Prediction of blast-induced rock fragmentation at Orapa Diamond Mine using hybrid ANN models

Abstract

This paper presents predictive models for blast induced fragmentation at Orapa Diamond Mine in Botswana using machine learning algorithms namely artificial neural networks (ANN), particle swarm optimization artificial neural networks (PSO-ANN), and genetic algorithm artificial neural networks. A dataset consisting of 50 blasts with eight blast design parameters such as burden, spacing, hole depth, hole diameter, maximum charge per delay, stemming length, powder factor, distance from the monitoring point as input parameters, and fragmentation as the output parameter are used. The main goal of production blasting is to achieve proper fragmentation. Rock fragmentation has a direct influence on the mill throughput and diggability which in turn affect the overall mine economics.Hence accurate prediction of fragmentation is crucial in arriving at an economical outcome. Root mean square error and determination coefficient (R2 ) indices were used to validate and compare the performance of the models. PSO-ANN demonstrated superiority over the other hybrid models in predicting fragmentation with the highest accuracy and lowest error. The results of sensitivity analysis showed that hole depth has the most influence on fragmentation while maximum charge per delay has the least influence on fragmentation.