Sunteți pe pagina 1din 8

A modular RANS approach for modelling laminarturbulent transition

in turbomachinery ows
Liang Wang
a,b,
, Song Fu
b
, Angelo Carnarius
a
, Charles Mockett
a,c
, Frank Thiele
a,c
a
Institute of Fluid Mechanics and Engineering Acoustics, Berlin University of Technology, Mller-Breslau-Str. 8, 10623 Berlin, Germany
b
School of Aerospace Engineering, Tsinghua University, Beijing 100084, China
c
CFD Software Entwicklungs- und Forschungsgesellschaft mbH, Wolzogenstr. 4, 14163 Berlin, Germany
a r t i c l e i n f o
Article history:
Received 10 October 2010
Received in revised form 1 September 2011
Accepted 16 January 2012
Available online 22 February 2012
Keywords:
Transition modelling
Elliptic approach
Intermittency factor
Turbomachinery
a b s t r a c t
In this study we propose a laminarturbulent transition model, which considers the effects of the various
instability modes that exist in turbomachinery ows. This model is based on a Kxc three-equation
eddy-viscosity concept with K representing the uctuating kinetic energy, x the specic dissipation rate
and c the intermittency factor. As usual, the local mechanics by which the freestream disturbances pen-
etrate into the laminar boundary layer, namely convection and viscous diffusion, are described by the
transport equations. However, as a novel feature, the non-local effects due to pressure diffusion are addi-
tionally represented by an elliptic formulation. Such an approach allows the present model to respond
accurately to freestream turbulence intensity properly and to predict both long and short bubble lengths
well. The success in its application to a 3-D cascade indicates that the mixed-mode transition scenario
indeed benets from such a modular prediction approach, which embodies current conceptual under-
standing of the transition process.
2012 Elsevier Inc. All rights reserved.
1. Introduction
It is well known that the aerodynamic performance of turboma-
chinery blades operating at chord-length-based Reynolds numbers
of less than one million, as is typical of aircraft cruise conditions, is
strongly dependent on the transition modes occurring on the blade
surface where a mixture of laminar, transitional and turbulent ow
occurs. The determination of the transitional region is thus a very
important task in the design process.
Currently, the Reynolds-Averaged NavierStokes (RANS) ap-
proach is still the main tool used for transition/turbulence model-
ling in engineering applications. It is well known that low Reynolds
number turbulence models, with the aid of damping functions to
characterise near-wall viscous effects on turbulence, have a certain
tendency to simulate typical transition proles, e.g. the sharp rise
in streamwise skin friction. However, it has recently been shown
by Rumsey et al. (2006) and Rumsey (2007) that using such mod-
els, the converged numerical solutions exhibit arbitrary depen-
dence on initial conditions or other solution parameters and can
be susceptible to pseudo-laminar states. These appear between
the leading edge and the transition onset, leading to the incorrect
conclusion that the low Reynolds number model somehow pre-
dicts the transition process when in fact, from a dynamical stand-
point, it does not. This conrms the viewpoint of the previous
review by Savill (1996), as summarised from calculation experi-
ences, that turbulence models which do not employ intermittency
prove to be very delicate and often extremely unreliable in the pre-
diction of transition.
As a result, many correlation-based transition models adopting
the intermittency concept have been proposed. The intermittency
factor, c, dened as the probability of the ow being turbulent in
a given spatial point, provides a general framework to model any
transition mechanism, from natural to by-pass and separation-
induced processes, as the streamwise distribution of c in the tran-
sitional region appears to be quite universal within a large range of
freestream Reynolds number and Mach number (Dhawan and
Narasimha, 1958; Mayle, 1996). Such distributions can be
modelled using either algebraic correlations (e.g. Dhawan and
Narasimha, 1958) or transport equations (e.g. Vicedo et al.,
2004). However, all such models must be coupled with a separate
transition-onset criterion involving non-local variables, such as the
momentum thickness, h, giving rise to serious implementation dif-
culties in modern CFD solvers. Models based on local variables are
thus highly preferable for the purposes of industrial applications.
A successful example is the work of Menter et al. (2006), which
has now been incorporated into commercial software packages. In
this model, the value of the transitional h-based Reynolds number,
which is used to determine the transition onset with an algebraic
criterion, is now obtained with a new transport equation instead
0142-727X/$ - see front matter 2012 Elsevier Inc. All rights reserved.
doi:10.1016/j.ijheatuidow.2012.01.008

