Académique Documents
Professionnel Documents
Culture Documents
Engine Noise
D. Casalino
A. F. P. Ribeiro
E. Fares
S. Nolting
F. Perot
Exa Corporation, 150 North Hill Drive, Brisbane, CA, 94005, USA
Y. Li
P.-T. Lew
C. Sun
P. Gopalakrishnan
R. Zhang
H. Chen
)
m, n Azimuthal and radial duct mode order
p Pressure
u Flow velocity
u
, v
, w
[f
i
(x, t) f
eq
i
(x, t)] , (2)
where is the relaxation time, which is related to the uid viscosity, and f
eq
i
is the equilibrium distribu-
tion, which is approximated by a third order expansion with constant temperature.
11
Once the distribution
3 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
function is computed, the density and linear momentum are simply determined through discrete integra-
tion: (x, t) =
i
f
i
(x, t) and u(x, t) =
i
f
i
(x, t) v
i
. All the other quantities are determined through
thermodynamic relationships for an ideal gas.
With the Chapman-Enskog expansion
12
expansion it is possible to retrieve the compressible Navier-Stokes
equations from LBM.
10, 13
However, in contrast to the Navier-Stokes equations, the current method is based
on linear formulation which relies on simple computational operations, allowing for ecient, accurate, and
highly scalable implementations. An explicit time advancement scheme is employed, which also facilitates
massively parallel simulations and the use of a very small time step, which is valuable when attempting to
simulate high frequency noise.
PowerFLOW solves the D3Q19 formulation of the Lattice-Boltzmann equation for direct numerical sim-
ulations. For high Reynolds ows, turbulence modeling is introduced
14
by solving a variant of the RNG
k model
15, 16
on the unresolved scales,
17
selected via a swirl model,
18
a method referred to as LBM Very
Large Eddy Simulation (LBM-VLES). An extended wall model including pressure gradient eects is used in
the near-wall region.
19
The LBM scheme is solved on a grid composed of cubic volumetric elements (voxels). A Variable Reso-
lution (VR) by a factor of two is allowed between adjacent regions. Consistently, the time step is varied by
a factor of two between two adjacent resolution regions. Solid surfaces are automatically facetized within
each voxel intersecting the wall geometry using planar surface elements
20, 21
(surfels).
Although PowerFLOW has intrinsic CAA capabilities and can compute the noise propagation directly
from the unsteady ow simulations, this is generally computationally expensive and limited to the near eld.
Therefore, in order to compute the far eld noise, an integral extrapolation based on the FW-H acoustic
analogy
22
is used, both using solid and permeable
23
surface formulations. A forward-time solution
24
of the
FW-H equation based on Farassats formulation 1A
25
is employed. The solver is embedded in the post-
processing tool PowerACOUSTICS, which is also used to perform statistical and spectral analysis of any
unsteady solutions generated by PowerFLOW (volume elds, surface elds, and probe signals).
The numerical methods herein described were extensively validated for a wide variety of applications
ranging from academic cases using direct numerical simulation
26
to industrial ow problems in the elds of
aerodynamics,
27
thermal management,
28
and aeroacoustics.
2, 1, 4, 6, 7
II.B. High-speed non-isothermal LB model
For the solution in the high subsonic Mach number range, e.g, ows with local Mach number greater than
0.5, a standard D3Q19 LBM is applied. The BGK collision operator in Eq.2 was replaced by a regularized
collision operator which can signicantly increase both numerical stability and accuracy when local ow
Mach number is high. An interaction force was introduced in Eq.1,
29
which can modify the equation of
state thus the speed of sound so that high Mach number ows can be simulated by a low order LB scheme.
Moreover, in order to take into account the ow heating due to compression work and viscous dissipation,
a hybrid approach was applied for the thermodynamics of energy eld, by solving the entropy equation
through a Lax-Wendro nite dierence scheme on the Cartesian LB mesh.
II.C. Acoustic porous medium model
In the LB method, external forces can be included in the uid dynamics by altering the local-instantaneous
particle distributions during the collision step. This technique can be used to model, for example, buoyancy
eects due to gravity. The solver used in the present study implements a porous medium model by applying
an external force driven by the ow resistivity and as function of the local ow velocity.
30
This model can
be used to predict pressure losses that aect the time-averaged ow eld solution and, at the same time, the
instantaneous acoustic uctuations.
The APM model in PowerFLOW, a recently patented technology by Exa Corporation,
31
is characterized
by three main parameters: the viscous resistance R, the porosity and the APM thickness d. The char-
acteristic surface impedance of the APM, say Z
APM
(f, R, , d), can be fully determined with these three
variables via an analytical model, where f is the frequency in Hertz; Z
APM
is also parameterized through
quantities that are directly related to the LB formulation, but these details are beyond the present scope.
The characteristic surface impedance Z(f) of an acoustic liner can be obtained via experiments, or via nu-
merical simulations of the complete geometry, in a normal or grazing impedance tube setup.
32
The three
4 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
parameters are then educed by minimization of the cost function J(R, , d):
J(R, , d) = max
f
(|Z(f) Z
APM
(f, R, , d)|) (3)
The cost function is minimized using a genetic algorithm for (R, , d) constrained on specic variation
ranges. The optimal results (R
0
,
0
, d
0
) are then validated with the simulation of the resulting APM in a
normal impedance tube and the computation of its numerical characteristic impedance. The target acoustic
resistance and reactance curves, their analytical tting of the LBM educed impedance curves are then plotted
for visual verication, as shown in subsection III.A.
