+ All Categories
Home > Documents > [American Institute of Aeronautics and Astronautics 41st AIAA Fluid Dynamics Conference and Exhibit...

[American Institute of Aeronautics and Astronautics 41st AIAA Fluid Dynamics Conference and Exhibit...

Date post: 14-Dec-2016
Category:
Upload: bassam
View: 212 times
Download: 0 times
Share this document with a friend
24
Near-wall second moment closure based on DNS analysis of pressure correlations G.A. Gerolymos, * C. Lo, I. Vallet Faculty of Engineering, Universit´ e Pierre-et-Marie-Curie (UPMC), 75005 Paris, France B.A. Younis § Department of Civil and Environmental Engineering, University of California, Davis, CA 95616, USA The purpose of the present paper is to use recent DNS results for the quadruple decomposition of pressure fluctuations in wall turbulence (rapid/slow and volume/wall terms) in the develop- ment of a new near-wall wall-normal free second-moment closure for wall-bounded flows. Based on the comparison of existing models with DNS results we propose the addition of new inho- mogeneous terms in the tensorial representations for pressure diffusion and for pressure-strain redistribution, and calibrate the proposed representations both a priori and a posteriori. The resulting model is then assessed by comparison with experimental data for flat-plate boundary- layer, separating diffuser and shock-wave/boundary-layer interaction flows. The overall behav- ior of the model is quite satisfactory and we analyze in detail the particular points which can be improved. These drawbacks, mainly related with the wall-layer asymptotic behavior of the Reynolds-stresses, are common to many existing closures, and it is suggested that to improve this behavior a model including transport equations for the components of the dissipation ten- sor should be developed. I. Introduction Experience with a previously developed second-moment closure 1 in complex flows around or inside com- plex geometries 2–6 has shown satisfactory prediction of separation, with a slightly slower than experiment reattachment behavior. Furthermore the Reynolds-stress model developed by Gerolymos and Vallet (GV RSM 1 ) follows Lumley’s 7 suggestion to model together redistribution and the anisotropy of dissipation (φ ij ε ij + 2 3 εδ ij ). In a recent research effort, 8–10 we have developed a DNS processing algorithm based on a Green’s function approach 11 to separate the rapid and slow pressure fluctuations obtained from the Poisson equation for fluctuating pressure into volume and wall echo terms p (x,y,z,t)= p (r;V) (x,y,z,t)+ p (r;w) (x,y,z,t) p (r) (x,y,z,t) + p (s;V) (x,y,z,t)+ p (s;w) (x,y,z,t) p (s) (x,y,z,t) +p (τ ) (x,y,z,t) (1) with corresponding splitting for redistribution φ ij , pressure transport p u i , pressure diffusion d (p) ij , and veloc- ity/ pressure-gradient correlation Π ij = φ ij + d (p) ij . These new DNS data 8, 9 indicate that wall-echo in pressure diffusion is quite weak, and have been used to evaluate different proposals for d (p) ij . 7, 12–14 The DNS data have also been used 8, 9 to assess several standard Reynolds-stress models typical of different approaches for the representation of φ ij , 15–19 separately evaluating quasi-homogeneous and wall (inhomogeneous) terms in the closure. * Institut d’Alembert, case 161, 4 place Jussieu, e-mail: [email protected], AIAA senior member Institut d’Alembert, case 161, 4 place Jussieu, e-mail: [email protected] Institut d’Alembert, case 161, 4 place Jussieu, e-mail: [email protected], AIAA senior member § [email protected] 1 of 24 American Institute of Aeronautics and Astronautics 41st AIAA Fluid Dynamics Conference and Exhibit 27 - 30 June 2011, Honolulu, Hawaii AIAA 2011-3574 Copyright © 2011 by G.A. Gerolymos. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
Transcript

Near-wall second moment closure based on

DNS analysis of pressure correlations

G.A. Gerolymos,∗ C. Lo, † I. Vallet‡

Faculty of Engineering, Universite Pierre-et-Marie-Curie (UPMC), 75005 Paris, France

B.A. Younis§

Department of Civil and Environmental Engineering, University of California, Davis, CA 95616, USA

The purpose of the present paper is to use recent DNS results for the quadruple decomposition

of pressure fluctuations in wall turbulence (rapid/slow and volume/wall terms) in the develop-

ment of a new near-wall wall-normal free second-moment closure for wall-bounded flows. Based

on the comparison of existing models with DNS results we propose the addition of new inho-

mogeneous terms in the tensorial representations for pressure diffusion and for pressure-strain

redistribution, and calibrate the proposed representations both a priori and a posteriori. The

resulting model is then assessed by comparison with experimental data for flat-plate boundary-

layer, separating diffuser and shock-wave/boundary-layer interaction flows. The overall behav-

ior of the model is quite satisfactory and we analyze in detail the particular points which can

be improved. These drawbacks, mainly related with the wall-layer asymptotic behavior of the

Reynolds-stresses, are common to many existing closures, and it is suggested that to improve

this behavior a model including transport equations for the components of the dissipation ten-

sor should be developed.

I. Introduction

Experience with a previously developed second-moment closure1 in complex flows around or inside com-

plex geometries2–6 has shown satisfactory prediction of separation, with a slightly slower than experiment

reattachment behavior. Furthermore the Reynolds-stress model developed by Gerolymos and Vallet (GV

RSM1) follows Lumley’s7 suggestion to model together redistribution and the anisotropy of dissipation (φij −εij + 2

3εδij).

In a recent research effort,8–10 we have developed a DNS processing algorithm based on a Green’s function

approach11 to separate the rapid and slow pressure fluctuations obtained from the Poisson equation for

fluctuating pressure into volume and wall echo terms

p′(x, y, z, t) = p′(r;V)(x, y, z, t) + p′(r;w)(x, y, z, t)︸ ︷︷ ︸

p′(r)(x, y, z, t)

+ p′(s;V)(x, y, z, t) + p′(s;w)(x, y, z, t)︸ ︷︷ ︸

p′(s)(x, y, z, t)

+p′(τ)(x, y, z, t) (1)

with corresponding splitting for redistribution φij , pressure transport p′u′i, pressure diffusion d

(p)ij , and veloc-

ity/ pressure-gradient correlation Πij = φij + d(p)ij . These new DNS data

8,9 indicate that wall-echo in pressure

diffusion is quite weak, and have been used to evaluate different proposals for d(p)ij .7,12–14 The DNS data have

also been used8,9 to assess several standard Reynolds-stress models typical of different approaches for the

representation of φij ,15–19 separately evaluating quasi-homogeneous and wall (inhomogeneous) terms in the

closure.

∗Institut d’Alembert, case 161, 4 place Jussieu, e-mail: [email protected], AIAA senior member†Institut d’Alembert, case 161, 4 place Jussieu, e-mail: [email protected]‡Institut d’Alembert, case 161, 4 place Jussieu, e-mail: [email protected], AIAA senior member§[email protected]

1 of 24

American Institute of Aeronautics and Astronautics

41st AIAA Fluid Dynamics Conference and Exhibit27 - 30 June 2011, Honolulu, Hawaii

AIAA 2011-3574

Copyright © 2011 by G.A. Gerolymos. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.

The purpose of the present paper is to develop a second-moment closure, separately modeling the anisotropy

of dissipation and redistribution tensors, maintaining the satisfactory prediction of separation of the GV

RSM1 and incorporating a specific model for pressure diffusion d(p)ij , which was shown in previous studies

14,20

to improve the prediction of reattachment and relaxation zones.

II. Flow Model and Turbulence Closures

The flow is modeled by the Favre-Reynolds-averaged Navier-Stokes equations,1,5 coupled with the ap-

propriate modeled turbulence-transport equations. Turbulent kinetic energy k = 12 u′′

i u′′i is linked to the

mean-energy transport equation through the source term by convecting the total enthalpy of mean flow

ht = h + 12 uiui rather than the Favre-averaged total enthalpy ht = ht + k (h is the specific enthalpy). The

symbol (.) is used to denote a function of average quantities that is neither Reynolds-averaged (.) nor Favre-

averaged (.), (.)′′are Favre-fluctuations and (.)

′are Reynolds-fluctuations. All computations were performed

for air considered thermodynamically and calorically perfect. The density-fluctuation effect and the influence

of temperature fluctuations were neglected in the mean-flow transport equations.5 Numerical method which

is accelerated by multigrid technique and numerical limiters for eddy-viscosity and Reynolds-stress models,

used for all computations are described by Gerolymos and Vallet.21,22

II.A. Reynolds-stress models

Second-moment closures solve the Favre-Reynolds-averaged Reynolds-stress transport equation, the exact

form of which is

Cij = dij + Pij + φij − ρεij + Kij +2

3φpδij (2)

Cij =∂ρu′′

i u′′j

∂t+

∂(ρu′′i u′′

j uℓ)

∂xℓ; Pij = −ρu′′

i u′′ℓ

∂uj

∂xℓ− ρu′′

j u′′ℓ

∂ui

∂xℓ(3)

where, the convectionCij and the production Pij terms are exact contrary to the Launder-Sharma k−εmodel

which requires closure for the Reynolds-stress tensor −ρu′′i u′′

j and uses the Boussinesq hypothesis to close

Pij . The diffusion dij (Eq. 4), the pressure-strain redistribution φij (Eq. 5) which is zero in k − ε models(φℓℓ = 0) and the dissipation ρεij (Eq. 5) terms require modeling

dij = d(u)ij + d

(p)ij + d

(µ)ij = d

(T)ij + d

(µ)ij =

∂xℓ

(−ρu′′

i u′′j u′′

ℓ − p′u′jδiℓ − p′u′

iδjℓ + µ∂u′

iu′j

∂xℓ

)(4)

φij = p′(

∂u′i

∂xj+

∂u′j

∂xi− 2

3

∂u′k

∂xkδij

); ρεij =

(τ ′jℓ

∂u′i

∂xℓ+ τ ′

iℓ

∂u′j

∂xℓ

)(5)

where τij = µ( ∂ui

∂xj+

∂uj

∂xi− 2

3∂um

∂xmδij) is the viscous stress tensor. The viscous diffusion term d

(µ)ij is approximated

by neglecting the influence of temperature fluctuations

d(µ)ij =

∂xℓ

∂u′′i u′′

j

∂xℓ

); µ = µ(T ) (6)

For the three Reynolds-stressmodels used in the present study, direct compressibility effectsKij and pressure-

dilatation correlation φp terms were neglected,

Kij =

(−u′′

i

∂p

∂xj− u′′

j

∂p

∂xi+ u′′

i

∂τjℓ

∂xℓ+ u′′

j

∂τiℓ

∂xℓ

)∼= 0 ; φp = p′

∂u′k

∂xk

∼= 0 (7)

The turbulence-length-scale ℓT was determined by solving the Launder-Sharma23 modified dissipation-rate

ε∗ = ε − 2ν(grad√

k)2 transport equation with the exception of the diffusion term where a tensorial diffusioncoefficient is used.24,25

∂ρε∗

∂t+

∂ (uℓρε∗)

∂xℓ=

∂xℓ

[Cε

k

ε∗ρu′′

mu′′ℓ

∂ε∗

∂xm+ µ

∂ε∗

∂xℓ

]

︸ ︷︷ ︸dε = d

(T)ε + d

(µ)ε

+Cε1Pkε∗

k− Cε2ρ

ε∗2

k+ 2µCµ

k2

ε∗∂2ui

∂xℓ∂xℓ

∂2ui

∂xm∂xm(8)

Pk =1

2Pℓℓ ; Cε = 0.18 ; Cε1 = 1.44 ; Cε2 = 1.92(1 − 0.3e−Re∗

T2

) ; Cµ = 0.09e− 3.4

(1+0.02Re∗T)2 ; Re∗

T=

ρk2

µε∗(9)

2 of 24

American Institute of Aeronautics and Astronautics

where Re∗Tis the turbulent Reynolds-number based on ε∗.

II.B. The Gerolymos-Vallet Wall-Normal-Free RSM (GV RSM)

Although the distance from the wall vector can be easily determined, even in complex flows, by solving

the eikonal equation,26,27 the use of geometric parameters (often fitted on flat-plate boundary-layer flow to

reproduce the mean-velocity logarithmic law) is directly responsible to the confinement close to the wall

of the inhomogeneous part of the model. As a result, anisotropic flows such as large recirculation zones are

underestimated, and improvement of RSM over linear eddy-viscosity closure is less spectacular than expected.

