This action might not be possible to undo. Are you sure you want to continue?
Modeling and Control of Top Tensioned
Risers
Anne Marthine Rustad
THESIS FOR THE DEGREE OF PHILOSOPHIAE DOCTOR
Department of Marine Technology
Norwegian University of Science and Technology
2007
NTNU
Norwegian University of Science and Technology
Thesis for the degree of philosophiae doctor
Faculty of Engineering Science & Technology
Department of Marine Technology
c _Anne Marthine Rustad
ISBN 9788247140116 (printed ver.)
ISBN 9788247140253 (electronic ver.)
ISSN 15038181
Theses at NTNU, 2007:183
Printed at Tapir Uttrykk
Abstract
This doctoral thesis presents new research results on the control of top tensioned
risers in deep waters. The main motivation for this work is the development of
a riser control system which control the top tension of the individual risers in
an array, such that collisions between adjacent risers are prevented. The risk of
collision increases with increasing water depth, and as the oil and gas industry is
moving to ever deeper waters, riser interaction has become an issue of considerable
concern. The two main design parameters to avoid collision are riser spacing and
top tension, which both are expensive alternatives. Hence, new solutions are
needed and dynamic control of the top tension is the solution proposed here.
A mathematical model of the riser system is developed. It includes two risers
in a tandem arrangement which are connected to a tension leg platform (TLP)
through their top nodes, forcing the risers to follow the prescribed motions of
the TLP in the horizontal direction. In the vertical direction the riser motion is
decided by the actuator which is a tensioner system. The risers are also exposed
to current forces which are found by considering hydrodynamic interaction. The
risers are modeled using the ﬁnite element method (FEM). This model is veriﬁed
by the commercial software RIFLEX.
For the purpose of control applications, the model needs to be computationally
fast, but still be able to describe the main physics of the real system. The number
of elements needed to keep a desired level of accuracy is therefore investigated.
Diﬀerent means to measure the performance of the model are considered, and
among these, payout measurements reﬂected the displacement of the riser in the
best manner. In addition measurements at the wellhead, like payout and tension,
are both accurate and available at most installations today. Hence, the proposed
control objectives are mainly based on keeping the top tension, payout or total
riser length equal for the risers of concern.
A simulation study showed that the dynamic variation of the riser elonga
tion needed to be taken into account in the control algorithm for deep water
risers. The reason was that the large diﬀerence between tension in the two ris
ers resulted in a signiﬁcant length diﬀerence that had to be compensated for by
diﬀerent payout. In shallow waters the diﬀerence between the riser elongations
i
Abstract
is smaller, and equal payout is shown to be appropriate. Equal tension, which
is used by the industry today, could be applied for small current velocities. A
second control objective principle was to measure the relative horizontal distance
between the risers keeping it at a desired distance. This method showed promis
ing results, but is dependent on measurements that are not easily available today.
The best control objective used in the controller design was found to be equal
eﬀective length, meaning that the sum of the payout and the actual riser length
should be equal for all risers. Introducing this method may reduce the needed
spacing between the risers and thereby reduce the wellbay area on the TLP. A
simulation study with model based supervisory switched control showed how the
diﬀerent controller structures and parameters could be included in the feedback
loop depending on the operational conditions and riser behavior.
The main contributions in this work are the mathematical modeling of the
riser/TLP system, the model analysis, and the proposed controller architecture,
including the control objectives and the supervisory switched control concept.
ii
Acknowledgements
This thesis represents the main results of my doctoral studies from August 2004
through August 2007. The work has primarily been carried out at the Norwe
gian University of Science and Technology (NTNU) and partly at Massachusetts
Institute of Technology (MIT). My funding has been provided by the Research
Council of Norway through the Center for Ships and Ocean Structures (CeSOS)
at NTNU and the Leiv Eiriksson mobility programme.
The work of this thesis has been a true joint project, supervised by Professor
Asgeir J. Sørensen and Professor Carl Martin Larsen. I would like to thank Asgeir
for his guidance and enthusiastic encouragement, for always backing me up and
providing advices both within research and life in general when needed. I deeply
appreciate the guidance of and collaboration with Carl Martin. His expertise on
modeling and understanding of the physical system has been of invaluable help
and inspiration. I would like to thank him for always having his door open for
discussions throughout this project.
I am grateful to Professor Michael S. Triantafyllou and Dr. Franz S. Hover for
their hospitality and advices during my research stay at MIT from October 2005
through April 2006. Dr. Elizabeth Passano should be thanked for her patience
in questions concerning RIFLEX. Another thanks to the MSc students who have
suﬀered under my supervision; Andr´e H. Jacobsen for working on TLP modeling,
and Simen F. Stølen for cooperation on the riser veriﬁcation in shallow waters
which is included in this thesis. I appreciate the company of my oﬃcemates An
ders S. Wroldsen at NTNU and Kristoﬀer H. Aronsen at MIT. I would also like
to thank my colleges at CeSOS and the Department of Marine Technology in
particular the administration by Sigrid B. Wold and Marianne Kjøl˚as for organi
zation, Dr. Susan Swithenbank and Dr. Jamison Szwalek for proofreading, and
Kari Unneland, Jon Refsnes and Per Ivar Barth Berntsen for fruitful discussions.
Finally, I would like to express my gratitude to my friends and family for their
support and continuous motivation. A special thanks to my parents and my sister
Lene Annette for their love and support during these years with research.
Trondheim, August 2007 Anne Marthine Rustad
iii
Notation
Abbreviations
ACPM Accurate control plant model
CeSOS Center for Ships and Ocean Structures.
CFD Computational ﬂuid dynamics
CPM Control plant model
DNV Det Norske Veritas
DOF Degree of freedom
DP Dynamic positioning
ESM Error state maneuvering
FCPM Fast control plant model
FEM Finite element method
FPSO Floating production storage oﬄoading
GoM Gulf of Mexico
LF Low frequency
PDE Partial diﬀerential equation
PPM Process plant model
R1, R2 Riser 1, riser 2
RC Riser characteristics
ROC Riser operational conditions
SCR Steel catenary riser
TLP Tension leg platform
VIV Vortex induced vibrations
WF Wave frequency
WIO Wake induced oscillations
Characters
The Roman and Greek letters most frequently used throughout the thesis are
given here. Bold types are used exclusively to denote vectors and matrices. Bold
uppercase denotes matrices and bold lowercase denotes matrices for one riser
v
Notation
element or vectors. All symbols used in the speciﬁc contexts are explained when
ﬁrst introduced.
Roman
A, A
e
, A
i
Crosssectional, external and internal riser area
A
TLP
Amplitude of the harmonic TLP motion
C, c
i
System and element damping matrices
C
D
Drag coeﬃcient
C
M
, C
m
Inertia and added mass coeﬃcients
D, D
e
, D
i
Diameter, external diameter, internal diameter
E Young’s modulus of elasticity
f
ext
, f
int
External and internal force vectors
K, k
i
System and element stiﬀness matrices
K
D
, K
I
, K
P
Controller gains
l, l
r
Riser length
l
i
Riser element length
l
t
Tendon length
M, m
i
System and element mass matrices
n, N Number of elements
P
i
, P
top
Eﬀective axial tension in element i and the applied top tension
r Riser position vector
R
f
g
, R
y,x
Rotation matrices
T
f
i
Transformation matrix from i to fframe
T
j
Top tension of riser j
T
TLP
Period of the harmonic TLP motion
v
i
Current velocity in node i
x
i
Horizontal riser position of node i
x
TLP
TLP surge position
z
i
Vertical riser position of node i
w
eff
Eﬀective weight
vi
Greek
∆ Diﬀerence
γ Norm function
µ Monitoring error
ρ Process switching signal
ρ
f
, ρ
s
, ρ
w
Density of the internal ﬂuid, steel and water
σ Controller switching signal
τ Controller contribution
θ
i
Inclination of element i
ξ Payout
vii
Contents
Abstract i
Acknowledgements iii
Notation v
Contents ix
1 Introduction 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Previous Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Main Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.4 Organization of the Thesis . . . . . . . . . . . . . . . . . . . . . . . 10
2 Tension Leg Platform 13
2.1 Background for the TLP Solution . . . . . . . . . . . . . . . . . . . 13
2.2 The TLP Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.3 Existing Platforms . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3 Mathematical Modeling 23
3.1 Current Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.2 Hydrodynamic Interaction . . . . . . . . . . . . . . . . . . . . . . . 27
3.2.1 Shielding Eﬀects . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2.2 Wake Induced Oscillation . . . . . . . . . . . . . . . . . . . 28
3.2.3 Vortex Induced Vibrations . . . . . . . . . . . . . . . . . . . 29
3.3 Kinematics and Coordinate Systems . . . . . . . . . . . . . . . . . 30
3.4 Tension Leg Platform . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4.1 TLP Surge Modeling . . . . . . . . . . . . . . . . . . . . . . 32
3.4.2 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
3.4.3 Riser Stroke Calculations . . . . . . . . . . . . . . . . . . . 33
3.5 Riser FEM Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . 35
ix
Contents
3.5.1 Transformations for the Riser Elements . . . . . . . . . . . 36
3.5.2 System Mass Matrix . . . . . . . . . . . . . . . . . . . . . . 37
3.5.3 System Stiﬀness Matrix . . . . . . . . . . . . . . . . . . . . 39
3.5.4 Structural Damping . . . . . . . . . . . . . . . . . . . . . . 39
3.5.5 Hydrodynamic Forces . . . . . . . . . . . . . . . . . . . . . 40
3.5.6 Load and Equilibrium Iteration . . . . . . . . . . . . . . . . 41
3.5.7 TLP Prescribed Forces . . . . . . . . . . . . . . . . . . . . . 42
3.5.8 Dynamic Equation of Motion . . . . . . . . . . . . . . . . . 42
3.5.9 QuasiStatic Equation of Motion . . . . . . . . . . . . . . . 43
3.6 Actuator and Constraints . . . . . . . . . . . . . . . . . . . . . . . 43
4 Model Veriﬁcation 47
4.1 SetUp . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
4.2 QuasiStatic Veriﬁcation with Increasing TLP Oﬀset . . . . . . . . 49
4.3 QuasiStatic Veriﬁcation with Increasing Top Tension . . . . . . . 49
4.4 Veriﬁcation with Dynamically Moving TLP . . . . . . . . . . . . . 54
4.5 Veriﬁcation with Dynamically Varying Tension . . . . . . . . . . . 55
4.6 QuasiStatic Veriﬁcation at Shallow Waters . . . . . . . . . . . . . 58
4.6.1 Increasing TLP Oﬀset . . . . . . . . . . . . . . . . . . . . . 58
4.6.2 Increasing Tension . . . . . . . . . . . . . . . . . . . . . . . 59
4.7 QuasiStatic Veriﬁcation with Stress Joint . . . . . . . . . . . . . . 60
4.7.1 Shallow Water . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.7.2 Deep Waters . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.8 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
5 The Riser Control System Overview 67
5.1 Implementation Overview . . . . . . . . . . . . . . . . . . . . . . . 67
5.2 Actuator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.3 Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
5.3.1 Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5.3.2 TLP Motions . . . . . . . . . . . . . . . . . . . . . . . . . . 72
5.3.3 Tension and Payout . . . . . . . . . . . . . . . . . . . . . . 72
5.3.4 Top and Bottom Angles . . . . . . . . . . . . . . . . . . . . 72
5.3.5 Relative Horizontal Distance . . . . . . . . . . . . . . . . . 73
5.4 Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
6 The Controller Architecture 77
6.1 Control Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.1.1 Control Objectives Based on Measurements at the Top . . . 77
6.1.2 Control Objectives Based in Measurements Along the Riser 80
6.1.3 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
x
6.2 Riser Operational Conditions . . . . . . . . . . . . . . . . . . . . . 81
6.2.1 Riser Characteristics . . . . . . . . . . . . . . . . . . . . . . 81
6.2.2 Riser Operational Conditions . . . . . . . . . . . . . . . . . 83
6.3 Switched Systems  Concept and Properties . . . . . . . . . . . . . 84
6.3.1 System Properties . . . . . . . . . . . . . . . . . . . . . . . 85
6.3.2 Switching Logic . . . . . . . . . . . . . . . . . . . . . . . . . 85
6.3.3 ScaleIndependent Hysteresis Switching . . . . . . . . . . . 86
6.3.4 Model Concept Deﬁnitions . . . . . . . . . . . . . . . . . . 87
6.3.5 Switched System Stability Analysis . . . . . . . . . . . . . . 88
6.4 Supervisor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
6.4.1 Accurate Control Plant Model . . . . . . . . . . . . . . . . 90
6.4.2 Monitor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90
6.4.3 Switching Logic . . . . . . . . . . . . . . . . . . . . . . . . . 94
6.5 Controller Set . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.5.1 Guidance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6.5.2 Reference Model . . . . . . . . . . . . . . . . . . . . . . . . 96
6.5.3 Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
6.5.4 Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . 99
6.5.5 Integrator AntiWindup . . . . . . . . . . . . . . . . . . . . 99
7 Control Plant Model Analysis 101
7.1 Analysis Input Data and SetUp . . . . . . . . . . . . . . . . . . . 101
7.2 QuasiStatic Analysis with Increasing TLP Position . . . . . . . . 102
7.2.1 Error Norms for Horizontal Positions . . . . . . . . . . . . . 103
7.2.2 Riser Top Angle . . . . . . . . . . . . . . . . . . . . . . . . 106
7.2.3 Area Under Curve . . . . . . . . . . . . . . . . . . . . . . . 107
7.2.4 Payout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
7.2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
7.3 Analysis of QuasiStatic Tension Variations . . . . . . . . . . . . . 112
7.4 Dynamically Moving TLP . . . . . . . . . . . . . . . . . . . . . . . 118
7.5 Analysis of Dynamic Variation in Top Tension . . . . . . . . . . . 119
7.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
8 Simulation Results 125
8.1 Set Up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
8.2 Control Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
8.2.1 Constant Equal Tension . . . . . . . . . . . . . . . . . . . . 126
8.2.2 Equal Payout . . . . . . . . . . . . . . . . . . . . . . . . . . 127
8.2.3 Equal Eﬀective Length . . . . . . . . . . . . . . . . . . . . . 129
8.2.4 Desired Horizontal Distance . . . . . . . . . . . . . . . . . . 133
8.3 Eﬀect of Shallow Water . . . . . . . . . . . . . . . . . . . . . . . . 135
xi
Contents
8.4 TLP Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
8.5 Supervisory Switched Controller . . . . . . . . . . . . . . . . . . . 140
8.5.1 Case SetUp . . . . . . . . . . . . . . . . . . . . . . . . . . . 140
8.5.2 Simulations without Control . . . . . . . . . . . . . . . . . 141
8.5.3 Simulations with Control . . . . . . . . . . . . . . . . . . . 142
8.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
9 Concluding Remarks 149
9.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
9.2 Proposals for Further Work . . . . . . . . . . . . . . . . . . . . . . 151
Bibliography 153
A Mathematics and Algorithm for the FEM Model 165
A.1 Wake Field of a Single Cylinder . . . . . . . . . . . . . . . . . . . . 165
A.2 System Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
A.2.1 Mass . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
A.2.2 Stiﬀness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
A.2.3 Damping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169
A.3 Load on the Risers . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
A.4 QuasiStatic Algorithm . . . . . . . . . . . . . . . . . . . . . . . . 174
A.5 Metrical Norms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
B Simulation Data 183
B.1 Environmental Data . . . . . . . . . . . . . . . . . . . . . . . . . . 183
B.1.1 Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
B.1.2 Tide . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
B.1.3 TLP Oﬀset and Dynamics . . . . . . . . . . . . . . . . . . . 184
B.2 Riser Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
B.3 Controller Gains . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
xii
Chapter 1
Introduction
Oil and gas are by far Norway’s largest export industries, being more than 50%
of the export value in 2006 (Ministry of Petroleum and Energy and the Nor
wegian Petroleum Directorate, 2007). The ﬁrst oil in Norwegian territory was
found in 1969 at the Ekoﬁsk ﬁeld, located at 7075m water depth. The ﬁrst
production started there in 1972. Since then, knowledge and experience have
developed, and the petroleum industry in Norway presently has expertise on
oﬀshore installations in harsh environment and deep water, which is based on
disciplines like hydrodynamics, structural mechanics, oceanography, automatic
control, material science, etc. However, as new oilﬁelds are explored and devel
oped, new challenges arise and new knowledge and research are needed. New
contributions are searched for within each discipline. It is believed that by inte
grating the diﬀerent disciplines, further progress can be made. At the Center for
Ships and Ocean Structures (CeSOS) at the Norwegian University of Science and
Technology (NTNU), the disciplines of hydrodynamics, structural mechanics and
automatic control are integrated as a strategy to contribute to the innovation of
ships and ocean structures. New approaches will hopefully result in new solu
tions to currently unsolved problems. This thesis is situated at the intersection
between structural mechanics and automatic control, but also some hydrodynam
ics are included. This is illustrated in Fig. 1.1 with a cross marking where this
thesis can be found disciplinewise. Working in a multidiscipline ﬁeld creates
new problems, but also opens for new possibilities.
1.1 Motivation
Oﬀshore petroleum production began in the most shallow waters. As these
reservoirs are exploited together with an increasing demand for oil and gas, the
petroleum industry has been moving to ever deeper waters. Riser technology is an
1
1. Introduction
Structural
Mechanics
Hydro
dynamics
Automatic
Control
Figure 1.1: Intersection of the diﬀerent disciplines found in this work.
important issue both when considering ﬁeld development costs and technological
feasibility. In deeper waters interference between adjacent top tensioned risers in
an array is an issue of considerable concern (DNV, 2005). Collision may lead to
dents in the riser pipe and also damage in the coating, with fatigue and corrosion
as possible consequences. Even a single collision event may be damaging if the
collision takes place with suﬃciently high impact.
If the riser spacing and properties are kept constant, the risk of collision
will increase with increasing water depth, since the static deﬂection due to the
uniform current drag is proportional to the square of the length (Huse, 1993).
In deep waters this means that even a relatively small diﬀerence in static forces
may lead to mechanical contact. In addition ﬂow separation and shielding eﬀects
between risers in an array can change the local ﬂow velocity. This diﬀerence
in current forces may cause large relative motions and lead to contact between
neighboring risers. There are mainly two design parameters that will prevent
collision between risers:
1. Increasing the riser spacing.
2. Increasing the top tension.
Both may result in signiﬁcant cost penalties. Increasing the riser spacing means
increasing the size of the wellbay (Fig. 1.2). This has inﬂuence on the global
platform parameters like deck space and load carrying capacity. The other option
is increasing the top tension to a high and constant level, which will increase
the wear and tear on the cylinders in a heave compensation system. The main
objective of this work is to prevent, or at least reduce, the number of collisions
by applying control of the riser top tension.
For deep water production systems, riser solutions are traditionally divided
into two main groups; (1) subsea wellheads (wet trees) with ﬂexible risers up
to a ﬂoater like a semisubmersible or a production ship (ﬂoating production
2
Motivation
Figure 1.2: The wellbay on the TLP Snorre (www.statoil.com).
storage oﬄoading  FPSO), and (2) tensioned risers with wellhead on a compliant
platform, like tension leg platform (TLP), spar
1
or deep draft ﬂoater (DDF).
These are the dry tree solutions for ﬂoating production systems. Dry tree systems
are often the preferred solution for production as they provide easy access to the
well for maintenance, intervention and workover. Alternative platform solutions,
while still using dry tree systems, are proposed by Often (2000) for a semi
submersible with heave compensation system, and by Mortazavi et al. (2001)
for deep draft caisson vessel with buoyancy cans. Pollack et al. (2000) suggest
a weight based tension leg deck (TLD) to which the risers are locked for a dry
tree solution on FPSOs. This solution allows long stroke for spread moored
shipshaped vessels. Wanvik and Koos (2000) present a two tier well riser tension
system, which is a mechanical alternative to hydraulic cylinders working in series.
This system separates the low frequency stroke motion from the wave induced
stroke motion.
Top tensioned risers operated from spars and TLPs are arranged in clusters
of (near) vertical riser arrays. The number of individual risers in an array may be
20 or more, which may consist of diﬀerent risers applied for production, drilling,
workover, export, etc. The problem of riser collision can be restricted to tensioned
risers operated from a TLP or a spar located in deep waters. This is due to the
small space available for the high number of risers, in addition to the increased
1
A spar is a vertical, cylindrical buoyant platform, usually manned.
3
1. Introduction
deﬂection for long risers.
A TLP is chosen as the platform concept for this work. The main reason is
that tensioned risers can be applied at a TLP with a relatively small requirement
for stroke capacity even at large TLP oﬀsets. This is a consequence of the ge
ometric restrictions for heave motions caused by the axially rigid tendons. For
other ﬂoaters like spars, FPSOs and semisubmersibles, the demand for stroke
capacity will be much higher, which means that other riser solutions like steel
catenary risers (SCR) or ﬂexible risers are preferred. Hence, the control system
presented in this thesis is most likely to be applied on a TLP.
Deep water TLPs are mainly found in the Gulf of Mexico (GoM), but there
are also some west of Africa and oﬀ the coast of Brazil. A TLP solution was
considered at the Ormen Lange ﬁeld 100km west of midNorway at 850m water
depth, but a system with subsea wells and pipelines to the onshore plant was
chosen. This solution is possible for a reservoir mainly consisting of gas. For a
reservoir with heavier hydrocarbons this solution may not be possible. Ellida is
a large oilﬁeld 60km north of Ormen Lange and located at 1200m water depth.
The ﬁeld was discovered in 1997, but it is unknown if the ﬁeld is commercial
viable. However, if it is decided to develop the ﬁeld, a TLP could be a possible
solution here.
1.2 Previous Work
Most research within the ﬁeld of collision between risers focuses on the hydro
dynamic interaction between them. In addition some work has been done on
the actual collision forces and the damage it may cause. Little or nothing has
been done to actively control the riser motions. Dynamic behavior and collision
of risers have been studied by several scientists the last decade. DNV (2005)
gives a thorough introduction to riser interference, and the deﬁnitions given in
the recommended practice will be followed here.
Each top tensioned riser in an array is exposed to environmental loads from
waves, currents, and forced ﬂoater motions, both at wave frequency (WF) and low
frequency (LF). Whether or not collision between two neighboring risers occurs
depends on factors like top tension, riser spacing, ﬂoater oﬀset, environmental
loads, hydrodynamic interaction, and riser properties to mention some. Direct
wave loads on the upper part of the riser and ﬁrst order ﬂoater motions are
assumed insigniﬁcant for such heave compensation systems, because the collision
is most likely to occur in the midsection of the riser, where the deﬂection is
largest due to current loads. The assessment of riser interference is therefore
mainly based upon the assumption of a steady state current proﬁle.
Little information regarding interaction eﬀects due to wave loading is found
4
Previous Work
(Duggal and Niedzwecki, 1993), while signiﬁcant eﬀort has been applied to inves
tigate hydrodynamic interaction in steady current, see Huse (1987, 1993, 1996),
Huse and Kleiven (2000) and Kavanagh et al. (2000). A reference should also
be made to Blevins (1994) and Zdravkovich (2003) for numerous experiments on
hydrodynamic interaction between cylinders. Note that within the ﬁeld of hy
drodynamics, cylinders is the general term used to denote marine risers, tendons,
cables, etc. For more industrial studies, the term riser is often used. We will
strive to use the terms in the same way as in each referred paper. Zdravkovich
(1977) gives a careful review on ﬂow between two circular cylinders in various ar
rangements in steady ﬂow. Zdravkovich (2003) presents arrangements with three
or more cylinders, and an extensive list of references may be found here.
Experiments by Tsahalis (1984), Bokaian and Geoola (1985), and recent stud
ies by DNV (2003) and Kalleklev et al. (2003) have shown that interaction be
tween neighboring risers will not have any hydrodynamic inﬂuence on the up
stream riser beyond a certain distance. Furthermore, the literature distinguishes
between three diﬀerent kinds of forces acting on the downstream riser in the wake
of an upstream riser. These eﬀects are (DNV, 2005):
• Mean force and shielding eﬀects bringing the risers closer together.
• Wake induced oscillation (WIO) on the downstream riser.
• Vortex induced vibrations (VIV) resulting in ampliﬁed drag coeﬃcients on
both risers.
The mean forces are modeled by two diﬀerent methods; the parametric wake ﬁeld
model and the parametric mean force model. Both can be applied in ﬁnite element
method (FEM) models. The parametric wake ﬁeld model is a semiempirical
static wake formulation which accounts for the interaction between stationary
individual cylinders in steady current. This was studied by Huse (1987, 1993,
1996), and is based on the analytic expression of a turbulent wake by Schlichting
(1968). The inﬂow of a cylinder situated in the wake of an upstream cylinder
can be computed at any location. The drag force is computed by taking the
actual velocity into account, but keeping the drag coeﬃcient constant. A mean
transverse force is formulated for cylinders more arbitrarily placed in the wake.
In Huse (1993), a scheme for calculating the inﬂow of an array of cylinders is
proposed.
The parametric mean force model was derived by Blevins (1994). In this
method the hydrodynamic forces are accounted for by introducing mean drag
and lift forces on the downstream cylinder as a function of the distance from
the upstream cylinder. The mean lift force is directed towards the wake cen
terline, where the current velocity is smallest. The drag and lift coeﬃcients in
5
1. Introduction
the parametric mean force model need to be established by model tests or by
two dimensional numerical methods like computational ﬂuid dynamics (CFD).
Gopalkrishnan et al. (1994) and Wu et al. (1999) have investigated the drag and
lift forces in steady current. The purpose of both studies was to gain necessary
information in order to assess the onset of potentially interference problems and
collision. Blevins (2005) developed a model for steady lift and drag forces on
a cylinder in the wake of an upstream one by using theoretically based equa
tions ﬁtted to data from experiments. Wu et al. (2001a,b) investigated the onset
criteria for wake induced instability, assuming two identical cylinders in steady
uniform ﬂow. The interest for tendon and riser interference in the work of Tsa
halis (1984) and Gopalkrishnan et al. (1994) is related to Shell’s activity on deep
water TLPs like Auger and Mars. An ongoing research program has been in
progress since 1989 (Allen et al., 2005). Most of these results are published in
conﬁdential reports, and have not been accessible to the public.
The parametric mean force model is also able to capture WIO. The study
of WIO dates back to the early 1970’s and the study of power transmission
lines. References could be made to for instance Simpson (1971), Price (1975)
and Tsui (1977). For marine applications, these motions were ﬁrst observed
and described by Huse (1996) and later by Wu et al. (2002). The risers are
exposed to current over a very long length relative to the diameter, and the
ﬂow around the risers will cause large hydrodynamic interaction forces. These
may cause slowly varying, large amplitude motions of very irregular behavior,
occurring about the ﬁrst natural mode of motion. Such motions may also occur at
signiﬁcantly lower frequencies, which means that they are quasistatic and caused
by unstable hydrodynamic forces. WIO are in any case found at frequencies
substantially lower than the vortex shedding frequency. Lately, Fontaine et al.
(2007) reproduced WIO in model scale and identiﬁed the diﬀerent regimes (stable,
unstable and critical). The onset criteria for WIO and clashing between SCRs
were proposed.
A slightly diﬀerent approach on the parametric wake force model was proposed
by Sagatun et al. (2002). The basis in this model was a parametric force rep
resentation for both upstream and downstream riser based on coeﬃcients found
from CFD. In addition to mean forces, dynamic forces are described in the time
domain using position dependent force parameters. The objective was to assess
whether or not adjacent risers move in the wake of an upstream riser. Sagatun et
al. (2002) postulated that the slowly varying WIO controlled the relative position
of the risers, while VIV with higher frequencies and velocities accounted for most
of the energy in a riser collision. However, note that the large riser displacement
governed by mean force and WIO are found to be unaﬀected by VIV, except
from the magniﬁed drag force (Tsahalis, 1984, DNV, 2003). Tsahalis (1984) also
found that the VIV responses are diﬀerent for the upstream and downstream
6
Previous Work
risers. The VIV response for the downstream riser is given by the mean water
velocity of the riser, which is diﬀerent from the velocity on the upstream riser
due to the wake eﬀect.
Kavanagh et al. (2000) did VIV model tests and outlined how riser interference
for deep water risers was assessed using a combination of wake ﬂow modeling,
global interference analysis and VIV prediction. Allen et al. (2005) investigated
the riser interference on ﬂexible risers, with focus on VIV and Reynolds number.
VIV is in itself subject to research as it is important for fatigue as well as to
calculate drag forces on the riser, see for instance Kaasen et al. (2000) and Halse
(2000). For the ampliﬁed drag coeﬃcient, see Sarpkaya and Shoaﬀ (1978) and
Vandiver (1983).
The design practice of today does not allow collision neither under normal
nor extreme conditions (DNV, 2005). As keeping the risers clustered in an array
with small riser spacing is an advantage with respect to both economy and op
eration, the industry is considering accepting that riser collision occurs. Nyg˚ard
et al. (2001) investigated the stress levels generated by the impact between two
neighboring risers. The result indicated acceptable stress levels and served as
a ﬁrst step to allow some impacts to occur. Herfjord et al. (2002) presented a
numerical model for the simulation of the interaction and clashing of ﬂexible ris
ers. The work includes methodology for contact dynamics of interacting risers
and recorded riser impacts, collision statistics and stress analysis. The impact of
collision and participating mass for colliding risers is also treated in Sagatun et
al. (1999) and Neto et al. (2002). Leira et al. (2002) have used the method from
Sagatun et al. (2002) in a case study to estimate extreme response and fatigue
damage for colliding risers. Kalleklev et al. (2003) proposed a design practice
that properly accounts for collision between risers in operating conditions typical
for major deep water locations. Lately, Baarholm et al. (2005) did model tests
to better understand the mechanics that drive the risers to collision, including
relative velocity at the time of collision which is a measure on its intensity. Both
bare risers and risers with VIV suppression (strakes) or bumpers along the col
lision zone were tested. These measured results provided benchmarks for code
validation of riser collisions. Little has been done on experiments for more than
two risers. However, Huang et al. (2004) examined the wake shielding for three
risers in a towing tank. Signiﬁcant drag reduction on the downstream cylinders
was found.
