Evaluation of Analysis Techniques for Low Frequency ... · Three design analysis techniques were...
Transcript of Evaluation of Analysis Techniques for Low Frequency ... · Three design analysis techniques were...
NASA Contractor Report 18"185"1
Evaluation of Analysis Techniques for LowFrequency Interior Noise and Vibrationof Commercial Aircraft
A. E. Landmann, H. F. Tillema, and S. E. Marshall
Boeing Commercial Airplanes
RO. Box 3707
Seattle, Washington 9812.4-2207
Contract NASI-18027
October 1989
N/ .SANational Aeronautics andSpace Administration
Langley Research CenterHampton, Virginia 23665-5225
(NASA-CF_-I61_,I) LVALUATION OF ANALYSIS
TECHNI(_UES FOR. LOW FRF_ULNCY INTEQIOR NnlSt
ANO VI_RATIqN uF C_]MMERCIAL AIRCRAFT
(hoeing Commercial Airplane Co.) -/3 pCSCL 20A c3/rl
NgO-I k o66
uncl as
0234097
https://ntrs.nasa.gov/search.jsp?R=19900005550 2020-01-15T21:32:59+00:00Z
H
8U
9
dd
U|HU
|
|
ii
H|
|
|
i A
f"
_-" ",1
FOREWORD
m
m
L
m
m
i
w
L
u
- 1
|
m
K_
m
m
This report contains comparisons of predicted and measured vibration and interior noise levels for an
aircraft, with aft mounted propeller engines. Based on these comparisons, recommendations are made
for modeling and analysis techniques for evaluation of low-frequency engine noise and vibration in the
passenger cabin. This work was conducted under NASA contract NAS1-18027 from January, 1987
through February, 1989. Work was managed by the Acoustics Division at the NASA Langley Research
Center. During 1987 work, Mr. H. Morgan was the chief of the Acoustics Division and Mr. D. G. Ste-
phens was the technical monitor for the contract. During the 1988 and early 1989 work, Mr. D. G. Ste-
phens was chief of the Acoustics Division at NASA and Dr. K. Shepherd was the technical monitor for
the contract.
All analytical work was performed under the direction of the Noise Research staff of the Boeing Com-
mercial Airplanes. A number of engineering organizations and subcontractors contributed to the
successful completion of the project as planned. Key contractor personnel responsible for this effort
were
Noise Technolo_,
L. M. Butzel
L. W. Craig
C. G. HocJgeA. E. Landmann
S. E. Marshall
G. K. Oueitzsch
P. M. Serati
Slructures Technolo_
K. H. Dickenson
R. L. Dreisbach
R. R. EnsmingerM. T. Hutchinson
H. J. Jamshidiat
E. E. MeyerH. E Tillema
Propulsion Technolotw.M. A. Heidari
R. L. Martin
J. L. White
T. E Yantis
Weights Technolo_D. E. Cook
D. J. Roberts
Subcontractors
A. C. Aubert - Cambridge Collaborative, Inc.
J. E. Manning - Cambridge Collaborative, Inc.
L. D. Pope
m
-.L
w
wi
m
i
[]
|Hi
i-imdI
m
iil !
mE _
i
J
ii
R i|mR
el
m
a
m
I
i !|
n m
|i
m
M
u
TABLE OF CONTENTS
FOREWORD ........................................................................... i
TABLE OF CONTENTS ................................................................ iii
LIST OF TABLES ....................................................................... v
LIST OF FIGURES .................................................................... vii
1.0 SUMMARY ....................................................................... 1
2.0 INTRODUCTION ................................................................. 3
i
W
L-
3.0
4.0
APPROACH ...................................................................... 5
3.1 Noise and Vibration Tests ..................................................... 5
3.2 Analysis Model Development .................................................. 73.2.1 Finite Element Model ................................................. 8
3.2.2 Statistical Energy Analysis Model ....................................... 93.2.3 PAIN Model ......................................................... 13
RESULTS
4.1
4.2
4.3
4.4
4.5
4.6
4.7
4.8
4.9
4.10
4.11
4.12
oo° ..... o,oo °,o,°°°o°o°o,o°°o°°o,oooo,ooo,°°°ooo.o°o°o°,,o° ,.° o.°° o°ooo 15
Finite Element Analysis Results CFY 1987) ..................................... 15
Finite Element Model Improvements (FY 1988) ................................. 20
Finite Element Analysis Results (FY 1988) ..................................... 22
Statistical Energy Analysis Results (FY 1987) ................................... 33
Review of SEA Procedures for Low-Frequency Problems ........................ 38
4.5.1 Acoustic Space Modal Density Estimate ................................ 38
4.5.2 Structure Modal Density Estimates .................................... 39
4.5.3 Use of Finite Element Models to Verify SEA Parameters ................. 39
SEAM TM Code Improvements ................................................ 41
SEA Low-Frequency Model Improvements ..................................... 43
4.7.1 Input Power Estimate ................................................. 43
4.7.2 Modal Density Estimates ............................................. 44
4.7.3 Strut Connection Modeling ............................................ 44
4.7.4 Strut Loss Factor .................................................... 46
Midfrequency Model Improvements ........................................... 47
Statistical Energy Analysis Results (FY 1988) ................................... 49
PAIN Analysis Resu]ts (FY 1987) .............................................. 55
PAIN Analysis Results (FY 1988) .............................................. 57
PAIN Improvements and Feasibility Studies ..................................... 62
°,o
111
PRECEDING PAGE BLANK NOT FILMED
5.0
eag_
CONCLUSIONS AND RECOMMENDATIONS ..................................... 65
5.1 Finite Element Analysis ....................................................... 65
5.2 Statistical Energy Analysis .................................................... 66
5.3 PAIN Analysis ............................................................... 66
m
===]
J
Z[]
z
III
|mm
iII
!Ei
1!
m,BIW
III
m
m
tll
w=
I
R _;l
!
n •
el -
Ill
iv
wl, •|
|
w |
!-=
i
L
L
m
E ¸J
RI
Table
LIST OF TABLES
r_ag_
727 Demonstrator Airplane Ground Vibration Test (GVT) Shake Conditions ..... 5
727 Demonstrator Airplane Flight Conditions ................................. 7
Modeling Time and Cost Comparison for 727 Demonstrator Airplane FiniteElement Models .......................................................... 33
L
w
l
u
F
vi
wl
i
U
ii
imnl
[]iimlr
iNI
!1
m
Ii
il
Iii
|_ J
Ii
i
RI
ill
m
il
m _i
El -"
Iil -
Iw
M
m
w
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
LIST OF" FIGURES
Ea_
Prediction Program Comparisons and Comments ............................. 1
727 Demonstrator Airplane--General Arrangement ........................... 6
BS 1010 Microphone and Accelerometer Locations ............................ 7
727 Demonstrator Airplane--Finite Element Model Finite Element Method ...... 8
GE36 Demonstrator Engine Finite Element Model ........................... 10
25-Element Low-Frequency SEAM TM Model (20- to 100-Hz Range) ............ 11
100-Element Midfrequency SEAM TM Model (100-to 400-Hz Range) ............ 12
Propeller Aircraft Interior Noise Model ..................................... 14
727 Demonstrator Airplane Predicted Versus Measured Frame Mode
Shape at BS 1010--21.5 Hz ................................................ 15
727 Demonstrator Airplane GVT--Roving Microphone Data .................. 16
727 Demonstrator Airplane GVT Accelerometer Data Frame Versus
Ceiling Panel Response at Stringer 4 ........................................ 17
Finite Element Prediction Versus 727/GE36 Demonstrator Test;
GVT--Front Mount Vertical Shake ......................................... 18
Finite Element Prediction Versus 727/GE36 Demonstrator Test;
GVT--A.ft Lower Mount Lateral Shake ..................................... 20
Original and Revised Ceiling Panel Weight Distribution ....................... 21
Original and Revised Hat Rack Weight Distribution .......................... 21
Acoustic Mesh Typical Cross Section ....................................... 23
727 Demonstrator Airplane--Reduced Finite Element Models ................. 24
Finite Element Prediction Versus 727/GE36 Demonstrator Test;
Ground Vibration Test--Front Mount Vertical Shake ......................... 25
Longitudinal Variations Around BS 1010--GE Side Window Seat
Microphone Prediction .................................................... 26
Finite Element Prediction Versus 727/GE36 Demonstrator Test;
Ground Vibration Test--Aft Lower Mount Lateral Shake ..................... 27
Finite Element Predictions Versus 727/GE36 Demonstrator Test;
Ground Vibration Test--Front Mount Vertical Shake ......................... 28
Finite Element Predictions Versus 727/GE36 Demonstrator Test;
Ground Vibration Test--Aft Lower Mount Fore-Aft Shake .................... 29
Finite Element Predictions Versus 727/GE36 Demonstrator Test;
Ground Vibration Test--Aft Lower Mount Vertical Shake ..................... 30
Finite Element Predictions Versus 727/GE36 Demonstrator Test;
Ground Vibration Test--Aft Lower Mount Lateral Shake ..................... 32
SEA2kfr,_ Predicted Levels Versus 727 Demonstrator Test Data:
GVT--Forward Mount Lateral Shake ....................................... 34
PRECEDING PAGE
vii
BLANK NOT FILMED
I
i2Je,
26
27
28
29
3O
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
5O
Lae,
Low-Frequency SEAM TM Model Predicted Levels Versus 727 Demonstrator
Test Data; GVT--Forward Mount Vertical Shake ............................. 35
Midfrequency SEAM TM Model Predicted Levels Versus 727 Demonstrator
Test Data; GVT--Forward Mount Vertical Shake ............................. 37
Asymptotic Modal Densities for the Low-Frequency SEA Model ............... 40
Input Power Comparisons .................................................. 42
Response Predictions for the Low-Frequency Baseline Model
Using Measured Input Power--Vertical Shake ................................ 43
Composite Strut Modal Density Based on Spacing Between Peaksin the Measured Vertical Shake Input Power ................................... 45
Instantaneous Modal Density for the Cabin Acoustic Space .................... 46
Comparison of Modal Density Prediction Procedures--Interior
Acoustic Space ............................................................ 47
Midfrequency SEAM Model Variable Panel Density ........................... 48
Response Predictions for the Low-Frequency Improved Model--Vertical Shake ............................................................ 50
Response Predictions for the Low-Frequency Improved Model
Using Measured Input Power Vertical Shake ................................. 51
Response Predictions for the Low-Frequency Improved Model--Fore-Aft Shake ............................................................ 52
Response Predictions for the Low-Frequency Improved Model
Using Measured Input Power Fore-Aft Shake ................................. 53
Response Predictions, Midfrequency Model Side-of-Body Shake,
Improved Model .......................................................... 54
Measured Minus Predicted SPL Using Impi'oved Midfrequency Model .......... 55
Power Flow Diagram for Elements of PAINUDF Model ....................... 56
Measured Versus PAINUDF Predicted Levels ................................ 57
Average of Seat Measurements Versus PAINUDF Space Average Prediction ..... 58
Propeller Wavefronts for Blade Downsweep (Forward Rotor) and Blade
Upsweep (Aft Rotor) at _ = 105 ° 59
Propeller Wavefronts for Blade Downsweep (Forward Rotor) and Blade
Upsweep (Aft Rotor) at O _ 800 59
Measured and PAINUDF Predicted Levels for Open Panel and
Conical Empennage Models ................................................. 60Measured and PAINUDF Predicted Levels Phase Sensitivity Study ............. 60
Comparison of Propeller Field Phase Contours ANOPP Phase Versus
Dummy Phase ..... ., ---,. ,.. .... . ............. . .......................... 61Measured and PAINUDF Predicted Levels PAINUDF Sensitivity
Study to Excitation Models (Mach -- 0.80 at 35 000 ft) ........................ 62
Example of Sidewall Layer Configurations ................................... 63
B,I
Vlll
M
m
f
M
mJ
n
N
l
j i
=__-.a
m
- i!
i
|. , =
!B
m
I IB
1.0 SUMMARY
L_
=
t_
_J
N
This document summarizes a 2-year effort to evaluate the application of selected analysis techniques to
low-frequency cabin noise associated with advanced propeller engine installations. Work was funded by
a NASA contract, NAS1-18027.
Three design analysis techniques were chosen for evaluation including finite element analysis, statistical
energy analysis (SEA), and a power flow method using elements of SEA (computer program Propeller
Aircraft Interior Noise (PAIN)). An overview of the three procedures is provided in figure 1. Data from
tests of a 727 airplane (modified to accept a propeller engine) were used to compare with predictions.
Comparisons of predicted and measured levels at the end of the first year's effort showed reasonable
agreement leading to the conclusion that each technique had value for propeller engine noise predictions
on large commercial transports. However, variations in agreement were large enough to remain cautious
and led to recommendations for further work with each technique. This recommended work was accom-
plished in the second year's effort.
Assessment of the second year's results leads to the conclusion that the selected techniques can accu-
rately predict trends and can be useful to a designer, but that absolute level predictions remain unreli-
able due to complexity of the aircraft structure and low modal densities. The extremely complex nature
of the modified 727 demonstrator airplane may be largely responsible for this conclusion. It would be
worthwhile to apply these techniques to more conventional aircraft structure cases.
= _
,l=w
L
m
Frequency range
(this study)
Calculations
Advantages
Disadvantages
Use
FEM
Rotor
Modes, display, SPL
versus position and
frequency
Good detail versus
frequency, position
Expensive, hard
to change model
Detailed predictions
design work for
exi_ng or new
airplanes
SEAM
Low-frequency model High-frequency model
Rotor and rotor Rotor harmonics and BPF
harmonics ( < 100 Hz)
Power, modal densities coupling loss
factors, space average display and SPL
Inexpensive/quick model
changes/power flow analysis/variance estimates
Lack of spatial details/highestimate vadance for low modal
density cases
Power flow analysis for existing
airplanes
Preliminary design
PAIN
BPF
Power structural modes
space average SPL
Moderate exp
Good trim definition
lack of spatial details,excitation field
complexity
Preliminan/design
Figure 1. Prediction Program Comparisons and CommentsUgO 196RI-4
w
i
i
N
i
mM
(In !
i
m
u
Ig
U
i
ug
n
i
ii
J
i
Lu
= =
r
m
m
m
----Sm
Z-
L_
_z
2.0 INTRODUCTION
The primary sources of cabin noise on advanced propeller-powered aircraft are low- to mid-frequency
(i.e., below 500 Hz) structure borne noise (caused by engine unbalances) and engine-radiated propeller
tones. Conventional sound proofing treatments, such as damping tapes or fiberglass blankets, are not
very effective in this frequency range. Also, most low- to mid-frequency design tools tend to be based on
idealized cylindrical models of the aircraft fuselage, which ignore effects of tapered empennage sections,
pressure bulkheads, floors, etc. Effective reduction of low- to mid-frequency noise requires the develop-
ment of improved design analysis tools. These improved tools will lead to a better understanding of the
mechanisms involved and provide guidance for developing new suppression concepts.
The objective of the FY 1987 task under contract NAS 1-18027 was to evaluate the ability of three analysis
techniques (contained in existing computer programs) to predict cabin noise and vibration. Specific
steps of this effort were as follows:
1. Predict cabin noise, together with vibration levels of the floor, cabin sidewall, pressure bulkhead,
empennage, and strut at rotor (once per revolution) and propeller-blade-passage frequencies.
. Compare predicted levels with measured levels from a 727 demonstrator aircraft modified to accept
a GE36 counter-rotating propeller engine. Make predictions and comparisons for ground vibration
test and inflight conditions.
3. Determine strengths and shortcomings of prediction program procedures and/or modeling tech-
niques and identify potential means for improvements.
The objectives of the FY 1988 task were to incorporate recommended improvements, repredict noise
and vibration levels, compare the new predictions to measured levels, assess results, and make recom-
mendations for future use of these procedures.
L_=
k
t =
PRECEDING PAGE BLANK NOT FILMED
4
m
I
Um
m
m
B
l
m
mgl
J
i
mE
B
JI
m
!
nBm
m
Iim
m
|
Rii
|
|:=:_
k :=
IBw
w
3.0 APPROACH
While the NASA contract focused on predictions and comparisons with test data, preliminary work
including 727 test descriptions and analysis model development is also included in this report.
