Proceedings of ASME Turbo Expo 2012

GT2012
June 11-15, 2012, Copenhagen, Denmark

GT2012-60

TURBULENCE LEVELS ARE HIGH AT THE COMBUSTOR-TURBINE INTERFACE

Chong M. Cha ∗ ,1 Peter T. Ireland2 Paul A. Denman3 Vivek Savarianandam3

1
Turbine Systems, Rolls-Royce plc, P.O. Box 31, PCF-2, Derby DE24 8BJ, UK
2
Dept. Engineering Science, University of Oxford, Parks Road, Oxford, OX1 3PJ, UK
3
Dept. Aeronautical and Automotive Engineering, Loughborough University, LE11 3TU, UK

ABSTRACT tor. The impact of the limitations of the combustor-turbine

Turbulence measurements are made in a novel gas tur- rig on these findings is discussed.
bine rig facility recently used to study combustor-turbine in-
teractions in jet engines [1]. The rig is capable of numer-
ous area traverses surrounding engine turbine nozzle guide INTRODUCTION
vanes (NGVs). The rig is unique in that complete engine A method of improving the efficiency of a turbine is to
hardware of the annular combustion subsystem is used to control secondary flow development. The secondary flows
simulate the upstream flow entering the turbine. The rig developing in the stator and rotor passages remove total en-
operates at cold, near-atmospheric conditions. The tur- ergy from the primary flow which is used to produce all the
bulence measurements include both the turbulence intensi- useful work. Understanding the underlying physical mecha-
ties and lengthscales and span an area over a single com- nisms behind the secondary flows has been based primarily
bustor sector. Axial measurement planes include locations on rig tests of a linear cascade of stators. The seminal work
both upstream and downstream of the real engine hardware on this front is now over three decades old [2]. To date, this
NGVs. The upstream plane corresponds to a conventional now classic, basic physical picture still represents the cur-
combustor-turbine interface plane. In [1], pressure, ve- rently accepted view [3–5]. Both turbine and compressor
locity, and passive scalar mixing measurements were pre- engineers have adopted this picture of the secondary flow
sented along with RANS CFD predictions. Here, in addi- physics.
tion to the newly measured turbulence quantities, large-eddy Figure 1 reviews this basic physical picture (copied from
simulations (LES) are performed for the complete, coupled Langston [2]). The streamwise vortices originate from the
combustor-turbine system. normal vorticity generated by the upstream boundary layer,
Good agreement between rig data and CFD is seen at where the bulk of the turbulence and therefore losses exist
the combustor-turbine interface, with LES yielding improved up to the cascade of stator vanes or rotor blades. This
predictions over RANS. For the flow through the NGV pas- vortex line resembles a horseshoe as it wraps around the
sages, vortex visualizations of the simulated flowfields show vane leading edge. The pressure side leg of this “horseshoe
significant differences to the classic, commonly accepted pic- vortex” grows due to turbulent entrainment as it convects
ture of Langston [2] and others [3]. The difference is at- across the vane passage, forced across by the cross-stream
tributed to the high turbulence levels created by the combus- pressure gradient. This vortex will interact with the suction
side leg of the horseshoe vortex under the right conditions.
Description of this interaction varies with researcher. In any
∗ Corresponding author. Current contact info: Turbine Aerody- case, a resulting, dominating streamwise vortex is called the
namics, Rolls-Royce Corp., 2001 S. Tibbs Ave., Speed Code T-63, “passage vortex”. Transition to turbulent flow over the vane
Indianapolis, IN 46241, USA. E-mail: chong.m.cha@gmail.com, Tel:
(317)230-4581.

1 Copyright © 2012 by Rolls-Royce Corporation

 νt /ν

FIGURE 2. Motivation of the work. Measurements of

turbulent-to-laminar viscosity ratio (νt /ν) just upstream of the
FIGURE 1. Classic physical picture of vortex dynamics in a NGVs in the present combustor-turbine interaction rig. The
stator vane or rotor blade passage (taken from Langston [2]). paper presents the measurements of the turbulence intensities
(u0 /U ) and lengthscales (`) used to construct νt for this figure,
and investigates the resulting impact on the flow physics through
airfoils (away from both endwalls) is deemed important due the nozzle guide vane passages using LES.
to the low turbulence levels (typically ∼1% intensity) in the
freestream (i.e., outside the boundary layers).
Today’s turbine CFD analyst adopts turbulence model- generated turbulence to simulate combustor exit turbulence
ing wisdom and simulation “best-practices” from his or her conditions. For their stator impact study, this results in an
compressor colleague based on the physical picture of Fig. 1. inflow turbulence intensity of 20% and a mesh size to stator
Given the heuristic nature of RANS turbulence modeling, chord length (turbulence lengthscale) of ∼2%.
the validity of this approach is questionable, especially for In the present paper, the turbulent intensity, u0 /U∞ ,
the high-pressure (HP) turbine. This is because the turbu- and lengthscales, `, are measured behind complete en-
lence levels entering the turbine are expected to be higher gine hardware of a RQL (Rich-burn–Quench–Lean), or
than those entering the compression subsystem due to the more simply, “rich-burn” combustion system. The two-
high levels of turbulence created by the combustion subsys- dimensional distributions of u0 and ` are given at both an
tem. upstream and downstream plane of uncooled engine geom-
Just how large are the actual turbulence levels from etry turbine NGVs. The low-speed, atmospheric rig is run
modern combustion systems? What is the resulting im- isothermally and reaches a peak Mach number in the NGV
pact on the flow development through the turbine? Are the of 0.23.
secondary flow differences from the classic picture signifi- Using complete combustion sub-system hardware to de-
cant enough to result in a different, more optimum coupled scribe the inflow turbulence characteristics into the tur-
combustor-turbine system design? The purpose of this work bine is the first study of this kind. The turbine subsystem
is to shed some light on these questions, specifically to bet- is represented by a set of engine hardware NGVs and is
ter guide the modeling and CFD simulation developments needed to replicate the upstream impact of the turbine on
for turbine design. the combustor-turbine interface plane [1].
Little attention has been paid to measuring and charac- Figure 2 shows the turbulent or eddy-viscosity, νt ∼ u0 `,
terising the turbulence from combustion systems. Radom- with u0 and ` taken from the present experimental measure-
sky & Thole [6] and Ames [7] review past studies which ments. The results have been normalized by the laminar
measure the turbulence levels exiting some combustors. value, ν, and are shown at the combustor-turbine interface
(Regrettably, lengthscales are reported as absolute values location and over an area encompassing a single fuel injec-
rather than normalized by a vane chord or combustor an- tor or burner sector. The figure illustrates that, in contrast
nuli height, or they are not reported at all.) As a point- to the inflow turbulence characteristics of turbomachinery
of-reference for the present work, these reviews put com- cascade tests, the turbulence levels entering the turbine
bustor exit turbulence intensity levels between 10–30%. In from a real combustor are many orders-of-magnitude larger.
their own experiment, Radomsky & Thole [6] use mesh- Furthermore, the distribution is significantly different, with

2 Copyright © 2012 by Rolls-Royce Corporation

 FIGURE 3. Annular test rig cross-section.

