Large-Eddy Simulation of

a Round Jet in Crossflow

JO

RG ZIEFLE*
1
- LEONHARD KLEISER**
*
Institute of Fluids Dynamics
2
, ETH Zurich, Switzerland
**
Institute of Fluids Dynamics, ETH Zurich, Switzerland
Abstract. A numerical simulation of a compressible round turbulent jet issuing
perpendicularly into a laminar boundary layer (jet in crossflow, JICF) is performed at
a jet-to-crossflow momentum ratio of 3.3 (defined with the bulk momentum of the jet
and the free-stream momentum of the boundary layer) and a Reynolds number of 2100
(based on the free-stream momentum, jet diameter and dynamic viscosity at the wall).
To be able to compare the results with incompressible reference data while still allowing
for anefficient time integration, a Mach number of 0.2 is chosen. The mixing behaviour
of the JICF is investigated by computing the evolution of a passive scalar at a Schmidt
number that equals the Prandtl number of the flow.
The spatial discretisation of the computational domain consists of a block-
structured multi-block grid with 58 blocks and a total of 8.9 million cells. The
simulation code NSMB uses a finite-volume discretisation with a skew-symmetric
fourth-order central scheme and a second-order Runge-Kutta method for time integra-
tion. To account for the subgrid scales, the approximate deconvolution model (ADM) is
employed in a multiblock formulation.
Of special interest in this investigation are the mixing behaviour of the jet with
the crossflow and the complex vortex systems in the mixing region. Results are
compared with incompressible LES data at similar flow parameters as mentioned
above.
The evolution of a jet under the influence of a crossflow (jet in crossflow, JICF)
has been subject to extensive research for more than half a century [1]. While many
earlier investigations have focussed on aerospace applications such as vertical and/
or short take-off and landing (V/STOL) aircraft, steering of rockets, film cooling of
turbine blades or fuel injection into combustors, in more recent times ecological
aspects such as plumes of smokestacks, volcanoes or (tunnel) fires have also gained
attention.
The goals of our study of a jet in crossfloware threefold. In the course of the recent
work, the approximate-deconvolution subgrid-scale model (ADM) [2, 3, 4] has been
implemented into a semi-industrial flow solver, supporting domain decomposition and
parallel computing. After a validation study using a simpler configuration [5], the
complex JICF case provides a means to validate the domain-decomposition features of
the new model code. Additionally, we want to assess the suitability of ADM for
separated flows in a low-order numerical code. Finally, the experience and infra-
structure developed with this flow case is applicable for the simulation of a flow
configuration of great industrial interest, namely film-cooling of turbine-blades, a work
currently in progress.
We selected the experimental JICF setup of Sherif and Pletcher [6], which was
also thoroughly investigated numerically by LES in the Ph. D. thesis of Yuan [7]. In
contrast to many other works, the Reynolds number is quite low here, rendering well-
resolved incompressible LES possible even with the compressible simulation ap-
proach we used. The two investigated jet-to-crossflow velocity ratios of 2 and 3.3 are
low enough for a rather small vertical size of the computational domain, but still large
enough for a highly complex and in both cases differing flow structure. The specific
flow case, see figure 1, consists of a turbulent jet issuing perpendicularly into a laminar
boundary layer. All relevant flow parameters and geometric dimensions are listed in
table 1.
FIGURE 1. Schematic of the jet in crossflow configuration. k1 Pipe simulation, k2 nozzle
and k3 boundary layer of mainJICFsimulation. n1 / n2 Data transfer betweenpipe and main
JICF simulations (see text).
470 JO

RG ZIEFLE - LEONHARD KLEISER

TABLE 1. Geometric and calculation parameters. Unless otherwise noted, all boundary-layer
quantities refer to the hypothetical laminar flow state in the centre of the jet nozzle (x 0)
without the presence of a jet.
L
x
= 16 R (r ~ w)
n
b
=(r~ u)
1;BL
= 3.3
L
y
= 10 Re
JICF
(r~ u)
1;BL
D= m
wall
= 2100
L
z
= 10 Re
jet
(r ~ w)
nb
D= m
wall
= 6930
L
upstream
= 5.5 Re
BL
(r~ u)
1;BL
d
99%;BL
= m
wall
= 1050
L
nozzle
= 5 Re
d

