Académique Documents
Professionnel Documents
Culture Documents
crossow
v
2
crossow
. (1)
180 J.U. SCHLTER AND T. SCHNFELD
Figure 3. Vortex system for a jet in crossow (from Fric and Roshko [13]).
In most cases the jet uid and the crossow uid consist of the same species and
have the same temperature, hence
jet
=
crossow
. The momentum ratio simplies
then to the velocity ratio:
r =
v
jet
v
crossow
. (2)
Different ow regimes can be determined based on the velocity ratio r. JICF
with r < 0.5 play a special role. The jet ow is weaker than the crossow. It is not
able to break through the wall boundary layer of the crossow and plays more the
role of an obstacle for the crossow. The far eld of the jet is primarily governed
by the oncoming boundary layer. This fact introduces a large number of additional
parameters to the investigations. Dependent on the wall boundary layer thickness
this ow regime can occur up to a velocity ratio of r = 1. This ow conguration
is especially important for turbine blade coolings [7].
Velocity ratios r between 1 and 10 are common ow regimes for combustion
applications. The jet is then able to push through the boundary layer, which plays
a minor role. The JICF is now determined by free turbulence characteristics and
is easier reproducible. Andreopoulos [5] measured velocity proles for JICF with
velocity ratios from 0.5 to 2, where the described transition from wall boundary
layer to free turbulence can be seen.
JICF with velocity ratios higher than 10 have additional effects as they behave
more and more like free jets with increasing velocity ratio.
LES OF JETS IN CROSS FLOW 181
Figure 4. Left: jet shear layer vortices and trajectory of the horseshoe vortex, right: horseshoe
vortex and recirculation bubble at the upstream edge of the jet orice (from Kelso [18]).
2.3. VORTEX SYSTEMS
The mechanism for the formation of the counterrotating vortex pair (CVP) is not
fully understood. It can be taken as certain, that the vorticity of the CVP has its
origin at the sidewalls of the jet. Haven [14] investigated different nozzle geome-
tries for the jet. Rectangular jets with a low aspect ratio (the edges at the sides of
the jet are longer than the upstream and downstream edge) amplied the CVP. The
longer sidewalls produce more vorticity, which can be found later in the CVP. On
the other side, jets with a high aspect ratio have a weaker CVP.
Toy [30] investigated two closely spaced jets issuing into a crossow. The jets
were either side-by-side or in-line. Hot-wire measurements on the centerline of
the jets were made. As a major feature found, only one CVP, instead of two,
was produced, when the jets were in a side-by-side setup. The data obtained, was
not sufcient to give a reason for this behaviour. To the authors knowledge, a
mechanism which leads to a single CVP is not reported in literature.
The jet shear layer vortices are produced directly at the jet orice (Figure 4).
The two streams (jet stream and crossow) form a mixing layer with a Kelvin
Helmholtz instability, which causes a roll-up near the edges of the jet.
The horseshoe vortex forms upstream of the jet at the crossow wall. The
adverse pressure gradient at the crossow wall forces the wall boundary layer
to separate and to form a vortex. It is then convected and stretched by the ow
and wraps around the jet nozzle like a necklace. The same kind of vortices can
be observed for ows where a boundary layer hits on an obstacle, e.g. a cylinder
mounted on a wall [6]. The vortex is in interaction with the upstream edge of the
jet orice, causing a separation bubble inside the jet (see also [17, 20]). Figure 4
shows a sketch of the horseshoe vortex and the separation bubble.
The existence of the separation bubble was rst proposed by Andreopoulos [2].
He measured the velocities in the interior of the jet-pipe and found non-symmetric
proles. In this investigation, the crossow affects the jet-pipe ow up to 2D up-
stream of the nozzle.
182 J.U. SCHLTER AND T. SCHNFELD
The wake vortices were rst believed to be the consequence of a shedding
process behind the jet, with the jet acting like a solid cylinder and the wake vortices
behaving like a von Krmn vortex street. However, Perry [20] pointed out, that in
an incompressible uid no vorticity can be produced inside the ow. The vorticity
transport equation for incompressible uids:
D
Dt
= u +
2
(3)
shows a convection term on the left-hand side, a term which describes the reori-
entation and stretching of the vorticity due to the velocity gradients and a viscous
term on the right-hand side, but no explicit production term. That means, in incom-
pressible ows vorticity can only enter a ow by initial conditions and imposed
wall boundaries, but not be generated in the interior, which would be the case, if
the wake vortices would be shed from the jet like from a solid cylinder. Fric and
Roshko [13] connected the wake vortices to a separation event at the cross ow
wall at the sides of the jet nozzle. The end of the vortex string, which is close to
the nozzle is convected by the jet and follows the jet trajectory, while the other end
stays close to the wall, bringing the vortex in an upright position.
