Mean Cutting Force Prediction of Conical Picks Using Ensemble Learning Paradigm

The conical pick is the most essential tool of excavation machinery such as roadheaders, continuous miners, and shearers for breaking rock in mining and civil engineering operations. For rock cuttability, however, the geometry of conical picks and mechanical parameters of rocks are the most important factors. This study aims to construct an optimized data-driven predictive model to establish a quantitative correlation between strength of rock, geometry of tool, and cutting action data with the mean cutting force (CF). For this purpose, 157 datasets of 47 different materials including rocks, ores, coals, and artificial rocks with uniaxial compressive strength (sigma(c)), tensile strength (sigma(t)), cone angle (theta), attack angle (gamma), cutting depth (d), and mean CF (MCF) are accumulated from the literature. Then, extreme gradient boosting (XGBoost) model is constructed by fine-tuning hyperparameters using grid search, random search, genetic algorithm (GA), particle swarm optimization (PSO), and differential evolution (DE). Based on performance indices that are calculated for each model, i.e., coefficient of determination (R-2), root mean square error (RMSE), and mean absolute percentage error (MAE) for the best performed model, i.e., DE-XGboost are R-2 = 0.812, RMSE =2256.90N, and MAE =1313.66 N for training stage and R-2 = 0.875, RMSE =2104.86N, and MAE =1140.42 N for testing stage, respectively. The findings also suggest that using a metaheuristic algorithm to fine-tune the hyperparameters of the XGBoost model can increase prediction accuracy. In the last step, three model interpretation methods viz., the permutation-based variable importance, H-statistic-based variable interaction, and accumulated local effects are applied to sensitivity analysis of the input parameters to predict MCF, providing key insights to model and researchers. The ALE plot demonstrated a complex non-linear relationship between predictors and the response variable. It is revealed that parameters d and theta have the highest and lowest impact on the MCF, respectively. Finally, the successful implementation of this approach provides a solid platform for future studies and can be an alternative to complicated conventional and theoretical methods.

Automated summary

This study aims to construct an optimized data-driven predictive model to establish a quantitative correlation between strength of rock, geometry of tool, and cutting action data with the mean cutting force (CF). The extreme gradient boosting (XGBoost) model is constructed by fine-tuning hyperparameters using various algorithms. The findings suggest that using a metaheuristic algorithm to fine-tune the hyperparameters of the XGBoost model can increase prediction accuracy, and sensitivity analysis provides key insights into the impact of different parameters on the mean cutting force.