Center for Turbulence Research 143

Annual Research Briefs 2002

Consistent boundary conditions for integrated

LES/RANS simulations: LES inflow conditions
By J. U. Schlüter

1. Motivation
Currently, a wide variety of flow phenomena are addressed with numerical simulations.
Many flow solvers are optimized to simulate a limited spectrum of flow effects effectively,
such as single parts of a flow system, but are either inadequate or too expensive to be
applied to a very complex problem.
As an example, the flow through a gas turbine can be considered. In the compressor
and the turbine section, the flow solver has to be able to handle the moving blades,
model the wall turbulence, and predict the pressure and density distribution properly.
This can be done by a flow solver based on the Reynolds-Averaged Navier-Stokes (RANS)
approach. On the other hand, the flow in the combustion chamber is governed by large
scale turbulence, chemical reactions, and the presence of fuel spray. Experience shows that
these phenomena require an unsteady approach (Veynante and Poinsot, 1996). Hence,
the use of a Large Eddy Simulation (LES) flow solver is desirable.
While many design problems of a single flow passage can be addressed by separate
computations, only the simultaneous computation of all parts can guarantee the proper
prediction of multi-component phenomena, such as compressor/combustor instability and
combustor/turbine hot-streak migration. Therefore, a promising strategy to perform full
aero-thermal simulations of gas-turbine engines is the use of a RANS flow solver for the
compressor sections, an LES flow solver for the combustor, and again a RANS flow solver
for the turbine section (figure 1).

2. Interface
The simultaneous computation of the flow in all parts of a gas turbine with different
flow solvers requires an exchange of information at the interfaces of the computational
domains of each part. Previous work has established algorithms, which ensure, that two
or more simultaneously running flow solvers are able to exchange the information at the
interfaces efficiently (Shankaran et al, 2001, Schlüter et al, 2002).
The necessity of information exchange in the flow direction from the upstream to the
downstream flow solver is self-explanatory: the flow in a passage is strongly dependent
on mass flux, velocity vectors, and temperature at the inlet of the domain. However,
since the Navier-Stokes equations are elliptic in subsonic flows, the downstream flow
conditions can have a substantial influence on the upstream flow development. This can
easily be imagined by considering that, for instance, a flow blockage in the turbine section
of the gas turbine can determine and even stop the mass flow rate through the entire
engine. This means that the information exchange at each interface has to go in both,
downstream and upstream, directions.
Considering an LES flow solver computing the flow in the combustor, information on
the flow field has to be provided to the RANS flow solver computing the turbine as
144 J. U. Schlüter

Figure 1. Gas turbine engine

LES to RANS
Provide time
averaged data
RANS to LES
Upstream influence
of pressure very
important
RANS to LES
Create turbulent
fluctuations
LES to RANS
Provide time
averaged data

Figure 2. Gas turbine combustor with interfaces. (LES of combustor from Mahesh et al, 2001)

well as to the RANS flow solver computing the compressor, while at the same time,
the LES solver has to obtain flow information from both RANS flow solvers (figure 2).
The coupling can be done using overlapping computational domains for the LES and
RANS simulations. For the example of the compressor/combustor interface this would
imply that inflow conditions for LES will be determined from the RANS solution at the
beginning of the overlap region, and correspondingly the outflow conditions for RANS
are determined from the LES solution at the end of the overlap region.
However, the different mathematical approaches of the different flow solvers make
the coupling of the flow solvers challenging. Since LES resolves large-scale turbulence
in space and time, the time step between two iterations is relatively small. RANS flow
solvers average all turbulent motions over time and predict ensemble averages of the
flow. Even when a so-called unsteady RANS approach is used, the time step between
two ensemble-averages of the RANS flow solver is usually larger by several orders of
magnitude than that for an LES flow solver.
The specification of boundary conditions for RANS from LES data is relatively straight-
 LES inflow conditions 145
forward. The LES data can be averaged over time and used as boundary condition for
the RANS solver.
Since LES computations have shown to be sensitive to outflow conditions (Moin, 1997,
Pierce & Moin, 1998a), the adjustment of the LES solution near the outlet to the RANS
solution of the downstream computation is of importance. Previous work investigated in
detail LES outflow conditions (Schlüter and Pitsch, 2001, Schlüter et al, 2002). A virtual
body force is employed to drive the LES mean velocity field to the RANS solution in an
overlap region. The turbulent quantities and the pressure field adjusts accordingly to the
mean velocity field.
In the present study, the boundary conditions provided from an upstream RANS flow
solver to a downstream LES flow solver is investigated. The inflow conditions have to be
created such, that the time-averaged mean values of all computed quantities match the
RANS solution at a given plane and meaningful turbulent fluctuations are added. LES
computations of validation test cases are performed to assess the influence of the inflow
boundary conditions.