2.4. PREVIOUS NUMERICAL INVESTIGATIONS OF THE JICF
The rst numerical investigation of a JICF has been made by Sykes [29]. He
simplied the calculation by using a slip wall as the crossow wall. Hence, the
crossow boundary layer is neglected and the horseshoe vortex and the wake vor-
tices cannot be calculated. His results agree qualitatively with the measurements of
Andreopoulos [3, 5], besides a large error near the wall.
A numerical investigation from Chiu [11] tested the applicability of different
algebraic turbulence models with the result, that the turbulence models do not
improve the calculations compared with a laminar calculation. It showed the limits
of an eddy viscosity model applied to free turbulence.
The investigation of Alvarez [1] used the k model and a direct closure. The
direct closure improved the calculations, but the error was still high.
The strong unsteady behaviour of the JICF leads to the conclusion, that an
unsteady LES approach might be useful for reproducing and predicting the JICF.
Yuan [32, 33] made an LES calculation of a JICF. He calculated a JICF with a
small Reynolds number (Re = 2100) on meshes with 1.3 million mesh points.
The results agree quite well with measurements, although no direct comparison
is possible , because the Reynolds number is lower and the velocity ratio is not
exactly the same as in the experimental references.
He pointed out, that it is necessary to mesh the pipe which supplies the jet, in his
case on a length of one diameter D upstream. But, as Andreopoulos [2] mentioned,
there is an interaction between the crossow and the jet-ow. In the examined case
(r = 2) the crossow affects the pipe-ow around two diameters upstream the
LES OF JETS IN CROSS FLOW 183
pipe, so that his extension was probably not sufcient. The high number of mesh
points and the low Reynolds number make it difcult to apply this computation to
industrial congurations.
3. Mathematical Formulation of LES
3.1. GOVERNING EQUATIONS
The basic idea of LES is to resolve the larger scales of motion of the turbulence
while approximating the smaller ones. To achieve this, a lter is applied to the
continuity equation and the transport equations of momentum, energy and species.
For reacting ows, often Favre ltering is used, which is dened as:
Q = Q =
+
_
Q(x, t )G(x x
) dx
, (4)
leading to the following equations for momentum u
i
and species Y
i
:
momentum (j = 1, 2, 3)
u
i
t
+
u
i
u
j
x
i
+
p
x
i
=
ij
x
j
+
T
ij
x
j
; (5)
species mass fraction (k = 1, . . . , N)
Y
k
t
+
u
i
Y
k
x
i
=
x
i
_
D
k
Y
k
x
i
_
+
k
+
kj
x
j
. (6)
3.2. SUBGRID SCALE MODELS
The terms T
ij
and
ik
result from the convective terms u
i
u
j
and
Y
k
u
i
, which are
split into a resolved part on the left-hand side of the equation, directly delivered by
the LES calculation, and an unresolved part on the right-hand side, which needs to
be modeled.
We used an eddy viscosity approach for the subgrid scales:
T
ij
= 2
t
S
ij
+
1
3
T
ll
ij
(7)
with
S
ij
=
1
2
_
u
i
x
j
+
u
j
x
i
_
. (8)
184 J.U. SCHLTER AND T. SCHNFELD
Although the eddy viscosity approach is not valid for free turbulence, its simplic-
ity allows faster computations and by this a higher spatial discretization and an
increase of the resolved part of the spectrum.
Subgrid mixing is modeled by an eddy diffusity approach with a turbulent dif-
fusity based on the turbulent viscosity
t
of the subgrid stress model and a constant
Schmidt number Sc:
k
j =
t
Sc
Y
k
x
j
. (9)
3.3. PRESENT IMPLEMENTATION
For our LES calculations we used the AVBP parallel solver developed at CERFACS
and the Oxford University [21], based on the parallel library COUPL [22]. The
program handles structured and unstructured meshes and is second-order accurate
in space and time.
Two models were used to determine the eddy viscosity
t
. The rst one is the
Standard Smagorinsky Model [23]:
t
= (C1
x
)
2
_
2
S
ij
S
ij
(10)
with C
1
= 0.18, which has the advantage of simplicity and speed.
The second model is the Filtered Smagorinsky model [12] dened on a high-
pass lter HP:
t
= (C2
x
)
2
_
2HP(
S
ij
)HP(
S
ij
) (11)
and a constant C
2
= 0.37. This model offers a better behaviour in transitional ows
and was optimized to work in wall boundary layers.
