HERON is jointly edited by:

of the
faculty of Civil Engineering,
Delft University of Technology,
Delft, The Netherlands
and

vol. 36
1991
no. 3

STEVIN-LABORATORY

TNO BUILDING AND

CONSTRUCTION RESEARCH.

Rijswijk (ZH), The Netherlands

HERON contains contributions
based mainly on research work
performed in these laboratories
on strength of materials, structures
and materials science.

Contents

AN EXPERIMENTAL AND NUMERICAL STUDY

ON JACK-UP DYNAMIC BEHAVIOR
P. Liu

ISSN 0046-7316

Offshore

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1 linirolill!llctim:n........... . . . . . . . . . . . . . . . . . . . . .
1.1
Total problem survey................
1.2
Scope of work. . . . . . . . . . . . . . . . . . . . . . .
1.3
Notation. . . . . . . . . . . . . . . . . . . . . . . . . . . .
:2 Software ilevelllIimenil . . . . . . . . . . . . . . . . . . . . . . .
2.1
Nosda package . . . . . . . . . . . . . . . . . . . . . .
2.2
Randa package ..................... ,
3 Physical moilel tests. . . . . . . . . . . . . . . . . . . . . . . ..
3.1
Models and test seiup. . . . . . . . . . . . . . ..
3.2
Test program. . . . . . . . . . . . . . . . . . . . . . ..
3.3
Typical results. . . . . . . . . . . . . . . . . . . . . ..
Analysis of s[mtic !Ind free viiJlI'!diollll tests. . . . . ..
4.1
Static stiffness .......................
4.2
Free vibration. . . . . . . . . . . . . . . . . . . . . ..
4.2.1 Natural period. . . . . . . . . . . . . . . . . . . . . ..
4.2.2 Inferred stiffness . . . . . . . . . . . . . . . . . . ..
4.2.3 Structural damping .. . . . . . . . . . . . . . . ..
4.3
Summary. . . . . . . . . . . . . . . . . . . . . . . . . ..
EDITORIAL BOARD:

A C. W. M. Vrouwenvelder,
editor in chief
R. de Borst
J. G. M. van Mier
R. Polder
J. Wardenier

Secretary:
J. G. M. van Mier
Stevinweg 1
P.O. Box 5048
2600 GA Delft, The Netherlands
TeL 0031-15-784578
Fax 0031-15-611465
Telex 38151 BUTUD

3
5
6
7

9
9

9
11
12
12
15
18
20
20
21
21
22
24
26

26
Structural nonlinearities . . . . . . . . . . . . .. 26
5.2
Hydrodynamic nonlinearities. . . . . . . . .. 28
5.3
Summary. . . . . . . . . . . . . . . . . . . . . . . . . .. 29
iii RegIDlhu wa'le test !maiysis :.m.@ computeI!'
silflmiatiolls .. . . . . . . . . . . . . . . . . . . . . . . . .. .... 29
6.1
Introduction . . . . . . . . . . . . . . . . . . . . . . .. 29
6.2
Computational model. . . . . . . . . . . . . . .. 30
6.2.1 Hydrodynamics. . . . . . . . . . . . . . . . . . . . .. 30
6.2.1.1 Wave kinematics. . . . . . . . . . . . . . . . . . .. 31
6.2.1.2 Hydrodynamic loads ................ , 32
5 Model l1Iloniil!llemities expected . . . . . . . . . . . . . . ..

5.1

This publication has been issued with financial support

from the Netherlands Technology Foundation (STW).

6.2.2
6.2.2.1
6.2.2.2
6.2.2.3
6.2.2.4
6.3
6.4
6.5
6.5.1
6.5.2
6.5.3
6.5.4
6.5.5
6.5.6
6.5.7
6.6

Structural model establishment .......

Initial structural computational models.
Discretization in time ................
Calibration using experimental data ....
Structural computational models
(in air) .............................
Measured versus simulated results .....
Computational effort. ................
Further computational results .........
Absolute versus relative velocities .....
Results of linearized model ...........
Free surface effects ..................
Hydrodynamic cancellation ...........
Airy versus Stokes 2nd order wave
theories ............................
Results of different connection
modeling ...........................
P-O effect ...........................
Summary ...........................

7 Irregular wave test analysis lind computer

sinmlations .................................
Introduction ........................
7.1
Data collection and preprocessing .....
7.2
7.2.1 Data recording ......................
7.2.2 Data digitalization ...................
7.2.3 Data preprocessing ...................
7.3
Probability analysis results ............
7.3.1 Relative motion type .................
7.3.2 Drag and nonlinear structure type .....
7.4
Spectral analysis results ..............
7.4.1 Relative motion type .................
7.4.2 Drag and nonlinear structure type .....
7.5
Measured versus simulated results .....
Summary ...........................
7.6
7.6.1 Data analysis ........................
7.6.2 Computer simulations ................

32
33
34
34
37
41
45
46
46
47
48
48
49
49
51
52
54
54
56
56
57
57
58
60

61
63
65
69
74
78
78
79

8 Conclusions ................................. 80
8.1
Model testing and experimental data
processing .......................... 80
8.2
Computer simulations ................ 81
8.3
Closing remarks ..................... 82
SlIlmmary ..................................... 83
Acknowledgement. . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 84
Symbols amI. notation . . . . . . . . . . . . . . . . . . . . . . . . .. 85
References. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .. 88
Appendix I. Static test results .. . . . . . . . . . . . . . .. 93
Appendix II!. Free vibration test results . . . . . . . . .. 95
Appendix III. Hydrodynamic analysis theory
selection ......................... 105
Appendix IV. Treatment of nonlinearities
and P-O effect .................... II 0
Appendix V Structural modeling ............... 119

Publication in HERON since 1970

Abstract
This paper presents the more salient results of an experimental and numerical study on
jack-up dynamic behavior.
The laboratory studies of three principle jack-up platform models were carried out in
both regular and irregular waves. The data from irregular wave tests were analyzed in
both the probability domain and frequency domain supported by a careful error analysis.
Computer simulations were carried out in the time domain using a nonlinear, dynamic,
multiple degree of freedom software which includes various hydrodynamic interaction
options.
The experimental results and associated computer simulations demonstrate that
nonlinearities are important even with the present simplified model testing and different
nonlinearities have different (sometimes compensating) influences on the structure's
dynamic behavior. Some more specific results include: (1) The stiffness obtained from
static tests can be significantly lower than that inferred from dynamic vibration tests; (2)
relative motions from structural compliance are such that they cannot be responsibly
neglected in the hydrodynamic computation; and (3) inclusion of the P-O effect in the
structural schematization is essential for the jack-up simulations.

Key words
Jack-up, Dynamics, Hydrodynamics, Nonlinear, Model, Experiment, Simulation, Random.

An experimental and numerical study on

jack~up dynamic behavior
1 Introduction
Common offshore units can be categorized into two types: fixed structures (such as jacket
platforms and gravity platforms) and mobile structures (drill barges, drill ships,
semisubmersibles, for example). The fixed structures are held stable either by piles or
their own weight, providing ultimate stability for offshore operations. The mobile
structures maintain their locations at the sea by either anchoring or dynamic positioning,
offering mobility and reusability. Combining the advantages of the above two concepts,
a jack-up rig is a hybrid type platform with both stability and mobility. Basically, a jackup is a self-elevating pontoon with retractable legs. When the legs are pulled up by
means of a jacking mechanism, the jack-up rig is effectively a barge and can be towed
by tugboats or carried by a heavy transport vessel to another location. When the
pontoon is elevated above the sea level with legs extended down to the sea bed, the rig
enters the platform mode (elevated operation condition), furnishing a relatively steady
and stable working place offshore. Because of this unique combination of properties,
jack-up platforms have been used extensively in the offshore industry for more than 30
years. There are about 440 of them at present, engaged primarily in hydrocarbon drilling
operations.
The present work was carried out as a part of a Delft University of Technology Jack-up
Project conducted by the Workgroup Offshore Technology (WOY), with objective to
increase the detailed knowledge of the behavior of such platform components as well as
the prediction of the overall structure's elevated behavior and (remaining) lifetime.
The need for such a study is demonstrated by the relatively high rate of structural failure
for jack-up rigs as compared to fixed platforms and the considerable discrepancy existing
among present var~ous industry assessment methods and criteria for elevated jack-up
platforms. The failure statistics of jack-up platforms based upon data from the
Worldwide Offshore Accident Databank shows that jack-up platforms are at least 20
times more 'accident-prone' than fixed offshore structures. Additionally, when the
present program was initiated in 1988, the industry criteria and procedures then in use
were so inconsistent that they could easily result in failures rates which differ by a factor
of 50 to 100 - see Efthyrniou (1988). (References are listed in the text by author and
year; a complete reference list is to be found at the end of the main text of the paper.)
The reasons for this seem to be rooted in too simple an approach to the computational

schematization of such platforms for design or evaluation purposes.

Since then,
considerable efforts from the worldwide jack-up industry have been made to bring about
some degree of harmonization for the jack-up assessment. While a substantial consensus
has been achieved after three years of joint industry activity, a spectrum of questions
remain to be answered - see Anon (1990). Further investigations on various aspects such
as spudcan fixity, hydrodynamic coefficient determination, assessment criteria selection,
etc., are still to be carried out - see Anon (1991).
With jack-ups venturing into deeper water - say 130 m or more - for longer term use
such as for production from marginal fields in more exposed locations, the adequate
performance assessment and analysis of these platforms become even more cruciaL
A price paid for the mobility is that a jack-up platform is much less rigid as compared
with a fixed platform. This flexibility comes from its weaker connections at both the
upper end (to the deck via the deck-leg clamping system) and lower end (to the sea bed
via the spudcans) as well as the independence of the separate legs (there are no braces
connecting one leg to another). Because of this flexibility, dynamic effects become
remarkably more important. This will be true for survival condition analyses and
especially for fatigue analyses. Additionally, the natural frequency of such rigs in sway
can enter an energy-rich exciting wave frequency band. This, combined with dynamic
influences, is expected to make overall structural responses even greater and damping
precision critical.

1.1 TOTAL PROBLEM SURVEY

Numerous investigations have been conducted to analyze the dynamic behavior of

elevated jack-up platforms and assess their structural safety - see, for example, Anon
(1981 - 1983 and 1989), Boon (1986), Bradshaw (1988), Brekke et al. (1989 and 1990),
van Haaren and Boon (1988), Manschot and Mommaas (1988), Lagers (1990), Leijten
and Efthymiou (1989), Sliggers (1990), etc. The total jack-up durability problem
definition and associated literature study were carried out in the earlier phase of this
project by Massie, Liu and Boon (1989). They came to the conclusion that elevated jackup platforms can be significantly nonlinear in their dynamic structural behavior. The
most important of these involve interactions of the legs with:
the sea bed via a spudcan,
the deck via the deck-leg clamping system, and
the sea itself: waves and currents acting on the moving structure.
Within the TV Delft Jack-up Project a series of investigations have been performed to
attack these various nonlinear interaction problems. For spudcan-soil interaction, the

readers are referred to Holtrop (1989), Spaargaren (1989), Stuit (1989), and Klaver
(1990), for deck-leg interaction Griindlehnler (1989) and Michels (1990), for
hydrodynamic interaction Zeelenberg (1990) and Massie, Liu and Zeelenberg (1991).
An overview of the progresses made so far in this program has been given by Massie and
Liu (1990).
Another report by Liu (1989a) inventoried and compared the (mathematical) methods
available for the analysis of jack-up platforms. It was concluded that the extrapolated
use of traditional analysis methods (such as quasi-static approach, design wave approach,
etc.) is no longer sufficiently dependable for predicting the nonlinear behavior of
elevated jack-up rigs. A more advanced, stochastic, nonlinear, dynamic, time domain
analysis approach must be chosen to simulate the nonlinear physical response of a jackup platform.

1.2 SCOPE OF WORK

While retaining the overall vision of the total jack-up durability problem, the author's
work has been concentrated on the investigation of the influence of hydrodynamic and
structural nonlinearities on elevated jack-up rigs. The hydrodynamic study focuses on
the wave load on the legs. The influences of currents are not included in the present
work. The examination of the structural nonlinearities concentrates on the jack-up
structure itself; its interaction with soil is excluded from the present study.
The research was done following two tracks: On the one hand model tests on the jack-up
platforms were carried out, and these were complemented on the other hand by
numerical modeling of such rigs.
The different test models have been chosen such that they segregate the several types
of hydrodynamic and structural nonlinearities. As for hydrodynamics the influence of
drag is important and the question arises whether one should use the absolute motion
of the water particles or the relative motion between leg and water particles to compute
the hydrodynamic loads on the legs. The structural nonlinearities came from the legdeck connection and possibly the P-O effect (second order effect). The physical models
were tested in both regular and irregular waves.
The numerical simulation required the development of a software package that accounts
the development of a software package that accounts for the nonlinear hydrodynamic
interaction and nonlinear structural behavior. As explained before, this program starts
from a time domain approach.

Not restricted to simulating the behavior of the present physical models only, the
software development is aimed to make available a more precise, verified, dependable
and commonly accepted computational model, that will make it possible to properly and
conveniently evaluate less exact but more efficient routine procedures for jack-up analysis
and assessment. The first validation of this computational model was done using the
experimental data from the present tests.
The mere fact that a numerical simulation will be successful does not necessarily mean
that it is understood which nonlinearities are dominant and under which circumstances.
To gain such insight the random wave test data were analyzed in two ways: Probability
analysis was performed to study the distortion of statistical distributions caused by
nonlinearities; frequency analysis exposed the influences of nonlinearities on the energy
distribution and helped determine which nonlinearities had major impact on the system
behavior. The software developed for these analyses is also supported by a responsible
error analysis in both the probability and the frequency domain.
In conclusion, the work presented in this paper includes the following three aspects:
Software Development
Two software packages have been developed for the project: (1) NOSDA
simulation software for the Nonlinear Offshore Structure Dynamic Analysis; (2)
RANDA software for RANdom Data Analysis. These codes are briefly described
in Chapter 2.
Physical Model Tests
Testing on three jack-up models was carried out in the wave tank of the Ship
Hydromechanics Laboratory, TU Delft. The models were not scaled to reproduce
actual field conditions exactly but they do retain the some important characteristics
of prototypes. The models and test program are discussed in Chapter 3.
Experimental Result Analysis and Computer Simulations
The processing of the measured data from the irregular wave tests was supported
by a careful error analysis using RANDA software. The model tests in regular and
irregular waves were simulated using NOSDA software. The experimental data
analyses and associated computer simulations are presented in Chapter 4 through
7.
The main conclusions of entire work are presented in Chapter 8.

This paper is structured in such a way that whenever possible, the main body of the text
is kept concise and descriptive; only the principles and essential results are presentedo
The detailed data and mathematics are described in the appendiceso More complete
theoretical aspects have been given by Liu (1991b)0

1.3 NOTATION

The present work lies on the interface between disciplines such as hydrodynamics and
structural mechanics (inclusion of statistical analysis complicates the notation system
further)o Each of these disciplines has its own, independent notation convention; it is
unavoidable that they conflict at timeso
in notation are necessary in this
paper. Consistency has been maintained, however, and - where possible - with an
international standardo A symbol table is included at the end of the main text of the
paper.

2 Software
Two software packages have been developed and used as
tools for this
of each package is given in the
study: NOSDA and RANDA A principle
remainder of this chapter.

2.1 NOSDA PACKAGE

NOSDA was developed as a special purpose software package for
nonlinear,
dynamic analysis of offshore structureso More details of this software have been
documented in a separate report by Liu and Massie
The structural analysis kernel of this software package has a strong heritage in another
nonlinear dynamic analysis program, TILLY, developed by the Mechanics and Structures
of Civil Engineering of the TV Delft - See
Department within the
Blaauwendraad (1989)0
The dynamic analysis is
in the time domain so that various types of
nonlinearities associated with jack-up dynamic behavior mentioned in Section L 1 can be
accommodatedo These nonlinearities can result from fluid particle kinematics, material
properties, geometric deformations, fluid-structure and soil-structure interactionso A
principle flowchart of NOSDA is included in figure 201.

The primary uniqueness of NOSDA involves the computation of hydrodynamic forces on

a moving structure in waves and! or currents.
As the price of its precision and flexibility, NOSDA shares the disadvantage of all time

domain nonlinear dynamic programs - they are computer time costly.

The NOSDA software is used as the computer simulation tool in Chapters 6 and 7.
Some details of the implementation of NOSDA are also to be found in these chapters.

~
timer

absolute water kinematics

compute relative
kinematics
1 -_ _ _ _ _ _ _ _

f no

hydrodynamic load

no
~

integration in time domain

structure response

Figure 2.1 Principle Flowchart of NOSDA

10

2.2 RANDA PACKAGE

The RANDA software analyzes random data in the both probability domain and the
frequency domain.
The probability analysis involves the computation of the statistical distributions of
instantaneous values, peak values and extreme values of measured data at different
transfer steps (wave elevation - wave kinematics - hydrodynamic loads - global structural
response - detailed structural response, for example). Existence of nonlinearities will
cause distortion in the statistical distributions from one step to another. The probability
analysis results provide information about how the energy is distributed among the
motion levels. Knowledge of the distortion caused by nonlinearities and thus the
resulting response distribution after each transfer step is important for both extreme and
fatigue analysis of a jack-Up.
The frequency analysis examines the autospectral properties of an individual measured
time series and cross-spectral properties between two time series. With a nonlinear
system the cross-spectral quantities will generally not be invariant, instead, they will be
dependent upon the input energy level as well as energy distribution. The spectral
analysis results shed light on the energy distributions and their transfer relationship as
a function of frequency.
The random data processing in both domains mentioned above is supported by a
responsible error analysis. This associated error estimate procedure is often essential for
such type of analysis, since an irresponsible processing can cause so big an error in the
results that any attempt to interpret them becomes totally meaningless.
A principle flowchart of the RANDA software is shown in figure 2.2. The two routes on
the left hand side of the flowchart (namely, the spectral analysis and probability analysis)
are employed for the random data analysis in Chapter 7. More details about this
software package have been given by Liu (1991a).

11

"

data preparation
- filtering
- scaling

'"

(.procellse o1 time
series storage
I

~
- segmenting,
- overlapping,
- windowing

data normalization
(oplional)

- segmenting
- zero padding

probabilty distributions,
distribution moments
and associated errors

auto- & crosllcorrelation functions

--

auto- & crollsspectral quantities

and associated errors,
confidence intervals

via the FFT

Figure 2.2 Principle Flowchart of RANDA

3 Physical model tests

3.1 MODELS AND TEST SETUP

Two principle physical models of three-legged jack-up structures - named Model I and
Model n, respectively - were designed and fabricated. These were tested in Towing
Tank I of the Ship Hydromechanics Laboratory of the Faculty of Mechanical Engineering
and Marine Technology. These tests were carried out using instrumentation from and
by personnel of this laboratory.
These models each had three identical circular cylindrical legs. Model I was designed
with relatively large diameter legs yielding inertia-dominated hydrodynamic forces; Model
II had more slender legs and thus more drag-dominated forces.

12

For each model, the deck was placed about 2.4 m above the tank bottom and was
assumed to be relatively rigid with (initially designed) completely damped deck-leg
connections.
The legs were hinged at their lower end with force meters located between the hinges
and the model base plate on the tank floor.
A convenient tank water depth, d, was 2.0 m.
Additional testing of Model II with extra deck masses - then denoted as Model II-M was carried out to expose the effects of deck load eccentricity - the P-13 effect and the
effects of a variation in the natural period of the model. Figure 3.1 illustrates the model
geometry.

T
1

Wave Probe

l/i:

I
o~

700

A~X

4.2 m

!
__

~G~g.o~m~ ~~
__

--------

a. Overall Plan of Towing Tank

b. Model Dimensions

Figure 3.1 Physical Model Setup

The coordinate system is chosen as follows: The origin is located at the base of the bow
leg, the x-axis is directed along the tank (away from the wavemaker), the z-axis is vertical
(positive upwards) and the y-axis is perpendicular to the x-z plane according to a righthand axis rule.
Necessary simplifications were made in the model design to concentrate attention on the
physical processes to be studied. While some discussion of model scales is relevant, no
attempt has been made to reproduce actual field conditions in the models. Instead, the
physical models should be seen as full scale structures, themselves.

13

The structure's natural frequency, in , was chosen to be around 1 Hz for both Model I
and Model II (the natural frequency of Model U-M became considerably lower due to
the extra deck mass). The model leg spacing was chosen to include a reasonable
hydrodynamic force cancellation effect. The design approach, further, was to choose the
leg stiffness such that the model platform has a quasi-static deflection of 2% of the water
depth at deck level if the peak force resulting from a design wave was applied to all 3
legs simultaneously. By choosing different leg materials and adjusting deck masses, it
proved possible to essentially retain the natural frequency and quasi-static deflection (as
outlined above) while using two quite different types of legs.

The most important

physical parameters for each of the three models are listed in table 3.1.
More details of the model set-up and test program can be found in a separate report by
Journee et al. (1988).
Three dynamometers were mounted at the base of each leg to measure the force
components along three axes. The forces measured by the dynamometers were labeled
as FAx ' FAy, F Az , F Bx , F By , F Bz , F Cx , F Cy and F Cz , where the first subscript denotes the
location of the dynamometers - see figure 3.1 - and the second refers to the direction.
A 5-g accelerometer was mounted at location D on the deck to measure x and y
components of the acceleration there, uD and iiD (Note that the displacements along
the x, y and z axes are denoted as u, v and wand the associated subscripts indicate the
location.)
Additionally, the horizontal displacements of the deck were measured at locations A and
C, denoted by u A , VA' Uc and Vc so as to doublecheck the acceleration measurements
and detect possible rotations around the vertical axis.
A two-wire conductance wave probe was mounted adjacent to the platform in the same
line perpendicular to the tank wall as the windward leg A. This wave elevation was
indicated by 17A .

14

Table 3.1 Physical Parameters of the Three Models

Item

Model I

Model II

Model II-M

Unit
kg

Construction mass

18.20

5.90

5.90

Additional deck mass

15.72

0.52

3.67

kg

Total model mass

33.92

6.42

9.57

kg

Deck materia!

alum./PVe

aluminum

aluminum

Leg material

hard pve

red copper

red copper

2118.0

133.1

133.1

N.m2

clamped

clamped

clamped

hinged

hinged

hinged

Leg outer diameter

0.090

0.016

0.016

Leg spacing (triangular)

0.700

0.700

0.700

Deck (topside)

2.373

2.403

2.403

Displacement meter

2.373

2.403

2.403

Leg stiffness, EI
Deck-leg connection
Leg-bottom connection

Elevation from tank floor:

Accelerometers

2.373

2.403

2.403

Still water surface

2.004

2.004

2.004

Leg cylinder base

0.143

0.143

0.143

Leg hinge

0.Q78

0.Q78

0.078

0.87

0.80

0.50

Hz

Natural freq., fn ' (approx.)

3.2 TEST PROGRAM

The model testing program included exposing the models to regular and irregular unidirectional, long crested waves as well as static and free vibration tests. As a special
case, some tests were completed with a superposition of two regular waves. The
experiments of this type in the past have often been concentrating on the regular wave
situation. Inclusion of irregular wave tests will help gain insight into the jack-up behavior
in a real random sea.

15

Totally 230 wave runs were carried out (including 9 runs for the instrumentation control).
The duration of each regular wave run was about 5 minutes (excluding transient motion)
and that of each irregular wave run was about 20 minutes.
All of the experimental data were recorded in an analog form on magnetic tapes (JR
recorder). Some data were also recorded on paper using a UV recorder. The UV
recording provides sufficient data for further processing with the static, free vibration and
regular wave tests, while before the irregular wave test results can be processed and
analyzed the analog data on the tapes need to be digitized.
The static tests were carried out for each model by exerting static, horizontal loads at the
deck level and recording the corresponding displacements.
The free vibration tests were carried out by giving a initial displacement at deck level
then releasing the deck and recording the deflection trace.
During the model testing the pen recorder and analog magnetic tape recorder were
connected in parallel to the sensors; the visual observation of the trace on paper could
not guarantee the
of recording on the magnetic tape. When digitizing the data
on the tapes, severe truncations have been found in the recorded data with paired
regular waves; no effort has, therefore, been dedicated to process this group of data
further.
With regular wave tests, possible wave frequencies in the basin range from about 0.6 to
1.3 Hz with wave heights up to 0.080 m. (Higher frequencies were reached for
lowerwave
The three models were tested in 103 regular wave runs. The wave
states used are listed in table 3.2,
In the tests, the wave heights actually generated were often slightly different from their
nominal values listed in the table. The measured wave heights were used in the later
analysis.

16

Table 3.2 Regu/ar Waves Tested

Model No.

Run No.

