Explosive Blast Prediction using MLP Network based Training Algorithm

Abstract

The blast wave profile produced by detonations has long been the subject of research. The propagation profile of blast waves can be predicted given certain parameters after significant experimentation. However, prior research has mostly concentrated on the center of initiation for spherical explosive forms. This study compares the accuracy of blast peak overpressure predictions according to the kind, shape, and location of the explosive detonation. The experiment required creating a prediction model using a Multilayer Perceptron (MLP) network and detonating 500 grammes of Plastic Explosive 4 (PE-4) and Emulex at various ranges (from 0.5 m to 4.0 m) to do this. When modelling the prediction of explosive blasts using Tansig and Logsig training algorithms, Lavenberg Marquardt (LM) training method outperforms Backpropagation (BP). The MSE and regression scores of 1.1348 and 0.9512, respectively, using the LM training algorithm show the best performance.