Professional Documents
Culture Documents
2 Numerical Results and Experimental Validation
2 Numerical Results and Experimental Validation
2 Numerical Results and Experimental Validation
Journal of Fluids Engineering Copyright © 2008 by ASME SEPTEMBER 2008, Vol. 130 / 091102-1
Numerical Methodology high Reynolds number version is obtained by neglecting all the
terms containing the kinematic viscosity. In the proximity of solid
Unsteady Viscous Flow Computation. A full three- walls, viscous effects become important and this assumption no
dimensional simulation of the unsteady flow in the centrifugal fan longer holds. Several modifications have been proposed: In the
described above was carried out. The calculations have been per- two-layer formulation 关14兴, a simpler model is used close to the
formed with a commercial software package, FLUENT®. This code wall 共usually a one-equation model兲 and then the eddy viscosity is
uses the finite volume method and the Navier–Stokes equations patched at a certain distance from the wall; FLUENT® offers this
are solved on an unstructured grid. The unsteady flow is solved option. More precisely, the additional transport equations in direc-
applying a sliding mesh technique, which has been successfully tion xi at time t that are solved for k and are given by
冋冉 冊册
applied to turbomachinery flows 关11,12兴.
The basic equations describing the flow in FLUENT® are the t k
共k兲 + 共uik兲 = + Gk − 共3兲
conservation of mass and the conservation of momentum. In the t xi xi k xi
冋冉 冊册
general form, the conservation of mass in direction xi, i = 1 , 2 , 3
共x1 = x, x2 = y and x3 = z兲 at time t, is given by t 2
共兲 + 共ui兲 = + C1 Gk − C2 共4兲
t xi xi xi k k
+ 共 u i兲 = 0 共1兲
t xi Here, and ui denote the density and the velocity in direction i,
where is the density and ui is the velocity in direction i. The respectively. Furthermore, the turbulent viscosity is given by
conservation of momentum is described by k2
t = C 共5兲
p ij
共 u i兲 + 共 u iu j 兲 = − + 共2兲
t xi xi x j and
in which p denotes the pressure and ij is the stress tensor.
Turbulence is simulated with the standard k- model. The stan-
dard k- model is an eddy-viscosity model consisting of two equa-
Gk = t 冉 + 冊
u j ui u j
xi x j xi
共6兲
tions for the turbulent kinetic energy k and its dissipation rate . represents the rate of production of the turbulent kinetic energy. In
The k- model was introduced by Launder and Spalding 关13兴. The these equations, the coefficients C1, C2, C, k, and are pa-
rameters of the standard k- model, which have the following
empirically derived values: C1 = 1.44, C2 = 1.92, C = 0.09, k
Table 1 Impeller dimensions
= 1.0, and = 1.3.
The time-dependent term of all the equations is discretized with
Outlet diameter 共mm兲 400
a second order, implicit scheme. Second order, upwind discretiza-
Inlet diameter 共mm兲 280
Outlet width 共mm兲 130 tion has been used for convection terms and central difference
Impeller-tongue distance 共mm兲 50 schemes for diffusion terms.
Impeller-tongue distance 12.5% The momentum equations and the continuity equation are
共% of outlet diameter兲 solved sequentially. Once the components of velocity have been
Fig. 6 Mesh details around the radial gap between the impeller front
shroud and the casing
unstructured mesh; some details of the geometric features of the values correspond to zones into the blade channels located near
model are shown in Fig. 5 and the mesh used in the modeled gap the impeller shroud whereas the minimum ones appear in the
is shown in Fig. 6. The minimum cell volume is 2.21⫻ 10−11 m3 blade wakes and close to the back impeller plate. The case of the
and the maximum cell volume is 6.47⫻ 10−5 m3. radial component is different: Zones with very low values 共even
The modeled boundary conditions are those considered with negatives, with a recirculation pattern兲 are present near the impel-
more physical meaning for turbomachinery flow simulations, that ler shroud, whereas the maximum values are concentrated near the
is, total pressure at the domain inlet and a pressure drop propor- hub. This feature indicates that the main fraction of the flow rate
tional to the kinetic energy at the domain outlet. The flow rate is passes through the rear middle part of the impeller.
