Scholarlink

<p>This study developed a machine learning model based on a multi-layer feedforward neural network to predict the settlement and ultimate load capacity of bored piles using data from three-dimensional finite element simulations. A finite element model was established for sandy clay soil with elastic modulus ranging from 14,000 to 26,500 kN/m², internal friction angle from 28° to 34°, and cohesion from 3 to 6 kN/m². A bored pile with a diameter of 800&#xa0;mm, a length of 43&#xa0;m, an elastic modulus of 34.5 × 10⁶ kN/m², and a unit weight of 25 kN/m³ was analyzed under incremental static loading. The simulation results were used to train a neural network with three hidden layers (64–64–32 neurons) and a ReLU activation function. The model achieved high predictive accuracy, with R² = 0.9983, MAE = 6.60&#xa0;mm, and RMSE = 11.43&#xa0;mm, indicating effective representation of the nonlinear load–settlement relationship. The ultimate load capacity was estimated using piecewise linear regression, curvature-based analysis, and the slope ratio method, yielding values between 14,400 and 15,900 kN/m², with deviations of less than 10% compared to field static load test results. The results indicate that the proposed FEM–MLP framework can supplement finite element analyses.</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 FEM-MLP Approach for Evaluating the Ultimate Load Capacity of Bored Piles in Sandy Clay Soil

  • 期刊论文

  • 发表日期：

    2026年2月9日

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

摘要

This study developed a machine learning model based on a multi-layer feedforward neural network to predict the settlement and ultimate load capacity of bored piles using data from three-dimensional finite element simulations. A finite element model was established for sandy clay soil with elastic modulus ranging from 14,000 to 26,500 kN/m², internal friction angle from 28° to 34°, and cohesion from 3 to 6 kN/m². A bored pile with a diameter of 800 mm, a length of 43 m, an elastic modulus of 34.5 × 10⁶ kN/m², and a unit weight of 25 kN/m³ was analyzed under incremental static loading. The simulation results were used to train a neural network with three hidden layers (64–64–32 neurons) and a ReLU activation function. The model achieved high predictive accuracy, with R² = 0.9983, MAE = 6.60 mm, and RMSE = 11.43 mm, indicating effective representation of the nonlinear load–settlement relationship. The ultimate load capacity was estimated using piecewise linear regression, curvature-based analysis, and the slope ratio method, yielding values between 14,400 and 15,900 kN/m², with deviations of less than 10% compared to field static load test results. The results indicate that the proposed FEM–MLP framework can supplement finite element analyses.