III. Verication and validation
This section deals with the unitary verication and validation studies carried out to build up the
required knowledge in terms of ow physics and simulation setup towards a comprehensive turbofan engine
aeroacoustic analysis. Fan and jet noise sources are addressed, whereas the internal turbomachinery ow
features (compressor, combustion chamber and turbine) are not in the present scope.
Subsection III.A deals with the propagation of a spinning duct mode through an acoustically treated
intake. LBM results are compared with a FEM solution of a convected 2nd order wave equation in the
frequency domain.
33, 34
The main goal of this study is to verify the LBM propagation capabilities in the
presence of an impedance wall modelled via an APM.
Subsection III.B deals with the propagation of a spinning duct mode through a bypass duct. LBM results
are compared with a FEM solution of a convected 3rd order wave equation in the frequency domain.
35
The
main goal of this study is to perform a cross validation of sound propagation in the presence of a shear layer.
Subsection III.C deals with the problem of tonal and broadband noise generation from rotating fan
blades. The Advanced Noise Control Fan (ANCF) conguration
36
is used as benchmark validation case.
The complete 3-D in-duct rotor/stator model and the rotating rotor is simulated for the nominal value of
Mach rotor tip of 0.33 used in the experiments. Far-eld measurements conducted at the NASA Glenn
research center are used to validate the simulation results.
Finally, subsection III.D focuses on the prediction of single stream jet ow noise. The NASA Glenn
research center hot jet experiments
37
carried out with the SMC000 nozzle geometry are used to validate the
LBM solution for high speed cold and hot jets.
Throughout, the size of the simulation problems is reported in terms of equivalent number of voxels and
surfels as the solution in the whole simulation domain would be updated every time step. In other words,
one voxel or surfel belonging to the second nest mesh resolution level, for instance, would count as a half
in the overall count of Fine Equivalent Voxels (FEV) and Fine Equivalent Surfels (FES). Simulation time is
reported in terms of overall computational hours (CPUh) on a cluster of Intel Xeon X5570 2.93GHz CPUs
connected by a Mellanox FDR Inniband 56Gb/s network.
All simulations presented in this section were performed using the recently released PowerFLOW version
5.0c, with the exception of the jet ow simulations presented in subsections III.B and III.D, which were
performed using a beta release of PowerFLOW 5.1, providing the high-speed non-isothermal functionality.
III.A. Intake fan noise radiation
In this subsection we show results for the transmission of a spinning mode through an idealized aero-engine
intake. The Helmholtz number, based on the fan plane outer radius, is kR
f
=5.257. The acoustic mode (6, 1)
is considered, with a magnitude corresponding to an overall inlet acoustic power of 130 dB. The acoustic
mode is introduced in the LBM simulation through a time-varying pressure/velocity boundary condition on
the fan plane. LBM results without and with mean ow convective eects and without and with an acoustic
liner are compared against a frequency-domain FEM solution of a convected wave equation for the acoustic
potential
34, 38, 39
put forward by Pierce.
40
Simulations are carried out for a three-dimensional rigid and acoustically treated (soft) nacelle. A zero-
splice liner is considered, located on the outer wall of the intake, as shown in Fig. 1. Both for the LBM and
the FEM computations, the near-eld solution is extrapolated to the far eld through a permeable FW-H
approach. In order to reduce the impact of the FW-H extrapolation to the far-eld noise prediction, the
same integration surface is used for the two simulations, as depicted in Fig. 1.
5 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
Figure 1. Nacelle intake simulation setup.
The LBM simulations are performed using a mesh resolution that guarantees 14 points per acoustic
wavelength (without convective eects) in the near eld volume inside the nacelle up to the FW-H integration
surface. Views of the mesh for the rigid and soft cases are shown in Fig. 2. The volume occupied by the
APM is also meshed, as this is treated as a uid material with dierent properties. Quantities that are useful
to estimate the computational eort for the present case are reported in Table 1.
Figure 2. Nacelle intake simulation mesh. Rigid case (left), soft case (right).
VRs x FEV (10
6
) FES (10
6
) T
t
T
s
CPUh
9
0
/56 4.999 2.222 377/f
0
10/f
0
430
Table 1. Nacelle intake simulation properties. VRs denotes the number of variable mesh resolution levels.
As already mentioned, the properties of the APM are dened by using an analytical expression that relates
the surface liner impedance to resistivity, porosity and thickness of the APM. Then, a genetic algorithm is
used to minimize the dierence between a given (target) impedance and the one corresponding to a prescribed
set of APM parameters. Finally, an LBM simulation is performed for the optimal APM in order to educe
the eective value of the impedance to be compared with the target one and thus verify the accuracy of the
best t procedure. As an example, in this study we have tted the impedance of the well-known Ceramic
Tubular liner (CT57) measured by NASA.
41
Fig. 3 shows the comparison between the measured impedance,
its analytical tting, and the educed impedance using an APM with the optimal parameters and a simple
numerical setup similar to a Kundts tube. The comparison between the analytical tting and the educed
values is quite good, meaning that the use of the analytical impedance model is appropriate for a search
involving thousands of designs. More interestingly, the agreement between the educed impedance and the
target experimental impedance is satisfactory, as resulting from a best t over a wide frequency range. The
properties of the equivalent APM made dimensionless by the acoustic wave properties are: R
0
=11.16/f
0
,
0
=0.5742 and d
0
=0.132
0
.