changed by modifying the constant for that pressure drop at the The numerical model described above has been employed to
outlet condition, which simulates the closure of a valve. calculate the time-dependent pressure both in the impeller and in
The walls of the model are stationary with respect to their re- the volute. In this way, the pressure fluctuations in some locations
spective frame of reference, and the nonslip condition is applied. over the volute wall have been obtained. The measurement posi-
The code was run in a cluster of 8 Pentium 4 共2.4 GHz兲 nodes. tions shown in Fig. 10 and detailed in Table 2 have been selected
The time step used in the unsteady calculation has been set to in order to make comparisons between the numerical and experi-
1.34⫻ 10−4 s seconds in order to get enough time resolution for mental results. The z-coordinate has been normalized by the vo-
the dynamic analysis. The impeller grid movement is related to lute width B 共248 mm兲. The impeller shroud corresponds to z / B
this time step and the rotational speed imposed 共 = 157 rad s−1兲, = 0.54, while z / B = 0 is the volute rear casing and z / B = 1 is the
so a complete revolution is performed each in 300 steps 共i.e., one volute front casing. Figures 11 and 12 show the evolution of pres-
blade passage each in 30 time steps兲. sure fluctuations with time obtained both by 3D-numerical model
The number of iterations has been adjusted to reduce the re- and experimentally, for two different flow rates: the BEP and
sidual below an acceptable value in each time step. In particular, 1.35⫻ BEP. The experimental pressure fluctuations were obtained
the ratio between the sum of the residuals and the sum of the with B&K 4138 1 / 8 in. microphones flush mounted on the volute
fluxes for a given variable in all the cells is reduced to the value of surface. The uncertainty of these two types of microphones has
10−5 共five orders of magnitude兲. Initializing the unsteady calcula- been established by the manufacturer in ⫾0.2 dB, with a confi-
tion with the steady solution, over 17 impeller revolutions 共ap- dence level of 95%. As these microphones are only able to mea-
proximately 5000 time steps兲 are necessary to achieve the conver- sure pressure fluctuations, only the fluctuating part of the signals
gence to the periodic unsteady solution. is compared.
In Fig. 11, the results obtained at Position P02 共angular position
Flow Field Results. The method described above has been em-
ployed to make a comparison for both the numerical and experi-
mental performance curves for the tested fan. The numerical data
are obtained after averaging the values of the unsteady calcula-
tion. In Fig. 7, the numerical and experimental performance
curves for the tested fan are compared. The best efficiency point
共BEP兲 corresponds to a flow rate Q = 0.92 m3 / s 共 = 0.093兲, with a
total pressure rise PT = 500 Pa 共 = 0.105兲.
The experimental and 3D-numerical simulated curves agree for
flow rates equal and higher than the BEP. At partial load, the
matching between 3D-numerical and experimental results is not
so high, probably due to the presence of flow separation in the
blade channels, which has not been correctly captured by the nu-
merical procedure.
The three-dimensional effects of the flow are illustrated in Figs.
8 and 9. These figures show the 3D-numerical results of the rela-
tive tangential component and of the radial component of velocity
over a cylindrical surface around the impeller outlet, with the fan
operating at the BEP. Both components of velocity exhibit impor-
tant gradients in the axial direction. In the tangential component, Fig. 9 Contours of the radial component of velocity at the im-
where the negative values are clockwise, the maximum absolute peller outlet
at 2 deg from the tongue, z / B = 0.30兲 have been represented. The amplitudes at the BPF in z / B = 0.40 are similar and slightly dif-
passing of the ten blades in front of the selected position is clearly ferent in z / B = 0.15. Second, great peak and broadband amplitudes
observed. The amplitude of the pressure fluctuation increases with at low frequencies appear in the experimental spectra. A peak at
the flow rate in this case. The numerical code has reproduced in a 25 Hz stands out, corresponding to the impeller rotational fre-
reasonable way both the order of magnitude and the temporal quency. Also, some peaks at 275 Hz, 300 Hz, and 325 Hz appear
pattern of the pressure fluctuations found experimentally. besides the BPF at 250 Hz with comparable amplitudes, suggest-
In Fig. 12, the results obtained at Position P10 共180 deg from ing the existence of mechanical sources of noise. In order to
the tongue, z / B = 0.30兲 have been represented. The amplitude of clarify the origin of the peaks observed at 275 Hz, 300 Hz, and
the pressure fluctuation at this point diminishes strongly with re- 325 Hz in the experimental spectra, some vibration signals at the
spect to the precedent case, shown in Fig. 11. The passing of the volute front casing were obtained with a 4384 B&K piezoelectric
blades is still clearly observed in the numerical results and the accelerometer, connected to a 2635 B&K amplifier. The ampli-
amplitude of the pressure fluctuations is similar to the experimen- tudes at these frequencies did not vary with the flow rate, thus
tal ones. However, the experimental signals show other sources of indicating its mechanical origin. Moreover, an impact test demon-
pressure fluctuation besides the blade passage, which distorts the
strated that the vibration signal at 300 Hz was due to a casing
clear sinusoidal pattern shown at the tongue. The origin of these
resonance caused by the excitation of a natural frequency.
distortions will be discussed later on.
