Blast Vibration Prediction using a Flexible Scaling Law Derived using Artificial Neural Network Techniques
The reliable prediction of blast induced vibration is crucial to the design of safe and economic blasts at surface mines, particularly when adjacent underground workings impose additional safety related design constraints. One of the objective in surface mine planning is to achieve large production blasts, which entails using higher explosive charge per round (or per delay). In contrast nearby underground operations cannot tolerate any damage to the underground structures due to surface blasting. Currently, the prevailing techniques for vibration prediction assume fixed scaling laws to model and estimate the attenuation characteristics of seismic waves through the transmitting rock mass. In fact, such fixed scaling laws fail to account for the significant variability with respect to geology and rock mass conditions, that is encountered between different mine sites. Therefore the desirability of developing a flexible, site specific, scaling law has long been recognized. Until recently, limitations in computing power, and lack of appropriate analytical methods, inhibited such development. This paper utilizes artificial neural networks (ANN’s) to identify site specific scaling law based on an input-output data set, where the input data consists of the explosive charge per delay as well as the distance from the blast site to the point of observation, and the output data set consists of observations of induced peak particle velocity (PPV). The performance of this technique is validated using case studies carried out at two mine sites, operated by Coal India Limited, where surface mines and underground mines are situated in close proximity, within a vertical distance of 80 metres to 130 metres. It is demonstrated that blast vibration prediction using the flexible scaling law derived using this technique is more accurate than fixed scaling laws. In addition the ANN-based technique is also more accurate than flexible scaling laws derived using statistical modelling.
training, optimization, challenges, simulation