INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING. VOL.

22,575-595 (1986)

HYDROSTATIC LOADING A N D STABILITY ANALYSIS

OF STRUCTURES WITH LARGE DISPLACEMENTS

P. G. BERGAN AND K. M. MATHISEN

Division of Structural Mechanics, The Norwegian Institute OJ Technology, Trondheim. Norwciy

SUMMARY
A method for the computation of hydrostatic forces on cable, truss, beam, plate and shell type structures
is presented. A general large displacement formulation is assumed which allows for displacements and
rotations of unlimited size. Equilibrating forces for nodal points as well as incremental force-displacement
relationships are given. Various ways of making the geometric load matrix symmetric for incremental
analysis is discussed. Alternative strategies for the incremental-iterative solution is also treated. A method
for the determination of the hydrostatic stability of floating structures is suggested. Metacentres and centres
of buoyancy for rotations about principal axes in the water plane are automatically computed. The paper
presents three examples of applications: a pipe which is partially submerged, a full overturning of a floating
space frame and, finally, a hydrostatic stability analysis of a ship.

INTRODUCTION
The theory of fluid pressures on structures and stability of floating structures is very well
understood today, at least for the static case. Analytical methods and computer programs for
hydrostatic analysis of a wide range of problems are readily available. However, in most cases
the hydrostatic analyses are based on the assumption that the structure is completely rigid, and
these analyses are completely separated from the structural analyses. Typically, as in the case
of the uprighting of the Alexander L. Kielland platform,' the hydrostatic computer program
generates a data file of hydrostatic forces which is subsequently read and utilized in a separate
analysis by a structural analysis program. Similar procedures are commonly used in, for instance,
studies of the launching of platforms. However, in these type of problems, there is much to be
gained, both in terms of accuracy and computational convenience, by integrating the
computations of pressure forces fully into the structural analysis. Such integrated analyses are
also necessary for highly flexible structures in order to account for the fact that the pressure
interacts with the structures and always works in the perpendirection to the deformed surface.
It is thus important to account for the truly deformed configuration in cases like mooring, pipe
laying, flexible riser, implosion problems, etc.
Most large-scale structures are today being analysed by means of the finite element method.2
A fully integrated fluid-structure analysis requires that the computation of hydrostatic forces is
formulated in such a way that it fits directly into the nonlinear finite element analysis. This
implies that nodal reaction forces and incremental force-displacement matrices must be
established. The present paper discusses how this may be done in an efficient and accurate
manner. Further, a method for fully integrated static or dynamic analysis of the hydrostatic
stability of floating, flexible structures is suggested.

0029-598 I /86/060575-2 I $10.50 Received July 1985

0 1986 by John Wiley & Sons, Ltd.
576 P. G. BERGAN AND K. M. MATHISEN

HYDROSTATIC FORCES O N BAR AND BEAM ELEMENTS

Preliminaries
The present formulation assumes that the hydrostatic forces are described in a global,
Cartesian reference system I , , I,, I,, where I, is opposite to the direction of gravity (see Figure 1).
The water surface intersects the I, axis at distance h, from the origin. Let P be an arbitrary
point with position vector X on the surface of a submerged body. The outward unit normal at
this point is n and the submerged depth is h. The pressure vector (traction) t at P is then
t = -pn= -pwgkn (1)
where p is the water pressure, p w the mass density of the water, and y the acceleration of gravity.
t is a null vector for negative h. Note also that t is a Cauchy type stress measure involving the
current, deformed areas and directions.
The resulting hydrostatic force on a partially or totally submerged body is obtained by
integrating the pressure vector over the wetted surface A

F, = tdA

F, is also denoted the buoyancy force, and it is directly related to the volume of the displaced

water surface

~
~ -

Figure 1. Reference system for hydrostatic pressure

Figure 2. Position of fully submerged element

 HYDROSTATIC LOADING AND STABILITY ANALYSIS 577

water Vw through Archimedes’ law

F, = P W S VWI, (3)
In the case of a floating body the buoyancy force may be considered as acting through the
centroid of the volume of displaced water.

Nodul ,forces
The computation of nodal forces on a straight bar or beam element with constant cross-section
will now be considered. The location of the beam is defined by the position vectors X , and X,
for the centroids of the end sections. The length direction is defined by a unit vector i, through
these end points (see Figure 2). The element forms an angle CI with the I , -I2 plane and the total
length L is projected onto this plane with length z.
The unit vector along this projection is
denoted 7. The element is considered as being capped at both ends and the outer area of the
cross-section is A , .
The force resultants at the capped ends are considered first. Denoting the water pressure at
the centroidal end points p , and p , , these axial resultants are
Fa1= P I A w i l (4)
Fa2= -P2Awi1
In addition to this, distributed pressure acts along the element in directions perpendicular to
the length axis i,.
The hydrostatic force resultant per unit length of the element is

f, =
i tds = pwgAwcosai,

where s is the external boundary line co-ordinate for the cross-section and i, is the unit vector
which is normal to the i, axis and which lies in the il-13 plane. It is easily shown that the total,
resulting transverse force for an element which is totally submerged is

