Vous êtes sur la page 1sur 25

Journal of Sound and Vibration (1995) 180(4), 583607

PREDICTION OF STRUCTURE-BORNE SOUND


TRANSMISSION IN LARGE WELDED SHIP
STRUCTURES USING STATISTICAL ENERGY
ANALYSIS
P. H.NN:, P. KiiNcr :Nb J. VioisiNrN
Technical Research Centre of Finland, Ship Laboratory, P.O. Box 114, SF-02151 Espoo,
Finland
(Received 21 February 1992, and in nal form 29 November 1993)
An ecient method is presented for the prediction of structure-borne sound transmission
in large welded ship structures. SEA (Statistical Energy Analysis) is used, and the equations
used for the SEA parameters are also presented. Traditionally, the SEA method requires
a great deal of work when steel structures are modelled. It is almost impossible to prepare
models manually for large structures such as ships. In the method developed, the
preprocessing programs used in the context of the nite element method (FEM) are applied
to reduce the modelling work. To date, one-dimensional beam, two-dimensional triangular
and quadrilateral plate, and three-dimensional volume elements have been implemented.
The assemblage of the loss factor matrix is made in a manner analogous to the stiness
matrix in FEM. In the computer implementation of the SEA program, standard FEM
techniques are used to reduce calculation time, including the skyline matrix technique and
LDL
T
-matrix decomposition of the loss factor matrix. The eectiveness of the present
method is illustrated by the computer run of the model of the passenger cruise vessel, which
contained over 5000 elements and 17 000 coupling branches. It took only 17 min in the Iris
Indigo R4000 workstation. Application calculations for an echo sweeping vessel, a
timber-container carrier, and a passenger cruise vessel are discussed, and comparison is
made with full scale measurements.
1. INTRODUCTION
The noise level criteria of the International Maritime Organization (IMO) [1] were
accepted during the 1980s in many countries by local authorities, national organizations
and shipowners. At the same time, requirements concerning greater ship power connected
with reduced weight and better economy tended to increase noise levels [2]. These often
conicting requirements led to extensive research and development work, producing
accumulated knowledge and possibilities for putting into practice low cost acoustically
ecient solutions [3]. Noise prediction in the ship context requires consideration of the
noise sources, the transmission paths along the ships hull and the receiver spaces.
Structure-borne sound transmission, being able to carry sound to remote places, is often
the main factor. Its contribution can be predicted by using semi-empirical or analytical
methods, such as the wave guide method or the SEA method.
Semi-empirical methods (see, e.g., references [49]) used to estimate the structure-borne
sound transmission in a ship require data on the most important noise sources, including
diesel engines, gears, propellers, bow thrusters, as well as on isolation measures, e.g.,
resilient mountings, compliant layers and oating oors. In these methods the estimation
583
0022460X/95/090583 +25 $08.00/0 7 1995 Academic Press Limited
i. n.NN: ET AL. 584
of the decrease in vibration level along a ships hull is based on measured data. It can be
seen as a set of model rules able to transform the acoustics of previous ships into that of
future similar vessels [4]. Often, prediction is made in octave bands with centre frequencies
from 315 Hz to 8 kHz. For a ship which is near the statistical average used in the
calculation, standard deviations between measured and predicted data are a maximum
45 dB for octave bands or 35 dB for the A-weighted sound pressure level in
accommodation spaces [5]. The other approach is to use a statistical model of multiple
regression, in which many descriptive variables, obtained mainly by measurements
onboard ships, are used to predict noise spectra in the superstructure of a ship [10]. The
standard deviation of this method is 3045 dB for octave bands with centre frequencies
from 63 Hz to 2 kHz, thus assuring a prediction accuracy of the order of 268 dB in the
95% condence interval [10]. Also the optimum linear parameter estimation theory has
been successfully tried for predicting the air-borne noise in ship superstructures [11]. In
general, semi-empirical methods can give good results for future similar ships, but they
can fail when predicting noise levels for new ship types because they have not the ability
to predict the eect of structural changes.
The waveguide model [12, 13] can be used when estimating the structure-borne sound
transmission in the vertical section of a ship structure. The methods [12, 13] have been
developed further to include frequencies below the cut-o frequency, typically 100 Hz in
a ship, of the waveguide system with use of the Galerkin variational method [14]. Thus,
by using these two methods, noise prediction can be made in the octave bands with centre
frequencies from 315 Hz to 8 kHz. The propagation of structure-borne sound in the
longitudinal direction is estimated in these methods by using experimentally determined
transmission loss data (see reference [14]). The relative importance of exural and
longitudinal waves in power transmission was discussed in reference [12]. The results of
this study showed that the main power transmission was by exural waves in the ship
structure. This means also that conversion to other wave types during propagation is of
less importance. The accuracy of predicted A-weighted sound pressure levels in cabins is
in most cases better than 23 dB and the standard deviation can be expected to be less
than 15 dB [14]. With this method the relative importance of the various noise sources
and structures can be evaluated. It is also possible to determine the eects of structural
changes in constructions.
SEA began to develop in 1959, when Lyon and Smith independently made their rst
calculations concerning power ow and response of linearly coupled resonators [15]. In
1968 Budrin and Nikiforov proposed a method based on energy balance between
substructures to calculate the structure-borne sound transmission in a ship [16]. This was,
according to reference [6], the rst attempt to apply SEA in the ship context. Sawley was
among the rst to apply the SEA method to ship noise problems [17]. After 1970 a rapid
period of development followed with new SEA applications: see, e.g., references [1822].
The rst applications were to structures and scale models typical of a ship. Thereafter
larger ship structures were analyzed and, also, special problems were considered in more
detail: see, e.g., references [2326]. In addition, a model of a whole ship containing 327
elements was used to predict structure-borne sound transmission when predicting the
sound pressure level in cabins [27]. To data, the largest model reported has contained 1647
elements and 7496 coupling branches [28].
Up to now, eective methods of modelling a large steel structure for SEA applications
or the FEM-type assemblage of the loss factor matrix and solution of the SEA equation
have not been presented. The rst aim of the work described in what follows here has been
to reduce the amount of work required to model a large steel structure of a ship for SEA
calculations and to make possible applications for large structures comprising thousands
s1ric1irr-norNr soiNb iN sniis 585
of elements. The second aim has been to test how the SEA method works in the ship
context when large models and even low frequencies are used. The modelling method and
the equations of SEA parameters used are described in detail. In addition, application
calculations for a hydrographic echo sweeping vessel, a timber-container carrier, and a
passenger cruise vessel are discussed, and a comparison is made with measured results. The
methods used to reduce computer time are also briey discussed.
2. SOME BASIC CONCEPTS OF SEA
2.1. crNrr:i
In the SEA method, systems are considered to be divided into subsystems, which are
linearly coupled together, and which exchange energy via resonant vibration modes. In
many applications geometrical structural elements are the subsystems, while sometimes
dierent wave types, e.g., bending or longitudinal waves, form the subsystems. The net
amount of energy input to each subsystem, which comes from external excitation forces
and from couplings with other subsystems, equals the energy dissipated in this subsystem.
In addition, there are several basic assumptions [15]: there is linear coupling between the
subsystems; energy ows between oscillator groups via resonant vibrations in the frequency
band considered; the oscillators are excited by broadband random excitations; each of the
modes is equally energetic within a frequency band in a subsystem; and the ow of energy
between subsystems is proportional to the modal energy dierence.
2.2. iovrr n:i:Ncr rqi:1ioNs
For the two groups of oscillators or for the two-element SEA model (see Figure 1), where
elements are excited by statistically independent broadband sources, the power balance
equations are [15]
P
in 1
=vh
1
E
1
+vh
12
E
1
vh
21
E
2
, P
in 2
=vh
2
E
2
+vh
21
E
2
vh
12
E
1
, (1, 2)
where P
in i
is the time-averaged input power to element i, v is the angular frequency of
the band of interest, h
i
is the internal loss factor of element i, h
ij
is the coupling loss factor
from element i to element j and E
i
is the time-averaged total energy (sum of potential and
kinetic energy) of element i. In matrix form the power balance equations (1) and (2) can
be presented as
v
$
h
1
+h
12
h
12
h
21
h
2
+h
21%6
E
1
E
27
=
6
P
in 1
P
in 27
. (3)
An important relationship in SEA is the reciprocity relationship [15],
h
ij
n
i
=h
ji
n
j
, (4)
where h
ij
is the coupling loss factor from element i to element j, and n
i
is the modal density
of element i, respectively for h
ji
and n
j
.
