1

1N09- 086/ A

Lecture 2

Axial Fans: Noise Prediction

Th. Carolus

UNIVERSITÄT SIEGEN Institut für Fluid- und Thermodynamik

2N09- 086/ A

UNIVERSITÄT SIEGEN Institut für Fluid- und Thermodynamik

2.1 Motivation

2.2 Summary: Fan noise mechanisms

2.3 Classification of prediction methods

2.4 Examples

2

3N09- 086/ A

ZX

Y

Cm [m/s]

14.0012.5011.009.508.006.505.003.502.000.50

-1.00-2.50-4.00-5.50-7.00-8.50

-10.00Quelle: H. Reese 2001

Autokühlgebläse (Bosch)

1. Motivation (I)

4N09- 086/ A

Cooling system for the Diesel-electric high speed train ofFirst Great Western (Voith)

Motivation (II)

3

5N09- 086/ A

023 A

Motivation (III)

Reichstagklimatisierung (Howden)

Geräte/Klima (Ziehl Abegg)

6N09- 086/ A2.2 Summary: Fan noise mechanisms

Fluid displacement

Forces on surfaces

Turbulence in fluid

4

7N09- 086/ AForces on Blade Cascade

spatially non-uniform inflow ⇒ tonal noise („Unsteady loading noise“) unsteady inflow ⇒ broad band noise

8N09- 086/ A

Flowseparation(FS)

Turbulent boundary layer/blade surface interaction(TBS)

A detail: Airfoil self-noise

Turbulent boundary layer/trailing edge interaction(TBTE)

⇒ mostly broad band noise

5

9N09- 086/ ASecondary sources, e.g. tip clearence flow

10N09- 086/ A2.3 Classification of prediction methods

CLASS IBasic maschine parameters• type• diameter• speed• flow rate• ressure rise

CLASS II•Separate consideration of various noise generationmechanisms

•Simplified fan geometry, flowfield (e.g. blade ⇒ flat plate)

CLASS III•Separate considerationof various noisegeneration mechanisms

•Detailed fan geometryand flow field (e.g. fromCFD-computation)

Acoustic models for all noisegeneration mechanisms

(SPECTRAL) SOUND POWER

Simple algebraic function(correlation)

Classification of fan noise prediction methods

6

11N09- 086/ A2.4 Examples

*, , ,

0 0 0

110lg 1 10 lg dBt aW ges W ges Wspez R

V p uL L L mV p c

∆∆ η

≡ − − = + ⋅

Class IRegenscheit-method; VDI-Richtlinie 3731

⇒ Specific sound power level for various types of fans

12N09- 086/ A

Class I

7

13N09- 086/ AA class II – noise prediction method (I)

Sharland/Költzsch/Schneider-method

14N09- 086/ A

Velocity fluctuations of turbulent flow: curve fit to dimensionless experimental results from various turbulence generators

Example: Turbulent ingestion (TI)

⇒ Lift force fluctuations in terms of modeled turbulent velocity fluctuations

( ) ∞≈ ⋅ ⋅ ⋅ ⋅ ⋅, 43

2

0

'ρak TIdPf const B w C L

df cdwdf

ΛΛ

∞

=fSrw

Λ( )F Sr

∞= ⋅ ⋅ ⋅12 ( )2 10' 10 ΛΛF Srdw w Tu

df

A class II – noise prediction method (II)

8

15N09- 086/ A

100 500 1000 5000 10000f [Hz]

0

10

20

30

40

50

60

70

80

L W [

dB]

Messung: LW, ges = 79.5 dBMessung: LW, ges = 79.5 dBRechnung: LW, ges = 80.7 dBRechnung: LW, ges = 80.7 dB

GA, ϕ = 0.179GA, ϕ = 0.179

Typical result

Prediction „smooth“Only broad bandVery fast method

Schneider, M.: Der Einfluss der Zuströmbedingungen auf das breitbandige Geräusch eines Axialventilators. Fortschritt-Berichte VDI Reihe 7: Strömungstechnik (Dr.-Ing. Diss. Univ. Siegen). Vol. Nr. 478. Düsseldorf: VDI Verlag GmbH, 2006. - ISBN 3-18-347807-2

