Determination of Jones-Wilkins-Lee equation of state parameters using artificial neural network;Accuracy verification by coupled with experiment and numerical simulation

Abstract

A method for determining the JWL parameters and the detonation parameters using artificial neural networks is presented. The JWL EOS parameters and detonation parameters were collected according to composition of explosives, density and formation heat for 91 materials. The neural network model consists of the input, layer, output layer and hidden layer, and used the Neural Network Fitting Tool of Matlab, and Levenberg-Marquardt backpropagation algorithm was used for network training.The rational neural network architecture has 237, 179 and 72 neurons in three hidden layers, respectively, and the fit is reasonably good for for training, validation, and test sets, with R values in each case of above 0.99. The JWL EOS parameters were predicted for some explosives, 95RDX/3WAX/2stearic acid, 60TNT/24RDX/16Al, and 50AP/25Al/25RDX.Using trained neural network, we evaluated the JWL state equation parameters and detonation parameters for three explosives with two densities, respectively, and coupled with numerical simulations and experiments to evaluate the generalization performance of the neural network. The relative error between the predicted detonation velocity values and detonation velocity values determined by Pin Oscillographic Technique is less than 5%.And also, the relative error between the experimental values and the explosion shock overpressure values obtained by numerical simulations using AutoDyn is less than 5%.