Scholarlink

<p>This study addresses a critical aspect of geoenvironmental engineering by evaluating the potential for breach in an embankment (starter dyke), which could result in significant environmental contamination in the surrounding areas. The study investigates the variability of ash dyke materials, which includes clayey sand used in the starter dyke and clay used in the foundation. A detailed database of geotechnical and hydraulic properties is compiled from literature spanning the years 1939 to 2024 to support the probabilistic modelling. The current study considers the variability of pond ash and dyke ash by determining the best-fit probability distributions, including Normal, Lognormal, Weibull, Inverse Gaussian, Gamma, and Gumbel distributions, using Kolmogorov–Smirnov statistics. The study incorporates a steady-state seepage analysis as a parent simulation, with transient seepage and slope stability analyses as child simulations. The reliability analysis in GeoStudio is limited by its consideration of only a single slip surface and its restriction to Normal and Lognormal probability distributions. The present study overcomes these limitations by proposing an automated framework that integrates GeoStudio with Python, Excel, XML, and HTML files, thereby significantly enhancing its functionality. The convergence study of the surrogate models concludes that the optimal accuracy is achieved with 600 data points, resulting in leave-one-out errors, root mean square errors, and coefficient of determination values of 0.003, 99.1%, and 0.019 for the polynomial chaos expansion model, and 0.007, 99.7%, and 0.012 for the Kriging model, respectively. These trained 600 datapoints provide an efficiency and accuracy of 10<sup>6</sup> simulations in Monte Carlo Simulations. A parametric study is conducted specifically for the starter dyke to analyse the influence of changing mean values and coefficients of variability of cohesion and friction angle on the reliability index. The findings underscore that restricting analysis to a single slip surface and limited probability distributions significantly underestimates the failure probability (<i>p</i><sub><i>f</i></sub> ~ 2.16%), whereas the proposed automated framework, incorporating multiple slip surfaces and broader variability, yields a more realistic <i>p</i><sub><i>f</i></sub> of 9.48%. This emphasizes the framework’s capability to deliver efficient and reliable risk assessments for ash dyke stability.</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

Probabilistic Stability Assessment of Ash Dykes Using Surrogate Models and Multi-Slip Surface Analysis

  • 期刊论文

  • 发表日期：

    2026年1月6日

查看全文
  • Kandikonda Abhinai,
  • Kenue Abdul Waris,
  • B. Munwar Basha

摘要

This study addresses a critical aspect of geoenvironmental engineering by evaluating the potential for breach in an embankment (starter dyke), which could result in significant environmental contamination in the surrounding areas. The study investigates the variability of ash dyke materials, which includes clayey sand used in the starter dyke and clay used in the foundation. A detailed database of geotechnical and hydraulic properties is compiled from literature spanning the years 1939 to 2024 to support the probabilistic modelling. The current study considers the variability of pond ash and dyke ash by determining the best-fit probability distributions, including Normal, Lognormal, Weibull, Inverse Gaussian, Gamma, and Gumbel distributions, using Kolmogorov–Smirnov statistics. The study incorporates a steady-state seepage analysis as a parent simulation, with transient seepage and slope stability analyses as child simulations. The reliability analysis in GeoStudio is limited by its consideration of only a single slip surface and its restriction to Normal and Lognormal probability distributions. The present study overcomes these limitations by proposing an automated framework that integrates GeoStudio with Python, Excel, XML, and HTML files, thereby significantly enhancing its functionality. The convergence study of the surrogate models concludes that the optimal accuracy is achieved with 600 data points, resulting in leave-one-out errors, root mean square errors, and coefficient of determination values of 0.003, 99.1%, and 0.019 for the polynomial chaos expansion model, and 0.007, 99.7%, and 0.012 for the Kriging model, respectively. These trained 600 datapoints provide an efficiency and accuracy of 106 simulations in Monte Carlo Simulations. A parametric study is conducted specifically for the starter dyke to analyse the influence of changing mean values and coefficients of variability of cohesion and friction angle on the reliability index. The findings underscore that restricting analysis to a single slip surface and limited probability distributions significantly underestimates the failure probability (pf ~ 2.16%), whereas the proposed automated framework, incorporating multiple slip surfaces and broader variability, yields a more realistic pf of 9.48%. This emphasizes the framework’s capability to deliver efficient and reliable risk assessments for ash dyke stability.