In the steady state single-frequency vibration of an individual oscillator, the input power
has to be in balance with the power dissipated. The time-averaged power dissipated P
d
is
Figure 1. A two-element SEA model.
i. n.NN: ET AL. 586
related to the time-averaged sum of the kinetic and the potential energy E stored in the
oscillator via the damping [29],
P
d
=c
v
x
2
=2zv
n
mx
2
=2zv
n
E=v
n
E/Q=v
n
Eh, (5)
where c
v
is the viscous damping coecient, x is the vibration velocity, z is the ratio of the
viscous damping coecient to the critical viscous damping coecient, v
n
is the natural
angular frequency, m is the mass of the oscillator, E is the time-averaged stored energy,
Q is the quality factor, and h is the internal loss factor. The concept can be extended to
a collection (group) of oscillators in a frequency band [29]: i.e.,
P
d
=vE/Q=vEh, (6)
where v=2pf, f is the centre frequency of the band, and h is now the mean loss factor
of the modes within the band.
In the general case of N oscillator groups, there are N simultaneous power balance
equations. These can be presented in matrix form, where the loss factor matrix is
symmetrical owing to the reciprocity relationship (4) as [29, 30]
K LF J
0
h
1
+ s
N
i $1
h
1i
1
n
1
h
12
n
1
h
13
n
1
h
1N
n
1
E
1
n
1 G GG G
G GG G
G GG G
h
21
n
2
0
h
2
+ s
N
i $2
h
2i
1
n
2
h
23
n
2
h
2N
n
2
E
2
n
2 G Gj f
G GJ F
v .
.
.
.
.
.
.
.
.
G GG G
G GG G
h
N1
n
N

0
h
N
+ s
N
i $N
h
Ni
1
n
N
E
N
n
N
G GG G
k lf j
P
1
P
2
=g
G
G
G
G
F
f
.
.
.
h
G
G
G
G
J
j
, (7)
P
N
where E
i
and P
i
are the time-averaged total energy and the input power of the element
number i, respectively. Noise analysis by using equation (7) requires that the angular
frequency v, the input powers and the SEA parameters (the modal densities, the internal
loss factors and the coupling loss factors associated with oscillator groups), are known.
These oscillator groups are subsystems, which are suitably selected (see, e.g., references
[15, 29]). In general, weak coupling between them (h
ij
Wh
i
and h
j
) is required. In addition,
it is assumed that most of the energy exchange is by resonant modes of subsystems. Solving
equation (7) at the angular frequency v concerned, one obtains the time-averaged energies
E of dierent oscillators.
2.3. sr: i:r:xr1rrs
2.3.1. Modal density
The mode count N, which is the fundamental quantity, is, for a subsystem, the number
of modes of that subsystem that resonate in the frequency band Df considered. It may
sometimes be estimated as a product of a modal density, n( f ), and the frequency
s1ric1irr-norNr soiNb iN sniis 587
bandwidth Df. As a derived quantity, the modal density is dened as the number of modes
per unit frequency [15]. Sometimes it is given per unit angular frequency as n(v), these
being related by
n( f ) =2pn(v). (8)
Theoretically derived modal densities are available in the literature (see, e.g., references
[15, 3032]) for idealized structural elements. Here only the analytical modal densities for
the structural elements used in the program package are given.
Modal densities of uniform beams in exural vibration are given by [15, 32]
n( f ) =L/2pc
B
={L/(2pf )
1/2
}(rS/EI)
1/4
, (9)
where L is the length of the beam, f is the frequency, c
B
=(2pfk
b
c
L
)
1/2
is the phase velocity
for exural waves, k
b
=(I/S)
1/2
is the radius of gyration, c
L
is the longitudinal wave velocity
for the beam, r is the density of the material, S is the cross-sectional area of the beam,
E is Youngs modulus of elasticity and I is the second moment of area of the beams
cross-section.
Modal densities n( f ) of uniform at plates in exural vibration are given by [15]
n( f ) =Sz12/2c
L
h, (10)
where S is the surface area of the plate, h is the thickness of the plate and c
L
is the
frequency-independent, quasi-longitudinal phase speed in the plate [32, 33]
c
L
=[E/r(1 n
2
)]
1/2
, (11)
where E is Youngs modulus of elasticity, r is the density of material and n is the Poisson
ratio. For steel one obtains from equation (11) c
L
=5270 m s
1
by using the values
E=2 10
11
N m
2
, r =78 10
3
kg m
3
and n =028; for other materials see references
[32, 33]. The phase speed of longitudinal waves c
L
is frequency independent, whereas the
phase speed of bending waves depends on frequency and is thus dispersive. For a thin plate
of thickness h the phase speed is obtained from [32, 33] c
B
=(2pf )
1/2
(B/rh)
1/4
1(18c
L
hf )
1/2
,
where the bending stiness B=Eh
3
/12(1 n
2
). For example, for a steel plate of thickness
10 mm the phase speed c
B
=98 m s
1
when f =100 Hz, and c
B
=310 m s
1
when
f =1 kHz. Equation (10) is used also for a triangular plate element, because modal density
is an additive property [15] and directly proportional to the surface area of the plate
element.
Modal densities n( f ) of a three-dimensional acoustic volume element are given by [15]
n( f ) =(4pf
2
V/c
3
) +(pfS/2c
2
) +(l/8c), (12)
where f is the frequency, V is the volume of the element, c is the speed of sound in the
uid (air), S is the total wall area of the element and l is the total edge length of the element.
The density of air and the sound velocity in air at the standard air pressure [34] at a given
Celsius air temperature and at a given relative humidity are calculated according to
references [35, 36]. This kind of accuracy is not of great importance, although subroutines
needed in other calculations were utilized also in this SEA program.
The mode count N will not deviate very much from the estimate as a product of a modal
density n( f ) and the frequency bandwidth Df, if the modal density is great enough or the
bandwidth is wide enough for the mode count estimate to be at least ten modes or so [15].
If a subsystem has a small mode count, it may still be modelled as a SEA subsystem, even
though the concept of modal density is inappropriate in this case [23, p. 198]. McCollum
and Cuschieri have shown in a previous paper that SEA provides a good approximation
to the mean level of the energy transmission or response even in those frequency regions
i. n.NN: ET AL. 588
where the modal density is low and the response of the structure is dominated by a few
resonances [26]. In building acoustics it has been found for plates that the condence limits
derived by Lyon [15] for SEA predictions considerably overestimate the actual error [37].
However, comparison with measured results gives the condence limits of estimates. In
building acoustics it was found that if the error of 3 dB in the prediction of sound
transmission was taken as a limit, then the requirement would be for the mode count
N=n( f )Df q05 and for the modal overlap M=fhn( f ) q04 for plates in one-third
octave bands, where h is the total loss factor [37]. As to the walls and oors of buildings
the modes have a large bandwidth, which typically may be half a one-third octave band
at 100 Hz. Thus there is a physical meaning to 05 modes per band, since a mode will often
have a signicant response reaching to more than one contiguous band [37]. This also
means that there are interacting modes of subsystems in contiguous frequency bands
exchanging energy even when the modal density Q1. In the ship context the global
behaviour of a large ship hull extends to frequencies of about 1015 Hz. Thereafter
substructures, such as decks and bulkheads, are excited into resonant vibrations. The
drastic increase in modal density for the entire structure in a ship hull structure happens
at a frequency of about 50 Hz [21, p. 109].
The requirement that mode count should be about ten means that suciently large
structural and acoustic elements should be selected and that the frequency bandwidth
considered should be wide enough. For example, if it is required that the mode
count within the band N=n( f )Df =10, this means that the area of a plate element from
equation (10) S=5z12c
L
h/(3Df ) should be 26.5 m
2
at the 50 Hz one-third octave band
and 68 m
2
at the 63 Hz octave band for a 10 mm thick steel plate. If in a SEA model the
length of a plate element is eight frame spacings of 06 m then the breadths should be 55 m
and 14 m, respectively. These breadths are usually easily obtainable in practice in the
modelling. The bandwidth between the upper and lower limiting frequencies has the
following values: Df 1023f
m
for a one-third octave band and Df 1071f
m
for an octave
band lter, where f
m
is the mid-band frequency of the one-third octave or octave band.
When the height between decks is 27 m, then the acoustic volume element with
V=55 48 27 m
3
will have n( f ) 1015 modes/Hz, giving N=17 in the 50 Hz
one-third octave band. This is clearly a small number, although acoustic volume elements
are used if the air volume is modelled or if the air-borne input power must be taken into
account.
When the calculation includes low frequency bands, the mode count within a band
clearly is less than ten for many subsystems used in the modelling. This is a relaxation of
the Lyons previous requirement [15]. However, theoretically calculated estimates of modal
density for ideal structural elements are used in the program package instead of measured
values. The estimates of sound pressure or vibration velocity levels thus obtained should
be compared with measured values, as has been done in building acoustics [37]. In practice,
many other simplications are also made during the modelling of a ships steel structure
with ideal triangular or quadrilateral plate and beam elements.