To improve the prediction of the inhomogeneous part of the flow, which may be important very far away from

solid walls, it is preferable to split the redistributive term into an homogeneous and an inhomogeneous parts

following the idea proposed by Craft and Launder.28

Gerolymos and Vallet1,29 introduced a unit-vector ~e I pointing in the direction of inhomogeneity of theturbulent field to replace the geometric wall-normals.

~e I = eIi~ei =

grad

{ℓT[1 − e−

Re∗

T

30 ]1 + 2

√A2 + 2A16

}

∣∣∣∣∣

∣∣∣∣∣grad{

ℓT[1 − e−Re∗

T

30 ]1 + 2

√A2 + 2A16

}∣∣∣∣∣

∣∣∣∣∣

; ℓT =k3/2

ε(10)

This relation can be used in any existing RSM to replace the geometric unit-normals19 which are often used

in the redistribution or dissipation tensors closures.30–32

The redistribution term φij is modeled with the dissipation ρεij term where the homogeneous slow part

φSHij contains also the anisotropic part of the dissipation tensor (εij − 23δijε) following the suggestion proposed

by Lumley7 (Eq. 14).

φij − ρεij =[φSHij − ρ

(εij − 2

3δijε)]

+ φRHij + φSIij + φRIij − 23δij ρε

= −CSH

φ ρεaij − CRH

φ

(Pij − 1

3δijPmm

)

+CSI

φ

ε

k

[ρu′′

nu′′meIneImδij − 3

2 ρu′′nu′′

i eIneIj − 32 ρu′′

nu′′j eIneIi

]

+CRI

φ

[φRHnmeIneImδij − 3

2φRHineIneIj − 32φRHjneIneIi

]− 2

3δij ρε (11)

where the coefficients CSI

φ and CRI

φ mimic distance from the wall effects but without using any wall-topology

parameter

CSI

φ = 0.83 [1 − 23 (CSH

φ − 1)]

∣∣∣∣∣

∣∣∣∣∣grad{

ℓT[1 − e−Re∗

T

30 ]

1 + 2A0.82

}∣∣∣∣∣

∣∣∣∣∣ (12)

CRI

φ = max

[2

3− 1

6CRH

φ

, 0

] ∣∣∣∣∣

∣∣∣∣∣grad{

ℓT[1 − e−Re∗

T

30 ]

1 + 1.8Amax(0.6,A)2

}∣∣∣∣∣

∣∣∣∣∣ (13)

Contrary to the RSM developed in the present study (cf II.C), the GV RSM used the form proposed by

Rotta33 for the homogeneous slow term (Eq. 15) and the coefficient CSH

φ developed by Launder and Shima34

(Eq. 16).

φSHij − ρεij =[φSHij − ρεij + 2

3 ρεδij

]− 2

3 ρεδij (14)

= −CSH

φ (ReT, A2, A3)ρεaij − 23 ρεδij (15)

CSH

φ = 1 + 2.58AA142

[1 − e

ReT150

«2]; ReT =

ρk2

µε(16)

whereA2 andA3 are related to the second and the third invariants of the Reynolds-stresses anisotropy tensor

aij , A is the flatness parameter proposed by Lumley7 and ReT is the turbulent Reynolds number

aij =u′′

i u′′j

k− 2

3δij ; A1 = aii = 0 ; A2 = aikaki ; A3 = aikakjaji ; A = 1 − 9

8(A2 − A3) (17)

3 of 24

American Institute of Aeronautics and Astronautics

Despite a non-optimal asymptotic behavior of Reynolds-stresses at the wall (in particular the Reynolds-

stress u′′u′′ and w′′w′′ components for flat-plate boundary-layer flow; Fig. 13) this simplified form (Eq. 14)

is numerically very stable at the wall and is very practical when computing complex flows over complex

geometries.3

The rapid-homogeneous term φRHij corresponds to the form proposed by Naot et al.35,36 (Eq. 11). The

rapid-part coefficient CRH

φ was developed by Gerolymos and Vallet1

CRH

φ = min [1, 0.75 + 1.3 max [0, A − 0.55]] × A[max(0.25,0.5−1.3max [0,A−0.55])][1 − max(0, 1 − ReT50 )] (18)

and calibrated both for flat plate boundary-layers and transonic channel flows.1 The particular form of the

coefficientCRH

φ (Eq. 18) in conjunction with the wall-topology-free inhomogeneous terms (Eqs. 10, 11), improve

the prediction of detached29 and secondary flows.20 The rapid-part coefficient CRH

φ of Launder-Shima (CRH

φ =

0.75√

A) was modified by Gerolymos and Vallet (Eq. 18) to sharply raise to a value of one when the flatnessparameter of Lumley A approaches unity (Fig. 1) , which corresponds to separation region (A ∼ 0.9).1,29

0

0.2

0.4

0.6

0.8

1

0 0.2 0.4 0.6 0.8 1

Gerolymos-Vallet (2001)Launder-Shima (1989)

-A

6

CRH

φ

Figure 1. Rapid-part coefficient CRHφ (A, ReT = ∞) in the high turbulence Reynolds number limit

The pressure-velocity correlation in the turbulent-diffusion term was neglected (p′u′i ≃ 0) and the Hirt12

proposal with the Hanjalic and Launder24 coefficient (Eq. 19), which respects the tensorial symmetry of

u′′i u′′

j u′′ℓ , was used to model the triple-velocity correlation

−ρu′′i u′′

j u′′ℓ = CS

u

k

ε

(ρu′′

i u′′m

∂u′′j u′′

∂xm+ ρu′′

j u′′m

∂u′′ℓ u′′

i

∂xm+ ρu′′

ℓ u′′m

∂u′′i u′′

j

∂xm

); CS

u = 0.11 (19)

II.C. Present Wall-Normal-Free RSM (present RSM)

Previous studies with the GV Reynolds-stress model1 indicate that separation is quite accurately predicted,

but also that there is room for improvement in the reattachment and relaxation region.5 Extensive testing

suggests that the modeling of the pressure terms in the Reynolds-stress transport equations has the greatest

impact on the prediction of both separation and reattachment.

We maintain, in the modeling approach, the splitting of the velocity/pressure-gradient tensor Πij into a

pressure-diffusion term d(p)ij and a redistribution term φij and neglect the pressure/dilatation correlation φp.

Πij = −u′i

∂p′

∂xj− u′

j

∂p′

∂xi(20)

= d(p)ij + φij +

2

3φpδij

=∂

∂xℓ

(−p′u′

jδiℓ − p′u′iδjℓ

)+ p′

(∂u′

i

∂xj+

∂u′j

∂xi− 2

3

∂u′m

∂xmδij

)+

2

3φpδij (21)

DNS database of plane channel flow, where the slow and rapid part of pressure fluctuation were slip into

volume and wall-echo terms,8,9 were used for a priori and a posteriori evaluation of modeling proposals for

redistribution φij and pressure-diffusion d(p)ij terms. DNS data were generated for two Reynolds-numbers

but for quasi-incompressible and supersonic Mach numbers (Tab. 1), using the DNS solver described and

validated in Gerolymos et al.37 The DNS database at the highest Reynolds and Mach numbers (Reτw=

4 of 24

American Institute of Aeronautics and Astronautics

226, MCL = 1.5) was used only for the a posteriori assessment because the four-part decomposition for theredistribution and the pressure-diffusion terms is not available, as in the compressible flow case the splitting

of the Poisson equation for p′ includes 10 terms.38

Table 1. Parameters of the DNS computations

ReτwMCL Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+CL ∆z+ ∆t+ t+OBS ∆t+s

179 0.34 193 × 129 × 169 4πδ 2δ 43πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 2329 6.47 × 10−3

226 1.50 241 × 137 × 217 4πδ 2δ 43πδ 11.9 0.23 19 5.6 4.4 16.5 × 10−3 1952 16.5 × 10−3

Lx, Ly , Lz (Nx, Ny , Nz) are the dimensions (number of grid-points) of the computational domain (x = homogeneous streamwise,

y = normal-to-the-wall, z = homogeneous spanwise direction); δ is the channel half-height; ∆x+, ∆y+w , ∆y+

CL, ∆z+ are the mesh-

sizes in wall-units; (·)w denotes wall and (·)CL centerline values; Ny+≤10 is the number of grid points between the wall and y+ = 10;

Reτw := uτ δν−1w ; uτ is the friction velocity; δ is the channel half-height; νw = is the kinematic viscosity at the wall; MCL is the centerline

Mach-number; ∆t+ is the computational time-step in wall-units; t+OBSis the observation period in wall units over which statistics were

computed; ∆t+s is the sampling time-step for the single-point statistics in wall-units.

Despite DNS databases over a backward-facing step39 or behind a rectangular trailing edge40 and previ-

ous studies14,20 which have shown the influence of the turbulent diffusion due to pressure fluctuations d(p)ij

in recirculating flows, the pressure-diffusion term is often neglected. Most of the second-moment closures

which take into account the pressure diffusion tensor, model the pressure-velocity correlation p′u′i and then

compute the divergence of this term. The slow part of the pressure-velocity correlation is generally modeled

through the triple-velocity correlation modeling7,13,18,20 following the only available theoretical closure es-

tablished by Lumley7 for weakly inhomogeneous flows, while very few proposal for the rapid-part have been

developed.13,14,18

In the present second-moment closure, the pressure-diffusion model contains a Lumley-type7 slow quasi-

homogeneous term, with slow and rapid inhomogeneous terms containing gradε∗ ⊗ gradε∗ and gradk ⊗ gradkrespectively

d(p)ij = CSP1ρ

k3

ε3

∂ε∗

∂xi

∂ε∗

∂xj+

∂xℓ

[CSP2(ρ ˜u′′

mu′′mu′′

j δiℓ + ρ ˜u′′mu′′

mu′′i δjℓ)

]+ CRPρ

k2

ε2Skℓaℓk

∂k

∂xi

∂k

∂xj(22)

Sij = 12

(∂ui∂xj

+∂uj

∂xi

)(23)

where aij is the Reynolds-stresses anisotropy tensor (Eq. 17) and Sij is the meanflow rate-of-strain tensor.

The triple-velocity correlation is approximated with the Hanjalic and Launder24 proposal (Eq. 19) instead

of the Lumley model7,14 which has a tendency to minimize the positive effect of the pressure-diffusion term

by increasing the relaxation zone in detached flows14 and explains the relatively unsatisfactory prediction

obtained with the proposal developed by Vallet14 in the diffusing duct of Wellborn et al.41

The coefficients CSP1 and CRP were calibrated unsing the DNS data base of Gerolymos-Senechal-Vallet42

for developed plane channel flow, with splitting of the slow and the rapid parts of the pressure-diffusion ten-

sor,8,9,43 while the coefficient of the slow quasi-homogeneous term CSP2, already used in a previous study20

was recalibrated. Notice that, the DNS database of Moser-Kim-Mansour44 does not include all of the triple-

velocity-correlation component which appears in the slow quasi-homogeneous Lumley model for fully devel-

oped plane channel flow (Eq. 25).

d(p)xy =

∂y

[CSP2(ρu′′u′′u′′ + ρv′′v′′u′′ + ρ ˜w′′w′′u′′)

](24)

d(p)yy = CSP1ρ

k3

ε3

∂ε∗

∂y

∂ε∗

∂y+

∂y

[2CSP2(ρu′′u′′v′′ + ρv′′v′′v′′ + ρw′′w′′v′′)

]+ CRPρ

k2

ε2Sxyaxy

∂k

∂y

∂k

∂y(25)

Therefore, the coefficient function CSP2 which was calibrated in the previous study14,20 on a priori assessment

using the Lumley model for the triple-velocity correlation instead of triple-velocity-correlation from DNS data

base, was overestimated. We suggest a constant-value CSP2 = 0.022 instead of CSP2 = 0.2 proposed by Lumley7

which is too high (Fig. 4).