Corresponding author at: Institute of Fluid Mechanics and Engineering Acous-


tics, Berlin University of Technology, Mller-Breslau-Str. 8, 10623 Berlin, Germany.
E-mail address: Liang.Wang@cfd.tu-berlin.de (L. Wang).
International Journal of Heat and Fluid Flow 34 (2012) 6269
Contents lists available at SciVerse ScienceDirect
International Journal of Heat and Fluid Flow
j our nal homepage: www. el sevi er. com/ l ocat e/ i j hf f
of being calculated by integration over the boundary layer. Another
type of local formulation related to triggering transition is based on
the concept of laminar uctuation energy, k
L
, introduced by Mayle
and Schulz (1997). A one-equation model is used to describe such
turbulent-like uctuations, which are very different from true tur-
bulent uctuations and have been demonstrated both in experi-
ment (Leib et al., 1999) and analysis (Jacobs and Durbin, 2001) in
the pre-transitional region. Recent examples of such models have
been proposed e.g. by Walters and Leylek (2002) and Lardeau
et al. (2004). It is noted that these methods still make use of the
wall distance, which is strictly a non-local variable. However, the
wall distance is routinely employed in industry-standard turbu-
lence models.
Such local-variable-based models are however not validated
for transition in supersonic ows or for crossow-induced
transition. One reason is that they rely strongly on empirical cal-
ibration rather than deduction from the fundamental physical
phenomena responsible for the actual transition process. For
example, the ow instability mode can vary considerably between
supersonic boundary layers and incompressible ows. A new
local-variable-based model formulated in terms of instability
modes has therefore been proposed recently, which can success-
fully simulate 3-D high-speed aerodynamic ow transition
(Fu et al., 2009; Wang and Fu, 2011). The goal of this work is to
extend this model to capture the bypass and separation-induced
transition effects common in separated ows, especially in turbo-
machinery applications.
Studies on separated ow transition have revealed multiple
transition modes that arise:
(1) Separation-induced mode: downstream of the separation
location (x
s
), due to the KelvinHelmholtz (KH) instability
of the separation velocity prole, transition occurs in the
free shear layer, whose turbulent part then entrains higher
momentum uid from the freestream more effectively than
the laminar shear layer, leading to reattachment of the tur-
bulent shear layer.
(2) Bypass mode: the freestream turbulence intensity (FSTI), e.g.
the periodic passage of wakes from upstream blade rows,
also affects transition and could even cause the boundary
layer to undergo transition ahead of any laminar separation.
(3) Natural mode: it is found by Hughes and Walker (2001) that
the ow transition on compressor blades is likely to be pre-
dominantly a natural growth process, rather than of the
bypass type.
Due to the complexity of such mixed-mode transition scenarios,
classical attempts to correlate the general transition-onset loca-
tion, x
tr
, with one or two parameters, such as FSTI and the dimen-
sionless pressure gradient, always encounter problems in cases
that deviate fromthose previously tested. Despite their lack of gen-
erality however, such empirical correlations for x
tr
are still
widespread.
Even if the x
tr
is accurately predicted, the calculation of transi-
tional separation using conventional turbulence models coupled
with c consistently leads to over-predicted bubble lengths (Mayle
and Schulz, 1997), because the turbulent spot formation rate here
is much larger than in pure bypass transition, which cannot be
identied by such models. To address this issue, many approaches
have recently been proposed. Some introduce additional source
terms to the transport equation for c (e.g. Vicedo et al., 2004),
while others force the distributions of c to exceed unity inside
the separated ow region (e.g. Koezulovic et al., 2007). These
methods pragmatically achieve the prediction of realistic reattach-
ment locations, however, they are not formulated on the basis of
physical reasoning.
Disturbances in the freestream ow in fact penetrate into the
laminar boundary layer not only by convection and viscous diffu-
sion, referred to as local effects in spectrum, but also by pressure
diffusion as the non-local mechanism (Mayle and Schulz, 1997).
This offers an explanation for the delayed response of intermit-
tency models to the freestream turbulence: they only consider
the transition triggered by the diffusion of the freestream turbu-
lence into the shear layer. Mayle and Schulz (1997) thus proposed
the transport equation for k
L
, as further rened by Walters and
Leylek (2002) and Lardeau et al. (2004), which considers the effect
of pressure diffusion and which appears to be a promising means
of simulating both the onset and length of transition.
However, it is doubtful whether the non-local effect, which does
not appear explicitly in any single-point model, can be modelled
with such a parabolic equation whose solution is only in the weak
form. Therefore, with reference to the work by Durbin (1993) in
which the non-local wall effect in the near-wall region is modelled,
an elliptic equation for c is proposed here. This is expected to pro-
vide a more natural way to introduce the effects of pressure uctu-
ation in the boundary layer: they enter into the boundary layer via
the solution of the governing equations, rather than via the trans-
verse variations derived from the empirical correlations. Conse-
quently, c is calculated by such an elliptic model coupled with
the original transport equation to take into account both the local
and non-local effects outlined above.
2. Model formulation
2.1. Modelling of the effective viscosity
In the existing study, the total uctuating kinetic energy, K, is
composed of two parts: one relating to the non-turbulent uctua-
tions, k
L
, and one to the turbulence k
T
, i.e. K = k
L
+ k
T
. Thus a sepa-
rate transport equation for k
L
is proposed, albeit in the same
manner as the conventional form for k
T
(Mayle and Schulz,
1997). However, the work of Rumsey et al. (2006) demonstrates
that the employment of two separate equations for k
T
and k
L
is
not necessary. It is found that a single equation for the total
uctuating kinetic energy K is more cost-effective, which is
adopted in the present work as
Nomenclature
a local sound speed
C
f
skin friction
c
r
phase velocity of disturbances
d wall distance, m
FSTI freestream turbulence intensity, FSTI = (2k/3)
0.5
/U

P
K
production term of K equation
Pr Prandtl number
S
ij
mean strain tensor
x
t
transition onset location, m
U

freestream velocity
c intermittency factor
f length scale in transitional ows
l
eff
effective viscosity
s
nt
characteristic timescale in the ow transition
L. Wang et al. / International Journal of Heat and Fluid Flow 34 (2012) 6269 63
@(qK)
@t