peak turbulence levels appearing near mid-span rather than measurement technique described in detail. The Simulation
in the endwall boundary layers. Full details are described section focuses on the description of the LES CFD. The Re-
in this paper. sults and Discussion section compares the LES and existing
Although these high levels and complex distributions RANS simulation results to the newly measured turbulence
of turbulence have been predicted by CFD [1], Fig. 2 repre- data. This section also presents the coherent vortex struc-
sents the first time experimental measurements of the tur- tures predicted by the LES developing through the turbine
bulence using real engine “hot-section” hardware have been NGV passages and compares them to the classic Langston
seen. picture (Fig. 1). The paper is summarized and conclusions
Another contribution of the present work stems from are given in the final section.
accompanying numerical simulations which are used to vi-
sualize the resulting vortex structures in the NGV passages.
LES is the CFD tool of choice here due to its well-known su- EXPERIMENT
periority of describing the behavior of large-scale turbulent The experiments are carried out at near atmospheric,
structures, e.g., as produced by the dilution jets of rich- isothermal conditions using a full annular test facility origi-
burn combustion systems in the present application. The nally developed to study the external aerodynamics of com-
entire combustor geometry upstream of the turbine inlet bustion systems. In this section, a description of the test
guide vanes is included in the CFD simulations to prop- rig, measurement techniques and traverse locations is given.
erly capture the complex turbulence features entering the
turbine.
The paper is organized as follows. The next two sec- Test Rig
tions describe the rig test experiment and numerical simu- The annular test facility from which the experimental
lations. (The present experimental rig and operating condi- data are recorded is shown in Fig. 3. The rig includes a 1.5
tions are identical to the case with NGVs installed from [1].) stage axial flow compressor designed specifically to provide
Here, the experiment is summarized with the turbulence representative flow structures to the test section containing

3 Copyright © 2012 by Rolls-Royce Corporation

the diffuser and combustion system. Filtered ambient air is
delivered to the compressor via a 150 m3 under-floor plenum
and flow straightener using a high-capacity centrifugal fan
(all not shown). Although originally developed to study the
aerodynamic performance of aero-engine combustor diffuser
systems, the test section is modified to accommodate a fully
featured RQL engine combustor. Consequently, the fuel in-
jectors, cowl, heatshield, combustor liner and cooling tiles
are all gas turbine engine hardware. The combustor inves-
tigated here is canted at 12 degrees to the engine center
line and contains a total of 20 symmetrically spaced fuel
injectors, fuel spray nozzles (FSNs) or “burners”. A single
FIGURE 4. Rapid prototyped NGV row and RIDN/RODN
burner sector, therefore, is defined by an 18 degree annular
cooling rings (viewed looking downstream from the combustor).
segment, containing one burner centered in the circumfer-
ential direction.
The test section is also designed to be operated with and let guide vane (OGV) chord Reynolds number in excess of
without a full set of 40 NGVs (i.e., two NGVs per burner the critical value. In fulfilling this condition for the com-
sector). The uncooled NGV pairs are manufactured using a pressor, the air mass flow rate delivered to the NGV passage
rapid prototyping process. Figure 4 shows a photograph of is 2.51 kg/s, resulting in a Mach number approaching the
the vane row in situ. The figure also shows the rear inner passage of 0.04. The Reynolds number based upon the NGV
discharge nozzle (RIDN) and rear outer discharge nozzle chord is 0.60 × 105 .
(RODN) arrangements. Both the RIDN and RODN are
comprised of multiple rows of discrete, angled holes on the
Instrumentation
inner and outer endwalls, respectively, just upstream of the
The experimental data presented in this paper relate to
NGV row leading edge. As indicated in Fig. 6, the RODN
turbulence information recovered using a miniature single-
jets are fed from the outer annulus while the RIDN flow
sensor 5 µm wire probe (Dantec 55P11), driven by a Dantec
is sourced from the inner annulus of the combustor. The
Dynamics StreamLine constant temperature anemometer
purpose of the RIDN and RODN jet flows is to cool the
(CTA) system. The wire overheat ratio is 1.8 and for the
vane platforms and adjust the combustor exit temperature
velocity magnitudes presented here, the overall frequency
profile prior to the hp turbine.
response (−3 dB) is in excess of 50 kHz. For air under am-
As indicated in Fig. 3, the air that would normally be bient conditions, the cooling of the CTA wire is a function
bled from the combustor feed annuli to cool the NGV row of flow velocity, temperature and direction. The response of
and turbine disc is instead metered through a series of ori- the probe to velocity is usually described using the following
fice rings located towards the rear of each annulus. The simple power law
size of the orifices contained in each ring can be adjusted
to distribute the air exiting the pre-diffuser in the correct
proportions for the combustor and its feed annuli. For ex- E 2 − A = BUeff
2
, (1)
ample, when the NGV row is fitted into the test section,
the increased blockage presented to the flow at combustor where E is the bridge voltage, Ueff is the effective cooling ve-
exit requires readjustment of the orifice rings to return the locity, and A and B are the calibration coefficients. For the
flow to the correct distribution. experiments described here, the coefficients in Eq. (1) are
derived over an appropriate range of velocities and constant
known flow temperature (Tamb,cal ) using a Dantec Stream-
Operating Conditions Line Pro calibrator. Although the CTA wire responds to
A full description of the control and instrumentation all 3 velocity components, these cannot be resolved into
system developed to run the test rig is provided by Den- orthogonal flow directions using a single-wire sensor. The
man [8]. In brief, the 1.5 stage axial flow compressor
√ is op- sensitivity of the 55P11 probe, however, is greatest when the
erated at a constant non-dimensional speed (N πD/ γRT ) total velocity vector is aligned normal to the wire/support
(±0.08%) and at a design flow coefficient (Va /Ublade ) of prongs. For this reason, wire calibration is performed with
0.560 (±0.18%). At this condition, the inlet Mach num- the reference flow vector coincident with this direction of
ber to the compressor stage remains constant. The design maximum sensitivity. Using this approach, the quality of
speed of the machine, however, is chosen to provide an out- fit to the model described by Eq. (1) is better than 99.9%.

4 Copyright © 2012 by Rolls-Royce Corporation

 (a) Spectra be applied to compensate for temperature changes up to
±5◦ C, for the experiments described here the air tempera-
ture was never more than ±3◦ C from the calibration con-
dition.
The software used to capture and process the output
signals from the CTA unit is developed in LabVIEW 7.1.
The system processes CTA data online to provide contours
of mean wire velocity, turbulence intensity and integral
length scale. In forming this information, a statistically in-
dependent data set is obtained by recording 16 × 32k blocks
of data sampled at a rate of 40 kHz. Power spectral den-
sity (u02 /Hz) spectra of the fluctuating velocity component
(u0 ) are formed from each block using the discrete Fourier
transform (DFT) technique. The spectra are ensemble av-
eraged and integrated to provide u0rms and local turbulence
(b) Autocorrelation intensity (u0 /U∞ ≡ Tu) at every point in the flow. The
space-time autocorrelation function (R(ξ, τ )) is derived at
ξ = 0 from the inverse Fourier transform of the ensemble av-
eraged power spectra and the integral time scale calculated
as

