Intelligent Prediction of Rock Drillability Using Mesoscopic Structure and Ensemble Learning Methods

Rock drillability, a critical indicator for evaluating rock fragmentation efficiency, plays a pivotal role in drilling and deep mining operations. Traditional physical measurement methods (e.g., micro-drilling) suffer from high costs and low efficiency, while existing numerical prediction approaches exhibit limited parameters and insufficient accuracy. The microstructural characteristics of rock play a fundamental role in determining its physicochemical properties and are strongly correlated with drillability. This study proposes a petrology-based feature set encompassing 21 particle texture parameters extracted from thin sections, combined with image analysis to establish a quantitative calculation model. Feature selection and dimensionality reduction were conducted using Pearson correlation analysis and PCA, followed by the development of a Stacking ensemble learning model for drillability prediction. The key findings are as follows: (1) Significant correlations exist between particle textures and drillability, with the highest correlation coefficients observed for shortest-axis variance (0.42) and area standard deviation (0.37); (2) The Stacking model demonstrated superior predictive performance, with E_\mathrmM\mathrmA\mathrmP\mathrmE and APE errors of only 14.1% and 12.6%, respectively, reducing errors by 4.7% and 2.5% compared to the best single-model benchmark; (3) The proposed framework enables automated thin-section analysis and drillability prediction within one minute. This approach provides a novel paradigm for efficient drillability assessment.