(
F, = p w g A w L -S (
. I CI~ x2 - Xl I , + = )
L
I, +c0s2aI3
)
This reference force is distributed to the ends using distribution factors k, and k,. For a fully
submerged element k, = k2 = 0.5. Note that a lumped force concept without fixed end moments
is used for beam as well as bar elements. The main reason for this is to avoid superposition of
particular solutions in the non-linear stress calculation for the beam element.
By combining the axial end forces (4) with the distributed transverse force ( 6 ) the total end
forces may be written

+ Aw(plsincr+ k,pwgLcos2cc)I,
578 P. G. BERGAN AND K. M. MATHISEN

=A, x2 -x,
, (-p2-k2pwgLsina)11

+ A,( - p2 sina + k,p,glcos2 a)l,

These expressions are valid for totally submerged elements as well as partially submerged
elements. There are four different cases that must be considered:
Fully submerged element; h, 2 0 and h, 2 0

L, = ~ h1 L
h1- h2

Only end 2 submerged; h, < 0 and h, 3 0

L,=- h2 L
h, - hl
Entire element above water; h , < 0 and h, < 0
pl=O, p2=0, k,=0, k,=O (84
L, is the submerged length of the element (see Figure 3a). It can be checked that the sum
of F, and F, statisfies the buoyancy law (3).

Certain refinements of the present theory are possible. First of all, the end reaction forces d o
in fact act slightly offset to the centroidal points unless the element stands vertically. This force
eccentricity, which can be calculated accurately, is normally of minor importance, and hence it
is neglected in the present study. It should also be noted that the end forces cancel each other
for joining elements in a straight, continuous member. The end forces are only important when
there are variations in the cross-sectional area A , and for real end sections of members.
The end force calculation may also be refined for elements that are cutting the water surface
at a low angle (see Figure 3b). In the case shown in the figure, the transverse force per unit
length in ( 5 ) should be calculated on the basis of the part of the cross-section A , that is actually
submerged. Modifications of the end forces for these effects have been carried out for circular
members, and these expressions may be found in Reference 3.
In the present large displacement formulation, the end co-ordinates are continuously updated
in the force computations. However, change of A , as a function of external pressure on hollow
sections or due to Poisson's ratio effects are not accounted for.
 HYDROSTATIC LOADING AND STABILITY ANALYSIS 579

a)

-A

Figure 3. Partially submerged elements: (a) a large, (b) a small

Incremental force-displacement relationship

Nonlinear analyses normally utilize the incremental relationship between forces and dis-
placements. In this connection it is important to account for variation of the external hydrostatic
forces during change in position of the body on which they act. The incremental variation of
the end forces on an element (7a, b) is obtained by differentiation:

where ox,, u y , , vz,, ux2, u y 2 , uz2 are the nodal displacements at the two ends. Differentiation
of F, and F, with respect to the nodal co-ordinates gives highly complicated expressions. This
complexity is mainly due to the influence of displacement increments on the element length L
when each displacement component is differentiated separately. In practice, however, axial
deformations have only little influence on the buoyancy forces. Most of all, these effects are
insignificant in the incremental formulation. The primary effect on the force variation is due to
the rigid body motion and, in particular, the pure translational motion in the 2-direction. This
means that L and c( may be assumed to remain constant, and that partial derivatives with respect
to these quantities may be set equal to zero. It is also seen that rigid-body motion in the X and
Y directions do not change the hydrostatic forces; hence, the partial derivatives of F, and F,
580 P. G. BERGAN A N D K. M. MATHISEN

with respect to these quantities may also be set equal to zero. This gives

az, az,

+Aw["';"'( dP2 ak,

-az,-az,

Four different cases are considered for the partial derivatives of end pressures and distribution
factors:
Fully submerged element: h, 2 0 and h, 2 0
dP1 - dP2 -
--pwg, akl -
--
- dk2 - dk2 - 0
az, az, az, az, az, az,
Only end 1 submerged: h, 2 0 and h, < 0

Only end 2 submerged: h , < 0 and h, 2 0

akl - ak, - hlh, ak,

-- h:
--
a ~ ,az, (h,-h,),' az, (h,-h,),
Entire element in air: h, < 0 and h, < 0

aP1 - 3 P 2 - 0, -a k-, - a=k 2

, = a2k = 0a k
az, az, az, az, az, az,
 HYDROSTATIC LOADING A N D STABILITY ANALYSIS 58 1

Equation (10) may also be written on matrix form

(124
with the components

The non-zero terms in [ k b ] are easily retrieved from (10) and (1 1). [ k b ] is a ‘geometric stiffness
matrix’ associated with the loads. It is important to note that it has a non-symmetric form. This
has to do with the fact that the principle of reciprocity does not apply to these incremental loads.
Typically, the vertical resulting forces do not change when the element is moved horizontally,
while the horizontal forces do change when the element is moved vertically.

