HUSSEIN A. KAMELt
University of Arizona
and
DONALD LIU:
American Bureau of Shipping
AbstractThe application of the finite element method to the analysis of a ship hull structure is discussed.
An analysis of vessel types, the appropriate computational approaches and critical loading patterns are
provided. The various types of elements available for modeling the structure and thetr limitations are de+
cribed. A summary of techniques designed to obtain the overall behaviorofthe vessel as well as local stress
details is provided. Examples illustrate the use of automatic data generating routines. The incorporation
of decision making within the computer program relieves the user of many unnecessary details and insures
against human error.
I. INTRODUCTION
THE SUCCESSFUL use of the electronic digital computer in the analysis of complex structures
of arbitrary configuration can be primarily attributed to the advances made during the
last twenty years in the finite element method. Although this approach was originally
developed with a view towards analysis of aerospace and civil engineering structures
[l3], it is now extensively used in the field of ship structures. The purpose of this paper
is to discuss the various aspects of application of the finite element method to ship struc
tures, some of the problems involved, and to suggest methods that may be useful in solving
them.
In adopting the finite element method an organization has to undertake a certain amount
of preparation and change. As in most engineering applications, the theory provides the
framework within which the engineer must exercise judgement in order to reach a valid
and economical solution. Computational tools and techniques as well as experience must
be accumulated before the theory can be meaningfully applied. This necessitates a re
education of the technical personnel. From the managements* point of view, the dollar
cost of performing a finite element computation relative to the value of the expected
results is a major factor in determining the desirability of the analysis. Another vital
consideration is the speed by which an analysis can be performed, so that a deadline
may be met. The time required to perform a computation depends not only on the available
programs and computer type but also on the experience and training of personnel and the
complexity of the problem.
103
104 HUSSEINA. KAMEL and DONALD LIU
In the application of the method to ship structures, original efforts were directed
mainly at twodimensional analyses of a webframe or three dimensional analyses of a short
segment of the ship. It was only recently that a demand for a threedimensional analysis
of a complete vessel emerged [4]. Traditionally ship structural design criteria are based
on long experience as set forth in the rules of ship classification societies such as given by [5].
The last ten years have seen many departures from conventional ship design with respect to
ship size and types. For example, oil tankers have increased in size from a typical 30,000
tons to well over 300,000 tons. The period saw the birth of the containership and the great
increase in the size of bulk carriers. Since for these vessel sizes little or no experience has
been accumulated, more rational methods of analysis have to be employed. Should these
new methods prove reliable, experimentation in design will be encouraged and the current
revolution in ship construction will be further enhanced.
In contrast to the aerospace industry, a small number of ships, often only one, of a
particular design are built. This limits the amount of funds available for the analysis. It is
therefore more imperative to seek efficient, inexpensive tools in performing the computation.
The total expense for problem solution consists of both computer charges and man hours,
Contrary to the belief of many, the man hours constitute a major part of the expenditure.
and more automation is called for. A reduction of the necessary man hours will also help
to meet production deadlines.
Automation of model generation and interpretation of the results will reduce man hours
while increasing computer charges. The net effect is usually a decrease in the total costs.
New numerical techniques are continuously devised in order to speed up existing programs.
New refined finite elements deliver, in general, more accurate solutions for the same com
puter time expenditures but may result in a more complex modelling procedure. Alter
natives such as substructuring or the super element technique, briefly described later,
achieve similar or often better results, using simpler elements.
Parallel to developments in available elements, programs and numerical methods,
improvements and innovations in computer hardware have also a profound effect. As
computational speeds greatly increase from one generation of computers to the next, com
puter charges are bound to decrease. Interactive graphics terminals seem to be the ultimate
answer to the man hour problem.
Of the large merchant ships, (about 1000 ft in length) being built today, several types
predominate. They are the oil tankers, and open deck type vessels such as bulk carriers
and container vessels. Because of their large size, their structural strength must be assured.
Typical structural configurations are shown in Figs. 1 and 2. The areas of emphasis in the
structural analysis depends on the vessel type as described below.
II. 1. Oil tankers
Oil tankers are weight limited ships, in the sense that the amount of liquid cargo that
may be carried by the vessel is limited by weight rather than by volume. They may displace
up to 300,000400,000 tons of sea water with their bottoms being perhaps 70 ft below the
sea water surface. Because of the large pressure heads due to the draft and liquid cargo in
the tanks, high shear forces develop in the web plating of girders and transverses near the
bottom, particularly if a cargo tank is empty, Fig. 3. Because of the high shear forces,
the depth of the webs has to be large thereby increasing the danger of elastic instability.
