Вы находитесь на странице: 1из 8

# Paper No. 2001-JSC-153 Cassidy, M.J., Houlsby G.T. and Eatock Taylor, R.

Page 1 of 8
Application of Probabilistic Models to the Response Analysis of Jack-Ups
M.J. Cassidy
Centre for Offshore Foundation Systems, The University of Western Australia, Australia
G.T. Houlsby and R. Eatock Taylor
Department of Engineering Science, Oxford University, United Kingdom
ABSTRACT
Confidence in the long-term use of jack-up platforms in deep water
and harsh environments requires appropriate models for their
probabilistic models are used to develop further understanding of
this assessment. Particular emphasis is placed on achieving a
balanced approach in considering the non-linearities and
method of calculating short-term extreme response statistics, while
including variability in parameters, is briefly outlined, and a
numerical experiment for typical central North Sea conditions
detailed. Long-term extreme response statistics are also evaluated,
and the quantitative influence of the probabilistic formulations of
the variables (as opposed to deterministic values) shown.
KEY WORDS: Jack-up, Reliability, Strain hardening plasticity,
NewWave, Extreme response statistics, Spudcan
foundations, Random variable
INTRODUCTION
Most analysis of jack-up units use deterministic models, i.e. the material
specified. It is known, however, that there are parameters within the
models that are not unique, but have a range of possible values. By
using a probabilistic formulation of one or more of the material
properties, geometric dimensions or the actions on the structure, the
likelihood that the jack-up behaves in a certain way can be more
realistically evaluated. Within this paper, probabilistic methods are
used to develop an understanding of the response behaviour of jack-
ups.
In structural reliability theory, the failure probability of one component
is defined as
[ ] x d X f X G P P
X G X f

0 ) (
) ( 0 ) ( (1)
where ) ( X G is the failure function ( 0 ) ( X G is a failure state and
0 ) ( > X G a safe state), X is a set of k random basic variables, i.e.
[ ] [ ]
k
X X X X , , ,
2 1
K and ) (X f
X
the multi-variant density function
of X. For a component reliability analysis, failure criteria are usually set
on the limiting factors of strength or behaviour of the jack-up and are of
the form:
S R X G ) ( (2)
where R is the components resistance (or upper limit of
strength/behaviour) and S its serviceability (or calculated response
distribution from load effects). This failure region is shown in Figure 1
in a diagrammatic comparison of a deterministic and a probabilistic
analysis. In this paper response is defined as the horizontal movement
of the deck and therefore the limiting behaviour is the probability that
the deck displacement (S) will be greater than a specified value (R).
Confidence in this probabilistic approach depends on the following
factors:
The ability to evaluate the integral in Equation 1 accurately.
The accuracy of the failure function. In any reliability analysis,
the results can only be judged by the accuracy of the individual
modelling components used in the analysis. This is especially true
for highly interactive and non-linear processes, as seen in jack-ups.
With inappropriate and highly conservative assumptions, such as
use of pinned footings to represent the spudcan-soil interaction,
not only are the reliability results inaccurate, but the level of
uncertainty in them can be unacceptably high.
The probabilistic modelling of the uncertainty in the basic random
variables. The statistical spread assumed for random variables
needs to reflect their inherent variability, and this will be
investigated for application to jack-up dynamic response in this
paper.
NUMERICAL ANALYSIS MODEL JAKUP
A dynamic structural analysis program written at the University of
Oxford and named JAKUP has been used for the numerical experiments
detailed in this paper. The motivation behind JAKUP is the
development of a balanced approach to the analysis of jack-up units,
Paper No. 2001-JSC-153 Cassidy, M.J., Houlsby G.T. and Eatock Taylor, R. Page 2 of 8
models taken into account.
Surface elevations and wave kinematics are evaluated using NewWave
theory (Tromans et al., 1991), with the extended Morison equation used
of NewWave theory is that although it is deterministic, it accounts for
the spectral composition of the sea. The surface elevation around an
extreme event is modelled by the statistically most probable shape
associated with its occurrence, and is given by the autocorrelation
function of the Gaussian process defining the sea-state. The continuous
time autocorrelation function is defined as
( )

0
2
) (
1

d e S r
i
(3)
with the time history of the extreme wave group proportional to ) ( r at
the region around 0 . The time relative to the initial position of the
crest is represented by
1
t t , where
1
t is the time when the extreme
event occurs. For a time lag of 0 , the autocorrelation function of
Equation 3 reduces to one, allowing the surface elevation of the
NewWave to be scaled efficiently. This is shown below in Equation 4.
The NewWave shape as defined by the autocorrelation function can be
discretised by a finite number (N) of component sinusoidal waves. As
there exists a unique relationship between wave number and frequency,
spatial dependency can also be included, leading to the discrete form:

N
n
n n n
X k d S X
1
2
) cos( ] ) ( [ ) , (

(4)
where is the surface elevation and
n
k and
n
the wavenumber and
frequency of the n
th
component respectively. As defined previously, is
the crest elevation,

d S
n
) ( the surface elevation spectrum and
the standard deviation corresponding to that wave spectrum. X = x - x
1
is the distance relative to the initial position with X = 0 representing the
wave crest. This allows the positioning of the spatial field such that the
crest occurs at a user-defined position relative to the structure, a useful
tool for time domain analysis.
The deterministic formulation of NewWave allows it to be conveniently
and efficiently implemented into structural analysis programs, such as
JAKUP. Furthermore, because it is based on linear theory, the water
wave particle kinematics can be easily obtained once the water surface
has been established. In the analyses described in this paper Wheeler
stretching was used to determine the kinematics within the wave crest
(Wheeler, 1970).
Figure 2 shows a NewWave wave profile evaluated in JAKUP for the
upwave and downwave legs of the example structure. A peak NewWave
crest of 12 m has been focussed on the upwave leg at the reference time
( 0 t ). For this case the sea-state is described by the JONSWAP wave
energy spectrum, with a significant wave height (H
s
) of 12 m and a
mean zero crossing period (T
z
) of 805 . 10 s.
appropriate nodes.
Structural Model
JAKUP uses the finite element method with dynamics modelled by time
stepping and solved using the Newmark ( 25 . 0 , 5 . 0 ) solution
method. The structure is idealised as a two-dimensional bar-stool
model, with Figure 3 showing a diagram of the jack-up used. The mean
water depth was assumed to be 90 m, with the rig size and properties
typical of a three-legged jack-up used in harsh North Sea conditions.
The leg and hull are represented by elastic finite elements, with the
corresponding equivalent beam stiffnesses and masses shown in Figure
3. Example structural node locations and hydrodynamic modelling
coefficients for the leg sections are also shown. Non-linearities due to
P , Euler and shear effects are considered. However, one important
limitation in JAKUP is that no account is taken of non-linearities at the
leg-hull connection. Further details of the structural model can be found
in Martin (1994), Thompson (1996) and Williams et al. (1998).
Foundation Model
Allowing for a level of foundation fixity of spudcan footings reduces
critical member stresses (usually at the leg/hull connection) and other
critical response values (Chiba et al., 1986; Norris and Aldridge, 1992).
Furthermore, the natural period of the jack-up is reduced, usually
improving the dynamic characteristics of the rig. One of JAKUP's major
advantages is the implementation of Model C - a strain hardening
plasticity model describing the load-displacement relationship for
spudcan footings on dense sand.
1
Model C is based on a series of
experimental tests described by Gottardi et al. (1999).
The plasticity framework models the load-displacement relationship for
spudcans in essentially the same way a constitutive law for a metal (or
the numerical plasticity model computes updated tangent stiffnesses for
each step. In Model C an empirical expression defines a yield surface in
( H R M V , 2 , ). (Note: moment is normalised by the radius of the
spudcan, R). The surface is defined by the best fit of experimental data.
It has a functional form of
( )
2
0 0 0
2
0 0
2
0 0
2 2 2
, 2 ,
V m h
R M aH
V m
R M
V h
H
H R M V f

,
_

,
_

( )
( )
0 1
2 1
2 1
2 1
2
0
2
0
2
2 1
2 1

,
_

,
_

1
1
]
1

+

V
V
V
V
(5)
This surface may be described as an eccentric ellipse in section on the
planes of constant V, and approximately parabolic on any section
including the V-axis (the term a determines the eccentricity of the
ellipse).
0
V determines the size of the yield surface and indicates the
intersection of the yield surface with the V-axis (H = 0 and 0 2 R M ).
The dimensions of the yield surface in the horizontal and moment
directions are determined by
0
h and
0
m respectively. The parameters
1
and
2
round the surface near 0
0
V V and 1
0
V V
respectively.
Any changes of load within this surface will result only in elastic
deformation. However, plastic deformation can result when the load
state touches the surface, with the plastic footing displacements
calculated from a flow rule. Although the shape of the Model C yield
surface is assumed constant, the size may vary, with the yield surface
expanding as the footing is pushed further into the soil and contracting
with footing heave. This expansion (or contraction) is defined by an