Symmetric forms
The buoyancy stiffness is added to the material and geometric stiffness for the structural
system. [kb], as given in (12), thus destroys the symmetry of the system stiffness matrix, a fact
that in practice is unacceptable. This problem may be dealt with by replacing [ k b ] by its symmetric
component

Another alternative is to use the terms

0 0 0
0 0 0
k33 k36
0 0 0 0 0
0 0 0 0 0
k63 k66.

This matrix is symmetric because k,, is equal to k,,, see (10) and (1 1).
The simplest form that can be obtained is to assume that Auz, is equal to AuZr, as in a pure
vertical translational motion and, thus, k,, can be added to k , , and k,, to k66. This gives a
third alternative [kb,] with only two diagonal stiffness terms:
kb3(3?3) = k33 + k36
= - - ~ , az,
( ~ s i n o l az,
+ ( ~ +az,
~)pWg~cos~Lr)

This last form corresponds to the well-known principle of using vertical ‘buoyancy springs’.
However, the present expression is somewhat more comprehensive than the usual form because
582 P. G. BERGAN A N D K. M. MATHISEN

it adjusts the vertical component of the end forces (and axial stressing) also for fully submerged
members. The most important effect, however, is the adjustable springs for the elements that
cut the water surface. Note that (15) accounts for redistribution effects between the end nodes
for inclined members that move through the surface.
The accuracy of the three forms (13), (14) and (15) depends on the type of problem considered.
The most extensive form (13) was tried out, but it was soon found that the simpler forms (14)
and (1 5) generally resulted in just as good convergence.
In some numerical comparisons between [kbJ and [kb,] it was found that [ k b , ] in fact gave
the least total computer time (faster convergence). Thus, [kb,] of equation (15) may be
recommended for most practical applications. It should be stressed that the form of the geometric
load matrix, be it symmetric or non-symmetric, only has consequences for the rate of convergence.
It is the equilibrium equations alone that govern the final solution. The equilibrium equations
are only influenced by the loads in (7), and no approximations concerning rigid body motion,
etc., have been made in these equations.
It is possible to account for the difference between symmetric and non-symmetric forms of
the geometric load matrix by introducing corrective terms in the equations used for equilibrium
iterations. These aspects are discussed in the section on solution of nonlinear equations.
The computation of the parameters in (11) is based upon checking of the position of the
centroidal points of the end sections in relation to water surface. This gives a rather abrupt
discontinuity of stiffness when an end point breaks through the water surface, particularly for
nearly horizontal members. A more extensive formulation has also been deviced in which the
element is replaced by an equivalent perfect cylinder and the stiffness is computed in an exact
way for the cylinder. Forces and stiffness are introduced as soon as a corner of the cylinder
touches the surface, and a more smooth transition is thereby achieved (see also previous section).
The equations for this case are rather involved and are not included in the present report;
however, they may be found in Reference 3. The numerical examples presented later are also
based on these complete expressions.

Highly flexible beams

Some engineering problems involve use of highly flexible beams submerged in liquids, e.g
rubber, plastic or composite tubes in water. The methods described in the preceding text are
based on the assumption that the hydrostatic forces may be computed on co-rotated, straight
beam segments. The validity of this assumption is, to a large extent, dependent upon the ratio
between the length of the beam elements and the radius of curvature of the bent beam. Experience
has shown that many cases, particularly those involving highly flexible beams with very little
pretensioning (axial tension), are numerically extremely sensitive, and that the straight element
assumption can lead to inaccuracies or convergence problems. For such cases it is necessary to
account for curved element effects in the hydrostatic force calculations.
There are essentially three effects from bending on the hydrostatic forces: (a) the capped end
pressure forces act on planes that are rotated in relation to the secant axis of the beam element;
(b) the distributed pressure resultant per unit length act normal to the curved mid-axis, not the
secant axis; (c) the bending leads to a structural surface on one side of the beam and a compressed
surface on the other. The end rotational effect is easily accounted for by way of the rotational
freedoms.
These end forces will also cancel between elements, except for at free ends and at abrupt
changes of the cross-section. The transfer of the distributed transverse force along the curved
axis to the end nodes may be done by way of integrals. However, a much simpler method is to
 HYDROSTATIC LOADING AND STABILITY ANALYSIS 583

