Towards certification by analysis : large-eddy simulations of commercial aircraft across the flight envelope
- This dissertation presents a summary of numerical simulations performed using several realistic aircraft models, both in landing and cruise configurations and describes the current state and predictive capabilities of high-fidelity Large Eddy Simulation (LES) applied to realistic commercial aircraft configurations. The objective of the selected cases is to demonstrate a level of capability and cost effectiveness of LES for the prediction aeronautical flows of engineering significance that will make LES a useful tool for routine use in the aerospace industry. One of the selected model problems is the Japanese Exploration Agency Standard Model (hereafter JSM). This configuration was selected due to the recent interest garnered by its featuring in the AIAA Third High-Lift Prediction Workshop where 35 participants submitted a total of 79 data sets of CFD results predicting the integrated forces and moments across the lift curve. Except for a few participants who used unsteady techniques (unsteady RANS, Lattice-Boltzman VLES, or delayed DES), most calculations presented in the workshop were steady RANS simulations deploying variants of the Spalart-Allmaras or SST models for the Reynolds stress closure. A key takeaway from this workshop series has been that the accuracy of steady RANS techniques has plateaued, particularly near stall and in off-design conditions. Large variations from different solvers were also observed, even when the same models/gridding strategies were employed. The calculations described in this dissertation leverage an LES approach in which prohibitive cost requirements associated with wall-bounded turbulence at realistic Reynolds numbers necessitate the introduction of wall models to ameliorate the stringent grid requirements associated with wall-resolved LES. Equilibrium wall models in which the unsteady, convective, and pressure gradient terms appearing in the turbulent boundary layer equations are assumed to be in balance have been shown to work well in flows of interest and are used in this work. The combination of these physics-based modeling choices along with recent advances in computing hardware (such as GPU-based high-performance computing clusters) and low-dissipation numerical methods for LES have made LES a powerful tool for use in informing design decisions in industry. Validation efforts such as the one presented in this dissertation are a key component towards building confidence in the predictive capability of this emerging computational fluid dynamics (CFD) paradigm. The studies presented herein which use the JSM configuration in general showed good prediction of the CL across the lift curve with the coefficient of lift at maximum lift, CL, max, being predicted to within 3 lift counts of the experimental value (i.e. within the tolerances required by the aerospace industry of Delta CL less than or equal to 0.03 at maximum lift). The grid point requirements to achieve this level of accuracy are reduced compared to recent estimates (even for wall modeled LES), with the solutions showing systematic improvement upon grid refinement on grids numbering up to 150 Mcv. Investigations which included the wind tunnel facility were made in order to address one of the key deficiencies of the free air calculations: the incorrect prediction of the stall mechanism which in the free air cases was missing a large inboard separation. It was found that inclusion of the tunnel facility in the simulations did indeed cause the flow to separate at the leading edge of the wing root and the presence of a "break" in the pitching moment curve associated with this phenomenon was captured, building confidence in the predictive capabilities of LES for high-lift flows, especially in the presence of wind tunnel effects. Another key finding from the JSM calculations was the fact that dynamic subgrid-scale turbulence models such as the DSM model in which the coefficient of the subgrid viscosity is modulated in space and time based on the finest resolved length scales of turbulence on the LES grid were shown to outperform static coefficient models such as the Vreman model in their ability to predict both trailing edge flap separation and juncture flow separation. In order to establish the robustness of these LES methods with respect to aircraft configuration and Reynolds number, further explorations were carried out in the high-lift flow regime with the High-Lift Common Research Model (hereafter CRM-HL), which has become the new benchmark flow in the community following the Fourth High-Lift Prediction Workshop. The CRM-HL experiment is run at nearly $3 \times$ the Reynolds number of JAXA and features a more challenging geometry, with slat suction peaks that reach up to $-20$ in $C_p$ compared with suctions up to $-10$ in the case of the JSM. Both of these differences impose stricter grid resolution requirements and pose a greater predictive challenge to LES models (both SGS and wall models). In this effort, we sought to apply lessons related to modeling choices and gridding approach from the JSM case to accurately predict aircraft maximum lift. For the CRM-HL configuration, a more comprehensive grid refinement study was performed compared with the JSM case. In this case, full alpha sweeps were carried out on grids numbering up to 1.5 Bcv. The solutions exhibited a systematic decrease in their sensitivity to grid resolution with each successive refinement until the 1.5 Bcv and 400 Mcv were nearly grid-insensitive in their predictions of lift, drag, pitching moment, and surface pressure, giving indication for the first time of what the grid resolution requirements for convergence of engineering quantities of interest for practical aircraft flows are. The 400 Mcv calculations are