Adaptive Sampling to Improve Aerodynamic Predictions for Reusable Rocket-Powered Vehicles

Abstract

Any vehicle which will be subjected to a wide range of flight conditions must have sufficient control authority to maintain vehicle attitude at all times, favoring designs which exhibit small pitching moments at each relevant flight condition. Designing such a vehicle becomes more difficult when the rapid analysis tools common in the early design process are not sufficiently accurate, forcing the application of more accurate analyses with the associated increases in computational expense. As a result, design space exploration and surrogate modeling becomes difficult or impossible for most computational budgets. In this effort, an adaptive sampling technique is described which identifies cases that improve prediction uncertainty in regions of the design space with desirable characteristics – i.e., where all relevant pitching moments are small. This technique enables creation of surrogate models which are more accurate over the response ranges of interest without excessive data requirements, and in particular enables the preferential sampling of the regions of greatest interest: the families of designs which are likely to trim for all relevant flight conditions.