High-Speed Motion Analysis-Based Machine Learning Models for Prediction and Simulation of Flyrock in Surface Mines

Abstract

Blasting is a cost-efficient and effective technique that utilizes explosive chemical energy to generate the necessary pressure for rock fragmentation in surface mines. However, a significant portion of this energy is dissipated in undesirable outcomes such as flyrock, ground vibration, back-break, etc. Among these, flyrock poses the gravest threat to structures, humans, and equipment. Consequently, the precise estimation of flyrock has garnered substantial attention as a prominent research domain. This research introduces an innovative approach for demarcating the hazardous zone for bench blasting through simulation of flyrock trajectories with probable launch conditions. To accomplish this, production blasts at five distinct surface mines in India were monitored using a high-speed video camera and data related to blast design and flyrock launch circumstances including the launch velocity (vf) were gathered by conducting motion analysis. The dataset was then used to develop ten Bayesian optimized machine learning regression models for predicting vf. Among all the models, the Extremely Randomized Trees Regression model (ERTR-BO) demonstrated the best predictive accuracy. Moreover, Shapely Additive Explanation (SHAP) analysis of the ERTR-BO model unveiled bulk density as the most influential input feature in predicting vf, followed by other features. To apply the model in a real-world setting, a user interface was developed to aid in flyrock trajectory simulation during bench blast designing.