Hypernetwork Based Surrogate Modeling of Hypersonic Glide Vehicle Aerothermodynamics

The complexity and computational expense of simulating high speed flows around a three-dimensional vehicle geometry to calculate aerothermodynamic loads can be immense. These limitations often restrict the use of high-fidelity simulation to a handful of selected points of interest along a trajectory, from which the worst-case design criteria can be estimated. In scenarios which require comprehensive coverage of the aerothermal environment, such as end-to-end trajectory modeling, surrogate models can be used to approximate the desired solutions at points which lay between sampled intervals. However, the efficacy of surrogate modeling is often complicated by the need to map many unique combinations of input parameters to an output solution, particularly when the desired solution is a distribution of physical field values which may consist of tens of thousands of points surrounding the vehicle's surface. This task becomes further complicated in the hypersonic regime by the presence of shocks and other nonlinear phenomena, as well as the potential for high temperature gas effects. In this work, we present a hypernetwork-based surrogate modelling strategy for predicting the distribution of physical field values, such as temperature and resultant stress, on a vehicle's surface for a variety of hypersonic flow conditions and vehicle orientations. We show that hypernetworks are a flexible surrogate modeling framework for CFD simulations that parameterize freestream flow conditions-including Mach number, angle of attack, and altitude. Furthermore, our results demonstrate the viability of using hypernetworks to predict stress and temperature distributions across tens of thousands of mesh elements surrounding the surface of the HIFiRE-5 flight test vehicle for unseen combinations of input parameters.