Nominal Height, H

VVave Frequency,J

(em)

(Hz)

0.7 - 1.7

0.7 - 1.2

0.7 - 1.1

15 - 50

II

78 - 123

II-M

162 - 182

0.6 - 1.2

0.5 - 1.15

0.5 - 1.0

12

0.5 - 0.8

0.55 - 0.8

0.3 - 0.9

0.3 - 0.7

36 successful
wave runs were performed with the three models: runs 55
63 for Model I, runs 133
140 for Model n and runs 210
218 for Model
II-M. Truncations - "~IJ'-""LUH in the wave elevation channel - occurred also in a few
runs with this group of tests. Excluding the truncated runs, 22 wave state combinations
listed in table 3.3 were
in the
the table
is the '''I",HHH..<UH
wave
and Ip the peak
Table 3.3 Irregular Waves Tested

Model I

I\,fodel II-M

Model II

--

~-

Ip

Run

(em)

(Hz)

no.

141

3.216

0.739

210

143

2.262

0.739

0.800

144

2.384

0.856

2.930

0.800

145

59

3.490

0.800

147

3.388
2.610

60

3.992

0.800

149

5.204

0.895

61

3.356

0.800

151

5.852

0.817

62

3.894

152

6.300

0.934

63

4300

2.262

0.739

2.328

0.934

4.906

Jp

Run

(em)

(Hz)

no.

t------.-- r--'
55
3.154

0.800

56

4.444

0.800

57

3.928

58

Run

no.

0.800

0.800
0.800
~

0.800

Ip

1)

(Hz)

211

3.300

0.778

212

4.622

0.739

0.817

215

4.906

0.661

0.934

216

3.160

0.545

I
I

6.300

0..545

0.778

17

3.3 TYPICAL RESULTS

Only a small representative part of the test results will be presented here, more results
are to be presented in the following chapters.
The static test results are plotted as force (exerted at the deck level) versus (deck)
displacement. An example is given in figure 3.2.
60

Force (N)

'0
30

20

O~~----~~-------d

0,02

0.04

0.06

0.06

Deck Disple.cement (rn)

Figure 3.2 Measured Overall Static

Constitutive Relation (Model J)

The free vibration tests result in decay curves such as shown in figure 3.3.

Deck Displacement (m)

O.06r~;;",;,:,:;,,;;;:,,~=;,;,,:=---------,

Time (sec)

Figure 3.3 Free Vibration Trace Record in Air

(Model J)

18

As an example, the Response-Amplitude-Operator (RAO) curves of Model I for

different wave heights derived from regular wave tests are superimposed in figure 3.4.
The RAOs in the regular wave case are determined by normalizing the deck
displacement amplitude with respect to the input wave amplitude.

2.5 r:RA""O-,(_-.:....)- - - - - - - - - - - - - ,

Model I
1.5

0.5

o~~~~~~~~
0,5

0.6

0,7

O.B

0.9

1.1

1.2

1.3

1.4

1.5

1.6

1.7

Frequency (Hz)
~ H ""

2 em

-+-- H '" 4 em

-+-- H '" B em

Figure 3.4 Measured RAGs for Deck

Displacement with Various Wave Heights
(Regular Waves)

As the typical results from the spectral analysis of the irregular wave test data, a wave
elevation spectrum, its corresponding deck displacement spectrum and the associated
RAO curve are presented in figure 3.5. The RAO with irregular waves is defined as the
gain factor between the wave elevation and the deck displacement. (A gain factor is the
modulus of the frequency response function which is determined here as the crossspectrum divided by the input spectrum). The notation system as shown in this figure
will be used frequently in the graphic presentations later in this paper: the horizontal-axis
is the frequency, f ; the solid curve is the value of interest (the spectrum, gain factor,
coherence function, and so forth), embraced by the 95% confidence interval (shown in
the figure as the two fine dashed curves); and the coarse dashed curve down at the
bottom of the figure is the normalized random error as a percentage. In the figure (-)
denotes that the quantity is dimensionless. G ~q is the wave spectrum, Guu is the deck
displacement spectrum and er is the normalized random error.

19

:~I'''(-'') /"'\

,,(-It:

~/:/
i f " ' . \.

...... \\.

0\)0%

. ------:5==0-=----------0-

I-~~

().~

0.6

0.5

0.7

o.a 0.9 1 U 1.2 1.3

Frequency (-)

~.~

o0.4

['<-)1:%

./\

M~!\

}(~~

:.;

______

10-

0.-'&

erm
_____ m

~~

0.6 G.a: 0.7

g.a 0.9

0.3 !lAO (-)

. .-.

~ !<,~~~!~

!leck DI"pl",ce",,,,,t Spectrum

~:IG""(cm..

RAO ("'"''''' el"""U"" to oide sway)

2<)%
~O%

1.6

G,\)

E.(-) "'''''

--.:>/

~
""

o,s 0.7 0.9 0.9 ~ 1.1 1.2 1.3 iA 1.6

Froquency (Hz)

ModoJ II (Ru" UH)

_ _ _ _ _ _ _ _. :

U:

1.:3

ft.4

-0%

U5

Frequency (Hz)

Figure 3.5 Measured Wave Spectrum, Deck Displacement Spectrum and Derived RAOs

4 Static and free vibration test analysis

The data resulting from the static tests and free vibration tests in air are analyzed in this
chapter. The results will be used to shed light on the establishment of the structural
computational models in Chapter 6. Many global properties of the models such as
structural tiffnesses, damping ratios, natural periods, etc., are derived from these two
groups of tests.

4.1 STATIC STIFFNESS

The global lateral stiffness of each model (defined as the force exerted at the deck level
divided by the resulting deck displacement) from static tests, Ks , is listed in table 4.1.
- more detailed data are given in Appendix 1.

20

Table 4.1 Model Static Stiffnesses

Model No.

Model I

Model II

Model II-M

Static Stiffness, Ks (N/m)

508.00

19.90

16.82

The only difference between Model II and Model U-M is that Model II-M has extra deck
mass and therefore extra P- /) effect. The P- /j effect reduces overall structural stiffness;
this is confirmed by the larger stiffness of Model II in the above table.
Note that the static stiffness of Model I in the table is calculated from the test data
before the deck to leg connection of this model was modified - see Section 4.2.1.

4.2 FREE VIERAnON

The detailed experimental results and associated analysis of the free vibration tests in
air are given in Appendix n. Only important results are summarized here.

4.2.1 Natural Period

During the free vibration tests in air, the response periods between two successive upcrossings of the deck displacement were found to decrease with increasing vibration
cycles (in fact with decreasing response levels) for all models. This variation is primarily
attributed to the imperfect deck-leg connections. These connections were different from
their original (rigidly clamped) design.
The deck-leg connections of Model I were glued to improve their mechanical behavior
(making the clamping more rigid).
The materials used in Model H( -M) were not suited for gluing, even though the
imperfection in the deck-leg connection is expected to have a more significant impact on
the structure's behavior with this model since its legs and deck beams are smaller than
those of Model I - see Appendix n. Consequently, during a free vibration run, different
natural periods were obtained for different response cycles - in fact for different response
amplitudes just as was the case initially with Model I. These natural periods within one
run were averaged over a few cycles to yield the 'representative' period.
Strictly speaking, a natural period for a nonlinear system does not exist and many
'mature' techniques developed for a linear system are not applicable to a nonlinear
21

system. However, the output of commonly encountered slightly nonlinear systems can
be seen to be composed of a 'fundamental' linear part plus a nonlinear modification.
The techniques normally used for linear systems can be 'borrowed' to approximately treat
a nonlinear system in a piece-wise (incremental) form or in an average sense. Using this
analogy between linear and slightly nonlinear systems, the response period in free
vibration will be called the natural period (the influence of damping on period is of
minor importance; even a damping as high as 20% causes only a variation less than 2%
in response period) and the virtual lateral stiffness of the structure will be called simply
the structural lateral stiffness. This will be discussed further in the following section.
Representative natural periods, Tn , for each of the models obtained from the free
vibration tests in air are listed in table 4.2.
Table 4.2 Model Natural Periods

Model No.
Natural

Model I

Model II

Model II-M

As Built

Glued

As Built

As Built

1.16

1.02

1.25

1.93

Period, Tn
(sec)

2.2 Inferred Stiffness

The stiffness of each of the models can be inferred from its dynamic response if it is
considered to be a single degree of freedom system. Its global 'dynamic' stiffness, Kd ,
can be derived from the natural period obtained in the free vibration tests and the
model's equivalent mass.
On the other hand, by assuming that the connections ideally represent the original
design, the theoretical structural overall stiffness, K t , can be computed analytically using
the construction material properties as given in table 3.1.
Furthermore, the global static stiffnesses of the models, Ks , have been derived in table
4.1 from the static tests.
The stiffnesses of the models obtained from these three approaches are compared in
table 4.3; the detailed calculations of ~ and Kd are given in Appendix n.

22

Table 4.3 Stiffness Comparison

Kt (N/m)

Ks (N/m)

Theoretical

As Built

As Built

Glued

1568.1

508.0

786.57

1017.0

II

82.4

19.9

88.4

--

U-M

65.8

16.8

70.5

--

Model No.

Kd (N/m)

The inconsistency is apparent. The observed natural period in Section 4.2.1 has already
led to distrust of the theoretical design values, Kt The data in table 4.3 show two
tendencies:
1. Kd is systematically larger than Ks ; this is especially evident with models II and

U-M. This deviation indicates that the models behave more stiffly in a dynamic
situation than in a static situation. This phenomenon is primarily attributable to
the connection imperfections (or more specifically, locally concentrated damping).
As will be shown in the next section, (especially with Model II and Model H-M)
a large amount of damping is (locally) concentrated in the deck-leg connections;
relative dynamic movement between the deck and legs generates remarkable
resistance. This resistance increases with increasing relative velocities between
the deck and legs. Hence, the effect of the high damping in the connections is
analogous to a fixation against dynamic loading and thus equivalent to a large
'dynamic stiffness'. When the damping is high enough, the connection will behave
dynamically as if it were clamped. As such, the localized high damping at the
connections has significant influences not only on the overall structural damping
behavior but also on the structural natural period and thus the inferred dynamic
stiffness, K d However, this fixing mechanism exists only when the structure is
experiencing a dynamic movement. If a loading is static, the structure shows
appreciably lower stiffness, since only the stiffness of the connection counts then.
This stiffness enhancement phenomenon in the dynamic situation has also been
discovered in field measurements. The field tests done by Chiba et at. (1986)
showed that the dynamic stiffness of a jack-up platform can be 2 times its static
stiffness.
2. With Models II and U-M the average dynamic stiffness values, Kd , seem quite in
agreement with the theoretical ones, K t This, however, does not indicate the
agreement of these models with their original designs. From the discussion in
point 1, above, it is clear that the calculated dynamic stiffness, Kd , generally does
not represent the structural (static) stiffness, but an apparent (dynamic) stiffness.
23

In
this gives extra supporting evidence for the assumption that the behavior
of the deck-leg connection is close to a rigid clamping (the original design) under
dynamic loading as a consequence of localized high damping.

4.2.3 StruchmllR Dampi.ng

The structural damping of the models tested is mainly attributed to the following
damping mechanisms:
- Viscous damping
- Dry friction
- Internal material damping
- Plastic deformations
Viscous damping is the only linear damping mechanism; the rest involve a nonlinearity
indicated by their dependency upon the response amplitude. Because of the convenience
of linear viscous damping in analysis, much effort has been invested (in the literature)
in the conversion of other damping mechanisms to 'equivalent' viscous forms by
averaging the damping values over several cycles.
The damping values for each of the models are computed in Appendix II. The results
are summarized in table 4.4 where r is the structural equivalent damping coefficient, (
the structural damping ratio, defined as the structural damping coefficient, r, divided by
the critical damping coefficient, rc
= rjrJ, andA the corresponding deck displacement
amplitude.

as with the
The damping values of the Models II and U-M show strong nonlinearity
global stiffnesses; they are heavily dependent upon the structural response level. This
dependence relation is, however, rather scattered. In contrast to this, the damping values
of Model I are much lower and more consistent; it shows only a relatively slight decrease
with decreasing response amplitude levels. This consistency is expected to result from
the improved deck-leg connection.

24

Table 4.4 Slmctural Damping Ratio

Model

II

H-M

r (kg/s)

t; (%)

2.55

16.14

5.0

2.15

13.78

4.2

1.25

12.82

3.8

1.00

10.46

3.2

Average

13.37

4.1

1.65

5.99

18.8

1.10

8.54

21.2

Average

7.27

20.0

(em)

1.6

9.15

25.6

0.9

12.67

27.8

0.35

9.61

17.5

Average

10.48

23.6

More specifically, the following phenomena can be observed from the above table:
1. The damping ratios are surprisingly large especially for Model II and Model II-M.
These values are much larger than the normally found structural internal
damping. The only possible source of these high damping percentages is the
imperfect connection at both ends. The lower end was linked to the bottom by
hinges; this connection is easier to realize than the clamping at the upper end.
It

therefore, considered that the deck-leg connection is most likely the cause

responsible for the high structural damping.

2. The average damping coefficient of Model U-M seems slightly higher than that
of Model

although both models are identical except for the deck weight. This
result from extra (dry friction) damping caused by that

extra deck
the contact forces between the clamping

of the

rings - this increased

and the deck connecting plates at

the upper end as well as the contact forces in the leg bottom hinges at the lower
end.

25

4.3 SUMMARY
The important observations from the discussion of the static and free vibration tests in
air are summarized as follows:
L The behavior of Model I is quite consistent. Gluing improved the connection. The
data recorded with this model are reliable.
2. An obvious scatter in the data exists with Model II and Model II-M. The deck-leg
connections with these models are found to be different from their original designs
and highly complicated. This imperfection in the deck-leg connections results in the
dependency of structural response periods (and thus inferred structural dynamic
stiffnesses) as well as structural damping on the response level. The general tendency
is that the inferred stiffness decreases with increasing response level; this indicates
structural nonlinearities. These connections also cause a surprisingly high structural
damping.
3. The apparent dynamic stiffness is substantially larger than the static stiffness with
all models.

5 Model nonHnearities expected

The analysis of the data from the static and free vibration tests in the previous chapter
has shown that the model structures tested are highly nonlinear. The nonlinearities
originate from various sources. An inventory of the nonlinearities will provide an
overview and shed light for the analysis later in the present work. The evaluation of the
relative importance of the influences of various nonlinearities on dynamic behavior will
be performed in the following chapters after thorough data analyses and computer
simulations have been carried out.

5.1 STRUCTURAL NONUNEARITIES

The models tested mainly include the following two forms of structural nonlinearities:
- Imperfect Connections
The deck-leg connections especially with Models II and H-M were different from
their originally intended (rigid clamping) design and had a complex mechanical

26

behavior. The imperfection of the deck-leg connections resulted in the

dependency of structural natural periods (and thus structural apparent stiffnesses)
as well as structural damping on the response level. It also causes a high overall
structural damping. However, the deck-leg connections of Model I have been
glued; this model showed a quite linear structural behavior.

P-O Effect
A second-order moment will be resulted as the deck load becomes eccentric to
the vertical reaction forces during horizontal displacements - the so-called P- 0
effect. Physically, the P- 0 effect decreases the structure's stiffness and increases
its response to the hydrodynamic load. It should be noted that when the vertical
deck load is constant, the P- 0 effect does not introduce extra nonlinearities - the
lateral deflection of the structure is linearly related to the lateral loading if the
system is otherwise completely linear. The lateral deformation of the structure
is, however, nonlinearly related to the vertical load. The resultant normal forces
along the legs of the models change with the variation of the overturning
moment. This will cause nonlinearity, although its influence on the overall
structural response in the investigated case is expected to be marginal. As such,
the P-O effect now manifests itself mainly as an enhancement of the structural
flexibility (Euler amplification). The ratio of the equivalent deck weight to the
Euler critical load gives an indication about the degree of the P- 0 influence. In
fact, this ratio roughly determines the reduction of the structure's stiffness due
to the P- /; effect The P- /; reduction ratios for each of the models have been
calculated in Appendix II where they were needed to estimate the models'
theoretical stiffnesses. Here, the ratios are summarized in table 5.1. For
comparison purpose, an approximate value of the P- 0 reduction ratio for a
prototype jack-up is listed in the table as well.
Table 5.1 P-/3 Stiffness Reduction Ratio

Model No.

II

II-M

Prototype

Stiffness reduction due

8.8

20.7

36.8

10.0

to P-/3 effect (%)

This table clearly shows that the P- 0 effect is of importance in the present tests.
27

5.2 HYDRODYNAMIC NONLINEARITIES

The hydrodynamic nonlinearities stem from the waves themselves and their interactions
with the structure. The water-related nonlinearities in the present model tests include
the following four primary aspects:
- Wave Kinematics
According to the analytical criterion of validity given by Dean & LeMehaute
(1970), the waves for all three models are best described by the (nonlinear) 2nd
Order Stokes Theory. Based upon Chakrabarti's experimental results (1980),
however, the Airy Theory is still applicable (for more details, see Appendix HI).
- Free Surface Effect
Obviously, neither the local force in the splash zone nor the total resulting force
on the legs at wave crests will be the same as those at troughs. When the
contribution to the hydrodynamic load from wave motion above the still water
level (SWL) up to the instantaneous surface is counted, the total hydrodynamic
force on the structure is no longer proportional to the input wave elevation even
for otherwise completely linear situations.

Another difficulty arising from

inclusion of actual wave surface instead of constant SVVL is the correct prediction
of wave kinematics near the free surface zone when the linear wave theory is
used. The linear wave theory satisfies the governing wave field equation (the
Laplace equation), but it assumes infinitesimal wave height in the free surface
boundary. It is, therefore, natural that the predictive capacity of the linear theory
is least satisfactory in the trough to crest zone when the infinitesimal wave height
assumption is violated. Many techniques have been developed to adjust the
kinematics prediction to achieve greater accuracy in this region - further
discussion of this is given in Appendix IV.I.
Since the model legs consist of vertical elements only, any slamming effect is
expected to be negligible.
- Quadratic Drag
Drag, which is quadratically linked to the wave elevation, plays an important role
with Models nand U-M, while Model I is fairly inertia-dominated - see Appendix
HI for more details.

28

- Relative Motion
When the structure response is not negligible compared with the absolute
water flow motion, the structural motion should be taken into
consideration in the hydrodynamic force computation. Note that the
relative motion generates nonlinearity only in combination with the
nonlinear drag term. The drag force depends quadratically on the
resultant velocity in this case; a resulting 10% increase in velocity, for
example, increases the drag force by more than 20%. With model I, the
typical value of the ratio between the deck displacement and wave
elevation - which gives all indication about the ratio of the model leg
horizontal motion to the water
horizontal motion - is around 1.5
with regular wave tests
resonance) and 1.0 with
wave tests
the root mean square sense). With Models II and U-M this ratio is
around 0.3 with
wave tests
and 0.15 with irregular
wave tests
the root mean square sense). It is,
anticipated
that the relative motion will be of more
for Model I and of
less "HY-,.CHU."O-"'-" for Models nand II-M.

5.3 SUMMARY

The models tested involved both '"",'r",I,,,,",,,,.,,,.. and structural nonlinearities. The
different models have different
of nonlinearities.
Model I
includes a significant relative
a
complicated deck-leg
with an extra mass on the deck Model H-M
demonstrates the influences of the P- {; effect further, TIlis segregation of nonlinearities
with different models
isolate and thus better expose the influences of an individual
nonlinearity on the behavior of the structures.

wave test

6.1 INTRODUCTION
The
models for the structures tested will be established in this ",",0,,","'"
They will involve discrete elements and
will be carried out in the time
domain. The
HH'-'Hc.,U results from the
here
wave tests will also be
together with the
simulation results.

29

6.2 COMPUTATIONAL MODEL

The computational simulation is done using the special purpose program NOSDA. The
modeling involves two facets:
- Hydrodynamics
- Structural modeling
The special NOSDA possibilities important for the description of the above two facets
include:
Hydrodynamic interaction options:
Wave theory choice
Free surface choice
Relative or absolute velocity field
Linearized (Borgman) or quadratic drag
Structural dynamics options:
P-O element
Local damping
The discussion in this section is aimed at establishing the most complete computational
models for the structures tested. This is checked against laboratory test data in Section
6.3. Some other options or simplifications will be used in Section 6.4 to expose their
influences.
The detailed treatments of several nonlinearities together with the P- 0 effect are
collectively discussed in Appendix IV.

6.2.1 Hydrodynamics
Determination of hydrodynamic loading on the structures tested consists of two steps.
The first step is the computation of wave kinematics. This describes the motion of the
water due to waves. The second step is the calculation of the forces on the model legs,
given the water motions. These two aspects are separable here because it is assumed
that the presence of the model structures has a negligible effect on the water motions.
This assumption is justified by the fact that the model legs are widely spaced and their
diameters are less than 1/8 the wave length of interest - in other words, the latter
30

criterion allows a wave frequency of up to 1.5 Hz with Model I and 35 Hz with Models
II and II-M.
These two steps of hydrodynamic force determination are discussed respectively in the
following two subsections.

6.2.1.1 Wave Kinematics

As noted in Section 5.2, the models worked in the area where the waves are best

described by the 2nd Order Stokes Theory according to the analytical criterion of validity
while the Airy Theory is still applicable based upon Chakrabarti's experimental results.
For simplicity, the Airy Linear Wave Theory is chosen to describe flow kinematics for
all wave states used; the 2nd Order Stokes Wave Theory will also be employed with
some steeper regular wave conditions for comparison. Since the models were tested in
intermediate to deep water, the complete form of linear wave theory is used.
The linear Airy Wave Theory describes the water motion only up to the (constant
elevation) still water level (SWL). Much effort has been made in the offshore industry
to modify the linear wave theory to improve the wave kinematic prediction near the free
surface where the correct kinematic information is most essential for the offshore
structure analysis and discrepancies between different wave theories are also most
obvious. Common approaches for computing the water motion kinematics up to the
instantaneous actual wave surface include: (1) 'primitive' functional extrapolation
represented by application of the Airy wave theory almost exponentially up to the
instantaneous wave level; (2) vertical uniform extrapolation that is realized by Airy Wave
Theory up to the SWL and constant kinematics above the SWL - see Steele et al. (1988);
(3) linear extrapolation which consists of using Airy wave prediction up to the SWL then
linearly extrapolating the kinematic value of interest using the rate of change of that
kinematic quantity with respect to z at the SWL as the slope - see Rodenbusch and
Forristall (1986); and (4) stretching approach whereby the Airy kinematic profile
between seabottom and the SWL is stretched to the instantaneous wave surface - see
Wheeler (1970) and Chakrabarti (1971). More detailed mathematical formulations for
the free surface treatment are to be found in Appendix IV. All four wave kinematic
modification options as well as standard Airy Theory are included in NOSDA. Note that
besides the modification models mentioned above, a great deal of other work has been
done in attempt to improve the prediction of the kinematics near the free surface.
Among these, Forristall (1981) demonstrates that the Wheeler stretching and the linear
extrapolation provides a lower and upper bound respectively for horizontal velocities in
the crests of waves. A combination of these two approaches leads to the Delta stretching
profile - see Rodenbusch and Forristall (1986). Other schemes proposed for the free

31

surface treatment include Gudmestad model (1990), Gamma extrapolation model - see
Borgman et al. (1989), and so forth. No single modification model seems universally
superior for predicting the kinematics in the crest-trough zone for all wave fields; the
accuracy of the prediction of each approximate method depends on the wave conditions see Zhang, et at.
The present test setup was not designed to evaluate these cresttrough kinematic models (the wave kinematics were not recorded.) The waves tested
were relatively low. The choice of the crest-trough Idnematic model is, therefore, not
expected to be vital for the model behavior simulation in the present case.
The Wheeler stretching profile is adopted here as the reference case for the model
simulations.

6.2.1.2 Hydrodynamic loads

Wave forces per unit length acting on each leg, based upon the modified Morison
Equation (including relative velocities and quadratic drag), are calculated at structural
model nodes. These forces are then integrated using linear interpolation between two
adjacent nodes.
Since the water particle Idnematics and the corresponding hydrodynamic forces per unit
length were not recorded during the tests, 'actual' Morison coefficients, Cd and em,
cannot be derived. The best solution,
is to extract these values from other tests
dedicated to the determination of the hydrodynamic coefficients under similar conditions
and reported in the literature. In the present tests, the Reynolds number (Re = aD/v,
where u is the amplitude of the water particle
at the SWL, D the outer diameter
of the leg and v the fluid viscosity) ranges from 4 . 103 to 2 ' 104 with Model I, 1 . 103
to 7 . 103 with Model II and 1 . 103 to 3 . 103 with Model H-M. These Reynolds
number ranges are rather
little experimental data are available. The closest test
series so far found are those by Chakrabarti (1982) which were carried out in a wave
tank with Re varying from 2 . 104 to 3 . 104 His results, therefore, are used as a basis
for later hydrodyna~ic coefficient determination.

6.2.2 Structural Model Establishment

The model subjected to time-dependent hydrodynamic loads is discretized both spatially
and temporally to perform a numerical structural dynamic analysis.
A multiple-degree-of-freedorn Discrete Element Method (DEM) is used to discretize the
structure in space. The DEM schematizes the physical object as if it were composed of
32

a finite number of discrete, undeformable elements interconnected by massless,

deformable springs and dampers. The degrees of freedom (DOFs) are defined at the
interconnections (the nodes). Lumped masses (or, more generally inertias) correspond
with the DOFs of the modeL
The DEM schematization results in a group of (differential) equations of motion. These
equations are solved in NOSDA using a direct time integration - the Kok-y method. The
direct integration, in fact, discretizes the equations in time and turns them into a set of
algebraic equations. The responses are then obtained through matrix manipulations.