4. Grid Resolution
The spatial discretization of the ow is based on structured meshes. Although the
AVBP solver allows unstructured meshes, the structured mesh approach provides
a better control of the point distribution in the ow. Figure 5 shows an exploded
view of the mesh. At the bottom is the plenum chamber of the jet (A) passing over
into a pipe (B). The jet nozzle is at the upper end of the pipe. An 0-grid is put in
the jet trajectory and the vicinity of the nozzle (C). A block behind the nozzle (D)
describes the ow downstream of the nozzle and several coarse blocks (E) are put
around the jet trajectory to mesh the nearly undisturbed outer ow.
Meshing the jet pipe ow is important. As already pointed out, there is an inu-
ence of the crossow to the jetow in the pipe. The existence of a recirculation zone
at the edge of the jet nozzle (see Figure 4) makes this area sensible to mesh point
distribution. The mesh has to be ne enough to capture this recirculation zone.
LES OF JETS IN CROSS FLOW 185
Figure 5. Exploded view of grid blocks.
In the LES computation of Yuan [33] the inuence of an extension of the mesh
into the jet pipe was examined. He found out, that the ow behaviour is much better
reproduced with such a mesh extension.
We found out, that a simple extension might not be sufcient. Our rst calcula-
tions have been carried out with a 3D long pipe leading to the jet nozzle and the
velocity prole u was imposed at the entry of the pipe. This led to strong pressure
oscillations in the pipe. As a numerical artefact, the pipe acts as a Helmholtz res-
onator, because the inlet below the wall forms a velocity node. The frequency of
the oscillations are determined by the length of the pipe. The jet shear layer roll-up
locks into the oscillations and the jet acts like a forced jet. Kelso [17] found, that
the jet trajectory is affected by the forcing. In our computations the trajectory is
higher than in the case where the oscillations are suppressed.
In order to avoid pressure oscillations we use a combination of two counter-
measures. The rst one is to impose the mass ux u instead of the velocity u
as a boundary condition to change the acoustic wave reections at the inlet. The
second is to extend the jet pipe mesh into the plenum chamber in front of the jet
pipe (Block A in Figure 5). The sudden change in diameter between jet pipe and
plenum chamber makes it more difcult for the system pipe/plenum chamber to act
186 J.U. SCHLTER AND T. SCHNFELD
Figure 6. Smoke visualization of the test case of Andreopoulos and Rodi, r = 2, Re = 81000.
as a resonator. This approach is more expensive, but offers as a by-product more
certainty on the jet velocity prole.
Furthermore a renement of the mesh in the low pressure region downstream
of the jet nozzle is necessary. It inuences the jet trajectory and higher trajectories
were obtained with a low resolution mesh. This mesh is automatically ne enough
to capture the wake vortices. But, additionally the wall region, where the wake
vortices have their origin, has to be well resolved as well.
However, it turned out that the number of mesh points was still too high, when
applied to a real conguration of the burner. To assess the accuracy of even coarser
meshes, additional computations were performed on an even simpler mesh. Here,
the jet pipe is meshed by a 6 6 H-mesh. The crossow is meshed with one single
block of H-type topology. The plenum chamber stays part of the computed domain
and is meshed by an H-block as well.
5. LES Validation Test Cases
In order to obtain meaningful statements on the validity of the LES computations
of the injectors of the real burner geometry, test cases are needed which are close
to the real problem with respect to the Reynolds number and the velocity ratio r.
In the real problem the Reynolds number is Re 8000, based on the exit velocity
of the injecting fuel and the injector diameter, and the velocity ratio r 2 at the
nozzle. Three cases have been chosen:
LES OF JETS IN CROSS FLOW 187
1. The rst one is a series of experiments carried out by Andreopoulos and Rodi
[25]. They provide detailed hot wire measurements of the mean velocity
components, the turbulent kinetic energy, the Reynolds stresses and measure-
ments on the turbulent kinetic energy budget. Furthermore they made measure-
ments on a slightly heated jet to obtain statements on the mixing behaviour by
measuring the temperature eld.
In the experiments the velocity ratio varies from r = 0.5 (Re = 20500) to
r = 2 (Re = 82000). Since the behaviour of the jet changes dramatically for
velocity ratios lower than r = 1 the experiments with a velocity ratio r = 2
were chosen to be reproduced despite the high Reynolds number.