3. Inflow Boundary conditions

The following section presents some possible LES inlet conditions, which are tested for
use in integrated RANS-LES computations

3.1. Creation of database from auxiliary LES computation

The formulation of LES inflow conditions from time-averaged RANS data is similar to
the definition of LES inflow conditions from experimental data, which is usually given
in time-averaged form. An established procedure to create inflow boundary conditions is
to perform an auxiliary LES computation prior to the actual LES of the desired domain
(Pierce & Moin, 1998b). The auxiliary LES computation computes a periodic pipe and
uses virtual body forces inside the domain to drive the flow to the desired velocity profiles.
The time history of one plane of this computation is written into a database. The actual
LES computation of the desired geometry then reads this database and uses its transient
velocity field to define its own inlet velocity field.
This method is a well established procedure and shows good results in reproducing
experiments (e.g. Duchamps & Pitsch, 2000, Schlüter, 2001). Hence, it will be used as a
benchmark for all following proposed boundary conditions.
The advantage of this method is that the representation of the inlet turbulence is taken
from a fully developed turbulent flow, which means all temporal and spatial correlations
of the turbulent fluctuations are actual representations of eddies. The energy spectra in
time and space have a natural energy distribution especially in the long wave range.
The creation of the database implies additional computational costs. However, using
flow solver specialized to this task, the auxiliary computation usually takes less than an
hour wall-clock time on a single processor, which is less than 1% of the computational
costs for the LES of the actual geometries used in the current investigation.
The disadvantage of this procedure is that the mean velocity field at the inlet has to be
known prior to the LES computation. In integrated RANS-LES computations the mean
velocity field at the inlet of the changes in time, and hence, is unknown. This makes it
impossible, to apply this procedure in this form for integrated RANS-LES computations.
146 J. U. Schlüter
3.2. No-fluctuations inflow conditions
The most simple way to define the inflow boundary conditions from RANS data is to ne-
glect the turbulent fluctuations entirely. The velocity field at the inlet is then defined by
the ensemble-averaged mean profiles of the RANS computation. This means, the incom-
ing flow is laminar, but the shape of the velocity profiles is still that from a turbulent flow.
It is easy to adapt this inflow condition during the LES computation to take variations
of the ensemble-averaged flow field delivered by the RANS computation into account.

3.3. Random fluctuations

Early work in LES inlet boundary conditions report the usage of random fluctuations
superposed to the mean velocity field at the inlet to simulate inflow turbulence. However,
due to the lack of correlations in space and time of the fluctuations, these lack the energy
in the long wave range. As a result of that, the fluctuations are usually in the high wave
spectrum and hence, dissipated very quickly. The few tests in the current investigation
which were made with this inlet condition agree with these findings. The flow laminarizes
quickly behind the inlet and flow fields obtained with this inlet condition were indistin-
guishable from the flow fields obtained with the no-fluctuations inflow condition. Hence,
they will not be shown here.