622.1 Initial Structural Computational Models

The initial computational model for each of the structures tested is established using the
building blocks available in NOSDA which are described in Appendix V. The structure
stiffness is modeled by springs and the inertia
lumped mass elements. The P- 15 effect
is included as a negative extension spring linking two nodes of an element in the
horizontal direction; the details about this type of special spring are given in Appendix
IVA. The rotational spring and dashpot can be considered to be a pair of extension
springs and dampers, respectively. An example of such nodes is illustrated in figure 6.1.
More general descriptions about structural schematization will be given in Section 622.4
after the
structural computational models are established.

Mass

Massless

rigid bar

Figure 6.1 Nodes, Elements, Springs and

Dampers in a Leg Section

33

All the internal damping coefficients along the legs as well as spring and damping
coefficients at the upper and lower ends of the legs remain undetermined in these initial
models. It is already known from the experimental data processing in Chapter 4 that the
physical models more or less deviated from their original design. Some major differences
were evident in the connections especially with Models II and II-M. These deviations
introduce a stiffness and damping uncertainty at the connections at both ends of each leg.
Additionally, the internal structural damping values along the legs and even the overall
internal structural damping ratio are also unknown, although they are expected to be
small and not to play an important role in the response analysis.

6.2.2.2 Discretization in Time

The DEM spatial discretization yields a set of ordinary differential equations of motion.
In NOSDA these equations are solved numerically using the Kok - r direct integration
method in the time domain - see Blaauwendraad and Kok (1987). In the actual
computation, the integration parameter r is chosen to be zero - see Liu and Massie
(1988). The system then works using a constant displacement field and works identically
to the Newmark - f3 method. This numerical method is unconditionally stable for a
linear system. For the present nonlinear case, the stability is not automatically assured;
its assumption is commonly considered to be reasonable, however. Luckily, divergence
of an unstable simulation is usually quite obvious.
6.2.2.3 Calibration Using Experimental Data
The unknown damping coefficients along the legs as well as spring coefficients at the
upper and lower connections in the initial models will be determined using the
information obtained from the free vibration tests in the air.
Since the free vibration data recorded are generally overall structural responses, they do
not shed much light on the detailed damping distribution within the structures. Instead,
the decays of the free vibration responses give an indication of the overall damping for
each structure. The detailed choice of the damper locations and the relative magnitude
of the damping coefficients is somewhat subjective. The internal damping ratio
(commonly not larger than 1%) can be converted to the internal damping element
coefficients in the computational model using the procedure given in Liu (1989b).
Unfortunately, even this internal damping ratio is unknown for the model materials used.
Nevertheless, it has already been assumed that the actual structural damping was largely
concentrated at the deck-leg connection; the internal structural damping and leg bottom

34

damping only play a minor role; this relative proportion is qualitatively taken into
consideration in the structural modeling.
The general approach of model calibration is to fit the simulated free vibration response
traces to the measured ones by adjusting the model damping coefficients and the
connection stiffness parameters. This is a 'try and correct' iteration process and will be
done for each of the models until the natural period and decay of the simulated response
match those of the measured response.
It has already been established from the analysis in Chapters 4 and 5 that in terms of
structural behavior Model I is reasonably linear, while Models II and II-M show
remarkable nonlinearity. It is straightforward to use simple linear rotational springs and
dashpots to model the upper connection for Model I. As for Models II and II-M, it will
be more scientifically reliable if realistic nonlinear (elasto-plastic) springs and dash-pots
are used to model the deck-leg connections. However, since very little is known about
the detailed mechanical properties of the connections for these two models, the choice
of the nonlinear springs and dampers will be too subjective. Any attempt to 'speculate'
connection nonlinearity is considered inappropriate here. Each of the three models is
modeled, therefore, using mass, linear spring and linear damping elements with an extra
group of P- 0 elements.
It should be noted that the damping and stiffness are interrelated if plasticity occurs. If

realistic elasto-plastic springs were used, hysteretic damping would be simulated under
cyclic loadings.
.
The detailed damping and connection stiffness distribution so determined is somewhat
arbitrary. For instance, two (and more) different sets of computational model
coefficients for Model II could result from the calibration as shown in figure 6.2.
The deck-leg connection with data set 1 in figure 6.2 consists of soft springs with low
stiffness and hard dampers, while in set 2 the connection springs have appreciably higher
coefficients (twice the field spring coefficient value - see Appendices V.2 and V.3, in fact,
this is the ideal clamping situation) and the dampers have lower coefficients. These two
data sets differ only in the deck-leg connection elements (as listed in the table on the left
side of the figure). The rest of the elements are identical. (For brevity their coefficients
are not shown in the figure.) Both models generate almost identical free vibration
response in terms of the decay and natural period; the only perceivable difference is that
the free vibration response trace resulting from set 1 shows somewhat more asymmetry
with respect to the time axis. This asymmetry was also observed in some of the
measured response traces, by the way. As will be shown later, these models also result
in almost the same dynamic response under wave loads. It is interesting to note the fact
35

that these two models have quite different static stiffnesses, while their apparent dynamic
stiffnesses derived from the free vibration simulation are the sameo The numerical
results are given in table 601,

Connection Modeling
Data Set SPtlng CoatI,

Damping Coello

(Nom)

(NJns)

20600

103 010 6

133100

162000

Figure 62 Two Computational Models (Mode/II)

Table 6,1 shows that data set 1 yields a static stiffness much closer to the measured
valueo Hence, this modeling set is used for the later simulation,
Table 601 Two Sets of Modeling for Model II

36

Data

Deck Connection

Static Stiffness

set

Modeling

(N/m)

Dynamic Stiffness

(N/m)

High Damping

38,0

8804

High Stiffness

79,6

8804

It should be noted that the phenomenon that the static stiffnesses are much lower than

those derived from vibration tests has also been discovered in field measurements at
several locations and with different jack-up platforms. The work done by Chiba et al.
(1986) showed that the dynamic stiffness of a jack-up platform can be 2 times its static
stiffness. Those authors attributed this discrepancy to the soil interaction. It seems
reasonable from the analysis in this section that the stiffness enhancement in the dynamic
situation could be also attributable, at least partially, to the tradeoff of local deck-to-Ieg
damping and stiffness.

6.2.2.4 Structural Computational Models (in air)

The computational model for each structure is completed using the calibration procedure
above. Note that the schematizations established so far are 'dry models'; their mass
elements will be modified to include water inertia effect when simulating stmctural
response in waves.
The 'dry' schematization for Model I is given in figure 6.3 and the associated lumped
masses, spring coefficients and damping coefficients are listed in table 6.2. Each leg of
Model I is discretized into 11 massless rigid elements connecting 12 nodes. Each of the
nodes includes a rotational spring, a rotational dashpot and a mass. Since both the
structure and flow are symmetric, the system is modeled one-dimensionally in the x
direction. The DOFs correspond to the nodes indicated as arrows in the figure. The
highest two elements of each leg are slightly longer than the rest of elements, so the
associated stiffness coefficients of the rotational springs are slightly lower. Additionally,
a negative spring is placed between two nodes of each element to represent the P- ()
effect. The contribution of the leg weight to the P- () effect is included by summing all
node weights above the investigated segment. As a result of this, the coefficients of the
P-{) spring decrease (become more negative) downwards along a leg. It should be noted
that the coefficients of these springs are so determined that they only account for the P- ()
effect due to the structure gravity; the additional dynamic axial load along the legs
induced by the wave forces are not included. Since the attention in this work is
concentrated on the global (deck) displacement, this negligence can be justified by the
fact that the stiffness lost in the leeward leg(s) is approximately compensated by the extra
stiffness gained in the windward leg(s). Inclusion of the effect of the axial load variation
would cause the P- () spring coefficients dependent on the instantaneous leg axial load;
this would introduce a nonlinearity and subsequently increase computational effort.
37

Node
Elevation
11
10

8
1

6
5
4

2
0

degree of freedom

===:> :

Figure 6.3 Schematization for Model I in Air

Table 6.2 Schematization Parametersfor Model I in Air

Node
Elevation
0
1
2
3
4
5
6
7
8
9
10
11

Foot
Restraint

38

Mass
(kg)
Leg A
1.282
0.227
0.207
0.207
0.207
0.207
0.207
0.207
0.207
0.218
0.228
7.951

Damping
(N.s/m)

Legs A, B, C

Legs A, B. C

Leg A

Legs B, C

0.0
10590.0
10590.0

29.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
120.0

-488.9

-493.1
-482.9
-472.B
-462.6
-452.5
-442.3
-432.2
-422.0
-411.9
-364.7
-354.5

Legs B, C
1.282
0.227
0.207
0.207
0.207
0.207
0.207
0.207
0.207
0.207
0.228
7.866

Spring

6 Spring
(N/m)

Spring Coeff.
(N.m)

10590.0
10590.0
10590.0
10590.0
10590.0
10590.0
10085.7
9627.3
4550.0

(N/m)

7.0 10 7

P -

-478.8
-468.6
-458.5
-448.3
-438.2
-428.0
-417.9
-407.7
-360.9
-350.7

Damping (N.s/m)
100.0

Assuming a relatively high damping at the deck-leg connection, the parameter set 1 in
figure 6.2 is used for Model n. The computational schematization is quite similar to that
of Model I, except one more leg element is used here in order to maintain a convenient
element length (Model II has a slightly different total leg length from Model I). The
computational schematization for Model II is shown in figure 6.4 and the associated
parameters are listed in table 6.3. Just as with Model I, the schematization for Model
II is also one dimensional. For simplicity, the DOFs are not indicated in the figure.
The schematization for Model II-M is almost identical to that for Model

n. Higher

deck weight requires an adjustment of the P- (3 springs as well as the mass elements at
the deck corners. The damping level is slightly higher; this can be attributed to extra
connection friction at the upper and lower ends - see Section 4.2.3. The schematization
parameters are given in table 6.4.

Node
Elevation
72

77
70

8
7

6
5
4
3

2
7
0

-0

-0

Figure 6.4 Schematizatian for Model II in Air

39

Table 6.3 Schematization Parameters/or Modelll in Air

Node
Elevation.
0
1
2
3
4
5
6
7
8
9
10
II
12

Mass

(kg)

Damping

Legs A. 8. C

Legs A. B. C

Leg A

Legs 8. C

38.7
170.0
170.0
170.0
170.0
170.0
170.0
170.0
170.0
170.0
170.0
170.0

-82.7
-7B.l
-74.0
-69.9
-65.8
-61.7
-57.5
-53.4
-49.3
-45.2
-38.0
-34.0

-86.7
-82.6
-78.5
-74.4
-70.3
-66.1
-62.0
-57.9
-53.8
-49.7
-42.2
-38.1

Leg A

Legs 8. C

0.524
0.078
0.078
0.078
0.078

0.524
0.078
0.078
0.078
0.078
0.078
0.078
0.078
0.078
0.078
0.081

0.0
715.6
715.6
715.6
715.6
715.6
715.6
715.6
715.6
715.6
689.6

0.084
0.777

665.5
205.0

0.076
0.078
0.078
0.078
0.078
0.061
0.084
0.692

Spring

Foot

(N/m)

(N.s/m)

1.3 10 6

5.0 10 .

Spring
(N/m)

(N.s/m)

Damping

"

Restraint

P -

Spring Coeff.
(N.m)

100.0

Table 6.4 Schematization Parameters/or Modelll-M in Air

Node
Elevation
0
I

2
3
4
5
6
7
8
9
10
II

12
Fool
Restraint

40

Mass

(kg)

Spring Coe!f.
(N.m)

Damping
(N.s/m)

P -

Spring
(N/m)

Leg A

Legs 8. C

Legs A. 8. C

Legs A. B. C

Leg A

Legs 8. C

0.524
0.078
0.078
0.076
0.078

0.524
0.078
0.078
0.078
0.078
0.078
0.078
0.076
0.078
0.078
0.081
0.084
1.827

0.0
715.6
715.6
715.6
715.6
715.6
715.6
715.6
715.6
715.6
689.6

67.0
170.0
170.0
170.0
170.0
170.0
170.0
170.0

-136.4
-132.3
-128.2
-124.1
-120.0
-115.6
-111.7
-107.6
-103.5
-99.4
-89.6
-85.4

-142.1
-138.0
-133.9
-129.7
-125.6
-121.5
-117.4

0.078
0.078
0.076
0.078
0.078
0.081
0.084
1.742

Spcing

(N/m)

5.0 10 4

665.5
205.0

170.0
170.0
170.0
170.0
9.B 10 6

Damping
100.0

-!i3.3
-109.2
-105.1
-93.7
-90.0

(N.s/m)

6.3 MEASURED VERSUS SIMULATED RESULTS

The hydrodynamic coefficients, Cd and Cm , are selected after considering the expected
relative movement of the model with respect to the water using Chakrabarti's (1982)
experimental results.
Model I was tested at low Keulegan-Carpenter Number (KC = aT/D, where a is the
amplitude of the water particle velocity at the SWL, T the wave period and D the outer
diameter of the leg; this parameter indicates the ratio of the water particle orbit
diameter to the structure diameter and provides a measure of the relative importance
of the drag force) with a smallKC variation range (KC = 0.7 ~ 2.1). From Chakrabarti's
results, the Cm value should be somewhere around 2.3 and Cd should be 0.5. However,
these hydrodynamic coefficients were determined for a fixed cylinder. With Model I, the
structural response is significant compared with the water motion (the ratio of the deck
displacement to wave elevation reaches 2.2 near resonance); the influence of the relative
motion on the hydrodynamic coefficients should be taken into consideration. The
flexible cylinder tests by Delft Hydraulics at De Voorst show that structure motion can
significantly increase Cd and correspondingly decrease Cm values - see Bearman (1988).
The Cd values and Cm values actually used in the simulation for all wave states with
Model I have (somewhat arbitrarily) been chosen to be 0.8 and 1.8 respectively.
The KC Numbers with Models II and U-M vary appreciably from one wave condition to
another. The KC values range from 8.0 to 24.0 with Model II and from 8.0 to 25.0 with
Model II-M. The Cm and Cd values are extracted from Chakrabarti's results for each
wave state tested (the influence of relative motion on the hydrodynamic coefficients is
considered of minor importance with these tests).
Knowing the Cm value, the computational model for Model I established in Section 6.2
is further modified to account for the water 'added mass' (about one third of the total
equivalent mass). In the modeling, this distributed mass is lumped to the corresponding
nodes and added to the nodal structure mass; this modifies the existing dry model given
in figure 6.3 and table 6.2 to a new 'wet' computation model. The schematization
remains basically the same, only the masses of the submerged nodes need to be changed.
The new parameter set is listed in table 6.5 where the modified node masses are
indicated by italic letters. These data will be used to simulate the dynamic response of
Model I in waves.

41

Table 6.5 Schematization Parameters for Model I in Water

Node
Elevation
0
1
2
3
4
5
6
7
8
9
10
11
Foot
Restraint

Spring Coe!f.
(N.m)

Mass
(kg)
Leg A Legs B, C
1.791
1.245
1.225
1.225
1.225
1.225
1.225
1.225
1.225
0.422
0.228
7.951

6 Spring
(N/m)

Legs A, B, C

Leg A

Legs B, C

0.0
10590.0
10590.0
10590.0
10590.0
10590.0
10590.0
10590.0
10590.0
10065.7
9627.3

29.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
10.0
120.0

-488.9
-478.8
-468.6
-458.5
-448.3
-438.2
-428.0
-417.9
-407.7
-360.9
-350.7

-493.1
-482.9
-472.8
-462.6
-452.5
-442.3
-432.2
-422.0
-411.9
-364.7
-354.5

4550.0

7.866

(N/m)

7.0 10 7

P -

Legs A, B, C

1.791
1.245
1.225
1.225
1.225
1.225
1.225
1.225
1.225
0.422
0.228
Spring

Damping
(N.s/m)

Damping

(N.s/m)

100.0

In contrast to the case of Model I, the water 'added mass' plays only a minor role for the
remaining two models (especially for Model H-M); it is now one order lower than the
equivalent dry structural mass. The hydrodynamic mass is, therefore, neglected; the dry
models given in tables 6.3 and 6.4 will be used as wet models for Model II and Model
H-M, respectively.
Various NOSDA options are used in the complete model simulations. The water
kinematics is calculated using linear Airy Theory. The modified Morison Equation
(including relative velocities and quadratic drag) is employed to compute hydrodynamic
forces. The stretched wave profile is adopted to include the free surface effect. Leg
shear and axial flexibilities are considered unimportant for the overall dynamic response
on which the main attention in the present simulation is concentrated and thus ignored.
In the actual computation, the iteration error tolerance is set to be 10-7 m (compared
with the magnitude of the model response at deck level of the order of 10-3 to 10-2 m).
The integration time step, Lit , is chosen to be 0.03 s to guarantee the numerical
convergence for all waves and a local truncation error - O(L1t4) = 10-7, The number of
vibration cycles needed for filtering out the transient response depends heavily on the
system damping level. With the damping data listed in table 4.4, the number of cycles
for the response amplitude to decay to 1% of its initial value is about 18 for Model I and
4 for model II( -M). Since the present study concerns the structural steady state response,
the transient response is excluded from the bookkeeping.

42

Inclusion of the free surface effect, as discussed in Appendix IV, introduces skewness to
the total hydrodynamic forces and therefore shifts the response trace from the standard
sinusoidal shape. In the following simulations, the maximum magnitudes of the
responses are taken as the steady state peak responses.
Most of the results in this work are presented via Response-Amplitude-Operator (RAO)
curves. A RAO curve for the deck displacement with regular waves is constructed here
as follows: let a series of monochromatic wave trains, with the same wave height but
each with a different wave frequency pass the structure individually; normalize the
obtained amplitude of the response displacement at deck level by half the input wave
height; plot this ratio for each wave frequency input of interest. The correct
determination of the deck displacement is vital in the offshore structural design and
assessment; the present work will concentrate mainly on this overall response parameter.
For brevity, the RAO curve for the deck displacement is often called simply 'RAO curve'
in the following text. This type of curve is a very general indication of structural
response behavior. From an analysis point of view it includes three major transformation
stages: wave surface elevation ~ water particle kinematics ~ hydrodynamic loads ~
overall structure response. Nonlinearity at any transformation stage will cause the
resulting curve to be dependent upon the input level. In other words, unlike a structural
resonance function in the usual linear sense (invariant with the input level at a
frequency), RAO curves for a system that is nonlinear (either hydrodynamically or
structurally) for varying inputs are no longer identical. This is an indicator of system
nonlinearity.
The RAO curves of Model I for three different wave heights are superimposed in
figure 6.5.
2.5 RAO (-)

Model I

1.5

0.5

0.5

0.3

0.7

0.9

1.1

1.3

1.5

1.7

Frequency (Hz)
-

:=

2 em

-+-- H

=4

em

H = 6 em

Figure 6.5 Measured RAas for Deck Displacement with

Different Wave Heights (Model 1, Regular Waves)

This figure shows that higher waves result in lower RAO values. The system thus shows
a definite nonlinearity. The deviation is especially obvious in the resonant area; this

43

leads to a hypothesis that the variation is mainly caused by the different hydrodynamic
damping level for different wave heights with this model. It is known that the
hydrodynamic damping is generated by the structural response (a fixed structure has,
obviously, no hydrodynamic damping.) This implies a need to use relative velocity in the
computational model.
The RAO curves of Model II for various wave heights are compared in figure 6.6.
O.4;.;.RA;;,:O;.,,;<...
-.:..)_ _ _ _ _ _ _ _ _ _ _- - ,

Model II
0.3

0.2

0.1

Ob-_~

0.3

0.5

__

0.7

__

__

0.9

__

1.1

1.3

__

__

1.5

1.7

Frequency (Hz)
~

H = 4 em

--+- H

=6

em

............. H = B em

Figure 6.6 Measured RAGs for Deck Displacement with

Different Wave Heights (Model II, Regular Waves)

Again, the RAO magnitude shows a definite dependency on the input wave heights. The
trend is, however, just opposite to that with Model I - the RAOs now increase with
increasing wave height and the RAO peaks shift to the left with increasing wave heights.
This dependency is probably caused by other types of nonlinearities. There are at least
two contributing effects in this case: (1) the structure's stiffness decreases with increasing
loading level and (2) the drag term (which increases quadratically with increasing wave
height) plays a more dominant role in the hydrodynamic interaction.
The RAO curves for three different wave heights have been calculated using the
complete computational model for each structure tested. Comparisons with the
corresponding measured data show a reasonable agreement. Only a representative part
of such comparisons (with a 6 cm wave height) is included here; each of figures 6.7
through 6.9 is for a different modeL The detailed analyses and results have been
reported by Liu (1989b).

44

O.4r
"'O:....('-'-l'--_ _ _ _ _ _ _ _ _ _ _--.
RA

1.8r
"'O:....('-'-l'--_ _ _ _ _ _ _ _ _ _ _--,
RA

Modell

~.H'6em

1.5
1.2
0.9

0.6

Model II
H 6 em

0.3
0.2

0.1

0.3

0L---~--~--~--~--~--~--~
0.1

0.3

0.6

0.7

t1

0.9

1.3

O~--~--~--~--~--~--~--~
0.1

1.6

0.3

0.5

Mea8ured Data

0.7

0,9

t1

1.3

t6

Frequency (Hz)

Frequency (Hz)
~

-+- Computed Data

Figure 6.7 Measured and Computed RAGs

(Modell, Regular Waves)

Measured Data

-+- Computed Data

Figure 6.8 Measured and Computed RAGs

(Model II, Regular Waves)

0.4r
"'O:....(:...:-lc--_ _ _ _ _ _ _ _ _ _ _- ,
RA

Model II - M
H' 6 em

0.3

0.2

0.1

O~--~--~-~--~--~--~--~
0.1

0.3

0.6

0.7

o.e

1.1

1.3

Frequency (Hz)
~

Measured Data

-+- Computed Data

Figure 6.9 Measured and Computed RAGs

(Model II-M, Regular Waves)

These results demonstrate that the behavior of Model I is best represented by the
computational model. The discrepancy between the computed results and measured
results with the last two models is expectable: their behavior is more nonlinear - both
structurally and hydrodynamically - and thus more complicated. But still, the simulated
results are quite acceptable.

6.4 COMPUTATIONAL EFFORT

The computational effort needed to simulate the dynamic behavior of a model generally
depends upon the following factors:
- Model size and complexity
- Incident wave frequency components
- Nonlinearity

45

When only the steady state response with regular wave simulations is of interest, the
overall damping level of the structure also influences the total computing time.
Obviously, for different input wave frequencies and different structures, the simulation
durations are quite different. It is, therefore, difficult to give a general evaluation of the
computation efforts. Nevertheless, experience with computations for this study can give
some indication of the computing time involved. A more detailed evaluation of the
computation efficiency has been given by Liu (1991b). With the present regular wave
simulation, for an excitation period near the structural fundamental natural period
(around 1.2 s), using a time step of 0.03 s (40 time steps per cycle), the DECstation 3100
Computer needs about 39 s of CPU time to simulate a clock time duration of 40 s; this
The ratio between the
gives a rough indication of the computational efficiency.
simulation time and the physical time is an efficiency of about 1:1.

6.5 FURTHER COMPUTATIONAL RESULTS

Certain simulations are carried out further to 'zoom in' on some particular modeling
features.

6.5.1 Absolute versus Relative Velocities

The absolute velocity approach is found to over-estimate the resulting peak response by
up to 50% near resonance. The need to use the relative velocity model is also
demonstrated by comparing the relative positions of RAO curves for different wave
heights. With Model I, unlike measured results given in figure 6.10 (extracted from
figure 6.5 by cutting off the higher frequencies in order to concentrate on the resonant
area), the RAO curves computed for different wave heights using absolute velocities
shown in figure 6.11 are almost identical. In contrast to this, the relative velocity model
properly simulates the variations of the RAO curves near the resonance - see figure 6.12.
This indicates that the drag term combined with relative velocity behaves somewhat like
a hydrodynamic damper in a large inertia situation; a higher wave causes a lower peak
at the RAO curve. When dealing with a fixed cylinder in the inertia dominant range, it
is commonly assumed that the drag term plays only a minor role and that therefore the
choice of Cd is not important. This is, however, not true with a flexible structure because
the drag coefficient and relative velocities now will determine the hydrodynamic damping
level in the simulation. Therefore, a correct choice of Cd is essential for the success of
such simulations of flexible structures even in the inertia dominant situations (0.7 < KC
< 2.1 as with Model I).

