Noise prediction via large eddy simulation: Application to radial fans Esra Sorguvena), Yilmaz Doganb), Faruk Bayraktarc) and Kenan Y. Sanliturkd) (Received: 23 September 2008; Revised: 11 March 2009; Accepted: 11 March 2009)
In this study, aerodynamics and aeroacoustics of two radial fans are investigated by using a hybrid computational aeroacoustics method. Unsteady turbulent flow field of both fans is simulated with large eddy simulation (LES).Acoustic sources are computed based on the pressure fluctuations. Inhomogeneous wave equation, which accounts for the propagation, diffraction and scattering of the acoustic waves, is solved to determine the far field sound pressure level with the boundary element method. Numerical results are validated with experimental data. Sound pressure level distribution in narrow band frequency spectrum and directivity of the acoustic waves are predicted numerically with a high accuracy. Results of the LES provide an insight into the turbulent flow and noise generation mechanisms, which can be utilized to reduce fan noise. © 2009 Institute of Noise Control Engineering. Primary subject classification: 11.4.1; Secondary subject classification: 75
1
INTRODUCTION
Flow induced noise prediction in industrial applications is essential in order to control the noise emission and to comply with the noise regulations and consumer demands. Experimental methods involve drawbacks like time and investment expenses and measurement errors like reflection problems. Thanks to the advancement in computational technology, computational aeroacoustics (CAA) became a valuable method for noise prediction and reduction. Hybrid CAA methods are especially efficient and relatively inexpensive, since they solve for the different scales of aerodynamics and aeroacoustics separately. Studies on aeroacoustics began after the 2nd World War, as the civil aircraft technology has evolved. In his acoustic analogy, Lighthill1,2 derived an inhomogeneous wave equation to describe the jet noise, which arises due to turbulent pressure fluctuations. Curle3 enhanced the analogy by introducing the effect of solid surfaces on sound generation. Later, Ffowcs Williams and Hawkings4 added the effect of moving solid surfaces on sound generation. Most of the modern CAA methods apply the acoustic analogy of Ffowcs a)
b)
c)
d)
Mechanical Engineering Department, Yeditepe University, 34755 Istanbul TURKEY; email:
[email protected]. Vibration and Acoustic Technologies, Arcelik AS 34950 Istanbul TURKEY; email:
[email protected]. Vibration and Acoustic Technologies, Arcelik AS 34950 Istanbul TURKEY; email:
[email protected]. Mechanical Engineering Department, ITU, 34437 Istanbul TURKEY,
[email protected].
Noise Control Eng. J. 57 (3), May-June 2009
Williams and Hawkings (FWH) to predict the noise radiated directly from a source to the observer in absence of any barriers in-between. If there are barriers between the source and the observer, like the casing of a radial fan, then wave equation is to be solved to take propagation, diffraction and scattering of the waves into account. A detailed review on computational aeroacoustics has been provided by Colonius and Lele5. Studies involving analysis, prediction and reduction of fan noise are active research areas because of the widespread use of axial and centrifugal fans in industry. Gutin6, Carolus7, Bommes et al.8 and Neise9–11 have made notable contributions to aerodynamics and acoustics of fans. Numerical methods for the prediction of fan noise usually solve for the flow field first, and then predict the acoustic field based on the flow data. A wide range of models are used for both flow simulations and acoustic computations. Jeon et al.12 calculated the aeroacoustic signal of a centrifugal fan by solving the inviscid, unsteady flow-field via vortex method and applying an acoustic analogy. Maaloum et al.13 coupled a 2D vortex surface analysis with the FWH analogy to predict the acoustic field of an axial fan. Khelladi et al.14 obtained the acoustic sources by solving the Reynolds Averaged Navier-Stokes (RANS) equations and calculated tonal noise of a centrifugal fan via FWH analogy. Nallasamy and Envia15 studied the broadband noise generated by the rotor wake turbulence stator interaction with the help of RANS simulations. Wu and Su16 developed a semi-empirical formula to predict noise of axial fans in free-field. BallesterosTajadura et al.17 studied the acoustic characteristics of a 169