In Position P10 共180 deg from the tongue兲, the amplitude of the
In Figs. 13–15, the power spectra of pressure fluctuations at
pressure fluctuations is lower than in the previous positions men-
Points P02 共2 deg from the tongue兲, P06 共60 deg from the
tioned 共Fig. 15兲, as a result of the increase in the radial distance
tongue兲, and P10 共180 deg from the tongue兲 have been
between the impeller and the volute wall. The amplitudes at the
represented.
The peak corresponding to the BPF exhibits high amplitude in BPF are quite similar in the experimental and the numerical spec-
Position P02 near the volute tongue, both in the numerical and the tra. On the other hand, in the experimental spectra important
experimental signals 共Fig. 13兲 basically due to the interaction be- broadband levels at low frequencies appear, especially in the axial
tween the flow leaving the impeller and the tongue. The numerical position z / B = 0.15.
and experimental amplitudes coincide in the axial position z / B In Fig. 16, the amplitudes of volute pressure fluctuations at the
= 0.40, while in the position z / B = 0.15 they are slightly different. BPF have been represented, both 3D numerical and experimental
This disagreement was expected because the position z / B = 0.15 is with the fan operating at the BEP 共 = 0.093兲. In this case 共and for
very close to the impeller hub and it is not easy to simulate pre- other flow rates tested 关12兴兲, the maximum values appear concen-
cisely the small axial gap between the impeller and the volute rear trated in a small zone very close to the volute tongue, as it was
casing. expected. These pressure fluctuations are generated by the inter-
In Position P06 共60 deg from the tongue兲, the situation is quite action between the unsteady flow leaving the impeller and the
different 共Fig. 14兲. First of all, the amplitudes corresponding to the fixed volute tongue. In the rest of the volute, noticeable ampli-
BPF have strongly diminished with respect to the previous case tudes are also present due to the jet-wake pattern associated with
共Fig. 13兲, although they contribute largely in the spectra, both the continuous blade rotation around the volute.
numerically and experimentally. In this position, the interaction The 3D-numerical model can reproduce in a reasonable way the
between the impeller and the volute tongue does not appear, be- trend and the order of magnitude of the pressure fluctuations ob-
sides the radial distance from the impeller to the volute is greater tained experimentally. The agreement between experimental and
than in the previous case. These two reasons explain the great numerical results is especially good in the axial position z / B
reduction in the amplitude of the pressure fluctuations with re- = 0.40. As the small axial gap between the impeller and the volute
spect to the previous case. Also, the experimental and numerical rear casing was not modeled, the 3D-numerical and experimental
40
20
-20
-40
-60
-80
0 0.01 0.02 0.03 0.04
t [s]
80
P02-1.35xBEP-EXP
Pressure Fluctuation [Pa]
60 P02-1.35xBEP-NUM
40
20
-20
-40
-60
-80
0 0.01 0.02 0.03 0.04
t [s]
Fig. 11 Evolution of volute pressure fluctuations with time at Point P02 „at
the tongue, z / B = 0.30…
results corresponding to low values of z / B are slightly different, in the unsteady flow numerical simulation constitute the basis for
especially near the volute tongue. Some other differences are a second step in which the sound field will be computed by a
probably due to the presence of flow separation in the blade chan- numerical solution of an appropriate system of acoustic equations
nels at partial load, which has not been correctly captured by the based on the acoustic analogy of Lighthill.
numerical procedure.
