Scholarlink

<p>Laboratory model tests were conducted to investigate the uplift behavior of rough concrete piles in unsaturated sand–clay mixtures under varying moisture conditions and slenderness ratios. Three piles with different length-to-diameter ratios (L/D = 5.3, 6.3, and 7.9) were tested at five soil moisture levels (1–9%) to capture geometry–moisture interactions. The experimental results show that the load–displacement response is highly nonlinear, with both strength and stiffness strongly influenced by moisture. Uplift resistance is lowest in very dry and wet states and reaches a maximum near 5% water content. Increasing slenderness improves capacity, but the benefit diminishes outside the suction-controlled range. To enable rapid prediction of design-critical parameters and full load–displacement curves, three machine learning (ML) models, namely, Random Forest, Gradient Boosting, and Support Vector Regression, were trained on the experimental dataset using physics-informed features and peak-weighted sampling. The dataset comprises 15 controlled laboratory uplift tests, downsampled to approximately 500 force–displacement points per test (7500 points total) after cleaning. Among these, Random Forest achieved the best performance on unseen tests (R² ≈ 0.92; RMSE ≈ 35&#xa0;N), accurately reproducing peak force and curve shape. This integrated experimental–ML framework provides a practical and interpretable tool for early-stage geotechnical design under seasonal moisture variability, supporting serviceability checks and geometry optimization without additional testing.</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

Experimental and data-driven modeling of uplift load–displacement response of concrete piles in unsaturated soils

  • 期刊论文

  • 发表日期：

    2026年4月7日

查看全文
  • Meghdad Bagheri

摘要

Laboratory model tests were conducted to investigate the uplift behavior of rough concrete piles in unsaturated sand–clay mixtures under varying moisture conditions and slenderness ratios. Three piles with different length-to-diameter ratios (L/D = 5.3, 6.3, and 7.9) were tested at five soil moisture levels (1–9%) to capture geometry–moisture interactions. The experimental results show that the load–displacement response is highly nonlinear, with both strength and stiffness strongly influenced by moisture. Uplift resistance is lowest in very dry and wet states and reaches a maximum near 5% water content. Increasing slenderness improves capacity, but the benefit diminishes outside the suction-controlled range. To enable rapid prediction of design-critical parameters and full load–displacement curves, three machine learning (ML) models, namely, Random Forest, Gradient Boosting, and Support Vector Regression, were trained on the experimental dataset using physics-informed features and peak-weighted sampling. The dataset comprises 15 controlled laboratory uplift tests, downsampled to approximately 500 force–displacement points per test (7500 points total) after cleaning. Among these, Random Forest achieved the best performance on unseen tests (R² ≈ 0.92; RMSE ≈ 35 N), accurately reproducing peak force and curve shape. This integrated experimental–ML framework provides a practical and interpretable tool for early-stage geotechnical design under seasonal moisture variability, supporting serviceability checks and geometry optimization without additional testing.