Investigation of flow structures in a turbulent separating flow using hybrid RANS-LES model

Gaurav Kumar; Ashoke De; Harish Gopalan

doi:10.1108/HFF-03-2016-0134

Investigation of flow structures in a turbulent separating flow using hybrid RANS-LES model

Kumar, Gaurav;De, Ashoke;Gopalan, Harish 2021-02-18 00:00:00 Purpose: Hybrid RANS-LES methods have become popular for simulation of massively separated ﬂows at high Reynolds numbers due to their reduced computational cost and good accuracy. In the current study, a comparison has been made to examine the performance of Large Eddy Simulation (LES) and Hybrid RANS-LES model for a given grid resolution. Design/methodology/approach: For better assessment and contrast of model performance, both mean and instantaneous ﬂow ﬁelds have been investigated. For studying instantaneous ﬂow, Proper Orthogonal Decomposition has been used. Findings: Current analysis shows that hybrid RANS-LES is capable of achieving similar accuracy in prediction of both mean and instantaneous ﬂow ﬁelds at a very coarse grid as compared to LES. Originality/value: Focusing mostly on the practical applications of computation, most of the attention has been given to the prediction of one-point ﬂow statistics and little consideration has been put to two-point statistics. Here, two-point statistics has been considered using POD to investigate unsteady turbulent ﬂow. KEY WORDS: Hybrid RANS-LES models; Periodic Hills; RANS; LES; Proper Orthogonal Decomposition. 1. INTRODUCTION Navier-Stokes equation is an excellent mathematical model for prediction of ﬂuid dynamics in continuum limit and a numerically accurate computation using Direct Numerical Simulation (DNS) can mimic experimental results to a high accuracy. Due to non-linearity of Navier-Stokes equation and multiplicity of scales present in the ﬂow, it becomes computationally very expensive or sometimes impossible to compute using currently available technology and resources in many cases. This has led to the development of turbulence models to minimize computational cost while providing reasonably accurate results. For a long time, Large Eddy Simulation (LES) and Reynolds Averaged Navier- Stokes (RANS) have been very popular among ﬂuid dynamics research and engineering community. However, the limitations of LES (high computational cost in near body region) and RANS (poor accuracy for many ﬂow conﬁgurations) have led to the development of hybrid RANS-LES methods. Correspondence to: Department of Aerospace Engineering, Indian Institute of Technology Kanpur, 208016, Kanpur, UP, India, E-mail:[email protected] arXiv:2102.09151v1 [physics.flu-dyn] 18 Feb 2021 2 Hybrid RANS-LES models are designed to take advantage of the best features of RANS and LES. Hybrid RANS-LES models employ RANS modeling in the near-wall region and LES in the oﬀ-body region. As a result, these models combine the advantage of RANS methods in the near-wall region (lower computational cost) and LES in the region away from the wall (improved accuracy). For example, the use of hybrid methods reduces the cost of LES by 100 times at Re = 10 (Gopalan et al., 2013). This reduction in computational cost will be higher as the Reynolds number is increased. In spite of this decrease in the computational cost, the accuracy of hybrid simulations has been found to be quite comparable to LES predictions (Abe, 2014; Breuer et al., 2008; Davidson and Peng, 2003; De Langhe et al., 2005; Fadai-Ghotbi et al., 2010; Fasel et al., 2006; Girimaji et al., 2003; Gopalan et al., 2013; Hamba, 2003; Han and Krajnovi´ c, 2013; Hedges et al., 2002; Menter and Egorov, 2005; Shur et al., 1999; Tucker and Davidson, 2004). This is possible as the size of the RANS region is relatively small compared to LES and it can be expected that it does not signiﬁcantly aﬀect the computational accuracy. This success has given rise to an increased usage of hybrid methods for simulations of ﬂows at Reynolds number which were not possible before. A detailed review of the various types of hybrid methods and their applications can be found in Refs. (Fr¨ ohlich and Terzi, 2008; Spalart, 2009). Focusing mostly on the practical applications of computation, most of the attention has been given to prediction of one-point ﬂow statistics and little consideration has been put to two-point statistics. Two-point velocity statistics contains information about vortical structures present in turbulent ﬂows (Pope, 2000). For applications in aeroacoustics and vortex induced vibration of bluﬀ bodies, dynamics and evolution of the unsteady ﬂows can be understood by considering two-point statistics to devise methods for active ﬂow control, noise modeling and Reduced Order Modelling (ROM). Numerous experimental observations and DNS computation results indicate the important eﬀect of so-called coherent structures in the ﬂow dynamics (Cantwell, 1981). Two-point statistics helps in understanding unsteady ﬂow dynamics which considers greater importance in the prediction of ﬂows which are not statistically stationary. This motivates us to study unsteady structures along with statistics for the development of a more economical and accurate model. In homogeneous ﬁelds, information can be extracted by subjecting two-point correlations to the Fourier analysis (Tennekes and Lumley, 1972). However, most of the practically important ﬂow cases like mixing layers, wakes or wall-bounded ﬂows are strongly inhomogeneous. To investigate an inhomogeneous turbulent ﬁeld, we have to use other approaches. Many turbulent ﬂows are characterized by recurrent structures that are collectively called coherent structures. These are energetically dominant in many ﬂows. Proper Orthogonal Decomposition (POD) is a statistical analysis method, based on the two-point correlation functions and helps in detecting coherent structures. This is a statistical technique that can be applied for the extraction of coherent structures of a turbulent ﬂow ﬁeld in ﬂuid dynamics. The method has been independently suggested by Kosambi (Kosambi, 1943), Loeve (Loeve, 1945), Karhunen (Karhunen, 1946), Pougachev (Pugachev, 1953) and Obukhov (Obukhov, 1954) and has been ﬁrst introduced in turbulent ﬂow analysis by Lumley (Lumley, 1970). The method is extensively presented in Sirovich (Sirovich, 1987) and Berkooz et. al. (Berkooz et al., 1993). This is a subject of great interest, as it also leads, via the Galerkin projection (Aubry, 1991; Panton, 1997), to a low-dimensional set of ordinary diﬀerential equations governing the evolution of vortical structures which is the key idea in understanding the turbulent ﬂows. This can be used to identify the large energy containing structures resolved by hybrid RANS-LES models to assess the model accuracy as compared to experiment or more accurate computational models (DNS, LES). Periodic hills geometry has been investigated both numerically and experimentally over a wide range of Reynolds number by several authors. One of the ﬁrst works was the experimental investigation of (Almeida et al., 1993). The domain size of Almeida’s experiment made it computationally expensive to perform a suitable numerical simulation. 3 There were also concerns if the periodic boundary conditions may be applicable for his experiments. To overcome this issue, (Mellen et al., 2000) performed experiments on the periodic hills to provide validation data for simulations. Parallely, computational work using LES was also performed to provide validation data for RANS and hybrid methods (Breuer et al., 2009; Fr¨ ohlich et al., 2005). (Fr¨ ohlich et al., 2005) performed highly resolved LES at Re = 10595, based on hill height. They presented detailed results of mean ﬂow quantities, Reynolds stresses, and budget for the Reynolds stresses. (Breuer et al., 2009) performed combined numerical and experimental work for a range of Reynolds number. They reported results about the existence of small recirculation zone at hill top and size of the recirculation th zone. Recently, (Diosady and Murman, 2014) performed DNS at Re = 10595 using an 8 th order scheme in space and 4 order scheme in time. The simplicity of the geometry coupled with complex ﬂow physics has made periodic hills simulations a popular choice for testing hybrid RANS-LES models in the literature. Some of the early works to demonstrate the reduction in computational cost using the hybrid models used periodic hills as a test case (Davidson and Dahlstr¨ om, 2005; Davidson and Peng, 2003; Temmerman et al., 2005; Tessicini et al., 2006). The general observation in all these studies was that the use of hybrid RANS-LES models reduces the computational cost compared to LES while providing better accuracy than RANS. To compare the performance of the diﬀerent hybrid models, a joint study was performed (Sari´ c et al., 2007) by various groups to study the ﬂow over periodic hills at Re = 10595 and numbers of diﬀerence in the location of the separation and reattachment points had been observed. In this paper, unsteady ﬂow ﬁeld prediction has been compared for a channel ﬂow with periodic constrictions at a Reynolds number Re=10595 for a set of ﬁne and coarse grids using local dynamic k-equation subgrid-scale LES model (Kim and Menon, 1995). Also, a comparison has been made with simulation for the coarse grid using a hybrid RANS-LES model for statistics and unsteady ﬂow dynamics. POD has been used to verify the cascading of energy into diﬀerent POD Eigen-modes. The rest of the paper is organized as follows. The mathematical models used for the simulations are presented in Sec. 2. Section 3 provides the details of the grid generation, numerical solver and the list of simulations that have been performed. The statistical comparison of results is obtained in Sec. 4.1 and in section 4.2, energy cascading has been compared using POD for LES and Hybrid RANS-LES models. Conclusions are presented in Section 5. 2. MATHEMATICAL MODEL 2.1. Flow Field Equations The governing equations for the conservation of mass and momentum for incompressible ﬂows are given by ∂U = 0 (1) ∂x ¯ ¯ ¯ ¯ ∂U ∂U U ∂p ¯ ∂ U ∂τ i i j i ij + = − + ν + (2) ∂t ∂x ∂x ∂x ∂x ∂x j i j j j Here ¯ is used to represent the hybrid variables. It is assumed that the variables seamlessly switch between the ensembled RANS and ﬁltered LES variables. U is the ﬂuid velocity, p ¯ is the ﬂuid pressure, ν is the kinematic viscosity and τ is the turbulent stress tensor. The ij turbulent stress tensor is given by τ = kδ − 2ν S + N (3) ij ij ij ij Here k is the turbulent kinetic energy (TKE), ν is the hybrid turbulent viscosity, S is the ij symmetric part of the velocity gradient tensor and N is the non-linear part of turbulent ij 4 stress tensor. For linear eddy-viscosity models, N = 0. In this study, the constitutive ij relation proposed by Abe et. al. (Abe, 2005) is used for N in hybrid RANS-LES model. ij The turbulent kinetic energy is calculated using the following equation ¯ ¯ ∂k ∂kU ∂U ∂ ν ∂k j i h t + = −τ − + ν + (4) ij ∂t ∂x ∂x ∂x σ ∂x j j j k j The system of equations are closed once the hybrid turbulent viscosity ν and dissipation rate are provided. This is discussed in section 2.4. 2.2. RANS Model In this section details of the baseline RANS models are provided. K-ω SST model (Menter, 1992) is used as the baseline linear RANS model. The model solves two transport equation for the TKE and speciﬁc turbulent dissipation rate ω (Menter, 1992) ∂k ∂k ∂ ∂k + u ¯ = P − β ωk + (ν + σ ν ) (5) j k k t ∂t ∂x ∂x ∂x j j j ∂ω ∂ω γ ∂ ∂ω + u ¯ = P − βω + (ν + σ ν ) (6) j k w t ∂t ∂x ν ∂x ∂x j t j j σ ∂k ∂ω w2 + 2 (1− F ) ω ∂x ∂x j j ∗ ∗ where P = max(τ ∂u ¯ /∂x , 10β ωk), is the kinetic energy production term, β = 0.09 is k ij i j a model constant and the last term in the ω equation is the cross-diﬀusion term. F is a blending function which has a value of one inside the boundary layer and zero outside. The limiting of the production term is an alternative to the use of damping function in the near-wall region. The turbulent stress tensor and viscosity are computed in this model as follows τ = kδ − 2ν S + N (7) ij ij t ij ij a k ν = (8) t q ¯ ¯ max a ω, 2S S F 1 ij ij 2 Here N is the non-linear part of the turbulent stress tensor. For more details on k-ω SST ij RANS models and non linear part of turbulent stress tensor, see Appendices 6.1. 2.3. LES Model In this study, local dynamic k-Equation Subgrid-Scale (LDKSGS) Model(Kim and Menon, 1995) is used for LES computations. Here, subgrid stresses τ are modelled in terms of the ij SGS eddy viscosity ν as: τ = −2ν S + δ k (9) ij t ij ij sgs where, e represents ﬁltering, ﬁlter width is taken to be the cube-root of the cell volume and ν = c k Δ (10) sgs ν sgs A one-equation model for the subgrid-scale kinetic energy, k is given in the following sgs form: f e ∂k ∂k U ∂U ∂ ∂k sgs sgs j i + = −τ − + ν (11) ij sgs ∂t ∂x ∂x ∂x ∂x j j j j 5 Table I. In the expressions given, β is model constant. F is the blending function. r l r l Model L L L L F F ν ν √ √ k k 1 r l SST-Blended Δ Δ F L + (1− F )l ∗ ∗ r l b b ν ν β ω β ω F L +(1−F )L b b Equation 11 is closed by providing a model for dissipation rate term as: sgs = c (12) Here two new coeﬃcients c and c are calculated dynamically using the method proposed in (Kim and Menon, 1995) along with local averaging of model coeﬃcients for stability. 2.4. Hybrid Model A general formulation for the viscosity and dissipation in hybrid methods can be written as follows h r l 3/2 r l = F ( , ) = k F (L , L ) (13) h r l r l ν = F (ν , ν ) = kF (L , L ) (14) ν ν t t t ν ν In these equations, superscripts h, r, and l denotes hybrid, RANS and LES, respectively. L and L denote the turbulent viscosity and dissipation length scales. Both these scales do not have to be the same (Breuer et al., 2008). F and F denote the hybrid RANS-LES switching function based on the length scales. It is also possible to formulate the models in terms of the turbulent time-scales. In this study, we propose to investigate the performance of SST- blended hybrid RANS-LES model which uses hyperbolic tangent switching function for both dissipation and viscosity. The expressions for the length scales and switching functions for the hybrid model are given in Table I. h h In the blended model, both ν and are modiﬁed from their RANS values. The switching occurs over a number of grid cells in the blended model (also called buﬀer region). Blended model uses complex switching function. l r 1− L /L ν ν F = 1− 0.5 1 + tanh (15) The amount of blending in SST-Blended is controlled by the model constant λ. This parameter is set to 0.25 in the current study (Gopalan et al., 2013). The ﬁlter width is taken to be the square-root of the maximum face area of the cell. r r Characteristic length scale of turbulence is used for L and L in the two-equation model. The shear stress transport (SST) model of Menter (Menter, 1992) is adopted as the baseline two-equation RANS model for hybrid model. Non-linear hybrid model is denoted by adding preﬁx “N” to the linear models (eg. NBSST). The hybrid model includes shielding function to avoid modelled stress depletion and grid induced separation. The model uses the SST blending function F (Eq. 29) as the shielding function (see Appendices). The deﬁnition of a generic framework for hybrid turbulent viscosity and dissipation (Eqs. 13-14) allows the construction of a model free hybrid approach and makes it easier to modify existing RANS codes to create the new hybrid models. 