Application of the Finite Element Method to Ship Structures 105
Section AA
In addition to the local transverse web frames, the longitudinal strength of the entire
vessel deserves equal attention. Because of the extreme length and depth of these vessels,
the distributed wave and cargo loading along the length causes appreciable bending of
the hull, resulting in high tensile and compressive stresses in the outer fibres of the hull
girder. This, combined with lateral pressure from sea and cargo render the plating sus
ceptible to buckling. The problem of instability is further enhanced by the use of higher
strength steel in these regions for their weight saving characteristics, resulting in thinner
plating.
PIor of symmetry ._
Although it is desirable to model the entire hull in order to correctly evaluate the res
ponse of the structure, less exacting procedures are often employed For example, in tanker
structures where the objective of the analysis is to obtain stresses in the transverse strength
members such as web frames and bulkheads, three tank lengths of the hull may constitute
the primary structure for analysis. The stresses calculated in the transverse members wili
depend therefore primarily upon the local tank loading and the relative vertical displace
ments of the center line, longitudinal bulkhead and side shell. To analyze only three tank
lengths, displacements must be assumed at the boundary of the tanks where they have been
cut from the entire hull structure. These boundary displacements are customarily
approximated by assuming the hull to behave as a simple beam under vertically applied
cargo and sea pressure loads. Using the engineers beam theory, the translational and rota
tional displacements of the beam neutral axis at the location of the cuts can be imposed
onto the 3D model as translational displacements, assuming that the tanker cross section
remains plane in bending about the neutral axis. At a transverse section, the centerline,
longitudinal bulkhead and side shell may be assumed to deflect by an equal amount in the
108 HUSSEIN
A. KAMEL and DONAH) LIU
vertical direction due to the beam bending behavior of the hull structure or relative dis
placements may be estimated using a separate analysis. By modeling only three tank spaces,
(one tank on either side of the tank of interest), the local effects of cargo loading in adjacent
tanks is included.
More drastic simplifications are sometimes made in finite element analysis of local
areas. Twodimensional models of webframes using membrane elements are often em
ployed, and in some instances beams and rods. Plate and shell structures have also been
solved using the Hrennikoff model made out of rods.
Although the authors do not doubt the usefuiness of results obtained through intelli
gient nse of simplified models, it would still seem desirable to conduct a computation rc
presenting the physical structure as closely as possible. The amount of simplifying assump
tions that can be tolerated in an analysis depends on the type of vessel, the nature of the
loading and the purpose of the computation.
Symmetry of loading and structure can always be used to reduce the problem size
prior to computation. in some cases even an arbitrary loading on a symmetric structure
can be converted into symmetric and antisymmetric loading patterns thus reducing the
computational effort significantly.
Oil cargo is symmetrically loaded about the centerline plane of a tanker. Although
waves oblique to the longitudinal axis of the vessel are present, only longitudinally directed
sinusoidal waves are considered critical. Due to the overall symmetry of the problem,
the size of the model analyzed can be reduced by half, thereby decreasing the computa
tional time.
For a container type vessel, the critical sea induced torsional loading is caused by waves
obliquely directed to the longitudinal axis. This results in a nonsymmetrical loading about
the centerline plane. This arbitrary loading can be represented by the superposition of two
loading modes, one symmetric and the other antisymmetric. In considering the two load
cases, only one half of the vessel need be modeled, symmetry and antisymmetry being im
posed by selecting the appropriate boundary constraints at the centerline plane. In order to
illustrate the savings involved let us consider a vessel modeled using s stations, each in
volving u degrees of freedom. The total number of unknowns, U, is equal to su and the half
band width, b, is 2~. Using Gaussian elimination, taking symmetry into account and as
suming L loading cases, the number of multiplications and additions involved in the
operation is
This number is representative of the computing effort necessary to solve the problem.
It is seen that, by reducing the number of freedoms per station by a factor of 2, N is reduced
by approximately 8. The above figure does not take into account the presence of mass
storage operations, which may have a significant influence. It is also interesting to note
that the addition of a loading case increases the time by a factor of 2/(21+ 2L). For example,
if u = 150 and L = 20 the additional time required to solve one more loading case is approxi
mately one percent of the original time.
Application of the Finite Element Method to Ship Structures 109
stress distributions must satisfy certain criteria so that the solution converges to the correct
answer as the finite elements are made smaller in size. In the case of kinematically idealized
elements, the correct solution is an upper bound since the element deformation is constrained
by assuming certain displacement patterns. The resulting strain energy and flexibility of
the structure are both underestimated. On the other hand, a static idealizationoftheelement
(assumed stress distribution) results in an overestimated structural flexibility so that the
correct solution is a lower bound [2].
With an assumed displacement or stress function, it is not possible to satisfy both
conditions of equilibrium and compatibility simultaneously. The hybrid method of element
formulation [8] combines features of both methods. Equilibriun: stress fields are chosen
within the element and displacement functions are cilosen on the boundary of theelement
in order to satisfy interelement compatability. The hybrid element formulation is currently
receiving widespread attention and appears promising.
