You are on page 1of 22

Journal of Turbulence

ISSN: (Print) 1468-5248 (Online) Journal homepage: http://www.tandfonline.com/loi/tjot20

Prediction of transitional and fully turbulent flow


using an alternative to the laminar kinetic energy
approach

Maurin Lopez & D. Keith Walters

To cite this article: Maurin Lopez & D. Keith Walters (2015): Prediction of transitional and
fully turbulent flow using an alternative to the laminar kinetic energy approach, Journal of
Turbulence, DOI: 10.1080/14685248.2015.1062509

To link to this article: http://dx.doi.org/10.1080/14685248.2015.1062509

Published online: 03 Dec 2015.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at


http://www.tandfonline.com/action/journalInformation?journalCode=tjot20

Download by: [University of Nebraska, Lincoln] Date: 04 December 2015, At: 11:03
JOURNAL OF TURBULENCE,
http://dx.doi.org/./..

Prediction of transitional and fully turbulent flow using an


alternative to the laminar kinetic energy approach
Maurin Lopeza and D. Keith Waltersb
a
Center for Advanced Vehicular Systems, Mississippi State University, Starkville, MS, USA; b Department of
Mechanical Engineering, Mississippi State University, Starkville, MS, USA
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

ABSTRACT ARTICLE HISTORY


This paper presents a new eddy-viscosity turbulence model for pre- Received February
diction of laminar, transitional and turbulent flow, as an alternative Accepted May
to the popular laminar kinetic energy framework. The model repre-
KEYWORDS
sents a conceptually different description of the transition process, turbulence modelling;
in which the pressure-strain terms of the Reynolds stress budget are computational uid
suppressed in pretransitional flow regions, and transition is viewed as dynamics; transition; shear
activation of the pressure-strain terms. This concept has been imple- ows
mented using an existing linear-eddy viscosity Reynolds-averaged
NavierStokes transitional model as a baseline, which has been mod-
ified to more accurately capture the physics of fully turbulent free
shear flows. The model is tested for several boundary layer and free
shear flow test cases. The simulations show engineering accurate
results, qualitatively equivalent to the baseline transition-sensitive
model for boundary layer test cases, and substantially improved over
the baseline model for fully turbulent free shear flows. The results
suggest that the new model formulation may have a wider range of
application than the baseline model for complex transitional and tur-
bulent flows.

Introduction
Transitional flow phenomena are observed in a wide range of engineering applications,
including aerospace, aeronautics, biomedical, wind turbines, etc. Prediction of transitional
flow using computational fluid dynamics (CFD) is of vital importance in aerodynamics sim-
ulations. For example, in some cases, a boundary layer may remain laminar throughout a
significant portion of the domain, and fully turbulent simulations may produce results that
can lead to inaccurate conclusions or inefficient design. The inherent behaviour of transi-
tional phenomena is very complex and the physics of transition is still not fully understood.
Nevertheless, the development and application of transition-sensitive turbulence models in
CFD simulations is an active research area, with significant progress made over the past two
decades.
A number of modelling approaches have been adopted to predict boundary layer
transition. These include direct numerical simulations, [1] low Reynolds number eddy
viscosity turbulence models, [25] coupling of empirical correlations [6,7] with fully

CONTACT D. Keith Walters walters@me.msstate.edu


Taylor & Francis
2 M. LOPEZ AND D.K. WALTERS

turbulent Reynolds-averaged Navier-Stokes (RANS) models, or modifications of fully tur-


bulent RANS models in order to include the effects of the transition process.[814]
The most popular RANS-based transition modelling approaches make modifications
to existing transport equations and/or add additional equations within an RANS-based
turbulence modelling framework to yield a local (i.e. single-point) model form. Models
adopting this approach can usually be classified as either correlation-based [812] or
physics-based.[1316] In practice, both approaches rely on a significant level of inherent
empiricism, since no simple and accurate theoretical description of the transition process
has yet been derived within the context of the Reynolds-averaged equations. The dis-
tinction between correlation- or physics-based models lies in the manner in which this
empiricism is introduced. In the former, explicit correlations are adopted based on existing
experimental data-sets for specific transitional flow conditions. In the latter, model terms
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

and coefficients are proposed to represent physical mechanisms in a heuristic manner,


and are calibrated based on applications of the model to a prescribed set of validation test
cases.
Suzen and Huang [8] proposed a correlation-based transition model that includes a
transport equation for an intermittency factor. Steelant and Dick [9] developed a transport
equation for the intermittency factor and incorporated it into conditioned NavierStokes
equations. Menter et al. [10] proposed a single-point, correlation-based transition model
that includes two different transport equations: one for the intermittency factor and the
other for the transition onset Reynolds number. Wang and Perot [13] applied additional
equations for turbulence potential terms to formulate a single-point, physics-based transi-
tion model. Walters and Leylek [15] developed a phenomenological RANS-based, single-
point transitional model that addresses in-depth transitional flow physics without intermit-
tency factors, based on an approach originally proposed by Mayle and Schulz [16,17]. The
newest version of this model was introduced by Walters and Cokljat [18]. For these model
types, a transport equation for the laminar kinetic energy (LKE) is added and the transition
process is represented as a transfer of energy from the streamwise non-turbulent velocity
fluctuations to the fully turbulent three-dimensional velocity fluctuations. The single-point
transition models of Menter and co-workers [1012], Wang and Perot [13] and Walters and
co-workers [15,18] have achieved relatively wide acceptance due in large part to their easy
to implement and generalised nature.
This paper adopts a description of the transition process based on a modification of
the LKE concept. Instead of a decomposition of velocity fluctuations into turbulent and
non-turbulent, as embedded in the LKE concept, this research proposes the introduction
of a new variable that represents the wall-normal turbulent velocity fluctuations which are
responsible for the initiation of transition. In this viewpoint, the new variable is a mea-
sure of the degree of isotropy of velocity fluctuations, where non-turbulent fluctuations are
assumed to be highly anisotropic relative to fully turbulent fluctuations. This new concept
has been discussed in previous papers, [19,20] but the current work represents the first time
that a complete and fully generalised model based on it has been proposed.