Z τ0
Tint = R(0, τ )dτ . (3)
0

Examples of typical power spectral density and auto-

correlation distributions measured using this approach are
shown in Fig. 5.
As suggested by Hinze [10], the integral time scale of
the dominant, energy containing eddies can be obtained by
integration of the autocorrelation up to the first crossing
FIGURE 5. Typical (a) turbulence spectra and (b) autocorre- point of the time axis (τ 0 ). For modest turbulence intensity
lation function measured using the CTA. levels up to about 25%, the stream turbulence is assumed to
be convected at the measured flow velocity (Bradshaw [11]).
The corresponding integral length scale (`) is then formed
During the experiment, the output voltage (E) from from the product of the integral time scale (Tint ) and the
the CTA unit is separated into mean and fluctuating com- mean flow velocity (Ueff ) measured by the CTA wire.
ponents before being filtered (anti-aliased) and amplified Under isotropic turbulence conditions, Bruun [9] dis-
appropriately using a National Instruments SCXI-1143 sig- cusses the errors that occur in the indicated value of Ueff
nal conditioner module. The signals are sampled at 40 kHz in the presence of moderate to high levels of turbulence
using a PCI-6052E 16-bit data acquisition card and recom- intensity when using the simple power law calibration rela-
bined digitally. The effects of drift in the ambient temper- tionship described by Eq. (1). For Tu < 30% the turbulence
ature are compensated for using the following expression correction factor applied to Ueff is less than 4%. However,
described by Bruun [9]: as the turbulence intensity increases beyond 35% the flow
begins to reverse (Ueff < 0) for some of the time and at Tu
0.5
= 50% the frequency of flow reversal occurrence increases

Tw − Tamb,cal
Ecor = E . (2) to ∼ 2% and the turbulence correction factor approaches
Tw − Tamb
12%.
Here, Ecor is the corrected wire voltage, Tw is the wire op-
erating temperature and Tamb,cal and Tamb are the ambient Measurement Planes
temperatures recorded during calibration and experiment The CTA measurements are made with and without
respectively. Although this simple correction method may the NGV row fitted at the two traverse planes shown in

5 Copyright © 2012 by Rolls-Royce Corporation

 components in the axial/radial plane but is insensitive to
vectors in the tangential direction.
For the experiments described here, a complete area
traverse is formed at each traverse plane by moving the CTA
probe radially and circumferentially with respect to the test
rig centre line. Radial probe traversing is achieved using a
stepper motor powered linear guide mounted to the inner
casing of the test section shown in Fig. 3. The positional
resolution of the traverse is ±0.025 mm, which, together
with the data acquisition system, provides a radial posi-
tional accuracy of better than ±0.1 mm. Circumferential
probe movement is achieved by rotating the inner casing
about the rig centreline using a high-power stepper. Using
this arrangement, the sensor is positioned to within ±0.05◦
of the desired location. The spatial resolution in the cir-
cumferential and radial directions is 21 × 21 respectively,
resulting in a total of 441 points distributed across the two
NGVs of a single burner sector.

SIMULATION
The large-eddy simulation (LES) of the coupled com-
bustor and turbine system is described. Detailed here are
the computational domain, mesh, boundary conditions, and
solution parameters.
Figure 7 shows the computational domain of the cal-
culations. The domain is a single periodic sector encom-
FIGURE 6. Measurement location details and traverse planes.
passing one combustor fuel injector, both inner and outer
combustor annuli, and a pair of turbine nozzle guide vanes.
Cha et al. [1] argue that Fig. 7 is the simplest do-
Fig. 6. Traverse Plane A corresponds to the combustor- main that can be used to describe boundary conditions
turbine interface plane at which the boundary conditions into a turbine CFD model (or turbine rig test) and, on the
are normally defined for the hp turbine. Located immedi- combustion-side, properly describe the combustor exit flow
ately downstream of the uncooled NGV row, traverse Plane conditions due to the upstream influence of the NGVs up
B is coincident with the location at which combustor exit to and even beyond the combustor-turbine interface plane
data (i.e., CO2 , CO, NOx , etc.) are normally recorded to (Plane A in Fig. 6).
characterise both the emissions performance and the exit Care is taken to generate the single mesh for the coupled
temperature traverse delivered by the combustor. In the combustor-turbine domain. This is not trivial as meshing
reacting flow case, these data are recorded in the absence best-practices are inconsistent for the combustor and tur-
of the NGV row and extrapolated upstream to Plane A to bine taken separately. For the combustor, little attention is
provide the hp turbine entry boundary conditions. placed on resolving the boundary layers (with exception of
As indicated earlier, the single-wire CTA probe is re- the injector swirler passages). Also, mesh smoothing is de-
sponsive to all three velocity components. Its sensitivity, structive to the generation of a hex-dominant core, required
however, is at a maximum for velocity vectors that are nor- for accuracy of an LES computation in the combustor. For
mal to the sensor wire but falls almost to zero for compo- the turbine NGVs, resolving the boundary layers is key and
nents that are parallel to the wire. For this reason, the therefore good smoothing necessary to generate the prism
alignment of the CTA probe at both traverse planes is set layers over the airfoil shapes, fillets, and other complex end-
to provide maximum sensitivity in the predominant flow wall geometric features. ICEM CFD [12] is used to generate
direction (i.e., normal to the wire/support prongs). The the optimal, coupled combustor-turbine domain mesh.
latter is then coincident with the orientation of the cali- For the turbine vanes, an iterative approach is required
bration vector described earlier. At Plane A, for example, to generate a sufficiently refined boundary layer mesh. To
this alignment means that the sensor responds to velocity simplify generation of this mesh, a constant, six prism layer

6 Copyright © 2012 by Rolls-Royce Corporation

FIGURE 7. Computational domain of the numerical experiment: The simplest-no-simpler geometry that must be used to properly
model the inflow to the HP turbine.

mesh with constant first-layer mesh thickness (∆wall ) and The Taylor lengthscale is different for the combustor
geometric expansion ratio is used over the entire NGV as- and turbine due to the characteristic length and velocity
sembly. The assembly includes the NGV platform and outer differences between the two subsystems. Assuming locally
endwall, the vanes, and fillets. The iterative procedure in- isotropic turbulence, the Taylor scale is estimated using [13]
volves generating a new mesh with a change of the boundary
layer mesh parameters and performing RANS CFD simu-  1/2
lations to yield target values for y + = Uτ ∆wall /ν. For the λ 15 `
= , (4)
present rig operating condition, y + ∼ O (1) upstream of the L Re` L
NGVs and increases to O (10) approaching the exit of the
domain. where Re` ≡ u0 `/ν is the outer scale turbulence Reynolds
A global cell size of ∆global = 3.0 mm is used for number and L is a characteristic geometric length.
the ICEM mesh throughout the coupled combustor-turbine For the combustor, λ ≈ 1.4 mm taking `/L ≈ 0.1 and
computational domain. The ICEM octree algorithm is used u0 /U ≈ 0.2. Here, L is the combustor duct height and U
to generate the background mesh and so ∆global character- the area-averaged mean axial velocity, both measured at the
izes the square hex cell size in the fluid regions sufficiently combustor-turbine interface (Plane A in Fig. 6).
far from solid surfaces. For the combustor, this is most of With hindsight, the illustrative values of `/L = 0.1 and
the domain. To simplify the mesh generation, no density u0 /U = 0.2 underestimate the actual turbulence levels as
points are used to reduce the cell size in the turbine vane measured in the present experiment. Following Eq. (4), the
passages. However, since the volume of these passages are corresponding estimates for λ therefore yield the approx-
small and the surface mesh size of the turbine NGV as- imate lower bound for λ. This then corresponds to the
sembly much less than 3.0 mm, this yields a more refined most pessimistic grid resolution requirements of resolving
mesh in the turbine, although no longer hex-dominant. Fig- the Taylor lengthscale in the LES.
ure 8 shows the hex-dominant core of the combustor mesh For the turbine, λ ≈ 0.5 mm, again taking `/L ≈ 0.1
and relative fidelity of the surface mesh on and around the and u0 /U ≈ 0.2. Here, L is now the NGV chord length and
NGVs. U the peak velocity through the NGV passage.