Fig. 1—Geometry of the investigated fans a) Fan I, b) Fan II. centrifugal fan based on a three-dimensional numerical simulation. Özyörük and Alpman18 predicted sound fields of ducted fans based on the solution of the frequency-domain form of the Euler equations, which are linearized about an axisymmetric non-uniform background flow. Merits and limits of these uncoupled aeroacoustic predictions basically depend on the accuracy of the flow simulation. More complex flow simulations result in more accurate acoustic source terms; however, increase the computational cost. Low noise emission has become one of the design criteria for fans. Therefore studies with an emphasis on numerical noise prediction and reduction in early design stages attract attention. Lin and Huang19 designed a small forward curved centrifugal fan for a notebook computer with the emphasis on the blade shape, blade inlet angle and the outlet geometry of the housing. Gérard and Berry20 developed an inverse aeroacoustic model to reconstruct the aerodynamic forces (dipole strength distribution) acting by the fan blades on the fluid. In this paper, a hybrid CAA method is employed to investigate the aerodynamics and aeroacoustics of radial fans. Aeroacoustic computations are performed in two steps: i) computing the unsteady flow field and ii) computing the acoustic pressure fluctuations in the far field. The flow field is solved with large eddy simulation (LES) where the large and energetic scales of turbulence are resolved and the small and dissipative scales are modeled. LES resolves the dominant noise sources without any modeling error and therefore provides a reliable basis for acoustic computation. Acoustic sources are computed based on the turbulent pressure fluctuations. The inhomogeneous wave equation is solved via multi domain boundary element method (BEM) analysis by taking propagation, diffraction and scattering of the acoustic sources into account. Finally, the far field sound pressure level and sound power level are obtained. This methodology is applied to two radial fans. 170
Noise Control Eng. J. 57 (3), May-June 2009
Detailed experimental measurements are performed and compared to numerical results to validate the numerical methodology. Sound pressure level distributions at discrete microphone points are compared in narrow band frequency spectrum. Prediction of the directivity of the acoustic waves is validated by comparing the sound intensity maps on fictitious surfaces surrounding the fans. These comparisons show that the numerical method can predict the directivity and sound pressure level in narrow band spectrum satisfactorily. Computing the aerodynamic and aeroacoustic data of two fans also shows how the CAA methods provide an insight to the turbulent flow and the noise generation mechanisms. By comparing the detailed numerical data of both fans, regions of high noise emission can be identified. Such information can be utilized to decrease the overall sound pressure level of a fan and to design low noise fans.
2
INVESTIGATED FANS
Two Sirocco type radial fan systems consisting of a radial impeller, volute, inlet and outlet pipes are investigated. Both of the investigated radial fan systems are nearly 50 cm long and have a rotational speed of 2800 rpm (Fig. 1). The impeller of Fan I, with 37 forward curved blades has an outer diameter of 130 mm and a depth of 55 mm. Accordingly, Reynolds number based on the blade tip diameter and speed is Retip = 136,000 and Mach number at the tip is Mtip = 0.05. The impeller of Fan II, with 25 forward curved blades, has an outer diameter of 120 mm and a depth of 85 mm. Accordingly, Reynolds number based on the blade tip diameter and speed is Retip = 110,000 and Mach number at the tip is Mtip = 0.05. Performance curves of both fans show that Fan II provides slightly higher pressure heads throughout the whole flow rate range (Fig. 2). At the operating point, the flow rate of Fan I is 72 lt/ s and of Fan II is 80 lt/ s. Numerical
350
Fan I - experimental Fan II - experimental
300
Fan I - numerical Fan II - numerical
P ressu re Head [P a]
250 200 150 100 50 0 0,00
10,00
20,00
30,00
40,00
50,00
60,00
70,00
80,00
90,00
The governing equations employed for LES are obtained by filtering the time-dependent Navier-Stokes equations. The filtering process effectively filters out the eddies whose scales are smaller than the filter width or grid spacing used in the computations. The resulting equations thus govern the dynamics of large eddies. The filtering operation of any flow variable is denoted with an overbar in Eqn. (1), and defined as the convolution integral of the field with a filter kernel G. Filtering substantially reduces the amplitude of the high-frequency spatial Fourier components.