3.4. Mean velocity profiles with turbulent fluctuations from database

The method proposed here for integrated RANS-LES computations, uses the mean flow
field from the RANS solution and adds meaningful turbulence from a database created
by an auxiliary LES computation. This allows to vary the mean flow field during the LES
computation in order to take temporal variations of the RANS solution into account. As
for the inlet condition proposed in section 3.1, an auxiliary LES computation of a pipe
flow is performed to define the turbulent fluctuations. The inlet condition is then defined
as: q
u02
(i) (τ )
ui,LES (t) = ūi,RANS (τ ) + (ui,DB (t) − ūi,DB ) · q RANS (3.1)
| {z } | {z } u02
(i)
I II
| {z DB }
III
with RANS denoting the solution delivered by the RANS computation and DB properties
delivered by the database. The time-scale t is the time-scale used by the LES computa-
tion, and τ the time-scale used by the RANS computation. The RANS time-step ∆τ is
usually much larger than the LES time-step ∆t, which means, that multiple LES inlet
conditions are computed before the RANS solution is updated.
Term II of Eq. 3.1 computes the velocity fluctuation of the database. This turbulent
fluctuation is scaled to the needed value with term III. Here, it is assumed that the
value of u02
(i) RANS (τ ) is a known quantity. However, most RANS turbulence models do not
compute the single components of the Reynolds-stress tensor, but more general turbulent
quantities such as the turbulent kinetic energy k. In this case, the axial components of
the Reynolds-tensor have to be approximated by:
2
u02
(i) k = (3.2)
3
RANS

Once the turbulent fluctuation is computed, it is then added to the time-averaged

velocity field (term I) and a meaningful inlet velocity field is recovered.
The quality of the database can be measured in the necessity of term III to scale the
 LES inflow conditions 147
20D

































































































1D 2D

































































































Figure 3. Geometry of test case

turbulent fluctuation. Since this scaling is linear, it is desirable, that the scaling factor
is close to unity. Hence, in the creation of the database it is important to reproduce the
expected inlet condition as closely as possible in order to keep scaling in the bounds of
the validity of a linear extrapolation.
In the following test cases, inlet conditions computed with eqn. 3.1 are using databases
with flow statistics strongly different from the desired values. This is solely to prove the
robustness of the method and not advisable for the final application.

4. Validation test case

In order to validate the influence of different inflow boundary conditions, LES compu-
tations of a confined jet with and without swirl were performed. The considered geometry
corresponds to the experiments of Dellenback (Dellenback, 1986, Dellenback et al, 1988).
The experiment investigates the flow at an axi-symmetric expansion (figure 3). Measure-
ments upstream of the expansion allow a proper description of the inflow statistics and
multiple measured velocity profiles downstream of the expansion give a good picture of
the flow development.
Three different flow configurations are computed:
(a) no swirl (S = 0.0) at a Reynolds-number Re = 30, 000
(b) strong swirl (S = 0.6), at Re = 30, 000 and
(c) weak swirl (S = 0.3), at Re = 20, 000
with the swirl number S defined as:
RR
1 0 r2 ūx ūφ dr
S= RR , (4.1)
R rū2x dr
0

where ux is the axial velocity component, uφ the azimuthal velocity component, and R
the radius of the nozzle.
For the first two cases measurements are available and LES predictions can be com-
pared with experimental data. For the last case, no measurements are available.
The computational meshes contain 1.58 million points (1.52 million cells) and are
identical for all LES computations.

5. LES flow solver

In order to investigate the effects of different inflow boundary conditions, the various
boundary conditions were implemented in an LES flow solver and tested. For this task,
the LES flow solver developed at the Center for Turbulence Research (Pierce & Moin,
148 J. U. Schlüter
x/D=−2.0 −0.5 0.5 1.0 2.0 3.0 4.0 6.0 8.0 10.0 12.0
1.0

Ux

0.5

0.0
0.0 1.0
1.0

RMS(u’x)

0.5

0.0
0.00 0.25
Figure 4. LES results for a confined jet. Squares: experiment; Dots: LES with inflow entirely
from database (section 3.1); Dashed line: LES with no-fluctuations inflow (section 3.2); Solid
line: LES with mean-flow + fluctuations from data-base (section 3.4)

1998a) has been used. The flow solver solves the filtered momentum equations with
a low-Mach number assumption on an axisymmetric structured mesh. A second-order
finite-volume scheme on a staggered grid is used (Akselvoll & Moin, 1996).
The subgrid stresses are approximated with an eddy-viscosity approach. The eddy
viscosity is determined by a dynamic procedure (Germano et al, 1991; Moin et al, 1991).