2.5. Proper Orthogonal Decomposition This method is adopted from the formulation presented in (Berkooz et al., 1993). In the current analysis, POD is applied on real valued scalar and vector ﬁelds and a simpliﬁed mathematical formulation is described as follows. 6 Consider a real valued scalar ﬁeld f deﬁned on an interval Ω. The inner product (f,g) and norm ||f|| deﬁned as (f, g) = f (x)g(x)dx (16) =< f ∗ g > where < . > means ensemble ||f|| = (f ∗ f ) (17) The problem of ﬁnding a single deterministic function most similar, on an average, to the set of values observed for the function u(x) mathematically translates to seeking a function φ(x) such that 2 2 < ||(u, ψ)|| > < ||(u, φ)|| > max = (18) (ψ, ψ) (φ, φ) That is, we ﬁnd the member of the ψ(= φ) which maximises the normalised inner product with the ﬁeld u, which is most nearly parallel in function space. A necessary condition for (18) to hold is that φ is an eigen-function of the two-point auto-covariance matrix 0 0 R(x, x ) =< u(x)u(x ) >. 0 0 0 < u(x)u(x ) > φ(x )dx = λφ(x) (19) The maximum in (18) is obtained for largest eigenvalue λ of (19), However, Hilbert Schmidth theory assures that there is denumerable inﬁnity of solutions of (19) as long as Ω is bounded. These are called empirical eigen-functions and we denote these by {φ } and normalise them so that ||φ || = 1. We order the eigenvalues by λ > λ . Observing k k k+1 the non-negative deﬁniteness of R(x,x’) assures that λ > 0. Also, the ensemble may be reproduced by a modal decomposition in the eigen-functions: u(x) = Σ a φ (x) (20) k k k The diagonal decomposition of the two-point auto-covariance matrix R ensures that the modal amplitudes are uncorrelated: 0 0 R(x, x ) = Σ λ φ (x)φ (x ) k k k k (21) 0 0 < a a > = δ λ k k kk k This method can be used to reduce a dynamical system with inﬁnite degrees of freedom represented by inﬁnite number of eigen-modes, possible to be obtained using equation (19), to a system with ﬁnite number of modes (N) which is most similar to the original system on an average. This ﬁnite number of modes N will depend on the smallness of eigen-values of the higher modes and could be seen as a loss of the information about the system. This reduction of the dynamical system to a ﬁnite degrees of freedom also provides a ﬁnite number of empirical eigen-function which represents the dynamics of the POD modes obtained and can be used to understand the evolution of the system considered. 3. NUMERICAL DETAILS Computational setup for the simulations is shown in Fig. 1(a). The size of the domain is deﬁned in terms of the height of the hill H. The domain size is taken to be 9H×3.035H×4.5H in streamwise (x), wall-normal (y) and spanwise (z) directions, respectively. The size has been chosen to match the benchmark LES results from the literature (Breuer et al., 2009). 7 (a) Domain setup (b) Grid used for the computations Figure 1. Computational setup for the simulations Table II. Grids used for the simulations. Grid n n n Δx/H Δy/H Δz/H x y z G1 127 94 40 0.058 - 0.078 0.002 - 0.117 0.113 G2 159 117 50 0.046 - 0.062 0.002 - 0.087 0.09 G3 200 146 63 0.036 - 0.049 0.002 - 0.065 0.071 G4 200 146 186 0.036 - 0.049 0.002 - 0.065 0.024 G5 252 184 235 0.036 0.002 - 0.031 0.019 3.1. Grid Generation Grids used in the current study are generated using commercial meshing software ANSYS- ICEMCFD . Figure 1(b) shows a cross-section (x-y) of the grid. Five diﬀerent grids are generated and the details of the grids are given in Table II. The grids G1, G2 and G3 are used to perform Hybrid RANS-LES simulations and grids G2, G3, G4 and G5 are used to compute LES simulations. The average value of y is found to be in the range 0.2-0.3 for all the grids. 3.2. Numerical Solver All the simulations are performed using the open-source CFD toolbox OpenFOAM . The hybrid models have been implemented and linked as a user deﬁned library at run-time. Pressure-velocity coupling is achieved using the PIMPLE algorithm available in the code, which is a blend of the PISO and SIMPLE algorithm. The convection term in the momentum equation is discretized using second-order central diﬀerence scheme and for the turbulent variables using bounded second-order schemes. Time marching is performed using a second- order backward diﬀerence scheme. All other terms are discretized using central diﬀerence −6 schemes. The tolerance has been set to 10 for all the variables. Once a statistically steady state is achieved, time averaging is performed over 20 ﬂow-through times (L /U ) x b to compute the statistics. Periodic boundary conditions are used in streamwise direction and spanwise directions. No-slip boundary conditions are enforced at the top and bottom wall. As the value of y lies in the viscous region, no wall-function was employed and the turbulence variables are directly integrated up to the wall. 8 Table III. Separation and reattachment points prediction for simulations and computational cost. −3 Grid Model X /H X /H Grid size Computational Time for 10 sec. s r physical time with 1 processor LES (Breuer et al. (2009)) 0.19 4.69 12.4×10 - G1 NBSST 0.23 5.0 0.48×10 1119 seconds G2 NBSST 0.22 4.65 0.93×10 1526 seconds G3 NBSST 0.21 4.7 1.84×10 4450 seconds G2 LES 0.2 4.73 0.93×10 1256 seconds G3 LES 0.2 4.6 1.84×10 1590 seconds G4 LES 0.17 4.7 5.43×10 7488 seconds G5 LES 0.17 4.7 10.9×10 144800 seconds 4. RESULTS 4.1. Grid Comparison Figures 2(a) and 2(b) show the comparison of the skin-friction and wall pressure coeﬃcients along the channel for the NBSST simulations on G1, G2 and G3 grids and Figures 2(c) and 2(d) depict the comparison of diﬀerence in skin-friction and wall pressure coeﬃcients with respect to data provided in (Breuer et al., 2009). From these ﬁgures, it can be clearly seen that G2 and G3 grid have very close prediction of skin-friction and wall pressure coeﬃcients except minor diﬀerence in the middle of the recirculation region near X/H=2.5. Also, G2 and G3 grids show signiﬁcant improvement in predictions as compared to G1 grid. Hence, G3 grid can be considered as grid with optimum resolution for Hybrid RANS-LES simulation and only G3 grid is chosen for the detailed analysis. Similarly, grid comparison has been performed on G2, G3, G4 and G5 grids from Table II for LES simulations. Figures 3(a) and 3(b) show the comparison of the skin-friction and wall pressure coeﬃcients along the hill for the LES on G2 - G5 grids and Figures 3(c) and 3(d) depict the comparison of diﬀerence in skin-friction and wall pressure coeﬃcients with respect to data provided in (Breuer et al., 2009). Comparing G2 and G3 and G4 and G5, it is evident that increasing grid resolution in streamwise and wall-normal directions does not result in much improvement in predictions of skin-friction and wall pressure coeﬃcients as grid is resolved to y <1 in the near wall region. However, a signiﬁcant improvement is observed with spanwise direction grid resolution among G3 and G4. The slight deviation in the skin-friction and wall pressure coeﬃcients prediction in LES simulation on G4 grid and predictions of Breuer et al. may be partly attributed to the use of smagorinsky based LES computation in (Breuer et al., 2009), whereas Dynamic one-equation subgrid scale model is used here and partly to small diﬀerence in grid resolution. Since, G4 grid predictions appear to be very close to the earlier results of Breuer et al. (2009) and G5 grid does not exhibit much improvement in the results, G4 grid is considered to be optimum for LES calculations and used for further comparison and analysis of results. In Table III, the separation and reattachment points for all the simulations along with the computational cost have been tabulated. Location of separation and reattachment point converges to the values reported in (Breuer et al., 2009) using G3 grid for NBSST simulations and G4 and G5 grids for LES simulations. Computational cost for Hybrid RANS-LES simulations on G3 grid is much smaller than LES simulation on G4 grid which provides similar prediction of skin-friction and wall pressure coeﬃcients. 4.2. Mean Statistics Figures 4(a) and 4(b) depict the comparison of streamwise and wall-normal mean velocity components from LES simulation on G4 grid and NBSST simulation on G3 grid with the experimental results. Diﬀerence in streamwise mean velocities among the two simulations 9 (a) Skin-friction coeﬃcient (b) Wall pressure coeﬃcient (c) Diﬀerence in skin-friction coeﬃcient (d) Diﬀerence in wall pressure coeﬃcient Figure 2. Comparison of the Skin-friction and wall pressure coeﬃcient among G1, G2 and G3 grid with NBSST and previous LES data from Breuer et al.(Breuer et al., 2009). Diﬀerence is calculated with respect to the LES data from Breuer et al. is indistinguishable, but LES simulation on G3 grid shows under-prediction of wall-normal velocity at X/H=0.5. Also, NBSST simulation on G3 grid shows slight over-prediction of wall-normal velocity at X/H=2. Figures 4(c), 4(d), 4(e) and 4(f) show comparison of mean streamwise and wall-normal Reynolds normal stress, mean Reynolds shear stress and mean turbulent kinetic energy with experiment. All the predictions by NBSST simulation on G3 grid is in close agreement with experiment and LES simulation on G4 grid, whereas LES simulation on G3 grid shows overprediction of Reynolds normal and shear stresses at X/H=0.5 in the vicinity of the wall. It should be noted here that the correct Reynolds stress components prediction is only possible here due to the use of non-linear corrections in the hybrid RANS-LES model, otherwise a signiﬁcant underprediction will be observed for linear Hybrid RANS- LES models. Figure 5 shows the comparison of Mean and mean square of pressure ﬂuctuations between LES simulation on G4 and NBSST simulation on G3 grid. Mean pressure comparison between the two cases does not show any signiﬁcant diﬀerence, however comparison of mean square of pressure ﬂuctuations shows considerable diﬀerence at the top of the hill near separation point and in the middle of the domain near reattachment point where ﬂuctuating pressure is less than LES computation. 10 (a) Skin-friction coeﬃcient (b) Wall pressure coeﬃcient (c) Diﬀerence in skin-friction coeﬃcient (d) Diﬀerence in wall pressure coeﬃcient Figure 3. Comparison of the Skin-friction and wall pressure coeﬃcient among G2, G3, G4 and G5 grid with Dynamic k-SGS model and previous LES data from Breuer et. al.(Breuer et al., 2009). Diﬀerence is calculated with respect to the LES data from Breuer et al. 4.3. Coherent Structures In Figures 6(a), 6(c) and 6(e), iso-contours of spanwise vorticity component are shown. In the previous studies (Fr¨ ohlich et al., 2005), existence of Kelvin-Helmholtz vortices and Gortler vortices at the top of the hill along with Helical pairing of vortices in the middle of the domain has been suggested at Re = 10595. Due to very small turbulent length and time scales present in these simulations due to high Reynolds number, only K-H vortices can be identiﬁed at the top of the hill at this Reynolds number as marked in Fig 6. However, K-H vortices can be very easily identiﬁed at low Re cases such as Re=600 as shown in Figure 7. In comparison to iso-contours of spanwise vorticity for LES simulation on G4, LES on G3 shows smaller length of the vortex instability region as soon as it breaks into turbulent eddies. However, NBSST simulation on G3 shows comparable length of vortex instability region and then breaks into turbulent eddies. Similar trend is observed in Figures 6(b), 6(d) and 6(f), where pressure iso-contours are plotted for three simulations. The pressure rolls built up from the top of the hill disappears quickly in case of LES on G3 grid as compared to LES on G4 grid. However, it follows up to a similar distance for NBSST simulation on G3 grid. Another point to note is the size of smaller eddies present in LES simulations on G4 and G3 grids are similar whereas NBSST shows paucity of the smaller structures probably due to RANS modeling in the near wall region where length scales used for modelling of small turbulent structures is much larger than length scales used in LES. This is one of the reasons for better performance (poor performance) of hybrid methods for separated ﬂows (attached ﬂows). This suggests that care is required in the use of hybrid methods for 11 (a) Streamwise velocity (b) Wall-normal velocity (c) Mean streamwise Reynolds normal stress (d) Mean wall-normal Reynolds normal stress (e) Mean Reynolds shear stress (f) Mean turbulent kinetic energy Figure 4. Mean ﬂow statistics attached and weakly separated ﬂows because this issue will not be observed for massively separated ﬂows. This also provides an advantage to the users to obtain similar coherent structures using hybrid RANS-LES model on a coarser grid. 12 (a) LES (G4) (b) LES (G4) (c) NBSST (G3) (d) NBSST (G3) Figure 5. Mean pressure and mean square of pressure ﬂuctuation 4.4. Proper Orthogonal Decomposition Analysis Instantaneous ﬂow ﬁeld provides insight into the ﬂow structures, but it is also important for a model to predict the dynamics correctly. Here, POD has been used to asses the two-point statistical diﬀerence between predictions of these three simulations. POD calculations have been performed over 201 snapshots taken at regular intervals in a period of approximately 8 ﬂow through times (= 8*(Length of domain/average bulk velocity over the domain)). Flow ﬁelds have been decomposed into 201 eigen-values and eigen-functions corresponding to instantaneous pressure and instantaneous velocity ﬁeld. Let us consider a scalar ﬁeld φ(x, t ) with ﬁnite number of snapshots (here n = 1, 2, ..., 201) Then, two-point auto-covariance matrix of dimension n× n can be constituted using ensamble (< . >) of the scalar ﬁeld over the spatial co-ordinates as: R(t , t ) =< φ(x, t ).φ(x, t ) > m n m n = Σ λ φ (x)∗ φ (x) k k k k (22) φ(x, t ) = Σ a φ (x) n k k k 0 0 where, < a a > = δ λ k k kk k th Here, φ (x) are the empirical eigen-functions, λ are the eigen-values corresponding to k k k th eigen-function and a are the time coeﬃcients corresponding to k eigen-function. Since, two-point correlation matrix is constituted by the ensemble of square of instantaneous ﬁeld th th over the domain, k eigen-values corresponds to the partition of energy associated with k eigen-function. Therefore, the list of eigen-values has been arranged in decreasing order, so that ﬁrst mode corresponds to the eigen-function containing maximum energy. Since, the correlation matrix is computed using instantaneous ﬁeld, ﬁrst eigen-function obtained is the time averaged mean of the ﬁeld and corresponding eigen-value correspond to the energy associated with the mean ﬂow. So, for easier representation, eigen-values have been normalized with the eigen-value of the ﬁrst mode, so that, any eigen-value in ﬁgure 8 represents the energy associated with that eigen-function with respect to Mean ﬂow energy. 