Abstract
Traditional methods to assess ship loadings are mostly based on static or quasi-static methods. When
dynamic fluid loads are incorporated they are usually restricted to those originated in head seas,
therefore only symmetric loads, such as vertical bending, are allowed into the analysis. It is a fact
that in conventional monohulls major stresses arise from symmetric loads, however when a ship
operates in oblique seas the entire hull experiences combined symmetric and antisymmetric loads such
as vertical bending (symmetric) and horizontal bending and torsion (antisymmetric). From the safety
point of view, when a stress analysis of certain areas of the hull is carried out, in addition to
symmetric loads it is desirable to include antisymmetric loads too. For instance, the direct
longitudinal stress at the hull-deck shell junction will have contributions from both, vertical and
horizontal bending moments.
The hydroelastic analysis of marine structures allows any type of loadings to be included in a stress
investigation. The theory is traditionally separated into two parts a dry analysis where a Finite
Element Modal Analysis is carried out to determine the dynamic in vacuo characteristics of a
structure. In this stage, a study of modal stresses can reveal potentially dangerous areas due to high
stresses being concentrated at hull discontinuities. Having established the dynamic characteristics
(natural frequencies, modal loads and shapes) of the dry hull a wet analysis is conducted to
introduce all fluid effects such as wave loadings and hydrodynamic damping and added mass. This
paper shows the modal stress analysis of a typical Chilean fishing vessel. Particular attention is given
to highly stressed areas. As a general rule, antisymmetric loads have been traditionally neglected in a
dynamic structural analysis, however, in this paper it is shown that they could contribute to increase
stresses at specific places of the hull.
1.- Introduction
The structural reliability of marine structures is paramount for ship safety at sea, yet the complexity of
these structures and the relative difficulty in determining realistic sea loads prevents, generally, first
principle calculations to be carried out. The ever growing capacity of computers and the evolution of
modelling techniques is allowing an increased use of the Finite Element Method in the analysis of
marine structures. Still the main difficulty consists on applying realistic loads to the Finite Element
Model.
Hydroelasticity theory is a modern approach to the rational analysis of marine structures. The theory
requires the calculation of modal properties (natural frequencies, modal shapes, stresses, etc) of a
vessel, to do so a Finite Element Modal Analysis has proven a suitable method. In the solution process
loads are incorporated in a consisted way onto the finite element model. Two different models were
created to study the hydroelastic behaviour of a typical Chilean Fishing Vessel, a beam three-
dimensional model and a plate three-dimensional model. Modal results obtained from both models
were analysed and compared. When generating the models, similar distributions of properties were
aimed at, as far as this was attainable. Advantages and disadvantages of these models and different
approaches employed are highlighted trough the paper.
A fundamental aspect of a modal analysis of ship structures is whether or not to include the fluid
effects. Inclusion of them (added mass for instance) complicates the mathematical model. On the other
hand, the assumption of all the hydrodynamic effects as being external applied loads, simplifies the
modal analysis and allows the use of well-known methods to determine the dry dynamic
characteristics. The disadvantage of this method is that the properties obtained correspond to a
structure vibrating in vacuo, i.e. they must be corrected to obtain those of a floating structure.
Despite this apparent limitation, the advantages of ignoring the fluid effects in the modal analysis were
early identified (Bishop and Price 1974, 1976) and the method has been employed successfully in the
modal analysis of all kind of floating structures ranging from sailing yachts (Louarn and Temarel
1999) to large barges (Price et al. 1996). The ship vibrating in vacuo is the approach adopted in this
paper, in order to perform a modal analysis of an undamped free vibrating fishing vessel. This process
is usually referred as to a dry modal analysis.
Both symmetric and antisymmetric loads affect the hydroelastic behaviour of a ship travelling in
oblique waves, therefore it will be interesting to evaluate the effect of antisymmetric loads on the
stress level of certain areas of a hull. Results from the hydroelastic analysis of a different mohohull are
shown to illustrate the relative effects of antisymmetric loadings in the total stress magnitudes.