3.1 NOISE AND VIBRATION TESTS
Noise and vibration testing on a 727 aircraft, modified for installation of a GE36 counter-rotating pro-
peller engine, provided the database with which to compare predictions. During ground vibration tests,
before engine installation, the strut was excited by a shaker to determine airframe and cabin response to
engine mount vibration. Response to shaker inputs at each engine mount location was recorded. In addi-
tion, side-of-body shake testing was performed on the fin, empennage, and aft passenger cabin to simu-
late high-level acoustic loading from the propellers. Ground vibration test conditions are given in table 1.
The 727 was configured with an acoustic test arena aft of body station 890. Standard 727 interior trim
and four passenger seat rows were installed in this area. General layout of the aircraft is shown in
figure 2.
Instrumentation locations were reviewed to ensure adequate data for comparison with predicted levels.
Cabin microphones were placed in the acoustic arena to monitor noise at passenger seated and standing
locations as well as next to interior vibration sensors (fig. 3). Also shown are accelerometers positioned in
the passenger cabin to monitor fuselage sidewall, floor, and cabin interior trim response.
Accelerometers were placed on the pressure bulkhead, empennage, and strut to evaluate the transmis-
sion path to the passenger cabin. A total of 128 response channels were measured for each ground vibra-
tion test condition.
Due to data channel limitations during flight testing, fuselage and cabin accelerometer numbers were
reduced. Further reductions were required to add flush mounted exterior microphones for the purpose
of measuring propeller noise excitation levels. Different cruise and low-power flight conditions were
completed (table 2).
Table 1. 727 Demonstrator Airplane Ground Vibration Test (GVT) Shake Conditions
Side-of-body shakes, 10 to 400 Hz Engine mount shakes, 2 to 200 Hz
BS 1160 WL 320
BS 1270 WL 190
BS 1300 WL 250
BS 1310 WL 340
B,S1400 WL 400
BS 1050 WL 23O
Front mount vertical
Front mount lateral
Aft upper mount vefdcal
Aft upper mount lateral
Aft upper mount fore to aftAft lower mount vertical
Aft lower mount lateral
Aft lower mount fore to aft
w
=
7 ¸ PRECEDING PAGE BLANK NOT FILMED
UlClOlg6RI-1
5
i
i
!
_t
v ,-_.L.J
Plan _qew
Right Side View
=mm
!
E
i
I
N
g
Jm
Ii
B
'II
Figure 2.
engine
nil finRear _qew
727 Demonstrator Airplane - General ArrangementUg01_R!-5
EE
i
!m
i
t
|
|
!
I ,
T_
E
_-!!i!_;-
GE36 Side
S-12
O
O
O
S-1
0 0 0
0
_4
==
z i
Legend:X Accelerometer0 Microphones
S-30
Figure 3. BS 1010 Microphone and Acce/erometer Locations
Table 2. 727 Demonstrator Airplane Flight Conditions
_U_1_-6
E
L
L
mkii'
Mach No,
0.80
0.72
0.42
0.25
Altitude, ft Blade-passagefrequency, Hz
35000 169
35000 170
10000 180
4200 180
UGO1Q6R1.;
3.2 ANALYSIS MODEL DE .VELOPMENT
Three design analysis techniques were chosen for evaluation including finite element analysis, statistical
energy, analysis (SEA), and a power flow method using elements of SEA (computer program PAIN). An
existing general application SEA computer program, SEAM TM, was chosen for evaluation based in part
on past successful applications to helicopter, ship, and automobile problems. The developer of
SEAM TM, Cambridge Collaborative, Inc., was retained as a subcontractor for this effort. Computer
7
program PAIN was specifically developed for propeller interior noise predictions. L. D. Pope, a co-
author of PAIN was also retained as a subcontractor.
3.2.1 FINITE ELEMENT MODEL
As a starting point for finite element analysis, an existing 727 aircraft finite element model suitable for
loads analysis was selected. Changes to model the modified structure (to accommodate the propeller
engine), the mass of the interior trim and systems, and an interior cabin acoustic volume were then incor-
porated. The model is shown in figure 4. Detailed finite element models of the fuselage containing the
acoustic arena and structural elements along the transmission path, such as the pressure bulkhead, em-
pennage, vertical fin, and strut, were used. In the fuselage section, body frames, floor beams, stringers,
and skin were all individually modeled. Body frames and floor beams were represented as beams, string-
ers as rods, and skin as membrane plates. In the test arena section, nodes were located at each frame and
Horizontal tailbeam model
STA 1183IIBq B
Ruddermass model
Wing beam element STA 870 •(typical) ,P
Center and left
engine massmodels"
mass (typicaT)_ Detailed stiffness and mass
model of the aft fuselage
I
Hl
[]
![]
Jim
a
mI
n
m
II
m
Fuselage beam element(typical)
Wing concentrated mass(typical)
• Right engine was not included in the ground vibration test (GVT) configuration.
Figure 4. 727 Demonstrator Airp/ane-Finite E/ement Mode/Finite E/ement Method
8
_U_I_R!-7
U
lB
m
m
II
|
L-
r
l
stringer intersection. Nodes of the cabin acoustic model interfaced with existing sidewall, floor, and
bulkhead structural nodes.
The model was completed by including beam models of the forward fuselage, wings, and horizontal tail.
The complete aircraft model was generated because the contribution of gross airplane modes to trans-
mission of low-frequency rotor noise was unknown, but potentially important.
Due to model size and existing node spacing, it was decided to limit predictions to the rotor frequency
range. The fuselage nodes were at each frame and stringer intersection, which would have been too
coarse a grid to predict skin panel modes that occur in the propeller-blade-passage frequency range
(-200 Hz). Because no dynamic nodes of the interior trim were expected below 30 Hz, the interior trim
was modeled by lumped mass only. Seat absorption was expected to be minimal below 30 Hz; therefore,
seats were also represented by lumped mass only. Additional tap tests of interior trim surfaces and rov-
ing microphone measurements were made around seats to validate these assumptions.
For inflight calculations, an engine model was also developed consisting of a beam element representa-
tion of the core engine and a shell model of the propulsion section (fig. 5).
Predictions using this model were made for unit force inputs and compared to measured ground vibra-
tion test (GVT) and flight transfer functions.
3.2.2 STATISTICAL ENERGY ANALYSIS MODEL
The primary task within the SEA procedure is to formulate power balance expressions for the substruc-
tures (energy storing mode groups) of the model. The substructures are typically simple representations
(e.g., plates, beams, and acoustic volumes) for sections of the modeled system. Physical information de-
scribing the system is used to define the substructure properties such as modal density (the average
number of resonances per unit frequency), characteristic wave number (a quantity directly related to the
average wavelength of a mode), characteristic impedance (the ratio of the maximum force to the maxi-
mum velocity for a mode of vibration), and coupling loss factor (a parameter controlling the average flow
of energy between two groups of modes). The power expressions are then solved for the modal energies
within the substructures. From the modal energies, the response variables of interest for the specified
substructures are determined. The input excitation for a SEA model is applied as the power input to a
subsystem of modes.
Because of the smaller model size requirements for SEA, it was decided to use this technique to predict
levels for both the rotor and propeller-blade-passage frequency ranges. The first model developed as
part of the SEA effort was intended to predict the structural and acoustic response near the propeller
tone frequency (100- to 400-Hz frequency range). This model consisted of approximately 400 SEAM TM
type subsystems and more than 2000 junctions, which, up to this point, was the largest SEAM TM model
developed by Cambridge Collaborative. This model contained elements representing the composite
bending and inplane activity of frame and stringer stiffened structure, elements that represented skin
panels without stiffeners (bending and inplane), trim elements, cabin acoustic space elements, and
9
f/
/I
//
//
/ Front rnoun_I
lI
Figure 5. GE36 Demonstrator Engine Finite Element Model_U_I_RI-g
acoustic elements that represented the space between the fuselage sidewall and trim panel. Initial pre-
dictions with this model indicated good agreement at the drive point, but overprediction at points away
from the drive point. Cambridge Collaborative concluded that the inability of the model to predict the
measured rolloff in response was due to the inclusion of inplane subsystems in the model, which lead to
overcoupling between subsystems.
In addition, it was found that the small element sizes resulted in modal densities that were too low based
on Cambridge Collaborative's past experience. As a result of these efforts, new models for rotor and
propeller tone predictions were generated (figs. 6 and 7) using larger substructure elements and elimi-
nating inplane activity. Data from manufacturing drawings were used to generate separate models in
each frequency range using basic pipe, plate beam, and airspace elements available in SEAM TM.
I
Z
m
W
Q
W
mil
10 i
i
z
m
L
r_
_J
._ :z
18 WINGORG
25 OUTA
11 REP
3 AFTFUSP
4 AFTFUSA
2 VI"C
17 WINGIRC
19 WINGILC
15 STRUTLB
12 LEP
16 PBULKC
5 MIDFUSP
6 MIDFUSA
21 MIDFLC
9 FWDFUSP
10 FWDFUSA
23 FWDFLC
Elements:
25 pipes, plates, and acoustic cavities
21 structural
4 acoustic
Figure 6.
7 WINGFUSP
8 WINGFUSA
22 WINGFLC
24 KBEAMC20 WINGOLC
25-Element Low-Frequency SEAMTM Model (20- to lO0-Hz Range)_U_1_R!-lO
The low-frequency model was developed to predict the interior acoustic response to mechanical excita-
tion of the engine strut. The transmission of vibration from the unducted fan (UDF) engine caused by
fan rotor imbalance was expected to dominate cabin noise within the low-frequency range (20 to 100 Hz).
As with the finite element model, the low-frequency SEA model (for rotor frequencies) did not account
for cabin trim or seats (other than mass). The substructure elements shown in figure 6 were selected
based on natural boundaries between areas of different structural properties or construction. Rather
than the detailed finite element modeling of skin, stringers, and frames, composite bending stiffnesses of
the plate or pipe elements were calculated to include the effects of the aircraft structural elements as
described below.
The material properties required for the SEA model are density., longitudinal wavespeed, shear wa-
vespeed, and damping loss factor. The substructures used in the low-frequency model consist of
composite bending modes. Effective material properties for these substructures were obtained from a
composite bending calculation that includes the mass and stiffness of frames and stringers and the mass
of fuselage sidewall trim panels. The properties and thickness of a homogeneous structure were set to
obtain the same bending wavespeed and surface density. Two approaches can be used: (1) an equivalent
material density approach (or stiffness thickness approach) in which the thickness of the structure is
11
(fourplates)
Roofcomposite
bending 2
/" space _ Side II_wer ar__ustic _ composite Ikt
I ;p=-'....... .I =!i ..X t fyloor acoustic] _'_"'_' _4
_,,__ Baggagecompartment Rear Viewcompositebending
Rear Vlew
torqueboxesand frames
frameplates
Aftfuselage
I _" At_iihgrnentpipeelements
i
II
i
Figure 7. lO0-Element Midfrequency SEAM TM Mode/(100- to 400-Hz Range)_U_1_R1-11
il
modified to give the correct bending wavespeed and the density is modified to give the correct density
per unit area and (2) an equivalent material density approach (or mass thickness approach)in which the
thickness of the structure is modified to give the correct density per unit area and the longitudinal wa-
vespeed is modified to give the correct bending wavespeed. For the low-frequency model the equivalent
material density approach was used for all shell or pipe substructures and the equivalent wavespeed
approach was used for plate structures. Use of the equivalent material density approach for shell or pipe
substructures allowed the longitudinal wavespeed to be held constant so that the ring frequency of the
substructure was unchanged.
I
m
I
l
i
The mid-frequency SEA model (for propeller frequencies-fig. 7) differed from the low-frequency model
in substructure element size, composition, and handling of the trim and seats. The substructure element
boundaries were chosen to reflect finer variations in construction (both axially and eircumferentially)
that were felt might effect response in this frequency range. Composite bending stiffnesses as well as skin
membrane stiffnesses were included in the fuselage plate elements. Because independent dynamic
modes of the interior trim were expected in this frequency range, the trim was modeled separately as
plates. Acoustic spaces between the skin and trim panels were also included. The effect of seats was
m===l
z
J
J
II
mm
m
12!
II
i
r-.--
=
K..d/
w
_2
w
_I
i •
w
included by separately modeling acoustic volumes above and below seatback levels. Coupling between
acoustic volumes below seatback level was adjusted to account for seat blockage. Material properties
were generated in the same fashion as for the low-frequency model.
Model predictions for both cases were again made for unit force inputs and compared to measured GVT
and flight transfer functions.
3.2.3 PAIN MODEL
The Propeller Aircraft Interior Noise (PAIN) prediction method (theory and computer implementation)
was previously developed by L. D. Pope, et al., under the auspices of NASA. The PAIN model estimates
the space average sound level in an airplane cabin resulting from an excitation field incident on the exte-
rior fuselage (fig. 8A). The excitation field consists of the pressure time history for a wing-mounted rotat-
ing propeller as defined by a grid of points on the fuselage surface.
The computer program PAIN was used to predict inflight cabin noise levels at the propeller-blade-
passage frequency. As mentioned above, PAIN was originally developed for general aviation aircraft
with wing-mounted propeller engines. In this form, only one uniform fuselage structural description was
allowed. In contrast, the 727 demonstrator aircraft had an aft fuselage-mounted propeller engine, and
most of the propeller tone noise fell on the empennage rather than the passenger cabin fuselage. There-
fore, before PAIN could be used, a significant modification for 727 predictions was completed by refor-
mulating the problem to allow representation of both the passenger cabin fuselage and empennage
structures (fig. 8B). The 727 demonstrator airplane empennage consists of a very stiff sidewall supported
by torque boxes at the top and bottom. To model this behavior, the empennage was represented by open,
curved panels.
The capability to model both fuselage and empennage structures is contained in computer program
Propeller Aircraft Interior Noise Unducted Fan (PAINUDF). The empennage element allows for the
region on which the excitation field acts to differ in both construction and location from the fuselage
region inside which cabin noise estimates are desired. Additionally, a junction between the two regions is
defined by a structurally significant (but nonacoustically radiating) bulkhead.
With PAIN, the fuselage sidewall is modeled in a simpler manner than with finite element modeling. The
sidewall was represented as a cylinder with an equivalent skin thickness based on skin dimensions and
smeared frame and stringer properties. The floor was modeled in a similar manner. The interior trim
was modeled as an airgap with absorption and a limp mass trim. The interior airspace was modeled as
one uniform volume. Data from manufacturing drawings were used to determine equivalent skin and
floor panel thicknesses, insulation properties, and trim mass.
Data from flush-mounted microphones were then used to describe excitation forces, allowing inflight aft
cabin noise level predictions to be made. Changes to the computer code as well as predictions and com-
parisons were made under the NASA contract.
13
W
I
Trim insula_on and lining
m
Ii
mII
l
Structurally integral floor 2O
mloI 0 I
-10 "20 To " BPF-1
(a) PAIN Mode/for Wing-Mounted Englne
Talloone and passenger cabinstructure markedly different ._
\ \
Most prop-tone energy falls
on this surface
D
B
!
m
W
l
i
Note: Reformulate problem to allow twoseparate coupled structures.
(b) 727/GE36 Demonstrator Airplane PAIN Mode/
Figure 8. Propeller Aircraft Interior Noise Mode/
mmm
U
l
i
[]
14 mmm
i
B
!1
4.0 RESULTS
F
w
E
w_=w
4.1 FINITE ELEMENT ANALYSIS RESULTS (FY 1987)
Predictions were made to encompass the rotor frequency range ( < 30 Hz) and were compared to GVT
and flight test conditions. A modal solution was used, meaning a limited number of structural and
acoustic modes were computed and used to generate forced response predictions. In this case, 150 struc-
tural modes (0 to 73 Hz) and 100 acoustic modes (0 to 230 Hz) were used.
