ARTIFICIAL NEURAL NETWORK FOR DAMAGE PREDICTION AT HYPERVELOCITY IMPACTS

Abstract

Impact of a projectile on a thin bumper at hypervelocity causes perforation. Primary fragment in residual projectile debris becomes a threat to main structure. Complexities in damage assessment arise due to changing phenomenology with impact conditions. Conventionally, damage assessment by empirical models has been developed based on experiment and simulation data. However, models have limited versatility. Artificial neural network (ANN) has advantages in adaptive learning from data and in solving complex problem to reach approximate solution. In this article, networks are trained for estimation of perforated hole dimensions and primary fragment characteristics (mass, velocity and exit angle). In certain range of impact parameters, estimated values of those models are deviating, while ANN indicates changing damage trend as per simulation data. To improve ANN estimates, additional simulations are performed. The networks are retrained and compared. It is observed that ANN requires less data for training with better accuracy.