Вы находитесь на странице: 1из 8

9th International Exergy, Energy and Environment Symposium (IEEES-9), May 14-17, 2017, Split, Croatia

Numerical analysis of energy efficiency performance and noise emissions of building


roof fan

1*Ivo Marinić-Kragić, 2Zoran Milas, 3Damir Vučina


1 University of Split, FESB, Department of Computer Application in Engineering, Ruđera Boškovića 32, Split, 21000, Croatia
2 University of Split, FESB, Department of Fluid Mechanics, Ruđera Boškovića 32, Split, 21000, Croatia
3 University of Split, FESB, Department of Computer Application in Engineering, Ruđera Boškovića 32, Split, 21000, Croatia

* E-mail: imarinic@fesb.hr

Abstract

Centrifugal roof fans are traditionally considered as ‘small’ machines and as such did not attract much attention.
They can be categorized as low power fans as their installed power is usually bellow 2kW. Nevertheless, they make
approximately 30% of fans used in non-residential ventilation, which when considered on mass-scale makes them
large energy consumers. Since they have relatively low efficiency (ranging between 0.3-0.5), considerable space
for improvement exists. This resulted in various studies which include optimization coupled with CFD flow models.
The most often used criterion was fan efficiency at single flow regime. Recent studies have shown that this is not a
good approach since there can be multiple different solutions with almost the same single-regime efficiency.
A better approach is to conduct optimization for multi-regime operating conditions or to use multi-objective
optimization with the minimization of noise emission used as the second criterion. The predictions of numerical flow
model must be robustly accurate (for various in impeller shapes and flow regimes).
In this paper, enhanced CFD models for the prediction of fan energy performance and noise emissions are
developed. First, the RANS based models with the frozen-rotor approach and k- / k- turbulence models were
reviewed. The application of more advanced models of frozen-rotor RANS, URANS and LES models is investigated.
The CFD predictions are compared to our experimental data from a previous study. The results show that
improved prediction of energy efficiency is obtained with LES based CFD models for the off-design flow regimes.
Meanwhile, the RANS and URANS models can also provide good results when using first-order upwind
discretization scheme. The results show that using high-order discretization schemes deteriorates the CFD
prediction of the fan performance.
The fan noise emission was predicted using LES data and this allowed for the investigation of wide-spectra noise
in comparison to simple models used in previous papers. The models developed in this paper can be used as the
basis for shape optimization studies which require multi-fidelity simulations.

Acknowledgment. This work has been supported by the Croatian Science Foundation under the project IP-2014-
09-6130.

Keywords: Centrifugal fan, noise, energy efficiency, CFD, multi-criteria optimization

I. Introduction Experimental and numerical study of a fan similar in


size (centrifugal fan for a refrigerator) was conducted
Until recently, centrifugal roof fans did not attract much by Heo et al. (2011). They adopted an URANS CFD
attention since they were considered as “small” model for performance analysis. The results from CFD
machines (installed power is usually bellow 2kW). simulation were subsequently used for acoustic noise
Nevertheless, they account for 30% of all non- calculation using Ffowcs Williams-Hawkings
residential ventilation (Radgen et al.). If we consider equations Williams et al. (1969). This noise calculation
them on a large-scale, this makes them large energy model in combination with URANS is also used for
consumers. The centrifugal roof fans have efficiency larger centrifugal fans for example in Zhang et al.
between 30% and 50%, which means that large space (2016) for a fan which has 510 mm outer diameter.
for improvement exists. Some authors have also used a similar but
These facts attracted researchers mostly from areas computationally more efficient noise estimation model
of fluid mechanics and numerical optimization. Recent developed by Lowson (1970). Example is the study by
numerical studies consider various centrifugal fan Buccieri and Richards (2016). Additionally, for flow
sizes (and designs), ranging from very small (outer simulation they used SAS turbulence model (Menter
diameter smaller than 100 mm) to large (outer and Egorov 2005). This is a hybrid LES-RANS
diameter up to 500 mm). Numerical simulation of a turbulence model which uses the RANS k-SST
very small centrifugal fan (used for CPU cooling) was model in the areas near the solid walls and the LES far
conducted in Tsai and Wu (2007). This study includes away.
numerical simulation of performance and noise Recent studies also include numerical optimization
emissions and a comparison with experimental results. of energy conversion efficiency and acoustic

