Assessing excavatability in varied rockmass conditions using real-time data and machine learning technique

Abstract

This study investigates shovel excavation performance across various rockmass conditions by integrating real-time performance assessments, rockmass property analysis, and machine learning techniques. Correlation analysis revealed significant positive relationships between Total Loading Time (TLT) and selected rock properties, specifically uniaxial compressive strength (UCS), tensile strength (TS) cohesion (C), and moisture content (M), while a negative correlation was observed with wet bulk density (WBD). Pareto analysis further highlighted C, UCS, and TS as the most impactful factors, cumulatively accounting for 56% of the total effect on excavation performance. A multiple linear regression model, using TLT as the dependent variable and significant rock properties (C, UCS, M) as predictors, achieved a strong correlation (R=0.76) and explained 76% of the variance, demonstrating the model’s effectiveness in estimating shovel performance. K-nearest neighbors (KNN) classification, optimized with a k-value of 7 and Manhattan distance, achieved a high accuracy of 99.43% in categorizing the excavation difficulty into four distinct classes. The frequency distribution of TLT data indicates that most materials in the pit fall under “Very Easy” and “Easy” classes, simplifying excavation processes. This research underscores the importance of the key rock properties in evaluating the excavation performance predictions and support optimized operational strategies in mining. Future work could expand on these findings by using additional machine learning techniques and exploring non-linear models to capture complex relationships.