13 (a) LES (G4) (b) LES (G4) (c) LES (G3) (d) LES (G3) (e) NBSST (G3) (f) NBSST (G3) Figure 6. Instantaneous Flow Fields at Re=10595 Figure 7. Instantaneous spanwise vorticity using LES at Re=600 The same analysis can be performed for vector ﬁeld such as velocity ﬁeld, by considering a new scalar ﬁeld formed by concatenating the three components in a sequence. LES simulation on G4 grid, which is very accurate due to high resolution of grid used for computation, has been used here as a reference for the assessment of the hybrid RANS- LES model simulation on G3 grid. In Figures 8(a) and 8(b), eigen-values for instantaneous pressure and velocity ﬁelds have been plotted. Here, energy associated with mean mode of 14 (a) Pressure based POD (p) (b) Velocity based POD (V ) Figure 8. Eigen values corresponding to Proper Orthogonal decomposition modes of pressure and velocity vector. Eigen-values are normalised with the eigen-value of ﬁrst mode of the ﬁeld itself. Here, ﬁrst POD mode corresponds to the mean ﬁeld. (a) Mode1(LES(G4)) (b) Mode1(NBSST(G3)) Figure 9. Two-dimensional vector plot for ﬁrst eigen-mode for velocity based POD. pressure POD is 92% of the total energy whereas the energy associated with the mean mode of velocity ﬁeld is 82%. In Figure 8(a), ﬁrst two modes have much higher energy than the rest of the modes which follow continuous distribution of energy in higher modes; whereas in Figure 8(b), only ﬁrst mode has signiﬁcantly higher energy whereas all the higher modes show gradual decrease in energy. In comparison with the eigen-values plot of LES simulation on G4 grid, the nonlinear BlendedSST on G3 grid has very close predictions of distribution of eigen-values for high energy modes and shows small diﬀerence in low energy modes. In ﬁgures 9(a) and 9(b), two dimensional vector plot for ﬁrst eigen-mode of velocity based POD is shown. Both the plots show same ﬂow structures, indicating NBSST simulation on G3 grid is capable of predicting the coherent structures accurately. Also, Iso-contours of ﬁrst two modes of pressure based POD is shown in 10. The structures seen in both the modes for NBSST is is found to be similar as LES simulation except magnitude diﬀerence. Since, these eigen-modes are vector and considered only to visualize the high energy coherent structures in the ﬂow, the magnitude diﬀerence may not add too much value for the comparison. 5. CONCLUSION In the current study, a comparison has been made to assess the solution inaccuracy due to lower grid resolution in Large Eddy Simulation (LES) and improvement of solution using Hybrid RANS-LES model. LES seems to provide very accurate results, for turbulent separating ﬂows as in case of ﬂow over periodic hills considered here, using highly resolved grid. However, the results deteriorates very rapidly for computations performed using same model on a coarser grid. At the same time, hybrid RANS-LES model (non-linear 15 (a) Mode1(LES(G4)) (b) Mode1(NBSST(G3)) (c) Mode2(LES(G4)) (d) Mode2(NBSST(G3)) Figure 10. Iso-contours for ﬁrst and second eigen-mode for pressure based POD blended SST model considered here), provides signiﬁcantly improved results than pure LES calculations on a coarse grid. The performance of the models have been assessed in terms of 1-point mean statistics, energy distributions and coherent structures in diﬀerent modes of the ﬂow obtained using proper orthogonal decomposition. LES and hybrid RANS-LES has a close agreement for the ﬁrst and second order 1-point statistics while performing hybrid RANS-LES simulation on very coarse grid. Hence, the predictions of skin-friction and wall pressure coeﬃcients and separation and reattachment locations show high dependence on grid in case of LES but the results converse for hybrid RANS-LES on a coarser grid. Since the ﬂow quantities associated with the wall depends on complete ﬂow information in the domain and can only be matched exactly if the ﬂow structures and their dynamics are same in the domain. POD as an analysis tool for two- points statistics shows that hybrid RANS-LES model has close prediction of ﬂow dynamics as compared to LES simulation on a highly reﬁned grids. Also, looking at the computational cost, Hybrid RANS-LES simulation is very cheap as compared to the highly resolved LES simulation for the amount of inaccuracy we have to accept in Hybrid RANS-LES simulations. Hence, in all aspects, Hybrid RANS-LES model provides a good prediction of mean and instantaneous ﬂow ﬁelds on a coarser grid compared to the LES simulation on highly reﬁned grid. ACKNOWLEDGMENTS Simulations are carried out on the computers provided by the Indian Institute of Technology Kanpur (IITK) (www.iitk.ac.in/cc) and the manuscript preparation as well as data analysis has been carried out using the resources available at IITK. This support is gratefully th acknowledged. This paper is an invited contribution for the 10 International Symposium on Numerical Analysis of Fluid Flow and Heat Transfer - Numerical Fluids 2015, Rhodes, Greece. 16 REFERENCES K. Abe. A hybrid LES/RANS approach using an anisotropy-resolving algebraic turbulence model. International Journal of Heat and Fluid Flow, 26(2):204–222, 2005. doi: 10.1016/j.ijheatﬂuidﬂow.2004.08.009. K. Abe, Y.-J. Jang, and M. A. Leschziner. An investigation of wall-anisotropy expressions and length-scale equations for non-linear eddy-viscosity models. International Journal of Heat and Fluid Flow, 24(2):181–198, 2003. doi: 10.1016/S0142-727X(02)00237-0. K.-i. Abe. An advanced switching parameter for a hybrid les/rans model considering the characteristics of near-wall turbulent length scales. Theoretical and Computational Fluid Dynamics, 28(5):499–519, 2014. doi: 10.1007/s00162-014-0328-3. G. Almeida, D. Durao, and M. Heitor. Wake ﬂows behind two-dimensional model hills. Experimental Thermal and Fluid Science, 7(1):87–101, 1993. N. Aubry. On the hidden beauty of the proper orthogonal decomposition. Theoretical and Computational Fluid Dynamics, 2(5-6):339–352, 1991. G. Berkooz, P. Holmes, and J. L. Lumley. The proper orthogonal decomposition in the analysis of turbulent ﬂows. Annual review of ﬂuid mechanics, 25(1):539–575, 1993. M. Breuer, B. Jaﬀr´ ezic, and K. Arora. Hybrid LES-RANS technique based on a one-equation near-wall model. Theoretical and Computational Fluid Dynamics, 22(3):157–187, 2008. doi: 10.1007/s00162-007-0067-9. M. Breuer, N. Peller, C. Rapp, and M. Manhart. Flow over periodic hills–numerical and experimental study in a wide range of reynolds numbers. Computers & Fluids, 38(2): 433–457, 2009. B. J. Cantwell. Organized motion in turbulent ﬂow. Annual review of ﬂuid mechanics, 13 (1):457–515, 1981. L. Davidson and S. Dahlstr¨ om. Hybrid les-rans: computation of the ﬂow around a three- dimensional hill. Engineering Turbulence Modeling and Experiments, 6:319–328, 2005. L. Davidson and S.-H. Peng. Hybrid les-rans modelling: a one-equation sgs model combined with ak–ω model for predicting recirculating ﬂows. International Journal for Numerical Methods in Fluids, 43(9):1003–1018, 2003. C. De Langhe, B. Merci, K. Lodeﬁer, and E. Dick. Hybrid RANS/LES modelling with an approximate renormalization group. II: Applications. Journal of Turbulence, 6(14):1–16, 2005. doi: 10.1080/14685240500149765. L. T. Diosady and S. M. Murman. Dns of ﬂows over periodic hills using a discontinuous- galerkin spectral-element method. AIAA Paper, 2784, 2014. A. Fadai-Ghotbi, C. Friess, R. Manceau, and J. Bor´ ee. A seamless hybrid rans-les model based on transport equations for the subgrid stresses and elliptic blending. Physics of Fluids, 22:055104, 2010. doi: http://dx.doi.org/10.1063/1.3415254. H. F. Fasel, D. A. V. Terzi, and R. D. Sandberg. A methodology for simulating compressible turbulent ﬂows. Journal of Applied Mechanics, 73(3):405–412, 2006. doi: 10.1115/1.2150231. J. Fr¨ ohlich and D. V. Terzi. Hybrid LES/RANS methods for the simulation of turbulent ﬂows. Progress in Aerospace Sciences, 44(5):349–377, 2008. doi: 10.1016/j.paerosci.2008. 05.001. 17 J. Fr¨ ohlich, C. P. Mellen, W. Rodi, L. Temmerman, and M. A. Leschziner. Highly resolved large-eddy simulation of separated ﬂow in a channel with streamwise periodic constrictions. Journal of Fluid Mechanics, 526:19–66, 2005. S. S. Girimaji, R. Sreenivasan, and E. Jeong. PANS turbulence model for seamless transition between RANS, LES: Fixed-point analysis and preliminary results. In FEDSM2003- 45336, Proceedings of ASME FEDSM, volume 3, pages 13–16, 2003. H. Gopalan, S. Heinz, and M. K. St¨ ollinger. A uniﬁed RANS–LES model: Computational development, accuracy and cost. Journal of Computational Physics, 249:249–274, 2013. doi: 10.1016/j.jcp.2013.03.066. F. Hamba. A hybrid RANS/LES simulation of turbulent channel ﬂow. Theoretical and Computational Fluid Dynamics, 16(5):387–403, 2003. X. Han and S. Krajnovi´ c. An eﬃcient very large eddy simulation model for simulation of turbulent ﬂow. International Journal for Numerical Methods in Fluids, 71(11):1341–1360, 2013. doi: 10.1002/ﬂd.3714. L. S. Hedges, A. K. Travin, and P. R. Spalart. Detached-eddy simulations over a simpliﬁed landing gear. Journal of Fluids Engineering, 124(2):413–423, 2002. doi: 10.1115/1.1471532. K. Karhunen. Zur spektraltheorie stochastischer prozesse. 1946. W.-W. Kim and S. Menon. A new dynamic one-equation subgrid-scale model for large eddy simulations. In AIAA, Aerospace Sciences Meeting and Exhibit, 33 rd, Reno, NV, 1995. D. Kosambi. Statistics in function space. J. Indian Math. Soc, 7(1):76–88, 1943. M. Loeve. Functions aleatorie de second ordre. CR Acad. des Sci, page 220, 1945. J. Lumley. L. 1970 stochastic tools in turbulence. Academic Press, New York., 1970. C. P. Mellen, J. Fr¨ olich, and W. Rodi. Large Eddy Simulations of the ﬂow over periodic hills. In M. Deville and R. Owens, editors, IMACS World Congress, 2000. F. Menter and Y. Egorov. A scale adaptive simulation model using two-equation models. In Aerospace Sciences Meetings, pages 2005–1095. American Institute of Aeronautics and Astronautics, Reno, Nevada, 10-13 January 2005. doi: 10.2514/6.2005-1095. F. R. Menter. Improved two-equation k-omega turbulence models for aerodynamic ﬂows. NASA STI/Recon Technical Report N, 93:22809, Oct. 1992. A. Obukhov. Statistical description of continuous ﬁelds. Tr. Geophys. Int. Akad. Nauk. SSSR, 24:3–42, 1954. R. L. Panton. Self-sustaining mechanisms of wal l turbulence, volume 15. Computational Mechanics, 1997. S. Pope. Turbulent ﬂows, 771 pp, 2000. V. S. Pugachev. The general theory of correlation of random functions. Izvestiya Rossiiskoi Akademii Nauk. Seriya Matematicheskaya, 17(5):401–420, 1953. S. Sari´ c, S. Jakirli´ c, M. Breuer, B. Jaﬀr´ ezic, G. Deng, O. Chikhaoui, J. Fr¨ ohlich, D. Von Terzi, M. Manhart, and N. Peller. Evaluation of detached eddy simulations for predicting the ﬂow over periodic hills. In ESAIM: proceedings, volume 16, pages 133–145. EDP Sciences, 2007. 18 M. Shur, P. R. Spalart, M. Strelets, and A. Travin. Detached-eddy simulation of an airfoil at high angle of attack, 4th int. In Symp. Eng. Turb. Model ling and Measurements, pages 669–678, 1999. L. Sirovich. Turbulence and the dynamics of coherent structures. part i: Coherent structures. Quarterly of applied mathematics, 45(3):561–571, 1987. P. R. Spalart. Detached-eddy simulation. Annual Review of Fluid Mechanics, 41:181–202, 2009. doi: 10.1146/annurev.ﬂuid.010908.165130. L. Temmerman, M. Hadˇ ziabdi´ c, M. Leschziner, and K. Hanjali´ c. A hybrid two-layer urans– les approach for large eddy simulation at high reynolds numbers. International journal of heat and ﬂuid ﬂow, 26(2):173–190, 2005. H. Tennekes and J. L. Lumley. A ﬁrst course in turbulence. MIT press, 1972. F. Tessicini, L. Temmerman, and M. Leschziner. Approximate near-wall treatments based on zonal and hybrid rans–les methods for les at high reynolds numbers. International Journal of Heat and Fluid Flow, 27(5):789–799, 2006. P. G. Tucker and L. Davidson. Zonal k–l based large eddy simulations. Computers and Fluids, 33(2):267–287, 2004. doi: 10.1016/S0045-7930(03)00039-2. 6. APPENDICES 6.1. k-ω SST RANS Model In this section details of the baseline RANS models are provided. K-ω SST model (Menter, 1992) is used as the baseline linear RANS model. The model solves two transport equation for the TKE and turbulent dissipation ω (Menter, 1992) ∂k ∂k ∂ ∂k + u ¯ = P − β ωk + (ν + σ ν ) (23) j k k t ∂t ∂x ∂x ∂x j j j ∂ω ∂ω γ ∂ ∂ω + u ¯ = P − βω + (ν + σ ν ) (24) j k w t ∂t ∂x ν ∂x ∂x j t j j σ ∂k ∂ω w2 + 2 (1− F ) ω ∂x ∂x j j where P = max(τ ∂u ¯ /∂x , 10β ∗ ωk), is the kinetic energy production term, β = 0.09 is k ij i j a model constant and the last term in the ω equation is the cross-diﬀusion term. F is a blending function which has a value of one inside the boundary layer and zero outside. The limiting of the production term is an alternative to the use of damping function in the near-wall region. The turbulent stress tensor and viscosity are computed in this model as follows τ = kδ − 2ν S + N (25) ij ij t ij ij a k ν = (26) ¯ ¯ max a ω, 2S S F 1 ij ij 2 Here N is the non-linear part of the turbulent stress tensor. a = 0.31 is a model constant ij 1 and F is a blending function similar to F . The expressions for the blending functions are 2 1 19 given by   ( " # )! k 500ν 4σ k w2   F = tanh min max , , (27) ∗ 2 2 β ωd d ω CD d kw σ ∂k ∂ω w2 −10 CD = max 2 , 10 (28) kw ω ∂x ∂x i i   k 500ν   F = tanh 2 , (29) ∗ 2 β ωd d ω The model constants are calculated by blending K-ω model near the wall and K- away from the wall using the blending function F . The form is given by φ = φ F + (1− F )φ (30) 1 1 1 2 The model constants are given in Table IV. For the non-linear K-ω SST model (NSST), a modiﬁed form of the non-linear constitutive relation proposed by Abe et al. (Abe et al., 2003) is used. The non-linear term is deﬁned as follows 3ν 2 ¯ ¯ ¯ ¯ ¯ ¯ N = f f 2S S − S S δ − S Ω − S Ω ij NL s ik kj nk kn ij ik kj jk ki k 3 + 2kd (31) ij f = C C (1− f (26)) (32) NL D B w C = (33) 2 2 2 2 2 1 + 22/3(C ν /k) Ω + 2/3(C ν /k) (Ω − S ) f D t D t B f (η) = exp − y /η 2 2 2 n o S Ω − S f = 1− 1 + C C (Ω− S) (34) s s2 D 2 2 (Ω + S ) ¯ ¯ where Ω = Ω Ω is the characteristic rotation rate, S = S S is the characteristic ij ij ij ij strain rate, f = 1 + C C ν / (k (Ω− S)), y = u d /ν is a non-dimensional wall distance, B η D t τ y η is a parameter, and C = 0.8, C = 100, and C = 7 are model constants. Finally, the D η s2 expression for the term d can be written as ij 1 δ ij d = −α f (26) (d d − d d ) w w i j k k ij 2 3 ( ) β C w w 2 2 ¯ ¯ + f (26)(1− f )T − √ S Ω − Ω S w ik kj ik kj r1 d 2 2 1 + C T S Ω γ C δ w w ij 2 2 2 ¯ ¯ + f (26)(1− f )T S S − S (35) w ik kj r1 d 1 + C T S 3 where d = ∂N /∂x , N is the unit-normal, and i i j i 2 2 2 2 f = (Ω − S )/(Ω + S ) (36) r1 T = {1− f (15)} k/ + f (15)δ ν/ (37) d w w w ∗ w Here = β ωk is the turbulence dissipation term.The model constants for the term d are ij given by α = 1, β = , C = 0.5, γ = 1.5, and δ = 1.0 (38) w w w w w 4 20 Table IV. Model constants for the SST model. σ = 0.85 γ = 5/9 β = 0.075 σ = 0.5 k w1 σ = 1.0 γ = 0.44 β = 0.0828 σ = 0.856 k w2 The non-linear model NSST is used as the baseline model for the non-linear hybrid model. A damping function was used in Ref. (Abe et al., 2003). However, this term is not used in the current formulation as production limiter is used. In the original formulation of Abe et al. (Abe et al., 2003), the cross-diﬀusion term is not included. It has been included based on the recent improvement in results obtained with the inclusion of this term. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Physics arXiv (Cornell University) http://www.deepdyve.com/lp/arxiv-cornell-university/investigation-of-flow-structures-in-a-turbulent-separating-flow-using-HNP8POwQ9x