-1-
9th International Exergy, Energy and Environment Symposium (IEEES-9), May 14-17, 2017, Split, Croatia

emissions (Marinić-Kragić et al. 2016; Milas et al. M i  M fd (3)


2014). It was shown that single objective (fan
where M i is the torque due to the aerodynamic
efficiency) optimization case has a very low sensitivity
in the area near the optimal solution. This means that forces acting on the inside impeller surface and M fd
multiple different solutions exist with almost the same the disc friction torque. In reality, additional power is
efficiency. The first approach in solving the low required due to the volumetric and mechanical losses
sensitivity problem was to define the efficiency as an (which have only a minor impact on the overall
average over multiple operating regime. The second efficiency). To calculate the torque, pressure and
approach implemented multi-objective optimization shear stress are integrated along the fan impeller walls
with noise emission as a second criterion. according to:
The use of an appropriate numerical model is    
important in all optimization scenarios. In this paper, Mi  M fd   ( r  n ) y pdA  ( r   w ) y dA (4)
we tested various CFD models regarding the A
prediction of fan pressure, efficiency and noise
emission. Steady-state RANS, URANS and LES II. 3. Experimental results
turbulence models were investigated. Frozen rotor In this paper, experimental results (Milas et al. 2014)
approach was used in all cases. were used. The experiments include measurements of
pressure and efficiency as defined by equations (1)
II. Roof Fan Design and Performance and (2) respectively. Table 1 defines the geometric
characteristics of the investigated fan. This geometry
II. 1. Design and Geometry was used in this paper for constructing the numerical
Since roof fans expel the air directly into the ambient, model.
they do not require a scroll housing. Roof fan housing
is usually a simple axisymmetric cap, used merely as Tab. 1: Fan geometric characteristics (Milas et al. 2014).
a weather shield as illustrated in Fig. 1. Electric motor
usually drives the impeller by a direct drive without a Impeller diameter
D (mm) 325
belt or gears. In any case, the far field flow conditions (outlet)
are axisymmetric. Impeller eye Do (mm) 235
External rotor
impeller electric der (mm) 138
(el. motor)
motor
housing Impeller width b (mm) 75
Number of vanes z 14
air Vane outlet angle 𝛽 (O) 45
air
outflow outflow
Vane thickness t (mm) 1,5

building roof The impeller outer diameter and the impeller width are
shown is shown in figure 2. while the impeller eye is
only visible from the bottom view (not shown).
air inflow

Fig. 1: Simplified roof fan schematic.

II. 2. Performance Definition


The most important fan performance parameters are
the total pressure increase and the corresponding fan
D der
efficiency. Since roof fans do not have a pressure duct,
the outlet dynamic pressure is ignored. Hence, the
total pressure increase p t is defined as: b
 1
p t   po ( v o dAo ) / V   ( pi  2 v i )( v i dAi ) / V
2
(1)
Ao Ai
hub
where po is static pressure at outlet and p i is the
static pressures at inlet. The absolute velocities at vanes
  y
outlet and inlet are designated by v o and v i shroud
respectively. Area of fan outlet section is designated air
Ao while the inlet section area is designated by Ai , inlet x

 is the air density, and V the volume flow rate. z

The fan efficiency is defined as: Fig. 2: Impeller geometry.


  Vpt / P (2)
Measurements were conducted for multiple flow rates
where P is the fan power. For this paper, the fan and the results are shown in figure 3. This small roof
power P is calculated by: fan was tested in laboratory conditions. The fan
“housing” was only a simple axisymmetric horizontal

-2-
9th International Exergy, Energy and Environment Symposium (IEEES-9), May 14-17, 2017, Split, Croatia

cap and far field conditions can also be regarded as frame while the rest of the computational domain was
axisymmetric. The peak efficiency (42%) was stationary. Mass flow boundary condition was set at
measured at flow rate of 918 m3/h, with the pressure the pipe inlet section while the pressure outlet (zero
increase 81 Pa. Measurements were conducted up to pressure) was prescribed at the outer atmospheric
flow rate of 1650 m 3/h where efficiency drops to near sub-domain.
zero. Rotational speed was constant in all tests =103
rad/s.
weather
shield outer
atmospheric
impeller fan sub-domain
outlet
inner
atmospheric
fan sub-domain
y inlet