deduce these forces from the vectorial difference between the total buoyancy force resultant on
the element and the capped end forces. This method is described in further detail in another
paper by the a ~ t h o r s . ~
The effect of the element bending on the incremental force-displacement relationship is also
important. In a typical non-linear, co-rotational formulation the tangential (incremental) stiffness
[ k T ] for an element may be written
IIkTI = Ckrnat1-t Ckgeornl+ [kl (16)
where [k,,,] is the material stiffness, [kgeom] is the geometric stiffness and [ k b ] is the geometric,
buoyancy stiffness which was derived earlier for a straight element.
The interaction between the beam bending and the hydrostatic forces may be illuminated by
imagining a slender beam lying on the bottom of the ocean. The water pressure on the ends of
the beam leads to axial forces. However, the beam will never buckle, regardless of its slenderness
or the water depth. This is simply due to the fact that the water pressure can never exent a
moment which leads to bending (buckling). Incrementally, this means there is no geometric
buckling effects stiffness associated with the axial force caused by the water pressure. This can
easily be adjusted for by computing the geometric stiffness for the element [kgeom] by use of an
effective axial compression force
Peff= P - P , = P - pwg A,h (17)
This adjustment has proved to be absolutely crucial in the analysis of problems like pipe laying
at deep waters and submerged flexible risers for oil and gas.

HYDROSTATIC FORCES ON ARBITRARY BODIES

Nodal forces for triangles and quadrilaterals

The computation of hydrostatic forces on arbitrary bodies is now considered. It is assumed
that the surface exposed to the water may be approximated by triangular or quadrilateral planar
facets or ‘elements’. A local co-ordinate system is introduced for each element in such a way
that the local i, axis follows local nodal line 1-2 and the local i, axis is perpendicular to the
plane of the element (see Figure 4).The i, axis is always pointing in the outward direction, that
is, the water pressure traction t acts in negative i, direction.
The consistent nodal forces ( F ) may be obtained by way of virtual work. When fixed or
co-rotated reference configurations are used it is necessary to utilize a ‘pseudo-pressure’, which
is real pressure (traction) transformed to the reference area by way of the deformation gradient.
The virtual work is then integrated over the undeformed reference area. Further details about
this may be found in References 5-7. The same consistent nodal forces may also be obtained
in a more direct fashion through integration of the virtual work from true pressure on the

Figure 4. Pressure on triangular and quadrilateral elements

584 P. G. BERGAN AND K. M. MATHISEN

Figure 5. Triangular elements with one or two nodes above the water surface

deformed area

(Gv)*{F)= -
I GwpdA

(Sv} is the virtual nodal displacements and w is the normal displacement of the deformed
(1 8)

surface.
For a triangular element the transverse displacement w as well as the pressure is taken to
vary linearly between the corners. Using (18) it is easy to show that the consistent nodal loads are

(F)=[ 1;1 =-&pwgAi3[1

2 1 1
1 21 2I ] [;I
where hi is the submerged depth for node i, and A is the area of the triangle. In addition, two
(19)

special cases of element that are intersected by the water surface have to be considered (see
Figure 5). The expressions for these cases are rather involved and may be found in Reference 3.
Quadrilateral elements are treated in a similar way by assuming bilinear interpolation in natural
co-ordinates of displacements and pressures between the corners. It can be shown that the integral
over the area results in the following explicit expression for the consistent nodal forces for the
quadrangle:

where

When the local co-ordinate system for the quadrilateral is chosen as in Figure 4, the C

Figure 6. Quadrilateral elements with one, two or three nodes above the water surface
 HYDROSTATIC LOADING AND STABILITY ANALYSIS 585

coeficients become

In addition to the fully submerged quadrilateral element, the cases of only one, two or three
submerged corners must be considered (see Figure 6). The expressions for these cases may be
found in Reference 3.

Incremental force-displacement relationship for triangles and rectangles

The incremental force-displacement relationship for water pressure on planar elements is
obtained by partial differentiation of the expressions (19) and (20) for consistent nodal forces.
This leads to a non-symmetric load matrix for elements crossing the water surface.
The geometric stiffness matrix for hydrostatic pressure acting on triangies and quadrilaterals
may be greatly simplified by utilizing the fact that the element area does not change appreciably
during the incremental motion and that the most important movement is the pure translational
motion decomposed along the global axes. In this way,just as for the beam element, the geometric
stiffness matrix may be simplified to a diagonal form where the only non-zero terms are associated
with the I, (depth) direction. The 9 x 9 matrix for the triangle simply have the following equal
terms:
k33=k,G=k99= -iPwgAT33 (22)
where T33 is the direction cosine between the i3 and I, axes
T33= i3-13 (23)
Special expressions for triangles with only one or two submerged corners have also been worked

Differentiation of (20) gives the following non-zero terms for the 12 x 12 geometric stiffness
matrix of the quadrilateral:

Note that the sign convention of (12a) still applies. Again, special expressions for quad-
rilaterals with only one, two or three submerged corners have been derived, and may be found
in Reference 3.