CSP2 = 0.022 (26)

The slow inhomogeneous term which contains gradε∗ allows to modify the incorrect opposite sign of the

maximum peak located in the buffer-layer zone at y+ ∼= 10 for the normal vertical component d(p)yy observed

5 of 24

American Institute of Aeronautics and Astronautics

for all turbulence closures assessed7,13,14 (Fig. 2), which is due to the slow quasi-homogeneous term proposed

by Lumley.7 The rapid term was added to correct the sign of the peak very close to the wall at y+ ∼= 6 for d(p)yy .

Notice that, it is easier to use gradε∗ and gradk to model directly the turbulent-diffusion term instead of thepressure-velocity correlation. Furthermore, since the echo-term was not taken into account, which means

that only the sum Πij = φij + d(p)ij was correctly modeled near the wall and not the individual terms, so that

both the redistribution and the turbulent pressure-diffusion closures erroneously vanish to zero at the wall

for all of the tensors components. The value of the coefficients CSP1 and CRP were calibrated on DNS database

of Gerolymos et al.8 (Figs. 2, 3), and we propose

CSP1 = −0.005 ; CRP = −0.005 (27)

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

1 10 100 1000

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

0.5

1 10 100 1000

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1 10 100 1000

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1 10 100 1000

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

1 10 100 1000

-0.04

-0.02

0

0.02

0.04

0.06

0.08

0.1

1 10 100 1000

[d(p;s;w)xy ]+

[d(p;s;w)yy ]+

[d(p;s;V)xy ]+

[d(p;s;V)yy ]+

[d(p;s)xy ]+

[d(p;s)yy ]+

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obsgf

∆t+sgf

179 0.34 193 × 129 × 169 4πδ 2δ 43πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 1648 6.47 × 10−1

dns

present model Vallet (2007)

drdl (1996)

Hirt (1969)

Lumley (1978)

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

Figure 2. A priori comparison of various models (present; Hirt;12 Lumley;7 DRDL;13 Vallet14) for the slow pressure diffusion d(p;s)ij

=

d(p;s;V)ij

+ d(p;s;w)ij

with DNS database of Gerolymos et al.8,9 for the volume d(p;s;V)ij

and the wall-echo d(p;s;w)ij

terms, in wall units [d(p)ij

]+ =

d(p)ij

νw/(ρwu4τ ), plotted against the nondimensional distance from the wall y+ = yuτ /νw .

The influence of the pressure-strain correlation φij = φRij +φSij for the prediction of confined complex flows

led to numerous near-wall closure proposals.1,15,34,45–48 The rapid-part closure of the redistribution term

φRij = φRHij + φRIij developed and assessed by Gerolymos-Vallet1 to improve the prediction of secondary and

detached flows, was not modified in the present study.

φRij = −CRH

φ

(Pij − 1

3δijPmm

)+ CRI

φ

[φRHnmeIneImδij − 3

2φRHineIneIj − 32φRHjneIneIi

](28)

CRH

φ = min [1, 0.75 + 1.3 max [0, A − 0.55]] × A[max(0.25,0.5−1.3max [0,A−0.55])][1 − max(0, 1 − ReT50 )] (29)

CRI

φ = max

[2

3− 1

6CRH

φ

, 0

] ∣∣∣∣∣

∣∣∣∣∣grad{

ℓT[1 − e−Re∗

T

30 ]

1 + 1.6Amax(0.6,A)2

}∣∣∣∣∣

∣∣∣∣∣ (30)

The slow-part φSij = φSHij + φSIij is usually modeled in the Lumley basis7 for the homegeneous part φSHij =

φSHij (ε, aij , a2ij) and the inhomogeneous part φ

SI

ij (or wall terms) use the form proposed by Shir49 with distance-

from-the-wall and normal-to-the-wall34,47 or without any wall-topology parameter.1,28,50 The two-component

limit realizability constraint, that a near-wall closure for φSHij should satisfy, is usually obtained by including

the anisotropic part of the dissipation term1,34,51 as suggested by Lumley,7 or by using a coefficient function of

the Reynolds-stress anisotropic tensor invariants (especially the Lumley flatness-parameterA (Eq. 17) whichvanish to zero at the two-component limit) and of the Reynolds number45,47 instead of constant values.15,52

6 of 24

American Institute of Aeronautics and Astronautics

-0.02

-0.015

-0.01

-0.005

0

0.005

0.01

1 10 100 1000

-0.004

-0.003

-0.002

-0.001

0

0.001

0.002

0.003

0.004

1 10 100 1000

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

1 10 100 1000

-0.05

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

0.03

1 10 100 1000

-0.05

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

1 10 100 1000

-0.05

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

0.03

1 10 100 1000

[d(p;r;w)xy ]+

[d(p;r;w)yy ]+

[d(p;r;V)xy ]+

[d(p;r;V)yy ]+

[d(p;r)xy ]+

[d(p;r)yy ]+

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obsgf

∆t+sgf

179 0.34 193 × 129 × 169 4πδ 2δ 43πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 1648 6.47 × 10−1

dns

present model Vallet (2007)

drdl (1996)

Hirt (1969)

Lumley (1978)

y+ -

y+ - y+ - y+ -

Figure 3. A priori comparison of various models (present; Hirt;12 Lumley;7 DRDL;13 Vallet14) for the rapid pressure diffusion d(p;r)ij

=

d(p;r;V)ij

+ d(p;r;w)ij

with DNS database of Gerolymos et al.8,9 for the volume d(p;r;V)ij

and the wall-echo d(p;r;w)ij

terms, in wall units [d(p)ij

]+ =

d(p)ij

νw/(ρwu4τ ), plotted against the nondimensional distance from the wall y+ = yuτ /νw .

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

1 10 100 1000-0.04

-0.02

0

0.02

0.04

0.06

0.08

0.1

1 10 100 1000

[d(p)xy ]+ [d

(p)yy ]+

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obs

∆t+s179 0.34 193 × 129 × 169 4πδ 2δ 4

3πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 2329 6.47 × 10−3

dns

present model Vallet (2007)

drdl (1996)

Hirt (1969)

Lumley (1978)

y+ - y+ -

Figure 4. A priori comparison of various models (present; Hirt;12 Lumley;7 DRDL;13 Vallet14) for the pressure diffusion d(p)ij

= d(p;V)ij

+d(r;w)ij

with DNS database of Gerolymos et al.,8,9 in wall units [d(p)ij

]+ = d(p)ij

νw/(ρwu4τ ), plotted against the nondimensional distance from the wall

y+ = yuτ /νw .

Notice that So-Aksoy-Yuan-Sommer46 (SAYS) proposed a near-wall correction φSIij to the high-Reynolds num-

ber quadratic closure for φSHij developed by Speziale-Sarkar-Gatski17 (SSG) to ensure the two-component limit.

Since the model for the inhomogeneous slow-part proposed by Shir49 is function of the Reynolds-stress tensor,

it also satisfies the realizability. A formulation where φij is modeled separately from the dissipation term εij

7 of 24

American Institute of Aeronautics and Astronautics

was developed instead

φij = − CSH1

φ ρε∗aij + CSI1

φ

ε∗

k

[ρu′′

nu′′meIneImδij − 3

2 ρu′′nu′′

i eIneIj − 32 ρu′′

nu′′j eIneIi

]

− CSI2

φ ρk

ε∂k∂xℓ

[aik

∂u′′ku′′

j

∂xℓ+ ajk

∂u′′ku′′

i∂xℓ

− 23δijamk

∂u′′ku′′

m∂xℓ

]+ CSI3

φ

[φSI2nmeIneImδij − 3

2φSI2ineIneIj − 32φSI2jneIneIi

]

− CRH

φ

(Pij − 1

3δijPmm

)+ CRI

φ

[φRHnmeIneImδij − 3

2φRHineIneIj − 32φRHjneIneIi

](31)

-0.01

0

0.01

0.02

0.03

0.04

1 10 100 1000

-0.04

-0.02

0

0.02

0.04

1 10 100 1000

-0.04

-0.035

-0.03

-0.025

-0.02

-0.015

-0.01

-0.005

0

0.005

1 10 100 1000

-0.005

0

0.005

0.01

0.015

0.02

1 10 100 1000

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

1 10 100 1000

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

1 10 100 1000

-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

1 10 100 1000

0

0.05

0.1

0.15

0.2

1 10 100 1000

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

1 10 100 1000

-0.2

-0.15

-0.1

-0.05

0

0.05

0.1

0.15

0.2

1 10 100 1000

-0.04

-0.02

0

0.02

0.04

0.06

0.08

0.1

0.12

1 10 100 1000

0

0.05

0.1

0.15

0.2

1 10 100 1000

[φ(r;w)xx ]+

[φ(r;w)xy ]+

[φ(r;w)yy ]+

[φ(r;w)zz ]+

[φ(r;V)xx ]+

[φ(r;V)xy ]+

[φ(r;V)yy ]+

[φ(r;V)zz ]+

[φ(r)xx]+

[φ(r)xy ]+

[φ(r)yy ]+

[φ(r)zz ]+

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obsgf

∆t+sgf

179 0.34 193 × 129 × 169 4πδ 2δ 43πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 1648 6.47 × 10−1

dns

present model gv (2001)

s (2004)

jejk (2007)

lrr (1975)

gl (1978)

ssg (1991)

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

Figure 5. A priori comparison of various models (present; LRR;15 GL;16 SSG;17 S;18 JEJK19) for the rapid pressure-strain redistribution

φ(r)ij

= φ(r;V)ij

+φ(r;w)ij

with DNS database of Gerolymos et al.8,9 for the volume φ(r;V)ij

and the wall-echo φ(r;w)ij

terms, in wall units [φ(r;w)ij

]+ =

φ(r;w)ij

νw/(ρwu4τ ), plotted against the nondimensional distance from the wall y+ = yuτ /νw .

8 of 24

American Institute of Aeronautics and Astronautics

-0.015

-0.01

-0.005

0

0.005

0.01

1 10 100 1000

-0.01

0

0.01

0.02

0.03

1 10 100 1000

-0.02

-0.015

-0.01

-0.005

0

0.005

1 10 100 1000

-0.02

-0.015

-0.01

-0.005

0

0.005

0.01

0.015

0.02

0.025

1 10 100 1000

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

1 10 100 1000

-0.02

-0.01

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

1 10 100 1000

0

0.05

0.1

0.15

0.2

1 10 100 1000

0

0.02

0.04

0.06

0.08

0.1

0.12

1 10 100 1000

-0.3

-0.25

-0.2

-0.15

-0.1

-0.05

0

1 10 100 1000

-0.02

-0.01

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

1 10 100 1000

-0.05

0

0.05

0.1

0.15

0.2

1 10 100 1000

0

0.02

0.04

0.06

0.08

0.1

0.12

1 10 100 1000

[φ(s;w)xx ]+

[φ(s;w)xy ]+

[φ(s;w)yy ]+

[φ(s;w)zz ]+

[φ(s;V)xx ]+

[φ(s;V)xy ]+

[φ(s;V)yy ]+

[φ(s;V)zz ]+

[φ(s)xx]+

[φ(s)xy ]+

[φ(s)yy ]+

[φ(s)zz ]+

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obsgf

∆t+sgf

179 0.34 193 × 129 × 169 4πδ 2δ 43πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 1648 6.47 × 10−1

dns: φ+ij

dns: φ+ij − [ρε

(µ)ij − 1

3ρε(µ)ℓℓ δij︸ ︷︷ ︸

ρε(µ;dev)ij

]+

gv (2001)

s (2004)

jejk (2007)

lrr (1975)

gl (1978)

ssg (1991)

present model

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

y+ -

Figure 6. A priori comparison of various models (present; LRR;15 GL;16 SSG;17 S;18 JEJK19) for the slow pressure-strain redistribution φ(s)ij =

φ(s;V)ij

+ φ(s;w)ij

with DNS database of Gerolymos et al.8,9 for the volume φ(s;V)ij

and the wall-echo φ(s;w)ij

terms, and slow redistribution

augmented by the anisotropy of dissipation φ(s)ij

− ρε(µ;dev)ij

, in wall units [φ(s)ij

− ρε(µ;dev)ij

]+ = (φ(s)ij

− ρε(µ;dev)ij

)νw/(ρwu4τ ), plotted against

the nondimensional distance from the wall y+ = yuτ /νw .