During the last decade, some doctoral theses on modeling and control of
cables have been written at NTNU. Fard (2000) investigated vibration control of
ﬂexible mechanical systems (i.e. VIV on marine risers) with focus on passivity
of nonlinear beams. Aamo (2002) applied FEM to model and analyze passivity
properties of mooring lines, and found that passive controllers could be applied for
dynamic line tensioning in a mooring system. T¨ urkyilmaz (2004) modeled a towed
7
1. Introduction
seismic cable and investigated its passivity properties. Ersdal (2004) developed
a FEM model for towed ﬂexible cylinders and investigated the hydrodynamic
forces on cylinders in axial ﬂow through experiments. Fredheim (2005) worked on
current forces on net structures, including experiments on drag and lift forces on
an array of risers. Johansen et al. (2006) presented a new model for dynamics of
inextensible cables, by separating axial and transverse motion. Lately, Wroldsen
(2007) has been working on cable models based on diﬀerential algebraic equations
for eﬃcient computation and realtime purposes.
Control of risers is in its infancy and little is published. An exception is the
work of Sabri et al. (2003) who use active control to make it possible to connect
a riser to a wellhead in harsh weather. A reduced model for the behavior of the
bottom riser end and a control law were developed. Only position measurements
at the bottom riser end and the control action at the top are needed to follow
the reference trajectory. This method is patented, see Sabri et al. (2006).
For drilling operations, the top and bottom riser angles are of concern, and
should not exceed an upper limit of 24 degrees during drilling. To achieve this,
the position of the drilling vessel is optimized by including the inclination mea
surements in the feedback loop for the dynamic positioning (DP) system. The use
of riser angle positioning system (RAPS) was ﬁrst investigated by Dean (1980).
The system was successfully tested aboard the drillship Discoverer Seven Seas
in 1979. Suzuki et al. (1995) used modal expressions for a riser at 4000m water
depth with current loads in the upper 1000m. To control the riser angles, actua
tors like thrusters or remote operated vehicles (ROV) were suggested to be used
along the riser. The DP system for drilling vessels with automatic control of the
riser end angle (REA) was proposed by Imakita et al. (2000). The objectives
were (1) to minimize the REA by vessel positioning and (2) to perform ship po
sition based on REA measurements with riser angle sensor and neural network.
Ship positioning sensors were not used as the proposed system should act as a
backup for commercial available positioning systems, either satellite based or hy
droacoustics with ﬁxed reference systems on the seabed. Sørensen et al. (2001)
proposed an optimal setpoint chasing algorithm for drilling vessels in deep wa
ters. The main objectives were to minimize the bending stiﬀness along the riser
and the riser angles at the top and bottom joints. Sørensen et al. (2002) were
also motivated to use riser angle as a reference system for drilling risers, in ad
dition to the existing systems. Improving the quality of position measurements
will increase the availability of the drilling vessel in bad weather, and give a fast
payback of the investment. The design principle of a riser angle positioning ref
erence system (RAPR) with an observer model is proposed. The riser model is a
quasistatic FEM model and accounts for varying current loads and top tension.
The requirement for redundant positioning references motivated also Høklie et
al. (2002). They estimated vessel positioning by measuring the riser angles and
8
Main Contributions
using Kalman ﬁltering technics. A dynamic mathematical model is tuned by
combining the vessel position measurement and riser inclinations and use this to
estimate the current proﬁle. After the initial tuning, the model is able to estimate
the vessel position based on the riser inclination measurements and the estimated
current proﬁle. The system was initially tested with success aboard the drill ship
Saipem 1000 west of Africa at 1100m and 2300m water depth in June 2002.
Active heave compensation is used today under drilling and in crane opera
tions for oﬀshore equipment. The ﬁrst is commercially available, for instance by
Aker Kværner (2007) and National Oilwell Varco (2007a). Korde (1998) investi
gated an actively controlled heave compensation aboard deep water drillships sub
jected to irregular waves. Active control of a heave compensated crane during the
water entry phase for the equipment is addressed by Sagatun (2002). Johansen
et al. (2003) made model experiments where the strategy was to synchronize the
load with the waves in the moon pool. Recently Perez and Steinmann (2007)
performed an analysis of operability and constraints in terms of vessel motions
and heave compensator stroke capacity.
However, heave compensation of top tensioned production risers has not ear
lier been investigated. The top tension is today kept constant by passive systems.
In order to avoid collisions Huse and Kleiven (2000) proposed a strategy based on
equal payout for risers in an array. All risers were connected to a common frame
in the top end, and no collision could be observed in steady current. The work
in this thesis is motivated by the work of Huse and Kleiven (2000). But instead
of keeping the risers ﬁxed to a plate, the top tension of each riser is controlled
individually using the payout of the heave compensator as the measured input.
1.3 Main Contributions
The main contributions found in this thesis are summarized here. Parts of this
thesis have earlier been published. References to these publications are given
below.
• Firstly, the work in this thesis has shown that it is possible to prevent
collision between neighboring risers by use of top tension control and mea
surements at the riser end and wellhead onboard the TLP. This was ﬁrst
shown in Rustad et al. (2006), and later in Rustad et al. (2007b,c). Control
of risers by changing the top tension has not previously been published.
• A twodimensional model for the TLP and riser system exposed to current
is developed. How the speciﬁed motion from the TLP acts on the risers is
not earlier published. This modeling is found in Chapter 3 and is published
in Rustad et al. (2007c).
9
1. Introduction
• The mathematical system model is veriﬁed by the commercial software
RIFLEX (Fylling et al., 2005) in Chapter 4. Parts of this are found in
Rustad et al. (2007c). An extensive analysis on how many elements are
needed for the various applications and purposes is found in Chapter 7.
This has not been published earlier.
• Four diﬀerent riser control objectives are formulated. These include two
existing objectives; one used in the industry today and one earlier proposed
by Huse and Kleiven (2000). In addition, two new control objectives are
proposed here. These are presented and investigated in Chapters 6.1 and
8, respectively, and could to a large extent also be found in Rustad et al.
(2007b,c). The control objectives are presented for two risers in a tandem
arrangement, but can be expanded to three dimensions and a matrix of
risers.
• A model based, supervisory switched control system is proposed and simu
lated in Sections 6.4 and 8.5, respectively. This case study is also presented
in Rustad et al. (2007a).
1.4 Organization of the Thesis
The outline of the thesis is as follows:
Chapter 2 introduces the TLP concept and the background for the TLP solu
tion. The most famous existing platforms are presented.
Chapter 3 presents the mathematical modeling of the system. This consists of
two risers in a tandem arrangement, exposed to TLP motions in the top
end and current forces. The riser tensioner system, being the actuator in
the system is also modeled.
Chapter 4 veriﬁes the riser model from Section 3.5. The validation of the model
for diﬀerent water depths and bottom end connections is investigated.
Chapter 5 gives an overview of the physical implementation of the system being
the actuator, measurement and controller system design.
Chapter 6 presents the controller architecture. This includes the control objec
tives, riser operational conditions and the diﬀerent controllers.
Chapter 7 analyses the control plant model thoroughly. Both the quasistatic
and dynamic models are simulated with varying TLP position and top
tension.
10
Organization of the Thesis
Chapter 8 shows the simulation results. These includes investigation of the dif
ferent control objectives and a case study with supervisory switched control.
Chapter 9 summarizes the main conclusions of this thesis and proposes further
work.
Appendix A includes additional details of the mathematical modeling, not in
cluded in Chapter 3.
Appendix B gives the environmental data, riser data and controller gains used
in the simulations in Chapter 8.
11
Chapter 2
Tension Leg Platform
The main purpose of the TLP model in this thesis is to represent the platform
motions at the wellhead area where the risers are linked to the platform. The
surge motions will act as a prescribed dynamic boundary condition in the riser
analysis, while heave relative to vertical motions of the upper riser end is sub
jected to active control. Heave motions of a TLP have two components. One
is cause by dynamic elastic strain and deﬂections of the tendons, which will in
ﬂuence the true distance between its ends. The other component is referred to
as surge induced heave, and is easily understood by realizing that the inclined
tendon will have another vertical distance from the seaﬂoor than its true length.
The motions caused by local tendon dynamics are signiﬁcantly smaller than the
surge induced component  in particular in extreme wave conditions  and is ne
glected in this work. This was done for the case of simplicity, but could have been
included without any conceptual changes in the approach. Conclusions from this
study are hence valid even if this simpliﬁcation was made. This chapter is mainly
based on Demirbilek (1989b), Faltinsen (1990) and Larsen (1995).
2.1 Background for the TLP Solution
Oﬀshore platform concepts are usually classiﬁed into two major categories; ﬁxed
and compliant. Fixed platforms stand at the seabed and remain in place by a
combination of their weight and/or piles driven into the soil. Little or no motions
are observed for such structures. The ﬁxed platforms resist the environmental
forces like wind, wave and current by generating large internal reaction forces.
The ﬁrst oﬀshore oil and gas ﬁelds were found in shallow waters, and mainly
ﬁxed structures were build in the ﬁrst decades of oﬀshore oil production. Fixed
platforms will normally have natural periods shorter than 5s and their responses
which are caused by extreme waves will therefore be quasistatic since such waves
13
2. Tension Leg Platform
typically will have periods above 10s.
As oﬀshore development moved towards deeper waters, the application of con
ventional ﬁxed jackets approached its limits principally imposed by the dynamic
behavior of the structures (Litton, 1989). With increasing water depths, ﬁxed
platforms become more ﬂexible, and their natural periods started to enter the
high energy levels of the ocean waves. To keep the eigenperiods away from this
damaging range, ﬁxed structures had to be designed to be stiﬀer, requiring more
steel and exponentially increasing costs.
The obvious alternative to ﬁxed platforms in deep water is to use compliant
platforms. The basic idea is to allow for rigid body motions with eigenperiods
longer than wave periods (T > 30s). Compliant platforms may be divided into
three types; ﬂoaters, towers and TLPs.
A ﬂoating platform will have all its structural eigenperiods well below 5s,
which means that ordinary wave loads will not give any structural dynamic re
sponse. However, dynamics of rigid body motions for ﬂoaters are of concern.
Eigenperiods are controlled by the geometry of the waterplane (heave, pitch and
roll) and design of the anchor system (surge, sway and yaw). These eigenperiods
are normally above wave periods, but resonance may still occur due to wind and
higher order wave forces. Hydrodynamic damping is the key to reduce these types
of response. Any anchored vessel, semisubmersible, spar and FPSO belong to
this category.
Towers are ﬁxed to the seaﬂoor by an arrangement that eliminates, or at least
reduces, the bending moment at the bottom end. Horizontal motions of the deck
are compliant, but vertical motion components are equivalent to a ﬁxed platform.
A TLP acts like a ﬂoater with regard to inplane motion components (surge,
sway and yaw), but like a ﬁxed platform for the outofplane components (heave,
roll and pitch). The governing parameters for the inplane stiﬀness are the tension
and length of the tendons, while the cross section area, modulus of elasticity and
length of the tendons decides outofplane stiﬀness. The design premises for the
tendon system are hence to determine these parameters so that desired values of
the eigenfrequencies can be obtained.
In addition to considering the sensitivity of the platform to external environ
mental forces, the choice of platform system depends upon other considerations.
Amongst these are the technical and economic factors, including water depth,
production rate, reservoir size, service life and removal requirements. However,
in deep waters the major factor in selecting the platform for a ﬁeld development
plan is the cost which is correlated to its weight. The cost of the ﬂoating sys
tem, especially that of the TLP, are relatively insensitive to changes in the water
depths compared to the cost of the compliant towers and ﬁxed structures (Lit
ton, 1989). Hence, economic factors make the compliant platforms and the TLPs
in particular one of the leading candidates for major deep water developments,
14
The TLP Concept
especially when a dry tree solution is preferred.
2.2 The TLP Concept
The basic idea behind the TLP concept was to make a platform that is partly
compliant and partly rigid. In order to maintain a steel riser connection between
the seaﬂoor and the production equipment on the platform, heave, roll and pitch
motions had to be minimized and are rigid degrees of freedom. The horizontal
forces due to waves on the vertical cylinders will always be larger in the horizontal
plane than in the vertical direction. Hence, by making inplane motions compli
ant, the largest environmental forces can be balanced by inertia forces instead of
by forces in rigid structural members. This idea can be realized by a pendulum
using buoyancy to reverse gravity forces (Larsen, 1995).
The TLP is deﬁned as a compliant structure, but might also be classiﬁed as
a moored structure. In general, the TLP is similar to other column stabilized
moored platforms with one exception; the buoyancy of a TLP exceeds its weight,
and thus the vertical equilibrium of the platforms requires taut moorings con
necting the upper structure to a foundation at the seabed. These taut mooring
are called tension legs, tethers or tendons. Drilling and production risers con
necting the platform to the wellhead template on the seaﬂoor are in general not a
part of the TLP mooring system (Demirbilek, 1989a). The diﬀerent parts of the
TLP are illustrated in Fig. 2.1, and deﬁned by the American Petroleum Institute
(1987) as follows:
• The hull consists of the buoyant columns, pontoons and the intermediate
structure bracings.
• The deck structure supports operational loads. It is a multilevel facility
consisting of trusses, deep girders and deck beams.
• The platform consists of the hull and deck structure.
• The foundation is found at the seabed and consists of templates and piles,
or even a gravity system.
• The tendons connects the platform to the foundation at the seabed.
• The mooring system consists of the tendons and the foundation.
• The risers include drilling, production and export risers.
• The well system includes ﬂowlines, risers, riser tensioners, wellhead and
the subsea well templates.
15
2. Tension Leg Platform
Figure 2.1: The layout of the Brutus TLP in the GoM (www.oﬀshore
technology.com).
• The tension leg platform includes all the above, in addition to all deck
equipment and the hull system.
2.3 Existing Platforms
Interest in TLPs dates back to 1960 and many studies have examined the appli
cability of this concept for deep water developments (Demirbilek, 1989a). During
the next two decades, and especially after the installation of the Hutton TLP in
1984, the TLP concept began attracting more attention from the oﬀshore indus
try as an appropriate structure for deep water applications.
There are at the moment installed approximately ﬁfteen TLPs, of which three
are found in the North Sea. The others are found in the Gulf of Mexico, on the
west coast of Africa and oﬀ the coast of Brazil. The increase in water depth for
TLPs is illustrated in Fig. 2.2. The deepest installed TLP today is the Magnolia
platform. Below follows a list with short descriptions of the some of the most
famous platforms.
16
Existing Platforms
0
500
1000
1500
Hutton Jolliet Auger Mars RamPowell
Ursa
Magnolia
W
a
t
e
r
d
e
p
t
h
[
m
]
Installation year
1984 1989 1994 1996 1997 1999 2003
Figure 2.2: The development of TLPs and riser depths.
Hutton (UK) The Hutton TLP was the ﬁrst of its kind and was installed by
Conoco in the British sector of the North Sea in 1984. The installation serves
as a combined drilling/wellhead, production, and quarter platform on a medium
sized oil and gas ﬁeld at 150m water depth. A conventional steel jacket would
be a less costly alternative, but the TLP solution was chosen to gain experience
with this concept for future applications on larger water depths. The tendons are
made of thickwalled tubes with conical threaded couplings. These tendons have
large submerged weight which will contribute to unwanted deﬂections when the
tendons are inclined (Larsen, 1995).
Snorre (Norway) The Snorre ﬁeld lies in the central North Sea approximately
200km west of Florø in Norway at 310m water depth. It was installed in 1988 by
Saga Petroleum, and the production started in 1992 (Ministry of Petroleum and
Energy and the Norwegian Petroleum Directorate, 2007). Snorre was developed
in two phases and the production facilities comprise a steel TLP on the southern
part of the ﬁeld and a subsea production system tied back to the TLP (Oﬀshore
Tecnology, 2007). 36 wells are drilled from the TLP. The Snorre TLP is moored
to the seabed by sixteen steel tendons 0.8m in diameter. The platform is an in
17
2. Tension Leg Platform
Figure 2.3: The Snorre TLP (www.statoil.com).
tegrated drilling, production and living quarter installation. The main diﬀerence
between Snorre and Hutton is that Snorre is much larger, it has four columns
instead of six, and the tendons have larger diameter and smaller wall thickness
(Larsen, 1995). Fig. 2.3 shows a picture of the Snorre TLP.
Jolliet (USA) The Jolliet platform was the ﬁrst real deep water TLP and was
installed by Conoco on the Green Canyon ﬁeld in the GoM at 536m water depth
in 1989. This platform is signiﬁcantly smaller than both Hutton and Snorre, and
is often referred to as a tension leg wellhead platform (TLWP) (Larsen, 1995).
Auger (USA) By the Auger platform, the TLP concept was brought to a new
frontier. The platform is of the same size as Hutton, but the water depth is even
larger than for Jolliet with 872m. Many of the design principles from Jolliet are
found at Auger. Note that the tensioned risers on Auger are inclined in order
to decrease the spacing distance between the wellheads on the deck compared to
the needed spacing at the seaﬂoor. Auger has catenary anchor lines in addition
to the tendons. These serve two main purposes; (1) to increase lateral stiﬀness
and thereby reduce static oﬀset due to wind and current, and (2) to enable the
platform to have a wanted oﬀset position during well drilling and maintenance
(Larsen, 1995).
18
Existing Platforms
Figure 2.4: The Mars TLP after the hurricane Katrina (www.worldoil.com).
Heidrun (Norway) The Heidrun platform was installed in 1995 at 345m water
depth. It is the ﬁrst and only concrete TLP, and has signiﬁcantly larger displace
ment than the earlier TLPs. The hull and tendons system have been tuned to
minimize ﬁrst order wave motions, but the result was that ringing (see Sec. 3.4)
became a bigger problem for this TLP than for the others (Larsen, 1995).
Mars (USA) Mars was installed in 1996 by Shell at 950m water depth and had
the water depth record when installed. It is also the largest producing TLP in
the GoM. The Mars platform’s drilling rig was heavily damaged during Hurricane
Katrina in August 2005 (see Fig. 2.4). It was shut down, as it was exposed to over
four hours of sustained winds of 270km/h, with gusts over 320km/h (Paganie,
2006). The production at Mars resumed on May 22 2006, after undergoing a
series of repairs to its damaged platform rig and export pipelines.
Ursa (USA) Shell began the ﬁeld production at Ursa in 1999, at approxi
mately 1200m water depth. This TLP is designed to simultaneously withstand
hurricane force waves and winds. The deck is composed of six modules: wellbay,
quarters, power, drilling and two process modules. The deck modules are an open
truss frame design (90x90x15m) with a total steel weight of 12,500 tons. There
are twelve well slots, and the well layout on the seaﬂoor is arranged in a rectan
gular pattern (30x18m). The TLP supports a single modiﬁed API platformtype
19
2. Tension Leg Platform
(a) Before the hurricane Rita. (b) Upside down following the hurri
cane Rita.
Figure 2.5: The Typhoon TLP of SeaStar design (www.upstreamonline.com).
drilling rig (leased) equipped with a surface blow out preventer (BOP) and high
pressure drilling riser.
Morpeth (USA) The threelegged Morpeth SeaStar is a miniTLP and the
ﬁrst TLP without surface completions. It was successfully installed in 1998 in
518m water depth at a cost of less than $100 million (Rigzone, 2007).
Brutus (USA) Brutus is a Shell installation at 910m water depth with pro
duction start in 2001. The batch setting of the eight wells was completed on
January 3 2000. Four of the planned development wells for the eightslot TLP
were subsequently predrilled, with the well layout on the seaﬂoor arranged in a
rectangular pattern. Brutus is Shell’s ﬁfth TLP in the GoM and the seventeenth
deep water project in the GoM in which Shell was involved (Rigzone, 2007).
Typhoon (USA) Chevron’s miniTLP Typhoon of SeaStar design was in
stalled in July 2001 at 700m water depth. It severed from its mooring and
capsized following the hurricane Rita fall 2005. It was ﬂoating upside down when
found after the hurricane. Fig. 2.5 shows the platform before and after the
hurricane Rita struck.
Matterhorn (USA) This miniTLP is an Atlantia SeaStar design of the type
previously installed on such deep water projects as Chevron’s Typhoon and
BritishBorneo’s Morpeth and Allegheny ﬁelds. At 1350m, however, the Mat
terhorn SeaStar was the largest when built in November 2003, and doubled the
20
Existing Platforms
size of the previous units. It was the ﬁrst unit of this design to incorporate dry
trees.
Marco Polo (USA) The Marco Polo ﬁeld is located at 1300m water depth in
the GoM. The ﬁeld was discovered in April 2000. In 2003, the last two develop
ment wells were drilled. The ﬁeld was developed with a dry tree TLP installed
in January 2004. It came online in July 2004 and is expected to reach peak
production of 50,000 barrels of oil equivalent per day (Rigzone, 2007).
Magnolia (USA) The Magnolia ﬁeld by Conoco Phillips was installed at
1433m water depth, a new record depth for this type of ﬂoating structure in
the GoM. The production started in December 2004. The ﬁeld is located approx
imately 50 km from the existing infrastructure, which will enable Magnolia to be
a regional oﬀtake point for future developments or third party tieins located in
Southeastern Garden Banks area (Oﬀshore Tecnology, 2007).
Kizomba (Angola) The Kizomba deep water project is located oﬀ the west
coast of Angola. The ﬁeld was installed in August 2005 at 1260m water depth.
Kizomba A consists of three main components. The drilling center is a TLP that
includes a rig and 36 slots for drilling the project’s oil and gas wells. The TLP
contract was awarded to ABB in partnership with Heerema of Holland. Moored
nearby is a FPSO, designed to take all of the oil produced from the platform,
process it and store it until the oil can be oﬄoaded onto waiting tankers (Rigzone,
2007).
21
Chapter 3
Mathematical Modeling
Mathematical models may be formulated in various levels of complexity. We usu
ally distinguish between process plant models and control plant models (Sørensen,
2005). The process plant model (PPM) is a comprehensive model of the actual
physical process. The main purpose of this model is to simulate the real plant dy
namics, including environmental disturbances, control inputs and sensor outputs.
A successful numerical simulation for design and veriﬁcation of our control sys
tem require a suﬃciently detailed mathematical model of the actual process. The
control plant model (CPM) is simpliﬁed from the process plant model containing
only the main physical properties. It may be a part of the modelbased controller.
The CPM is often formulated such that the analytical stability analysis becomes
feasible.
This chapter will focus on the process plant model. The system modeled here
consists of two risers connected to a TLP through the top nodes, forcing the top
nodes to follow the prescribed motion from the TLP in the horizontal direction.
The top nodes are free in the vertical direction, only aﬀected by the top tension
acting as a vertical force. In addition the risers are exposed to current forces that
are found by considering hydrodynamic interactions between the risers.
Interaction between two cylinders is often classiﬁed into two categories accord
ing to the space between them; the proximity interference when the two cylinders
are close to each other, and wake interference when one cylinder is in the wake of
an upstream cylinder (Zdravkovich, 1985). Top tensioned marine riser systems
normally fall into the latter category, although proximity interference also can
take place. It is also found that the inline motions are much larger than the
transverse motions. We will focus on two risers in a tandem arrangement, were
the centers of the two risers are aligned parallel to the free stream (Fig. 3.1). It is
assumed that a twodimensional model will capture the most important dynamics
of the riser array system. Analysis of the threedimensional eﬀects are subject
23
3. Mathematical Modeling
R1 R2
Incoming ﬂow Upstream riser
Downstream riser
Figure 3.1: The risers in a tandem arrangement.
for further research.
3.1 Current Models
Current models could be divided in two: (1) Surface current, needed for ships
and ﬂoaters, and (2) full current proﬁle, for use in load models for risers, tendons,
anchor lines, etc. Two diﬀerent geographical areas are chosen to investigate the
risers behavior under various current velocity conditions. These are:
• The Ormen Lange ﬁeld in the North Sea.
• The Gulf of Mexico (GoM).
Common to these selected areas are that they are:
• Located in deep waters.
• Tend to have strong current velocities, with diﬀerent proﬁles.
• TLPs are already situated here or considered situated here.
The current proﬁles from these areas will be explained in more detail. In addi
tion, some linear, theoretical proﬁles will be investigated. There are three main
components in the resulting current proﬁle:
• Wind generated currents.
• Tide generated currents.
• Major ocean currents.
The formulas for these current components can be found in Faltinsen (1990).
Here, we have used inplane proﬁles which may vary in velocity through the
water column. The proﬁles from the North Sea and GoM are design proﬁles,
24
Current Models
−1 0 1
0
200
400
600
800
1000
1200
W
a
t
e
r
d
e
p
t
h
[
m
]
Uniform
−1 0 1
Linearly Sheared
−1 0 1
Ormen Lange 1
−1 0 1
Ormen Lange 2
−1 0 1
0
200
400
600
800
1000
1200
GoM1
W
a
t
e
r
d
e
p
t
h
[
m
]
−1 0 1
GoM2
−1 0 1
Bidirectional
Figure 3.2: The simulated current velocity proﬁles in m/s.
based on measurements and dimensioned for 1200m water depth. All current
velocity proﬁles are illustrated in Fig. 3.2. The current velocity data are given
in Appendix B.1.1, and can be summarized as:
• Uniform current (theoretical) meaning constant current velocity for
the entire water column.
• Linearly sheared current (theoretical) which denotes a linearly de
creasing current velocity throughout the water column, with a given veloc
ity at the sea surface and 0m/s at the seaﬂoor.
• Ormen Lange 1 which is a one year return period current proﬁle for the
Ormen Lange ﬁeld based on Herfjord et al. (2002) and extended to 1200m
water depth by linearly decreasing the velocity from 0.65m/s at 850m down
to 0m/s at the seabed.
• Ormen Lange 2 which is a one year return period current proﬁle found
in Aker Maritime (2002), originally given in Norsk Hydro (2001). It is
25
3. Mathematical Modeling
extended from 850m to 1200m water depth. The current between these
depths are considered constant at 0.5m/s.
• Gulf of Mexico 1 (GoM1) which is a wind driven current velocity proﬁle
in the GoM, based on a nondimensional proﬁle in Nowlin et al. (2001). It
is dimensioned for 1200m water depth and 1m/s current velocity at the sea
surface.
• Gulf of Mexico 2 (GoM2) which is a current proﬁle with a loop eddy in
the top layer, also based on Nowlin et al. (2001) and scaled the same way
as GoM1.
• Bidirectional current (theoretical) which represents a linear shear cur
rent with opposite current directions at the sea surface and the seabed.
For the current proﬁles related to a geographical area, a further explanation is
given. The Ormen Lange ﬁeld is close to the shelf edge. This gives often strong
and variable current, and might potentially cause operational challenges (van
Smirren et al., 1999). In Fig. 3.2 the current velocities are seen to be large at
the sea surface due to wind generated current, but also large all the way down to
the seabed since the ﬁeld is close to the shelf edge. For the rest of the analysis
and simulations in this thesis, the Ormen Lange 2 is used. It will be referred to
as the Ormen Lange design current proﬁle or just the Ormen Lange proﬁle.
For GoM2 a loop eddy is seen to reduce the current at 200400m water depth,
whereas the velocity increases again for even larger water depths. The loop
currents found in the GoM are due to a large ﬂow of warm water that dominates
the circulation within the eastern part of the GoM. It is due to the Gulf Stream
which ﬂows northwards between Cuba and the Yuc´atan peninsula, Mexico. Some
of the current moves straight north into the GoM, and then loops east and south
before it exits to the east through the Florida Straits. Some loop currents tear oﬀ
the main stream and ﬂow clockwise westward into the GoM (Wikipedia, 2007a,
NOAA Coastal Services Center, 2007). This is illustrated in Fig. 3.3.
Last, a bidirectional shear current is presented. A bidirectional current proﬁle
can be found west of Shetland, and is due to a residual warm ﬂow northeastward
usually in the upper layer, and a southwestern cold ﬂow in the lower layer. In
addition to the residual ﬂow there are tidal currents. Here a simple linear, bidi
rectional shear current is introduced to investigate the riser behavior and verify
the model for a wider range of environmental conditions.
26
Hydrodynamic Interaction
Figure 3.3: Loop current in the Golf of Mexico (www.oceanexplorer.noaa.gov).
3.2 Hydrodynamic Interaction
Recall form Section 1.2 that interaction between two neighboring risers will not
have any hydrodynamic inﬂuence on the upstream riser (R1) beyond a certain
distance, and that R1 can be treated as an isolated riser (Kalleklev et al., 2003,
DNV, 2003). The attention will therefore be given to the hydrodynamic inﬂuence
on the downstream riser (R2). Also remember the three most important eﬀects
for assessment of riser interaction (DNV, 2005):
1. Mean force and shielding eﬀects, tending to bring the risers closer.
2. WIO on the downstream riser.
3. VIV leading to ampliﬁed drag coeﬃcients for both risers.
In this thesis we will focus on shielding eﬀects, which is of main importance when
calculating the mean current force, and hence the position of the second riser in
the wake. The two latter eﬀects will not be included in the simulations, but are
brieﬂy described here for completeness. Inclusion of these eﬀects may have an
inﬂuence on the relative riser displacements, but to a lesser extent on the control
strategy.
27
3. Mathematical Modeling
Since the riser model is twodimensional in a tandem arrangement, the lift
force is not included, and Huse’s model for wake shielding eﬀects is an appropriate
model choice.
3.2.1 Shielding Eﬀects
R2 experiences reduced mean drag force due to shielding eﬀects, depending on
the location in the wake. A semiempirical static wake formulation to account
for the hydrodynamic interaction between individual risers in steady current was
proposed by Huse (1993). The reduced velocity ﬁeld in the wake of the upstream
cylinder, see Fig. 3.4, is given by
V
r
(x, y) = k
2
V
c
_
C
D
D
e
x
s
e
−0.693(
y
b
)
2
, (3.1)
where V
c
is the incoming current velocity on R1, C
D
is the drag coeﬃcient, D
e
is
the diameter of the upstream riser, and y is the distance away from the centerline
of the incoming velocity proﬁle. x
s
and b are deﬁned as
x
s
= x +
4D
e
C
D
, b = k
1
_
C
D
D
e
x
s
, (3.2)
where k
1
= 0.25 and k
2
= 1.0 for a smooth circular cylinder. x is the distance
behind the upstream riser R1, and x
s
is the distance between the downstream
riser R2 and a virtual wake source upstream of R1. The wake may be inﬂuenced
by VIV of the upstream riser, but this eﬀect is considered insigniﬁcant for the
present use of the wake model. The mean inﬂow on the downstream riser is hence
given by
V
mean
= V
c
−V
r
. (3.3)
It is assumed that mechanical contact occurs when the distance between the
riser centers is equal to one riser diameter (1D). The parametric wake model is
only applicable for the far wake region larger than two diameters (2D) behind
R1. The behavior of the ﬂow in the near region is not adequately described as
this is a highly nonlinear phenomenon, where R2 might experience negative drag
forces (Kavanagh et al., 2000). The centertocenter distance should therefore
preferably be kept larger than 2D. A more extended presentation of how (3.1) is
derived, is found in Appendix A.1.