1
Model C follows Model A and Model B which are strain hardening
plasticity models described by Martin (1994) for spudcans on clay.
Paper No. 2001-JSC-153 Cassidy, M.J., Houlsby G.T. and Eatock Taylor, R. Page 3 of 8
empirical strain-hardening expression and is shown diagrammatically in
Figure 4. Details of an appropriate strain-hardening expression that
accounts for the conical shape of spudcans on dense sand can be found
in Cassidy and Houlsby (1999).
In strain-hardening plasticity theory, the response of the foundation is
expressed purely in terms of force resultants on the footing. It can
therefore be coupled directly to the numerical analysis of a structure.
2
Plasticity models, such as Model C, are particularly attractive in
modelling jack-ups as they allow the movement of the spudcans to be
evaluated, with differentiation between the actions of the three legs
possible.
A full description of Model C and its implementation into JAKUP can
be found in Cassidy (1999).
ANALYSIS OF JACK-UPS USING PROBABILISTIC METHODS
Response Surface Method (RSM)
If many combinations of probabilistically distributed variables are to be
examined, the number of analyses required in a Monte Carlo type
analysis becomes very large. This makes it unrealistic to carry out
sufficient analyses using a program such as JAKUP. An efficient way to
proceed is as follows. A specially selected set of JAKUP analyses is
used to determine a Response Surface which effectively encodes the
results of these analyses. The Response Surface (RS) is then used in
combination with probablilistically distributed variables to calculate
response statistics. Cassidy (1999) demonstrated that the RS method
provides an accurate method in this context, and can serve to reduce the
number of computations enormously.
The form of the response surface that has been chosen to model extreme
jack-up response is the versatile second-order polynomial with mixed
terms:
+ + + +
<
k
i
k
i
k
j i
k
j i ij i i i i
X X d X c X b a X S
1 1
2
) (

(6)
where
i
X and
j
X are the i
th
and j
th
components respectively of the set
of random variables. The terms a, b
i
, c
i
and d
ij
are the free parameters
needing evaluation
3
and the error of fit. The term S

represents the
service response predicted by the RS and in this paper the extreme
response calculated is horizontal hull displacement. However, the
method is equally applicable to other response criteria. The free
parameters are determined using the composite design method (see, for
instance, Myers and Montgomery, 1995) and regression analysis
(Cassidy, 1999).
This form of the RS was chosen for its ability to model response with
significant system curvature. Unfortunately, it carries no formal
resemblance to the actual surface resulting from the mechanical
modelling of jack-ups. However, with further developments it should be
possible to identify a surface shape resembling the physical processes
more closely.

2
Though first used as a geotechnical solution to another problem by Roscoe and
Schofield (1956), the use of force resultant models has recently been used in the
examination of jack-up performance (for instance by Schotman (1989), Martin
(1994), Thompson (1996) and Cassidy (1999)).
3
Total of ( ) 2 1 2 1 + + k k k parameters.
Distributions of Random Variables used in Numerical Experiments
Seven basic random variables are used in the numerical experiment
detailed here. Table 1 outlines these variables, their distribution type,
mean values and Coefficient of Variations. These variables were
selected based on the experience of previous studies found in the
literature (see Cassidy (1999) for details) and knowledge from the
authors experience. There are three types of basic random variables
used:
structural modelling, and
foundation modelling.
A short description of the set of random variables (X) is given in Table
1, with further explanation below.
in individual components used in formulating that load. Variation in the
Morison equation is considered by using the drag coefficient (
d
C ) as a
random variable. There is considerable uncertainty in its application
and this is reflected in values previously used in the literature, with
CoVs quoted around 20-25% (see for instance Thoft-Christensen and
Baker, 1982; Lseth and Bjerager, 1989; Lseth and Hauge, 1992;
Sigurdsson et al., 1994). Uncertainty due to the effect of marine growth
was considered to be included as a component of the uncertainty within
the Morison coefficient.
Wind and current are other variables influencing the force on a jack-up.
In reliability studies, wind can be described either in terms of wind
velocity (Thoft-Christensen and Baker, 1982) or, as is used here, a wind
force (Morandi et al., 1997).
Structural Modelling
An important variable found in the literature for the response of jack-
ups was the mass of the deck (Baker and Ramachandran, 1981;
Karunakaran, 1993; Morandi et al., 1997). Deck mass influences the
dynamic response and geometric non-linearities, as well as the pre-load
applied to the foundations. As Model C is very dependent on the pre-
load level (as it determines the initial yield size), in the numerical
experiments described here these two effects have been separated into
Elasto-Plastic Foundation Model (Model C)
Due to limited data (in terms of both quantity and quality) there is a
large measure of subjective judgement when determining geotechnical
uncertainty (Gilbert and Tang, 1995). When probabilistic methods for
geotechnical models have been used, it has often been as an overall
uncertainty on the deterministic models results (see for example Nadim
and Lacasse (1992)). However, this was not the approach adopted here.
As was the case for the environmental and structural variability,
foundation uncertainty has been included as variation in components of
Model C, rather than uncertainty in the model itself. Incorporating
uncertainty or bias to the whole model is extremely difficult to quantify
and reduces any attempt to reflect the physical processes occurring. The
Model C parameters listed in Table 1 can be described as follows.
Shape of Yield Surface (
0
m ): The size of the widest section of the yield
surface in the moment plane ( V R M : 2 ) is determined by
0
m . This is
as defined in Equation 5 and shown in Figure 5. As rotational fixity at
Paper No. 2001-JSC-153 Cassidy, M.J., Houlsby G.T. and Eatock Taylor, R. Page 4 of 8
the foundations contributes to jack-up dynamic response levels, the
yield surface shape in the moment plane was considered important to
study. Therefore, the non-dimensional Model C parameter
0
m has been
expressed as a random variable.
Size of Yield Surface (pre-load factor): Before use at a site, jack-ups are
vertically pre-loaded by pumping sea-water into their ballast tanks. For
a JAKUP analysis using Model C this causes the yield surface for each
spudcan to expand to the size defined by the vertical pre-load. Before
from the tanks) and this is shown in Figure 6. The pre-load factor is
defined as the ratio of pre-load to operational weight per spudcan.
Shear Modulus Factor (g): Elastic response of the soil needs to be
defined for any increments within the yield surface. Finite element work
has shown that cross coupling exists between the horizontal and
rotational footing displacements (Bell, 1991; Ngo-Tran, 1996), with a
linear elastic incremental force (V, M, H) -displacement (w, , u)
relationship of the form:

,
_

1
1
1
]
1

,
_

du
Rd
dw
h
k
c
k
c
k
m
k
v
k
GR
dH
R dM
dV
2
0
0
0 0
2 2 (7)
G is a representative shear modulus and R the radius of the footing.
Equation 7 is implemented in Model C with the elastic constant values
(
h m v
k k k , , and
c
k ) as evaluated by Bell

(1991)

for a Poissons ratio of
0.2. The shear modulus is estimated as:
a a
p
R
g
p
G

2
(8)
where
a
p is atmospheric pressure and the submerged unit weight
of sand. A non-dimensional shear modulus factor g has been defined
and is used here as a random variable to scale all of the elastic
coefficients.
The statistical distributions ascribed to the Model C parameters should
be considered as best judgements for this example numerical
experiment, not as definitive results.
Other Uncertainty
Uncertainty in the fit of the RS has not been included as it is expected
to be minimal. This assumption has been tested for a series of short-
term sea-states in Cassidy (1999). Additional basic variables were
investigated in a sensitivity study in Cassidy (1999), including the
inertia coefficient (C
m
), structural damping and parameters affecting the
strain-hardening expression of Model C. They did not significantly
affect the extreme jack-up hull displacement and therefore were not
used in the calculations presented in this paper.
NUMERICAL EXPERIMENT RESULTS
Short-Term Results (for a three-hour storm)
An example of the methodology to compile short-term extreme
response statistics for probabilistic random variables is outlined here.
The sea-state is characterised by the JONSWAP spectrum with
parameters H
s
= 12 m and T
z
= 805 . 10 s, representing conditions that
could describe a 100-year event in a central North Sea location.
For five discrete NewWave crest elevations that are representative of
the full range of wave heights in a three-hour storm, five separate
Response Surfaces were estimated using the central composite design
and regression analysis. In this 100-year case, the five NewWave crest
elevations were 3.5, 7, 10, 12 and 15 m. A total of 143 JAKUP runs
were used to fit each surface (for k = 7). Because of variability of
extreme wave amplitudes (described by the Rayleigh distribution for
short-term conditions), as well as the probabilistic occurrence of the
basic random variables, a method to incorporate both in the evaluation
of extreme response distributions is necessary, and follows these steps:
4
Step 1: In a Monte Carlo calculation, for the same set of random
variables (X) a response is calculated for each of the five NewWave
elevations using the Response Surfaces previously determined. A line
of best fit is evaluated for these five responses allowing interpolation
for intermediate crests.
Step 2: With an extreme wave elevation randomly calculated,
5
one
extreme response can be estimated from the random-set best-fit
polynomial of Step 1.
By repeating these steps for many sets of X, a statistical distribution of
the extreme response can be obtained. In this paper 10 000 sets of basic
random variables were simulated and 10 000 extreme wave amplitudes
used to evaluate the short-term distribution of extreme deck
displacements. Figure 7 shows this distribution and also compares it
with the distribution of extreme deck displacements calculated with
variable NewWave amplitudes, but with the basic variables at their
mean values (previously described in Cassidy (1999) and Cassidy et al.,
2000). The additional variation caused by the uncertainty in the basic
random variables is clearly shown. Two more observations from Figure
The increase in displacement caused by the basic random variables
at large deck displacements is greater than the reduction at low
deck displacement levels. It is believed this is due to the non-linear
response to the basic random variables, and this is discussed
further in the following section on long-term distributions.
The two curves intersect below the 50% exceedence level, at about
38 . 0 ) ( x Q when 36 . 0
deck
m. Again, reasons for this are
explored in the following section.
Long-Term Distributions
With the short-term RSM methodology established, example long-term
conditions were considered in order to:
estimate probability of exceedence values by convoluting the
short-term Response Surfaces with return period; and
compare the probability of exceedence estimates using
probabilistic basic random variables with the values calculated
using just their deterministic mean values (previously evaluated
and described in Cassidy (1999) and Cassidy et al. (2000)).
The example long-term conditions are shown in Table 2. Furthermore,
the five NewWave elevations (
5 2 1
, K ) used to evaluate the
Response Surfaces are detailed. The experiments included wind and
current, with their mean ( ) and standard deviations ( ) also shown
in Table 2.