2.3.2. Internal loss factors
The loss factor is proportional to the ratio of energy dissipated per cycle to the
time-averaged energy stored [15]; see equation (5). Sometimes it is dened as the phase
angle of a complex Youngs modulus of elasticity [32, 15]. The internal loss factor of a
structural element includes several dierent damping or energy-loss mechanisms. Com-
monly accepted forms of linear damping are structural (hysteretic or viscoelastic) damping
and acoustic radiation damping [29]. In practice, other non-linear damping mechanisms
are also present at the structural joints. These include gas pumping, squeeze-lm damping
s1ric1irr-norNr soiNb iN sniis 589
and frictional forces [29]. The internal loss factor of a structural element forming part of
a built-up structure is given by [29]
h =h
s
+h
rad
+h
j
, (13)
where h
s
is the structural loss factor, h
rad
is the radiation loss factor and h
j
is the loss factor
associated with energy dissipation at the boundaries of the structural element. Typically,
engineering structures are lightly damped and 25 10
4
Qh Q50 10
2
[29, 33]. For
most structures h tends to decrease with frequency, roughly as f
1/2
[33]. Theoretical
estimates of loss factors are not generally available for structural elements. In practice,
measured values are used (see, e.g., references [15, 29, 32, 3840]). In this context it is very
important to know the measurement conditions. This means that one must know what
components are included in equation (13). For many engineering structures h
j
is zero and
h
rad
may or may not be included. When only h
s
is required, it is measured in a vacuum.
This point is not always clearly stated when measured results are presented, although for
very thin plates the radiation loss factor is of the same order as the internal loss factor.
An approximate expression for the internal loss factor of the steel plate, where the
dissipation due to sound radiation is excluded, is given in reference [41],
h =041f
07
, (14)
where f is the frequency in Hz; from equation (14) h =0016 when f =100 Hz, and
h =00032 when f =1 kHz. The radiation loss factor is given by [15]
h
rad
=r
0
cs/vr
s
, (15)
where r
0
is the uid density, c is the speed of sound in the uid, s is the radiation ratio
of the structure, v is the centre frequency of the band and r
s
is the surface mass of the
structure.
The internal loss factor for an acoustic volume element is obtained from [42]
h =(c/pf )[a
i
(S/8V) ln (1 a)], (16)
where c is the speed of sound in air, f is the frequency, a
i
is the total pure tone atmospheric
absorption coecient (nepers per metre), S is the total surface area of the volume element,
V is its volume and a is the average sound absorption coecient of the conning surfaces
of the element, excluding the eects of sound transmission through the surfaces. The
atmospheric absorption coecient (dB per metre) is calculated by using the equations
presented in reference [43]. Some representative values at a pressure of one standard
atmosphere (101325 kPa) when the air temperature is +20C and relative humidity is 50%
are a
i
=0466 dB/100 m when f =1 kHz, and a
i
=161 dB/100 m when f =10 kHz.
Equation (16) includes the atmospheric absorption and the absorption at the conning
surfaces of the volume element. The atmospheric absorption is small, especially at low
frequencies, and often it is not necessary to take it into account.
2.3.3. Coupling loss factors
The coupling loss factor, h
ij
, is related to energy ow from subsystem i to subsystem
j. Theoretical expressions for couplings between dierent structural element types are
available: see, e.g., references [15, 31, 32]. The coupling loss factor of a beam cantilevered
to a at plate is given by [44]
h
bp
=
(2r
b
c
Lb
k
b
S
b
)
2
vm
b b
Z
p
Z
p
+Z
b b
2
Re (Z
1
p
), (17)
i. n.NN: ET AL. 590
where h
bp
is the coupling loss factor from the beam to the plate, r
b
is the density of the
beam material, c
Lb
is the longitudinal wave speed in the beam, k
b
=(I
b
/S
b
)
1/2
is the radius
of gyration of the beam, I
b
is the second moment of area of the beams cross-section, S
b
is the cross-sectional area of the beam, v is the angular frequency, m
b
is the beam mass,
and Z
p
is the moment impedance of the plate, given by [44]
Z
p
=16r
s
k
2
p
c
2
Lp
/v(1 +iG), (18)
where r
s
is the surface density of the plate, k
p
=h/z12 is the radius of gyration for the
plate cross-section of thickness h, c
Lp
is the longitudinal wave speed in the plate, i =z1,
and [45]
G=(4/p) ln (11/k
p
a), (19)
where k
p
is the bending wavenumber of the plate and a is the eective distance of the pair
of point forces making up the moment on the plate. For rectangular and circular beam
cross-sections, respectively, a
r
=d/3 and a
c
=059r, where d is the side dimension of the
rectangular beam cross-section (in the direction of bendings) and r is the radius of the
circular beam cross-section. The moment impedance of the beam is [44]
Z
b
=r
b
c
2
Lb
S
b
k
2
b
k
b
v
1
(1 +i), (20)
where r
b
is the density of the beam material, S
b
is the cross-section and k
b
is the exural
wavenumber of the beam. The coupling loss factor h
pb
from plate to beam is calculated
by using the reciprocity relation (4).
The coupling loss factor for a line junction, which is often encountered in a welded
structure such as a ship, is obtained from [15, 32]
h
ij
=2c
Bi
Lt
ij
/pvS
i
, (21)
where c
Bi
is the bending wave velocity (or phase velocity) of exural waves in the rst
subsystem, L is the length of the coupling line, t
ij
is the power transmission eciency
corresponding to the wave type and the type of junction in question, v is the centre
frequency of the band and S
i
is the surface area of the subsystem i. Transmission
eciencies t
ij
for pure exural waves of direct incidence are calculated by using the
equations presented in references [32, 18]. The equations for normal incidence of exural
waves to the straight-line junction presented in reference [18] for a right-angle junction,
a T-junction and a right-angle cross-junction between plates are used for calculation in
the program package because of their simplicity. Of course, more exact equations can
easily be implemented should it be necessary to obtain better accuracy.
The coupling loss factor h
ij
from the structural element i to the acoustic volume element
j is given by [15]
h
ij
=r
0
cS
i
s/vm
i
, (22)
where r
0
is the uid density, c is the speed of sound in the uid, S
i
is the surface area of
the structural element i, s is the radiation ratio of resonant vibration, v is the centre
frequency of the band and m
i
is the mass of the structural element i. In the other direction,
the reciprocity relation (4) is used. The radiation ratio is calculated in the program package
developed according to Maidanik [46].
The coupling loss factor between acoustic volume elements is [15]
h
ij
=cSt
ij
/4vV
i
, (23)
where c is the speed of sound in the uid, S is the surface area of the panel between spaces,
t
ij
is the sound intensity transmission coecient (sound reduction index R=10 log (1/t))
s1ric1irr-norNr soiNb iN sniis 591
from the source volume element i to the receiving volume element j, v is the centre
frequency of the band and V
i
is the volume of the acoustic volume element i. The
transmission coecient t
ij
for a single simply supported panel between acoustic volume
elements is calculated by using the SEA method as presented in reference [47]. This gives
better agreement with measured results than the mass law at frequencies near and higher
than the coincidence frequency of the panel. Even measured results could be used, but it
would not be eective from the calculation point of view. When two volume elements are
connected by an open aperture, then equation (23) is valid with t
ij
=1. Dierent edge
conditions of panels can be taken into account by using the knowledge that, at low
frequencies, the radiation resistance of a clamped panel should be twice that of a simply
supported panel, as Maidanik has stated [46].
2.4. rNrrc. :Nb iNii1 iovrr
When the SEA equation (7) is solved at the given angular frequency v, one obtains the
time-averaged energies E of dierent oscillators. The time-averaged energy can be related
to other quantities: e.g., the energy of the acoustic volume element in a diuse eld is
related to the mean squared sound pressure by [15]
E=p
2
V/rc
2
=p
2
rms
V/rc
2
, (24)
where p
2
is the mean squared pressure, p
rms
is the root-mean-squared (r.m.s.) pressure, V
is the volume of the acoustic volume element, r is the uid density, c is the speed of sound
and means space averaging. In practice, space averaging is needed because subsystems
such as a reverberant volume are not ideal. The time-averaged energy of a structural
element is related to its vibrational velocity as [15]
E=mv
2
, (25)
where v
2
is the mean squared vibration velocity averaged over the element surface and
m is the mass of the structural element.
The air-borne input powers of sound sources are derived by using measured values
obtained by standardized methods based on sound pressure measurements (see, e.g.,
reference [48]) or on sound intensity measurements: see, e.g., reference [49]. When
measured values are not available, semi-empirical estimation formulas may be used: see,
e.g., references [6, 7]. The sound power level of a diesel engine can also be estimated by
using the sound pressure level it produces [50]. These air-borne input powers are used for
the acoustic volume elements where the sound sources are located. In the ship context,
these include the air-borne sound powers of the main and auxiliary engines in the machine
room and of other noise sources: for example, ventilation fans in the ventilation machine
rooms.