New model concept


The LKE concept was first introduced by Mayle and Schulz [17] and it has been used
to construct RANS-based transitional models since then. [1518] However, the LKE
JOURNAL OF TURBULENCE 3

concept has some theoretical weaknesses: (1) the dynamics of LKE production is not
entirely understood.[18] Walters and Leylek [15], for example, used a different LKE produc-
tion mechanism compared with that proposed by Mayle and Schulz [17]. (2) The descrip-
tion of the transition process is not clearly defined. The transition process is due to a transfer
of energy from the laminar to the turbulent kinetic energy which is simulated in [18] via
equal and opposite terms included in the respective transport equations, which is a fun-
damentally ad hoc approach. Furthermore, there is no precise definition for LKE in the
research community.
The new model concept presented in this paper is intended to provide a less ad hoc
description of the turbulence process, by defining a new variable tied to the dynamics of
the Reynolds stress equations, which are well defined mathematically and (relatively) well
understood physically. The new model concept explicitly ties the transition process to the
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

pressure-strain terms in the Reynolds stress equations. Assuming that the pretransitional
region develops due to a suppression of the pressure-strain terms in the Reynolds stress
transport equation, transition is initiated with the activation of these terms due to non-
linear instability mechanisms, which results in a transfer of energy from the streamwise
to the wall normal and spanwise components, and leads to a rapid rise in turbulence pro-
duction. In most LKE models, such as [18] for example, this transfer of energy is made
from the laminar to the turbulent kinetic energy via equal and opposite terms included in
the respective transport equations. The version of the model presented here includes the
initiation of the transition process by the rapid increase of energy in the fully turbulent,
three-dimensional velocity fluctuations which are represented by the model variable v 2 .
This modelling approach, initially introduced by Lopez and Walters in [19], leads to slow
growth of fluctuating energy in the pretransitional region and relaxation towards a fully
turbulent model form downstream of transition.
The motivation for using the variable v 2 comes from several sources in the open lit-
erature. The variable was originally introduced into the k- model in order to represent
velocity fluctuations normal to mean flow streamlines, which most directly influence the
mean momentum transfer in turbulent shear layers.[21] It is therefore inherently tied to the
anisotropy of the Reynolds stress tensor. Likewise, the v 2 variable has been used to incorpo-
rate the effects of the pressure-strain terms within the context of linear-eddy viscosity mod-
els.[22] The connection between v 2 and transition has also been noted.[23] For example,
the LES results of Voke and Yang [24] show a positive contribution of the pressure-strain
term to the streamwise Reynolds stress component in the pretransitional region, appar-
ently a wall reflection effect, which works to suppress rather than increase the wall-normal
component represented by v 2 . This inhibits the growth of the shear stress component and
causes relatively low production of turbulent kinetic energy. Other studies [2527] show the
same behaviour, and report no peak in wall-normal or spanwise fluctuation energy within
the pretransitional region of the boundary layer. Furthermore, these studies have demon-
strated that only the wall-normal component of entrained freestream turbulence leads to
the growth of the Klebanoff modes, and that the onset of transition coincides with a sudden
increase in wall normal energy. In fact, [24] showed that pure streamwise disturbances at
the inflow are ineffective at forcing transition, while wall-normal disturbances alone are as
effective as full isotropic disturbances. These considerations have led to the adoption of the
k v 2 framework as an alternative to the k kL model approach in [18] for the
development of a phenomenological transition-sensitive model.
4 M. LOPEZ AND D.K. WALTERS

New model formulation


The model presented in this section is a modification of the transitional model developed
by Walters and Cokljat in [18]. The new version incorporates the new variable v 2 discussed
above. Note that the new model variables are related to the previously defined variable kL
in [18] as kL k v 2 .
For simplicity, the equations are presented here in their incompressible form. The model
equations have the compact form:

  
Dk T k
= Pk min(k, v 2 ) Dk + + (1)
Dt x j k x j
  2
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Dv 2 T v
= Pv 2 + RBP + RNAT v 2 Dv 2 + + (2)
Dt x j k x j
 