The first term of the slow-part φSH1ij (Eq. 31) corresponds to the linear Rotta model33 with the modified

dissipation-rate ε∗ = ε − 2ν(grad√

k)2 instead of the dissipation-rate ε to reach automatically the two-component limit realizability constraint whatever the coefficient CSH1

φ . The coefficient function CSH1

φ was

modified from the Launder-Shima proposal (Eq. 16) to remove the anisotropic part of the dissipation term

9 of 24

American Institute of Aeronautics and Astronautics

(Eq. 32)

CSH1

φ = 3.7AA142

[1 − e

ReT130

«2](32)

while the corresponding inhomogeneous part φSI1ij is scaled byε∗

k and the coefficient CSI1

φ was re-calibrated on

flat-plate boundary-layer flow to recover the limit values (at the wall and the outer part of the boundary-

layer) of the wall-echo coefficient developed by Launder and Shima.34

CSI1

φ =

[− 2

4.5

(CSH1

φ − 4.5

2

)] ∣∣∣∣∣

∣∣∣∣∣grad{

ℓT[1 − e−Re∗

T

30 ]

1 + 2.9√

A2

}∣∣∣∣∣

∣∣∣∣∣ (33)

A new inhomogeneous φSI2ij (Eq. 31) term, function of aij and the Reynolds-stress gradients was added to the

linear model to improve mainly the normal streamwise φxx and transverse φzz components (Fig. 7) both near

the wall and at the outlet of the boundary-layer for channel flows. The coefficient of this new term was

calibrated on DNS data and we propose

CSI2

φ = 0.002 (34)

A corresponding wall-normal-free echo-like inhomogeneous term φSI3ij (Eq. 31), which corrects the normal

vertical and the cross components φyy and φxy, was developed based on the Gibson-Launder form for the

inhomogeneous rapid-term. The use of the modified turbulent length scale gradient gradℓ∗Tinstead of the

turbulent length-scale gradient gradℓT allows to be active very close to the wall and still goes to zero withoutany damping function. The coefficient CSI3

φ mimics distance-from-the-wall effects but without using any wall-

topology-related parameter

CSI3

φ = 0.14∣∣∣∣∣∣gradℓ∗T

∣∣∣∣∣∣ ; ℓ∗

T=

k23

ε∗(35)

However, these two supplementary terms φSH1ij +φSI1ij are inhomogeneous terms because they contain gradients

of turbulence quantities and therefore added to [φ(s;w)ij ] in the a priori assessment of various models (Fig. 6).

Since Rotta,33 numerous near-wall closures for the dissipation term εij were proposed, most of which con-

tained geometric normals to the wall without satisfying the limit value of each component of the dissipation

tensor.30,31,53,54

Shima55,56 developed a wall-normal-free model, which has the correct asymptotic behavior in the viscous

sub-layer zone, by adding a viscous term ε∗ij to the general form of the Rotta closure introduced by Hanjalic-

Launder57

εij =2

3ε (1 − fw) δij + fw

ε

ku′′

i u′′j + ε∗ij (36)

ε∗ij =1

2

(d(ν)ij −

u′′i u′′

j

kd(ν)k

); fw = 1 −

√AE2 (37)

d(ν)ij =

∂xℓ

∂u′′i u′′

j

∂xℓ

); d

(ν)k =

∂xℓ

∂k

∂xℓ

)=

1

2d(ν)

mm (38)

where the flatness parameter of the dissipation anisotropy tensor E is used in the damping-function fw

proposed by Hanjalic-Jakirlic.31

Jakirlic and Hanjalic32 showed the importance of the viscous diffusion of the turbulent kinetic energy d(ν)k

through the homogeneous part εh = ε − 12d

(ν)k of the turbulent-kinetic-energy dissipation-rate which should

be used instead of ε

εij =2

3εh (1 − fw) δij + fw

εh

ku′′

i u′′j +

1

2d(ν)ij ; fw = 1 −

√AE2 (39)

However, because all of the proposed models for the dissipation tensor εij are functions of the Reynolds-

stress tensor u′′i u′′

j , the prediction of which should be improved, it is mathematically impossible to predict

independently the near-wall asymptotic behavior of the individual components of both tensors by solving

10 of 24

American Institute of Aeronautics and Astronautics

-0.4

-0.3

-0.2

-0.1

0

1 10 100 1000

-0.05

0

0.05

0.1

0.15

0.2

0.25

1 10 100 1000

-0.04

-0.02

0

0.02

0.04

0.06

0.08

0.1

1 10 100 1000

0

0.05

0.1

0.15

0.2

0.25

1 10 100 1000

φ+xx

φ+yy

φ+xy

φ+zz

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obs

∆t+s179 0.34 193 × 129 × 169 4πδ 2δ 4

3πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 2329 6.47 × 10−3

dns: φ+ij

dns: φ+ij − [ρε

(µ)ij − 1

3ρε(µ)ℓℓ δij︸ ︷︷ ︸

ρε(µ;dev)ij

]+

gv (2001)

s (2004)

jejk (2007)

lrr (1975)

gl (1978)

ssg (1991)

present model

y+ -

y+ -

y+ -

Figure 7. A priori comparison of various models (present; LRR;15 GL;16 SSG;17 GV;1 S;18 JEJK19) for the pressure-strain redistribution φij with

DNS database of Gerolymos et al.8,9 for redistribution φij and redistribution augmented by the anisotropy of dissipation φij − ρε(µ;dev)ij

, in

wall units [φij − ρε(µ;dev)ij

]+ = (φij − ρε(µ;dev)ij

)νw/(ρwu4τ ), plotted against the nondimensional distance from the wall y+ = yuτ /νw .

only one tensorial transport equation (Eq. 2). As a result, a calibration of near-wall unclosed terms which

appear in the second-moment closure to obtain a correct a posteriori prediction of the turbulent-kinetic-

energy dissipation-rate ε in a channel flow leads to a very good balance of the longitudinal normal Reynolds-

stresses component u′′u′′ but to an overestimation of the vertical normal Reynolds-stresses component v′′v′′

and the shear-stress u′′v′′ near the wall. This deficiency of algebraic closures, also observed by Jakirlic-Hanjalic,32 can have an impact on the mean-velocity prediction for channel flow. In order to preserve the

mean-velocity profile prediction in a channel flow, all of the damping functions of the second-moment closure

developed in the present study, were calibrated to obtain a correct asymptotic behavior at the wall of the v′′v′′

and u′′v′′ Reynolds-stresses components. Therefore, the benefit of using the homogeneous dissipation-rate εh

became poor and despite an erroneous wall-limit values for the dissipation-tensor component the very simple

model proposed by Rotta was used with a damping function calibrated on DNS data of Gerolymos et al.42

ρεij =2

3ρε (1 − fw) δij + fw

ε

kρu′′

i u′′j ; fw = 1 − A[1+A2]

[1 − e−

ReT10

](40)

This amounts to modeling the dissipation εij and the viscous diffusion d(µ)ij tensors as a whole. The prediction

of the modified dissipation-rate ε∗, which tends to zero at the wall, is not affected. Note that the best way(and probably the only way) to accurately predict the dissipation tensor in order to improve the Reynolds-

stress tensor, in particular in the immediate vicinity of the wall, is to solve a tensorial transport equation for

11 of 24

American Institute of Aeronautics and Astronautics

0

0.05

0.1

0.15

0.2

0.25

0.3

1 10 100 1000

0

0.005

0.01

0.015

0.02

0.025

0.03

1 10 100 1000

-0.02

-0.015

-0.01

-0.005

0

0.005

1 10 100 1000

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

1 10 100 1000

ε+xx

ε+yy

ε+xy

ε+zz

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obs

∆t+s179 0.34 193 × 129 × 169 4πδ 2δ 4

3πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 2329 6.47 × 10−3

dns present model

Jakirlic-Hanjalic (2002)

y+ -

y+ -

y+ -

y+ -

Figure 8. A priori comparison of the closure proposed by Rotta33 with a damping function (Eq. 40) developed in the present study, and

the model (Eq. 39) developed by Jakirlic and Hanjalic32 for the dissipation εij with DNS database of Gerolymos et al.8,9 in wall units

ε+ij

= εij νw/u4τ , plotted against the nondimensional distance from the wall y

+ = yuτ /νw .

εij58–60 instead of a scalar transport-equation for its trace ε or for the modified dissipation-rate ε∗.

II.D. The Wall-normal-free Reynolds-stress Model of Launder and Shima

A low-Reynolds-number Reynolds-stress model (hereafter WNF-LSS-HL RSM) was developed to analyze the

rapid-redistribution-term influence. This second-moment closure corresponds to the GV RSM1 described§II.Bwith the rapid-redistribution coefficient proposed by Launder-Shima34 (Fig. 1)

CRH

φ = 0.75√

A[1 − max(0, 1 − ReT50 )] (41)

where the coefficient CSI

φ which mimics distance from the wall effects was recalibrated to reach the correct

logarithmic-law on flat-plate (Fig. 13)

CSI

φ = 0.83 [1 − 23 (CSH

φ − 1)]

∣∣∣∣∣

∣∣∣∣∣grad{

ℓT[1 − e−Re∗

T

30 ]

1 + 2A2

}∣∣∣∣∣

∣∣∣∣∣ (42)

The WNF-LSS-HL RSM can be also considered as a wall-normal-free (WNF) version of the Reynold-stress tensor

transport-equation closure developed by Launder-Shima34 (LSS) but using Hanjalic-Launder57 (HL; Eq. 19)

model for the turbulent-diffusion term and the modified dissipation-rate transport equation of Launder-

Sharma (Eq. 8). This model is different from the WNF-LSS model29 which uses Daly-Harlow61 diffusion.

II.E. The Linear k − εModel of Launder and Sharma

The low-Reynolds-number k − ε closure of Launder and Sharma23 (hereafter LS k − ε) is used as a baselineeddy-viscosity model prediction. The substantial well-known disadvantages of the k − ε model are mainlydue to the use of the Boussinesq hypothesis

−ρu′′i u′′

ℓ∼= µT

(∂ui

∂xℓ+

∂uℓ

∂xi− 2

3

∂um

∂xmδiℓ

)− 2

3ρkδiℓ ; µT = CµµRe∗

T(43)

12 of 24

American Institute of Aeronautics and Astronautics

which is based on the analogy with the mean viscous stress tensor law, and to the impossibility to take into

account the turbulence anisotropy through the redistribution term (φmm = 0).This two-equation closure, widely used especially for industrial configurations,62 solves a transport equa-

tion for the turbulent-kinetic energy k (Eq. 44) and a transport equation for the modified dissipation-rate ε∗

(Eq. 8) where the turbulent diffusion dTε (Eq. 45) is computed by using the eddy viscosity µT (Eq. 43).

∂ρk

∂t+

∂ρkuℓ

∂xℓ=

∂xℓ

[(µ +

µTσk

)∂k

∂xℓ

]+ Pk − ρε ; σk = 1 (44)

dTε =∂

∂xℓ

(µTσǫ

∂ε∗

∂xℓ

); σε = 1.3 (45)

III. Validation

III.A. Fully developed plane channel flow

The DNS data for plane channel flow at quasi-incompressible MCL = 0.34 and supersonic MCL = 1.5 Machnumbers (Tab. 1) of Gerolymos-Senechal-Vallet9,42 were used for a posteriori assessment of the second-

moment closure developed in the present study (§II.C).The present model captures the correct shape and sign of both pressure-diffusion components (Fig. 9) in

accordance with the a priori assessment (Fig. 4), but severely underpredicts the level especially for y+ > 20 of

the shear component d(p)xy . However, contrary to previous developments for the pressure-diffusion closure,

14,20

the normal vertical component d(p)yy is quite well predicted for y+ > 10 whatever the Mach number (Fig. 9).

Notice the wall-echo part of the pressure-diffusion term is not taken into account so that, in practice, Πij =

d(p)ij + φij is modeled as a whole near the wall (y

+ ≤ 3). Therefore the asymptotic behavior of all componentsof both pressure-diffusion and redistribution terms vanish to zero at the wall.

