Scholarlink

<p>The construction of infrastructure on gypseous soils is challenging because gypsum dissolves upon wetting, creating additional voids and inducing settlement and collapse. This study proposes a hybrid framework that couples Monte Carlo simulation (MCS) with Random Forest (RF) regression to quantify and predict leaching strain, defined here as the vertical strain accumulated during continuous leaching under sustained load due to gypsum dissolution. Leaching tests were conducted using an Oedometer-Permeability Leaching Test (OPLT; 75&#xa0;mm diameter, 19&#xa0;mm height) and a Modified Permeability Leaching Test (MPLT; 150&#xa0;mm diameter, 50&#xa0;mm height) on remoulded silty clay-gypsum mixtures (20–60% gypsum by weight) compacted at dry unit weights of 14–16 kN/m<sup>3</sup> and loaded to 200&#xa0;kPa, followed by 24&#xa0;h saturation and 7 days of leaching. The measured permeability, total dissolved solids (TDS), and leaching strain were used to parameterize a probabilistic model. MCS (100,000 samples) provided the distribution of leaching strain and associated reliability metrics (beta and Pf) for a failure threshold of Ls = 1%. An RF model trained on the generated dataset achieved high predictive accuracy (R approximately 0.96 on testing) with low RMSE/MAE; slight overfitting is attributed to limited experimental coverage and the use of synthetic data generation. The reliability analysis yielded beta = 0.19 and Pf = 0.43, indicating an unacceptably high failure likelihood under the analysed conditions and motivating mitigation measures (stabilisation, drainage, and conservative design).</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

Mitigating infrastructure risks in gypseous soils: combining random forest modeling with monte carlo simulation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月19日

查看全文
  • Hasan Ali Abbas,
  • Saif M. Hassan Al-Riahi,
  • Mohammed Y. Fattah

摘要

The construction of infrastructure on gypseous soils is challenging because gypsum dissolves upon wetting, creating additional voids and inducing settlement and collapse. This study proposes a hybrid framework that couples Monte Carlo simulation (MCS) with Random Forest (RF) regression to quantify and predict leaching strain, defined here as the vertical strain accumulated during continuous leaching under sustained load due to gypsum dissolution. Leaching tests were conducted using an Oedometer-Permeability Leaching Test (OPLT; 75 mm diameter, 19 mm height) and a Modified Permeability Leaching Test (MPLT; 150 mm diameter, 50 mm height) on remoulded silty clay-gypsum mixtures (20–60% gypsum by weight) compacted at dry unit weights of 14–16 kN/m3 and loaded to 200 kPa, followed by 24 h saturation and 7 days of leaching. The measured permeability, total dissolved solids (TDS), and leaching strain were used to parameterize a probabilistic model. MCS (100,000 samples) provided the distribution of leaching strain and associated reliability metrics (beta and Pf) for a failure threshold of Ls = 1%. An RF model trained on the generated dataset achieved high predictive accuracy (R approximately 0.96 on testing) with low RMSE/MAE; slight overfitting is attributed to limited experimental coverage and the use of synthetic data generation. The reliability analysis yielded beta = 0.19 and Pf = 0.43, indicating an unacceptably high failure likelihood under the analysed conditions and motivating mitigation measures (stabilisation, drainage, and conservative design).