The equation of motion for a freely floating flexible structure can be written as (Bishop, Price and Wu
(1986))
ap&&(t ) + bp& (t ) + cp(t ) = Z(t ) . (1)
In this equation a, b and c are the NN generalised mass, structural damping and stiffness matrices of
the dry or in vacuo structure, with elements arr , brr = 2 r r arr and crr = r2 arr for r = n,...,N
respectively. n denotes the index of the first flexible mode. r and r denote the natural frequency of
the structure in vacuo and the structural damping factor, respectively. The part of matrix a
corresponding to the rigid body modes contains the mass and moments and products of inertia of the
structure. N is the number of degrees of freedom, associated with rigid body motions and distortions,
allowed for in the analysis. p(t) is the N1 principal coordinate vector and Z(t) represents the N1
external force vector describing, in this case, the fluid actions. The equilibrium axis system xyz is
placed at the calm water level with x, y and z denoting the longitudinal (+ve to bow), athwartships
(+ve to port) and vertical (+ve upwards) axes respectively.
The dry or in vacuo analysis is carried out (in the absence of external forces and structural damping) to
obtain the natural frequencies r , r = 1,...,N and corresponding principal mode shapes as well as other
modal characteristics such as modal bending moments, modal stresses, etc. associated with the
distortions of the structure. The dry or in vacuo analysis for three-dimensional hydroelasticity,
involves discretisation of the structure using shell and, if required, beam finite elements. It should be
noted that other types of finite elements (e.g. mass elements, contact elements, link elements) may also
be used in the dry structure idealisation, as shown by Louarn and Temarel (1999), depending on the
type of structure and structural detail under consideration.
For a freely vibrating structure, i.e. with no restrains in space, the first 6 principal modes obtained
from a modal analysis represent the rigid body motions associated with zero frequencies (r = 0 ; r =
1,2,.....,6). These rigid modes are the three components of the displacement of the centre of mass,
namely: surge, heave and sway. The remaining three rigid body motions are rotations: yaw, roll and
pitch.
Having obtained the distortion mode shapes of the dry hull, all actions can be applied as external
forces, including hydrodynamic actions and structural damping, in order to perform the wet hull
analysis. For floating structures of complex shape, a three-dimensional hydrodynamic model, with
suitable Greens functions, is recommended.
Assuming the flow is ideal, thus represented by a velocity potential, the external fluid force in regular
waves contains contributions associated with the incident, diffracted and radiated wave due to the
motions and distortions of the vessel, namely
N
Z r (t ) = exp(i e t ) or + dr + ( e2 Ark i e Brk Crk ) pk . (2)
k =1
Substituting eq.(2) into eq.(1) the generalised linear equations of motion for a vessel travelling in
regular oblique waves is obtained, in matrix form, as
(a + A)p
&&(t ) + (b + B)p& (t ) + (c + C)p(t ) = exp(iet ) (3)
and are solved to obtain the principal coordinates in the form p (t)= p exp (iet), where p denotes the
complex amplitude of the principal coordinate vector. In these equations the NN matrices A, B and
C represent the generalised added mass, hydrodynamic damping and restoring coefficients, with
elements Ark, Brk and Crk, respectively. The generalised excitation vector contains the incident (or)
and diffracted (dr) wave contributions. e = - k U cos is the wave encounter frequency, in deep
water, for waves of frequency , wave number k = 2/g, encountered at heading ( = 180o, head
waves) when the vessel is proceeding with a forward speed U.
The ocean surface can be represented by the linear addition of several regular waves of random height,
length and if required, direction. The temporal variation of the seaway is then represented by
R R
( x , y , t ) = j ( x, y , t ) = a j e
i ( j t + j )
(4)
j =1 j =1
where aj is the wave amplitude of the jth wave; j and j represent the wave frequency and phase
angle respectively, the wave frequencies and amplitudes are chosen according to a given sea spectrum.
R represents a sufficiently large number of regular waves, Lloyd (1998) suggests that 50 waves are
able to reproduce the ocean surface with sufficient accuracy.
