Marine Technology, Vol. 17, No. 3, Oct. 1980, pp.

341-350

Eigenvalue Analysis as an Approach to the Prediction of Global

Vibration of Deckhouse Structures
Robert E. Sandstrom 1 and N. Pharr Smith 1
There are several approaches to the prediction of global deckhouse vibration where the deckhouse natural
frequencies lie close to the blade rate exciting frequencies. This paper discusses several approaches and
recommends the eigenvalue analysis procedures as the most useful design tool for the early prediction of
deckhouse vibration problems. Three eigenvalue analysis procedures are presented: the Hirowatari
method--an empirical approach; the simplistic modeling method--a simple analytical approach; and the
finite-element method--a more sophisticated analytical approach. The need for simplicity and accuracy
is emphasized. Two example problems are included to demonstrate the merits of the eigenvalue analysis
procedures.

Introduction
WITH CONTINUING INCREASES both in ship structure flexibility and in propeller loading, shipboard vibration has emerged
in recent years as a major concern of ship designers, builders, and
operators alike [1-3] .2 Moreover, as ships have evolved into the
massive present-day structures with superstructures and machinery aft, deckhouse vibration, in particular, has become an
i m p o r t a n t problem. For many ships with the superstructure in
the afterbody, unsteady propeller forces have been found to excite the deckhouse in a global fore-and-aft mode of vibration, and
it is this mode with which this paper is primarily concerned. The
effects of this vibration can be widely varied, resulting not only
in the discomfort of the crew but also in the cracking of welds and
the fracture of strength members in the hull. Consequently,
regulatory agencies are beginning to address the question of
habitability and structural integrity; more and more investigations are being undertaken to correlate experimental d a t a and
analytical predictions; and symposiums and technical papers are
gradually establishing a repository of theory and methodology
for attacking the problem. Certainly, modern ships are highly
susceptible to vibrations; however, by judicious design these
problems can be reduced.
A basic problem confronting the design engineer is the selection of an analysis procedure which will lead to an accurate prediction of the ship's vibratory characteristics. The preferred
procedure should lead to the prediction of the vibratory characteristics of the vessel early in the design, so that the ship can
be tailored to avoid harmful resonances.
Four methods can be used to determine the vibratory characteristics. They include full-scale measurements, model tests,
empirical calculations, and analytical procedures. Full-scale tests
include such procedures as shaker tests and trial run measurements. Although these tests can provide guidelines for afterthe-fact repairs, they are of little use in the design stage. The
construction and testing of experimental scale models has met
with some success; however, this approach tends to be expensive.
T h e chosen method of prediction need not always be involved
or complex, and some empirical methods have proven to be rem a r k a b l y accurate when compared with far more complicated,
1 Graduate students, Department of Naval Architecture and Marine
En~gineering, The University of Michigan, Ann Arbor, Michigan.
z Numbers in brackets designate References at end of paper.
Presented at the October 3, 1979 meeting of the Hampton Roads
Section of THE SOCIETY OF NAVAL ARCHITECTS AND MARINE ENGINEERS.

OCTOBER 1980

analytical, computerized solutions. If a more indepth investigation of the problem is desired, however, one must resort to some
form of analytical technique. Again, the complexity of the procedure can be matched to the necessary accuracy of the results,
and solutions can range from fairly simple to quite cumbersome
ones. In most cases empirical and analytical solutions can provide
the designer with crucial information during the early stages of
the design.
In this paper we will discuss those empirical and analytical
procedures which we believe to be the most beneficial during the
early design stages.

Forced-response analysis versus eigenvalue

analysis
There are two basic types of analytical solutions: forced response and eigenvalue analysis. For reasons given in the following,
the authors prefer to rely on an eigenvalue analysis rather than
a forced-response analysis.
The fore-and-aft vibratory response of the deckhouse, which
is produced primarily by propeller excitation forces, is the ultimate measure of a superstructure's vibration quality. The
forced-response analysis is used to predict the actual response
amplitudes due to unsteady propeller exciting forces. These
propeller excitation forces are determined using model test data
from a nominal wake survey and a propeller cavitation test in
combination with the propeller and hull geometry. The excitation
forces must then be applied to a structural model of the ship, with
the output attempting to predict the actual response amplitudes.
It is this response on which designers must rely to predict the
ship's vibration quality.
An eigenvalue analysis, on the other hand, is used to determine
the natural vibratory characteristics of the ship as are embodied
in its natural frequencies and mode shapes. Mode shapes represent the relative amplitudes of vibration associated with each
natural frequency and, therefore, are considered to be a more
generalized, pure definition of the ship's characteristic vibrational
nature.
An eigenvalue analysis offers insight into the vibratory characteristics of the ship not available through a forced-response
analysis. The natural frequencies and mode shapes calculated
from the eigenvalue analysis can be used to identify and examine
the following traits:
1. Global and local modes of vibration--The global modes
define the general motion of the structure at a particular fre-

0025-331618011704-0341500.45/0

341

quency. Global modes of the deckhouse structure include the

bending motions in the transverse as well as the fore-and-aft
direction, vertical motions, and torsional motions. Identification
of these modes is most important, since their excitation must be
avoided.
Local modes, on the other hand, describe the motion of isolated
portions of the structure such as the flapping of bridge wings on
a deckhouse. Typically, local modes are difficult to predict analytically because most structural models do not contain sufficient
detail at the local level. Moreover, it is generally considered easier
to fix the local modes after they have been discovered on the ship
rather than to try to predict them analytically.
2. Nodes of the global modes--The nodes of a mode can be
defined as the points within the structure which experience zero
displacement. This is particularly helpful since forces which act
through the nodes will not be capable of producing an excitable
response.