D CR
= P + 1 (RBP + RNAT ) C2 2 fW2
Dt fW v2
  


1 k T
+ 2 1 F1 2 + + . (3)
x j x j x j x j

As a result of the change of variables, note that Equation (1) does not include the tran-
sition terms RBP and RNAT as in [18]. In the k equation, those terms were used to repre-
sent the transfer of energy from the non-turbulent pretransitional fluctuations (kL ) to the
fully turbulent fluctuations (kT ). In the new model, production of v 2 is suppressed in the
pretransitional region. Transition initiates when the value of the term RBP becomes non-
negligible, representing activation of the pressure-strain terms, followed by the growth of
three-dimensional, fully turbulent fluctuations (v 2 ). Equation (3) also incorporates a cross
diffusion term as in the SST (Shear Stress Transport) k- k model that helps ensure the
correct prediction of the boundary layer wake region, in an effort to address deficiencies in
the original model [18] as evidenced in [28].
The definitions of most of the model terms are similar to the version of the model in [18],
and are used without modification in the new model. The production terms are expressed
as
 

Pk = T S , Pv 2 = T,s S ,
2 2
P = C1 T,s S2 . (4)
v2

The turbulent viscosity T used in the momentum equation is the sum of the small and
large-scale contributions

T = T,s + T,l . (5)

The small-scale eddy viscosity and the effective small-scale turbulence are expressed as

T,s = f fINTC v s2 eff and v s2 = fss fW v 2 . (6)


JOURNAL OF TURBULENCE 5

The effective (wall limited) turbulent length scale eff and the damping function fW are
defined as

v s2
eff = min (C d, T ) , T = (7)

 2/
eff 3
fW = . (8)
T
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

The viscous wall effect is incorporated as


 
ReT fW2 v 2
f = 1 exp , ReT = . (9)
A

The shear sheltering effect responsible for inhibiting the production of three-
dimensional velocity fluctuations in the pretransitional region is expressed as
  
CSS
2
fss = exp . (10)
v2

The turbulent viscosity coefficient and the intermittency damping function for the tur-
bulence production are defined as

1 v2
C =  , fINT = min ,1 . (11)
A0 + AS S CINT k

The large-scale eddy viscosity in Equation (5) is defined as


 
 
 2  0.5 (k v 2

2eff deff
s
T,l = min f,l C11 v l2 eff + T SC12 2
deff
, . (12)
S

The limit is applied to satisfy the realisability constraint for the total Reynolds stress con-
tribution. Note that the second part of this term is slightly different from the corresponding
term in [18]. Instead of the wall distance variable d, the variable deff is used and is defined
as

eff
deff = . (13)
C

This change is intended to limit the production of natural pretransitional modes in zones
far from the wall in fully turbulent flows, where this mechanism should not be present.
The variable v l2 represents the energy contained in large-scale turbulent motions, and is
6 M. LOPEZ AND D.K. WALTERS

represented as

v l2 = v 2 v s2 (14)

Other terms in (12) are defined as


 
v2
f,l = 1 exp C,l 2 l 2 (15)
eff

 
max (Re
CT S,crit , 0)2 d2

T S = 1 exp , Re
= .
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

(16)
AT S

The near-wall dissipation terms for k and v 2 are expressed as


 
k k v2 v2
Dk = 2 , Dv 2 = 2 . (17)
x j x j x j x j

The effective scalar diffusivity T in the turbulent transport terms is defined by


T = f v s2 eff . (18)

The cross diffusion term in the equation is similar to the one in the fully turbulent SST
k- model. The term is included to improve the behaviour of the model in the wake region
and in separated shear layers, where the SST k model has proven to be effective.[29,30]
To accommodate transitional flow, the blending function F1 is defined as

F1 = 1 [(1 F1 ) fSS ] (19)

in which the shear-sheltering damping function inhibits the fully turbulent effects of the
F1 function in the pretransitional region of the boundary layer. The F1 function is defined
similar to the SST k model
   4
v 2 500 42 k
F1 = tanh min max , , (20)
d d2 CDk d 2

with
 
1 k 10
CDk = max 22 , 10 . (21)
x j x j

The terms representing the natural and bypass transition are defined as

RNAT = CR,NAT NAT (k v 2 )


, RBP = CR BP (k v 2 )/ fW . (22)
JOURNAL OF TURBULENCE 7

Table . Model constants.


A0 = 4.04 Equation () CINT = 0.95 Equation () C1 = 0.44 Equation ()

AS = 2.12 Equation () CT S,crit = 1000 Equation () C2 = 0.92 Equation ()


A = 3.8 Equation () CR,NAT = 0.02 Equation () CR = 1.15 Equation ()
ABP = 0.2 Equation () C11 = 3.4 106 Equation () C = 2.495 Equation ()
ANAT = 200 Equation () C12 = 1.0 1010 Equation () = 0.09 Equation ()
AT S = 200 Equation () CR = 0.32 Equation () Pr = 0.85 Equation ()
CBP,crit = 1.5 Equation () C, = 0.035 Equation () k = 1 Equation ()
CNC = 0.1 Equation () CSS = 3.0 Equation () = 1.17 Equation ()
CNAT,crit = 1450 Equation () C,1 = 4360 Equation () 2 = 1.856 Equation ()

As discussed in [18], transition is initiated when the characteristic time-scale for tur-
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

bulence production is smaller than the viscous diffusion time scale of pretransitional fluc-
tuations. This process is controlled by the transition initiation terms NAT and BP . The
expressions for these two terms are

