Detached-Eddy Simulation With Compressibility Corrections Applied to a Supersonic Axisymmetric Base Flow

The separated turbulent flow past an inclined flat plate with sharp leading and trailing edges was computed based on three different simulation approaches for a reynolds number Rec= 20000 and a high angle of attack α=18°. The simulation techniques applied were the Reynolds‐averaged Navier–Stokes (RANS) equations combined with a one‐equation Spalart–Allmaras turbulence model, the large‐eddy simulation (LES) based on an algebraic eddy‐viscosity model, and a hybrid approach known as detached‐eddy simulation (DES) applying a slightly modified Spalart–Allmaras model in the entire integration domain. DES is a non‐zonal coupling technique of RANS and LES developed in the hope of reducing the large computational resources required for LES computations of turbulent flows with practical relevance. However, the objective of the present study was not to compare the resources (CPU time, memory) required for all three techniques but to investigate and evaluate the quality of the predicted results for RANS, DES and LES. For this purpose, a test case which is favourable to the basic concept of DES and which places special emphasis on the LES part of the DES concept was chosen. This last issue was important since the modified Spalart–Allmaras model applied as a subgrid scale model in the LES part is not as well validated as other models usually applied in LES. For this purpose, an LES prediction on a very fine grid served as a reference case for the evaluation. For all three techniques, predictions with different grid resolutions were carried out and compared with each other based on important integral parameters (e.g. Strouhal number, mean drag and lift coefficients and their standard deviations), the instantaneous and time‐averaged flow structures, and higher‐order statistics. As expected, the pure RANS calculation, although applied as unsteady RANS, failed to predict the unsteady characteristics of the separated flow. In contrast, the DES approach yielded reasonably the shedding phenomenon and some integral parameters. However, analysing the results in more detail led to remarkable deviations between the DES and LES predictions also when the same grid resolution was applied. Especially the free shear layer originating from the leading edge of the plate was not well reproduced by DES, showing strong deficiencies of the model applied as a subgrid scale model. The reasons for this behaviour of the model were analysed in detail. Two basic causes were identified; the first is given by some near‐wall corrections in the finite Reynolds number version of the model which are not working properly in the LES mode of DES. The second is a modified definition of the filter width typically applied for DES, leading to strongly increased values of the eddy viscosity. A revised version of the S–A model taking both issues seriously into account was used as a subgrid scale model in the LES mode. As a direct consequence, much better agreement with the reference LES solution was found for the DES prediction on the coarse grid, eliminating the deficiencies of the original formulation. Copyright© 2003 John Wiley & Sons, Ltd.

This paper presents results of a Computational Fluid Dynamic (CFD) study of a sharp-edged symmetric flat strut at Reynolds number 500,000 based on chord at zero degrees angle of attack, and the subsequent estimation of far-field noise generated at the trailing-edge. Flow field results obtained using Detached Eddy Simulation (DES) modeling and Reynolds-averaged Navier Stokes (RANS) modeling techniques are compared with empirical wind-tunnel data. The flow is observed to be physically complex in nature, exhibiting numerical solutions that are sensitive to the mesh grid and freestream turbulence intensity. Although originally developed for use specifically with RANS-generated flow data, the RANS-based Statistical Noise Model (RSNM) technique, which estimates far-field noise from mean turbulence data via an acoustic Green’s function and a statistical turbulence correlation model, is used here to estimate far-field noise spectra from both RANS and DES flow data. Far-field noise is also estimated from the DES m...

Unsteady cavitating flow with high Reynolds number and significant instability commonly exists in fluid machinery and engineering system. The high-resolution approaches, such as direct numerical simulation and large eddy simulation, are not practical for engineering issues due to the significant cost in the computational resource. The objective of this paper is to provide the approach with Detached-Eddy Simulation (DES) model based on the Reynolds-averaged Navier–Stokes (RANS) equations for predicting unsteady cavitating flows. The credibility of the approach is validated by a set of numerical examples of its application: the unsteady cavitating flows around the two-dimensional (2D) Clark-Y hydrofoil and the three-dimensional (3D) blunt body. It is found that the calculated cavity shapes, cavity lengths and unsteady characteristics by DES model agree well with the experimental measurements and observations. Further analysis indicates that the turbulent eddy viscosity around the cavity and wake region is well predicted by the DES model, which results in the development of large-scale vortexes, and further cavitation instability. The DES model, which exhibits a significantly unsteady 3D behavior, is a more comprehensive turbulence model for unsteady cavitating flows.