3. Relative amplification of principal modes--Amplification

can be defined as the ratio of the structural displacements x (t)
to the support-base displacements y(t) as shown in Fig. 1:

~_

~y(t)

Fig. 1

Single-degree-of-freedom

spring mass system

~-----[-x
(t)

XI
=

Wn2

~n 2 _

(1)

(22

where the damping is taken to be equal to zero and

w, = natural frequency of the system
= frequency of the excitation
The authors have used this idea of relative amplification to
compare the effects of various propeller excitation frequencies
on the vibration levels of a deckhouse in the near-resonance
condition. We have not used this idea to determine actual vibration amplitudes.
Suppose, for example, that a ship has a design shaft rate of 120
rpm and a deckhouse fore-and-aft natural frequency at 650 cpm.
The amplification factor as computed from equation (1) gives
values of 2.2 for a four-bladed propeller and of 6.8 for a fivebladed propeller. This suggests that use of the five-bladed propeller on this ship may in fact triple the vibration amplitudes.
We must mention that this concept as used here is intended
to be qualitative in nature and, to the best of our knowledge, has
not been confirmed by experimental studies. What this concept
does emphasize is the relative dangers of a near-resonance condition.
Furthermore, if the actual vibratory amplitudes are required,
knowledge of both the natural frequencies and mode shapes
obtained from an eigenvalue analysis and the exciting forces and
corresponding frequencies can be used in an efficient forcedresponse calculation commonly referred to as a "modal analysis"
or "mode summation analysis" [8]. If the mode shapes and natural frequencies are not known, then the forced-response calculation can be performed, using less-efficient computational
techniques such as direct inversion of the equations of motion
and time-stepped integration schemes. It should be noted that
it may be difficult to determine a near-resonance condition with
a forced-response computation without performing an eigenvalue
analysis, since one would never know where the resonances occur.
Consequently, an eigenvalue analysis should be performed prior
to the forced-response analysis.
Recent reports indicate that the results of forced-response
analyses correlate poorly with experimental results [4, 5]. Pro342

peller-induced forces have been calculated with reasonable

precision; however, the application of these forces to structural
models has not consistently produced results of acceptable accuracy. On the other hand, the results of eigenvalue analyses have
been found to correlate reasonably well with experimental results.
Moreover, an eigenvalue analysis offers insight into the vibratory
characteristics of the ship not available through a forced-response
analysis. For these reasons, the authors endorse the use of eigenvalue analysis over forced-response analysis in the design
process.

Eigenvalue analysis of deckhouse structures

The vibratory characteristics of a deckhouse structure can be
determined through an eigenvalue analysis. In considering the
specific case of propeller-excited fore-and-aft vibration, the
natural frequencies of the deckhouse and of the hull are virtually
uncoupled. In most cases it is necessary to model only the deckhouse arid its supporting structure [6]. This procedure has proven
successful in several investigations by the authors; addition of
the hull to the structural model has resulted in less than 10 percent reduction in the natural frequencies (less than I cps). This
phenomenon has also been observed in reference [6].
An eigenvalue analysis of the deckhouse and supporting
structure can be carried out in three steps. The proposed method
is as follows. First, one should determine the fixed-base deckhouse natural frequencies where the house is cantilevered from
the main deck (Fig. 2). This step establishes an upper bound on
the deckhouse natural frequencies and, in addition, provides a
means of comparing the fixed-base natural frequencies of the
deckhouse model with the natural frequencies of an equivalent
cantilever beam. Second, one should determine the flexibility of
the deckhouse base support (Fig. 3). This flexibility can be
computed empirically or from a static analysis of the deckhouse
base support structure. Third, one should compute the deckhouse
frequencies from a combination of the fixed-based deckhouse
model and the flexible support base model (Fig. 4). This procedure is illustrated later.

/-

Flexible house

~_!/./~_(///~(.//I
f FixedBase
Fig. 2

Fixed-base deckhouse

r-"%

Rigldhouse

/iI"
--6A/b---{/

i A//*-~in ae~
/

Flexiblesupportbase

Flexibledeckhouse base support

Fig. 3

~/~-- Flexiblehouse

~__~iIlllil/l/MaindeCk
.........I.
.............
... !,,.,.,,,,,/Flexible
supportbase
Fig. 4

Combined

fixed-basedeckhouse and flexiblebase support

MARINE TECHNOLOGY

Three eigenvalue analysis procedures are described in the remaining portion of this section. They include the Hirowatari
method, the simplistic modeling method, and the finite-element
method. The purpose of this discussion is to present the general
concepts behind each method rather than the specific details of
each.
Hirowatari method. The Hirowatari method [7] is an empirical method for the estimation of the first fore-and-aft deckhouse natural frequency. This method was developed from
comparisons with observed fore-and-aft deckhouse vibration.
In this method the fixed-base natural frequency of the deckhouse
is determined according to deckhouse type and height. The
fixed-base natural frequency is then reduced by a correction
factor to account for the rotational flexibility of the deckhouse
support base. The application of this correction factor produces
an estimate of the first fore-and-aft deckhouse natural frequency.
The procedure for this empirical method is given as follows:
1. Select deckhouse structure type from Fig. 5.
2. Determine deckhouse height, h.
3. Read f~ (fixed-base natural frequency) from Fig. 6.
4. Read fe/f~ (the correction factor) from Table 1.
5. Compute [e (the expected deckhouse natural frequency
in the first fore-and-aft bending mode) from

fe = [~ *felf=

(2)