Fig. 3: Experimental fan characteristics from (Milas et al. z x


2014).
Fig. 5: Computational (sub-) domains.
The numerical results obtained in this paper were
compared to these experimental results as shown later III. 2. Domain discretization
in results section. The computational domain is discretized by the finite
volumes using the ANSYS Meshing. Regarding the
III. Numerical Modeling number of elements, two different domain
The numerical modeling of the fan flow was conducted discretization were used. A courser mesh is used for
by using the commercial software ANSYS Fluent. the RANS simulations while finer mesh was used for
LES. Structured mesh with hexahedral elements was
III. 1. Computational domain used in the atmosphere domains. Intake pipe domain
Computational domain is selected such that it was constructed with unstructured swept mesh with
encapsulates only one vane. This is sufficient since the inflation zone near the pipe wall. In the impeller
periodic boundary conditions can by applied. A single domain, tetrahedral elements were used with prism
segment of impeller computational sub-domain is inflation layers near the wall surfaces. In RANS model,
illustrated in figure 4. 17 inflation layers were used. The first layer height was
set to 0.015 mm and growth rate was set to 1.17. For
LES model, inflation zone with 30 layers and growth
rate of 1.035 was used. Figure 6 illustrates the mesh
at half-height of the impeller.

impeller
diameter, D
 y x
y x
fan
outlet

z
z
detail

Fig. 4: Single computational domain (CD) segment a)


with periodic boundary conditions (BC).

In addition to impeller sub-domain, intake pipe and


outflow atmospheric domain has to be modelled. Again, inflation
only one segment is modelled. Intake pipe segment layers
with length twice the impeller diameter (D) was used,
while the diameter of the atmospheric domain was 6
times the impeller diameter. Atmospheric sub-domain
is divided in two domains so that different element
sizes can be used near the fan outlet and far away.
The impeller sub-domain and the inner atmospheric
sub-domain were analyzed in a rotating reference b)

-3-
9th International Exergy, Energy and Environment Symposium (IEEES-9), May 14-17, 2017, Split, Croatia

     

t x j
 
 ui    uj    
x j  x j 
(8)

The presence of fluctiatomg terms in the conservation


equations namely Reynolds stresses and turbulent
scalar flux, means that the equations are not closed.
The closure requires the use of some approximations,
which usually take the form of prescribing the
Reynolds stress tensor and turbulent scalar fluxes in
terms of the time-averaged flow quantities. The effect
c) of turbulence can be represented as an increased
Fig. 6: Mesh at half-height of impeller domain: a) viscosity (which is called turbulence viscosity  t )
RANS model - whole impeller view, b) RANS model - (Ferziger and Peric 2002). This leads to the
vane leading edge and c) LES model – vane leading Boussinesq eddy-viscosity model for the Reynolds
edge.
stress Rij :
While the total number of elements required for RANS   u  u j 2  uk  2
simulation is about 0.5 million, LES (LES-WALE) Rij    uiuj  t  i       k (9)
 x j xi 3 xk ij  3 ij
requires a mesh with about 50 million elements  
(almost 700 million elements without the periodic In (9), k is the turbulence kinetic energy:
boundary conditions). The distribution of elements per 1
sub-domain is shown in table 2. k  uiui (10)
2
Tab. 2: Number of finite volume elements per sub-dp,aom The eddy-viscosity hypothesis is widely used due to
for mesh used with RANS and LES models. easy to implement and its application has shown
reasonably good results for many types of flows.
Number of elements Turbulent viscosity  t and turbulence kinetic energy
Sub-domain RANS model LES model k are determined by the turbulence model.
Inlet pipe 31416 186873 In this paper, the system of time-averaged equations
Impeller 336812 40780027 with time term is also termed as unsteady RANS
(URANS). Specifically, this paper used k-SST
Atmosphere 90600 2213843 (Menter 1993) as RANS turbulence model. In addition
(inner) to RANS turbulence model, this paper uses model
Atmosphere 167859 6322836 LES (large eddy simulation). LES equations are
(outer) derived by averaging (in space and time) only on
scales with a prescribed limit (usually defined indirectly
III. 3. Flow and Acoustics Modeling by the mesh size). The LES equations result in a
The averaged continuity and momentum equations in system similar to RANS equations but with different
the case of incompressible flows (and body forces definition of turbulence viscosity and Reynolds
excluded), can be written in index notation and stresses. The LES-WALE model (Nicoud and Ducros
Cartesian coordinates as: 1999) was selected for LES.
Convergence criteria was set to 10-4 for residuals in all
 ui
0 (5) cases. In unsteady simulations, time-step was 2°/.
xi
The selected method for modeling the noise problem
 ui   p  ij
t

