Session M28: Computational Fluid Dynamics: General I

Modeling Shock-Separated Flow with the Active Model Split

The predictive accuracy of typical hybrid RANS/LES models is limited by several shortcomings, such as modeled-stress depletion and a dependence on scalar grid measures. The "active model-split" (AMS) hybrid RANS/LES model was developed by Haering, Oliver, and Moser to address these shortcomings. In this model, the mean and fluctuating portions of the stress are modeled separately; this allows for a consistent treatment of the mean flow even in the presence of resolved fluctuations.  Part of this hybrid model is an active forcing, which generates turbulent fluctuations in regions that support LES-type fluctuations.  In this presentation, some recent results are shown using the active model-split for shock-separated flows. The effects of the active forcing are examined, including the effects on anisotropy and acoustics. Superior predictive accuracy over RANS is shown for some quantities of interest, such as the shock location.   

Large-eddy simulation of flow around a 6:1 prolate spheroid.

The flow around a 6:1 prolate spheroid at angle of attack is characterized by a range of pressure gradients from favorable on the windward side to adverse on the leeward side, crossflow on the side, and open and close separation of the windward side. Experimental campaigns often trip the flow in order to ensure uniform transition and repeatability of the measurement; however, designing a trip that effectively triggers transition on both sides of the spheroid without over-tripping is challenging due to the range of pressure-gradient. In this context, the wall-resolved, trip-resolved flow around the spheroid at 20^o angle of attack and Reynolds number of 4.2 million is studied using large-eddy simulation. The trip consists of cylindrical posts uniformly distributed azimuthally and placed close to the nose of the spheroid. The sensitivity of the flow to location, geometry and azimuthal distribution of the trip will be evaluated. The results will be compared to the existing experimental data.   

Optimal wall-modeling strategies using data.

Traditional wall modeling approaches for LES rely on a RANS-like description near the wall or computation of wall stresses using an equilibrium mean velocity profile. These models do not adapt to the underlying numerical method and are susceptible to sub-grid modeling errors and under-resolution. To resolve these inconsistencies, variational multiscale (VMS) decomposition of the flow-field is performed by the formal projection of the DNS solution on coarse-scale finite element basis functions. A direct consequence of the VMS decomposition is that the optimal coarse-scale solution and slip velocity are naturally obtained, thus defining an optimal wall model. To reach near-optimal wall-modeling performance, a model consistent training approach is employed to infer the parameters of slip and wall stress-based wall models that allow the recreation of velocity statistics present in the optimal coarse-scale solution.   

Filtering, averaging, and scale dependency in homogeneous variable density turbulence

We investigate relationships between filtering and Reynolds-averaging statistics as a function of length scale. Generalized central moments are expressed as inner products of generalized fluctuating quantities q′(ξ, x) = q(ξ) − q(x), representing fluctuations of a field q(ξ) with respect to its filtered values at x. These expressions provide a scale-resolving (SR) framework, with statistics, governing equations and realizability conditions at any length scale. In the small-scale limit, SR statistics become zero. In the large-scale limit, SR statistics and realizability conditions are the same as in the Reynolds-averaged description. Using DNS of homogeneous variable density turbulence, we diagnose Reynolds stresses T_ij, resolved kinetic energy k_r, turbulent mass-flux velocity a_i, and density-specific volume covariance b, defined in the SR framework. At intermediate scales, the governing equations for these variables exhibit interactions between terms that are not active in the Reynolds-averaged limit: in the Reynolds-averaged limit, b follows a decaying process; b peaks at intermediate length scales, where it is a balance between production, redistribution, destruction, and transport. This work supports the notion of a hybrid, length-scale adaptive model.   

Dynamics of the transition region (grey area) in sensitized hybrid (U)RANS/LES modeling framework

Typically, a simple extension of RANS (Reynolds-Averaged Navier-Stokes) turbulence models, designed and calibrated in the steady framework, to an un- steady calculation (URANS) will result in a steady flow field, particularly for stable flows, i.e., attached and mildly separated flows. URANS methods fail to capture energetic and essential unsteady dynamics for globally unstable flows, even for massively separated flows. Therefore, an (instability) sensitized eddy- resolving model is designed based on the k-ζ-f model [1] and applied to attached and separated wall-bounded flow configurations – channel and periodic hill flows. Specifically, the model maintains the effects of vortex stretching, which appear to capture a departure from equilibrium more effectively. Particular focus has been placed on identifying the requirements for accurate description of grey re- gion dynamics between URANS and Large-eddy simulation mode. Additionally, the sensitivity of the resulting hybrid model regarding mesh resolution, mesh design, and the underlying RANS model has been investigated and compared to the IDDES methodology. The model demonstrates a robust and consistent be- havior when applied to different flow regimes at different mesh resolution/design and Reynolds numbers.

[1] DR Laurence, JC Uribe, and SV Utyuzhnikov. A robust formulation of the v2- f model. Flow, Turbulence and Combustion, 73(3):169–185, 2005.   