2. Because the work on the gas turbine burner focuses on mixing, we have chosen
as a second test case the experiments by Smith and Mungal [2427]. They
seeded the jet air with acetone and made LIF measurements of the mixing
behaviour. Here, we have chosen to reproduce their measurement of a JICF
with a Re = 16400 and a velocity ratio r = 5.
3. An additional problem in the gas turbine concerns the interaction between
adjacent jets. Hence, the third test case is a measurement of a twin jet from
Toy et al. [30]. He measured velocity proles in the far eld of the jets with
Re = 31800 and a velocity ratio r = 6. Here the establishment of a single
CVP in the far eld of the jets shall be reproduced.
Despite the requirements on the mesh, a dexterous point distribution in the mesh
limits the number of points to 90.000 for the test cases of Andreopoulos and Rodi
and Smith and Mungal and 200.000 for the case of Toy. Tests with coarser meshes
have been done, but the results were poor, so that it can be assumed, that this is a
lower limit in terms of mesh points. The rst cell on the surface of the crossow
wall has a thickness of y
+
= 90 in the case of Andreopoulos and Rodi and Toy and
a thickness of y
+
= 50 in the case of Smith and Mungal.
6. Computational Results
6.1. REPRODUCTION OF GENERAL FLOW CHARACTERISTICS
All ow visualizations in this section have been made from the Reynolds-averaged
ow eld of the unsteady computation of the test case of Andreopoulos and Rodi.
The features shown can be found in all other computations as well.
Figure 7 shows the streamlines computed by LES in the initial region of the
CVP. It can be seen, that the CVP starts to develop very early. Haven [14] showed,
that the vorticity of the CVP has its origin at the side walls of the jet pipe. It seems
from the ow visualization of the LES computation, that this can be conrmed.
The vortices start very early, right behind the orice.
Figure 8 juxtaposes the streamline pattern proposed by Perry and Kelso [20]
with the computed streamlines. The resolution of the mesh upstream is not ne
enough to resolve all vortices in this region. Baker [6] showed that the two sec-
188 J.U. SCHLTER AND T. SCHNFELD
Figure 7. Streamlines at the jet orice, viewer is downstream of the orice looking up-
stream. The developing counterrotating vortex pair can be seen. Streamlines created from
Reynolds-averaged ow eld from the unsteady LES computation from the case of An-
dreopoulos and Rodi.
Figure 8. Side view on the computed streamlines at the jet orice fromthe case of Andreopou-
los and Rodi juxtaposed with the proposed streamline pattern of Perry and Kelso [20]. The
horseshoe vortex system is not fully resolved by the LES computation and thus simpler. The
recirculation zone at the upstream edge of the jet pipe can be seen. Streamlines created from
Reynolds-averaged ow eld from the unsteady LES computation.
ondary vortices of a horseshoe vortex are weak. The weakness and the small size
of these vortices make it difcult for a numerical investigation to capture these
structures. Nevertheless a simplied horseshoe vortex develops upstream of the
orice.
Figure 9 juxtaposes the streamline pattern in the xy plane on the centerline of
the jet found by Kelso [17] and the streamline pattern of the LES computation of
the case of Andreopoulos and Rodi. The comparison can be done only qualitatively,
because of the different experimental setups. In both ow elds a velocity node in
the wake, close to the wall can be found.
In Figure 10 the side view and top view of the streamline patterns are compared.
The streamline patterns resemble, except for one important difference: the vortex
in the xz plane directly behind the jet nozzle turns in opposite direction. Kelso
LES OF JETS IN CROSS FLOW 189
Figure 9. Qualitative comparison of streamline patterns in the xy plane on the centerline
of the jet, left Kelso [17] with a v
r
= 2.2 and the LES computation of Smith and Mungal
with a v
r
= 5. Streamlines created from Reynolds-averaged ow eld from the unsteady LES
computation.
Figure 10. Qualitative comparison of streamline patterns in the xy plane on the centerline
of the jet, left Kelso [17] with v
r
= 2.2 and the LES computation of Andreopoulos and Rodi
with v
r
= 2, the half-circle denotes the position of the jet nozzle. Streamlines created from
Reynolds averaged ow eld from the unsteady LES computation.
admitted in his investigation, that the resolution of his measurement grid is quite
coarse in this region. Because of the higher resolution in the LES computation we
believe, that the orientation of this vortex has to be clockwise. The direction of the
vortex can be conrmed from Figure 7. The vortex pair developing directly behind
the nozzle is in a nearly upright position at the nozzle itself and bends over with
the jet. That means, the vortex seen behind the nozzle in Figure 10 must have the
same orientation as the CVP. This is only the case, when the vortex behind the
nozzle spirals out clockwise. The origin of this vortex is unclear. Neither the jet
nor the wall boundary layer of the crossow carries the vorticity of this sign. A
computation, where the plane wall from the crossow has been replaced by a slip
wall, still showed this vortex. This indicates, that the origin of this vortex is located
in the pipe wall.
