Enhancing Blast Design Efficiency for Rock Fragmentation with Gradient Descent and Artificial Neural Networks: An Optimization Study

This paper presents a predictive model for blast-induced fragmentation at Jwaneng Diamond Mine in Botswana using a machine learning algorithm namely artificial neural networks (ANN). A dataset consisting of 70 blasts with seven blast design parameters was used. The ANN is optimized based on Monte Carlo method to explore the solution space that is modelled using the input parameters, and the overall fragmentation has been improved to at-least 80%. Root mean square error (RMSE) and determination coefficient (R2) indices were used to validate and compare the performance of the the different ANN models. ANN one model, with architecture 7-10-1 demonstrated superiority over the other ANN models in predicting fragmentation with the highest R2 value of 0.880 and lower RMSE of 0.481. The results of sensitivity analysis showed that spacing has the most influence on fragmentation while hole diameter has the least influence on fragmentation.