Building a predictive soft armor finite element model combining experiments, simulations, and machine learning

Abstract

Despite its relevance for law enforcement applications, the design of soft armor has mainly been based on a trial-and-error approach. In this paper, a combined experimental, machine learning, and finite element analysis framework is used to build a predictive numerical model for the analysis and hence, design of soft armor. The material models for major components of the soft armor certification system—bullet, shoot pack, straps, and clay backing, are first constructed using laboratory tests and publicly available data. Next, three metrics, namely, back face signature (BFS), number of penetrated shoot-pack layers, and mushrooming of the bullet are established to gauge the model’s accuracy with respect to the laboratory ballistic test data. A machine learning (ML) model is used as a surrogate to predict the BFS and the number of eroded elements. Finally, optimized material model parameters are obtained through ML-based surrogate model and Bayesian optimization algorithm. The final validation of the developed framework is carried out using laboratory ballistic test data involving multiple shots on the shoot pack. The results indicate that reliable predictive data can be obtained using the developed process, and likely, can be extended for use in modeling other impact simulations.