The structure-borne sound power is dicult to evaluate and no standardized methods
are available [51]. However, research workers have developed semi-empirical and
theoretical methods which may be used to obtain approximate values of structure-borne
input powers: see, e.g., references [7, 5257]. Usually the source strength is described by
measuring the vertical vibration velocity at the machine footing or at the shell plating
immediately above the propeller. This is insucient, as the power transfer also depends
on the mobility properties of the seating structure and on machine footings. Also, the
driving point impedance of the shell plating above the propeller aects the power transfer
from the propeller-induced pressure eld into the shell plating.
In this program package the propeller input power is estimated by using semi-empirical
methods or by using the power spectral density calculated for full scale from a model test.
Measured values have better accuracy, and hence are preferred if available. The power
i. n.NN: ET AL. 592
spectrum density is used to estimate the velocity levels of the shell plating above the
propellers by using the theory of the response of a ships hull to acoustic pressure in water
induced by the propeller [58]. The input power is then obtained approximately by using
the driving point impedance Z for an innitely thin isotropic plate [32],
P=
1
2
=v=
2
Re (Z), (26)
where v is the vibration velocity and Z, which now is real, is given by [32]
Z=8[Eh
3
/12(1 n
2
)]
1/2
(rh)
1/2
, (27)
where E is Youngs modulus of elasticity, h is the plate thickness, n is the Poisson ratio
and r is the material density.
3. MODELLING THE STEEL STRUCTURE
In the earlier applications, the amount of work needed to model the structure was found
to restrict the applicability of SEA for large structures such as ships. Since in FEM very
large steel structures are modelled by using the preprocessor programs, an idea was
developed to apply the geometrical modelling typical of FEM with SEA [59]. This rst
approach showed that it was possible to proceed in this way by dening the geometry with
the aid of nodes and elements. Later, the method was further developed to include the
indexing technique of FEM, the skyline matrix and bandwidth optimization: that is, the
loss factor matrix was assembled as in FEM [27]. Subsequently, beam and triangular plate
elements have been added, and intermediate nodes allowed for elements. These intermedi-
ate nodes need not be in the middle of sides of elements as in FEM. Moreover, calculation
algorithms were changed to allow for distorted elements. In this way, the modelling may
better follow the real geometry of the steel structure. Also bandwidth optimization based
on geometry has been added. The following variable-number-nodes elements (see Figure
2) are implemented: two- to three-noded beam element, four- to nine-noded quadrilateral
plate element, three- to nine-noded triangular plate element and eight- to 26-noded
acoustic volume element. At least the key nodes must be non-zero. Intermediate nodes give
Figure 2. Variable-number-nodes elements used in the SEA program package: (a) one-dimensional beam
element; (b) two-dimensional quadrilateral plate element; (c) two-dimensional triangular plate element;
(d) three-dimensional acoustic volume element. W, Key nodes; w, intermediate nodes.
s1ric1irr-norNr soiNb iN sniis 593
exibility in modelling work when dierent element types are connected in common nodes.
The triangular plate element is obtained by letting one side of a plate element collapse to
a point. This is shown by giving the same node to two corners of a plate element (see
Figures 2(b) and 2(c)).
The geometry of the steel structure is dened with nodes and elements. The element
boundaries are selected to join with the structural junctions. The side length of elements
usually used is foureight frame spacings. Only the main features of a ship are modelled,
including the double bottom with longitudinal and transverse girders, bulkheads,
supporting pillars, hull plating and decks excluding girders and small stieners. The
supporting pillars, which are often used in passenger cruise vessels, are modelled by using
beam elements. The acoustic volume elements are used in the machine room and in the
other spaces where air-borne input power is brought into the structure. Otherwise, volume
elements are not used between decks, their contribution to sound transmission as air-borne
sound being negligible compared to the structure-borne sound transmission. In addition,
this reduces the modelling work and the size of the model. After the geometry of the steel
structure is dened with the aid of nodes and elements, the program determines the
couplings between elements. The program also calculates the geometrical quantities and
other properties of elements and couplings.
4. ASSEMBLAGE OF LOSS FACTOR MATRIX
The symmetrical SEA equation (7) is similar to the equilibrium equation obtained in
the structural analysis when FEM is used [60],
[K]{U} ={F}, (28)
where [K] is the structure stiness matrix corresponding to the loss factor matrix [Y
s
], {U}
is the displacement vector corresponding to {E
i
/n
i
}, and {F} is the applied force vector
corresponding to {P
i
} in SEA. In FEM programs the global stiness matrix is not directly
formed. Rather, element stiness matrices are formed separately and assembled into the
global stiness matrix. This is symbolically written as [60]
[K] =s
i
[K]
(i)
, (29)
where [K]
(i)
is the stiness matrix of the ith element. In the SEA application the assemblage
of the symmetrical loss factor matrix [Y
s
] is obtained from
[Y
s
] =[Y]
(d)
+s
i
[Y
s
]
(i)
, (30)
where the summation goes over all coupling branches, which correspond to elements in
FEM,
[Y]
(d)
=[diagonal h
ii
n
i
], h
ii
=h
i
, (31)
and the equation for one coupling branch is
[Y
s
]
(i)
=
$
h
ij
n
i
h
ji
n
j
h
ij
n
i
h
ji
n
j %
=h
ij
n
i
$
1
1
1
1
%
. (32)
5. PRACTICAL IMPLEMENTATION
In the current implementation of the SEA program, as many subroutines of a FEM
program [61] as possible, 50 altogether, have been used. The SEA program part itself now
i. n.NN: ET AL. 594
contains 59 subroutines and about 9400 lines of FORTRAN 77 code (without comment
lines), and about 14 000 lines in all. At the moment the FEM program package contains
about 135 000 lines of code including comments and the SEA part.
The FEM-type approach requires a list of coupling branch elements. The assemblage
of local loss factor matrices into the global loss factor matrix is made with the aid of this
list. Before the list is made, bandwidth optimization based on geometry is applied to
arrange the elements into suitable order. This usually results in so dense a loss factor matrix
that bandwidth optimization is not necessary. Geometrical optimization also reduces the
time required to determine the couplings between elements, because elements can be
connected to each other within certain distances in space. Here bandwidth optimization
based on geometry is made in the longitudinal direction of the ship.
Since the loss factor matrix is symmetric, only the upper half of the matrix needs to be
stored. For eective storage and fast computation it is useful to store only the minimum
number of zero matrix elements. One way of storing this kind of matrix is the skyline
technique, which is commonly used in FEM programs [60]. The loss factor matrix is
banded especially if large structures are modelled. When using the skyline technique,
the solution of equation (7) requires about
1
2
nm
2
K
operations, where n is the order of the
matrix and m
K
is the average half-bandwidth (see reference [60]). The loss factor matrix
is not really inverted, only its LDL
T
-factorization is determined by using Gauss
elimination.
The overall solution proceeds as follows. After some initial checks are made, e.g., for
triangular and quadrilateral plate elements, a check is made to ensure that all nodes are
in the same plane within a given tolerance. Also the geometrical dimensions of elements
are checked to be within tolerances. After checks, the average global co-ordinates of
elements are calculated and elements are sorted into suitable order, that is geometrical
optimization is made. Then coupling branches are determined. Subsequently, the
frequency-independent properties of the elements and coupling branches are calculated.
Then the frequency-dependent properties are calculated in one-third octave bands with
centre frequencies from 50 Hz to 10 kHz. Thereafter the global loss factor matrix is
assembled and the energies per modal density of elements are solved. The program package
contains an error handler that is used to check that the calculated values are within the
allowed range during the computer run. If a discrepancy is observed, this gives an error
message with location information.
6. SOUND PRESSURE LEVELS IN ACCOMMODATION SPACES
Both the air-borne and structureborne sound transmission, as well as local sound
sources, are included in the prediction of sound pressure levels. The air-borne sound is
included as sound power levels both for space elements and for cabins. For structure-borne
sound, the sound radiated into the space is determined by the velocity levels, radiation
ratios and dimensions of the structures facing the cabin. The calculation follows the scheme
presented in references [4, 14]. The velocity levels of a ships steel structure are determined
after solving the SEA equation (7). Thereafter measured or semi-empirically estimated
transmission loss spectra are used to obtain the velocity levels of the cabin surfaces. By
using measured or estimated sound radiation ratio spectra for cabin surfaces, the
contribution of structure-borne sound radiated into a cabin is calculated. The sound
absorption is estimated by using measured or estimated sound absorption spectra for every
surface instead of using the average 05 s reverberation time typical of cabins. A xed
reverberation time does not allow alternative design calculations for dierent space types
including hard walled spaces. The sound pressure level in a space is calculated by using
s1ric1irr-norNr soiNb iN sniis 595
diuse eld approximation in octave bands with centre frequencies from 63 Hz to 8 kHz,
while in addition the A-weighted sound pressure level is determined.