4
This method is similar to that described in Cassidy (1999) and Cassidy et al.
(2000) for evaluating short-term extreme response statistics of random waves
(but with all other properties deterministic).
5
Evaluated by Monte Carlo simulation of the number of waves in the short-term
time period (Ncrest) using the Rayleigh distribution of wave elevations.
Paper No. 2001-JSC-153 Cassidy, M.J., Houlsby G.T. and Eatock Taylor, R. Page 5 of 8
Figure 8 shows the extreme deck displacement distributions evaluated
with statistical variation in the basic random variables (or random-
variable distributions) and compares them with the short-term
distributions calculated for the mean values of the basic variables (or
mean-variable distributions). Table 3 outlines all of the statistical
properties of the extreme response distributions (mean values and
CoVs). A considerable increase in CoVs for the random-variable
distributions is observed for all sea-states. Although for a return period
of 1-year the mean deck displacement is virtually the same (for the
random-variable and mean-variable curves), there is a steady increase in
the mean response for the random-variable extreme deck displacement
distributions for longer return periods, as is shown by the percentage
increases in Table 3.
For the 1-year return period, the intersection of the mean-variable and
the random-variable distributions on the cumulative distribution plot of
Figure 8 is at approximately 5 . 0 ) ( x Q . For the other short-term
distributions, however, as the sea-states become less probable the
intersection is at progressively lower ) (x Q values, crossing at around
26 . 0 ) ( x Q for the 10
5
-year sea-state. An explanation considering the
linearity (or otherwise) of the response to each of the random variables
is explored further here.
Firstly, assume the change in response to all of the random variables is
linear, i.e. for the same probability of occurrence of the random
variables value at a distance from its mean, either lower or higher,
the reduction in response is the same as the increase. This is shown in
Figure 9 (a). For this case, the mean-variable and the random-variable
extreme response distributions should intersect at 5 . 0 ) ( x Q .
Furthermore, as shown in Figure 9 (b), the variation of the random-
variable curve away from the mean-variable extreme response
distribution should be of the same magnitude for both 5 . 0 ) ( < x Q and
5 . 0 ) ( > x Q . On the other hand, if the response to a random variable is
non-linear, the random-variable extreme response distribution becomes
skewed. If, as in Figure 9 (c), the additional response is relatively larger
for the same probability away from the mean as it is smaller, then the
distributions should intersect at 5 . 0 ) ( < x Q (assuming more than one
random variable). Additionally, the random-variable distribution will
not have the same difference in variability, but will show a larger
change in response at high ) (x Q values, as indicated in Figure 9 (d).
This is a simplistic explanation and one which becomes more complex
with competing non-linearities and cross-term effects.
For the 1-year sea-state the majority of runs used to evaluate the RS
were within the initial Model C yield surface (only a few of the runs for
the highest NewWave elevation of 11m caused expansion of the yield
surface). Therefore, the runs were all within the foundations elastic
region and the footings were acting as linear springs. In this situation,
the Model C parameters
0
the response. The remaining parameters, especially
d
have an approximately linear effect on the horizontal deck
displacement, and this is reflected in the extreme response distribution
for the 1-year sea-state shown in Figure 8.
As the sea-states become harsher, the Model C parameters become more
important and more non-linear. For example, with all of the other
variables at their mean level and only
0
m varied for the 10
5
-year sea-
state and a NewWave amplitude of 19 m, the deck displacement