Inspection of predicted mode shapes showed gross airplane modes involving the wings, tail, and fuselage
from 0 to 30 Hz; local frame and floor bending modes occurred above 20 Hz. Comparison with test data
showed good agreement. Below 20 Hz, predicted gross airplane mode shapes matched results from mea-
sured data, with predicted resonant frequencies typically within 5% of those measured. From 20 to 30
Hz, fuselage modes occur in which the frame exhibits a maximum of three half-waves around the passen-
ger cabin and the floor one half-wave across the cabin. Measured data showed similar response charac-
teristics (fig. 9). The measured data in figure 9 includes both frame and floor bending plus rigid body
translation associated with gross airframe bending. Since only radial accelerometers were used during
testing, the local bending and rigid translation responses could not be separated. It was recommended
that future rotor frequency shake tests use biaxial accelerometers. A few modes related to strut response
Prediction Test
a
21.501 Hz 21.48 Hz
Figure 9. 727 Demonstrator Airplane Predicted Versus Measured Frame ModeShape at BS 1010-21.5 Hz
UgO196RI-14
M
t=,,*
w
15
g
were also identified, but either did not appear in measured data or were shifted in frequency by a signifi-
cant amount. The cause of these modes was investigated as part of the FY 1988 task.
One cabin acoustic mode was predicted in the frequency range below 30 Hz. At 23 Hz a longitudinal
mode was predicted and represented by a standing half-wave between the pressure bulkhead and a par-
tition at body station 890 used to separate the test arena from the rest of the airplane. Cross-cabin modes
occurred at more than 50 Hz.
As stated in the Approach section, additional tests were performed to validate seat and trim modeling
assumptions for the rotor frequency range. Noise measurements made at several locations around one
seat during a body shake condition showed only small variations at less than 50 Hz (fig. 10). Trim and
ceiling panel tap tests showed no modes in the rotor frequency range. Shake test data showed that the
ceiling panel followed the primary structure response to slightly more than 30 Hz (fig. 11).
Example comparisons of test with analysis for the strut, pressure bulkhead, fuselage sidewall, floor, and
acoustic space are given in figures 12A through 12F for one GVT condition. Predicted peak levels fell
SLP, dB
T10
±
_j,,,
•! I
i;:J_]
l
li:
.9
I
o
Figure 10.
I
II
i
li
III
U
Ill
ml
|mi
,°_.w*
,-_r m;,;?- =, -.,,, ,, /;_'_. i..:._._'_ I'/_"_,, "",-;.-_,/[
2f
10 2O
Frequency, Hz
727 Demonstrator Airp/ane GVT-Rovlng Microphone Data
u| 1 I i L
3O 4O 5O
aiu
16MmII
i
'_,v
U
F
w
L
Y_.....
w
O0
0Z
70
65
60
50
45
4O
35
3O
25
20
15
i
..., • ;
.:, .o ! e
: p,_.•, I
ii:1It
.I
o, iO,-.: :!
iI •
:; "'I
Ir ,
II- I
.o
t
"l
• J\ "
I tt tI tI I
I II t
,t"'*.
Ceilingpanel , ",
at slTinger 4 ---_./ ....'....
io
| : : tI : "'_• J
""_ .__
I I-,.
I I !
-" it: I !
: I !
',%
,,-"'-'.'_...X I '_ _''
,,,. ,,.._°_,
Frame at
stringer 4
I_o _ I _ I _ I _ I , I , I10 20 30 40 50 60 70 80
Frequency, Hz
Figure 11. 727 Demonstrator Airplane GVT Accelerometer Data Frame VersusCeiling Panel Response at Stringer 4
9-Ug_196Rl-16
within 10 dB of the measured levels with frequency shifts up to 1.5 Hz. It was felt that agreement could be
improved with the additional modeling studies and recommendations discussed below.
Figure 12A compares measured and predicted strut response for the vertical strut shake condition. Gen-
erally, levels were overpredicted at less than 19 Hz and underpredicted at more than 19 Hz. The mode
predicted at 14.5 Hz and shown shifted in the test data to near 16 Hz illustrates some of the questions
about strut modeling that were recommended for investigation.
Figure 12B shows radial response of a stringer at the intersection of the fo_'ard strut spar and the pres-
sure bulkhead. Agreement was very good although the trend of overprediction at low frequencies and
17
80
70 ,-
6oh& ;.,"40--
legend:- - - GVT data
.... FEM prediction
•"_._/.- ... .'..
30--
2O
10
I I I I% 14 18 = 26 _3oFrequency. Hz
(a) Front Mount Vertical Drive Point
100
9O8O
70
z
Legend:m - -- GVT data
.... FEM prediction
60 m
50 m
0 ,,:,_ L ,,. I,_. -,,_..._.r\ // ,.-f,, \..,._/'" -' ",X-j , ,' , ,
20 _- " "
10
o I 1 1 110 14 18 22 26
Frequency, Hz(¢) BS 1183 Aft Pressure Bulkhead
Right Side Fore-Aft Responae
m 100
9O80
70
t_
_oZ
•\\
3O
Legend:- -- GVT data.... FEM prediction
6O
5O
4O
10
Oo
w
• ° •
] I J /_14 18 22 26
Frequency. Hz3O
(e) BS 1010 Right Side Outboard SeatTrack Vertical Floor Response
Figure 12.
100
"_ 90L Legend:
/ --- av'rdata80 _.-- .... FEM prediction/
!:f6O ° ._.
• "f.',.... ./'_-_.../ .,',.._
N _J "', "'
lO ',Jb-z ol, l I r I
10 i4 i8 22 26Frequency. Hz
(b) BS 1183 Right Side Stringer 12Radial Frame Response
30
m 100
i 908O
70
_ 5O
_ 4O30
_ 2010Z
!
Legend:--- GVT data.... FEM prediction
m
-.-- ,_., ,, ,v , .-'X'4.,,f,,., - .. _ ." - ,- k/
[ I 1 r010 ' 14 18 22 26
Frequency, HZ
(d) BS 1010 Right Side Stringer 12Radial Frame Response
3O
1!f Legend:
- - - GV'r"data.... FEM prediction
70
z
6O B
'_.'," -,, / \X /40-- {/ \ / ", . .1130
2O-- J AJ,
10m
, I 1 I t(3
_10 14 18 22 26 30Frequency, Hz
(1')BS 1010 Right Side Wlndow Seat Microphone
Finite Element Prediction Versus 727/GE36 Demonstrator Test;GVT-Front Mount Vertical Shake
]8
i
B
mii
|
|
t
I
|
imIB
lib
;
t
I
ml
!
|
|
E
_m
F
=
_i___E
underprediction at higher frequencies seen in the strut response was repeated. The drop out at 22.5 Hz
appeared to be related to a spurious strut mode.
Figure 12C compares predicted and measured pressure bulkhead response. Predicted levels were very
close to test with a 1.5-Hz shift in frequency believed to be related to the above noted strut modes.
Figures 12D, E, and F show body frame, floor, and acoustic response comparisons at body station 1010,
approximately the middle of the test arena. While comparisons are reasonable, some of the largest varia-
tions between prediction and test were noted at floor and frame locations. Also, the differences varied
considerably between individual measurement points. It was believed that inappropriate lumping of in-
terior masses might have shifted the predicted locations of modal node points, which could have a sub-
stantial effect when comparing individual points. At the beginning of the modeling effort, it was assumed
that the interior trim mass could be uniformly spread. From these results, it appeared that this assump-
tion was not valid. Interior trim and seats are in fact mounted at discrete locations with the baggage
racks and seats being cantilevered from the frames and floor beams. It was recommended that sensitivity
studies of mass representation and distribution be completed using the 727 demonstrator airplane mod-
el to define mass loading representation requirements.
Figure 12F compares predicted and measured noise levels at a passenger seat. The major trends were
predicted and good agreement at the peak response point (for the half-wave axial mode) was observed. It
was noted, however, that the predicted response peak was sharper and shifted to a higher frequency than
the data. A partition located at body station 890 to separate the test arena from the rest of the airplane
was assumed to be rigid for modeling purposes. Inspection of the data suggested that the partition was
providing some low-frequency absorption in the form of mechanical vibration. This tended to broaden
peak response and shift it to a lower frequency as seen in the test data.
In the initial finite element model, only the cabin airspace was represented. Previous analysis on simpli-
fied fuselage section models indicated that the cargo compartment airspace could have a significant
effect on passenger cabin acoustic response, particularly at acoustic resonances. It was recommended
that this be included in the full 727 demonstrator airplane model.
Comparisons of predicted and measured levels for other shake conditions generally led to the same con-
clusions as the vertical shake condition with the following one exception. For shake directions in the
plane of the strut, a predicted strut mode at 23 Hz coupled with the cabin acoustic mode at the same
frequency resulted in an overprediction of interior noise levels. Measured data showed a similar strut
mode with lower response near 28 Hz (fig. 13A). The resulting overprediction in noise is shown in figure
13B. Another factor possibly contributing to the overprediction at the acoustic resonance was the rigid
boundary condition assumed for the acoustic curtain.
Flight data predictions did not match well with measured levels. The model showed a great deal of sensi-
tivity to both input unbalance levels and phase between propellers. Unfortunately, measured response
data did not include phase information referenced to the propellers that made assessment of flight
19
m
Legend:..... Gv'r data
" ......... FEM predic'don
b-:-,'<:"....
Z
10
18 22Frequency. Hz
10 14
(e) Aft Lower Engine Mount Lateral Response
Legend:..... GVT data......... FEM prediction
70
:L ".'* "°%
G)
z
J= \ e #_
40_.-r.... , ........ ,.,,=, • _ .._ ....
'°mo -' I , I , I , I ,10 14 18 22 26 30
Frequency, Hz
(b) BS 1010 Loft Side Aisle Seat Microphone
Figure 13. Finite Element Prediction Versus 727/GE36 Demonstrator Test;GVT-Aft Lower Mount Lateral Shake
E)-UgO106R1-18
predictions difficult. Phase measurements referenced to propellers were recommended for any future
testing.
Agreement between predicted and GVT data showed the finite element modeling procedure to be a
valid tool for predicting low-frequency structural-acoustic behavior. Based on the above results, it was
recommended that studies of strut and acoustic curtain modeling and trim mass distribution be
completed.
Use of the finite element model was found to be time consuming and expensive. The sheer size of the
model made assembly and checkout a very lengthy process. Turnaround of results was also slow. It was
also recommended to systematically reduce the size and simplify the model to define the minimum mod-
el size requirements versus frequency. As stated previously, the entire aircraft was modeled because the
effect of gross airplane modes was not known.
4.2 FINITE ELEMENT MODEL IMPROVEMENTS (FY 1988)
The existing finite element model was refined by improving the mass distribution of interior nonstruc-
tural elements and improving the acoustic curtain representation. The propeller engine strut model was
checked for modeling errors due to spurious modes identified during 1987 work. Some errors were found
and corrected. Details are given below.
The initial finite element modeling procedure simply smeared the mass of interior trim and seats evenly
as shown for a ceiling panel example in figure 14A. In fact, this panel is supported only at discrete points
along its edges. The mass of the ceiling panel was redefined as concentrated masses at the attachment
points only (fig. 14B). The hat rack and seats were originally defined as lumped masses at the fuselage
2O
II
imI
J
R
II
u
m
tm
!J
n
u
rail
N
m
I
m
m
m
II
i
m
J
m!
Ii
U
L_
W
U
mw
W
F
z :
S-11 R
S-1
___- Tdm panel
, Y Y
S-I
_- Floor
_ Idealized Idealiz_lframe Oypic_)
(a)Orlglnal_illng PanelWeight Distribution (b)Revised CeilingPanelWeight Dls_ibutlon
Figure14. Odglnaland RevisedCeilingPanelWeight DistributionUgOI_RI-19
mold line and in the plane of the floor, respectively. Their masses were redefined as concentrated masses
located at the center of gravity. The hat rack change is portrayed in figure 15.
The test airplane configuration included a partition to separate the forward portion of the aircraft, which
was filled with test equipment from the aft acoustic test arena. This partition was constructed from a
honeycomb board with fiberglass and lead vinyl installed on both sides. Original modeling assumed this
S-1 S-1
S-8 L_S-8 L S-7_7 L
..... C'S-11 R 'f/'Wtl wt2 \_/"S-11 Rff./ I W'Zl w_2 T \_S-11 L
(a) Original Hat Rack Weight Distribution (b) Revised Hat Rack Weight Distribution
Figure 15. Original and Revised Hat Rack Weight DistributionUgO196R1-20
21
D
partition to be rigid, Predicted acoustic response showed the acoustic resonance to be 1 to 2 Hz more
than seen in the test. This may be the result of ignoring the effects of mechanical vibration in the initial
model. Therefore, the acoustic partition model was changed to a flexible partition using brick elements
to represent the mass and stiffness of the fiberglass and lead vinyl septum on the test arena side.
It was noted previously that strut modes were predicted that either did not occur ifi the measured results
or were at significantly different frequencies. It was suspected that the strut and empennage connection,
very complex for the modified demonstrator aircraft, might be in error. The model was compared to
drawings and a few changes were made with the net effect of increasing strut stiffness.
The above changes constituted the improved finite element model that was used to repredict ground
vibration test conditions. In addition, further studies were conducted by representing the cargo com-
partment airspace and by systematically reducing the size of the improved model. A finite element mesh
of the cargo compartment airspace was generated that interfaced with the lower sidewall and floor
structural nodes. The cabin acoustic model interfaced with the same floor nodes allowing coupling be-
tween the acoustic spaces (fig. 16).
As was noted previously, the size of the finite element model made its use time consuming and expensive.
Figure 17 shows model representations that were studied to define minimum model size requirements.
In the first case, the stiffness and mass substructures aft of BS 1183 were dynamically reduced to the
retained degrees of freedom at the aft pressure bulkhead and the engine attach points. In the second
case, the empennage and tail section aft of the strut rear spar were replaced by a beam model. Finally, a
representation that ignored structure aft of the strut rear spar and forward of the acoustic test arena wasconsidered.
4.3 FINITE ELEMENT ANALYSIS RESULTS (FY 1988)
Predictions made using the refined baseline model are given in figures 18 through 24. Figures 18 and 20
compare 1987 and 1988 predictions to GVT data. In this case, results are plotted up to 40 Hz in order to
better understand response characteristics around 30 Hz. Overall, the refined finite element model
yields reasonable predictions of the GVT response. The most notable improvement in predictions asso-
ciated with this model is the acoustic response at the acoustic resonance for lateral and axial shake
conditions.
Figure 18 compares predictions to test for a front mount vertical shake condition at several of the posi-
tions discussed previously. At the drive point (fig. 18A) the new prediction follows the trends seen in the
GV'E As a direct result of the improved strut and bulkhead modeling, a closer match of prediction and
test is obtained at less than 24 Hz. At more than 24 Hz, the model still underpredicts the data.
Figures 18B, C, and D show body frame, floor, and acoustic response comparisons at approximately the
middle of the test arena (BS 1010). It was this area that was noted to have some of the largest variations
between prediction and test when evaluating 1987 results. This led to the review and improvement of the
interior trim mass distribution in the refined model. Although the refined model yieIds reasonable
22
i
mi
|
|m
mU
|I
I
!
i
|
I
m
U
i
i
|
g
i
L_
i :
72037202 7201 17202
172O4
172O5
= :
_---m_
w
--=-=4
=__ibm
W
f--
PassengerCabin
7208
7209
7210 _41 ........
7211
7212
7213 .....
7214
7ZG;
7_83
17208
• . _72
11 17211
.............. 1724g'1"7_._17212
,_,_ ,_ ............. .._._--- - 17213
17214
''_^ "_ "_=" _¢ _?_ 17215
721E 17216
7217 J72177233 7232 7231 17232 17233
Floor 107233 107232 107231 117233 117232
721_ II .............. _ ..... I '_ 1721g
c,,0o i \ \\7,=/ / \ \ 7,,=,
7""" 7:74 17275 _ "727__ \I.....
7228 7229 723O 17229
Figure 16. Acoustic Mesh Typical Cross Sectiong-UgO196R 1-21
predictions of trends and levels, there is no clear improvement over 1987 results. In fact, the response
prediction for the window seat microphone appears worse. However, at this frequency the model is very
sensitive to axial position because of the node line of the first acoustic mode. Changing the acoustic
partition properties has moved the predicted response node position slightly. Figure 19 compares the
predicted noise levels at three sequential axial positions near the GVT measurement point. As can be
seen, there is a 20 dB difference at approximately 22 Hz in the three predictions with the prediction for
BS 990 closely matching the data at BS 1010. This illustrates one of the difficulties in comparing the test
data to prediction on a point-to-point basis.