Detached Eddy Simulation (DES) is performed for developing turbulent flow of the 270° curved duct at a Reynolds number of 56,690. The curvature ratio on the basis of a centric radius R c and a duct height H is 3.357. Turbulence models adopted are k- ω model for Reynolds Average Navier-Stokes (RANS) equation Simulation and Shear Stress Transport (SST) model for DES. DES is used as the hybrid computation technique combined with RANS-SST and Large Eddy Simulation (LES). Predicted results are compared with measured results including the distributions of Reynolds stresses and the flow characteristics on the symmetric plane of curved duct are presented. Judging from the comparison between the predicted and the measured results, the DES approach is applicable to calculate the developing turbulent flow in a 270° curved duct. 기호설명 U b : 입구단면 평균속도: 덕트의 높이 k : 난류운동에너지 S : 변형률 텐서( S ij ) l t : 난류길이척도 u + : 벽면 마찰 속도 y + : 벽면으로부터 무차원거리 F 2 : 혼합함수(blending function) Δ t : 계산시간간격(time-step)

Flow and heat transfer inside an idealized electronics system is simulated using large‐eddy simulation (LES)‐related approaches that have been validated for simpler canonical flows. These include: Yoshizawa LES, detached eddy simulation (DES), limited numerical scales (LNS) and other hybrid LES–RANS (Reynolds‐averaged Navier–Stokes) approaches including a new ILES (implicit LES)–RANS method. For the ILES, dissipation from the one legged temporal discretization is used to drain turbulence. The use of differential equations, including the Hamilton–Jacobi and Eikonal, to model turbulence distance functions is explored. The Hamilton–Jacobi is shown to be especially compatible with the zonal RANS–ILES approach and the Eikonal with DES. Performances of the LES‐related methods are compared with explicit algebraic stress unsteady RANS (URANS) results and also measurements. Considering the problem complexity, generally, for all methods, predicted mean velocities and turbulence intensities are in a reasonable agreement with measurements. Average errors are 15 and 25%, respectively. With the exception of the zonal ILES–RANS method, turbulence intensities are under‐predicted. For heat transfer, none of the models performs well giving circa 100% errors. Notably, the LNS performs poorly for both the flow field and heat transfer giving a highly complex RANS–LES interface with inappropriate upstream LES boundary conditions. DES is found impossible to converge. This is partly attributed to the irregular LES–RANS interface arising with the method. All the LES approaches significantly underpredict heat transfer and the URANS over‐predicts. Even the increased flow activity arising from use of the less dissipative ILES element does not prevent the significant heat transfer under‐prediction. Copyright © 2005 John Wiley & Sons, Ltd.

Numerical simulations of the flow and noise generated by two different automotive outside rear-view mirrors are investigated. The mirrors were mounted on the top of a specially designed table for simulations and wind tunnel tests. A decoupled two-stage methodology is applied that requires CFD simulations to collect pressure variations at solid surfaces followed by an acoustic post-processing of the data recorded. For the flow simulations, URANS (Unsteady RANS), DES (Detached Eddy Simulation) and LES (Large Eddy Simulation) turbulence modeling approaches are used. This study is part of an on-going project to develop a transient CFD procedure to accurately predict wind noise generated by the exterior surfaces of a moving ground vehicle. The Curle's formulation in low Mach number limit is used to calculate the acoustic pressure generated by the pressure fluctuations at solid walls. Reasonable agreement (within 5 dB) between measurements and predictions has been achieved both for the flow and sound characteristics (with denser mesh quality). It is found that the URANS and DES approaches underestimate surface pressure fluctuations and therefore, result in underestimated sound pressure levels. The LES approach on the other hand (particularly with the dynamic subgrid viscosity model) provides better agreement for the time-averaged surface pressure distributions and for the sound pressure levels.

Flows around maneuvering ships, submarines and underwater vehicles are usually quite complicated, and often experience three-dimensional open flow separation resulting in unsteady forces and moments that may be detrimental to ship performance. The objective of this investigation is to support more applied maneuvering studies by investigating the computational model performance with respect to approaches to flow simulation methodologies, turbulence modeling, grid resolution and the effects of tripping devices often used in experimental studies. To study these issues simulations are performed of the flow past a 6:1 prolate spheroid, tripped at the nose, and experimentally studied in a series of papers by Simpson et al. Here, we compare predictions from Reynolds Averaged Navier Stokes (RANS), Detached Eddy Simulation (DES) and Large Eddy Simulation (LES) models with experimental data. For the LES computations, different subgrid models are utilized and since the experimental study is carried out using a tripped model, a simple trip model is also developed and tested together with the LES models. Large scatter between the predictions is found, with the DES model and one of the trip-ped LES model showing very good agreement with the data. For LES, the modeling of the trip, which is usually not considered, appears as important as that of the near wall handling and modeling in wall-modeled LES.