-0.06

-0.05

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

1 10 100 -0.01

-0.005

0

0.005

0.01

0.015

0.02

0.025

0.03

1 10 100

[d(p)xy ]+ [d

(p)yy ]+

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obs

∆t+s179 0.34 193 × 129 × 169 4πδ 2δ 4

3πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 2329 6.47 × 10−3

226 1.50 241 × 137 × 217 4πδ 2δ 43πδ 11.9 0.23 19 5.6 4.4 16.5 × 10−3 1952 16.5 × 10−3

dns Mcl = 0.34

dns Mcl = 1.50

present model Mcl = 0.34

present model Mcl = 1.50

y+ - y+ -

Figure 9. A posteriori comparison of present model for the pressure diffusion tensor d(p)ijwith DNS data of Gerolymos-Senechal-Vallet,9,42

in wall units [d(p)ij

]+ = d(p)ij

νw/(ρwu4τ ), plotted against the nondimensional distance from the wall y+ = yuτ /νw .

Figure 10 presents the a posteriori comparison of turbulence closures for the pressure-strain correlation

φij . The present model is in good agreement with DNS database in the outer regions (y+ ≥ 30) for all

components (Fig. 10), but is less accurate near the wall (y+ ≤ 10). The shear φxy and the normal vertical φyy

components return zero value at the wall contrary to DNS data, which do not necessarily affect the global

RSM performance since pressure-diffusion d(p)ij and redistribution φij terms cancel one another in this region.

Although the present model was developed a priori to improve the normal streamwise φxx and the transverse

φzz components (Fig. 7), the asymptotic behavior of φzz near the wall is still strongly underestimated (Fig. 10).

On the other hand, the present model predicts satisfactorily the normal streamwise φxx except for the

lower Reynolds-number, which corresponds to the subsonic Mach number, in the near-wall region (y+ ≥ 10).Because of the overestimation of the shear component of the dissipation tensor (Fig. 11) in the buffer-layer

zone at y+ ∼= 10 − 30 by about 0.3, the maximum level of the shear component φxy is underestimated by the

same value (Fig. 10).

The a posteriori assessment of the dissipation-tensor closure is presented figure 11. The normal vertical

component of the dissipation tensor εyy presents an overall good agreement with DNS data (Fig. 11) contrary

13 of 24

American Institute of Aeronautics and Astronautics

-0.06

-0.05

-0.04

-0.03

-0.02

-0.01

0

0.01

1 10 100

-0.03

-0.02

-0.01

0

0.01

0.02

0.03

1 10 100

-0.02

0

0.02

0.04

0.06

0.08

0.1

1 10 100

0

0.01

0.02

0.03

0.04

1 10 100

φ+xx

φ+yy

φ+xy

φ+zz

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obs

∆t+s179 0.34 193 × 129 × 169 4πδ 2δ 4

3πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 2329 6.47 × 10−3

226 1.50 241 × 137 × 217 4πδ 2δ 43πδ 11.9 0.23 19 5.6 4.4 16.5 × 10−3 1952 16.5 × 10−3

dns Mcl = 0.34

dns Mcl = 1.50

present model Mcl = 0.34

present model Mcl = 1.50

y+ -

y+ -

y+ -

Figure 10. A posteriori comparison of present model for the pressure-strain redistribution φij with DNS database of Gerolymos-Senechal-

Vallet,9,42 in wall units φ+ij

= φij νw/(ρwu4τ ), plotted against the nondimensional distance from the wall y+ = yuτ /νw .

to the asymptotic behavior at the wall of the two other normal components εxx and εzz. However, the profile

of all components far away from the wall y+ ≥ 50 , as well as the location of the maximum peak in thebuffer-layer region y+ ∼= 10− 30, are quite well predicted with a correct effect of the Mach-number variation.Nevertheless, it is important to notice the huge difference between the a priori (Fig. 8) and the a posteriori

(Fig. 11) assessments especially near the wall. Indeed, the wall-asymptotic behavior of the Reynolds stresses

corresponds to the very important anisotropy of the dissipation tensor εij in the viscous-layer zone. That

means that only one component wall-asymptotic behavior can be satisfied if using a linear algebraic closure

instead of a tensorial transport-equation for εij . In the present model, the prediction of the normal vertical

εyy and the shear εxy components were prioritized to the detriment of two other normal components (Fig. 10)

to obtain the best mean-velocity profile estimation (Fig. 12). As a consequence, an underestimation of the

normal streamwise εxx and tranversal εzz components is observed near the wall.

Concerning the Reynolds-stress tensor (Fig. 12), the effect of Mach-number variation is well predicted

by the present model for y+ > 30, but the wall-asymptotic behavior and as a result, the peak level, are

underestimated for the normal streamwise u′′u′′ and transverse w′′w′′ components. However, the maximum

peak of these two components predicted by the present second-moment closure, is in better agreement with

the DNS data than the RSM GV. despite a slight underestimation of the maximum peak of the shear u′′v′

and the normal vertical v′′v′′ components, their asymptotic behavior near the wall is correctly estimated.Thereby, both second-moment closures give a good agreement with the two DNS databases of the mean-

velocity profiles (Fig. 12). However, for the lowest Reynolds-number (Reτw= 180), which corresponds to the

lowest Mach-numberM = 0.34, the location of the logarithmic-law is too high (Fig. 12).

14 of 24

American Institute of Aeronautics and Astronautics

0

0.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

1 10 100

0

0.005

0.01

0.015

0.02

0.025

0.03

1 10 100

-0.04

-0.03

-0.02

-0.01

0

0.01

1 10 100

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

1 10 100

ε+xx

ε+yy

ε+xy

ε+zz

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obs

∆t+s179 0.34 193 × 129 × 169 4πδ 2δ 4

3πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 2329 6.47 × 10−3

226 1.50 241 × 137 × 217 4πδ 2δ 43πδ 11.9 0.23 19 5.6 4.4 16.5 × 10−3 1952 16.5 × 10−3

dns Mcl = 0.34

dns Mcl = 1.50

present model Mcl = 0.34

present model Mcl = 1.50

y+ -

y+ -

y+ -

y+ -

Figure 11. A posteriori comparison of present model (Eq. 40) for the dissipation εij with DNS database of Gerolymos-Senechal-Vallet?,42 in

wall units ε+ij

= εij νw/u4τ , plotted against the nondimensional distance from the wall y

+ = yuτ /νw .

0

2

4

6

8

10

12

1 10 100

0

0.2

0.4

0.6

0.8

1

1 10 100

-1

-0.8

-0.6

-0.4

-0.2

0

1 10 100

0

0.2

0.4

0.6

0.8

1

1.2

1.4

1.6

1 10 100 0

5

10

15

20

25

1 10 100

u′′u′′+

v′′v′′+

u′′v′′+

w′′w′′+ u+

ReτwMcl Nx × Ny × Nz Lx Ly Lz ∆x+ ∆y+

w Ny+≤10 ∆y+cl

∆z+ ∆t+ t+obs

∆t+s179 0.34 193 × 129 × 169 4πδ 2δ 4

3πδ 11.7 0.22 20 4.9 4.4 6.47 × 10−3 2329 6.47 × 10−3

226 1.50 241 × 137 × 217 4πδ 2δ 43πδ 11.9 0.23 19 5.6 4.4 16.5 × 10−3 1952 16.5 × 10−3

dns Mcl = 0.34

dns Mcl = 1.50

present model Mcl = 0.34

present model Mcl = 1.50

gv rsm Mcl = 0.34

gv rsm Mcl = 1.50

y+ -

y+ -

y+ -

y+ - y+ -

Figure 12. A posteriori comparison of present model and the GV RSM1 for the Reynolds stresses u′′

iu′′

jand mean streamwise velocity with

DNS database of Gerolymos-Senechal-Vallet9,42 in wall units u′′

iu′′

j

+= u′′

iu′′

j/u2

τ , u+ = u/uτ , plotted against the nondimensional distance

from the wall y+ = yuτ /νw .

15 of 24

American Institute of Aeronautics and Astronautics

III.B. Flate-Plate Boundary-layer Flow

The first 2-D configuration assessed by the second-moment closure developed in the present study, is the

incompressible flat-plate boundary-layer flow measured by Klebanoff.63 The computational domain Lx ×Ly = 2m × 0.3m, where Lx is the flat-plate length in the streamwise direction x and Ly is the height in the

vertical direction y, is discretised with a fine grid Ni × Nj = 261 × 801. In the y-direction normal to the wall,Njs

= 80%Nj grid points were stretched geometrically with a ratio rj = 1.007 to obtain a first node at the walllocated at y+

w ≃ 0.3.Reservoir inlet condition (Tti

= 288 K, pti= 101325 Pa) with a turbulence-intensity Tui

= 0.3% and aturbulence length-scale ℓTi

= 0.2 m, were imposed while the inflow boundary-layer thickness and the outflowpressure were adjusted to obtain a close fit to the experimental boundary-layer thickness δ = 0.99ue ≃67.2 mm and free-stream mean-velocity ue = 15.24m/s respectively.Computations were compared against measurements of Klebanoff63 at a streamwise location x which

corresponds to the experimental Reynolds number Reθ = 7800 (Fig. 13). The three Reynolds-stress modelsgive similar results with an overall good agreement with experimental data except for the maximum value

of normal streamwise u′u′ and transverse w′w′ components (Fig. 13). The underestimation of these two

components is probably the consequence of the underestimation of the asymptotic behavior at the wall of the

dissipation-tensor components εxx and εzz observed in the channel flow (Fig. 11). Indeed, as a linear k − εmodel, which solves the transport-equation for the trace of the Reynolds-stress tensor k, is not able to predictthe Reynolds-stress anisotropy (Fig. 13), the transport-equation for the trace of the dissipation tensor ε hasnot the possibility to predict with accuracy the anisotropy of the dissipation tensor εij especially because the

algebraic εij-model is function of the Reynolds stresses u′iu

′j . On the contrary, the excellent agreement with

measurements of the normal vertical component v′v′ and the shear-stress u′v′ yields a perfect prediction ofthe mean-velocity u profile (Fig. 13) where the constants of the logarithmic-law were determined by So etal.64 from Klebanoff measurements.63

√u′u′

ue

√v′v′ue

√w′w′

ue

−2u′v′

u2e

√u′u′

ue

√v′v′ue

√w′w′

ue

−2u′v′

u2e

0

0.02

0.04

0.06

0.08

0.1

0.12

0 0.2 0.4 0.6 0.8 1 1.2 0

0.01

0.02

0.03

0.04

0.05

0.06

0 0.2 0.4 0.6 0.8 1 1.2 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0 0.2 0.4 0.6 0.8 1 1.2 0

0.001

0.002

0.003

0.004

0 0.2 0.4 0.6 0.8 1 1.2

0

5

10

15

20

25

30

1 10 100 1000 10000

0

0.02

0.04

0.06

0.08

0.1

0.12

0.0001 0.001 0.01 0.1 1 0

0.01

0.02

0.03

0.04

0.05

0.06

0.0001 0.001 0.01 0.1 1 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.0001 0.001 0.01 0.1 1 0

0.001

0.002

0.003

0.004

0.0001 0.001 0.01 0.1 1

y/δ - y/δ - y/δ - y/δ -

u+

6

y+ -

10.43 ln y+ + 5.34

6

y/δ - y/δ - y/δ - y/δ -

experiment

ls k − ε (1974)

wnf–lss–hl rsm

gv rsm (2001)

present model

Figure 13. Comparison of measured Reynolds-stresses and mean-velocity profiles along the distance from the wall, and grid-converged

computations using three RSMs and a linear eddy-viscosity closure for a flat-plate boundary-layer flow63

TheWNF-LSS-HL RSM, which differs from the two others RSMs only by the homogeneous rapid-redistribution

coefficient, predicts the best results for the Reynolds-stresses profiles of u′u′ and u′v′ for y/δ > 0.4, while the

16 of 24

American Institute of Aeronautics and Astronautics

present model and the GV RSM are better at y/δ ∼= 0.4 for the normal transverse w′w′ component. Notice the

outer-region prediction of the mean-velocity profile is also slightly different due to the CRH

φ coefficient. Never-

theless, it is quite difficult to distinguish the impact of different models for unclosed terms and the flat-plate

boundary-layer flow is therefore not very useful for comparing statistical turbulence closure. However, the

ability of a turbulence model to accurately predict this flow is a necessary condition (but unfortunately not

sufficient) to obtain good results in more complex flows.

