Drilling and blasting is a process frequently used in rock-surface and deep excavation. For a proper drilling plan, accurate prediction of the amount of explosive material is essential to reduce the environmental effects associated with blasting operations. This study introduces a series of tree-based models, namely extreme gradient boosting machine (XGBoost), gradient boosting machine (GBM), adaptive boosting machine (AdaBoost), and random forest (RF), for predicting powder factor (PF) values obtained from blasting operations. The predictive models were constructed based on geomechanical characteristics at the blasting site, blasting pattern parameters, and rock material properties. These tree-based models were designed and tuned to minimize system error or maximize accuracy in predicting PF. Subsequently, the best model from each category was evaluated using various statistical metrics. It was found that the XGBoost model outperformed the other implemented techniques and exhibited outstanding potential in establishing the relationship between PF and input variables in the training set. Among the input parameters, hole diameter received the highest significance rating for predicting the system output, while the point load index had the least impact on the PF values.
Read full abstract