   
NAT BP
NAT = 1 exp , BP = 1 exp (23)
ANAT ABP
  
CNAT,crit kd
NAT = max Re
, 0 , fNAT,crit = 1 exp CNC (24)
fNAT,crit
 
v2
BP = max CBP,crit , 0 . (25)

To include heat transfer effects, the turbulent heat flux vector can be modelled using a
turbulent thermal diffusivity

T
u i T = (26)
xi

v 2 T,s  
= fW + 1 fW C, v 2 eff . (27)
k Pr

Model constants are given in Table 1. Constants related to the fully turbulent k- turbu-
lence model take common values (e.g. C1 = 0.44, C2 = 0.92, etc.). Many of the constants
specific to the transition aspects of the model are identical to their counterparts in Ref. [18].
Constants influencing the transition process that take different values in the current model
from Ref. [18] include A , ABP , CBP,crit , CNAT,crit , CINT , CR and CSS . Similar to the approach
taken in [18], these constants were calibrated based on the zero-pressure flat plate boundary
layer simulations described in the results section below.
In practice, the model will yield the result v 2 = k in fully turbulent regions, i.e. down-
stream of the transition location in attached boundary layers outside of the viscous sublayer,
and in far-field regions. In the pretransitional region of attached boundary layers, the model
predicts v 2 < k. The ratio v 2 /k may therefore be viewed as an indicator function for iden-
tifying fully turbulent regions. For regions in which v 2 /k = 1, the equations governing v 2
and k are identical, and the model effectively reduces to a two-equation k- model form
8 M. LOPEZ AND D.K. WALTERS

given by
  
Dk k k T k
= T S k 2
2
+ + (28)
Dt x j x j x j k x j

  
D  1 k T
= C1 T S2 C2 2 fW2 + 2 1 F1 2 + + .
Dt k x j x j x j x j
(29)

Boundary conditions
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

The boundary conditions for the model variables follow those of [18]. At solid boundaries,
the no slip condition enforces

k = v2 = 0 (30)

and a zero normal gradient condition is used for


= 0, (31)

where is the coordinate direction normal to the wall. The zero-gradient boundary condi-
tion for is chosen as an alternative to the more traditional boundary condition for which
2 in the near-wall limit as 0. The choice allows the RANS model to characterise
relatively low-frequency fluctuations present in the pretransitional boundary layer as well as
the high-frequency fluctuations present in the fully turbulent boundary layer. Dissipation
due to the direct action of viscosity on near-wall fluctuating gradients is instead addressed
through the inclusion of the terms defined in Equation (17).
At the inlet, the values for k and are calculated as for other k- type models. The
turbulence kinetic energy is usually obtained based on the inlet turbulence intensity T u ,
assuming isotropic freestream turbulence

2
3
k
T u = . (32)
U

The value of is chosen to match the available freestream information. For example,
if the turbulent length scale or the decay rate is known, then is chosen to appropriately
reproduce the freestream conditions. Since a good representation for v 2 is given by v 2
k kL and the appropriate inlet boundary condition for kL is kL = 0,[18] then it is suggested
to use v 2 = k as the boundary condition at the inlet in regions of freestream flow.

Test cases
The model described above was implemented as a user-defined function in the commercial
finite volume CFD solver ANSYS FLUENT version 14.0.[31] The pressure-based solver
option was used with the SIMPLE (Semi-Implicit Method for Pressure Linked Equations)
JOURNAL OF TURBULENCE 9

method for pressure-velocity coupling. This approach has been well demonstrated to be
appropriate for incompressible single-phase flows. All results presented in this document
used a second-order upwind-based spatial discretisation scheme for convective terms, and
a second-order central difference scheme for diffusive terms. The test cases selected include
standard boundary layer cases along with an airfoil test case in order to check the behaviour
of the model for transitional boundary layers. Since any transition-sensitive model must
also be able to predict the flow in fully turbulent attached and separated boundary layers,
a fully turbulent backward facing step case is considered. The model is expected to behave
qualitatively equal to the kT kL model in these cases. A round jet flow case is included
as well, to demonstrate the improvements in accuracy due to the modifications presented
in this document, for regions of separated shear layer flow.
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Boundary layer test cases


The model was initially tested and calibrated for boundary layer cases. The new model was
expected to have a similar behaviour for transitional boundary layers to its baseline model.
Flat plate cases with and without pressure gradients, as well as an airfoil test case, were
selected.