Loading next page...

References (53)

Shenghong Huang, Qiusheng Li (2009)
A new dynamic one‐equation subgrid‐scale model for large eddy simulations
International Journal for Numerical Methods in Engineering, 81
Zonal k–l based large eddy simulations
Computers and Fluids, 33
(1945)
Functions aleatorie de second ordre
A. Fadai-Ghotbi, R. Manceau, J. Borée (2010)
A SEAMLESS HYBRID RANS-LES MODEL BASED ON TRANSPORT EQUATIONS FOR THE SUBGRID STRESSES AND ELLIPTIC BLENDING
Proceeding of Fifth International Symposium on Turbulence and Shear Flow Phenomena
(2005)
Hybrid les-rans: computation of the flow around a threedimensional hill. Engineering Turbulence Modeling and Experiments
Lars Davidson, S. Peng (2003)
Hybrid LES‐RANS modelling: a one‐equation SGS model combined with a k–ω model for predicting recirculating flows
International Journal for Numerical Methods in Fluids, 43
M. Shur, P. Spalart, M. Strelets, A. Travin (1999)
Detached-eddy simulation of an airfoil at high angle of attack
(1993)
The proper orthogonal decomposition in the analysis of turbulent flows. Annual review of fluid mechanics
H. Gopalan, S. Heinz, M. Stöllinger (2013)
A unified RANS-LES model: Computational development, accuracy and cost
J. Comput. Phys., 249
S. Girimaji, R. Srinivasan, Euhwan Jeong (2003)
PANS Turbulence Model for Seamless Transition Between RANS and LES: Fixed-Point Analysis and Preliminary Results
D. Kosambi (2016)
Statistics in Function Space
S. Šarić, S. Jakirlic, M. Breuer, B. Jaffrézic, G. Deng, Oussama Chikhaoui, J. Fröhlich, D. Terzi, M. Manhart, Nikolaus Peller (2007)
Evaluation of Detached Eddy Simulations for predicting the flow over periodic hills
Esaim: Proceedings, 16
L. Sirovich (1987)
Turbulence and the dynamics of coherent structures. I. Coherent structures
Quarterly of Applied Mathematics, 45
Xingsi Han, S. Krajnović (2013)
An efficient very large eddy simulation model for simulation of turbulent flow
International Journal for Numerical Methods in Fluids, 71
(1999)
Detached-eddy simulation of an airfoil at high angle of attack, 4th int
(2009)
Spalart. Detached-eddy simulation
M. Breuer, Nikolaus Peller, C. Rapp, M. Manhart (2009)
Flow over periodic hills – Numerical and experimental study in a wide range of Reynolds numbers
Computers & Fluids, 38
L. Diosady, S. Murman (2014)
DNS of Flows over Periodic Hills using a Discontinuous-Galerkin Spectral-Element Method
L. Sirovich (2016)
TURBULENCE AND THE DYNAMICS OF COHERENT STRUCTURES PART
J. Lumley (1970)
Stochastic tools in turbulence
K. Abe (2014)
An advanced switching parameter for a hybrid LES/RANS model considering the characteristics of near-wall turbulent length scales
Theoretical and Computational Fluid Dynamics, 28
J. Fröhlich, C. Mellen, W. Rodi, L. Temmerman, M. Leschziner (2005)
Highly resolved large-eddy simulation of separated flow in a channel with streamwise periodic constrictions
Journal of Fluid Mechanics, 526
L. Hedges, A. Travin, P. Spalart (2002)
Detached-Eddy Simulations Over a Simplified Landing Gear
Journal of Fluids Engineering-transactions of The Asme, 124
S. Pope (2000)
Turbulent Flows
Measurement Science and Technology, 12
G. Almeida, D. Durão, M. Heitor (1993)
Wake flows behind two-dimensional model hills
Experimental Thermal and Fluid Science, 7
C. Langhe, B. Merci, K. Lodefier, E. Dick (2005)
Hybrid RANS/LES modelling with an approximate renormalization group. II: Applications
Journal of Turbulence
N. Aubry (1991)
On the hidden beauty of the proper orthogonal decomposition
Theoretical and Computational Fluid Dynamics, 2
Michael Breuer, B. Jaffrézic, K. Arora (2008)
Hybrid LES–RANS technique based on a one-equation near-wall model
Theoretical and Computational Fluid Dynamics, 22
(1953)
The general theory of correlation of random functions
K. Abe (2005)
A hybrid LES/RANS approach using an anisotropy-resolving algebraic turbulence model
International Journal of Heat and Fluid Flow, 26
J. Fröhlich, D. Terzi (2008)
Hybrid LES/RANS methods for the simulation of turbulent flows
Progress in Aerospace Sciences, 44
李栋, 焦予秦, Igor, Men’shov, 中村佳朗 (2005)
Detached—Eddy Simulation方法模拟不同类型翼型的失速特性
P. Burattini, R. Antonia, L. Danaila (2005)
Scale-by-scale energy budget on the axis of a turbulent round jet
Journal of Turbulence
(2005)
Hanjalić. A hybrid two-layer urans– les approach for large eddy simulation at high reynolds numbers
L. Temmerman, M. Hadžiabdić, M. Leschziner, K. Hanjalic (2005)
A hybrid two-layer URANS–LES approach for large eddy simulation at high Reynolds numbers
International Journal of Heat and Fluid Flow, 26
F. Menter (1992)
Improved two-equation k-omega turbulence models for aerodynamic flows
, 93
R. Panton (1997)
Self-Sustaining Mechanisms of Wall Turbulence
H. Tennekes, J. Lumley (1972)
A First Course in Turbulence
F. Hamba (2003)
A Hybrid RANS/LES Simulation of Turbulent Channel Flow
Theoretical and Computational Fluid Dynamics, 16
C. Mellen, Jochen Froehlich, W. Rodi (2000)
Large eddy simulation of the flow over periodic hills
K. Karhunen (1946)
Zur Spektraltheorie stochastischer prozesse
B. Cantwell (1981)
Organized Motion in Turbulent Flow
Annual Review of Fluid Mechanics, 13
(1954)
Statistical description of continuous fields
(2005)
Hybrid RANS/LES modelling with an approximate renormalization group
H. Fasel, D. Terzi, R. Sandberg (2006)
A Methodology for Simulating Compressible Turbulent Flows
Journal of Applied Mechanics, 73
(1970)
1970 stochastic tools in turbulence
L. Davidson, S. Dahlström (2005)
Hybrid Les-Rans: Computation of the Flow Around a Three-Dimensional Hill
F. Tessicini, L. Temmerman, M. Leschziner (2006)
Approximate near-wall treatments based on zonal and hybrid RANS–LES methods for LES at high Reynolds numbers
International Journal of Heat and Fluid Flow, 27
K. Abe, Y. Jang, M. Leschziner (2003)
An investigation of wall-anisotropy expressions and length-scale equations for non-linear eddy-viscosity models
International Journal of Heat and Fluid Flow, 24
F. Menter, Y. Egorov (2005)
A Scale-Adaptive Simulation Model using Two-Equation Models
(2003)
A hybrid RANS/LES simulation of turbulent channel flow. Theoretical and Computational Fluid Dynamics
G. Berkooz, P. Holmes, J. Lumley (1993)
The Proper Orthogonal Decomposition in the Analysis of Turbulent Flows
Annual Review of Fluid Mechanics, 25
P. Tucker, L. Davidson (2003)
Zonal k-l based Large Eddy Simulations