Flow rate [lt/s]
Fig. 2—Performance curves of Fan I and Fan II (solid lines: experimental data; dots: numerical data). simulations and experimental measurements are performed at these operating points. Overall sound power level of the fans increase with the flow rate. At the operating point, the first radial fan system (Fan I) has a 6 dB higher overall sound power level than the second one (Fan II).
3 3.1
COMPUTATIONAL METHODS Computational Fluid Dynamics
Modern hybrid CAA techniques can be separated into two steps; the first step being the determination of the unsteady flow data and the second step being the computation of the acoustic data. Time-dependent flow calculation can be performed via unsteady Reynolds averaged Navier-Stokes equations (uRANS), LES or direct numerical simulation (DNS). URANS requires relatively low computational time and power; however, because of intrinsic modeling errors, it cannot represent turbulent fluctuations accurately. On the other hand, DNS aims to solve the Navier-Stokes equations without any modeling by resolving the whole range of time and length scales; from integral scales to Kolmogorov scales. With DNS, one can resolve all the scales and obtain the mean, turbulent and acoustic parts of the flow parameters. The main disadvantage of such methods is the enormous computational cost. Therefore only relatively simple flow configurations at modest Reynolds numbers are studied with DNS. In this paper LES method, which resolves the large and energetic scales of turbulence and models the small and dissipative scales, is used to calculate the unsteady flow field. LES is a good compromise for acoustic calculations, since the dominant noise sources are generated by the energetic eddies and these are resolved without any modeling. Flow simulation is performed with the commercial software Fluent21. Noise Control Eng. J. 57 (3), May-June 2009
¯ 共x兲 =
冕 共x⬘兲G共x,x⬘兲dx⬘
共1兲
D
Here, D is the domain of concern. The filtered form of the incompressible, unsteady continuity and transport equations for LES are
u¯i =0 xi and
共2兲
冉 冊
u¯i 共u¯iu¯j兲 u¯i p¯ ij = µ − − + t xj xj xj xi xj
共3兲
where ui is the flow velocity in i direction, p is pressure, µ is the fluid viscosity and ij is the subgrid-scale stress, which is defined as
ij ⬅ uiuj − u¯iu¯j
共4兲
The subgrid-scale stresses resulting from the filtering operation are unknown, and require modeling. The majority of subgrid-scale models in use today are eddy viscosity models of the following form: 1 ij − kk␦ij = − 2µtS¯ij 3
共5兲
where ␦ij is the Kronecker delta function, µt is the subgrid-scale turbulent viscosity, and S¯ij is the rate-ofstrain tensor for the resolved scale.
冉
1 u¯i u¯j + S¯ij = 2 xj xi
冊
共6兲
A popular subgrid-scale model was proposed by Smagorinsky22 and further developed by Lilly23. In the Smagorinsky-Lilly model, the eddy viscosity is modeled by µt = Ls2兩S¯兩
共7兲
where Ls is the mixing length for subgrid scales and 兩S¯兩 ⬅ 冑2S¯ijS¯ij. In this study a modified version of the Smagorinsky-Lilly model is employed, where Ls is computed as 171
Fig. 4—Pressure distribution on the volute of Fan II (color map shows the results of the fine mesh (i.e. 11· 106 CV) and the solid contour lines show the results of the coarse mesh (i.e. 2.5· 106 CV)).