Reference [7] reports that this procedure generally produces

results that are within + 15 percent of measured shaker tests, but
the method is difficult to use when the superstructure type varies
substantially from those given in Fig. 5. Furthermore, there is
some uncertainty regarding the use of the correction factors given
in Table 1 to account for the flexibility of the deckhouse support
structure, since the supporting structure may vary from deep
beams to column supports to structural bulkheads. Despite these
difficulties, the method seems to work quite well considering the
limited input which is required. This feature makes the method
particularly attractive in the early design stages when the design
data are sparse or unknown.
Simplistic modeling method. In some cases a simplistic
modeling approach can be taken to estimate the first fore-and-aft
natural frequencies fe of the deckhouse. In this method the lowest
natural frequencies/e of the deckhouse are computed by modeling the deckhouse as a cantilever beam supported at the base
by springs as shown in Fig. 7.
Naturally, one should not expect the cantilever model to exhibit the same vibratory characteristics as the three-dimensional
deckhouse structure; however, the first bending modes of the
deckhouse do in fact resemble beamlike behavior. Figure 8 shows
the first two beamlike modes of a 2-D deckhouse model as computed from a finite-element analysis. These modes can certainly
be modeled with a simple cantilever beam.
The fixed-base natural frequencies f~ of the equivalent cantilever beam can be computed either from beam theory or
through an approximate technique involving energy principles
and Dunkerley's equation. The latter method yields a good approximation to the lowest natural frequency f~ and is easy to
apply to beams where the mass and section properties vary along
the length of the beam. The results from this method will be given
here with the deviations given in the Appendix.
For the case where the section properties and mass vary continuously along the length of the beam (Fig. 9), the lowest
fixed-base natural frequency can be computed from

/~=2~

TYPE A

TYPE B

l
TYPE C

TYPE D

Deckhouse types

Fig. 5

1500

i000

J
500
I

l0

15

20

h in meters

Fixed-base deckhouse frequencies

Fig. 6

Table 1

Flexible base correction factors

Type

fe/f~,

A,C

.625

.602

.751

Cantilever beam

Fig. 7

Equivalent deckhouse model

Mode 1

Fig. 8

Y~de2

Beamlike deckhouse modes

dx

60

/f0

fo

OCTOBER 1980

1
[(X - ~)2

d dx

(cpm)

(3)
Fig. 9

Continuous beam

343

to produce an estimate of the actual deckhouse frequencies re.

Second, a more theoretical correction can be made through the
use of Dunkerley's equation. Here, the natural frequencies of a
rigid house on a flexible base fe and fv are computed and combined with. the natural frequency of a flexible house on a fixed
base f~ to give fe using Dunkerley's equation

Ii' El' "'"

[e =

Fig. 10

Lumped-mass beam

Fig. 11

1
1

Simple house model

(cpm)

(8)

+ +f7
where

=60 /~o

where

[o

re(x)
G(x)
As (x)
E(x)
I(x)

=
=
=
=
=

mass per unit length

shear modulus
shear area
Young's modulus
moment of inertia

Iv

This formulation includes bending and shear effects. If the mass

and the moment of inertia are assumed to be constant, and if the
shear effect term is ignored, then the fixed base natural frequency
given by equation (3) is
/

[ ~ = 3.46X (60__)27rq m ~ 4 (cpm)

(4)

In this case, a computation using beam theory [8] gives

(60) q E1
[~ = 3.52 x ~
~-~ (cpm)

(5)

which shows that the approximate method produces only a -1.7

percent error!
For the case where the section properties are considered
piecewise constant and the masses are lumped at various points
along the beam (Fig. 10), the lowest fixed-base natural frequency
can be computed from

2~ V

d (cpm)

60,/F
27r Y M

(cpm)

rotational frequency of a rigid house on a flexible base

vertical frequency of a rigid house on a flexible base
rotational stiffness of support base (Pig. 7)
vertical stiffness of support base (Fig. 7)
mass moment of inertia of deekhouse about the center
of rotation
M = total mass of deckhouse
In this simplistic modeling method it is important to include
the stiffness of the deckhouse base support. Both the linear and
rotational components of the stiffness must be considered, with
the latter component being the most important for an accurate
evaluation of the fore-and-aft bending modes. This stiffness can
easily be computed from static analysis procedures. Also, an estimate of the rotational stiffness Ko can be found from the
Hirowatari method by working through Dunkerley's equation
in reverse, neglecting [~, whose contribution is generally small.
This estimation of Ko may be of interest for comparisons with the
static analysis. By computing Or,/~, and/v, Ko can be computed
from
fo
[~
Ko
k~
Or

=
=
=
=
=

/ ~ = --60 x
2~r

V~/~1N

[~ ((i-//-1) 3- 1
Mi t/=l
Eilj

(li-/])3+

(l/ -- /j-l)]
AsjGj ]

'

(cpm)

(6)

where
Mi
Gi
Asi
Ei
Ii

=
=
=
=
=

total mass at i
shear modulus
shear area
Young's modulus
moment of inertia

If the deckhouse can be modeled as a cantilever beam with both

uniform mass along its length and a lumped mass at the tip of the
beam as shown in Fig. 11, then a quick estimate of the lowest
fixed-base natural frequency f~ of the deckhouse can be obtained
from
60
f~ = ~
where
M =
m =
E =
I =

3EI
( M + 0.25ml)l a (cpm)

(7)

lumped mass
mass per unit length
Young's modulus
moment of inertia

Once the fixed-base natural frequency f~ has been calculated,

it must be corrected to account for the flexibility of the deckhouse
base support structure. This correction can be accomplished in
two ways. First, the correction factor from the Hirowatari method
(f~/f~) can be applied to the fixed-base natural frequencies f~

344

Or

L!