7. APPLICATION EXAMPLES
7.1. n.brocr:inic rcno svrriiNc vrssri
The SEA program was applied for the rst time to a hydrographic echo sweeping vessel
built by Rauma-Repola (now part of Finnyards Ltd) in Finland (see Figure 3). The main
dimensions of the ship are length overall 329 m, width 104 m and draught 28 m. The
propulsion is produced by a pair of main engines, each generating 220 kW when the
rotational frequency is 1500 min
1
, coupled through reduction gears to two controllable
pitch propellers rotating at 370 min
1
. Accommodation is for 19 persons in 12 cabins. At
the shipyard the input data for the SEA calculation was prepared manually, dening the
steel structure with the aid of nodes and elements. The whole steel structure was modelled
with 241 plate elements and 86 space elements [59, supplement]. The coupling branches,
in all 1148, were calculated by the SEA program. In the design stage the shipyard
independently calculated the sound pressure levels by using empirical methods and
measured these during a sea trial. The practical implementation of the program version
was discussed and the SEA estimates were compared with measured and empirically
calculated values [59, supplement]. The mean absolute dierence between SEA estimates
and measured A-weighted sound pressure levels L
pA
was (41 211) dB (x 2s/zn) when
eight spaces of dierent type were evaluated. In the octave bands with centre frequencies
from 63 Hz to 8 kHz the mean absolute dierence varied between 33 and 85 dB. The
individual error curves and the arithmetic mean error are presented in Figure 4. The mean
arithmetic error for L
pA
was 21 dB and the standard deviation 50 dB.
Figure 3. The hydrographic echo sweeping vessel.
i. n.NN: ET AL. 596
Figure 4. SEA estimate compared with measured octave band sound pressure level Lp on board the
hydrographic echo sweeping vessel. , The eight individual curves; , arithmetic mean value. The
measurements were made by the shipyard.
7.2. 1ixnrr-coN1:iNrr c:rrirr
The second version of this SEA program package was developed to include the skyline
technique and bandwidth optimization which are widely used in similar problems in FEM.
The improved SEA program was applied to analyze a timber-container carrier (see Figure
5) with the main dimensions length overall 131 m, width 20 m and draught 68 m [27]. The
accommodation was located in the afterbody of the ship. The model of the steel structure
was prepared with the aid of the preprocessing program. The model of the whole steel
structure of this ship comprised 2445 elements. The SEA program determined 7930
coupling branches between these elements. The ship was built in Spain at the shipyard
Astilleros Reunidos Del Nervion, S.A. The shipyard independently measured the sound
pressure levels on board the ship during a sea trial. During the trial the ship was in the
ballast condition and the main propulsion machinery was running at the maximum
continuous rating (85%). The meteorological conditions were good: sea state and wind
force varied from 1 to 2 on the Beaufort scale (28 March 1990). The estimated sound
pressure levels were in reasonable agreement with measured results. The mean absolute
dierence x between SEA estimates and measured A-weighted sound pressure levels L
pA
was 55 207 dB (x 2s/zn) when 25 spaces of dierent type were evaluated. The
individual error curves in the octave bands with centre frequencies from 63 Hz to 8 kHz
and the arithmetic mean error are presented in Figure 6. For these 25 spaces the arithmetic
mean error for L
pA
was 19 dB and the standard deviation 64 dB.
7.3. i:ssrNcrr criisr vrssri
The latest version of the program package has been applied to the passenger cruise vessel
Crown Jewel with the main dimensions length overall 164 m, draught 54 m, and moulded
breadth 225 m (see Figure 7). The number of passenger decks is eight and the number of
cabins 410. The ship was built in Spain at the Shipyard Union Naval de Levante, S.A. The
whole steel structure of this ship was modelled for vibration analysis by using FEM. This
same model, with some alterations, was used for this SEA calculation. The model was
generated by using the preprocessor. Many simplifying assumptions were made to reduce
the number of elements and modelling work. The SEA model consisted of 5143 elements.
These included 3107 plate elements, 1357 triangular plate elements, 611 beam elements and
68 acoustic volume elements. Not all of the beam elements were necessary for the SEA
s1ric1irr-norNr soiNb iN sniis 597
Figure 5(a). The timber-container carrier Igor Ilinskiy; (b) the SEA model of the timber-container carrier.
Figure 6. SEA estimates compared with measured octave band sound pressure levels Lp on board the
timber-container carrier Igor Ilinskiy. , The 25 individual curves; , arithmetic mean value. The
measurements were made by the shipyard.
i. n.NN: ET AL. 598
Figure 7(a). The passenger cruise vessel Crown Jewel; (b) the SEA model of the passenger cruise vesselthe
viewing angle is selected so that the decks are distinguishable.
calculation. In the node mesh used to dene the geometry of the steel structure with these
elements were 7401 nodes. The couplings between elements, 17 449 altogether, were
determined by the program. In the model real material thicknesses were used in accordance
with the steel structure.
The propulsion machinery consisted of four main engines, four auxiliary engines and
two highly skewed propellers. For the main engines and the auxiliary engines the measured
vibration velocity level under resilient mountings was available. These gures were
transformed to power levels with the aid of the point mobility (see section 2.4). The
air-borne input powers were obtained by using sound pressure levels measured at a
distance of 1 m. The propeller input power was obtained by using the power spectral
density calculated for full scale from a model test by using equations (26) and (27). The
air-borne noise of ventilation fans in the ventilation rooms on dierent decks was included
in the calculation model. The structure-borne noise of resiliently mounted fans was
estimated to be negligible compared to the noise of other sources. The air-borne sound
power into the engine casing was taken into account and estimated with the estimated
sound pressure levels.
The acoustical data on cabins and other accommodation areas was obtained partly from
measured data and partly from semi-empirical data. The local air-borne sound power of
air ventilation units in cabins and other areas was included as measured values.
7.3.1. Estimated sound pressure levels
The sound pressure levels on board the passenger cruise vessel Crown Jewel were
estimated at 96 places on dierent decks where it was assumed to be necessary to check
the conformity with the requirements agreed. Clearly, the inuence of structure-borne
sound transmission was seen in the calculated estimates, both in the horizontal and in the
vertical directions. In the cabins facing the engine casing and the ventilation rooms higher
noise levels, as expected, were estimated owing to the inuence of more structure-borne
s1ric1irr-norNr soiNb iN sniis 599
sound. The spaces located near propellers or above the machine rooms showed higher
values, as expected. Also, the relative estimated dierences between the conning places
were in agreement with experience. When making repeated computer runs, the eectiveness
of some noise abatement measures was estimated. Whatever the absolute accuracy of the
estimates, the relative dierences provided good guidance during design. The shipyard
independently measured the sound pressure levels on board the ship during sea trials in
June and July 1992. During the trials the main propulsion machinery was running at the
maximum continuous rating (85%). The meteorological conditions were good during
trials. In June the wind force was 4 and sea state 3 (moderate sea) on the Beafort scale.
In July the wind force and sea state were both 2. The ship was not fully completed during
trials. In some cases the measurements were not made in the places for which the sound
pressure levels were estimated. Hence only 22 measurement points could be evaluated.
The mean absolute dierence between SEA estimates and measured A-weighted sound
pressure levels L
pA
was 30 205 dB (x 2s/zn) when 22 spaces of a dierent type
were evaluated. The individual error curves in the octave bands with centre frequencies
from 63 Hz to 8 kHz and the arithmetic mean error are presented in Figure 8. For these
22 spaces the arithmetic mean error for L
pA
was 16 dB and the standard deviation was
34 dB.
7.3.2. Calculated vibration velocity level dierence between decks
The performance of this SEA method in predicting the vibration velocity level dierence
between decks in the vertical direction was tested numerically by exciting only one element
in the double bottom under one of the main engines (see Figure 9). The shape of the input
power spectrum was similar to that of the main engine spectrum, but its level was increased
to obtain reasonable response. The selection of the place of the input element also
simulates the situation faced in practice when one applies a semi-empirical method. The
calculated vibration level dierences DL between decks vary considerably (see Figure 10).
Vibration is more easily transmitted from the double bottom up to the rst deck via
primarily vertical bulkheads surrounding the machine room and shell platings, rather than
in the horizontal direction along the complicated double-bottom structure, where the
transmission loss per frame is higher (see Figure 11(b)). The transmission path was veried
by comparing the calculated vibration velocity levels of the elements concerned. This
same explanation is also valid concerning the dierence between the double bottom and
Figure 8. SEA estimates compared with measured octave band sound pressure levels Lp on board the passenger
cruise vessel Crown Jewel. , The 22 individual curves; , arithmetic mean value. The measurements were
made by the shipyard.
i. n.NN: ET AL. 600
Figure 9. A cross-section of the SEA model of the passenger cruise vessel showing the element (darkened)
used for the power input when the horizontal and vertical vibration transmission properties were numerically
tested: viewed from behind double bottom, engine casing, location of main engines, and only decks 1, 2, and
3; elements of the engine foundation are removed for clarity.
deck 2. The vibration energy is more evenly distributed over longer distances, because
structure-borne sound travels around the discontinuities along many dierent paths in the
structure. This is clearly seen when one looks at the averaged vibration level dierences
per deck between the double bottom and deck 2 or 7 (see Figure 10).