2.2.- Steady State Loads in a Seaway
Having determined principal coordinates by solving equation (3) it is possible to find any relevant
response such as bending moments, shearing forces, twisting moments or stresses using the
appropriate characteristic function of the dry structure. According to a theorem due to Rayleigh
(1894), any distortion of the structure may be expressed as an aggregate of distortions in its principal
modes. The same principle applies to any modal characteristic function. Finally the dynamic response
to a seaway, since the model is linear, is simply the addition of the responses to each individual
component wave of the seaway. For example any steady state load in an irregular sea is defined by
R m
S ( x, y, z, t ) = a j p r (t ) S r ( x, y, z ) (5)
j =1 r = 7
where Sr denotes any modal steady state load such as bending or torsion moment, direct stress, shear
stress or any known modal load. pr(t) represents the principal coordinate of the rth mode and aj
represents the amplitude of the jth component wave of the seaway. The corresponding modal
properties of the dry structure Sr may be obtained from suitable modal nodal stresses. It is interesting
to note that no contribution to these loadings arise from the rigid body modes (r = 1,...,6). This is not
the case with displacements, where all the modes, rigid and flexible, contribute to the total
displacements.
In order to investigate the application of the FEM modal analysis applied to a ship structure, modal
analyses of a ship vibrating in vacuo were carried out using two different finite element idealisations, a
plate finite element model and a beam model. The test case hull corresponds to a typical Chilean steel
fishing vessel shown in figure 1.
The Plate finite element model was creating using SHELL63 ANSYS elements to represent the hull,
deck and bulkheads. This plate element has bending and membrane capabilities. The hull was divided
into 50 sections of different lengths, shorter sections were defined at the bow and the stern in order to
better represent the hull forms. This allowed the assignation of different thicknesses and material
densities in order to control the modelling of adequate hull cross sectional area, second moments of
areas and mass distribution. Longitudinal and transverse bulkheads were modelled at their actual
positions. Due care was given to the correct assignment of element thicknesses, densities and material
mechanical properties. The resultant Plate model has 3781 elements and 1660 nodes, as shown in
figure 2.
Figure 2: Three-dimensional Plate Finite Element Idealisation. The model was created using
1660 nodes and 3781 SHELL63 elements
Since no lumped masses were modelled, all masses must be adequately represented by the plate
elements. It is important to define a proper mass distribution, otherwise it is very difficult to achieve
agreement in the results obtained from the dry analyses of the different models.
The required data is easily available and corresponds to the mass per section, material properties,
vertical and horizontal second moments of area and the effective shear areas. This Beam element
assumes that the section is symmetric in both transverse directions, therefore the neutral axis is placed
in the middle of the section in both directions, this is correct in one direction (port starboard
symmetry) but is not the case in the vertical direction (no deck bottom symmetry). As a result
horizontal bending modes are not coupled to torsion modes and vice versa.
General information about thickness of steel plates of the hull, decks and bulkheads of the fishing
vessel was available, using this information the Plate Finite Element Model was defined. Beamlike
properties for this model were calculated, i.e. cross sectional area and second moments of area.
Torsional inertia for the beam model was assumed as G(Iyy+Izz), where G denotes the Shear Modulus.
Iyy and Izz are the vertical and horizontal second moments of area respectively.
The mass distribution of the vessel including all non-structural weights and lumped masses (such as
engines) is shown in figure 3. Distances from the bottom to the vertical neutral axis are shown in
figure 4. Cross sectional areas are given in figure 5 and second moments of area in the vertical (Iyy)
and horizontal (Izz) directions are shown in figure 6.
40 8
30 6
ton/m
20 4
m
10 2
0 0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x/L x/L
Figure 4: Vertical distance from the keel to the
Figure 3: Mass distribution cross section centroid
1.0
0.6 12
m2
m4
0.4 8
0.2 4
0.0 0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x/L
x/L
Figure 5: Cross sectional area Figure 6: Second moment of area
As described in section 2, to perform a free vibration modal analysis, damping and forcing terms are
ignored in the equation of motion (1). Through the discrete idealisation of the hull structure described
for the different models in previous section, a Finite Element modal analysis can be carried out in
order to obtain all the dynamic modal properties of each model. The eingen value problem was solved
using the Block Lanczos method available as a standard solution method of ANSYS.