46

3rRA~O~(~-)~

____________________--,
Modell
Regular Waves

2.5

~--

1.5

0.5

o~~--~--~--~~--~--~~
0.5
0.6
0.7
0.8
0.9
1
1.1
t2
1.3

Frequency (Hz)
-H~2cm

--+-H"4cm

............... H"Scm

Figure 6.10 Measured RAOs for Various Wave

Heights (Modell, Regular waves)

______________________- - .

3rRA~O~(~-)

Modell
Regular Waves

2.6

3rRA~O~(~-)______________________- - ,

1.5

1.5

0.5

0.6

oL-~--~--~--~~~~--~~

0.6

0.6

0.7

0,8

0.9

1.1

1.2

Modell
Regular Waves

2.6

1.3

oL-~

0.6

__

0.6

__

0.7

-b-H"4cm

O.S

__

~~

0.9

__- L_ _- L - - J

1.1

1.2

1.3

Frequency (Hz)

Frequency (Hz)
-~H"2cm

-H"6cm

Figure 6.11 Use of Absolute Velocity Model for

Various Wave Heights

~H"2cm

--+-H"4cm

-*-H~6cm

Figure 6.12 Use of Relative Velocity Model for

Variolls Wave Heights

Neglecting the effect of structural velocity will eliminate the hydrodynamic damping.
When using an absolute velocity model, this damping is often compensated by adding an
'artificial' equivalent damping to the structural damping. However, the choice of this
damping is somewhat SUbjective. Further, it should be noted that this damping is
dependent upon the input wave and structural response level. Generally speaking, a
higher wave will cause a higher level of hydrodynamic damping. Use of relative
velocities avoids the associated guesswork at the cost of a greater computational effort.

6.5.2 Results of Linearized Model

The simulation using a linearized model is carried out by choosing the following NOSDA
options: absolute velocity, Borgman-type dr<l:g term linearization and exclusion ofthe free

47

surface effect. It should be noted that using Borgman linearization with regular
waves is not a common practice; this is done here only for comparison purpose. In fact,
a regular wave can be considered as a special case of irregular waves. The results show
that the linearization overestimates the response by about 61 % with Model I (inertia
type) and about 70% with Model n (drag type) near resonance.
Note that various nonlinear effects - neglected in the linearized approach - can have
compensating influences. For example, there are two factors increasing the response:
1. using absolute velocity rules out the hydrodynamic damping; and 2. the Borgman
Linearization applied to monochromatic waves overestimates the drag force peak by
about 12.8% - see Liu (1989b). On the other hand, leaving out the free surface effect
underestimates the total hydrodynamic exciting force to some extent.

6.5.3 Free Surface Effect!>

Five types of common mathematical treatments of the free surface effect are available
in the NOSDA software, namely, (1) standard Airy profile (integrated up to the still
water level, SWL), (2) functional extrapolation profile (exponentially extended to the
actual water surface), (3) vertical uniform profile (the kinematics being kept equal to
those at the SWL up to the wave crest), (4) linear extrapolation profile (linearly extended
to the instantaneous water surface) and (5) Wheeler stretching profile (the kinematics
at the instantaneous free surface are considered identical to those originally calculated
for the SWL) - see Appendix IV for more details. With the present (low) waves, the
difference in the results computed using different free surface treatments is found to be
negligible.

6.5.4 Hydrodynamk Cancellation

With the present model ~etup and incident wave direction, a simplified theoretical
analysis given by Liu (1989b) shows that when a wave length is twice as long as the
distance between the bow leg and the aft leg plane, the total hydrodynamic force on the
three model legs is minimum and equal to one third of what it would be if the forces on
all legs were in phase; this results in a cancellation frequency of roughly 1.15 Hz. On
the other hand, with both the measured and computed RAO curves as shown in figure
6.13, a slight 'dent' is found in the neighborhood of 1.2 Hz. This dent is more obvious
in tabulated data; this confirms the theoretical prediction.

48

Modell
H' 2 em

2.5

1.5

0.5

oL-~~~~~~==~I=~=L~
M
~
M
M
1
U
U
U
U
U
W
u

Frequency (Hz)
-

Measured Oats

-+-- Computed Data

Figure 6.13 Measured and Computed RAGs

(Model I, Regular Waves)

Theoretical (numerical) studies also show that when the incident wave direction
coincides with the line connecting two legs (30 degrees for the present case), the true
cancellation (sum of the wave forces on three legs remains zero during the entire wave
period) can be predicted at certain input wave lengths (1.5 times the leg spacing, for
example) using the linearized drag term and excluding the structure response and the
,quasi' cancellation
free surface effect; when the quadratic drag term is used there is
in which the sum of the force is minimum but not zero - see Spaargaren

6.5.5 Airy versus Stokes 2nd Order Wave Theories

According to the analytic criterion of wave theory
given by Dean (1968), all three
models work in the hydrodynamic area where Stokes' 2nd Order Theory is the most
suitable for the wave description; the Airy Theory is still applicable based upon
Chakrabarti's experimental results, however. For validation purposes, the Stokes' 2nd
Order Theory is used with Model II for a somewhat higher wave tested in the lab and
the results are compared with measurements and those obtained using the Airy Theory.
It is found that with the present waves, use of these two wave theories makes negligible
difference in terms of the resulting structural response,

6,5,6 Results of Different COl:mection Modeling

Two sets of computational models for
simplifying the deck-leg connection as
(set 1), and another as a hard spring
computed using these two models are

Model II have been presented in figure 6.1: one

a heavy damper combined with a soft spring
with a light damper
2). The RAO curves
compared in figure 6,14. The results generated

49

by these different computational models are almost identical till very low frequencies.
It seems that more than one schematization can simulate the dynamic behavior of a

physical model if only the overall dynamic response is examined.

This result also confirms the hypothesis that localized (large) damping can function like
a stiff spring (or even rigid connection) in a dynamic situation. Actually, the
phenomenon that static stiffnesses can be much lower than those derived from vibration
tests has also been discovered in field measurements at several locations and with
different jack-up platforms. This has already been discussed in Section 6.2.2.3.
0.3F'RA;..;,O;;..,;..(-.;..)_ _ _ _~_ _~_~_ _ _..,
Model II
H 6 em
Computed Results
0.2

0.1

0.2

0.4

O.S

0.8

1.2

1.4

Frequenoy (Hz)
-

Data Sell-+- Dala S.t 2

Figure 6.14 Comparison of Different Connecting Modeling

A" a matter of fact, the damping stiffness tradeoff can be demonstrated more vividly by
a simpler system with 2 degrees of freedom excited by a sinusoidal force, F - see figure
6.1Sa. The values of k2 and are kept constant for both data sets. Data set I has a
weak spring (k1) and a heavy damper ('1) betweenMJ and M 2 Data set II is constructed
by swapping the arithmetic values of k1 and '1' Theses two data sets obviously have

'2

different static behavior. They are, however, dynamically identical over a wide range of
frequencies as shown in figure 6.1Sb where the RAO values along the vertical axis are
obtained by normalizing u (the displacement amplitude of M j ) with respect to the force
amplitude, P. Note that the RAO curve here is different from structure resonance curve
which is determined by normalizing dynamic response amplitude with respect to static
response.
It should be pointed out that both the physical models tested - Model II(-M) - and the
simple system illustrated above are extreme cases. Their damping is excessively high and
locally concentrated. Further computation shows that with a lower damping concentrated
at a certain location, the tradeoff phenomenon will still occur. However, unlike the
extreme cases above, the RAO curve calculated using a high damping schematization and
that using a high stiffness schematization are often not identical while both the

so

schematizations yield the same apparent dynamic 'natural frequency' (thus the same
apparent dynamic stiffness). This indicates that a unique computational model can not
be guaranteed by calibrating its natural frequency computed against that measured alone.

Two Data Sels

Set

k,
(Nim)

(,

(N,sim)

40

10 5

II

10 5

40

k2

(2

(Nim)

(N,sim)

80

a. Structure and Parameters

RAO (cmlN)

18~~~~~-----------------------,

15
12
9

3~
O~~~~--~--~--~--~~--~---'

0.1

0.2

0.3 0.4 0.5 0.6

Frequency (Hz)
Data Set I

---+-

0.7

0.8

0.9

Data Set II

b. Result Comparison
Figure 6.15 Further lllustration of Damping Stiffness Tradeoff

6.5.7

p-~

Effect

A RAO curve for Model I simulated without the P- 0 effect is compared with the
corresponding results including this effect in figure 6.16. This figure shows that the effect
of including P- 0 is two-fold:

51

a.

b.

Firstly, it decreased the system stiffness and hence decreases the natural frequency
of the system. It can be seen from the figure that the peak of the RAO curve shifts
to the left when the P- 13 effect is included.
Secondly, an increase of peak structural response accompanies the reduction in
stiffness. Note that in spite of the structural linearity of Model I, this peak value
increment is not proportional to the reduction of the global stiffness, since a RAO
curve includes more than the structural dynamic amplification. For example, the
transformation from the wave surface elevation to the water particle kinematics is
frequency dependent; in the higher frequencies (say, f > 0.5 Hz for the present
case, approximately), with waves of the same height, the wave velocities decrease
linearly and the wave accelerations decrease quadratically with decreasing wave
frequencies. The Morison Equation transformation strengthens this trend further.
On the other hand, the cancellation effect of total hydrodynamic force would, in the
investigated frequency range, raise the peak.

2rRA~O~(~-)~__~__________~~~-,
Modell
H 4 em
Computed Resu Its

1.6

0.6
O~~~~~--~~~~~

__~-d
W
U

Frequency (Hz)
~

With p- 6

Without P-6

Figure 6.16 Influence of p-t5 Effect

6.6 SUMMARY
The regular wave test results have been presented and analyzed in this chapter.
Simulations have been carried out using the computational models established with the
NOSDA software. The work in this chapter can be summarized as follows:
1. The physical models tested in regular waves show a definite nonlinearity. With
Model I higher waves cause lower RAO values as a result of hydrodynamic damping
generated by relative motion. In contrast to this, the trend of RAO variations with
Models II and H-M is to increase with increasing input level; this dependency is

52

2.

3.

4.

5.
6.

7.

8.

attributable mainly to two factors: (1) the structure's stiffness decreases with
increasing loading level and (2) the drag excitation increases quadratically with
increasing wave heights.
The results from the NOSDA simulations which include the P-O effect and hydrodynamic nonlinearities are generally in agreement with the measured data. This
justifies the computational models used.
The computational intensity for use of NOSDA is acceptable; the ratio between the
computer time and the physical time is about 1:1 with regular wave simulations using
a DECstation 3100 computer.
Relative velocity, instead of absolute water particle velocity, is required for
simulating the behavior of a compliant structure. This allows the straightforward
modeling of hydrodynamic damping. Near resonance this (extra) damping level is
important even though the contribution of the structural velocity to the
computation of the hydrodynamic force might otherwise be of minor importance.
Drag, when combined with significant structural response, then remains important,
even at low KC Number conditions.
Using Airy Wave Theory or Stokes' 2nd Order Wave Theory makes negligible
difference for the (low wave) cases investigated.
Discrepancies between the stiffness obtained from static tests and that derived from
dynamic vibration tests have been observed both in the field (by others) and in the
present lab models. Connection damping and stiffness at the deck-leg connection
can - within certain limits - be 'traded off. Numerical investigation using NOSDA
shows that identical dynamic lateral deflection at deck level can be obtained over
a wide range of frequencies from models which differ only in the damping and
stiffness values at the deck-leg connection. Such models have quite different static
properties. Since the degree of the stiffening phenomenon is structure and sea-state
dependent, this tradeoff of damping and stiffness will need considerable additional
study.
A unique dynamic model of a jack-up rig cannot be determined by calibration with
lateral deck deflection or measured natural frequency alone. This must be
augmented by precise knowledge of deck-leg connection and spudcan behavior. An
alternative for an existing platform is to calibrate the model against recorded
internal loadings in the top and bottom connections as well.
Inclusion of the P- 0 effect is essential for the success of jack-up simulations. This
effect can be well simulated using a group of specialP-o elements (negative springs).

53

7 Irregular wave test analysis and computer simulations

7.1 INTRODUCTION
The analysis of the data obtained from the static tests and regular wave tests in the
previous chapters has shown that the model structures are rather nonlinear. The
nonlinearities originate from two sources: (1) Structural - mainly caused by imperfect
deck to leg connection; and (2) Hydrodynamic - including wave kinematics, free surface
effects and relative motion between waves and structure acting with quadratic drag.
With the relatively low waves used in the regular wave tests, the nonlinearities caused
by free surface effects and wave kinematics have proven to be of minor importance
(Chapter 6); this statement is expected to be valid for the irregular wave tests as well,
since their hydrodynamic characteristics are quite similar to those of the regular wave
tests - see Appendix III for details. As such, the models tested can be categorized into
two types according to their nonlinear properties:
a. Relative motion type: Model I belongs to this category. With this model,
the structural displacement and water particle motion are of the same
order of magnitude and relative motion is, therefore, obviously of
importance, while its structural and hydrodynamic behavior is otherwise
predominantly linear.
b. Drag and nonlinear structure type: Model II and II-M fall into this
category; their deck-leg connections have a complicated nonlinear behavior
and their hydrodynamic forces include an important contribution from drag
(due to slenderness of their legs), while relative motion only plays a minor
role (the structural response is roughly one order of magnitude lower than
the water particle displacement).
The random data from these two types of model tests will be analyzed using the RANDA
software supported by a careful error analysis. The data analysis will be carried out in
two different domains or stages:

54

- Probability Domain
This involves computing the statistical distributions of measured data at different,
separate transfer steps (wave elevation - wave kinematics - hydrodynamic loads structural response, for example). Existence of nonlinearities will cause distortions
in the statistical distributions from one step to another. For example, quadratic drag
will convert a Gaussian distribution (wave kinematics) to a Pierson-Holmes type of
distribution (Wave loads) - see Pierson and Holmes (1965) and Burrows (1979).
Consequently, the ratio of the Most Probable Maximum (MPM) force to the root
mean square (rms) force from the short-term statistics will be significantly increased;
assuming 1000 peaks which corresponds approximately to a three-hour storm, in a
pure inertia condition, this ratio is about 3.7 (Gaussian Distribution), while with a
pure drag case, this ratio is increased to 8.6 (an extreme case of Pierson-Holmes
Distribution). Other forms of nonlinearities will complicate this problem further.
Knowledge of this distortion effect and thus the resulting response distribution after
each transfer step is important for both extreme and fatigue analysis of a jack-up.
Besides, variations of the statistical distributions at different steps can be used to
detect nonlinearities. The probability domain analysis involves one time series at a
time and does not (directly) relate anyone time series to another.
The probability analysis results provide information about how the energy is
distributed among the motion levels. For example, two loading histories can contain
the same energy spectrum: one consists of a series of cycles with medium force while
another has a portion of low force and a portion of high force. These two loading
series will have obviously different probability distributions and different impact on
the structural behavior, however.
- Frequency Domain
This involves the following computations: the autospectrum of an individual measured
time seriesl the gain factor (the modulus of the frequency response function, which
is of primary interest for the jack-up analysis) and phase factor (the phase angle of
the frequency response function) between a pair of measured time series, the
associated coherence function and so forth. With a nonlinear system the gain factor
as well as other inter-step parameters will generally not be constant; instead, they
will be dependent upon the input energy level. Nonlinearities can also be exposed
(to some extent) or in other words isolated by comparing the coherence functions
between various transform steps. (The coherence is always unity for a perfectly
linear transformation.)

55

The spectral analysis results shed light on the energy distributions and their transfer
relationship as a function of frequency.
The estimates of statistical quantities either in the probability domains or the frequency
domain are inevitably accompanied by errors. There exist two kinds of errors:
bias
error which is a systematic error occurring with the same magnitude in the same direction
when measurements are repeated under identical circumstances and (2) random error
which is that portion of error that is not systematic and can occur in either direction with
different magnitudes from one measurement to another. The statistical errors (both bias
and random errors) should be estimated carefully; an irresponsible processing of random
data can cause so big an error in the results that any interpretation becomes totally
meaningless.
The analyses of the measured data in the probability domain and the frequency domain
are discussed separately in Sections 7.3 and 7.4.
In light of the insight gained from the data analysis, the dynamic behavior of the models
in the irregular waves is simulated with NOSDA using the schematizations established
in Chapter 6; the results are presented in Section 7.5.
Limited by space, only a few representative results are included in this chapter. More
detailed presentations and interpretations are to be found in a separate report by Liu
(1991a).

7.2 DATA COLLECTION AND PREPROCESSING

This section discusses the gathering and preliminary processing of the model test results
using irregular waves. These data provide the input to the statistical analyses later in this
chapter.

7.2.1 Data Recording

Twelve channels were used to record the measured signals using an analog instrumentation recorder (IR): 6 channels were used for the bottom reaction forces (x and z
components for each leg), 4 channels for the x and y components of the deck
displacement at locations A and C (see figure 3.1), 1 channel for the x direction deck
acceleration at location D, and 1 channel for the wave elevation.

56

Besides the IR recording, a UV recorder was used to record ten channels (six of them
were the same as the IR recording). The UV recording was mainly used for the on-site
visual control and for providing a first group of data for static, free vibration and regular
wave test processing as indicated in the previous chapters. The present chapter will
focus on the processing of the irregular wave data recorded on the instrumentation
recorder.

7.2.2 Data Digitalization

Before the analog data were digitized, they were low-pass filtered at 5 Hz using 12
hardware filters in order to suppress the measurement noise.
The analog signals were digitized using a Data Acquisition System (DAS). The sample
frequency of the DAS was set to be 20 Hz. The choice of 20 Hz sample frequency was
made based upon the consideration that the same digitized data could also be used for
the time domain analysis where a finer grid would be required. For the present
processing in both the probability domain and the frequency domain the sampling
frequency actually used will be 10 Hz. This means a 2nd-order decimation will be
applied to reduce the amount of data to one half after the data are digitized and
converted to the proper physical units.
The DAS system used has 14 bits (with sign); the full scale of input between -10 V and
+ 10 V was equally divided into 32766 intervals, corresponding to 32766 equally spaced
ievels.
The relative time delay phase shift due to filtering and digitalization has been checked
and proven to be negligible (the total shift from the first channel to the last channel is
less than 200 ILS.)

7.2.3 Data Prepmcessing

The measured time series are preprocessed by using a high-pass (numerical) filter.
waves longer than half of an individual record segment (defined in Section 7.4)
filtered out. Attention in the present study is focused on the vicinity of resonance;
frequency secondary waves are expected to be unimportant for the response of

All
are
low
the

57

structures tested. All the series have already been analogously filtered at 5 Hz low pass.
This choice of upper cutoff frequency leaves possible 3rd harmonics (generally lower
than 3 Hz) intact.
The filtering is carried out in the frequency domain; this corresponds to multiplying the
Fourier Transform of the data record by the frequency response function of the desired
filter and then taking the inverse transform. The software used to do the FFT does not
require the number of the input data be an exact power of 2. The author's experience,
however, shows that the quality of filtering increases significantly when this number is
a power of 2. Therefore, the time history of each run is divided into two sections: one
contains 16384 (= 214) data points (= 819.2 s) and the other contains 4096 (=212) data
points (= 204.8 s). After filtering, the two sections are merged together again.
The digitized data so far obtained do not represent physical units. A unit conversion
procedure is applied for each individual channel of each individual run to make the data
physically meaningful.
After the conversion, a 2nd-order decimation is employed to all time series to reduce the
amount of the data to half; the 20480 data points of each channel resulting from the
conversion are cut down to 10240 points (= 1024 s). The decimated series (10 Hz
sampling frequency) will be used as the input data for the statistical and frequency
analyses in the following sections.

7.3 PROBABILITY ANALYSIS RESULTS

The traditional approach to this problem emphasizes the comparison between theoretical
distributions and actual distributions derived from each measured time series: testing the
normality of the instantaneous values, comparing the distribution of the measured peak
values with the Rice distribution, verifying whether the distribution of the extreme values
is of Poison type, and so forth. Many researchers have dedicated considerable effort to
this type of analysis; much valuable information is already available - see Anon. (1983)
and Battjes and van Heteren (1983), for example. A preliminary check following this
traditional line has been performed using the observed data. Both the direct frequency
histogram comparison and the Kolmogorov-Smirnov (or K-S) test (the most generally
accepted test for continuous data - see, for example, Press et al. (1986)) involving even
the very first 'primitive' group of data - instantaneous wave surface elevation - showed

58

deviations from the expected theoretical Gaussian Distribution. Further processing along
this line was not expected to lead to any new or conclusive results. Since the main
objective of the present work is to investigate the influence of the nonlinearities involved,
statistical analysis here will focus instead on the distortions in the statistical distributions
from one step to another caused by the existence of nonlinearities.
Since most of the possible skewness has been excluded by high-pass filtering, the crests
and troughs are not distinguished in this analysis. It should be noted that skewness (or
asymmetry) could result from both the (true) physical process (such as structural
plasticity, dry friction, secondary waves, free surface effect, etc.) and the (false)
instrumentation shift. Apparent instrumentation shift was observed in the time series
record. Since it is difficult to differentiate this shift from the realistic physical
asymmetry, the whole skewness is indiscriminately excluded from the time record by the
high-pass filtering. Consequently, this could eliminate some effects - especially on the
response statistical distributions caused by nonlinearities.
The distortion in the probability distributions (with the exception of the mean shift which
is ruled out by the high-pass filtering and data normalization) caused by various
nonlinearities is demonstrated by comparing the curves of the chance of exceedance for
two different quantities (the wave elevation versus side sway or the side sway versus
bottom reaction, for example).
Higher order harmonics in a response introduced by nonlinearities are the primary cause
of its statistical distribution being different from that of its input. Quadratic drag
introduces higher order wave force components. In the present tests, the natural
frequency of the structure is close to the wave peak frequency (see tables 3.1 and 3.3);
the first order effect is dominant in the response while higher order terms are suppressed
(filtered out) to some extent in the response. Therefore, the response often tends to be
more Gaussian-like than the hydrodynamic force excitation. The distortions of
probability distributions found between the wave surface and structure-related quantities
(such as the deck displacement and bottom reactions) are a net effect of physical
nonlinearities counteracted by dynamic amplification filtering.
An implication of this phenomenon is that a linear looking overall system can contain
significant internal nonlinearities. This is also discussed by Massie, Liu and Zeelenburg
(1991) from another angle.

59

The choice of interval between two succeeding histogram steps or levels is a compromise
between bias suppression and random error suppression. A large interval is desirable
to reduce the random error, while a small interval is needed to suppress the bias error.
This interval is selected here to minimize total error of estimates. With the parameters
chosen, the normalized bias error associated with (cumulative) probability distribution
estimates is restricted to less than 1% and normalized random errors are limited to less
than 5% with all models. This lends confidence to the results obtained from the present
frequency analysis.

7.3.1 Relative Motion Type

When relative motion combined with quadratic drag is the only important nonlinearity
involved (Model I), neither the comparisons of chance of exceedance between the water
related quantity and structure-related quantities (wave elevation versus deck
displacement and wave elevation versus bottom reaction forces), nor those among the
inter-structural quantities (deck displacement versus bottom forces, vertical force versus
horizontal force) show noticeable difference - see figures 7.1 and 7.2.

Model I
Run 55

607.
407.
207.
07.+-~~~--~--~~---F~~~FF~~~

0.3

0.6

0.9

1.2

1.5

1.8

2.1

2.4

2.7

Normalized Values (-)

-

Wave Elevation

Deck Displacement

Figure 7.1 Chance of Exceedance: Wave Elevation and

Horizontal Force

60

100% Chance of Exceedance

(%)
Model I

Run 55

80%
60%
40%

0.3

0.6

0.9

1.2

1.5

1.8

2.1

2.4

2.7

Normalized Value (-)

-

Deck Displacement

--+- Vertical Force

Figure 7.2 Chance of Exceedance: Deck Displacement

and Vertical Force

These comparisons indicate that relative motion does not have a significant impact on
the response probability distribution.

7.3.2 Drag and Nonlinear Structure Type

When drag and structural nonlinearity are important, the comparisons between the
chance of exceedance of the wave elevation and those of the response show a clear
deviation. A typical example of this is given in figure 7.3. This figure shows that when
compared with the chance of exceedance of the wave surface elevation, the deck
displacement response chance of exceedance drops more rapidly at the lower range, then
slows down gradually and at a certain point becomes higher. (This means that there are
more extreme response data than corresponding excitation data.) If the whole physical
process (the structure standing in waves) were seen as a filter, the function of this filter
would be to stretch an input (wave elevation distribution) to a more extreme response
(structural displacement or bottom force distribution). The transfer within the structure
itself (side sway to bottom force, for example) also distorts the distribution; the degree
of this distortion was found to be less profound than that from the water surface
elevation to any structural response quantity, however. It seems that the stretching effect
is primarily caused by the hydrodynamic drag while the structural nonlinearity
(complicated deck-leg connections) plays a less significant role.

61

Model !l
Run 141

0.6

1.6

2.5

3.5

4.5

Normalized Value (-)

-+-

- - Wave Elevation

Deck Displacement

Figure 7.3 Chance of Exceedance: Wave Elevation and Deck

Displacement

The influence of the p- /j effect on the probability distribution is examined by comparing

the curves of chance of exceedance of inter-structural quantities calculated for Model
II-M. Although this model has an exaggerated P-O effect (the ratio of the equivalent
deck weight to the Euler critical load is 36.8% with this model), the distributions of the
measured deck displacement and reaction force are still quite similar - see figure 7.4; the
influence of the P-O effect on the response probability distribution is marginal. This
indicates that the influence of the P- 0 effect on the overall dynamic behavior of the
model is basically linear and the nonlinear contribution of this effect caused by the
varying axial forces along the legs is negligible.