7 Copyright © 2012 by Rolls-Royce Corporation

 (a) Mesh cutplane TABLE 1. Taylor lengthscale and Reynolds numbers for the
current rig test compared to a representative, high-power engine
test condition.

Current rig test Engine test

Case CTI HPT CTI HPT
λ 1.4 mm 0.5 mm 0.7 mm 0.2 mm
Reλ 2 × 102 4 × 102 3 × 102 10 × 102
ReL 1 × 105 2 × 105 2 × 105 20 × 105

(b) Pressure-side and U is the mean, area-averaged velocity there. For the
HPT or high pressure turbine, L is the NGV chord length
and U the local peak velocity magnitude through the vane
passage. Also tabulated are the corresponding values scaled
to a representative high-power case or “Engine test” condi-
tions. This includes the effect of high pressure and tempera-
ture on ρ and µ and choked conditions in the vane passage.
Table 1 shows the Taylor lengthscale is O (1) mm for all
cases and conditions except the HP turbine at engine con-
ditions, where it decreases to O (0.1) mm.
For the presently used global mesh size of 3 mm, the
resulting mesh size is 10 million cells. (This is 3 million
more than the RANS mesh in [1].) Reducing the gobal
(c) Suction-side mesh size to 1 mm would result in a mesh of 270 million
cells, making unsteady simulations impractical. Typically,
LES of combustors are computed using RANS meshes of
∼2 million cells [14, 15]. The fidelity of the present LES
simulations are a factor of ∼5 larger than this.
The LES flow equations are integrated using the
pressure-based solver in Fluent 6.3 [16]. The mass-flow
boundary conditions are obtained from an in-house code.
The flow-field is initialized with steady-state RANS calcu-
lations. The unsteady LES is run for a single flow-thru time
(Tflow ) to purge the initial conditions before time-averaged
statistics are computed. Tflow is the combustor subsytem
residence or flow-thru time. Beyond this time, a timestep
FIGURE 8. Some details of the unstructured mesh for the cou- of ∆t = 5 µsec is used with 20 iterations per timestep.
pled combustor-turbine domain: (a) Cutplane showing the hex- The wall-adapting local eddy-viscosity (WALE) model with
dominant mesh of the combustor volume, (b) resolution of the Cw = 0.5 [17] is used for the subgrid-scale modeling. A total
surface mesh on the NGV pressure-side, and (c) the surface mesh simulation time of ∼5 × Tflow has been used in computing
on the NGV suction-side. the time-averaged statistics presented in this paper.

Table 1 summarizes these values. Also given are the RESULTS AND DISCUSSION
Reynolds numbers, given by Reλ ≡ ρu0 λ/µ and ReL ≡ Figure 9 shows contours of the normalized steady-state
ρU L/µ. “CTI” corresponds to reference values taken at axial velocities at Plane A. Plane A is the combustor-turbine
the combustor-turbine interface plane, i.e., L is the duct interface plane, just upstream of the NGVs (cf. Fig. 6). The
or annuli height at the combustor-turbine interface plane view here and throughout the remainder of the paper shows

8 Copyright © 2012 by Rolls-Royce Corporation

 U ∗ (r, θ) U ∗ (r, θ)
Rig data

RANS
FIGURE 10. Relative circumferential locations of the NGVs
with respect to Plane A and B.
RANS

per, the turbulence intensities to be presented are multiplied

by the CTA velocity and renormalized by the mean axial
velocity. In Cha et al. [1], the different combustor annulus
heights at planes A and B were used to normalize the tur-
bulence lengthscales. Here, they are normalized by a fixed
constant, C, the NGV chord length. The diameters of the
ports are ∼0.2C. The duct heights are comparable to C,
e.g., H ∼1.3C at Plane A.
The LES CFD results in Fig. 9 show the time-
averaged filtered axial velocity solutions. Typically,
LES

circumferentially-averaged quantities, like the radial tem-

perature distribution function (RTDF) [1], are exchanged
between Combustion and Turbine designers. Although a
single Tflow is sufficient to yield statistical stationarity for
the RTDF due to the additional circumferential spatial av-
erage, complete stationarity is not reached for the two-
dimensional fields. Even statistics computed over 5Tflow ,
as in the LES results of Fig. 9, exhibit some variation
when additional realisations are taken in the time averag-
ing. Meaningful trends however can still be inferred from
FIGURE 9. Contours of normal velocity at the combustor- the first-order, two-dimensional statistics. For this system,
turbine interface (Plane A in Fig. 6). Values have been nor- the non-stationarity at even 5Tflow is likely caused by the
malized by the steady-state, area-averaged axial velocity at this large turbulent lengthscales created by the combustor (pre-
plane. The view is looking upstream from the turbine. sented below).
For reference, RANS CFD predictions are also shown
in Fig. 9 and throughout the paper. The RANS simula-
results looking upstream. tion results are obtained from the realizable k–ε turbulence
The rig data is presented showing the discrete radial model. (Details of the RANS CFD are given in [1].)
and circumferential measurement locations. This highlights Figure 10 shows the relative circumferential locations
the fidelity of the experimental data, useful in later discus- of the NGVs with respect to Planes A and B (cf. Fig. 6
sions. The rig data in Fig. 9 has been obtained using a for their axial locations). The RANS results in Fig. 10 are
five-hole pneumatic probe [1]. In the remainder of the pa- identical to those in Fig. 9.

9 Copyright © 2012 by Rolls-Royce Corporation

 u0 /U `/C
Plane A
Plane B

FIGURE 11. RANS CFD predictions of the turbulence from Cha et al. [1]. Contours of turbulence intensity (left column) and
lengthscale (right) at Plane A (top row) and Plane B (bottom). Plane A is the combustor-turbine interface plane. Note, in the present
paper, Plane B is the post-NGV plane. Results are normalized by the area-averaged mean axial velocity (U ) and NGV chord (C).