Fig. 3—Detail of the computational mesh. Ls = min共d,CsV1/3兲
共8兲
where = 0.42, d is the distance to the closest wall, Cs is the Smagorinsky constant and V is the volume of the computational cell. Lilly derived a value of 0.23 for Cs from homogeneous isotropic turbulence in the inertial subrange. However, this value was found to cause excessive damping of large-scale fluctuations in the presence of mean shear or in transitional flows. Cs = 0.1 has been found to yield the best results for a wide range of flows; hence in this study the default value for Cs is set to 0.1. Computational mesh for individual fans comprises approximately 2.5· 106 control volumes (Fig. 3). Density of the control volumes is increased in the vicinity of the fan blades, where most of the sound emission occurs. Cell distribution is forced to be finer on the walls to resolve the boundary layer and in the neighborhood of the blade. The dimensionless wall distance y+ is kept about 1 over the whole propeller surface and the use of a wall model is omitted. Mesh elements surrounding the impeller are structured and hexahedral, whereas unstructured tetrahedral elements are used in the volute. For aeroacoustic computation, flow has to be simulated with small time steps for a large time interval. This causes even with parallel processing a sudden growth in computational time. In order to prevent unfeasible CPU-time requirements, the number of control volumes is kept as low as possible. In order to check for the numerical error due to grid dependency, an additional simulation is performed with a finer grid involving 11· 106 control volumes. Both time-averaged and instantaneous results of the coarse mesh (i.e. 2.5· 106) are in agreement with the results of the fine mesh (i.e. 11· 106). For Fan II, the coarse mesh resulted in a pressure head of 50 Pa at a flow rate of 80 lt/ s, 172
Noise Control Eng. J. 57 (3), May-June 2009
whereas the fine mesh resulted in 48 Pa. Figure 4 shows that the instantaneous pressure distributions predicted with the coarse and fine meshes concur. The computational domain is divided into two zones, one surrounding the rotating impeller and other surrounding the stationary volute. Zones are coupled via a sliding interface and mass balance is forced across the sling interface. In order to minimize interpolation errors, the ratio of the control volumes across the sliding interface is kept below 4:1. The employed boundary conditions are no-slip at the walls, constant total pressure at the inlet and constant static pressure at the outlet. The computational domain is initialized with the flow data obtained from a steady RANS simulation, in order to accelerate convergence. Spatial discretization is performed with 2nd order central differencing scheme and temporal discretization with 2nd order implicit dual time stepping scheme. The aerodynamical and acoustic time steps are set equal as 1 · 10−4 s, i.e. about 1° of rotation of fan is simulated at each time step. Accordingly, the maximum frequency resolved is 5000 Hz. Acoustic data are extracted from the LES results and recorded after nearly ten rotations of the fans, so that only statistically steady data are evaluated in the acoustic computation. After the statistically steady state is achieved, flow simulations are continued further for about 5 revolutions of the fans, i.e. for 0.107 seconds.
3.2
Acoustic Computation
Flow induced noise of a fan consists of monopole, dipole and quadrupole sound sources. Monopole sources are originated by the volume displacement of the rotating blades. Monopoles create pure tones, which are negligible for flows with Mach numbers less than 0.6. The quadrupole sources are emitted due to the
turbulent fluctuations of fluid elements away from the solid surfaces. Quadrupoles inherit the characteristics of the broadband noise and these are negligible as long as the Mach number remains less than 0.8. Pressure and viscous stress fluctuations on solid surfaces create dipole sources. Dipole sources involve both tonal and broadband components. In this paper, only the dipoles are computed, since among the three types of sound sources (i.e. monopoles, dipoles and quadrupoles) the dipole terms dominate the sound emission in turbomachinery operating at low Mach numbers24. The input of aeroacoustic computation is the time-dependent pressure fluctuations on surfaces, which are obtained from the flow simulation. The flow data are used to calculate the acoustic source terms on the right hand side of the inhomogeneous wave equation (Eqn. (9)).
