Using Artificial Neural Networks for the Prediction of Bond Work Index from Rock Mechanics Properties

Abstract

The resistance shown to the grinding process and energy consumption can be determined using the work index. The Bond method is widely used in design of grinding circuits, selection of comminution equipment, determination of the power requirement and performance evaluation. Therefore, it is important to predict the Bond work index (BWi) using some easy and practical rock mechanics tests without the need to use a mill. In this study, rock mechanics and Bond tests were carried out on seven different marble and travertine samples. These rock mechanics tests are uniaxial compressive strength (sigma(c)), Brazilian tensile strength (sigma(t)), ultrasonic velocity (V-p), Schmidt hardness (R-L), point load index (I-S(50)) and density (rho). The BWi value was tried to be predicted using these rock mechanics test results by the feature selection method which is one of the artificial neural networks (ANN) methods. ANN has been used successfully for years in a very broad range of area such as classification, clustering, pattern recognition, prediction, etc. It was found out that the prediction of the BWi value by sigma(c), R-L, rho and I-S(50) values is reliable based on the obtained correlation coefficients by ANN feature selection method.