The equivalent surface liner impedance corresponding to the reference frequency of the present study is
6 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
Figure 3. CT73 liner impedance tting with APM. Resistance (left), reactance (right). Experiments (red), APM
analytical tting (blue), LBM educed impedance (black).
extracted from the best t curve plotted in Fig. 3 and used to perform FEM computations in axi-symmetric
modality. The mesh used for the FEM simulation is the one required to run a case at a Helmholtz number
about 12 times higher than the present one, thus guaranteeing a mesh independent reference solution. The
time-averaged ow solution used for the FEM computations is computed by setting slip wall conditions in
the same setup used to perform the LBM acoustic simulations. The free-stream Mach number for the cases
with ow is M
=0.2.
Fig. 4 shows the comparison between the near-eld acoustic pressure at zero cycle computed using LBM
and the reference FEM solution for cases in the absence of convective eects (M
)
x/Rf [-]
LBM - liner
FEM/GIC-beta=1 - liner
FEM/GIC-beta=0 - liner
-4
-3.5
-3
-2.5
-2
-1.5
-1
-0.5
0 0.5 1 1.5 2 2.5
lo
g
1
0
(
v
)
x/Rf [-]
LBM - liner
FEM/GIC-beta=1 - liner
FEM/GIC-beta=0 - liner
Figure 6. Pressure and velocity uctuation magnitudes along the line y=4.47 10
2
R
f
parallel to the intake outer wall.
Comparison between LBM and FEM results in the presence of mean ow.
8 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
III.B. Exhaust fan noise radiation
This subsection is focused on the prediction of noise transmission through a turbofan bypass duct and the
propagation through the shear layer. The LBM solution is compared with frequency domain FEM solution
of a 2nd order wave equation for the acoustic potential,
40
the same wave model used in subsection III.A, and
to the solution of a 3rd order wave equation
35
derived from the Goldstein-Lilley
43
acoustic analogy for the
logarithmic pressure perturbation. It is important to mention that, both FEM solutions are not theoretically
suited to address the present model problem: the acoustic potential equation does not allow to take into
account the Kutta condition at the bypass nozzle edge, whereas the Goldstein-Lilley equation is strictly valid
only for a unidirectional transversely sheared mean ow, which is not the case of the present problem. The
comparison between LBM and FEM results is therefore only made for the sake of completeness, keeping in
mind that the LBM solution is the one without theoretical limitations.
The three-dimensional turbofan bypass geometry considered for both the LBM and FEM computations
is shown in Fig. 7(a). The LBM mesh is shown in Fig. 7(b). The mesh resolution in the jet shear layer is
suciently ne to resolve the time-averaged shear layer ow, but not enough to promote the occurrence of
large scale hydrodynamic instabilities that would result in high jet noise contributions, thus making dicult
to isolate the tonal noise radiation from the bypass exhaust. The acoustic mode (3, 1) is considered, with
a magnitude corresponding to an overall inlet acoustic power of 140 dB. The acoustic mode is introduced
in the LBM simulation through a time-varying pressure/velocity boundary condition on the fan plane.
The choice of this mode is due to the fact that, as discussed in Ref.,
44
this is the most energetic mode
for the fundamental Blade Passing Frequency (BPF) for the ANCF fan rotor-stator conguration used
in subsection III.C. Moreover, since the turbofan conguration simulated in section IV was derived by
inserting the ANCF fan system into a realistic nacelle whose exhaust prole is the same as the one used in
this subsection, a certain consistency among all cases can be preserved by considering the same dominant
spinning mode as for the ANCF conguration at the 1
st
BPF. The Helmholtz number, based on the fan
plane outer radius, is kR
f
=10.37.
(a) Geometry sketch (b) LBM mesh
Figure 7. Turbofan exhaust conguration.
The mean ow solution used for the FEM simulations is obtained by time-averaging the LBM solution
computed using the non-isothermal solver. LBM simulations are performed by prescribing a Mach number
of 0.3 both on the fan plane inlet boundary and on the core jet inlet boundary. The ow temperature on
the fan plane is equal to the free-eld temperature, whereas a temperature ratio of 2.5 is prescribed on the
core jet inlet boundary. A quiescent free-eld is assumed. Fig. 8 shows the contour levels of the velocity
magnitude made dimensionless by the ambient speed of sound (acoustic Mach number) and the temperature
made dimensionless by the ambient temperature (temperature ratio). Quantities useful to estimate the
computational eort for the present case are reported in Table 2.
Fig. 9 show contour levels of the real part of the Fourier component at the fan plane excitation frequency.
A qualitative comparison is made with the two FEM solutions. Both the FEM solution of the Lilley-
Goldstein equation and the LBM solution exhibit hydrodynamic perturbations in the shear layer. However,
9 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
Figure 8. Turbofan exhaust LBM time-averaged ow solution: acoustic Mach number (left) and temperature ratio
(right).
VRs x FEV (10
6
) FES (10
6
) T
t
T
s
CPUh
9
0
/80 13.3 3.2 528/f
0
10/f
0
3563
Table 2. Nacelle exhaust simulation properties.
as expected for a linear wave model, the correctness of the perturbations taking place in the FEM solution
is quite questionable. The uctuations levels in the shear layer for the LBM solution are signicantly
higher and cause a more pronounced interference pattern with the waves radiated from the bypass duct.
Interestingly, the FEM solution of the Pierce equation exhibits a weak discontinuity of the wave fronts along
a direction almost normal to the axial direction, in correspondence of the bypass duct nozzle. Such a feature
is a consequence of the fact that a potential wave model is not able to satisfy the proper condition for the
acoustic pressure at the edge in the presence of a discontinuous mean ow about the edge.