190 J.U. SCHLTER AND T. SCHNFELD
Figure 11. Comparison of the momentum elds on the centerline of the jet, circles: mea-
surements, solid line: LES Filtered Smag. Model, dashed line: LES Standard Smag. Model,
averaging time 0.25 s.
In the literature this vortex is unmentioned, with the exception from the investi-
gation of Kelso [17]. It seems, further investigation on the origin and further history
of this vortex is necessary, especially to shed light into the development of the CVP.
6.2. TEST CASE OF ANDREOPOULOS AND RODI
6.2.1. Momentum Field
Comparisons between the hot wire data obtained from Andreopoulos and Rodi and
our LES computations were made to quantify the reproducibility. In order to show
the ability of LES to simulate the experiment on low-resolved meshes, all LES
computations are made on meshes with 90000 mesh points. The u component of
velocity is compared in Figure 11 for different positions downstream on the jet
centerline. Regarding the prole at x/D = 2 the measurements and the LES com-
putation with the Filtered Smagorinsky model agree well. The LES computation
using the Standard Smagorinsky model shows a wrong trajectory, the location of
the velocity maximum is too high by 0.4D, but the right order of magnitude. As
already mentioned, the Filtered Smagorinsky model was optimized for boundary
layers. Hence, the oncoming wall boundary layer is better described and the mo-
mentum ratio close to the wall is better predicted. This has an inuence on the jet
trajectory. The u velocity proles downstream show, that the Filtered Smagorinsky
LES OF JETS IN CROSS FLOW 191
Figure 12. Comparison of the turbulent kinetic energy elds on the centerline of the jet. The
graphs concerning LES contain only the resolved part of the turbulent motion. Circles: mea-
surements, solid line: LES Filtered Smag. Model, dashed line: LES Standard Smag. Model,
averaging time 0.25 s.
model has advantages over the Standard Smagorinsky model, although the trajec-
tory is slightly too high. The measured velocity decit below the jet trajectory is
leveled out in the LES computation too early.
The turbulent kinetic energy (TKE) k
2
= u
2
+v
2
+w
2
was chosen to compare
dynamic variables. The TKE of the LES calculations presented here represent only
the TKE of the resolved spectrum of turbulence. This means, that the subgrid
turbulence does not appear in the graphs. In regions, where the level of subgrid
turbulence is high, the TKE of the LES computations is underestimated. This is
especially the case in wall boundary layers, where the high shear stress at the wall
implies a production of small scale structures, which are not captured by the mesh
resolution. In free turbulence far off the wall, the LES mesh is ne enough to allow
comparisons with measurements.
Figure 12 shows a comparison of TKE. The prole at x/D = 4 shows a
good agreement of the LES calculation with the Filtered Smagorinsky model.
The Standard Smagorinsky model shows a too high trajectory (which conrms
the observation from Figure 11) and overestimates the TKE. Obviously the Stan-
dard Smagorinsky model is not dissipating sufciently the turbulent energy. The
constant C
1
of the Standard Smagorinsky model could be increased to adapt the
model to the ow, but this would question the universal validity of the model. The
downstream proles show a quite good agreement far off the wall, but close to the
192 J.U. SCHLTER AND T. SCHNFELD
Figure 13. Comparison of the velocity proles at the jet orice. Circles: measurements, solid
line: LES, averaging time 0.25 s.
wall the TKE of the LES computations is well below the measured TKE. Here, as
mentioned above, the level of subgrid turbulence is not neglectable.
A look at the measured velocity prole at the jet orice (Figure 13) from
Andreopoulos and Rodi shows a major discrepancy in the recirculation bubble
between the hot wire measurements and the LES computation, which can be ex-
plained as follows: although Andreopoulos already proposed the possibility of a
temporal recirculation zone at this location for very short time spans [2], he did
not expect a steady recirculation zone. But the major limitation of hot wire mea-
surements is, that they cannot measure the direction of the ow and thus always
pretend to measure a positive velocity. This explains, that in the region of the recir-
culation bubble Andreopoulos and Rodi still measure a positive velocity. The LES
computations in contrast, show the recirculation zone and conrm the observations
of Perry and Kelso [20] (see also Figure 8).
6.2.2. Scalar Field
Due to the lack of reliable measurement techniques for mixing in the 1980s,
Andreopoulos heated the jet slightly by 4