Scholarlink

<p>The Cerchar Abrasiveness Index (CAI) is a vital parameter in geotechnical engineering, especially when it comes to tunneling and mechanized excavations. The study employed a comprehensive dataset of 163 samples representing various rock types, including igneous, sedimentary, and metamorphic formations. The methodology included three base algorithms (XGBoost, LightGBM, and Random Forest), improved by three distinct metaheuristic techniques: Arithmetic Optimization Algorithm (AOA), Reptile Search Optimization (RSO), and Harris Hawks Optimization (HHO). The Brazilian tensile strength (BTS), uniaxial compressive strength (UCS), equivalent quartz content (EQC), and brittleness index (BI) were the four main rock parameters used to make the predictive models. The model was further evaluated by splitting the data into 80% training and 20% testing sets. Subsequently, the model was compared to 17 real-world hard rock TBM projects in different countries and geological conditions. The AOA-optimized versions performed nicely, with AOA-LightGBM doing the best on the held-out test set (R² = 0.952, RMSE = 0.290, MAE = 0.208, VAF = 0.952). External validation showed that AOA-XGBoost performed properly, with the highest correlation coefficient of 0.8308 compared to field measurements from international tunneling projects. Also, the AOA-XGBoost did well on tests with R² = 0.951, RMSE = 0.296, MAE = 0.223, and VAF = 0.951. Using SHAP values to examine feature importance revealed unique parameter influence signatures. EQC was the most important parameter in XGBoost models, while UCS had the greatest impact in LightGBM and Random Forest-based models. The new method described here is an important advancement in CAI prediction methodology. It is more accurate and efficient than traditional experimental testing methods, and it works well on different types of rock. Its engineering applicability has been proven through real-world operational scenarios.</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

Cerchar abrasiveness index prediction based on rock properties leveraging hybrid soft computing techniques

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月27日

查看全文
  • Mohammad Matin Rouhani,
  • Alireza Dolatshahi,
  • Mahdi Hasanipanah,
  • Yewuhalashet Fissha,
  • N. Rao Cheepurupalli,
  • Taoufik Saidani

摘要

The Cerchar Abrasiveness Index (CAI) is a vital parameter in geotechnical engineering, especially when it comes to tunneling and mechanized excavations. The study employed a comprehensive dataset of 163 samples representing various rock types, including igneous, sedimentary, and metamorphic formations. The methodology included three base algorithms (XGBoost, LightGBM, and Random Forest), improved by three distinct metaheuristic techniques: Arithmetic Optimization Algorithm (AOA), Reptile Search Optimization (RSO), and Harris Hawks Optimization (HHO). The Brazilian tensile strength (BTS), uniaxial compressive strength (UCS), equivalent quartz content (EQC), and brittleness index (BI) were the four main rock parameters used to make the predictive models. The model was further evaluated by splitting the data into 80% training and 20% testing sets. Subsequently, the model was compared to 17 real-world hard rock TBM projects in different countries and geological conditions. The AOA-optimized versions performed nicely, with AOA-LightGBM doing the best on the held-out test set (R² = 0.952, RMSE = 0.290, MAE = 0.208, VAF = 0.952). External validation showed that AOA-XGBoost performed properly, with the highest correlation coefficient of 0.8308 compared to field measurements from international tunneling projects. Also, the AOA-XGBoost did well on tests with R² = 0.951, RMSE = 0.296, MAE = 0.223, and VAF = 0.951. Using SHAP values to examine feature importance revealed unique parameter influence signatures. EQC was the most important parameter in XGBoost models, while UCS had the greatest impact in LightGBM and Random Forest-based models. The new method described here is an important advancement in CAI prediction methodology. It is more accurate and efficient than traditional experimental testing methods, and it works well on different types of rock. Its engineering applicability has been proven through real-world operational scenarios.