Scholarlink

<p>Precise measurement and prediction of rock slope stability remain crucial for ensuring the safety of civil infrastructure in challenging geological environments. This study develops a measurement-oriented framework integrating finite element modeling with machine learning (ML) techniques to predict the Factor of Safety (FoS) in reinforced rock slopes. A comprehensive dataset of 432 numerical simulations was constructed through systematic parametric investigation, comparing the generalized Hoek-Brown (GHB) and Mohr-Coulomb (MC) failure criteria across varying Geological Strength Index (GSI) values (15–25) and disturbance factors (D = 0–1.0). Results demonstrate that the GHB criterion exhibits higher sensitivity to disturbance-induced strength degradation, with FoS reductions reaching 34% under maximum disturbance conditions, while the MC criterion consistently overestimates stability by 40–70% in jointed rock masses. Six ML algorithms were evaluated through five-fold cross-validation. The XGBoost model achieved superior performance with R² = 0.94 and minimal error metrics, effectively capturing nonlinear geomechanical relationships. Permutation-based feature importance analysis identified second bench height and major principal stress as dominant parameters governing FoS predictions. Monte Carlo uncertainty propagation analysis quantified how measurement errors propagate through predictive models, revealing that GSI measurements require high precision (± 2–3%) to maintain FoS prediction reliability. The surrogate model exhibited CV = 15.07% with consistent performance across stability classes. A measurement priority classification framework was developed, translating quantitative sensitivity analyses into practical field data acquisition strategies. The integrated framework provides efficient FoS estimation for parametric studies, while the measurement priority classification enables optimized field investigation strategies by identifying critical parameters requiring high-precision instrumentation.</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

Parameter measurement sensitivity in disturbed rock masses: A machine learning framework for slope stability prediction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月24日

查看全文
  • Yesim Tuskan,
  • Aybike Özyüksel Çiftçioğlu

摘要

Precise measurement and prediction of rock slope stability remain crucial for ensuring the safety of civil infrastructure in challenging geological environments. This study develops a measurement-oriented framework integrating finite element modeling with machine learning (ML) techniques to predict the Factor of Safety (FoS) in reinforced rock slopes. A comprehensive dataset of 432 numerical simulations was constructed through systematic parametric investigation, comparing the generalized Hoek-Brown (GHB) and Mohr-Coulomb (MC) failure criteria across varying Geological Strength Index (GSI) values (15–25) and disturbance factors (D = 0–1.0). Results demonstrate that the GHB criterion exhibits higher sensitivity to disturbance-induced strength degradation, with FoS reductions reaching 34% under maximum disturbance conditions, while the MC criterion consistently overestimates stability by 40–70% in jointed rock masses. Six ML algorithms were evaluated through five-fold cross-validation. The XGBoost model achieved superior performance with R² = 0.94 and minimal error metrics, effectively capturing nonlinear geomechanical relationships. Permutation-based feature importance analysis identified second bench height and major principal stress as dominant parameters governing FoS predictions. Monte Carlo uncertainty propagation analysis quantified how measurement errors propagate through predictive models, revealing that GSI measurements require high precision (± 2–3%) to maintain FoS prediction reliability. The surrogate model exhibited CV = 15.07% with consistent performance across stability classes. A measurement priority classification framework was developed, translating quantitative sensitivity analyses into practical field data acquisition strategies. The integrated framework provides efficient FoS estimation for parametric studies, while the measurement priority classification enables optimized field investigation strategies by identifying critical parameters requiring high-precision instrumentation.