As expected, the linear eddy-viscositymodel of Launder-Sharma23 fails to predict the anisotropy of Reynolds-

stress tensor and gives a maximum peak of about u′u′ = v′v′ = w′w′ ∼= 0.6. The shear-stress is overestimatedand the logarithmic-law of the mean-velocity is too low.

III.C. Asymmetric Plane Diffuser

The Reynolds-stress model developed was also evaluated by comparison with experimental data for sepa-

rating and reattaching flow due to adverse pressure gradient in a diffuser.65,66 The two-dimensional plane

asymmetric diffuser has an opening angle of 10 degrees with a diffusing-section length of 21H and an expan-sion ratio of 4.7H , whereH is the inlet channel height. The Reynolds number is ReB = uBHν−1=20000, where

uB is the inlet bulk velocity, H is the inlet height channel and ν is the kinematic viscosity.66 The asymmetricdiffuser studied experimentally by Obi et al.65 and Buice and Eaton66 was previously assessed by numerous

authors.67–73 In particular, Apsley and Leschziner69 compared various turbulence closures ranging from lin-

ear eddy-viscosity models to second-moment closures, while Kaltenbach et al.68 used large-eddy simulation

(LES) to make a thorough study of flow physics and of validity of experimental data, very useful for future

turbulence closure assessments and developments.

Simulations were performed on a 481 000 points grid (Ni×Nj=Nx×Ny=801×601), which is grid-convergedfor all of the turbulence closures assessed, discretizing the computational domain extended from x/H =−6 to x/H = 73.71 where measurements from Buice and Eaton66 are available. The grid was stretchedgeometrically in the wall-normal direction (with a ratio rj = 1.01 up to diffuser mid-height which correspondsto Nj/2 points) to obtain a first node at the wall located approximately at y+

w = 0.28A fully-developed channel flow using the same turbulence closure from a preliminary 1-D computation74

was imposed as inlet condition at x/H = −6, and the outflow pressure was determined from experimentaldata of Buice and Eaton66 Cp = 7.36 10−1 at x/H = 73.71. Dimensional profiles (Figs 14-16) uses the valuesfrom Buice and Eaton’s experiments and configuration (H = 0.015m, uCL = 1.14uBm/s, uB = 20m/s).Comparison of the measured and computed mean-velocity u profile (Fig. 14) indicates that the present

RSM maintains the ability of the GV RSM1 to correctly predict separation6 while it improves the reattachment

behavior. Nevertheless, at the beginning of the diffuser (x = 2.6H ; 6H), the maximum peak of the mean-velocity is slightly overestimated by the present and the GV RSMs.

The GV RSM slightly overestimates the mean-velocity profile at x = 13.6H and x = 17H and fails tocorrectly predict relaxation downstream of x = 40H (relaxation is too slow). On the contrary, the presentmodel, which has the same rapid-redistribution term but with an additional slow part and furthermore

includes a closure for the pressure-diffusion tensor, slightly underestimates the mean-velocity u after x =20H but is close to experimental data near the bottom wall after x = 40H . The discrepancy between allcomputations and measurements at x = 53H for y − yw ≥ 0.25m is obviously a consequence of measurementswhich were not corrected to ensure mass flow conservation. Indeed, since computations, which conserve mass

flow, are in good agreement with experimental data for y−yw ≤ 0.25, the discrepancy between measurementsand computations for y − yw ≥ 0.25 is clearly due to measurement inaccuracies.Results obtained by the k − ε model of Launder-Sharma23 are typical of those predicted by linear eddy-

viscosity models69 which fail to predict separation. As a result, the blockage of the separation-region is

underestimated inducing too low values for the maximum peak of the mean-velocity profile u. The WNF-LSS-HL RSM, which differs from the GV RSM only by the rapid redistribution coefficient (Fig. 1), gives a better

than k − ε agreement with experiments but also fails to predict flow detachment on the opening wall.The prediction of Reynolds-stresses profiles is quite challenging since most of turbulence closures67,69,75

have a tendency to overpredict Reynolds-stresses in the first part of the diffuser and to underpredict them in

the second part, despite a good prediction of the mean streamwise velocity.

Figures 15-16 shows a comparison of measured and computed Reynolds-stresses profiles. Because, the

Reynolds-stresses v′v′ and u′v′ were not available in the separated flow region in the Buice and Eaton’sexperimental database, measurements of Obi et al.65 were added (Figs 15-16), with appropriate dimension-

alisation and at approximately the same stations (location variation ∆x/H ≤ ±0.5).The normal streamwise Reynolds-stress (Fig. 15) presents a double-peak structure rather well predicted

by the three second-moment closures. The present model improve the prediction of the GV RSM especially

17 of 24

American Institute of Aeronautics and Astronautics

x=−6 H

x=2.6 H x=6 Hx=13.6 H

x=17 H

x=20 H x=24 H

x=30.5 H x=40 Hx=53 H

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

-5 0 5 10 15 20 25 0

0.005

0.01

0.015

0.02

0.025

-5 0 5 10 15 20 0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

-5 0 5 10 15 20 0

0.01

0.02

0.03

0.04

0.05

-5 0 5 10 15 0

0.01

0.02

0.03

0.04

0.05

0.06

-5 0 5 10 15

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-5 0 5 10 15 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-5 0 5 10 15 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-4 -2 0 2 4 6 8 10 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0 1 2 3 4 5 6 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0 1 2 3 4 5 6

y−

y w(m

)

6

u (m s−1) -

y−

y w(m

)

6

u (m s−1) -

����������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

@@

@@I 6

�������* 6 6

��

��� ? ? ?

AAAAAU

Buice-Eaton (2000)

ls k − ε (1974)

wnf–lss–hl rsm

gv rsm (2001)

present model

Figure 14. Comparison of measured mean-velocity profiles with computations using three RSMs and a linear eddy-viscosity closure for the

asymmetric plane diffuser flow66 (iso-contours computed with the present RSM)

near the wall, for instance at x = 6H and x = 19.5H (Fig. 15). Except for the station x = 6H , the maxi-mum value of the normal streamwise Reynolds-stress component u′u′ given by the present RSM, is in better

agreement with experimental data but is overestimated upstream of x = 17H . Downstream of x = 17H , thepresent turbulence closure is very close to experiments, especially to the measurements of Obi et al.65 The

normal vertical Reynolds-stress v′v′ profile presents a similar shape of the normal streamwise componentu′u′ with a lower maximum peak, which is quite well predicted by the present RSM contrary to the k − εmodel. However, an overestimation of the v′v′-profile is observed at stations upstream of x = 19.5H . Noticealso how the present and GV RSMs improve the predicton of the locations of the peaks in the u′u′ and v′v′

y-wise distributions (Fig. 15).The linear k − ε model23 completely fails to reproduce the turbulence-anisotropy of the flow predicting

everywhere an identical maximum value for the two normal components u′u′ = v′v′ ≃ 2.5 m2/s2 (Fig. 15).

The shear-stress u′v′ profile is presented figures 16. Again, the present model and the GV RSM are veryclose and give the overall best agreement with experimental data, especially in the second part of the diffuser

at x ≥ 17H . However, the present RSM is slightly better near the wall at the station x = 23H .The eddy-viscosity closure overestimates the shear-stress profile for x ≤ 6H and strongly underestimates

it downstream (Fig. 16). The same prediction is obtained with the WNF-LSS-HL RSM. Results obtained by

the RSM developed in the present study are overall in good agreement with experimental data, in particular

near the wall and with a better level of the maximum peak of Reynolds-stresses for stations downstream of

x ≥ 17H .

18 of 24

American Institute of Aeronautics and Astronautics

x=−6 H

x=2.6 H x=6 H

13.6 Hx=17 H

19.5 H

23.7 H

30.5 H

x=40 H x=53 H

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

-2 0 2 4 6 8 10 0

0.005

0.01

0.015

0.02

0.025

-2 0 2 4 6 8 10 0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

-2 0 2 4 6 8 10 0

0.01

0.02

0.03

0.04

0.05

-2 0 2 4 6 8 10 12 14 0

0.01

0.02

0.03

0.04

0.05

0.06

-2 0 2 4 6 8 10 12

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 0 2 4 6 8 10 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 0 2 4 6 8 10 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 0 2 4 6 8 10 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 0 2 4 6 8 10 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 0 2 4 6 8 10

y−

y w(m

)

6

u′u′ (m2 s−2) -

y−

y w(m

)

6

u′u′ (m2 s−2) -

����������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

@@

@@I 6

�������* 6 6

��

��� ? ? ?

AAAAAU

Obi et al. (1993)

Buice-Eaton (2000)

ls k − ε (1974)

wnf–lss–hl rsm

gv rsm (2001)

present model

-6 H

2.6 H x=6 H

13 H17 H

19.5 H23 H

30 H 40 H

53 H

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

-2 -1 0 1 2 3 0

0.005

0.01

0.015

0.02

0.025

-2 -1 0 1 2 3 4 5 0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

-2 -1 0 1 2 3 4 5 6 0

0.01

0.02

0.03

0.04

0.05

-2 0 2 4 6 8 0

0.01

0.02

0.03

0.04

0.05

0.06

-2 0 2 4 6 8

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 -1 0 1 2 3 4 5 6 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 -1 0 1 2 3 4 5 6 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 -1 0 1 2 3 4 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-1 -0.5 0 0.5 1

y−

y w(m

)

6

v′v′ (m2 s−2) -

y−

y w(m

)

6

v′v′ (m2 s−2) -

����������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

@@

@@I 6

�������* 6 6

��

��� ? ? ?

AAAAAU

Obi et al. (1993)

Buice-Eaton (2000)

ls k − ε (1974)

wnf–lss–hl rsm

gv rsm (2001)

present model

Figure 15. Comparison of measured streamwise and vertical normal Reynolds-stress profiles with computations using three RSMs and a

linear eddy-viscosity closure for the asymmetric plane diffuser flow65,66 (iso-contours computed with the present RSM)

19 of 24

American Institute of Aeronautics and Astronautics

x=−6 Hx=2.6 H

x=6 H x=13 Hx=17 H

x=19.5 H x=23 H x=30 H40 H 53 H

0

0.002

0.004

0.006

0.008

0.01

0.012

0.014

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0

0.005

0.01

0.015

0.02

0.025

-3 -2 -1 0 1 2 0

0.005

0.01

0.015

0.02

0.025

0.03

0.035

-4 -3 -2 -1 0 1 2 3 4 0

0.01

0.02

0.03

0.04

0.05

-6 -4 -2 0 2 4 6 0

0.01

0.02

0.03

0.04

0.05

0.06

-6 -4 -2 0 2 4 6

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-4 -3 -2 -1 0 1 2 3 4 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-3 -2 -1 0 1 2 3 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-3 -2 -1 0 1 2 3 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2 0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

-1 -0.5 0 0.5 1

y−

y w(m

)

6

u′v′ (m2 s−2) -

y−

y w(m

)

6

u′v′ (m2 s−2) -

����������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

@@

@@I 6

�������* 6 6

��

��� ? ? ?

AAAAAU

Obi et al. (1993)

Buice-Eaton (2000)

ls k − ε (1974)

wnf–lss–hl rsm

gv rsm (2001)

present model

Figure 16. Comparison of measured shear Reynolds-stress profiles with computations using three RSMs and a linear eddy-viscosity closure

for the asymmetric plane diffuser flow65,66 (iso-contours computed with the present RSM)

III.D. Shock-Wave/Boundary-Layer Interaction Flow

Finally, the present turbulence closure was assessed by comparison with available experimental data for an

incident oblique-shock-wave/turbulent-boundary-layer interaction on a flat-plate studied experimentally by

Schulein,76 with a Mach-numberMSW=5, a flow deflection induced by the shock-wave∆ϑSW=14 deg and a unitReynolds number Re∞ = 37 × 106 m−1 . The computational domain Lx×Ly = 0.5×0.25m with xin = −0.2 m(xin corresponds to the beginning of the domain and x = 0 is the location of shock-wave impact on theflat-plate) is discretized with a very fine mesh because of the quite small recirculation zone and the high

Mach-number. The computational grid Ni×Nj = 801×801 points is stretched by using geometric progressionratio rj = 1.0111 on Njs

= 0.8Nj points in the normal-from-the-wall direction y to obtain a nondimensionaldistance of the first-grid-node away from the wall y+

w ≤ 0.1 at the beginning of the interaction.Experimental inlet total conditions (Tti

= 408.2 K, pti= 2136066 Pa) with a turbulence-intensity Tui

= 1%and a turbulence length-scale ℓTi = 0.05m were applied at the inlet of the computational domain xin − 0.2m.The inflow boundary-layer thickness δi was adjusted to obtain a close fit to the measured integral boundary-

layer parameters (Tab. 2) especially the kinematic boundary-layer momentum-thickness δ2k= 4.73×10−4.