HYDROSTATIC STABILITY

Centre of mass
The mass matrix for the structural system is here constructed from lumped element mass
matrices and additional concentrated nodal mass matrices. The system mass matrix thus has a
diagonal form. An analysis of hydrostatic stability requires that the centre of the mass is known.
This is easily established through moment of intertia consideration about the global reference
586 P. G. BERGAN AND K. M. MATHISEN

system. The co-ordinates X,, Y, and Z , of the mass centre is obtained from
1
X --CMnX,
,-Mn
yG=GTMnyn1

1
z G ' z c n MnZn

where M is the total mass of the system. The sums are taken over all nodes n. The resulting
gravity force vector F, acts through the centre of gravity and is equal to
F, = - MgI, (26)

Resulting buoyancy force

The resulting buoyancy force is
Fb = Fbr3 = P w g 'dr3 (27)
where Vd is the volume of the water which is displaced by the system.
F, may also be obtained by summing the vertical components of the pressure forces at all
nodes. Here, however, V, is computed by way of the divergence theorem:

The sum is taken over all surface elements. The integrals over area A , of triangular and
quadrilateral elements corresponds to the volume V, of a prism formed by the element area and
the water plane and with sides parallel to the I, axis. The volume for beam elements is computed
in a more direct manner by product of cross-section A , and length L.
It is also necessary to establish the position of the resulting buoyancy force Fb. This force
acts in the I, direction and its X and Y co-ordinates may be found by a simple moment
consideration

where FZn is the vertical component of the pressure force at node n. In practice, however, it is
of interest also to find the 'point' where this force is acting. This point is taken as the centroid
of the displaced water volume (point B in Figure 7), and it is used in connection with practical
stability and potentially energy considerations. It is found from
X,=-
4 1 -

Xd V =
4

Vb e
c X, V,
where V, is the volume of the prism associated with the element, confer equation (28), and X, is the
 HYDROSTATIC LOADING A N D STABILITY ANALYSIS 587

Figure 7. Hydrostatic stability of unsymmetric body

co-ordinate for the centroid of this prism. The displaced water volume and the midpoint
co-ordinates are used directly for the beam elements. Accurate calculations are also made for
elements that intersect the water surface.
The variation of F , during a constant vertical displacement Auz is

where k,, is the vertical buoyancy spring associated with node n, see equation (12a). The
corresponding change of displaced water volume is

The change of volume associated with a small vertical motion Auz is computed on an
element-by-element basis in connection with the establishment of V,.

Finding the metacentre

The buoyancy force F, is equal and opposite to the gravity force F, for a floating body in
equilibrium. The stability of this state depends on what happens with the equilibrium when the
body is rotated a small angle 4. The gravity force always acts through the centre of mass G (see
Figure 7), while the resulting buoyancy force is displaced from point B to B’ during this rotation.
These two points are the current centroids of the displaced water. It is this offset of the buoyancy
force that is decisive for the stability.
The point of intersection between the line of the displaced buoyancy force and the initial line
of the gravity axis is called the metacentre M . It is seen that if this point lies above the centre
of gravity G, the force couple F, and Fb forms a positive, stabilizing moment that drives the
588 P. G . BERGAN A N D K. M. MATHISEN

body back towards its initial position. The distance ?%? is called the metacentric height, and
this distance must be positive in order for the floating body to be stable against rotations.
The metacentric height is normally computed in an indirect way by considering a destabilizing
force couple formed by F, and F, at B and G, respectively, against a stabilizing moment provided
by the change of water plane in relation to the body (see shaded area in Figure 7). This stabilizing
moment may be derived in a simple way by looking at the second area moment (moment of
inertia) of the displaced water surface in relation to the axis of rotation. The metacentric height
and, hence, the stability condition, is easily derived from these moments, see e.g. Reference 8.
The classical way of finding the metacentre, as described above, is acceptable for rigid, doubly
symmetric bodies with vertical sides. However, this method is insufficient for flexible bodies of
arbitrary shape. Here, the body in its current deformed configuration is given a small rotation
4 about the rotation axis through F (see Figure 7). It has been found that setting 4 equal to
at least 0-001rad gives good numerical stability. It may be seen for non-symmetric structures
or structures with sides that are not vertical, that such a rotation disturbs the vertical force
balance (Vd changes). It is thus necessary also to correct the vertical position of the floating
body. This is done through a Newton-Raphson iteration

where j is the iteration index. The iterations are terminated when

I/d,j+ 1 - vd,j
<& (34)
Vd

The convergence criterion E is typically set equal to 0.01 4. Point F’ in Figure 7 is the location
where the initial water line intersects the new water line. It is about this point that the pure
rotational motion has taken place.
Having found the rotated position which is in vertical force balance, the metacentric height
is easily retrieved from
~

1 -
GM =-GZ (35)
sin 4
Note that for three-dimensional problems there will be different metacentres for rotation about
the two principal axes of the displaced water plane. Both centres are automatically computed
in the present procedures.
The metacentre computed for a small 4 may be denoted the ‘initial metacentre’. The point
M does in fact move for large 4 . By computing for a set of angles 4 of increasing magnitude,
it is possible to determine the stability for large rotational motion. A critical angle is obtained
for the 4 that makes GM zero. For this state the floating structure is about to overturn.

