and <i>Extra Trees</i> (et), with gbr showing the best performance (R² = 0.965, MSE = 6.21&#xa0;kPa). A sensitivity analysis of the models showed that ultimate bearing capacity (q<sub>u</sub>) was consistently the most influential parameter (41–53%), followed by the undrained shear strength of the surrounding soil (C<sub>us</sub>) at 18–31%. This suggests that enhancing soil strength before installing DCM may be more effective than simply increasing the area improvement ratio or column size. The model was packaged into a deployable pipeline to allow practical use by engineers, bridging the gap between advanced computational modelling and real-world geotechnical design of DCM-enhanced pavement foundations.</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

Prediction of Deep Cement Column Strength for Pavement Improvement Using Machine Learning Models

  • 期刊论文

  • 发表日期：

    2025年11月4日

查看全文
  • Ahmad Safuan A Rashid,
  • Ali Dehghanbanadaki,
  • Iman Golpazir,
  • Shervin Motamedi

摘要

This study introduces a machine learning (ML) approach to predict the required undrained shear strength of Deep Cement Mixed (DCM) (Cuc) columns used in pavement foundations. It presents a data-driven design method as an alternative to traditional techniques for estimating DCM column strength, marking a shift in geotechnical design practices. For this, two experimental datasets from laboratory investigations of soft soil improvement were combined to create a training database. To overcome limited experimental data, SMOTE-like interpolation methods were used to expand the database while maintaining geotechnical accuracy. Using the Pycaret library, a thorough analysis of different regression algorithms identified four top models: Gradient Boosting Regression (gbr), AdaBoost (ada), Random Forest (rf), and Extra Trees (et), with gbr showing the best performance (R² = 0.965, MSE = 6.21 kPa). A sensitivity analysis of the models showed that ultimate bearing capacity (qu) was consistently the most influential parameter (41–53%), followed by the undrained shear strength of the surrounding soil (Cus) at 18–31%. This suggests that enhancing soil strength before installing DCM may be more effective than simply increasing the area improvement ratio or column size. The model was packaged into a deployable pipeline to allow practical use by engineers, bridging the gap between advanced computational modelling and real-world geotechnical design of DCM-enhanced pavement foundations.