Comprehensive Models for Estimating Drag Force of Conical Picks Using Relieved Small-Scale LCM Tests Data

During excavation with roadheaders (RHs), the mounted conical picks (CPs) experienced three distinct cutting forces: drag (FD), normal, and side. Specifying these forces for designing or selecting a given RH is critical, and this may be determined with the linear cutting machine (LCM) tests. Nevertheless, to take advantage of soft computing technologies, a dataset based on the relieved small-scale LCM tests is compiled from the literature to develop Machine Learning (ML)-based predictive models for FD. The dataset consists of four groups of parameters: (1) rock properties, (2) CPs' specifications, (3) CP's operation variables, and (4) cutting conditions. The dataset undergoes extensive preprocessing to ensure its reliability by deleting outliers using three outlier detection approaches and removing redundant parameters through collinearity and multicollinearity checks. A cleansed dataset containing 224 data points is obtained, and five ensemble ML techniques, including AdaBoost, CatBoost, Gradient Boosting, LightGBM, and Random Forests, are employed to develop the FD predictive models, which are then fine-tuned and assessed using three evaluation metrics, such as R-squared. An applied ranking score on the models' prediction performance reveals that the CatBoost model yields the highest accuracy, with an R-squared value of 0.9389 for the testing dataset. Employing three eXplainable Artificial Intelligence (XAI) tools, the importance and contribution of input parameters to the prediction are assessed, demonstrating that rock compressive strength, tip angle, and spacing are key parameters verified through an ablation study. Therefore, these models can determine the CPs' FD applicable to RHs selection and cutter head design toward efficient cutting.