MLP Network Prediction for Blast Explosive based Training Algorithm

Abstract

For many years, researchers have been examining the profile of blast waves resulting from detonations and using experimentation to make predictions based on specific parameters. However, previous studies have mainly focused on the central point of initiation for spherical explosive shapes. The aim of this study is to compare the accuracy of predicting the blast peak overpressure based on various factors, including the type and shape of the explosive and the location of detonation. The experiment involved detonating 500 grams of PE-4 and Emulex at different distances (ranging from 0.5 to 4.0 meters) and creating a prediction model using a Multilayer Perceptron (MLP) network. Bayesian Regularization (BR) proved to be more effective than Backpropagation (BP) when modelling Explosive Blast Prediction. The BR training with Logsig training algorithm shows the best performance with 0.9280 and 0.9658 for MSE and regression, respectively.