6. Validation
6.1. Confined jet without swirl
The first test of the inlet boundary conditions is made for a confined jet without swirl. It
is well known, that the spreading rate of the jet is dependent on the turbulence present
in the jet flow.
Figure 4 shows the velocity fields obtained for this case. Experimental results are shown
as square symbols. The two velocity profiles on the left are taken upstream of the step,
and the leftmost profile defines the inlet conditions for the LES computations.
The first LES computation (black dots in figure 4) determines the inlet conditions
entirely from a database (compare section 3.1). The database was created by an auxiliary
computation with body forces driving the mean velocity field to the desired inlet velocity
profile on the left. The usage of this boundary condition is possible in this case, since
the time-averaged velocity field defining the inlet condition does not vary in time. It
 LES inflow conditions 149
can be seen, that this LES computation reproduces the experimental data well in mean
values and turbulent fluctuations. The reattachment of the flow behind the step is well
captured.
The second LES computation uses no-fluctuations inflow conditions (compare section
3.2). Since the initial turbulence in the flow does not reach the desired level near the
step, the spreading rate of the jet is underestimated and the jet penetrates much further
into the chamber (dashed lines in figure 4). The reattachment length is overestimated.
As a result of the neglect of turbulence at the inlet, the axial turbulent fluctuations are
underestimated throughout the domain.
The third LES computation uses the mean velocity field and turbulent fluctuations
from a database (compare section 3.4). The database used is the database created for
a swirling flow (first LES computation of the following chapter, section 6.2) and hence,
is not adapted to this particular case. The correction of Eq. 3.1 is used to obtain the
desired mean statistics. The results of this computation (solid lines) show a good agree-
ment with the experiment and the first LES computation in the mean values. There are
some discrepancies in turbulent fluctuations between this LES computation and the LES
computation using a matching database due to the different description of turbulence at
the inlet. However, both LES results are reasonably close to the experimental data.
This test case shows, that the proposed inflow condition (Eq. 3.1) is capable of repro-
ducing the desired flow field, even when a low-quality database is used. The importance
of turbulent fluctuations at the inlet is underlined with the failure of the no-fluctuations
inlet condition to reproduce the flow field properly.

6.2. Confined jet with strong swirl

As a second test case a swirl flow at an expansion with a swirl number S = 0.6 is
considered. Swirl flows with high swirl (S > 0.25) create central recirculation zones, and,
as a result of that, flows with high shear are created which have a high level of turbulence
production.
Figure 5 shows the results of this series of computations. The LES computation using
a data-base with matching inlet velocity statistics (black dots) agrees well with the
experiments (square symbols).
Surprisingly, the LES computation using the no-fluctuations inflow conditions (dashed
lines) yields a comparable flow field and, despite some discrepancies, agrees reasonably
well with the experiments. This can be explained with the fact, that the level of turbulence
production is very high behind the expansion. The origin of the inner recirculation zone
in highly swirling flows is fixed at the location of the expansion. This means, the zones
of the turbulence production – the shear layers created by the recirculating fluid and
the issuing jet – is well determined and independently of the inflow conditions. The
turbulence level is then nearly entirely defined by the turbulence production behind the
step.
The third LES computation uses a combined approach and the database from the
previous test case of the flow without swirl (section 6.1). The particular shape of the
axial velocity profile and the entire swirl component is imposed using Eq. 3.1. The LES
computation recovers the LES solution using a matching database exactly.
This second test case shows, that situations exist, where the inlet turbulence plays a
minor role, even when complex flow configurations are considered. In this special case, the
high level of turbulence production inside the LES domain is dominant and its location
and level are not determined by the inlet conditions.
150 J. U. Schlüter
x/D= −0.5 0.25 0.5 0.75 1.0 1.5 2.0 3.0 4.0 6.0 8.0
1.00