Of the three methods of clement formulation, the assumed displacement method is the
most common. The resultant force deformation relationship is in terms of the element
stiffness which can be used directly in :he finite element matrix displacement method.
then assembling them in a suitable manner to obtain the stiffness of the entire element.
In subdividing the element, internal node points are specified at the midsides so that cubic
polynomial displacement functions can be employed. This increases the number of degrees
of freedom for the basic element. The degrees of freedom are then reduced by enforcing
compatibility of the internal nodes and by a static condensation process eliminating internal
degrees of freedom [18, 191.
1 w, Uz, wy 9 3 No 23
3 4 1 n
Node 1
II, Wr,
Node 2 :
w, Wn
wy 1.5 4 YCS 36
27

Nodt! I
II, wr, WI 24
4 l~ZZ, WYY, WLV 21 5 Yes 28
Node 2 :
Wfl 29
5 w, WI, WY 9 3 No 18
I
Node 1
II,ws. WY 12 3 Yes 19
6 n
Node 2
Hn
112 HUSSEIN A. KAMEL and DONALIILw
TABLE 2
t 1
7 w, wr, wy, wzy 16 5 YCS 30
i i 31
,I
D
.
, _I
30 ,y_._
12 3 Yes 19
,
I
Table 2 illustrates some rectangular and quadrilateral plate bending elements. Elements
9 and 10 are macro elements composed of preassembled triangular plates.
The so called isoparametric elements present another example of kinematic idealiza
tion [20221. Polynomial displacement functions are chosen in terms of coordinates natur
ally associated with the geometry of the element boundaries, called element or intrinsic
coordinates. By so doing, the polynomial is uniquely defined by values of the function at
the boundary nodes, hence ensuring compatibility of adjacent elements. The same functions
chosen to describe the intrinsic coordinate system are also used to describe the displace
ment field within the element, The isoparametric concept is useful for developing families
of one, two and three dimensional plane stress and bending elements of irregular geometry
from a basic, or parent, element. By specifying more nodes on the element boundaries
and increasing the order of the displacement polynomial accordingly, elements of greater
complexity can be formulated.
In the derivation of kinematically idealized elements using polynomial functions, the
degree of polynomial selected is dependent upon the number of degrees of freedom to be
allowed for the element, which is a function of element type and number of boundary
Applicationof the Finite Element Method to Ship Structures 113
points. By using a higher degree displacement polynomial, more deformation patterns can
be represented, thus improving the structural behavior. For this purpose additional node
paints must be specified, either within the element or along its edges. Alternatively, more
freedoms must be used per node (curvature, twist, etc.). Improvements in accuracy thus
obtained must be balanced against some less desirable features. In particular, intermediate
node points along the element sides complicate automatic mesh generating procedures
and increase the band width due to the element. By including mare than six freedoms per
paint, we limit our abiiity to handle complex structures in general, and multiconnected
shells, in particular. The question whether it is preferable to use a large number of elements
having loworder displacement polynomials or a smaller number of elements having higher
order displa~me~t polynomials is still to be resolved,
The foregoing element presentation is intended as a brief review of some of the more
typical elements potentially useful for ship structural analysis. In selecting the element
types for a particular model, consideration must be given to such factors as element ac
curacy, reliability, and simplicity as well as computational costs. If more than one element
type is to be used interelement compatibility must be borne in mind.
The idea of conducting local analyses subsequent to the analysis of a larger area in
order to zoom on a local area of interest was mentioned and used in structural analysis
prior to the advent of the finite element method. An excellent example of its use in the area
of finite element analysis of ship structures can be found in [34]. The applicatian of a local
analysis subsequent to an analysis of the complete hull is given in [4]. There, the purpose
of the overall analysis is to obtain the displacements for the whale vessel under varying
loading conditions. Guided by the displacement results, local areas of interest are chosen,
isolated and remodeled. A finer analysis is then conducted using a mare detailed local
loading configuration together with boundary displacements from the first analysis. A
related but more promising procedure is that of the super element which has been success
fully employed in 1969 and 1970 at the University of Arizona and Standard Oil of California
[351/. This procedure can be described as a local analysis of a refined made1 (super element)
within the overall analysis. An automatic process far reducing the interaction freedoms
between the super element and the rest of the structure using interpolation functions is
employed. The more refined local model is integrated within the computation of the overall
vessel and local results can be automatically generated.
An alternative to a displacement interface between a local analysis and a larger one is a
force interface. The authors believe that the tist procedure is easier to apply and is mare
consistent with the nature of the displacement method.