calculated by JAKUP is increased by 219 . 0 m for
0
m at two standard
deviations below its mean, but only decreased by 122 . 0 m for
0
m at
two standard deviations above its mean. This influence would be
increased when cross terms are considered, especially due to the
increase in load caused by higher
d
C values. It is these non-linear
effects which are thought to create the skewed curves described by
Figure 8.
Long-Term Extreme Responses
Convolution of the short-term distribution with the assumed logarithmic
distribution of sea-state occurrence gives long-term probability
predictions of response. However, as sea-states do not occur in discrete
intervals, the extreme response distribution of any intermediate sea-state
must be adequately estimated to evaluate this convolution numerically.
A method of scaling the normalised (by the 50% exceedence value) 1 in
100 year distribution is used to estimate values in intermediate sea-
states. Details of this scaling method can be found in Cassidy (1999) or
Cassidy et al. (2000).
Long-term extreme exceedence probabilities have been calculated for
the horizontal deck displacement of the example jack-up for the
statistical distributions of random variables outlined in Table 2, and are
shown in Figure 10. The estimates are significantly larger than the
annual probability of exceedence values calculated for just the mean
deterministic values of the basic random variables (Cassidy, 1999). This
can be explained by the increased variability observed in the short-term
random-variable extreme response distributions. For all sea-states (apart
from the 1-year return period) the variations in the basic random
variables caused larger mean deck displacements for the short-term
distributions (as outlined in Table 3). This, as well as the fact that the
short-term response is relatively larger at high ) (x Q , and that the
curves cross at lower ) (x Q levels as the sea-state return period
increases, means that the long-term exceedence prediction is
significantly increased. It is believed that if a linear foundation model
was used, the difference in long-term exceedence estimates would not
be as large.
CONCLUSIONS
This paper is concerned with the dynamic analysis of jack-up units. A
balanced approach to non-linearites was taken. This included use of a
plasticity model for the load-displacement behaviour of spudcans on
P , Euler and shear effects in the structural model. Physical and
modelling uncertainties were considered using a probabilistic analysis
approach and the RSM. It was found that accounting for the uncertainty
in the values of a set of basic random variables significantly affected the
extreme response statistics.
For short-term statistics there was an increase in CoV values due to the
probabilistic formulations. Furthermore, for increasing sea-state
severity, the 50% exceedence response value increased in comparison
with the equivalent deterministic approach. Both of these affected long-
term estimates, giving increased annual probability of exceedence
results. Accounting for the probabilistic distributions of random
variables was therefore shown to be important.
ACKNOWLEDGEMENTS
Support from the Rhodes Trust for the first author is gratefully
acknowledged.
REFERENCES
Baker, M.J. and Ramachandran, K. (1981). Reliability analysis as a tool in the
design of fixed offshore platforms. Proc. of 2
nd
Int. Symp. on The Integrity
of Offshore Structures, Glasgow, pp. 135-153.
Bell, R.W. (1991). The analysis of offshore foundations subjected to combined
Paper No. 2001-JSC-153 Cassidy, M.J., Houlsby G.T. and Eatock Taylor, R. Page 6 of 8
Cassidy, M.J. (1999). Non-linear analysis of jack-up structures subjected to
random waves, D.Phil Thesis, University of Oxford.
Cassidy, M.J., Eatock-Taylor, R., Houlsby, G.T. (2000). Analysis of jack-up
units using a constrained NewWave methodology. Report No. 2224/00,
Oxford University Engineering Laboratory.
Cassidy, M.J. and Houlsby, G.T. (1999). On the modelling of foundations for
jack-up units on sand, Proc. 31
st