100%

Chance of Exceedance (%)

Model I!-M
Run 210

80ll

60:\

40%

1.5

0.6

2.5

3.5

4.5

Normalized Value (-)

-

Deok Dlaplacement.

-+-

Vertical Force

Figure 7.4 Chance of Exceedance: Deck Displacement and

Vertical Force

62

7.4 SPECTRAL ANALYSIS RESULTS

It should be pointed out first that the spectral method theory was originally developed

for analyzing a constant-parameter linear system. With the system under investigation
the constant-parameter assumption is valid while the linearity assumption is apparently
violated. However, the application of this approach to determine system cross
characteristics (coherence function, frequency response function and thus gain factor and
phase factor, for example) will produce the best linear approximation (in the least square
sense) of those characteristics associated with the specific input and output conditions. For
different inputs, the frequency response functions so determined are generally different.
It is also worthwhile to note that recent developments in spectral analysis techniques

make it possible to identify a nonlinear system in more detail provided the nonlinearities
are well formulated in principle. The basis of the more sophisticated spectral approaches
is to decompose a nonlinear system into linear, bilinear and trilinear parts. In turn, the
bilinear part is modeled as a zero-memory squarer followed or preceded by a linear
operation with finite-memory and the trilinear part as a zero-memory cuber followed or
preceded by a linear operation with finite-memory. For example, the hydrodynamic wave
forces on a fixed small diameter cylinder are first split into inertia and drag parts; the
inertia part is treated by a linear operation and drag part is replaced by the sum of a
linear operation plus a cubic operation and this sum is again put through a linear finitememory operation - see Bendat (1990) for more details. The application of these new
techniques involves much more computational work and demands precise knowledge and
realistic mathematical formulations of the nonlinear physical processes that are far from
well known in the present case. The attention in this work, therefore, is aimed at
qualitative identification of nonlinearities and their influence on the dynamic behavior
of the structure by employing the more mature 'linear' spectral technique.
The time series have been preprocessed as described as in Section 7.2. Additional
preparations of the data are necessary for the frequency analysis. These preparations
include three steps: segmenting, overlapping and windowing. All of them are carried out
to improve the accuracy of the resulting estimates. This is only briefly recapitulated
here; for more details, see Liu (1991b).

63

In order to obtain smooth spectral estimates, each time record is divided into segments.
The choice of the number of data segments in the spectral analysis is critical to the
overall error of the results especially when the spectra concerned are narrow-banded.
Random error increases and bias error decreases with a decreasing number of segments
in a fixed total record length. The number of segments is chosen here to minimize the
total error. The number of segments actually used is 20 for Model I and 40 for Models
II and H-M. The frequency resolution bandwidth resulting from this segmentation
guarantees that there are at least ten grid points within the energy-rich range of
frequencies, while the degree of smoothing is nearly optimal as well.
The Hanning window is employed to taper the time series. 50% overlapping is used to
improve the accuracy of estimates as well as to compensate for the information loss due
to windowing. Accordingly, the equivalent number of segments after overlapping is
increased to 32 and 64 for Model I and Model H(-M) - see Press, et ai. (1986).
The computation principles used in RANDA generally follow the line given by Bendat
and Piersol (1971 and 1986) and will not be extensively discussed here.
The computations involve estimates of autospectra and joint record spectral functions.
The term 'joint record spectral functions' refers to the coherence function, the frequency
response function and thus the gain factor as well as the phase factor; these all link one
time series to another.
Interpretation of the results obtained in the following frequency analysis focuses on
exposing nonlinear influences. These show up most prominently in joint record
functions. Note that bias error suppression with joint record function estimates reduces
only that portion caused externally due to either the computation procedure or
instrumentation. The bias error caused by nonlinearities is inherent in the system being
investigated and, in fact, is the phenomenon being sought; this bias gives an indication
of the influences of various nonlinearities - see also Liu, et al. (1991).
Besides the normalized bias and random errors, a 95% confidence interval is also
computed for each spectral estimate to give a vivid illustration of the scope of likely true
values.
A general tendency common with all models and all runs is that the response spectra are
systematically narrower than those of their excitation. This signal filtering effect can be

64

physically explained by the structural dynamic amplification. Note this phenomenon is

not universally true; when the dynamic amplification causes a twin-peaked response
spectrum, the spectral width parameter (ern) defined by spectral moments (m;) can well
be wider than that of input spectrum. Additionally, the waves generated in these tests
were relatively narrow-banded; this was especially true with the Model I tests. It should
be emphasized that it is the narrowness of the spectra that makes the present frequency
analysis extra difficult; bias suppression requires such a high bandwidth resolution that
one has very little room left for random error suppression; an optimum balance is vital
for success. Fortunately, spectra measured in a real sea are generally wider; error
suppression is expected to be less critical with prototypes.

7.4.1 Relative Motion Type

The coherence between the waves and the deck displacement as well as waves and the
bottom reaction force with Model I is rather high (up to 0.98) in the vicinity of the peak
frequency of the input waves - see figure 7.5. The notation system is chosen as follows:
the horizontal-axis is the frequency, f ; the solid curve is the value of interest (the
coherence function in this case), encompassed by the 95% confidence interval (shown
in the figure as the two fine dashed curves); and the coarse dashed curve down at the
bottom of the figure is the normalized random error as a percentage. In the figure (-)
denotes that the quantity is dimensionless and er is the normalized random error. Note
that the results are plotted only in the range where the spectral values are significant.
The generally high coherence values in figure 7.5 show that the influence of nonlinearity
(here primarily relative motion) on the structure dynamic behavior is generally small.
On the other hand, the coherence has a dip in the neighborhood of resonance near f =
0.87 Hz.

This indicates that relative motion has a more profound influence near

resonance. A logical explanation for this is that the nonlinearity caused by relative
motion manifests itself as damping which is most apparent only near resonance. Since
in the present case the input energy level at true resonance is relatively low, the impact
of this relative motion damping on the overall dynamic behavior is expected to be less
significant. Figure 7.5 also shows that the 95% confidence interval is narrow. The
normalized random error for the coherence estimate is less than 5% in the energy-rich
range of frequencies; this value is also representative for other joint record estimates
between the wave elevation and structure-related quantities for Model I.

65

0.8
0.6
0.4
0.2

0.4

0.6

0.6

0.7

0.8

0.9

1.1

1.2

1.3

1.4

1.6

Frequency (Hz)
95% ConH. luten-al

- - Coherenoe

Normal. Random Error

Model J (Run 61)

Figure 7.5 Coherence Between Wave Elevation and Deck

Displacement

The coherence between inter-structural quantities is even higher; an example is given in

figure 7.6. This perfect coherence should be expected since the structure is reasonably
linear (The deck to leg connection of Model I was glued). This figure also shows that
the normalized random errors associated with this estimate is rather low (typically within
1%); this error range is also representative for other joint record estimates of interstructural quantities. Although the record segmentation with this model is relatively
coarse (to suppress bias error), the final random errors of the joint record estimates are
still low; this comes from the high coherence.

0.6
0.6

0.'

40%

0.2

20%

0.4

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

1..

1.5

Frequency (Hz)
95% Conti. Interval

Cohe!r"ence

Normal. Random Error

Model I (Run 61)

Figure 7.6 Coherence between Deck Displacement and Horizontal

Force

66

RAO for deck displacement with irregular waves is defined as the gain factor between
the wave elevation and the deck displacement. (A gain factor is the modulus of the
frequency response function which is determined here as the cross-spectrum between the
input and output divided by the input spectrum.) Superimposing the RAOs for three
most representative wave heights (Run 58 with the lowest significant wave height Hs =
2.93 cm, Run 56 with the highest Hs = 4.44 ern and Run 59 with the middle value Hs =
3.49 cm) yields figure 7.7. It shows that irregular waves with different significant heights
result in different RAOs - especially near resonance. This deviation is not as obvious
as with regular waves shown in figure 6.5 which is repeated here for better comparison.
This disparity can be explained by the fact that a sinusoidal wave with a frequency
coincident with the resonant frequency will generate a larger structure response than
irregular waves. The relatively lower irregular wave structural response compared to the
wave elevation causes a more modest relative motion effect as well. Even so, the
general tendency that a higher wave causes a higher level of hydrodynamic damping
(thus a lower RAO) remains valid with irregular waves; it is less apparent, however.
Just as with the regular wave tests in figure 6.5, a slight 'dent' can also be observed in
the neighborhood of 1.2 Hz with the irregular wave tests of figure 7.7 - this dent is
clearer in tabulated data. This effect is caused by hydrodynamic cancellation.
The nonlinearity caused by relative motion seems not to have a significant impact on the
average magnitude of the RAOs in irregular waves compared with that in regular waves.
Since a different input level will cause a different RAO curve, a comparison between
results with regular versus irregular waves should be done on a comparable wave height
basis. However, the definition of an irregular wave comparable with a regular wave is
inevitably subjective. There are two simple approaches in use: (1) Assume that the
significant wave height of irregular waves equals the wave height of a regular wave; (2)
The energy contained in the irregular waves is the same as that contained in the regular
waves, (in other worps, their standard deviations are identical). The first approach
provides an irregular wave height that is visually about 'equal' to that of the regular

67

Model!

0.3

0.5

0.7

0.9

L1

!.5

1.3

1.7

Frequency (Hz)
Run 52 (!is;::; 2.9 em)
4-

Run 56 (Hg

=:;;

--I--

Run 59 (H".

3.5 em)

4.4 em)

Fig;.1re 7.7 RADs for Different Wave Heights

(Model 1, In'egular Waves)

Model I

0.5

t~~~:::::::::;:::::~

0.3

0.5

0.7

0.9

1.1

1.3

1.5

1.7

Frequency

Figure 6.5 IUDs for Different Wave Heights

(Modell, Regular Waves)

wave; the

wave will contain about 12 times as much energy as its irregular

The second definition

the conservation of energy, while it

underestimates the contribution from

waves. It seems to the author that the most
reasonable definition should be somewhere III between these. For example, the
the results from each of these
definitions as
deviation of the wave surface. A.n

is

68

in

7.8.

vh.'UUf+nv

'" 0.85 Hs ~ where a'l is the standard

using this approach

of the irregular waves of

2.93 cm, then the corresponding

vvave
'closest' regular waves recorded are those with
that the RAOs computed from
and

2.5

be around 2.5 cm, the

of 2 em. T11is
demonstrates
waves can be
similar.

RAO (-)

1.5

0.5

OA.

0.5

0.6

0.'"

0.6

0.9

.i.

Ll

1.2

loS

lA,

1..5

Frequency (Hz)

Figure 7.8 Comparison of RAOs: Regular

(Modell)

Waves

7.4.2 Drag and Nonlinear Structure

Models II and U-M are
nonlinear deck to leg connection. The wave

dorninated

have a somewhat

and the RAOs with these models

are noticeably wider than those with Modell. The response

is less criticaL TIle bandwidth

too. Consequently, bias

these models is twice as wide as that used with Model I;

number of equivalent record

sel~ments

used for

means that a

with 32 with IVlodel I) are available

for estimate smoothing.

The RAOs for various wave
shows a definite
increase with increasing

'The

RAOs

wave heights; this is

wave tests - see figure 6.6
There are
structure's stiffness decreases
increases quadratically with

69

Model II
0.3

0.2

0.1

0.5

0.7

0.9

1.1

1.3

1.6

1.7

Frequency (Hz)
-+- Run

RUIll 143 (H!!l ;;; 2.3cm)

~

1~9

(Bs ;; 502cm)

Run 132 (Us;;; a.Scm)

Figure 7.9 RAOs for Different Wave Heights

(Model II, Irregular Waves)

