2:04 PM - 2:08 PM
[2107-24-07] Utilizing Machine Learning for Enhanced Rock Fragmentation Prediction in Open Pit Mines
Chairperson：鳥屋 剛毅（秋田大学）
Keywords:Rock Fragmentation, Machine Learning, Artificial Neural Networks, Predictive Modeling, Sensitivity Analysis
Predicting rock fragmentation is crucial for optimizing blasting operations in an open-pit mining setting. By accurately forecasting the size and distribution of fragmented rocks, this process aids in minimizing operational costs and energy consumption, while maximizing production efficiency. This strategic approach ensures that the blasting process is both cost-effective and environmentally conscious, aligning with the goals of sustainable mining practices. This study harnesses machine learning algorithms to forecast fragmentation outcomes in a limestone quarry mine, focusing on pivotal rock mass characteristics and blast geometry parameters. A comprehensive array of variables including spacing, drill hole diameter, burden, average bench height, powder factor, number of holes, charge per delay, uniaxial compressive strength, specific drilling, and stemming was initially considered. To enhance model efficiency, correlation techniques were employed, and variables with weak correlations were subsequently excluded. The refined predictive model utilizes a dataset of 219 entries, concentrating on key five features such as the number of holes, spacing, burden, specific drilling, and powder factor. Several predictive models, including Random Forest Regression, Support Vector Regression, XGBoost, and Neural Network Regression, were applied and evaluated using metrics like R-squared, RMSE, MSE, MAPE, and MAE. The analysis reveals that the Neural Network Regression model outperforms others, demonstrating the highest efficacy. Sensitivity analysis further identifies the powder factor as the most influential variable on fragmentation outcomes, while the burden shows minimal impact.