冉
2 f 2 ⬘ 2 ⬘ pij␦共f兲 2 − c0 2 =− t xi xj xi
冊
共9兲
Here, ⬘ is density fluctuation, c0 is the speed of sound in the mean flow and ␦ is the dirac delta function. The solid surfaces, on which dipole sources are calculated, are defined with the function f共xជ , t兲 as
冦
⬍0 in the solid body on the boundary f共xជ ,t兲 = =0 ⬎0 outside the solid body
冧
共10兲
The compressive stress tensor is defined as pij = p⬘␦ij − ij
共11兲
On the solid surfaces pressure forces are about ten times larger than the viscous forces. Therefore the compressive stress tensor is approximated to pij ⬇ p⬘␦ij. Dipole sources are created on the surfaces of the blades and casing. Since the blades are rotating at a constant speed, the dipoles created on the blades cause a tonal noise in far field at the blade passing frequency and at its harmonics. On the casing, dipoles are created due to turbulent motion with various length scales, like the periodic excitation of the blade motion, vortex separation at the trailing edge and boundary layer separation. Sources on the casing are stationary and contribute both to the tonal and broadband noise. The experimental measurements of the investigated fans show that the tonal noise at the blade passing frequency is negligibly small compared to the broadband noise. Hence, in this study, the rotating dipole sources are not taken into account and only the stationary dipoles on the casing are assigned as the acoustic sources. The Noise Control Eng. J. 57 (3), May-June 2009
Fig. 5—Acoustic coupling of the interior and exterior multi domain BEM models. stationary dipoles are fed into the software LMS Sysnoise25, where the acoustic computation is performed. Aeroacoustic computation involves two steps: i) assigning the dipole sources on the acoustic mesh and ii) solving for the propagation of the sound sources inside and outside the casing. Approximately seven nodes are required to resolve a wavelength in aeroacoustic computation. Therefore, the numerical grid for the acoustic computation is much coarser than the LES grid. Hence, the time-dependent flow data at each node of the fine LES mesh are mapped onto the coarse aeroacoustic mesh via interpolation algorithms. The time-dependent data are then converted to frequency domain and acoustic dipole sources are calculated. Inhomogeneous wave equation is solved via a multi domain BEM analysis. The analysis consists of two models, which are the direct BEM interior and the direct BEM exterior models (Fig. 5). Dipole sources are assigned on the interior model as discrete sound sources on the nodes of the acoustic mesh. Both models are linked at the openings of the duct, through a fluidfluid coupling. The coupling satisfies the boundary condition at the openings, equivalent to the ambient pressure boundary condition. The boundary condition applied on the duct surface is the rigid wall boundary condition. Interior and exterior models are solved simultaneously. A fictitious surface, which is used for post-processing in both experiments and numerical calculations, is shown in Fig. 6. The fictitious surface is in form of a rectangular box and surrounds the fan system. Field points, i.e. nodes of the fictitious surface, are located in the acoustic far field. Acoustic signal are evaluated at the field points to compute sound pressure and sound intensity values. Sound pressure level is defined as: 173
Fig. 7—Experimental set-up for both sound intensity and sound pressure measurements (left: Fan I, right: Fan II). Fig. 6—BEM model and fictitious surface for acoustic post-processing.
冉 冊
SPL = 10 log10
p2 , 2 pref
共12兲
where p is the sound pressure at a field point. The reference pressure in air is chosen as pref = 2 · 10−5 Pa. The sound intensity, which shows the average rate at which sound energy is transmitted through a unit area, is defined as I=
p2 0c
共13兲
Where 0 is mean density and c is the speed of sound. Correspondingly, the sound intensity level is defined as:
冉 冊
IL = 10 log10
I2 2 Iref
共14兲
intensity probe with 50 mm spacer over the hypothetical surface and intensity measurements are performed using two-microphone technique. Microphone pair is matched for phase and amplitude response and calibrated for the frequency range of interest. Measurement setup is placed on a wire cage that divides the chamber in two vertical directions. Wire diameter and cage pattern provide a good suspension and also avoid scattering and refraction from the ground. Data from this measurement are evaluated in 1 / 12 octave frequency spectra. These data are used to validate the numerical prediction of directivity. Secondly, sound pressure levels (SPL) are measured at discrete microphone points in narrow band frequency spectra. For these measurements microphones are placed in acoustic far field. These data are used for a detailed validation, since SPL distribution in narrow band provides an insight to the characteristics of the sound signal and indicates the frequencies which are excited most by the acoustic sources.