Model II
0.3

0.2
0.1

06-__- b____
0.5

0.3

~~~~

0.7

__

0.9

&_~~~~~

1.1

1.3

_ __ d

1.5

1.7

Frequency (Hz)
-

~ 4,

em

-+- II

=6

em

.....;>-

II = 8 em

Figure 6.6 RAOs for Different Wave Heights

(Model II, Regular Waves)

Once more, cancellation is observed near 1.2 Hz in figure 7.9 (The regular wave tests did
not reach or pass this frequency with these models).
Furthermore, the RAOs also seem dependent of the input energy distribution as a
function of frequency. This is shown in figure 7.10. The significant wave height of Run
141 (Hs = 3.22 cm) is approximately equal to that of Run 145 (Hs = 3.39 cm), while Run
145 has a higher peak frequency (!p = 0.82 Hz) than Run 141 (Jp = 0.74 Hz). It can be
seen from this figure that Run 145 yields a higher RAO peak.
70

0.2

r-RA;.;;..O,--,-(--<)_ _ _--:_ _ _ _ _ _ _ _ _ _-,

Model II
0.15

0.1

0.05

oL-~

0.4

___ b_ _d -_ _

0.5

0.6

0.7

~_d

O.B

__

0,9

__

I.

1.1

__

~~

1.2

__

1.3

__

1.4

1..5

Frequency (Hz)
- - fp = 0.74 Hz

-+- fp == 0.82 Hz

Figure 7.10 RAGs for Different Peak Frequencies

The magnitude of the RAOs computed from irregular waves are found to be lower than
those computed from regular waves. For example, the significant wave height of Run
149 is 5.20 cm, and its comparable wave height is about 4.4 em. The closest wave height
in the regular wave tests is 4 cm; superimposing the two associated RAO curves yields
figure 7.11.

0.3 rR,;;.AO;....:.,,(-,.:.)_ _ _ _ _ _ _ _ _ _ _ _ _.....,

Model II

+
0.25
0.2
0.15
0.1

0.05

0.4

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

1.4

1.5

Frequency (Hz)
-

Irregular (Hs=5.2om)

-+-

Regular (H=4cm)

Figure 7.11 Comparison of RAGs: Regular and Irregular Waves

(Model II)

The possible reasons for this RAO reduction in irregular waves are:

71

a.

Inherent bias of the spectral analysis technique. The response will be

partly decomposed from the peak frequency to higher frequencies.
Assume x = sin 2nft and y =

I sin

2nft

I sin

2nft, (in fact, this is a

simplified pure drag case), then the gain factor computed from the time
domain is obviously 1, while the gain factor computed using the frequency
analysis technique is only about 0.85 at f. As such, the reduction in RAOs
implies the importance of the hydrodynamic drag term.
b.

A sinusoidal wave with a frequency identical to the natural frequency of

the structure will excite more response than comparable irregular waves.
This is caused by the fact that the wave components with different
frequencies in the irregular waves tend to cancel each other and the
structural dynamic amplification unequally enhances the response along
the frequency axis (only that portion of the response due to the irregular
wave components with frequencies near the structural natural frequency
is so strongly amplified as the regular wave counterpart). With the present
models, the structural stiffness decreases with increasing displacement.
The decrease in the stiffness will be fed back and show up as a higher
RAO value.

The coherence functions between the water-related quantity and structure-related

quantities are found to be lower than those with Model I and less than 0.9. As an
example, the coherence function between the water surface elevation and the deck
displacement as well as the coherence function between the water surface elevation and
bottom horizontal reaction force are plotted in figure 7.12. On the other hand, the
coherence functions between structure-related quantities remain close to unity - see
figure 7.13. All these indicate that the water related nonlinearities and wave-structure
interaction have a major impact on the coherence. Hydrodynamics is the dominant
nonlinearity, while the structural nonlinearity (mainly due to the imperfection
of the deck-leg connection) apparently plays a less significant role.

72

Coherence between Wave Elevation and

Horizontal Bottom Force

Coherence bewteen Wave Elevation and

Deck Displacement

: r-_

TC~o=h=er~e~n=c=.~(~-~)__________________~c.~(~~ 100%

oo

::;::::::::::::::::::::::::::::::::::::::::\""f:-

",--~:--~.,.,.//./

OAr

0,2
0

::::::: ... ,

0.4

\\.\

0.6

0.8

0.7

0.8

0.9

1.1

1.2

,.

1.3

1.4

,~/\, ,A, .,/,.,,:>::::::::~::::::::::.::::"""'''''';:::~''''''

v

.'

.' ,

0,2

..

~~

0.4

1.6

0.6

0.03

96% ConiL

l~hHFval

Nonn$l. Random

<0%
20%

... ...... ----------------------------

0.7

O.S

0.9

1.1

1.2

1.3

1.4

0%

1.6

Frequency (Hz)

Frequency (Hz)
......

80%

:::::

v \././
.x :'
, i..

40%

~,'- .. _________ .. --- _____________ ~/); 02:n

:::

9S% Con.ti. [nwrva!

CohE:lrElnoa

Cohersnc<8

Normal. Random Errol!"

ErI!"Ol"

Figure 7.12 Coherence between Water-Related Quantity and Structure-Related Quantities

Coherence bet",,,,,n Deck Displacement and

Vertical Bottom Force

Coherence 'bewleen Deck DisplacemGnt and

Horizontal BoHom Force

'l::

':1~OO(-'

o,a

Q,S

0.4

0.4

0,2

0,2

o --0.4

o~

0.6

0.18

0.7

O.S

0.9

"

1.1,

1.2

Frequency (Hz)
96% Cont!. Ini.@rvaR
Norma!.

Rc.n~om

Elrro!r'

Jl.3

1.4

1.6

0.4

0.6

0.03 0.7

0.8

0,9

1.1

1.2

1.3

1.4

1.6

Frequency (Hz)

ConeNlllce
Normal. Ran.dom Error

Figure 7.13 Coherence between Inter-Structural Quantities

It can also be seen from figures 7.12 and 7.13 that the random errors are rather low (restricted

to less than 5% in the energetic area) associated with the joint record estimates between the
wave elevation and structure-related quantities and even lower (less than 1%) between interstructure quantities. These error analysis results indicate the reliability of the joint record
estimates.

73

7.5 MEASURED VERSUS SIMULATED RESULTS

The analyses in the previous sections are factual; they focused on gaining insight into the
dynamic behavior of the models by examining the recorded data from the tests. The
present section deals with the NOSDA simulations of the model behavior using the
schematizations established in Chapter 6.
The autospectra and probability distribution of the input wave surface elevation and the
corresponding responses have been obtained earlier in this chapter. The quality of the
NOSDA irregular wave simulation will be checked by comparing the spectra and

probability distribution of the response time series simulated using the measured wave
spectra as the input with those of the measured response time series.
The irregular wave surface profile is reproduced using wave superposition (also known
as Random Phase Theory). The phase information lost in the spectrum representation
is compensated by supplying a group of randomly generated 'artificial' phases from a
uniform distribution in the range (0, 211:).

The amplitude of each wavelet follows

(deterministically) from the wave spectrum. It should be noted that the wave surface
(and thus the kinematics) reproduced using deterministic amplitude (also called
constrained wave simulation) does not strictly satisfy the condition of a Gaussian process
unless the number of wave components approaches infinity. An alternative scheme is
to generate Rayleigh random amplitudes combined with uniform random phases - see
Tucker, et al. (1984). An important limitation of the constrained model is that it may
incorrectly reproduce wave group statistics - or the 'groupiness' of the waves which can
have a profound effect on ships, moored structures, etc. However, the models tested in
the present study are relatively stiff and thus not sensitive to such low frequency wave
excitation, therefore. The deviation from the Gaussian distribution caused by the
constrained wave reproduction scheme is expected to be unimportant for the present
model simulation. In fact, the wave surface measured in the present tests is not strictly
Gaussian, either. An additional advantage for using the constrained wave reproduction
model is that it guarantees a stricter conservation of the total input wave energy.
The spectrum and probability distribution of the wave surface so reproduced are checked
with those of the wave surface measured (the target spectrum and probability
distribution). The comparison is satisfactory.

74

The wave kinematics are predicted using linear wave theory (summing the contributions
from all wave components)o The validity of such a linear wave model for kinematics
prediction in unidirectional irregular waves has been confirmed in the MaTS
investigation (the Netherlands program for Marine Technological Research) - see Anono

(1983)0
Note that the wave surface and the corresponding wave kinematics so simulated will
repeat themselves after a time segment, Ts
used in discretizing the wave spectrum)
random phases after each Ts

llBe (where Be is the frequency resolution

This repetition is avoided by regenerating

Just as with the regular wave simulations, the free surface effect on the wave kinematics
is included using the Wheeler stretching approacho
Given the (resultant) velocity and acceleration, the hydrodynamic load is computed using
the modified Morison Equationo The extension of the Morison Equation to irregular
waves has been validated in a project jointly performed by SIPM (Shell International
Petroleum Maatschappij) and MaTS - see Vugts and Bouquet (1985)0
As discussed in Chapter 5 and Section 704,2, irregular waves will excite less response than

a comparable sinusoidal wave with a frequency identical to the natural frequency of the
structureo With Model I, the RAG value - which gives an indication about the ratio
between the model leg motion to the water particle motion - is up to 200 with regular
wave tests (near resonance), while the typical value of the ratio between the root mean
square deck displacement and rms wave elevation is around LO with irregular wave testso
Therefore, the influences of the structural motion on the hydrodynamic coefficients are
expected to be less significant with the irregular wave tests than with the regular wave
testso In light of this, the Cd and Cm coefficients for the irregular wave simulations are
chosen to be 007 and 200, respectively; these are closer to those given by Chakrabarti
(1986) for a fixed cylinder (Cd = 005, em = 23), compared with 008 and L8 with the
regular wave simulationso With Model I simulations, 25 harmonics (005 to 1 Hz with a
resolution of 0002 Hz) are used to reproduce the irregular wave profile and kinematicso
The spectrum and chance of exceedance of the simulated deck displacement are
compared with those of the measured deck displacement in figures 7014 and 70150

75

Model I
Run 56

Ha = 4.44 em

1\

30

20
10

0.4

0.5

0.0

0.7

0.6

0.9

1.a

1.2

1.3

1.4

1.6

r,equency (Hz)

Figure 7.14 Measured and Computed Response Spectra

(Mode! l, Deck Displacement)

100r. Chance of Exceedance (%)

Model I
Run 56

Bar.
sor.
40%

20%

O%+-~~~--~~~~~--~~~~~~~~

0.3

0.6

0.9

1.2

1.5

1.6

2.1

2.4

2.7

Normalized Displacement (-)

- - Computed

-+- Meesured

Figure 7.15 Chance of Exceedance: Measured and Computed

Deck Displacement for Model I

The results from other models (II and II-M) are presented here before conclusions are
drawn. These models are of the drag and structurally nonlinear type; the relative motion
plays only a minor role. The hydrodynamic coefficients for these models are extracted
from Chakrabarti's results (1986). With these models, 27 sinusoidal waves (0.45 to 1.5
Hz with an interval of 0.04 Hz) are used for irregular wave representation. A
comparison of the computed and simulated deck response is given in figure 7.16 and 7.17
in terms of the spectrum and chance of exceedance.

76

Modell!
Run 151
Ha = 5.85 em

0.5
0.4
0.3
0.2

0.1

0.4

0.5

0.6

0.7

0.8

0.9

1.1

1.2

1.3

1.4

1.5

Frequency (Hz)
- - Measured

--t-

Computed

Figure 7.16 Measured and Computed Response Spectra

(Model II, Deck Displacement)
100%

Chance of Exceedance (%)

Model II
Run 151

80%
60%

40%

20%

O%+-~~~~~~~~~~~~~W+~*H~J

0.5

1.5

2.5

3.5

4.5

Normalized Displacement (-)

- - Computed

-+-- Meaauwed

Figure 7.17 Chance of Exceedance: Measured and Computed

Deck Displacement for Model II

The conclusion for all of these models is that the probability and spectral properties of
the response from the NOSDA simulation match wen with those from the physical
models with statistically equivalent input. This is true for the relative motion type as
well as the drag plus nonlinear structure type models; this validates the NOSDA
simulation in irregular waves.
The NOSDA simulation with irregular waves is obviously more time consuming than with
regular waves. Since more waves (instead of one single wave) are superposed to
calculate the instantaneous wave surface and wave kinematics, more computing time is
needed in the hydrodynamic part. More specifically, in the present study, the time step
is chosen to be 0.05 s and the wave peak frequency is around 0.8 Hz. Therefore, there
are about 25 data points per primary cycle. Using the same structural models as used
with regular wave simulations and 25 to 27 waves representing spectra, a simulation of
1034 s clock time uses around 8000 s of CPU time on the DECstation 3100. The ratio

77

of the simulation time to the physical time is about 8:1, which is roughly 8 times as costly
as compared with the corresponding regular wave simulation.

7.6 SUMMARY
The experimental data from three principle jack-up models under irregular waves have
been processed and analyzed in a responsible way. The data were examined both in the
probability domain and frequency domain using RANDA. The results increase the
insight about the behavior of such rigs at a random sea. Furthermore, the model
behavior was well simulated using NOSDA. More specifically:

7.6.1 Data Analysis

Probability Domain
1. Relative motion does not noticeably distort the response probability distribution.
An additional influence of structural motion manifests itself in the spectral analysis,
however; see item 1 in the frequency domain results below.
2. The deck-leg connection nonlinearity plays only a minor role in distorting the
statistical distributions of response.
3. Existence of drag apparently stretches the response distribution as compared with
that of the water elevation - the response distribution contains larger extremes than
does the water surface elevation.
Frequency Domain
1. Relative motion when combined with (even minor) quadratic drag manifests itself
as damping. The general trend in RAO caused by relative motion is that the RAO
peak decreases with increasing wave height - this is in agreement with the
observation from the regular wave tests. When relative motion is the main
contributing factor of nonlinearities, the average magnitude of RAOs computed
from irregular waves and that computed from regular waves are of the same order.
2. Existence of drag causes a definite input energy level dependency of the RAOs;
higher waves result in higher RAOs both in regular and irregular waves. In this
case, RAOs computed from irregular waves are generally lower than those
computed from comparable regular waves.
3. The magnitude of RAOs is also (weakly) dependent upon the relative locations of
wave peak frequencies and structure resonance frequencies.

78

4.

S.

Limited hydrodynamic cancellation is observed around 1.2 Hz for all three models
in irregular waves; this confirms the results from the regular wave tests and
theoretical prediction.
The inter-structural coherence of the measurements is noticeably greater than that
between waves and structural response. This indicates that nonlinearities are
primarily of a hydrodynamic nature.

Data Analysis Aspects

1. In the probability domain analysis, the choice of interval between two succeeding
histogram steps or levels is a compromise between bias suppression and random
error suppression. A large interval is desirable to reduce the random error, while
a small interval is needed to suppress the bias error. This interval is determined
here to minimize total error of estimates. The normalized bias error associated
with the distribution estimate is less than 1%, and the normalized random error is
restricted to less than 5% with all models.
2. The choice of the number of data segments in the spectral analysis is critical to the
overall error of the results. Random error increases and bias error decreases with
a decreasing number of segments. The number of segments is chosen here to
minimize the total error. The normalized random error associated with estimates
of spectral quantities relating hydrodynamic to structural response is usually less
than 5% and that for inter-structural estimates is less than 1% over the energy-rich
range of frequency for all models.
3. If the wave spectra and RAO are narrow, extra care is needed in the spectral
analysis.
4. An important experience gained through the present random data processing is that
a careful error analysis is essential in this type of study. Computer software for
analyses such as these wiU always produce results. Blind analysis of random data
can lead to equally random results.

7.6.2 Computer Simulations

1.

2.

The comparison between the simulated and measured response is satisfactory; this
confirms the applicability of NOSDA to a stochastic sea.
The computational effort for the NOSDA simulation in irregular waves is
acceptable with the present models. Using a DECstation 3100 computer, the
ratio of the simulation time to the physical time is about 8:1.

79

8 Conclusions
The work included in this paper is aimed at investigating the influence of nonlinearities
on elevated jack-up rigs. The nonlinearities studied here originates from hydrodynamic
interaction and structural behavior. Both experimental and computational approaches
have been used. Testing on three principle jack-up models (I, II and II-M) has been
carried out in a wave tanle Two software packages, RANDA and NOSDA, have been
developed parallel to the laboratory studies. The RANDA software was used for
processing the random data from the irregular wave tests. NOSDA was developed as a
software package for stochastic, nonlinear, dynamic analysis of general, moving, slenderelement offshore structures. As a specific application, NOSDA was used to simulate the
dynamic behavior of the models tested in the lab. More specific conclusions from this
investigation are drawn in the following sections.

8.1 MODEL TESTING AND EXPERIMENTAL DATA PROCESSING

The models tested involved both hydrodynamic and structural nonlinearities, but the
different models have different types of nonlinearities. Model I includes a significant
relative motion, Model II has a high drag contribution plus a complicated deck-leg
connection; with an extra mass on the deck, Model II-M demonstrates the influences of
the P- () effect further. This segregation of nonlinearities with different models helps
isolate and thus better expose the influences of an individual nonlinearity on the
behavior of the structures.
The experimental data have been carefully processed and analyzed. The measured data
from the irregular wave tests were examined both in the probability domain and
frequency domain using the RANDA software. This "'\lith the error analysis lends
confidence to the conclusions concerning the model behavior when subjected to irregular
waves. The irregular wave test results, especially when compared with those from the
regular wave tests, increase the insight about the behavior of such rigs in a random sea.
The following more specific conclusions can be drawn from the present experimental
study (the most relevant sections in the previous text are indicated at the end of each
item):
Discrepancies between the stiffness obtained from static tests and that inferred
from dynamic vibration tests have been observed both in the field (by others) and
in the present lab models. The apparent dynamic stiffness of a model was found

80

to be up to 4 times its static stiffness. This dynamic stiffness enhancement of the

present models is caused by the large local damping at the deck-leg connection,
which effectively makes the connection rigid. (See section 4.2.2.)
When structure motion combined with quadratic drag is the main nonlinearity,
higher waves cause lower Response-Amplitude-Operator (RAO) values as a result
of hydrodynamic damping generated by relative motion - this is true for both
regular and irregular wave situations. In this case, the average magnitude of RAOs
computed from irregular waves and that computed from regular waves are of the
same order. This type of nonlinearity, however, does not noticeably deform the
response probability distribution in irregular waves. (See sections 6.3, 7.3.1 and
7.4.l.)
When drag and structural nonlinearities are important, the trend of RAO variations
in the investigated cases is to increase with increasing input level in both the
regular and irregular waves; this dependency is attributable to two factors: (1) the
structure's stiffness decreases with increasing loading level because of structural
nonlinearities and (2) the drag excitation increases quadratically with increasing
wave heights. Another consequence of the drag plus structural nonlinearities is
that RAOs in irregular waves are generally lower than in comparable regular
waves.
Additionally, existence of drag apparently stretches the response
distribution - the response distribution contains larger extremes than does the water
surface elevation. (See sections 6.3, 7.3.2 and 7.4.2.)
An extra parenthetic observation is that a linear-looking overall response (a
sinusoidal output resulting from a sinusoidal input, for example) does not
necessarily mean that the system is linear; a linear-looking overall system can
contain significant internal nonlinearities. (See section 7.3.)

8.2 COMPUTER SIMULATIONS

NOSDA is a multiple degree of freedom, nonlinear, dynamic, time domain analysis

program for offshore structures. It allows the accurate representation of the nonlinear
phenomena involved in jack-up behavior. Application of this software to simulate the
dynamic behavior of the models tested in both regular and irregular waves resulted in
a satisfactory comparison with the measurements. This validates the computational
schematizations and confirms the applicability of NOSDA.
The computational effort for NOSDA simulations of the models tested is acceptable for
research purposes. The ratio between the computer time and the physical time is about
1:1 with regular wave simulations and 8:1 with irregular wave simulations using a
DECstation 3100 computer. Such an approach is expected to be still inefficient for
routine prototype design practice, however. Even so, the availability of a more precise,
81

verified, dependable and commonly accepted computational model will make it possible
to properly and conveniently evaluate less exact but more efficient routine design
procedures, The overall purpose of this total project is to develop, document and verify
this computational modeL Further research can focus on reduction of the number of
degrees of freedom and thus computational intensity,
Some other salient results found in the computer simulations are recapitulated as
follows:
Structure compliance should be included in the hydrodynamic force determination,
Relative velocity, instead of absolute water particle velocity, is required for
simulating the behavior of a jack-up structure, This allows the straightforward
modeling of hydrodynamic damping. Near resonance this (extra) damping level is
important even though the contribution of the structural velocity to the
computation of the hydrodynamic force might otherwise be of minor importance.
Drag, when acting on a flexible structure, then remains important even under low
KC Number conditions. (See section 6.5.1.)
Linear wave theory is sufficient for predicting the wave kinematics with the low
wave cases investigated; using linear wave theory or Stokes' 2nd Order Wave
Theory makes negligible difference, here, (See sections 5.2 and 6.5.5.)
Numerical investigation using NOSDA shows that identical overall dynamic
responses over a wide range of frequencies can be obtained from models which
differ only in the damping and stiffness values at the deck-leg connection. Such
models have quite different static properties, however. (See sections 6.2.2.3 and
6,5.6.)

Effects of weight eccentricity (P- b) cannot responsibly be neglected. This effect

can be well simulated in the NOSDA software. (See sections 6.2.2.4 and 6.5.7.)

8.3 CLOSING REMARKS

As concluding remarks it should be emphasized that no lab tests can exactly reproduce
the physical process of large, complex systems such as a prototype jack-up standing in a
random sea. Comparing the model tests with a real sea situation, the Reynolds Numbers
are too low, the structural damping of model II( -M) is excessively high, spudcan fixity is
neglected, structural response relative to water particle motion is exaggerated, and so
forth. All these deviations or simplifications will certainly restrict the applicability of the
results obtained from the present study. On the other hand, these models isolate
(although often exaggerate) various important physical processes involved in the jack-up
behavior and thus help expose and pinpoint the consequence of individual parameters.
Besides, the modeling procedure developed for and validated by the structures tested

82

provides a solid basis for the further study of prototype jack-up behavior. In fact, the
similar schematization has, in the meantime, been successfully applied for a case study
of a prototype jack-up. The outcome demonstrates that the results and insight gained
from the present model study are also qualitatively valid with the prototype, although the
of various parameters can differ from the lab
quantities and relative
situation.

Summary
The present work was carried out as a
of a project with objective to increase the
detailed knowledge of the behavior of
platform components as well as the
predicticn of the overall structure's elevated behavior and (remaining) lifetime.
The need for such a study is demonstrated
the relatively high rate of structural failure
for jack-up rigs as
to fixed platforms and the ~onsiderable discrepancy existing
among various industry assessment methods and criteria for elevated jack-up platforms.
The work presented in this paper concentrates on the investigation of the influence of
hydrodynamic and structural nonlinearities on the
behavior of elevated jack-up
rigs. The work involves the LV"V"H>"'" three aspects:
- Software Development
the prosecution of the
Two software
have been
investigation: (1) NOSDA simulation software for the Nonlinear Offshore Structure
RANDA software for RANdom Data Analysis.
Dynamic Analysis;
- Physical Model Tests
The experimental studies of three !JllUo.!fYHO jack-up models were carried out in a
wave tank of the Hydromechanics
of the Faculty of the Mechanical
Engineering and Marine Technology, TV Delft. The model testing program
included exposing the models to
and irregular uni-directional, long crested
waves as well as static and free vibration tests.
- Experimental Result
and
Simulations
The processing of the measured data from the irregular wave tests was carried out
a careful error analysis. The model
using RANDA software and
behavior in regular and
waves was simulated using NOSDA software.
The experimental results and associate
- Hydrodynamic forces include an
motions from structural
in the

simulations demonstrate that:

drag element. Relative
are such that they cannot be neglected

,"UJlHIJ'!,w"r

83

- The stiffness obtained from static tests can be significantly lower than that
inferred from dynamic vibration tests; this discrepancy has been observed
both in the field (by others) and in the present lab models. This apparent
stiffness enhancement in the present testing is caused by the large local
damping at the deck-leg connection.
- Inclusion of the P- 0 effect in the structural schematization is essential for the
jack-up simulations. This effect can be well modeled with the NOSDA
software.
Nonlinearities are important even with the present simplified model testing and different
nonlinearities have different (sometimes compensating) influences on the structure's
dynamic behavior. Therefore, the scientifically responsible type of computer model for
jack-up analysis must be capable of reproducing a wide range of nonlinear, dynamic
phenomena. Use of a nonlinear, dynamic, stochastic computer model based upon a
discrete element schematization and working in the time domain has proven to be a
success for simulating the dynamic behavior of the models tested. While the
computational effort of such an approach is acceptable for the present models, further
improvements in the computational efficiency are needed for its application to routine
prototype design practice. In spite of this, the availability of a more precise, verified and
dependable computational model is essential as a tool with which to concisely check the
performance of more approximate, efficient routine design procedures.

Acknowledgment
Help from many people has made the present work possible. The research included in
this paper was carried out in the Workgroup Offshore Technology of the Delft University
of Technology, under the supervision of Prof. J. G. Wolters and Prof. J. Blaauwendraad.
The direct guidance came from Mr. W. W. Massie. These three persons are from the
Faculty of Civil Engineering. Their sustained help and encourage are the driving force
for this work.
The author is grateful to other members of his steering committee: Prof. B. Boon, Prof.
M. van Holst, Prof. J.A. Pinkster (from the Faculty of Mechanical Engineering and
Marine Technology), Mr. G.H.G. Lagers from Marine Structure Consultants, Mr. P.G.F.
Sliggers from Shell International Petroleum Maatschappij and Mr. W.J. van Tiggelhoven
from Neddrill. Their valuable support and constructive suggestions have greatly
improved the quality of the present work.

84

The author is also indebted to the colleagues in the Hydromechanics Laboratory of the
Faculty of Mechanical Engineering and Marine Technology where the three jack-up
models were tested and the colleagues in the Fluid Mechanics Laboratory of the Faculty
of Civil Engineering where the analog data were digitized.
The author wishes to thank the Dutch Technology Foundation (STW) for their financial
support for this study.

Notation and symbols

The most common symbols used in this paper are listed in this section. International
standards of notation have been used where available except for occasional uses where
a direct conflict of meaning would result. Certain symbols have more than one meaning,
however. This is only allowed when the local context of a symbol used is
sufficient to define its meaning explicitly.
The numbers in the right hand of this table indicate the sections where the
corresponding symbol first appears in this paper.

Roman letters
A

cross section area

Ai

average response amplitude at cycle i

displacement-strain relation matrix

II.I

or kinematic matrix
resolution bandwidth

V.I

B
Be
Bg
C

generalized kinematic matrix

structural damping matrix

Y.2.2

7.5
IVA

V.I

drag coefficient
inertia coefficient

6.2.1.2
6.2.1.2

outer diameter of leg

generalized constitutive relation
constitutive relation
water depth

E
E1

elastic modulus
leg bending stiffness

3.1

structural load vector

V.I

FAx

x component force at leg A

3.1

Cd
Cm
D
Dii

De

6.2.1.2
IVA

V.l
3.1

3.1

85

FAy
F Az
F Ex
FEy
FEz

Fc:r

Fe\'
Fez
Fmax
F min

I
In
Ip

Y component force at leg A

z component force at leg A
x component force at leg B
Y component force at leg B
z component force at leg B
x component force at leg C
y component force at leg C
z component force at leg C
maximum hydrodynamic wave force on a cylinder
minimum hydrodynamic wave force on a cylinder
cyclical frequency
primary natural frequency of structure

H,

peak wave frequency

local wave load causing maximum total load
local wave load causing minimum total load
autospectral density function (one-sided)
acceleration due to gravity
wave height
comparable wave height

Hs

significant wave height

moment of inertia
structural stiffness matrix
Keulegan-Carpenter parameter
structural lateral stiffness from dynamic tests

r
G<,
g

KC
Kd
Ki
K'b
Km"
K,
Kt
k

L
Ld
I
M
Meq

tn,

86

incremental stiffness
leg theoretical pure bending stiffness
model theoretical pure bending stiffness
struct\lral stiffness obtained from static tests
theoretical structural stiffness
spring coefficient
wave number
leg length
leg spacing
element length
structural mass (or inertia) matrix
structural equivalent mass
ith moment of spectrum

3.1
3.1
3.1
3.1
3.1
3.1
3.1
3.1
IV.l
IV.1
3.2
3.1
3.2
IV.l
IV.l
3.3
3.1
3.2
7.4.1
3.2
3.1
V.l
6.3
4.2.2
I
H.2
II.2
4.1
4.2.2
V.2.2
IV.1
II.2
V.3.1
IV.4
V.I

H.2
7.4

interpolation function matrix

decrement coefficient
vertical force

Pe
Re

r
rc
r'
T
Tm
Tn
Ts

Euler critical load

Reynolds number
viscous damping coefficient
critical damping coefficient
viscous damping coefficient per unit length
wave period
free vibration response period
structural natural frequency
segment length

flD

time
thickness
displacement
water particle horizontal velocity
nodal displacement vector
x direction deck displacement at location A
x direction deck displacement at location C
displacement field
x direction deck acceleration at location D

amplitude of water particle horizontal velocity

u
UA
Uc
Uc

displacement amplitude

V.I
II.3
IV.4
II.2
6.2.1.2
4.2.3
4.2.3

V.l
6.2.1.2
ILl
4.2.1

7.5

V.l
V.2.1
6.5.6
IV.1
V.I
3.1
3.1
V.I
3.1
6.2.1.2

6.5.6

VA

Y direction deck displacement at location A

3.1

Vc

Y direction deck displacement at location C

vD

y direction deck acceleration at location D

3.1
3.1

Weq

structural equivalent weight for the P- 0 effect

II.2

structural displacement vector

coordinate direction
coordinate direction
vertical coordinate direction

V.l
3.1
3.1
3.1

Greek letters
L1

increment
horizontal eccentricity

1.3

log decrement

H.3
87

e
g

em
er
(

IJA

e
A
v

e
11:

p
PI'>'

p'
U

ug
(Un

strain vector
generalized strain vector
spectral width parameter
normalized random error
coefficient
instantaneous wave surface elevation at location A
rotational angle
wave length
fluid viscosity
structural damping ratio
3.1415926536
leg mass density
water density
leg mass density per unit length
stress vector
generalized stress vector
circular natural frequency

V.l
V.I
7.4
3.3

IV.4
3.1
V.2.2
HI.I
6.2.1.2
4.2.3
B.2
Y.2.l
V.2.1
V.I
V.l
V.I
II.3

Acronyms
DAS
DEM
DOF

FIT
IR
MPM
RAO
rms
SWL
UV

data acquisition system

discrete element method
degree of freedom
fast Fourier transform
instrumentation recorder
most probable maximum
response-amplItude-operator
root mean square
still waver level
ultraviolet light

7.2.2
6.2.2
6.2.2
7.2.3
3.2
7.1
3.3
7.1
5.2
3.2

References
(Anon.) 1981-1983. Dynamics of Jack-up Platforms - A Joint Industry Project, Part
Reports 1-4 and Final Report, Det Norske Veritas, Oslo, Norway.
(Anon.) 1983. Wave Kinematics in Irregular Waves, MaTS Program Report, M1628
/MaTS VM-I-4, Delft Hydraulics Laboratory, April.

88

(Anon.) 1989. Practice for the Site-Specific Assessment of Jack-Up Units, Shell
International Petroleum Maatschappij, The Hague, May.
(Anon.) 1990. Guideline for Site Specific Assessment of Mobile Jack-Up Units, the
Working Group of the Joint Industry Sponsored Project "Jack-Up Site Assessment
Procedures and Establishment of an International Technical Guideline", June.
(Anon.) 1991. Jack-Up Site Assessment Guided onto Common Ground, Offshore
Engineer, January.
Battjes, l.A and van Heteren,l., 1983. Measurements of Wind Wave Kinematics, Report
WWKZ-G007, Rijkswaterstaat, The Netherlands.
Bea, RG. and Lai, N.W., 1978. Hydrodynamic Loadings on Offshore Platforms, Paper
3064, Offshore Technology Conference, Houston, Texas, May.
Bearman, P.W., 1988. Wave Loading Experiments on Circular Cylinders at Large Scale,
BOSS '88, Trondheim, Norway.
Bendat, 1.S. and Piersol, AG., 1971. Random Data: Analysis and Measurement
Procedures, Willey-Intersience, New York.
Bendat, J.S. and Piersol, AG., 1986. Random Data: Analysis and Measurement
Procedures (2nd Edition), Willey-intersience, New York.
Bendat, J .S., 1990. Nonlinear System Analysis & Identification from Random Data,
WiUey-Intersience, New York.
Blaauwendraad, J. and Kok, AW.M., 1987. Numerical Techniques for Engineering
Analysis and Design, NUMETA 87 Conference, Swansea, u.K.
B1aauwendraad, J., 1989. TILLY, Introduction Manual, Research Report, Faculty of
Civil Engineering, Delft University of Technology. (in Dutch).
Borgman, L.E., Allender, J., Krogstad, H., Barstow, S. and Audunson, T., 1989.
Conditional Simulation of Ocean Wave Kinematics and Comparisons with Storm
Field Measurements, NATO Advanced Research Workshop, Molde, Norway, May.
Brekke, J.N., Murff, J.D., Campbell, R.B and Lamb, W.C, 1989. Calibration of Jackup
Leg Foundation Model Using Full-Scale Structural Measurements, Paper 6127,
Offshore Technolog'j Conference, Houston, Texas, May.
Brekke, J.N., Campbell, R.B., Lamb, W.C and Murff, J.D., 1990. Calibration of Jackup
Structural Analysis Procedure Using Field Measurements from A North Sea
Jackup, Paper 6465, Offshore Technology Conference, Houston, Texas, May.
Burrows, R., 1979. Probabilistic Description of the Response of Offshore Structures to
Random Wave Loading, Mechanics of Wave-Induced Forces on Cylinders, Pitman,
San Francisco.
Carlsen, CA, Kjeoy, H. and Eriksson, K., 1986. Structural Behavior of Harsh
Environment Jack-ups, The Jack-up Drilling Platform - Design and Operation,
Collins, London.
Carter, G.C, Knapp, CH. and
A.H., 1973. Estimation of the MagnitudeSquared Coherence via Overlapped Fast Fourier Transfer Processing, IEEE
Transactions on Audio and Electroacoustics, Vol. AU-21, P337, Institute of
Electral and Electronics Engineering, August.

89

Chakrabarti, S.K, 1971. Dynamics of single point mooring in Deep Water (Discussion),
Journal of Waterway, Port and Coastal Division, Vol. 97, American Society of Civil
Engineers, August.
Chakrabarti, S.K, 1980. Laboratory Generated Waves and Wave Theories, Journal of
Waterway, Port and Ocean Division, Vol. 106, American Society of Civil
Engineers, August.
Chakrabarti, S.K., 1982. Wave Force Coefficients for Rough Vertical Cylinders,
Journal of Waterway, Port, Coastal and Ocean Division, American Society of Civil
Engineers, November.
Chakrabarti, S.K., 1986. Hydrodynamics of Offshore Structures, Springer Verlag,
Computational Mechanics Publications, Southampton, Boston.
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.
Dean, R.G., 1968. Relative Validity of Water Wave Theories, Proc. on Civil Engineering
in Ocean, American Society of Civil Engineers, San Francisco.
Dean, R.G. and LeMehaute, B., 1970. Experimental Validity of Water Wave Theories
Structural Engineering Conference, American Society of Civil Engineers, Portland,
Oregon.
Forristall, G.Z., 1981. Kinematics of Directionally Spread Waves, the Conference on
Directional Wave Spectra Applications, American Society of Civil Engineers,
Berkeley, California, September.
Grundlehner, G.1., 1989. Simple Analytical Jack-Up Leg - Hull Connection Model,
Report P 8525-1428, Faculty of Mechanical Engineering and Marine Technology,
Workgroup Offshore Technology, Delft University of Technology, March.
Griindlehner, GJ., 1989. A Simple Model for Jack-Up Leg - Hull Deformation
Behavior, Report PF 8525-1470, Faculty of Mechanical Engineering and Marine
Technology, Workgroup Offshore Technology, Delft University of Technology,
August.
Gudmestad, O.T., 1990. A New approach for Estimating Irregular Deep Water Wave
Kinematics, Applied Ocean Research, Vol. 12, No. 1.
Holtrop, E.F., 1989. Model Testing of Spudcan Behavior Under Decreasing Vertical
Load, Report 313, Geotechnical Lab, Faculty of Civil Engineering, Workgroup
Offshore Technology, Delft University of Technology, July.
Jenkins, G.M and Watts, D.M., 1968. Spectral Analysis and its Applications, Holden-I?ay,
San Francisco.
Journee, 1.MJ., Massie, W.W., Boon, B. and Onnink, R., 1988. Model Experiments on
Jack-Up Platform Hydrodynamics, Report no. 809: Ship Hydromechanics Lab,
Faculty of Mechanical Engineering and Marine Technology, Workgroup
Offshore Technology, Delft University of Technology, November.
Klaver, J.P., 1990. The effect of Footing Length on Footing Bearing Capacity,
Geotechnical Lab, Faculty of Civil Engineering, Workgroup Offshore Technology,
Delft University of Technology, April.

90

Lagers, G.H.G, 1990. Morison Coefficients of Jack-up legs, MSC ref SP 8603-1505,
Marine Structure Consultants
bv, February.
1989. A Philosophy for the Integrity Assessment of Jack-Up
Leijten, S.F., Efthymiou
19236, Society of Petroleum Engineers.
Units,
1988. NOSDA - Nonlinear Dynamic Analysis Program for
Liu, P. and Massie,
Offshore Structures, Research Report, Faculty of Civil Engineering, Workgroup
Offshore Technology, Delft University of Technology, December.
Liu, P., 1989a. Elevated Jack-Up Platform Analysis Methods, Research Report, Faculty
Offshore Technology, Delft University of
of Civil Engineering,
Technology,
Liu, P., 1989b. Nonlinear
Simulation of Jack-Up Platform Models, Research
Report, Faculty of Civil Engineering, Workgroup Offshore Technology, Delft
University of Technology, November.
J.G. and
J., 1990. Nonlinear Simulation
Liu, P., Massie,

Models to Irregular Waves, Research

Offshore Technology, Delft

Liu,

J., 1991. Response of Jack-Up

Offshore Technology

Delft University Press, Delft, The

Liu,

Offshore

Elevated Jack-Up Platform Durability - A

Delft University of Technology,

October,
Durability Problem, Paper 20911,
of Petroleum Engineers, The Hague, October.
1991. Jack~Up Leg Hydrodynamic
6592, Offshore Technology Conference,
Michels,

V'.A~.r '

Leg to Hull Connections, Faculty of

Offshore Technology, Delft University of
Wave Forces on a Pile, Journal of
American Society of Civil Engineers.
and Vetterling, W.T., 1986. Numerical
Cambridge University Press, Printed in

91

Rienecker, M.N, and Fenton, J.D., 1981. A Fourier Approximation Method for Steady
Water Waves, Journal of Fluid Mechanics, VoL 104.
Rodenbush, G. and Forristall, G.Z., 1986. An Empirical Model for Random Directional
Wave Kinematics Near the Free Surface, Paper 5097, Offshore Technology
Conference, Houston, Texas, May.
Sliggers, P,G.F., 1991. SIMP Practice for Site Specific Structural Fitness for Purpose
Assessment of Jack-Up Rigs, Paper 21979, IADC, Society of Petroleum Engineers.
Sabey, RJ., 1989. Wave Theory Predictions of Crest Kinematics, NATO Advanced
Research Workshop, Molde, Norway, May.
Spaargaren,
1988, Hydrodynamic Force Cancellation, Faculty of Civil Engineering,
Workgroup Offshore Technology, Delft University of Technology, April.
Spaargaren, PJ., 1989. Spud can Modeling for Jack-Up Dynamics, Faculty of Civil
Engineering, Workgroup Offshore Technology, Delft University of
Technology, September.
Spaarman, R., 1989. Verwerking, Analyse en Evaluatie van Hydrodynamische Modelexperimenten met een Jack-up, Faculty of Mechanical Engineering and Marine
Technology, Workgroup Offshore Technology, Delft University of Technology,
March.
Steele, K.M., Finn, L.D., and Lambrakos, K.F., 1988. Compliant Tower Response
Prediction Procedures, Paper 5783, Offshore Technology Conference, Houston,
Texas, May.
Stu it, H,G., 1989. Spud-can Limit Loads in Sand under Decreasing Vertical Loads,
Geotechnical Lab, Faculty of Civil Engineering, Workgroup Offshore Technology,
Delft University of Technology, May.
Tucker, MJ., Challenor, P.G. and Carter, DJ.T., 1984. Numerical Simulation of a
Random Sea, a Common Error and its Effect upon Wave Group Statistics,
Applied Ocean Research, Vol. 6, No.2.
Vugts, J.R. and Bouquet, A.G., 1985. A Nonlinear Frequency Domain Description of
Wave Forces on an Element of Vertical Pile in Random Sea, Behaviour of
Offshore Structures, Elsevier Science Publishers, Amsterdam.
Wheeler, J.D., 1970. Method for Calculating Forces Produced by Irregular Waves,
Journal of Petroleum Technology, March.
Yuen, R.c. and Lake, B.M., 1982. Nonlinear Dynamics of Deep-Water Gravity Waves,
Advances in Applied Mechanics, Vol. 22.
Zeelenberg, B. L., 1990. Schematization of Hydrodynamic Interaction of Jack-Up Legs,
Faculty of Civil Engineering, Workgroup Offshore Technology, Delft University
of Technology, June.
Zeelenberg, B. L., 1990. Morison Coefficients for Jack-Up Legs, Faculty of Civil
Engineering, Workgroup Offshore Technology, Delft University of Technology,
July.
Zhang, J., Randall, R.E. and Spell, c.A., 1991. On Wave Kinematics Approximate
Methods, Paper 6522, Offshore Technology Conference, Houston, Texas, May.

92

Appendix I Static test results

The static tests were carried out for each model by exerting static, horizontal loads at the
deck level and recording the corresponding displacements. The results are plotted in
figures 1.1 through 1.3. Note that the static test with Model I was carried out before its
deck-leg connections were modified.

50rFo~c~c~e~(N~l,--_ _ _ _ _ _ _ _ _- ,

1.4 r-FO:::.:c.;:.C::..C~(N,,-,),--_ _ _ _ _ _ _...,

l.2

40

:w

O.B

0.6

20

0.4

'0

0.2

0.02

0.04

0.06

0,08

o~~--~--~~-~~--~
o 0.01 0.02 0.03 0,04 0.05 0.06 0.07

Deck Displacement (m)

Deck Displacement (m)

~UA

Figure !.1 Measured Overall Static

Constitutive Relation (Model f)

-B-u c

Figure 1.2 Measured Overall Static

Constitutive Relation (Model II)