Figure 9. Real part of the pressure uctuation: FEM solution of Pierce equation (left), FEM solution of Lilley-Goldstein
equation (middle) and LBM solution (right).
A more quantitative comparison between LBM and FEM solutions is made in Fig. 10, where the noise
Sound Pressure Level (SPL) directivity plots are shown. For all solutions, the SPL was computed by using
pressure uctuations directly extracted from the near-eld solution, without any integral extrapolation. For
the LBM solution, a very narrow band integration around the tonal frequency was performed to lter out the
jet noise contribution. As depicted in Fig. 7(a), the radiation arc is centered in the midpoint of the bypass
duct outlet section and has a radius of 2.06 R
f
. Interestingly, both for the LBM and the Lilley-Goldstein
FEM solutions, the sound pressure levels up to about 40 deg are dominated by the jet perturbations. Away
from the jet inuence, the LBM and the Lilley-Goldstein FEM radiation pattern are in fairly good agreement,
with the largest discrepancies of about 4 dB taking place in the upward radiation arc. Conversely, the Pierce
FEM solution is in fairly good agreement with the other solutions only in the angular sector between 45 deg
and 75 deg; furthermore, the upward radiation levels are underestimated by about 20 dB, and this is related
to the incorrect treatment of the edge condition in the presence of a discontinuous mean ow about the edge.
This pitfall of an acoustic potential wave model is well known in the literature and inferred to the violation
of the Kutta condition at the nozzle edge.
45
Having stated the physical reliability of the fan noise radiation from a turbofan bypass duct in the presence
of a complex non-isothermal sheared mean ow, it is worthwhile to highlight some interesting feature of the
LBM solution. Although the mesh resolution both in the primary and secondary jets is not ne enough to
accurately predict the jet noise contribution, which is indeed in the scope of subsection III.D, it is interesting
to investigate the eects of the tonal excitation on the shear-layer hydrodynamic instabilities. Excitation
levels of 140 dB are strong enough to promote non-linear eects that can be investigated by computing the
10 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
0
20
40
60
80
100
120
140
160
0 20 40 60 80 100 120 140
S
P
L
[
d
B
]
Observation angle [deg]
LBM/High-Speed Formulation
FEM/Pierce Equation
FEM/Lilley-Goldstein Equation
Figure 10. Exhaust fan noise directivity. Comparison between FEM and LBM solutions.
cross-bicoherence between a reference pressure signal (#1) extracted by a probe located in the laminar core of
the secondary jet, and a pressure signal (#2) extracted at three angular locations along the noise directivity
arc depicted in Fig. 7(a), say 45 deg (downstream), 90 deg and 135 deg. The following formula was used to
compute the cross-bicoherence:
12
(f
i
, f
j
) =
N
w
|
p
1
(f
i
) p
2
(f
j
) p
2
(f
i
+ f
j
)|
2
p
1
(f
i
) p
1
(f
i
)
p
2
(f
j
) p
2
(f
j
)
p
2
(f
i
+ f
j
) p
2
(f
i
+ f
j
)
, (4)
where summations are carried out over N
w
=68 windows with an overlapping coecient of 0.5 and by using a
spectral bandwidth of f
0
/10. Results are plotted in Fig. 11 as function of the harmonic counts. Interestingly,
signicantly high non-linear coherence levels can be observed for the microphone at 45 deg, which is more
strongly aected by the near-eld noise induced by the shear-layer hydrodynamic wave packets. Non-linear
coherence can be also observed in the very low frequency range for the microphone at 90 deg.
Signal #2 at 45 deg
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
f
j
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
f
i
0
0.02
0.04
0.06
0.08
0.1
Signal #2 at 90 deg
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
f
j
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
f
i
0
0.02
0.04
0.06
0.08
0.1
Signal #2 at 135 deg
0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
f
j
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
f
i
0
0.02
0.04
0.06
0.08
0.1
Figure 11. Cross-bicoherence 12 as a function of the rst fi/f0 and second harmonic count fj/f0.
III.C. Fan noise generation and radiation
In this subsection we report some results of a validation study recently carried out by simulating the NASA
Glenn ANCF
36
fan conguration with PowerFLOW 5.0c. The simulation setup was derived from an existing
one used in the past to perform pioneering fan noise simulations and analyses.
44, 46
Despite the low Mach
tip operating condition ( 0.33), which is not representative of a real turbofan engine, and the uncertainties
associated with some geometry detail in the experimental setup and to the installation of the so-called Inlet
Control Device (ICD) around the nacelle intake, the ANCF constitutes a very useful benchmark experiment
for the validation of CFD/CAA tools.
11 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
As shown in Fig. 12, the full rotor/stator ANCF geometry, i.e. fan blades, guide vanes, hub and duct are
included in the simulation. The main properties of the ANCF conguration are: a rotor diameter of 1.219 m,
16 rotor blades, 13 guide vanes, 28 deg rotor pitch angle and stator located one stator chord downstream the
rotor. The rotor blades are fully embedded in a rotating mesh volume, as indicated in Fig. 12.
Figure 12. Sketch of the ANCF computational setup.
Three main dierences between the simulated geometry and the experimental conguration should be
pointed out: no tip clearance is simulated, the outlet shaft system is not included in the simulation model
and the eects due to the presence of the ICD on both the ow eld and the far-eld noise are not modelled.
The nominal operating condition is simulated, corresponding to 1800 rpm. The overall simulation size is
summarized in Table. 3.
VRs x[ mm] FEV (10
6
) FES (10
6
) T
t
[ s] T
s
[ s] CPUh (10
3
)
11 0.65 114.3 33.4 0.333 0.5 25
Table 3. ANCF simulation properties.
