Applicability of a neural network model for forecasting ground vibrations in opencast mining

Abstract

Ground vibration and air-blast over pressure are two significant undesirables, among many environmental risks, in open-pit mining . Gaining control over the ground vibrations generated by rock blasts had been difficult mainly due to the complexities involved with local geology and properties of the blast. Accordingly, existing empirical equations are only capable of making vague approximations on the vibration frequencies based on site-specific parameters and attenuation factor. Therefore, the available models cannot be generalized to different geo-mining environments to obtain sufficiently reliable forecasts for ground vibration and airblast overpressure. Hence, this study attempts to employ an Artificial Neural Network (ANN) based feed-forward back-propagation algorithm to train a model, using a supervised learning technique to forecast possible ground v i b r a t i on frequencies. The main in-put parameters included in the model are noise level, number of boreholes per single blast, depth and diameter of a borehole, charge per hole, number of delays of the Electric Detonators (ED) in a single blast, burden and spacing. Airblast overpressure and the ground vibration levels will be the output by ANN model. The model was validated using 50 datasets, which were obtained from a quarry site. After adequate training, the model can determine Peak Particle Velocity (PPV) and frequency of Ground Vibrations (GV) for new input parameters with a statistically significant confidence level.