The decrease in vibration level TL in dB along a ships hull according to the
semi-empirical method of Janssen and Buiten is given by [4]
TL=m
0
10
057
for machinery
for propeller
1
+n
0
5,
2 +12/n,
if n Q4
if n e4
1
, (33)
where m is the number of transverse frames between the source and the receiver, and n
is the number of the deck on which the receiver is located. If the source is on deck number
1 or 2, instead of the tanktop, the corresponding 5 or 10 dB must be subtracted. Equation
(33), based on empirical data, gives a dierence of 5 dB per deck when n E4, 44 dB when
n =5, and thereafter a little less. The accuracy decreases rapidly when m exceeds 25 [4].
The same type of result is obtained with this model in the vertical direction. The large
vibration level dierences, up to 157 dB per deck, are explained by the fact that the pillars
supporting decks in large public areas and in some accommodation areas transmit very
little exural vibration. It seems that the calculation method is able to predict reasonably
the dierences caused by structural details in vibration transmission. The results have not
Figure 10. Calculated vibration level dierences DL (dB) at octave bands between the decks of the passenger
cruise vessel Crown Jewel, when noise was input into only one element in the double bottom at frame no. 66.
Average DL between two decks at frame numbers 10 (w), 50 (e), 130 (t), and 170 (T); average DL between
double bottom and deck 1 at frame number 130 (Q); average DL between double bottom and deck 2 at frame
numbers 50 (r) and 130 (R), average DL between double bottom and deck 7 at frame numbers 50 (E), 90 (q)
and 130 (W). (Frame spacing =065 m.)
s1ric1irr-norNr soiNb iN sniis 601
Figure 11. The calculated horizontal transmission loss TLH of structure-borne sound at octave bands
(dB/frame) along the decks of the passenger cruise vessel Crown Jewel, when noise was input into one element
in the double bottom at the frame number 66. Reference elements were selected from equivalent positions just
right of the centreline on the starboard side (frame spacing =065 m). (a) Values shown are TLH across a total
of 8 frames (q), 32 frames (r), 72 frames (w) and 120 frames (e); also shown are values calculated from
equations (34a, b) (equivalent black symbols). (b) Average TLH along double bottom (Q), deck 1 (W), deck 2
(R), deck 3 (E), deck 4 (q), deck 5 (t), deck 6 (r), deck 7 (e) and deck 8 (w).
been veried by vibration measurements, as the ship was manufactured abroad and the
laboratory could not aord these measurements.
7.3.3. Calculated decrease in vibration velocity level along decks
The same test run as in section 7.3.2 was used to obtain the vibration velocity level
decrease horizontally along the double bottom and decks (see Figure 11). The average
decrease in vibration velocity level in the double bottom was 17 dB per frame at the 63 Hz
octave band, decreasing thereafter to the value of 13 dB per frame at the 8 kHz octave
band. The mean frame spacing in this ship is 065 m. Near the source, the decrease in the
double bottom is almost twice the decrease on the decks. The structure of the double
bottom is very complicated and there are many small sized structural elements forming
more junctions, as compared to those for the decks. The bulkheads bounding the machine
room, the engine casing, ventilation rooms, stairs and lift wells from discontinuities in the
vibration transmission path. The values on decks 13 are fairly near the empirical value
of 10 dB per frame given by equation (33) for machinery. On deck 3 between frames 58
and 90 there is the engine casing and ventilation room; otherwise, the deck is supported
by pillars. This means that in the structure near the centreline there are only straight-
through connections and connections with pillars. This is seen as a lower decrease value.
Thereafter the vibration energy is distributed again more evenly. The behaviour is similar
on decks 48. These numerically obtained values can be compared with those used in the
i. n.NN: ET AL. 602
waveguide method, where the transmission loss TL
H
in dB in the horizontal direction
across a total of m frames at frequency f based on full-scale measurements is given by
[14, 62]
TL
H
=
8
mb
1
,
20b
1
+(m20)b
2
,
20b
1
+30b
2
+(m50)b
3
,
20 E
mE20
mE50
me50
9
(34a)
where
b
1
=07, b
2
=010 +0161 log f, b
3
=0032 +0052 log f. (34b)
The horizontal transmission loss TL
H
calculated from equations (34a) and (34b) gives
07 dB per frame for the rst 20 frames. The values in Figure 11 are in close agreement
with the empirical data of equations (34a) and (34b). One considerable dierence is that
when using this SEA method, the values are, in contrast, lower when the frequency is
increasing. This may be due to excessively low damping at higher frequencies because no
acoustic volume elements were used, excluding the machine room, engine casing and
ventilation rooms. The other reason is that the contributions of the in-plane compressional
and shear mode types were ignored. They have an eect when structure-borne sound
propagates over large distances [63]. However, this is not of great practical importance,
because just the low frequency sound is usually the main factor when estimating sound
pressure levels in the ship context. In addition, in the places far away from noise sources,
the local noise sources dominate: e.g., air distribution devices in cabins. One should also
bear in mind that the test ship here was a very special passenger cruise ship, when
comparing the estimated values with those obtained empirically or from semi-impirical
formulas. In the waveguide method, the transmission loss of structure-borne sound is
determined analytically in a vertical ship cross-section between the same two frames. The
horizontal transmission loss is determined by using experimentally determined values. This
is because most noise problems are encountered in spaces just above the machine rooms
or the propeller, and it is of less importance to predict accurately a noise level in a place
far away horizontally if it is known to be below any noise requirements [62].
7.4. crNrr:i rrx:ris oN soiNb irrssirr irvri birrrrrNcrs
In the SEA method the following quantities may be erroneous: input power estimates
of structure-borne or air-borne sound of sound sources, the SEA parameters, the simplied
model of a steel structure, and the simplied sound wave eld model, including here only
the exural waves. The conventional acoustic part used to estimate the sound pressure
levels in the accommodation includes erroneous estimates: e.g., sound transmission loss
between steel structure and cabin surfaces, sound radiation ratios, sound absorption and
local air-borne input powers. Errors due to these erroneous input data may partly cancel
each other out. Dierences in the nal construction from the drawings according to which
the calculations were made may have some eect. The eects of workmanship are also
reected in the measured sound pressure levels.
One diculty is to obtain high quality input data, which is essential for the accuracy
of the estimates. This is especially true for structure-borne input powers [51] or dierent
kinds of machinery and propellers. The acoustical properties of any noise abatement
measures should be known. These include, e.g., vibration-damping properties of
oating-oor constructions, sound radiation ratios of panels, sound-absorption properties,
and insertion loss of resilient mountings of wall and ceiling panels used. Measured values
are preferred, if available, rather than those estimated by using semi-empirical or empirical
methods.
s1ric1irr-norNr soiNb iN sniis 603
During sea trials the ship is usually not yet nished, and hence not all of the points
included in the estimates are measured or included. Thus the measured dierences between
the SEA estimates and measured sound pressure levels reect the inuence of all possible
errors in the whole estimation scheme. The mean absolute dierences of this SEA method
are of the same order at octave bands from 125 Hz up to 8 kHz, while at the lowest band
of 63 Hz the mean absolute dierence is about 2 dB larger.
7.5. coxii1rr 1ixr
The SEA model of the passenger cruise vessel which contains 5143 elements and 17 449
coupling branches has been calculated using three computers. The charged CPU (central
processor unit) time, including the reading of input data and all the subsequent
calculations, was about 8 h 45 min, while the elapsed real time was about 13 h on a Micro
VaxII. Subsequently, the program package was installed on a UNIX-based workstation,
Iris Indigo. The equivalent CPU time was now about 38 min and the elapsed real time was
about 43 min. With the workstation Iris Indigo R4000 the equivalent CPU time was
163 min and the real time 164 min. The given times depend on the device conguration
and on the computer load during the run time and, as such, give only an indication of
required computer resources and capabilities. These gures can be compared with the CPU
time of 3 h 8 min in a Micro VaxII which elapsed during the calculation with the SEA
model of the timber-container carrier containing 2445 elements and 7930 coupling
branches. Additional information on the computational eectiveness of the skyline
technique with bandwidth optimization is presented, e.g., in references [27, 60].
8. DISCUSSION
With the applications it was shown that the principles of modelling the geometry of a
large welded structure such as those in FEM are also applicable when the SEA approach
is used. It was also shown that the methodology typical of FEM for the assemblage and
solving of the SEA equation is very eective. The geometrical optimization for reducing
the bandwidth of the loss factor matrix also proved to be eective in the SEA program
package developed for ship applications. It also reduces the computation time when the
couplings between elements are determined.