R=Kr (2)
where 1yis the original matrix of the model. Let the reduced set of displacements be denoted
by q so that
r=Tq (3)
where T is a suitable reduction matrix obtained from the applied constraints. it can be
shown that the equivalent generalized load Q can be obtained from
Q = TR. (4)
Equation (2) is transformed into
Q=K,q (5)
where
Kc,= TKT. (5)
Equation (6) may be solved economically for q. If the constraints are chosen appro
priately, the solution will still be capable of indicating areas of local high stress. By re
moving constraints in these high stressed areas and rerunning the problem, a better estimate
of the detailed stress distribution may be obtained. Whereas this procedure may involve
the generation of a large number of structural elements, it may be automated to a great
degree. We believe that the convenience and ease of zeroing on the trouble spots without
necessarily conducting large scale computations can be of great advantage to the designer.
Assumed constraints can range from those of the theory of bending of beams where
cross sections remain plane and retain the geometrical shape on the one hand, to more
sophisticated behavior patterns which may include higher order polynomial representation
for the deformation of the sideshell, deck and bulkheads on the other. We particularly
recommend this procedure for use in conjunction with interactive graphics oriented methods.
load together with a finite element analysis, the resultant stresses are considered only as
representative stress levels rather than absolute stress values. In the absence of accurate
pressure distributions, the static calculation can be applied to a number of wave load
ing cases that create various modes ofcritical loading (maximum bending moment, maximum
shear, maximum torsion). If the results of the stress analysis are then compared to
measured maximum values, then the wave loading can be readjusted so that the maximum
stress levels are the same. It is almost certain that areas of stress concentration that may
give rise to failure during ship operation will also show high stress levels in the static
computations.
Recent advances in the analysis of waves and in defining the response characteristics
of a ship in waves has removed many of the difficulties in prediciting ship motions in realis
tic seas [36381. It is now possible to solve the equations of motion of a vessel moving in
regular waves with a high degree of accuracy. The response of a vessel in irregular waves
can then be considered as the summation of responses to regular waves of different fre
quencies [36]. A knowledge of the vessel response permits the calculation of the hydro
static and hydrodynamic pressure distribution along the length of the vessel [39]. It has
been shown analytically and from model tests [40, 411 that maximum values of the hull
bending moment as calculated from solutions of the equations of motion are less than those
calculated from the static method. This may be attributed to dynamic effects.
Of greater interest for design purposes is the ability to predict the maximum expected
loading on the vessel during its service life. A statistical approach must be taken in order to
establish this maximum. Beginning in 1960, a long range research program was started at
Webb Institute of Naval Architecture under the sponsorship of the American Bureau of
Shipping, for theoretically predicting the maximum expected hull bending stresses in a
vessel [4244].
In addition, several vessel types operating in various trade routes are instrumented to
provide realistic data for correlation and verification of the theory.
As a consequence of these recent advances, a better understanding of the loading on a
vessel will be achieved. A more realistic load description coupled with an accurate stress
analysis technique such as the finite element method will result in establishing a more ra
tional design and analysis procedure. Using such a procedure the designer can more
accurately develop and analyze new and unusual vessel types. It will also provide him with
a tool for exploring optimum designs with confidence.
The generation of the computational finite element model as well as the associ&d
nodal point loads necessary to perform a finite element analysis constitutes one of the
most timeconsuming portions of the total effort. The interpretation of the resultant
deflections and stresses represents the other expensive part of the procedure. In thi<!area
work has been initiated [4, 45471 and much more remains to be done. In our opinion,
the benefits in reducing cost and time will more than compensate for the initial investment.
The model generation is also perhaps that part of the computation most prone to errors.
In order to check for these errors and reduce them to a minimum, a process of automation
of model generation coupled with graphics representation is almost compulsory. Auto
matic and semiautomatic mesh generators can be written to speed up model generation
116 HUSSEINA. KAMEL and DONALD LIU
procedures. One example is the program EXAM [4] designed to produce finite element
models of complete vessels from scantling tables, The results of such a program can
typically be demonstrated by Fig. 6 where a plot of the deck of a container type vessel is
shown. Another example is the program MEG [48] designed to generate models for local
areas of interest as shown in the web frame of Fig. 5. The optimum mesh generator is,
in our opinion, a reasonably efficient near perfect automatic system which allows human
interference. For that purpose interactive graphics seem to provide the answer.
Automation also can be extended to other aspects of the computational procedure.
For example, the choice of freedoms for a complex structure as well as the optimum num
bering of nodal points in order to obtain the best stiffness matrix patterns, and automatic
detection and suppression of singularities during a solution will all be discussed in the next
sections.
XRX I5 [X10 I
It is assumed that structural nodes may each have a maximum of six degrees of freedom.
three translational and three rotational. Which of these six degrees of freedom are to be
allowed and which are to bc suppressed depends on many considerations. The freedoms
which a particular node is permitted are those along (or about) which the structure can
resist a force (or moment) if all other nodes of thestructure are fixed. The total numberwill
depend on the type and geometrical orientation of the elements attached to the node.