@(qu
j
K)
@x
j
= D
K


u
//
i
u
//
j
@
~
U
i
@x
j
..
P
K
e (1)
where D
K
, P
K
and e represent the energy diffusion, production and
dissipation terms, respectively. The Boussinesq hypothesis on the
second-order correlations is adopted in the same manner as in con-
ventional eddy-viscosity models. Thus, for the Reynolds stress, the
following constitutive relation holds:

u
//
i
u
//
j
= 2
l
eff
q

S
ij

1
3

S
kk
d
ij
_ _

2
3
d
ij
K (2)
Here, l
eff
is the effective viscosity that can be conjectured from the
following exact shear stress relationships

u
//
i
u
//
j
= c

u
/
i
u
/
j
(1 c)u
///
i
u
///
j
c(1 c)(

U
i
U
i
)(

U
i
U
j
) (ij)
(3)
where . and = denote the turbulent and the non-turbulent zone
averages, respectively. The rst and second terms on the right-hand
side stand for the momentum transport due to the turbulent and
non-turbulent uctuations, respectively, while the last term repre-
sents the mean velocity difference between the two uids that also
contributes to the momentumtransport by the bulk convection mo-
tion. Referring to the work by Cho and Chung for free shear ows
(1992), l
eff
is modelled as
l
eff
= (1 c)l
nt
cl
t
C
lg
q
K
2
x
3
c
3
(1 c)
@c
@x
k
@c
@x
k
(4)
where the subscript nt denotes the non-turbulent part in the effec-
tive viscosity and c bridges the non-turbulent and turbulent contri-
butions. C
lg
= 0.1 is the model coefcient given by Cho and Chung
(1992).
Since the modelling of the turbulent eddy viscosity l
t
can now
be considered well-established, the present work adopts the SST
model (Menter, 1994). However, other eddy-viscosity models
could readily be applied within this framework. Attention will be
focussed on the modelling of the non-turbulent uctuation, for
which the technique of McDaniel et al. (2000) is adapted that reads
l
nt
= C
l
qKs
nt
(5)
where C
l
is the model coefcient and s
nt
represents the character-
istic timescale associated with different instability-mode transition.
2.2. Modelling the characteristic timescales of the instability waves
Both experimental correlations and theoretical analyses (Arnal
and Casalis, 2000) demonstrate that the formulation of the time-
scale mentioned above would involve the boundary layer thick-
ness, which, regarded as non-local, is calculated via integration
over the boundary layer. In this study, a local-variable-based
length scale f is proposed as
f = d
2
X=(2E
u
)
0:5
(6)
Here, d is the distance to wall, Xthe absolute value of the mean vor-
ticity, and E
u
(=0.5 + |U|
2
) stands for the kinetic energy of the mean
ow relative to the wall. Because |U| is not a Gallilean invariant, this
formulation is not generally applicable to moving geometries.
The turbulence length scale l
T
adopts the conventional deni-
tion as K
0.5
/x and the bound of the length scale, l
B
= K
0.5
/(C
l
|S|),
is used to avoid the stagnation point anomaly (Medic and Durbin,
2002). The effective transition length scale f
eff
is then set as the
minimum value among f, l
T
and C
1
l
B
, where C
1
is a model constant.
Next, according to the stability analyses of the frequency of the
KelvinHelmholtz instability with the maximumamplication rate
in separated shear layers (Monkewitz and Huerre, 1982), its char-
acteristic timescale, s
sep
, can be set as
s
sep
= C
1
1:21(f
eff
=(2E
u
)
0:5
) (7)
Consequently, s
nt
is expressed as
s
nt
= s
nt1
s
nt2

1
2
[1 sgn(M
rel
1)[ s
cross
(W
s
) s
sep

1
2
[1 sgn(k
1
0:046)[ (8)
Here, the local relative Mach number M
rel
= (U c
r
)/a, a is the sound
speed and c
r
represents the phase velocity of disturbances with the
same value for all Mack modes. W
s
is the magnitude of the cross-
ow velocity perpendicular to the local inviscid streamline that is
generated by the combination of curvilinear inviscid streamlines
and the viscous no-slip condition at the wall (Reed and Saric,
1989). k
f
= (f
eff
)
2
/m(d|U|/ds) is the dimension-less pressure gradi-
ent parameter, sgn (x) = |x|/x the sign function. s
nt1
, s
nt2
and s
cross
stand for the timescales of rst-mode (Mack), second-mode (Mack)
and crossow instabilities, respectively, as:
s
nt1
= C
8
qf
1:5
eff
=[(2E
u
)
0:5
l[
0:5
(9)
s
nt2
= C
9
2f
eff
=[U[ (10)
tau
cross
= C
10
(4f
eff
=[U[) (W
s
=[U[)
0:5
exp C
11
(qf
eff
[U[=l 44)
2
_ _ _ _
(11)
However, since the test cases in the present paper are all sub-
sonic and non-swept, the second-Mack mode and the crossow
mode do not exert any inuence.
2.3. Calculation of the intermittency factor c
A transport equation for c has been developed by the authors as
@(qc)
@t