A class II – noise prediction method (III)

16N09- 086/ A

Detailed unsteadyflow field data

Fluctuationforces as sources, e.g. in a BEM acoustic field calculation

Sound power spectrum10

210

310

40

20

40

60

80

PSD

L sp [d

B]

f [Hz]

LES: OASPL = 70.76 dBExp: OASPL = 74.51 dB

Reese, H.: Anwendung von instationären numerischenSimulationsmethoden zur Berechnung aeroakustischerSchallquellen bei Ventilatoren. (Dr.-Ing. Dissertation Universität Siegen), Fortschritt-Berichte VDI Reihe 7, Nr. 489, VDI Verlag, Düsseldorf, 2007

Reese, H., Kato, C., Carolus, T.: Large eddy simulation of acoustical sources in a low pressure axial-flow fan encountering highly turbulent inflow. ASME J. of Fluids Engineering, March 2007, Vol. 129, pp. 263-272

A Class III – noise prediction method

9

17N09- 086/ AUnsteady CFD-Methods

Direct Numerical Simulation (DNS):• Basic equations are solved without any

additional models • Solution contains the acoustic field • High numerical costs ~ Re3

Large Eddy Simulation (LES):• Filtering of the basic equations • Large scales are solved directly • The numerical costs are still high ~ Re1.4.

Unsteady Reynolds Averaged Navier-Stokes Simulation (URANS):• Ensemble averaging of the basic equations• Turbulence completely modeled • The numerical costs are independent

of Re

18N09- 086/ AHybrid CFD-Methods

Detached Eddy Simulation (DES): • A combination of an URANS with parts of a LES • The near wall area is solved in URANS mode• The detached flow field is solved in LES mode• The numerical costs are comparable to URANS

Scale Adaptive Simulation (SAS):• Improved URANS method because the turbulence modeling depends on the flow field• Turbulence model is based on a physical length scale (von Kármán-length scale)• The model behaves like a LES in the detached flow region • The near wall region is solved in a stationary mode (similar to the DES method) • The model does not require a switching between the near wall region and the detached flow field • The numerical costs are comparable to URANS

LES RANS

10

19N09- 086/ AUnsteady CFD

LESSAS

DESURANS

Snap shot of the absolute velocity (v/uTIP) at 50% blade height

20N09- 086/ A

Quelle: Reese, Kato 2004

Example: LES

11

21N09- 086/ ACFD: Unsteady Blade Surface Pressure

Blade suction side (p / 0.5ρuTip2)

SAS LES

22N09- 086/ A

102

10320

40

60

80

100

120

140

PSD

L p [dB

]

f [Hz]

ExpLESDESSASURANS

CFD: Blade Surface Pressure Spectra

Hub

Leading edge

Suction sideMonitoring point

12

23N09- 086/ ANumerical Noise Prediction with BEM

(LES based acoustic BEM)

300 Hz= BPF

24N09- 086/ A

13

25N09- 086/ A

Acoustic results as good as the unsteady flow field dataIn principle capable to predict noise of realistic fan assembliesLES very costly - hybrid CFD-methods promisingFeeding in megabytes of source data into acoustic models maybe tricky

Essentials of Class III – noise prediction methods

26N09- 086/ A5. Summary and Conclusions

Aeroacoustic noise sources in low Ma number fans are flow induced forces

Several principle mechanisms can be identified, such as spatial non-uniforminflow, turbulent ingestion, blade self noise, tip clearance flow, etc.

Noise prediction methods range form simple correlations (class I) to complexcomputational aeroacoustics (CAA) methods (class III)

Confirmation of the classical rule: Detailed acoustic prediction requires excellent source data, e.g. the unsteady flow field in realistic fan assemblies

- High quality LES very costly and for daily tasks not yet an option- Hybrid CFD-methods are promising

Feeding in detailed source data into acoustic models maybe tricky