High Reynolds number wall bounded flow is here investigated using Large Eddy Simulation (LES), Detached Eddy Simulation (DES) and Reynolds Averaged Navier Stokes (RANS). The first case considered is the fully developed turbulent channel flow at Re 395, 590, 1800 and 10,000. This flow clearly indicates the development of the undisturbed boundary layer and related events, such as streaky structures, hairpin vortices and ejection events. The second case is the flow over an axisymmetric hill in a channel, here the flow contains complex structures such as a turbulent boundary layer with several unsteady separations and reattachments. It is three-dimensional due to both streamwise and spanwise pressure gradients on the lee-side of the hill. The shallowness of the separation region makes the flow a demanding test case for any computational fluid dynamics model. The third case is the flow past an axisymmetric submarine hull with an elliptic forebody and a smoothly tapered stern the DARPA Suboff model AFF-1. This flow case is highly demanding due to the long midship section, on which the boundary layer is developed, in combination with the elliptic forebody and the tapered stern. Both LES and DES performs well in all cases considered, while RANS has slightly lower accuracy in the channel flow and the axisymmetric hull, and fails to predict some flow features for the axisymmetric hill. Also DES has some problems with the axisymmetric hill case, related to the inlet condition of the modified eddy viscosity.

This article is focused on the choice of a suitable turbulence model for simulations of an industrial pump's intake, from the perspective of accuracy and, partially, also the CPU time. Twelve steady-state and transient simulations were made on a fine computational mesh, using turbulence models such as: the shear stress transport (SST), the scale-adaptive simulation (SAS), the Reynolds stress model, the explicit algebraic Reynolds-stress model, the detached eddy simulation and the large eddy simulation (LES). The curvature-correction (CC) option was assessed for the SST and SAS turbulence models. The results were compared with the LES and with published experimental results. Although all the models could predict the main floor vortex, there were still some substantial differences. It can be able to conclude that it is better to use either the SST-CC turbulence model, due to its low-computational resources and far better results than the SST model, or the SAS-CC turbulence model, since its predictions are quite similar to the LES results. In the final step, good agreement with experimental results was shown for a longer simulation with the SAS-CC turbulence model.

Turbulence model adaptability at critical Reynolds numbers and applications in wake control via fairings

Large-Eddy Simulation (LES), Detached-Eddy Simulation (DES) and Scale-Adaptive Simulation (SAS) are increasingly being used as engineering tools to predict the behaviour of complex industrial flows. Often the flows studied have not been examined previously and the required grid resolution is unknown. Industrial users studying these flows tend to be using commercial CFD codes and do not usually have access to high-performance computing facilities. Due to the significant computing times required, it is difficult to undertake systematic grid-dependence studies. There is therefore a risk that LES, DES and SAS will be performed using overly coarse grids which may lead to unreliable predictions. The present work surveys a number of practical techniques that provide a means of assessing the quality of the grid resolution in large-eddy simulations and related approaches. To examine the usefulness of these techniques, a gas release in a ventilated room is examined using DES and SAS. The grid resolution measures indicate that overall the grids used are relatively coarse. Both DES and SAS model predictions are found to be in poor agreement with experimental data compared to steady and unsteady Reynolds-averaged Navier–Stokes (RANS) results using the SST model. The SAS model also shows the greatest grid sensitivity of the four models tested. The work highlights the need for grid-dependence studies and the potential problems of using coarse grids.

