Improvement of small scale mine blast operation: A comparative application of hunter-point artificial neural network, support vector machine, and regression analysis models

Abstract

The blasting operation is one of the technologies used for breaking rock masses and reducing the rock mass into smaller sizes to improve transportation and further particle separation. The improvement of blast fragmentation supports the maximization of mining operation and productivity. Soft computing and regression model has been developed in this study to optimize small-scale dolomite blast operations in Akoko Edo, Nigeria. WipFrag software was used to analyze the results of 35 blasting rounds. As independent variables, one uncontrollable parameter and five controllable blast parameters were chosen to predict blast particle sizes using four mathematically motivated soft computing model approaches. The prediction accuracy of the developed models was tested using various model performance indices. The study revealed that rock strength influences blast fragmentation results, and based on the rock strength properties, the fragmentation block size increases with an increase in rock strength. The results of the model performance indices used for the evaluation of the proposed models showed that the modified Artificial Neural Network (ANN) called Hunter Point (HP-ANN) has the highest predictive accuracy. A new model evaluator was also developed in this study called the decision factor. Its application indicates that the HP-ANN model is the best model suitable for the prediction of blast fragment size distribution. Therefore, the developed models can be used to predict the blast result mean size (X50) and the 80% percentage passing size (X80) for mining engineering blasting practices.

Keywords

• artificial neural network
• blasting
• empirical modelling
• support vector machine
• wipfrag software