A comparison of hybrid RANS/LES models across the complete transition from URANS to DNS

Non-zonal hybrid RANS/LES models are derived from the mathematical invariance of the filtered Navier-Stokes equations such that a modified RANS closure may operate as a consistent LES model at higher resolutions than typical of a URANS. Despite this, however, very few individual or comparative studies have been conducted on hybrid models to determine how accurately and controllably each model transitions from URANS mode (model stresses, τ_ij, are independent of Δx) to LES mode (τ_ij ~ Δx²), to an effective DNS (τ_ij ≈ 0).  In this talk we will compare the performance of standard k-ε formulations of several hybrid models – including Partially-Averaged Navier-Stokes (PANS) and Very Large Eddy Simulation (VLES), among others – against the dynamic Smagorinksy model using initial conditions filtered from a pseudospectral DNS of incompressible decaying HIT on computational meshes ranging in size from a single-cell mesh (URANS) up to 512³ (DNS). We will compare each model against the DNS truth values of typical RANS and LES quantities of interest, as well as the grid-size dependence of the model stress, as we systematically vary both the simulation mesh size and “resolution parameter”, a tunable coefficient most models employ to hybridize the underlying k-ε model.   

Comparison of RANS, LES and wind tunnel experiments for the calculation of wind loads on a low-rise building.

Wind-resistant design of buildings and their components plays an important role to reduce losses due to extreme wind events. Computational Fluid Dynamics (CFD) provides a powerful tool to calculate wind loads on buildings, but validation is required to assess their performance. Reynolds-averaged Navier-Stokes (RANS) simulations support fast calculations, but have lower fidelity and require empirical relationships to calculate turbulent statistics, such as the root-mean-square pressure coefficients (C_p). In contrast, Large-eddy simulations (LES) are computationally more expensive, but have higher fidelity and all statistics can be calculated directly. The objective of this study is to compare mean and fluctuating C_p values between wind tunnel experiments, RANS and LES of a low-rise office building. The analysis considers 1) the isolated building, and 2) the building in its urban environment, to evaluate interference effects for different wind directions. Preliminary results for the dominant wind direction show good agreement for the mean C_p. Future work will extend the comparison to turbulent statistics.   

Large Eddy Simulation of a Realistic Aircraft Configuration

We discuss the application of Large Eddy Simulation (LES) to complex aircraft geometries and the prediction of quantities of interest in these settings. The test case is that of the NASA High Lift Common Research Model. Calculations are carried out using the charLES finite volume flow solver. Lift and drag forces as well as pitching moment data are validated against experimental data collected at the QinetiQ test facilities, with sectional static pressure measurements used to corroborate the accuracy of the integrated quantities. Grid convergence studies across the lift curve reveal systematic improvement of the solution with increasing grid density. The lift coefficient at maximum lift and the stall angle of attack are predicted to near the accuracy tolerances required by industry for use in design and certification by analysis, while drag also shows very good agreement with the experimental measurements on the finest grid. The key missing feature in the free air simulations is a large wing root separation bubble at stall. We find that inboard separation appears when wind tunnel effects are included in the simulations. The LES calculations, using the charLES code, leverage key technologies such as low dissipation and entropy preserving numerical schemes, physics-based subgrid and wall models, and rapid high-quality unstructured Voronoi grid generation to enable tractable simulation turnaround times. Calculations on the GPU accelerated charLES flow solver achieve statistical convergence within 7 hours on a grid numbering more than half a billion cells at a flight condition in the post-stall regime.   

Modeling Effects of Transonic Axial Compressor Performance Degradation due to Rotor Blade Damage

Particle ingestion into aircraft engines during operation causes considerable damage to the upstream axial compressor. This results in significant performance degradation of the compressor and hence, the entire gas turbine engine. Our computational study aims to model the effects of damage of the Stage 1 Rotor blade of the T700-401C engine on the performance of the axial compressor. Several damage modes of the Stage 1 Rotor blade were replicated using additive manufacturing and the resultant geometries were 3D optically scanned and incorporated into our CFD studies. The flow studies were performed in STAR-CCM+. Half-stage and full-stage studies were performed. The analyses focused on overall performance parameters namely, stagnation pressure ratio, isentropic efficiency, stagnation temperature ratio and power output.   

Data-Driven Unsteady Aerodynamic Modeling for Transient Blade Response Prediction

Aircraft intermittent combustion engines often incorporate turbochargers adapted from ground-based applications to improve their efficiency and performance. These turbochargers operate at off-design conditions and experience high-cycle fatigue brought on by aerodynamically-induced blade resonances. A reduced-order model of the aeroelastic response of general fluid-structural configurations is developed using the Euler-Lagrange equation informed by numerical data from uncoupled computational fluid dynamic (CFD) and computational structural dynamic (CSD) calculations. The structural response is derived from a method of assumed-modes approach. The unsteady fluid response is described by a modified version of piston theory augmented with an inhomogeneous source term developed utilizing data-driven methods. The reduced-order model is applied to a classical panel flutter scenario as well as a more complex turbocharger turbine wheel configuration. The capability of the reduced-order model to predict the presence of flutter in both scenarios from a subset of the uncoupled numerical simulation data is presented, including a critical discussion of the accuracy and validity of piston theory in this context.