Computational analysis of locally forced flow over a wall-mounted hump at high- Re number

DES method is applied to an axisymmetric base and base-bleed flows at supersonic mainstream. The model is based on the Spalart-Allmaras (S-A) turbulence model in the RANS mode, and is based on the sub-grid scale model in the Large-eddy simulation (LES) mode. Accurate predictions of the base flow field and base pressure are successfully achieved by using the DES methodology which is less expensive than LES. Flow properties at the edge of base, such as boundary layer thickness, momentum thickness and skin fraction are compared with Dutton’s experimental data to proper prediction of base flow filed. From the present results, The DES accurately resolves the physics of unsteady turbulent motions, such as shear layer rollup, large-eddy motions in the downstream region and small eddy motions inside the recirculation region. Moreover, the DES analysis of strong compressibility flow, LES mode is intentionally performed in boundary layer with the conventional empirical constant CDES value of 0.65. In this study, an expression is suggested to determine the CDES value dynamically by using a distribution function of the ratio of turbulence length scale and wall distance which is used in S-A DDES model for RANS mode protection. The application of the dynamic CDES presents better prediction than previous results those used constant but different CDES values.

In the present paper the Discontinuous Galerkin (DG) method is applied for the discretisation of the Reynolds-Averaged Navier-Stokes (RANS) equations. Turbulence modelling is done with the Detached eddy simulation (DES) model based on the Spalart-Allmaras (SA) turbulence equation. In this method a turbulent lengthscale based on the cell width of the mesh is used for the filter width of the large eddy simulation (LES) region. As DG employs higher-order polynomial basis functions inside a cell an adaptation of the “cell width definition” is needed so that the turbulent viscosity is not too high in the LES region of the simulation. Hence, the cell width is adapted with the polynomial degree of the employed basis functions. Simulations of a backward facing step as well as a circular cylinder show that the adaptation is successful.KeywordsDiscontinuous GalerkinDetached eddy simulation

Numerical simulations based on Reynolds-averaged Navier Stokes (RANS) models are still the work-horse tool in engineering design involving turbulent flows. Two decades ago, when LES started gaining popularity thanks to the increasing availability of computational resources, it was widely expected that it would have gradually replaced RANS methods in industrial CFD for decades to come. In the past two decades, however, while LES-based methods gained widespread applications and the earlier hope did not diminish, the predicted time when LES would replace RANS has been significantly delayed. Most industrial users are probably decades away from any routine use of scale resolving simulations, not to mention the cost, time and user skill it take to run these computations. In brief, RANS solvers, particularly those based on standard eddy viscosity models (e.g k-e, k-ω, S-A and k-ω SST) are expected to remain the workhorse in the CFD of high Reynolds number flows for decades. However, predictions from RANS simulations are known to have large discrepancies in many flows of engineering relevance, including those with swirl, pressure gradients, or mean streamline curvature. It is a consensus that the dominant cause for such discrepancies is the RANS-modeled Reynolds stresses. In light of the long stagnation in traditional turbulence modeling, researchers explored machine learning as an alternative to improve RANS modeling by leveraging data from highfidelity simulations. The goal is to make use of vast amounts of turbulent flows data, machine learning techniques and current understanding of turbulence physics to develop models with better predictive capabilities in the context of RANS simulations. Recently, in a seminar work, Ling et al (2016) developed a neural network architecture capable of embedding invariance properties into the Reynolds stress tensor predicted in output. Such a network, named the tensor basis neural network (TBNN), was applied to a variety of flow fields with encouraging results compared to both classical turbulence models and neural networks that do not preserve Galilean invariance. Yet, as in most data driven turbulence modelling approaches, the TBNN was used as a post-processing tool to correct the Reynolds stress tensor field predicted by a RANS simulation run with standard closure models. This means that, theoretically, the network can be applied only to correct the Reynolds stress tensor for the same RANS model on which it has been trained since, in general, different turbulence models yield different results depending on the flow type. Moreover, there is no physisical insight that suggests a relation between the RANS velocity gradients - used as inputs of the machine learning model - and the true Reynolds stress tensor. Differently, in this work a network with a similar architecture to the Ling’s one was trained and tested on a database of high-fidelity data of eight different flows to learn a functional mapping between the inputs of Pope’s General Eddy Viscosity Model and the anisotropic part of the Reynolds stress tensor. Then the network was embedded into a CFD RANS solver as a replacement of the standard closure model - and therefore called at every solver’s iteration. Lasty, the RANS solver with embedded TBNN was be tested on a canonical flow case - turbulent channel flow - to evaluate its performances. As for the organization of this work: in Chapter 1 further details on the data driven turbulence modelling will be given, RANS models and equations will be introduced and also an introduction to Neural Networks will be presented. In Chapter 2, it will be given a detailedexplanation of the RANS CFD solver and the neural network’s implementation. In Chapter 3, the method will be tested on a turbulent channel flow case and the results will be discussed. Lastly, in Chapter 4, some meaningful conclusions will be drawn.