PhD Defense
 

Jehan Dastoor

(Advisor: Dr. Dimitri Mavris)

Bayesian Calibration of Data-Driven Reduced-Order Models to Heterogeneous Data with an Entry Vehicle Application

Monday, August 17
1:00 pm
Weber (SST II) CoVE
 

Abstract::

Entry, Descent, and Landing (EDL) is one of the most crucial phases of a space mission, involving flight from atmospheric interface, deceleration through an atmosphere, and ultimately delivering a payload to a body's surface. Due to the exorbitant cost of flight tests, EDL aerodynamicists rely on ground-based tests and numerical simulations to understand aerodynamic phenomena. The culmination of possibly up to a decade of testing is an aerodynamic database, which characterizes the vehicle's aerodynamics at each flight condition of interest. These databases combine multiple sources of aerodynamic information and uncertainties to provide a single prediction at each point along a trajectory, but are often costly and time-consuming to construct, as well as limited in their ability to capture dynamic responses.

Reduced-Order Models (ROMs) have been suggested as an alternative to traditional databases, where instead of predicting aerodynamic coefficients, the entire surface pressure and shear field is predicted. These models have been shown to produce similar results to higher-fidelity CFD solutions with a lower online computational cost, allowing them to be used in trajectory simulations and dispersion analysis. However, ROMs currently only utilize data from CFD solutions, which are known to have discrepancies when compared to experimental data.

This motivates the research objective of this dissertation: to incorporate experimental trajectory measurements from a ballistic range, which is often considered the most trusted ground source of dynamic data for entry capsules, into an aerodynamic ROM. This is challenging because surface pressure fields are not currently measured experimentally for entry capsules, requiring heterogeneous fusion techniques to be used. This dissertation proposes the use of Bayesian inference, specifically black-box variational inference, to determine state-dependent corrections to a ROM with uncertainty.

The proposed method is developed through three key research areas. The first area determined where corrections should be applied within a ROM framework by comparing candidate field- and force-based correction formulations. The field-based approach provided improved numerical stability, parameter identifiability, and inference performance and was therefore selected. The second research area evaluates the feasibility of the heterogeneous inference problem and whether trajectory observations alone contain sufficient information to identify aerodynamic corrections. Using pseudo-ballistic range data, it was shown that aerodynamic corrections could be recovered from trajectory observations with greater accuracy than when using reconstructed aerodynamic loads. The final area investigated how the correction should be defined to capture nonlinear discrepancies. Single- and multi-trajectory inferences were performed using three experimental ballistic range trajectories of the Genesis capsule to compare different model forms and state inputs. A Gaussian process dependent on the Mach number, wind-axis angles, and wind-axis rates was shown to outperform the other models in most cases. However, the results were not entirely conclusive, given extrapolation concerns in the multi-trajectory results.

Finally, the derived method was demonstrated against a relevant aerodynamic database. The corrected ROM appears to capture amplitude growth more effectively and improves the characterization of aerodynamics for small attitude oscillations, where the dynamics are most nonlinear, but suffers from overfitting, limiting its performance on the validation.

Committee:
Dr. Dimitri Mavris (advisor), School of Aerospace Engineering
Dr. Graeme J. Kennedy, School of Aerospace Engineering
Dr. Lakshmi N. Sankar, School of Aerospace Engineering
Dr. Bradford E. Robertson, School of Aerospace Engineering
Dr. Christian Perron, School of Aerospace Engineering
Dr. Hisham M. Shehata, Atmospheric Flight Entry Systems Branch NASA Langley Research Center, Analytical Mechanics Associates