Geometric Design of Hypersonic Vehicles for Optimal Mission Performance using Machine Learning

Recent advances in highly efficient algorithms and high-performance computing allow the construction of an integrated design framework wherein the traditionally segregated disciplines of airframe design and trajectory design are coupled together in order to undertake the design and optimization of hypersonic vehicles as integrated systems. The particular interest in this paper is the potential approach to incorporating high-fidelity aerodynamic models in the hypersonic trajectory optimization problem, incrementally varying the geometric parameters of the vehicle to observe induced performance variations, and employing Bayesian optimization and machine learning algorithms to optimize the vehicle geometry for specified mission profiles. First, the exigency for considering high-fidelity aerodynamic models is justified. Then energy-based problem formulations for hypersonic trajectory optimization are introduced. A panel method based on the modified Newtonian flow theory and Eckert’s reference model is used to produce high-fidelity aerodynamic force and heating coefficients, based on which a pseudospectral optimal control package is used to solve for optimal trajectories. Finally, an iterative procedure employing the Bayesian optimization and machine learning is established to successively search for the geometry that enables the optimal mission performance. Preliminary results demonstrate the feasibility and advantage of the developed approach.