23
d
|Reduced model A- Reduced model B- Reduoed model C-
dynamic reduc_on of aft-end shell structure aft-end beam model passenger cabin and strut model
Figure 17. 727 Demonstrator A/rp/ane-- Reduced Fin�re E/ement Mode/sg,UQO106RI-22
For the lateral shake condition, the refined model did a better job than the original model based on
improvements to the strut and acoustic curtain models. At the drive point, the predicted response
follows the measured response trend at less than 24 Hz (fig. 20A). The refined model now predicts a
mode at 24.5 Hz (versus 23 Hz originally) that corresponds to a 28-Hz peak in the measured data. Simi-
larly, a 37-Hz peak in the measured data is also underpredicted by approximately 3 Hz. This
improvement in strut response prediction coupled with the more accurate representation of the acoustic
curtain led to a much improved prediction for the interior noise level (fig. 20B).
As stated earlier, point-to-point comparisons of test data and prediction can be difficult due to uncer-
tainties associated with exact location of test instrumentation and small differences in location between
predicted and measured modal node lines. An alternate way to assess the finite element model predic-
tions is to compare the range of response for all individual measurements of a common set of instrumen-
tation with the corresponding predicted levels. Figures 21 and 22 compare measured ranges of test data
to predictions for the front mount vertical shake and aft lower mount fore-aft shake conditions,
respectively. The areas shown for comparison are the bulkhead, the floor, the body frame at BS 1010, and
the cabin interior.
For the front mount vertical shake condition, the majority of predictions for the individual locations fall
within the range of measured data with the exception of the floor response. Predictions are particularly
good at more than 20 Hz. For comparison, data and predictions for an aft mount fore-aft shake condi-
tion are presented in figure 22. The same general comments for the front mount vertical shake condition
apply here as well. Note that the model accurately predicts the changing sensitivity of the cabin sidewall
and airspace to force inputs in the vertical and axial directions. That is, for the vertical shake condition
measured and predicted response remains fairly constant while for the fore-aft shake condition, both
clearly rise with increasing frequency.
i
ii
U
I
lm
II
U
Mg
I
m
Using the refined model, the effect of modeling the cargo compartment acoustic space was evaluated. A
finite element mesh of the cargo compartment was generated that interfaced with lower sidewall and
floor structural nodes. The cabin acoustic model interfaced with the same floor nodes (fig. 16) allowing
coupling between the two compartments. Predictions show small differences to the cabin acoustic re-
sponse most notably at the cabin acoustic resonance and at a cargo compartment acoustic resonance,
|
l
24 |m
J
L-
=__
- 7]tw
Im==tg
._-_
g_
N
w
= ....
100
9O
8Oen"oc" 70O
U
_ 40
z20
10
010
tO0
oo
0z
1987 FEM prediction
F GVT data
;"" _';'"_- "..... t1988 FEM prediction ""/
I I
2O 3O
Frequency, Hz
(a) Front Engine Mount Vertical Response
4O
100
9O¢0lO
8O
7o
_so
_30
z
10
_ GV'I" data
.,--,! ....-...--.' ' 'v' ', ",, ':'-" / ....... i./
987 FEM 1988 FEM prediction
predictionI I
10 20 30
Frequency, Hz
(b) BS 1010 GE Slde Stringer 12
Radial Frame Response
4O
9O -
80-
70
60
5O
4O
30
2O
10-
010
/._ GV'I" data/
I /--1988 FEM
- .J / prediction
I I
20 30
Frequency, Hz
(¢) BS 1010 GE Side Outboard Seat
Track Vertical Floor Response
I. t:i
100
90co
7O
4O
.N 30
z
10
40 10
/.-- GVT data
I C1987 FEM prediction
/" "-'__1_.. ,T ,'...# _','_x...' i_ o,
'"'/,_ 1988 FEM
prediction
I I
20 3O 4O
Frequency, Hz
(d) BS 1010 GE Side Window Seat Microphone
Figure 18. Finite Element Prediction Versus 727/GE36 Demonstrator Test;Ground Vibration Test- Front Mount Vertical Shake
_-U_01_-23
which occurs at 32 Hz. However, the general characteristics of the response are not changed and further
studies were completed without the cargo compartment model.
A study was also completed to help define minimum modeling size requirements. It was noted that the
tremendous size of the finite element model made model checkout and predictions time consuming and
expensive. A series of model size reductions were previously described (fig. 17). These models were used
to predict response to several shake conditions in order to assess prediction costs and associated accura-
cy. Table 3 gives cost and model size comparisons for each model as referenced to the baseline (refined)
25
I
100
NormaJized
sound pressure
level, dB
Figure 19.
90
8O
7O
6O
50
4O
30
2O
10 --
-- GVT data/-_ _ /-- BS 990 FEM predic_on
-- i/ .,_%J
..-...... / t "V_" --. ..... ,,._J*. " -- . "*:. "." • ." "_ /'" ". ,_x t " " ._ "" /_r.t / . ....... , .. _, .. . . , ,,i. .% .
_-_ ;)" .,_ -.._. / "'-:;x x I";'- "q"_'_'- ""'.,':';'"=II _, .."":" / " : _"_ _ I _ "." '_-"r"_<.."_-.':.::"/ _ i ! ",',, _. rJ v.,I ._ -._ ,'l i _! "_ L.,_:r? ,, t i a /" _: "_.r s;:
" '; Itit I"-- BS 1030 FEM ", ¢ /
-- predic_on _ ,"
..%
_ ": _--- _ 1010 FEM prediclJon
o I I10 20 30 4O
Frequency, Hz
Longitudina/ VariationsAround BS 1010-GE Side Window Seat Microphone PredictionO-UgO 196-24
model. As can be seen, 40% to 60% reductions in computing costs were achieved as well as up to 40%
reductions in model size. Figures 23A and 23B compare the baseline model to the reduced models for
passenger cabin sidewall and acoustic response, respectively, for an aft lower mount vertical shake con-
dition. The results shown in figure 23 are typical for all the shake conditions. It was felt that reduced
models A and B still gave reasonable results, but that complete elimination of the tail, wing, and forward
body structure (reduced model C) degrad e pred!ctions to unacceptable levels within the frequency range
studied. A further comparison that more clearly illustrates this conclusion iscontained in figure 24. The
importance of gross airplane modes in prediction transmission and radiation of engine vibration noise
within this frequency range has been noted previously.
The refined model reflects a valid representation of the 727 demonstrator airplane per production draw-
ings. Inclusion of trim structural and seat absorption models might offer some improvement although
tap tests and noise surveys completed during ground tests indicate the main effects would occur at more
26
I
I
I
II
i
I
i
i
i
i
i
I
i
[]
[]I
I
I
|
|
[ z
L.=u
i
R
!
-5"
i
= _w
M
=
100
90
80
i13_70
8
i°50
40
z
10
1987 FEM GVT data--_
prediction _ \
\ /--1988 FEM _.
I predicti°n ,._
.,.;o,o, js ,_. /" •
t'. ,-.i
i'_.... /" ... ! ,/ "./Y,D: il
"..." il
100
0 I I
10 20 30
Frequency. Hz
(a) Aft Lower Engine Mount Lateral Shake
Figure 20.
40 4O
9O
m 80 1987 FEM
prediction 1988 FEM
70 _.. predic_on-_
6o i.! .... ,• ,',.,.' .'.;,. ._... ,' .... "... ,_-,r:.... ?'.."_:'t '"....'#.... \ ' ', '
40 i;,_._<_.-_" _ i ._ . _ / ", /
,, ,., , - \j
",lz 20 "--- GV'I" data
10
0 _ !
10 20 30
Frequency, Hz
(b) BS 1010 Alile Microphone
Finite Element Prediction Versus 727/GE36 Demonstrator Test,"Ground Vibration Test-AftLower Mount Lateral Shake
_-U_O196-25
than 40 Hz. The extremely complex empennage and strut structure, modified for the GE36 engine, make
modeling difficult and may lead to some problems that would not be encountered in more conventional
structure. Given the above comments and the results presented in this section, it is concluded that finite
element models can be used to predict differences in cabin noise and vibration sensitivity to engine un-
balance forces for different force levels and directions and for different aircraft configurations. Howev-
er, the ability to predict absolute levels at discrete frequencies, with accuracy typically required to make
cabin noise guarantees, has not been demonstrated.
Other studies with the model for flight predictions have shown that the phase relationships between
multiple input forces can shift predicted cabin noise levels 20 dB or more. This result has beenvalidated
independently on flight tests of conventional turbofan aircraft. Thus, the analyst also needs a good un-
derstanding of engine response characteristics in order to assess cabin noise at engine rotor frequencies.
Model reduction studies indicate that within the frequency range studied in this effort, some representa-
tion of the entire aircraft is required. That is, low-frequency whole body and wing modes can contribute
to transmission and radiation of engine rotor noise. This effect diminishes with increasing frequency.
Therefore, the analyst should have some understanding of the aircraft response modes versus the fre-
quency range of his interest in order to determine model size requirements.
27
iJ
I 'it I '| Bulkhead Accelerometers Fore-Aft Response t Floor Accelerometers Vertical Response
_o 1 F &. I"_60 Range of GV'I"data "_ 60
i,_.:.t • _. . !, .. ..,,%. : .'_._.,,,-
I:' ! " F
o 1t'_ I , '_W." I10 20 30 40 10 20 30 40
100
°:LFrequency, Hz Frequency, Hz
BS 1010 Frame Accelerometer Radial Response
7o
50
40
_ 2oz
lO
010
-- Range of Gv'r data
F_eqicfi°ni_
-:...._:,,%,,
I
I I20 30 40
Frequency, Hz
100 1 Cabin Seat Microphones
9O
70 I--- Range of GVT dab
_- . 2"
10 -- FEM predictions
I I _t010 2O 30 40
Frequency. Hz
mm
II
|
i
m
i
i
Figure 21. Finite Element Predictions Versus 727/GE36 Demonstrator Test;Ground Vibration Test-Front Mount Vertical Shake
_UgO186-26
IB
N
m
tl
28M
iiL
L
= =
iii
i
W
=.......
100
9O
8O
70
50
_ 4o
z
20
10
.
0
z
010
IO0
9O
8O
70
60
50
40
3O
2O
10
Bulkhead Accelerometer= Fore-Aft Response
Range of GVT data
_:_ii_ ::'i:_: ,, ,.-..,. :,'._: ....../-V ""
20 30 40
Frequency, Hz
flO"0
z
100
9O
8O
70
6O
50
40
3O
2O
10
010
Floor Accelerometers Vertical Response
,/,, FEM predictions/
.., • .! • ._ =_
_ _. - , ,.',"_ -..-.;
,20 3O 4O
Frequency, Hz
010
BS 1010 Frame Accelerometers Radial Response
FEM predictions _,
p._,. o.
20 30 40
Frequency, HZ
133
z
100
9O
80
70
6O
5O
40
3O
20
t Cabin Seat Microphones
Range of GVT data,
,,.. ,,',_.,_ t- _,_i __
_',_'_k'"" /"- "" "(VL............. /
10 --
I I0
10 20 30
Frequency, Hz
Figure 22. Finite Element Predictions Versus 727/GE36 Demonstrator Test;Ground Vibration Test-Aft Lower Mount Fore-Aft Shake
4O
9-UgO196-27
L
--_._
29
rnnO
N
Z
100
I8O7O
6O
5O
4O
3O
2O
10--
0I0
Full Model
Range of FEM
GVT data predictions
e ;1',_
I I
100
9O
8O
_= 70
i°5(1
I°Z
20
10
|
Jmi
[]I
_ Reduced Model A
FEM
predictions_
Range of
GVT data
t I20 3O
Frequency, Hzo |20 30 40 10 40
Frequency, Hz
Reduced Model B i90 -- Reduced Model C
80 I 80 --== _ Range of Gv'r"data |
r _ .J FEM predictions,_.,..:.,i,_!::.il/_,-.".,.J_ 70 _ . ,-._:_'i,"
:1
I20 20
10 10
0 010 20 30 40 10 20 30
Frequency, Hz Frequency. Hz
(a) BS 1010 Frame Accelerometers Radial Response
Figure 23. Finite E/ement Predictions Versus 727/GE36 Demonstrator Test;Ground Vibration Test-Aft Lower Mount Vertical Shake
4O
g-ug0196.28 f
IB
30
g_
B
mmz
B
w
_qw
m
W
L
HH
E--
6
U
W
100 100
9O
o0•o 80
7O
I0.
))z
CO
nO
iI)
z
Full Model
90
m80
7o
iI
z
Range ofGv'r data
-_:2:.,,_at,
10 FEM predictions
o I I10 20 30 40
Frequency, Hz
100 100
_9_3t Reduced Model B 90co•o 80
| FEM pr_ictions.70_-- _,,_. Range of L_ _ 70
/ _'_'_%'_i_'" GVT data _..,K7" "¢
i,o4O
' ! 130
10 10
o I I o10 20 30 40 10
Frequency, Hz
(b) Cabin Seat Microphones
Figure 23.
m
60
5O
40
3O
2O
_o _-o II0 2O
Reduced Model A
Range of GVT data
"_ predictions
I3O
Frequency, Hz
Reduced Model C
m
• .,j FEM predictions
i.'_ _ _::>.. % _ , .'.._#- "
'+l ".. ,." •,-,¢,.,:::/... I
I l2O 3O
Frequency. Hz
Finite Element Predictions Versus 727/GE36 Demonstrator Test,Ground Vibration Test-Aft Lower Mount Vertical Shake
40
4O
9-Ug01g_28a
r_=,,,,,=,_
L
31
r_
100
9O
=08O
70
_6o
o20
10
010
lOO
9O
=08O
_70
C
_ 4o
N
7° 2o
10
010
Full Model
-- Range of GV'I"data---_L
,^..-. _-_,_.]
I 120 3O 40
Frequency. Hz
Reduced Model B
-- FEM predictions _
!" _!i;'_! :,..'
100
Reduced Model A9O
! ,o_ x_ 6°- .'__t,, -._'--
20 F _ _ " Ra_ngeof GVT datat;
10 --_
' I I01
10 20 30
=0_o
._ 70
_o0e_
-_ 5t?t-
_ 4o
z 20
10
I I o20 30 40 10
Frequency, Hz
Frequency. Hz4[
Reduced Model C
. ,,/,_/_-.-- FEM predictions
,'J ,;..,,\_ //'_J'-_..:, ,.. __ ,.-.
i& "-_ \'..... :..'_'_.,,.._,,/_', ','_ -...,_,_
_ta _'_ n
• I I20 30 ,_
Frequency, Hz
Cabin Seat Microphones
Figure 24. Finite Element Predictions Versus 727/GE36 Demonstrator Test;Ground Vibration Test-Aft Lower Mount Lateral Shake _U_01_8-2_
==
I
m
|
i
i
|
i
i
mm
i
|
II
32
Im
i
II
i
L_
-y-,
w
U
W
=--7
i
t---.
i
L
L
g
Table 3. Modefing Time and Cost Comparison for 727 Demonstrator Airplane Finite Element Models
Model
BaselineReducedmodel AReducedmodel B
Reducedmodel C
Scaled numberofstructuralnodes
1.00 I1.00 I0.67 I0.65 I
I
i
Scaled numberofstructuralelements
1.00
1.000.710.61
Scaledcostofmodalcalculation
0.64
0.440.330.28
i
i Scaledcostof one I Scaled cost ofmodalplusone
I forced response I forcedresponseI caJcula_on 1 calculationI II 0.36 I 1.00I 0.15 I 0.59I 0.29 I 0.62
I 0.16 I 0.44I I
U_1_R1-3
4.4 STATISTICAL ENERGY ANALYSIS RESULTS (FY 1987)
Statistical energy analysis models were developed to predict response for the rotor frequency range and
the propeller frequency range. Predictions were compared to engine strut shake, side-of-body shake, and
flight test conditions.