x j
 
 ui u j   uiu j   
xi x j
(6) is the Ffowcs-Williams and Hawkings (FW-H) method
(Williams et al. 1969). The FW-H method uses the
where  , p and ui are the time-averaged flow most general form of the Lighthill acoustic analogy.
This method is appropriate for the prediction of sound
density, pressure and the Cartesian velocity generated by rigid bodies moving at subsonic
components respectively: xi is the Cartesian velocities. FW-H method was implemented using the
ANSYS Fluent. Stationary observer was set at 1m
coordinate; the  ij are the time-averaged viscous distance from the rotational axis. In order to calculate
stress tensor components: the overall pressure caused by all vanes multiple
 u u  observers were to used (one for each vane).
 ij    i  j  (7)
 x j x j  IV. Results and discussions
 
The generalized transport equation for the time-
IV. 1. Steady state model
averaged of a scalar quantity  (velocity u for First test case was steady state RANS simulation with
momentum conservation and 1 for mass conservation) SST turbulence model. First-order upwind
can be written: discretization scheme was used. Figure 7 shows the
comparison of the results obtained by numerical

-4-
9th International Exergy, Energy and Environment Symposium (IEEES-9), May 14-17, 2017, Split, Croatia

simulation and experiment. A very good agreement is flow rate 918 m3/h is illustrated in figure 9a. This is the
evident especially near the maximum efficiency point acoustic pressure that would appear at the observer
(flow rate 900 m3/h). location if there was only a single vane. Since multiple
vanes exist, the FW-H equation has to be integrated
over the whole fan. Since periodic conditions are
assumed, this does not require additional CFD
simulation. To obtain the acoustic pressure caused by
the remaining vanes, the single-vane acoustic
pressure result is time-translated (t=2*/14/). By
summing the results over all vanes, the overall
acoustic pressure for the whole fan can be obtained as
shown in figure 9b. By conducting a Fourier
transformation, sound pressure level as a function of
frequency can be calculated as shown in figure 9c.
Fig. 7: Comparison of experimental results (exp) and
steady-state RANS numerical (CFD) simulation.

The following figure (Fig. 8) shows the flow field


velocities at two sub-domains for the flow rate 1650
m3/h. Apparently, no, or very little flow separation
appears at the vane surface. Looking at the outlet flow,
it can be observed that the flow direction is slightly
axial.
Vstf,t [m/s] Vstf [m/s]
13 20
11.7 18
10.4 16
9.1 14
7.8 fan 12
a)
6.5 outlet 10
5.2 8
3.9 6
2.6 4
1.3 2
0 0
a)
Vrel [m/s]
Vrel,t [m/s]
fan 20
20 18
18 outlet
vane 16
16 14
14 12
12 10
10 8
8 6
6 4 b)
4 2
2 0
0

b)
Fig. 8: Flow field velocities for RANS simulation at
flow rate 1650 m3/h (Vstf –velocity in stationary frame,
Vstf,t – velocity in stationary frame tangential to current
plane, Vrel – relative velocity, Vrel,t – relative velocity
tangential to current plane): a) x-y plane section at
z=0 and b) x-z plane section at half vane height.

The data from the CFD simulation can now be used for
calculation of the acoustic pressure at the observer c)
point. This requires “virtual” rotation of the vane since Fig. 9: Acoustic noise for observer at 1m distance
only steady-state solution exists. The observer point using RANS model at flow rate 918 m 3/h: a) acoustic
was set at 1m distance from the rotation axis. The pressure caused by single vane b) overall fan
result of the acoustic pressure for a single vane at the acoustic pressure c) sound pressure level as a
function of frequency.

-5-
9th International Exergy, Energy and Environment Symposium (IEEES-9), May 14-17, 2017, Split, Croatia