3.2.2 Wake Induced Oscillation
The WIO is described as a broad band buﬀeting force due to oncoming turbulent
ﬂow and vortices shed from the upstream riser. This may result in a LF, large
amplitude motion of the downstream riser which wander around in the wake
28
Hydrodynamic Interaction
V
c
V
r
(y)
x
y
Figure 3.4: The decrease in the inline water particle velocity in the wake region,
from Huse (1993).
and is likely to cause collision (Huse, 1996, Wu et al., 2002). In one experiment
Huse (1996) investigated the interaction between cylinders. It was observed that
in addition to the high frequency VIV response of amplitudes up to one half
of the diameter, the downstream cylinder also had LF inline oscillations of an
apparently very irregular nature with the peaktopeak stroke of these oscillations
up to 30 to 40 diameters or more. These motions occur at the ﬁrst mode of the
riser. Later Wu et al. (2002) found the peakto peak range to be 20D. A more
detailed review on existing models are found in DNV (2005). However, this kind
of oscillations and the physics behind it is so far not well understood. It will
therefore not be implemented in this work, and is subject for further research.
3.2.3 Vortex Induced Vibrations
Recall that the large displacements due to shielding eﬀects and WIO are found
to be independent of VIV, except from the magniﬁed drag coeﬃcient (Tsahalis,
1984, DNV, 2003, 2005). Therefore, an ampliﬁed drag coeﬃcient due to VIV
needs to be included. The ampliﬁed drag is modiﬁed with VIV amplitude A and
diameter D. One of the used relations presented by Vandiver (1983) is
C
′
D
= C
D
_
1 + 1.043
_
2
√
2A
D
__
0.65
. (3.4)
A conservative estimate is to select a high value for the upstream riser and a low
value for the downstream riser. This will bring the risers closer together in their
static positions. Anyhow, very little information is available for VIV behavior
for a downstream riser. In this study we have used equal and constant drag
coeﬃcients for both risers. The drag coeﬃcients are not used explicitly in the
controller algorithm, and will therefore not have impact on the conclusions as
such.
29
3. Mathematical Modeling
f
1
f
3
f
2
g
1
g
3
g
2
b
1
b
3
b
2
i
1
i
3
i
2
Figure 3.5: The coordinate systems: f  ﬁxed frame for ﬁrst riser, g  global
frame at the surface, b  body frame of TLP, i  element frames.
3.3 Kinematics and Coordinate Systems
Four orthogonal coordinate systems are used to describe the riser and TLP mo
tions, see Fig. 3.5. A bold letter with subscript i = ¦1, 2, 3¦ denotes the unit
vector along the x, y and z axes in the frame respectively, i.e. f
1
denotes the unit
vector along the xaxis for the seaﬂoor ﬁxed frame. The coordinate systems are
all righthand systems.
seabed ﬁxed frame (fframe): The fframe (o
f
f
1
f
2
f
3
) is considered inertial
and is ﬁxed to the seaﬂoor. The positions of all the riser nodes in the global
system are described relative to this frame. f
1
points to the right, f
2
points
into the plane and f
3
points upwards normal to the earth surface.
Bodyﬁxed frame (bframe): The bframe (o
b
b
1
b
2
b
3
) is ﬁxed to the body
of the TLP with axes chosen to coincide with the principal axes of inertia
for the body with origin in the center of gravity. b
1
is the longitudinal axis
pointing forward, b
2
is the transversal axis pointing starboard, and b
3
is
the normal axis pointing downwards.
Global frame (gframe): The gframe (o
g
g
1
g
2
g
3
) is ﬁxed to the sea surface
right above the fframe with a distance in heave direction equal to the
water depth and is also considered inertial. g
1
points to the front side of
the TLP, g
2
points to starboard, and g
3
points downwards normal to the
30
Tension Leg Platform
Earth surface. The position vector r
g
TLP,bg
= [x
g
bg
, y
g
bg
, z
g
bg
]
T
describes the
position of the bframe relative to the gframe expressed in the gframe.
The origin, o
g
, is located on the mean water freefree surface so that z
g
passes through the center of the TLP when it is in its static undisturbed
equilibrium.
Local riser frames (iframe): The iframes (o
i
i
1
i
2
i
3
) used in the ﬁnite ele
ment modeling (FEM) are located in the ith node of the riser and is a local
frame for each element i. i
1
is to the right of the element, i
2
is pointing
into the plane, and i
3
is along the axial direction of the element pointing
upwards. The forces acting on the element, such as tension, eﬀective weight
and current due to drag are computed in this frame for each element and
thereafter transformed to the fframe. There are as many iframes as there
are number of elements in the FEM model.
3.4 Tension Leg Platform
Recall that the TLP is a partly rigid and partly compliant structure. The eigen
frequencies for inplane motions must be below frequencies for wave energy and
the frequencies for the outofplane motions must be above. With respect to the
horizontal degrees of freedom, the TLP is compliant and behaves similar to other
ﬂoating structures. The horizontal degrees of freedom surge, sway and yaw are
inertia dominated with eigenperiods around 12 minutes, well above the range
of ﬁrst order waves with periods of 520s. With respect to the vertical degrees
of freedom, it is stiﬀ and resembles a ﬁxed structure. The vertical degrees of
freedom heave, roll and pitch are stiﬀness dominated with eigenperiods in the
order 24s, and well below the period of the exciting ﬁrst order waves (Larsen,
1995).
The eigenfrequencies in all six degrees of freedom (DOFs) are tuned relative
to the ﬁrst order wave loads. Thus, loads at the wave frequencies do not excite the
TLP at its natural frequencies. On the other hand, second or higher order loads
at the sum and diﬀerence frequencies can produce signiﬁcant resonant excitations
at the TLP natural frequencies because of the small amount of damping available
at these frequencies (Faltinsen, 1990, Mekha et al., 1996).
• Higher order components and sum frequencies (2ω
i
, 2ω
j
, ω
i
+ ω
j
) in waves
may give signiﬁcant resonance oscillations for the TLP in heave, roll and
pitch known as ringing and springing. The restoring forces are due to the
tendons and the mass forces due to the TLP, and they are excited by the
nonlinear wave eﬀects. Ringing is associated with transients eﬀects, while
springing is steadystate oscillations (Faltinsen, 1990).
31
3. Mathematical Modeling
• Loads on diﬀerence frequencies (ω
i
−ω
j
) will give slowly varying wave loads
that may give rise to resonant inplane motions. Excitation from wind gusts
may appear in the same frequency range and contribute signiﬁcantly to such
motions (Faltinsen, 1990, Mekha et al., 1996).
Both these nonlinear phenomena should be considered when designing a TLP.
The ringing and springing phenomena have impact on the upper and lower limits
of the tendon force. The mean wave drift and current loads on the hull will induce
a mean oﬀset force. Assuming linear analysis as the water depth increases, the
eigenperiod of the riser system increases as well. The ﬁrst eigenperiod for a riser
at 1200m water depth is approximately 3040s depending of the top tension and
cross section properties. As the riser eigenperiod is approaching the eigenperiods
in surge and sway for the TLP, induced motions from these components may have
more inﬂuence on the riser dynamics than in more shallow waters.
3.4.1 TLP Surge Modeling
Floater models and analysis are usually divided into two groups; separated or de
coupled analysis and coupled analysis. For separated analysis the vessel motions
are found ﬁrst. The eﬀects from moorings and risers are included as nonlinear
position dependent forces or stiﬀness. The damping or velocity dependent forces,
which are important for estimation of the LF motion may be neglected. In the
second step, the dynamic response on the risers and moorings are analyzed, using
the vessel response from step one as a forced displacement on the top node. The
main problem with this method is that inertia and drag forces on the risers
and tendons are not accounted for. These eﬀects could be large in deep waters
(Ormberg et al., 1998, Chen et al., 2002). The term coupled analysis means
simultaneous analysis of vessel motions, mooring systems and riser dynamics
where the full interaction is taken into account. The main drawback with this
method is that it is very time consuming since a nonlinear time domain simulation
is required. Much eﬀort has been put into the investigation of coupled and
separated analysis and solutions in between, see Mekha et al. (1996), Ormberg
and Larsen (1997), Ormberg et al. (1997, 1998), Ma et al. (2000), Chaudhury
and Ho (2000), Chen et al. (2002). For an extensive description of TLP, tendons
and mooring, see Demirbilek (1989b).
In this study we assume that the TLP motions inﬂuence the riser behavior.
However, the risers do not aﬀect the TLP motion. The TLP represents the
speciﬁed motion in surge for the top node of each riser. Hence, a decoupled
analysis model is applied. Since the system model is twodimensional, only the
32
Tension Leg Platform
surge motion for the TLP is needed. It is modeled as a LF harmonic motion
x
TLP
= A
TLP
sin
_
2π
T
TLP
t
_
+ x
off
, (3.5)
where x
off
is the static TLP oﬀset, A
TLP
is the amplitude, and T
TLP
is the
period of the TLP motion.
3.4.2 Kinematics
The motions of the TLP are given as the motion of the bframe relative to
the gframe in the gframe with the position vector position vector r
g
TLP,bg
=
[x
g
bg
, y
g
bg
, z
g
bg
]
T
. As the riser motions are expressed in the fframe, the rotation
matrix between the f and gframes is needed. The transformation from g to
fframe is given by the diagonal matrix
R
f
g
= diag (1, −1, −1) . (3.6)
Since both frames are considered inertial and ﬁxed with a translation in heave
direction only, all axes are parallel, and the xaxes are always pointing in the
same direction. In a two dimensional system, only the surge motion of the TLP
is needed to deﬁne the prescribed motion of the riser. Hence, the motion in x
direction for the TLP described in the gframe is equal to the xposition given
in the fframe. Thus the frames are omitted for the surge position of the TLP,
x
TLP
.
3.4.3 Riser Stroke Calculations
Marine risers made of steel have very low structural strength against lateral
loading unless they are tensioned. It is therefore important to maintain the upper
end tension at a certain level under all realistic conditions, irrespective of platform
motions, dynamic riser response and internal ﬂow parameters. The tensioner
system will therefore act as a heave compensating system with an adequate stroke
capacity and ability to maintain a near constant tension (Larsen, 1993).
If the relative platform/riser motion exceeds the stroke capacity, unwanted
loss of tension or tension increase will occur. Such situations may result in exces
sive bending stresses in the riser, excessive rotations of the ball joint or excessive
riser tension. Such loads may cause damage to the well template or the riser ten
sioning system. Other unwanted eﬀects might be mechanical interaction between
neighboring risers or between the riser and the platform. When designing the riser
tensioner system, consequences of such events must be taken into considerations
(Larsen, 1993).
33
3. Mathematical Modeling
x
TLP
∆s
r
∆s
t
l
r
l
t
Figure 3.6: Setdown due to tendon/riser geometry.
The platform vertical motion is the most important parameter that deﬁnes
stroke. For a TLP with vertical and parallel tendons in the initial zero oﬀset
position, motions in surge direction will not induce any pitch motion, but will
be coupled to the vertical motion referred to as surge induced heave or setdown.
Using the Pythagorean theorem and assuming both risers and tendons to be
straightlined, the relative setdown between the riser and the platform can be
found. The position of the platform is controlled by the tendons. The oﬀset in
surge, x
TLP
, is equal for the risers and the tendons, but as the risers are longer
and hence have larger radius, their setdown is smallest. The relative setdown,
∆s, is found as
∆s = ∆s
t
−∆s
r
=
_
l
t
−
_
l
2
t
−x
2
TLP
_
−
_
l
r
−
_
l
2
r
−x
2
TLP
_
, (3.7)
where ∆s
t
and ∆s
r
are the tendon and riser setdown respectively, and l
t
and l
r
are the tendon and riser lengths. This is illustrated in Fig. 3.6. For given l
t
and
l
r
, we can calculate the setdowns for the upper end of the risers and tendons. The
relative setdown is for normal TLP designs small, even for large water depths and
oﬀsets, due to the tendon and riser geometry. Hence, the requirement for heave
compensation due to surge induced oﬀset is small.
34
Riser FEM Modeling
i
1
i
3
1, x
1
2, z
1
3, x
2
4, z
2
Figure 3.7: The bar element with four DOFs. Node 1 to the left with the two
ﬁrst DOFs. Node 2 to the right with the third and fourth DOFs.
3.5 Riser FEM Modeling
The partial diﬀerential equation (PDE) governing the static and dynamic be
havior of a riser can not be solved exactly for arbitrary riser problems and load
patterns. Hence, a numerical method is required, such as the ﬁnite element
method (FEM). The stiﬀness matrix in the FEM model will have an elastic and
a geometric component. The elastic stiﬀness matrix accounts for the axial and
bending stiﬀness as present in any beam, while the geometric stiﬀness matrix
will take into account the changes of the global geometry and the stiﬀening eﬀect
from the tension. As the depth is increasing, the riser will behave more and more
like a cable, and the geometric stiﬀness will become more important than the
elastic stiﬀness. Hence, the geometric stiﬀness gives the main contribution to
lateral resistance against the static and dynamic forces acting perpendicular to
the longitudinal riser axis.
At larger water depths a simpliﬁcation of the riser model can be made by
neglecting the bending stiﬀness and assuming free rotations at the ends. In
cases where the global geometry is of major importance this will only introduce
a small error. Hence, a model consisting of bar elements is suﬃcient. Each bar
element in a two dimensional model can be described with four DOFs, that is two
translational DOFs in both ends of the element. x is transverse of the element,
positive to the right along i
1
, and z is along the element, positive upwards along
i
3
, see Fig. 3.7.
The riser is ﬁxed to the seabed, and the top node displacement is prescribed
in the horizontal direction, whereas it is free to move vertically. Note, however,
that the top tension, P
top
, is introduced at the upper end, which prevents the
riser from collapsing. The tension vector, f
top
, is an external force and will be
found on the right hand side of the equation of motion according to
f
top
=
_
0
1×(2n+1)
P
top
¸
T
. (3.8)
In later chapters top tension is referred to as T, while P
top
is used in the modeling
35
3. Mathematical Modeling
1
2
i+1
i
1
2
i
n
n+1
v
1
v
2
v
i
v
n
v
n+1
θ
1
θ
2
θ
i
θ
n n
Figure 3.8: Numbering of the elements (encircled), nodes (boxes), current v
i
in
each node i and the inclination θ
i
of each local element i, relative to the global
frame.
as tension applied at the top node.
3.5.1 Transformations for the Riser Elements
The risers are modeled with n elements and n+1 nodes, with node number 1 at the
seabed and node n+1 at the wellhead area on the platform. θ
i
is the inclination
of element i relative to the global coordinate system fframe, which is situated
at the seabed in the ﬁrst node of the upstream riser. The numbering of elements,
nodes and the inclination of each local element are illustrated in Fig. 3.8. v
f
i
is the current in node i expressed in the inertial fframe. Note that i is used
as the numbering of the elements, nodes and the corresponding inclination. The
positions x
i
and z
i
of the nodes along the riser are found through equilibrium
iterations, and are used to calculate the sine and cosine of the inclination θ
i
of each element, needed for use in the transformations between the global and
local coordinate systems. The length of each element i is found by use of the
Pythagorean theorem
∆x
i
= x
i+1
−x
i
, ∆z
i
= z
i+1
−z
i
, l
i
=
_
∆x
2
i
+ ∆z
2
i
, (3.9)
cos θ
i
=
∆z
i
l
i
, sin θ
i
=
∆x
i
l
i
. (3.10)
36
Riser FEM Modeling
The transformation between the local iframe and the global, ﬁxed fframe is
described by a rotation about the yaxis (Fossen, 2002)
R
y,θ
=
_
_
cos θ 0 sin θ
0 1 0
−sinθ 0 cos θ
_
_
. (3.11)
Since we are only considering a twodimensional system with two DOFs in each
node, the transformation matrix from i to f is written as
T
f
0,i
(r) =
_
cos θ
i
sin θ
i
−sinθ
i
cos θ
i
_
, (3.12)
where r is the displacement vector, as the inclination in each element is given as
a function of the positions, given in the seaﬂoor ﬁxed fframe. For notational
simplicity the superscript f is omitted as r is always given in the fframe. The
full displacement vector is given as
r =
_
x
1
z
1
x
2
z
2
x
i
z
i
x
n
z
n
x
TLP
z
n+1
¸
T
. (3.13)
For all four DOFs related to an element we have
T
f
i
(r) =
_
T
f
0,i
(r) 0
2×2
0
2×2
T
f
0,i
(r)
_
, (3.14)
as the inclination of each element is a function of the positions of its nodes.
3.5.2 System Mass Matrix
For each element a mass matrix is deﬁned based on the local coordinate system.
The local mass matrix consists of three terms; the structural mass of the riser
m
s
, the internal ﬂuid m
f
, and the hydrodynamic added mass m
a
. The structural
mass is given by
m
si
=
ρ
s
Al
i
6
_
¸
¸
_
2 0 1 0
0 2 0 1
1 0 2 0
0 1 0 2
_
¸
¸
_
, (3.15)
where ρ
s
is the density of the riser material steel, A is the cross sectional area of
the riser, and l
i
is the length of element i found from (3.9). The internal ﬂuid is
included as well
m
fi
=
ρ
f
A
int
l
i
6
_
¸
¸
_
2 0 1 0
0 2 0 1
1 0 2 0
0 1 0 2
_
¸
¸
_
, (3.16)
37
3. Mathematical Modeling
where A
int
is the internal area of the riser, and ρ
f
is the density of the internal
ﬂuid. The derivation of the consistent mass matrix structure is found in Appendix
A.2.1. The local added mass in the axial zdirection is assumed to be zero, while
the added mass for a circular cylinder in the lateral xdirection is equal to the
displaced water volume (Sumer and Fredsøe, 1997). The matrix for the added
mass is then given as
m
ai
=
ρ
w
C
m
A
e
l
i
6
_
¸
¸
_
2 0 1 0
0 0 0 0
1 0 2 0
0 0 0 0
_
¸
¸
_
, (3.17)
where A
e
is the area found from the exterior diameter D
e
, and ρ
w
is the density
of water. The coeﬃcient C
m
is called the hydrodynamic added mass coeﬃcient.
The total local mass matrix m
i
for each element i in its own frame is then the
sum of the three terms written as
m
i
= m
si
+m
fi
+m
ai
. (3.18)
Each local matrix m
i
is transformed to the global fframe ﬁxed to the seabed,
and all the local mass matrices are assembled to form the global mass matrix M
for the riser. Each local mass matrix is transformed to fframe by
¯ m
i
= T
f
i
m
i
T
fT
i
=
_
¯ m
i
11
¯ m
i
12
¯ m
i
21
¯ m
i
22
_
, (3.19)
where ¯ m
i
is the mass matrix for element i in the global fframe. For ¯ m
i
11
the
superscript is used to recognize the element number, and the subscript is used
to identify the diﬀerent parts of the mass matrix, needed for the concatenation
to the global matrix. T
i
f
is the transformation matrix from the fframe to the
iframe. The full, global mass matrix in the fframe for n elements is then
M =
_
¸
¸
¸
¸
¸
¸
¸
_
¯ m
1
11
¯ m
1
12
¯ m
1
21
¯ m
1
22
+ ¯ m
2
11
¯ m
2
12
¯ m
2
21
¯ m
2
22
+ ¯ m
3
11
.
.
.
¯ m
n−1
22
+ ¯ m
n
11
¯ m
n
12
¯ m
n
21
¯ m
n
22
_
¸
¸
¸
¸
¸
¸
¸
_
. (3.20)
The global mass matrix M is always given in the fframe, and the superscript f
is hence omitted. The size of the global matrix is 2(n + 1) 2(n + 1).
38
Riser FEM Modeling
3.5.3 System Stiﬀness Matrix
The local stiﬀness matrix consists of two terms; the elastic stiﬀness, k
E
, and the
geometric stiﬀness, k
G
. The geometric stiﬀness matrix includes the eﬀective axial
tension P
i
acting on the particular element i. The elastic stiﬀness matrix works
in the axial direction, whereas the geometric stiﬀness in the lateral direction. The
resulting stiﬀness matrix k
i
for the element i in its own frame becomes
k
i
= k
E
i
+k
G
i
=
1
l
i
_
_
_
_
EA
_
¸
¸
_
0 0 0 0
0 1 0 −1
0 0 0 0
0 −1 0 1
_
¸
¸
_
+ P
i
_
¸
¸
_
1 0 −1 0
0 0 0 0
−1 0 1 0
0 0 0 0
_
¸
¸
_
_
_
_
_
,
(3.21)
where E is Young’s modulus of elasticity. The derivation of the consistent stiﬀness
matrix is found in Appendix A.2.2. The eﬀective axial tension P
i
in each element
is according to Appendix A.4 related to the elongation of the element by
P
i
=
EA
l
0
∆l
i
, ∆l
i
= l
i
−l
0
, (3.22)
where l
0
is the initial length of an element in a stressfree riser. To calculate the
global stiﬀness matrix for the riser, we transform each local stiﬀness matrix from
its local iframe to the global fframe, similar to what was done with the mass
matrix. These are concatenated to the global stiﬀness matrix K in the same way
as the mass matrix. The full, global stiﬀness matrix in the fframe for n elements
is then
K =
_
¸
¸
¸
¸
¸
¸
¸
_
¯
k
1
11
¯
k
1
12
¯
k
1
21
¯
k
1
22
+
¯
k
2
11
¯
k
2
12
¯
k
2
21
¯
k
2
22
+
¯
k
3
11
.
.
.
¯
k
n−1
22
+
¯
k
n
11
¯
k
n
12
¯
k
n
21
¯
k
n
22
_
¸
¸
¸
¸
¸
¸
¸
_
. (3.23)
3.5.4 Structural Damping
The damping experienced by a riser is a combination of the structural damping
and hydrodynamic damping, resulting from both radiation and viscous dissipa
tion of energy, see Faltinsen (1990) and Mekha et al. (1996). The structural
damping is due to the strain and elasticity properties of steel. The damping
matrix for a tensioned steel riser is based on proportional Rayleigh damping and
assumed proportional to the global stiﬀness matrix (Berge et al., 1992)
C = α
2
K. (3.24)
39
3. Mathematical Modeling
The coeﬃcient α
2
can be found from simple equations if the total damping level
is known at a speciﬁed frequency. The term proportional to mass is neglected in
order to avoid damping from rigid body motions. In this work we have assumed
that the structural damping is 1.5% of the critical damping at the eigenperiod
of T = 10s. Frequencies lower than this will have smaller damping, and the
damping is assumed linearly increasing with logarithmic increasing ω, such that
λ = 0.15% for T = 100s. Thus, the material damping is small relative to the
hydrodynamic damping. A further discussion on use of the Rayleigh damping
can be found in Appendix A.2.3.
3.5.5 Hydrodynamic Forces
The hydrodynamic forces on the risers are computed using Morison’s equation
modiﬁed for a moving circular cylinder. The horizontal hydrodynamic force on a
strip of the cylinder can be written
df
hyd
= ρ
w
C
M
πD
2
e
4
adz −ρ
w
C
m
πD
2
e
4
¨ rdz +
1
2
ρ
w
C
D
D
e
(v − ˙ r) [v − ˙ r[ dz, (3.25)
where v and a are the undisturbed water velocity and acceleration, ˙ r is the
response velocity, and ¨ r is the structure’s acceleration. C
M
= C
m
+ 1 is the
inertia force coeﬃcient. The acceleration of the water has a material derivative,
a =
Dv
Dt
, as the water velocity ﬁeld is varying in space due to shielding eﬀects, but
this is assumed to be small. This and the fact that the wave induced water motion
is assumed negligible for a riser in deep waters, means that the ﬁrst term of (3.25)
can be neglected. The added mass term is already included in the mass term on
the left hand side of the equation of motion. Hence, the only hydrodynamic
force included on the right hand side is the drag force df
drag
, which is the last
term of (3.25). This drag force is calculated for each element and summarized
as concentrated forces in each node. The force vector f
drag
is found at the right
hand side in the equation of motion given in (3.27) later in the text.
The added mass C
m
for a circular cylinder is known to be 1.0. However, this
value may for the downstream cylinder be inﬂuenced by the wake. VIV for both
risers may also cause some variations of C
m
. For this application such eﬀects are
considered to be insigniﬁcant, and a constant value is therefore applied in this
study. The drag coeﬃcient is inﬂuenced by similar eﬀects as added mass, and
C
D
is known to depend on the ﬂow velocity for a speciﬁc cross section. However,
it is not easy to describe such variations, and it is therefore common practice
for engineering purposes to assume a constant value. This approach is taken in
the present study, and a value of C
D
= 1.0 for both risers is applied. It is also
important to note that variations of C
D
will not alter any conclusions regarding
40
Riser FEM Modeling
P
i
f
drag
f
ext
w
eff
P
i−1
f
int
Figure 3.9: Balanced internal and external force.
the control system design for this study since the estimates of drag forces are not
needed for control purposes.
3.5.6 Load and Equilibrium Iteration
The modeled system contains nonlinearities in the mass, stiﬀness and damping
matrices in addition to the nonlinearities due to response dependent drag forces.
The top tension deﬁnes geometric stiﬀness according to (3.21) and will vary
in time as it is the control input. The forces acting on each element, such as
tension, eﬀective weight and drag are computed in the iframe for each element
and thereafter transformed to the fframe. A detailed description is given in
Appendix A.3.
As in traditional FEM, the nonlinearities are solved numerically by incre
mental formulation with the Newmarkβ time integration method and Newton
Raphson equilibrium iteration. For each time step the force equilibrium between
the internal force f
int
, due to tension, and the external force f
ext
, due to drag
forces f
drag
and eﬀective weight w
eff
, is found, as illustrated in Fig. 3.9. A
combined load and equilibrium iteration is needed since drag forces depend on
the relative velocity v − ˙ r, and the lateral stiﬀness of the riser will vary due to
dynamic tension variation. Iteration continues until an equilibrium solution is
found with desired accuracy.
41
3. Mathematical Modeling
3.5.7 TLP Prescribed Forces
The complete stiﬀness, damping and mass matrices found for the riser can be
divided into submatrices that contain the free and prescribed (ﬁxed or with spec
iﬁed motion) DOFs respectively. The columns and rows corresponding to the
prescribed DOFs are removed from the original system matrices, meaning that
only the free DOFs are present in the dynamic equation of motion. The ﬁxed
DOFs are the positions at the bottom (both x
i=1
and z
i=1
), whereas x
n+1
= x
TLP
at the top node is the only node with speciﬁed motion. The speciﬁed motion will
give contribution to the vector with dynamic loads found on the right hand side of
the equation of motion, see (3.27). These contributions are found from elements
in the original riser system matrices that links the free DOFs to the DOFs with
speciﬁed motions (spe). For the present case this contribution will be given by
f
TLP
= m
spe
¨ x
TLP
+c
spe
˙ x
TLP
+k
spe
x
TLP,
(3.26)
where f
TLP
is the force vector that originates from the TLP motions. m
spe
, c
spe
and k
spe
are columns in the original riser system matrices that correspond to
x
TLP
, and hence are removed from the riser system matrices used in the dynamic
equation of motion.
3.5.8 Dynamic Equation of Motion
In the dynamic equation of motion, the massdamperspring system for the free
DOFs in the riser system are found on the left hand side. On the right hand side
the external forces from top tension, current and drag forces, and the speciﬁed
motion from the TLP is included
M(r) ¨r +C(r) ˙ r +K(r) r = f
top
+f
drag
−f
TLP
, (3.27)
where r is the riser position vector, f
top
is the top tension, and f
drag
is the drag
forces from current and riser motions. The superscript f is omitted as the equa
tion of motion is always given in the fframe. In this equation the ﬁxed and
prescribed DOFs are removed from the equation, and the inﬂuence from speci
ﬁed DOFs is included on the right hand side as a force acting on the system. The
riser position vector is then
r =
_
x
2
z
2
x
i
z
i
x
n
z
n
z
n+1
¸
T
. (3.28)
Note that this vector corresponds to the entire position vector found in (3.13),
but the prescribed DOFs are left out. The dynamic simulation algorithm could
be found in Fylling et al. (2005).
42
Actuator and Constraints
3.5.9 QuasiStatic Equation of Motion
The quasistatic equation of motion includes the position dependent terms of
(3.27). This means that only the nonlinear stiﬀness for the riser system is found
on the left hand side. On the right hand side the static external forces are given.
These include the top tension, the nonlinear viscous drag forces due to the current,
while the relative velocity is left out, and the speciﬁed, position dependent force
from the TLP. This gives
K(r) r = f
top
+f
cur
−f
TLP,qs
, (3.29)
where
df
cur
=
1
2
ρ
w
C
D
D
e
v[v[dz, (3.30)
f
TLP,qs
= k
spe
x
TLP
. (3.31)
The algorithm for simulation of the quasistatic model is found in Appendix A.4.
3.6 Actuator and Constraints
The riser tensioner system or heave compensator can be implemented as a hy
draulic cylinder with a piston. Today, this setup strives to keep the tension close
to constant. This is obtained by using a compressed air volume as a soft spring
in the hydraulic system. Hence, no active control is needed. Designing the heave
compensator such that the payout is controlled, will still give the same physi
cal constraints. These could be divided into two groups; (1) constraints due to
stroke and (2) constraints due to tension. The stroke parameters are based on
the deﬁnitions by Larsen (1993), slightly modiﬁed, and illustrated in Fig. 3.10.
The initial position refers to a riser and platform condition without oﬀset or en
vironmental forces, and a desired level of top tension. The static position is a
particular case with deﬁned environmental and operational conditions. Payout is
the distance between the bottom of the cylinder and the top of the riser, positive
downwards. The stroke variation is the maximum length variation the tensioner
system can provide. The dynamic stroke is the length variation needed to tension
the riser. The dynamic stroke must compensate for the relative motion between
the platform and the riser subjected to all environmental conditions.
