Predicting Hypersonic Vehicle Heating Using a Multi-Model Deep Learning Framework

The design of hypersonic vehicles is driven by the harsh thermal environment in which the vehicle must operate. This thermal environment drives key design decisions like outer mold line (OML), flight envelope, trajectory, and material selection. Being able to quickly iterate over this design tradespace is becoming increasingly more important as hypersonic vehicles gain prevalence not only with military systems but also in the commercial sector. Current approaches primarily rely on two primary methods: analytic solutions, which have limitations with accuracy across the entire tradespace of Mach, Altitude, and Angle of Attack, and numerical solutions, which suffer from high computational cost and time required. This work evaluates analytic solutions, conducts CFD analysis, and develops a data-driven surrogate modeling approach for the prediction of the surface heating across various operating conditions while also accounting for real gas effects. A novel multi-model deep learning based surrogate modeling framework is developed to predict physical flow properties, which are then used for the temperature and heat flux predictions based on the given initial conditions. Kestrel is utilized to generate the CFD based training dataset for model development and for validation. A validation study is presented for the developed modeling framework comparing predictions with analytic solutions and CFD results. The results highlight superior performance as compared to the analytic solutions and good agreement with the CFD results (including unseen data) showing promise as a surrogate modeling framework to quickly predict the surface heating across a wide tradespace of Mach, Altitude, and Angle of Attack for hypersonic applications.