eISSN: ARCH-3341
DOI: 10.1108/HFF-03-2016-0134
Publisher site: See Article on Publisher Site

Abstract

Purpose: Hybrid RANS-LES methods have become popular for simulation of massively separated ﬂows at high Reynolds numbers due to their reduced computational cost and good accuracy. In the current study, a comparison has been made to examine the performance of Large Eddy Simulation (LES) and Hybrid RANS-LES model for a given grid resolution. Design/methodology/approach: For better assessment and contrast of model performance, both mean and instantaneous ﬂow ﬁelds have been investigated. For studying instantaneous ﬂow, Proper Orthogonal Decomposition has been used. Findings: Current analysis shows that hybrid RANS-LES is capable of achieving similar accuracy in prediction of both mean and instantaneous ﬂow ﬁelds at a very coarse grid as compared to LES. Originality/value: Focusing mostly on the practical applications of computation, most of the attention has been given to the prediction of one-point ﬂow statistics and little consideration has been put to two-point statistics. Here, two-point statistics has been considered using POD to investigate unsteady turbulent ﬂow. KEY WORDS: Hybrid RANS-LES models; Periodic Hills; RANS; LES; Proper Orthogonal Decomposition. 1. INTRODUCTION Navier-Stokes equation is an excellent mathematical model for prediction of ﬂuid dynamics in continuum limit and a numerically accurate computation using Direct Numerical Simulation (DNS) can mimic experimental results to a high accuracy. Due to non-linearity of Navier-Stokes equation and multiplicity of scales present in the ﬂow, it becomes computationally very expensive or sometimes impossible to compute using currently available technology and resources in many cases. This has led to the development of turbulence models to minimize computational cost while providing reasonably accurate results. For a long time, Large Eddy Simulation (LES) and Reynolds Averaged Navier- Stokes (RANS) have been very popular among ﬂuid dynamics research and engineering community. However, the limitations of LES (high computational cost in near body region) and RANS (poor accuracy for many ﬂow conﬁgurations) have led to the development of hybrid RANS-LES methods. Correspondence to: Department of Aerospace Engineering, Indian Institute of Technology Kanpur, 208016, Kanpur, UP, India, E-mail:[email protected] arXiv:2102.09151v1 [physics.flu-dyn] 18 Feb 2021 2 Hybrid RANS-LES models are designed to take advantage of the best features of RANS and LES. Hybrid RANS-LES models employ RANS modeling in the near-wall region and LES in the oﬀ-body region. As a result, these models combine the advantage of RANS methods in the near-wall region (lower computational cost) and LES in the region away from the wall (improved accuracy). For example, the use of hybrid methods reduces the cost of LES by 100 times at Re = 10 (Gopalan et al., 2013). This reduction in computational cost will be higher as the Reynolds number is increased. In spite of this decrease in the computational cost, the accuracy of hybrid simulations has been found to be quite comparable to LES predictions (Abe, 2014; Breuer et al., 2008; Davidson and Peng, 2003; De Langhe et al., 2005; Fadai-Ghotbi et al., 2010; Fasel et al., 2006; Girimaji et al., 2003; Gopalan et al., 2013; Hamba, 2003; Han and Krajnovi´ c, 2013; Hedges et al., 2002; Menter and Egorov, 2005; Shur et al., 1999; Tucker and Davidson, 2004). This is possible as the size of the RANS region is relatively small compared to LES and it can be expected that it does not signiﬁcantly aﬀect the computational accuracy. This success has given rise to an increased usage of hybrid methods for simulations of ﬂows at Reynolds number which were not possible before. A detailed review of the various types of hybrid methods and their applications can be found in Refs. (Fr¨ ohlich and Terzi, 2008; Spalart, 2009). Focusing mostly on the practical applications of computation, most of the attention has been given to prediction of one-point ﬂow statistics and little consideration has been put to two-point statistics. Two-point velocity statistics contains information about vortical structures present in turbulent ﬂows (Pope, 2000). For applications in aeroacoustics and vortex induced vibration of bluﬀ bodies, dynamics and evolution of the unsteady ﬂows can be understood by considering two-point statistics to devise methods for active ﬂow control, noise modeling and Reduced Order Modelling (ROM). Numerous experimental observations and DNS computation results indicate the important eﬀect of so-called coherent structures in the ﬂow dynamics (Cantwell, 1981). Two-point statistics helps in understanding unsteady ﬂow dynamics which considers greater importance in the prediction of ﬂows which are not statistically stationary. This motivates us to study unsteady structures along with statistics for the development of a more economical and accurate model. In homogeneous ﬁelds, information can be extracted by subjecting two-point correlations to the Fourier analysis (Tennekes and Lumley, 1972). However, most of the practically important ﬂow cases like mixing layers, wakes or wall-bounded ﬂows are strongly inhomogeneous. To investigate an inhomogeneous turbulent ﬁeld, we have to use other approaches. Many turbulent ﬂows are characterized by recurrent structures that are collectively called coherent structures. These are energetically dominant in many ﬂows. Proper Orthogonal Decomposition (POD) is a statistical analysis method, based on the two-point correlation functions and helps in detecting coherent structures. This is a statistical technique that can be applied for the extraction of coherent structures of a turbulent ﬂow ﬁeld in ﬂuid dynamics. The method has been independently suggested by Kosambi (Kosambi, 1943), Loeve (Loeve, 1945), Karhunen (Karhunen, 1946), Pougachev (Pugachev, 1953) and Obukhov (Obukhov, 1954) and has been ﬁrst introduced in turbulent ﬂow analysis by Lumley (Lumley, 1970). The method is extensively presented in Sirovich (Sirovich, 1987) and Berkooz et. al. (Berkooz et al., 1993). This is a subject of great interest, as it also leads, via the Galerkin projection (Aubry, 1991; Panton, 1997), to a low-dimensional set of ordinary diﬀerential equations governing the evolution of vortical structures which is the key idea in understanding the turbulent ﬂows. This can be used to identify the large energy containing structures resolved by hybrid RANS-LES models to assess the model accuracy as compared to experiment or more accurate computational models (DNS, LES). Periodic hills geometry has been investigated both numerically and experimentally over a wide range of Reynolds number by several authors. One of the ﬁrst works was the experimental investigation of (Almeida et al., 1993). The domain size of Almeida’s experiment made it computationally expensive to perform a suitable numerical simulation. 3 There were also concerns if the periodic boundary conditions may be applicable for his experiments. To overcome this issue, (Mellen et al., 2000) performed experiments on the periodic hills to provide validation data for simulations. Parallely, computational work using LES was also performed to provide validation data for RANS and hybrid methods (Breuer et al., 2009; Fr¨ ohlich et al., 2005). (Fr¨ ohlich et al., 2005) performed highly resolved LES at Re = 10595, based on hill height. They presented detailed results of mean ﬂow quantities, Reynolds stresses, and budget for the Reynolds stresses. (Breuer et al., 2009) performed combined numerical and experimental work for a range of Reynolds number. They reported results about the existence of small recirculation zone at hill top and size of the recirculation th zone. Recently, (Diosady and Murman, 2014) performed DNS at Re = 10595 using an 8 th order scheme in space and 4 order scheme in time. The simplicity of the geometry coupled with complex ﬂow physics has made periodic hills simulations a popular choice for testing hybrid RANS-LES models in the literature. Some of the early works to demonstrate the reduction in computational cost using the hybrid models used periodic hills as a test case (Davidson and Dahlstr¨ om, 2005; Davidson and Peng, 2003; Temmerman et al., 2005; Tessicini et al., 2006). The general observation in all these studies was that the use of hybrid RANS-LES models reduces the computational cost compared to LES while providing better accuracy than RANS. To compare the performance of the diﬀerent hybrid models, a joint study was performed (Sari´ c et al., 2007) by various groups to study the ﬂow over periodic hills at Re = 10595 and numbers of diﬀerence in the location of the separation and reattachment points had been observed. In this paper, unsteady ﬂow ﬁeld prediction has been compared for a channel ﬂow with periodic constrictions at a Reynolds number Re=10595 for a set of ﬁne and coarse grids using local dynamic k-equation subgrid-scale LES model (Kim and Menon, 1995). Also, a comparison has been made with simulation for the coarse grid using a hybrid RANS-LES model for statistics and unsteady ﬂow dynamics. POD has been used to verify the cascading of energy into diﬀerent POD Eigen-modes. The rest of the paper is organized as follows. The mathematical models used for the simulations are presented in Sec. 2. Section 3 provides the details of the grid generation, numerical solver and the list of simulations that have been performed. The statistical comparison of results is obtained in Sec. 4.1 and in section 4.2, energy cascading has been compared using POD for LES and Hybrid RANS-LES models. Conclusions are presented in Section 5. 2. MATHEMATICAL MODEL 2.1. Flow Field Equations The governing equations for the conservation of mass and momentum for incompressible ﬂows are given by ∂U = 0 (1) ∂x ¯ ¯ ¯ ¯ ∂U ∂U U ∂p ¯ ∂ U ∂τ i i j i ij + = − + ν + (2) ∂t ∂x ∂x ∂x ∂x ∂x j i j j j Here ¯ is used to represent the hybrid variables. It is assumed that the variables seamlessly switch between the ensembled RANS and ﬁltered LES variables. U is the ﬂuid velocity, p ¯ is the ﬂuid pressure, ν is the kinematic viscosity and τ is the turbulent stress tensor. The ij turbulent stress tensor is given by τ = kδ − 2ν S + N (3) ij ij ij ij Here k is the turbulent kinetic energy (TKE), ν is the hybrid turbulent viscosity, S is the ij symmetric part of the velocity gradient tensor and N is the non-linear part of turbulent ij 4 stress tensor. For linear eddy-viscosity models, N = 0. In this study, the constitutive ij relation proposed by Abe et. al. (Abe, 2005) is used for N in hybrid RANS-LES model. ij The turbulent kinetic energy is calculated using the following equation ¯ ¯ ∂k ∂kU ∂U ∂ ν ∂k j i h t + = −τ − + ν + (4) ij ∂t ∂x ∂x ∂x σ ∂x j j j k j The system of equations are closed once the hybrid turbulent viscosity ν and dissipation rate are provided. This is discussed in section 2.4. 2.2. RANS Model In this section details of the baseline RANS models are provided. K-ω SST model (Menter, 1992) is used as the baseline linear RANS model. The model solves two transport equation for the TKE and speciﬁc turbulent dissipation rate ω (Menter, 1992) ∂k ∂k ∂ ∂k + u ¯ = P − β ωk + (ν + σ ν ) (5) j k k t ∂t ∂x ∂x ∂x j j j ∂ω ∂ω γ ∂ ∂ω + u ¯ = P − βω + (ν + σ ν ) (6) j k w t ∂t ∂x ν ∂x ∂x j t j j σ ∂k ∂ω w2 + 2 (1− F ) ω ∂x ∂x j j ∗ ∗ where P = max(τ ∂u ¯ /∂x , 10β ωk), is the kinetic energy production term, β = 0.09 is k ij i j a model constant and the last term in the ω equation is the cross-diﬀusion term. F is a blending function which has a value of one inside the boundary layer and zero outside. The limiting of the production term is an alternative to the use of damping function in the near-wall region. The turbulent stress tensor and viscosity are computed in this model as follows τ = kδ − 2ν S + N (7) ij ij t ij ij a k ν = (8) t q ¯ ¯ max a ω, 2S S F 1 ij ij 2 Here N is the non-linear part of the turbulent stress tensor. For more details on k-ω SST ij RANS models and non linear part of turbulent stress tensor, see Appendices 6.1. 