Review of Cha et al. [1] relative penetration of the inner and outer annuli jets con-
Cha et al. [1] describe how the complex flow patterns trol the midspan position of the highest normal velocities
at the combustor-turbine interface plane evolve from the at Plane A. For each burner sector, there are two secondary
interaction of the dilution jets in the upstream combustor. dilution ports or jets which are staggered off the burner cen-
Briefly, two rows of large dilution ports (cf. Fig. 7 in this terline (cf. Fig. 7). This has a relatively small impact on the
paper) are used to stage the combustion processes. The first circumferential pattern of the velocities at Plane A. Rather,
stage occurs in the primary combustion zone, created by the the circumferential variation seen consistently in all three
swirling fuel nozzle flow and primary dilution port jets. It is subplots of Fig. 9 is mainly due to the upstream impact of
in this overall rich, primary zone wherein the fuel is burned the downstream NGVs at this, the combustor-turbine in-
and flame stabilized. The first row of jets created by the terface plane [1].
primary ports are used to rapidly quench the flame, thereby Both primary and secondary dilution ports are arranged
inhibiting thermal NOx and smoke production. Control of to promote fast mixing, which the combustor relies on to
emissions is the main goal of this second, or intermediate meet many of its design requirements. The ensuing tur-
combustion zone defined by the region between the primary bulence characteristics at the combustor-turbine interface
jets and the downstream, second row of ports. The row of plane due to the interaction of these upstream opposed jet
jets created by the second row of ports further oxidize the flows have been predicted by CFD in [1]. These predictions
soot and mixes out the traverse. This third region between are summarized here in Fig. 11.
the secondary ports and combustor-turbine interface plane Figure 11 shows turbulence intensity (left column of
is called the dilution zone. subplots) and turbulence lengthscale (right column) at
Although the dilution jets are in cross-flow, they pen- Plane A (top row) and Plane B (bottom). (The axial loca-
etrate deep into the combustion chamber [1]. This cre- tions of planes A and B with respect to the NGVs have been
ates the surge of momentum near the radial midspan at shown in Fig. 6.) The corresponding area-averaged mean
the combustor-turbine interface plane seen in Fig. 9. The normal velocities at planes A and B are used to normal-

10 Copyright © 2012 by Rolls-Royce Corporation

ize the turbulence intensities, while a constant NGV chord Pre-NGV plane
length is used to normalize the turbulence lengthscales for Figure 12 shows the experimental measurements of u0
both planes A and B. and ` at Plane A. The turbulence intensity and integral
The turbulence intensity from the RANS simulations lengthscale are calculated from the hot-wire spectra and
are computed using time autocorrelation function, respectively, at each of the 21
× 21 measurement locations shown (cf. the Experiment sec-
r tion for details). The data shows the high turbulence levels
0 2 responsible for the large eddy-viscosities at the combustor-
u = k , (5) turbine interface plane that were computed for Fig. 2 in
3
this paper. (The turbulence viscosities in Fig. 2 have been
calculated using the values shown in Fig. 12.)
where k is the turbulence kinetic energy. k is obtained di- The turbulence is not perfectly periodic over a burner
rectly from the solution of its transport equation in the sector as the mean fields are not periodic. The causes for the
realizable k–ε model. breaking of symmetry in the mean fields is detailed in [1].
The turbulence lengthscale from RANS is approxi- Comparing Fig. 12 with Fig. 11 (top row) reveals that
mated using the RANS model overpredicts the peak turbulence inten-
sities by an absolute margin of ∼15% (peak intensities of
∼50% are seen in Fig. 11 vs. ∼35% in Fig. 12). The mag-
k 3/2 nitudes of turbulence lengthscale are overpredicted by the
`= , (6)
ε RANS by a factor of ∼2 (∼50% in Fig. 11 vs. ∼25% in
Fig. 12).
The distribution of the turbulence in Fig. 12 suggests
where ε is the dissipation of the turbulence kinetic energy,
that the upstream impact of the NGVs on the turbulence
obtained directly from the solution of its transport equation.
development is stronger than predicted by the RANS mod-
Figure 11 shows a large band of high turbulence inten- eling. The RANS simulations show, for example, that the
sities varying mainly in the radial direction and occuring at peak values of u0 redistribute themselves along the leading
mid-span. This is created by the large-scale velocity gra- edge of the vanes leading to a circumferentially “banded”
dients produced by the upstream collision of the opposed pattern [1]. This pattern is created by the flow redistribut-
dilution jet flows. These velocity gradients produce the tur- ing itself to navigate around the NGV blockages. Here,
bulence leading to the high values of k. the rig data measurements of Fig. 12 suggest this upstream
The RANS model predicts peak turbulence intensities influence of the NGV potential field on the turbulence oc-
∼50% of U at the combustor-turbine interface (Plane A). curs further upstream than the RANS modeling predicts.
By NGV exit (Plane B), the intensities relative to an in- This interpretation is consistent with the expected decrease
creasing mean axial velocity have decreased but not by the in the magnitudes of the turbulence levels as the flow ap-
factor of ∼2 increase in U . Rather, turbulence kinetic en- proaches the NGVs, as predicted by the RANS in [1].
ergy is produced in the vane passages, created by the veloc- Figure 13 shows the LES predictions of turbulence in-
ity gradients from the bulk flow redistributing itself through tensity and lengthscale at Plane A. For the LES, the tur-
the NGV passages [1]. The result is a hub-biased distribu- bulence intensity is also calculated from Eq. (5) but with k
tion of peak turbulence intensities of ∼35% of the larger U computed from
at Plane B. The highest turbulence levels appear behind the
NGV passages and not the turbulent NGV airfoil wakes. 1 02
u + v 02 + w02 ,


At Plane A, the RANS CFD modeling also predicts k≡ (7)
2
large magnitudes of the turbulence lengthscale, ∼50% of
the NGV chord length. That ` scales with the large dis- the trace of the time-averaged Reynolds stress tensor.
tances between dilution ports and not the much smaller For the LES, the turbulence lengthscale is approxi-
port diameters follows from the production of turbulence mated by
by the large-scale interactions of the dilution jets flows [1].
By NGV exit (Plane B in this paper), the RANS modeling √
predicts that the absolute turbulence lengthscales have de- `= kτ 2 , (8)
creased. This is physically reasonable, as the outer geomet-
ric lengthscale now becomes the smaller-scale vane passage where k is given by Eq. (7) and τ is the time-scale character-
size. izing the large-scale velocity gradients, estimated here using

11 Copyright © 2012 by Rolls-Royce Corporation

 u0 /U `/C
Plane A

FIGURE 12. Rig data contours of turbulence intensity (left column) and lengthscale (right) at Plane A. Results normalized by the
area-averaged mean axial velocity (U ) at this, the combustor-turbine interface plane, and NGV chord (C).

u0 /U `/C
Plane A

FIGURE 13. LES contours of turbulence intensity (left column) and lengthscale (right) at Plane A. Results normalized by the area-
averaged mean axial velocity (U ) at this, the combustor-turbine interface plane, and NGV chord (C).