Another factor influencing the choice of freedoms at a point is the presence of symmetry
or antisymmetry about an axis or a plane passing through the point. The presence of
symmetry will override the element considerations and eliminate some of the freedoms
already allowed. The presence of kinematic boundary conditions such as rigid supports or
prescribed displacements would override previous results. Of all three factors the first is
the more difficult to determine. The following describes a procedure which has been
successfully tested for automatic generation of freedoms in a complex structure composed
of homogeneous elements with a maximum of six displacement unknowns at each nodal
point.
Figure 7 describes two basic freedom vectors. The first is a translation vector in an
arbitrary position in space having direction cosines 1,, ~1, and II]. The second is a basic
rotation vector which resists moments around an axis, the direction of which is described
by the three direction cosines, 12, r~?~and n2. Each finite element is replaced by a number
of basic translation and rotation vectors acting at its corner points. Figure 8 shows typical
examples of how some element types may be handled in this manner.
, t
,_.,,,. ,
/,, *
,,/?
. 
e
/
Degrees of freedom at Q node
,;;
/ and associated direction cos,nes
Bending triangle
The flow diagram of Fig. 9 presents an overall description of the procedure for a struc
ture with P nodes and E elements. First, freedoms of all nodes are suppressed and for each
node two sets of direction cosines, one for translational and one for rotational freedoms
are initialized to zero. The program then scans all elements. For each element the type is
determined as well as the types and directions of the basic vectors required to represent it.
01
e= I, E _ Choose elemegt e
Replace by a number
of, basic translational
and~rototional vectors
t
All element  Update freedoms and
corner pts. directiancasines
of corner pts.
of elements
Fro. 9. Overall flow diagram to obtain basic freedom pattern of structural nodes.
120 HUSSEINA. KAMELand DONALV LIU
The vectors associated with each nodal point of that element are then applied to the cor
responding structural node and the freedom pattern of that node as well as the correspon
ding direction cosines are updated. Figure 10 demonstrates the logic involved in choosing
the direction cosines and the number of degrees of freedom for either transfational or
rotational displacements at a node. IT,,represents the current number of freedoms at the
node p. C, represents a set of direction cosines associated with the point in question.
C, represents the direction cosines associated with a basic vector being applied to the node.
It is understood that basic transIation vectors operate on the translational freedoms of
the node whereas basic rotation vectors modify the rotational freedoms.
rStart
If the point has no degrees of freedom so far, nP is set to unity and the direction cosines
are then those of the applied vector. Should the point already possess one degree of freedom
a test is performed to see if the new vector is in the same direction as the previous freedom
If it is, no action will be taken. If it is not the number of dgerees of freedom is increased to
two and the new direction cosines of the point are taken to be those of a perpendicular to
the plane in which the basic vector and the old freedom are located. Should the point
already possess two degrees offreedom, a test is performed to see if the vector lies within the
plane in which the two previous freedoms lie. If it does, no action will be taken. Tf it does
not, nP is set to the full value of three. From now on the direction cosines have no signifi
cance since all freedoms will be present, and the presence of further basic vectors of this
type will not affect the freedom pattern. Figure 11 describes the logic used in the selection
of the appropriate freedoms.
Application of the Finite Element Method to Ship Structures 121
freedom corresponding
corresponding to
smallest two of
The point may be reached where a freedom should be suppressed in a direction not
coinciding with any of the Cartesian freedoms. [4, 491 prove that the choice of the nearest
Cartesian freedom produces the correct stresses, the displacement requiring a slight modifi
cation. Should the program be capabIe of handling oblique freedoms the problem is
simplified and Fig. 1I can be adjusted accordingly.
In describing a grid a user or an automatic program may assign node numbers in the
order most convenient for the purpose of their generation. The resulting numbering
scheme produces, in general, a stiffness matrix with a nonoptimal band width. It is pos
sible to renumber the points in such a manner so as to reduce the band width to aminimum.
Figure 12 illustrates a suggested technique first described in [471 which can be used to
renumber the nodes using the element connection list prior to stiffness computations.
In the flow diagram reference is ma& to two dictionaries which represent fists relating one
node numbering system to another. The interstep dictionary relates the numbering of
nodal points between one step and the next. The accumulative dictionary relates final
num~ng of the system to the initial nurn~~ng scheme. Convergence of the scheme is
reached when no renumbering of the nodes occurs during an iteration.
122 HUSSEIN A. KAMEL and DONALD LIU
updote _
interstep
dictionary
accumulative
dictionary 1
 element
connections
which produ~s the deflections in terms of the forces, presents a formidable computat~ona1
task that may occupy hours of computer time. The matrix K may be decomposed into two
triangular matrices (the socalled Choleski method) or the expression K 'R may be evalu
ated directly using Gaussian elimination or one of its variations. The two methods are
equivalent for any one particular solution, although the decomposition, or triangularization,
method will have a distinct advantage in evaluating any new set of loading conditions,
assuming the decomposed triangular matrices are saved.