In addition to the boundaries given by the limitations for payout and stroke,
the top tension forces on the riser is physically constrained with upper and lower
boundaries. If this tension is too low, i.e. less than the eﬀective weight of the
riser, the riser will experience buckling. Hence, the lower limit for tension should
be the eﬀective weight plus a safety margin. The upper tension limit is restricted
43
3. Mathematical Modeling
Cylinder
P
i
s
t
o
n
+
Dynamic
stroke
Static
position
position
Initial
Initial
payout
payout
payout
Follows TLP motion
Minimum
Payout
variation
Maximum
Follows the riser motion
z
init
ξ
min
ξ
max
Figure 3.10: Stroke parameters of one individual cylinder.
by the yield stress for the riser material and chosen such that the maximum axial
stress is less than 40% of the yield stress for steel. A given tensioning system will
also have limitations regarding maximum tension due to limitations of pressure
in the hydraulic system, and a tension rate limit on how fast tension may be
varied.
The noncompensated initial position, z
0
, is deﬁned with the design tension
and no environmental forces or oﬀsets, and it will be equal for all risers in the
present study. When the TLP is in an oﬀset position, the setdown of the initial
position can be found from the tendon geometry, such that the initial position
compensated for setdown, ∆s
T
is found from
z
init
(t) = z
0
−∆s
T
(t). (3.32)
The measured parameter used in the control loop is the payout for each piston,
denoted ξ
j
. Payout is deﬁned positive when it adds elongation to the riser and
negative in the opposite direction. Minimum payout, ξ
min
, means that the piston
is as far into the cylinder as possible, for simplicity assumed zero here. Maximum
44
Actuator and Constraints
payout, ξ
max
, refers to the position with maximum free piston length. The initial
payout, ξ
0
, is the distance between the lower end of the cylinder and the initial
position. The total dynamic payout is given as
ξ
j
(t) = ξ
0
+ z
init
(t) −z
j,n+1
(t), j = 1, 2, (3.33)
where z
j,n+1
is the vertical top position of riser j.
45
Chapter 4
Model Veriﬁcation
RIFLEX is a commercial FEM program for static and dynamic analysis of slender
marine structures (Fylling et al., 2005). For the purpose of control system design,
a more convenient numerical procedure and software code has been implemented
in Matlab/Simulink. The motivation behind this chapter is to validate the riser
model from Chapter 3 and verify that it is appropriate as a model of the real world
used in the simulations and for the purpose of control system design, despite the
simpliﬁcations made in the modeling. The model should give a good picture of the
global geometry of the riser, with focus on the maximum horizontal displacement
and the vertical riser top position, exposed to varying TLP positions and tensions.
Features often included in riser analysis, which are of less importance in this case
are:
• Bending stiﬀness EI in the riser model.
• End conditions, i.e. bending stiﬀeners in the top and bottom end points.
• Stress.
Hence, in the presented Simulink model, the simpliﬁcations can be summarized
as:
• No bending stiﬀness included.
• Free rotations in the ends.
• Few elements compared to typical structural analysis, 220 versus 400.
The ﬁrst two of these items are assumed to be insigniﬁcant at large water depths
(large length to diameter ratio), as long as the global geometry is of importance.
In Section 4.7, a bending stiﬀener is implemented at the bottom end node for
47
4. Model Verification
the RIFLEX model, and compared to the Simulink model with free rotation to
investigate the eﬀect of such a simpliﬁcation. Note that bending stiﬀness has been
included in the RIFLEX model in all analyses presented herein. In the veriﬁcation
of the code, a relatively large number of elements (N = 20) are chosen to increase
the resolution and accuracy. Four tests are run:
1. Quasistatic veriﬁcation with increasing TLP oﬀset.
2. Quasistatic veriﬁcation with increasing top tension.
3. Dynamic veriﬁcation with harmonic TLP motions.
4. Dynamic veriﬁcation with harmonic top tension variations.
The quasistatic analyses are run with the static model given in (3.29). Only
the terms proportional to the position r are included, while the dynamic terms
proportional to velocity ˙ r and acceleration ¨r are ignored. The static riser con
ﬁguration is calculated for each increment in TLP position or top tension. The
dynamic analysis are run with the full dynamic model from (3.27). The TLP
position and the top tension are harmonically varied.
4.1 SetUp
The equilibrium solution in each time step is found numerically by incremental
equilibrium correction. The NewtonRaphson method is used for equilibrium
iterations within each time step. The integration method for the dynamic riser
model is the Newmarkβ method with ﬁxed step size and constant acceleration
in each time step. The implemented algorithm is found in Fylling et al. (2005).
For the quasistatic algorithm, see Appendix A.4.
Most of the tests are run at 1200m water depth. For simplicity, the riser top is
assumed to be at the level of the free surface, i.e. 1200m above the seaﬂoor. The
model run in RIFLEX consists of 400 elements, each of length 3m. The model
implemented in Simulink is run with 20 elements which gives an element length of
60m. In order to limit the computation time for realtime control applications, it
is of interest to minimize the number of elements applied in the riser model, while
still maintaining a suﬃcient level of accuracy, such that the implemented model
can represent the geometry of the real world. In Chapter 7, this model with
the number of elements varying between 2 and 20 are analyzed and compared to
investigate the size of the error introduced.
To verify the riser model, the analyses were run with seven diﬀerent current
velocity proﬁles, representing diﬀerent geographical areas. A wide range of cases
were analyzed to better investigate the robustness of the model. The current
48
QuasiStatic Verification with Increasing TLP Offset
velocity proﬁles are illustrated in Fig. 3.2 and described in Section 3.1. These
current proﬁles were run with various current velocity amplitude, with surface
velocity ranging from 0.3m/s to 2.5m/s, with most cases about 1m/s. The current
velocity at the seabed varied from 1.0m/s for the bidirectional current proﬁle
to 1.2m/s for the ampliﬁed GoM2. Most cases were run with between 0m/s and
0.5m/s at the seabed. Current proﬁle data are found in Appendix B.1.1. Physical
riser data are found in Appendix B.2.
The Newmarkβ method is unconditionally stable for linear systems. How
ever, for nonlinear systems the step length the simulations must be small enough
to capture system dynamics also for the stiﬀ vertical DOFs. A step length of 0.1s
is seen to be stable in all DOFs for all simulations.
4.2 QuasiStatic Veriﬁcation with Increasing TLP
Oﬀset
The quasistatic riser model was veriﬁed ﬁrst. The TLP oﬀset was increased in
steps of 5m from 0m to 70m, corresponding to 0 to 5.8% of the water depth. The
static equilibrium solution is found for each TLP position for both the RIFLEX
and the Simulink models. The models were run for all the diﬀerent current proﬁles
and with various current velocity amplitudes. Deﬂections of the riser with the
design current from the Ormen Lange ﬁeld with surface velocity 1.15m/s are
seen in Fig. 4.1a). x and o mark the position along the riser with the maximum
horizontal displacement for RIFLEX and Simulink, respectively. The riser models
match close to perfect. The deviation in the vertical position of the node with
max displacement is due to the relatively low number of elements in the Simulink
model. The deviation for the riser conﬁguration is seen to be small. 20 elements
as seen in the Fig. 4.1, gave as expected a better correspondence between the
RIFLEX and Simulink calculations than fewer elements, like 5 or 10, see Section
7.2.
The setdown is given in Fig. 4.1b). It is seen to correspond nicely for the
two models. The setdown curve reﬂects the setdown due to surge, which also
is used to describe the TLPs motion on a sphere surface. We achieved similar
correspondence in snapshots and setdown for the other current proﬁles.
4.3 QuasiStatic Veriﬁcation with Increasing Top
Tension
In the second veriﬁcation of the quasistatic riser model, the tension was increased
from 1200kN to 2700kN in steps of 50kN. All current proﬁles were tested with
49
4. Model Verification
0 20 40 60
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Snapshots
0 20 40 60
1197.5
1198
1198.5
1199
1199.5
1200
1200.5
TLP position [m]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Setdown
RIFLEX N = 400
Simulink N = 20
Figure 4.1: Deﬂections (a) and setdown (b) for the quasistatic riser model with
TLP oﬀsets from 0m to 70m and Ormen Lange design current. RIFLEX(– x)
and Simulink ( o).
TLP positioned in zero and 30m oﬀset, x
off
= ¦0, 30¦m. The lower tension limit
was the eﬀective weight plus a safety margin for the structure connection at the
seabed. The upper tension limit was given by a percentage of the yield stress for
steel.
Fig. 4.2 shows the deﬂections of a riser exposed to the Ormen Lange current
and increasing tension. As before, x and o mark the position along the riser with
the maximum horizontal displacement for the RIFLEX and Simulink model, re
spectively. The curve to the right with horizontal displacement of approximately
20m corresponds to the lowest tension of 1200kN. For each curve the tension
increases with 50kN and the deﬂection decreases correspondingly. Note that the
vertical position of the max horizontal displacement increases when the tension
increases. The reason is that the relative diﬀerence between top and bottom
tension will decrease with increasing tension. This will lead to a more symmetric
deﬂection shape for increasing tension in uniform current. The tension along the
riser is decreased from top to bottom due to the eﬀective weight of the riser.
Fig. 4.3 shows the deﬂections of the quasistatic riser model exposed to the
bidirectional current velocity proﬁle. Eﬀects of the reduced tension along the riser
are clearly seen at low tensions; the deﬂection is much larger in the lower half than
the upper half, even with a symmetric current proﬁle. This is particularly clear at
low tensions. Also the deviation between the models in RIFLEX and Simulink is
larger than for the other current proﬁles, and is clearly seen from the ﬁgure. The
curvature is large, especially about 200m above the seabed. 20 elements seems to
50
QuasiStatic Verification with Increasing Top Tension
0 2 4 6 8 10 12 14 16 18 20
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Figure 4.2: Deﬂections of the quasistatic riser model with increasing top tensions
from 1200kN to 2700kN in steps of 50kN, and current proﬁle from the Ormen
Lange ﬁeld. RIFLEX(– x) and Simulink ( o).
be too few to describe the riser conﬁguration accurately, i.e. more elements are
needed to describe the conﬁguration satisfactory, if the curvature is large. The
model is seen to be closer to the RIFLEX model for large tensions giving less
deﬂections. However, the important characteristics are kept and the model can
picture the main riser conﬁguration.
The relations between top tension, vertical top position and maximum hori
zontal displacement are investigated. The relations for the Ormen Lange design
current is illustrated in Fig. 4.4. The veriﬁcation showed good agreement between
the models within the tension limits. For lower tensions, nonlinear eﬀects like
buckling may appear. These eﬀects are not implemented in the Simulink model.
Top tension less than the lower saturation limit is therefore not simulated.
The physical correspondence between the vertical top position, maximum
horizontal displacement, top tension and geometric stiﬀness should be explained
more thoroughly. This is best done by considering the numerical derivatives
51
4. Model Verification
−5 −4 −3 −2 −1 0 1
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Figure 4.3: Deﬂections of the quasistatic model with increasing tension exposed
to the bidirectional shear current. RIFLEX(– x) and Simulink ( o).
1199 1200 1201
5
10
15
20
25
M
a
x
d
i
s
p
l
a
c
e
m
e
n
t
[
m
]
Top position [m]
Displacement vs top position
1500 2000 2500
5
10
15
20
25
Top tension [kN]
Displacement vs tension
1500 2000 2500
1199
1199.5
1200
1200.5
1201
Top tension [kN]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Top position vs tension
Figure 4.4: From left: a) Maximum horizontal displacement vs vertical top posi
tion, x
max
(z
max
), b) Maximum horizontal displacement vs top tension, x
max
(T),
c) Vertical top position vs top tension, z
max
(T). All plots are for the quasistatic
riser model exposed to the Ormen Lange proﬁle. RIFLEX(– x) and Simulink
( ).
52
QuasiStatic Verification with Increasing Top Tension
1199.5 1200 1200.5 1201
−12
−10
−8
−6
−4
Vertical top position [m]
∆
x
/
∆
z
1500 2000 2500
−0.03
−0.02
−0.01
0
Top tension [kN]
∆
x
/
∆
T
1500 2000 2500
0
1
2
3
4
x 10
−3
Top tension [kN]
∆
z
/
∆
T
Figure 4.5: From left: a) The derivative of x w.r.t. z,
∆x
∆z
(z) b) The derivative of
x w.r.t. tension,
∆x
∆T
(T), c) The derivative of z w.r.t. tension,
∆z
∆T
(T). All for the
design current at the Ormen Lange ﬁeld, RIFLEX(– x) and Simulink ( ).
∆x
∆z
(z),
∆x
∆T
(T) and
∆z
∆T
(T) in Fig. 4.5. The eﬀect of the increase in tension could
be divided into three diﬀerent stages depending on the tension level:
1. Low tension, large deﬂection.
2. Medium tension, medium deﬂection.
3. Large tension, small deﬂection.
The ﬁrst of these may or may not be clearly seen in the analysis, depending on
the current proﬁle and the top tension. In the exploration of all stages, the top
tension is increased with a ﬁxed magnitude ∆T. For stage 13 above this results
in:
Stage 1: Small tension, large deﬂection
• Large decrease in deﬂection.
• Large increase in top position, due to geometric ﬂexibility.
• A ﬁxed increase of z by ∆z will give a medium decrease in displacement,
∆x.
Stage 2: Medium tension, medium deﬂection
• Medium decrease in deﬂection.
• Medium increase in top position, reduced geometric ﬂexibility; some con
tributions from elastic ﬂexibility.
53
4. Model Verification
• A ﬁxed increase in z gives a large decrease in displacement.
Stage 3: Large tension, small deﬂection
• Small decrease in the deﬂection.
• Small increase in top position, controlled by elastic ﬂexibility, negligible
geometric ﬂexibility.
• A ﬁxed increase in z gives a small decrease in displacement.
The geometric stiﬀness is proportional to the increase in tension, ∆K
G
∝ ∆T,
meaning that a ﬁxed increase in tension gives a proportional change in the geomet
ric stiﬀness. For low tensions, a ﬁxed increase has large eﬀects on the geometry
and the displacements. For large tensions, and already large geometric stiﬀness,
an additional increase in tension, does not have the same eﬀect.
There were also run tests with decreasing top position in RIFLEX. The top
tension started at 2900kN and decreased corresponding to steps in the vertical
top position of 5cm. These result are similar to what is presented here, and can
be found in Rustad et al. (2007c).
4.4 Veriﬁcation with Dynamically Moving TLP
The dynamics of the riser model and the inﬂuence from the drag forces were
tested by harmonic motions in surge for the TLP, with amplitude A
TLP
= 20m,
and periods in the range T
TLP
= 60 − 300s. The static oﬀset was x
off
= 30m,
and the riser tension was kept constant at 1800kN. Fig. 4.6 shows snapshots of
the dynamic riser motion at the Ormen Lange ﬁeld simulated in Simulink with
T
TLP
= 60s. The dynamic envelope curves are found from RIFLEX. The snap
shots ﬁt nicely between the envelope curves given by RIFLEX. The top position
versus time is plotted in Fig. 4.7a), while Fig. 4.7b) shows the correspondence
between the top position and the TLP surge position. The change in top position
is bigger for larger curvature of the riser. This was best seen when the TLP was
moving from its maximum to its minimum oﬀset, with the smallest surge period,
T
TLP
= 60s, seen in Fig. 4.6a).
Let us for simplicity assume modal representation of the riser. Then the riser
top position could be said to be a function of time and space, i.e. TLP motion
and deﬂection
z(t, x) = Σ
i
φ
i
(x) q
i
(t), (4.1)
where q
i
(t) is dependent of time, being the TLP motion and the ﬁrst mode shape.
φ
i
(x) is higher modes due to the deﬂection and resonance for the riser at its
54
Verification with Dynamically Varying Tension
0 10 20 30 40 50
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Snapshots of TLP moving from right til left
0 10 20 30 40 50
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Snapshots of TLP moving from left til right
Figure 4.6: Snapshots of the TLP moving from right to left (a), and from left to
right (b) for the Ormen Lange design current. The snapshots are from Simulink
and the thick, blue lines are envelope displacement curves from RIFLEX.
eigenperiods. Depending on the period of the TLP, more resonance frequencies
can be excited at higher modes. When the TLP is moved from left to right,
Fig. 4.6b), the second mode is seen. This second mode can also be found in
the top position versus time, in Fig. 4.7a), and is even more clearly seen in the
simulations with small current velocities in the top layer, like the GoM2 velocity
proﬁle. The top node position in time for GoM2 is seen in Fig. 4.8a) and the
top node position versus TLP surge in Fig. 4.8b). The ﬁrst order mode shape
is less dominating due to the smaller current velocities and smaller deﬂection.
This will give smaller setdown of the top node, such that the second order mode
shape is more visible. The second order mode shape is due to the TLP motion,
causing resonance in the higher order modes. For longer surge periods, like
T
TLP
= 300s, the motions were close to quasistatic giving a smaller curvature
and only the ﬁrst mode shape was observed, not shown here. Also the setdown
due to deﬂection was smaller, giving less need for stroke capacity than with faster
TLP motions. To summarize, the TLP velocity has a large inﬂuence on the riser
deﬂection, including the second mode shape. The riser setdown can be said to
be dependent on the TLP motions, and hence the need for payout and stroke.
4.5 Veriﬁcation with Dynamically Varying Tension
The fourth and last of the riser model veriﬁcation tests included harmonic ten
sion variations. The initial tension was 1950kN, with an amplitude of 750kN
55
4. Model Verification
0 100 200 300
1198.2
1198.4
1198.6
1198.8
1199
1199.2
1199.4
1199.6
1199.8
1200
Time [s]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Top node position
10 20 30 40 50
1198.2
1198.4
1198.6
1198.8
1199
1199.2
1199.4
1199.6
1199.8
1200
TLP position [m]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Top node xz−trajectory
Figure 4.7: Top position as a function of time (a) and TLP motion (b) with period
of 60s. RIFLEX(–) and Simulink ( ) for the Ormen Lange design current.
0 100 200 300
1199.2
1199.4
1199.6
1199.8
1200
1200.2
Time [s]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Top node position
10 20 30 40 50
1199.2
1199.4
1199.6
1199.8
1200
1200.2
TLP position [m]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Top node xz−trajectory
Figure 4.8: Top position as a function of time (a) and TLP motion (b) with period
of 60s. RIFLEX(–) and Simulink ( ) for GoM2 current and TLP dynamic period
of 60s.
56
Verification with Dynamically Varying Tension
0 5 10 15
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Snapshot
Figure 4.9: Snapshots with harmonically changing top tension for the Ormen
Lange design current (black). The blue lines are the static conﬁguration and
envelope displacement curves from RIFLEX.
and periods of 60s and 120s. This gives maximum tension rates of 75kN/s and
37.5kN/s, respectively. This is far more than the limit of a conventional tension
system. Hence, if the riser model is valid for these large and fast changes in
tension, it will also be valid for slower limit rates. Simulations were run with
currents corresponding to the Ormen Lange ﬁeld, GoM2 and the bidirectional
current proﬁle. Fig. 4.9 shows the snapshots with dynamical tension variation of
period 120s from Simulink. The thick lines are the static conﬁguration for the ini
tial tension and the envelope displacement curves from RIFLEX. The snapshots
from Simulink are seen to be within these envelope curves. Fig. 4.10 shows the
dynamic variation of the top tension (a) and the top position (b). The motion
trajectory for vertical position versus tension is seen in Fig. 4.10 c). A transient
period from the static initial condition is seen. The results from simulations with
60s are similar.
There were also run tests where the motion of the top node in RIFLEX was
57
4. Model Verification
0 100 200 300
1500
2000
2500
Time [s]
T
o
p
t
e
n
s
i
o
n
[
k
N
]
Top tension vs time
0 100 200 300
1200
1200.5
1201
Time [s]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Top position vs time
1000 2000 3000
1200
1200.5
1201
Top tension [kN]
Top position vs top tension
Figure 4.10: Top tension (a) and position (b) as functions of time, and top
position vs top tension (c). RIFLEX(–) and Simulink ( ) for Ormen Lange
design current.
a prescribed vertical sinusoidal. In the corresponding simulations in Simulink,
sinusoidal tension was used. The correlation between tension and vertical top
position is nonlinear and asymmetric, meaning that more tension is needed to
lift the riser 0.5m than the reduction in tension when lowering the top with 0.5m.
This is caused by hydrodynamic drag forces that will increase needed tension for
an upwards motion, but decrease the tension reduction for a downwards motion.
However, the results presented in this section are better suited for the purpose
of veriﬁcation. The results with prescribed motions are not presented here, but
can be found in Rustad et al. (2007c).
4.6 QuasiStatic Veriﬁcation at Shallow Waters
In deep waters, the lateral stiﬀness is assumed to be dominated by the geometric
stiﬀness, and the riser behaves more like a cable. In shallow waters, the bending
stiﬀness is assumed to be of considerable importance. To investigate the area
of application for the riser model with free end rotations, it is tested in shallow
water, being 300m and 600m. In even shallower waters, TLP and tensioned risers
are not likely to be applied. Hence, it is not necessary to investigate the validation
at even smaller water depths. This section is written in cooperation with Stølen
(2007), and the test setup is similar to the corresponding veriﬁcations for 1200m
water depth. A linearly sheared current is used in all tests.
4.6.1 Increasing TLP Oﬀset
At 300m water depth, the TLP oﬀset is varied from 0m to 30m in steps of 2.5m.
At 600m water depth the oﬀset is varied from 0m to 40m in steps of 5m. The
58
QuasiStatic Verification at Shallow Waters
0 10 20 30
0
50
100
150
200
250
300
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Snapshots
0 10 20 30
298
298.5
299
299.5
300
300.5
TLP position [m]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Setdown
RIFLEX N = 100
Simulink N = 20
Figure 4.11: Deﬂections (a) and setdown (b) for the quasistatic riser model with
TLP oﬀsets from 0m to 30m, 300m water depth and linearly sheared current.
RIFLEX(– x) and Simulink ( o).
top tensions are 450kN and 900kN at 300m and 600m, respectively. This is the
same top tension factor compared to the eﬀective weight as for 1200m, with
T
f
= 1.94. The results for 300m and 600m are found in Fig. 4.11. Similar results
are found for 600m. The deviation is small, both for the lateral deﬂections and
the setdowns, similar to what is found at 1200m water depth.
4.6.2 Increasing Tension
The quasistatic analysis is now run with increasing tension at 300m and 600m
water depth. For the 300m case the tension is increased from 250kN in steps
of 25kN up to 700kN. The lower tension limit should be set to give a minimum
tension at the bottom connection of the riser. The eﬀective weight is 230kN which
results in a minimum residual tension of 20kN at the bottom. This is less than
the safety margin, and the boundaries for the validation of the Simulink model.
The eﬀect of bending stiﬀness is not included in the Simulink model such that
the total stiﬀness is too small in particular at low tensions, and hence too large
deﬂections must be expected. The deviations in Fig. 4.12a) for the rightmost
curves are seen to be large (0.12m and 3.2%) for the lowest tension. For higher
tensions, the bending stiﬀness is of less relative importance, and the agreement is
better. Except for the two lowest tensions, which are out of range of the validity
of the Simulink model, the relative error was less than 1%, and decreased for
increased tensions.
59
4. Model Verification
0 1 2 3 4
0
50
100
150
200
250
300
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
300m
0 1 2 3 4 5
0
100
200
300
400
500
600
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
600m
Figure 4.12: Deﬂections for the quasistatic riser model with increasing tension
in (a) 300m and (b) 600m water depth and shear current. RIFLEX(– x) and
Simulink ( o).
At 600m water depth the tension is varied from 600kN to 1200kN in steps
of 50kN. The eﬀective weight is 450kN, giving a safety margin of 150kN. The
snapshots in Fig. 4.12b) show satisfactory performance. The relative error was
less than 0.1%.
4.7 QuasiStatic Veriﬁcation with Stress Joint
The Simulink riser model has free rotation at the ends, and the end conditions are
assumed to be insigniﬁcant for the global geometry in deep waters. The solution,
with respect to stresses, is to tie down a marine riser to the seaﬂoor through a
ball joint. This has zero rotation stiﬀness and will eliminate bending stresses at
the riser end. However, in most cases this is not a feasible solution. The main
purpose of the stress joint is to provide a gradual transition between the relatively
ﬂexible riser and the rigid wellhead. The joint must have suﬃciently ﬂexibility
to keep the bending stresses in the lower part of the riser at an acceptable level,
but also suﬃciently strong to resist the forces and moments introduced by the
riser at the top of the wellhead. In this section the eﬀect of a bending stiﬀener at
the bottom connection will be analyzed. This section is written in cooperation
with Stølen (2007).
A stress joint, with the same material speciﬁcations as the riser, is implement
in RIFLEX. The risers will be located at 300m, 600m and 1200m water depths
60
QuasiStatic Verification with Stress Joint
and tested with increasing tensions like in Sections 4.3 and 4.6.2. No stress joint is
implemented in the Simulink model. The objective is to test what eﬀect the stress
joint will have on the global geometry, and if the assumption of insigniﬁcance of
the end connections is valid for the various water depths. These tests are needed
since it has been assumed that the maximum horizontal displacement and the
vertical riser top position are unaﬀected by the presence of a bending stiﬀener
in deep waters, and that a ball joint can be used at the bottom connection.
Fig. 4.13 shows the stress joint model, and Table 4.1 summarizes the geometric
parameters. The internal diameter is the same for all segments and equal to the
riser. The external diameter is increased from the riser thickness at the top to 4t
at the bottom. Only one element is applied for each segment.
Riser
Segment
i
D
S
i
h
S
i
t
S
i
Figure 4.13: Stress joint conﬁguration.
4.7.1 Shallow Water
The stress joint conﬁguration was ﬁrst tested at 300m water depth. The stress
joint model is seen to give the riser a vertical orientation close to the bottom, see
Fig. 4.14. The stress joint is seen to have a large impact on the curvature of the
riser and also the maximum horizontal displacements. Recall from Section 4.6.2
that the lowest tensions (and rightmost curves) are not valid for the Simulink
model in 300m, which does not include the eﬀects from bending stiﬀness. The
two lowest tension cases give a relative error of 17.3% and 10.1%. For higher
61
4. Model Verification
h
si
[m] t
si
[m] D
si
[m]
Riser  t D
Segment 1 2.7 1.5 t D + t
Segment 2 1.8 2.0 t D + 2 t
Segment 3 1.0 3.0 t D + 4 t
Segment 4 0.5 4.0 t D + 6 t
Total stress joint 6  
Table 4.1: Stress joint data.
0 1 2 3 4
0
50
100
150
200
250
300
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
300m
0 1 2 3 4 5
0
100
200
300
400
500
600
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
600m
Figure 4.14: Snapshots for quasistatic riser model with stress joint of 6m, in
creasing top tension and linearly sheared current at (a) 300m and (b) 600m water
depth. RIFLEX(– x) and Simulink ( o).
tensions the deviations are much smaller decreasing from 6.2% down to less than
1% for the highest tensions.
The snapshots for 600m are shown in Fig. 4.14b). As expected the results are
now in better agreement than for the 300 meter case. The stress joint imposed
at the lower end has less inﬂuence on the curvature due to the increased riser
length. The horizontal displacements are also less aﬀected by the stress joint.
For the lowest tension, the relative error in maximum horizontal displacement is
3.2% and less than 1% for tension higher than 850kN.
62
Discussion
0 2 4 6 8 10
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
6m joint
0 2 4 6 8 10
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
12m joint
Figure 4.15: Snapshots for quasistatic riser model with increasing top tension
at 1200m water depth including stress joint of (a) 6m and (b) 12m exposed to a
linearly sheared current. RIFLEX(– x) and Simulink ( o).
4.7.2 Deep Waters
The riser at 1200m water depth is tested in linearly sheared current with stress
joints of 6m and 12m, see Fig. 4.15. For the 12m stress joint, all dimensions
from Table 4.1 are doubled. The importance of the stress joint is seen to be less
signiﬁcant at this water depth, even for the longest stress joint. The correspon
dence between the RIFLEX and the Simulink model was better with the ball
joint conﬁguration. However, the deviations with the stress joint are still seen to
be small, even for the lowest tensions. The RIFLEX model with 6m stress joint
was also exposed to the Ormen Lange current. This gives larger deﬂection and
larger horizontal deviations in meters (see Fig. 4.16), but the relative deviations
are smaller. In Fig. 4.17 the relative errors in horizontal maximum deﬂection for
each of the three cases are shown. The importance of a stress joint and inclusion
of bending stiﬀness is seen to give largest error for tensions less than 1600kN. For
tension higher than 1600kN, the error is less than 0.5%, even for the 12m stress
joint.
4.8 Discussion
In this chapter the mathematical model developed in Chapter 3 has been veriﬁed.
The veriﬁcation was performed for both quasistatic and dynamic models. Both
models were exposed to TLP motions and tension variations, and several diﬀerent
63
4. Model Verification
0 2 4 6 8 10 12 14 16 18 20
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Figure 4.16: Snapshots for quasistatic riser model with a 6m stress joint and in
creasing top tension in 1200m waters exposed to the Ormen Lange design current.
RIFLEX (–x) and Simulink ( o).
64
Discussion
1200 1400 1600 1800 2000 2200 2400 2600
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
Top tension [kN]
[
%
]
Relative error in horizontal displacement
6m stress joint shear
12m stress joint shear
6m stress joint OL
Figure 4.17: Relative error in horizontal maximum displacement for stress joints
of 6m (–x) and 12m ( o) in linearly sheared current and 6m (.*) stress joint in
Ormen Lange current.
current proﬁles with a wide range of current velocities have been analyzed.
The simulations have shown that there is a good agreement between the
model from Chapter 3 and the RIFLEX model. The quasistatic model was
tested in shallow waters, where the model with free rotations could be used for
drilling purposes. It showed good performance at suﬃciently high tension levels.
For tension levels lower than the limit for the Simulink model, the eﬀect of the
neglected bending stiﬀness was seen. Hence, when applying this model, one
should be aware of this eﬀect, and avoid use of the model for low tension values.
A stress joint was implemented in the RIFLEX model. The eﬀect was large in
shallow waters, especially at the lowest tensions, acting together with the bending
stiﬀness. In deep waters, the eﬀect of the stress joint on the global geometry is
small and the assumption of free rotations at the ends is acceptable. The main
focus has been deep waters and second order LF induced motions, as this is the
65
4. Model Verification
main motivation for the thesis. Note that WF motions have not been included
in this veriﬁcation.