τ = 1/|S| with |S| the mean strain-rate (the time-average of tration of the dilution jet flows into the combustion chamber
the trace of the anti-symmetric part of the filtered velocity resulting in a highly unsteady and unstable interaction with
gradient tensor). the opposed jet, as was inferred by the time-averaged fields
Equation (8) is the third definition of turbulence length- from RANS results dissected in [1]. In terms of the distri-
scale used in the paper, motivated by convenience in esti- bution of u0 , the circumferentially-continuous band of high
mating ` from the available rig test, RANS, or LES data. turbulence intensity levels near mid-span suggests that the
Although these definitions follow from scaling arguments, LES is converging more towards the steady RANS solution
the proportionality constants have not been explicitly writ- (cf. Fig. 11) and thus the upstream impact of the NGVs on
ten. Implicit is that all scaling constants are identically the turbulence characteristics at Plane A is also weaker in
one. the LES than what the rig data suggests.
Figure 13 shows the peak magnitudes of u0 predicted by For the turbulence lengthscale, the LES shows an un-
the LES are ∼35% of the mean axial velocity at Plane A. derprediction of the peak magnitudes seen in the rig data
This is in better agreement to the rig test data (Fig. 12) than by a factor of ∼2 (note the different contour scales of `/C
the RANS predictions (top row of Fig. 11). This is some- between Fig. 12 and Fig. 13). Although this is within reach
what surprising as animations of the upstream combustor of an order-unity multiplicative constant in Eq. (8), we do
flow from the LES show, at instances, a much deeper pene- not discount the potential need to continue the ensemble

12 Copyright © 2012 by Rolls-Royce Corporation

 (a) P rig data Fig. 14 (b) are also shown. The experimental data shows
that the turbulence levels for both u0 and ` have approx-
imately doubled from Plane A: Compared to Fig. 12, the
peak values of the turbulence intensity have increased to
∼65% at Plane B while the turbulence lengthscale has in-
creased to ∼50%. The largest magnitudes of u0 at Plane B
are distributed following the NGV wake shapes but do not
coincide with their circumferential location, implying that
they are not a result of the turbulence generated by the air-
foil wakes. Similarly, the distribution of the high ` levels in
(b) Linearly interpolated contours of P Fig. 15 also coincide with the NGV passage flows. The high
turbulence lengthscales are distributed more evenly across
the NGV passage whereas the high turbulence intensities
are all localized close to the NGV wake positions but do
not directly coincide with them.
Returning to Fig. 11, the RANS modeling predictions of
u0 and ` at Plane B (bottom row of subplots in Fig. 11) show
limited similarity to Fig. 15. The RANS modeling under-
predicts the peak values of u0 , but most importantly, shows
a very different distribution of u0 . Recall, from the discus-
FIGURE 14. Normalized total pressure contours at Plane B
sion surrounding Fig. 11, the RANS modeling also predicts
(NGV exit). Subplot (a) shows P for each discrete measurement
turbulence energy to increase from Plane A to Plane B.
location and (b) linearly interpolated contours using the same P
At Plane B, however, the regions of largest u0 predicted by
data. Subplot (b) is used to generate the black contour lines of
RANS are relatively diffuse compared to even the coarse
constant P to locate the NGV wake locations.
resolution in the rig-test data and has predicted them to
peak at a location in the NGV mid-passages. The RANS
averaging over the ∼5 Tflow simulation time reached by the model does a better job of predicting distributions of ` than
present LES calculations. This is because the turbulence u0 . Discounting any multiplicative constant in the RANS
lengthscale characterises a second-order statistic. The LES scaling estimation, the peak levels of the turbulence length-
predictions of ` magnitudes are more than a factor of ∼4 scale are underpredicted by a factor of ∼2.
less than the RANS results (Fig. 13 vs. Fig. 11). As the LES It is useful to consider an axial location just upstream
is able to resolve smaller scales than the RANS, a complex of the NGV trailing edge to better highlight the strengths
distribution of ` is revealed which admits no apparent pat- and weaknesses in the CFD modeling. This plane is a small
tern on scales less than C. distance upstream of Plane B and removes any obfuscation
in the predicted turbulence levels that might be due to the
component from the NGV wake flows.
Post-NGV plane Figure 16 shows the time-averaged turbulence charac-
Pivotal for discussions of the turbulence characteristics teristics from the CFD modeling at a plane near the NGV
toward NGV exit (Plane B) are the coincident total pressure trailing edge. The NGV geometry has been included in the
measurements made at NGV exit. Figure 14 (a) shows the figure to provide a spatial point-of-reference. The figure
rig test measurements made for total pressure, P , at the shows the highest turbulence levels predicted by RANS are
discrete measurement locations of Plane B. The deficit of diffusely distributed at the center of the NGV passages and
pressure corresponds to the NGV airfoil wake flows while biased towards the hub, produced by the larger mean ve-
the high values of P show the result of the migration of locity gradients there. (The asymmetry between the two
the NGV passage flow towards the hub. (The surge of flow passages is also due to a larger mean velocity scale but one
near the casing is due to the upstream RODN flows but which varies circumferentially, as shown in Fig. 9.)
is characterized by a smaller local velocity scale than the In stark contrast, the LES results in Fig. 16 show the
bulk passage flow.) Figure 14 (b) shows a trace of the NGV largest levels of u0 to be highly localized and appearing
wakes constructed by linearly interpolating the 2-d pressure mainly near the suction surface of the NGVs. The predicted
field. magnitudes are also in better agreement to the rig test mea-
Figure 15 shows the rig test measurements of u0 and ` surements over the RANS results. However, the LES again
made at Plane B. The interpolated contour lines of P from underpredicts the peak turbulence lengthscale magnitudes

13 Copyright © 2012 by Rolls-Royce Corporation

 u0 /U `/C
Plane B

FIGURE 15. Downstream of NGVs plane: Rig data of turbulence intensity (left column) and lengthscale (right) contours.

u0 /U `/C
RANS
LES

FIGURE 16. Contours of turbulence intensity (left column) and lengthscale (right) from CFD at a position just upstream of Plane B.

as was also seen at Plane A, this time by a larger factor tensor, ∇u, and is defined as
as the LES predicts a decrease in ` by a factor ∼2 moving
from Plane A to Plane B. Unlike Plane A, the LES now 1
|Ω|2 − |S|2 ,


reveals some physical pattern to the turbulence lengthscale Q≡ (9)
2
distributions in the form of radially-stacked, ovular struc-
tures which are aligned circumferentially and almost span
where Ω and S are the symmetric and anti-symmetric parts
the entire width of the NGV passages.
of ∇u, respectively. With direct numerical solutions of
u, i.e., from the unmodelled Navier-Stokes equations, the
physical interpretation of the “Q-criterion” follows directly
Visualisation of passage vortices from kinematics: Sufficiently large values of Q represent re-
gions of the flowfield where rotation (Ω) dominates strain
The coherent vortex structures which physically charac- (S), thereby defining a coherent vortex or an eddy of some
terize the turbulence are visualized using the LES solutions. particular size in a turbulent flow. Jeong & Hussain de-
This is done using the Q-criterion of Hunt et al. [18]. scribe the “inadequacy” of the Q-criterion [19].
Briefly, Q is the second invariant of the velocity gradient In the present application, Q is constructed with the fil-