As already mentioned, this solution was obtained by solution is required. Since FW-H equation only uses
using the first-order upwind discretization scheme. surface integration, this requires storage of transient
When the same simulation was conducted using solution only at the selected source surfaces which is
higher order discretization schemes, residuals did not a considerable memory saving in comparison to
converge to the defined criterion 10-4. Upwind scheme storing full 3D transient solution. The observer point
is known to introduce lots of numerical dissipation was again set at 1m distance from the rotation axis.
(stability). This means that physically unsteady The result of the acoustic pressure for a single vane at
solution can become a steady-state solution. Thus, the the flow rate 918 m 3/h is illustrated in figure 12a. In
next case conducts an unsteady RANS simulation with total, 14 observer points were set at equidistant angles
a higher order discretization scheme. for calculating the acoustic pressure. Assuming
periodic conditions, the data from these 14 points can
IV. 2. URANS model be used to integrate the acoustic pressure for the
Next, transient RANS simulation was conducted with overall fan. Next, sound pressure level as a function of
the same settings as earlier but with 2°/ time-step. frequency is shown in figure 12b. In comparison to
Simulation was conducted for approximately 2 earlier case, URANS results in wider spectrum of
rotations and the results were averaged. The results acoustic noise and much higher sound pressure levels.
are illustrated in figure 10. Compared to the previous Experimental data for the fan noise emissions is not
case, CFD prediction capability for pressure and available but this is clearly a much more realistic result.
efficiency has deteriorated. For example, earlier the
maximum fan flow rate was about 1650 m 3/h while now
it is approximately 10% lower and amounts
approximately 1500 m 3/h. The pressure obtained at
maximum efficiency point has also reduced from about
80 Pa to 65 Pa. Since this was not an expected result,
various test cases with different mesh sizes and
domain shapes were conducted but the results did not
improve.

a)

Fig. 10: Comparison of experimental results (exp)


and averaged transient URANS numerical (CFD)
simulation.
In comparison to earlier case, flow field shows a bit
more separation near the vane leading edge as
illustrated in figure 11.

Vrel,n [m/s] Vrel [m/s]


fan 20
20
18 outlet 18
16 vane 16
14 14
12 12 b)
10 10 Fig. 12: Acoustic noise for observer at 1m distance
8 8 using URANS model at flow rate 918 m 3/h: a)
6 6 acoustic pressure caused by single vane b) overall
4 4 fan sound pressure level as a function of frequency.
2 2
0 0
IV. 3. LES model
Next, the LES WALE turbulence model is used. This
model requires a very fine computational mesh with
Fig. 11: Flow field velocities for URANS simulation at about 50 million elements which is about one hundred
flow rate 1650 m3/h (Vrel – relative velocity, Vrel,t – time more than required for RANS simulation. The
relative velocity tangential to current plane) in x-z same time step 2°/ was used. The pressure and
plane section at half vane height. efficiency results are shown in figure 13. Still, the
In order to evaluate acoustic pressure, now a transient maximum fan flow is approximately 10% lower than

-6-
9th International Exergy, Energy and Environment Symposium (IEEES-9), May 14-17, 2017, Split, Croatia

experimental, but the result at the maximum efficiency visible that up to about 100 Hz, the solution is
point is considerably improved in comparison to approximately the same as was when using URANS.
URANS simulation. While the URANS simulation predicted that peak
acoustic emissions would be generated at 100 Hz, the
LES predicts that the most noise will be generated at
800 Hz frequency. In order to confirm these results,
future work would have to include experimental
acoustic measurements.

Fig. 13: Comparison of experimental results (exp)


and averaged transient LES numerical (CFD)
simulation.
The flow field for the LES simulation at the flow rate
1650 m3/h is shown in figure 14. Now, the flow field
contains lots of turbulent eddies. Some of these eddies
are apparently generated near the leading edge at the
pressure side of the vane. This kind of flow field is
expected to produce more acoustic emissions
especially at higher frequencies. a)
Vstf,n [m/s] Vstf [m/s]
13 20
11.7 18
10.4 16
9.1 14
7.8 12
6.5
fan 10
5.2 outlet 8
3.9 6
2.6 4
1.3 2
0 0

a)
Vrel [m/s]
Vrel,n [m/s]
20 20
18 fan 18
16 16
outlet 14 b)
vane Fig. 15: Acoustic noise for observer at 1m distance
14
12
12 using LES model at flow rate 918 m 3/h: a) acoustic
10
10
8 pressure caused by single vane b) overall fan sound
8 pressure level as a function of frequency.
6 6
4 4
2 2 IV. Conclusion and future work
0 0
This paper has tested RANS, URANS and LES
b) turbulence models for simulating the centrifugal roof
Fig. 14: Flow field velocities for LES simulation at flow fan performance and noise emissions and the results
rate 1650 m3/h (Vstf –velocity in stationary frame, Vstf,t were compared to the available experimental results.
– velocity in stationary frame tangential to current The best performance prediction model is obtained by
plane, Vrel – relative velocity, Vrel,t – relative velocity using the steady-state RANS model with the first order
tangential to current plane): a) x-y plane section at upwind discretization. The first order scheme
z=0 and b) x-z plane section at half vane height. introduces lots of numerical dissipation thus higher
order schemes are usually recommended. But, when
As was the case in URANS simulation, in order to using higher order schemes, it was shown that the
calculate the acoustic pressure by FW-H equation, the solution does not converge to a steady state. Instead
transient solution at the selected source surfaces of steady state, high-order scheme can be used solved
needs to be stored. Figure 15a shows the acoustic in a transient simulation (URANS). The results
noise for observer at 1m distance using the LES model. obtained by the URANS do not achieve as good
Flow rate 918 m 3/h was used. Figure 15b shows the performance prediction as the RANS with the first-
sound pressure level as a function of frequency. It is order discretization scheme. In comparison to the