In the special applications considered, the SEA method proved to be applicable to the
analysis of the structure-borne sound transmission in large ship structures, even when
many approximations were made during modelling of the structure. Comparison with
semi-empirical estimation methods showed that the SEA method was able to predict
reasonably the structure-borne sound (vibration) transmission loss when only the
subsystem of exural sound waves was included. This supports the ndings of Nilssons
study [12]: i.e., that the main power ow is by exural waves in a ship structure. In the
SEA method the prediction is made both in the horizontal and in the vertical direction
of propagation, whereas in the waveguide method the sound transmission in the horizontal
direction is estimated semi-empirically with data based on measured results. It is obvious
that the SEA method is able to predict the relative inuence of structural changes and
details on structure-borne sound transmission.
The fact that modal densities n( f ) are low is often seen as a problem. This may be partly
circumvented by using suciently large elements, as must be done for the economical
modelling of large structures. If a subsystem has a small mode count, it may still be
modelled as a SEA subsystem, even though the concept of modal density is inappropriate
in this case [23]. SEA provides a good approximation to the mean level of the energy
transmission or response even in those frequency regions where the modal density is low
i. n.NN: ET AL. 604
and the response of the structure is dominated by a few resonances [26]. In building
acoustics it was found that if the error of 3 dB in the prediction of sound transmission
was taken as a limit, then the requirement would be for the mode count N=n( f )Df q05
and for the modal overlap M=fhn( f ) q04 for plates in one-third octave bands, where
h is the total loss factor [37]. The results of this paper also support the fact that in
engineering calculations the SEA calculation can be extended to lower frequencies despite
low modal densities. However, thorough research is required to establish the uncertainty
in the applications of the SEA method at low frequencies, where there are only a few
interacting modes per frequency band.
The importance of in-plane vibration transmission has been emphasized by many
authors (see, e.g., references [23, 63, 25]), although so far, in many applications of the SEA
method, only relatively small models have been used. In the applications considered in this
paper, the SEA method worked reasonably even for the large model of the passenger cruise
ship without the inclusion of the in-plane vibration transmission. For engineering purposes
it seems feasible to develop further this program package, including more developed theory
and new element types for new applications. Implementing acoustic volume elements
where the uid is a liquid would allow the inclusion of the eects of liquid storage tanks
and the water surrounding the ships hull in the model [24]. However, it is worth noting
that the modelling accuracy must be adequate for describing the phenomena considered.
As such, the SEA program package developed can very well be used in the estimation of
sound pressure levels in the ship context for octave-band frequencies from 63 Hz to 8 kHz.
The mean absolute dierence between estimated and measured L
pA
was in the range of
3055 dB (55 measurement points) when three dierent types of ship were analyzed.
General limits of accuracy cannot be given based on these limited statistics.
The SEA approach is especially applicable to new ship types when the structure-borne
sound transmission and the eects of structural details are to be estimated. This method
estimates only the transfer function from the point of excitation to the receiving point.
However, problems are encountered when the structure-borne input powers are
determined by using the usually available information on measured vibration velocities at
the machine foundations or at the footings measured either at the workshop or in situ on
board the ship. The accuracy of input power estimates is of vital importance for the
calculated SEA estimates of vibration or sound pressure levels. After the vibration levels
of the steel structure are obtained by using the SEA method, the sound pressure levels in
accommodation spaces are estimated by using the same methods as in the wave guide
method or in semi-empirical methods. Semi-empirical methods are cheap to apply and are
suitable if scant data is available, as in the early design stage. Hence the choice between
methods depends on the design stage as well as on the data available.
ACKNOWLEDGMENTS
The authors are grateful to Mr M. Vahteri who worked at Rauma-Repolas Rauma
Shipyard and helped in the preparation of input data of the hydrographic echo sweeping
vessel. They also thank Mr P. Olkinuora of Finnyards Ltd for his consent to publish data
and for the photograph of this ship.
The authors would like to express their gratitude to Mr G. Asuar and to the personnel
of the shipyard Astilleros Reunidos del Nervion, S.A. in Spain for their permission to
publish the results of the SEA calculations and sound pressure level measurements of the
timber-container carrier. The authors acknowledge the assistance provided by Mr P. Arkke
from Deltamarin, Ltd (part of Elomatic Ltd) in Finland in the preparation of data for
the SEA calculations of this ship.
s1ric1irr-norNr soiNb iN sniis 605
The authors would like to express their gratitude to Mr J. Poblet and to the personnel
of the shipyard Union Naval de Levante, S.A. in Spain for their patient help during the
preparation of data for the SEA calculations of the passenger cruise vessel. For the
measured results they also deserve considerable thanks. The authors owe thanks to Mr
H. Bache r and the personnel of EJohn International in Finland for valuable help during
the SEA calculations and for their permission to publish the data.
The authors are indebted to Mr Jyrki Kullaa Lic. Tech., Mr Sauli Liukkonen
M.Sc.(Tech.), Mr Jukka Airaksinen M.Sc.(Tech.), Mr Kai Katajama ki M.Sc.(Tech.) and
Mr Jari Kivela from VTT for preparing the model of the passenger cruise vessel with the
preprocessor and for help during the programming and computer runs. The authors Klinge
and Vuoksinen specialize in the nite element method and have written the computer code
of the FEM package. Mr Hynna is responsible for the acoustic part as well as for the SEA
program code. The nancial support provided by the Division for Manufacturing
Technology of VTT is gratefully acknowledged.
REFERENCES
1. IN1rrN:1ioN:i M:ri1ixr Orc:Niz:1ioN 1982 Noise Levels on Board Ships: Code on Noise
Levels on Board Ships and Recommendation on Methods of Measuring Noise Levels at Listening
Posts.
2. D. M:ixbrr and H. Gor1nr 1983 Schi & Hafen 35, 3437. Statistical investigations into noise
situation in deep-sea shipping (in German).
3. H. F. S1rrNnori 1988 ICMES 87, Proceedings of the Fourth International Symposium on Marine
Engineering Systems Application of Technological Advances. Stockholm, Sweden: Inter-
national Cooperation on Marine Engineering Systems (ICMES). Marine engineering and noise
control.
4. J. H. J:NssrN and J. Bii1rN 1973 Proceedings of INTER-NOISE 73. On acoustical designing
in naval architecture.
5. J. Bii1rN 1977 Proceedings of the International Symposium on Shipboard Acoustics 1976.
Amsterdam: Elsevier. Experiences with structure-borne sound transmission in sea-going
ships.
6. J. PiiN1 1980 Doctoral Thesis, Chalmers University of Technology, Gothenburg, Sweden.
Methods for predicting noise levels in ships. Experiences from empirical and SEA calculation
methods, part I: noise level prediction methods for ships, based on empirical data.
7. R. W. Fisnrr, C. B. Birroicns and D. L. NrisoN 1983 The Society of Naval Architects and
Marine Engineers, Technical and Research Bulletin 337. Design guide for shipboard airborne
noise control.
8. Handbook for Shipboard Airborne Noise Control 1974 U.S. Coast Guard Technical Publication
073-0100.
9. Guide to Ship Noise Control 1986. Tokyo: Nippon Kaiji Kyokai.
10. E. Szczrrnicii and A. Sziv:rz.Nsii 1983 Archives of Acoustics 3, 119130. Prediction of
octave noise spectra in accommodations in the superstructure of a ship.
11. P. C:ic:cNo, R. M:i1rsr and F. PiN:zzi 1986 Shipboard Acoustics, Proceedings of the 2nd
International Symposium on Shipboard Acoustics, ISSA 86. Dordrecht: Martinus Nijho.
Applications of two mathematical approaches to predict airborne noise levels is ship
superstructures.
12. A. C. NiissoN 1977 Journal of Sound and Vibration 55, 7191. Attenuation of structure-borne
sound in superstructures on ships.
13. A. C. NiissoN 1978 Journal of Sound and Vibration 61, 4560. Reduction of structure-borne
sound in simple ship structures: results of model tests.
14. A. C. NiissoN 1984 Journal of Sound and Vibration 94, 411429. A method for the prediction
of noise and velocity levels in ship constructions.
15. R. H. L.oN 1975 Statistical Energy Analysis of Dynamical Systems: Theory and applications.
Cambridge, Massachusetts: MIT Press.
16. S. V. BibriN and A. S. Niiirorov 1969 Joint Published Research Service Translation, JPRS
47.803. Propagation and damping of sound vibrations on ships.
i. n.NN: ET AL. 606
17. R. J. S:vir. 1969 Proceedings of the Symposium on Stochastic Processes in Dynamical Problems.
New York: American Society of Mechanical Engineers. The evaluation of a shipboard noise and
vibration problem using statistical energy analysis.