The matrix K will have been assembled using the structural model described by the
analyst. The program must contain as many safeguards as possible, since the description
of the model is prone to human error. One major source of such errors is the choice of
freedoms at the individual nodal points which should be greatly reduced if the procedure
Application of the Finite Element Method to Ship Structures 123
where 2 is the a~umulative operator leading eventually to the triangula~zation, and then
diagonalization of the product, 2X. Should a row, i, disappear at this point, the corres
ponding freedom must be dependent on the preceding displacements. It is therefore not
to be considered as an independent unknown, and may be eliminated from the equation.
It can be proven that the correct way to handle this situation would be to strike out the
corresponding row and column from the product TK, and the corresponding load entry from
R. DAISY prints out a warning stating the node number and the corresponding freedom.
This valuable feature may also be extended to warn of the loss of numerical accuracy
which may be of great use in pointing out areas of poor modeling.
AcknowledgementTheauthors are indepted to Miss JAN SCHMERBECK for typing the manuscript and to
Messrs. RICHARDB~ROATTIand THOMASMCGRATH for preparing the drawings.
REFERENCES
111M. J. TURNER,R. W. CU)UGH, H. C. MARTINand L. J. TOPP, Stiffness and deflection analysis of
complex structures. J. Aerosuace Sci. 23. 805823 (1956%
121J. H: ARGYRISand S. KEL~E+,Energy theorems ad structural analysis. Parts I and II, Aircraft Engi
neerinsc. 19541955.
1313. LA:GEFORS,Analysis of elastic structures by matrix ~nsfo~ation with special regard to semi
monocoque structures. J. Aero. Sci. 19,45X458 (1952).
[41 H. A. K.&EL, W. BIRCHLER, D. LIU, J.W. McKr&~~~and W. REID,An automated approach to ship
structure analysis. Trans. Sm. Naval Architects Marine Engrs 77, 233268 (1969).
Rules for Building and Classing Steel Vessels, The American Bureau of Shipping (1971).
:; 0. C. LZmwmmz and Y. K. c%RXJNO, The Finite Element Method in Structural and Continuum Mecha
its, 1st ed. McGrawHill, London (1967).
J. S. PRZ~~, Theory of Matrix Stru~ut~~ A~~~sis, 1st ed. M~rawHiiI, New York (1968).
t: T. H. II. PIAN, Derivation of element stiffness matrices by assumed stress distributions. AZAA Jl. 2,
13331336 (19641.
[91 J. H. ARGYRIS, Continua and discontinua, Pm. 1st Conf. Matrix Methods in Structural Mechanics.
WrightPatterson AFB, Ohio. AFFDL TR6680. (Nov. 1965).
1101B. I%AEUS DEVEUBEK~ and G. SANDER, An eq&ibrium model for plate bending. Int. J. SoIids Struc
tures 4,447468 (1968).
1111L. S. D. MORLEY,A trianguiar ~ui~ib~~ element with Iinearly varying bending moments for plate
bending problems. b. Royal Aero. Sot. 71,715 (1967).
[I21 T. H. H. PIAN, Element stiffnessmatrices for boundary compatibility and for prescribed boundary
stresses. Proc. 1st Co& Matrix Methods in Structural Mechanics, WrightPatterson AFB, Ohio,
AFFDL TR6680 (November 19651.
1131R. D. COOKand J. K. A~DULLA,Some plane quadrilateral hybrid finite elements. AIAA JI. 7,
21842185 (1969).
124 HUSSEINA. KAMELand DONALDLru
[14] F. K. BOGNER,R. L. Fox and L. A. SCHMIT,The generation of interelement, compatibie stiffness and
mass matrices by use of interpolation formulae. Proc. 1st Conf. Matrix Methods in Structural Mechanics.
WrightPatterson AFB, Ohio, AFFDL TR6680, (November 1965).
[1511. M. SMITH, A finite element analysis for moderately thick rectangular piates in bending. int. J. Me&.
Ski. 10,563570 (1968).
I161I. HOLAND, Stiffness matrices for plate bending elements, Finite E[ement Methods in Stress Analysis,
edited by HOLANDand BELL. TAPIR, Technical University of Norway, Trondheim, Norway (1969).
1171 R. H. GALLAGHER,Analysis of plate and shell structures, Proc. Symp. Application of Finite Element
Methods in Civil Engineering. Vanderbilt University (November 1969).
V81 R. CLOU~H and J. TOCHER,Finite element stiffness matrices for the analysis of plate bending. Proc.
1st Co& Matrix Methods in Structural Mechanics. WrightPatterson AFB, Ohio AFFDL TR6680
(November 1965).
WI R. CL~UGH and C. FEUPPA, A refined quadrilateral element for analysis of plate bending, Proc. 2&
Conf. Matrix Methods in Structurai Mechanics. WrightPatterson AFB, Ohio, AFFDLTR68150
(October 1968).
