Calibrating hypersonic turbulence flow models with the HIFiRE-1 experiment using data-driven machine-learned models

Abstract

In this paper we study the efficacy of combining machine-learning methods with projection-based model reduction techniques for creating data-driven surrogate models of computationally expensive, high-fidelity physics models. Such surrogate models are essential for many-query applications e.g., engineering design optimization and parameter estimation, where it is necessary to invoke the high-fidelity model sequentially, many times. Surrogate models are usually constructed for individual scalar quantities. However there are scenarios where a spatially varying field needs to be modeled as a function of the model's input parameters. We develop a method to do so, using projections to represent spatial variability while a machine-learned model captures the dependence of the model's response on the inputs. The method is demonstrated on modeling the heat flux and pressure on the surface of the HIFiRE-1 geometry in a Mach 7.16 turbulent flow. The surrogate model is then used to perform Bayesian estimation of freestream conditions and parameters of the SST (Shear Stress Transport) turbulence model embedded in the high-fidelity (Reynolds-Averaged Navier-Stokes) flow simulator, using

Artificial Intelligence for Earth System Predictability (AI4ESP) (2021 Workshop Report)

“…Salter et al (2019) demonstrated the scalability and accuracy of these approaches to the Canadian atmosphere model, CanAM4. Chowdhary et al (2021) have successfully applied these surrogate techniques to high-fidelity exascale simulations in aerodynamics, which have similar computational resource burdens to large-scale climate models.…”

“…This is oftentimes neglected, because training is much more expensive, i.e., requiring multiple fits with different models. Swischuk et al (2019), Chowdhary et al (2021), andPenwarden et al (2021) are among few authors addressing this problem, and even fewer provide automated ML tools and software to implement methodologies such as tesuract (Chowdhary 2022). Furthermore, hyperparameter tuning can sometimes reveal surprising results-taking existing approaches like multivariate polynomial regression and showing that, in some cases, it can be competitive with or even better than state-of-the-art approaches like neural networks (Chowdhary 2022).…”