COMPUTATIONAL ASPECTS

Finite element formulation

The methods of hydrostatic analysis described earlier have been implemented in a general,
nonlinear finite element formulation. This formulation allows for displacements and rotations
of unlimited magnitude as well as other types of nonlinearities, like inelastic material behaviour,
contact between moving bodies, nonlinear forces, etc. The nodal points are thought of as containing
 HYDROSTATIC LOADING A N D STABILITY ANALYSIS 589

infinitely rigid balls to which all adjacent finite elements are attached. The motion of these balls
is uniquely defined by their current position vector as well as rotation of an orthogonal base
vector triad which is initially parallel to the global base vectors. The rotations of this triad is
defined in terms of the base vector transformation

The transformation Tij is continuously updated for small incremental rotations during incre-
mentation and iteration, and it maintains its orthogonality properties throughout. Elements may
also be attached eccentrically in relation to the node. Further details about this method may
be found in References 9 and 10.
The numerical examples that follow involve beam elements as well as triangular and
quadrilateral membrane elements. The spatial beam element is based on the normal bending
theory for beams, and it is formulated in a co-rotational co-ordinate frame that allows for unlimited
average rotations.
The triangular and quadrilateral membrane elements are also formulated in co-rotational
co-ordinate frames with co-referential Green strains. Linear and bilinear interpolation are used
for the triangle and quadrilateral elements, respectively. The latter element utilizes selective
integration of shear strains and it has improved in-plane bending response.

Solution of nonlinear equations

The nonlinear, static equations for the system are solved in an incremental-iterative fashion
involving simple Euler forward integrations and Newton-Raphson iteration. The incremental
equations may be written

where [ K T ] is the tangential stiffness (see (16)) for the assembled system, {Ar} is the incremental
displacements and {AR} is the increment of external loads. The incremental buoyancy forces
{ A R b } for the system depends interactively on the incremental displacements through

This equation shows that the buoyancy matrix for the system may be split into one symmetric
and one non-symmetric contribution, just as for the single elements.
The symmetric part is now transferred to the left-hand side of equation (37):

An approximate solution to this incremental equation is obtained simply by neglecting the

buoyancy term on the right-hand side. Improved accuracy is achieved by substituting this first
{ A r } on the right-hand side and solving again for the {Ar} on the left-hand side. Finally, a
sequence of this type of substitutions may be performed until {AT} has converged.
Experience has shown that it normally does not pay to carry out iterations in order to obtain
an 'exact' solution to the approximate, incremental equations. Secondly, equilibrium iterations
are always carried out anyway in a more direct fashion through
(r}"' = { r } j + [[KT]'+ [Kb,]']-'{ {Rf'+' + {Rb)'" - {Rint)') (40)
In this Newton-Raphson type iterative equation ( j is the iteration index) the residual load term
consists of the applied external forces { R } j " , the buoyancy forces { R b } j + ' and the internal
reaction force vector ( R i n t ) j computed for the last configuration { r ) j . The gradient
590 P. G. BERGAN AND K. M. MATHISEN

operation is symmetric, and it is normally updated only once or twice for a sequence of iterations.
Note that [ K T ] is based on a geometric stiffnesses that are adjusted for the hydrostatic pressure
effect, see (17). The iterations normally converge to an accurate solution after a few cycles, even
for quite sensitive systems.
The incremental-iterative equations may be solved with prescribed load or displacement
increments. It is also possible to use automatic solution schemes like the ‘hyperplane displacement
control method’ described in References 11 and 12.
Certain classes of problems, in which real instabilities are involved, have to be solved
dynamically. A standard, nonlinear incremental-iterative dynamic solution algorithm l 3 has been
used. Rayleigh type damping proportional to structural mass and tangent stiffness is assumed.
In addition, drag damping for bar and beam members may be prescribed. The time integration
is carried out with the constant average acceleration scheme (Newmark with y = 1/2, = 1/4).
Further information about the dynamic solution may be found in References 11, 14 and 15.

Computer program
The methods described herein have been implemented in the general-purpose finite element
program FENRIS (Finite Element NonlineaR Intergrated System). This program is described
in References 9 and 10. Some particular capabilities for marine applications are that the program
accounts for Morison type hydrodynamic forces and linear as well as nonlinear wave forces in
addition to the hydrostatic forces.