@(qu
j
c)
@x
j
=
@
@x
j
(l l
eff
)
@c
@x
j
_ _
P
c
e
c
(12)
Here P
c
and e
c
represent the intermittency production and dissipa-
tion term, respectively, which are modelled as follows:
P
c
= C
3
qF
onset

ln(1 c)
_
1 C
4

k
2E
u
_
_
_
_
d[\E
u
[
m
10
C
6
k
C
7
f
e
c
= cP
c
(13)
where the function F
onset
is set to trigger the onset of transition and
is given by
F
onset
= 1:0 exp C
5
f
eff
K
0:5
[\K[
m[\E
u
[
_ _
(14)
It is seen that F
onset
is determined by the development of K and
the mean ow in the pre-transitional region.
Next, we rename c as c
l
, where the subscript l stands for the lo-
cal. As mentioned in the introduction, the effect of disturbances pe-
netrating into the laminar boundary layer by pressure diffusion is
non-local, which is equivalent to say that the information speed is
innity. This means that a given point M
0
in space will feel the
inuence of all points inside the sphere S
0
of centre M
0
and with
a radius L
c
. L
c
is the correlation length scale of pressure effects
on the transition process.
With reference to Durbins model (1993) in which the non-local
wall effect in the near-wall region is modelled, we propose an
64 L. Wang et al. / International Journal of Heat and Fluid Flow 34 (2012) 6269
elliptic equation for c
nl
to represent the non-local pressure diffu-
sion effect as
c
nl
L
2
c
\
2
c
nl
= C
c
nl
(K=E
u
)
1=2
F
d
(15)
Here, the length scale L
c
= C
L
f
eff
, C
L
is a model constant, the sensor
function F
d
is used to turn off the source term in the boundary layer
and allows c
nl
to diffuse in from the freestream. F
d
is equal to zero in
the boundary layer and one in the freestream, as dened as follows:
F
d
= 1 tanh (l
eff
= qd
2
X)
3
_ _
(16)
The motivation for Eq. (15) is simply the notion that the non-lo-
cal effect might indirectly be represented by an elliptic model
equation, because the exact representation of non-local processes
would require knowledge of two-point correlations. Furthermore,
such a methodology can also reproduce the strong non-homogene-
ity of the near-wall region.
Consequently, c is obtained as the union of these two probabil-
ities as:
c = c
l
c
nl
= c
l
c
nl
c
l
c
nl
(17)
2.4. Summary of the present model
The present model consists of a transport equation for c
l
, i.e. the
local component of the intermittency factor, an elliptic equation
for the non-local component c
nl
, and the equations for uctuating
kinetic energy K and specic dissipation rate x that are modied
from the SST Kx eddy-viscosity model, all of which are listed as
follows:
D(qc
l
)
Dt
=
@
@x
j
l
l
eff
r
c
_ _
@c
l
@x
j
_ _
P
c
l
(F
onset
) e
c
l
c
nl
L
2
c
\
2
c
nl
= C
c
nl
(K=E
u
)
1=2
F
d
D(qK)
Dt
=
@
@x
j
l
l
eff
r
k
_ _
@K
@x
j
_ _
P
K
(l
eff
) e
D(qx)
Dt
=
@
@x
j
l
l
eff
r
x
_ _
@x
@x
j
_ _
P
x
e
x
Cd
x
(18)
Here, c = c
l
+ c
nl
c
l
c
nl
is set as a weighting function between the
non-turbulent and the turbulent components of the effective vis-
cosity l
eff
, i.e. Eq. (4). In the pre-transitional region, l
eff
~ l
nt
and
different instability modes dominate the development of K. The
developments of K and the mean ow then determine the value
of the transition-onset trigger F
onset
, i.e. Eq. (14), in the production
term of the c
l
equation. F
onset
rapidly goes from zero to unity after
the onset point. The increase of c is also affected by the FSTI in a
non-local manner via the solution of the elliptic equation for c
nl
.
In the transitional region, since l
nt
l
t
, the ow develops accord-
ing to the distribution of c. In the fully turbulent region, l
eff
= l
t
, the
SST model is recovered. All the model constant values are shown in
Table 1.
3. Results and discussion
The model proposed above has been calibrated and validated
for a reasonably wide range of transitional cases involving incom-
pressible ows past sharp and semicircular leading-edge at plates
and blades exposed to different levels of freestream turbulent
intensities (FSTI) either under zero or non-zero pressure gradients
(PG) and a 3D stator compressor cascade.
The transition behaviour of the model is rst tested for simple
zero-PG at-plate (sharp leading-edge) cases, in order to assess
the response to FSTI, and to compare the prediction of transition
onset and length with experimental data and available empirical
correlations. Following this, the zero-PG semicircular leading-edge
at-plate cases (where the natural mode is fairly weak compared
to the separation-induced and bypass modes) are considered to
calibrate the model constants quoted in Sections 2.1 and 2.2. Next,
to assess the applicability of the model to more realistic geome-
tries, at plates with adverse pressure gradients and exposed to
varying FSTI are computed. Finally, the steady RANS simulation
of a 3D stator compressor cascade is performed.
All of the simulations presented here are performed using an
in-house code where the RANS equations are solved on the non-
staggered H-type and O-type grid systems. The convection and
diffusion terms in RANS are discretised with the UMIST-TVD
scheme (Liou, 1996) and the central difference scheme respec-
tively. To solve the pressure eld, the SIMPLE algorithm is used
and in order to eliminate the pressure uctuation associated with
the use of a non-staggered grid system, a momentum-interpolation
technique is utilised to calculate the velocity on the nite volume
faces.
For all test case geometries, an initial mesh and a second mesh
rened by a factor of 1.5 in each direction are constructed with a
rst cell y
+
value of one or less and a structured body-tted mesh
in the boundary layer region. In all cases, steady-state solutions
have been obtained on the both meshes, which shownegligible dif-
ference and are therefore judged to be mesh independent. All cases
are run to full convergence, determined based on a drop in residu-
als of typically ve orders of magnitude, as well as a attening of
all residuals indicating that machine accuracy has been reached.
The various results reported herein correspond to the four dif-
ferent models employed during the study. Results labelled SST cor-
respond to those predictions in which no transition-prediction tool
is used. Those of the present transition model are labelled SST-c
l
-c
nl
and those when the elliptic approach is excluded SST-c
l
. Finally, the
results labelled by SST-I have been obtained using the standard SST
model with a manually-specied transition onset, x
tr
, obtained
from the SST-c
l
-c
nl
results, i.e. with l
eff
= 0 upstream while l
eff
= l
t
downstream of x
tr
.
3.1. Sharp leading-edge at plate with zero pressure gradient
The test cases chosen include the T3 series experiments (Coup-
land, 1990a,b) from the European Research Consortium on Flow,
Turbulence and Combustion (ERCOFTAC) database, and the well
Table 1
Model constants.
C
l
C
1
C
2
C
3
C
4
C
5
C
6
0.09 0.7 0.8 8e5 0.07 1.2 0.03
C
7
C
8
C
9
C
10
C
11
C
cnl
C
L
0.6 0.46 0.005 1e3 5.0 0.2 0.1
Table 2
Flow inlet conditions and computed transition-onset locations for the sharp leading-
edge at plates with zero pressure gradient.
Case S&K T3A T3B T3A-
U inlet (m/s) 50.1 5.4 9.4 19.8
l
t
/l 5.0 12.5 100.0 8.72
FSTI (%) inlet 0.18 3.5 6.5 0.874
Computed x
tr
(mm) (SST-c
l
-c
nl
) 0.873 0.466 0.105 1.243
Computed x
tr
(mm) (SST-c
l
) 0.873 6.255 3.328 2.221
Measured x
tr
(mm) 0.86 0.46 0.10 1.31
L. Wang et al. / International Journal of Heat and Fluid Flow 34 (2012) 6269 65
known Schubauer and Klebanof (S&K) experiment (1955), all of
which were conceived specically for the validation of transition
models and have become a recognised standard in the research
community. The T3 series (T3A-, T3A, and T3B) cases have zero
pressure gradient with freestream turbulence levels of 0.874%,
3.5% and 6.5% that correspond to transition in the bypass regime.
The S&K test case has low freestream turbulence intensity and cor-
responds to purely natural transition. The inlet quantities for all
the cases computed, besides q = 1.2 kg/m
3
, l = 1.8e5 kg/ms, are
summarised in Table 2. Here the inlet values of l
t
/l are chosen
in order to correctly reproduce the streamwise decay of freestream
turbulence reported in the experiments. An example of typical
agreement in the freestream is shown in Fig. 1a and this step has
been carried out for all the cases individually.
Fig. 1b gives the streamwise skin friction coefcient for the test
case T3B. It is seen that the ow transition prole is well captured
with the present model, while this is missed entirely when the
elliptic approach is excluded.
The transition-onset locations, x
tr
, as dened by the local min-
ima of shear stress in the skin friction distribution, are listed for
all cases in Table 2. It is seen that the intermittency model coupled
with the elliptic approach shows a signicantly improved perfor-
mance in the sensitivity of x
tr
to FSTI, with excellent agreement
with the experimental data. In contrary, exclusion of the elliptic
c
nl
equation leads to a weak sensitivity to FSTI and the transition
occurs approximately when Re
x
exceeds a critical value. This dem-
onstrates that the additional modelling of the pressure diffusion of
freestream turbulence into the boundary layer via the elliptic for-
mulation is responsible for the improvements achieved.
3.2. Semicircular leading-edge at plate with zero pressure gradient
The test cases chosen match the T3L1 to T3L4 validation cases
from the ERCOFTAC database (Coupland, 1995). The ow condi-
tions specied are listed in Table 3. Here, the xed Reynolds num-
ber, Re
d
, is based on the leading-edge diameter (0.01 m) and the
freestream velocity U