Unlike the finite element analysis, which used specific structural and acoustic modes, the SEA models
relied on predicted mod.' densities within each substructure to calculate power transfer to the passen-
ger cabin structure and airspace. Specific mode shapes for frequencies were not calculated, rather a
statistical description of modes and interaction of modes with adjacent substructures was used. Thus,
predictions for structures with higher modal densities tend to be better. As part of this effort, it was
found that the dimensions and physical properties of aircraft structures tend to give low modal densities
for the frequency ranges studied.
Low-Frequency Model
Low-frequency (rotor) model predictions were made up to 100 Hz for eight strut shakes, one side-of-
body shake, and one flight test condition. The model was felt to be valid over this frequency range despite
lumped mass representation of the interior trim. Results from the midfrequency (propeller) model
showed energy transfer from the trim to be small compared to that from sidewall structure modes.
Example comparisons of ground test data with low-frequency model predictions for lateral and vertical
strut shake condition are given in figures 25A through 25D and 26A through 26C for the strut, fuselage
sidewall, floor, and acoustic space. The predicted average represents the space average response for the
subsystem. The 95% confidence intervals represent the range or variance predicted for response at indi-
vidual locations on or within the substructure (up to 95% confidence). All of the available measured data
on or within the subsystem were compared to the prediction.
The following general observations were made:
1. For the most part, the data fell within the predicted confidence interval, although the interval ap-
peared to be somewhat greater than the corresponding ranges of measured data for the lateral shake
condition and somewhat less for the vertical condition.
33
813
m
•o 70==
==6o
40
0 0 20
70_
¢0
_6O°
4oZ
30
00 20
I
(a) Strut Vibration Levels ,q Measured] 70 Co)Section 46 Sidewall Vibration /, _ data I I
I i '• • $
40 60 80 100 120 140 160 180 200 O0 20 40 60 80 100 120 140 160 180 200
Frequency, Hz Frequency, Hz
(c) Section 46 Floor Vibration Levels
70
40 60 80 100 120 140 160 180 200Frequency. Hz
Legend:
95% confidenceJ :[ SEAMTM predicted average-intervals I ]: 25-element model
6O0313
¢50
.i
z_
2O :120
(d) Section 46 Interior Noise Levels
h,..:.',• g/<a.2, ',., ,
.\=.,,, .-.
\i: "_-, , ,,.:, v,'.' , , ,
40 60 80 100 i20 i40 160 180 200Frequency, Hz
Figure 25. SEAM TM Predicted Levels Versus 727 Demonstrator Test Data;GVT-Forward Mount Lateral Shake
|miiB
mB
_-Ug01 _30
2. The data peaks were in generally good agreement with the upper limit of the confidence interval, but
the predicted average exceeded the data average at many frequencies. iil
3. The rise and fall of the data mean with frequency was not predicted by the model, particularly for
strut and acoustic response. I
For the lateral strut shake condition, the large confidence interval was due to uncertainty in predicted
power input to the strut inplane subsystem. This uncertainty results from the low modal density of the
strut inplane subsystem• For any structure, the range between maximum and minimum levels will in-
crease as the modal density is decreased.
I
=34
m
I
1
L
=.=
r _
r_w
w
7-i
w
2
=
w
80
". I,', I
8.i
Ii ,,./z
/,f_\ s Measured
,'.-.', \ / data, '_ -,\ ,\t ,
ill v
'k/J.-- .I. "",.,"
,07
0 20 40
95% | T SEAMTM predicted
confidence | T average- 25-elementintervals ( model
Figure 26.
I I I I I60 80 100 120 140
Frequency, Hz
(a) Strut Vibration Levels
70
iiE0Z
ao_-.Z0
I I I160 180 2OO
95% confidenceint rvals
+7
,. ,°
1
80
==:_ r° f
1
z40
0
95% confidence I T SEAMTM predicted
intervals l.L average - 25-element model
& , • , °
o-° i.. l+ _ • /a; • ' ,'"-• ,- ;_,!_,_-"_, "_"",.... .--_" _..._,\.'-"x_,-:,"'-..-:A_,¢i,, ,. I, ,- ;, , ,..,.,..,U.,::"_,,.. " /
20 40 60 80 100 120 140 1613 180 200
Frequency, Hz
(b) Section 46 Sidewall Vibration Level=
t SEAMTM predictedaverage- 25-element model
,,! %'...
Measuredda_
0 20 40 60 80 100 120 140 160 180 200
Frequency, Hz
(c) Section 46 Interior Noise Levels
Low-Frequency SEAM TM Mode/Predicted Leve/s Versus 727 Demonstrator Test Data;GVT-Forward Mount Vertica/ Shake
g-UgOlg6-31
It was found that for this and other test conditions, the response at locations away from the drive point
were overpredicted by a biased amount. This indicted that predicted power inputs, corresponding to the
measured input force, were too high although the transmission modeling is correct• (This may seem at
odds with the underprediction of strut response shown in figure 25A. However, the measured data are at
the end of the strut near the drive point and does not accurately represent average strut response or
power.) The overprediction of power input was related to modeling the drive point structure as an un-
bounded system.
Finally, the failure of the model to predict variations of the mean was attributed to low modal densities in
the actual aircraft structure in this frequency range• This led to the possibility of individual modes domi-
nating response. In this case high strut response near 40 Hz and a cross-cabin acoustic mode near 50 Hz
appeared to combine to give the broad peak in acoustic levels•
35
Bycomparingfigures25and26,it wasnotedthatthemodelpredictedtheverylargemeasureddifferencein strut responsefor the lateralandverticaldriveconditions,aswellasthemuchlowerinteriornoise.Thisagainindicatedthetransmissionmodelingwascorrectandthat SEAis capableof predictingair-craft responsetrendsat lowfrequencies.
MidfrequencyModel
Midfrequencymodelpredictionswere made for two strut shake, three side-of-body shake, and one flight
test condition. Originally, it was planned to make predictions for more GVT conditions. However, the
additional time and budget expended on the initial modeling effort (400-element model) and subsequent
revisions precluded this.
Evaluation of the model concluded that composite-bending stiffness modes of thi_ sidewall dominated
energy transfer to the passenger cabin when compared to trim-panel-bending and skin-membrane-
bending modes. The same observations made for the low-frequency model generally held for compari-
sons of mid-frequency model predictions to test. Predictions tended to be somewhat better and the
influence of individual modes was minimized in this frequency range. The test arena acoustic space was
represented by three space subsystems rather than one (in the low-frequency model) and falloff in level
versus body station was accurately predicted. Figure 27A presents sample midfrequency model predic-
tions for a GVT strut shake condition. Again, the strut response appears underpredicted, but note that
all the measurement points are at the end of the strut. Figure 27B shows predictions for sidewall vibra-
tion levels, and figure 27C shows predictions for aft interior noise levels.
Flight test predictions were hampered again by the inability to estimate input power, a situation also
noted when using PAIN as described later. In this case, measured accelerations were used to develop
input forces. Predicted responseSdownstream of the strut shoWed consistent bias levels (overprediction)
with respect to the measured data. This indicated that estimated power was again overpredicted but that
energy flow was accurately modeled.
Based on the results presented, it was concluded that SEA shows promise in providing relatively simple
models for evaluation of low-frequency noise and vibration transmission. The mean and variance esti-
mates provided by SEAM TM are useful to the designer in that the upper bound of predicted response
can be compared to requirements for maximum allowable noise and vibration. The power flow analysis
feature can be useful in guiding design type and application of suppression hardware.
Two areas of study were recommended including investigation of power input estimation techniques and
improved modal density estimates for low-frequency problems.
In this effort, excitation loading (input power) was developed from input forces and conductance of the
strut (or body) derived from the SEA model. Results showed the response to be consistently
overpredicted, while the rate of response falloff (with distance from the drive point) was accurately pre-
dicted. This suggested that the input power was overpredicted. In the model, the input power was pre-
dicted in terms of the average conductance of the subsystem (associated with the drive point), with the
36
m
l
il
Z
m
m!
I
J
l
[]
[]
I
mm
[]
i
m
u
m
m
I
I
|
[]
|
n
i
D
i
L_
'iW
• _ f//,_,_ .,, "., __ confidence| average- 100-elementmodel70 -- ; f \ ) i _ a/ % _ _' _ /._/-" _ 70 i._nntervals - (prediction normalized at aft fuselage]
-", , ',, ,:,,,,,,,! ,xj I l i'"'.,' .. ,-, t,, r
50- I ', i,_ Measured / ..L 50- ..--. ,.. ; ... ,-.. , ,.,
,,//data 95% l''J average -IO0-eIementSEAMTMpredicted :::!. /: '...........,, /: "'... : .....',k,'I.. . ..,-
Z4_0_7_ _. ¢onnaence [ j model (prediction z40{ j"" i'"" 7intervals normalized at aft fuselage) M u
I I 1 1 I I I I I I0 20 40 60 80 100 120 140 160 180 200 0 20 40 60 80 100 120 140 160 180 200
Frequency, Hz Frequency. Hz
(a) Strut Vibration Levels (b) Section 48 Sidewall Vibration Levels
{_ SEAMTMpredicted
80 !T 95% confidence average - l O0-element model
intervals (prediction normalized at aft fuselage)
L,A ii Measured
-_ _ , _." [<_i'..,.
oz I I I 1 I _ I I I I '_10 20 40 60 80 100 120 140 160 180 200
Frequency. Hz
(c) Section 46 Aft Interior Noise Levels
Midfrequency SEAM TM Mode/Predicted Leve/s Versus 727 Demonstrator Test Data;GVT"- Forward Mount Vertical Shake
0-Ug0106-32
Figure 27.
average encompassing at least one resonance frequency of the subsystem. At frequencies below the first
subsystem resonance, the input power will be incorrectly predicted. Some method for handling these
cases must be developed.
It was noted that the SEA predictions do not follow variations with frequency of the measured mean
response. A review of data indicated that low modal density in the aircraft structure may result in re-
sponse that can be dominated by a few individual modes. Because SEA does not calculate specific
modes, it will not tend to predict this behavior. For example, a cross-cabin acoustic mode near 50 Hz
seemed to influence measured noise levels, but the SEA model only predicted modal densities based on
volume of the acoustic space. In this low-frequency range, it was recommended to study how SEA could
be augmented by modal density estimates from simple models of selected subsystems. These models
would represent closed form or finite element solutions that couid predict specific modes.
37
m
i
4.5 REVIEW OF SEA PROCEDURES FOR LOW-FREQUENCY PROBLEMS
Before modifying the SEA models to improve predictions, a number of studies regarding SEA proce-
dures and potential enhancements for low-frequency problems were undertaken. Tasks were defined for
evaluating methods to improve modal density estimates (including the use of finite element modeling), to
improve input power estimates, and to improve variance predictions.
The modal density is an important element in the calculation of statistical energy analysis parameters.
For continuous systems, the number of modes with resonance frequencies in a given band (the mode
count) is determined from the modal density. The mode count relates the modal energy to the mean-
square response. Impedance functions, which, along with prescribed mechanical or acoustical sources,
determine input power to SEA subsystems, are calculated from the modal densities. The modal density
also enters directly into the relationship between coupling loss factor and the coupling parameter use in
the matrix formulations of the SEA power balance equations. Finally, the modal density forms the basis
for a statistical description of the resonance frequencies. In such a description, the resonance frequen-
cies are assumed to be random "events." The modal density specifies the mean rate of occurrence for
these events per unit frequency. Methods to improve modal density estimates are described below.
At high frequencies, the modal density of a continuous dynamic system can be shown to approach an
asymptotic limit that depends on the dimensions of the system. These asymptotic limits are written as--
Ln'-D(w) = -- (la)
_Cg
n2_D(w ) = kA2_Cg (lb)
k2V
n3-D(0_) .= 2_2Cg (lc)
where Cg is the group speed for the system, k is the wavenumber, and the geometric parameters L, A, and
V refer to the length, area, and volume of the system, respectively. The asymptotic forms for the modal
density are valid at high frequencies regardless of the boundary conditions and can be applied to both
structural and acoustic systems as long as the correct wavenumber and group speed are used.
4.5.1 ACOUSTIC SPACE MODAL DENSITY ESTIMATE
The modal density for acoustic spaces can be determined quite simply from the asymptotic forms for the
modal densities because the wavenumber and group speed are related to the radian frequency and the
speed of sound in the acoustic media. The asymptotic forms for the modal densities are useful for acous-
tic spaces that can be idealized as one, two, or three dimensional. Most spaces, however, can be idealized
in this way only over limited ranges of frequency. In general, the modal density of acoustic spaces can be
described as the sum of one-, re'o-, and three-dimensional asymptotic modal densities,
g
ii
i
I
m
m
I
m
l
E
,!
I
mU
!
[]
mm
U
U
m
[]
38 mi
II
lIII
L_
k_
IRi
U
i
w
i
i
i
L oaA oa2V= -- + + -- (2)n(co) 2 =cg
where the parameters are selected to provide the best representation of the modal density for the acous-
tic space over the entire frequency range of interest. This form of the modal density provides three geo-
metric parameters that can be adjusted according to modeling rules for specific types of acoustic spaces.
Generally, the first term is selected to obtain the desired modal density at low frequencies, where one-
dimensional modes dominate. The second term is selected to obtain the desired modal density at midfre-
quencies where two-dimensional modes dominate, and the third term is selected to obtain the desired
modal density at high frequencies where three-dimensional modes dominate. In some cases only two of
the three terms in the general expression will be needed. Example calculations for the 727 demonstrator
airplane are given in section 4.7.2.
4.5.2 STRUCTURE MODAL DENSITY ESTIMATES
The modal density of a structural subsystem can also be obtained from the asymptotic forms for the
modal density given by equations 1. In general, a structural subsystem may have different types of
modes, e.g., bending, inplane compression, inplane shear, and torsional vibration modes. The different
types of modes are usually placed in separate SEA subsystems. As part of this study, general modal
density expressions were developed for each type of mode. To use these expressions requires that a wave-
number and group speed be determined for each type of mode. However, the calculation ofwavenumber
and group speed is more complex than for acoustic spaces. For the 727 demonstrator airplane model,further studies were limited to the use of finite element models or measured data.
4.5.3 USE OF FINITE ELEMENT MODELS TO VERIFY SEA PARAMETERS
Statistical energy analysis is an effective technique for modeling complex dynamic systems having many
modes of vibration. SEA is generally perceived to complement finite element modeling by providing a
prediction procedure at high frequencies where the great number of degrees of freedom required for
finite element modeling is prohibitive. Finite element models can also complement SEA modeling by
providing data to determine parameters such as modal density., power input, and coupling loss factors
for complex structures.
As discussed previously, the modal density is used in the calculation of several SEA parameters. Thus,
an immediate use for finite element modeling is to determine the subsystem modal densities. A finite
element model of an SEA substructure can be used to calculate the substructure resonance frequencies.
The asymptotic values of the modal densities for the substructures of the 727/GE36 demonstrator air-
plane as computed for the low-frequency SEA model are shown in figure 28 over the frequency range 20to 100 Hz. The modal densities of the cabin acoustic space (subsystem 6), the fuselage sidewall (subsys-
tem 5), and the cabin floor (subsystem 21) show an average resonance frequency spacing (inverse modal
density) of approximately 8 Hz. The modal densities for these subsystems are sufficiently large that the
use of the asymptotic form of the modal density should be valid.
39
m
0.5
0.2
0.10.05
0.020.01
0.0050.002
0.001
0.0005
Legend:
0.002
-- I 0.001 __
I 0.0005
i 0.0002
0,00005 I
10 20 30 40 50 60 70 80 90 loo 10 20 30 40 50 60 70 80
Frequency, Hz Frequency, Hz
1 Subsystem 3 (afffusp) 4 Subsystem 16 (pbulkc) 7 Subsystem 16 (pbulkc) * Sb-at bending (sb'ub_)
2 Subsystem 4 (aftfusa) 5 Subsystem 5 (midfusp) 8 Subsystem 21 (miclflc) 0 Strut in plane (strutri)
3 Subsystem 14 (strutrb) 6 Subsystem 6 (midfusa)
Figure 28. Asymptotic Moda/ Densities for the Low-Frequency SEA Mode/
I90 IO0
g-UgO196-33
The predicted modal densities for the pressure bulkhead (subsystem 16), the aft fuselage empennage
structure (subsystem 3), and the empennage acoustic space (subsystem 4) show an average resonance
frequency spacing of approximately 50 Hz. The complexity of the structures and the empennage acoustic
space is sufficiently great that the predicted modal densities should be checked using a finite element
model to obtain a modal density. Because of the fairly large frequency spacing, the use of the asymptotic
forms to predict the modal densities for these subsystems is questionable.