Ux

0.50

0.00
0.00 1.00
1.00

Uφ

0.50

0.00
0.00 1.00
1.00
RMS(ux’)

0.50

0.00
0.00 0.50
1.00
RMS(uφ’)

0.50

0.00
0.00 0.50
Figure 5. LES results for a jet with strong swirl (S = 0.6). Squares: experiment; Dots: LES
with inflow entirely from database (section 3.1); Dashed line: LES with no-fluctuations inflow
(section 3.2); Solid line: LES with mean-flow + fluctuations from data-base (section 3.4)
 LES inflow conditions 151
6.3. Confined jet with weak swirl
While in the previous case the strong swirl ensured a certain universality of the extent of
the recirculation zone, weakly swirling flows are much more sensitive to inflow conditions
(Gupta et al, 1984). Since it is desirable for most flow applications, e.g for gas turbine
combustors, to keep the swirl number low in order to minimize the pressure drop over the
swirler, these kinds of flows are of particular interest for industrial applications. Hence, a
proper definition of LES inflow boundary conditions is crucial for the prediction of these
flows and the optimization of swirler geometries.
The third test case considered in this investigation is a weakly swirling flow at a
swirl number S = 0.3. The swirl number is just supercritical, meaning that an inner
recirculation zone develops. Unfortunately, no experimental measurements are available
for this case, but since the traditional method of generating inflow conditions by an
auxiliary LES computation was very successful in the previous two cases, the results of
this computation can be used as a reference.
Figure 6 shows the results of this series of LES computations. The LES computation
with a database with matching mean flow statistics (black dots) shows the onset of the
recirculation zone near the location of the expansion.
Using the no-fluctuations inflow condition (dashed lines), the location and extent of
the recirculation zone changes dramatically. As a result of that, the mean flow field differs
substantially from the previous LES computation. Due to the neglect of turbulence at
the inlet, the turbulent fluctuations are underestimated throughout the near field of the
expansion. As a result of the displacement of the zones of turbulence production, not
even the shape of the profiles of turbulent fluctuations is reproduced.
Using Eq. 3.1 and the non-swirling database from the test case in section 6.1 (solid
lines), all flow features are recovered. The origin and the extent of the recirculation zone
are identical to the LES with the matching database. As a result of that, the turbulence
production is also well represented and the prediction of turbulent fluctuations coincide.
This last test case shows most dramatically how the choice of LES inflow boundary
conditions may alter the results of a computation. While the previous test case of the
strongly swirling flow was remarkably robust to different inflow conditions, the present
case shows that only a little change in flow parameters, the decrease of the swirl number,
may result in flow configuration much more sensitive to inflow conditions.

7. Conclusions
The definition of LES inflow conditions from time-averaged statistical data has been
subject of research for some time. The current investigation focuses on LES inflow bound-
ary conditions for integrated RANS-LES computations, where the LES inflow conditions
are prescribed by the solution of an upstream RANS solver. Here, the flow statistics,
which have to be prescribed at the inlet of the LES domain, may vary in time.
A modification of the already widely used procedure, where an auxiliary LES compu-
tation creates a database for inflow conditions, was proposed. This inflow condition uses
the unsteady mean velocity profiles at the inlet and superposes turbulent fluctuations
from a database.
The proposed inflow condition was validated on three different flows. While one case,
a strongly swirling flow, was surprisingly robust against different inflow conditions, the
other two cases, a confined jet and a weakly swirling flow, underlined the necessity of a
proper turbulence description at the inlet.
152 J. U. Schlüter
x/D= −0.5 0.25 0.5 0.75 1.0 1.5
1.00
r/D

Ux

0.50

0.00
1.0
1.00
r/D

Uφ

0.50

0.00
0.5
x/D= −0.5 0.25 0.5 0.75 1.0 1.5
1.00
r/D
RMS(ux’)

0.50

0.00
0.5
1.00
r/D
RMS(uφ’)

0.50

0.00
0.5

Figure 6. LES results for a jet with weak swirl (S = 0.3). Dots: LES with inflow entirely from
database (section 3.1); Dashed line: LES with no-fluctuations inflow (section 3.2); Solid line:
LES with mean-flow + fluctuations from data-base (section 3.4), no experimental data available
 LES inflow conditions 153
The inflow boundary condition proposed in the present study shows equivalent results
to the commonly applied procedure using a database from an auxiliary LES computation.
Its advantage over other methods is its flexibility to accommodate for time-dependent
flow statistics at the LES inlet.
The definition of LES inflow boundary conditions for integrated RANS-LES computa-
tions is an important step towards integrated flow computations.