2.3. LES Model In this study, local dynamic k-Equation Subgrid-Scale (LDKSGS) Model(Kim and Menon, 1995) is used for LES computations. Here, subgrid stresses τ are modelled in terms of the ij SGS eddy viscosity ν as: τ = −2ν S + δ k (9) ij t ij ij sgs where, e represents ﬁltering, ﬁlter width is taken to be the cube-root of the cell volume and ν = c k Δ (10) sgs ν sgs A one-equation model for the subgrid-scale kinetic energy, k is given in the following sgs form: f e ∂k ∂k U ∂U ∂ ∂k sgs sgs j i + = −τ − + ν (11) ij sgs ∂t ∂x ∂x ∂x ∂x j j j j 5 Table I. In the expressions given, β is model constant. F is the blending function. r l r l Model L L L L F F ν ν √ √ k k 1 r l SST-Blended Δ Δ F L + (1− F )l ∗ ∗ r l b b ν ν β ω β ω F L +(1−F )L b b Equation 11 is closed by providing a model for dissipation rate term as: sgs = c (12) Here two new coeﬃcients c and c are calculated dynamically using the method proposed in (Kim and Menon, 1995) along with local averaging of model coeﬃcients for stability. 2.4. Hybrid Model A general formulation for the viscosity and dissipation in hybrid methods can be written as follows h r l 3/2 r l = F ( , ) = k F (L , L ) (13) h r l r l ν = F (ν , ν ) = kF (L , L ) (14) ν ν t t t ν ν In these equations, superscripts h, r, and l denotes hybrid, RANS and LES, respectively. L and L denote the turbulent viscosity and dissipation length scales. Both these scales do not have to be the same (Breuer et al., 2008). F and F denote the hybrid RANS-LES switching function based on the length scales. It is also possible to formulate the models in terms of the turbulent time-scales. In this study, we propose to investigate the performance of SST- blended hybrid RANS-LES model which uses hyperbolic tangent switching function for both dissipation and viscosity. The expressions for the length scales and switching functions for the hybrid model are given in Table I. h h In the blended model, both ν and are modiﬁed from their RANS values. The switching occurs over a number of grid cells in the blended model (also called buﬀer region). Blended model uses complex switching function. l r 1− L /L ν ν F = 1− 0.5 1 + tanh (15) The amount of blending in SST-Blended is controlled by the model constant λ. This parameter is set to 0.25 in the current study (Gopalan et al., 2013). The ﬁlter width is taken to be the square-root of the maximum face area of the cell. r r Characteristic length scale of turbulence is used for L and L in the two-equation model. The shear stress transport (SST) model of Menter (Menter, 1992) is adopted as the baseline two-equation RANS model for hybrid model. Non-linear hybrid model is denoted by adding preﬁx “N” to the linear models (eg. NBSST). The hybrid model includes shielding function to avoid modelled stress depletion and grid induced separation. The model uses the SST blending function F (Eq. 29) as the shielding function (see Appendices). The deﬁnition of a generic framework for hybrid turbulent viscosity and dissipation (Eqs. 13-14) allows the construction of a model free hybrid approach and makes it easier to modify existing RANS codes to create the new hybrid models. 2.5. Proper Orthogonal Decomposition This method is adopted from the formulation presented in (Berkooz et al., 1993). In the current analysis, POD is applied on real valued scalar and vector ﬁelds and a simpliﬁed mathematical formulation is described as follows. 6 Consider a real valued scalar ﬁeld f deﬁned on an interval Ω. The inner product (f,g) and norm ||f|| deﬁned as (f, g) = f (x)g(x)dx (16) =< f ∗ g > where < . > means ensemble ||f|| = (f ∗ f ) (17) The problem of ﬁnding a single deterministic function most similar, on an average, to the set of values observed for the function u(x) mathematically translates to seeking a function φ(x) such that 2 2 < ||(u, ψ)|| > < ||(u, φ)|| > max = (18) (ψ, ψ) (φ, φ) That is, we ﬁnd the member of the ψ(= φ) which maximises the normalised inner product with the ﬁeld u, which is most nearly parallel in function space. A necessary condition for (18) to hold is that φ is an eigen-function of the two-point auto-covariance matrix 0 0 R(x, x ) =< u(x)u(x ) >. 0 0 0 < u(x)u(x ) > φ(x )dx = λφ(x) (19) The maximum in (18) is obtained for largest eigenvalue λ of (19), However, Hilbert Schmidth theory assures that there is denumerable inﬁnity of solutions of (19) as long as Ω is bounded. These are called empirical eigen-functions and we denote these by {φ } and normalise them so that ||φ || = 1. We order the eigenvalues by λ > λ . Observing k k k+1 the non-negative deﬁniteness of R(x,x’) assures that λ > 0. Also, the ensemble may be reproduced by a modal decomposition in the eigen-functions: u(x) = Σ a φ (x) (20) k k k The diagonal decomposition of the two-point auto-covariance matrix R ensures that the modal amplitudes are uncorrelated: 0 0 R(x, x ) = Σ λ φ (x)φ (x ) k k k k (21) 0 0 < a a > = δ λ k k kk k This method can be used to reduce a dynamical system with inﬁnite degrees of freedom represented by inﬁnite number of eigen-modes, possible to be obtained using equation (19), to a system with ﬁnite number of modes (N) which is most similar to the original system on an average. This ﬁnite number of modes N will depend on the smallness of eigen-values of the higher modes and could be seen as a loss of the information about the system. This reduction of the dynamical system to a ﬁnite degrees of freedom also provides a ﬁnite number of empirical eigen-function which represents the dynamics of the POD modes obtained and can be used to understand the evolution of the system considered. 3. NUMERICAL DETAILS Computational setup for the simulations is shown in Fig. 1(a). The size of the domain is deﬁned in terms of the height of the hill H. The domain size is taken to be 9H×3.035H×4.5H in streamwise (x), wall-normal (y) and spanwise (z) directions, respectively. The size has been chosen to match the benchmark LES results from the literature (Breuer et al., 2009). 7 (a) Domain setup (b) Grid used for the computations Figure 1. Computational setup for the simulations Table II. Grids used for the simulations. Grid n n n Δx/H Δy/H Δz/H x y z G1 127 94 40 0.058 - 0.078 0.002 - 0.117 0.113 G2 159 117 50 0.046 - 0.062 0.002 - 0.087 0.09 G3 200 146 63 0.036 - 0.049 0.002 - 0.065 0.071 G4 200 146 186 0.036 - 0.049 0.002 - 0.065 0.024 G5 252 184 235 0.036 0.002 - 0.031 0.019 3.1. Grid Generation Grids used in the current study are generated using commercial meshing software ANSYS- ICEMCFD . Figure 1(b) shows a cross-section (x-y) of the grid. Five diﬀerent grids are generated and the details of the grids are given in Table II. The grids G1, G2 and G3 are used to perform Hybrid RANS-LES simulations and grids G2, G3, G4 and G5 are used to compute LES simulations. The average value of y is found to be in the range 0.2-0.3 for all the grids. 3.2. Numerical Solver All the simulations are performed using the open-source CFD toolbox OpenFOAM . The hybrid models have been implemented and linked as a user deﬁned library at run-time. Pressure-velocity coupling is achieved using the PIMPLE algorithm available in the code, which is a blend of the PISO and SIMPLE algorithm. The convection term in the momentum equation is discretized using second-order central diﬀerence scheme and for the turbulent variables using bounded second-order schemes. Time marching is performed using a second- order backward diﬀerence scheme. All other terms are discretized using central diﬀerence −6 schemes. The tolerance has been set to 10 for all the variables. Once a statistically steady state is achieved, time averaging is performed over 20 ﬂow-through times (L /U ) x b to compute the statistics. Periodic boundary conditions are used in streamwise direction and spanwise directions. No-slip boundary conditions are enforced at the top and bottom wall. As the value of y lies in the viscous region, no wall-function was employed and the turbulence variables are directly integrated up to the wall. 8 Table III. Separation and reattachment points prediction for simulations and computational cost. −3 Grid Model X /H X /H Grid size Computational Time for 10 sec. s r physical time with 1 processor LES (Breuer et al. (2009)) 0.19 4.69 12.4×10 - G1 NBSST 0.23 5.0 0.48×10 1119 seconds G2 NBSST 0.22 4.65 0.93×10 1526 seconds G3 NBSST 0.21 4.7 1.84×10 4450 seconds G2 LES 0.2 4.73 0.93×10 1256 seconds G3 LES 0.2 4.6 1.84×10 1590 seconds G4 LES 0.17 4.7 5.43×10 7488 seconds G5 LES 0.17 4.7 10.9×10 144800 seconds 4. RESULTS 4.1. Grid Comparison Figures 2(a) and 2(b) show the comparison of the skin-friction and wall pressure coeﬃcients along the channel for the NBSST simulations on G1, G2 and G3 grids and Figures 2(c) and 2(d) depict the comparison of diﬀerence in skin-friction and wall pressure coeﬃcients with respect to data provided in (Breuer et al., 2009). From these ﬁgures, it can be clearly seen that G2 and G3 grid have very close prediction of skin-friction and wall pressure coeﬃcients except minor diﬀerence in the middle of the recirculation region near X/H=2.5. Also, G2 and G3 grids show signiﬁcant improvement in predictions as compared to G1 grid. Hence, G3 grid can be considered as grid with optimum resolution for Hybrid RANS-LES simulation and only G3 grid is chosen for the detailed analysis. Similarly, grid comparison has been performed on G2, G3, G4 and G5 grids from Table II for LES simulations. Figures 3(a) and 3(b) show the comparison of the skin-friction and wall pressure coeﬃcients along the hill for the LES on G2 - G5 grids and Figures 3(c) and 3(d) depict the comparison of diﬀerence in skin-friction and wall pressure coeﬃcients with respect to data provided in (Breuer et al., 2009). Comparing G2 and G3 and G4 and G5, it is evident that increasing grid resolution in streamwise and wall-normal directions does not result in much improvement in predictions of skin-friction and wall pressure coeﬃcients as grid is resolved to y <1 in the near wall region. However, a signiﬁcant improvement is observed with spanwise direction grid resolution among G3 and G4. The slight deviation in the skin-friction and wall pressure coeﬃcients prediction in LES simulation on G4 grid and predictions of Breuer et al. may be partly attributed to the use of smagorinsky based LES computation in (Breuer et al., 2009), whereas Dynamic one-equation subgrid scale model is used here and partly to small diﬀerence in grid resolution. Since, G4 grid predictions appear to be very close to the earlier results of Breuer et al. (2009) and G5 grid does not exhibit much improvement in the results, G4 grid is considered to be optimum for LES calculations and used for further comparison and analysis of results. In Table III, the separation and reattachment points for all the simulations along with the computational cost have been tabulated. Location of separation and reattachment point converges to the values reported in (Breuer et al., 2009) using G3 grid for NBSST simulations and G4 and G5 grids for LES simulations. Computational cost for Hybrid RANS-LES simulations on G3 grid is much smaller than LES simulation on G4 grid which provides similar prediction of skin-friction and wall pressure coeﬃcients. 