20 elements have shown to be suﬃcient to describe the geometry of the riser
in most cases. However, in cases with large curvature, like the bidirectional
current proﬁle, with zero oﬀset and low tension, 20 elements did not describe
the riser accurately, but could still draw the main conﬁguration. Anyhow, in
fully developed, undisturbed bidirectional current and the lowest top tension, the
largest deﬂection is less than 2m in the upper half and less than 6m in the lower
half, see Fig. 4.3. This is far less than the maximum deﬂection of approximately
20m with the lowest tension for the Ormen Lange current (Fig. 4.2). The small
deﬂections are due to the second order mode shape. As the deﬂections are small,
the risk of collision is limited and hence, no need for control. The Ormen Lange
proﬁle, on the other hand, gives a large deﬂection, which will increase the risk of
collision. There is a good agreement with the RIFLEX model in all the relevant
veriﬁcation tests for this proﬁle. It will therefore be used as the current proﬁle in
the simulations with control. However, in the riser analysis in Chapter 7, some
additional current proﬁles will also be used.
66
Chapter 5
The Riser Control System
Overview
This chapter aims to develop the control architecture and system design for the
riser tensioning system. The main goal of this system is to prevent the risers
from colliding subjected to the given constraints. This includes identiﬁcation of
the various stages of riser behavior depending on the environmental conditions,
the control architecture, i.e. how the various parts are working together and the
controller system design, including actuators and measurements.
5.1 Implementation Overview
The implementation of the system is shown in Fig. 5.1. The diﬀerent parts of
the system, is brieﬂy described below, starting with the physical components.
The environment consists of wind, waves and current. The entire current pro
ﬁle has inﬂuence on the riser conﬁguration. The current proﬁles are de
scribed in Section 3.1.
The TLP is described in Chapter 2 and modeled in Section 3.4. The TLP
motions depend on the environmental conditions current, waves and wind.
The risers are modeled in Section 3.5. The behavior of the two risers depends
on the TLP motions as well as the current velocity proﬁle. On the other
hand, the TLP motions are assumed independent of the risers.
The actuator is the riser tensioner system and a part of the implementation.
By increasing or decreasing the tension and payout, the risers are tightened
or loosened, and the lateral deﬂections are thereby aﬀected.
67
5. The Riser Control System Overview
Controller Environment Actuator Risers
Measurements
TLP
Figure 5.1: Outline of the implementation of the system.
The measurements used by the controller could be the current proﬁle mea
surements, TLP motions, riser inclinations, relative horizontal distance,
tensions and payouts for the risers.
The controller takes input from an operator or computes new optimal setpoints
based on the measured inputs. The output is the control input for the
actuator.
The environment, TLP and risers are described in Chapter 3. This chapter
focuses on the actuator, measurements and controller system.
5.2 Actuator
The actuator in this system is the riser tensioner system, which is designed to
carry the load of the risers and transfer it to the platform structure. In oﬀshore
operations the tension must be maintained independent of the movements of the
platform. The TLP motions will induce motions on the risers and impact the
tension. Current, wind gusts and second order LF wave forces cause the horizontal
motions of the platform, introducing setdown in an oﬀset positions. The WF
oscillations of the riser are caused by ﬁrst order wave loads. The compensation
for the relative vertical motion between the TLP and the risers is called heave
compensation. Heave compensators are usually divided into three groups; passive,
active and a combination of these. The aim of the heave compensation system
is to keep the riser tension unaﬀected by the vertical TLP motions. For riser
systems today, this means to keep the tension close to constant.
Passive heave compensation systems can be described as springdamper sys
tems, and do not require any input of energy under operation. Most passive
heave compensation systems are pneumatichydraulic systems with glycol and
air in compression, and the load is balanced by the pressure of ﬂuid volume act
68
Measurements
(a) Wireline. (b) Nline.
Figure 5.2: The diﬀerent riser tensioner systems (National Oilwell Varco, 2007b).
ing on a piston. The spring is nonlinear due to the compressed gas volume. See
Nielsen (2004) for more details.
There are two types of heave active compensators: (1) Wireline and (2)
Nline riser tensioner systems, see Fig. 5.2. Both riser tensioner systems give
a nearly constant tension to the marine risers and compensate for rig motions.
The Nline system has six cylinders in a ring. The system is symmetrical, so if
one cylinder fails, the opposite cylinder is set free, and the remaining four keep
on working. In drilling operations, the stroke length could be as much as 15m.
In our application 23m could be suﬃcient. A tensioner system for production
risers at Jolliet is shown in Fig. 5.3. The cylinders are installed in frames on the
cellar deck and connected to a tensioner ring on the riser.
5.3 Measurements
To be able to run a supervisory, iterative FEM model we need to know the
incoming current, the TLP motions and the top tension for each riser. To ensure
that this monitoring model is giving a correct result, we could measure the payout
and the top and bottom riser angles. The measured payout and the relative
horizontal positions are used by the controller to calculate feedback.
69
5. The Riser Control System Overview
Figure 5.3: Production riser tensioner system at the TLP Jolliet (National Oilwell
Varco, 2007b).
5.3.1 Current
The incoming undisturbed current proﬁle can be measured using an acoustic
Doppler current proﬁler (ADCP). The current proﬁle can be measured for water
depths down to 2000m, depending on the conﬁguration including the placement
of the sensors, working frequency for the unit and cell size. ADCP measurement
units are currently deployed on environmental monitoring buoys, oﬀshore oil rigs
and polar research moorings (RD Instruments, 2006). With increased operating
depths, a single downward looking ADCP may not provide enough coverage.
Fig. 5.4 shows how current proﬁles can be obtained in water depth larger than
1000m. Three diﬀerent placements of the ADCPs may be used to cover the surface
current, the midcolumn current and the bottom current. Near the surface, the
platform itself introduces changes in the current and wave ﬁeld, due to its size
and the use of thrusters for station keeping for ﬂoaters like the semisubmersible.
Hence, the near surface current can not be reliably accomplished by using a
current meter that measures the current in the direct vicinity of the rig. By
using a horizontally directed ADCP the undisturbed, near surface current can be
70
Measurements
Figure 5.4: Using ADCP to cover the full current proﬁle, including near surface,
midcolumn and bottom current (www.rdinstruments.com).
measured. While the near surface current is of main importance for the oﬀset
of the rig or platform, the midcolumn velocity is aﬀecting the riser deﬂections.
The ADCP measuring the midcolumn velocity is mounted on the rig or near the
surface and directed downwards.
To measure the bottom current, an upwardlooking ADCP is attached to the
bottom structure or moored to the seabed. This bottommounted ADCP system
collects current data and transmits it to the surface in realtime via an acoustic
modem. In this way, the data of the upwardlooking ADCP is combined with the
downlooking and near surface data, to form the a continuous, fullwatercolumn
current proﬁle.
71
5. The Riser Control System Overview
5.3.2 TLP Motions
Several position measurement systems are commercially available, such as the
local hydroacoustic position reference (HPR) system and the global navigation
satellite system (GNSS). The two commercial satellite systems available today
are the US system NAVSTAR GPS (NAVigation Satellite Timing And Ranging
Global Positioning System) and the Russian GLONASS (GLObal NAvigation
Satellite System), see Fossen (2002) and references therein. In addition a Euro
pean satellite navigation system, Galileo, is under construction, estimated to be
gin operational services in 2010 (European Commission, 2007). The GPS system
provides position measurements with an accuracy less than 1m in the horizontal
direction, which is the typical accuracy for ship positioning systems today.
To ﬁnd the acceleration and velocity of the TLP we can use available com
mercial inertial measurement technology. An inertial navigation system (INS)
consists of a measurement part and a software part. The measurement part is
called an inertial measurement unit (IMU), measuring the linear and angular ac
celerations. The software part is a stateobserver, which computes the position,
velocity and attitude from the measurements. To prevent the system from drift
ing, a GNSS is used as a reference for the position, and the resulting system is
known as a strapdown inertial navigation system (Fossen, 2002).
It should also be noted that the company SeaFlex in cooperation with Kongs
berg Maritime has developed a riser position reference system (RPR) which is
an independent riser position reference system for a dynamic positioned drilling
vessel. The position estimation is based on measured top tension and the top and
bottom riser inclinations. The RPR is calibrated with other available position
reference measurements (Høklie et al., 2002).
5.3.3 Tension and Payout
The top tension can be found by measuring the pressure inside the pneumatic
hydraulic pistons. When we know the area of the piston stamp, the tension force
can be calculated. The payout is found by counting magnetic coils along the
piston. This is an accurate method down to the mmlevel.
5.3.4 Top and Bottom Angles
The top and bottom angles for each riser could be measured by inclinometers,
either electronic riser angle (ERA) or acoustic riser angle (ARA) measurements.
The riser end connections are in this study modeled with free rotations. Recall
that it in the industry is common to use bending stiﬀeners to prevent large
rotations in the connections. We may therefore calculate larger end inclinations in
this study than what is actually experienced in the industry. This simpliﬁcation is
72
Controller
made as the main objective is to calculate the global geometry, and in deep waters
this simpliﬁcation will only introduce small errors in the end point position, see
Section 4.7.
5.3.5 Relative Horizontal Distance
To determine the relative horizontal distance between two risers, one could use
acoustic location. In water this technique is known as sonar (SOund Navigation
And Ranging). There are two types; passive sonars, which listens without trans
mitting, and active sonars which creates pulses (pings) in order to produce echo.
The time from transmission of a pulse to the reception is measured and converted
to distance when we know the speed of sound in water (approximately 1500m/s).
For a travel distance of 15m, the time will be 10ms, which is satisfactory for
the sampling time here. The active sonar consists of a transmitter and receiver.
These could be placed at the same location (monostatic operation) or at separate
locations (bistatic). When more transmitters are used, spatially separated, it is
called a multistatic operation (Wikipedia, 2007b).
5.4 Controller
To prevent collision between risers, control of the top tension is introduced. This
dynamically control output could either be directly based on one or two of the
measurements presented above, or there could be a model of the system (CPM)
running in parallel, which use measurements as inputs to decide which action to
take and calculate the feedback. Here, the ﬁrst approach is used in investigation
of the various control strategies, while the second approach is used in a more
realistic case scenario.
Marine systems of today are required to operate in a wide range of envi
ronmental conditions while still giving acceptable safety and performance. For
permanent installations, like the TLP, the allyear weather changes must be taken
into consideration, also when designing the control system. In the North Sea the
weather conditions can be harsh, and the TLP will experience large LF and WF
motions, which again aﬀect the risers. When installed in deep waters, the risers
are more ﬂexible than in shallow waters and behave more like cables. In addi
tion, diﬀerent static and/or dynamic properties of the risers in the array should
be taken into consideration in the controller design. Hence, there could be large
variations in the requirements for the controllers depending on both the opera
tional conditions, riser properties and the water depth. We will later on denote
this as regimes.
There may be two implementations of controller designs; one nonlinear con
troller covering all possible regimes or a combination of various controllers into
73
5. The Riser Control System Overview
Environment
System Controller Set
Supervisor w
u
y
y
Figure 5.5: The hybrid control system.
one. The ﬁrst solution may give a complicated controller, which could be almost
impossible to design if the dynamics changes considerably. The second solutions is
a concept of supervisory control switching between a set of controllers. Each con
troller is made for a prespeciﬁed operational regime. Designing a controller for
each regime is easier than one for all regimes. On the other hand, the drawback
could be chattering between controllers, and complex logics and synchronizing
functions handling the switching.
This switching between controllers may lead to instability. More general so
lutions are therefore developed to secure stability. A systematic approach can be
found in Hespanha (2001), Hespanha and Morse (2002), Hespanha et al. (2003)
and references therein. This strategy is called supervisory control, and makes it
possible to switch between the diﬀerent controllers corresponding to the current
operational condition. A hybrid system often consists of continuous controllers
and discrete switching between them. Nguyen (2006) and Nguyen et al. (2007)
have proposed a hybrid control structure for dynamic positioning systems of ma
rine vessels. The work on supervisory control in this thesis is motivated by the
work of Nguyen (2006).
The two main blocks in the supervisory switched control system are the super
visor and the controller set (Fig. 5.5). The block representing the environment,
w, aﬀects the physical marine system, consisting of the TLP, the riser array, the
actuator and the measurements. The measured values, y, are used as inputs in
the supervisor and the controller. The supervisor monitors the conﬁguration of
the diﬀerent risers and decides which controller set to use. The controller switch
ing signal, σ, is sent to the controller. The correct controller is activated, and the
output, u, is then sent to the actuator, forcing the risers to behave as desired.
74
Controller
The supervisor and the controllers are described in detail in Chapter 6.
75
Chapter 6
The Controller Architecture
This chapter aims to give a more thorough presentation of the controller ar
chitecture. This includes speciﬁcations of the diﬀerent control objectives and a
description of the riser operational conditions. The concepts and properties for
a switched system in general are presented. A more in depth deﬁnition of terms
and description of the supervisor, the controller and their components are given.
All in all this describes the controller architecture and its components used in
the simulations.
6.1 Control Objectives
The control objectives may be formulated based on two diﬀerent principles sub
jected to available measurements. This can be measurements at the TLP well
head, such as payout and top tension, or available measurements along the riser.
The ﬁrst control objective principle is based on wellhead measurements, whereas
the second principle uses measurements of the relative horizontal distance be
tween the risers, at one or more predeﬁned depths, keeping it at a desired dis
tance. For simplicity the objectives are presented for two risers, but they could
easily be extended to an array of risers.
6.1.1 Control Objectives Based on Measurements at the Top
Today the top tension is kept close to constant by a passive heave compensation
system for each riser
T
1
= T
2
, (6.1)
where T
j
is the tension of riser j = 1, 2. Keeping the tensions constant and equal
in both risers may lead to the following scenario: For the two risers in a tandem
arrangement, R2 is in the wake of R1. Due to the shielding eﬀects on R2, R1 will
77
6. The Controller Architecture
T
1
= T
2
ξ
1
= ξ
2
ξ
1
+ l
R1
= ξ
2
+ l
R2
Figure 6.1: Eﬀect of equal tension (a), equal payout (b), and equal eﬀective length
(c).
experience larger drag force than its downstream neighbor. If both risers have
the same top tension, the deﬂection of R1 will exceed the deﬂection of R2, and
the two risers may collide, shown in Fig. 6.1 a).
Another strategy is equal payout by connecting all risers to a common frame,
proposed by Huse and Kleiven (2000), see Fig. 6.1b). This will give varying top
tension on the risers depending on the drag forces and the position in the riser
array. Equal payout by adjusting the top tension through automatic control is
studied in this work, with the control objective
ξ
1
= ξ
2
, (6.2)
where ξ
j
is the payout of riser j. Further work proposed in Rustad et al. (2007b,c)
has shown that due to this tension variation, two equal risers will experience
diﬀerent length due to axial strain according to
∆l
R
=
∆T
EA
l
R0
, ∆T = T
1
−T
2
, (6.3)
78
Control Objectives
where ∆l
R
is the length variation due to the diﬀerence in top tension ∆T of the
two risers. l
R0
is the untensioned initial riser length. Applying equal payout
collision can still occur, but less frequent and in a smaller riser segment than for
the equal tension strategy under the same environmental conditions. The risk of
collision increases with increasing depth. In addition to the eﬀect of axial strain
increases for longer riser lengths and with larger tension variations.
We therefore propose a new control strategy letting the risers have equal
eﬀective length. By using automatic control of the heave compensators and
top tension, the sum of payout and riser length should be equal such that the
controller objective can be formulated as
ξ
1
+ l
R1
= ξ
2
+ l
R2
, (6.4)
where l
Rj
is the length of riser j. This means that in contrast to the strategy of
equal payout we also compensate for the axial elasticity due to the tension vari
ation, see Fig. 6.1c). By introducing this way of controlling the top tension, the
risers may be placed with closer spacing without increasing the risk of collision.
Equation (6.3) is a simpliﬁcation valid for equal risers. For two or more risers
with diﬀerent characteristics with respect to diameter, riser material or ﬁlling, a
more general expression is needed. The riser length can be formulated as
l
Rj
= l
Rj
(T
0j
) + ∆l
Rj,
∆l
Rj
=
T
j
−T
0j
EA
l
R0
, (6.5)
where l
Rj
(T
0j
) is the initial length of riser j with the initial tension T
0j
. ∆l
Rj
is the elongation of riser j relative to its initial tension. The riser material is
assumed linear as long as the tension is much smaller than the yield stress. Hence,
an increase ∆T in top tension will give the same increase for all elements along
the riser, and (6.5) is a good estimate on the riser elongation due to tension.
Equation (6.4) could be then be rewritten as
ξ
1
+ l
R1
(T
0j
) + ∆l
R1
= ξ
2
+ l
R2
(T
0j
) + ∆l
R2
. (6.6)
The initial length and static payout can be found individually. However, note
that the payouts need to have the same initial positions for the equation to be
valid, otherwise this diﬀerence needs to be included in the equation.
To summarize, we propose three diﬀerent control objectives based on top
measurements:
1. Equal tension: T
1
= T
2
.
2. Equal payout: ξ
1
= ξ
2
.
3. Equal eﬀective length: ξ
1
+ l
R1
= ξ
2
+ l
R2
.
79
6. The Controller Architecture
6.1.2 Control Objectives Based in Measurements Along the
Riser
The second control objective principle is based on available measurements of the
relative horizontal distance between the risers. In order to achieve near parallel
risers, the horizontal distance between them should be constant and equal along
the entire length of the risers. With only one measurement along the riser length,
the distance between the risers ∆x
R12,m
(z) should be equal to the distance at the
top and bottom being the desired distance ∆x
d
. Hence,
∆x
R12,m
(z) = ∆x
d
. (6.7)
The measurement should be placed where the risers are likely to be closest, which
is determined by the current proﬁle. Note that we in this control objective do
not need to consider the elasticity of the riser material directly.
6.1.3 Discussion
For the ﬁrst principle of control objectives, reliable and accurate measurements
of payout and tension are available today. Furthermore, the measurements and
the actuator (top tension force) are found at the same location, thus preserving
passivity properties of the closed loop system more easily due to collocated con
trol. To calculate the total riser length as a function of time and tension, the
initial riser length corresponding to the initial tension needs to be known.
If supervisory switched control is used, a system model is required and ad
ditional measurements of TLP motions, undisturbed current velocity proﬁle and
a good model for the hydrodynamic interaction are needed. These are used to
calculate the smallest relative distance between the risers.
For the second principle of control objectives, the relative distance between the
risers is measured directly. Hence, we do not need the additional measurements
or a good process model for the linear controllers. The drawback of this method is
that we are limited to a ﬁnite number of measurements, with predeﬁned location
along the riser. Also, we do not know if these are placed at the water depth
where the risers are closest. As the measurements will be along the riser and
the actuator is located at the top end, this setup is not collocated. However,
the system is much slower in the horizontal than the vertical direction, such that
this may not be a problem after all. This may introduce some scattering in the
vertical direction if the reference is not slow enough, but the horizontal direction
will still be stable. We will in this work focus on the ﬁrst principle of control
objectives, as described in Section 6.1.1.
80
Riser Operational Conditions
6.2 Riser Operational Conditions
The changes in the environmental conditions, in addition to the various riser
types during operation and production, may require diﬀerent control algorithms
during the lifetime of a TLP/riser system. The purpose of the control system
is to prevent collision for all environmental conditions, riser types and water
depth, denoted as regimes or riser operational condition (ROC). A classiﬁcation
of the various regimes could help the design of appropriate controllers and smooth
switching between them. The risers in an array may often have diﬀerent phys
ical properties. The external diameter, wall thickness, material and density of
the internal ﬂuid could vary, depending on whether the riser is used for drilling,
production, export or workover to mention some. Together these factors decide
the dynamics of the riser. The applications and properties of the riser, the riser
characteristics (RC), also aﬀect the controller gains. A speciﬁcation of the dif
ferent ROCs and the controllers based on these leads to a supervisory system
where the time parameters and the necessary controller components are included
according to the prevaling ROC. This concept of ROC is motivated by the work
on vessel operational conditions (VOC) by Perez et al. (2006).
The aim of this section is to highlight the essential characteristics of the
conditions that aﬀect the dynamics of the riser system and use this information
to decide which control action to perform.
6.2.1 Riser Characteristics
During drilling and production on oﬀshore ﬁelds diﬀerent riser types are used
for diﬀerent operations and purposes. Flexible risers or SCRs are not considered
here. All risers referred to in this work are vertical steel risers, connected to a
TLP. Some typical riser applications are:
• Drilling.
• Production.
• Workover/Maintenance.
• Export.
• Import.
• Injection.
The risers are specially made for each purpose, giving diﬀerent properties which
decides the risers physical behavior, both statically and dynamically. The most
important parameters aﬀecting the dynamics of a steel riser are:
81
6. The Controller Architecture
• External diameter.
• Cross sectional area and wall thickness.
• Density of the riser contents.
• Riser length.
• Top tension.
• Elasticity.
Some of these parameters are closely related. The top tension level is dependent
on the weight of the riser, i.e. the cross sectional area, the length of the riser and
the density of the internal ﬂuid.
The riser elongation is proportional to the modulus of elasticity times the cross
sectional area. Hence, diﬀerent cross sectional area will give diﬀerent elongation
according to the stiﬀness EA. The cross sectional parameters also contribute to
a weight diﬀerence, increased or decreased by the diﬀerence in the density of the
internal ﬂuid between the risers. The eﬀective weight decides the tension level
along the riser, which together with the riser length is of major importance for
the eigenperiods of the risers. For two otherwise equal risers, a diﬀerence in the
variation of the contents aﬀectes the total eﬀective weight and the eﬀective weight
gradient. This could be seen statically as at which water depth the deﬂection is
largest. A larger eﬀective weight will have its maximum deﬂection at larger water
depths than for otherwise equal risers.
The upper tension limit is decided by the yield stress for steel, whereas the
lower limit is given by the eﬀective weight plus a safety margin. Hence, for the
otherwise equal risers with diﬀerent contents, the upper tension limit is the same,
but the lower limit is smaller for the riser with lighter contents.
Recall that the drag forces on the riser are proportional to the external di
ameter. For two risers with diﬀerent external diameter, but equal weight per
unit length, the riser with the largest diameter needs a larger top tension to
compensate for the horizontal displacement due to the drag forces.
In extreme cases the longest riser eigenperiod can be close to the typical LF
motions. For another riser case, the ﬁrst eigenperiod could be close to the slowest
WF motions. Hence, depending on the diﬀerent dynamic properties and corre
sponding eigenperiods, resonance may occur at diﬀerent frequencies. In addition
it should be noted that the current ﬁeld behind a riser and the shielding eﬀect,
is depending on the diameter of the upstream riser. The eﬀects of the riser prop
erties are not further investigated here, but should be taken into consideration
when deciding the control strategy and the controller gains.
82
Riser Operational Conditions
Current
TLP
Dependening on
environmental conditions
Profile and velocity variation
Calm
Strong
Stationary Dynamic
Control
No
Control
Riser
Characteristics
Initial
Conditions
Figure 6.2: Riser operational conditions and inﬂuence on the controller choice.
6.2.2 Riser Operational Conditions
The ROC may be described by a triplet of attributes for a given riser
ROC := ¸Current, TLP, Riser Characteristics) (6.8)
where
Current refers to the current proﬁle and velocity, acting on the riser array.
TLP refers to the motion of the ﬂoater which is aﬀected by the environment;
wind, waves and current.
Riser Characteristics (RC) refers to the static properties like riser length,
diameters and crosssectional area as well as the density of the internal
ﬂuid which may vary during the risers lifetime depending on the operation
purpose.
The diﬀerent RC along the outofplane axis decides which control objective to
apply and an appropriate set of control gains. Figure 6.2 shows how these at
tributes are linked to the demand for control. The motion of the TLP due to
the environmental conditions is the main parameter to compensate for. In harsh
weather conditions the TLP will experience larger forces and motions and will
induce more motions on the risers than in calm weather. Going to the right along
the TLP axis will result in more motions and demand for a controller.
83
6. The Controller Architecture
Large ocean currents give larger riser displacements. In addition, the char
acteristics of the current proﬁle decide the riser conﬁguration and where the
deﬂection is largest. Going upward with increasing current velocity and possibly
large changes in the velocity ﬁeld may also require a controller. Hence, the fur
ther we move up and to the right, the larger is the demand for a controller. The
decision is made in the switching logic block, see Section 6.4.3.
6.3 Switched Systems  Concept and Properties
This section is based on Hespanha (2001), Hespanha and Morse (2002) and Hes
panha et al. (2003). The main idea with the supervisory switched controller
is to automatically switch between diﬀerent controllers depending on the situa
tion. Note that the supervisor and the controller are decoupled. Hence, between
switching times, the process is connected to one of the candidate controllers only,
and the dynamics of the supervisor is not visible in the closedloop system. This
simpliﬁes the stability analysis of the system. Detailed behavior of the supervisor
or the controller set can be abstracted and we can concentrate on a small set of
properties for the system.
Estimatorbased supervisors continuously compare the behavior of the process
with the behavior of the estimators or admissible process models to determine
which model is best describing the actual process. The set of admissible process
models considered in estimatorbased supervision is
/:=
_
p∈P
/
p
. (6.9)
Each process model maps to a controller, which provides satisfactory performance
for each model. The controller set is denoted
( :=
_
q∈Q
(
q
. (6.10)
p and q are the parameters taking values on the set T and Q, respectively. A
controller selection function χ : T →Q maps each parameter value of p ∈ T with
the corresponding index q = χ(p) ∈ Q of controller (
q
which provides satisfactory
performance when connected to model /
p
. The estimator and controller may
be described by linear or nonlinear systems according to
/:= ¦˙ x
E
= A
E
(x
E
, u, y), y
p
= C
E
(p, x
E
, u, y),
e
p
= y
p
−y : p ∈ T¦, (6.11)
( := ¦˙ x
q
= F
q
(z
q
, y), u = G
q
(z
q
, y) : q ∈ Q¦, (6.12)
84
Switched Systems  Concept and Properties
where x
E
is the estimation of the state vector, u is the control input vector, y is
the process output vector, y
p
is the output estimate vector, e
p
is the estimation
error vector, and z
q
is the state vector of the controller. The decision logic
compares all the estimation errors and e
p
can be regarded as a measure of the
likelihood that the actual process is inside the ball /
p
.
6.3.1 System Properties
For the formal stability proofs of supervisory switched system, the following def
inition of a switched system (Hespanha, 2001) is used
˙ x = A
σ
(x, w), (6.13)
e
p
= C
p
(x, w), p ∈ T, (6.14)
where x denotes the state vectors of the process, the multicontroller and the
multiestimate, and w is the environmental disturbance. The two basic properties
for the switched system are the matching and detectability properties.
Matching Property means that the set of estimators should be designed such
that each particular y
p
provides a good approximation to the process output
y. This means that e
p
is small whenever the actual process is inside the
corresponding /
p
.
Detectability Property means that for every ﬁxed estimator, the switched
system must be detectable with respect to the estimation error e
p
when the
value of the switching is frozen at σ = χ(p) ∈ Q.
6.3.2 Switching Logic
The index σ of the controller in the feedback loop is determined by the switch
ing logic which takes the estimation error vector e
p
as an input. According to
the certainty equivalence principle, when a particular estimation error e
p
is the
smallest element in the error vector e
p
, the index p ∈ T is the likely value of
the parameter. Hence, the corresponding p ∈ T describes the ongoing process
best and is mapped to the controller switching signal σ = χ(p) ∈ Q. To prevent
chattering, a delay is introduced in the switching process. This could for instance
be dwelltime switching or hysteresis switching. Since the value of p which cor
responds to the smallest e
p
varies, a process switching signal ρ : [0, ∞) → T is
introduced to indicate the current estimate ρ(t) ∈ T of the index p used in the
feedback loop. The mapping from the process switching signal to the controller
switching signal is written σ = χ(ρ). Note that this output for the switching
logic is the one that determines which controller should be used. The small error
85
6. The Controller Architecture
and nondestabilization properties need to be satisﬁed by the monitoring signal
generator and the switching logic.
Small Error Property means that there is a bound on e
ρ
in the terms of the
smallest signal e
p
for a process switching signal for which σ = χ(ρ). The
norm of e
ρ
should be guaranteed, by the switching logic, to be smaller than
a constant times the norm of the smallest e
p
.
NonDestabilization Property The switching signal σ is said to have the non
destabilization property if it preserves the detectability in a timevarying
sense, i.e. if the switched system is detectable with respect to the switched
output e
ρ
for ρ and σ = χ(ρ). This property is satisﬁed if:
• The switching is slow on the average by using dwelltime switching
logic which strictly is used among linear models and controllers.
• The switching stops in ﬁnite time using scaleindependent switching
logic which can be used in switching between both linear and nonlinear
models and controllers.
For the switching logic there is a conﬂict between the desire to switch to the
smallest estimation error to satisfy the small error property, and the concern of
too much switching which may violate the nondestabilization property.
6.3.3 ScaleIndependent Hysteresis Switching
Since the modeling of the riser system is nonlinear, the scaleindependent hystere
sis switching based on Hespanha (2001) will be used. The hysteresis switching
logic in general slows down the switching based on the observed growth in esti
mation error vector e
p
instead of for a ﬁxed dwelltime. The monitoring signal
µ
p
for each process is based on the norm of the estimation error and is deﬁned
by Hespanha (2001) as
˙ µ
p
= −λµ
p
+ γ(e
p
), p ∈ T, (6.15)
where λ is a nonnegative forgetting factor, µ(0) > 0, γ is a class /function
1
and   is a norm. The switching procedure is illustrated in Fig. 6.3. h is a
positive hysteresis constant, and ρ = arg min µ
p
returns the index of the smallest
monitoring signal µ
p
. The monitoring signal for the prevailing model, µ
ρ
, is
compared to the other monitoring signals. When a new signal µ
p
times the
contribution from the hysteresis is smaller than the prevailing signal µ
ρ
, the
system switches. Otherwise, the actual value is held.
1
A continuous function α : [0, a) → [0, ∞) is said to belong to a class K if it is strictly
increasing and α(0) = 0 (Khalil, 2000).
