Predic5ng)Turbofan)Fan:Stage)Noise) - Boston University · 2016-12-22 ·...
Transcript of 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
Department of Mechanical Engineering
Predic5ng Turbofan Fan-‐Stage Noise
What is it? Who cares?
Department of Mechanical Engineering
Turbofans
Military Low bypass Multistage
Commercial High bypass Single stage Large axial gap
Department of Mechanical Engineering
Turbofans
Commercial High bypass Single stage Large axial gap
Strict noise regulations Getting stricter
Department of Mechanical Engineering
Department of Mechanical Engineering
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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?
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
1st building block --- inAlow
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
2nd building block --- interaction model
Really high level overview of unsteady aerodynamics
Department of Mechanical Engineering
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)
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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)
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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, describe caveats
• Wrap up
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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)
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
CFD wake accuracy: SDT (Approach)
Passage-averaged axial velocity Passage-averaged circumferential velocity
Streamwise velocity Upwash velocity
(mid-span) (mid-span)
Department of Mechanical Engineering
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
MeasuredCFD1CFD2CFD3CFD4
SDT approach case
Department of Mechanical Engineering
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
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
MeasuredCFD1CFD2CFD3CFD4
o HW based prediction higher: o potential effect maybe
o Largest variations come from largest length scale differences
Department of Mechanical Engineering
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
Nondimensional radius
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
Nondimensional radius
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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)
MeasuredCFD1CFD2CFD3CFD4
103 104−3
−2
−1
0
1
2
3
4
5
6
7
8
frequency (Hz)
Diff
eren
ce (d
B)
MeasuredCFD1CFD2CFD3CFD4
Take-off – Cutback
103 104
−4
−2
0
2
4
6
8
10
12
14
16
frequency (Hz)
Diff
eren
ce (d
B)
MeasuredCFD1CFD2CFD3CFD4
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?
Department of Mechanical Engineering
End goal: use asymptotic method of Peake et al. for cascade gust response extend method to calculate the unsteady surface pressure
Mean loading effects
Department of Mechanical Engineering
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
Department of Mechanical Engineering
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