Predicting peak particle velocity by artificial neural networks and multivariate regression analysis - Sarcheshmeh copper mine, Kerman, Iran

Abstract

Ground vibrations caused by blasting are undesirable consequences in the mining industry and can cause serious damage to the nearby buildings and facilities. Hence, such vibrations have to be controlled to reduce the damage to the environment and this may be achieved once blasting peak particle velocity (PPV) is predicted. In this study, PPV is predicted and compared in a case study in Kerman using three methods of artificial neural network (ANN), multivariate regression analysis (MVRA) and empirical relations. The data gathered belonged to 11 blast operations in Sarcheshmeh copper mine, Kerman. The neural network input parameters include: distance from blast point, maximum charge weight per delay, spacing, stemming and the number of drill-hole rows in each blasting operation. The network is of the multi-layer perception (MLP) type, with 24 sets of training data including 2 hidden layers, 1 output layer with the network architecture being {5-11-12-1}, and Sigmoid tangent and linear transfer functions. To ensure adequate training accuracy, the network was tested by 6 data sets; the determination coefficient and the average relative error were found to be 0.977 and 8.85% respectively, indicating the MLP network’s high capability and precision in estimating the PPV. Comparison of the predicted PPV’s with those obtained from MVRA and the empirical relations revealed low capabilities of these two in estimating the PPV parameter.