(9)

where fe is computed from [~ and the Hirowatari correction

factor (fJ[~).
It should be added here that Dunkerley's equation produces
a lower-bound estimate of the lowest system natural frequency.
Although the solution errors given in the preceding are remarkably small, the authors have experienced errors as large a s - 1 5
percent. If the first natural frequency of the system is much lower
than the second, then Dunl~erley's equation can be expected to
produce fairly accurate results. In the case of the cantilever beam,
the second natural frequency is 6.36 times the first; hence only
a small error is introduced. For deckhouse structures, we have
typically encountered a second bending frequency at about 2.5
times the first bending frequency, producing errors from - 7 to
- 1 0 percent. A discussion regarding these errors is given in the
Appendix.
Despite the errors inherent in this method, the predicted
frequencies always produce lower-bound estimates. This feature
is quite desirable since in most applications it is preferred to have
the lowest deckhouse natural frequencies above the operating
excitation frequency. Knowledge of the lowest natural frequency,
based on a lower-bound estimate, can be used in the early design
stages to flag potential problems and to indicate the need for a
more in-depth analysis.
Finite-element method. The finite-element method provides
a more general approach for computing the vibratory characteristics of the deckhouse structure [9, 10]. This method has been
used by the authors in all of their vibration studies to verify the
results of the previously discussed methods as well as to inves-

MARINE TECHNOLOGY

tigate the higher-order modes of the deckhouse. Knowledge of

the higher modes is particularly i m p o r t a n t when the lower
deckhouse natural frequencies lie below the design blade rate
frequency (where blade rate is equal to the propeller r p m multiplied by the number of blades) or where the higher harmonics
of the blade rate exciting force are significant.
In this method the deckhouse can be represented by a one, two,
or three-dimensional model as shown in Fig. 12. These models
can then be input to a structural finite-element program to perform an eigenvalue analysis. The output from these programs
produces accurate and detailed natural frequency and mode
shape data. Of course, the accuracy of the finite-element analysis
is only relative to the representative model and not to the actual
deckhouse structure. Therefore, to obtain useful results the
representative model must accurately characterize the global
characteristics of the deckhouse: mass distribution, moments of
inertia, shear area, and axial area. In this section we concern
ourselves only with the development of the representative
deckhouse model.
In condensing an actual ship's superstructure down to a finite-element representation, it is crucial t h a t the designer keep
the model's purpose in mind. When investigating the global
fore-and-aft vibration of the deckhouse, for example, the number
of elements and nodes need only be sufficient to describe the first
few beamlike mode shapes; local structural details need not be
included so long as the appropriate cross-sectional areas, shear
areas, and moments of inertia that represent the global stiffness
are conserved. Likewise, a high degree of mass discretization is
usually not necessary to preserve the fundamental inertial (mass)
properties of the house.
The great value of a one-dimensional model lies in its simple,
condensed form. A vertical beam, if modeled correctly, can yield
surprisingly accurate results with a minimum amount of work
and only modest computer capabilities. In the analysis of foreand-aft bending, masses may be lumped at each deck, and the
cross-sectional properties for the beam elements are easily obtained from plan views of each level of the deckhouse structure.
A vertical linear spring and a transverse torsional spring may be
used to represent the deckhouse supporting structure. From the
one-dimensional model, the beamlike mode shapes are easily
recognizable; these natural frequencies and corresponding mode
shapes can be used both as an aid to identify global modes of the
more complex models and as a check against the results of these
models.
A two-dimensional representation of a deckhouse requires a
somewhat closer look at the structure involved, a bit more time
to set up, and more computer time and storage to solve. The
house can now be modeled with a combination of truss, beam, and
plane-stress elements, where the trusses may represent transverse
bulkheads, the beams may represent principal deck stiffeners,
and the plane-stress elements may represent longitudinal bulkheads. Once again, vertical springs can be used where stanchions
and transverse bulkheads support the house. The rotational
stiffness is accounted for by the longitudinal distribution of
vertical springs.
Three-dimensional models go one step further in the refinem e n t of the deckhouse finite-element representation. It is imp o r t a n t to realize, however, that a three-dimensional model
should not be a t t e m p t e d without first gaining the insight available from the simpler models. The effort involved in developing
and "debugging" a three-dimensional model can be greatly reduced with the guidance available from the simpler computations.
Certainly, the modeling techniques outlined here are not by
any means the only ones available, and for t h a t reason the preceding description has been kept to a minimum. Sound engineering judgment and a good feel for the actual structure involved
are far more i m p o r t a n t than any predetermined modeling procedure, and, if used wisely, a procedure utilizing a series of inOCTOBER 1980

i!

,'2

2-D Model

I-D Model

3-D Model

Fig. 12

Finite-element deckhouse models

creasingly complex models can converge on a dependably accurate solution with a minimum of effort and wasted time.