Predicted modal densities for the strut are even smaller. The obvious solution to the SEA modeling
problem for a small modal density is to increase the size of the subsystem. In the case of the strut, a
subsystem could be created that includes the strut, a major section (perhaps all) of the aft fuselage em-
pennage, and the pressure bulkhead. Although the problem could potentially be alleviated in this man-
ner, the procedure of calculating SEA parameters for such a subsystem have yet to be developed.
However, the use of a finite element model would appear to be a reasonable approach to obtaining the
modal density for such subsystem.
As part of this task, a procedure for using finite element data to calculate a modal density was estab-
lished. Briefly, a single substructure is selected to be modeled. Simple boundary conditions are chosen to
approximate bordering subsystems. Eigenvalues are then calculated to a frequency somewhat more than
the upper frequency of interest. At each eigenvalue the frequency spacing to the next eigenvalue is calcu-
lated. The frequency spacings are then inverted to find an "instantaneous modal density," which is
plotted to compare with the asymptotic modal density.
An example of this procedure using the cabin acoustic space is given in section 4.7.2. A new strut modal
density is also calculated in section 4.7.2, but rather than using the FEM predicted frequencies,
4O
m
i
m
I
m
.m
l
t
i
m
m
nl
i
m
Ii
i
.m
i
I
z
m
[]
I
=
|
t_J
i
.:==
m
2__
"7
B_
=
u
F
i
Ez_
K_
=
measured modal densities of the composite system were used in a like fashion. Obviously, it is desirable
that the complexity of subsystem finite element models be kept to a minimum otherwise the advantages
of SEA, such as small model size and quick result turnaround, are lost.
A second use for finite element modeling is to determine the input power from prescribed excitations. In
SEA the input power is obtained from a statistical impedance analysis. Finite element models can also
be employed to compute power input and, thereby, provide a means to check the validity of the SEAcalculations.
A third use for finite element modeling is to provide data for calculating the coupling loss factor. Be-
cause coupling loss factors are formulated in terms of modal densities and impedances, improved esti-
mates for the modal densities and impedances will lead to improved coupling loss factor predictions.
4.6 SEAM TM CODE IMPROVEMENTS
In the baseline low-frequency model, the power input to the strut is calculated assuming that the strut
subsystem has infinite boundary conditions. Based on studies completed, the finiteness of the subsystem
and "loading" effects caused by other subsystems connected to the strut should be accounted for.
The input power calculation in the SEAM code has been improved. The input power at a given frequency
or in a band of frequencies is the product of the mean-square force applied to the system and the drive
point conductance (real part of the mobility). When a mass is added to the structure at the point of
excitation, the impedance (inverse of mobility) of the combined system is the sum of the structure imped-
ance plus the mass impedance. In general, the structure impedance is a complex number with real and
imaginary parts, while the mass impedance has only an imaginary part. In earlier versions of SEAM the
mass impedance was improperly set to be real. This resulted in incorrect power input calculations, al-
though the error was not great. The current SEAM code properly calculates the power input using the
correct imaginary value for the mass impedance.
In SEA the predicted variance depends on the inverse of the product of the modal density and the loss
factor. At low frequencies, where the modal densities are small, the variance can be quite large. Again,
considering the effect of bordering subsystems on the subsystem in question, an "effective loss factor"
has been defined that includes both energy dissipation to damping within the subsystem as well as ener-
gy transmitted to connected subsystems.
In order to evaluate the improvements to the SEAM computational procedure, the original and new
input power estimates are shown in figure 29 along with the measured input power for the vertical and
lateral drives, respectively. The SEA mean for the vertical drive condition is 1 to 2 dB more than the
lateral drive condition and in somewhat better agreement with the measured power input. The SEA
mean for the lateral drive conditions is not significantly changed.
The SEAM code provides a prediction for the variance (standard deviation squared) of the power input.
In SEA it is generally assumed that the power input is a log normal distribution, i.e., the variable 10
41
-20
-30
-soi
-60
-7O
;aseqne SEAM VI.861
I, I I I I I I I 1
-20 Improved Version
i
_o1 -7o-- I.... 1 I I I 1 I I
i
i
i0 100 200 0 100 200 __
Frequency, Hz Frequency. Hz H
(a) Vertical Shake Condition
-20 B_.aasellne SEAM V1.861 .... 20 Improved Version _ _
-30 -- _ -30
;-8.-5o _ _50_ "_
f
fI I 1 I 1 I I I I I -7o [ I I 1 I [ I I I I _-
100 200 0 100 2O0
Frequency, Hz Frequency, Hz
(b) Lateral Shake Condition
SEA mean
SEA mean ± 2 sigma
Measured data
Figure 29. Input Power Comparisons• .UgOlgS-34 j
-60--
-700
Legend:
LOG10 [power input] is normally distributed. Given this assumption, 95% of the power input data in
decibels should fall within a range defined by the mean ,,2 times the standard deviation. Thus, this range
can be used as a 95% confidence interval. In figure 29, the confidence interval has increased approxi-
mately 5 dB to encompass the data for the vertical drive and has decreased up to 5 dB at less than 60 Hz
as desired for the lateral drive, so that the standard deviation is now essentially constant with frequency.
The same trend is true of the response prediction standard deviations.
iI
wE
42
i
m
BI
W
m
w
w
k_
m
4.7 SEA LOW-FREQUENCY MODEL IMPROVEMENTS
A number of improvements to the low-frequency model were made including the input power estima-
tion, new modal density estimates for the cabin acoustic space and the strut, a new estimate for the strut
loss factor, and an improvement to the modeling of the strut connection. Details are given below.
4.7.1 INPUT POWER ESTIMATE
During the ground test measurements, impedances were measured at the drive points. It was therefore
possible to use conductance, the real part of the mobility (inverse impedance), to obtain a measured
value for the power input. The variance in the SEA predictions includes a variance in the input power. By
specifying directly the measured input power in the SEA model, the overall prediction variance can be
reduced thereby improving the prediction accuracy.
The predictions obtained when measured input powers are used directly are given in figure 30 for the
vertical shake condition. The SEA mean is plotted with the mean of the measured data for clarity of
presentation. The predicted results are too high, which indicates that the coupling between the strut and
the empennage is too high. Clearly, simply changing the input power level without correcting the low
strut modal density is not satisfactory.
80
701
_6o__8.
40m
3O
0
\\
I I I I I I I I I20 40 60 80 100 120 140 160 180 200
Frequency. Hz
(a) Section 46 Sidewall Vibration (mldfusp)
Legend:SEAM
_ = Average of measured data
Figure 30.
70
6O
(o
= 40¢6
2:
30
¢t
x._._/'x\',,j\.
20 I I I I I I I I I0 20 40 60 80 100 120 140 160 180 200
Frequency, Hz
(b) Section 46 Interior Noise Levels (midfusa)
Response Predictions for the Low-Frequency Baseline Model Using Measured InputPower- Vertical Shake
9-Ug0196-35
43
m
il
4.7.2 MODAL DENSITY ESTIMATES
The modal densities for the strut do not include the effect of boundary conditions or connected struc-
tures. A composite modal density, which lumps together both the strut and sections of the empennage,
was determined from the spacing between peaks in the measured input power. These frequency spacings
were inverted to calculate modal density in the same way that spacing between eigenvalues predicted
from a finite element model would be used (fig. 31A). Because modal density cannot be input directly in
the SEAM program, it was necessary to modify the strut properties to simulate the modal density
derived from the measured data. This was accomplished by increasingthe strutlength by a factor of four
while decreasing the strut density by the same factor to maintain a constant strut mass. The comparison
of the SEAM calculated modal density with that calculated from the measured input power peak spac-
ings is shown in figure 3lB.
The use of a finite element model to validate a modal density estimate is illustrated for the cabin acoustic
space. A method for determining the modal densities from finite element models has been discussed. To
demonstrate the procedure, the first 100 modes for the acoustic interior of the Boeing 727/GE36 demon-
strator a_rcraft were computed from a finite element model with 1525 three-dimensional elements. The
resonance frequencies were sorted in ascending order, adjacent frequencies subtracted to compute the
interval between resonance frequencies, and the intervals inverted to obtain an "instantaneous" modal
density. The result of the calculation is shown in figure 32. When presented in this manner the resonance
frequency data appear to be random.
Figure 33 presents the modal density data smoothed over a 10-Hz bandwidth. The data in this plot have
been obtained by counting the number of resonance frequencies in each 10-Hz-wide band and dividing
by bandwidth. The result is a series of data points with a 10-I-Iz spacing, which can now be compared
with predicted modal densities.
In order to more accurately represent the demonstrator airplane test environment, the dimensions of the
interior acoustic space were re-evaluated and modified from those used in the 1987 SEA model predic-
tions. The modal density prediction using the cabin acoustic space volume (52m 3) in equation lc is
shown in figure 33. The estimate is seen to underpredict the modal density obtained from the finite ele-
ment data. The second procedure is to use the general form of the modal density given by equation 2. The
volume was selected to equal the volume of the cabin (52m3), the area to equal the area of the floor
(22.3m2), and the length to equal the length from the lavatory walls to the acoustic barrier (6.1m). The
resulting modal density is also shown in figure 33. The use of the area and length terms improve the
agreement between the finite element data and the prediction. This estimate was used in all further
studies.
4.7.3 STRUT CONNECTION MODELING
It was noted that in the lateral drive model, the strut inplane subsystem is connected to both the empen-
nage and pressure bulkhead bending mode groups. In fact, the strut should only be connected to the
inplane modes of the pressure bulkhead. However, there are no inplane modes of the pressure bulkhead
44
m
am
iim
m
m
m
In
!i
H
J
lU
mmm
gl
EII
iIII
L.-_
III
iI
|
|
; c
L_r_
L_
w
I_ip
E
!
J
k
.=m"10
E_=
"O
Eo"
C)
-20
-30 -
/
-40 --
-50
-60-
-70-
-8O0
-10 -
-20 -
-30 -
-40-
.60 0
Figure 31.
A O) 1 =246.8
0_ 1 =253.4
Legend:Measured input powerforward stnJtddve vertical
z_ _ 2 =341.6
(_ 2 = 547.6
n((_ 1) = n (253.4) = 1/ ZX (_ 1-1/246.8 = 4.1E-3
n( {o2) = n (547.6) = 1/ /k (d 2•=1/341.6 = 2,9E-3
I2O
I I I I I I40 60 80 100 120 140
Frequency, Hz
(a) Modal Density Estimation Calculation Using Measured Data
I I160 180
Legend:Baseline modelStrut 1_2"4, rho/4
* Determined from frequency spacingpeaks in power input data
n(_ I) n(co2)
_.=
I I 1 I I I I I I20 40 60 80 100 120 140 160 180
Frequency, Hz
(b) Strut Modal Density Comparison (strutrb)
Composite Strut Modal Density Based on Spacing Between Peaks in the Measured
Vertical Shake Input Power
20o
2-uo
U90169-36
45
I.__ 727 Demonstrator Cabin AcousUc Modes NASA and Boeing FEM Da_
20
0,1
i0.05
p
0.02" -
0.010
I I I I50 100 150 200
Frequency, Hz
instantaneous Moda/ Density for the Cabin Acoustic SpaceFigure 32.
25O
g-Ug0106-37
to excite at the frequencies involved. A better model of the strut connection to the fuselage is a junction
with the strut inplane and the empennage through the mass of the bulkhead, so that the nonresonant
impedance of the bulkhead is taken into account.
4.7.4 STRUT LOSS FACTOR
At this point, predictions were made to determine the effect of the above described changes. It was noted
that predictions were all about 5 dB high including the strut vibration level. Because most of the strut
energy is lost to damping (approximately 80% of the power input to the strut is damped internally), the
high predictions can be lowered by increasing strut damping. This will allow a better comparison of the
power flow prediction capabilities of the model. The damping was increased to a value of 0.24 (12% of
critical), which while arbitrary is not unreasonable considering the strut is a highly bolted structure filled
with hydraulic system components. Results are given in section 4.9.
46
=
i
i|
imi
m
m
o
l
im
a
i
I
m
U
I
t_
w
r ....
= .
w
|iiiiii_
t=,,,m_
w
0.20
0,18
0.16
0.14
_'_ 0.12
._ 0.10
0.08
0.08
0.04
0.02
0
Legend:StattstJcaJ results using volume term only
0---"-0 Modal density calculated from three-dimensional FEM results
_ Sta_stJcal results using length, area, and volume terms
20 40 60 80 100 120 140 160 180 2(
Frequency, Hz
Figure 33. Comparison of Modal Density Prediction Procedures- Interior Acoustic Spaceg-Ug01 g6-3
4.8 MIDFREQUENCY MODEL IMPROVEMENTS
The midfrequency model was originally developed to predict cabin noise from 100 to 400 Hz for both
strut vibration and empennage acoustic excitation. However, during the effort to improve the low-fre-
quency model, it was found that this model could be used up to 200 Hz for strut vibration conditions.
Thus, efforts to improve the midfrequency model have focused on side-of-body shaker and inflight
acoustic excitation.
The first attempt to improve the accuracy of the midfrequency model was to use the measured power
input. As with the low-frequency model, some improvement was observed although further work on im-
proving modeling accuracy was felt to be needed.
In setting up the midfrequency model, the empennage was modeled using the properties of the skin pan-
els, stringers, and added honeycomb panels. The frames and torsion boxes were not included in deriving
equivalent homogeneous cross-section properties. Above the frequency of the first subpanel resonance,
the modeling assumption is correct. However, at low frequencies the properties must be changed to in-
clude the stiffening effect of the frames and torsion box structure. The complexity of the empennage
structure precludes development of a modeling procedure without use of measured data. The approach,
47
I
NASAN_IOPali A_oP, abt_ _rlcl
Report Documentation Page
1. Report No.
NASA CR-181851
2. Government Accession No.
4. Title and Subtitle
Evaluation of Analysis Techniques for Low FrequencyInterior Noise and Vibration of Commercial Aircraft
7. Author(s)
A. E. Landrnann, H. E "nllema, andS. E. Marshall
9. Performing Organization Name and Address
Boeing Commercial AirplanesP. O. Box 3707Seattle, WA 98124-2207
12. Sponsoring Agency Name and Address
National Aeronautics and Space AdministrationLangley Research Center
Hampton, VA 23665-5225
3. Recipient's Catalog No.
5. Report Date
October 1989
6. Performing Organization Code
8. Performing Organization Report No.
D6-54871
10. Work Unit No.
1I. Conkact or Grant No.
NAS I - 18027
13. Type of Report and Period Covered
Contractor Report
14. Sponsoring Agency Code
t5. Supplementary Notes
Langley Technical Writer: Kevin R ShepherdFinal Report
16, Abs_act
This report documents the application and evaluation of selected analysis techniques for theprediction of low-frequency cabin noise and vibration associated with propeller engine installations.Three analysis techniques were chosen for evaluation including finite element analysis, statisticalenergy analysis (SEA), and a power flow method using elements of SEA (computer program PAIN).Data from ground and flight tests of a 727 airplane modified to accept a GE36 counter-rotatingpropeller engine were used to compare with the predictions. Generally, the comparisons showreasonable agreement, leading to the conclusion that each technique has value for estimatingpotential impact of candidate noise suppression features. However, variations were large enough toremain cautious about prediction of absolute levels for new aircraft. Recommended improvements"
for each procedure are made.