86
Switched Systems  Concept and Properties
Do switch
Do not switch
µ
ρ
≤ (1 + h)µ
p
, ∀p
No
ρ := arg minµ
p
σ := χ(ρ)
Start
Yes
Figure 6.3: Scaleindependent hysteresis switching logic (Hespanha, 2001).
To summarize, the matching and detectability properties are important for
the multiestimator and the multicontroller, respectively. The smallerror prop
erty makes sure that the selected controller is the best. When the process is
between two regimes, fast switching may aﬀect the stability of the system. Non
destabilization will prevent chattering by providing switching logic such as hys
teresis switching.
6.3.4 Model Concept Deﬁnitions
The supervisor described in Hespanha (2001), Hespanha and Morse (2002) and
Hespanha et al. (2003), consists of a set of admissible models which are compared
to the actual process. The model which best describes the ongoing process is used.
The terms admissible process model and actual process used in their works, and
also given here in Section 6.3, can not be directly transferred to the setup and
mathematical models used in this work. Furthermore, the diﬀerence between an
observer and an independent model needs to be clariﬁed.
Observers or state estimators are aiming to obtain an estimate of the current
state of the dynamic system by using available measurements online (each
time step). An observer often copies the dynamics of the system and add
87
6. The Controller Architecture
an injection term constructed such that the state estimate should be a
reconstruction of the unmeasured state.
Independent models are also used to estimate the states of dynamic systems.
The states are estimated based on inputs and the system model. These
models do not use online measurements to correct the states of the system.
They are therefore dependent on accurate system models and inputs to
calculate good results. Notice that also here the model will be updated
based on available information and measurements. However, this is not
done by an injection term.
What Hespanha (2001), Hespanha and Morse (2002), Hespanha et al. (2003) refer
to as the actual or ongoing process, corresponds to what we denote the physical
process or the process plant model (PPM). The main diﬀerence is found for the
supervisor. In this work we have used only one nonlinear model type, denoted
ACPM (accurate control plant model), covering all the operational regimes. This
follows a similar approach as in Reite (2006).
However, the desired distance between the risers is deﬁned according to the
operational regimes. Depending on the deviation from the desired riser conﬁg
uration, a model error e
p
, corresponding to the estimation error, is deﬁned and
calculated for each regime p ∈ T.
Hence, this independent process model (ACPM) and the resulting error mod
els e
p
correspond to the set of admissible process models or multiestimators in
the deﬁnitions in Section 6.3.1. The matching and detectability properties are
deﬁned for observers, and are not directly applicable for the independent model,
in sense of global exponential observers. However, the matching property with
respect to the accuracy of the model and the boundaries for the diﬀerent regimes
are meaningful. The switching logic is used, even though the observer properties
are not relevant. The second part with the small error and nondestabilization
properties are of importance.
6.3.5 Switched System Stability Analysis
The scaleindependent hysteresis switching logic guarantees the nondestabilization
and the small error properties by the following theorem:
Theorem 6.1 (ScaleIndependent HysteresisSwitching, Hespanha (2001))
Let σ be the switching signal and N
σ
(τ, t), t > τ ≥ 0 be the number of discontin
uous of σ in an open interval (t, τ). Let T be a ﬁnite set with m elements. For
any p ∈ T we have that
N
σ
(τ, t) ≤ 1 + m +
mlog(
µp(t)
ǫ+e
−λtǫ
0
)
log(1 + h)
+
mλ(t −τ)
log(1 + h)
, (6.16)
88
Supervisor
No control
Stationary Controller
ACPM
Monitor
Switching
logic
Dynamical Controller
Error State Manuevering
Marine
System
Environment
Controller Set
Supervisor Model
w
y
u
y
Controller 1
Controller 2
Controller 3
Controller 4
Figure 6.4: Structure of the switching control concept.
and
_
t
0
e
−λ(t−τ)
γ
p
(e
ρ
(τ))dτ ≤ (1 + h)mµ
p
(t) (6.17)
where ǫ, ǫ
0
are nonnegative factors, with at least one of them strictly positive.
Proof. See Hespanha et al. (2000). (6.16) guarantees the nondestabilization
of switching, while (6.17) guarantees the small error property. For details on the
stability analysis, see Hespanha et al. (2000).
6.4 Supervisor
The supervisor’s main task is monitoring of the riser array, deciding which control
action to perform and trigging the correct controller. The supervisor consists of
three parts; the riser model, the monitor and the switching logic, see Fig. 6.4.
The ACPM calculates the conﬁguration of the risers and is the most accurate
estimate of the riser system behavior. As inputs it takes the measured
89
6. The Controller Architecture
undisturbed incoming current, the TLP motions and the top tension for
each riser. One model is running for each riser.
The monitor is continuously comparing the calculated riser conﬁguration data
from the ACPM and the measured tension with a set of values describing
the diﬀerent regimes. Based on the regimes and the deviation from the
desired riser conﬁguration, a monitoring signal µ
p
, is calculated. p ∈ T is
the regime index number.
The switching decides which controller to use and which action to perform
based on the inputs from the monitor.
6.4.1 Accurate Control Plant Model
The ACPM is a nonlinear, iterative FEM model of the riser system based on the
PPM from Chapter 3. It is used to accurately estimate the state of the riser
system and its behavior. It is characterized as an independent model, as it is not
taking any measurements, in terms of an injection term, to update its estimated
states.
As inputs it takes the measured undisturbed incoming current, the TLP po
sition, velocity and acceleration, and the top tension for each riser. When we
know these inputs, we can easily calculate the conﬁguration of R1. For R2 we
know its top tension, the TLP motion, the position of R1 and the undisturbed
current. The current acting on R2 is a function of the incoming current on R1
and the distance between the two risers. We will therefore need iterations to ﬁnd
the right position of R2. For the risers further down the array, the inputs and
procedures are similar.
The ACPMs are simulated with fewer elements (812) than the PPMs (1520).
This allows the ACPMs to run faster, but some of the accuracy is lost. However,
the number of elements used for the ACPM is depending on the accuracy needed
in the monitor. This is further analyzed and discussed in Chapter 7.
6.4.2 Monitor
The diﬀerent regimes can be divided into ﬁve: (12) normal regimes where all
riser nodes are close to their desired positions, or the deviations are small and the
tension variations have been small for a given time period, (3) the risers have a
medium deviation from the desired conﬁguration due to slow variations from LF
TLP motions and tide, (4) the risers are subject to fast and large disturbances
due to TLP motions or currents, (5) the risers are in a (near) collision situation
called an error state.
90
Supervisor
R1 R2 R2
d
∆x
d
∆e
rel
∆x
R12
Figure 6.5: Actual and desired riser positions.
The switching conditions are mainly deﬁned by the deviation from the desired
horizontal distance between the risers
∆e
rel
= ∆x
d
−∆x
R12
, (6.18)
∆x
R12
= x
R2
−x
R1
, (6.19)
where ∆x
d
is the vector of the desired distance between the risers in each node,
∆x
R12
is the horizontal distance between two neighboring risers, and x
Rj
is the
horizontal position vector of riser j. For a riser segment, the scalar values are
found in Fig. 6.5. ∆e
rel
is the (scalar) deviation from the desired R2 position,
R2
d
, in the riser segment shown here. The maximum deviation along the riser
is deﬁned as ∆e
rel

∞
. The max deviation is given a value in diameters, D,
depending on the regime. The monitor comparing values in Table 6.1 is based on
a desired relative riser distance ∆x
d
= 15D, which is the distance at the top and
bottom end points used in the simulations here at 1200m water depths. The
model error vector e
p
used in the calculations of the monitoring signals is mainly
found from the maximum deviation from the desired riser distance. In some cases
additional conditions on the horizontal R2 velocity, ˙ x
R2
, or the maximum tension
variations for a given period of time [∆T(t, t
0
)[
max
are included. The switching
conditions for the error calculations are found in Table 6.1.
The relative velocity vector ∆˙ e
rel
tells how fast the risers are moving towards
or away from each other, where how fast they are moving towards each other is of
main importance. This number will be small in most cases since the risers move
91
6. The Controller Architecture
Regimes Monitor comparing values µ
p
ρ σ
Normal 1 ∆e
rel

∞
= 0D µ
1
1 1
Normal 2
∆e
rel

∞
= 1.5D
µ
2
2 1
& [∆T(t, t
0
)[
max
≤ 20kN
Slowly varying ∆e
rel

∞
= 2.5D µ
3
3 2
Fast changing
∆e
rel

∞
= 2.5D
µ
4
4 3
& [ ˙ x
R2
[
max
≥ 0.6m/s
Error state ∆e
rel
 ≥ 13D µ
5
5 2
Table 6.1: The switching conditions.
slowly, such that the variance in position will be captured by considering the
relative distance. Furthermore, if both risers are moving fast, for instance due to
WF TLP motions, ∆˙ e
rel
would not necessarily capture these motions, since the
relative motion might be small. On the other hand, if ∆˙ e
rel
is small, it may not
lead to an instant collision. By considering only ˙ x
R2
, the fast TLP motions are
captured. If R2 is exposed to varying drag forces due hydrodynamic interaction,
this could be captured both by considering ∆˙ e
rel
and ˙ x
R2
. For activation of the
dynamical controller, we will consider ˙ x
R2
.
The scalar model error e
p
for each regime p ∈ T is given below. The model
error closest to the deﬁnition of each regime is the smallest, and its index number
is used to decide the controller.
Normal 1, p = 1
This regime is deﬁned as when the risers are parallel, with the desired horizon
tal distance between the nodes at the each water depth. Using the monitoring
comparing values from Table 6.1 gives the following model error and γfunction:
e
1
= ∆e
rel
−0D, (6.20)
γ
1
= γ(∆e
rel
) = ∆e
rel

∞
. (6.21)
Hence, the maximum deviation from the desired position determines the error
and the monitoring signal.
Normal 2, p = 2
In addition to the deviation for the desired relative horizontal distance shown
above, the variation in tension is also included. The function T(t, t
0
) is the
tension for a given time window, ses Fig. 6.6. t is the current time, and t
0
is
the oldest time for which information is kept, for instance t
0
= t − 100s. If the
92
Supervisor
T
max
T
min
∆[T(t, t
0
)[
max
t
0 t
T
o
p
t
e
n
s
i
o
n
T
(
t
)
Time
Time window
Figure 6.6: Time window for the top tension.
tension diﬀerence between maximum and minimum tension in the time window
is smaller than the boundary, here 20kN, the error contribution from tension is
set to zero. Otherwise, the errors are weighted and summarized.
[∆T(t, t
0
)[
max
= [T(t, t
0
)[
max
−[T(t, t
0
)[
min
, (6.22)
e
2,T
= 0 for [∆T(t, t
0
)[
max
≤ 20kN, else (6.23)
e
2,T
= k
T
[∆T(t, t
0
)[
max
, (6.24)
e
2,∆x
= ∆e
rel
−1.5D
∞
, (6.25)
e
2
= [ e
2,T
e
2,∆x
]
T
, (6.26)
γ
2
= γ
2
(e
2
) =
_
(e
2,T
)
2
+ (e
2,∆x
)
2
, (6.27)
where k
T
= 1e
−5
, such that for [∆T(t, t
0
)[
max
> 20kN, the error contribution for
tension will be e
2,T
> 0.2, whereas the contribution from the relative horizontal
distance with deviation 0.5D gives e
2,∆x
= 0.15. For a smaller tension deviation,
the error from the relative horizontal distance is the only contribution.
Slowly varying, p = 3
The slowly varying regime is for LF motions and tide giving medium deviation
from the desired relative distance. Like for the regime normal 1, this regime does
93
6. The Controller Architecture
only consider the relative horizontal distance. Hence,
e
3
= ∆e
rel
−2.5D, (6.28)
γ
3
= γ
3
(e
3
) = e
3

∞
. (6.29)
Fast changing, p = 4
For the fast changing regime, the risers experience faster motions due to the TLP
motions or current. This can be monitored through the horizontal velocity vector
of R2, ˙ x
R2
, which is included in this error model in addition to the relative riser
distance. Since we want to capture the fast motions, we use a highpass ﬁlter,
corresponding to the WF motions with periods T = 5 − 20s. We are interested
in the largest velocity and how fast it changes. The largest horizontal velocity is
found by
˙ x
R2,max
= ˙ x
R2

∞
. (6.30)
If ˙ x
R2,max
is larger than the boundary value given in Table 6.1, we have e
4, ˙ x
= 0.
Otherwise the error is given by e
4, ˙ x
= 0.6m/s − ˙ x
R2,max
. Since the risers are
changing the direction according to the speciﬁed motion from the TLP, the re
sulting error will be chattering between zero and the calculated value. A lowpass
ﬁlter is therefore introduced to give a mean error contribution from the veloc
ity. The contribution from relative distance is weighted to be of less importance.
Hence,
e
4,∆x
=
1
2
∆e
rel
−2.5D
∞
, (6.31)
e
4
= [ e
4,∆x
e
4, ˙ x
], (6.32)
γ
4
= e
4

1
= [e
4,∆x
[ +[e
4, ˙ x
[. (6.33)
Error State, p = 5
Instead of ﬁnding the maxdeviation from a desired conﬁguration, we here inves
tigate if the closest node is in a (near) collision situation. This could be written
e
5
= ∆e
rel
−13D, (6.34)
γ
5
= γ
5
(e
5
) = [e
5
[
min
. (6.35)
This regime will also be activated if the risers are far apart, for instance if the
second riser has a large deﬂection.
6.4.3 Switching Logic
The switching logic decides which controller to use through the controller switch
ing signal σ, based on the value of the monitoring signal µ
p
in (6.15). The
94
Controller Set
scaleindependent hysteresis switching logic from Section 6.3.3 is used. The state
diagram (Fig. 6.3) shows how the process switching signal ρ is found based on
the smallest monitoring signal µ
p
. The mapping σ = χ(ρ) is found in Table 6.1.
λ
p
is the forgetting factor for µ
p
. A larger factor λ gives faster forgetting,
while a smaller λ gives longer memory. To decrease the detection time when
active control is needed, the memory is slightly decreased by choosing λ
p
slightly
bigger for p ∈ ¦3, 4, 5¦. Rewriting (6.15) on vector form we have
˙ µ = −λµ +γ(e
p
), (6.36)
where ˙ µ is the vector of monitoring errors, λ is the diagonal matrix of non
negative constants (here λ = diag
_
1 1 1.5 2 1.5
¸
T
), and γ(e
p
) is the
resulting vector of the class /functions. The initial monitoring value is set to
µ(0) =
_
0.9 1 1 1 1
¸
T
> 0.
6.5 Controller Set
The controller set’s main task is to prevent riser collision, and the controller best
suited for the current operational conditions is determined by the switching logic.
Other inputs in the controller block are the measured top tension and payout.
The controller block in Fig. 6.4 consists of a set of controllers designed for each
of the regimes described in Section 6.4.2 based on the ROCs from Section 6.2.
The active controllers comprise of the following components:
The guidance block calculates the guidance trajectory for each riser in the
array based on the control objectives.
The reference model generates the trajectory based on the guidance model.
To prevent collision in an array of risers during transit to reach the desired
tension, the sequence and how fast the risers are tensioned are of impor
tance. This could be formulated as a maneuvering problem (Skjetne, 2005),
and is a subject for further research.
One controller is designed specially for each operational regime. Which con
troller to use is decided by the supervisor which sends the controller switch
ing signal σ to the controller block such that the most appropriate controller
is used in the feedback loop. To provide a smooth transition between the
controllers, a weighting function is introduced.
6.5.1 Guidance
The guidance trajectory determined by the actual control objective, described in
Section 6.1. Depending on the control objective it takes relative distance, payout
95
6. The Controller Architecture
and/or tension as inputs. In the guidance formulations below, we have assumed
two risers, with R1 being the reference of R2. However, these principles are easily
expanded to an array of risers.
Equal Payout
The equal payout guidance block needs the payout measured for the reference
risers. For the payout to be equal for both risers, the guidance is written
ξ
r,2
(t) = ξ
1
(t), (6.37)
where ξ
1
is the measured payout of R1 and ξ
r,2
is the guidance for R2.
Equal Eﬀective Length
When the riser elongation due to changes in top tension is included, we need
to know the prevailing top tensions, the initial top tensions and the initial riser
length for both risers, in addition to the measured payout to the reference riser.
Using R1 as a reference for R2 we rewrite (6.6) to
ξ
r,2
(t, T) = ξ
1
(t) + l
R,1
(T
0,1
) + ∆l
R,1
(T
1
) −[l
R,2
(T
0,2
) + ∆l
R,2
(T
2
)] . (6.38)
If R2 is used as a reference, the indexes are switched. For an array of risers, one
could be chosen as the leader, and the payout for all the other risers could be
calculated relative to this one.
Relative Horizontal Distance
Here we measure the relative horizontal distance between the risers at one pre
deﬁned water depth z. The desired distance is ∆x
d
, and the guidance is written
∆x
R12,r
(z) = ∆x
d
. (6.39)
The error (6.51) later, indicates that when R1 approaches R2, the relative dis
tance is reduced. If only R2 is controlled, the top tension of R2 needs to be
decreased to maintain the same relative distance.
6.5.2 Reference Model
To provide a smooth trajectory and high performance, a reference model is de
signed for the diﬀerent guidance models. The reference is introduced to calculate
96
Controller Set
a feasible trajectory for the payout or the relative horizontal distance decided in
the guidance block. The following third order ﬁlter is demonstrated appropriate
¨
ξ
d
+ 2ζ
d
ω
d
˙
ξ
d
+ ω
2
d
ξ
d
= ω
2
d
ξ
ref
, (6.40)
˙
ξ
ref
= −
1
t
d
ξ
ref
+
1
t
d
ξ
r
, (6.41)
where ξ
d
and its derivatives are the desired payout position, velocity and acceler
ation trajectories. ξ
r
is the new reference coordinates in the same frame, and ξ
ref
is the low pass ﬁltered coordinate. ζ
d
is the relative damping ratio, ω
d
is natural
frequency, and t
d
is the cutoﬀ period period of the low pass ﬁlter in (6.41). This
provides a smooth transfer between diﬀerent setpoints.
For the relative horizontal distance the corresponding third order ﬁlter is
written
∆¨ e
rel,d
+ 2ζ
d
ω
d
∆˙ e
rel,d
+ ω
2
d
∆e
rel,d
= ω
2
d
∆e
rel,ref
, (6.42)
∆˙ e
rel,ref
= −
1
t
d
∆e
rel,ref
+
1
t
d
∆e
rel,r
, (6.43)
where ∆e
rel,d
and its derivatives are the desired relative horizontal distance, ve
locity and acceleration trajectories. ∆e
rel,r
is the new reference coordinates in
the same frame, and ∆e
rel,ref
is the low pass ﬁltered coordinate.
The main objective of this block is to generate a safe and legal reference
trajectory for the top tension to follow. For an array of risers, changing the
tension too fast for one riser and too slow for another might lead to a near
collision situation between risers not originally conﬂicting. To avoid this problem,
we have to ﬁgure out how fast each riser should follow its trajectory. This may
be expressed as a maneuvering problem, with a geometric and a dynamic task.
The geometric task is the path to follow, while the dynamic task is related to
how fast we follow this path. This is a subject to further study. The deﬁnition
of the maneuvering problem is found in Fossen (2002) and Skjetne (2005).
6.5.3 Controllers
Each riser has a pretension T
0,j
. The total top tension in each riser is equal to
the pretension, plus the contribution from the controller
T
j
= T
0,j
+ τ
c,j
. (6.44)
The controllers are designed for each of the regimes deﬁned in Section 6.2.
Controller 1, σ
1
is for normal conditions and low current velocities, where the
top tensions are kept constant, to reduce unnecessary wear and tear. The
tensions are equal to the previous controller in time, but possibly at diﬀerent
values for the two risers. The initial tension is equal to the pretension.
97
6. The Controller Architecture
Controller 2, σ
2
is for slow changes and stationary conditions, calculating the
new top tensions based on the payout, top tension and riser length. A
PIcontroller could be appropriate.
Controller 3, σ
3
is for faster changing conditions, and the derivative properties
should be included. Hence, a PIDcontroller could be used.
Controller 4, σ
4
is called error state maneuvering (ESM). It is used when we
experience fault conditions and is structurally diﬀerent from the other con
trollers. For two risers only, the ESM controller can be the same PI
controller as for controller 2. For an array of risers, the controller algorithm
and the calculation of the reference trajectory may be more complicated to
avoid collisions.
For the ﬁrst control objective principle the controllers can be written
τ
c1,j
= const, (6.45)
τ
c2,j
= −K
P2,j
e
j
−K
I,j
_
e
j
dt, (6.46)
τ
c3,j
= −K
P3,j
e
j
−K
I,j
_
e
j
dt −K
D3,j
˙ e
j
, (6.47)
τ
c4,j
= τ
c2,j
, (6.48)
e
j
= ξ
d,j
−ξ
j
, j = 1, 2, (6.49)
where e
j
is the error between the desired payout ξ
d,j
and the actual payout ξ
j
for riser j. K
I,j
=
K
P2,j
T
I,j
is the integrator gain, and K
D3,j
= K
P3,j
T
D3,j
is the
derivation gain. The integrator is the same in both controllers 2 and 3, and
integrator antiwindup is included to avoid saturation of the actuator.
For the second control objective principle based on the relative horizontal
distance we propose a PIcontroller
τ
c2,j
= −K
P,j
∆e
rel
−K
I,j
_
∆e
rel
dt, (6.50)
∆e
rel
(z) = ∆e
rel,d
−∆x
R12,m
(z), (6.51)
where K
P,j
and K
I,j
are tuned for this control objective.
98
Controller Set
0 2 4 6 8 10 12 14 16 18 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Time [s]
W
e
i
g
h
t
i
n
g
v
a
l
u
e
α
1
(t)
α
2
(t)
Figure 6.7: Weighting functions α
1
and α
2
.
6.5.4 Transition
To ensure smooth transition when switching between the controllers, a weighting
function is used such that
τ
c,j
= α
1
(t)τ
old
+ α
2
(t)τ
new
, (6.52)
α
1
= exp
_
−10 ((t/20))
2
_
, (6.53)
α
2
= 1 −exp
_
−10 ((t/20))
2
_
, (6.54)
where τ
c,j
is the resulting controller, τ
old
the previous and τ
new
the new controller
output in time, respectively. α
1
and α
2
are weighting functions dependent on the
time from the switching moment. Their sum is always 1. The functions are seen
in Fig. 6.7. The switching between the controllers is tuned to approximately 20s
here. In real life the switching may take longer time and is dependent on how
fast the environmental conditions change.
6.5.5 Integrator AntiWindup
A controller with integral action in combination with an actuator which may
become saturated might give undesirable eﬀects. If the error is so large over a
period of time that the integrator saturates the actuator, the feedback path will
99
6. The Controller Architecture
be broken. The integrator may continue to integrate up to a large value. When
the error ﬁnally is reduced, the integrator value may be so large that it takes
time to discharge it. This eﬀect is called integrator windup. To prevent it, we can
stop updating the integrator when the actuator saturates, or we can discharge it.
The second method is implemented here.
The error diﬀerence between the actuator output and the controller output
times a gain is fed back to the system. When the actuator is in saturation, this
error signal works to discharge the integrator such that the controller output is
at the saturation limit. When the actuator is not in saturation, the error is zero.
This could be written
˙ u
I
=
K
p
T
i
e(t) −
1
T
i
[v(t) −u(t)], (6.55)
v(t) = sat(u(t)), (6.56)
where u is the controller output and v is the actuator output (Egeland, 1993,
˚
Astr¨om and Wittenmark, 1997).
100
Chapter 7
Control Plant Model Analysis
In Chapter 3, a mathematical model for a riser is derived. This model was
implemented in Simulink and veriﬁed by comparing it to the commercial FEM
software RIFLEX (Fylling et al., 2005) in Chapter 4. The model had the same
physical behavior as RIFLEX in all deep water cases, and could be said to be
close to the real world and a good model of the riser process. The model used in
the veriﬁcations had 20 elements, and satisﬁed the desired level of accuracy. As
deﬁned in Chapter 3, the purpose of the PPM is to describe the actual physical
process as accurately as possible. For the purpose of control applications, the
CPM should be computationally fast, however, still describe the main physics.
This is motivated by the realtime requirements for control systems. A low order
model gives small system matrices and keeps the number of numerical operations
down. A simple processor could then be able to run the model online. Small,
simple computers help to keep the costs to a minimum, while large and fast
computers cost more.
In this chapter we investigate how many elements are needed to keep a desired
level of accuracy, both for the PPM and for the CPM. Diﬀerent means to measure
the performance of a model are considered. Generally, better performance is
expected for more elements. How many elements are good enough? And which
cases might need more elements to keep the desired level of accuracy?
7.1 Analysis Input Data and SetUp
The riser top is assumed to be at the level of free sea surface, i.e. 1200m
above the seabed. The RIFLEX model has 400 elements, each of length 3m.
The number of elements for the Simulink riser models are common multiples
of 400, ranging from 2 to 20, i.e. N = ¦2, 4, 5, 8, 10, 16, 20¦. This is done
to get the nodes at the same height above the seabed, and hence more easily
101
7. Control Plant Model Analysis
compare the horizontal displacements, especially of interest in the quasistatic
analyses. With a water depth of 1200m, the element lengths were respectively
l
0
= ¦600, 400, 240, 150, 120, 75, 60¦m for the diﬀerent models and increasing num
ber of elements. The Simulink models are hereafter called CPMs. These 7 ver
sions of the model are identical except for the number of elements, and are com
pared under the same environmental conditions. Two current proﬁles were run
in all tests. These are:
• Uniform current with velocity 0.7m/s.
• One year return period Ormen Lange current proﬁle, with surface velocity
1.15m/s and velocity close to the seabed 0.5m/s, illustrated in Fig. 3.2.
For some of the cases, additional proﬁles were run to investigate and illustrate
special phenomena. The test setup was otherwise similar to what was used in
Sections 4.24.5. The tension and the TLP motions are speciﬁed in each section.
The simulation results for the RIFLEX model are the same as in Sections 4.24.5
and are used in the analysis to illustrate the assumed correct solution.
7.2 QuasiStatic Analysis with Increasing TLP
Position
The quasistatic CPMs were run with increasing TLP position in steps of 5m,
from 0m to 70m. Three of these positions (0m, 30m and 50m) were subject for
more detailed investigations. The top tension was kept constant at 1800kN. To
investigate the error introduced with few elements, diﬀerent measurements and
methods were used. These could be summarized as:
• Deﬂections/Deformation shapes.
• Error norms for horizontal positions.
• Riser top angle.
• Area under the curve.
• Payout.
Most of these methods are mainly of theoretical interest. The measurement of
the (absolute) horizontal positions along the riser is expensive and not accurate
enough to be a part of the control loop. On the other hand, both top and bottom
riser inclinations and payout are already measured today.
102
QuasiStatic Analysis with Increasing TLP Position
7.2.1 Error Norms for Horizontal Positions
The idea behind this method is to compare the horizontal displacement for nodes
at the same vertical position, i.e. all nodes for each CPM is compared to the
RIFLEX model. To be able to compare the models better, we apply metrical
norms, see Appendix A.5. The three norms used here are:
1. l
1
norm: x
1
= ([x
1
[ +[x
2
[ + +[x
n
[).
2. l
2
norm: x
2
=
_
(x
2
1
+ x
2
2
+ + x
2
n
).
3. l
∞
norm: x
∞
= max
j
[x
j
[.
Here x is replaced by the horizontal error vector given by
∆x
N=i
(j) = x
R
(j) −x
N=i
(j), (7.1)
where i is the number of elements, j is the height of the actual node, x
R
is the
RIFLEX solution, and x
N
the horizontal positions for the CPM with N elements.
Two of the norms above could then be more commonly explained as:
• l
1
norm  sum of errors in all nodes.
• l
∞
norm  maximum displacement error.
Both the l
1
 and l
2
norms are dependent on the number of elements and are
therefore divided by the number of nodes, except for the top and bottom nodes,
which are always located at the same place. Hence, the normalized l
1
 and l
2

norms, accounted for the number of nodes can be written
l
1
:
x
1
N −1
, l
2
:
x
2
N −1
. (7.2)
This makes it easier to compare the various models. Fig. 7.1 shows the riser
conﬁguration without TLP oﬀsets for all CPMs. The models with 2, 4, and 5
elements are all seen to give rather rough estimates of the riser conﬁguration.
The 10 elements model give a good impression of the riser curve. For more
elements, the deviation is even smaller. The same color code as in Fig. 7.1
is used throughout this chapter, unless otherwise speciﬁed. The diﬀerent error
norms for this TLP position are given in Fig. 7.2. All the error norms converge
to zero when N is increasing, and the norms are close to zero when N ≥ 16.
However, at N = 10, the error norm is increasing, which is clearest seen for the
l
1
 and l
∞
norms, i.e. the sum of errors and the maximum error. The maximum
deﬂection error were found at 600m above the seabed, and is larger for N = 10
than for N = 8, which is opposite of what is expected. Zooming in on the
103
7. Control Plant Model Analysis
0 2 4 6 8 10 12
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
N=2
N=4
N=5
N=8
N=10
N=16
N=20
N=400
Figure 7.1: Riser conﬁguration without TLP oﬀset for an increasing number of
elements exposed to the Ormen Lange design current.
2 4 6 8 10 12 14 16 18 20
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Number of elements
Norms for horizontal positions; TLP 0m
l
1
l
2
l
∞
Figure 7.2: The diﬀerent error norms for the horizontal nodes without TLP oﬀset,
exposed to the Ormen Lange design current.
104
QuasiStatic Analysis with Increasing TLP Position
9.5 10 10.5 11
400
450
500
550
600
650
700
750
800
850
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
N=2
N=4
N=5
N=8
N=10
N=16
N=20
N=400
Figure 7.3: Zoomed riser conﬁguration without TLP oﬀset for an increasing
number of elements exposed to the Ormen Lange design current.
riser conﬁguration (Fig. 7.3) we see that the model with 8 elements match the
RIFLEX solution better in the node points, while N = 10 follow the curve better
in general. Also note that each norm method gives approximately the same norm
curve independent of the TLP oﬀset, see Fig 7.4.
