Predic5ng)Turbofan)Fan:Stage)Noise) - Boston University · 2016-12-22 ·...

Department of Mechanical Engineering

Predic5ng Turbofan Fan-‐Stage Noise

Sheryl Grace

Boston University TuCs CEEO


Predic5ng Turbofan Fan-‐Stage Noise

What is it? Who cares?


Turbofans

Military Low bypass Multistage

Commercial High bypass Single stage Large axial gap


Turbofans

Commercial High bypass Single stage Large axial gap

Strict noise regulations Getting stricter


Tonal

Broadband

Types of noise from fan stage

Nallasamy and Envia, JSV, 2005

than rotor wake turbulence noise (see also Ref. [13]). However, it is to be noted that the inlet/exhaust noise separation at higher speeds encountered problems as discussed in Section 2.3.

3.3. Variation of noise with vane count

As the vane count is reduced, the rotor-wake turbulence generated broadband noise is expectedto reduce substantially. The cut-on fan design is expected to produce less broadband noise thanthe cut-off design to counteract the cut-on tone noise levels.Fig. 11(a) shows computed exhaust power spectra for 54 radial vane and 26 radial vane

configurations at approach and the corresponding measured spectra are shown in Fig. 11(b). It isseen that the experimentally observed noise reduction due to reduced vane count is clearly shownin the predictions. The acoustic power for 26 radial vanes is substantially lower than for the 54radial vanes configuration. The power is expected to vary as 10 logN2

V (see Ref. [10]) where NV isthe vane count. This should result in 6.3 dB, (20 LOG (54/26)), reduction in power levels for the26-vane OGV compared with the 54-vane OGV. The predicted reduction in power due to thereduction in vane number is higher than expected at high frequencies. At low frequencies!o3 kHz"; the experimental result is insensitive to the vane count. The noise levels computed forthe two configurations are not significantly different, probably because at low frequencies thereare additional noise sources present, which are not modeled in the present theory. A similar trendwith vane count was observed in Boeing broadband noise experiments on the effect of vane counton noise spectra [14].The reduction in acoustic power level due to a reduction in vane count was further explored to

understand the contributing factors, in an effort to explain the more than expected reduction inacoustic power. An examination of inputs and computed power levels indicated a strongdependence on the vane stagger angle. Fig. 12 shows the power levels of the exhaust approach

ARTICLE IN PRESS

Fig. 10. Effect of fan tip speed on fan exhaust duct acoustic power spectrum for the baseline stator. (a) Computedspectra: solid line, approach; dashed line, cutback; and dotted line, takeoff. (b) Measured spectra: solid line, approach;dashed line, cutback; and long dashed line, takeoff.

M. Nallasamy, E. Envia / Journal of Sound and Vibration 282 (2005) 649–678 663

approach

cutback

takeoff


rotor FEGV

exhaust plane

fan FEGV

Fan-stage tonal noise, explained

Frequency depends on number of blades and rate of rotation Propagation depends on number of blades and vanes, the frequency, the axial

Glow speed, the shape of the annulus, etc.

Engine designers have this pretty well modeled Use blade counts and duct liners to control


Fan-stage broadband noise, primer

Rotor-FEGV interaction noise considered dominant source (especially at exhaust plane)

Fan Broadband Noise Sources

7

Fan-OGV interaction

Fan-IGV interaction Inlet turbulence and distorsion fan interaction

Fan clearance

Fan self noise

Inlet Boundary layer fan interaction

Typical broadband noise sources

Fan broadband noise may have different origins Fan-OGV interaction is considered as the main contributor of total broadband noise on modern civil engines. Other BBN sources cannot be neglected and should be also studied and controlled.

Envia, AIAA workshop presenta5on, 2014

than rotor wake turbulence noise (see also Ref. [13]). However, it is to be noted that the inlet/exhaust noise separation at higher speeds encountered problems as discussed in Section 2.3.

