Speaker
Description
Predictive models are a cornerstone of NASA’s ability to design missions to outer planets. For hypersonic entry into an atmosphere, chemical rate data, excitation cross sections, and emission probabilities are a necessary precursor for these models to compute the radiative heat flux on a vehicle. This data is inferred from ground based test facilities which attempt to reproduce certain aspects of flight conditions in order to best inform these computational models.
The NASA electric arc shock tube (EAST) facility represents one such data source. EAST generates high velocity shockwaves in various gas compositions. The emission for this shock is imaged using high speed spectroscopic techniques, yielding two-dimensional images of the emitted spectra at various positions around the shock.
Computational models are be employed to make a prediction of the images that EAST generates. Rarely do the predictions match the experiment data at all positions and wavelengths. It is expected that much of this discrepancy is due to heritage chemical data which is outdated. However, because this chemical data has such a non-linear coupling, it has been found that using improved rates can increase modeling error. Furthermore, a rigorous comparison of the model variance to the EAST uncertainties has not been performed.
In order to improve the modelling accuracy, it is necessary to take a more holistic, statistical approach. First, a sensitivity analysis of the computation model is performed, analyzing all chemical rate data. Once the most important parameters are isolated, a Bayesian inference is performed against the EAST data to determine a improved probability distribution for these parameters.
The computational models used in this study are DPLR and NEQAIR. DPLR is a flow solver used at NASA Ames to predict hypersonic flow environments. For this study, DPLR is run in ‘space marching’ mode, a one dimensional technique which calculates the temperature and number density of a fluid behind a shock. This data is passed to NEQAIR, a line-by-line radiative solver which predicts the emitted spectrum and radiative heating from a flow. NEQAIR uses an internal database to find the quasi-steady-state (QSS) excited state populations of each species in the flow. It then constructs each spectral line from all allowed transitions between these states and performs the necessary spectral convolutions to yield a prediction of the EAST facility.
Monte Carlo techniques are using to sample the DPLR and NEQAIR chemical databases and create a database of predictions of EAST’s pure Nitrogen shots. From these, the Sobol indices are calculated for each chemical rate parameters. It is shown that parameters both in DPLR and NEQAIR are responsible for the variance of the model outputs. The largest five are then used in a Monte Carlo Bayesian inference, using an adaptive delayed rejection Markov chain Monte Carlo (DRAM-MCMC) to calculate posterior probability distributions on these parameters.
Summary
A Bayesian inference is performed on the chemical parameters used in NASA’s hypersonic fluid and radiation solvers. The chemical data is calibrated to experimental data from NASA’s Electric Arc Shock Tube (EAST) facility.