= 5 m/s. The different FSTI values are mea-


sured at 0.006 m downstream of the leading edge.
Figs. 2 and 3 show the separated ow predicted by means of
streamlines and turbulence kinetic energy contours for the T3L1
and T3L4 cases, respectively. It is rstly seen that the onset of sep-
arated ow is detected at the point where the curved surface
merges with the horizontal plate, at roughly x
s
= 0.0048 m, which
remains approximately constant for all cases.
For the T3L1 case, the SST model alone predicts the transition
onset at x
tr
= 0.0110 m, which is in clear disagreement with the va-
lue of x
tr
= 0.0237 m obtained by the proposed transition model. In
this model calculation, as a consequence of the x
tr
further down-
stream, the turbulence generated due to the shearing effect of
the separated ow takes longer to develop. Reattachment thus oc-
curs further downstream (x
r
= 0.0390 m) than with the baseline
SST model, resulting in an 85% increase in bubble length and im-
proved agreement with the experimental data. Furthermore, since
the bypass mode in this case is fairly weak compared to the sepa-
ration-induced mode, the good agreement with experiment indi-
cates that the modelling of s
sep
, i.e. Eq. (7), is effective.
x (m)
F
S
T
I
(
%
)
0.2 0.4 0.6 0.8 1
Experimental data
Present calculation
T3B flat plate
Re

= 6.26E5 m
-1
FSTI = 6.5%
(a)
Re
x
C
f
200000 400000 600000 800000
SST -
l
-
nl
Experimental data
SST -
l
FSTI = 6.5% T3B flat plate
Re