3.3. Variation of noise with vane count

As the vane count is reduced, the rotor-wake turbulence generated broadband noise is expectedto reduce substantially. The cut-on fan design is expected to produce less broadband noise thanthe cut-off design to counteract the cut-on tone noise levels.Fig. 11(a) shows computed exhaust power spectra for 54 radial vane and 26 radial vane

configurations at approach and the corresponding measured spectra are shown in Fig. 11(b). It isseen that the experimentally observed noise reduction due to reduced vane count is clearly shownin the predictions. The acoustic power for 26 radial vanes is substantially lower than for the 54radial vanes configuration. The power is expected to vary as 10 logN2

V (see Ref. [10]) where NV isthe vane count. This should result in 6.3 dB, (20 LOG (54/26)), reduction in power levels for the26-vane OGV compared with the 54-vane OGV. The predicted reduction in power due to thereduction in vane number is higher than expected at high frequencies. At low frequencies!o3 kHz"; the experimental result is insensitive to the vane count. The noise levels computed forthe two configurations are not significantly different, probably because at low frequencies thereare additional noise sources present, which are not modeled in the present theory. A similar trendwith vane count was observed in Boeing broadband noise experiments on the effect of vane counton noise spectra [14].The reduction in acoustic power level due to a reduction in vane count was further explored to

understand the contributing factors, in an effort to explain the more than expected reduction inacoustic power. An examination of inputs and computed power levels indicated a strongdependence on the vane stagger angle. Fig. 12 shows the power levels of the exhaust approach

ARTICLE IN PRESS

Fig. 10. Effect of fan tip speed on fan exhaust duct acoustic power spectrum for the baseline stator. (a) Computedspectra: solid line, approach; dashed line, cutback; and dotted line, takeoff. (b) Measured spectra: solid line, approach;dashed line, cutback; and long dashed line, takeoff.

M. Nallasamy, E. Envia / Journal of Sound and Vibration 282 (2005) 649–678 663


Full RANS CFD of rotor stator interaction (even if modeled as one-on-one) takes several months to start up and then weeks to obtain enough data to perform the statistics. May not even capture turbulence statistics correctly.

Some think only LES might be able to do the job

Can CFD be used?


This is where my research comes in…

Motivation •  Create a computational tool for predictions of fan-stage noise that can be used in design

•  Focus on interaction noise problem and exhaust noise •  Must be a low-order model

Outline

•  Describe two essential building blocks

•  Give overview of how the pieces Ait together to develop the model

•  Describe two experiments used for validation

•  Show some broadband noise predictions from the method

•  Wrap up


Outline





•  Wrap up


1st building block --- inAlow


Physically, you have turbulence coming into the FEGV and interacting with it 81% Span 25% Span

97% Span 81% Span 25% Span

97% Span

PowerSpectralDensity

(ft2/sec2/Hz)

101

100

10!1

10!2

10!3

10!4

101

100

10!1

10!2

102 104 102 104 102 104

102 102 102104 104 104

frequency (Hz)

frequency (Hz)

Figure 26. PSDs computed from upwash velocities measured in the rotor wake with the rotoroperating at 61.7% speed (7808 RPMC).

Figure 27. Comparison of experimental and von Karman model PSDs. Experimental PSDs were computedfrom same time traces used in Fig 26., but after removing periodic component of the signal.

PowerSpectralDensity

(ft2/sec2/Hz)

29NASA/TM—2003-212329

Data from hot-‐wire probe upstream of vane

Data averaged over 129 wheel rota5ons

Power Spectral Density: think square of the Fourier Transform

Full signal

Periodic part removed


We need a useful model of the turbulent inGlow

Hot wire data is not going to be available LDV data may be available (averaged, not time accurate) CFD calcs just give statistics – like turbulent kinetic energy and dissipation

SimpliAications homogeneous (same throughout space, deAinitely not true, wake Alow) isotropic (same in all directions u1u1 = u2u2 = u3u3)

turbulence convected by the meanAlow (relates space and time through velocity) (x,t) x + U t k1 and ω related through U

Correla5on tensor defined by correla5ons

3-‐D energy spectrum defined by correla5on tensor

Liepmann model for energy spectrum

ω e-‐iωt dω

ω


Example of spectrum from experimental turbulence intensity and length scale

25% span

50% span

81% span

mid-‐gap streamwise

upwash

Hotwire Von Karman Liepmann Gaussian

Experimentally based spectrum Related 1D form (Λ = Ls)

