Scholarlink

<p>This study presents a machine learning-based framework for predicting failure envelopes of typical and modified caisson foundations under combined general loading (<i>V</i>, <i>H</i>, and <i>M</i>) in nonhomogeneous clays. Parametric analyses are conducted using 3D finite element limit analysis (FELA) with the Tresca failure criterion to evaluate the effects of embedment depth ratio (<i>L/D</i>) and nonuniform strength ratio (<i>κ</i>) on 2D and 3D failure envelopes and failure patterns of two caisson foundation types. Special emphasis is placed on 2D failure envelopes in the <i>H-M</i> space (<i>H/s</i><sub><i>u</i></sub><i>LD</i>, <i>M/s</i><sub><i>u</i></sub><i>L</i><sup><i>2</i></sup><i>D</i>) under varying vertical load levels (<i>V/V</i><sub><i>0</i></sub>), considering full-tension conditions at the foundation–soil interface. To address the complexities of geotechnical stability modeling, machine learning techniques such as adaptive boosting (AdaBoost), support vector machines, and K-nearest neighbors are employed to develop surrogate models for constructing failure envelopes. Among these, the AdaBoost model demonstrates superior performance, achieving the highest accuracy (R<sup>2</sup> = 98.9%) in predicting failure envelopes. This integration of machine learning and 3D FELA represents a robust and efficient approach to understand and model the behavior of caisson foundations under general loading conditions.</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

Predicting 3D Failure Envelopes of Caisson Foundations in Nonhomogeneous Clay: FELA, AdaBoost, SVM, and KNN

  • 期刊论文

  • 发表日期：

    2025年6月16日

查看全文
  • Katavut Vichai,
  • Jim Shiau,
  • Duy Tan Tran,
  • Suraparb Keawsawasvong,
  • Pitthaya Jamsawang

摘要

This study presents a machine learning-based framework for predicting failure envelopes of typical and modified caisson foundations under combined general loading (V, H, and M) in nonhomogeneous clays. Parametric analyses are conducted using 3D finite element limit analysis (FELA) with the Tresca failure criterion to evaluate the effects of embedment depth ratio (L/D) and nonuniform strength ratio (κ) on 2D and 3D failure envelopes and failure patterns of two caisson foundation types. Special emphasis is placed on 2D failure envelopes in the H-M space (H/suLD, M/suL2D) under varying vertical load levels (V/V0), considering full-tension conditions at the foundation–soil interface. To address the complexities of geotechnical stability modeling, machine learning techniques such as adaptive boosting (AdaBoost), support vector machines, and K-nearest neighbors are employed to develop surrogate models for constructing failure envelopes. Among these, the AdaBoost model demonstrates superior performance, achieving the highest accuracy (R2 = 98.9%) in predicting failure envelopes. This integration of machine learning and 3D FELA represents a robust and efficient approach to understand and model the behavior of caisson foundations under general loading conditions.