Assessment of blast induced ground vibrations by artificial neural network

Abstract

Blast-induced ground motion is analyzed by means of two prediction methods. First conventional approach assumes several types of nonlinear dependence of peak particle velocity on scaled distance from the explosion charge, while the second technique implements a feed-forward three-layer back-propagation neural network with three nodes in input layer (total charge, maximum charge per delay and distance from explosive charge to monitoring point) and only one node in output layer (peak particle velocity). As a result, traditional predictors give acceptable prediction accuracy (r>0.7) when compared with registered values of peak particle velocity. Regarding the forecasting accuracy estimated by neural network, model with nine hidden nodes gives reasonable predictive precision (r>0.9), with much lower standard error in comparison to conventional predictors.