The Plate Finite Element model presented some difficulties and modifications were needed to ensure
the generation of beamlike global distortions. Initially the modal analysis of the FE Plate model was
unable to produce mode shapes comparable to those expected in a beam-like hull, and only local
vibrations were obtained at low frequencies, as those shown in figure 7.
It was observed that local vibrations were developed at any nodal position without support normal to
the hull shell, in other words, at any of the stations along the hull without bulkheads. In a real hull
these local vibrations are not observed because structural stiffeners, frames for example, prevent such
local distortions. Therefore the modal shapes obtained with the Plate model were unrealistic and
impossible to compare with those obtained from a Beam model.
Since it was necessary to constrain nodes transversely, new elements, shown in figure 8, were
incorporated to the Plate model. These elements acted as artificial bulkheads. These were created at
every unsupported station. As a result, a bulkhead at each end delimited all the 50 sections in which
the hull is divided. In order not to modify the modal properties of the hull, i.e. the mode shapes and
frequencies, the added or virtual elements had their properties of density and thickness set to very
low magnitudes, hence they do not affect the global mode shapes and frequencies, in contrast the
actual bulkheads were modelled using real thicknesses and material properties.
Fig. 7: Local vibration mode. Freq. = 1.045 Hz Fig. 8: Virtual elements of arbitrarily low density.
Mass densities of the hull plates had to be manipulated to take into account lumped masses, e.g.
engines. The final total mass and mass distribution so defined result very close to those of the actual
ship. The added elements simple enabled the Plate model to adequately represent beam-like distortion
modes and reasonable mode shapes were obtained for a wide range of frequencies. Several types of
flexible modes were identified in the modal analyses: vertical (symmetric) bending, horizontal
(antisymmetric) bending, twisting (antisymmetric) and longitudinal (symmetric) modes. These are
easily identifiable in figure 9, showing modal shapes of the Plate model.
a) 2-node vertical bending mode. Freq. 5.47 Hz b) 2-node horizontal bending mode. Freq. 7.39 Hz
c) 1-node torsion mode. Freq. 10.53 Hz d) 3-node vertical bending mode. Freq. 12.35 Hz
e) 3-node horizontal mode. Freq. 15.35 Hz f) 2-node torsion mode. Freq. 18.40 Hz
The Beam model modal analysis was carried out using BEAM4, a three-dimensional beam finite
element available in the ANSYS elements library. BEAM4 allows for shear deformation and rotatory
inertia effects to be accounted for. It is important to accurately estimate the shear effective transverse
area of each beam element of the model. The shear deformation effect is dependent on the effective
shear area, a method to calculate this is illustrated by Chalmers and Price (1979). The effective shear
area can be laborious to calculate and a common practice is to assume the effective shear area is a
fraction of the total (longitudinally) effective cross section area. After validation through comparison
with results obtained for the Plate model Finite Modal Analysis, 25% of the transverse area of the
monohull was considered to be shear effective in the vertical direction, and 45% in the horizontal
direction.
The rotatory inertia is not easily available for non-uniform beams, involving calculations of mass
distribution about the neutral axis for every cross section of the hull. Since this effect is relatively
small for low frequency modes, no rotatory inertia correction was included. The model itself presented
no difficulties and the modal solutions were obtained with considerably less computing time expense.
3.4.- Natural frequencies
Table 1 shows natural frequencies obtained from the finite element modal analyses for the 13 initial
flexible vibrating modes.
It must be noted that in the Plate model, antisymmetric bending modes are always coupled with
torsion. The same can be said about the torsion modes, where lateral displacements are also present.
The latter case is less evident to the eye when observing the mode shapes. Thus the mode shape
descriptions for the antisymmetric mode shapes of the Plate model in table 1 denote the dominant
nature of the distortion. The coupling is produced because the hull is vertically asymmetric. On the
other hand, the Beam model uses a symmetric element (ANSYS BEAM4), therefore, all modes are
uncoupled, in other words, bending and torsion modes are pure. In the Plate model the amount of
coupling between horizontal bending and torsion modes is very small. For example, the torsion related
rotation angles x present in a horizontal bending dominant mode are small when compared with the
rotation angles x of a torsion dominant mode.