;BL
j
inlet
= 309.28
L
pipe
= 5 Ma
1
1=

gR
gas
p
= 0.2
D = 1 d
99%;BL
=D = 0.5
The jet nozzle consists of two independent parts, marked k1 and k2 in figure 1.
The upper part k2 , which ends at the boundary layer, obtains its inflow data from the
lower part k1 by the usual block coupling interface, symbolised by connection n1 . The
lower part of the nozzle, however, is not coupled to the upper one but assumes
streamwise-periodic boundary conditions (connection n2 ). In these so-called ``blind
blocks'', an independent simulation of turbulent pipe flow is carried out concurrently
with the main JICFsimulation. This one-way block-coupling mechanismin conjunction
with the simultaneous inflow data generation is a convenient and easy-to-implement
way of providing unsteady turbulent inflow data.
The current computational grid was generated with the commercial mesh
generator ICEM CFD Hexa. Its topology consists of 58 structured blocks with
matching interfaces, of which five blocks comprise the pipe region (blind blocks)
and another five blocks are used for the pipe nozzle. The total number of cells is
approximately 8:9 million, of which 4:5% each belong to the pipe simulation and the
jet nozzle.
A complete simulation cycle consists of the following steps. First solely the pipe
flow simulation is run, until a statistically stationary flow state is obtained. During
this stage, no simulation takes place in the blocks of the main JICF simulation
domain (k2 and k3 ). When the pipe flow simulation reaches a stationary state, the
full JICF simulation setup is turned on. After a sufficiently long time, in which the jet
penetrates the crossflow and the characteristic vortex systems have evolved,
statistical sampling is started. The simulation is continued until the statistics have
converged to the desired degree.
Time-averaging was performed by sampling the instantaneous flow field every 30
time-steps during a period of more than 250 time units L=~ u
1;BL
. This averaging time is
significantly longer than in the reference computation [7], where available computa-
tional resources limited sampling to 80 time units. However, our statistics were still
found to be not perfectly symmetric, especially in the far-field downstream of the jet
exit, where the flow is highly unsteady and irregular.
The computation was carried out on four processors of the NEC SX/8 vector
machine at HLRS, using more than 3000 CPU hours (including initial transients).
During time-averaging, 12 GB of memory were necessary. One instantaneous flow
field (including mesh and restart data) consumes 1.4 GB disk space, while the time-
LARGE-EDDY SIMULATION OF A ROUND JET IN CROSSFLOW 471
averaged flow field (73 quantities) requires 5.9 GB disk space. The total accumu-
lated data (including a series of instantaneous flow fields for animation and
frequency analysis purposes) comprises approximately 700 GB.
In the following a few sample results are shown. More details are found in the
corresponding publication [8]. The jet trajectory is one of the most important
characteristics of a JICF. Naturally, its determination is to some extent a matter of
definition. Here we compute the jet trajectory in three different ways that are
commonly used in literature [7]. As the first possibility, see figure 2(a), the streamline
of the mean flow field through the centre of the jet nozzle ((x; y; z) 0) marks the jet
trajectory. Additionally, the streamlines starting at the upstream and downstream
edges of the jet nozzle are shown as an approximation of the jet boundary in the centre
plane. Secondly, as depicted in figure 2(b), the trajectory can be defined as the locus
through all points with maximum scalar concentration in the centre plane y 0.
Alternatively, the maximum velocity magnitude can be used as defining quantity, as
FIGURE 2. Jet trajectory determined in different ways: a) Streamline trajectory, b) maximum
passive-scalar concentration trajectory and c) maximum velocity magnitude trajectory.
Present LES, LES of Yuan [7], ----bounding streamlines in present LES, bounding
streamlines in LES of Yuan.
472 JO

RG ZIEFLE - LEONHARD KLEISER

FIGURE 3. Vertical profiles of time-averaged quantities at four streamwise positions in the
symmetry plane y 0. a)-d) Mean velocity magnitude hj ~ uji, e)-h) normalised rms fluctuations
of the velocity magnitude hj ~ uji
rms
=hj ~ uji. Present LES, -- -- LES of Yuan [7],
experiments of Sherif and Pletcher [6].
LARGE-EDDY SIMULATION OF A ROUND JET IN CROSSFLOW 473
shown in figure 2(c). All trajectories exhibit generally good agreement with the
computational results of Yuan [7] (experimental results were not available for
comparison), even though the sensitivity to the particular method of determination
is quite strong for the latter two methods.
To allow for a more quantitative comparison of our LES results with the incom-
pressible reference simulation [7] and experiment [6], vertical profiles of the mean flow
fieldinthe symmetry plane are depictedat four different streamwisepositions infigure 3.
The first series of pictures, figures 3(a)-(d), shows the average velocity magnitude hj ~ uji.
The overall agreement with the reference data is quite good. The first local minimum fits
the experimental data better inour LESthaninthe one of Yuan. Inthe next rowof figures,
3(e)-(h), the normalised rms fluctuations of the velocity magnitude hj ~ uji
rms
=hj ~ uji are
plotted. Again our LES generally reproduces the comparison results well.
The isocontour of the passive-scalar concentration in figure 4(a) gives a good
impression of the unsteady and irregular instantaneous flowstate caused by the mixing
of the two streams. The roll-up and breakdown of the jet shear layer is also observable:
the initially smooth contour surface above the jet exit shows a regular pattern of ripples
after a distance along the jet trajectory of approximately one jet hole diameter. A few
diameters farther downstream the structure changes to an irregular form, which
coincides with the breakup of the shear-layer vortices [8].
Figure 4(b) depicts an isocontour of the instantaneous vortex-identification
quantity

