Scholarlink

<p>Fully grouted rock bolts are extensively used in geotechnical engineering applications to improve the stability of fractured rock masses. The ultimate shear capacity of these bolts has traditionally been estimated using empirical formulas or analytical models. However, due to the complexities of stress–strain interactions and boundary conditions between the bolt and surrounding materials under varying conditions, accurate prediction of bolt ultimate shear performance remains a significant challenge. To overcome this limitation, this study, for the first time, uses machine learning (ML) techniques for predicting the ultimate shear performance of fully grouted rock bolts, specifically the dimensionless ultimate shear resistance, <i>f(T)</i>, and the corresponding joint shear displacement at bolt failure, <i>f(s)</i>. A comprehensive database consisting of 110 bolted rock joint shear tests is established and used, including 10 input parameters and experimentally measured values of <i>f(T)</i> and <i>f(s)</i>. Four ML models—Random Forest (RF), Extreme Gradient Boosting (XGB), Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP)—are developed and evaluated. The results indicate that among these models, XGB exhibits the highest predictive accuracy for both <i>f(T)</i> and <i>f(s)</i>. Furthermore, the XGB model significantly outperforms the existing empirical formula and theoretical model, demonstrating superior accuracy and stability across different input parameter ranges. SHapley Additive exPlanations (SHAP) analysis confirms that the XGB model effectively captures the influence of key input variables on bolt shear behavior, offering deeper insights into the underlying mechanics. Predictions of ultimate bolt resistance using the XGB model in an engineering case study indicate that conventional design methods, which consider only axial force, may considerably underestimate bolt resistance. Moreover, the XGB model provides insights into optimizing anchoring angles by accounting for the coupled tensile-shear force, demonstrating its potential as a robust tool for improving anchorage system performance in practical applications.</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

Ultimate Shear Performance of Fully Grouted Rock Bolts: Insights from Machine Learning

  • 期刊论文

  • 发表日期：

    2025年8月20日

查看全文
  • Linfeng Zhu,
  • Murat Karakus,
  • Liangqing Wang,
  • Zihao Sun,
  • Changshuo Wang,
  • Luobin Zheng,
  • Binqiang Fan

摘要

Fully grouted rock bolts are extensively used in geotechnical engineering applications to improve the stability of fractured rock masses. The ultimate shear capacity of these bolts has traditionally been estimated using empirical formulas or analytical models. However, due to the complexities of stress–strain interactions and boundary conditions between the bolt and surrounding materials under varying conditions, accurate prediction of bolt ultimate shear performance remains a significant challenge. To overcome this limitation, this study, for the first time, uses machine learning (ML) techniques for predicting the ultimate shear performance of fully grouted rock bolts, specifically the dimensionless ultimate shear resistance, f(T), and the corresponding joint shear displacement at bolt failure, f(s). A comprehensive database consisting of 110 bolted rock joint shear tests is established and used, including 10 input parameters and experimentally measured values of f(T) and f(s). Four ML models—Random Forest (RF), Extreme Gradient Boosting (XGB), Support Vector Regression (SVR), and Multi-Layer Perceptron (MLP)—are developed and evaluated. The results indicate that among these models, XGB exhibits the highest predictive accuracy for both f(T) and f(s). Furthermore, the XGB model significantly outperforms the existing empirical formula and theoretical model, demonstrating superior accuracy and stability across different input parameter ranges. SHapley Additive exPlanations (SHAP) analysis confirms that the XGB model effectively captures the influence of key input variables on bolt shear behavior, offering deeper insights into the underlying mechanics. Predictions of ultimate bolt resistance using the XGB model in an engineering case study indicate that conventional design methods, which consider only axial force, may considerably underestimate bolt resistance. Moreover, the XGB model provides insights into optimizing anchoring angles by accounting for the coupled tensile-shear force, demonstrating its potential as a robust tool for improving anchorage system performance in practical applications.