The wall is heated at a temperature Tw = 300K and a supersonic inflow is applied at the upper boundaryLy = 0.25m.Linear two-equation closures are particularly inappropriate for shock-wave/boundary-layer interaction

flow at high Mach number77 contrary to Reynolds-stress models. Indeed, comparison of wall-pressures and

skin-friction distributions (Fig. 17) indicates that the Launder-Sharma23 k − ε computations gives the worstprediction of upstream influence. The WNF-LSS-HL RSM performs better than the eddy-viscosity model and

predicts a good shape of the wall-pressure profile but the recirculation-zone is still underestimated. The

present RSM and the GV RSM improve the prediction of upstream influence which can be directly attributed

to the optimized rapid-redistribution coefficient (Fig. 1). However, the present RSM and the GV RSM underes-

timate the height of recirculation-zone which is probably a consequence of compressibility effects neglected

20 of 24

American Institute of Aeronautics and Astronautics

in the turbulence closure.78

Considering the skin-friction distribution (Fig. 17), it is noticeable that all turbulence closures fail to

predict the correct shape of wall-friction in the reattachment flow region (x ≥ 0m). Furthermore, the presentReynolds-stress model overestimates the maximum peak of the skin-friction coefficient in the relaxation zone

while the k − ε model strongly underestimates the level of Cf after x = 0m.

Table 2. Measured and computed integral boundary-layer parameters

experiment present model GV RSM1 WNF-LSS-HL RSM LS23 k − ε

δi × 103 m – 4.3 4.3 4.3 4.8

δ × 103 m 4.102 4.867 4.868 5.138 5.53

δ1 × 103 m 1.786 2.357 2.356 2.40 2.416

δ2 × 104 m 1.78 2.157 2.152 2.16 2.274

H12 10.027 10.92 10.95 11.13 10.62

δk1 × 104 m 6.26 6.61 6.57 6.558 6.464

δk2 × 104 m 4.73 4.74 4.72 4.728 4.724

Hk12 1.325 1.394 1.392 1.387 1.368

δi = inflow boundary-layer thickness; δ = boundary-layer thickness at the beginning of the interaction; δ1 =R δ0

(1 − ρu/ρu∞)dy =

boundary-layer displacement-thickness at the beginning of the interaction; δ2 =R δ

0ρu/ρu∞(1−u/u∞)dy = boundary-layer momentum-

thickness at the beginning of the interaction; H12 = δ1δ−1

2= boundary-layer shape-factor at the beginning of the interaction; δk1

=R δ0

(1 − u/u∞dy; δk2=

R δ0

u/u∞(1 − u/u∞)dy; Hk12= δk1

δ−1

k2= kinematic boundary-layer shape-factor

IV. Conclusions and Discussion

In the present work, we developed a Reynolds-stress model using recently obtained DNS data splitting

the pressure fluctuation p′, and as a consequence all correlations containing p′ not only into slow and rapidterms, but also into volume (quasi-homogeneous) and wall (strongly inhomogeneous) terms. The new closure

models separately the anisotropy of dissipation εij − 23εδij and the pressure correlations φij + d

(p)ij = Πij . The

tensorial representations for φij and d(p)ij include supplementary inhomogeneous terms extending previously

used tensorial representations with coefficients calibrated against DNS data. The proposed closure performs

quite satisfactorily in a complex separating and reattaching flow in a subsonic asymmetric diffuser and

for an oblique shock-wave/boundary-layer interaction on flat-plate at Mach number MSW=5. Furthermore,

improvement is possible in the viscous wall region (y+ < 10), but the authors believe that such improvementrequire, differential transport modeling of the dissipation tensor.

Acknowledgments

The computations were performed using HPC ressources from GENCI-IDRIS (Grant 2010-020218). The

authors are listed alphabetically.

References

1Gerolymos, G. A. and Vallet, I., “Wall-Normal-Free Near-Wall Reynolds-Stress Closure for 3-D Compressible Separated Flows,”

AIAA J., Vol. 39, No. 10, Oct. 2001, pp. 1833–1842.2Gerolymos, G. A. and Vallet, I., “Wall-Normal-Free Reynolds-Stress Model for Rotating Flows Applied to Turbomachinery,” AIAA

J., Vol. 40, No. 2, Feb. 2002, pp. 199–208.3Gerolymos, G. A., Neubauer, J., Sharma, V. C., and Vallet, I., “Improved Prediction of Turbomachinery Flows using Near-Wall

Reynolds-Stress Model,” ASME J. Turbom., Vol. 124, No. 1, Jan. 2002, pp. 86–99.4Gerolymos, G. A., Sauret, E., and Vallet, I., “Oblique-Shock-Wave/Boundary-Layer Interaction using Near-Wall Reynolds-Stress

Models,” AIAA J., Vol. 42, No. 6, June 2004, pp. 1089–1100.5Vallet, I., “Reynolds-Stress Modelling ofM = 2.25 Shock-Wave/Turbulent-Boundary-Layer Interaction,” Int. J. Num. Meth. Fluids,

Vol. 56, No. 5, Feb. 2008, pp. 525–555.6Gerolymos, G. A., Joly, S., Mallet, M., and Vallet, I., “Reynolds-Stress Model Flow Prediction in Aircraft-Engine Intake Double-S-

Shaped Duct,” J. Aircraft, Vol. 47, No. 4, July 2010, pp. 1368–1381.7Lumley, J. L., “Computational Modeling of Turbulent Flows,” Adv. Appl. Mech., Vol. 18, 1978, pp. 123–176.8Gerolymos, G. A., Lo, C., Senechal, D., Vallet, I., and Younis, B. A., “Wall Effects on Pressure Fluctuations in Quasi-Incompressible

Turbulent Plane Channel Flow,” Proceedings 7. ERCOFTAC International Symposium on Engineering Turbulence Modelling and Mea-

surements (ETMM7), 4–6 jun 2008, Limassol, [CYP], edited by M. A. Leschziner, S. Kassinos, J. P. Bonnet, D. Laurence, W. Rodi,

21 of 24

American Institute of Aeronautics and Astronautics

-x (m)

6

pw × 103 Pa

-inviscid theory

0

10

20

30

40

50

60

-0.1 -0.05 0 0.05 0.1��������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

M

��������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������������

-x (m)

6

Cf∞× 103

-4

-2

0

2

4

6

8

10

12

-0.1 -0.05 0 0.05 0.1

Schulein (2006)

ls k − ε (1974)

wnf–lss–hl rsm

gv rsm (2001)

present model

∆ϑSW=14 deg

Figure 17. Comparison of grid-converged computations with measurements76 of wall-pressure and skin-friction x-wise distributions, for

Schulein76 incident-shock-wave interaction (MSW=5, Re∞=37106/m, ∆ϑSW=14 deg) using three RSMs and a linear eddy-viscosity closure

(iso-Mach number computed with the present RSM).

A. Tomboulides, and C. Tropea, Vol. 3, June 2008, pp. 634–639.9Senechal, D., DNS des Ecoulements Turbulents Compressibles: Fluctuations de Pression, de Masse Volumique et de Temperature,

Doctorat, Universite Pierre-et-Marie-Curie, Paris [FRA], Jan. 2009.10Gerolymos, G. A., Lo, C., Senechal, D., and Vallet, I., “Fluctuating pressure correlations in wall turbulence,” ArXiv, Vol. 2011,

2011, pp. 1101.1876v1(1–62), (http://arxiv.org/pdf/1101.1876v1; submitted for publication).11Kim, J., “On the Structure of Pressure Fluctuations in Simulated Turbulent Channel-Flow,” J. Fluid Mech., Vol. 205, 1989,

pp. 421–451.12Hirt, C. W., “Generalized Turbulence Transport Equations,” Int Sem. of Int. Center for Heat and Mass Transfer, Herceg Novi,

Yugoslavia, 1969.13Demuren, A. O., Rogers, M. M., Durbin, P., and Lele, S. K., “On Modelling Pressure Diffusion in non-Homogeneous Shear Flows,”

Studying Turbulence Using Numerical Simulation Databases - VI, Proceeding of the CTR Summer Program 1996, Center for Turbulence

Reasearch, NASA Ames/Stanford University [USA], Dec. 1996.14Vallet, I., “Reynolds-Stress Modelling of 3-D Secondary Flows with Emphasis on Turbulent Diffusion Closure,” ASME J. Appl.

Mech., Vol. 74, No. 6, Nov. 2007, pp. 1142–1156.15Launder, B. E., Reece, G. J., and Rodi, W., “Progress in the Development of a Reynolds-Stress Turbulence Closure,” J. Fluid Mech.,

Vol. 68, 1975, pp. 537–566.16Gibson, M. M. and Launder, B. E., “Ground Effects on Pressure Fluctuations in the Atmospheric Boundary-Layer,” J. Fluid Mech.,

Vol. 86, 1978, pp. 491–511.17Speziale, C. G., Sarkar, S., and Gatski, T. B., “Modelling the Pressure-Strain Correlation of Turbulence: An Invariant Dynamical

Systems Approach,” J. Fluid Mech., Vol. 227, 1991, pp. 245–272.18Suga, K., “Modeling the Rapid Part of the Pressure-Diffusion Process in the Reynolds Stress Transport Equation,” ASME J. Fluids

Eng., Vol. 126, July 2004, pp. 634–641.19Jakirlic, S., Eisfeld, B., Jester-Zurker, R., and Kroll, N., “Near-wall Reynolds-stress model calculations of transonic flow configu-

rations relevant to aircraft aerodynamics,” Int J. Heat Fluid Flow, Vol. 28, 2007, pp. 602–615.20Sauret, E. and Vallet, I., “Near-Wall Turbulent Pressure Diffusion Modelling and Influence in 3-D Secondary Flows,” ASME J.

Fluids Eng., Vol. 129, No. 5, May 2007, pp. 634–642.21Gerolymos, G. A. and Vallet, I., “Mean-Flow-Multigrid for Implicit Reynolds-Stress-Model Computations,” AIAA J., Vol. 43, No. 9,

Sept. 2005, pp. 1887–1898.22Gerolymos, G. A. and Vallet, I., “Robust Implicit Multigrid Reynolds-Stress-Model Computation of 3-D Turbomachinery Flows,”

ASME J. Fluids Eng., Vol. 129, No. 9, Sept. 2007, pp. 1212–1227.