Flat plate
The transition prediction behaviour of the model was initially tested for zero-pressure and
variable pressure gradient flat plate boundary layers, with different freestream turbulence
intensities. The test cases chosen match the T3 validation cases from the ERCOFTAC clas-
sic database.[32] The T3 test cases were developed specifically for validation of transition
models and have become a recognised standard in the research community.
For each of the flat plate test cases, the computational domain was constructed to match
the experimental geometry. A symmetry plane was applied at the bottom of the domain,
upstream of the plate leading edge. This was done to allow a natural stagnation and bound-
ary layer initiation. The other boundaries were set as velocity inlet, pressure outlet, wall, and
symmetry planes, as appropriate. For the zero-pressure-gradient (ZPG) cases, the extent
of the domain in the vertical (y) direction was chosen to be far enough from the plate to
ensure negligible acceleration of the freestream due to the finite plate thickness and bound-
ary layer development. For the cases with applied favourable and adverse pressure gradient,
the upper wall was contoured to ensure the appropriate variation in the streamwise velocity.
The meshes, shown in Figure 1, consisted of block-structured quadrilaterals clustered
in the near-wall and leading edge regions, and unstructured triangular elements far from
the wall for the pressure gradient case. The first near-wall cell was placed such that y+ was
less than one over the entire plate surface for all cases considered. The number of grid
cells in the two-dimensional meshes was 30,196 and 24,650. The top wall for the second
geometry was contoured in order to produce a varying (favourable and adverse) streamwise
flow acceleration. The contoured top of the geometry was built to match the experimental
pressure distribution on the plate. Mesh independence was verified using the procedure
outlined in [18].
The inlet conditions for the T3A, T3A-, T3B and T3C2 test cases were identical to those
reported in [18]. Inlet conditions for the T3C1, T3C3 and T3C5 test cases were chosen in
10 M. LOPEZ AND D.K. WALTERS
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Figure . Meshes used for at plate test cases. (a) ZPG at plate for TA-, TA, TB. (b) Pressure gradient
at plate for TC, TC, TC, TC.

order to reproduce the correct freestream decay of turbulent kinetic energy in agreement
with the experimental data, using a similar procedure as that discussed in [18]. The inlet
values for the dimensionless turbulence variables are listed in Table 2, and the effective
turbulent viscosity used to determine the inlet value of is defined as

v2
T = . (33)

Table . Leading edge freestream conditions for at plate test


cases.
Test case Tu (%) T /

TA- . .
TA . .
TB . .
TC . .
TC . .
TC . .
TC . .
JOURNAL OF TURBULENCE 11
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Figure . Skin friction coecient for ZPG at plate test cases: (a) TA-, (b) TA, (c) TB.

Figure 2 shows the skin friction coefficient for the ZPG flat plate versus the dimensionless
downstream distance predicted by the new model (v 2 model) compared with the experi-
mental data,[32] the fully turbulent SST k model [29] and the kT kL transitional
model.[18] It is apparent that the v 2 model predicts well the transition location in all cases.
The shear stress levels in the laminar and transitional region show excellent agreement with
the experimental data. The predicted skin friction coefficients for the varying pressure gra-
dient test cases are shown in Figure 3, and while a small overprediction is apparent in the
pretransitonal region for all of the T3C cases, it is evident that the transition location is
better predicted by the v 2 model than the kT kL model, even when the transition is
early or delayed. All cases show a smooth transition rather than a sudden jump in shear
stress levels. The predicted wall shear stresses are quite similar for both models in the pre-
transitional and fully turbulent regions.
Figures 4 and 5 show the velocity and total fluctuating kinetic energy profiles for the new
v 2 model and the kT kL model. The profiles of the SST model are not included due
to that models inability to resolve laminar and transitional flow regions. The profiles were
obtained at a dimensionless location Rex = 105 for the laminar region, Rex = 2.25 105
for the transitional region and Rex = 5 105 for the fully turbulent region. The predicted
velocity profiles are slightly better for the v 2 model in all regions, while the kT kL
model better captures the peaks of the total kinetic energy in all three regions. In general,
the results of the two transitional models are very similar in all four cases, and it is concluded
12 M. LOPEZ AND D.K. WALTERS
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Figure . Skin friction coecient for variable pressure gradient at plate test cases: (a) TC, (b) TC, (c)
TC, (d) TC.

that the proposed model correctly inherits the transitional behaviour of the baseline model,
as expected.

VPI cascade
A more complex test case that illustrates the importance of the transitional modelling is
the airfoil test case performed at the Virginia Polytechnic Institute and State University,

Figure . Velocity proles in the (a) laminar, (b) transitional and (c) turbulent regions for the TA case.
JOURNAL OF TURBULENCE 13

Figure . Turbulent kinetic energy proles in the (a) laminar, (b) transitional and (c) turbulent regions for
the TA case.
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

and documented for the express purpose of validating transitional flow CFD simulations.
As shown previously, transition in the boundary layer affects the skin friction distribution,
which indirectly affects the separation or reattachment of the flow on airfoils, and it can
dramatically alter the force and moment distribution of lifting bodies. The experiments
used here for comparison were documented by Radomsky and Thole [33,34].
The mesh and boundary conditions for this test case are identical to those in [18]. The
hybrid mesh in Figure 6 was built with 24,386 cells. The inlet air velocity was 5.85 m/s,
which corresponds to a Reynolds number of 230,000 based on a chord length of 59.4 cm.
Two test cases, corresponding to relatively high freestream turbulence levels of 10% and
19.5%, were run. For the two cases, the specific dissipation rate was chosen to correspond
to a turbulent viscosity ratio T / of 900 and 2100, respectively.
A constant heat flux boundary condition was applied on the airfoil surface and the heat
transfer coefficient was calculated using the three models previously discussed for compar-
ison. Figure 7 shows heat transfer coefficient versus distance along the airfoil surface (from
the stagnation point) normalised by the chord length (s/C). Negative values of s indicate
the pressure surface; positive values indicate the suction surface. Transition is predicted

Figure . Periodic domain and mesh for the VPI cascade.