17. Key Words (Suggested by Author(s))
1. Advanced Propeller Engines
2. Analytical Techniques3. Cabin Vibration4. Interior Noise
18. Distribution Statement
Unclassified - Unlimited
19. Security Classif. (of this report)
Unclassified
20. Security Classif. (of this page)
Unclassified
21. No. of pages
76
22. Price
i
i
|U
m
B
I
BU
i
|m!
lJ
_m
i
m
mB
m
B
m
i
I
lm
_IASA FORM 1626 OCT 66ug01g,6-Recort DO<: Form
U
w
w
_=
g
therefore, was to change the properties of the empennage structure so that the predictions of input
power and empennage response would match the data.
Using this procedure the drive point impedance was modified by constructing a model with a frequency
dependent empennage density (fig. 34). Note the very large and abrupt shift that was required to match
data. This seems to validate the above comments regarding modeling accuracy of the empennage, as the
break in density occurs where the first subpanel resonances would be expected to occur. This would
result in a dramatic shift in input power impedance reflected by the large change in density. Compari-
sons with data are presented in section 4.9.
,¢
toE
r_
35OOO
30 O3O
25 000
2OOOO
15OOO
1OOOO
50OO
Midfrequency 727/GE36 Demonstrator Aircraft SEAM Model VariableDensity Used for Subsystem 25, aslfrp
I I I100 150 200 250 300 3505O
Frequency, Hz
Figure 34. Midfrequency SEAM Model Variable Panel Density
!400 450
9-U9019_-3,g
4S
n
4.9 STATISTICAL ENERGY ANALYSIS RESULTS (FY 1988)
The low-frequency model incorporates the new airspace modal density estimate, the composite strut
modal density, revised strut damping, and the improved strut connection model. Results obtained using
this model are described below.
Response predictions for the vertical shake condition are given in figure 35. Note that predictions have
been extended to 200 Hz. Comparisons with measured data show very good agreement with some
overprediction of acoustic levels at high frequencies. The data falls within the predicted confidence in-
terval, the data peaks are in good agreement with the upper limit of the confidence interval, and in most
cases the predicted average matches the measured average. As was noted earlier, these results were ob-
tained by increasing strut damping to an arbitrary but not unreasonable level. An almost exact match of
the data is obtained when measured input powers are used (fig. 36). Comparisons for fore-aft (figs. 37
and 38) and lateral strut shake conditions are similar to the vertical shake condition.
Because data from the vertical and lateral shake conditions were used in determining parameters for the
improved model, good agreement with data for these conditions is not unexpected. However, the cofn-
parison of the prediction and data for the fore-aft shake condition also shows good agreement, thereby
supporting the general validity of the improvements.
Flight predictions at the rotor frequency fundamental and harmonics were also made. Predictions using
measured GVT transfer functions as the "best" possible prediction were compared with predictions
using SEA estimated transfer functions. In both cases, measured inflight vibration levels were combined
with measured GVT strut impedances to infer forces acting on the strut. Both predictions overestimated
flight levels (by approximately 4 and 6 dB, respectively), but it is noteworthy that the SEA prediction was
within 2 dB of the "best" possible prediction using GVT measured transfer functions.
Comparison of predicted (using the midfrequency model) and measured results for a side-of-body shake
condition are given in figure 39. Because the empennage properties were modified to match measured
response, levels for section 48 (empennage), the bulkhead, and section 46 (passenger cabin sidewall)
match closely. Acoustic levels show good agreement at low frequencies, but are overpredicted at higher
frequencies, a trend seen with the low-frequency model as well. One possible explanation is the absence
of interior trim panels in the models. In 1987 work, it was concluded from preliminary midfrequency
model results that energy transfer through the trim (as compared to composite sidewall modes) was not
important. In light of the most recent findings, this conclusion should be re-examined.
Predictions for flight conditions were again evaluated by comparing results using GVT measured trans-
fer functions and SEA predicted transfer functions. Levels were estimated at fundamental and first har-
monic blade-passage frequencies. Input power was determined by using measured exterior sound
pressure levels and the GVT determined transfer functions. A power input was obtained by multiplying
the measured exterior pressure by an effective excitation area to obtain an equivalent force input. This
area was selected so that the average interior sound pressure level, predicted using the GVT transfer
functions, was equal to the average interior level measured in flight. A prediction was then made using
49
i
J
|
m
J
mm
iimn
II
|
|m
11U
R
[]i
n
Sn
|mI
U
|
i
,= ....
=.==,
_ =
u
,o /t
Sir -- _z
301 I I I I I I s __
z
BO
70
6O
5O
4O
3O
8O
70
6O
5O
40
o 2o
0Z
Pressure Bulkhead
I I I I I I I I I I
Section 46 Sidewall
fI
4O 60 80 100 120 140 160 180 200 220
Frequency, HzLegend:
SEA mean
80
70£o"o
4O
Figure 35.
30
80
7O,-r,,
z
"O
.N
Z
50
4O
3O
8O
70
6O
300
Section 48 Sidewall
I I I I I I ! I I I
Section 46 Floor
I I I I I t I I I I
Section 46 Interior
I
20 40 60 80 100 120 140 160 180 200220
Frequency, HZ
SEA mean + 2 sigma-- -- -- Measured data
Response Predictions for the Low-Frequency Improved Model- Vertical ShakeL_O196R1-40
50
"O
Z
8O
,0t6O
,50 m
40--
3O8O
70--
J. I I I 1 I I 1 I I
Pressure Bulkhead
O_"G
i _-
Z
[ I8O
70
6O_
_n10
Section 46 Sidewall
I 1 I I I I I 1 I I20 40 60 80 100 120 140 160 180 200 220
Frequency, HzLegend
SEA prediction using measured power input_ dB average ot measured data
g
Section 48 Sidewall
Section 46 Floor I
1 I 18O
iiZ
Section 46 Interlor
I 1 I I 1 I I I I I20 40 60 80 100 120 140 160 180 200 220
Frequency, Hz
Figure 36. Response Predictions for the Low.Frequency Improved ModelUsing Measured Input Power Vertical Shake
|
l
l
J
i
i
|M
mg
=
r_
-.,..
.==.
£ iii_
: :::.
=
F_
,==_=
E_
8O
7O
°E'10
_._oco
,=
_o-
Z
40 m
308O
7Orn'10
d
60
i °°4O
3O
Strut
\
I I I I_ I I I I I I
Pressure Bulkhead
- ,_/_v _[_
I I I I I I I I I I8O
m 70 t Section 46 Sidewall
_-
0Z
40
30 I0 20 40 60 80 100 120 140 160 180 200 220
Frequency,HzLegend
SEAmeanSEAmean _ 2 sigma
_ clBaverageof measureddata
80
Section48 Sidewall
7O
ii3o I I I l I I I I I I80
El
Z
Section46 Floor
70_
60--
40_
3O I I I I I I I I I I80
t,i-1Z
30 [ I l I I_ I'\Jl _1 '*_L1FI0 20 40 60 80 100 120 140 160 180 200 220
Frequency,Hz
Figure 37. Response Predictions for the Low-Frequency Improved Model-Fore-Aft Shakeg-UgO 196-42
52
d
50_
Z
40
3O8O
I I I I I I I I I 1
¢0
OZ
Pressure Bulkhead
70--
6(3 --
3o I I I I ] I I ] I [80
80
m 70 1
J
Section 48 Sidewall
60_
50
40
3O
80
Z
7O
50 m
50--
°I30
80
Section 46 Floor
70--m
d
_50-
Z
4O
300
Section 46 Sidewall
l 1 I I I 1 I ].. I I20 40 50 80 100 120 140 160 180 200 220
Frequency. HzLegend
, SEA prediction using measured power input-- dB average of measured data
tn"ID
OZ
70
50 --
Section 46 Interior
%.
I I I 1, I I [ I I I20 40 60 80 100 120 140 150 180 200 220
Frequency. Hz
Figure 38. Response Predictions for the Low-Frequency Improved Mode/Using Measured InputPower Fore-Aft Shake
g*Ug01 @6-43
53
|
U
!m
E
[]
|
i
!m
U
|Im
D
mIzmm
m
mB
m
Ii
|
mmQ
|
l
_c
=,_Lm
= .F
=.++
FII;II_
-20
z
C:
-3om"10
8" -40
-5o
"_. -60
D
-70
8O
z
"o
i-
S.
z
54
Drive Point Conductance
J
] I I I I I I
100
9Om
7o
z 60
5O
Section 48 Sidewall
n
I I I
Bulkhead
60
50
+f3o I I I I I I I I
813
_ 7O
z 40
3O
Section 46 Sidewall
I
7O
60
50
4O
3O
20
t Section 46 Interior Stations 4 to 5
I I ] I I I I I0 50 100 150 200 250 300 350 400 450
Frequency, Hz
LegendSEAM meanSEAM mean _ 2 sigma
+
-- -- Measured data
,lo
+
z
7O /
60 L Section 46 Interior Stations 5 to 7
50_ I
2o I t I I0 50 100 150 200 250 300 350 400 450
Frequency, Hz
Figure 39. Response Predictions, Midfrequency Model Side-of-Body Shake, Improved Model9-U90196R 1-,44
the same area but with the SEA transfer functions. With this procedure, SEA predictions were reason-
able when compared to measured results at the blade-passage frequency, but high at the first harmonic
(fig. 40). This is comparable to results obtained for GVT comparisons. One interesting result of this
study is the small size (2400 in2) of the empennage area used to convert measured pressure to power
input. This result is very similar to conclusions from PAIN studies (sec. 4.11). That is, cabin propeller
noise seems to be controlled by excitation of a small area by intense sound pressure levels opposite the
fan blades. This result could be important to the designer in locating and optimizing acoustic
treatments.
As with finite element modeling, SEA has not demonstrated an ability to predict absolute levels in the
aircraft. However, as a result of these studies, a number of important conclusions were reached. First,
the method can accurately predict energy flow within the structure as well as trends shown with changing
frequency and excitation directions. Subsystems should be reviewed for modal density levels. Estimates
for those subsystems with very low modal densities might be re-evaluated, perhaps by considering
expanding the subsystem size or by considering the effects of neighboring subsystems. A procedure for
validating modal density estimates using subsystem finite element models was developed.
Particular attention should be paid to the subsystem subjected to excitation forces. That is, the input
power calculation has been shown to be critical and a more detailed understanding (than offered by
SEA) of this subsystem may be required for low-frequency predictions. Again, selected use of finite
element models to define or validate SEA parameters may be appropriate.
4.10 PAIN ANALYSIS RESULTS (FY 1987)
Predictions were made and compared with measured propeller-blade-passage frequency interior noise
for three flight conditions. These included both cruise and holding conditions as shown in table 2. En-
gine power was set accordingly.
PAINUDF allows prediction of explicit empennage and passenger cabin structure modes. Structure re-
sponse modes are based on a floor stiffened cylinder for the passenger cabin fuselage or stiffened plates
for the empennage. In principle, power transfer can be calculated on a mode-by-mode basis. In this
analysis, power transfer from the exterior pressure field to the empennage, from the empennage to the
Blade-passage frequencies and harmonics
m
I
i[]
B
J
!
|iiII
iiII
!
II
l
. =
Q
R
162 167
Measured transfer functions
Aft cabin + 2 -7
Forward cabin -5 -7
SEA transfer functions
Aft cabin + 3 -6
Forward cabin + 1 -2
323
+6
-1
+18
+15
333
+3
-1
+16
+15
m
i
Figure 40. Measured Minus Predicted SPL Using Improved Midfrequency Mode/_Ut_196R1--45
55
Il
Q
|im
J
f
3-=._.=F-
l.d
FL_
_=_
E_
passenger cabin fuselage, and from the passenger cabin fuselage to airspace was calculated using the
structural modes. A power flow diagram for PAINUDF is shown in figure 41. The bulkhead junction
between the empennage and fuselage influences the flow of energy between these elements. However, the
bulkhead itself is incapable of transferring power (such as between the empennage and fuselage acoustic
spaces).
For this effort, noise levels measured on the exterior of the empennage of the airplane were used in con-
junction with estimated exterior noise phase behavior based on propeller systems having only one row of
blades. Dummy phase data were required because the density of exterior test microphones was insuffi-
cient to produce accurate measurements. For the first two flight conditions (mach--0.8 and 0.72), the
two rotor speeds were sufficiently separated to extract data for the tone levels associated with the for-
ward rotor only. For the holding condition (mach = 0.42), contributions from the forward and aft rotors
could not be separated. Thus, predictions for each rotor were made using phase estimates for a forward
downsweeping blade row and an aft upsweeping blade row.
Comparison of predicted and measured cabin noise levels is given in figure 42. For the cruise conditions,
agreement between predicted and measured space average values was very good, within 1 dB. At the
holding power condition, a large difference was noted that was attributed to the contribution of the aft
rotor. Time and contract budget constraints did not allow further investigation of this difference. A
fus
Wdis sFuselagestructure"fus"
®
wemp/fus
. emp
Wdiss
®
Empennagestructure"emp"
W prop/emp
wfUS h (_)
W idiss
Cabininterior"i"
Figure 41. Power Flow Diagram for Elements of PAINUDF Model_U_l_R1-46
56
W
check was made that showed no statistical difference (bias) between the measured and predicted results.
The standard deviation was calculated to be 4.7 dB.
I
iFlight condition
Mach = 0.80 at 35 000 ft
Mach = 0.72 at 35 000 ft
Mach =0.42 at 10 000 ft
Rotor
Forward
Forward
Forwardand aft
BPF frequency, Hz
169
170
180
Predicted minus measured levels, dB
0.2
-0.3
Forward 0.2 "] =8.i totalAft 8.0 J
.mi
iFigure 42. Measured Versus PAINUDF Predicted Levels
Q-UgO196R!-47
It was concluded that PAINUDF showed promise in providing models to predict propellei'-blade-
passage frequency interior noise for large commercial transports. Based on the FY 1987 results, a num-
ber of recommendations were made regarding further analytical studies and exploration of potential
PAINUDF improvements. Description of these recommendations follows.
It was recommended that predictions for aft rotor noise levels for the holding condition be reviewed to
determine why predicted levels for the aft rotor were so much higher than for the forward rotor.
Experience gained in using PAINUDF showed that flight measured exterior levels were on a much spars-
er grid than used by the program and not of sufficient quality to obtain phase relationships for program
input. A fair amount of work involving interpolation of amplitude measurements and generation of
dummy phase data from a simple propeller blade row engine model were required to generate program
inputs. A study using the current 727 demonstrator airplane model was recommended to evaluate sensi-
tivity to propeller signature amplitude and phase. This would help define input data requirements for
future applications.
It was noted that the empennage representation in PAIN-tSDF was for the rather unique 727 configura-
tion. It was therefore recommended that a more general conical empennage section model be developed.
While predicted and measured space-averaged acoustic levels were within 1 dB, inspection of the datashowed more than a 10-dB falloff from the back to the front of the compartment (fig. 43). The figure
compares the individual measurement points and their relative location to the predicted and measured
space average. Some estimate of peak levels and distribution would be useful to the designer for
placement of acoustic treatments. Thus, a procedure for estimating noise gradients within an acoustic
airspace was recommended.
4,11 PAIN ANALYSIS RESULTS (FY 1988)
Initial activity focused on understanding the anomalous prediction for the aft rotor contribution to the
interior noise of the demonstrator airplane. It was felt that an evaluation of program coding was
57
II
iI
imiI
i
i
I
iI
I
Ii
I
I
i
I
I
_I
!I
w
I
7. ;
r
L
r_
r--D
L:._9.
E
=
P
|i!'_!!!'..-
E_
_L5
IB)
PAINUDF
prediction
1
II
,!
-.1.
Power average
of seat levels
Seat row number
!2
,
_ | !
.fat ....
.c-O
,,I I I l l I900 950 1000 1050 1100 1150
Body Station
Seat level (1 of 10)
Average of Seat Measurements Versus PAINUDF Space Average PredictionFigure 43.
I .,
1200
g-UgO196R!-4_
warranted. As a result of this evaluation, two coding errors were discovered in the PAINUDF program
and were corrected. The results of the code correction changed levels slightly, but did not significantly
affect the large difference between forward and aft rotors.
From this point, the excitation field characteristics for forward (downsweeping) and aft (upsweeping)
rotors were studied. The geometry for the open, curved panel model of the empennage and the relative
position of the UDF engine is shown from the axial view in figure 44. Note that the line of normal inci-
dence is nearer the center of the panel for blade upsweep than the downsweep. This means that in both
cases, some energy from the excitation field cannot be represented, but for blade upsweep a greater
portion of the excitation field is accounted for. Thus, it is reasonable to expect that the blade upsweep
case would give the higher prediction for this particular geometry.
