Bayesian Calibration of Aerothermal Models for Hypersonic Air Vehicles

Predictive capabilities of coupled fluid-thermal-structural models for aerodynamic pressure and heating are investigated for a flat plate and spherical dome protuberances subjected to hypersonic flow conditions. Aerothermal test data from hypersonic wind tunnel experiments are used in a Bayesian network to calibrate uncertain model parameters and model discrepancies for aerodynamic pressure and heat flux predictions. A segmented Bayesian model calibration approach is explored as an alternative to full, simultaneous calibration for reduced computational cost. To quantify the viability and potential benefit of isolating calibrations of models in the network, segmented and simultaneous calibration are compared using the Kullback-Leibler distance and Bayes factor metrics. For model calibration using the aerothermal data, the segmented approach yielded greater prediction uncertainty than the simultaneous approach due to inherent correlations lost through the segmentation. However, the segmented approach resulted in calibration convergence with fewer samples, and the Bayes factor metric indicated good prediction agreement with simultaneous calibration.