WI J. G. ERGATOUDIS, B. M. IRONS and 0. C. ZIENKIEWICZ,Curved, isoparametric, quadrilateral elements
for finite element analysis. Int. J. Solids Structures 7, 3142 (1968).
WI B. M. IRONS and 0. C. ZIENKIEWICZ,The isoparametric element systemA new concept in finite
element analysis. Conf. Recent Advances in Stress Analysis, J.B.C.S.A., Royal Aero Sot., London
(March 1968).
1221 0. C. ZIENKIEWICZ, B. M. IRONS, J. ERGATOUKXS, S. AHFAADand F. C. SCOTT, &oparametric and
associated element families for two and three dimensional analysis. Finite Eiement Mefhods in Stress
An&y&, edited by HOLANDand BELL. TAPIR, Technical University of Norway, Trondheim, Norway
(1969).
WI G. P. BAZELEY,Y. K. CHEUNG, B. M. IRONS and 0. C. ZIENKIEWICZ,Triangular elements in plate
bendingconforming and nonconforming solutions. Proc. 1st Co& Matrix Methods in Structural
Mechanics. W~ghtPatterson AFB, Ohio AFFDL TR6680 (Novem~r 1965).
1241 K. BELL, A refined triangular plate bending finite element. Int. J. ~iimerical ~ethoc~s in Engng 1,
(1969).
WI G. BUTLINand R. FORD, A compatible triangular plate bending finite element. University of Leicester,
Engineering Dept., Report 6815 (October 1968).
P61 K. BELL, Triangular plate bending elements. Ftnite Element Methods in Stress Analysis, edited by
HOLANDand BELL. TAPIR, Technical University of Norway, Trondheim, Norway (1969).
v71 B. IRONS,A conforming quartic triangular element for plate bending. Int. .T. Numerical Methods Engng
1, (1969).
WI J. H. ARGYRXS,I. FRIED and D. SCHARPF,The TUBA family of plate elements for the matrix displace
ment method. Aeronaut. J. 72, (1968).
WI W. BOSSHARD,Ein neues, vollvertragliches endliches Element fur Plattenbiegung. ht. Assoc. Bridge
Structural Engng Bulletin 28, 2740 (1968).
[30] H. HANSTEEN,Finite element displacement analysis of plate bending based on rectangular elements.
Int. Symp. Use of Electronic Digital Computers in Structural Engineering, Newcastle upon Tyne (1966).
f31J G. A. BUTLINand F. A, LECKIE,A study of finite elements applied to plate flexure. ,~un~eri~alMethods
for Vibration Problems Symposium at Southampton University (July 1966).
[32] J. H. ARGYRISand 1. FRIED, The PUBA family of plate elements for the matrix displacement method.
LSD Report No. 56 (August 1968).
[33] B. FRAEIJSDE VEUBECKE,A conforming finite element for plate bending. Inr. J. So/ids Structures 4,
95108 (1968).
[34] J. R. PAULLINGand H. G. PAYER, HullDeckhouse interaction by finiteelement calculations. Trans.
Sot. Naval Architects Marine Engrs 76, 281308 (1968).
1351 H. A. KAMEL, Analysis of large, closed ship structures. Seminar on Finite E~enlent Methods of Ship
Structure Analvsis at Webb Institute of Naval Architecture, New York, 2426 (June 1970).
[36] M. ST. DENISand W. J. PIERSON,On the motions of ships in confused seas, Trans. Sec. Naval Architects
Marine Engrs 61, 280357 (1953).
[37] B. V. KORVINKROUKOVSKY, Theory of Seakeeping. The Society of Naval Architects and Marine
Engineers (1961).
1381 R. B. ZUBALYand E. V. LEWIS, Ship bending moments in irregular seas predicted from model tests.
Webb Jnstitute o~Nuva~ Architecture Report (1963).
[39] D. HOFFMAN,Lecture notes on conformal mapping techniques in ship hydrodynamics, Webb Institute
of Naval Architecture Report No. 41lg (June 1969).
[40] E. V. LEWIS, A study of midship bending moments in irregular head seas. J. Ship Res. l(l), (1957).
[41] z95tj JACOBS,The analytical calculations of ship bending moments in regular waves. J. Ship Res. 2(l),
Application of the Finite Element Method to Ship Structures 125
1421E. V. Lewrs, Predicting longterm distributions of waveinduced bending moment on ship hulls,
paper presented at Spring Meeting of the Society of Naval Architects and Marine Engineers, Montreal,
Canada, 2326 (July 1967).
1431E. 0. U. BAND,Longterm trends of huh bending moments, American Bureau of Shipping (1966).
[44] Scientific advances in ship design. Surveyor (quarterly publication of the American Bureau of Shipping),
3, No. 3 (August 1969).