4.2. Mean Statistics Figures 4(a) and 4(b) depict the comparison of streamwise and wall-normal mean velocity components from LES simulation on G4 grid and NBSST simulation on G3 grid with the experimental results. Diﬀerence in streamwise mean velocities among the two simulations 9 (a) Skin-friction coeﬃcient (b) Wall pressure coeﬃcient (c) Diﬀerence in skin-friction coeﬃcient (d) Diﬀerence in wall pressure coeﬃcient Figure 2. Comparison of the Skin-friction and wall pressure coeﬃcient among G1, G2 and G3 grid with NBSST and previous LES data from Breuer et al.(Breuer et al., 2009). Diﬀerence is calculated with respect to the LES data from Breuer et al. is indistinguishable, but LES simulation on G3 grid shows under-prediction of wall-normal velocity at X/H=0.5. Also, NBSST simulation on G3 grid shows slight over-prediction of wall-normal velocity at X/H=2. Figures 4(c), 4(d), 4(e) and 4(f) show comparison of mean streamwise and wall-normal Reynolds normal stress, mean Reynolds shear stress and mean turbulent kinetic energy with experiment. All the predictions by NBSST simulation on G3 grid is in close agreement with experiment and LES simulation on G4 grid, whereas LES simulation on G3 grid shows overprediction of Reynolds normal and shear stresses at X/H=0.5 in the vicinity of the wall. It should be noted here that the correct Reynolds stress components prediction is only possible here due to the use of non-linear corrections in the hybrid RANS-LES model, otherwise a signiﬁcant underprediction will be observed for linear Hybrid RANS- LES models. Figure 5 shows the comparison of Mean and mean square of pressure ﬂuctuations between LES simulation on G4 and NBSST simulation on G3 grid. Mean pressure comparison between the two cases does not show any signiﬁcant diﬀerence, however comparison of mean square of pressure ﬂuctuations shows considerable diﬀerence at the top of the hill near separation point and in the middle of the domain near reattachment point where ﬂuctuating pressure is less than LES computation. 10 (a) Skin-friction coeﬃcient (b) Wall pressure coeﬃcient (c) Diﬀerence in skin-friction coeﬃcient (d) Diﬀerence in wall pressure coeﬃcient Figure 3. Comparison of the Skin-friction and wall pressure coeﬃcient among G2, G3, G4 and G5 grid with Dynamic k-SGS model and previous LES data from Breuer et. al.(Breuer et al., 2009). Diﬀerence is calculated with respect to the LES data from Breuer et al. 4.3. Coherent Structures In Figures 6(a), 6(c) and 6(e), iso-contours of spanwise vorticity component are shown. In the previous studies (Fr¨ ohlich et al., 2005), existence of Kelvin-Helmholtz vortices and Gortler vortices at the top of the hill along with Helical pairing of vortices in the middle of the domain has been suggested at Re = 10595. Due to very small turbulent length and time scales present in these simulations due to high Reynolds number, only K-H vortices can be identiﬁed at the top of the hill at this Reynolds number as marked in Fig 6. However, K-H vortices can be very easily identiﬁed at low Re cases such as Re=600 as shown in Figure 7. In comparison to iso-contours of spanwise vorticity for LES simulation on G4, LES on G3 shows smaller length of the vortex instability region as soon as it breaks into turbulent eddies. However, NBSST simulation on G3 shows comparable length of vortex instability region and then breaks into turbulent eddies. Similar trend is observed in Figures 6(b), 6(d) and 6(f), where pressure iso-contours are plotted for three simulations. The pressure rolls built up from the top of the hill disappears quickly in case of LES on G3 grid as compared to LES on G4 grid. However, it follows up to a similar distance for NBSST simulation on G3 grid. Another point to note is the size of smaller eddies present in LES simulations on G4 and G3 grids are similar whereas NBSST shows paucity of the smaller structures probably due to RANS modeling in the near wall region where length scales used for modelling of small turbulent structures is much larger than length scales used in LES. This is one of the reasons for better performance (poor performance) of hybrid methods for separated ﬂows (attached ﬂows). This suggests that care is required in the use of hybrid methods for 11 (a) Streamwise velocity (b) Wall-normal velocity (c) Mean streamwise Reynolds normal stress (d) Mean wall-normal Reynolds normal stress (e) Mean Reynolds shear stress (f) Mean turbulent kinetic energy Figure 4. Mean ﬂow statistics attached and weakly separated ﬂows because this issue will not be observed for massively separated ﬂows. This also provides an advantage to the users to obtain similar coherent structures using hybrid RANS-LES model on a coarser grid. 12 (a) LES (G4) (b) LES (G4) (c) NBSST (G3) (d) NBSST (G3) Figure 5. Mean pressure and mean square of pressure ﬂuctuation 4.4. Proper Orthogonal Decomposition Analysis Instantaneous ﬂow ﬁeld provides insight into the ﬂow structures, but it is also important for a model to predict the dynamics correctly. Here, POD has been used to asses the two-point statistical diﬀerence between predictions of these three simulations. POD calculations have been performed over 201 snapshots taken at regular intervals in a period of approximately 8 ﬂow through times (= 8*(Length of domain/average bulk velocity over the domain)). Flow ﬁelds have been decomposed into 201 eigen-values and eigen-functions corresponding to instantaneous pressure and instantaneous velocity ﬁeld. Let us consider a scalar ﬁeld φ(x, t ) with ﬁnite number of snapshots (here n = 1, 2, ..., 201) Then, two-point auto-covariance matrix of dimension n× n can be constituted using ensamble (< . >) of the scalar ﬁeld over the spatial co-ordinates as: R(t , t ) =< φ(x, t ).φ(x, t ) > m n m n = Σ λ φ (x)∗ φ (x) k k k k (22) φ(x, t ) = Σ a φ (x) n k k k 0 0 where, < a a > = δ λ k k kk k th Here, φ (x) are the empirical eigen-functions, λ are the eigen-values corresponding to k k k th eigen-function and a are the time coeﬃcients corresponding to k eigen-function. Since, two-point correlation matrix is constituted by the ensemble of square of instantaneous ﬁeld th th over the domain, k eigen-values corresponds to the partition of energy associated with k eigen-function. Therefore, the list of eigen-values has been arranged in decreasing order, so that ﬁrst mode corresponds to the eigen-function containing maximum energy. Since, the correlation matrix is computed using instantaneous ﬁeld, ﬁrst eigen-function obtained is the time averaged mean of the ﬁeld and corresponding eigen-value correspond to the energy associated with the mean ﬂow. So, for easier representation, eigen-values have been normalized with the eigen-value of the ﬁrst mode, so that, any eigen-value in ﬁgure 8 represents the energy associated with that eigen-function with respect to Mean ﬂow energy. 13 (a) LES (G4) (b) LES (G4) (c) LES (G3) (d) LES (G3) (e) NBSST (G3) (f) NBSST (G3) Figure 6. Instantaneous Flow Fields at Re=10595 Figure 7. Instantaneous spanwise vorticity using LES at Re=600 The same analysis can be performed for vector ﬁeld such as velocity ﬁeld, by considering a new scalar ﬁeld formed by concatenating the three components in a sequence. LES simulation on G4 grid, which is very accurate due to high resolution of grid used for computation, has been used here as a reference for the assessment of the hybrid RANS- LES model simulation on G3 grid. In Figures 8(a) and 8(b), eigen-values for instantaneous pressure and velocity ﬁelds have been plotted. Here, energy associated with mean mode of 14 (a) Pressure based POD (p) (b) Velocity based POD (V ) Figure 8. Eigen values corresponding to Proper Orthogonal decomposition modes of pressure and velocity vector. Eigen-values are normalised with the eigen-value of ﬁrst mode of the ﬁeld itself. Here, ﬁrst POD mode corresponds to the mean ﬁeld. (a) Mode1(LES(G4)) (b) Mode1(NBSST(G3)) Figure 9. Two-dimensional vector plot for ﬁrst eigen-mode for velocity based POD. pressure POD is 92% of the total energy whereas the energy associated with the mean mode of velocity ﬁeld is 82%. In Figure 8(a), ﬁrst two modes have much higher energy than the rest of the modes which follow continuous distribution of energy in higher modes; whereas in Figure 8(b), only ﬁrst mode has signiﬁcantly higher energy whereas all the higher modes show gradual decrease in energy. In comparison with the eigen-values plot of LES simulation on G4 grid, the nonlinear BlendedSST on G3 grid has very close predictions of distribution of eigen-values for high energy modes and shows small diﬀerence in low energy modes. In ﬁgures 9(a) and 9(b), two dimensional vector plot for ﬁrst eigen-mode of velocity based POD is shown. Both the plots show same ﬂow structures, indicating NBSST simulation on G3 grid is capable of predicting the coherent structures accurately. Also, Iso-contours of ﬁrst two modes of pressure based POD is shown in 10. The structures seen in both the modes for NBSST is is found to be similar as LES simulation except magnitude diﬀerence. Since, these eigen-modes are vector and considered only to visualize the high energy coherent structures in the ﬂow, the magnitude diﬀerence may not add too much value for the comparison. 5. CONCLUSION In the current study, a comparison has been made to assess the solution inaccuracy due to lower grid resolution in Large Eddy Simulation (LES) and improvement of solution using Hybrid RANS-LES model. LES seems to provide very accurate results, for turbulent separating ﬂows as in case of ﬂow over periodic hills considered here, using highly resolved grid. However, the results deteriorates very rapidly for computations performed using same model on a coarser grid. At the same time, hybrid RANS-LES model (non-linear 15 (a) Mode1(LES(G4)) (b) Mode1(NBSST(G3)) (c) Mode2(LES(G4)) (d) Mode2(NBSST(G3)) Figure 10. Iso-contours for ﬁrst and second eigen-mode for pressure based POD blended SST model considered here), provides signiﬁcantly improved results than pure LES calculations on a coarse grid. The performance of the models have been assessed in terms of 1-point mean statistics, energy distributions and coherent structures in diﬀerent modes of the ﬂow obtained using proper orthogonal decomposition. LES and hybrid RANS-LES has a close agreement for the ﬁrst and second order 1-point statistics while performing hybrid RANS-LES simulation on very coarse grid. Hence, the predictions of skin-friction and wall pressure coeﬃcients and separation and reattachment locations show high dependence on grid in case of LES but the results converse for hybrid RANS-LES on a coarser grid. Since the ﬂow quantities associated with the wall depends on complete ﬂow information in the domain and can only be matched exactly if the ﬂow structures and their dynamics are same in the domain. POD as an analysis tool for two- points statistics shows that hybrid RANS-LES model has close prediction of ﬂow dynamics as compared to LES simulation on a highly reﬁned grids. Also, looking at the computational cost, Hybrid RANS-LES simulation is very cheap as compared to the highly resolved LES simulation for the amount of inaccuracy we have to accept in Hybrid RANS-LES simulations. Hence, in all aspects, Hybrid RANS-LES model provides a good prediction of mean and instantaneous ﬂow ﬁelds on a coarser grid compared to the LES simulation on highly reﬁned grid. ACKNOWLEDGMENTS Simulations are carried out on the computers provided by the Indian Institute of Technology Kanpur (IITK) (www.iitk.ac.in/cc) and the manuscript preparation as well as data analysis has been carried out using the resources available at IITK. This support is gratefully th acknowledged. This paper is an invited contribution for the 10 International Symposium on Numerical Analysis of Fluid Flow and Heat Transfer - Numerical Fluids 2015, Rhodes, Greece. 