14 M. LOPEZ AND D.K. WALTERS
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Figure . Heat transfer coecient on airfoil surface for the VPI test case: (a) T u = 10%, (b) T u =
19.5%.

for both transitional models with comparable accuracy in both cases, while the SST model
again fails to reproduce this result. As above, it is concluded that the new model reproduces
the performance of the baseline model [18] for this attached boundary layer case.

Backward facing step


The third test case in this study is the backward facing step, which is a widely used bench-
mark test case for turbulence model validation. The purpose of this test case is not to high-
light the transitional prediction capabilities of the model, but rather to demonstrate that
the model can be used for cases with fully turbulent attached and separated boundary layer
flow and still produce reasonable results in agreement with previously published models.
In this test case, the boundary layer separates at the step with a reattachment and con-
tinued boundary layer development farther downstream. The details of the experimental
configuration are found in [35], and the computational domain and inlet conditions were
developed to match as closely as possible as that case. Figure 8 shows the geometry and
mesh. In order to allow the inlet boundary layer to correctly develop, the domain upstream
of the step was built to measure 100D, where D = 1.27 cm is the step height. With an inlet
velocity of 44.2 m/s and T u = 3.0%, the flow is fully turbulent at the step location, and
the boundary layer height is equal to that reported in the experiments.
Figure 9 shows the pressure distribution and the skin friction coefficient calculated at the
bottom wall of the domain. Note that x = 0 corresponds to the streamwise location of the
step. The attached and separated flow is fully turbulent in this section of the domain and the
SST model shows better performance relative to the transitional models. The negative and
positive peaks in the pressure coefficient are very well captured by the SST model, followed
in accuracy by the proposed v 2 model. The three models show good accuracy in predicting
the skin friction coefficient. The reattachment point is very well predicted by the SST model,
while it occurs slightly early for the v 2 model and even earlier for the kT kL model.
Figure 10 shows the mean streamwise velocity profiles at different locations downstream
of the step. In the recirculation zone (x/D = 1.0, 3.0), the SST model accurately predicts
the velocity profiles in the boundary layer. For the two transitional models, the negative
JOURNAL OF TURBULENCE 15
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Figure . Geometry and mesh for the backward facing step test case.

peak close to the wall is over predicted. The two transitional models display comparable
behaviour away from the wall. After reattachment, all of the models show qualitatively cor-
rect prediction of the developing boundary layer, though the v 2 model seems to be slightly
more accurate than the other models, while all show a similar behaviour outside of the
boundary layer.
The test cases presented above demonstrate the ability of the v 2 model to predict bound-
ary layer flows both with and without separation with reasonable accuracy, at least compa-
rable to the kT kL model upon which it is based. Obviously, both of the transitional
models show improved performance versus the SST model for cases in which attached
boundary layer transition occurs. Significantly, the v 2 model shows improved performance

Figure . (a) Pressure coecient and (b) skin friction coecient distribution on the bottom wall of the
backward facing step.
16 M. LOPEZ AND D.K. WALTERS
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Figure . Mean velocity proles at several streamwise stations for backward facing step test case.

versus the kT kL model for the fully turbulent separated flow region of the back-
step case, though both models underperform the SST model in this region. That the v 2
model also shows relatively accurate prediction downstream of reattachment is also sig-
nificant, since any general purpose transition-sensitive model must also be able to predict
flow in fully turbulent boundary layers even under non-equilibrium conditions such as
wake recovery.

Round jet flow


In this section, the models are compared for a fully turbulent round jet flow. Although
the test case therefore does not highlight any transitional capability of the model, it does
demonstrate that the model can be used for general purpose flow solutions, including those
in separated and far-field turbulent shear layers, more effectively than the kT kL on
which it is based.
There are several experimental and numerical studies of axisymmetric round jet flows in
which the performance of RANS models is mixed.[28,30,36] Of particular interest are the
results of Ghahremanian and Moshfegh in [28]. They studied the behaviour of several tran-
sitional and fully turbulent RANS models on a three-dimensional fully turbulent, round jet.
Their results show that the kT kL transitional model performs remarkably poorly for
JOURNAL OF TURBULENCE 17
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Figure . Axisymmetric computational domain conguration for the round jet test case.

this particular case. One of the key goals of this research is to correct these apparent failures
in free shear flows of the kT kL model in the new v 2 model.
A sketch for the domain and the mesh used here are shown in Figures 11 and 12. An
axisymmetric planar domain was used in the calculations, taking the centreline of the
domain as a symmetry axis. The flow conditions follow the values reported in the experi-
mental study in [36], with a jet exit velocity of 56.2 m/s and turbulence intensity of 0.58%.
In the experimental study, the flow was manipulated to transition to fully turbulent flow
upstream of the jet exit. For this reason, two grids were used in this study. In the first, the
length of the channel upstream of the exit is 3H. The fully turbulent SST k model was
run on this mesh, leading to an exit profile identical to the experiments in terms of bound-
ary layer height. From this simulation, profiles of velocity, turbulent kinetic energy and

Figure . Mesh used for round jet ow test case.


