Digital Twin Machine Learning for Hypersonic and Asymmetric Warhead Characterization

Accurate lethality and effectiveness predictions for ordnance systems is a critical objective of the U.S. Air Force. Characterization of new hypersonic weapon systems and asymmetric warheads requires innovative testing, diagnostics, and fragment modeling methods. The traditional approach to weapon testing, diagnostics, and fragment modeling will not work for asymmetric weapons since the fragment field differs around their azimuth. At best, asymmetric weapon testing will result in a patchwork of data collection at relatively few carefully chosen zones leaving the bulk of the fragment field uncharacterized. Digital engineered data will be the cornerstone of a new paradigm for hypersonic asymmetric weapon characterization. The ability to synthesize digital twin fragment fields with patchwork experimental measurements will be a transformative outcome of this effort.  For organizations such as AFLCMC/EB, AFRL/RWML, and AFRL/RWMW, the highest payoff of this new paradigm will be accurate next-generation fragment models for effective lethality predictions of asymmetric and hypersonic warheads. Protection Engineering Consultants (PEC) has teamed with Midland Research (MR) to address the lack of accurate fragment models for hypersonic and asymmetric weapons. This will be accomplished by developing a machine-learning paradigm to accurately synthesize and model data from physical and digital twin tests. The ML framework will consist of two primary modules: 1) a Generative Adversarial Network (GAN) to synthesis physical and digital twin data, and 2) a Bayesian Neural Network (BNN) to stochastically model fragment fields. GAN-based data synthesis will address the issue of limited experimental measurements, while BNN-based models will address the need for objective fragment model development that considers the uncertainties involved.