l.< ;-1'0::..:'..::.'0::..."-"(N"'l'--_ _ _ _ _ _ _ _-,

1.2

0.8
0.6

0.4

x (u)

/./
/;f/

0.2

/f/ ______

o"-~

0.01

0.02

~~

0.03

__

0.04

~~

0.05

__

0.06

0.07

Deck Displacement (m)

Figure 1.3 Measured Overall Static

Constitutive Relation (Mode/II-M)

Figure fA Deck Displacement Caused by

Rotation

is the displacement in the x direction recorded at location A and U c

at location C - see figure 104.
In these figures

UA

93

In the plots it can be seen that with Models nand H-M u A deviates from Uc (With
Model I only uA was measured.) Since the deck frame is stiff enough to be considered
as a rigid body, the differences between uA and Uc are caused by the deck rotations due
to load asymmetries, leg stiffness differences and/or connection stiffness differences. U A
is used to calculate the global stiffness so that the effect of rotation is avoided - see
figure 1.4.
The incremental global lateral stiffness is computed by:
(ll)

The results for all three models are given in table 1.1:
Table 1.1 Global Horizontal Stiffness at Deck Level

Model I

Model II

Model II-M

Ki

(N/m)

(N)
0.000

0.000

0.000

0.0000

0.010

490.00

0.195

0.Q108

18.06

0.195

0.0140

13.93

0.019

544,40

0.390

0.0185

25.32

0.390

0.0255

16.96

14.7

0.029

490.00

0.590

0.0293

18.52

0.590

0.0378

16.26

19.6

0.039

490.00

0.785

0.0385

21.20

0.785

0.0500

15.98

24.5

0.049

490.00

0.980

0.0495

17.73

0.980

0.0593

20.97

29.4

0.058

544.00

1.175

0.0600

18.57

uA
(m)

(N)
(J.O

4.9
9.8

O.(JO(J

Average

508.00

uA
(m)

Average

Ki

(N/m)

(N)

19.90

uA
(m)

Average

K;
(N/m)

16.82

It can be seen that the incremental stiffnesses fluctuate at different load levels. It is hard
to find a consistent relation that follows the changes. Possible explanations for the
fluctuations are: (1) equipment errors and (2) structural nonlinearities such as the
nonlinear connections at both upper and lower ends, etc.

At the beginning of the loading paths, the incremental stiffnesses for all three models are
systematically lower; this might be caused by (1) free play in the connections of both
ends and/or (2) structural dry friction which keeps the structures away from their true
equilibrium positions.

94

The only difference between Model II and Model II-Mis the deck mass and therefore
P-15 effect. The P-15 effect reduces overall structural stiffness; this is confirmed by the
lower average stiffness for Model U-M in the above table.
The average global horizontal stiffness of the model is considered to be its representative
,sta tic' stiffness.

Appendix II Free vibration test results

ILl NATURAL PERIOD

When testing Model I standing in air, an unexpected significant decrease of response
period, Tm , with succeeding vibration cycles (in fact with decreasing response levels) was
found - see figure n.l and the table derived from it.

Cycle

1.28

1-2

1.22

2-3

1.14

3-4.

1.09

4-{;

l.08

Average

Fig'.tre

lI, 1

Tm (sec)

0-1

1.16

Free Vibration Response Record

There are at least three factors which can influence the response period:
1. p- (; Effect
A lateral displacement results in an additional second order moment. This leads
to a relatively smaller horizontal restoring force of the structure (smaller virtual
stiffness) and in turn yields a longer response period.
2. Damping
The natural period, Tn , is expressed as:
(II. I)

95

3.

where Tm is the free vibration response period and I; the damping ratio. This
influence is of minor importance, however. Even when;; is as high as 20%, the
difference is within 2% - ,,1(1 - ;;2) = 0.98; the free vibration response period can
be used directly as the natural period.
Deck-leg and bottom-leg connections

Comparing these three possible causes, it is most likely that the scatter of the response
period data stemmed from the bad leg-deck connections. A sketch of these connections
is given in figure H2.
Clamping Ring
,~

Screw
Free Play Gap
Plate

Figure Il.2 Deck-leg Connection Detail

It is designed to be a perfectly clamped joint with infinite stiffness. The deck members

are connected by two parallel plates. Two parallel clamping rings screwed to the leg
outside the plates provide fixity. A more realistic process of connection deformation with
increasing load might be:
1. The connection remains undeformed in the horizontal direction due to the Coulomb
friction between the clamping rings and plates until the loading exceeds the critical
static friction; meanwhile the clamping rings impose a pair of vertical (normal) forces
on the upper and lower plates respectively due to the bending moment. Since the
plates are relatively weak in terms of bending stiffness, a significant deformation can
occur now and throughout the following loading phases; this can yield a much more
flexible connection than the originally intended rigid clamping.
2. The rings start to slide (relative to the plates) so that the leg undergoes a free play
till the leg touches the edges of the deck connecting plate holes.
3. The connection deformation follows the elastic rule.
4. It enters a plastic phase when the local leg and/or plate yielding stress is exceeded.

96

Note that since the contact area is relatively small, local plasticity is expected to be
reached easily. The constitutive curve of the whole process described above is
summarized in figure H.3.
M

Dry Fnction (1)

""'---'----'------'----'-8

Figure 11.3 Possible Deck-leg Connection Constitutive Relation

It can be seen that beyond a certain loading level, a larger displacement corresponds to

a lower resulting overall stiffness and thus a higher response period; this is qualitatively
in agreement with the measurements in figure II.1.
Later, the deck-leg connections of Model I were glued to improve their mechanical
behavior. Since the clamping rings, the plates and the leg itself of model I are all made
of PYC, the gluing was effective. The response period data with the glued connections
are tabulated below:
Table 11.1 Response Periods of Model I
after Gluing

Cycle

Tm (sec)

0-1

1.04

1-2

1.03

2-3

1.01S

3-4

1.00

It shows that the response periods after the gluing

are much more consistent. The structural
nonlinearities of the deck-leg connection has been
largely eliminated.

Strictly speaking, the definition of natural period is

not valid for a nonlinear system and many 'mature' techniques developed for a linear
system are not applicable to a nonlinear system. However, the output of commonly
encountered slightly nonlinear systems can be seen to be composed of a 'fundamental'
linear part plus a nonlinear modification. The techniques normally used for linear
systems can be transplanted to approximately treat a nonlinear system in a piece-wise
(incremental) form or in an average sense. Using this analogy between linear and

97

slightly nonlinear systems, the response period in free vibration will be called the natural
period and the virtual stiffness of the structure will be called simply the structural
stiffness.
Models II and H-M have the same basic deck-leg connections as Model!. A worse
situation could be expected now since their legs and deck are of smaller sizes. It is
obvious from figure IIA that a more severe free play can result from the same clearance
with Model U(-M).

Model I

\
/Le

Model Jl(-M)

Figure 11.4 Different Influences of Clearance

In order to avoid extra structural uncertainty the deck-leg connections of Model II(-M)
should have been improved, too. However, the combination of materials now used (PVC
clamping rings, copper legs and aluminum plates) made gluing unattractive. Therefore,
during a free vibration run, different natural periods were obtained for different response
cycles - in fact for different response amplitudes - just as was the case initially with
Model L These natural periods within one run were averaged to yield the
'representative' period.
When the free vibration tests with Models II and n-M were carried out, it was found that
the decay was so fast that it was difficult to record the response traces. As a remedy although not scientifically responsible - a sort of 'hand help' was used to obtain readable
oscillatory response traces. These results are less accurate but are still used further (with
care!).
Natural periods associated with different deck displacement amplitude, A, for each of the
models obtained from the free vibration tests in air are listed in table H.2.

98

Table 1l.2 Model Natural Periods

Model I

Model II

A (em)

Tn (sec)

3.7

1.04

2.15

2.7

1.03

1.55

2.0

1.015

1.5

1.00

Average

1.02

(em)

Average

Model II-M

A (em)

Tn (sec)

1.38

4.1

2.44

1.09

0.8

1.83

0.4

1.52

Average

1.93

Tn (sec)

1.25

This table shows that while the results with Model I are rather consistent, those with
Models II and II-M are quite scattered. A general trend is that the natural periods
decrease with decreasing amplitudes; this nonlinear phenomenon can be explained, as
discussed above, mainly by the imperfect connections. Model U-M has longer periods;
this is due to greater deck mass and the extra P- 13 effect.
The natural period data for Models II and II-M should be used with caution.

n.2 MODEL STIFFNESS

There are three approaches to obtain model stiffnesses:
1.

Theoretical Approach
If the legs are completely clamped into the deck at the upper end and perfectly

hinged to the bottom at the lower end, then each of the legs can be schematized
as a cantilevered beam. The theoretical pure bending stiffness can be expressed for
3 legs then as:

(B.2)

where:
Kmb
Klb

L
El

= model theoretical pure bending stiffness

leg theoretical pure bending stiffness
= leg length
= bending stiffness
=

99

When there is an (equivalent) deck weight, the resulting P-O effect can be
expressed to be a reduction of the pure bending stiffness approximately by:
(II.3)

where:
Kt

Weq

theoretical model stiffness with the P- 0 effect

= equivalent deck weight

=

Euler critical load

column theory)

3-n?-EI/(2L)2 (from the slender compressional

Assuming ideal connections (clamped deck-deck connection and hinged leg-bottom

connection), an analytical derivation shows that in addition to the deck weight
11/16 of the leg weight should be lumped to the deck level for the P-O contribution
- see Liu (1989b). Using this result and data in table 3.1, the theoretical stiffnesses
for each of the models are given in table 11.3.
Table II.3 Theoretical Model Stiffnesses

Model No.

Kmb

Weq

Pe

P-{} reduction

Kt

(N/m)

(N)

(N)

(%)

(N/m)

1719

276.69

3152.70

8.8%

1568.1

104

40.02

192.90

20.7%

82.4

H-M

104

70.92

192.90

36.8%

65.8

2. Static Load Tests

The static stiffness, Ks , for each of the models has been obtained in table 1.1.
3. Derivation from Free Vibration Response
By simplifying each of the jack-up models to a single degree of freedom system, the
system global 'dynamic' stiffness, Kd , can be inferred from the natural period
obtained in the free vibration tests:
100

K =M
d

eq

[2rc]2
T

(HA)

where:
Meq

Tn

equivalent mass
average natural period (from table H.2)

The details for calculating equivalent mass are given by Liu (1989b). It has been
demonstrated that 17/35 of the leg mass should be counted in the model equivalent
masses for horizontal response, assuming that the legs move in accordance with
their static deflection curve. The equivalent masses in air for each of the models
are tabulated in table 11.4.
Table lI.4 Model Equivalent Masses (in air)

Model No.

Meq (kg)

26.81

II

3.50

II-M

6.65

The stiffnesses of the models obtained from these three approaches are compared in
table n.5.
Table II.S Stiffness Comparison

Kt (N/m)

Ks (N/m)

Theoretical

As Built

As Built

Glued

1568.1

508.0

786.57

1017.0

II

82.4

19.9

88.4

--

II-M

65.8

16.8

70.5

--

Model No.

Kd (N/m)

101

The inconsistency is apparent. The results in section n.1 have already shown that the
models were different from their original design and therefore, the theoretical design
values of the model stiffness, K( , were not trustworthy. It should also be noted that the
Kd results for Models II and H-M are no better than the natural period data upon which
they are based. Nevertheless, there seem to be two tendencies worth pointing out:
1. Kd is systematically larger than Ks ; this is evident with models II and II-M, This

deviation indicates that the models were stiffer dynamically than statically,
This phenomenon may possibly be explained by:
a, Material Properties
Metal materials tend to have a higher yield stress under a dynamic load than
under a static load; this leads to a higher equivalent, resultant 'dynamic' stiffness
- see figure US
b. Connection Imperfections - Locally Concentrated Damping
As will be shown in the next section, (especially with Model II and Model II-M)
a large amount of damping is (locally) concentrated in the deck-leg connections;
relative movement between the deck and legs generates remarkable resistance,
This resistance increases with increasing relative velocities between the deck and
legs. Hence, the effect of the high damping in the connection is analogous to
a fixation against dynamic loading. When the damping is high enough, the
connection will behave dynamically as if it were clamped, As such, the localized
high damping at the connections has a significant influence not only on the
overall structural damping behavior, but also on the natural period and thus the
inferred dynamic stiffness, Kd , However, this fixing mechanism only exists when
the structure is experiencing a dynamic movement. If a loading is static, the
structure exhibits appreciably lower stiffness.

Dynamic

Static

Figure IIS Influence of Loading Rate on Yield Stress

102

2. With Models II and II-M the average dynamic stiffness values, Kd , seem quite in
agreement with the theoretical ones, K,. This, however, does not indicate the
agreement of these models with their original designs. From the discussion in point
b, above, it is clear that the calculated dynamic stiffness, Kd , generally does not
represent the real structural (static) stiffness. In fact, this gives an extra supporting
evidence for the assumption that the behavior of the deck-leg connection is closer
to a rigid clamping under dynamic loading.

II.3 STRUCTURAL DAMPING

The models tested involve the following structural damping mechanisms:
1. Viscous Damping

This type of damping is often found at lubricated contact surfaces; the submerged
bottom hinge connection is an example of this although its contribution to the total
structural damping is of minor importance.
2. Dry Friction
This type of friction is likely to occur in the imperfect deck-leg connections where
a free play gap exists in its pure form; it results in a hysteresis damping with a
rectangular hysteresis loop.
3. Internal Material Damping
Deformations of the materials of the structure itself result in energy loss via
heating. Material damping is of minor influence for the structural behavior;
compared with the case of Model I (whose legs are fabricated from PVC), this type
of damping is even less important with Models II and II-M (whose legs are
fabricated from copper). The material damping is commonly considered to be not
more than 1% of the critical damping.
4. Plastic Deformations
Considerable plastic deformation can take place when the yield load of a member
is exceeded. Generally the initial portion of the unloading curve is again elastic
and not coincident with the loading curve just experienced; it results in a hysteresis
curve which looks much like a parallelogram. The energy lost in the deformation
will manifest itself as a type of hysteretic damping. Such plastic deformations are
likely to occur in the deck-leg connections, since the contacts between the deck and
legs are very local.
Viscous damping is the only linear damping mechanism; the rest involve a nonlinearity
indicated by their dependency upon the response amplitude. Because of the convenience
of linear viscous damping in analysis, much effort has been invested (in the literature)
in the conversion of other damping mechanisms to 'equivalent' viscous forms.

103

With viscous damping, the relation between the log decrement, 13, and the decrement
coefficient, n , is:
(I1.5)

And further the overall structural damping is expressed as:

(II.6)

r = 2nMeq

where:
r

Meq

structural equivalent viscous damping coefficient

structural equivalent mass (from table 4.4)

The damping ratio between the viscous damping coefficient and critical damping
coefficient, rc (= 2 Wn Meq , where wn = 2J! fn is the circular natural frequency), is:
n

o
211"

(II.7)

The damping data for each of the models (in air) are given in table II.6 where Ai is the
average deck response amplitude associated with cycle i. Note that the global damping
values listed in this table have been calculated as if they were of the equivalent linear
viscous form within one cycle. Just as for the natural period data processed in the
previous section, the reliability of the damping data for Models II and H-M is
questionable; the values should be used with caution. The data for model I are relatively
dependable.
The following phenomena can be observed from this table:
1. The damping ratios are surprisingly large especially for Model II and Model II-M.
These values are much larger than the internal structural damping normally found.
The only possible source of these high damping percentages is the imperfect
deck-leg connection or also partly the leg-bottom connection (although the lower
connection is designed to be a perfect hinge).
2. The damping values of the Models II and H-M show strong nonlinearity just as with
the global stiffnesses; they are heavily dependent upon the structural response level.
This dependence relation is, however, rather scattered. In contrast to this, the
damping values of Model I are much lower and more consistent; it shows only a
relatively slight decrease with decreasing response amplitude levels. This
consistency is expected to result from the improved deck-leg connection.

104

3.

The average damping coefficient of Model Il-M seems slightly higher than that of Model II, although these
two models are identical except for the deck weight. This deviation can possibly be attributed to extra
(dry friction) damping resulting from that extra deck weight which was placed on top of the clamping
rings - this increased the contact forces between the clamping rings and the deck connecting plates at the
upper end as well as the contact forces in the leg bottom hinges.

Table Il.6 Damping Data

Model

II

H-M

Ai

(em)

Ai IAi+l

(- )

"

(-)

Tn (sec)

n (lis)

r (kg/s)

( (%)
5.0
4.2

2.55

1.37

0.32

16.14

1.30

0.26

1.03
1.02

0.31

2.15

0.26

13.78

1.25

1.27

0.24

1.00

0.24

12.82

3.8

1.00

1.22

0.20

1.02

0.20

lD.46

3.2

Average

1.29

0.26

1.02

0.25

13.37

4.1

1.65

3.26

1.18

1.38

0.86

5.99

18.8

1.lD

3.78

1.33

1.09

1.22

8.54

21.2

Average

3.52

1.26

1.25

1.04

7.27

20.0

1.6

5.00

1.61

2.44

0.69

9.15

25.6

0.9

5.72

1.74

1.83

0.95

12.67

27.8

0.35

3.00

1.10

1.52

0.72

9.61

17.5

Average

3.72

1.48

1.93

0.79

lD.48

23.6

Appendix HI Hydrodynamic anaiyis theory election

m.l WAVE THEORY

The wave states tested are given in tables 3.2 and 3.3 for regular wave tests and irregular
wave tests, respectively. The same parameters are plotted in figures HI I through IH.3
in the form of wave steepness (HIT2) and wave depth to wave length ratio (dIT2) to
show their relationship to the region of validity for various wave theories as suggested
by Dean & LeMehaute (1968 and 1970). For irregular waves, Hand T are replaced by
Hs and Tp to give an indicative vision on the scope where the representative waves work.
Chakrabarti's experimental study results are superimposed on the figures as dots with
legends - see Chakrabarti (1980) and (1986)0

105

Chdu"~tI'liI

LAta0A4T,",Y

TEST DATA

O.II---------l-.J
~

Modell
Waves

VI

".

".

~I~~--------_t~----~~----_+----~~----~
IR'1

iMlOR';'
(LIM[NI'

.JL (FIISec')
T'

Regular wsw mats

Irregular wave issts

Figure II!.l Model I Waves

i $ t 4 A L L g w ' - - - . ; < I - - - - I M T P l M i D I A T i - - - - - . . ; r - - o ...

l.or----Tr--;:;~....w'=<-=::;;tS;.;;;;:::I
STOKES
14 t1 .. 5"' OftDEft'
ChBltrcb!uU'1!
LAI!IOftATOAY

TEST DAT...

O.II--------+--'

aiRY

THIIORY

ClolNEARI

Reguiar waw ts

Irregular wave tests

Figure 1l1.2 Model II Waves

106

Ch&krabtiU'o
LA!iOOAT~'t

TEST DATA

O.II-------+__'

\.HAIT SOLITAI'lY

...V<

THIfORY
U.IN[Aft)

Irregular wave teats

Figure lll.3 Modelll-M Waves

These figures show that:

a. The models were generally tested in intermediate to deep water waves.
h. According to the analytical criterion of validity, the waves for all three models are
best described by the 2nd Order Stokes Theory. Based upon Chakrabarti's
experimental results, however, the Airy Theory is still applicable.
c. The 'working areas' in the irregular wave tests are near those in the regular wave
tests.
Airy Linear Wave Theory is chosen to describe the flow kinematics for all of the wave
states used; the 2nd Order Stokes Wave Theory is also employed with some steeper
regular wave conditions for comparison.
Since the models were tested in intermediate to deep water, the complete form of linear
wave theory is used.

107

Note that the wave kinematics predicted using the chosen wave theory is only valid in
the fluid field. Since the Linear Wave Theory was developed on the basis of simplified
free surface boundary condition, it does not provide accurate kinematics in the cresttrough region. The treatment of the kinematics near the free surface is discussed in
Appendix IV.

HU WAVE FORCES

Wave force types can be plotted against the relative wave height HID and the diffraction
number rr:D/ A (where D is the leg diameter and A the wave length) to give a rough
indication about the relative importance of drag versus inertia and drag versus
diffraction. For irregular waves, H and A. are replaced by Hs and Ap (where Ap is the
wave length computed using the peak frequency,fp). A reasonable assumption of the Cd
and em pairs of values are 1.0 and 2.0 for Model I and 1.5 and 1.5 for Models II and lIM. Using these data the relative importance of drag to inertia is summarized in figures
HlA through HL6 for each of the three models tested.

Regular wave tests

IZI
50

Irregular wave tests

Large Drag

10

Deep Water

Inertia
and
Drag

Small Drag

WorkLng Area

5.0

Breaking Wa.ve
Curve

D
1.0

Drag and
Diffraction

0.5

0.1

Negligible Drag

ri1~t!~t~~~

.05

interUa

.01

.05 o.I'lD/:!/..5

Diffraction
Region

1.0

5.0

10 .

Figure II/,4 Relative Importance Drag vs Inertia, Model I

108

Regular wave tests

~

50

Irregular wave tests

Large Drag

10+---~

5.0

Area

H
D

Deep Water

iiod~Tij"\
Working \

.....

Breaking VIave
Curve

~.' In=~~a

Drag

1.0

Drag and
Diffraction

0.5

Small Drag
Large lnterlia

0.1

.05 Negligible Drag 61;fg;!~~~~';,

Diffraction

Region

[ntertia
.01

.Ot

5.0

10.

Figure III.S Relative Importance Drag vs Inertia, Model II

Regular wave tests

~ Irregular wave lests

50

Deep Water

Breaking Wave
5.0

H
D
1.0

Curve

Inertia
and
Drag and

Drag

Diffraction

0.5

Small Drag
Large Intertia

0.1

.05

Negligible
Diffraction

Diffraction
Region

Figure III.6 Relative Importance Drag vs Inertia, Modeill-M

109

These figures show that:

a. With Model I, the hydrodynamic force is essentially inertia dominated in both
regular and irregular wave tests.
b. Models II and H-M work in the area where drag force plays a significant role.
The drag/inertia ratio with Model II is slightly higher than that with Model II-M.
c. The diffraction effect can be ignored with aU three models and thus the Morison
Equation is valid for the hydrodynamic force description. Note that the
negligence of the diffraction effect here refers only to the exclusion of the water
elevation and wave kinematics caused by the diffracted waves. The diffraction
effect on the hydrodynamic force is included in the inertia term.

Appendix IV Treatment of nonlinearitie and p-/j effed

The nonlinearities with the present physical models originate from structures themselves,
the hydrodynamics (free surface, drag term) and the wave-structure interaction (relative
motion). Connection nonlinearity cannot adequately be treated here; the model tests
were designed to investigate other phenomena; this has been discussed to some extent
in Appendix 11.1, however. All of the other nonlinearities together with the P-O effect
are discussed in this appendix.

IV.l FREE SURFACE

Although there exist some numerical schemes based upon the finite-amplitude wave
theory which are capable of predicting quite accurate kinematics for certain wave fields see Rienecker & Fenton (1981), Yuen & Lake (1982) and Sobey (1989), these are not
presented as explicit solutions and far too sophisticated to apply in practice. In problems
where the waves are not extremely high or where great accuracy is not required, it is
more reasonable to use an approximate explicit solution, such as Cnoidal Theory for
shallow water or Stokes Theory for deeper water. For practical problem, it is especially
desirable to modify tpe linear wave theory to improve the wave kinematics prediction
primarily in the crest-trough region where the correct kinematics information is most
essential for the offshore structure analysis and discrepancies between different wave
theories are also most obvious.
The linear Airy Wave Theory describes the water motion only up to the (constant
elevation) still water level (SWL). However, when the wave height is large relative to
the water depth, the effect of the changing free surface elevation on the total wave loads
(base shear and especially the overturning moments) becomes significant.
110

Four common approaches for computing the water motion kinematics up to the
instantaneous actual wave surface are briefly described as follows:
a.

Exponential Extrapolation
The velocity profile continues exponentially to the actual water surface. For
shallow water and high waves this 'primitive' approach is believed to yield very
conservative results - the predicted velocities and accelerations near the wave
crest will be too large - see Chakrabarti (1986).

b.

Vertical Uniform Extrapolation

The kinematics are kept equal to those at the SWL up to the wave crest when the
actual wave surface is above the SWL. Otherwise, standard Airy Theory is used
up to the actual water level, just as in method a, above - see Steele et at. (1988).
This approach is formulated as:
u(x,z,t)

for d ,;; z ,;; d

u(x,d,t)

+ 'fJ

(IV.l)

where:
d

water depth
instantaneous wave surface elevation measured from the SWL

'fJ

This method should be applied with caution as it can lead to overestimation of

loads in random waves; this is particularly true for the overturning moment
calculation.

c.

Linear Extrapolation
Like the vertical extrapolation profile, the linear extrapolation approach modifies
the direct exponential extrapolation profile approach only in the region under the
instantaneous crest and above the SWL, by replacing it with the linear Taylor
expansion above the SWL - see Forristall (1981):
u (x, z, t)

d.

au (x, d, t )
az

u (x, d, t) + (z - d) -

for d ,;;

Z ,;;

d + 'fJ (IV.2)

Stretching
The kinematics at the instantaneous free surface are considered identical to those
originally calculated for the still water level. Wheeler (1970) first introduced a
modification in such a fashion by mapping the vertical coordinate z onto a
computational vertical coordinate Zs :

111

Zs =

(_d)

(IV3)

d+r)

It follows that:

rcH
T

u =
in which:
u
H
T
k
if!

cosh kzs

(IVA)

cos if!

sinh kd

water particle horizontal velocity

wave height
wave period
wave number
time dependent phase

A slightly different alternative has been suggested by Chakrabarti (1971):

u=

reH
T

cosh kz
sinh k (d +

cos

if!

(IV.5)

r)

With this formulation the effective water depth is changed to d + fl. The
remaining kinematics between that free surface and the sea floor follows from
traditional linear theory as if it were being applied in the actual (instantaneous)
water depth.
These two stretching approaches produce the same kinematics at the free surface,
while the Wheeler stretching results in slightly larger values at any other point
downwards.
All four wave kinematics modification options as well as standard Airy Theory can be
used in NOSDA. Note that besides the modification models mentioned above, a great
deal of other work has been done in attempt to improve the prediction of the kinematics
near the free surface. Among these, Forristall (1981) demonstrates that the Wheeler
stretching and the linear extrapolation provides a lower and upper bound respectively for
horizontal velocities in the crests of waves. A combination of these two approaches leads
to the Delta stretching profile - see Rodenbusch and Forristall (1986). Other schemes
proposed for the free surface treatment include Gudmestad model (1990), Gamma
extrapolation model - see Borgman et al. (1989), and so on.
By comparing the
kinematics predicted using various free surface treatment approaches with the measured
results, Zhang, et al. (1991) indicated that there is not a crest-trough kinematic model
112

universally superior for all wave fields; the accuracy of the prediction of each
approximate method depends on the wave conditions. The present test setup was not
designed to evaluate these crest-trough kinematic models (the wave kinematics were not
recorded.) The waves tested were relatively low. The choice of the crest-trough
kinematic model is, therefore, not expected to be vital for the structural response
simulation. The Wheeler stretching profile is adopted here as the reference case for the
model simulations.
Unlike the basic linear wave theory, above, nonlinear wave theories compute water
particle kinematics up to the actual free surface. It should be emphasized that a higher
order nonlinear wave does not necessarily furnish a better prediction for the wave
kinematics, although it generally reproduces a better wave surface profile. Irresponsible
use of higher wave theories such as Stokes' Second through Fifth Order Theories for the
prediction of wave kinematics often leads conservative results - see Sobey (1989). Data
obtained from a structure in the Gulf of Mexico has verified this trend - see Bea and Lai
(1978).
It should be noted that inclusion of a free surface effect will, even with a pure sinusoidal
input wave, cause a skewness in the total hydrodynamic force on a leg. A simple
illustration with horizontal forces on a rigid vertical cylinder is given in figure IV.I. The
two total wave force extremes are always 180 degrees out of phase and occur at
symmetric points in the sinusoidal water surface profile.
z

o H/2
H /2

a. Inertia Dominant

SWL
.."":---l==I-~'---------r--

=<>

Fe

<)=>

b. Drag Domina.nt

Figure TV.1 Hydrodynamic Force Skewness

113

In an inertia force dominated case, the extreme wave loads occur in the vicinity of the
zero-crossing of the wave profile - within some small distance - from the SWL. The
extreme total wave forces on the cylinder are calculated using the following integrals
which extend from the sea bed to the actual water surface at the moment that the total
loading is extreme:
d,

Fmax

Jf dz
-d

(IV.6)

-Ai

F.
mID

f f- dz

-d

Where f + and are the values of the local wave load at the moment that an extreme
total load occurs and A; is measured relative to the SWL.
Since A; is small, the magnitudes of the maximum and minimum loads are almost the
same, and they act almost co-linearly in opposed directions, so that I F max I ..
I F min I and I1F; = F max - I F min I is small.
In contrast to this, for a drag dominated case, the maximum load occurs in the
neighborhood of the wave crest and the minimum in the neighborhood of the wave
trough. The extreme total wave forces on the cylinder are calculated now by:
d.

Fmax =

f f+ dz

-d

(IV.7)

-e. d
F.
=
mID

f- dz

-d

where I1d is again measured relative to the still water level.

Obviously, Ad is nearly equal to H /2 and much greater than A; , so that AFd = Fmax I Fmin I will be larger than above. Also, the resultant lines of action of F max and Fmin
are certainly not co-linear.
The wave climate situation will be between these two extreme cases for the models
tested here.

114

Apparently, hydrodynamic force skewness can be expected to cause skewness in the

response to this force as well.
When the vertical cylinder is non-rigid and relative instead of absolute velocities are used
in the Morison Equation, the above discussion will become much more complex. The
general, qualitative results (I F max I > I F min I and response skewness) will remain
valid, however.

IV.2 RELATIVE MOTION

The Morison Equation was originally intended for use with a fixed vertical cylinder in
wave. The extrapolated application of this equation to a structure moving in waves leads
to several differently revised forms. A commonly accepted approach is to base the
hydrodynamic computation on the relative velocity and acceleration:
(IY.8a)
or
(Iy'8b)
where:
fw
u

u
i

x
AD

AI
D
Pw

wave force per unit length of the vertical cylinder

horizontal component of water particle velocity
horizontal component of water particle acceleration
cylinder velocity
cylinder acceleration
VzDpw
V4 rrD2pw
cylinder outer diameter
ambient water density

In the computer simulation, the third term on the right hand of equation (IV.8b) is
moved to the left side of the equation of motion becoming the hydrodynamic force due
to the so called 'water added mass'; this is accounted in the computational model by
adding this portion of 'mass' to the 'dry' structural mass. For practical 'bookkeeping'
reasons, this is done only up to a constant elevation, the SWL. This approach introduces
an error in the splash zone, where the hydrodynamic mass of a given cylinder element
is continually changing. However, this error can be neglected with confidence - see
Massie, Liu and Boon (1989).

115

It is clear from equation (IV.S) that inclusion of relative motion has major consequences
Indeed, the entire computation of the external
for the numerical modeling.

hydrodynamic interaction now becomes dependent upon the (unknown!) velocity of the
structure. The proper structure motion will be that for which the computed response
agrees with the assumed response used in the computation of the hydrodynamic force.
In NOSDA this proper value - in terms of velocity - is determined by iteration. These
iterations are carried out several times for each simulation time step, and thus it more
than doubles the computational effort.

Hydrodynamic damping influences are automatically included using the relative velocity
model of the modified Morison Equation. The difficult task of estimating a somewhat
artificial equivalent damping value for a linearized system is avoided.

IV.3 QUADRATIC DRAG TERM

Quadratic drag introduces several complications from an analysis point of view. It

introduces a number of higher frequency harmonics in the wave force. A Fourier Series
development yields a series in which all even-numbered harmonics are zero. It also
shows that the third harmonic has an amplitude which is still 1/5 of that of the first
harmonic.
Unlike a frequency domain analysis, the treatment of drag in a time domain NOSDA
simulation is simple and straightforward. It requires no extra modeling or significant
computational effort.

IVA P-IJ EFFECT

The P- (j effect is the consequence from secondary moments generated as the deck load
becomes eccentric to )the vertical leg reaction forces during horizontal displacements.
It is modeled by including an extra set of special springs as defined in this section.
Examination of one leg segment subject to an initial, vertical compression load, P - see
figure IV.2 - with the nodal displacements u 1 and U 2 shows that the vertical load becomes
eccentric and therefore generates an overturning moment. This moment is balanced by
a horizontal force pair (FI , F 2).

116

Figure IV2 Detennination of P-O Spring Coefficient

The equilibrium equations are readily obtained:

{FI} _-I [1-1 -11 1{UzI}

P

F2

(IV.9)
= {-/}

(-Pfl) [-1

1]

{:J

Comparing this to the equilibrium equation for an extension spring (see Appendix V.2.2)
shows that it is identical except for a sign. As such, the P- 0 effect within the segment
can be modeled by a spring with a negative generalized rigidity matrix, Dg == - P/l , and
a kinematic matrix, Bg == [-1 1].
The applicability of the P-O modeling can be demonstrated by a simple example.
Assume a cantilevered beam subjected to a compression load, P, and discretized into two
segments - as in figure IV.3.
The equilibrium equation is expressed as follows:
1 -2

EI
[3

-2 6 -4
1 -4 3

{~1-7

1 -1

-1 2 -1
0

-1

{~l' {g)

Using the substitution:

117

1
~

Us == 0

7717777777111
Figure IV.3 Applicability of p-{j Elements

where:
(

EI
I

=:

coefficient
bending stiffness
segment length

yields:

-2+(
1
-2+( 6 - 2( -4+(
1
-4+( 3-(
1- (

EI
13

Setting u 3

{~l" m

(IV. 12)

0 then yields a second order algebraic eigen value equation:

(IV. 13)

This equation has two roots:

0.586
(

(IV.14)

3.414

The smaller root leads to the first order critical loading:

P

118

0.586 El
{l.

(IV. IS)

Compared with the corresponding result from Euler theory:

p =

rrEI
16 z2

(IV. 16)

or:

p= 0.617 EI

z2

(IV.I7)

It shows that with two segments the predicted result is already only 5% in error relative

to the theoretical value. With an increasing number of segments, the result predicted
in this fashion will approach and finally converge to the theoretical value.
In the actual modeling, the contribution of leg weight to the P- 0 effect is included by
summing all the node weights above the investigated segment. As a result of this, the
coefficients of the P- 15 elements decrease (become more negative) downwards along a
leg.

Appendix V Structural modeling

V.I DISCRETE ELEMENT METHOD

A model structure subjected to time-dependent hydrodynamic loads can be discretized
both spatially and temporally to perform a numerical structural dynamic analysis.
A multiple-degree-of-freedom Discrete Element Method (DEM) is used here to
discretize the structure in space. The DEM schematizes the physical object as if it were
composed of a finite number of discrete, undeformable elements interconnected by
massless, deformable springs and dampers. Lumped masses (or, more generally inertias)
are located to correspond with the degrees of freedom of the model. This
schematization is generally accepted for the overall dynamic analysis of large complex
structures such as jack-up platforms. It has been proven that for a linear system any
order of desired numerical accuracy can be obtained; the approach yields converging
results as the element size is decreased. For an arbitrary nonlinear system, this
convergency is not automatically assured, but its use for such systems has often been
successful in practice.

119

The discrete element method of spatial discretization provides great freedom in

modeling. However, as a price of this versatility, the approach demands a sufficient
knowledge of structural mechanics combined with user creativity.
After spatial discretization one obtains a structural motion equation having the fonowing
form:

M d 2X
dt 2

C dX + K X ; F
dt

X, dX)
dt

(V.l)

where:
X
t

C
K
F

structural
time
structural
structural
structural
structural

displacement vector
mass (or inertia) matrix
damping matrix
stiffness matrix
load vector

Note that when the system is linear, F is only a function of time, t. The above equation
need not be linear. This can be accommodated either by stipulating that M, C and K
need not be constant or by including higher order response-related terms in F.
For a jack-up model, the structural load vector, F, is determined from the hydrodynamic
analysis. Now, the problem remains of how to determine the M, K and C matrix values.
Basically, the DEM is a stiffness method which treats the nodal displacements as the
fundamental unknowns.
The DEM can be seen as a small and specialized 'handicraft shop' next to a big and
general 'supermarket', the Finite Element Method (FEM) - see Blaauwendraad and Kok
(1987). In the standard Finite Element Method, the analysis procedure is as follows.
The structural displacement field is expressed as a function of the nodal displacements:
(V.2)

where:
Uc

120

displacement field
interpolation function matrix
nodal displacement vector

thus, the strain vector can be written as:

IE =

(V3)

Bu

where:
IE

strain vector
displacement - strain relation matrix or kinematic matrix

The constitutive relation,

, links the strain vector,

IE,

and stress vector, u:

(VA)

For example, with the above relations, using the principle of virtual work, the
mathematical formulations for M,
and C for one leg element in the investigated case
result fromHxhe following three integrals:
I

M=

Ndz

K=

B dz

(V.5)

0
I

C=

rl

N dz

where:
p'

r'

element length
mass density per unit length
viscous damping coefficient per unit length

The Discrete Element Method chooses a different approach. The main difference is that
generalized strains, eg , and generalized stresses, ug , are applied instead of e and u, such
that integration over the area of an element is no longer needed. Consider an element
with m degrees of freedom (DOFs) and generalized displacement vector, FAg. If this
element contains i rigid-body DOFs, then there are n "" m - i DOFs left to determine the
deformations.
These deformations are the generalized strains, @'g ,Hxhile the
corresponding stresses are the generalized stresses, O"g. The node displacements and the
generalized strains are related via the
kinematic matrix, Bg . The generalized
strains and generalized stresses are related via the generalized rigidity matrix, Dg , (the
generalized constitutive relation). All of these relations can be expressed in formulas as:
121

(V.6)

(V.7)

The element stiffness matrix can readily be derived:

(V.8)

Similarly, a system damping matrix, C, can be computed by:

C = B/ Cg Bg

(VI.9)

Further explanations and derivations have been given by Blaauwendraad (1989).

Vol STRUCTURAL MODEL BUILDING BLOCKS

The establishment of a structural computational model is equivalent to choosing a set

of mass, stiffness and damping elements with proper characteristics, placing them in
proper relative locations and determining proper linkage. The details of the building
blocks - namely mass, stiffness and damping elements - are given in the following
subsections.

V.2.1 Mass Elements

The distributed mass of the structures is lumped at the nodes.
The mass of each of the model jack-up leg elements is divided equally and attached at
its two nodes. The mass contribution from one adjacent element of the cylindrical model
legs is given by:
(V.lO)

where:
D
p

122

outer diameter of the leg

wall thickness of leg
leg material mass density
element length

When the node is not located at the ends of the leg (field node), this mass value is
doubled in case of equal element length beHxuse the final value is the sum of the
contributions from two adjacent elements, while only one element contributes to the
concentrated mass if the node is located in the leg ends (edge node).
PVC plugs roughly 0.1 m long were mounted in the lower ends of the legs of Model I.
This extra mass is taken into account, even though this has only a minor effect to the
global dynamic behavior of the model.
The hydrodynamic or 'water added' mass for a submerged cylindrical element is:
(V. 11)

where Pw is the ambient water density.

Similarly, this mass is also equally divided and added to the corresponding node masses.
The effect of instantaneous elevation in the splash zone on the mass lumping is
neglected; constant "masses are used throughout. When an element penetrates the still
water level - see figure V.1 ~ the water 'added mass' is lumped to the two nodes as
follows:
ml
Where m
element.

2l - Is I
-2-1- m

(V.12)

is the 'water added' mass of the submerged portion of the splash zone

SWL
.-----------~-----

----------

m'

Figure V 1 'Water Added' Mass Lumping on a Splash Zone Element

The deck mass of each model comes from the frame, clamps and accelerometers.
Besides, with Model I and Model H-M, extra masses were added to the deck to obtain
the desired fundamental natural frequency or enhance the P- 0 effect. The deck mass
123

is lumped at the three corner nodes where the deck is connected to the legs. With
Model U( -M) the frame mass was measured. With Model I, however, this mass was
calculated from its dimensions and material densities, since the deck had already been
connected to the legs before starting the experiments. As the accelerometer was
installed on the stem bar of the deck frame, its mass is lumped only to the two nodes at
the ends of that bar.

V.2.2 SHffnes Eiement

The stiffness of each structures tested is modeled by a group of springs. Three types of
springs are used:
1. Extension Springs

Figure V.2 shows a spring before and after axial deformation. The extension, Liu ,
is taken as the generalized strain and the normal force, N , as the generalized
stress.

Before

After

k>O

UI

U2

=~~=
I tq
:
Uz
l6==$j
:
:'
b
~~~~o