-7-
9th International Exergy, Energy and Environment Symposium (IEEES-9), May 14-17, 2017, Split, Croatia

URANS it was shown that the LES achieves improved


performance predictions.
The fan noise emission were also predicted using
RANS, URANS and LES models. It was shown that
steady-state RANS model results in unrealistically
small sound pressure levels. Meanwhile the URANS
and the LES models predict more noise emissions
especially at higher frequencies.
This paper included only the frozen rotor approach so
future work will investigate also the sliding mesh
approach. Since LES is very computationally
demanding, hybrid approaches such as DES will be
tested. These hybrid turbulence models might achieve
as good results but with considerably reduced
computational requirements.

Acknowledgements

This work has been supported by the Croatian Science


Foundation under the project IP-2014-09-6130.

References

Buccieri BM., Richards CM. Application of the aeroacoustic


analogy to a shrouded, subsonic, radial fan. J. Sound Vib.
385:125–37 (2016).
Ferziger JH., Peric M. Computational Methods for Fluid
Dynamics. Springer. Springer; (2002.).
Heo S., Cheong C., Kim T-H. Development of low-noise
centrifugal fans for a refrigerator using inclined S-shaped
trailing edge. Int. J. Refrig. 34(8):2076–91 (2011).
Lowson M V. Theoretical Analysis of Compressor Noise. J.
Acoust. Soc. Am. 47(1B):371–85 (1970).
Marinić-Kragić I., Vučina D., Milas Z. 3D shape optimization
of fan vanes for multiple operating regimes subject to
efficiency and noise-related excellence criteria and
constraints. Eng. Appl. Comput. Fluid Mech. 10(1):210–28
(2016).
Menter F. Zonal Two Equation k-w Turbulence Models For
Aerodynamic Flows. 23rd Fluid Dyn. Plasmadynamics,
Lasers Conf. Reston, Virigina: American Institute of
Aeronautics and Astronautics; (1993.).
Menter F., Egorov Y. A Scale Adaptive Simulation Model
using Two-Equation Models. 43rd AIAA Aerosp. Sci. Meet.
Exhib. Reston, Virigina: American Institute of Aeronautics
and Astronautics; (2005.).
Milas Z., Vučina D., Marinić-Kragić I. Multi-regime shape
optimization of fan vanes for energy conversion efficiency
using CFD, 3D optical scanning and parameterization. Eng.
Appl. Comput. Fluid Mech. 8(3):407–21 (2014).
Nicoud F., Ducros F. Subgrid-Scale Stress Modelling Based
on the Square of the Velocity Gradient Tensor. Flow, Turbul.
Combust. 62(3):183–200 (1999).
Radgen P., Oberschmidt J., Cory WTW. EuP Lot 11: Fans for
ventilation in non residential buildings Final Report.
Tsai BJ., Wu CL. Investigation of a miniature centrifugal fan.
Appl. Therm. Eng. 27(1):229–39 (2007).
Williams JEF., Hawkings DL., Ffowcs JE., And W.,
Hawkingst DL. Sound Generation by Turbulence and
Surfaces in Arbitrary Motion [ 321 ] SOUND GENERATION
BY TURBULENCE AND SURFACES IN ARBITFRARY
MOTION. Philos. Trans. R. Soc. London. Ser. A, Math. Phys.
Sci. 264(1151):321–42 (1969).
Zhang J., Chu W., Zhang H., Wu Y., Dong X. Numerical and
experimental investigations of the unsteady aerodynamics
and aero-acoustics characteristics of a backward curved
blade centrifugal fan. Appl. Acoust. 110:256–67 (2016).

-8-

Вам также может понравиться