As depicted in Fig. 12, two permeable surfaces, which encompass the nacelle intake and exhaust, are
used to perform far-eld FW-H computations. These surface are fully enclosed in a mesh resolution level
VR8 (three coarsening levels with respect to the nest VR11 one) that guarantees a space/time resolution
of 20 points per acoustic wavelength and 35 samples per period, respectively, at the 7th BPF harmonic
(3.36 kHz). Far-eld noise was computed at 15 microphones located on a forward arc, which covers the
angular sector [0 deg : 90 deg] (Mic. # 1 located at 0 deg), and at 15 microphones on an aft arc covering the
sector [90 deg : 160 deg]; the forward arc has a radius of 3.66 m and is centered in the midpoint of the intake
section, whereas the aft arc has a radius of 3.66 m and is centered in the midpoint of the exhaust section.
The time-averaged pressure and Mach number eld solutions are shown in Fig. 13. The highest values
of Mach number are achieved in proximity of the rotor blades. The ow compression induced by the fan
system is clearly evident in the pressure eld. The ow eld about the nacelle lip is very regular and does
not reveal any ow separation.
Fig. 14 shows an instantaneous view of the time derivative pressure eld, which is quite eective in
visualizing the instantaneous acoustic waves. Very interestingly, a dominant wave front takes place in the
acoustic eld radiated from the nacelle intake, and the radiation direction seems to be related to the main
tonal content (fundamental BPF). Conversely, the noise radiated from the nacelle exhaust exhibits at least
two dominant wave fronts, one associated with the fundamental BPF, the other one with 2
nd
BPF. This
is a clear diraction eect from the sharp edge of the exhaust nozzle. Another interesting feature of the
instantaneous acoustic pressure eld is the complex interference pattern taking place between the waves
radiated from the intake and exhaust openings.
Noise spectra at two angular positions are plotted in Fig. 15. Both for the measured and computed
acoustic pressure signals, Fourier transforms were computed with a bandwidth equal to 30 Hz; spectra are
plotted in terms of BPF harmonic count. Some discrepancies can be observed between numerical and
experimental results, both in the tonal and in the broadband spectral components. More precisely, the
broadband components exhibit a systematic underestimation ( 5 dB), which could be due to two reasons:
a lack of mesh resolution in the rotor/stator region, resulting in more coherent turbulent structures in the
rotor wake, and the absence, in the simulation, of the rotor tip clearance, where interactions take place
12 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
Figure 13. ANCF simulation: time-averaged pressure eld (left) and Mach number (right). Front view on the bottom
extracted from the rotating mesh, cutting through the rotor.
Figure 14. ANCF simulation: snapshot of pressure time derivative in one azimuthal plane.
13 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
between the nacelle boundary layer turbulent structures and the rotor blades. It is important to mention
that the ANCF is characterized by a predominant tonal character, with tones arising by about 30 40 dB
from the broadband levels, and that it is dicult to cover such a dynamic range in CFD simulations.
20
40
60
80
100
0 1 2 3 4
P
S
D
[
d
B
/
H
z
]
BPF harmonic count [-]
Measurements
LBM simulation
20
40
60
80
100
0 1 2 3 4
P
S
D
[
d
B
/
H
z
]
BPF harmonic count [-]
Measurements
LBM simulation
Figure 15. ANCF simulation: narrow-band noise spectra at two angular locations: Mic. #11 (left) and Mic. #23
(right). Comparison between measured and computed results.
Fig. 16 shows a comparison between measured and predicted noise directivity patterns for the rst two
BPF harmonics and for the Overall SPL (OASPL) covering a broad frequency range. It is important to
mention that, for the sake of computational eciency, the noise signals along the forward and aft arcs
have been computed by performing separate FW-H integrations on the two permeable surfaces depicted in
Fig. 12, thus neglecting the interference eects between the intake and exhaust radiation that are expected
to have a non negligible eect only around 90 deg. The numerical predictions exhibit a systematic OASPL
overestimation that is due to an overestimation of the tonal peaks. This is conrmed by the BPF SPL
directivity plots of Fig. 16. The overestimation of the tonal components can be also related to a lack
of resolution in the rotor/startor region, which results in a too deterministic rotor-wake/stator-vanes
interaction. It is interesting to observe that the discrepancies between measurements and predictions are
higher in the forward direction ( 5 dB) compared to the aft direction ( 3 dB). This can be related to the
presence of the ICD in the experiments, which results in a noise transmission loss from the nacelle intake to
the microphones. Indeed, the measured forward noise levels are 2 3 dB lower than the measured aft levels.
The reader is remanded to Ref.
44
for other comparisons between measurements and predictions.
55
60
65
70
75
80
85
90
95
100
105
0 20 40 60 80 100 120 140 160 180
S
P
L
[
d
B
]
Observation angle [deg]
SPL - 1st BPF (480 Hz)
Measurements - forward arc
LBM simulation - forward arc
Measurements - aft arc
LBM simulation - aft arc
55
60
65
70
75
80
85
90
95
100
105
0 20 40 60 80 100 120 140 160 180
S
P
L
[
d
B
]
Observation angle [deg]
SPL - 2nd BPF (960 Hz)
Measurements - forward arc
LBM simulation - forward arc
Measurements - aft arc
LBM simulation - aft arc
55
60
65
70
75
80
85
90
95
100
105
0 20 40 60 80 100 120 140 160 180
S
P
L
[
d
B
]
Observation angle [deg]
OASPL in band [450 Hz : 1470 Hz]
Measurements - forward arc
LBM simulation - forward arc
Measurements - aft arc
LBM simulation - aft arc
Figure 16. ANCF simulation: SPL noise directivity along forward and aft microphone arcs in dierent frequency bands.
