Physics-informed deep learning model for predicting ballistic coefficients of explosively driven fragments

Deep Learning models have the potential for accelerated predictions of physical phenomenon with an acceptable accuracy loss as alternatives to reduced-order models. In this work, we use a Deep Learning network to perform fast and accurate lift and drag predictions of explosively driven fragments traveling at hypersonic velocities. Specifically, we employed a generative adversarial network (GAN) to predict the total force on arbitrary shapes fixed in an external supersonic flow. The loss function of the generator was modified with additional terms based on the physics of the problem, heavily penalizing generated solutions that violated certain physical constraints. We trained our physics-informed network with a large set of flow fields from high fidelity aerodynamics simulations and show that drag was accurately predicted to within 2% average error. Sandia National Laboratories is a multimission laboratory managed and operated by National Technology and Engineering Solutions of Sandia, LLC, a wholly owned subsidiary of Honeywell International, Inc., for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-NA0003525.