= 6.26E5 m
-1
(b)
2.5
3
3.5
4
4.5
5
5.5
0.001
0.002
0.003
0.004
0.005
0.006
Fig. 1. Streamwise decay of the freestream turbulence intensity (a) and skin friction
distribution (b) for the test case T3B.
Table 3
Flow inlet conditions and computed transition-onset and reattachment locations for
the semicircular leading-edge at plates with zero PG. Here, (V) stands for the
calculations by Vicedo et al. (2004).
Case T3L1 T3L2 T3L3 T3L4
Re
d
3293 3293 3293 3293
FSTI (%) inlet 0.17 0.63 2.39 5.34
Computed x
tr
(mm) 23.7 20.2 18.1 16.0
Computed x
tr
(mm) (V) 12.4 12.2 11.8
Measured x
r
(mm) 40 27 23 21
Computed x
r
(mm) 39.0 29.2 24.2 20.5
Computed x
r
(mm) (V) 28.3 24.3 21.2
X (m)
Y
(
m
)
0.01 0.02 0.03
Level
k / U

2
:
1
0.005
3
0.015
5
0.025
7
0.035
9
0.045
11
0.055
Experimental data (T3L1):
Re

= 3.3E5 m
-1
FSTI = 0.17%
x
s
= 0.0049 m
x
r
= 0.040 m
(a)
SST -
l
-
nl
X (m)
Y
(
m
)
0 0.01 0.02 0.03
Level
k / U

2
:
1
0.003
3
0.009
5
0.015
7
0.021
9
0.027
11
0.033
13
0.039
Experimental data (T3L1):
Re

= 3.3E5 m
-1
FSTI = 0.17%
x
s
= 0.0049 m
x
r
= 0.040 m
(b)
SST
-0.005
0
0.005
0.01
-0.005
0
0.005
0.01
Fig. 2. Streamline (above) and turbulent kinetic energy contours (below) for (a) the
present model and (b) the baseline SST model.
66 L. Wang et al. / International Journal of Heat and Fluid Flow 34 (2012) 6269
Turning to the T3L4 case, despite the manual specication of the
transition onset to the same location, the extent of the separated
ow given by the present model is signicantly smaller than that
of the SST turbulence model. This indicates that the strongly accel-
erated turbulent spot formation in the transitional region is prop-
erly simulated by the consideration of the entrainment mechanism
in the l
eff
modelling, i.e. Eq. (4).
Finally, Table 3 summarises the computed x
tr
and x
r
and mea-
sured x
r
for all cases. The proposed model shows fairly good agree-
ment with experiment, and with an increase in FSTI, the transition
onset obtained by this model is shifted signicantly upstream
whereas that predicted by a traditional Kxc model (Vicedo
et al., 2004), marked as V in Table 3, varies only slightly. In the lat-
ter model, additional source terms are introduced to the transport
equation for c, which results in excessive production of turbulent
kinetic energy, hence forcing the boundary layer to reattach at
the location measured.
3.3. Flat plate with non-zero pressure gradient
In Section 3.2, the proposed model was shown to predict well
the short bubble that has a local effect on the ow and only slightly
affects the inviscid ow outside the bubble. Here, we assess its pre-
dictive capability for the long bubble, which by contrast exhibits a
signicant interaction with the external ow, such that the pres-
sure distribution deviates markedly from the inviscid case.
The test cases chosen include the T3C2T3C5 (at plates with
adverse PG) experiments from the ERCOFTAC database (Coupland,
1990a,b). The computational domain for T3 cases is constructed
with a contoured upper surface, where the contour is chosen to
match the experimental pressure distribution on the at plate.
The inlet quantities are summarised in Table 4.
Comparisons of computed and measured skin friction for the
T3C4 case is given in Fig. 4. The model results for the T3C4 case
indicate a laminar boundary layer separation due to the adverse
pressure gradient, followed by the transition and reattachment of
the boundary layer. All results show generally good agreement
with the experimental data.
3.4. 3D stator compressor cascade
The stator cascade chosen corresponds to the low speed cascade
test section at the Department of Aeronautics and Astronautics of
the Berlin Institute of Technology (Zander et al., 2009). The highly
loaded Controlled Diffusion Airfoil (CDA) has a chord length of
L = 0.375 m. The stagger angle is h = 20. In the experiment the
transition occurs in the free shear layer on its suction side, which
determines whether or not the separation bubble will reattach as
a turbulent boundary and, ultimately, whether or not the blade will
stall. Moreover, due to the relatively small pitch to chord ratio of
T/L = 0.4, the high turning angle of up to Db = 60, and the low
aspect ratio of h/L = 0.8, strong secondary ow structures are
observed in the experiment. An overview of the stator cascade
parameters is shown in Fig. 5.
The oil-ow visualisation in Fig. 6a shows the general ow pat-
tern over the projected suction side: the laminar boundary layer
separates at the blade surface S = 0.17, forming a quasi-2D laminar
separation bubble; the separated ow then undergoes transition,
reattaching on the blade suction surface as a turbulent boundary
layer (approximately 0.25S); downstream of the turbulent reat-
tachment, the ow is constricted by the occurrence of the corner
vortices at the end walls and a 3D separation line is formed be-
tween the secondary and the main ow, ending up at midspan
(0.7S).
Fig. 6b compares the calculated wall streamlines to the under-
lying oil-ow visualisation. It is seen that the laminar ow separa-
tion and the reattachment, even in the near-end-wall region, are
predicted well. The turbulent separation position at midspan is in
contrast predicted much later than the experiment (0.99S com-
pared to 0.7S). The cause for this is believed to be the underlying
linear eddy viscosity turbulence models, which are known to
over-predict the strength of such corner vortices. This in turn leads
X (m)
Y
(
m
)
0 0.01 0.02 0.03
Level
k / U