Comparison between measured and computed results.
III.D. Jet noise generation and radiation
The goal of this subsection is to validate the jet noise prediction capabilities of the high-speed non-isothermal
LBM formulation implemented in a beta release version of PowerFLOW 5.1. The NASA Glenn research
center hot jet experiment
37
performed by using the SMC000 nozzle was modelled for two ow conditions:
setpoint 07, characterized by an acoustic Mach number M
a
= U
j
/c
)
0.28
1 0.16 M
j
, and
U
U
j
= 1 e
1,43/(1x/Xw)
. (5)
It is important to mention that no tuning of the turbulent levels prescribed at the inlet nozzle boundary
condition was performed to get the proper value of the laminar core length, which is a genuine result
of the simulations; moreover, no random forcing at the nozzle inlet was employed to seed the turbulence
uctuations in the nozzle boundary layer, which develops naturally along the nozzle walls, thus providing the
proper boundary layer integral quantities at the nozzle exit. Therefore, the accuracy of the time-averaged
centerline velocity is a good indication of the proper bahavior of the turbulence model in the shear layer, as
well as in the wall region inside the nozzle. The streamwise velocity standard deviation velocity along the jet
centerline is compared with the curve plotted in Fig.(36) of Ref.
49
The agreement is very good for setpoint
46, whereas a signicant overestimation of the uctuation levels in the laminar core region is predicted
for setpoint 07. It should be mentioned that setpoint 07 is characterized by almost sonic conditions, thus
constituting a challenging test case for the high-speed D3Q19-based LBM ow solver. At the present stage
of the research, it is not clear whether the overestimation of the uctuation levels for setpoint 07 is due to the
computational setup or to the fact that we approached the usage limits of the high-speed LB formulation.
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 1 2 3 4 5
u
/
U
j
[
-
]
x / D [-]
Witze formula
LBM simulation
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0 1 2 3 4 5
u
/
V
j
[
-
]
x / Xw [-]
Measurements
LBM simulation
Figure 20. SMC000 simulation: setpoint 07.
37
Dimensionless centerline time-averaged (left) and standard deviation
x-velocity (right).
Finally, noise results are reported in Figs. 22 and 23 for setpoint 07 and 46, respectively. The computed
third-octave SPL spectra at 90 deg and 135 deg (downstream) are compared with measurements by Brown
& Bridges.
37
The estimated grid cuto frequency for the employed FW-H surface is about 20 kHz, and this
seems to be conrmed by the rapid drop-o of the noise levels around that frequency for both cases. The
low-frequency part of the predicted noise spectra exhibit a certain lack of statistical convergence. The noise
directivity for setpoint 07 is predicted within 1 dB accuracy up to 135 dB. For higher value (downstream)
16 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
1.1
0 1 2 3 4 5
u
/
U
j
[
-
]
x / D [-]
Witze formula
LBM simulation
0
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0 1 2 3 4 5
u
/
V
j
[
-
]
x / Xw [-]
Measurements
LBM simulation
Figure 21. SMC000 simulation: setpoint 46.
37
Dimensionless centerline time-averaged (left) and standard deviation
x-velocity (right).
noise levels are signicantly underestimated, with the maximum predicted levels about 5 dB lover than in
the experiments. This is due to the fact that we omitted the cup of the permeable surface from the FW-H
integration, thus neglecting part of the noise contribution generated by the largest wave packets that are
very eective at shallow radiation angles, in particular in almost sonic conditions. The same trend can be
observed for setpoint 46 in Fig. 23, but the underestimation of the maximum noise levels is about 2 dB.
Further work is required in order to develop a procedure that allows to extend the FW-H integration to the
surface cup.
55
60
65
70
75
80
85
90
95
100 1000 10000 100000
S
P
L
[
d
B
]
Frequency [Hz]
Measurements - 135 deg
LBM simulation - 135 deg
Measurements - 90 deg
LBM simulation - 90 deg
92
94
96
98
100
102
104
106
40 60 80 100 120 140 160 180
S
P
L
[
d
B
]
Observation angle [deg]
Measurements
LBM simulation
Figure 22. SMC000 simulation: setpoint 07.
37
One-third octave band SPL (left) and OASPL directivity (right).
IV. Prediction of turbofan engine noise
The previous section focused on the fundamental validation of the major turbofan components responsible
for noise generation. In this section the simulation results of a complete turbofan demonstrator geometry
with bypass and a rotor-stator fan/OGV conguration are discussed. These results are to be understood
as a capability demonstration to address a comprehensive turbofan noise problem, taking simultaneously in
one single simulation all the noise generation mechanisms into account that were validated separately in the
previous sections. The simulation was performed by using a beta release of PowerFLOW 5.1, which provides
the high-speed non-isothermal functionality.
17 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
50
55
60
65
70
75
80
85
90
95
100 1000 10000 100000
S
P
L
[
d
B
]
Frequency [Hz]
Measurements - 135 deg
LBM simulation - 135 deg
Measurements - 90 deg
LBM simulation - 90 deg
88
90
92
94
96
98
100
102
104
40 60 80 100 120 140 160 180
S
P
L
[
d
B
]
Observation angle [deg]
Measurements
LBM simulation
Figure 23. SMC000 simulation: setpoint 46.