Another source of discrepancies between the 3D-numerical and Aeroacoustic Noise Prediction. The sound pressure generated
the experimental pressure fluctuations can be taken into account: from the impeller blades and the volute tongue is predicted by the
the feasibility that the microphones placed on the volute wall mea- Ffowcs Williams–Hawkings equation 关3兴. The integration surfaces
sure noise from distant zones of the flow, i.e., pressure fluctuations of the Ffowcs Williams–Hawkings calculation are the ten impeller
of the “acoustic type,” which cannot be calculated in the three- blades and the volute tongue. It is well known that the sound
dimensional simulation. The computation of pressure fluctuations pressure p⬘, i.e., the solution of the inhomogeneous wave equation
of the acoustic type by CFD codes, which solve the unsteady com- following Ffowcs–Williams–Hawkings approach, comes for the
pressible Navier–Stokes equations, exceeds by far the current contribution of three types of sources: the monopole noise related
computational capabilities. In any case, the authors acknowledge to the blade moving volume 共i.e., blade thickness兲, the dipole
that these errors in the volute pressure fluctuations could result in noises related to the forces 共pressures兲 exerted on the surface of
errors in the computed far-field noise. the blades, and the quadrupole noise associated with flow turbu-
Regardless the exposed constraints, which would be possibly lences. In this work, the later contribution is neglected because the
overcome with a greater computational capability, the presented flow Mach number of the fan is low. The present numerical study
results permit to conclude that the 3D-numerical methodology de- does not account for the influence of the casing on the noise, and
veloped is a useful tool for the unsteady simulation of the three- as a result the predicted far-field noise could differ from the ex-
dimensional flow in a centrifugal fan. The application of this perimentally measured noise.
method to other alternative geometries would permit to establish Also, the noise sources are assumed compact, since the com-
design criteria for the improvement of the aerodynamic perfor- pact noise source conditions suggested by Farassat 关16兴 are satis-
mance of these machines. On the other hand, the results obtained fied. The maximum length of the noise source corresponding to a
-2
-4
-6
-8
0 0.01 0.02 0.03 0.04
t [s]
8
P10-1.35xBEP-EXP
Pressure Fluctuation [Pa]
6 P10-1.35xBEP-NUM
-2
-4
-6
-8
0 0.01 0.02 0.03 0.04
t [s]
Fig. 12 Evolution of volute pressure fluctuations with time at Point P10
„180 deg from the tongue, z / B = 0.30…
grid section along the blades and tongue is much smaller than the
minimum distance between the noise source and the observer con-
sidered and the time step that takes for a sound wave to cross that
p⬘fn,i共x,t兲 =
1 1
冋 1
4 ri2 共1 − M r兲2
冉
ri · fi
1 − Mi · Mi
1 − Mr
− fi · Mi 冊册
maximum length is also much smaller than the period of the BPF. 共8b兲
Using the compact noise source formula given by Cho and
冋 冊冊册
Moon 关5兴 and the geometric parameters shown in Fig. 17, contri-
butions of n noise sources are added as the following:
n
p⬘fn,i共x,t兲 =
1 1 1
冉
ri fi
· +
4 ri 共1 − M r兲2 c0 1 − M r c0
· 冉
ri · fi ri Mi
p⬘共x,t兲 = 兺
i=1
关p⬘th,i共x,t兲 + p⬘fn,i共x,t兲 + p⬘ff,i共x,t兲兴 共7兲 共8c兲
p⬘th,i共x,t兲 =
0V 0 1
冋 1 1
4 1 − M r 1 − M r ri兩1 − M r兩
冉 冉 冊冊册
is the speed of sound, V0 is the blade volume, ri = 共x − yi兲 / ri is a
unit vector from the noise source i to the observer, and fi is the
force vector acting onto the fluid. The local source Mach number
共8a兲 vector and the relative Mach number are also defined as
15 15
10 10
5 5
0 0
25 25
Wall Pressure Fluctuation [Pa]
15 15
10 10
5 5
0 0
0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700
Frequency [Hz] Frequency [Hz]
Fig. 13 Power spectra of volute pressure fluctuations in pascals „experimental, upper side; 3D-
numerical simulation, bottom side… at the measurement Point P02 „at 2 deg from the tongue, z / B
= 0.15 and z / B = 0.40…, with the fan operating at the BEP
1 yi ments were made for the following flow rates: 1.70⫻ BEP and
Mi = , M r = ri · Mi 共10兲 Qmax; the latter is the flow rate obtained when neither plate nor
c0
duct is connected at the fan exit. Obviously, this flow rate was not
where yi is a position vector to the noise source i.
reachable when the fan was connected to the normalized test
Aeroacoustic Results. The acoustic pressure p⬘共x , t兲 is calcu- installation.