18. J. O brc::rb JrNssrN 1977 Proceedings of the International Symposium on Shipboard Acoustics
1976. Amsterdam: Elsevier. Calculation of structureborne noise transmission in ships using the
statistical energy analysis approach.
19. Y. Irir and S. T:i:ci 1978 Proceedings of INTER-NOISE 78. Structure-borne noise
transmission in steel structure like a ship.
20. K. Fiiiz:v: and C. Y:sib: Proceedings of INTER-NOISE 79. Studies on structure-borne
sound in ship.
21. J. PiiN1 1980 Doctoral Thesis, Chalmers University of Technology, Gothenburg, Sweden.
Methods for predicting noise levels in ships. Experiences from empirical and SEA calculation
methods, part II: prediction of structure-borne sound transmission in complex structures with
the SEA method.
22. T. Yosnii:i, K. H:11ori and T. S:1o 1981 Proceedings of INTER-NOISE 81. Prediction of
noise level on board ship using statistical energy analysis.
23. J. Tr:1cn Jr. 1985 M.Sc. (Eng.) Thesis, Massachusetts Institute of Technology. Vibration
transmission through machinery foundation and ship bottom structure.
24. Y. Snixoxir: 1985 Ocean Engineering Thesis, Massachusetts Institute of Technology. The eect
of a liquid storage tank on sound transmission through ship structures.
25. R. H. L.oN 1986 Noise Control Engineering Journal 26, 2227. In-plane contribution to
structural noise transmission.
26. M. D. McCoiiix and J. M. Ciscnirri 1990 Journal of the Acoustical Society of America 88,
14801485. Bending and in-plane wave transmission in thick connected plates using statistical
energy analysis.
27. P. H.NN:, P. KiiNcr and M. NirxiNrN 1988 ICMES 87, Proceedings of the Fourth International
Symposium on Marine Engineering Systems Application of Technological Advances.
Stockholm, Sweden: International Cooperation on Marine Engineering Systems (ICMES).
Statistical energy analysis with nite element model for noise prediction in ships.
28. K. IiNo and I. HoNb: 1992 Proceedings of the 5th International Symposium on Practical Design
of Ships and Mobile Units, PRADS 92, vol. 1, 16481661. (J. B. Caldwell and G. Ward, editors).
London: Elsevier Applied Science. Total noise prediction system for a passenger cruise ship.
29. M. P. Nor1oN 1989 Fundamentals of Noise and Vibration Analysis for Engineers. Cambridge:
Cambridge University Press.
30. F. J. F:n. 1982 in Noise and Vibration (R. G. White and J. G. Walker, editors), 165186.
Chichester: Ellis Horwood. Statistical Energy Analysis.
31. F. D. H:r1 and K. C. Sn:n 1971 National Aeronautics and Space Administration Report NASA
CR-1773. Compendium of modal densities for structures.
32. L. Crrxrr, M. Hrcii and E. E. UNc:r 1973 Structure-borne Sound: Structural Vibrations and
Sound Radiation at Audio Frequencies. Berlin: Springer-Verlag.
33. F. F:n. 1985 Sound and Structural Vibration: Radiation, Transmission and Response. London:
Academic Press.
34. ISO 2533. Standard atmosphere.
35. G. S. K. WoNc and T. F. W. Exnir1oN 1985 Journal of the Acoustical Society of America 77,
17101712. Variation of the speed of sound in air with humidity and temperature.
36. G. S. K. WoNc 1986 Journal of the Acoustical Society of America 80, 12031204. Characteristic
impedance of humid air.
37. R. J. M. Cr:ii, J. A. S1rri and D. I. Ev:Ns 1991 Journal of Sound and Vibration 144, 95107.
Statistical energy analysis of structure-borne sound transmission at low frequencies.
38. J. C. SiN and E. J. Ricn:rbs 1985 Journal of Sound and Vibration 103, 109117. Prediction of
total loss factors of structures, I: theory and experiments.
39. J. C. SiN, H. B. SiN, L. C. Cnov and E. J. Ricn:rbs 1986 Journal of Sound and Vibration 104,
243257. Prediction of total loss factors of structures, part II: loss factors of sand-lled structure.
40. H. B. SiN, J. C. SiN and E. J. Ricn:rbs Journal of Sound and Vibration 106, 465479. Prediction
of total loss factors of structures, part III: eective loss factors in quasi-transient conditions.
41. Y. Irir and T. N:i:xir: 1985 Bulletin of the Marine Engineering Society in Japan 13, 6072.
Prediction of structure borne sound transmission using statistical energy analysis.
42. S. Uosii:iNrN and K. PrsoNrN 1983 Kari Pesonen Consulting Engineers Ltd, Structure-borne
sound research, Research Report 1. The calculation of structure-borne and airborne sound of
industrial buildings using the statistical energy analysis (in Finnish).
s1ric1irr-norNr soiNb iN sniis 607
43. ISO/DP 9613/1 1986 AcousticsPrediction of environmental noisePart 1: calculation of the
absorption of sound by the atmosphere.
44. R. H. L.oN and E. Eicnirr 1964 Journal of the Acoustical Society of America 36, 13441354.
Random vibration of connected structures.
45. I. L. Vrr and C. I. Hoixrr 1988 in Noise and Vibration Control (L. L. Beranek, editor), 270361.
Washington, D.C.: Institute of Noise Control Engineering; revised edition. Interaction of sound
waves with solid structures.
46. G. M:ib:Nii 1962 Journal of the Acoustical Society of America 34, 809826. Response of ribbed
panels to reverberant acoustic elds. Erratum: 1975 Journal of the Acoustical Society of America
57, 1552.
47. M. J. Crocirr and A. J. Pricr 1969 Journal of Sound and Vibration 9, 469486. Sound
transmission using statistical energy analysis.
48. ISO 37403747. AcousticsDetermination of sound power levels of noise sources.
49. ISO 9614-1. AcousticsDetermination of sound power levels of noise sources using sound
intensityMeasurement at discrete points.
50. W. Hrxiri and T. Sribi 1970 Motortechnische Zeitschrift (MTZ) 31, 153156. Statistic
investigations into diesel engine noises (in German).
51. T. 1rN Woibr and G. R. G:brrri1 1987 Noise Control Engineering Journal 28, 514.
Development of standard measurement methods for structureborne sound emission.
52. A. C. NiissoN and N. P. T.v:Nb (editors) 1981 NORDFORSK, Miljova rdsserien Publication
1981: 2. Noise sources in ships, I: propellers.
53. J. PiiN1 and J. O brc::rb (editors) 1983 NORDFORSK, Miljova rdsserien Publication 1983: 2.
Noise sources in ships, II: diesel engines.
54. B. Pr1rrssoN and J. PiiN1 1980 Chalmers University of Technology, Gothenburg, Sweden,
Department of Building Acoustics Report 80-19. Structure-borne sound transmission from
machinery to foundations.
55. J. W. Vrrnri 1982 Doctoral Thesis, Delft University of Technology, Delft, The Netherlands.
Multi-path sound transfer from resiliently mounted shipboard machinery.
56. J. Bii1rN (editor) 1986 Shipboard Acoustics, Proceeding of the 2nd International Symposium on
Shipboard Acoustics, ISSA 86 (and Supplement). Dordrecht: Martinus Nijho.
57. R. G. Wni1r 1982 in Noise and Vibration (R. G. White and J. G. Walker, editors), 685712.
Chichester: Ellis Horwood, Vibration control (II).
58. J. PiiN1 and D. McQirrN 1980 Acustica 47, 1621. A theory of the response of a ships hull
to propeller-generated acoustic pressure.
59. P. H.NN: 1986 in Shipboard Acoustics, Proceeding of the 2nd International Symposium on
Shipboard Acoustics, ISSA 86, 233243 (and Supplement, 535540) (J. Buiten, editor).
Dordrecht: Martinus Nijho. A literature survey concerning propeller as a noise source and
prediction methods of structure-borne noise in ships.
60. K.-J. B:1nr 1982 Finite Element Procedures in Engineering Analysis. Englewood Clis, New
Jersey: Prentice-Hall.
61. M. K. H:i:i: 1986 VTT Symposium 68, The 4th Marine Technology Symposium, Ship
Vibration, Noise & Hydrodynamics, 1314 January, Espoo, Technical Research Centre of Finland
(VTT). Solution of ship vibration problemscurrent situation and future prospects.
62. A. C. NiissoN 1983 Danish Acoustical Institute, The Danish Academy of Technical Sciences
Report no. 103. A method for the prediction of noise and velocity levels in ship constructions.
63. R. E. Povrii and J. E. M:NNiNc 1988 Proceedings of the 59th Shock and Vibration Symposium,
SANDIA REPORT SAND88-2473C. Albuquerque, New Mexico: Sandia National Labora-
tories. The importance of non-resonant and in-plane vibration transmission in statistical energy
analysis.

Vous aimerez peut-être aussi