The reference sound intensity is chosen as Iref = 10−12 Watt/ m2 in accordance with pref.
5
4
5.1
EXPERIMENTS
Acoustic measurements are performed in a full anechoic room, which has a background noise level of less than 20 dB. Dimensions of the anechoic room are so that the minimum measurable frequency is 63 Hz, which is beyond the frequency range of interest in this study. The experimental setup is placed in the geometrical center of the anechoic room. Both physical and numerical setups include a reflective surface underneath the fan. Figure 7 shows the experimental setups and the reflective surface underneath these setups. Two sets of measurements are performed for both fans. First, sound intensity maps are measured on the grid points of the measurement surface shown in Fig. 6. Sound intensity mapping is selected as a measurement technique, because the intensity vector defines the acoustic energy and radiation pattern which gives the maximum information about the acoustic field. In the experiment, a five axis robot arm is used to control the 174
Noise Control Eng. J. 57 (3), May-June 2009
RESULTS Aerodynamics
Vorticity is one of the main causes of flow-induced noise. Figure 8 shows the instantaneous distribution of the vorticity magnitude at a cross-section across the impeller. Vorticity is produced mainly around the blades; especially on the trailing edge. Vorticity magnitudes around the blades are of the same order in both fans. In Fan I vorticity is transported further with the flow through the volute; however, in Fan II vorticity is dissipated immediately. In Fan I, large vorticity magnitudes indicate a strong tongue—blade interaction and poor flow conditions through the volute due to the limited area increase in rotational direction. Hot spots upstream of the blades indicate that inflow is not axisymmetric, which causes an unbalanced loading on the blades. Figure 9 shows instantaneous vorticity contours on a longitudinal cross section. A helical flow is visible in the outlet pipe of the Fan I, which generates vorticity
Fig. 8—Instantaneous vorticity distribution across the impeller a) Fan I, b) Fan II. through the outlet pipe. On the other hand, large vorticity magnitudes are only visible in the neighborhood of the impeller in Fan II. Vorticity contours indicate that powerful sound sources are distributed on the whole geometry in Fan I, whereas these are located only on the impeller in Fan II. Accordingly, one can conclude that Fan II has fewer noise sources than Fan I. Experimental measurements are in agreement with that statement and show that the overall sound power level of Fan II is 6 dB less than Fan I.