l
2
. Similar to the flow around a wall-mounted cylinder, a horseshoe vortex
wraps around the jet exit hole. The mentioned shedding of shear-layer vortices from
the upstreamedge of the jet nozzle is directly visible. After a fewjet diameters along the
trajectory they loose their distinct appearance. This marks the location of their
breakup which was already observed in the isocontour of the passive-scalar concen-
tration in figure 4(a).
In figure 4(c) we display three sets of streamlines: two originating from the jet
nozzle and one from the crossflow. The red streamlines start from within the boundary
layer upstream of the jet exit at 10% boundary-layer thickness. The green and blue
streamlines are seeded across the jet nozzle throughout the two spanwise diameters
aligned with the x and y axes.
Following the red boundary-layer streamlines, we observe their deflection around
the jet exit, as the jet obstructs their downstream path, and the creation of a horseshoe
vortex. We can clearly recognise the bundling of the outer streamlines to form the two
legs of the horseshoe vortex. The streamlines close to the symmetry plane pass under
the other streamlines. After a wide circle around the jet exit, they finally join the
horseshoe leg or move downstream parallel to it.
At the downstream edge of the jet exit, a spanwise vortex caused by the roll-up of
the jet shear layer induces negative streamwise velocities near the wall. Therefore,
some of the boundary-layer streamlines within the horseshoe legs are forced to move
upstream toward the jet exit. There they fall into the upward influence of the emerging
jet and its counter-rotating vortex pair.
The path of the green and blue streamlines emanating from the jet nozzle is more
regular. The inner ones follow more or less the jet trajectory, while the blue ones at the
474 JO

RG ZIEFLE - LEONHARD KLEISER

spanwise edges of the jet are subject to the influence of a stationary inclined vortex pair
above the lateral sides of the jet exit. Farther downstream, those hanging vortices form
the characteristic counter-rotating vortex pair (CVP).
Acknowledgements. Calculations have been performed at the High Performance
Computing Center Stuttgart (HLRS). This work was carried out under the HPC-
EUROPA project (RII3-CT-2003-506079), with the support of the European
Community - Research Infrastructure Action under the FP6 ``Structuring the
European Research Area'' Programme. We are grateful to Prof. U. Rist and his group
for their hospitality.
FIGURE 4. a) Smoke visualisation of the instantaneous flow field. b) Visualisation of the
instantaneous flow field using the vortex identification criterion

l
2
. c) Streamlines of the
mean velocity field. Red: streamlines originating from the boundary layer upstream of the jet
exit hole at 10% boundary-layer height. Green and blue: streamlines seeded from the two
diameters aligned with the x and y axis across the jet nozzle.
LARGE-EDDY SIMULATION OF A ROUND JET IN CROSSFLOW 475
References
[1] MARGASON R.J., Fifty years of jet in cross flow research, Computational and Experimental
Assessment of Jets in Cross Flow, AGARD report CP-534, Nov. 1993.
[2] STOLZ S. and ADAMS N.A., An approximate deconvolution procedure for large-eddy
simulation, Phys. Fluids, Vol. 11, No. 7, 1999, pp. 1699-1701.
[3] STOLZ S., ADAMS N.A. and KLEISER L., An approximate deconvolution model for large-eddy
simulation with application to incompressible wall-bounded flows, Phys. Fluids, Vol. 13,
No. 4, 2001, pp. 997-1015.
[4] STOLZ S., ADAMS N.A. and KLEISER L., The approximate deconvolution model for large-eddy
simulations of compressible flows and its application to shock-turbulent-boundary-layer
interaction, Phys. Fluids, Vol. 13, No. 10, 2001, pp. 2985-3001.
[5] ZIEFLE J., STOLZ S. and KLEISER L., Large-Eddy Simulation of Separated Flow in a Channel
with Streamwise-Periodic Constrictions, 17th AIAA Computational Fluid Dynamics
Conference, Toronto, Canada, June 6-9, 2005, AIAA Paper 2005-5353.
[6] SHERIF S.A. and PLETCHER R.H., Measurements of the Flow and Turbulence Characteristics
of Round Jets in Crossflow, J. Fluids Eng., Vol. 111, 1989, pp. 165-171.
[7] YUAN L.L., Large Eddy Simulations of a Jet inCrossflow, Ph. D. thesis, Stanford University,
1997.
[8] ZIEFLE J. and KLEISER L., Large-Eddy Simulation of a Round Jet in Crossflow, 36th AIAA
Fluid Dynamics Conference, San Francisco, USA, June 5-8, 2006, AIAA Paper 2006-3370.
476 JO

RG ZIEFLE - LEONHARD KLEISER