14 Copyright © 2012 by Rolls-Royce Corporation

tered (modeled) velocity field from the LES solutions. The onto the suction side approaching the NGV exit may be re-
interpretation of the Q-criterion is assumed to be main- sponsible for the build-up of the high turbulence intensity
tained since the largest-scale structures directly computed there as these eddies also dissipate before reaching the NGV
by the LES are essentially inviscid. exit.
Figure 17 shows isocontours of Q = Q0 , an arbitrary Numerous studies [3, 5] suggest various detailed inter-
constant. The combustor domain has been truncated only actions of vortices in a single vane passage. More detailed
for the figure (recall the simulation has included the whole inspection of visualisations like Fig. 17 and Fig. 18 indeed
domain shown in Fig. 7). The constant Q0 contour has been reveal interactions like the merging of smaller vortices into
colored by radial distance to aid the eye in discerning the larger ones. Due to the large range of eddy sizes in the NGV
radial span location of a vortex. In this figure, the red and passages (due to the high Reynolds number turbulence from
blue eddies correspond to the combustor annuli flows—the the combustor), it is expected that no previously hypoth-
red and blue eddies do not enter the NGV passages. esized interactions can be precluded if pursued here with
The figure shows there is little organized structure to sufficient effort!
the eddies exiting the combustor: The eddies at the scales The seminal classic (inviscid) picture describes the key
shown appear to be tangled and with no apparent orienta- organization of two dominant passage vortices [20]. In a
tion (isotropic). Near the end-walls, there is no coherent given NGV passage, these two vortices counter-rotate with
vortex aligned in the circumferential direction as any sem- respect to each other while they co-rotate with their adja-
blance of a boundary layer is dispersed by the RIDN and cent neighbor in the next passage.
RODN jet flows. Indeed, it is the smallest (least coher- Figure 19 attempts to portray the bulk rotation of the
ent) eddies which appear near the end-walls at the NGV coherent vortices in the NGV passages and therefore the
leading edges due to the small RIDN and RODN jets. Up- secondary flows. The LES dataset has been rotationally
stream of the RIDN and RODN holes (in the combustor), copied to aid discussion (recall only an NGV pair is actually
there also does not exist well-defined vortices aligned in the simulated). The figure shows isocontours of Q = Q0 colored
cross-stream direction as the development of the boundary by the vorticity magnitude in the direction of NGV turning.
layers there are disrupted by the combustor liner cooling. This direction is aligned with the primary bulk flow path
Figure 17 also reveals that the largest eddies which have at NGV exit and, therefore, approximately aligned with the
entered the turbine have been stretched in the primary flow largest eddies by the exit of the NGV passage. In this figure,
direction as they are accelerated through the NGV passages. red signifies positive rotation, while blue is negative.
Nearly all the surviving vortices are aligned by NGV exit. The size of the vortices are clearly seen to scale with
The coherence of the largest vortices is maintained well be- NGV passage size (witness the vortices in the NGV pair at
yond the NGV exit plane, even reaching the exit of the the top of Fig. 19). This is consistent with the LES predic-
computational domain (included in the figure). The vor- tions of ` magnitudes already observed in Fig. 16. Further,
tices appear at all radial span locations due to the same, the vortices are densely packed all along the radial span.
dense distribution of eddies entering the turbine from com- Here, the coherent structures do not reveal two dominating,
bustor exit. counter-rotating passage vortices. Instead, the dense popu-
Following Eq. (9), larger values of Q represent larger- lation of eddies counter-rotate in no particular pattern but
scale and hence longer-life vortex structures. Figure 18 perhaps to minimize resistence between all the eddies of all
shows isocontours of Q = 30 × Q0 . The figure illustrates different sizes. The densely-populated, “radially-stacked”
that the strongest or most coherent vortices appear all along vortices shown are responsible for the turbulence length-
the radial span of the NGV passages. scale distributions already observed in Fig. 16.
Qualitative comparisons between Fig. 17 and Fig. 18 An interesting organized pattern does seem to emerge
show the smallest, and therefore most short-lived, eddies are when considering the dominant eddies in Fig. 19. On
dissipated by NGV exit. This occurs everywhere, but most vane passage-to-passage scales, the largest vortices seem to
noticeably in the vane passages. This is consistent to the counter-rotate between themselves at NGV exit. This is
predicted turbulence energy production there, which follows consistent with the strong coherence of the largest stretched
from the fundamental turbulence equilibrium assumption vortices which survive clear across the entire turbine do-
made by the LES (and RANS). main, as was revealed by Fig. 18.
Inspection of the vortex trajectories in the vane pas-
sages at other values of Q (figures not shown), reveals their
expected cross-stream convection due to the pressure gradi- SUMMARY AND CONCLUSIONS
ent, with the largest eddy trajectories being least impacted. The turbulence characteristics have been measured in
The modified trajectories of many of the smaller vortices a novel combustor-turbine interaction rig wherein full com-

15 Copyright © 2012 by Rolls-Royce Corporation

 (a) Pressure-side

(b) Suction-side

FIGURE 17. Coherent vortex visualizations using isocontours of Q = Q0 (constant). Contours have been colored by radial distance
to aid the eye is discerning the radial span location of an eddy. Blue is near the hub and red is near the casing.

16 Copyright © 2012 by Rolls-Royce Corporation

 (a) Pressure-side

(b) Suction-side

FIGURE 18. Coherent vortex visualizations using isocontours of Q = 30 × Q0 . Contours have been colored by radial distance to aid
the eye is discerning the radial span location of an eddy.

17 Copyright © 2012 by Rolls-Royce Corporation

 of-magnitude larger than previously perceived.
These newly discovered results estimate the effective
viscosity ratio to be νt /ν ∼ 104 (Fig. 2). For an engine
test, this would double in magnitude extrapolating to the
higher engine Reynolds number from the present rig test
conditions (Table 1). However, the turbulence levels are
not expected to change appreciably with heat-release as it
is generated mainly by the dilution jet dynamics (cf. dis-
cussion surrounding Fig. 9 or Ref. [1]).
To simulate the turbulence development approaching
the combustor-turbine interface, a RANS simulation can be
adequate. Because RANS turbulence modeling tends to be
overly dissipative, e.g., the realizable k-ε model used in the
present studies, the turbulence levels tend to be overpre-
dicted at the combustor-turbine interface plane. However,
the distributions are in reasonable agreement to the experi-
mental data. In exchange for the additional computational
costs of LES calculations, the magnitudes of the turbulence
intensity predictions improve.
The present measurements made at the exit plane of
the NGVs show a large increase in turbulence levels from
the combustor-turbine interface: Turbulence intensities in-
crease to u0 /U ∼ 65% and lengthscales to `/C ∼ 50%
(Fig. 15 in this paper). Distinguishing the wake regions
from the NGV passage flows using the coincident pressure
measurements suggest that high turbulence energy is local-
ized on the suction side of the NGVs near the trailing edge,
while the largest ` distributions occur across the entire span
of each NGV passage.
The LES is able to simulate most of these turbulence
characteristics. The realizable k-ε model does a relatively
poor job, completely misrepresenting where the highest tur-
bulence energy is produced.
Both LES and RANS CFD predict a decrease in ` by
FIGURE 19. Coherent vortices with Q = Q0 . The isocontours
NGV exit. Although it is physical to assume a decrease in
have been colored by the magnitude of vorticity approximately
the turbulence lengthscale due to the reduced outer geomet-
aligned with the primary gas path direction at NGV exit: Red
rical scale there (the NGV passage size), this trend in ` is
signifies positive rotation, while blue is negative.
inconsistent with that derived from experiment.
In the first application of this kind, the LES has been
used to visualize the coherent vortex structures developing
bustor engine hardware has been used to generate realistic through the vane passages. The classic physical picture of
inflow conditions into the turbine. The rig test includes en- secondary flow development is scrutinized in light of these
gine turbine NGVs to account for the coupling between the never-before-seen LES visualisations of the “passage vor-
combustion and turbine subsystems [1]. tices”. Simulation of the entire combustion subsystem up-
At the combustor-turbine interface plane, the current stream of the NGVs ensures that the turbulence entering
measurements place peak turbulence intensities u0 /U ∼ the turbine subsystem is representative of the real system.
35% and lengthscales ` ∼ 25% of the NGV chord length The classic picture of the vortical flows (Fig. 1) shows
(Fig. 12 in this paper). little resemblence to the vortex structures predicted by the
Compared to previous turbine rig test experiments LES (e.g., see any one of Figs 17–19). This is because there
which attempt to model combustor exit-representative is relatively little vorticity generated at the endwalls of a
turbulence, this puts the turbulence intensity on the combustor due to the disruptions in the boundary layer
“high” side while the turbulence lengthscale is an order- development by liner cooling (and for this particular case,