As can be observed in table 1, there is good agreement in natural frequencies, particularly for low
natural frequencies. Little differences are observed in the antisymmetric modes, indicating that the
coupling between horizontal bending modes and torsion modes, present in the Plate model, does not
introduce significant discrepancies when compared to the vibration modes of the Beam model, where
this modes are uncoupled.
Modal shapes of vertical bending, horizontal bending and torsion modes for both models are compared
in figures 10, 11 and 12 respectively. It is observed that good agreement exist for all kind of modes.
Regarding the Plate Model, the mode shapes correspond to the vertical distortions of the bottom nodes
at the relevant position along the hull. In the Plate model, as horizontal bending is coupled with
torsion, the lateral displacement, at a selected station along the hull is different depending on the
position of the selected node at this station. By definition lateral displacements should be taken at the
shear centre. Being this position variable for every station and lacking the definition of suitable nodes
coinciding with these positions, other row of nodes must be selected to measure the lateral
displacements. The row of nodes at the main deck-longitudinal bulkhead junction was chosen to
represent the horizontal displacements. The hull distortion so described agreed well with the
corresponding mode obtained from the beam models, as shown in figure 11. Regarding the torsion
modes, the angle of rotation for each station is very similar for beam and plate models, as can be
observed in figure 12 indicating there is little distortion associated to horizontal bending in the Plate
model.
1.0
0.5
0.0
-0.5 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x/L
-1.0
-1.5 2-node
-2.0
5-node
1.0
0.5
0.0
-0.50.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 x/L
1.0
-1.0
-1.5
2-node
-2.0
0.0
-0.5 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x/L
-1.0
-1.5
2-node
-2.0
a) Longitudinal direct stress for 2-node b) Longitudinal direct stress for 3-node
vertical bending mode. Freq. = 5.47 Hz. vertical bending mode. Freq. = 12.35 Hz
c) Longitudinal direct stress for 2-node d) Longitudinal direct stress for 3-node
horizontal bending mode. Freq. = 7.39 Hz. horizontal bending mode. Freq. = 15.35 Hz
e) Vertical shear stress for 1-node f) Vertical shear stress for 2-node
torsion mode. Freq. = 10.53 Hz. torsion mode. Freq. = 18.40 Hz
Figure 13: Modal stresses. Dark areas represent highly stressed regions of the hull
Using nodal stresses and the distance from the nodal position to the neutral axis, it is possible to
calculate bending moments for the plate model. The comparison of modal loadings so obtained is
presented in figure 14 where vertical bending is shown and figure 15 where horizontal bending
moment is presented. Using shear stresses it is possible also to calculate torsion modes, these are not
shown in the present paper.
0.005
MYr/L
-0.005 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
-0.015 x/L
2-node
-0.025
Figure 14: Vertical Bending Moments
_____ beam model
_ _ _ _ plate model
0.04
0.03 3-node 5-node
0.02
0.01
MZr/L
0.00
-0.01 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
-0.02 x/L
-0.03 2-node
-0.04
Figure 15: Horizontal Bending Moments
It is clearly shown in figure 14 that there are notorious differences in the vertical bending moment at
positions x/L =0.41 and x/L = 0.84. These positions correspond to the superstructure and forecastle
bulkheads respectively. It is observed in figure 15 that there are differences in the horizontal bending
moment as well, however, these are not remarkable as in the vertical case, this is expected since the
discontinuities are in the vertical direction. It is evident that the beam model cannot adequately
represent these discontinuities and therefore this model should be used as a mean of providing a
preliminary, non-detailed, stress distribution. Regarding the plate model, as we are interested in stress
magnitudes, the determination of loads in form of bending moments does not present any advantage
and therefore the dynamic analysis should be carried out using modal stresses directly.
An advantage of a three-dimensional finite element model is that it allows special attention to be given
to areas of the hull where high stresses are known or suspected to occur, this is achieved by studying
modal stresses at specific nodes, typically those placed at hull discontinuities. The fishing vessel
modelled has two distinctive features at the main deck, these corresponds to the bulkheads positions
defining the superstructure and forecastle decks. Direct modal stresses in the longitudinal direction for
the 3-node vertical mode are shown in figure 16 and details of longitudinal bulkhead stresses for the 2-
node horizontal bending mode are presented in figure17.
