Scholarlink

<p>Accurate estimation of the drilling rate (DR) of drilling machine, such as diamond core driller is critical for optimizing drilling efficiency and ensuring operational safety in subsurface exploration. In this study, eight machine learning algorithms, involving Random Forest (RF), Bagging, Gradient Boosting Machines (GB), XGBoost, CatBoost, AdaBoost, Decision Tree (DT), and Extra Trees (ET), were trained and tested on a dataset of 63 field‐measured drilling operations encompassing both mechanical parameters such as bit diameter, rotational speed, and thrust as well as geomechanical parameters such as uniaxial compressive strength (UCS), Brazilian tensile strength (BTS), density and brittleness index. Models were evaluated using <i>R</i><sup>2</sup>, RMSE, variance accounted for (VAF), and prediction accuracy (%). The RF model presented the highest accuracy, with a training <i>R</i><sup>2</sup> of 0.9727, RMSE of 8.71 mm/min, VAF of 97.21%, and accuracy of 99.71%, and a testing <i>R</i><sup>2</sup> of 0.9706, RMSE of 15.88 mm/min, VAF of 96.90%, and accuracy of 99.47%. Bagging followed closely (training <i>R</i><sup>2</sup> = 0.9583; testing <i>R</i><sup>2</sup> = 0.9507), while GB and ET ranked lowest (testing <i>R</i><sup>2</sup> = 0.9032 and 0.8639, respectively). SHAP (SHapley Additive exPlanations) analysis on the RF model revealed that rotational speed (mean |SHAP|= 23.76), thrust (10.72), UCS (6.62), and BTS (4.91) were the most influential predictors of DR. Dependence plots further showed that rotational speed above 700 RPM and thrust exceeding 2.5 kN contribute positively to DR, whereas UCS above 200 MPa and BTS above 15 MPa impede drilling. The integrated Graphic User Interface developed allows nonexpert users to perform “what‐if” scenario analyses and export results seamlessly. These findings demonstrate the efficacy of ensemble learning and SHAP‐based interpretability for reliable, data‐driven DR of diamond core driller forecasting under variable geology and subsurface conditions.</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

Explainable Ensemble Machine Learning for Predicting the Drilling Rate of Diamond Core Driller Using SHAP and Integrated Graphic User Interface

  • 期刊论文

  • 发表日期：

    2026年5月25日

查看全文
  • Shahab Hosseini,
  • Saffet Yagiz

摘要

Accurate estimation of the drilling rate (DR) of drilling machine, such as diamond core driller is critical for optimizing drilling efficiency and ensuring operational safety in subsurface exploration. In this study, eight machine learning algorithms, involving Random Forest (RF), Bagging, Gradient Boosting Machines (GB), XGBoost, CatBoost, AdaBoost, Decision Tree (DT), and Extra Trees (ET), were trained and tested on a dataset of 63 field‐measured drilling operations encompassing both mechanical parameters such as bit diameter, rotational speed, and thrust as well as geomechanical parameters such as uniaxial compressive strength (UCS), Brazilian tensile strength (BTS), density and brittleness index. Models were evaluated using R2, RMSE, variance accounted for (VAF), and prediction accuracy (%). The RF model presented the highest accuracy, with a training R2 of 0.9727, RMSE of 8.71 mm/min, VAF of 97.21%, and accuracy of 99.71%, and a testing R2 of 0.9706, RMSE of 15.88 mm/min, VAF of 96.90%, and accuracy of 99.47%. Bagging followed closely (training R2 = 0.9583; testing R2 = 0.9507), while GB and ET ranked lowest (testing R2 = 0.9032 and 0.8639, respectively). SHAP (SHapley Additive exPlanations) analysis on the RF model revealed that rotational speed (mean |SHAP|= 23.76), thrust (10.72), UCS (6.62), and BTS (4.91) were the most influential predictors of DR. Dependence plots further showed that rotational speed above 700 RPM and thrust exceeding 2.5 kN contribute positively to DR, whereas UCS above 200 MPa and BTS above 15 MPa impede drilling. The integrated Graphic User Interface developed allows nonexpert users to perform “what‐if” scenario analyses and export results seamlessly. These findings demonstrate the efficacy of ensemble learning and SHAP‐based interpretability for reliable, data‐driven DR of diamond core driller forecasting under variable geology and subsurface conditions.