37
One-third octave band SPL (left) and OASPL directivity (right).
IV.A. Geometry and ow conditions
The geometry shown in Fig. 24 was designed by considering a generic shape for the nacelle with a realistic full
scale outer diameter of about 3.5 m and a length of about 5.2 m. The core nozzle and the bypass diameters
were chosen to be about 1.5 m and 2.9 m, respectively. The fan and stator geometries are taken from the
ANCF simulations described previously in subsection III.C with a geometric scale factor of 2.3 to match the
overall geometrical size of the chosen turbofan.
Figure 24. Geometry of the complete turbofan demonstrator.
The simulations were conducted at a free-stream Mach number M
V
n
=0), but when the surface is vibrating the two expression are substantially dierent, and so we have to
try to identify the most appropriate denition to be used. In order to do that, it is useful to focus on the
expected properties for the impedance. It is clear that one important property would be that the impedance
27 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
of a given surface does not change with the surface motion. Let us look at Z
a
and Z
v
for an untreated
surface for which we expect to have Z . For an untreated vibrating surface we have u
r
=0 and so we
get Z
v
, but for Z
a
instead we get Z
a
and in addition Z
a
is a function of the surface velocity.
The denition in terms of the relative uid/structural velocity seems therefore preferable for our scopes.
This is further reinforced by considering other aspects, such as the physical mechanism of damping through
the liner orices that is driven by the relative velocity, and also the classical coupling of dynamical systems
where the relative velocity is used.
70
It can be therefore concluded that the more convenient denition for
the surface impedance is based on the relative velocity between uid and surface and we will therefore adopt
this denition in the following.
Expression of the relative velocity normal to a vibrating surface
In this section the expression of the relative velocity u
r
between uid and a vibrating surface is derived. The
expressions is valid for a generic material surface, both solid or uid (e.g. a vortex sheet).
Let us consider a generic curved surface dened by the equation f(x, t) =0. The unit normal vector to
the surface is given by:
n = f/|f|, (8)
while the surface velocity V is such that df/ dt =0, i.e.:
f
t
+V f =
f
t
+|f| V n = 0, (9)
from which the normal surface velocity V
n
can be derived, i.e.:
V
n
=
1
|f|
f
t
. (10)
Following Myers,
42
the surface f =0 can be modeled as a time-averaged shape (x)=0 with superimposed
the unsteady perturbationg(x, t), i.e. f =g, where the function g is of the same order as and 1.
Hence the moving surface is dened by the equation (x) = g(x, t). Considering the Fourier component
with angular frequency (+i t convention) of the surface motion and using Eq. 10, we obtain:
V
n
=
i g
||
+ O
_
2
_
. (11)
It is also useful to derive an expression for the unsteady unit normal vector, which reads:
n(x, t) =
g
||
+ O
_
2
_
=
||
g e
i t
||
+ O
_
2
_
; (12)
hence, making use of Eq. 11 and introducing the unit normal vector to the time-averaged surface n
0
, we can
write:
n(x, t) = n
0
(x)
V
n0
(x)
i
e
i t
+ O
_
2
_
on = g. (13)
Let us now suppose that the unsteady velocity eld can be decomposed into a time-averaged eld U(x)
and a superimposed acoustic perturbation u(x, t), such that |u| |U|. Then, let us introduce the relative
uid/surface velocity projected on the normal direction and dened on the deformed surface = g, which
reads:
u
r
(x, t) = [u(x, t) +U(x) V(x, t)] n(x, t)
=
_
u(x) e
i t
+U(x)
V(x) e
i t
_
_
n
0
(x)
V
n0
(x)
i
e
i t
+ O
_
2
_
_
(14)
= u(x) n
0
(x) e
i t
+U(x) n
0
(x) U(x)
V
n0
(x)
i
e
i t
V(x) n
0
(x) e
i t
+ O
_
2
_
.
By applying the perturbation analysis adopted by Myers,
42
the relative velocity u
r
on the moving surface
can be computed through a Taylor expansion applied to the time-averaged surface, i.e.:
u
r
(x, t) |
=g
= u
r
(x, t) |
=0
+ g
u
r
(x, t)
|
=0
+ O
_
2
_
. (15)
28 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
Hence, making use of Eq. 14, we can write:
u
r
(x, t) |
=0
= u(x) n
0
(x) e
i t
+U
0
(x) n
0
(x) U
0
(x)
V
n0
(x)
i
e
i t
V(x) n
0
(x) e
i t
+O
_
2
_
, (16)
and:
g
u
r
|
=0
=
g
||
u
r
n
|
=0
=
g
||
(U
0
n
0
) n
0
+ O
_
2
_
(17)
=
V
n0
i
(U
0
n
0
) n
0
e
i t
+ O
_
2
_
,
where use of Eq. 11 has been made, only the zero-th order term in Eq. 16 has been retained (u
r
(x, t) |
=0
=
U
0
(x) n
0
(x) + O()), and the hypothesis of smoothly curved surface has been introduced, i.e.:
u
r
= [U
0
(x) n
0
(x)] + O() [U
0
(x)] n
0
(x) + O() . (18)
Finally, substituting Eqs. 16 and 17 into Eq. 15, and making use of the slip condition U
0
n
0
=0, we obtain
the expression of the harmonic component of the relative velocity on the deformed surface:
u
r
= u
n0
U
0
V
n0
i
+
V
n0
i
(U
0
n
0
) n
0
V
n0
+ O
_
2
_
(19)
Impedance boundary condition for a vibrating surface
In this subsection, a GIC for a vibrating surface is derived, with reference to the conceptual scheme depicted
in Fig. 36. The eect of the boundary layer is modeled through a vortex sheet that supports an idealized
discontinuous time-averaged ow that below the vortex sheet is forced to be at rest (U=0). The liner and
vortex-sheet normal velocities are indicated as
V
n
and
V
I
n
, respectively.
l
i
n
e
r
s
u
r
f
a
c
e
v
o
r
t
e
x
-
s
h
e
e
t V
I
n
V
n
+
-
U=0
U 0
a(x)=0
a
I
(x)=0
n
I
0
n
0
Figure 36. Conceptual scheme for a moving surface in the presence of a discontinuous time-averaged velocity eld U
and a uctuating vortex sheet.
