Random forest and bayesian network techniques for probabilistic prediction of flyrock induced by blasting in quarry sites

Abstract

Flyrock is one of the most important environmental and hazardous issues in mine blasting, which can affect equipment and people, and may lead to fatal accidents. Therefore, prediction and minimization of this phenomenon are crucial to many rock removal projects. This study was organized in two sections. The first section is related to evaluation and selection of the most effective parameters of flyrock through the use of random forest technique. This resulted in the exclusion of “maximum charge per delay” from being used as input variable. The remaining input variables, i.e., hole diameter, hole depth, burden-to-spacing ratio, stemming, and powder factor, were utilized to develop the probabilistic prediction model using the Bayesian network (BN) technique. The learning and structure type of the BN model were maximum likelihood and tree augmented naïve Bayes, respectively. Many perfect probabilities were observed in the BN model for flyrock occurrence. The hole diameters between 97.5 and 127.5 mm appeared in four perfect probability conditions, which show that this hole diameter range is the most influential parameter. In addition, the combination of hole diameter and hole depth yielded three perfect probability predictions, suggesting that this factor combination is also influential on flyrock distance prediction. The results of this study can be used for optimum design of blasting pattern parameters for flyrock prediction.