2
:
1
0.005
3
0.015
5
0.025
7
0.035
9
0.045
11
0.055
Experimental data (T3L4):
Re

= 3.3E5 m
-1
FSTI = 5.34%
x
s
= 0.0049 m
x
r
= 0.021 m
(a)
SST -
l
-
nl
X (m)
Y
(
m
)
0.01 0.02 0.03
Level
k / U

2
:
1
0.005
3
0.015
5
0.025
7
0.035
9
0.045
11
0.055
Experimental data (T3L4):
Re

= 3.3E5 m
-1
FSTI = 5.34%
x
s
= 0.0049 m
x
r
= 0.021 m
(b)
SST - I
-0.005
0
0.005
0.01
-0.005
0
0.005
0.01
Fig. 3. Streamline (above) and turbulent kinetic energy contours (below) for (a) the
present transition model and (b) the SST-I model.
Table 4
Flow inlet conditions and computed transition-onset locations for the sharp leading-
edge at plates with non-zero pressure gradients.
Case T3C2 T3C3 T3C4 T3C5
FSTI (%) inlet 3.0 3.0 3.0 3.0
l
t
/l 11.0 6.0 8.0 15.0
Computed x
tr
(m) 0.834 1.173 1.173 0.352
Measured x
tr
(m) 0.795 1.195 1.395 0.345
Relative error of x
tr
(%) 4.90 1.88 2.50 2.03
Re
x
C
f
50000 100000 150000
-0.002
0
0.002
0.004
0.006
0.008
0.01
Experimental data
Present calculation
FSTI = 3% T3C4 flat plate
Fig. 4. Comparison of computed and measured skin friction (C
f
) for T3C4 case.
L. Wang et al. / International Journal of Heat and Fluid Flow 34 (2012) 6269 67
to excessive induced downwards ow at midspan and delayed sep-
aration. Future research will thus focus on predicting such turbu-
lent separated-ow region using Detached-Eddy Simulation
(Spalart et al., 2006) where the present transition model functions
in the RANS zones. Promising results for Detached-Eddy Simula-
tion combined with manually-prescribed transition have already
been obtained for this ow (Steger et al., 2011).
Fig. 7 shows the contours of dimensionless turbulent kinetic en-
ergy and intermittency factor at midspan. It is seen that K ramps up
when c exceeds 510%. The transition starts at 0.21S on the suction
side and at 0.08S on the pressure side, both of which agree well
with the experimental data.
4. Conclusions
A local-variable-based Kxc three-equation transition/turbu-
lence model considering different instability modes is proposed
and validated for a relatively wide range of transitional ow condi-
tions corresponding to cases of practical relevance to turbomachin-
ery ows. It takes into account not only the local effects by which
the disturbances penetrate into the laminar boundary layer,
namely convection and viscous diffusion as described by a trans-
port equation for c, but also the non-local effect by pressure diffu-
sion, as represented by an elliptic approach. This model, which
now additionally considers bypass and separation-induced transi-
tion effects, is an extension to the original transition-sensitive
model (Fu et al., 2009; Wang and Fu, 2011), which itself takes into
account 3-D high-speed aerodynamic ow transition by the mod-
elling of natural and crossow modes.
The results show generally good agreement with experimental
data, which applies both to global quantities such as separation
length and transition-onset location, as well as for local velocity
proles and turbulent intensity contours. For the zero-PG sharp
leading-edge at-plate cases, the elliptic approach results in a
proper response to FSTI. For the semicircular leading-edge cases,
the modelling of the separation-induced mode and the entrain-
ment in the transitional region are both effective. For the
nonzero-PG cases, all new modelling components perform reason-
ably well. For the 3D stator cascade case, the present model pre-
dicts fairly accurate onset and reattachment positions of the
laminar separated ow on the suction side, and transition onset
locations on both sides. The conclusion can therefore be made that
the mixed-mode transition scenario benets from such a modular
prediction approach that is based on the current conceptual under-
standing of the transition process.
With regard to the use of the intermittency model coupled with
the elliptic approach, as opposed to production term modications
and prescribed intermittency models (Vicedo et al., 2004), the
present approach has shown an improved performance when com-
pared to earlier work on the same test case. The differences be-
tween both approaches demonstrate that the inclusion of the
intermittency model for the pressure diffusion of freestream tur-
bulence into the boundary layer is the reason behind the improve-
ments achieved.
In short, the model is based on an extremely simplied view of
transition physics, but can nonetheless make useful predictions by
Fig. 5. Geometry of the stator cascade.
Fig. 6. General ow pattern over the projected suction side of the cascade. (a) Oil-
ow visualisation and (b) superimposed simulated wall streamlines to (a).
Fig. 7. Contours of dimensionless turbulent kinetic energy (top) and intermittency
factor (bottom) at midspan.
68 L. Wang et al. / International Journal of Heat and Fluid Flow 34 (2012) 6269
relying on a limited amount of simple and reliable experimental
data. It suggests that the physics-based approach adopted here al-