The derivation proceeds with the denition of the relative normal velocity in the dierent layers of the
domain indicated in Fig. 36. Immediately above the liner surface, Eq. 19 yields:
u
w
r
= u
w
n0
V
n0
. (20)
Hence, making use of the surface impedance denition, as given in Eq. 7, we obtain:
p
w
Z
= u
n0
+
V
n0
. (21)
Immediately below and above the vortex sheet, Eq. 19 yields:
u
r
= u
n0
V
I
n0
(22)
u
+
r
= u
+
n0
V
I
n0
U
0
V
I
n0
i
+
V
I
n0
i
_
U
0
n
I
0
_
n
I
0
, (23)
where U
0
is the time-averaged velocity immediately above the vortex sheet. Since the relative velocity u
r
is
continuous across the vortex sheet, by subtraction of Eqs. 22 and 23 we obtain:
u
+
n0
= u
n0
+U
0
V
I
n0
i
V
I
n0
i
_
U
0
n
I
0
_
n
I
0
. (24)
29 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1
Finally, making use of the hypothesis of continuity of pressure across the vortex sheet ( p
w
= p
= p
+
), and
by setting, for simplicity of notation, p
+
p and u
+
n0
u
n0
, the following condition relating the inviscid
acoustic normal velocity, the acoustic wall pressure, the liner normal velocity and the vortex-sheet normal
velocity can be derived:
u
n0
=
p
Z
+
V
n0
+U
0
V
I
n0
i
V
I
n0
i
_
U
0
n
I
0
_
n
I
0
. (25)
This equation models the eect on the acoustic propagation due to an impedance vibrating surface in the
presence of ow and a vortex sheet.
Eq. 25 cannot be used unless the normal velocity of the vortex sheet is not dened. Two dierent scenarios
can be dened. The rst one is based on the following hypotheses:
the unsteady ow displacement across the vortex sheet due to the viscous interaction between the
acoustic uctuation and the liner surface does not aect the vortex-sheet velocity or, equivalently, the
acoustic boundary layer is much thinner than the steady boundary layer, i.e.
a
_
/f ;
the vibration of the liner surface does not aect the vortex-sheet velocity or, equivalently, the normal
liner displacement is much smaller than the steady boundary layer, i.e. V
n
/f .
As a consequence, in the rst scenario, the vortex sheet is driven by the impinging acoustic wave, i.e.
V
I
n
= u
n0
, hence from Eq. 21 we have
V
I
n
= p/Z +
V
n0
, and the particle displacement is continuous across
the vortex sheet.
The second scenario results from the failure of one of the two hypotheses above: if the acoustic boundary
layer is larger than the steady boundary layer, then the acoustic velocity u
n
is continuous across the vortex
sheet or, equivalently, the vortex sheet has a vanishing eect; if the normal liner displacement is larger than
the steady boundary layer, then the inviscid velocity uctuation u
+
n
is directly aected by the liner velocity.
In both cases, the eect is mathematically equivalent to assume that the vortex sheet is rigidly driven by
the liner surface, i.e.
V
I
n
=
V
n
.
The boundary conditions associated with the depicted scenarios are:
u
n0
=
p
Z
+
V
n0
U
0
_
p/Z
V
n0
_
i
+
p/Z
V
n0
i
(U
0
n
0
) n
0
, (26)
u
n0
=
p
Z
+
V
n0
+U
0
V
n0
i
V
n0
i
(U
0
n
0
) n
0
. (27)
Since the above expressions represent two opposite conditions, it comes natural to dene a unique condition
by introducing a blending factor , thus writing:
u
n0
=
p
Z
+
V
n0
U
0
_
(1 ) p/Z
V
n0
_
i
+
(1 ) p/Z
V
n0
i
(U
0
n
0
) n
0
. (28)
The condition = 0 is equivalent to Eq. 26, whereas the condition = 1 is equivalent to Eq. 27. It is
worthwhile to notice that Eq. 27 can be obtained directly by inserting the normal relative velocity, as
dened in Eq. 23, into the impedance denition given in Eq. 7, and by letting the vortex sheet coincide with
the liner surface. In other words, the condition =1 is equivalent to neglect the inuence of the vortex sheet
in the present ow idealization. In the case of a non vibrating surface, Eq. 28 reduces to:
u
n
=
p
Z
U
[(1 ) p/Z]
i
+
(1 ) p/Z
i
(U n) n. (29)
30 of 30
American Institute of Aeronautics and Astronautics
D
o
w
n
l
o
a
d
e
d
b
y
E
h
a
b
F
a
r
e
s
o
n
J
u
n
e
2
4
,
2
0
1
4
|
h
t
t
p
:
/
/
a
r
c
.
a
i
a
a
.
o
r
g
|
D
O
I
:
1
0
.
2
5
1
4
/
6
.
2
0
1
4
-
3
1
0
1