To summarize, using this method with error norms on the horizontal dis
placement only to quantify the performance of each model is not reliable. It
is dependent on how good each node matches the RIFLEX solution for a given
current proﬁle and number of elements. For other current proﬁles, another num
ber of elements could indicate to be the best. The diﬀerent norms show various
sensitivity. The l
2
norm seems to be least sensitive to the single node error (Fig.
7.4).
105
7. Control Plant Model Analysis
2 4 6 8 10 12 14 16 18 20
0
0.2
0.4
0.6
0.8
l
1
n
o
r
m
Norms for horizontal positions
TLP 0m
TLP 30m
TLP 50m
2 4 6 8 10 12 14 16 18 20
0
0.2
0.4
0.6
0.8
l
2
n
o
r
m
TLP 0m
TLP 30m
TLP 50m
2 4 6 8 10 12 14 16 18 20
0
0.2
0.4
0.6
0.8
Number of elements
l
∞
n
o
r
m
TLP 0m
TLP 30m
TLP 50m
Figure 7.4: The diﬀerent norms for the TLP oﬀsets, exposed to the Ormen Lange
design current.
7.2.2 Riser Top Angle
Today the top and bottom riser angles are measurements available oﬀshore. These
are used to calculate the riser conﬁguration and may be used as inputs to the
DP systems of ﬂoaters (Sørensen et al., 2002, Høklie et al., 2002). Production
risers usually have bending stiﬀeners in their top and bottom end connections,
while our model assumes free end rotations, which is the case for drilling risers.
The riser inclination calculated here might therefore be larger than what could
be measured for production risers in the industry. However, we would like to
investigate the riser top angle as it tells something about the riser conﬁguration,
the deﬂection and the accuracy of the riser model.
In Fig. 7.5 the riser top angles are given for the three TLP oﬀset. For a
TLP without oﬀsets, all inclinations have correct sign. For a TLP oﬀset of 30m
(see Fig. 7.6), two elements are too few to describe the riser conﬁguration and
gives a positive sign, while the inclination should have been negative. At a TLP
oﬀset of 50m (Fig. 7.7) three models N = ¦2, 4, 5¦ are all seen to give the wrong
sign, while for N > 8 the top angle is approaching the correct value. Here we
have used the angle given by the straight line between the two upper nodes to
represent the top angle. An alternative method would be to assume constant
106
QuasiStatic Analysis with Increasing TLP Position
2 4 6 8 10 12 14 16 18 20
−2.5
−2
−1.5
−1
−0.5
0
0.5
1
1.5
Number of elements
T
o
p
a
n
g
l
e
[
d
e
g
]
TLP 0m
TLP 30m
TLP 50m
Figure 7.5: The riser top angles with diﬀerent TLP oﬀsets, exposed to the Ormen
Lange design current.
curvature in the upper element and apply the tangent at the upper node as the
top angle. This would give a better angle estimate for few element CPMs, but
converge to the same result for more elements. Anyhow, these analyses show
the importance of having enough elements in the model, especially when using
inclinations in the measurements in the control feedback loop. The phenomenon
described here is dependent on the TLP oﬀset, current proﬁle and velocity. It is
more likely to be observed in a large oﬀset and medium to large current velocity.
The max deﬂection will then be slightly larger than the TLP oﬀset and near the
top. However, this example is from the Ormen Lange current, with an oﬀset at
50m, which is a most realistic case.
7.2.3 Area Under Curve
A better way to compare the performance of the models is to consider the area
under the riser curves, recall Fig. 7.1. The area under the riser curve, A
N
, for
each CPM can be found by summarizing the area of the trapezoids between x = 0
107
7. Control Plant Model Analysis
0 5 10 15 20 25 30 35
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
N=2
N=4
N=5
N=8
N=10
N=16
N=20
N=400
Figure 7.6: Riser conﬁguration with TLP oﬀset of 30m for an increasing number
of elements exposed to the Ormen Lange design current.
49 49.5 50 50.5
900
950
1000
1050
1100
1150
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
N=2
N=4
N=5
N=8
N=10
N=16
N=20
N=400
Figure 7.7: Zoomed riser conﬁguration with 50 TLP oﬀset for an increasing
number of elements exposed to the Ormen Lange design current.
108
QuasiStatic Analysis with Increasing TLP Position
2 4 6 8 10 12 14 16 18 20
0.5
1
1.5
2
2.5
3
3.5
4
4.5
x 10
4
Number of elements
A
r
e
a
u
n
d
e
r
c
u
r
v
e
[
m
2
]
TLP 0m
TLP 30m
TLP 50m
Figure 7.8: The area under the riser curves for TLP oﬀsets.
and the riser curve for each pair of nodes
A
N
=
N
i=0
x
i
+ x
i+1
2
l
0
, (7.3)
where x
i
is the horizontal position of node i, and l
0
is the element length. Shorter
riser elements will capture the deﬂection better with more trapezoids. The riser
areas are seen to converge to the desired value for N ≥ 10 (Fig. 7.8). This
method is less dependent on how the selected nodes match the correct solution,
since all node values and not only the common multiplier are included in the
calculations. The relative error under the curve is illustrated in Fig. 7.9 and
given by
e
A
=
A
RIFLEX
−A
N=i
A
RIFLEX
100%, (7.4)
where A
R
is the RIFLEX area for 400 elements. For N ≥ 8, the relative error
is less than 3% for all TLP positions. It should be noted that N = 10 gives
better results than N = 8, which is expected. Hence, this method gives overall a
better evaluation of the convergence of the various riser models than the method
109
7. Control Plant Model Analysis
2 4 6 8 10 12 14 16 18 20
0
5
10
15
20
25
30
35
Number of elements
E
r
r
o
r
a
r
e
a
u
n
d
e
r
c
u
r
v
e
[
%
]
TLP 0m
TLP 30m
TLP 50m
Figure 7.9: The relative error in the area under the riser curves for TLP oﬀsets.
with horizontal error norms. For CPMs with few elements, applying Simpson’s
method to ﬁnd the area would give better estimates. For more elements, the
estimate of the area will converge to the same value for both methods.
7.2.4 Payout
The payout is a direct result of the deﬂection and how well the CPMs correspond
to the correct curve. It is similar to the method with area under the curve pre
sented in Section 7.2.3. However, while the area under the curve is a theoretical
calculation, the payout can actually be measured and used in the feedback loop.
The static payout for the diﬀerent TLP oﬀsets and various models are shown in
Fig. 7.10. The payouts converge to the desired values for increasing number of
elements, and at N ≥ 8 the payout error is less than 2%, see Fig. 7.11. Hence,
the payout is a good and robust method.
110
QuasiStatic Analysis with Increasing TLP Position
2 4 6 8 10 12 14 16 18 20
0.65
0.7
0.75
0.8
0.85
0.9
0.95
1
Number of elements
P
a
y
o
u
t
[
m
]
TLP 0m
TLP 30m
TLP 50m
Figure 7.10: The payout error for TLP oﬀsets.
7.2.5 Discussion
Four methods have been investigated; horizontal error norms, top angle, area
and payout. All methods showed, as expected, that increasing the number of
elements gave a better model closer to the correct solution. The horizontal error
norms were sensitive to the prevailing current proﬁle such that a lower number
of elements appeared to have better accuracy than the true situation. This was
especially seen for the l
∞
norm and the l
1
norm. The l
2
norm was the least
sensitive method.
Too few elements could give the opposite sign on the riser top angle for some
environmental conditions. When using inclinations in the feedback loop for riser
control or DP operations, one should be careful when deciding the number of
elements in the model. Calculation of the area showed good robustness. Anyhow,
this method could not be used online as it requires accurate measurements of the
horizontal positions. Finally, payouts proved to be a robust and accurate method,
which is easily measured and can be used in feedback. For the rest of the analysis,
only payout will be used. The results are summarized in Table 7.1.
111
7. Control Plant Model Analysis
2 4 6 8 10 12 14 16 18 20
0
2
4
6
8
10
12
14
16
18
20
R
e
l
a
t
i
v
e
P
a
y
o
u
t
E
r
r
o
r
[
%
]
Number of elements
TLP 0m
TLP 30m
TLP 50m
Figure 7.11: The relative error in payout for the diﬀerent TLP oﬀsets.
Method Robustness Possible to measure
Horizontal
error norms
Sensible for single node errors
and current proﬁles
Theoretical
Top angle Wrong sign for low order models Measurements
Area Robust Theoretical
Payout Robust Measurements
Table 7.1: The pros and cons for the diﬀerent methods to quantify the perfor
mance of the CPMs.
7.3 Analysis of QuasiStatic Tension Variations
The quasistatic CPMs were analyzed with increasing top tension from 1200kN to
2700kN in steps of 50kN. The TLP was positioned in zero oﬀset. The riser models
were exposed to the bidirectional current, in addition to the uniform and the Or
men Lange currents. The physics of the riser behavior and the relations between
the top tension, top position and the maximum deﬂection were explained in Sec
112
Analysis of QuasiStatic Tension Variations
1199 1199.5 1200 1200.5 1201
6
8
10
12
14
16
18
20
22
M
a
x
h
o
r
i
z
o
n
t
a
l
d
i
s
p
l
a
c
e
m
e
n
t
[
m
]
Vertical top position [m]
Displacement vs top position
N = 2
N = 4
N = 5
N = 8
N = 10
N = 16
N = 20
N = 400
1199 1199.5 1200 1200.5 1201
−14
−13
−12
−11
−10
−9
−8
−7
−6
−5
Vertical top position [m]
∆
x
/
∆
z
[
−
]
The derivative of x(z)
Figure 7.12: Maximum deﬂection as a function of top position (a), and its deriva
tive (b) for the CPMs exposed to the Ormen Lange current.
tion 4.3. In this section we mainly focus on the error introduced by decreasing
the number of elements. Fig. 7.12 shows the relation between the maximum hor
izontal deﬂection as a function of the vertical top position x(z) and its derivative
∆x/∆z. Note that the physical interpolation of
∆x
∆z
is “How much will the max
imum horizontal deﬂection change for a unit variation of vertical position of the
upper riser end.” For a low top position and small top tension, Fig. 7.12a) shows
that 2 elements are too few, such that the maximum horizontal displacement is
too small. N = ¦4, 5¦ match better. For all these low element order CPMs, the
maximum deﬂections at the lowest tensions are 12m too small such that the riser
end is assumed to be 0.251m higher than the correct value. For eight or more
elements, the x(z)curves match fairly well with the RIFLEX solution. For larger
tensions and higher top positions all CPMs match the solution better than for
lower tensions. The same is seen for the derivatives in Fig. 7.12b). For the three
lowest number of elements, the derivatives deviate noticeably from the RIFLEX
solution. For all higher number of elements, the curves match the RIFLEX so
lution satisfactorily. The values of the derivatives for N = ¦5, 10, 16, 20¦ have a
sudden discontinuity. This can be explained by looking at the snapshots in Fig.
7.13. For each step in tension the node with max deﬂection is found. This node
number will vary, depending on the tension. The change in vertical position ∆z
is dependent on the length of the element. This will give a drop in ∆x/∆z, and
is due to the numerical way of ﬁnding the derivatives. The RIFLEX model does
not have any noticeable discontinuities. The element length is only 3m, such
that the change ∆z is much smaller. Longer discontinuities were seen for longer
113
7. Control Plant Model Analysis
0 2 4 6 8 10 12 14 16 18 20
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
N = 10
N = 16
N = 20
Figure 7.13: Riser conﬁguration for 10 (opink), 16 (*green) and 20 (∆–blue)
elements and their maximum deﬂection.
element lengths. Note that these discontinuities are due to numerical errors in
calculation of the derivatives, while the simulation model experience continuous
derivatives.
Maximum deﬂection versus top tension x(T) and its derivatives are seen in
Fig. 7.14. The variation in tension from one step to another is constant. When
the node for which the maximum deﬂection is changed, ∆x is slightly larger than
when it is kept at the same height. Due to this, some smaller discontinuities
were noticed in Fig. 7.14b). At high tensions, the derivatives converge to the
same value. The x(T) curves all have the same shape. The three CPMs with
least elements do not have the same maximum deﬂection as for the higher order
models. The absolute error in top position as a function of tension relative to the
RIFLEX solution, is found in Fig. 7.15 and given by
[e
z(T)
[ = [z
R
(T) −z
N=i
(T)[, (7.5)
where z
R
is the RIFLEX top position and z
N
is the top position for each CPM.
Three tensions are chosen; a medium initial tension (1800kN), and the lower and
upper tension limits, 1200kN and 2700kN, respectively. The error is seen to de
114
Analysis of QuasiStatic Tension Variations
1500 2000 2500
6
8
10
12
14
16
18
20
22
Top tension [kN]
M
a
x
h
o
r
i
z
o
n
t
a
l
d
i
s
p
l
a
c
e
m
e
n
t
[
m
]
Displacement vs top tension
N = 2
N = 4
N = 5
N = 8
N = 10
N = 16
N = 20
N = 400
1500 2000 2500
−0.03
−0.025
−0.02
−0.015
−0.01
−0.005
0
Top tension [kN]
∆
x
/
∆
T
[
m
/
k
N
]
The derivative of x(T)
Figure 7.14: Maximum deﬂection as a function of top tension (a), and its deriva
tive (b) for the CPMs exposed to the Ormen Lange current.
crease with increasing number of elements and also increasing tension. Neglected
eﬀects in the CPM seem to be more signiﬁcant for low tensions. For instance
bending stiﬀness (EI) is more important for low tensions than for higher tensions,
where the geometric stiﬀness is dominating. For higher tensions the dominance
of elastic ﬂexibility over geometric ﬂexibility is noticed. In Fig. 7.16a) the rela
tionship between the vertical top position and top tension z(T) is plotted. For
low tensions, the graphs are spread. For high tensions, all curves approach the
same asymptote. The same is seen in Fig. 7.16b). The eﬀect on the top position
by increasing the top tension is largest for low tensions. For higher tensions, the
increase in the top positions is mainly due to the increase in length.
This is particularly obvious when considering the two element model with
bidirectional current, see Fig. 7.17. The two element model does not capture
the current forces and is vertical. The only increase in top tension is therefore
due to elasticity. In Fig. 7.18, z(T)
N=2
is an asymptote which the other CPMs
are approaching, when the tension increases and the contribution to z from the
geometry due to deﬂection decreases. The derivatives are seen in Fig. 7.18b).
As expected, ∆z/∆T
N=2
is constant and the others are approaching this level.
Approximately the same value for ∆z/∆T is seen for all current proﬁles as it is
only dependent on the elasticity of the riser material. This can be derived from
115
7. Control Plant Model Analysis
2 4 6 8 10 12 14 16 18 20
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0.4
0.45
[
m
]
Absolute Error  e
z(T)

Number of elements
1200kN
1800kN
2700kN
Figure 7.15: Error in top position for the various tensions.
1500 2000 2500
1199.2
1199.4
1199.6
1199.8
1200
1200.2
1200.4
1200.6
1200.8
1201
Top tension [kN]
V
e
r
t
i
c
a
l
t
o
p
p
o
s
i
t
i
o
n
[
m
]
Vertical top position vs top tension
N = 2
N = 4
N = 5
N = 8
N = 10
N = 16
N = 20
N = 400
1500 2000 2500
0
0.5
1
1.5
2
2.5
3
3.5
4
x 10
−3
Top tension [kN]
∆
z
/
∆
T
[
m
/
k
N
]
The derivative of z(T)
Figure 7.16: Maximum top position as a function of top tension (a), and its
derivative (b) for the CPMs exposed to the Ormen Lange current.
116
Analysis of QuasiStatic Tension Variations
−6 −5 −4 −3 −2 −1 0 1 2
0
200
400
600
800
1000
1200
Horizontal position [m]
V
e
r
t
i
c
a
l
p
o
s
i
t
i
o
n
[
m
]
Top tension 1200kN
N=2
N=4
N=5
N=8
N=10
N=16
N=20
N=400
Figure 7.17: Riser conﬁguration for the CPMs exposed to the bidirectional cur
rent velocity proﬁle.
(3.22).
T
0
+ ∆T =
EA
l
0
(z
0
−l
0
+ ∆z), (7.6)
⇒∆T =
EA
l
0
∆z, (7.7)
∆z
∆T
=
z
0
EA
(7.8)
=
1200m
2.06e
11
N/m
2
0.0134m
2
= 4.35e
−4
m/kN. (7.9)
This value is the same as in Fig. 7.18b).
Remark: The riser conﬁguration as shown in Fig. 7.17 is also of interest from
a mode shape point of view. The bidirectional current velocity proﬁle excites the
second mode shape. FEM can be considered as a spatial discretization of the
riser conﬁguration, similar to what can be found for time varying signals. For
time varying systems, the slowest sampling frequency is decided by Shannon’s
sampling theorem and the Nyquist frequency: If the sampling frequency is higher
117
7. Control Plant Model Analysis
1500 2000 2500
1200
1200.2
1200.4
1200.6
1200.8
1201
Top tension [kN]
V
e
r
t
i
c
a
l
t
o
p
p
o
s
i
t
i
o
n
[
m
]
Vertical top position vs top tension
N = 2
N = 4
N = 5
N = 8
N = 10
N = 16
N = 20
N = 400
1500 2000 2500
4
6
8
10
12
14
x 10
−4
Top tension [kN]
∆
z
/
∆
T
[
m
/
k
N
]
The derivative of z(T)
Figure 7.18: Maximum top position as a function of top tension (a), and its
derivative (b) for the CPMs exposed to the bidirectional current proﬁle.
than twice the frequency of the highest frequency component of the continuous
signal, it can be reconstructed completely from the sampled signal (Ogata, 1995).
This is the time reconstruction of a signal. A parallel could here be drawn to our
spatial model reconstruction. How many elements are needed to represent the
main riser conﬁguration? For N = 2, none of the deﬂections are captured. For
the other low order CPMs, the main conﬁguration is seen. For the highest order
CPMs, the riser is drawn with fair accuracy.
7.4 Dynamically Moving TLP
The dynamic CPMs were then exposed to harmonic TLP motions using the same
setup as in Section 4.4. Figs. 7.19 a) and b) show the payout versus time and
TLP position, respectively, for TLP periods 60s and the Ormen Lange current
proﬁle. All CPMs are close to the RIFLEX solution at large oﬀsets where the
risers are straight. When the TLP is moving from right to left (see Fig.4.6a))
the deﬂection is increasing, and also the payout. The two element model does
not capture the large deﬂection, and their change in payout is only half as much
as for the RIFLEX solution. The other few element CPMs (N = ¦4, 5¦) capture
the main deﬂection. The medium order CPMs (N = ¦8, 10¦) follow the solution
nicely, while the highest order CPMs (N = ¦16, 20¦) are hardly seen as they are
on the curve for the RIFLEX solution.
The payout in Fig. 7.19 a) shows that the ﬁrst mode corresponding to the
TLP motion is dominating. The TLP motion and setdown is seen in Fig. 7.20
118
Analysis of Dynamic Variation in Top Tension
0 100 200 300
0.5
1
1.5
2
2.5
3
Time [s]
P
a
y
o
u
t
[
m
]
Piston payout
10 20 30 40 50
0.5
1
1.5
2
2.5
3
Payout vs TLP
TLP position [m]
N = 2
N = 4
N = 5
N = 8
N = 10
N = 16
N = 20
N = 400
Figure 7.19: Payout as a function of time (a) and TLP oﬀset(b) exposed to the
Ormen Lange current.
as the bottom of the cylinder in the tension system (blue, thick line). The riser
tops are plotted in the same graph, such that the distance between them are the
payouts. The second order riser mode shape is clearly seen in the riser tops and
corresponds to the second order mode shape in Fig. 4.6b) where the TLP moves
from left to right. The payout error
[e
ξ
[ = [ξ
R
−ξ
N=i
[, (7.10)
is plotted in Fig. 7.21. For the low order CPMs, the ﬁrst mode dominates. The
error for N = 2 is up to 45% for the smallest TLP oﬀset with the largest deﬂection.
Such a large error on the ﬁrst mode shape dominates the second mode shape seen
in Fig. 7.20. The error from ﬁrst mode shape is also dominating for N = ¦4, 5¦,
with errors of approximately 20% and 15%, respectively. For the medium and
high order CPMs, the total errors are less than 8% and 5%, respectively. The
ﬁrst mode shape errors are small such the second mode shape errors are observed.
Note that these results presented here are those with largest relative error. At
larger TLP periods, like 120s, the errors were smaller, especially for the low order
models. At 300s, the risers were close to quasistatic and only the ﬁrst mode shape
was seen. Smaller current velocities gave also smaller errors.
7.5 Analysis of Dynamic Variation in Top Tension
Finally, the dynamical CPMs were analyzed with harmonically varying tension.
The setup was the same as in Section 4.5. The simulations shown here are
119
7. Control Plant Model Analysis
0 50 100 150 200 250 300
1198
1198.5
1199
1199.5
1200
1200.5
1201
1201.5
Time [s]
V
e
r
t
i
c
a
l
t
o
p
p
o
s
i
t
i
o
n
[
m
]
Cylinder bottom and riser tops
Cylinder
N=2
N=4
N=5
N=8
N=10
N=16
N=20
N=400
Figure 7.20: Cylinder bottom (blue line) and riser top positions for TLP motions
and Ormen Lange current.
with a tension period of 120s and a riser exposed to the Ormen Lange current.
Fig. 7.22 shows the top position as a function of time (a) and top tension (b).
The top position versus top tension is seen to be a hysteresis function due to
relative velocity and drag. Hence, a given tension does not correspond to one top
position, but depends on whether the tension is increasing or decreasing as well.
The deviations in top position between the models are largest for low tensions
and low top positions. For higher tensions, the deviations are smaller. This is
expected based on the results from Sections 7.3 and 7.4. Fig. 7.22 b) illustrates
how an increasing number of elements result in a better match between the CPM
model and RIFLEX. Fig. 7.23 shows the relative payout error. Maximum errors
are seen for the low order CPMs, with approximately 0.22m for N = 2, and
0.09m for N = 4. For the medium and high order CPMs, the maximum error is
± 0.02m. A small second order nonlinearity and phase shifts are seen. This is
due to the nonlinear viscous forces, i.e. the drag forces and diﬀerence in relative
velocity when the tension is decreased and increased.
120
Analysis of Dynamic Variation in Top Tension
0 50 100 150 200 250 300
−0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
Time [s]
[
m
]
Payout Error
N = 2
N = 4
N = 5
N = 8
N = 10
N = 16
N = 20
Figure 7.21: Relative payout error for harmonic TLP motions and Ormen Lange
current.
0 100 200 300
1199.6
1199.8
1200
1200.2
1200.4
1200.6
1200.8
1201
Time [s]
T
o
p
p
o
s
i
t
i
o
n
[
m
]
Top position vs time
1000 1500 2000 2500 3000
1199.6
1199.8
1200
1200.2
1200.4
1200.6
1200.8
1201
Top tension [kN]
Top position vs top tension
N = 2
N = 4
N = 5
N = 8
N = 10
N = 16
N = 20
N = 400
Figure 7.22: Riser tops as a function of time (a) and tension (b) exposed to the
Ormen Lange current.
121
7. Control Plant Model Analysis
0 50 100 150 200 250 300 350 400
−0.05
0
0.05
0.1
0.15
0.2
0.25
0.3
Time [s]
[
m
]
Top Position Error
N = 2
N = 4
N = 5
N = 8
N = 10
N = 16
N = 20
Figure 7.23: Relative payout error for a riser exposed to harmonic tension varia
tions with periods 120s and the Ormen Lange current.
7.6 Discussion
In Chapter 4, we veriﬁed the mathematical model. In this chapter, we inves
tigated what happens if the number of elements is decreased. From a control
point of view, the possibility of realtime operations are of great importance. We
therefore seek the smallest and least complex model which still aﬀord the desired
level of accuracy for the current operation. The analysis in this chapter could be
divided into three levels of accuracy. These correspond to the fast control plant
model (FCPM), the accurate control plant model (ACPM) and the process plant
model (PPM).
Recall that the size of the system matrices are 2(n+1)2(n+1) giving 4(n+1)
2
matrix elements, where n is the number of riser segments. Hence, doubling the
number of riser nodes increases the number of matrix elements by a multiple of
four. This slows down the computation speed considerably for the time consuming
matrix operations and iterations. The low order CPMs, with N = ¦4, 5¦, follow
the main conﬁguration. A small number of elements gives small system matrices.
The matrix operations are fewer, which gives shorter simulation times. The main
dynamic properties are still kept in this model. The medium order CPMs, with
N = ¦8, 10¦, have small quasistatic deviations, and follow the dynamics with
122
Discussion
Type N QS error Dyn error Description
FCPM 46 510% 1520% Main physics used in control
analysis
ACPM 812 25% < 8% Good model for monitoring of
the risers
PPM ≥ 15 < 1% < 5% High accuracy for the riser
process model
Table 7.2: The categorized results of the CPM analysis.
errors less than 8%. The high order CPMs with N = ¦16, 20¦, have very small
static deviations and follow the dynamics of the speciﬁed motion of the TLP and
tension with errors less than 5%. These results are categorized in Table 7.2. The
proposed limits and relative errors are based on the worst simulated case. For
slower TLP dynamics and smaller current velocities, the results are better with
smaller relative errors. Hence, the limit for the diﬀerent model classiﬁcations
could be set diﬀerently, such that the needed number of elements N are chosen
to correspond to the physics of the environment.
Note that the number of elements for the FCPMs and ACPMs does not need
to be a common multiple of the PPM, but can be chosen freely within the given
limits. Increasing the number of elements gives a more accurate model at the
expense of slower simulations due to larger system matrices.
123
Chapter 8
Simulation Results
The TLP/riser system was modeled in Chapter 3, and the controller system
design and architecture were presented in Chapters 5 and 6, respectively. In
this chapter the control objectives proposed in Section 6.1 are investigated and
compared in deep waters. Two of them, the equal payout and equal eﬀective
length, are also tested in shallow waters. The best working control objective is
thereafter simulated with dynamic TLP motions, and used in a case study with
changing environmental conditions and supervisory switched controller.
8.1 Set Up
The riser model from Section 3.5 was veriﬁed with a large variety of current
proﬁles in Chapter 4. However, in the simulations included in this chapter, a
current proﬁle with one year return period from the Ormen Lange ﬁeld is used
as a basis (Norsk Hydro, 2001). In addition a variance in the current is made
by ﬁltering white noise through a low pass ﬁlter with a period of 100s, and an
amplitude within the band of 5% of the current velocity in each node. The initial
centertocenter distance between the risers is 15D in 1200m waters, and each
riser model consists of 10 elements, unless otherwise speciﬁed. Some limitations
and assumptions are made in relation to the shielding eﬀect and the resulting
current on R2 to keep the riser and current models valid.
• The current is coming from one direction only, such that the risers are
always in a tandem position, with R2 in the wake of R1.
• Small TLP velocities such that the current is larger than the riser velocities
and the relative velocity always positive, keeping R2 in the wake of R1.
The latter will not be true for the last part of the case simulations with WF TLP
motions. However, this part of the simulation is included to demonstrate the
125
8. Simulation Results
eﬀect of control. The hydrodynamic interactions for the threedimensional case
and faster relative motions should be further investigated for industrial imple
mentations, but is outside the scope of this thesis. The riser data, current proﬁle
velocities and control gains are found in Appendix B.
8.2 Control Objectives
The four diﬀerent control objectives from Section 6.1 are tested and compared.
The equal tension control objective is run with two diﬀerent top tension levels and
0m or 30m oﬀset. The equal payout, equal eﬀective length and desired horizontal
distance control objectives are run with control of R2, using measurements from
R1 as the reference. For the equal eﬀective length control objective, control of
R1 with the payout of R2 as a reference is also simulated.
8.2.1 Constant Equal Tension
In the ﬁrst case, the top tension of the risers are constant and equal to a preten
sion. From (6.1) we have
T
1
= T
2
= T
0,j
. (8.1)
A medium top tension of T
0,j
= 1800kN is applied, and there is no TLP oﬀset.
The incoming current proﬁle is increased as a second order lowpass ﬁltered step
from zero to the design current proﬁle. As the current increases, the risers are
seen to slide out to the right in Fig. 8.1 (top left). The maximum horizontal
deﬂection is seen at 600m above the seabed. Collision occurs along most of the
riser (nodes 3 to 9), seen as relative horizontal distance smaller than 2D (bottom
left). The payout of R1 is larger than for R2 due to the larger deﬂection (Fig.
8.1, bottom right).
The TLP is then put in an oﬀset position of 30m, with increasing current,
seen top left in Fig. 8.2. The TLP is not very likely to have an oﬀset in the
opposite direction of the surface current, so oﬀsets are always simulated in the
positive direction. Collision is seen to occur at the same nodes independent of
the oﬀset position. The relative horizontal distance between three selected nodes
are shown in bottom left of Fig. 8.2, where it is seen that collision occurs at the
ﬁrst node after approximately 400s. The corresponding payout due to deﬂection
is seen bottom right in Fig. 8.2. R1 has the largest deﬂection and hence setdown,
which is clearly seen in the payout plot.
Increasing the top tension to the upper limit of T
0,j
= 2700kN and keeping
the TLP in a nooﬀset position, no collision occurs at any nodes, seen in the
snapshots (top left) and relative horizontal distance (bottom left) in Fig. 8.3.
Keeping the tension at the upper limit, collision could be avoided for these design
126
Control Objectives
0 2 4 6 8 10
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.45
0.5
0.55
0.6
0.65
0.7
0.75
0.8
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
1200
1400
1600
1800
2000
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−10
−5
0
5
10
15
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.1: Incrementing current velocity from zero to the Ormen Lange design
current. Snapshots of the riser conﬁguration (top left), top tension (top right),
relative horizontal distance (bottom left) and payout (bottom right).
currents. However, operation of the risers at this tension level is not desired due
to increased stress in the riser and excessive wear of the tension system. The
payouts (bottom right) are smaller than for the smaller top tensions.
For the rest of the simulations, the pretension is kept to T
0,j
= 1800kN for
both risers. A fully developed current proﬁle for the Ormen Lange ﬁeld is used.
8.2.2 Equal Payout
For the equal payout controller objective, the payout of all risers should be equal,
see (6.2). Using the measured payout of R1 as a reference for R2, we have from
(6.37)
ξ
r,2
(t) = ξ
1
(t). (8.2)
127
8. Simulation Results
0 10 20 30
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.4
0.5
0.6
0.7
0.8
0.9
1
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
1200
1400
1600
1800
2000
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−10
−5
0
5
10
15
20
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.2: Incrementing current and 30m TLP oﬀset. Snapshots of the riser con
ﬁguration (top left), top tension (top right), relative horizontal distance (bottom
left) and payout (bottom right).
