Scholarlink

<p>While shallow foundations are ubiquitous in global construction, the underlying soil-structure interaction remains computationally complex to analyze. This work introduces a Deep Neural Network (DNN) approach to accurately predict the displacement and stress fields of shallow foundations resting on sand. To establish a robust computational database, rigorous Finite Element Method (FEM) simulations were conducted. A comprehensive dataset for training and testing was generated by parametrically varying essential soil properties, including Young’s modulus, cohesion, friction angle and dilatancy angle. The proposed DNN architecture utilizes multiple hidden layers coupled with the ReLU activation function to capture the highly nonlinear behavior of the foundation system. Validation of the results indicates a high degree of fidelity between the DNN predictions and the traditional FEM outputs. Ultimately, this machine learning framework serves as a reliable and instantaneous surrogate model, effectively eliminating the computational bottleneck of time-consuming FEM analyses in iterative design optimization.</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

Deep Learning-Based Prediction of Shallow Foundation Behavior on Sandy Soils

  • 期刊论文

  • 发表日期：

    2026年5月30日

查看全文
  • Vi T. Huynh

摘要

While shallow foundations are ubiquitous in global construction, the underlying soil-structure interaction remains computationally complex to analyze. This work introduces a Deep Neural Network (DNN) approach to accurately predict the displacement and stress fields of shallow foundations resting on sand. To establish a robust computational database, rigorous Finite Element Method (FEM) simulations were conducted. A comprehensive dataset for training and testing was generated by parametrically varying essential soil properties, including Young’s modulus, cohesion, friction angle and dilatancy angle. The proposed DNN architecture utilizes multiple hidden layers coupled with the ReLU activation function to capture the highly nonlinear behavior of the foundation system. Validation of the results indicates a high degree of fidelity between the DNN predictions and the traditional FEM outputs. Ultimately, this machine learning framework serves as a reliable and instantaneous surrogate model, effectively eliminating the computational bottleneck of time-consuming FEM analyses in iterative design optimization.