18 Copyright © 2012 by Rolls-Royce Corporation

RIDN and RODN flows). Instead, the highest levels of tur- Tflow Combustor residence or flow-thru time
bulence are generated near midspan, by the unstable colli- Tint Integral time scale
sions of the opposed dilution jet flows. This high Reynolds Tw CTA wire temperature
number turbulence from the combustor creates eddies of Tu Turbulence intensity
practically all sizes (up to the peak magnitudes of `). The U Time-averaged axial velocity
result is a complex array of tangled turbulent eddies with no Ueff An effective velocity
preferred orientation that convect into the NGV passages. Uτ Boundary layer velocity scale
This physical picture is consistent with the success of k-ε U∞ A reference velocity
modeling for the rich-burn combustor subsystem only. u Velocity vector
As the eddies accelerate through the turbine, the vor- u0 Rms of fluctuating velocities
tices are stretched and, for the largest vortices, become or- x Axial direction or coordinate
ganized into a highly anisotropic pattern aligned along the y+ Boundary layer Reynolds number
primary flow direction. This picture puts the applicability ∆global Global mesh size
of any two-equation RANS turbulence model into question ∆wall Boundary layer wall mesh size
for the NGV flows. The smallest eddies are observed to Ω Rate-of-rotation tensor
be more prone to the cross-stream pressure gradient and ε Turbulence dissipation rate
mostly dissipate by NGV exit. By NGV exit, the largest θ Circumferential direction or coordinate
vortices seem to develop a counter-rotating pattern between λ Taylor lengthscale
themselves across vane passages since, for this particular µ Dynamic viscosity
case, there is no downstream rotor to disrupt their coher- ν Kinematic viscosity
ence. νt Turbulence viscosity
ρ Fluid density
τ Turbulence timescale
NOMENCLATURE

( )∗ Non-dimensional quantity ACKNOWLEDGMENT

( )0 Fluctuating (rms) quantity The lead author thanks Rolls-Royce plc for funding his
| | Magnitude of a quantity UK secondments, during which time this work was per-
A Calibation constant formed. The work would not have been possible without
B Calibation coefficient the help from numerous Turbine Systems and Combustion
C NGV chord length & Casings colleagues: Cheers to Emmanuel Aurifeille, Dou-
D Diameter gal Jackson, Mark Glover, Nick Brown, Mark Rogers, and
E CTA bridge voltage Marcus Foale.
Ecor Corrected CTA bridge voltage A special thanks also goes to Steve Gegg (Turbine Aero-
H Combustor duct height dynamics, Rolls-Royce Corp.) for reviewing the draft.
k Turbulence kinetic energy
L A reference lengthscale
` Turbulence lengthscale REFERENCES
N Rotational speed [1] Cha, C. M., Hong, S., Ireland, P. T., Denman, P. A.,
n Calibration coefficient and Savarianandam, V. “Experimental and numerical
NGV Nozzle guide vane investigation of combustor-turbine interaction using an
P Total pressure isothermal, non-reacting tracer”. ASME J. Engineer-
Q Second invariant of ∇u ing for Gas Turbines and Power. Accepted.
R Gas constant [2] Langston, L. S., 1980. “Crossflows in a turbine cascade
R(ξ, τ ) Space-time autocorrelation function passage”. Trans. ASME, J. Engineering for Power,
RIDN Rear inner discharge nozzle 102, pp. 866–874.
RODN Rear outer discharge nozzle [3] Sieverding, C. H., 1985. “Recent progress in the under-
r Radial direction or coordinate standing of basic aspects of secondary flows in turbine
S Rate-of-strain tensor blade passages”. J. Turbomachinery, 107, pp. 248–
Tamb Ambient temperature 257.
Tamb,cal Calibration ambient temperature [4] Sharma, O. P., and Butler, T. L., 1987. “Predictions of

19 Copyright © 2012 by Rolls-Royce Corporation

 endwall losses and secondary flows in axial flow turbine
cascades”. J. Turbomachinery, 109, pp. 229–235.
[5] Wang, H. P., Olson, S. J., Goldstein, R. J., and Eck-
ert, E. R. G., 1997. “Flow visualization in a linear
turbine cascade of high performance turbine blades”.
J. Turbomachinery, 119, pp. 1–8.
[6] Radomsky, R. W., and Thole, K. A., 2000. “Flowfield
measurements for a highly turbulent flow in a stator
vane passage”. J. Turbomachinery, 122, pp. 255–262.
[7] Ames, F. E., Proceedings of the 37th National & In-
ternational Conference on Fluid Mechanics and Fluid
Power, 2010. “Some aspects of the influence of flow
field turbulence on heat transfer and boundary layer
development related to gas turbines”.
[8] Denman, P. A., Proc. ASME Turbo Expo, The Nether-
lands, GT2002-30465 (2002). “Aerodynamic evaluation
of double annular combustion systems”.
[9] Bruun, H. H., 1995. Hot-Wire Anemometry—
Principles and Signal Analysis. Oxford.
[10] Hinze, J. O., 1975. Turbulence. McGraw-Hill.
[11] Bradshaw, P., 1971. Introduction to Turbulence and its
Measurement. Pergamon Press.
[12] http://www.ansys.com/products/icemcfd.asp.
[13] Tennekes, H., and Lumley, J. L., 1972. A First Course
in Turbulence. MIT.
[14] Mahesh, K., Constantinescu, G., Apte, S., Iaccarino,
G., Ham, F., and Moin, P., 2006. “Large-eddy simula-
tion of reacting turbulent flows in complex geometries”.
ASME J. of Applied Mechanics, 73, pp. 374–381.
[15] Boudier, G., Gicquel, L., and Poinsot, T., 2007. “Com-
parison of LES, RANS and experiments in an aeronau-
tical gas turbine combustion chamber”. Proc. Combust.
Inst., 31(2), pp. 3075–3082.
[16] http://www.fluent.com/software/flowizard/.
[17] Nicoud, F., and Ducros, F., 1999. “Subgrid-scale stress
modelling based on the square of the velocity gradient
tensor”. Flow Turb. Combust., 62, pp. 183–200.
[18] Hunt, J. C. R., Wray, A. A., and Moin, P., 1988.
“Eddies, stream, and convergence zones in turbulent
flows”. Proc. CTR Summer Program, p. 193.
[19] Jeong, J., and Hussain, F., 1995. “On the identification
of a vortex”. J. Fluid Mech., 285, pp. 69–94.
[20] Hawthorne, W. R., 1951. “Secondary circulation of
fluid flow”. Proc. R. Soc. Lond. A, 206(1086),
pp. 374–387.

20 Copyright © 2012 by Rolls-Royce Corporation