16 REFERENCES K. Abe. A hybrid LES/RANS approach using an anisotropy-resolving algebraic turbulence model. International Journal of Heat and Fluid Flow, 26(2):204–222, 2005. doi: 10.1016/j.ijheatﬂuidﬂow.2004.08.009. K. Abe, Y.-J. Jang, and M. A. Leschziner. An investigation of wall-anisotropy expressions and length-scale equations for non-linear eddy-viscosity models. International Journal of Heat and Fluid Flow, 24(2):181–198, 2003. doi: 10.1016/S0142-727X(02)00237-0. K.-i. Abe. An advanced switching parameter for a hybrid les/rans model considering the characteristics of near-wall turbulent length scales. Theoretical and Computational Fluid Dynamics, 28(5):499–519, 2014. doi: 10.1007/s00162-014-0328-3. G. Almeida, D. Durao, and M. Heitor. Wake ﬂows behind two-dimensional model hills. Experimental Thermal and Fluid Science, 7(1):87–101, 1993. N. Aubry. On the hidden beauty of the proper orthogonal decomposition. Theoretical and Computational Fluid Dynamics, 2(5-6):339–352, 1991. G. Berkooz, P. Holmes, and J. L. Lumley. The proper orthogonal decomposition in the analysis of turbulent ﬂows. Annual review of ﬂuid mechanics, 25(1):539–575, 1993. M. Breuer, B. Jaﬀr´ ezic, and K. Arora. Hybrid LES-RANS technique based on a one-equation near-wall model. Theoretical and Computational Fluid Dynamics, 22(3):157–187, 2008. doi: 10.1007/s00162-007-0067-9. M. Breuer, N. Peller, C. Rapp, and M. Manhart. Flow over periodic hills–numerical and experimental study in a wide range of reynolds numbers. Computers & Fluids, 38(2): 433–457, 2009. B. J. Cantwell. Organized motion in turbulent ﬂow. Annual review of ﬂuid mechanics, 13 (1):457–515, 1981. L. Davidson and S. Dahlstr¨ om. Hybrid les-rans: computation of the ﬂow around a three- dimensional hill. Engineering Turbulence Modeling and Experiments, 6:319–328, 2005. L. Davidson and S.-H. Peng. Hybrid les-rans modelling: a one-equation sgs model combined with ak–ω model for predicting recirculating ﬂows. International Journal for Numerical Methods in Fluids, 43(9):1003–1018, 2003. C. De Langhe, B. Merci, K. Lodeﬁer, and E. Dick. Hybrid RANS/LES modelling with an approximate renormalization group. II: Applications. Journal of Turbulence, 6(14):1–16, 2005. doi: 10.1080/14685240500149765. L. T. Diosady and S. M. Murman. Dns of ﬂows over periodic hills using a discontinuous- galerkin spectral-element method. AIAA Paper, 2784, 2014. A. Fadai-Ghotbi, C. Friess, R. Manceau, and J. Bor´ ee. A seamless hybrid rans-les model based on transport equations for the subgrid stresses and elliptic blending. Physics of Fluids, 22:055104, 2010. doi: http://dx.doi.org/10.1063/1.3415254. H. F. Fasel, D. A. V. Terzi, and R. D. Sandberg. A methodology for simulating compressible turbulent ﬂows. Journal of Applied Mechanics, 73(3):405–412, 2006. doi: 10.1115/1.2150231. J. Fr¨ ohlich and D. V. Terzi. Hybrid LES/RANS methods for the simulation of turbulent ﬂows. Progress in Aerospace Sciences, 44(5):349–377, 2008. doi: 10.1016/j.paerosci.2008. 05.001. 17 J. Fr¨ ohlich, C. P. Mellen, W. Rodi, L. Temmerman, and M. A. Leschziner. Highly resolved large-eddy simulation of separated ﬂow in a channel with streamwise periodic constrictions. Journal of Fluid Mechanics, 526:19–66, 2005. S. S. Girimaji, R. Sreenivasan, and E. Jeong. PANS turbulence model for seamless transition between RANS, LES: Fixed-point analysis and preliminary results. In FEDSM2003- 45336, Proceedings of ASME FEDSM, volume 3, pages 13–16, 2003. H. Gopalan, S. Heinz, and M. K. St¨ ollinger. A uniﬁed RANS–LES model: Computational development, accuracy and cost. Journal of Computational Physics, 249:249–274, 2013. doi: 10.1016/j.jcp.2013.03.066. F. Hamba. A hybrid RANS/LES simulation of turbulent channel ﬂow. Theoretical and Computational Fluid Dynamics, 16(5):387–403, 2003. X. Han and S. Krajnovi´ c. An eﬃcient very large eddy simulation model for simulation of turbulent ﬂow. International Journal for Numerical Methods in Fluids, 71(11):1341–1360, 2013. doi: 10.1002/ﬂd.3714. L. S. Hedges, A. K. Travin, and P. R. Spalart. Detached-eddy simulations over a simpliﬁed landing gear. Journal of Fluids Engineering, 124(2):413–423, 2002. doi: 10.1115/1.1471532. K. Karhunen. Zur spektraltheorie stochastischer prozesse. 1946. W.-W. Kim and S. Menon. A new dynamic one-equation subgrid-scale model for large eddy simulations. In AIAA, Aerospace Sciences Meeting and Exhibit, 33 rd, Reno, NV, 1995. D. Kosambi. Statistics in function space. J. Indian Math. Soc, 7(1):76–88, 1943. M. Loeve. Functions aleatorie de second ordre. CR Acad. des Sci, page 220, 1945. J. Lumley. L. 1970 stochastic tools in turbulence. Academic Press, New York., 1970. C. P. Mellen, J. Fr¨ olich, and W. Rodi. Large Eddy Simulations of the ﬂow over periodic hills. In M. Deville and R. Owens, editors, IMACS World Congress, 2000. F. Menter and Y. Egorov. A scale adaptive simulation model using two-equation models. In Aerospace Sciences Meetings, pages 2005–1095. American Institute of Aeronautics and Astronautics, Reno, Nevada, 10-13 January 2005. doi: 10.2514/6.2005-1095. F. R. Menter. Improved two-equation k-omega turbulence models for aerodynamic ﬂows. NASA STI/Recon Technical Report N, 93:22809, Oct. 1992. A. Obukhov. Statistical description of continuous ﬁelds. Tr. Geophys. Int. Akad. Nauk. SSSR, 24:3–42, 1954. R. L. Panton. Self-sustaining mechanisms of wal l turbulence, volume 15. Computational Mechanics, 1997. S. Pope. Turbulent ﬂows, 771 pp, 2000. V. S. Pugachev. The general theory of correlation of random functions. Izvestiya Rossiiskoi Akademii Nauk. Seriya Matematicheskaya, 17(5):401–420, 1953. S. Sari´ c, S. Jakirli´ c, M. Breuer, B. Jaﬀr´ ezic, G. Deng, O. Chikhaoui, J. Fr¨ ohlich, D. Von Terzi, M. Manhart, and N. Peller. Evaluation of detached eddy simulations for predicting the ﬂow over periodic hills. In ESAIM: proceedings, volume 16, pages 133–145. EDP Sciences, 2007. 18 M. Shur, P. R. Spalart, M. Strelets, and A. Travin. Detached-eddy simulation of an airfoil at high angle of attack, 4th int. In Symp. Eng. Turb. Model ling and Measurements, pages 669–678, 1999. L. Sirovich. Turbulence and the dynamics of coherent structures. part i: Coherent structures. Quarterly of applied mathematics, 45(3):561–571, 1987. P. R. Spalart. Detached-eddy simulation. Annual Review of Fluid Mechanics, 41:181–202, 2009. doi: 10.1146/annurev.ﬂuid.010908.165130. L. Temmerman, M. Hadˇ ziabdi´ c, M. Leschziner, and K. Hanjali´ c. A hybrid two-layer urans– les approach for large eddy simulation at high reynolds numbers. International journal of heat and ﬂuid ﬂow, 26(2):173–190, 2005. H. Tennekes and J. L. Lumley. A ﬁrst course in turbulence. MIT press, 1972. F. Tessicini, L. Temmerman, and M. Leschziner. Approximate near-wall treatments based on zonal and hybrid rans–les methods for les at high reynolds numbers. International Journal of Heat and Fluid Flow, 27(5):789–799, 2006. P. G. Tucker and L. Davidson. Zonal k–l based large eddy simulations. Computers and Fluids, 33(2):267–287, 2004. doi: 10.1016/S0045-7930(03)00039-2. 6. APPENDICES 6.1. k-ω SST RANS Model In this section details of the baseline RANS models are provided. K-ω SST model (Menter, 1992) is used as the baseline linear RANS model. The model solves two transport equation for the TKE and turbulent dissipation ω (Menter, 1992) ∂k ∂k ∂ ∂k + u ¯ = P − β ωk + (ν + σ ν ) (23) j k k t ∂t ∂x ∂x ∂x j j j ∂ω ∂ω γ ∂ ∂ω + u ¯ = P − βω + (ν + σ ν ) (24) j k w t ∂t ∂x ν ∂x ∂x j t j j σ ∂k ∂ω w2 + 2 (1− F ) ω ∂x ∂x j j where P = max(τ ∂u ¯ /∂x , 10β ∗ ωk), is the kinetic energy production term, β = 0.09 is k ij i j a model constant and the last term in the ω equation is the cross-diﬀusion term. F is a blending function which has a value of one inside the boundary layer and zero outside. The limiting of the production term is an alternative to the use of damping function in the near-wall region. The turbulent stress tensor and viscosity are computed in this model as follows τ = kδ − 2ν S + N (25) ij ij t ij ij a k ν = (26) ¯ ¯ max a ω, 2S S F 1 ij ij 2 Here N is the non-linear part of the turbulent stress tensor. a = 0.31 is a model constant ij 1 and F is a blending function similar to F . The expressions for the blending functions are 2 1 19 given by   ( " # )! k 500ν 4σ k w2   F = tanh min max , , (27) ∗ 2 2 β ωd d ω CD d kw σ ∂k ∂ω w2 −10 CD = max 2 , 10 (28) kw ω ∂x ∂x i i   k 500ν   F = tanh 2 , (29) ∗ 2 β ωd d ω The model constants are calculated by blending K-ω model near the wall and K- away from the wall using the blending function F . The form is given by φ = φ F + (1− F )φ (30) 1 1 1 2 The model constants are given in Table IV. For the non-linear K-ω SST model (NSST), a modiﬁed form of the non-linear constitutive relation proposed by Abe et al. (Abe et al., 2003) is used. The non-linear term is deﬁned as follows 3ν 2 ¯ ¯ ¯ ¯ ¯ ¯ N = f f 2S S − S S δ − S Ω − S Ω ij NL s ik kj nk kn ij ik kj jk ki k 3 + 2kd (31) ij f = C C (1− f (26)) (32) NL D B w C = (33) 2 2 2 2 2 1 + 22/3(C ν /k) Ω + 2/3(C ν /k) (Ω − S ) f D t D t B f (η) = exp − y /η 2 2 2 n o S Ω − S f = 1− 1 + C C (Ω− S) (34) s s2 D 2 2 (Ω + S ) ¯ ¯ where Ω = Ω Ω is the characteristic rotation rate, S = S S is the characteristic ij ij ij ij strain rate, f = 1 + C C ν / (k (Ω− S)), y = u d /ν is a non-dimensional wall distance, B η D t τ y η is a parameter, and C = 0.8, C = 100, and C = 7 are model constants. Finally, the D η s2 expression for the term d can be written as ij 1 δ ij d = −α f (26) (d d − d d ) w w i j k k ij 2 3 ( ) β C w w 2 2 ¯ ¯ + f (26)(1− f )T − √ S Ω − Ω S w ik kj ik kj r1 d 2 2 1 + C T S Ω γ C δ w w ij 2 2 2 ¯ ¯ + f (26)(1− f )T S S − S (35) w ik kj r1 d 1 + C T S 3 where d = ∂N /∂x , N is the unit-normal, and i i j i 2 2 2 2 f = (Ω − S )/(Ω + S ) (36) r1 T = {1− f (15)} k/ + f (15)δ ν/ (37) d w w w ∗ w Here = β ωk is the turbulence dissipation term.The model constants for the term d are ij given by α = 1, β = , C = 0.5, γ = 1.5, and δ = 1.0 (38) w w w w w 4 20 Table IV. Model constants for the SST model. σ = 0.85 γ = 5/9 β = 0.075 σ = 0.5 k w1 σ = 1.0 γ = 0.44 β = 0.0828 σ = 0.856 k w2 The non-linear model NSST is used as the baseline model for the non-linear hybrid model. A damping function was used in Ref. (Abe et al., 2003). However, this term is not used in the current formulation as production limiter is used. In the original formulation of Abe et al. (Abe et al., 2003), the cross-diﬀusion term is not included. It has been included based on the recent improvement in results obtained with the inclusion of this term.

Journal

Physics – arXiv (Cornell University)

Published: Feb 18, 2021

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Investigation of flow structures in a turbulent separating flow using hybrid RANS-LES model

Investigation of flow structures in a turbulent separating flow using hybrid RANS-LES model

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Investigation of flow structures in a turbulent separating flow using hybrid RANS-LES model

Investigation of flow structures in a turbulent separating flow using hybrid RANS-LES model

References (53)

Abstract

Journal

Recommended Articles

There are no references for this article.

Our policy towards the use of cookies