ω


2nd building block --- interaction model

Really high level overview of unsteady aerodynamics


Early unsteady aerodynamics – focus was Alutter

used Linearized Euler Eqs. as basis

considered response of Alat-plate airfoil in unsteady setting

“airfoil” moving : heaving and/or pitching (Wagner, Theodorsen)

considered response of stationary Alat-plate airfoil to Alow disturbance

“airfoil” with unsteady inAlow (gust) (Kussner, von Karman, Sears, Possio)


e-‐iωt dω

€

∫Each component looks like an individual gust-type disturbance

€

E2,2ei(k1x1 −ωt )+ ik2x2 + ik3x3

Later, real airfoil shapes were considered and the associated Aield acoustics (Atassi and others)

Also similar methods for determining the response of a cascade of “airfoils” were derived

ω

We have 2nd type of problem


Think about the fan-stage Slice it at a given radial position and unwrap it

!!"

!#"

" "

# "

!"

$"

%&"

# "

#"$'!("

%#"

%!"

&#"

&!"

'"

Rotor

Wake turbulence

FEGV

Cascade of flat plates

Use strip theory to build up unsteady pressure on entire vane


Outline





•  Wrap up


Method

Based on experimental or CFD data, 3D spectrum model

Aerodynamic core : 2-‐D, Alat-plate cascade response to 3D gust + strip theory

Acoustic pressure: Green’s function for annular duct (avoid singular values kmn 0)

InBlow wake turbulence

Vane unsteady pressure spectrum

Acoustic pressure and velocity spectrum in duct

Power spectrum at duct exit

Cascade response to oblique gust


Outline





•  Wrap up


22-‐inch diameter turbofan model ConBiguration •  22 blade rotor •  54 vane stator – baseline •  26 vane stator – low count •  26 vane stator – swept Design Parameters •  Maximum RPM: 12,656 •  Maximum tip speed: 1,215 ft/s •  Maximum fan weight Blow: 100.5 lbm/s

9 x 15 Foot Wind Tunnel (NASA Glenn)

Source Diagnostic Test (SDT)


Experiments Experimental wake data Laser Doppler Velocimetry (LDV) •  Optical, 2-‐component (axial, tangential) system •  Steady measurements not restricted by rotor speed Hot-‐wire Anemometer •  4-‐wire (5 micron diameter tungsten) anemometer •  3-‐component system •  Unsteady measurements restricted by rotor speed

LDV System

Experimental acoustic data Sideline Traversing Microphone System •  Translating microphone probe + 3 aft Bixed microphones •  Provides total PWL only – includes extraneous modes


FC2 benchmark

Fan Broadband Noise Benchmarking Panel Session FC2

Grid turbulence interaction with an annular cascade Based on data taken in the anechoic subsonic open jet facility at the Ecole Centrale de Lyon Posson & Roger, AIAA Journal, 49(9), 2011

Slide 2

2 Turbulence Grids T1 (~3%), T2 (~6%) 2 Cascades C1 (V=49), C2 (V=98) Giving four test cases T1C1, T2C1, T1C2, T2C2

Subsonic open-‐jet facility at Ecole Centrale de Lyon

Grid turbulence interaction with annular cascade

Hot wire measurements and Bield microphones

Coupland, AIAA workshop presenta5on, 2014


Outline




•  Show some broadband noise predictions from the method, describe caveats

•  Wrap up


Must select a stagger angle – Glat plate model

10480

85

90

95

100

105

frequency (Hz)

PWL

(dB)

Experimental90/1050/5010/90

0.5 0.6 0.7 0.8 0.9 10

5

10

15

20

25

30

35

40

Nondimensional radial location

Stag

ger a

ngle

It makes a difference! No clear winner

Use mid value for now


FC2 predictions – based on experimental input

Trends are captured reasonable well

C2 – C1 : too large of difference , dependence on number of vanes not perfectly modeled