8. Acknowledgments
Support by the US Department of Energy under the ASCI program is gratefully ac-
knowledged.

REFERENCES
Akselvoll, K., & Moin, P. 1996 Large-eddy simulation of turbulent confined coan-
nular jets. J. Fluid Mech. 315, 387–411.
Dellenback, P. A., Metzger, D. E. & Neitzel, G. P 1988 Measurements in
turbulent swirling flow through an abrupt axisymmetric expansion. AIAA J. 26,
669–681.
Dellenback, P. A. 1986 Heat transfer and velocity measurements in turbulent swirling
flows through an abrupt axisymmetric expansion. PhD thesis, Arizona State Univer-
sity.
Duchamps de Lageneste, L. & Pitsch, H. 2000 A level-set approach to large eddy
simulation of premixed turbulent combustion. Annual Research Briefs Center for
Turbulence Research, NASA Ames/Stanford Univ., 105–116.
Germano, M., Piomelli, U., Moin, P. & Cabot, W., 1991 A dynamic subgrid-scale
eddy viscosity model. Phys. Fluids A 3, 1760–1765.
Gupta, A. K., Lilley, D. G. & Syred, N. 1984 Swirl Flows. Abacus Press.
Mahesh, K., Constantinescu, G., Apte, S., Iaccarino, G. & Moin, P., 2001
Large-eddy simulation of gas turbine combustors Annual Research Briefs, Center
for Turbulence Research, NASA Ames/Stanford Univ. 3–18.
Moin, P. 1997 Progress in large eddy simulation of turbulent flows. AIAA Paper
97–0749.
Moin, P., Squires, K., Cabot, W. & Lee, S. 1991 A dynamic subgrid-scale model
for compressible turbulence and scalar transport. Phys. Fluids A (3), 2746–2757.
Pierce, C. D. & Moin, P. 1998a Large eddy simulation of a confined coaxial jet with
swirl and heat release. AIAA Paper 98–2892.
Pierce, C. D. & Moin, P. 1998b Method for generating equilibrium swirling inflow
conditions. AIAA J. 36, 1325–1327.
Schlüter, J., 2001 Large-eddy simulations of combustion instability suppression by
static turbulence control. Annual Research Briefs Center for Turbulence Research,
NASA Ames/Stanford Univ. 119—130.
Schlüter, J., Pitsch, H. 2001 Consistent boundary conditions for integrated
LES/RANS simulations: LES outflow conditions. Annual Research Briefs Center
for Turbulence Research, NASA Ames/Stanford Univ. 19–30.
Schlüter, J., Shankaran, S., Kim, S., Pitsch, H., Alonso, J. J. 2002 Integration
154 J. U. Schlüter
of LES and RANS flow solvers: Interface validation. Annual Research Briefs, Center
for Turbulence Research, NASA Ames/Stanford Univ.
Schlüter, J. U., Pitsch, H., and Moin, P. 2002 Consistent boundary conditions for
integrated LES/RANS simulations: LES outflow conditions. AIAA paper 2002-3121.
Shankaran, S., Liou, M.-F., Liu, N.-S., Davis, R., and Alonso, J. J. 2001 A
multi-code-coupling interface for combustor/turbomachinery simulations. AIAA pa-
per 2001-0974.
Veynante, D. and Poinsot, T. 1996 Reynolds averaged and large eddy simulation
modeling for turbulent combustion. In New Tools in Turbulence Modelling, Les
edition physique. Springer, Berlin. pp. 105–140.