Predicting Corner-Flow Separation in a Wing‑Body Juncture using Scale Resolving Simulations

Overview

The rapid growth of high-performance computing (HPC) in the past decade has enabled the use of state-of-the-art computational fluid dynamics (CFD) methods for predicting aerodynamic flows and has fundamentally changed the aerospace design process. However, as noted in several American Institute of Aeronautics and Astronautics (AIAA) and NASA workshops where CFD predictions are directly tested against highly precise experimental measurements, most Reynolds Averaged Navier-Stokes (RANS) models that solve for the steady state solutions struggle when non-equilibrium flow physics—such as boundary layer separation due to corners/junctions, shocks, and so on—plays a crucial role in overall aircraft performance. One such flow topology where legacy RANS modeling is incapable of accurately predicting flow kinematics occurs in the junctions where the aircraft’s wings meet the fuselage. In this project we address this flow state, underscoring the important role of HPC in enabling the use of scale-resolving simulations to accurately predict the complex fluid mechanics associated with corner-flow separation in a wing-body juncture.

"Nearly all the simulations were run on Electra and Aitken supercomputers. The service provided by the exceptional visualization team at the NAS facility was critical for both assessment of the flow physics and communicating results to a variety of audiences."

Project Details

The primary goal of this work, funded by NASA’s Transformational Tools and Technologies (T3) Project, is to evaluate two scale-resolving methodologies for predicting juncture flows: delayed detached eddy simulation (DDES) and wall-modelled large eddy simulation (WMLES). We performed a thorough comparative assessment with the state-of-the-art steady-state RANS modeling available today to highlight the successes and shortcomings of each approach. Experimental measurements obtained from the 14‑ by 22‑Foot Subsonic Wind Tunnel at NASA’s Langley Research Center were used to perform quantitative comparisons of first- and second-order single-point correlations at various locations of interest of the wing-body model studied. The project is the first of its kind to perform detailed comparative assessment of these two philosophically distinct scale-resolving methodologies on a non-canonical geometry.

Results and Impact

Our simulations successfully demonstrated that the DDES and WMLES methods show measurable accuracy improvements over steady-state RANS for prediction of corner flow separation, justifying their substantially higher computational expenditure. The DDES and WMLES simulations utilized computational grids with up to 196 million and 520 million cells, respectively. The best quality solutions for DDES were obtained at a computational overhead 13 times higher than RANS (on structured overset meshes), and the best quality WMLES solution was approximately four times as expensive as the DDES. However, WMLES with explicit time stepping—thereby lacking any elliptic/linear solves—used approximately the same computational resources as several unstructured steady state RANS solvers on numerical meshes of comparable size. This highlights the readiness of WMLES for predictive external aerodynamics simulations of practical configurations of interest on modern HPC systems.

Why HPC Matters

This project was made possible by the computational resources at the NASA Advanced Supercomputing (NAS) facility. During Fiscal Year 2020 (FY20), approximately 19.6 million core-hours were utilized to develop the modeling and simulation capabilities used in this project. Research and development included code optimization, database generation to study the role of tripping, and development of best practices for scale-resolving simulations of non-canonical geometries. This massive investment is expected to enable routine use of WMLES in a variety of NASA programs and upcoming AIAA/NASA workshops, such as the High Lift Prediction Workshop in FY21. Nearly all the simulations were run on the Electra supercomputer’s Intel Xeon Skylake nodes, and on the Aitken supercomputer’s Cascade Lake nodes. Additionally, the service provided by the exceptional visualization team at the NAS facility was critical for both assessment of the flow physics and communicating results to a variety of audiences.

Related Links

• Scale Resolving Simulations of Corner-Flow Separation in a Wing-Body Juncture
• Scale Resolving Simulations of the NASA Juncture Flow Model using the LAVA Solver
• An Experimental Investigation of a Wing-Fuselage Junction Model in the NASA Langley 14- by 22-Foot Subsonic Tunnel