lows engineers to signicantly extend RANS-based computational
analysis capability by providing realistic transitional modelling in
a relatively simple eddy-viscosity framework. Moreover, such a
framework has an inherent potential to extend directly to De-
tached-Eddy Simulation that can well resolve the both the bound-
ary-layer and the free shear ows.
Acknowledgements
The investigations presented in this paper have been obtained
within the European research Project TATMo (Turbulence and
Transition Modelling for Special Turbomachinery Applications).
References
Arnal, D., Casalis, G., 2000. Laminarturbulent transition prediction in three-
dimensional ows. Prog. Aerospace Sci. 36, 173191.
Cho, J.R., Chung, M.K., 1992. A kec equation turbulence model. J. Fluid Mech. 237,
301322.
Coupland, J., 1990. ERCOFTAC Special Interest Group on Laminar to Turbulent
Transition and Retransition: T3A and T3B Test Cases.
Coupland, J., 1990. ERCOFTAC Special Interest Group on Laminar to Turbulent
Transition and Retransition: T3C Test Cases.
Coupland, J., 1995. Transition Modelling for Turbomachinery Flows. ERCOFTAC
Bulletin 24, pp. 58.
Dhawan, S., Narasimha, R., 1958. Some properties of boundary-layer ow during
transition from laminar to turbulent motion. J. Fluid Mech. 3 (4), 414436.
Durbin, P.A., 1993. A Reynolds stress model for near-wall turbulence. J. Fluid Mech.
249, 465498.
Fu, S., Wang, L., Carnarius, A., Mockett, C., Thiele, F., 2009. Modelling supersonic and
hypersonic ow transition over three-dimensional bodies. In: 7th IUTAM
Symposium on LaminarTurbulent Transition. Springer.
Hughes, J.D., Walker, G.J., 2001. Natural transition phenomena on an axial
compressor blade. J. Turbo-mach. 123, 392400.
Jacobs, R.G., Durbin, P.A., 2001. Simulations of bypass transition. J. Fluid Mech. 428,
185212.
Koezulovic, D., Roeber, T., Nuernberger, D., 2007. Application of a multimode
transition model to turbomachinery ows. In: 7th European Turbomachinery
Conference. Springer.
Lardeau, S., Leschziner, M., Li, N., 2004. Modelling bypass transition with low-
Reynolds-number nonlinear eddy-viscosity closure. Flow Turbul. Combust. 73,
4976.
Leib, S.J., Wundrow, D.W., Goldstein, M.E., 1999. Effect of free-stream turbulence
and other vortical disturbances on a laminar boundary layer. J. Fluid Mech. 380,
169203.
Liou, M.S., 1996. A sequel to AUSM: AUSM+. J. Comput. Phys. 129, 364382.
Mayle, R.E., 1996. Transition in separation bubble. J. Turbomach. 118, 752759.
Mayle, R.E., Schulz, A., 1997. The path to predicting bypass transition. J. Turbomach.
119, 405411.
McDaniel, R.D., Nance, R.P., Hassan, H.A., 2000. Transition onset prediction for high-
speed ow. J. Space Rockets 37, 304309.
Medic, G., Durbin, P.A., 2002. Toward improved prediction of heat transfer on
turbine blades. J. Turbomach. 124, 187192.
Menter, F.R., 1994. Two-equation eddy-viscosity turbulence models for engineering
applications. AIAA J. 32, 15981605.
Menter, F.R., Langtry, R., Volker, S., 2006. Transition modelling for general purpose
CFD codes. Flow Turbul. Combust. 77, 277303.
Monkewitz, P.A., Huerre, P., 1982. Inuence of the velocity ratio on the spatial
instability of mixing layers. Phys. Fluids 25, 11371143.
Reed, H.L., Saric, W.S., 1989. Stability of three-dimensional boundary layers. Ann.
Rev. Fluid Mech. 21, 235284.
Rumsey, C.L., 2007. Apparent transition behavior of widely-used turbulence models.
Int. J. Heat Fluid Flow 28, 14601471.
Rumsey, C.L., Pettersson-Reif, B.A., Gatski, T.B., 2006. Arbitrary steady-state
solutions with the ke model. AI AA J. 44, 15861592.
Savill, A.M., 1996. Turbulence and transition model-ling. In: Turbulence and
Transition Modelling. McGraw.
Schubauer, G.B., Klebanof, P.S., 1955. Contributions on the Mechanics of Boundary
Layer Transition. NACA TN 3489.
Spalart, P.R., Deck, S., Shur, M.L., Squires, K.D., Strelets, M., Travin, A., 2006. A new
version of detached-eddy simulation, resistant to ambiguous grid densities.
Theor. Comput. Fluid Dyn. 20, 181195.
Steger, M., van Rennings, R., Gmelin, C., Thiele, F., Huppertz, A., 2011. Detached-
eddy simulation of a highly loaded compressor cascade with laminar separation
bubble. In: 9th European Conference on Turbomachinery, Turkey.
Vicedo, J., Vilmin, S., Dawes, W.N., Savill, A.M., 2004. Intermittency transport
modelling of separated ow transition. J. Turbomach. 126, 424431.
Walters, D.K., Leylek, J.H., 2002. A new model for boundary layer transition using a
single-point RANS approach. J. Turbomach. 126, 193202.
Wang, L., Fu, S., 2011. Development of an intermittency equation for the modeling
of the supersonic/hypersonic boundary layer ow transition. Flow Turbul.
Combust. 87, 165187.
Zander, V., Hecklau, M., Nitsche, W., Huppertz, A., Swoboda, M., 2009. Active control
of corner vortices on a highly loaded compressor cascade. In: 8th European
Turbomachinery Conference, Graz.
L. Wang et al. / International Journal of Heat and Fluid Flow 34 (2012) 6269 69

S-ar putea să vă placă și