[45] W. J. BATDORP, S. S. KAPURand R. B. SAYER,The role of computer graphics in the structural design
process. Proc. 2nd Co& Matrix Methods in Strucutral Mechanics, WrightPatterson AFB, Ohio,
AFFDLTR68150 (Obtober 1968).
1461J. L. TOCHERand C. A. FELIPPA,Computer graphics applied to production structural analysis. Paper
presented at Symp. High Speed Computiw of Hastic Structures, IUTAM, Liege, Belgium, 2328
{August 1970).
f47] H. A. KAMELand H. K. E~~ENSTEIN, Automatic mesh generation in two and three dimensional inter
connected domains. Paper presented at Symp. Nigh Speed Computing of Elastic Structures, IUTAM,
Liege, Belgium, 2328 (August 1970).
(481 H. A. CAMELand C. MEARS,MEG, mesh generator. Unpublished Report, University of Ariozna
(May 1970).
[49] J. H. ARGYRIS,S. KELSEY and H. A. KAMEL,Matrix Methods of Structural Analysis, A Precis ojRecent
Developments, edited by DE VEUBEKE. Pergamon Press, Oxford (1963).
[SO]H. A. WL, A. MILLERand A. ELnArowrr, DAISY programmers and engineers manuals. University
of Arizona internal reports, August, 1970 (previous edition, 1968).
APPENDIX A
FINITE ELEMENT DISPLACEMENT METHOD
A brief outline is given of the theory. For further details regarding history and late
developments, consult reference list, particularly [l3, 6, 71.
NOMENCLATURE
A [Air], matrix with m rows and n columns whose typical (i, jj element is A,, j
hxn)
A' transpose of matrix A
A' inverse of square matrix A
T X Y
matrix defining a linear transformation between two sets of variables x and y,
such that
I3 X Y
matrix differential operator defining a differentiation operation to obtain one
set of variables x from another y, such that
If both pe and r were measured with respect to the global axis system, OXyr,then TP,rl
reduces to a sparsely populated matrix with one nonzero entry (equal to unity) in every
row. TP,,., is then an identification matrix expressing the logical connection between the
element and the complete structure. It is therefore called the element connection (identifca
tion or incidence) matrix.
= Tu,
UC? P4Ps
Application of the Finite Element Method to Ship Structures 127
where T,, Ps is a matrix which is a function of the shape and nature of the element and
the position of the internal point. Every column of T., pB gives the displacements of the
internal point due to a unit boundary displacement.
The matrix TutPB constitutes a formal definition of the properties of the element.
The choice of this matrix is critical and should be made with great care. It is essentially
a matrix by which the internal element displacements may be found by interpolation
from the boundary displacements. Therefore, it is called the element interpolation matrix.
Once T., Ps has been chosen, all other properties of the element automatically follow.
The strain function e, at a point e, is found from the displacement function, ugr by
differentiation. In a threedimensional continuum, for example,
(3)
and
(3a)
1
EX a/ax 0 0
_
&Y 0 alay 0
U
5 0 0 d/a2
V
=
alax
&,= (4)
Y XY way 0
W
q, = De,u,. (5)
This essentially means that each column in Te_ is obtained by differentiating the
corresponding column in T,, pIIaccording to the usual definition of strain.
The loads on the element are described in terms of the components of the corner
forces as a matrix P, which corresponds to the corner deflection matrix pg. Assuming linear
128 HUSSEIN A. KAMELand DONALD Lrti
elasticity, the work done by the forces, P,, over the complete range of displacement, Py,
is given by half the sum of the products of each force component times the cotresponding
displacement component: i.e.
The internally stored energy within the element g is a function of the stresses sy and the
strains e,; i.e.
Since the forces P, produce the displacements pq, and since linear elasticity is assumed,
a relationship must exist in the form
where k, is a square symmetric matrix called the stiffness matrix of the element.
The stiffness matrix, k,, is a function of the element properties defined by the element
interpolation matrix, T, ps. To derive k,, the expression for the workdone on the boundary
of the element is equated to that of the strain energy. See equations (8) and (9).
p;pg=
s I,
n
e,s,dV . (12)
Introducing the concept of the stiffness matrix, equation (1 I), as well as Hookes law,
equation (IO)
Substituting the relationship between e, and pe as given by equation (6), and observing
that pg is a constant matrix, the following is obtained:
p&p, = P;
s V,
T, ,gET,,~QWS
. (12bj
Assembly of structure
Having obtained the element stiffnesses, equation (13), and the element connection
scheme given by equation (I), the stiffness matrix K of the assembled structure may be
computed. K is defined as the proportionality matrix between loads R and displacements
r of the assembled structure, i.e.
R==Kr. (14)
The relationship between K, kgs and Tp,rs3 is found by equating the work done
on the assembled structure to the internally stored strain energy for the individual elements
or
(15)
or
Having obtained the structure displacements, r, the stresses may be found in each element,
g, using
sg = EC,
=ET,, PnPE