Offshore Technology Conference, Houston,
OTC 10995.
Chiba, S., Onuki, T. and Sao, K. (1986). Static and dynamic measurement of
bottom fixity. The Jack-up Drilling Platform Design and Operation,
Collins, London, pp. 307-327.
Gilbert, R.B. and Tang, W.H. (1995). Model uncertainty in offshore
geotechnical reliability. Proc. 27
th
Offshore Technology Conference,
Houston, pp. 557-567, OTC 7757.
Gottardi, G., Houlsby, G.T. and Butterfield, R. (1999). The plastic response of
49, No. 4, pp.453-470.
Karunakaran, D. (1993). Nonlinear dynamic response and reliability analysis of
drag-dominated offshore structures. Dr. ing Thesis. Div. of Marine
Structures, NTH, Norway.
Lseth, R. and Bjerager, P. (1989). Reliability of offshore structures with
uncertain properties under multiple load processes. Proc. 21
st
Offshore
Technology Conference, Houston, pp. 107-114, OTC 5969.
Lseth, R. and Hauge, L. (1992). Probabilistic methods in calibration of partial
safety factors. In Recent Developments in Jack-up Platforms, Blackwell
Scientific Publications, London, pp. 228-245.
Martin, C.M. (1994). Physical and numerical modelling of offshore foundations
under combined loads. D.Phil. Thesis, University of Oxford.
Morandi, A.C., Frieze, P.A., Birkinshaw, M., Smith, D. and Dixon, A.T. (1997).
Reliability of fixed and jack-up structures: a comparative study. Proc. of
BOSS97 Behaviour of Offshore Structures, Vol. 3, Delft University of
Technology, pp. 111-126.
Myers, R.H. and Montgomery, D.C. (1995). Response surface methodology:
Process and product optimization using designed experiments. John Wiley
& Sons, New York.
Nadim, F. and Lacasse, S. (1992). Probabilistic bearing capacity analysis of
jack-up structures. Canadian Geotechnical Journal, Vol. 29, No. 4, pp.
580-588.
Ngo-Tran, C.L. (1996). The analysis of offshore foundations subjected to
Norris, V.A. and Aldridge, T.R. (1992). Recent analytical advances in the study
of the influence of spudcan fixity on jack-up unit operations. Recent
Developments in Jack-Up Platforms, Elsevier, London, pp. 424-450.
Roscoe, K.H. and Schofield, A.N. (1956). The stability of short pier foundations
in sand. British Welding Journal, August, pp. 343-354.
Schotman, G.J.M. (1989). The effects of displacements on the stability of jack-
up spud-can foundations, Proc. 21
st
Offshore Technology Conference,
Houston, pp. 515-524, OTC 6026.
Sigurdsson, G., Skjong, R., Skallerud, B. and Amdahl, J. (1994). Probabilistic
collapse analysis of jackets, Proc. of 13
th
Int. Conf. on Offshore Mechanics
and Arctic Engineering (OMAE), Vol. 2, pp. 367-379.
Thoft-Christensen, P. and Baker, M.J. (1982). Structural reliability theory and
its applications. Springer-Verlag, New York.
Thompson, R.S.G. (1996). Development of non-linear numerical models
appropriate for the analysis of jack-up units. D.Phil. Thesis, University of
Oxford.
Tromans, P.S., Anaturk, A.R. and Hagemeijer, P. (1991). A new model for the
kinematics of large ocean waves -applications as a design wave-. Proc. 1
st
Int. Offshore and Polar Engng Conf. , Edinburgh, Vol. 3, pp. 64-71.
Wheeler, J.D. (1970). Method for calculating forces produced by irregular
waves. J. Petroleum Technology, March, pp. 359-367.
Williams M.S., Thompson R.S.G., Houlsby G.T. (1998). Non-linear dynamic
analysis of offshore jack-up units. Computers and Structures, 69(2), pp.
171-180.
Table 1 - Set of seven basic random variables used in numerical
experiments
Random
variable
number
(Xi)
Basic
variable
Category Description Mean value
(X)
Distrib-
ution
CoV
(%)
with depth)
0.8 m/s Normal 20
(applied at deck)
1.35E6 N Normal 10
4 mass of
hull
structural Mass of hull 16.1E6 kg Normal 10
5 m0 Model C Dimension of
yield surface in
the moment
direction
0.086 Normal 15
factor
Model C Scales initial size
of yield surface
(V0)
1.925 Normal 10
7 g Model C Shear modulus
factor (affects
elastic constants)
4000 Log-
normal
37.5
Table 2 Sea-states and NewWave elevations used in the long-term
numerical experiments
Return
period
(year)
Hs
(m)
Tz
(s)
1
(m)
2
(m)
3
(m)
4
(m)
5
(m)