Some singular frequencies had to be included

102 103 104

50

60

70

80

90

100

110

120

130

140

150

160

T1C1

T1C2

T2C1

T2C2

frequency (Hz)

PWL

(dB)

103 1040

5

10

15

20

25

30

35

40

45

T2C2 − T1C1

T2C2 − T2C1

T1C1 − T1C2

frequency (Hz)

Diff

eren

ce (d

B)


SDT prediction (based on hot-wire)

Approach speed

Low-count prediction better Used mid-stagger for both

Spectral shape very good

Low frequency bump in low-count

Trend is captured

103 10480

85

90

95

100

105

frequency (Hz)

PWL

(dB)

Measured, baselineMeasured, low countPredicted, baselinePredicted, low count

103 104−4

−2

0

2

4

6

8

10

12

frequency (Hz)

Diff

eren

ce (d

B)

MeasuredPredicted


Summary and conclusions

o The possibility exists for predicting broadband noise trends from a fan stage using a low-order method

o  Parameter selection necessary due to simpliAied model ( stagger ) does not affect trend prediction

o  Fully computational prediction suffers from reliance on turbulence length scale modeled with CFD

o  Difference due to selected background turbulence intensity does not affect trends but does affect individual predictions

o  Low-count conAiguration is better predicted : potential effect

o  These are the best predictions shown in the literature for the broadband workshop case and for the SDT


Future directions

o Can more realistic vane geometry be modeled?

o Can “soft” vanes be modeled? Need to add impedance of surface to model.

Thickness Thickness and camber : Dlow turning


Questions

o NASA Glenn – Ed Envia, Gary Podboy, etc.

o  GE, P&W, NASA Glenn : CFD Ailes

o AARC for funding

o My collaborators Doug Sondak and Victor Yakhot

o Graduate students on this project Jeremy Maunus and Andy Wixom

Thanks


What if input based on CFD

Code Grid config.

Grid density* x,θ,r

Tip clearance

Turb. Model

Rotor config.

Vane config.

Comp. type

1 O-‐H 95x64x36 y k-‐ω hot baseline loosely coupled

2 C-‐H 86x89x81 y k-‐ω hot baseline loosely coupled

3 H 11x51x51 n k-‐ε approach baseline average passage

4 O-‐H 51x33x57 y/n k-‐ω hot none rotor alone

*between two measurement loca5ons in gap

4 simulations of the SDT rig obtained


CFD wake accuracy: SDT (Approach)

Passage-averaged axial velocity Passage-averaged circumferential velocity

Streamwise velocity Upwash velocity

(mid-span) (mid-span)


CFD wake accuracy: SDT (Approach)

Turbulence intensity Length scale (streamwise)

0.5 0.6 0.7 0.8 0.9 10

0.02

0.04

0.06

0.08

0.1

0.12

Nondimensional radius

Non

dim

ensi

onal

turb

ulen

ce in

tens

ity

MeasuredCFD1CFD2CFD3CFD4

0.5 0.6 0.7 0.8 0.9 10

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

Nondimensional radiusN

ondi

men

sion

al tu

rbul

ence

leng

th s

cale


SDT approach case


Exhaust sound power level prediction: approach

Baseline: 54 vanes Low count: 26 vanes

SDT approach case

103 10480

85

90

95

100

105

frequency (Hz)

PWL

(dB)

MeasuredPredicted with HWPredicted with CFD1Predicted with CFD2Predicted with CFD3Predicted with CFD4

103 10475

80

85

90

95

100

105

frequency (Hz)

PWL

(dB)

MeasuredPredicted with HWPredicted with CFD1Predicted with CFD2Predicted with CFD3Predicted with CFD4

