A Hybrid Data-Driven Machine Learning Framework for Predicting the Impact Resistance of Composite Armor

Abstract

Composite armor plays a crucial role as the primary defense against high-velocity impacts from fragments and projectiles. However, balancing the need for lightweight structures with the requirement for robust protection remains a significant engineering challenge. Traditional approaches for predicting the protective performance of armor typically involve a combination of experimental testing and numerical simulations, both of which can be resource-intensive and costly. In contrast, data-driven methods combined with machine learning have demonstrated the potential to significantly reduce both time and economic costs, highlighting their substantial advantages in various engineering domains. Unfortunately, a mature machine learning framework for predicting the performance of multilayer composite armor against high-velocity impacts from large fragments has yet to be established. In this paper, a novel data-driven framework for predicting the ballistic performance of composite armor using a hybrid model of Support Vector Machine and Deep Neural Network was established. This framework employed hyperparameter optimization to enhance predictive performance, yielding a model with excellent accuracy. The proposed model was adaptable to multilayered armor with varying layer thicknesses, enabling rapid predictions of armor penetration, residual projectile kinetic energy, and armor deformation.