Dimitri Marvis
Georgia Institute of Technology
Blunt-body entry vehicles exhibit complex, stochastic aerodynamic behavior due to the unsteady recirculating wake; this behavior is further exacerbated by deceleration and oscillation during flight. These oscillations are governed by the vehicle’s inherent dynamic stability which is historically quantified through data reduction and system identification from ballistic range campaigns. Recent progress in computational fluid dynamics (CFD), such as the POST2/FUN3D environment developed at the Georgia Institute of Technology (Georgia Tech), has enabled validated CFD-in-the-loop flight simulations which are capable of predicting blunt-body trajectories in six degrees of freedom (6DOF). This work will move beyond the paradigm of linearized aerodynamics and its associated simplifying assumptions by leveraging recent progress in machine learning and reduced order modeling techniques to predict blunt body aerodynamics. A family of semi-supervised Reduced Order Models (ROMs), which use Proper Orthogonal Decomposition (POD) and Kriging regression will be used to identify coherent structures within a flow field and leverage knowledge of these structures to predict flow fields at unseen flight conditions and vehicle motion states. These models predict surface pressure and shear force distributions on the blunt-body, which can be used to directly compute aerodynamic forces and moments through integration. This will enable ROM-in-the-loop flight simulation where a traditional aerodynamic database can be replaced. Finally, epistemic uncertainty will be quantified by anchoring the ROM to experimental results via Bayesian calibration.