Accuracy of the methods presented

To demonstrate the merits of the eigenvalue analysis procedures presented in this paper, the results from two example ships
are presented. The first example ship demonstrates the agreem e n t between the various eigenvalue analysis procedures presented herein. The second shows the correlation between our
simple modeling procedures and actual measured data from a
shaker test.
E x a m p l e S h i p 1. The eigenvalue analysis procedures discussed in this paper were applied to the deckhouse structure of
a 300-ft (91.4 m) bulk carrier. Three models shown in Fig. 13 were
developed for this study. They were designed to retain the global
characteristics of the designed deckhouse. This was accomplished
by preserving the axial area, the shear area, and the moments of
inertia between each deck level.
The one-dimensional deckhouse model (Fig. 13) was used to
predict the first bending mode in the fore-and-aft direction. This
model was composed of beam elements with varying section
properties. The deckhouse masses were lumped at the deck levels
and at the midpoints between the decks. The rotational stiffness
of the deckhouse support base was computed using the Hirowatari correction factor described earlier.
The two-dimensional deckhouse model (Fig. 13) was used to
predict the following effects:
1. first bending m o d e - - f o r e and aft
2. second bending m o d e - - f o r e and aft
3. first vertical mode
The principal advantage of this model over the one-dimensional model was its capability to account for the inertial effects
of the fore-and-aft distributed masses. It should be noted t h a t
this model eliminates the need to calculate the rotational stiffness
of the house supporting structure; this effect is accounted for
automatically by the fore-and-aft distribution of vertical
springs.
The two-dimensional model was composed of plane-stress,

345

LtmpedMass
(typ)

~
-

Plane Stress

Padar mast

-f

Masses~
(typ)~

4
Beam E l ~ n e n t ~
(typ)

~-- Truss
Elements

Truss
Elements

Main deck
irder

Rotational-A
Stiffness~
Sprin~~

>>

IAW-Vertical- ~
i
Springs ~

I-D Model
~

2-D Model

Plane Stress--~ . ~ . ~ ~
Element (typ) ~ < ~ x ~
i "~K

~~Masses

~~ "

I'~.~..~ / ~I'~

TM

Truss Element (typ)

k'/~37~.
k. /

"<'~

(typ)

" ~

"I'.. i /~

FlexibleSupportBase ~

~--Maindeck stiffeners

r.-4.~

~__9
,

3-D M e e l

Fig. 13 xample 1--deckhouse models

truss, and beam elements. The plane-stress elements modeled
the longitudinal side shells of the house and the vertical truss
elements modeled the transverse side shells. The beam elements
were used to transmit the inertial effects of the radar mast down
into the house and to model the local stiffness of the main deck
girders directly under the longitudinal side shells of the house.
The masses were lumped at the nodal points on the decks as
shown in Fig. 13.
The three-dimensional model (Fig. 13) was used to predict the
following effects:
1. first bending mode--fore and aft
2. second bending mode--fore and aft
3. first bending mode--transverse
4. second bending mode--transverse
5. first vertical mode
6. first torsional mode
The principal advantage of this model over the other models
was its ability to characterize the principal modes of vibration,
including possible coupling between modes, which could not be
ascertained with the lower-order models.
The three-dimensional model was composed of plane-stress,
beam, and truss elements. The plane-stress elements were used
to model both the side shell stiffness and the in-plane deck
stiffness of the house. The beam elements were used to model the
local stiffness of the main deck girders directly under the side
shell bulkheads of the house. The truss elements were used to

346

help retain the original moments of inertia of the deckhouse cross

sections by adding additional distributed area. The house masses
were lumped at nodes distributed throughout each deck. In the
three-dimensional model, the stiffness representations for the
stack and the radar mast were omitted in an effort to maintain
the simplicity of the model; their masses, however, were lumped
on the house top.
Natural frequencies for the first fore-and-aft bending mode
are given in Table 2. A few observations regarding the analysis
should be noted. First of all, Deckhouse Type A was used in the
application of the Hirowatari method. Secondly, the one-diTable 2

Example 1--fore-and-aft bending frequencies

HirowatariMethod

567 c7~

Simp]isticModelingMed]od
using Dunker]ey's
equation (6).

532 cpm

I-D FiniteE]ementMode]

551 c]gm

2-D FiniteElementMode]

627 cpm

3-D FiniteElementMode]

525 cTxn

Coup]c<]2-D housemode]
and hu]] girder.

577 c7~

MARINE TECHNOLOGY

FictitioUS~surge
~
sprlng

~2-D

House

_ ~ j/-Underdeck supporting structure

I 1~"
/-I-D Hull girder

Buoyancy

Fig. 14

"