5.2
Aeroacoustics
Figures 10 and 11 show the comparison of the numerical and experimental sound intensity maps at different frequencies. Sound intensity maps are sketched at the cavity mode and at its first harmonic for both fans, which are 210 and 420 Hz for Fan I, and 270 and 520 for Fan II. Directivity of the acoustic signal depends both on the type and position of the sound sources and on the material and shape of the casing, which surrounds the sound sources. At high frequencies (420 and 520 Hz) sound waves are directed through the two openings of the casing; i.e. inlet and outlet regions. Sound intensity levels at the middle of
the top, left and right surfaces are low. At low frequencies (210 and 270 Hz) noise is directed towards inlet rather than outlet. In addition to the sound intensity measurements, SPL are measured at discrete microphone points in narrow band spectrum, to perform a detailed validation. SPL spectra recorded at the middle field point of the outlet surface are shown in Figs. 12 and 13. Numerical calculations can predict the level of broadband noise correctly. In Fig. 12, a shift in the numerical SPL curve both in frequency and amplitude is visible, especially at low frequencies. The first peak at 520 Hz is overpredicted; however, frequency and amplitude of the second peak at 840 Hz is predicted accurately. Numerical SPL prediction of Fan II matched the experimental curve satisfactorily. Experimental SPL curves are smoother than numerical curves. The smoothness of the SPL curve is related to the amount of data used to obtain these results. In the experiments, the acoustic signal is measured for about 10 s. However, in the simulation the total time for the acoustic evaluation is about 0.107 s, which corresponds to about 5 rotations of the propeller. In the experiments the acoustic signal of about 500 rotations is evaluated for both fans. The frequency analysis is
Fig. 9—Instantaneous vorticity distribution on a longitudinal cross-section a) Fan I, b) Fan II. Noise Control Eng. J. 57 (3), May-June 2009
175
Fig. 10—Numerical and experimental sound intensity maps for Fan I (left: 270 Hz and right: 520 Hz (1 / 12 octave band)). performed with far less data in the simulation than the data available in the experiment. Therefore, the numerical SPL curve involves more fluctuations than the experimental curve. If the acoustic computation is carried out for a longer time, these fluctuations will disappear, but the general shape of the curve will remain the same.
6
CONCLUSION
In the frame of this work, far field noise of two radial fans is predicted numerically and the numerical method is validated with the help of experimental measurements. In the first step of the hybrid CAA method, the
time-dependent turbulent flow field is simulated via LES. The pressure fluctuations are extracted from the flow field to calculate the dipole sources. The wave equation is solved via multi domain BEM to compute the far field acoustic signal. Experimental data are used to validate both narrow band SPL spectra at discrete microphone locations and sound intensity maps at different frequencies. Numerical predictions for both fans are in good agreement with experiments. There are some discrepancies between the experimental and the numerical SPL curves, which can be attributed to the limited time span of the numerical acoustic signal and to the grid resolution in flow
Fig. 11—Numerical and experimental sound intensity maps for Fan II (left: 210 Hz and right: 420 Hz (1 / 12 octave band)). 176
Noise Control Eng. J. 57 (3), May-June 2009
Fig. 12—Numerical and experimental sound pressure level distributions in narrow band spectrum (Fan I). simulations. The main limitation of this study is the required computational power and CPU-time. For the acoustic computation LES is performed for about 10 full revolutions of the impeller to achieve the statistically steady state and than for 5 more revolutions to record the acoustic signal. Throughout the flow simulation, a total number of 5000 time steps are solved for both fans. Computations took nearly 2 months of CPU-time in a 16-processor HP computer. Despite the large computational effort, some numerical errors due to grid resolution and low acoustic time span are inevitable.
A merit of this methodology is that the acoustic prediction is based on detailed flow simulations. Since the aerodynamical mechanisms which generate the noise sources are simulated, this information can be used to compare the aeroacoustical characteristics of different designs. For example, the comparison of LES results of the investigated fans showed clearly that high vorticity levels appear through the whole flow domain in the noisier fan (Fan I). LES shows a correlation between the large vorticity magnitudes and the location of powerful dipole sources. Accordingly, dipole sources are located only on the impeller in Fan II; whereas in
Fig. 13—Numerical and experimental sound pressure level distributions in narrow band spectrum (Fan II). Noise Control Eng. J. 57 (3), May-June 2009
177
Fan I, powerful dipoles occur on impeller, tongue, volute and outlet pipe. As a future work, this information may be utilized to reduce noise of existing fans or in the inverse design of low-noise turbomachinery.