0.5 0.6 0.7 0.8 0.9 10

0.02

0.04

0.06

0.08

0.1

0.12


Non

dim

ensi

onal

turb

ulen

ce in

tens

ity


0.5 0.6 0.7 0.8 0.9 10

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04

0.045

0.05

Nondimensional radiusN

ondi

men

sion

al tu

rbul

ence

leng

th s

cale


o  HW based prediction higher: o  potential effect maybe

o  Largest variations come from largest length scale differences


Exhaust sound power level prediction: takeoff

Baseline: 54 vanes Low count: 26 vanes

0.5 0.6 0.7 0.8 0.9 10

0.005

0.01

0.015

0.02

0.025

0.03

0.035

0.04


Non

dim

ensi

onal

turb

ulen

ce le

ngth

sca

le

CFD1CFD2CFD3CFD4

0.5 0.6 0.7 0.8 0.9 10.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

0.11


Non

dim

ensi

onal

turb

ulen

ce in

tens

ity

CFD1CFD2CFD3CFD4

103 10480

85

90

95

100

105

110

115

frequency (Hz)

PWL

(dB)

MeasuredPredicted with CFD1Predicted with CFD2Predicted with CFD3Predicted with CFD4

103 10480

85

90

95

100

105

110

115

frequency (Hz)

PWL

(dB)

Predicted with CFD1Predicted with CFD2Predicted with CFD3Predicted with CFD4

Turbulence intensity Length scale (streamwise)

CFD4

Rotor alone

Hot geometry

No 5p gap at high speed

CFD2

Baseline, hot geometry

Largest 5p gap

Background turb. level really low


Exhaust sound power level prediction Use CFD1 length scale

Baseline: 54 vanes

103 10475

80

85

90

95

100

105

frequency (Hz)

PWL

(dB)

MeasuredHW with CFD1 length scaleCFD1CFD2 with CFD1 length scaleCFD3 with CFD1 length scaleCFD4 with CFD1 length scale

103 10480

85

90

95

100

105

110

115

frequency (Hz)

PWL

(dB)

MeasuredPredicted with CFD1CFD2 with CFD1 length scaleCFD3 with CFD1 length scaleCFD4 with CFD1 length scale

Approach Takeoff

Collapses predictions, except for hw based


Exhaust sound power level prediction Trends

Number of vanes (shape of vanes also)

Baseline - lowcount

o  Completely consistent o  Slightly high

103 104−4

−2

0

2

4

6

8

10

frequency (Hz)

Diff

eren

ce (d

B)

MeasuredHot wireCFD1CFD2CFD3CFD4


Exhaust sound power level prediction Trends

Baseline: 54 vanes

Cutback-Approach

Takeoff - approach

103 104−5

0

5

10

15

frequency (Hz)

Diff

eren

ce (d

B)


103 104−3

−2

−1

0

1

2

3

4

5

6

7

8

frequency (Hz)

Diff

eren

ce (d

B)


Take-off – Cutback

103 104

−4

−2

0

2

4

6

8

10

12

14

16

frequency (Hz)

Diff

eren

ce (d

B)


o  Trend with frequency : good

o  Trend with rotor speed : o  CFD2 less variation than expected

o  CFD4 gives closest, but driven by large tip length scale: physical?


End goal: use asymptotic method of Peake et al. for cascade gust response extend method to calculate the unsteady surface pressure

Mean loading effects


Thick, perfectly aligned Blow (t.e. stagger)

Cambered, Blow aligned with chord

Cambered, Blow aligned with chord, stagger simulated

Same unsteady response

7% thick 9% camber 13o stagger

30o aoa -‐ approach


Effect of inGlow modeling assumptions

Different turbulence spectra

Liepmann matched turbulence the best and gives good spectral shape for acoustics

103 10475

80

85

90

95

100

105

110

frequency (Hz)

PWL

(dB)

MeasuredActual predictionPrediction with 2 LrPrediction with 2Lr and 2Ls

103 10480

85

90

95

100

105

110

frequency (Hz)

PWL

(dB)

Measured, baselinePredicted, LiepmannPredicted, Gaussian

Turbulence length scale is Achilles heel

Both streamwise and radial length scales are in the model

Doubling radial lengthscale doubles pressure (3 dB)

Doubling streamwise tilts the spectrum

Predic5ng)Turbofan)Fan:Stage)Noise) - Boston University · 2016-12-22 ·...

Documents

Transcript of Predic5ng)Turbofan)Fan:Stage)Noise) - Boston University · 2016-12-22 ·...