NUMERICAL EXAMPLES

Partially submerged polyethylene pipe

Figure 8(a) gives the geometric and physical data for a floating 50m long polyethylene pipe
with capped ends. The pipe is modelled by 10 beam elements of equal length.
The initial state is taken as the static equilibrium configuration for the freely floating pipe
under own weight and buoyancy. The left end is then gradually submerged by prescribing a
negative displacement along the global Z-axis. 18 equal increments of 1 m are applied. The
problem is analysed statically and Newton-Raphson equilibrium iterations are performed for
each step.
Figure 8(b) shows the solution for the left end being submerged 0, 6, 12 and 18m. It was
found that the solution did not converge when the left end was forced deeper down than 19.8 m.
This is due to the fact that an instability occurs at this point and that the pipe suddenly swings
over to the left side. This critical depth of submerged H,, for a narrow rigid pipe is easily found
to be
.
J( 2)L
a .

Hcr =

where mp and m, is the mass per unit length of the pipe and the displaced water, respectively.
Substitution of the data in Figure 8 gives H,, equal to 18.82m. The small difference from the
computed results is caused by the effect of the pipe bending.
It should be mentioned that the dynamic solution for the swinging from the right to the left
side was also obtained after restart of the analysis with a dynamic solution algorithm.
Discretizations of 5 and 20 elements were also tried. It was found that the solution for the
finest mesh deviated very little from that of 10 elements. Even the 5-element model gave quite
 HYDROSTATIC LOADING AND STABILITY ANALYSIS 59 1

a)

t
1 1

L=50m E = 2.0.10~~/~~
R = 400mm ppvc = 1.2.103kg/m3
T = 25mm pw = 1025kg/m3
9 = 9-81m/s‘
T

-20 t
Figure 8. Floating polyethylene pipe: (a) initial position and data, (b) sequence of submerged positions

good results. This proves that not only is the finite element beam model quite good, but also
that the computation of hydrostatic forces on a pipe that intersects the water plane at a very
low angle, is highly accurate.
A three-dimensional model with a slight geometric imperfection in the horizontal plane
(Y-direction) was tried in order to investigate whether the ‘air-borne’ part of the pipe would like
to buckle out of the X-Z-plane. No such instability did occur in the analysis.
592 P. G. BERGAN A N D K. M. MATHISEN

M M

I
-

M M

L = 10m p = 7850kg/m’
D = 0.50m py= 1025kg/m’
R = 0.25m M = 1718kg

T = 0.01 m g = 9.81 m/s‘

E = 2.1 - 10”N/m’

Figure 9. Floating spaceframe: (a) geometry, (b) finite element model

t=o t = 2.4 sek. t = 4.8 sek. t = 7.2 sek.

t = 9.6 sek. t = 12.0 sek t = 14.4 sek. t = m

Figure 10. Sequence of rotated positions of the frame

 HYDROSTATIC LOADING AND STABILITY ANALYSIS 593

Complete overturning of a floating frame

Figure 9 shows a plane frame model of a floating structure. The frame is discretized by a total
of 20 beam elements and 18 nodal points. The main geometric and physical data are given in
the figure.
Figure 10 shows the prescribed, partially submerged, initial position of the frame and a sequence
of computed configurations after the frame was suddenly released. The response was computed
with a nonlinear dynamic algorithm with automatic computation of time steps the way it has
been suggested by Bergan and Mollestad.' 14, l 5 A high degree of mass-proportional damping
corresponding to 20 per cent of critical damping in heave motion was introduced. Hydrodynamic
drag and added mass were not introduced since the purpose here was mainly to find the final
stable equilibrium position and not to establish an accurate time history.
Some simple checks confirmed that the computed final, stable configuration was the correct
one. The present example demonstrates that the present method is capable of solving very large
rotational motion of floating structures. This type of problems are encountered in connection
with launching of structures into the ocean from barges.

Stability analysis of a ship

The final case considered is the hydrostatic stability of an idealized ship. Figure 11 shows the
spatial finite element model consisting of 17 triangular and quadrilateral membrane elements.
The data used in establishing the computational model are given in the figure.

b)

12 11 11.12 9,lO 8 9

5.6

Mass density of the water: p,, = 1025kg/m3

Acceleration of g r a v i t y : g = 9.81 m / s z
Mass lumped to nodal point 3 a n d 4: MI =2.1562-1O6kg
Mass lumped to nodal point 5 a n d 6: M 2 = 2.6005.106kg
Mass lumped to nodal point 9 and 10: Mg = 0 . 8 6 6 8 . 1 0 6 k g
Mass lumped to nodal point 1 1 and 12: M 4 = 0 . 7 1 8 7 - 1 0 6 k g

