Using Expected Information Gain to Design Aerothermal Model Calibration Experiments

Abstract

The design of hypersonic air vehicles involves coupled, multi-physics interactions, which are predicted through computational models of various levels of fidelity and accuracy. To reduce uncertainty and improve predictive capability, these models are calibrated with experimental data. Since the number of experiments is often limited, especially those conducted for structures undergoing the combined loading of hypersonic flight, optimal data collection is of great importance for uncertainty reduction and model validation. In this research, the maximum expected information gain is used to determine which wind tunnel specimen geometry, instrumentation locations, and observables are projected to be most informative for Bayesian calibration of the uncertain parameters of an aerothermal model. Higher fidelity simulations and synthetic experimental data are used to measure and compare the actual information gain from optimal designs to the expected information gain. It was observed that geometries and instrumentation locations at the limits of the design space provided the maximum expected information gain. Additionally, tests to measure the output of the furthest downstream model in the Bayesian network were favored due their ability to calibrate the full set of uncertain parameters.