Coupled house and hull girder model

mensional model shown in Fig. 13 was used in the simplistic

modeling approach. The short program given in the Appendix
was the main computational tool used in this analysis (Dunkerley's equation). Thirdly, all three models shown in Fig. 13 were
analyzed by the finite-element method. Finally, an analysis was
made coupling the two-dimensional deckhouse model with a
one-dimensional hull girder model as shown in Fig. 14 to demonstrate the minor coupling effects of the hull girder.
The accuracy of the results given in Table 2 can be evaluated
by considering the deviations in the computed natural frequencies about some norm. The results obtained in this analysis, excluding the coupled house and hull girder analysis, deviate from
-6.3 to +12.0 percent about an average frequency of 560 cpm.
The Hirowatari method gives an estimate which is quite close to
this computed average.
The trend indicated by the progression of frequencies computed from the finite-element models indicates that there is some
inconsistency in the development of the increasingly complex
models. Normally, one would expect the 1-D model to have the
highest natural frequency estimate and the 3-D model to have
the lowest estimate, since increasingly complex models tend to
be more flexible. In this case the inconsistency is resolved by
noting that the support base stiffness has been computed from
the Hirowatari method for the 1-D model and by a static analysis
procedure for the 2-D and 3-D models.
The simplistic modeling method produced results which agree
quite well with the results obtained from the 1-D finite-element
analysis. The success of the method, as used in this example, is
due to the fact that this deckhouse structure closely resembles
a cantilever beam.
Generally speaking, all of the results do show fair agreement
with each other, indicating the reliability of the models and the
methods.
Example Ship 2. The profile of the deckhouse structure from
the second example ship, an 800-ft (244 m) ore carrier, is shown
in Fig. 15. The basic shape of this superstructure--a tall house
and funnel linked by a longitudinal truss--is not easily identified
by one of the superstructure types presented by Hirowatari, and
for this reason we developed a one-dimensional finite-element
model for our first estimation of the primary fore-and-aft deckhouse natural frequency. The flexible base model is shown in Fig.
16.
The model consists of two vertical beams representing the
house and the engine casing-funnel. These two beams are linked
by a horizontal truss element as required by the deckhouse geometry. The beams are supported by vertical linear springs as
well as lateral torsional springs. These spring supports must accurately account for the stiffness of the deckhouse supporting
structure. The superstructure masses are lumped at the decks
while the sectional properties of the funnel and house are represented between decks by piecewise continuous beams.
An eigenvalue analysis using the one dimensional model described in the preceding was performed and compared with experimental data measured from shaker tests. Results are given
in Table 3. This analysis demonstrates both the accuracy obtainable from a careful eigenvalue analysis procedure as well as
OCTOBER 1980

springs

Funnel

Truss

I
Fig. 15

I Huse

l
Fig. 16

House profile

House model

the potential dangers involved in the misuse of simple analysis

procedures.
This deckhouse type does not fit one of the types studied by
Hirowatari; hence, in this case it is difficult to use this method.
One could argue that this house could be represented by a combination of Types A and B, with an average frequency of 513 cpm
computed from the two house types; however, this procedure is
foolhardy and could lead to disasterous conclusions.
The simplistic modeling method works well when the second-system natural frequency is at least twice that of the firstsystem frequency. It is doubtful that this house could satisfy this
criterion, since the funnel and the house may have separate first
bending frequencies lying close to each other. The finite-element
analysis of this deckhouse indicates that the first bending modes
do in fact differ by a ratio of only 1:1.7. Consequently, the error
introduced by applying the simplistic modeling method is substantial. An important point to bear in mind is that this method
does produce lower-bound estimates. If the natural frequency
produced by the method lies well above the operating excitation
frequencies, then the fore-and-aft bending mode of the house
probably will not be excited.
The finite-element method provides the best tool for handling
the eigenvalue analysis of this deckhouse. This method easily
accounts for the deckhouse configuration and produces accurate
estimates of the fore-and-aft natural frequencies as is indicated
from the shaker test measurements. The analysis comparing the
house-only analysis with the house-plus-funnel analysis shown
in Table 3 demonstrates the importance of including the funnel
in this investigation.

Conclusions
The discussion of the procedures given in this paper has been
Table 3

Example 2--fore-and-aft bending frequencies

Hirowatari - House only - Type A

453 cpm

Simplistic Modeling - House only

477 c~n

House-p] us-Funne ]

375 cvm

Finite E]eraent Method - House only

House-p] us-Funne ]
Shaker Test Measurement

559 cpm
518 cpm
528 cpm

347

intended to increase the designer's awareness of analysis procedures which will lead to the identification of global deckhouse
vibration problems during the early design stages.
The eigenvalue analysis can play a central role in the early
stages of design by providing the designer with valuable information regarding the natural vibratory characteristics of the
deckhouse structure. This information can provide guidance for
making design decisions as well as for simplifying the forcedresponse calculations, if required.
It is important t h a t the designer realize t h a t large complex
models are not always synonymous with correct results. Typically, the results obtained from these models are difficult to interpret, since the output may contain clusters of local and erroneous information which tend to obscure the more i m p o r t a n t
global information. The authors endorse the use of an eigenvalue
analysis procedure which utilizes a series of increasingly complex
models. The simpler models can be used to determine the need
for a more complex analysis as well as to provide guidance for the
development and evaluation of the more complex analysis.
Although the procedures described in this paper have been
used with success in all of the vibration studies performed by the
authors, it m u s t be k e p t in mind that for some studies an alternative approach, which may not have been covered in this paper,
may in fact be more appropriate. An understanding of the principal features of the deckhouse vibration phenomenon is the key
to the selection of a creative and successful analysis procedure.

Acknowledgment
The authors would like to express their sincere thanks to
Professor William S. Vorus for his valuable advice during the
development of this paper and for his cultivation of our interest
in structural dynamics.

