State-of-the-art computational tools are inadequate in their ability to provide accurate information in a time efficient manner for integration into the design of hypersonic weapons systems. This is especially true of new vehicle configurations that impose extreme challenges on predictive modeling of aerothermal loadings. In design applications of realistic weapons systems, it is well-recognized that Reynolds-Averaged Navier-Stokes models (RANS) offer the only feasible physics-based option for turbulent flow modeling. However, there are numerous examples where RANS based approaches (that have largely been constructed and calibrated in incompressible flows) are insufficiently accurate to predict critical quantities of interest in hypersonic environments, such as wall heat fluxes. However, recent advances in numerical methods for high speed flows in complex geometries, wall modeling, and use of contemporary compute architectures (e.g., GPUs) has made large eddy simulations (LES) and even direct numerical simulations (DNS) feasible in targeted situations of unit and subsystem problems. We seek to leverage the availability of selected high-fidelity data relevant to hypersonic flow phenomena (shock/turbulent boundary layer interaction, supersonic boundary layers with strong viscous heating) to make improvements in RANS closure models. Specifically, we will utilize novel data-driven and machine learning approaches to augment existing RANS models to improve their accuracy in a range of hypersonic flows. These data-driven augmentations critically consider constraints of model consistency with the governing equations and consider the impacts of erroneous or uncertain input data on the closure model training process. The efficacy of these improvements to RANS closures will be demonstrated in a variety of hypersonic test cases including shock/boundary layer interaction problems, diabatic turbulent boundary layers, and transitional flows.
Benefit: Cascade Technologies has had prior contact with several industrial clients who have expressed interest in the use of the tools for high speed and hypersonic flow applications. These companies (Boeing, SpaceX, Raytheon, Lockheed Martin) have suggested that they would be amenable to software benchmarking studies to showcase improvements in predictive accuracy within these environments. While LES/DNS offerings from Cascade have satisfied many accuracy targets, the cost envelope associated with a purely high-fidelity solution strategy would be untenable in a design environment. The multi-fidelity approach suggested herein that bootstraps the results of select high-fidelity calculations to remedy shortcomings in lower fidelity models would address existing obstacles. Moreover, as the US military is the largest driver of the commercial hypersonics market, successful use of the improved Cascade software by the US government would signal software readiness for major OEMs. There would additionally be immediate spillover effects for existing users of the Cascade flow solvers with the DoD (NAVAIR, Army Research Laboratory, Air Force Research Laboratory) who also have external contacts with the commercial aerospace industry.
Keywords: data-augmented RANS models, data-augmented RANS models, hypersonics flows, large-eddy simulation, direct numerical simulation