18 M. LOPEZ AND D.K. WALTERS
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Figure . Inverse centreline velocity decay for round jet case: (a) full domain; (b) close-up view in near
eld of jet exit.

specific dissipation rate were obtained at a location 1.5H downstream of inlet of the supply
tube. The second mesh was constructed such that the length of the channel upstream of the
exit is 1.5H. The v 2 and kT kL transitional models were run using this mesh, and the
profiles obtained from the SST model were used as inlet conditions for the simulations, in
order to reproduce a fully turbulent jet exit condition. Mesh independence for both meshes
and all three turbulence models was verified using a procedure similar to that outlined in
[18].
Computational results for the round jet case are compared with measurements in terms
of mean flow data, in order to highlight the ability of the models to accurately predict jet
mixing and decay. Figure 13 shows normalised centreline mean velocity, and very clearly
highlights the advantages of the new v 2 transitional model over the existing kT kL
model. The latter shows a significant error in predicting the centreline velocity decay of the
jet, while the v 2 model performs nearly like the SST model. This behaviour of the kT kL
model was previously documented in [28], along with results obtained from other RANS
turbulence models, where the SST k model was clearly superior. The combination of
these results could suggest that the proposed v 2 model is a competitive alternative among
RANS transitional models for this particular test case.
As seen in Figure 13(b), near the exit of the channel, the v 2 model predicts the virtual ori-
gin of the jet (the distance between the exit of the channel and the x-intercept of the straight
line representing the inverse velocity decay) to occur too early, which causes a small devi-
ation from the experimental data near the exit of the channel. This implies that the model
overpredicts the level of mixing in the early stages of the jet development, and is consistent
with the slight underprediction of reattachment length using the v 2 model for the back-
step case shown earlier. When compared to the baseline kT kL model, however, the
v 2 model shows significantly better agreement with the SST model and with experimental
results, even in the near field region. Farther downstream, the v 2 and SST models behave
asymptotically equal.
In Figure 14, cross-sectional mean velocity profiles are plotted versus the dimensionless
radial coordinate = y/(x x0 ), where x0 represents the virtual origin predicted by each
model. The results are compared with experimental data (stationary hot wire, laser-doppler
JOURNAL OF TURBULENCE 19

Figure . Mean axial velocity proles versus dimensionless radial coordinate: (a) x/H = , (b) x/H = ,
(c) x/H = .
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

anemometry and flying hot wire) reported in [36]. While the SST and the v 2 model produce
reasonably accurate results, the kT kL model shows qualitative and even physically
unrealistic discrepancy away from the centreline of the yet.
The jet flow results clearly demonstrate the advantages of the proposed v 2 model com-
pared with its baseline model. For free shear flows, the v 2 model behaves more like the
SST k model, while in transitional boundary layers it inherits the characteristics of the
kT kL model.

Conclusions
In this study, a new methodology for the description of the transition process in turbu-
lence models for use in CFD simulations has been proposed as an alternative to the LKE
approach. This new methodology has been used to further develop a physics-based, single-
point, transition-sensitive linear eddy-viscosity RANS model. The model uses the existing
kT kL transitional model presented in [18] as a baseline, along with a transformation
of variables to a k v 2 form. The introduction of the new variable v 2 agrees concep-
tually with the description of the transition process described in this paper and initially
introduced in [20].
One key modification to the new model is the introduction of a blended cross-diffusion
term in the equation, which yields closer agreement with SST model results and a better
ability to predict the behaviour of fully turbulent free shear flows. The cross-diffusion term
from the SST k model effectively incorporates the benefits of the SST k model on
the wake region into the new v 2 model. The new model formulation was tested for several
boundary layer cases and the fully turbulent round jet flow test case. The simulations show
accurate results, qualitatively equivalent to the baseline model for the transitional boundary
layer test cases, and substantially improved over the baseline model for free shear flows.
Even though the v 2 model is intended to closely mimic the baseline model when predicting
the transition process, the transition location predicted by the new model is more accurate
for the majority of the test cases presented here. The results for the free shear flow test case
(the round jet flow) confirm the SST-like behaviour of the new v 2 model, which leads to
more accurate results compared to the kT kL model. This suggests that the v 2 model
has a wider range of applications compared to the kT kL model, and can be more
confidently applied for complex test cases that contain features of both attached boundary
layers and separated shear flows.
20 M. LOPEZ AND D.K. WALTERS

Acknowledgments
This research was funded by the U.S. National Aeronautics and Space Administration (NASA) under
grant number NNX 10AN06A.

Disclosure statement
No potential conflict of interest was reported by the authors.

References
[1] Kalitzin G, Wu X, Durbin PA. DNS of fully turbulent flow in a LPT passage. Int J Heat Fluid
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