7
REFERENCES
1. J. M. Lighthill, “On sound generated aerodynamically. I. General theory”, Proceedings of the Royal Society of London (A), 211, 564–587, (1952). 2. J. M. Lighthill, “On sound generated aerodynamically. II. Turbulence as a source of sound”, Proceedings of the Royal Society of London (A), 222, 1–32, (1954). 3. N. Curle, “The influence of solid boundaries upon aerodynamic sound”, Proceedings of the Royal Society (A), 231, 505–514, (1955). 4. J. E. Ffowcs Williams and D. L. Hawkings, “Sound generation by turbulence and surfaces in arbitrary motion”, Philosophical Transactions of the Royal Society (A), 264, 321–342, (1969). 5. T. Colonius and S. K. Lele, “Computational aeroacoustics: progress on nonlinear problems of sound generation”, Prog. Aerosp. Sci., 40, 345–416, (2004). 6. L. Gutin, “On the sound of a rotating propeller”, Zhurnal tekhnichesko fiziki, 6, 899–909, Translated as NACA TM 1195, (1948). 7. Thomas Carolus, “Ventilatoren—Aerodynamischer Entwurf, Schallvorhersage, Konstruktion”, Teubner Verlag, Wiesbaden, (2003). 8. Leonhard Bommes, Jürgen Fricke and Reinhard Grundmann, “Ventilatoren”, Vulkan Verlag, Essen, (2003). 9. W. Neise, “Noise reduction in centrifugal fans: a literature survey”, J. Sound Vib., 45, 375–403, (1976). 10. W. Neise and G. H. Koopmann, “Reduction of centrifugal fan noise by use of resonators”, J. Sound Vib., 73, 297–308, (1980). 11. W. Neise, “Review of noise reduction methods for centrifugal fans”, J. Eng. Ind., 104, 151–161, (1982).
178
Noise Control Eng. J. 57 (3), May-June 2009
12. W. H. Jeon, S. J. Baek and C. J. Kim, “Analysis of the aeroacoustic characteristics of the centrifugal fan in a vacuum cleaner”, J. Sound Vib., 268, 1025–1035, (2002). 13. A. Maaloum, S. Kouidri and R. Rey, “Aeroacoustic performances evaluation of axial fans based on the unsteady pressure field on the blades surface”, Appl. Acoust., 65, 367–384, (2004). 14. S. Khelladi, S. Kouidri, F. Bakir and R. Rey, “Predicting tonal noise from a high rotational speed centrifugal fan”, J. Sound Vib., 313, 113–133, (2008). 15. M. Nallasamy and E. Envia, “Computation of rotor wake turbulence noise”, J. Sound Vib., 282, 649–678, (2005). 16. S. F. Wu and S. Su, “Modeling of the noise spectra of axial flow fans in a free field”, J. Sound Vib., 200, 379–399, (1997). 17. R. Ballesteros-Tajadura, S. Velarde-Suárez, J. P. Hurtado-Cruz and C. Santolaria-Morros, “Numerical calculation of pressure fluctuations in the volute of a centrifugal fan”, Helicobacter, 128, 359–369, (2006). 18. Y. Özyörük and E. Alpman, “Frequency-domain prediction of turbofan noise radiation”, J. Sound Vib., 270, 933–950, (2004). 19. S. C. Lin and C. L. Huang, “An integrated experimental and numerical study of forward-curved centrifugal fan”, Exp. Therm. Fluid Sci., 26, 421–434, (2001). 20. A. Gerard and A. Berry, “Control of tonal noise from subsonic axial fan. Part 1: Reconstruction of aeroacoustic sources from far-field sound pressure”, J. Sound Vib., 288, 1049–1075, (2005). 21. Fluent Version 6.2, User’s Guide, USA, (2005). 22. J. Smagorinsky, “General Circulation Experiments with the Primitive Equations. I. The Basic Experiment”, Mon. Weather Rev., 91, 99–164, (1963). 23. D. K. Lilly, “On the Application of the Eddy Viscosity Concept in the Inertial Subrange of Turbulence”, NCAR Manuscript 123, (1966). 24. Marvin E. Goldstein, Aeroacoustics, McGraw-Hill Book Company, New York, (1976). 25. Sysnoise rev.5.6, User’s Manual, LMS Numerical Technologies, Leuven, Belgium, (2005).