carried out on 600 V100 GPU's within 5 hours of walltime to reach statistical convergence of the forces/moments. For the CRM-HL again wind tunnel effects were explored, this time with a more complete representation of the wind tunnel facility which included not just the test section as in the case of JSM, but also the inlet, contraction, diffuser sections as well. This necessitated developing a workflow in LES for clearing the long transient timescales associated with the tunnel facility acoustics which was done by grid sequencing solutions from coarser to successively finer grids. The findings from including the calculations which included the tunnel facility were less clear-cut than in the case of the JSM (which is hypothesized to be due to the presence of discrepancies in the tunnel sidewall boundary layer development between the QinetiQ experiment and simulations), but did generally show a trend towards improving the character of the wing root separation pattern at stall. Explorations with numerical boundary conditions to delay nacelle lip separation are made and a novel non-Boussinesq subgrid model is explored to assess its impact on the persistent issue of lift over-prediction by LES at low angles of attack in high-lift flows. The model was shown to largely rectify similar issues in canonical flows, but its effect on the prediction of the trailing edge flap flow was to cause over-separation and a corresponding lift under-prediction compared with traditional SGS models such as the Dynamic Smagorinsky. Finally, an extension to the transonic flow regime was pursued in which LES was applied to the transonic NASA Common Research Model (CRM), which has been the focus of several Drag Prediction Workshops (DPW). Sensitivities to laminar-to-turbulent transition, symmetry plane treatment, and grid topology are established and suggestions for best practices in these simulations are made. Specifically, the inclusion of experimental trip dots is found to be an important component of accurately prediction surface pressures and drag coefficients of transonic low angle of attack flows as the flow in these cases is more streamlined than in their high-lift counterparts and the integration of surface skin friction accounts for nearly 50% of the drag coefficient compared to only 5% in high-lift flows. Additionally, an important observation is made regarding symmetry plane boundary conditions in LES. Simpler problems, such as a Blasius boundary layer flow encountering a trip dot on a flat plate are studied in order to draw conclusions pertaining to transition behavior in a controlled environment before deploying the trip dots on the full aircraft configuration. It is also found that the simulations exhibit strong sensitivity to whether they are run with the entire full-span geometry of the aircraft is including or whether a numerical symmetry plane boundary condition is used, with the full span simulations showing much better agreement with the experiment (which uses a full span aircraft). This suggests that in LES, the numerical shortcut of using a symmetry plane on the aircraft center plane may be inappropriate for aircraft simulations, potentially owing to the instantaneous asymmetries that arise when the component of turbulent fluctuations normal to the symmetry plane cause flow to cross this plane. Additionally, anisotropic stranded Voronoi meshes are used and found to outperform their isotropic Voronoi HCP counterparts in this flow as they allow for additional targeted resolution in the wall-normal direction for better grid support of near-wall turbulent structures. Overall, it is found that promoting transition to turbulence via an array of cylindrical trip dots, using the full span of the aircraft, including the experimental sting mounting system in the simulations, and leveraging prismatic boundary layer grids all tend to improve the quality of the LES solutions. Again, a novel non-Boussinesq subgrid-scale model and sensor-based wall models are applied to this complex external flow over an aircraft. This time, promising results are obtained, particularly on relatively
- coarse grids numbering about 100 million control volumes (100 Mcv). Finally, the angle of attack of the transonic flows is pushed to high values in order to assess the ability of LES to predict the onset of aerodynamic buffet. This study is performed on a NACA 0012 flow and shows promising early results, though there are discrepancies of 10-15% in the precise prediction of the angle of attack of buffet onset and of the frequency of the shock oscillations once buffet is sustained. Still, these efforts offer preliminary indication of the accuracy of LES in this flow regime and propose a tractable workflow for identification of buffet onset using rotating flow solver technology to dynamically modulate the angle of attack of the airfoil section with time. Overall, these efforts serve as important steps towards building confidence in the predictive capabilities of LES applied to complex external aerodynamic flows at the edge of the flight envelope, both in high-lift and transonic regimes, which is a key ingredient in the industry-wide push towards Certification by Analysis.
|Type of resource
|electronic resource; remote; computer; online resource
|1 online resource.
|Goc, Konrad Andrzej
|Lele, Sanjiva K. (Sanjiva Keshava), 1958-
|Degree committee member
|Degree committee member
|Lele, Sanjiva K. (Sanjiva Keshava), 1958-
|Stanford University, School of Engineering
|Stanford University, Department of Aeronautics and Astronautics
|Statement of responsibility
|Konrad Andrzej Goc.
|Submitted to the Department of Aeronautics and Astronautics.
|Thesis Ph.D. Stanford University 2023.
- © 2023 by Konrad Andrzej Goc
- This work is licensed under a Creative Commons Attribution 3.0 Unported license (CC BY).
Also listed in
Loading usage metrics...