Figure 11. Hydrostatic stability of a ship: (a) geometry, (b) finite element model
594 P. G. BERGAN AND K. M.MATHISEN

I I 1 I I I I
-1.5 -1.0 -0.5 0.5 1.0
l-s 9
(rad)

Figure 12. Transversal metacentric height computed by FENRIS

The first part of the analysis consisted of finding the freely floating position of the ship. The
ship was restrained against horizontal motion in this phase (see Figure 1lb). An initial guess of
the floating position was made (see Figure lla), and the correct vertical position was found
quasistatically after a few iterations. The internal forces in the ship was computed simultaneously.
Hydrostatic stability about the longitudinal axis was considered next. The boundary conditions
at points 1 and 2 were removed and a sequence of rotational angles 4 was prescribed. Figure
12 shows the computed metacentric height as a function of 4.
It is clear that ships in general are relatively stiff structures, and that secondary elastic
deformations are of little or no consequence for hydrostatic analyses. At the same time, an
integrated analysis that considers both elastic deformation and hydrostatic forces has the great
advantage that it gives information about internal stresses in the ship. The present, crude model
required about 30 sec computer time on a NORD computer for the entire analysis. Much more
complicated models can be analysed at moderate computer cost.
It is also possible to assume a rigid-body mechanics formulation and to suppress the
computation of structural deformation in the FENRIS program.

ACKNOWLEDGEMENT
The authors would like to thank their colleagues in the FENRIS project for developing large parts
of the computer program that was used.

REFERENCES
1. J. S. Clinton, J. A. Clarkson, S. J. Cook and S. Walker, ‘Project to upright the Alexander L. Kielland’ Proc. 1981
Offshore Technology Conference, Paper Otc 4027, pp. 175-194, Vol 11, Houston, 1981.
2. 0.C. Zienkiewicz, The Finite Element Method, 3rd edn, McGraw-Hill, London, 1977.
3. K. M. Mathisen, ‘Large displacement analysis of hydrostatic loading and stability of submerged structures, Dr. Iny.
dissert., Division of Structural Mechanics, The Norwegian Institute of Technology, Trondheim, Norway (1986).
4. K. M. Mathisen and P. G. Bergan, ‘Nonlinear static analysis of flexible risers’, Proc. 5th Int. OMAE Symp., Tokyo,
1986.
5. L. E. Malvern, Introduction to the Mechanics of a Continuous Medium, Prentice-Hall, Englewood Cliffs, 1969.
6. K. Schweizerhof and E. Ramm, ‘Displacement dependent pressure loads in nonlinear finite element analysis’, Comp.
Struct., 18(6), 1099-1114 (1984).
7. H. D. Hibbit, ‘Some follower forces and load stiffness’, Int. j. numer. methods eng., 14, 937-941 (1984).
8. J. F. Douglas, J. M. Gasiorek and J. A. Swaflield, Fluid Mechanics, Pitman, London, 1979.
 HYDROSTATIC LOADING A N D STABILITY ANALYSIS 595

9. P. G . Bergan and A. Arnesen, ‘FENRIS-a general purpose finite element program’, Proc. 4th Int. Conf. on Finite
Element Systems, July 6-8, 1983, in Finite Element Systems Handbook (C. Brebbia, Ed.), C M L Publications,
Southampton, U. K., 1983.
10. FENRIS-Finite Element NonlineaR Integrated System, System Manuals. Theory-Program Ourline-Dora Inpur,
NTH-SINTEF-VERITAS, A/S VERITEC, Hqhik, Norway, 1984.
11. P. G. Bergan and E. Mollestad, ‘Static and dynamic solution strategies in nonlinear analysis’. Proc. 2nd In!. Con/’. on
Numerical Methods for Nonlinear Problems, New Orleans, Pineridge Press, Swansea, U.K., 1980.
12. J. Simons, P. G. Bergan and M. K. NygHrd, ‘Hyperplane displacement control methods in nonlinear analysis’, in
Innovative Methods for Nonlinear Problems (W. K. Lin, T. Belytschko and K. C. Park, Eds.), Pineridge Press, Swansea,
U.K., 1984, pp. 345-364.
13. K. J. Bathe, Finite Element Procedures in Engineering Analysis, Prentice-Hall, Englewood Cliffs, 1982.
14. E. Mollestad, ‘Techniques for static and dynamic solution of nonlinear finite element problems’, Dr. Ing. dissert.,
Division of Structural Mechanics, Report 84-1, The Norwegian Institute of Technology, Trondheim, Norway ( IY84).
15. P. G. Bergan and E. Mollestad ‘An automatic time stepping algorithm for dynamic problems’, Comp. Meth. A p p l .
Mech. Eng., 49, 299-318 (1985).