Determination of ballistic resistance model of finite thickness material based on Artificial Neural Network

Ballistic resistance behavior of finite thickness material is affected by factors of target thickness, projectile nose shape, and impact angle. The widely used ballistic limit velocity (BLV) model can only reflect the ballistic resistance behavior of a target in a single given impact scenario. The collected BLV from different impact scenarios can only partially reflect the effect of those factors on the ballistic resistance. A general model of BLV considering these factors simultaneously is desired. In this study, numerical impact models are first verified by a significant number of experiments. A guided random sampling method is proposed to collect numerical BLV data for training/validating usage. BLV data from experiments and simulations are then assembled by this method and subsequently trained by Artificial Neural Network (ANN). k-Folder Cross Validation (K-CV) technique designed for evaluating small sample training network is introduced to validate and test the network. The ANN-predicted results are decomposed by proper singular value decomposition methods. It is found that the Rank-1 decomposition results (with highest singular value) can fully reveal the decoupled relationship between BLV and its independent factors (MAPE < 6 %), with which analytical ballistic models are established. The results show that our proposed model can quantitatively describe the fully decoupled effect of target thickness, impact angle and projectile nose shape on the BLV with high accuracy.