22 of 24

American Institute of Aeronautics and Astronautics

23Launder, B. E. and Sharma, B. I., “Application of the Energy Dissipation Model of Turbulence to the Calculation of Flows near a

Spinning Disk,” Lett. Heat Mass Transfer, Vol. 1, 1974, pp. 131–138.24Hanjalic, K. and Launder, B. E., “A Reynolds Stress Model of Turbulence and its Application to Thin Shear Flows,” J. Fluid Mech.,

Vol. 52, 1972, pp. 609–638.25Launder, B. E., Tselepidakis, D. P., and Younis, B. A., “A Second-Moment Closure Study of Rotating Channel Flow,” J. Fluid Mech.,

Vol. 183, 1987, pp. 63–75.26Tucker, P. G., “Differential-Equation-BasedWall-Distance Computation in DES and RANS,” J. Comp. Phys., Vol. 190, 2003, pp. 229–

248.27Tucker, P. G., Rumsey, C. L., Spalart, P. R., Bartels, R. E., and Biedron, R. T., “Computations of Wall-Distances Based on Differen-

tial Equations,” AIAA J., Vol. 43, 2005, pp. 539–549.28Craft, T. J. and Launder, B., “A Reynolds-Stress Model Designed for Complex Geometries,” Int. J. Heat Fluid Flow, Vol. 17, 1996,

pp. 245–254.29Gerolymos, G. A., Sauret, E., and Vallet, I., “Contribution to the Single-Point-Closure Reynolds-Stress Modelling of Inhomoge-

neous Flow,” Theor. Comp. Fluid Dyn., Vol. 17, No. 5–6, Sept. 2004, pp. 407–431.30Lai, Y. G. and So, R. M. C., “On Near-Wall Turbulent Flow Modelling,” J. Fluid Mech., Vol. 221, 1990, pp. 641–673.31Hanjalic, K. and Jakirlic, S., “A Model of Stress Dissipation in Second-Moment Closures,” Appl. Sci. Res., Vol. 51, 1993, pp. 513–

518.32Jakirlic, S. and Hanjalic, K., “A New Approach to Modelling Near-Wall Turbulence Energy and Stress Dissipation,” J. Fluid Mech.,

Vol. 459, 2002, pp. 139–166.33Rotta, J., “Statistische Theorie nichthomogener Turbulenz — 1. Mitteilung,” Z. Phys., Vol. 129, 1951, pp. 547–572.34Launder, B. E. and Shima, N., “2-Moment Closure for the Near-Wall Sublayer: Development and Application,” AIAA J., Vol. 27,

No. 10, Oct. 1989, pp. 1319–1325.35Naot, D., Shavit, A., and Wolfshtein, M., “Interactions between Components of the Turbulent Velocity Correlation Tensor due to

Pressure Fluctuations,” Israel J. Techn., Vol. 8, No. 3, 1970, pp. 259–269.36Naot, D., Shavit, A., and Wolfshtein, M., “2-Point-Correlation Model and the Redistribution of Reynolds-Stresses,” Phys. Fluids,

Vol. 16, No. 6, June 1973, pp. 738–743.37Gerolymos, G. A., Senechal, D., and Vallet, I., “Performance of Very-High-Order Upwind Schemes for DNS of Compressible Wall-

Turbulence,” Int. J. Num. Meth. Fluids, Vol. 63, July 2010, pp. 769–810.38Gerolymos, G. A., Senechal, D., and Vallet, I., “Wall-Effects on Pressure Fluctuations in Quasi-Incompressible and Compressible

Turbulent Plane Channel Flow,” AIAA Paper 2007–3863, 37. AIAA Fluid Dynamics Conference, 25–28 jun 2007, Miami [FL, USA], June

2007.39Le, H., Moin, P., and Kim, J., “Direct Numerical Simulation of Turbulent Flow over a Backward-Facing Step,” J. Fluid Mech.,

Vol. 330, 1997, pp. 349–374.40Yao, Y. F., Thomas, T. G., and Sandham, N. D., “Direct Numerical Simulation of Turbulent Flow over a Rectangular Trailing Edge,”

Theor. Comp. Fluid Dyn., Vol. 14, No. 5–6, 2001, pp. 337–358.41Wellborn, S. R., Reichert, B. A., and Okiishi, T. H., “Study of the Compressible Flow in a Diffusing S-Duct,” J. Prop. Power, Vol. 10,

No. 5, Sept. 1994, pp. 668–675.42Gerolymos, G. A., Senechal, D., and Vallet, I., “Very-High-Order WENO Schemes,” J. Comp. Phys., Vol. 228, Dec. 2009, pp. 8481–

8524.43Gerolymos, G. A., Senechal, D., Vallet, I., and Younis, B. A., “Comparison of Tensor Representaions of Velocity/Pressure-Gradient,

Pressure-Strain and Pressure-Velocity Correlations with Plane Channel Flow DNS Data,” 11. European Turbulence Conference, 25–28

jun 2007, Porto [PRT], June 2007.44Moser, R. D., Kim, J., and Mansour, N. N., “Direct Numerical Simulation of Turbulent Channel Flow up to Reτ = 590,” Phys.

Fluids, Vol. 11, No. 4, April 1999, pp. 943–945.45Launder, B. E. and Tselepidakis, D. P., “Application of a New Second-Moment Closure to Turbulent Channel Flow Rotating in

Orthogonal Mode,” Int. J. Heat Fluid Flow, Vol. 15, 1994, pp. 2–10.46So, R. M. C., Aksoy, H., Yuan, S. P., and Sommer, T. P., “Modeling Reynolds-number effects in wall-bounded turbulent flows,”

ASME J. Fluids Eng., Vol. 118, June 1996, pp. 260–267.47Hanjalic, K. and Jakirlic, S., “Contribution towards the Second-Moment Closure Modelling of Separating Turbulent Flows,” Comp.

Fluids, Vol. 27, 1998, pp. 137–156.48Craft, T. J. and Launder, B., “Principles and Performance of TCL-Based Second-Moment Closures,” Flow Turb. Comb., Vol. 66,

2001, pp. 355–372.49Shir, C. C., “A Preliminary Numerical Study of Atmospheric Turbulent Flows in the Idealized Planetary Boundary-Layer,” J. Atm.

Sci., Vol. 30, 1973, pp. 1327–1339.50Batten, P., Craft, T. J., Leschziner, M. A., and Loyau, H., “Reynolds-Stress-Transport Modeling for Compressible Aerodynamics

Applications,” AIAA J., Vol. 37, No. 7, July 1999, pp. 785–797.51Shih, T. H. and Lumley, J. L., “Modeling of Pressure Correlation Terms in Reynolds Stress ans Scalar Flux Equations,” Rep.

FDA-85-3, Sibley School of Mechanical and Aerospace Engineering, Cornell University, Ithaca, [NY], March 1985.52Shih, T. H. and Lumley, J. L., “Critical Comparison of 2-Order Closures with Direct Numerical Simulations of Homogeneous

Turbulence,” AIAA J., Vol. 31, No. 4, April 1993, pp. 663–670.53Kebede, W., Launder, B., and Younis, B., “Large-Amplitude Periodic Pipe Flow: A Second-Moment Closure Study,” 5. Symposium

on Turbulence Shear Flows, Cornell [USA], Aug. 1985.54Launder, B. E. and Tselepidakis, D. P., “Directions in 2-Moment Modelling of Near-Wall Turbulence,” AIAA Paper 1991–0219,

1991.55Shima, N., “Modeling of Asymptotic Near-Wall Behavior of the Reynolds-Stress Dissipation,” 10. Symposium on Turbulence Shear

Flows, the Pennysylvania State University, University Park [PA, USA], 1995, vol. 1, paper 2-13.56So, R. M. C. and Yuan, S. P., “A Geometry Independent Near-Wall Reynolds-Stress Closure,” Int. J. Eng. Sci., Vol. 37, 1999,

pp. 33–57.57Hanjalic, K. and Launder, B. E., “Contribution Towards a Reynolds-Stress Closure for Low-Reynolds-Number Turbulence,” J.

Fluid Mech., Vol. 74, 1976, pp. 593–610.58Tagawa, M., Nagano, Y., and Tsuji, T., “Turbulence Model for the Dissipation Components of Reynolds Stresses,” 8. Symposium

on Turbulence Shear Flows,Technical University of Munich [GER], Sept. 1991, Paper 29-3.59Speziale, C. G. and Gatski, T. B., “Analysis and Modelling of Anisotropies in the Dissipation Rate of Turbulence,” J. Fluid Mech.,

Vol. 344, 1997, pp. 155–180.60Oberlack, M., “Non-Isotropic Dissipation in Non-Homogeneous Turbulence,” J. Fluid Mech., Vol. 350, 1997, pp. 351–374.61Daly, B. J. and Harlow, F. H., “Transport Equations in Turbulence,” Phys. Fluids, Vol. 13, 1970, pp. 2634–2649.62Gerolymos, G. A., “Implicit Multiple-Grid Solution of the Compressible Navier-Stokes Equations using k− ε Turbulence Closure,”

23 of 24

American Institute of Aeronautics and Astronautics

AIAA J., Vol. 28, No. 10, Oct. 1990, pp. 1707–1717.63Klebanoff, P. S., “Characteristics of Turbulence in a Boundary-Layer with Zero Pressure Gradient,” Rep. 1247, NACA, 1955.64So, R. M. C., Zhang, H. S., Gatski, T. B., and Speziale, C. G., “Logarithmic Laws for Compressible Turbulent Boundary-Layers,”

AIAA J., Vol. 32, 1994, pp. 2162–2168.65Obi, S., Aoki, K., and Masuda, S., “Experimental and Computational study of turbulent separating flow in an asymmetric plane

diffuser,” Proceedings 9th Symposium on Turbulent Shear Flows, 16–18 Aug 1993, Kyoto, [J], edited by F. Durst, N. Kasagi, B. Launder,

F. Schmidt, K. Suzuki, and J. Whitelaw, Aug. 1993, pp. 305–1–305–4.66Buice, C. U. and Eaton, J. K., “Experimental Investigation of Flow through an Asymmetric Plane Diffuser,” ASME J. Fluids Eng.,

Vol. 122, June 2000, pp. 433–435.67Durbin, P. A., “Separated Flow Computation with the k − ε − v2 Model,” AIAA J., Vol. 33, No. 4, April 1995, pp. 659–664.68Kaltenbach, H.-J., Fatica, M., Mittal, R., Lund, T. S., and Moin, P., “Study of Flow in a Planar Asymmetric Diffuser using Large-

Eddy Simulation,” J. Fluid Mech., Vol. 390, 1999, pp. 151–185.69Apsley, D. D. and Leschziner, M. A., “Advanced Turbulence Modelling of Separated Flow in a Diffuser,” Flow Turb. Comb., Vol. 63,

1999, pp. 81–112.70Laurence, D. R., Uribe, J. C., and Utyuzhnikov, S. V., “A Robust Formulation of the v2-f Model,” Flow Turb. Comb., Vol. 73, 2004,

pp. 169–185.71Gullman-Strand, J., Tornblom, O., Lindgren, B., Amberg, G., and Johansson, A. V., “Numerical and Experimental Study of Sepa-

rated Flow in a Plane Asymmetric Diffuser,” Int. J. Heat Fluid Flow, Vol. 25, 2004, pp. 451–460.72Kobayashi, H., Ham, F., and Wu, X., “Application of a local SGS Model based on Coherent Structures to Complex Geometries,”

Int. J. Heat Fluid Flow, Vol. 29, 2008, pp. 640–653.73Rahman, M. M. and Siikonen, T., “An Eddy Viscosity Model with Elliptic Realxation Approach,” Int. J. Heat Fluid Flow, Vol. 30,

2009, pp. 319–330.74Gerolymos, G. A., Sauret, E., and Vallet, I., “Influence of Inflow-Turbulence in Shock-Wave/Turbulent-Boundary-Layer Interaction

Computations,” AIAA J., Vol. 42, No. 6, June 2004, pp. 1101–1106.75Okinaga, Y. and Obi, S., “Effect of Near-Wall Treatments on the Prediction of Turbulent Separating Flows,” 4th JSME-KSME

Thermal Engineering Conference, oct 1-6, Kobe, [J], 2000, pp. 479-484.76Schulein, E., “Skin-Friction and Heat Flux Measurements in Shock/Boundary-Layer Interaction Flows,” AIAA J., Vol. 44, No. 8,

Aug. 2006, pp. 1732–1741.77Pasha, A. A. and Sinha, K., “Shoc-unsteadiness Model apllied to Oblique Shock Wave/Turbulent Boundary-Layer Interaction,”

Int. J. Comp. Fluid Dyn., Vol. 22, No. 8, Sept. 2008, pp. 569–582.78Steelant, J., “Effect of a Compressibility Correction on Different Turbulence Models,” Proceedings 5. International Symposium on

Engineering Turbulence Modelling andMeasurements (ETMM5), 16–18 sep 2002, Mallorca [E], edited by W. Rodi and N. Fueyo, Elsevier,

Oxford [UK], Sept. 2002, pp. 207–216.

24 of 24

American Institute of Aeronautics and Astronautics


Recommended