lated at the observer time 共t兲 using the described aeroacoustic In order to have a more comprehensive outlook of the results,
noise prediction procedure. The sound pressure level 共SPL兲 spec- the algorithm was applied to observer points placed in the follow-
trum is obtained by a fast-Fourier transform 共FFT兲 algorithm. Fig- ing planes: 共x , y , 0兲, 共0 , y , z兲, and 共x , 0 , z兲. For instance, in Fig. 19
ure 18 shows a picture of the studied fan and the Cartesian coor- the total SPL is represented over those planes for the flow rate
dinates used. The origin of this coordinate system is placed on the corresponding to 1.70⫻ BEP. In this figure, it can be seen that the
rotation axis at the impeller hub. During the measurements of the
directivity pattern of the acoustic source is similar to the dipole
SPL around the fan, the exit duct was not used. Thus, a calibrated
plate was used to obtain the desired flow rate. These measure- source. The figure also shows that the data are symmetric in the
5 5
Wall Pressure Fluctuation [Pa]
3 3
2 2
1 1
0 0
5 5
Wall Pressure Fluctuation [Pa]
3 3
2 2
1 1
0 0
0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700
Frequency [Hz] Frequency [Hz]
Fig. 14 Power spectra of volute pressure fluctuations in pascals „experimental, upper side; 3D-
numerical simulation, bottom side… at the measurement Point P06 „at 60 deg from the tongue, z / B
= 0.15 and z / B = 0.40…, with the fan operating at the BEP
3 3
2 2
1 1
0 0
5 5
Wall Pressure Fluctuation [Pa]
3 3
2 2
1 1
0 0
0 100 200 300 400 500 600 700 0 100 200 300 400 500 600 700
Frequency [Hz] Frequency [Hz]
Fig. 15 Power spectra of volute pressure fluctuations in pascals „experimental, upper side; 3D-
numerical simulation, bottom side… at the measurement Point P10 „at 180 deg from the tongue, z / B
= 0.15 and z / B = 0.40…, with the fan operating at the BEP
Fig. 16 Amplitude „pascals… of volute pressure fluctuation at the blade passing frequency, 3D-numerical „white squares…
and experimental „dotted line…, with the fan operating at the BEP
c)
Fig. 19 Predicted SPL „decibels… around the fan: „a… plane
„x , 0 , z…, „b… plane „x , y , 0…, and „c… plane „0 , y , z…; flow rate: BEP
a)
(b)
tions from the fan, the effects of the motor casing and the floor
reflection, and the background noise significantly affect the ex-
perimental results.
Conclusions
In this study, a previously published aeroacoustic prediction
b) methodology 关5兴 has been extended to three-dimensional turbulent
flow in order to predict the noise generated by a centrifugal fan
using the Ffowcs Williams and Hawkings model extension of
Fig. 20 SPL „decibels… at the blade passing frequency around Lighthill’s analogy. The forces exerted by the fan blades on the air
the fan; plane „x , 0 , z…; flow rate: 1.70Ã BEP: „a… predicted and obtained from an unsteady viscous flow computation are used as
„b… measured
input data for the calculation of the acoustic pressure around the
fan.
Both the predicted and the measured SPL at the BPF around the
共x , 0 , z兲 for the flow rate corresponding to 1.70⫻ BEP. In both fan show a dipolar behavior. The experimental results were influ-
lines the results agree in the vicinity of the fan. In further posi- enced by the pavement reflection and the noise background level,
which are not present in the numerical results. These features also
mask the dipolar behavior in the experimental results. It has also
been observed that the strength of the dipole increases when the
flow rate increases. Given the approximations made, the agree-
ment between the predicted and the measured SPL distributions in
the vicinity of the fan is reasonable.
This methodology could be applied in the design stage in order
to optimize the aeroacoustic behavior of centrifugal fans.
Acknowledgment
This work was supported by the Research Projects TRA2007-
62708, TRA2004-04269, and DPI2001-2598 共Ministerio de Edu-
cación y Ciencia, España兲.
Nomenclature
B ⫽ volute width
C1 , C2 , C ⫽ parameters of the standard k- model
D ⫽ diameter
c0 ⫽ speed of sound
Fig. 21 Measured SPL „decibels… at the blade passing fre- f ⫽ frequency
quency around the fan; plane „x , 0 , z…; flow rate: Qmax fi ⫽ force vector acting onto the fluid