Scholarlink

<p>A machine-learning framework was established to predict the axial bearing capacity of fully grouted rock bolts using an experimental database of 84 pull-out tests on fibreglass and steel rock bolts. The models incorporated six input parameters, embedment length (EL, 50–150 mm), confinement diameter (CD, 23–50 mm), bolt diameter (BD, 16–25 mm), water-to-grout ratio (W/G, 30–40%), grout compressive strength (UCS, 35–84 MPa), and curing time (CT, 7–28 days), together with bolt type (BT). Seven algorithms were developed and benchmarked: random forest (RF), extreme gradient boosting (XGBoost), support vector regression (SVR), decision tree (DT), gene expression programming (GEP), multiple linear regression (MLR), and nonlinear multiple linear regression (NMLR). The XGBoost model achieved the highest predictive accuracy (R<sup>2</sup> = 0.985, RMSE = 2.8 kN in training; R<sup>2</sup> = 0.952, RMSE = 5.9 kN, MAE = 4.8 kN in testing), followed by RF and NMLR, while MLR performed the weakest (R<sup>2</sup> = 0.611). The final ranking, based on all metrics, was: XGBoost &gt; RF &gt; NMLR &gt; SVR &gt; GEP &gt; DT &gt; MLR. Shapley Additive Explanations (SHAP) analysis identified CT, W/G, and UCS as the most influential features, jointly explaining more than 60% of the total variance in PL. Longer curing time and higher grout strength increased axial capacity, while higher water content reduced it. The proposed ensemble–SHAP framework provides a robust, interpretable, and physically consistent tool for predicting load-transfer behaviour in fully grouted rock-bolt systems.</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

Data-driven prediction and SHAP-based interpretation of the axial bearing capacity of fully grouted fibreglass and steel rock bolts

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月31日

查看全文
  • Shima Entezam,
  • Behshad Jodeiri Shokri,
  • Alireza Entezam,
  • Hadi Nourizadeh,
  • Kevin McDougall,
  • Warna Karunasena,
  • Naj Aziz,
  • Ali Mirzaghorbanali

摘要

A machine-learning framework was established to predict the axial bearing capacity of fully grouted rock bolts using an experimental database of 84 pull-out tests on fibreglass and steel rock bolts. The models incorporated six input parameters, embedment length (EL, 50–150 mm), confinement diameter (CD, 23–50 mm), bolt diameter (BD, 16–25 mm), water-to-grout ratio (W/G, 30–40%), grout compressive strength (UCS, 35–84 MPa), and curing time (CT, 7–28 days), together with bolt type (BT). Seven algorithms were developed and benchmarked: random forest (RF), extreme gradient boosting (XGBoost), support vector regression (SVR), decision tree (DT), gene expression programming (GEP), multiple linear regression (MLR), and nonlinear multiple linear regression (NMLR). The XGBoost model achieved the highest predictive accuracy (R2 = 0.985, RMSE = 2.8 kN in training; R2 = 0.952, RMSE = 5.9 kN, MAE = 4.8 kN in testing), followed by RF and NMLR, while MLR performed the weakest (R2 = 0.611). The final ranking, based on all metrics, was: XGBoost > RF > NMLR > SVR > GEP > DT > MLR. Shapley Additive Explanations (SHAP) analysis identified CT, W/G, and UCS as the most influential features, jointly explaining more than 60% of the total variance in PL. Longer curing time and higher grout strength increased axial capacity, while higher water content reduced it. The proposed ensemble–SHAP framework provides a robust, interpretable, and physically consistent tool for predicting load-transfer behaviour in fully grouted rock-bolt systems.