Figure V.2 Extension Spring with Deformation Change

=*

=* =*

Bg

= [

-1

1]
(V.13)

D g =k

k Liu

thus:

1] =

124

[k -k]
-k k

(V. 14)

Note that since this spring element has only one generalized strain, Llu, the
generalized rigidity matrix, Dg , is a scalar.
An application of this type of spring is to model a bar with stiffness, k, loaded in
tension or compression:
k

EA

(V. IS)

=-

in which A is the cross section area, E the elastic modulus and I the length.
2. Bending Spring

This type of spring is mainly used to model the bending stiffness of a beam segment
located in the middle of the leg (field segment). (The treatment of edge segments located in the upper end of the leg - is given in section V.3.l.)
A beam section is replaced by a rigid bar which has two rotation springs at its ends.
In fact, each rotation spring can also be considered to be composed of two parallel
non-collinear extension springs.

1---112---1---123---1
Before

r------------ ---- -------

After

1____

.(M
----

--- - -- - - --- - - -- - -- -- - i
: U3

___ b

Figure V.3 Rotation Spring with Deformation Change

The generalized strain is now the angle, 8, and the corresponding generalized stress
is the moment M. For relatively small rotations:

125

e
M

U 2 -U 1

112

U Z -U 3

=> ==>

123

EI
1
1
-/12 + -l23
2
2

=> =>

1
Bg -_ [ - 112

1
123

-+112

-I~J

(V,16)

2EI

Dg

112

l23

When 112 is identical to /23 , then the element length, t, in the generalized difference
matrix, Bg , can be moved to the rigidity matrix, Dg , yielding:
B g =[-12

-1]

(V17)

D _ EI
g -

[3

and the stiffness matrix is:

1
2 -1]

-2

EI - 2 4
13

-2

(VI8)

1 -2 1

3, p- /) Spring

,..----O_p

Figure VA P- (j Spring with Deformation Change

Hence, again one has:

126

This type of virtual spring is used to model the

second order moment caused by the deck
weight (P-t; effect), Such a spring provides a
positive rather than a negative force as a result
of a positive displacement, It is of the same
form as the extension spring, except that its
elastic coefficient, k, and therefore the
generalized rigidity matrix, Dg (here this is also
a scalar), are negative,

=> -

=>

(Y.19)
-

=> =>

thus:

with k

= -

[-1]
1

k [-1

1]

[k-k -k]
k

(V.20)

PI!.

More details about the use of this type of spring and the derivation of k have
already been given in Appendix IV.4.

V.2.3 Damping Elements

Two types of specific damping elements are used:
1. Extension damper
2. Bending damper
Procedures similar to those used in the previous section to generate the stiffness matrices
for extension and rotational springs are also used for the generation of the damping
matrix. Here, displacements are replaced by velocities and strains by strain rates.

V.3 THEORETICAL MODELING OF CONNECTIONS

The connections at both ends of a leg have been highly simplified in the design of the
physical models, when compared to actual jack-up rigs. Even so, the preliminary
processing of the experimental data has already shown that the mechanical behavior of
these simplified connections was far more complicated than desired. Without losing the
vision of the connection deviation from their design, the modeling approach of ideal
connections is discussed in this section for the sake of theoretical completeness. In fact,
the idealized approach can be the most responsible approximation when the necessary
specific information on the connections is not available as in the present case.

127

V.3.1 The Deck ami Hs Leg ConnedHolll

The model deck consists mainly of a triangular frame of hollow, square bars. It is not
difficult to show that with all models both the extension stiffness and the bending
stiffness of the decks are at least one order of magnitude higher than those of the legs;
it is reasonable to consider the decks to be rigid - see Liu (1989b).
The deck is designed to be rigidly clamped to the legs. Under this ideal condition, the
bending spring linking a leg to the deck (edge node) is twice as stiff as a field spring
along the leg. TIle connections actually constructed are less rigid and more complicated
than the intended design; softer bending springs are used in the computational
schematizations for the models tested. Accompanying the bending springs, rotational
dampers are used to represent the (large) connection local damping.

V,3.2 lEouom COl!l.Il.ecHoD.

By design, the legs are perfectly hinged to the bottom plate. This is physically
implemented using universal joints. Theoretically the joint hinges provide no rotational
resistance (neither stiffness nor damping). In practice, it seems reasonable to model the
hinge as a rotational damper with a small damping coefficient. The connection between
the hinge and the bottom plate is modeled
two translational extension springs (one
vertical and another horizontal) and two corresponding translational extension dampers.
This is illustrated in figure V.5.

rotational
damper

Iranslalional
spring

Iranslalicmal
damper

Figure V5 Leg to Bottom Schematization

128