References
1 McFarland, R. and Lindquist, D., "Vibration from a Ship Owner's
Standpoint," SNAME Ship Vibration Symposium, Arlington, Virginia,
Oct. 16-17, 1978.
2 Maniar, N. M. and Daidola, J. C., "The Considerations of Vibration and Noise at the Preliminary and Contract Levels of Ship Design,"
SNAME Ship Vibration Symposium, Arlington, Virginia, Oct. 16-17,
1978.
3 Glasfeld, R. D. and MacMillian, D. C., "Vibration from a Shipbuilders Point of View," SNAME Ship Vibration Symposium, Arlington,
Virginia, Oct. 16-17, 1978.
4 Vorus, W. S., Stiansen, S. G., and Bertz, R. H., "Correlations of
Propeller Induced Forces and Structural Vibratory Response of the MW
Roger Blough," University of Michigan, Department of Naval Architecture and Marine Engineering Report 193, Ann Arbor, Mich., Oct.
1977.
5 Armand, J. L., Orsero, P., and Robert, 0., "Dynamic Analysis of
the Afterbody of a Ship--Towards the Successful Correlation Between
Analytical and Experimental Results," SNAME Ship Vibration Symposium, Arlington, Virginia, Oct. 16-17, 1978.
6 Hirowatari, T. and Matsumoto, K., "On the Fore and Aft Vibration of Superstructure Located at Aftship," Journal of the Society of
Naval Architects of Japan, Vol. 119, June 1966.
7 Hirowatari, T. and Matsumoto, K., "On the Fore and Aft Vibration of Superstructure Located at Aftship (Second Report)," Journal
of the Society of Naval Architects o/Japan, Vol. 125, June 1969.
8 Thomson, W. T., Theory of Vibration with Applications, Prentice-Hall, Englewood Cliffs, New Jersey, 1972.
9 Kavlie, D. and Aasjord, H., "Prediction of Vibration in the Afterbody of Ships," Norwegian Maritime Research, Vol. 5, No. 4, 1977.
10 Hansen, H. R. and Skaar, K. T., "Hull and Superstructure Vibrations-Design Calculations by Finite Elements," Det norske Veritas,
Publication No. 86, Oslo, Norway, Jan. 1975.

Dunkerley's equation for a system with N degrees of freedom

as

_1_1 = ,~.
i = 1 COi 2

(10)

i=10Jii 2

where 0:i represents the ith natural frequency of the system and
0:ii = ~ / K i / m i represents a natural frequency of the system when
m is considered to be the only mass on the system. I n the application of equation (10) to beams, 0:2 through 0:n do not contribute
substantially to the sum
N

i--~l 0:i 2

hence, an approximate form of Dunkerley's equation can be expressed in discrete form as

1

-0:i
- ~ -2-

(11)

1 = ~ mi

t.=
~ l ~ i i .. 2

i=1 h i

or in continuous form as
1
- -

(12)

dx

0:i 2 --

The lowest natural frequency of a cantilever beam with variable mass and variable section properties can be estimated
through the approximate form of Dunkerley's equation. Two
cases are considered: a continuous beam and a lumped-mass
beam. A short computer program has been provided for the
lumped-mass case. Finally, a discussion regarding the errors
produced by this method is given.

Case 1 - - c o n t i n u o u s beams
Consider the beam shown in Fig. 9 in the text with continuously varying properties. In this application we apply Dunkerley's equation in the continuous form as given by equation (12).
To apply this equation to the cantilever beam of Fig. 9, k (x) must
be evaluated. This stiffness is readily obtained by the displacement method, where the stiffness is equal to a force F applied at
x divided by the displacement 5 at x due to F. Applying Castigliano's theorem, the displacement 5(x), including bending and
shear terms, is given by

~v*

1
/

5M(~)

x/M(~)I-~-

where
U* = complementary strain energy
M(() = F(x - ~)
v(~) = F
Substituting values for M(~) and V(~) we get

Jf x [ ~ [ ( x ~)2 + As(~)-G(~)] d~
-

(14)

The stiffness k (x) is then computed by

F

k(x) = a(x-~ ~ f o x [[(x~- ~)2 + As(~)G(~)]

~
d~

(15)

Appendix
Dunkerley's equation provides an easy means for estimating
the lowest natural frequency of a given system. Thomson [8] gives

348

Finally, applying this result, equation (15), to equation (12) and

solving for wl we obtain

MARINE TECHNOLOGY

L
= ~/2

dx

L m(x) C

[(x

1~)2 +

Jo x [ ~

(16)

As(~)G(})

This equation gives, in radians per second, a lower-bound estimate for the lowest natural frequency of a cantilever beam with
continuously varying properties.

Case 2 - - l u m p e d - m a s s beam
Consider the lumped-mass cantilever beam shown in Fig. 10
in the text. In this case we apply Dunkerley's equation in the
discrete form as given by equation (11). As before, the stiffness
ki corresponding to mi must now be evaluated. Applying equation
(15) and noting that the section properties E, I, As, and G have
been assumed to be piecewise constant, ki can be written directly
as

ki =
1

(17)

lj (ll - ~)2d} + ~

j=l

d}

--1

-1

Noting that

(li - lj-1) a. (li - lj) a

f i ! l (li - ~)2 d~ -

(18)

and that

~lj!l d~
we can now write

lj

--

lj-1

(19)

lj) a + ~ ]
AsjGj ]

(20)

ki as
1

ki =

I(li

j=]