Figure 16: Direct longitudinal stress for the 3-node vertical bending mode.
Dark areas represent highly stressed regions of the hull.
Figure 17: Direct longitudinal stress for the 2-node horizontal bending mode. Detail
showing the longitudinal bulkhead. Dark areas represent highly stressed regions.
Usually only symmetric modes are considered relevant in a flexible analysis, however antisymmetric
modes can also contribute to high stresses at the very same position. Figure 18 shows modal stresses
for the 2-node horizontal mode. It is clearly seen that high stresses occur at the hull position where the
deck joins the superstructure bulkhead, the very same position where high stresses for the 3-node
vertical bending mode is found. Figure 19 shows vertical shear force, again stresses are high at the
superstructure bulkhead position.
Figure 18: Direct longitudinal stress distribution for the 2-node horizontal bending mode.
Figure 19: Vertical shear stress distribution for the 1-node torsion mode.
From the analysis of highly stressed regions of the hull it is evident that all kind of flexible modes can
contribute to the stresses at a particular position of the hull. The actual level of stress attributed to each
mode will depend on the magnitude of the respective principal coordinate.
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
x/L
Figure 20: Effect of antisymmetric loads on the stress at the hull-deck junction
5.- Conclusions
Finite Element Modal Analysis is a valuable tool in performing the dry analysis of a marine structure.
Plate models are definitively necessary for the investigation of stress distributions, this is a particularly
important advantage when a study of highly stresses areas, at hull discontinuities for example, is
required. A Beam model may be created in a fraction of the time required for a Plate model. Also the
computing time employed to perform the modal analysis is significantly lower. This suggests that a
Beam model should be preferred when non-detailed calculations are required.
The type of hull analysed, i.e. a closed hull with no large openings such as hatches, presented little
coupling between antisymmetric modes. This enabled the use of the Beam model, based on BEAM4
element of ANSYS. This model produced uncoupled antisymmetric modes, which compared very well
with results obtained from the Plate model where these modes were coupled. For other types of hulls,
especially those with large deck openings, such as container carriers, a high degree of coupling
between antisymmetric modes may be developed. In this, case it can be necessary to use alternative
beam finite elements.
From the analysis of highly stressed regions of the hull it is evident that all kind of flexible modes can
contribute to the stresses at a particular position of the hull. A structural dynamic analysis of marine
structures should allow for all kind of modes to be duly considered such as vertical bending,
elongation-shrinking (symmetric), horizontal bending and torsion (antisymmetric).
Acknowledgements
The authors are grateful to DID, University Austral of Chile for supporting research project S200067,
from which partial result are presented.
References
BISHOP, R.E.D. and PRICE, W.G. (1974) On Modal Analysis of Ship Strength. Proc. Royal
Society. London, A341, pp. 121-134.
BISHOP, R.E.D. and PRICE, W.G. (1976) On the Relationship Between Dry Modes and Wet
Modes in the Theory of Ship Response. Journal of Sound and Vibration (1976) , vol. 45(2).
CHALMERS, D.W. and PRICE, W.G. (1979) On the Effective Shear Area of Ship Sections.
Transactions of the Royal Institution of Naval Architects, 1979.
LOUARN, F.H. and TEMAREL, P. (1999) An Investigation of the Structural Dynamics of a Racing
Yacht. Proc. of the 14th Chesapeake Sailing Yacht Symposium, pp. 123-142.
PRICE, W.G., SALAS, M. and TEMAREL, P. (1996) The Hydroelastic Behaviour of Barge Type
Structures in Waves. Paper 39, International Workshop on Very Large Floating Structures (VLFS 96).
Hayama, Japan, Nov. 1996.
PRICE, W.G., SALAS, M. and TEMAREL, P. (2001) The dynamic behaviour of a mono-hull in
oblique waves using two- and three-dimensional fluid-structure interaction models. Accepted by
TRINA.
RAYLEIGH, LORD (1894) The Theory of Sound, ed. 2, art. 92. London Macmillan.
ZIENKIEWICZ, O.C. (1977) The Finite Element Method McGraw-Hill, New York