Flow. 2003;24(4):636644.
[2] Savill AM. By-pass transition using conventional closures. In: Launder BE, Sandham ND, edi-
tors. Closure strategies for turbulent and transitional flows. Cambridge (UK): Cambridge Uni-
versity Press; 2002. 464492.
[3] Wilcox DA. Simulation of transition with a two-equation turbulence model. AIAA J.
1994;32(2):247255.
[4] Rumsey CL. Apparent transition behavior of widely-used turbulence models. Int J Heat Fluid
Flow. 2007;28(6):14601471
[5] Abid R. Evaluation of two-equation turbulence models for predicting transitional flows. Int J
Eng Sci. 1993;31(6):831840.
[6] Dhawan S, Narasimha R. Some properties of boundary layer during the transition from laminar
to turbulent flow motion. J Fluid Mech. 1958;3:418436.
[7] Abu-Ghannam BJ, Shaw R. Natural transition of boundary layers: the effects of turbulence,
pressure gradient, and flow history. J Mech Eng Sci. 1980;2:213228
[8] Suzen YB, Huang PG. Modeling of flow transition using an intermittency transport equation.
J Fluids Eng. 2000;122:273284.
[9] Steelant J, Dick E. Modeling of laminar-turbulent transition for high freestream turbulence. J
Fluids Eng. 2001;123:2230.
[10] Menter FR, Langtry RB, Likki SR, et al. A correlation-based transition model using local vari-
ables part I: model formulation. J Turbomach. 2006;128:413422.
[11] Langtry RB, Menter FR. Transition modeling for general CFD applications in aeronautics.
Paper presented at: AIAA Aerospace Sciences Meeting and Exhibit. 43rd Annual Meeting; 2005
January 1013; Reno.
[12] Menter FR, Langtry R, Volker S. Transition modeling for general purpose CFD codes. Flow
Turb Combust. 2006;77:277303.
[13] Wang C, Perot B. Prediction of turbulent transition in boundary layers using the turbulent
potential model. J Turbulence. 2002;3:N22 115.
[14] Edwards JR, Roy CJ, Blottner FG, et al. Development of a one-equation transition/turbulence
model. AIAA J. 2001;39(9):16911698.
[15] Walters DK, Leylek JH. A new model for boundary layer transition using a single-point RANS
approach. J Turbomachinery. 2004;126:193202.
[16] Mayle RE. The role of laminar-turbulent transition in gas turbine engines. J Turbomachinery.
1991;113:509537.
[17] Mayle RE, Schulz A. The path to predicting bypass transition. J Turbomachinery.
1997;119:405411.
[18] Walters DK, Cokljat D. A three-equation eddy-viscosity model for Reynolds-averaged navier-
stokes simulations of transitional flow. J Fluids Eng. 2008;130.
[19] Lopez M, Walters DK. Laminar-to-turbulent boundary layer prediction using an alternative
to the laminar kinetic energy approach. In: ASME 2012 International Mechanical Engineering
Congress & Exposition. Paper no. IMECE2012-89433; 2012 November 915; Houston.
[20] Walters DK. Physical interpretation of transition-sensitive RANS models employing the lami-
nar kinetic energy concept. ERCOFTAC Bull. 2009;80:6771.
JOURNAL OF TURBULENCE 21

[21] Durbin PA. Near-wall turbulence closure modeling without Damping functions. Theor
Comp Fluid Dyn. 1991;3:113.
[22] Parneix S, Durbin PA, Behnia M. Computation of 3-D turbulent boundary layers using the
V2F model. Flow Turbululence Combustion. 1998;60:1946.
[23] Lien FS, Kalitzin G, Durbin PA. RANS modeling for compressible and transitional flows. In:
Proceedings of the Stanford University Center for Turbulence Research Summer Program.
Stanford (CA): Stanford University Press; 1998. p. 267286.
[24] Voke PR, Yang Z. Numerical study of bypass transition. Phys Fluids. 1995;7:22562264.
[25] Jacobs RG, Durbin PA. Simulations of bypass transition. J Fluid Mech. 2001;428:185212.
[26] Brandt, L., Schlatter, P., Henningson, D.S. Transition in boundary layers subject to free-stream
turbulence. J Fluid Mech. 2004;517:167198.
[27] Zaki TA, Durbin PA. Mode interaction and the bypass route to transition. J Fluid Mech.
2005;531:85111.
Downloaded by [University of Nebraska, Lincoln] at 11:03 04 December 2015

[28] Ghahremanian S, Moshfegh B. Evaluation of RANS models in predicting low Reynolds, free,
turbulent round jet. J Fluids Eng. 2014;136:011201 113.
[29] Menter F. R. Improved two-equation k turbulence models for aerodynamic flows. NASA
Technical Memorandum, 103975, Ames Research Center, Moffett Field. 1992.
[30] Heschl C, Inthavong K, Sanz W, et al. Evaluation and improvements of RANS turbulence mod-
els for linear diffuse flows. Comput Fluids. 2013;71:272282.
[31] User Guide FLUENT 6.3. Centerra resource park, 10 cavendish court, Lebanon, NH 03766.
Canonsburg (PA): Ansys, Inc.; 2011.
[32] Coupland J. 1990. ERCOFTAC special interest group on laminar to turbulent transition and
retransition. T3A and T3B Test Cases.
[33] Radomsky RW, Thole KA. Flowfield measurements for a highly turbulent flow in a stator vane
passage. J Turbomachinery. 2000;122:255262.
[34] Radomsky RW, Thole KA. Detailed boundary-layer measurements on a turbine stator vane
at elevated freestream turbulence levels. ASME Paper No. 2001-GT-0169. 2001 June 47; New
Orleans.
[35] Driver DM, Seegmiller HL. Features of a reattaching turbulent shear layer in divergent channel
flow. AIAA J. 1985;23:163171.
[36] Hussein HJ, Capp SP, George WK. Velocity measurements in a high-Reynolds number,
momentum-conserving, axisymmetric, turbulent jet. J Fluid Mech. 1994;258:3175.

You might also like