- -

/j--i) 3

(li

- -

3EjIj

- -

i
2
3
4
5
6
7
8
9
i0
ii
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60

C........................................................
-C
C
App]ication of Dunker]ey's equation (12a) to the
C
C
computation of fixed base natura] frequencies
C
C
of a cantilever beam with 1 ~ p e d masses and
C
C
varying section properties.
C
C
C
C
Variable ]ist:
C
C
N
- N~nber of ]u~ped points on the beam.
C
C
M(i) - L ~ p e d mass at the i-th point.
C
C
L(i) - Distance between the i-th point and the
C
C
fixed base.
C
C
IY (i) - Moment of inertia between the points
C
C
i and i-1.
C
C
AS(i) - Shear area between the.points i and i-l.
C
C
(This va]ue can be ignored by setting it
C
C
equa] to zero.)
C
C
E(i) - Young'smodu]us.
C
C
G(i) - Modu]us of rigidity.
C
C
C
C
Data format:
C
C
Card 1 - (i5)
C
C
N
C
C
Card 2 - (6FI0.3) - Need "N" cards (One per point)
C
C
M, L, IY, AS, E, G
C
C ............................................................. -C
C
DIMENSION M(20) ,L(20), IY(20) ,AS (20) ,E(20) ,G(20)
REAL M,L, IY
C
.READ (5,100) N
READ (5,200) (M(I),L(I),IY(1),AS(I),E(I),G(I),I=I,N)
SUM1 = 0.0
POLAR = 0.0
DO i0 I=I,N ,
SUM2 = 0.0
DO 20 J=l,I
IF(J-I.EQ.0) GO TO 30
SUM2 = SUM2 + ((LII)-L(J-I))**3 - (L(I)-L(J))*~3)/
+
(3.0*E(J)*IY(J))
IF(AS(J) .EQ.0) GO TO 20
SUM2 = SU~2 + (L (J) -L (J-l)) / (AS (J) *G (J))
C~D TO 20
30
SUM2 = Sb~2 + (L(I)**3 - (L(I)-L(J))**3)/(3.0*E(J)*IY(J))
IF(AS(J) .EO.0) GO TO 20
SUM2 = S[~42 + L(J)/(AS(J)*G(J))
20
CONTINUE
SL~I = SUM1 + M(I)*SUM2
POLAR - POLAR + M(I)*L(I)**2
I0 CO~rfINUE.
CPM = 9.5493 * SQRT(I/SUMI)
~TRITE (6,300) CPM
WRITE (6,400) POLAR
C
i00 FOR~@kT (I5)
200 FORMAT (6FI0.3)
300 FOPS[AT ('0The frequency in c~n = ',E12.6)
400 FORMAT ('0The mass moment of inertia of this beam = ',E12.6)
C
STOP
END

LND OF FILE

Applying this result (20) to equation (11) and solving f o r ~ l w e

obtain
~ l e

0~1~

i~

N
mi

=1

(li - l j - l V - (li - l y + ~ ]

3EjIj

(21)

Solution errors

~NDOF

Sample

execution

The

of the results obtained by this method of analysis. Errors in the

#EXECUTION

OCTOBER 1980

for

the

program.

FILE

The application of Dunkerley's equation to deckhouse structures, which can be modeled as cantilever beams, is a relatively
straightforward task. Since the method developed in this paper
is an approximation, it is important to determine the accuracy
method are generated from the fact that the approx,mation assumes

file

7
522.620
108.300 1.500E+08 1.410E+03
3.0E+07 1.15E+07
455.610
216.540 0.693E+08 1.010E+03
3.0E+07 1.15E+07
466.510
324.800 0.816E+08 0.775E+03
3.0E+07 1.15E07
462.360
433.070 0.650E+08 0.994E+03
3.0E+07 1.15E+07
466.780
541.340 0.629E+08 1.140E+03
3.0E+07 1.15E+07
386.890
649.610 0.560E+08 1.300E+03
3.0E+07 1.15E+07
062.030
751.970 0.019E+08 0.095E+03
3.0E+07 1.15E+07
123456789 123456789 123456789 123456789 123456789 123456789

A4Gj I

A short computer program (Fig. 17) has been developed to solve

this equation. The program, sample data file, and computed result 01 have been included. Program documentation is given in
the program listing.

data

#EXECUTION
The

the

program.

BEGINS

frequency
mass

of

in c p m

moment

of

= 0.858986E+03

inertia

of

this

beam

= 0.498553E+09

TERMINATED

Fig. 17

Computer program

349

('012 0)approx2

= ~,-

(22)

i=__ COi2

T h i s works only w h e n co1 << co2, . ,coN.

T o examine the error introduced by this approximation, write
co2 t h r o u g h coN as multiplies of co1
-- --~1

~0approx2

i= (0)10~i)2

(23)

5012i=10~i 2

where o~1 ----1.0 and ai < ai+l.

N o w the error can be expressed as:

%errr=lWapPrx-Wl}
l Oco1
Ot

(24)

Noting that

O)approx = CO1

we now write the error as

350

~-~

(25)

%error={
Several observations can now be m a d e regarding t h e error int r o d u c e d in coapprox:
1. As a2 a p p r o a c h e s infinity the error is r e d u c e d to zero.
2. T h e error will always be negative since

,/1
i=l

< 1

(27)

T h i s means t h a t Wapprox < co1; hence C%pproxwill always represent

a l o w e r - b o u n d e s t i m a t e of co1.
3. T h e errors are governed by the distribution of the natural
frequencies of the system (~i); hence there is little t h a t the analyst
can do to reduce t h e error p r o d u c e d by this m e t h o d . Obviously,
it is i m p o r t a n t to avoid the use of this m e t h o d on systems where
the s e c o n d - s y s t e m natural f r e q u e n c y lies near the first.

MARINE TECHNOLOGY