(wind)
(MN)

(wind)
(MN)

(curr.)
(m/s)

(curr.)
(m/s)
1 8.98 9.35 2.5 4.5 6.5 8.5 11 0.756 0.0756 0.599 0.1198
10 10.60 10.16 3.0 5.5 7.5 10 14 1.053 0.1053 0.707 0.1414
100 12.00 10.81 3.5 7.0 10 12 15 1.350 0.1350 0.800 0.16
10
3
13.25 11.36 4.0 7.5 10.5 13 17 1.647 0.1647 0.884 0.1768
10
4
14.40 11.84 4.5 8.0 11 15 18 1.944 0.1944 0.960 0.1920
10
5
15.46 12.27 4.75 10 13 16 19 2.241 0.2241 1.031 0.2062
Table 3 Statistical properties of the short-term extreme deck
displacement distributions
All Xi with mean values
(Cassidy, 1999)
All Xi include statistical
variability
Return
Period
(year)
( )
deck

(m)
CoV (%)
( )
deck

(m)
CoV (%)
Percentage
increase in
( )
deck

## 1 0.175 12.99 0.176 22.38 0.5

10 0.278 17.81 0.291 31.59 4.7
100 0.400 20.58 0.433 35.66 8.3
10
3
0.596 22.15 0.660 36.67 10.7
10
4
0.857 22.68 0.952 36.82 11.1
10
5
1.141 22.86 1.294 38.72 13.4
Vertical plastic
displacement
M/2R
H
penetration curve
Yield surface in
(V, M/2R , H) load space
V
0
Expansion
of surface
Basic Random Variables (X) Response Response
Analysis Model
Basic Random Variables (X) Response Response
Analysis Model
Resistance (R)
failure region
(S>R)
Input Parameters Component ServiceResponse( S) Failure Check
Deterministic Analysis:
Probabilistic Analysis:
p
r
o
b
a
b
i
l
i
t
y

d
e
n
s
i
t
y

f
u
n
c
t
i
o
n
p
r
o
b
a
b
i
l
i
t
y

d
e
n
s
i
t
y

f
u
n
c
t
i
o
n
Service (S) Resistance (R)
Service (S)
Figure 1 - Illustration of deterministic and probabilistic methods
-12
-8
-4
0
4
8
12
-100 -80 -60 -40 -20 0 20 40 60 80 100
time (s)
s
u
r
f
a
c
e

e
l
e
v
a
t
i
o
n

(
m
)
upwave legs
downwave leg
Figure 2 - NewWave surface elevation at the upwave and
downwave legs
M
e
a
n

w
a
t
e
r

d
e
p
t
h

9
0
m
80m
35.2m
Upwave Downwave
Two legs
Single leg
51.96m
Values:
For a single leg:
E = 200 GPa
I = 15 m
4
A = 0.6 m
2
M = 1.93x10
6
kg
A
s
= 0.04 m
2
G = 80 GPa
D
E
= 8.44 m
A
h
= 3.94 m
2
C
d
= 1.1 (mean value)
C
m
= 2.0 (mean value)
For hull:
I = 150 m
4
A
s
= 0.2 m
2
M = 16.1x10
6
kg (mean
value)
For spudcans:
R = 10 m
Structural Damping 2% of
Critical (mean value)
Figure 3 - General layout of idealised jack-up used in the analyses
Figure 4 - Expansion of yield surface with plastic vertical
displacement
M/2R
V
V
0
m
0
V
0
Figure 5 Size of the yield surface in the M/2R:V plane
V
H
Initial yield surface in
space
V
initial service
M/2R
V
0

0

/ V
initial service
Figure 6 Definition of basic random variable Pre-load
Hs = 12 m
T z = 10.805 s
= 12 m
upwave
leg
downwave
leg
x = 0m x = 51.96m
V
Paper No. 2001-JSC-153 Cassidy, M.J., Houlsby, G.T. and Eatock Taylor, R. Page 7 of 8
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x = response of deck displacement (m)
Q
(
x
)

=

P
(
d
e
c
k

d
i
s
p
l
a
c
e
m
e
n
t

<

x
)
mean variable values used (Cassidy, 1999)
variation in random variables
Figure 7 - Extreme deck displacement distributions for the
100-year sea-state with and without variation in the
basic variables
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
x = response of deck displacement (m)
Q
(
x
)

=

P
(
d
e
c
k

d
i
s
p
l
a
c
e
m
e
n
t

<

x
)
1 in 1 year
1 in 10 year
1 in 100 year
1 in 1000 year
1 in 1E4 year
1 in 1E5 year
thick line = random-variable dist.
thin line = mean-variable dist.
Figure 8 - Distributions for six short-term sea-states with and
without variation in the basic variables
r
e
s
p
o
n
s
e
r
e
s
p
o
n
s
e
p
r
o
b
.

d
e
n
s
i
t
y

f
u
n
c
t
i
o
n
p
r
o
b
.

d
e
n
s
i
t
y

f
u
n
c
t
i
o
n
mean
mean
Q(x)
x = response
Q(x)
x = response
1.0
0.5
1.0
0.5
(a) Linear response
(d) (c) Non-linear response
(b)
intersects at
Q(x) < 0.5
variation from mean
curve is greater at
Q(x) > 0.5
mean curve
random var.
curve
mean curve
random var.
curve
Figure 9 - Comparison of linear and non-linear response to
short-term statstics
1.E-09
1.E-08
1.E-07
1.E-06
1.E-05
1.E-04
1.E-03
1.E-02
1.E-01
1.E+00
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 6
horizontal deck displacement (m)
a
n
n
u
a
l

p
r
o
b
a
b
i
l
i
t
y

o
f

e
x
c
e
e
d
e
n
c
e
random variables
mean values of random variables
Figure 10 - Comparison of annual probabilities of exceedence of
deck displacements for variable input parameters and
their mean values (including wind and current)
Paper No. 2001-JSC-153 Cassidy, M.J., Houlsby, G.T. and Eatock Taylor, R. Page 7 of 8