58
J
/
IDown=weep(+ 1)
Figure 44.
1
wm
IUpsweep (-1)
Propeller Wavefronts for Blade Downsweep (Forward Rotor) and Blade Upsweep (Aft Rotor)at 0 = 105 °
9-U901 _R_-4_
To test this expectation, the axis of the propulsor was rotated downward so that the wavefronts for blade
upsweep and downsweep strike the panel in an identical (mirror image) fashion (fig. 45). Predictions for
this configuration show equal levels for blade upsweep and downsweep. However, predicted levels for
this case were still 5 dB more than those measured on the airplane.
i
!
I
i IDownsweep Upsweep
Figure 45. Propeller Wavefronts for Blade Downsweep (Forward RotoO and Blade Upsweep (Aft Rotor)at 0=80 °
_._01_RF-._
59
m
I
iR
m
i
|
|
|
I
[]U
i
g
i
i
|u
EU
ti
un
B
K...
L_.
=
t_
w
= =
i
Although predicted levels remain high, it was concluded that the model is working correctly. The original
aft rotor prediction, thought to be anomalous, has been shown to be consistent with the forward rotor
predictions and is associated with the position of the rotor with respect to the open panel boundaries of
the empennage representation.
Based on the above results, it was decided that the empennage representation should be improved be-
fore further analysis. On the 727 demonstrator airplane, energy from the propeller blade rows is trans-
mitted through the empennage sidewalls as well as the tail surface and bottom of the empennage, which
are not represented by the open shell structure. A conical empennage section was modeled using local
mass and stiffness properties similar to those of the curved open panels. Note that with the full conical
section the entire excitation field can now be represented regardless of engine and strut angle or upsweep
versus downsweep. Cabin interior noise levels were again estimated using the correct propulsor location.
Results are shown in figure 46 and are compared to baseline results using the corrected code. Note that
the differences between the aft and forward rotor are much smaller (but not zero because of the tapering
geometry of the empennage versus blade row position). Also, the aft rotor prediction is now independent
of the direc:ion of rotation. Note also that the expected increase in level, due to the conical structure
allowing full representation of the excitation field, is more than offset by the increased stiffness of the
conical structure. That is, the stiffer (conical versus open panel) geometry results in a lower modal densi-
ty and less efficient coupling with the exterior pressure field.
Because measured phase values are often difficult to obtain or estimate, it was desirable to determine
the level of accuracy necessary for phase description. For a preliminary assessment, predictions using
the model with the open panel empennage representation were made for constant phase, dummy phase,
and rapidly varying phase. Predictions varied dramatically as seen in figure 47. The case of constant
phase yields the highest interior levels and the case for rapidly varying phase yields the lowest
Predicted minus measured levels, dB
E
Flight condition
Mach = 0.80 at 35 000 ft
Mach = 0.72at 35000 ft
Mach = 0.42 at 10 OO0 ft
Mach = 0.42 at 10 0(30 ft
Mach = 0.42 at 10 000 ft
Rotor
Forward
Forward
Forward
Aft
Aft
Rotation
Downsweep
Downsweep
Downsweep
Upsweep
Down,weep
Open empennage
1.1
0.1
2.1
8.0
Conical empennagei
-5.3
-5.9
-4.6
-2.8
-2.8
::_=:
Figure 46. Measured and PAINUDF Predicted Levels for Open Pane/and Conical Empennage Models9-UgOlg6R!-51
w
Flight condition
Math = 0.80 at 35 0(30 ft
Mach = 0.72 at 35 000 ft
Mach = 0.42 at 10 000 ft
Mach = 0.42 at 10 (300 ff
Rotor
Forward
Forward
ForwardAft
Rotation
iDownsweep
Downsweep
Downsweep
Upsweep
Predicted minus measured levels, dB
Dummy phase
1.1
0.1
2.1
8.0
Constant phase
8.g
7.8
10.0
11.3
Rapidly varying phase
-13.1
-23.2
-14.1
-6.6
m
_7
6O
Figure 47. Measured and PAINUDF Predicted Levels Phase Sensitivity Study9-UgO'r 96RL52
predictions. Although constant phase over the entire grid is unrealistic, it appears that significant energy
flow from the propeller field to the empennage can be expected from regions where phase varies slowly
and pressure levels are high.
It should be noted that the very stiff empennage structure representation results in low modal densities
(long structural wavelengths) in the frequency range of interest. As a result, this particular model is very B
sensitive to phase variations. Further sensitivity studies to phase variation and receiving structure stir-2
fness are presented in following sections.i
It was felt that to improve 727 demonstrator airplane predictions an accurate estimation of the excitation m
field phase characteristics would be required. To this end, NASA supplied such data using Aircraft MNoise Prediction Program (ANOPP) predictions for the test propeller engine configuration. The =
ANOPP predictions showed more rapidly varying phase than the dummy data, particularly forward of
the propeller plane (fig. 48). A series of predictions were made using the 727 demonstrator airplane mod-
el with conical empennage, measured propeller excitation amplitudes, and either dummy or ANOPP
phase data (fig. 49).
As expected, the prediction using the more rapidly varying ANOPP data is much lower than the predic- R
tion using dummy phase data (case 1 versus case 2). To study phase effects further, predictions using only
Forward rotor prediction 0,8 math,
BPF = 16g Hz
Dummy phase data i___
....... ANOPP phase data Forward A.ftprop I. prop plane plane
(90 line Increments) .,. ,. ..................
/ / f
W "'- .... -II
iFigure 48. Comparison of Propeller Field Phase Contours ANOPP Phase Versus Dummy Phase
10-UgDlg6R1-53
|m
B
61
Ji
II
=
r
w
"Z_ 2
K.a
FE
_4W
w
E
r_
F_
m
E
Case
1
2
3
4
5
6
Phase data
Dummy
ANOPP
Dummy
ANOPP
Dummy
ANOPP
Grid size
Full
Full
Rear-haft
Rear-half
Full
Full
Receiving structure
Empennage
Empennage
Empennage
Empennage
Fuselage
Fuselage
Predicted minus
measured levels, dB
-4.6
-22.9
-11.1
-23.6
13.1
10.2
Figure 49. Measured and PAINUDF Predicted Levels PAINUDF Sensitivity Study to ExcitationModels (Mach = O.80 at 35 000 ft)
10-UgO196R 1-54
the rear-half of the excitation grid were made. For the ANOPP phase data where phase changes rapidly
in front of the propeller plane, nearly all the energy enters through the rear-half of the grid (case 4). For
the dummy phase data the effect is not nearly so dominant (case 3).
Finally, the effect of modal density was studied. As noted earlier, very stiff structures, such as the empen-
nage, will have low modal densities (long structural wavelengths), which will couple most efficiently with
excitation fields of slowly varying phase. Predictions were made by subjecting the fuselage section direct-
ly to the excitation field. The fuselage is much less stiff than the empennage. Cases 5 and 6 of figure 49
now show much less sensitivity to phase than before. Levels are significantly higher because losses
through the empennage structure are no longer accounted for.
Without additional test data to validate excitation field predictions it was felt that further attempts to
improve the PAINUDF 727 demonstrator airplane model would not be useful. The above studies show
that interior noise levels on propeller aircraft can be extremely sensitive to the excitation field phase and
the receiving structure stiffness (or modal density). The analyst must be cognizant of the modal charac-
teristics of the aircraft structure in the frequency range of interest and take pains to describe the excita-
tion field in corresponding detail.
4.12 PAIN IMPROVEMENTS AND FEASIBILITY STUDIES
The second part of the 1988 effort focused on expanding capabilities of the PAIN program. As a first
step, a five layer passenger cabin trim was incorporated. The original two layer model allowed an airgap
oi" layer of insulation and a limp mass trim. The five layer model allows combinations of insulation, air-
gaps, and septa (with damping). An example of a five layer trim is given in figure 50.
A users manual was written and delivered to NASA that describes the aft propulsor version (PAINUDF)
as well as the five layer cabin trim option.
62
w
J
Element
Skin Airgap 1
2 3
I Insulation ]
Figure 50.
4
Septum
/
Airga_o2
p Trimpanel(septum)
Airgap 1 = element 1 = type 2Insulation = element 2 = type 1Septum = element 3 = type 3Airgap 2 = element 4 = type 2Trim panel = element 5 = type 3
S-Layer Trim (Example)
Examp/e of S/dewa// Layer Conf/gurat/onslO-U_OI_6R 1-55
A number of potential features were studied that could further enhance the capability of the PAIN mod-
el. These features were bulkhead transmission and radiation, exterior pressure loading fore and aft of
the pressure bulkhead, cabin noise caused by engine unbalance, and cabin noise gradient.
To add radiation by the bulkhead, the bulkhead must be included as a fourth element of the system. A
fifth element, the acoustic cavity within the empennage must also be accounted for if transmission
through the bulkhead is to be included. It was shown to be feasible to include the bulkhead as an inde-
pendent element. It is also considered possible to include the bulkhead as part of the integral fuselage-
bulkhead-empennage composite structure as was done in the original PAIN formulation for the fuselage
and floor. Based on the expectation that low-frequency global modes will be important at propeller tone
frequencies, the latter method was recommended.
Two methods for accounting for pressure loading fore and aft of the bulkhead were outlined. In the first
method, it was pointed out that PAINUDF can currently run twice with separate pressure fields on the
fuselage and empennage elements. The resulting interior space average sound pressure levels (SPL) can
then be summed. In the second method, a power balance could be formulated on the three elements of
the PAINUDF system (the empennage, fuselage, and cabin volume). Because it has been shown that the
majority of power inflow occurs over small sidewall sections local to the propeller plane, no further work
in this area was recommended. PAINUDF should be run twice for those cases when significant power
inflow is expected through both the empennage and fuselage surfaces.
PAINUDF can be modified to predict cabin levels caused by engine unbalance in principle by simply
including the strut as an element just as the bulkhead would be included in the first study. However, the
difficulty in calculating the resulting modes of integral composite structure and the low modal density at
63
t
i[]
i
i
|I
i
;
i
i
ih
i
i
lid
=
w
L
engine once-per-revolution frequencies make this enhancement of questionable value and therefore, it
was not recommended.
The interior sound levels measured during flight varied significantly along the axis of the demonstrator
airplane fuselage (fig. 43). Therefore, the feasibility was considered of accounting for the sound level
gradient within the PAIN program. In order to account for the gradient, the power balance formulation
presently used would be replaced by a general solution in which a cavity Green's function is defined with
axial dependency. Space averages will then only be taken over the cross section.
U
ID
L
L_
L_
D
w
U
r_
m
64
_I
w
5.0 CONCLUSIONS AND RECOMMENDATIONS
At the end of the first year's effort, it was concluded that each analysis procedure showed promise for use
in predicting low-frequency noise associated with advanced propeller engine installations. It was noted,
however, that variations between predicted and measured levels were large enough to remain cautious.
The goal of the second year's effort was to explore modeling and analysis techniques to improve results.
Assessment of the second year's results indicate that predictions of absolute levels, with accuracy typi-
cally required to make cabin noise guarantees, have not been demonstrated. However, the models have
demonstrated the ability to predict trends and should be useful to a designer in assessing type and place-
ment of suppression treatments. Thus, it is recommended to proceed by using the models to identify
engine vibration and propeller noise transmission mechanisms and assess potential suppression con-
cepts. Other conclusions are summarized in the following sections.
5.1 FINITE ELEMENT ANALYSIS
One of the initial objectives of the finite element modeling task was to demonstrate that loads analysis
models could be modified for noise analysis at engine rotor frequencies. If possible, this would save a
large amount of modeling effort. The 727 demonstrator airplane model was developed from such a loads
model by adding mass elements for the structure and trim and a finite element representation of the
cabin airspace. The general agreement with measured GVT levels and the ability to predict response
sensitivity to different excitation directions indicate this objective has been met.
The 727 demonstrator airplane finite element model was found to be time consuming in terms of model
checkout and turnaround of predicted results due to its large size. Studies were completed in which the
model size was reduced by simplifying and then eliminating representation of wing, tail, and empennage
structure. Results indicate that whole body and wing modes can contribute to transmission and radi-
ation of engine rotor noise. Thus, some representation of all of the aircraft structure is required. Beamelements for structure not in the direct transmission path were found to be satisfactory. Inclusion of the
baggage compartment airspace did not significantly affect results.
Studies with the model for flight predictions show that the phase relationship between multiple input
forces can shift predicted levels by 20 dB or more. This result has been demonstrated independently on
flight tests of conventional turbofan aircraft and suggests that engine response modes can couple effec-
tively with strut modes to transmit vibration at engine rotor frequencies.
These results indicate that the analyst should have some understanding of potential engine and aircraft
modes in order to properly model the structure and assess cabin noise. It is noteworthy that SEA results,
described below, support these conclusions.
U
= =l
I
U
M
I
i
E
g
i
I
II
|il
MI
!!
mE1
m
|i
g
[]m
N
ig
[]
i
65
i
|
D
5.2 STATISTICAL ENERGY ANALYSIS
z
m
= =
m
U
w
H
U
K
_.m
H
Studies have shown that the complex structure and low modal densities of the strut and empennage
precluded accurate SEA predictions of input power and therefore, cabin noise levels. When measured
input power from GVT tests was used, the models were shown to accurately predict power flow through
the aircraft to the cabin. Procedures for estimating modal densities and input power using finite element
analysis were developed and demonstrated. It now appears that development of selected small finite
element models to determine input power coupled with SEA models to predict transfer functions to the
passenger cabin forms a viable prediction procedure. In this way, a relatively small model can be devel-
oped that allows noise control engineers to evaluate and influence structural design before configura-
tions are frozen. Final configuration details would be evaluated using finite element models.
Propeller tone vibration and noise levels in flight were overpredicted throughout the aircraft. This led to
the conclusion that input power was being overestimated. This was confirmed by reducing the empen-
nage excitation area used to convert measured pressure to force inputs so that measured empennage
response was matched, resulting in good agreement at all other locations. Another conclusion from this
exercise is that a significant portion of the propeller tone energy appears to enter through a relatively
small portion of the empennage. Studies using PAIN confirm this conclusion and are described below.
Finally, it was noted during these studies that the low-frequency model provided relatively good predic-
tions up to 200 Hz. This implies the more detailed subsystem representation in the midfrequency model
may not be required for predictions at propeller-blade-passage frequencies. Further work in this area is
recommended.
5.3 PAIN ANALYSIS
At the end of the first year's work, an apparent anomalous result regarding prediction of tone level_ from
the aft rotor was noted. These results were explored and found to be consistent with forward rotor pre-
dictions. This is a result of excitation field phase differences for forward (downsweep) and aft (upsweep)
blade rows mapped over the open panel representation of the empennage. The open panel empennage
was thus replaced by a conical representation.
Further studies with this model indicated that cabin levels were very sensitive to the excitation field
phase and that by varying phase, the measured flight tone levels could be bracketed. Because test mea-
surements were insufficient to describe excitation field phase (dummy values were used for all cases),
further efforts to match predicted and measured levels were discontinued. As a result of these studies
however, an important conclusion was reached. As with SEA results, it appears that most propeller tone
energy enters the aircraft through a relatively small area of the empennage where the excitation field
phase varies slowly. This is important because it implies that noise suppression treatments, which are
inherently heavy and expensive, can be concentrated over small areas where power flow is the greatest.
The empennage is a stiff structure with low modal density (long structural wavelengths). When the softer
fuselage structure was evaluated (by exposing it directly to the propeller noise field), it was found to be
66
M
U
I
less sensi',:ve to phase of the excitation field. That is, the higher modal density of the fuselage allowed
efficient excitation of several modes regardless of the propeller field phase. As with finite element analy-
sis and SEA, this indicates that for low-frequency evaluation of aircraft structures, some knowledge of
modal behavior is required.
m
U
m
[]
![]
|[][]
|
|
|
MIiI
uB
I
ll
[]
[]
[]II
im
1167
iNIm