Scholarlink

<p>The unconfined compressive strength of organic-rich clay shale is a fundamental parameter in geotechnical and energy applications, influencing drilling efficiency, wellbore stability, and excavation design. This study presents machine learning-based predictive models for unconfined compressive strength estimation, trained on a comprehensive dataset of 1217 samples that integrate non-destructive indicators such as ultrasonic pulse velocity, shale fabric metrics, wettability potential and destructive field-derived parameters. A dual-model framework was implemented using Support Vector Machine, Decision Tree, K-Nearest Neighbor, and Extreme Gradient Boosting (XGBoost) algorithms. Among these, the composite XGBoost model exhibited the highest accuracy (R<sup>2</sup> = 0.981; RMSE = 0.02; MAE = 0.02), and maintained strong generalization (R<sup>2</sup> = 0.91) on an independent validation set of 959 samples. Taylor diagram analysis and sensitivity evaluation identified ultrasonic velocity, void ratio, bedding angle, and temperature as critical predictors. This study offers a scalable, data-driven alternative to conventional unconfined compressive strength testing and enables rapid, reliable geo-mechanical characterization for complex shale formations.</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

Machine learning-based prediction of unconfined compressive strength of organic-rich clay shales using hybrid destructive and non-destructive inputs

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年8月29日

查看全文
  • Muhammad Ali,
  • Mubashir Aziz,
  • Aaqib Ali,
  • Usman Ali

摘要

The unconfined compressive strength of organic-rich clay shale is a fundamental parameter in geotechnical and energy applications, influencing drilling efficiency, wellbore stability, and excavation design. This study presents machine learning-based predictive models for unconfined compressive strength estimation, trained on a comprehensive dataset of 1217 samples that integrate non-destructive indicators such as ultrasonic pulse velocity, shale fabric metrics, wettability potential and destructive field-derived parameters. A dual-model framework was implemented using Support Vector Machine, Decision Tree, K-Nearest Neighbor, and Extreme Gradient Boosting (XGBoost) algorithms. Among these, the composite XGBoost model exhibited the highest accuracy (R2 = 0.981; RMSE = 0.02; MAE = 0.02), and maintained strong generalization (R2 = 0.91) on an independent validation set of 959 samples. Taylor diagram analysis and sensitivity evaluation identified ultrasonic velocity, void ratio, bedding angle, and temperature as critical predictors. This study offers a scalable, data-driven alternative to conventional unconfined compressive strength testing and enables rapid, reliable geo-mechanical characterization for complex shale formations.