Furthermore, a PIcontroller for the top tension could be formulated from (6.44),
(6.46) and (6.49) with parameters from Appendix B.3.
T
1
= T
0,1
, (8.3)
T
2
= T
0,2
+ τ
c,2
, (8.4)
τ
c,2
= −K
P,2
e
2
−K
I,2
_
e
2
dt, (8.5)
e
2
= ξ
d,2
−ξ
2
, (8.6)
where ξ
d,2
is the guidance trajectory passed through the third order ﬁlter in
(6.40)(6.41). The results are found in Fig. 8.4, with the top tension shown in
the top right ﬁgure.
This control algorithm gives equal payout for the risers, seen bottom right in
Fig. 8.4. Top left shows how R2 slides out to the right due to decreased tension.
However, collision still occurs, but in a smaller riser segment than with equal
128
Control Objectives
0 2 4 6 8
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.1
0.12
0.14
0.16
0.18
0.2
0.22
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
2000
2200
2400
2600
2800
3000
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
2
4
6
8
10
12
14
16
Time [s]
[
D
]
Relative horizotnal distance
Node 3
Node 6
Node 9
Figure 8.3: Incrementing current with top tension 2700kN. Snapshots of the
riser conﬁguration (top left), top tension (top right), relative horizontal distance
(bottom left) and payout (bottom right).
tension (nodes 5 and 6), also seen for the relative distance (bottom left). The
collision is caused by the lower tension in R2 compared to R1, which in turn
gives less axial elongation, and a shorter length of R2 than R1. Hence, due to the
elasticity of steel, collision may still occur for long risers when using the equal
payout control objective.
8.2.3 Equal Eﬀective Length
In the next control objective, the elasticity of the riser material is included. This
could be done by considering the riser length plus the payout. For equal risers
with the same pretension, the guidance trajectory from (6.38), using (6.3) and
(6.5) is simpliﬁed to
ξ
r,2
= ξ
1
+ ∆l
R
. (8.7)
129
8. Simulation Results
0 5 10
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.5
0.6
0.7
0.8
0.9
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
1200
1400
1600
1800
2000
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−10
−5
0
5
10
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.4: Equal payout algorithm and control of R2. Snapshots of the riser con
ﬁguration (top left), top tension (top right), relative horizontal distance (bottom
left) and payout (bottom right).
The controller is the same as in Section 8.2.2, (8.3) through (8.6). The results
are seen in Fig. 8.5. The snapshots found top left show how the midposition
of R2 starts in front of R1 when the pretensions are equal. R2 then slides out
to the right to avoid collision as the tension decreases, also seen in the relative
horizontal distance to the lower left. The payout of R2 approaches the reference
trajectory and both risers achieve similar deﬂection. The smaller variation in the
horizontal position is due to variance in the current. To the bottom right in Fig.
8.5 we clearly see that R2 has a larger payout to compensate for smaller riser
elongation and achieve equal eﬀective length.
The same algorithm is applied for 30m TLP oﬀset (Fig. 8.6). As before R2
slides out to the right with decreasing tension, increasing the horizontal distance
between the risers to avoid collision, see Fig. 8.6 top left. The payouts (bottom
right) are larger in the oﬀset case than without oﬀset due to the eﬀects from
weight and current. The top tension T
2
is seen to stabilize about 1400kN for
130
Control Objectives
0 5 10 15
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
1200
1400
1600
1800
2000
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−10
−5
0
5
10
15
20
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.5: Equal eﬀective length and control of R2. Snapshots of the riser con
ﬁguration (top left), top tension (top right), relative horizontal distance (bottom
left) and payout (bottom right).
the oﬀset case and 1300kN for the nooﬀset case, to the upper left in Figs. 8.5
and 8.6, respectively. Hence, a smaller tension diﬀerence is needed in the oﬀset
position. This is due to a longer eﬀective length (riser plus payout), and the
relation between tension, payout and horizontal deﬂection. For a long riser with
small tension, an increase in tension has larger eﬀect on the lateral deﬂection
than if the riser already has a high tension level.
Now the top tension of R1 is controlled using the payout of R2 as reference,
compensated for axial elasticity. R2 is not controlled. Hence,
ξ
r,1
= ξ
2
−∆l
R
, (8.8)
T
1
= T
0,1
+ τ
c,1
, (8.9)
T
2
= T
0,2
, (8.10)
e
1
= ξ
d,1
−ξ
1
. (8.11)
The controller is the same as for Section 8.2.2. Fig. 8.7 shows the snapshots of
131
8. Simulation Results
0 10 20 30
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.4
0.6
0.8
1
1.2
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
1200
1400
1600
1800
2000
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−10
−5
0
5
10
15
20
25
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.6: 30m TLP oﬀset, equal eﬀective length and control of R2. Snapshots
of the riser conﬁguration (top left), top tension (top right), relative horizontal
distance (bottom left) and payout (bottom right).
the risers. As the tension in R1 increases and it is tightened up, the deﬂection
decreases. In addition, R2 moves behind since R1 comes in front. This is due to
the increased current and reduced shielding eﬀect on R2 as the distance between
the risers increases. It should be noted that the simulation is valid only when the
downstream riser is more than 2D behind the upstream riser. The payouts, how
the piston end of R1 is pulled in, and how the payout of R2 is slightly increased
when the deﬂection increases, are seen (bottom right).
The TLP is placed at 30m oﬀset. The risers behave similarly as for the case
without oﬀset, see snapshots and payouts in Fig. 8.8. Top tension was seen to
decrease compared to the nooﬀset case and stabilize about 2650kN and 2400kN
with tension diﬀerences of 850kN and 600kN, respectively shown in Figs. 8.7
and 8.8. It should also be noted that the diﬀerence and payout is larger when
controlling R1 than R2. This is caused by the more straightlined conﬁguration at
higher tension levels, and a need of even more tension to straighten up to prevent
132
Control Objectives
0 5 10
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
1600
1800
2000
2200
2400
2600
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−10
−5
0
5
10
15
20
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.7: Equal eﬀective length and control of R1. Snapshots of the riser con
ﬁguration (top left), top tension (top right), relative horizontal distance (bottom
left) and payout (bottom right).
collisions, due to the nonlinear relationship (cf.
∆x
∆T
p. 53) between the tension
and the deﬂection.
8.2.4 Desired Horizontal Distance
The desired horizontal distance control objective requires one or more measure
ments of the relative distance between the riser at predeﬁned water depths. Keep
ing a desired relative distance, gives the following guidance trajectory from (6.39)
∆x
R12,r
(z) = ∆x
d
. (8.12)
133
8. Simulation Results
0 10 20 30
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.4
0.5
0.6
0.7
0.8
0.9
1
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
1600
1800
2000
2200
2400
2600
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−10
−5
0
5
10
15
20
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.8: TLP oﬀset of 30m, equal eﬀective length and control of R1. Snapshots
of the riser conﬁguration (top left), top tension (top right), relative horizontal
distance (bottom left) and payout (bottom right).
Controlling R2 with a PIcontroller gives
T
1
= T
0,1
, (8.13)
T
2
= T
0,2
+ τ
c,2
, (8.14)
τ
c,2
= −K
P,2
∆e
rel
−K
I,2
_
∆e
rel
dt, (8.15)
∆e
rel
(z) = ∆e
rel,d
−∆x
R12,m
(z), (8.16)
using (6.44), (6.50) and (6.51). The tension is found in Fig. 8.9 (top right). The
snapshots (top left) show how R2 slides to the right when the tension decreases
and removes the risers from the error state. The corresponding payout is seen
bottom right.
This control objective is seen to eﬀectively remove the risers from a collision
situation. However, it is dependent on measurement(s) along the riser, located
where the risers are most likely to collide.
134
Effect of Shallow Water
0 5 10 15
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800
0.5
0.6
0.7
0.8
0.9
1
1.1
1.2
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
1200
1400
1600
1800
2000
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−10
−5
0
5
10
15
20
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.9: Desired relative horizontal distance and control of R1. Snapshots
of the riser conﬁguration (top left), top tension (top right), relative horizontal
distance (bottom left) and payout (bottom right).
8.3 Eﬀect of Shallow Water
The control objectives equal payout and equal eﬀective length, are based on
payout and tension measurements at the wellhead and available today, and might
therefore more likely be implemented by the industry. Hence, the equal payout
and equal eﬀective length control objectives are subject for the next investigation
in shallow waters.
The risers are located at 300m water depth and the current proﬁle from the
Ormen Lange ﬁeld is scaled to this depth, keeping the same velocity in each node.
The risers have the same physical dimensions as before except for the length. The
centertocenter distance is decreased to 8D. Keeping the same riser diameter, the
upper tension limit could be the same due to stress considerations, whereas the
lower tension limit is due to eﬀective weight. Here, the water depth and also the
eﬀective weight is one fourth of the previously simulated depth of 1200m. The
135
8. Simulation Results
0 2 4 6
0
50
100
150
200
250
300
[m]
[
m
]
Snapshots
0 200 400 600 800
0.52
0.54
0.56
0.58
0.6
0.62
0.64
0.66
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
350
400
450
500
550
600
650
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−2
0
2
4
6
8
10
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.10: Equal payout control of R2 at 300m water depth. Snapshots of the
riser conﬁguration (top left), top tension (top right), relative horizontal distance
(bottom left) and payout (bottom right).
lower tension limit could be as low as 350kN at 300m water depth. The tension
level is much lower and the axial elongation therefore smaller, and hence excepted
to be of less importance. The equal payout and the equal eﬀective length control
objectives are tested with an initial top tension of 600kN, zero oﬀset and control
of R2, starting after 200s. Fig. 8.10 (top left) shows how R2 slides out to the right
due to the decreased tension when the control is turned on. The payout is seen
to be equal for the two risers in Fig. 8.10 (bottom right). For the equal eﬀective
length control objective in Fig. 8.11, R2 slides slightly more to the right, and the
payout is seen to be 23cm larger for R2 than for R1, due to the compensation
of the elasticity. The tension of R2 is right above 400kN using equal payout and
right below 400kN using the equal eﬀective length control objective. The smallest
relative horizontal distance between the risers is about 6.7D in the ﬁrst case and
7.2D in the latter. Hence, a small eﬀect of taking the elasticity into account is
seen even at 300m water depth. However, it is of far less importance than in
136
TLP Dynamics
0 2 4 6
0
50
100
150
200
250
300
[m]
[
m
]
Snapshots
0 200 400 600 800
0.5
0.55
0.6
0.65
0.7
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800
350
400
450
500
550
600
650
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800
−2
0
2
4
6
8
10
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.11: Equal eﬀective length control of R2. Snapshots of the riser conﬁg
uration (top left), top tension (top right), relative horizontal distance (bottom
left) and payout (bottom right).
deeper waters.
8.4 TLP Dynamics
For the simulations with dynamic TLP motions, 1200m water depth is used. In
this case, the TLP moves with harmonic motions in surge about a static oﬀset of
30m. The period is 120s, and the peaktopeak amplitude 40m. The controllers
are enabled after 400s. Fig. 8.12 shows the horizontal positions for three selected
nodes. It is seen in these plots, as expected, that the upper nodes are most
inﬂuenced by the TLP motion. In this case where R2 is controlled, we see that
the collision is avoided by increasing the horizontal position of R2. When R1 was
controlled, the horizontal positions of R1 decreased, and the horizontal positions
for R2 increased at the same time due to reduced shielding, not shown here.
137
8. Simulation Results
10
20
30
40
50
Node 9
[
m
]
10
20
30
40
50
Node 6
[
m
]
0 100 200 300 400 500 600 700 800 900 1000
10
20
30
40
50
Node 3
[
m
]
Time [s]
R1
R2
Figure 8.12: Dynamic TLP motions and control of R2 enabled after 400s. Hori
zontal positions for node 3, 6 and 9 for R1 and R2.
The riser slides out to the right when the payout ξ
2
increases, see Fig. 8.13.
The relative horizontal position between the same three nodes is seen in Fig.
8.13 (bottom left). The eﬀect of control is clearly seen. Before the controller
is turned on, collision occurs twice for each cycle. After 400s, the controller is
turned on, and the mean distance is about 15D, equal to the top and bottom
distance, for all nodes, giving a similar conﬁguration for the two risers. Also, the
variation in distance between corresponding nodes decreases signiﬁcantly. The
payout with control of R1 and R2 are given in Figs. 8.13 and 8.14, respectively.
For the uncontrolled ﬁrst 400s, the dynamic stroke of R1 is 1.6m and 0.8m for
R2. This diﬀerence is caused by larger drag and deﬂection of R1. R2 has a more
straightlined conﬁguration due to reduced drag forces. Note that when R2 comes
in front, the model is not valid and R2 keeps its straight conﬁguration.
When R1 is the reference, the stroke of R2 is increased, and the stroke of R1 is
slightly decreased, giving a dynamic stroke about 1.5m for both risers. The mean
tension of R2 is decreased to about 1450kN, giving a tension diﬀerence of 350kN
138
TLP Dynamics
0 10 20 30 40 50
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800 1000
0
0.5
1
1.5
2
2.5
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800 1000
1200
1400
1600
1800
2000
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800 1000
−20
−10
0
10
20
30
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.13: Dynamic TLP motions and control of R2 enabled after 400s. Snap
shot of the riser conﬁguration (top left), top tension (top right), relative horizontal
distance (bottom left) and payout (bottom right).
between the risers, see Fig. 8.13. The smaller mean tension compared to the
static case is due to longer payout and total eﬀective length, such that a smaller
tension diﬀerence is needed to avoid collision. The dynamic tension variation has
a period of 120s as for the TLP motions, and a peaktopeak amplitude about
120140kN.
When R2 is the reference for R1, its dynamic stroke decreases, giving both
R1 and R2 a dynamic stroke about 1m, see Fig. 8.14. The mean tension level is
increased with 400kN for T
1
, seen in Fig. 8.14. The dynamic tension variation was
about 120130kN. Increasing the TLP period to 300s, a more quasistatic riser
behavior is seen with less dynamic deﬂection. This gives less need for stroke, a
smaller payout and eﬀective length and a larger tension diﬀerence closer to the
static case.
139
8. Simulation Results
0 10 20 30 40 50
0
200
400
600
800
1000
1200
[m]
[
m
]
Snapshots
0 200 400 600 800 1000
0
0.5
1
1.5
2
2.5
Payout
[
m
]
Time [s]
ξ
1
ξ
2
0 200 400 600 800 1000
1600
1800
2000
2200
2400
2600
Top tension
[
k
N
]
Time [s]
T
1
T
2
0 200 400 600 800 1000
−20
−10
0
10
20
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 6
Node 9
Figure 8.14: Dynamic TLP motions and control of R1 enabled after 400s. Snap
shot of the riser conﬁguration (top left), top tension (top right), relative horizontal
distance (bottom left) and payout (bottom right).
8.5 Supervisory Switched Controller
For permanent installations, like the TLP, the allyear weather changes must be
taken into consideration, also when designing the control system. The diﬀerent
ROCs should be identiﬁed and the control actions adjusted to each regime. This
includes controlling the top tension of the risers depending on the prevailing
situation, and also turning oﬀ the controller in calm weather to avoid unnecessary
wear and tear of the cylinders. Switching between the diﬀerent regimes and
control actions are illustrated in this section.
8.5.1 Case SetUp
The initial current is of factor 0.8 of the Ormen Lange design current with the
same variation as used before. The TLP is in a nooﬀset position, and the center
tocenter distance between the risers is 15D at the top and bottom end points.
140
Supervisory Switched Controller
At 400s, an additional, uniform tide current increases from 0m/s to 0.3m/s
over 300s. This total current velocity is used for the rest of the simulations. At
500s, the TLP starts moving to an oﬀset position of 30m, which it uses approx
imately 300s to reach. When the TLP has almost arrived the oﬀset position, a
LF motion in surge with a period of 120s and a peaktopeak amplitude of 40m
starts (800s). At 1200s, an additional extreme WF motion is superimposed on
the LF TLP motion. This extreme wave has a 12s period and an amplitude of
20m. The linear waveinduced surge motion on the TLP can be found from the
response amplitude operator (RAO) for the TLP ASM600, see Faltinsen (1990).
This gives a WF TLP surge motion with the same period of 12s and a peakto
peak amplitude of 10m. The changes in current and TLP motions are given in
Table 8.1 and illustrated in Fig. 8.15. It should be noted that the changes in
mean current velocity and the diﬀerent environmental conditions happen faster
than in real life, but the magnitudes themselves are representative.
Disturbance Start time [s]
Tide current 400
TLP oﬀset 500
TLP LF 800
TLP WF 1200
Table 8.1: Case simulation.
The PPMs consist of 16 elements, while the ACPMs consist of 8 elements.
Only R2 is controlled, using the payout from R1, ξ
1
, as a reference. Due to the
deep water, the riser elongation ∆l
R
is included, giving the payout reference
ξ
r,2
= ξ
1
+ ∆l
R
, ∆l
R
=
∆T
EA
l
0
. (8.17)
8.5.2 Simulations without Control
First the simulation case is run without control for two diﬀerent pretensions, see
Fig. 8.16. To the left simulations with a pretension of 1800kN for both risers are
seen. The risers are close to collision from the start, but actual collision is seen
for node 9 after 200s, due to the small variations in current velocity. The relative
distance is smaller than 2D (bottom left). The collision is present until 400s. A
new collision is seen at 450s, and is lasting for the rest of the simulation. Recall
that the model is not valid for riser distances smaller than 2D, and we can only
say that for smaller relative distance than this, collision is likely to occur.
With a pretension of 2700kN for both risers (Fig. 8.16, right), collision ﬁrst
happens after 600s, when the current velocity is increased due to tide. For LF
141
8. Simulation Results
0 500 1000 1500
0
20
40
60
TLP position
[
m
]
Time [s]
0 500 1000 1500
0
0.5
1
1.5
[
m
/
s
]
Current velocities
Node 3 Node 9 Node 15
Figure 8.15: Disturbances from current (top) and TLP (bottom), acting on the
risers.
TLP motions, the risers are in a no collision situation in the rightmost TLP
oﬀset, where the risers are most straightlined. When the TLP moves to the left,
the risers experience larger relative velocities. R1 experiences larger drag forces
than R2, and hence a larger deﬂection and increased risk of collision. Notice the
diﬀerence in payout values in the top plots. The medium pretension of 1800kN
results in a payout approximately twice as big as the payout for the pretension
of 2700kN. The relative distance between the risers are similar.
8.5.3 Simulations with Control
The risers are exposed to the same environmental setup, but now control of
top tension is introduced. R2 is controlled by switching between the controllers
described in (6.45) to (6.49) as described in Section 6.4.
The initial top tension is 1800kN for both risers. Due to the shielding eﬀect
which gives diﬀerences in drag forces, the risers are starting in a state close to
collision, with 3D being the smallest distance between the risers (see Fig. 8.18).
The monitoring signal is smallest for process µ
5
, leading to the process switching
signal ρ = 5 which maps to the controller switching signal σ = 2 and the PI
controller (see Fig. 8.17). The tension for R2, T
2
, is decreasing such that the
distance between the risers is increasing to a level about 15D (Fig 8.18). The
payout for the two risers are also found in Fig. 8.18, showing a diﬀerence of 0.2m,
corresponding to the riser elongation due to tension.
142
Supervisory Switched Controller
0 500 1000 1500
0
1
2
3
4
[
m
]
Payout for T = 1800kN
R1
R2
0 500 1000 1500
−30
−20
−10
0
10
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 9
Node 15
0 500 1000 1500
0
1
2
3
4
[
m
]
Payout for T = 2700kN
R1
R2
0 500 1000 1500
−30
−20
−10
0
10
Time [s]
[
D
]
Relative horizontal distance
Node 3
Node 9
Node 15
Figure 8.16: Eﬀects of the constant top tension, 1800kN (left) and 2700kN (right),
payouts (top) and relative horizontal distance between the risers (bottom).
As the riser distance increases, the monitoring signal for the slowly varying
process, µ
3
, is the smallest, and the process switching signal switches to ρ = 3.
The same controller is kept, as it still maps to the same controller switching
signal σ = 2. After approximately 310s, R2 has reached a stable position near
the desired horizontal riser distance and a small variation in tension the last 100s.
The smallest monitoring signal is found for normal state µ
2
, giving ρ = 2 and
σ = 1 (Fig. 8.17). The top tension T
2
is set constant and equal the prevailing
value at the switching moment (Fig. 8.18).
From 550s, the smallest distance between the risers decreases, due to increas
ing current velocity from tide. The PIcontroller for T
2
is turned on when µ
3
is the least monitoring signal, giving ρ = 3 and σ = 2 (Fig. 8.17). The same
controller is kept when the TLP starts moving with LF motions. An additional
WF TLP motion is superimposed, starting at 1200s, see Fig. 8.19. At 1250s,
when the WF motions have achieved almost full amplitude, µ
4
is the smallest
monitoring signal, giving ρ = 4 and σ = 3 (Fig. 8.17), and the PIDcontroller is
143
8. Simulation Results
0 500 1000 1500
0
1
2
3
4
Controller switching signal
σ
1
2
3
4
5
6
Process switcing signal
ρ
0 500 1000 1500
0
2
4
Monitoring signal
µ
p
Time [s]
µ
1
µ
2
µ
3
µ
4
µ
5
Figure 8.17: From the top: Controller switching signal σ, process switching signal
ρ and switching error µ
p
.
phased in. The eﬀects of the TLP WF motions on the top tension and payout
are seen in Fig. 8.19. The top tension desired by the controller (–) and the satu
rated tension from the actuator ( ) are seen in the middle graph. The switching
from the PI to the PIDcontroller happens at 1250s, with a smooth transition
lasting 20s, such that at 270s the PIDcontroller is working alone. T
2
is in rate
saturation with the PIcontroller just before the PIDcontroller is phased in. The
tension variations are larger for the PIcontroller than the PIDcontroller due to
a larger proportional gain. Anyhow, the PIDcontroller is also seen to be in rate
saturation every 120s, corresponding to the LF TLP motions, where the TLP
moves from right to left, giving the largest deﬂections.
When the TLP moves against the current velocity, the relative velocity in
creases, and hence the drag forces increase with the square of the velocity. For
increasing relative velocity, the riser feels stiﬀer, and to straighten the riser or
even maintain the same curvature, a higher tension is needed. For a pure LF
motion the tension rates are not exceeded, but for an additional WF motion,
144
Supervisory Switched Controller
0 500 1000 1500
1200
1400
1600
1800
2000
Top tension
[
k
N
]
T1 T2
0 500 1000 1500
0
1
2
3
4
[
m
]
Payout
R1 R2
0 500 1000 1500
0
5
10
15
20
Time [s]
[
D
]
Relative horizontal distance
Node 3 Node 9 Node 15
Figure 8.18: Eﬀects of the controller. From the top: Top tension, payout and
relative riser distance.
large and fast changes in tension are requested, and rate saturation occurs. This
rate saturation is seen to aﬀect the payout of R2. The diﬀerence in payouts for
R1 and R2 corresponds to the riser elongation due to the tension diﬀerence in
the unsaturated periods. While the tension is in rate saturation, the payout for
R2 is not large enough, such that the payouts for R1 and R2 are close to equal.
However, this is not a problem with respect to collision for such fast motions here.
Regardless of this, we can not neglect that this could be a problem for risers with
diﬀerent physical properties and dynamics.
The relative distance between the upper nodes is naturally more aﬀected
by the WF motions than the nodes further down the riser (Fig. 8.18). For
instance, node 15 experiences more dynamics, with larger motions over a longer
period of time. It takes some time for the motions to travel down the riser.
The riser top seems most sensitive to the WF motions when it is in a nearly
Straightlined conﬁguration, corresponding to a TLP movement from left to right.
The least variation in relative distance is seen when the TLP moves in the opposite
145
8. Simulation Results
1200 1250 1300 1350 1400 1450 1500
1200
1300
1400
1500
1600
Top tension saturation T2
[
k
N
]
Actuator tension Controller tension
1200 1250 1300 1350 1400 1450 1500
0
1
2
3
4
Time [s]
[
m
]
Payout
R1 R2
1200 1250 1300 1350 1400 1450 1500
0
20
40
60
[
m
]
TLP position
Figure 8.19: The WF motion. From the top: TLP position, top tension and
saturation for T2 and payout.
direction and the deﬂection is largest. This means that when the ﬁrst mode is
the dominating mode shape, higher order mode shapes are less pronounced.
8.6 Discussion
The controller objective may change according to the ROC. For low current
velocities, constant tension may be a suﬃcient solution. Equal payout is shown to
be appropriate in shallow waters (∼300m). However, the eﬀects of elasticity and
riser elongation need to be included in the control strategy in order to prevent riser
collision in deep waters when top measurements are used. The desired relative
horizontal distance control objective indirectly includes the elasticity and showed
promising results, but the measurements along the riser used in this algorithm are
not easily available oﬀshore. Instead, a model is used to estimate the distance
between the risers, and used in the monitor to trigger switching between the
diﬀerent controllers. If measurements were easily available, they could be used
146
Discussion
instead. Note that the model used here is veriﬁed for LF TLP motions. For large
riser or TLP velocities, the wake model is no longer valid. However, WF TLP
motions are included in the simulation study to illustrate the eﬀect of control
also in this regime.
147
Chapter 9
Concluding Remarks
9.1 Conclusion
This thesis has focused on modeling, simulation and control of top tensioned ma
rine risers. More speciﬁcally, we have considered how we can prevent collision
between risers in an array, exposed to platform motions, current and hydrody
namic interaction forces by controlling the top tension. The main conclusion
from this study is that it may be possible to prevent riser collision by use of top
tension control.
A twodimensional mathematical model of a riser/TLP system was developed.
This included the current velocity proﬁles, hydrodynamic interaction between ris
ers, TLP motions, the riser model and the actuator. The hydrodynamic interac
tion could be described by three diﬀerent components; the mean shielding eﬀects,
WIO and VIV. Here, only the ﬁrst one was considered, as this was assumed to
be of major importance for the risers’ relative position. The risers were modeled
with FEM and integrated load and equilibrium iterations. Simpliﬁcations were
made by neglecting the bending stiﬀness and assuming free end rotations.
The riser model was veriﬁed with the commercial software RIFLEX (Fylling
et al., 2005). At 1200m water depth, the riser model with 20 elements was
compared to a RIFLEX model with 400 elements. These veriﬁcations showed
good agreement, both with respect to payout and horizontal deﬂections. The
agreement was better for high tensions and when the curvature was small. For a
larger curvature, more elements were needed to maintain the same accuracy. A
quasistatic model was also veriﬁed for shallow waters, and compared to RIFLEX
models with and without a stress joint at the seabed. The stress joint was seen
to be of importance in shallow waters, whereas it had less impact on the global
geometry in deep waters and could be considered insigniﬁcant. Hence, when
simulating production risers in shallow waters, the stress joint should be included.
149
9. Concluding Remarks
For a drilling riser which actually has a balljoint at the lower end, free end rotation
was a more correct way of modeling.
The riser model was then run with 2 to 20 elements and compared to the
RIFLEX model with 400 elements. This was motivated by the purpose of real
time control applications which require computationally fast models, while still
being able to describe the main physics with desired accuracy. The performance
of the quasistatic model with increasing TLP oﬀset was investigated through
error norms of the horizontal position, riser top inclination, area under the riser
curve and payout. Simulations showed that payout best reﬂected the models’
performance. This method was therefore used to compare the performance of
the dynamic model with diﬀerent number of elements. Based on the payout er
ror, the various element models were classiﬁed for diﬀerent applications. With
more than 15 elements, we achieved high accuracy and an accurate description of
the physical process (PPM). The models with 812 elements showed good agree
ment and rather accurate estimates of the process. They are appropriate for
ACPM purposes, where both fast simulations and high accuracy are required.
For few elements models (46), the errors were larger, but the main physics were
still captured. This model could be used in control analysis or fast simulations
(FCPM).
A control system design for control of top tensioned marine risers has been
proposed and needed instrumentation has been suggested. Instrumentation in
cludes payout and top tension at the TLP wellhead and the relative horizontal
distance along the riser. The diﬀerent control objectives to prevent riser collision
were proposed. These included equal top tension, equal payout, equal eﬀective
length and desired horizontal distance. Equal top tension is used by the industry
today, whereas equal payout was proposed by Huse and Kleiven (2000). We have
in this thesis extended these results to also include equal eﬀective length and
desired horizontal distance as new control objectives.
In deep waters equal top tension resulted in collision for medium current
velocities. However, for very low current velocities, equal tension can be appro
priate. Equal payout gave collision in a smaller riser segment, but collision still
occurred. By taking the riser elongation into account by introducing equal ef
fective length, collision was prevented. These two latter control objectives were
also simulated in shallow waters (300m). Here equal payout showed to be sat
isfactory, as the relative riser elongation was observed to be much smaller and
hence of less importance. However, in deeper waters the eﬀect of elasticity and
riser elongation needs to be included in the control strategy in order to avoid
riser collision. Desired horizontal distance included the elasticity indirectly and
showed promising results. Anyhow, the measurements used in this method is not
easily available oﬀshore today.
The diﬀerent regimes for riser operations were deﬁned, and a relation between
150
Proposals for Further Work
regime and control action was proposed. A simulation study with supervisory
switched control showed a case, with switching between the diﬀerent controllers
depending on the prevailing regime. These results may be of industrial interest,
and further studies should be carried out to investigate the possibility of physical
implementation oﬀshore.
9.2 Proposals for Further Work
This work has focused on a twodimensional model of risers in a tandem arrange
ment. A simple twodimensional wake model was applied. In these analyses we
have focused on two equal risers. Further analyses using the same model should
include riser for diﬀerent purposes (drilling, production, etc.) giving diﬀerent
riser characteristics (diameter, pretension, ﬂuid density, etc.) and thereby physi
cal behavior. As a next step, more than two risers should be included in an array
or matrix. To simulate risers placed in a matrix structure a threedimensional
model is needed. The largest challenge will be to make a good threedimensional
wake model, meaning that lift forces should be included, but also other hydro
dynamic interaction eﬀects such as WIO and VIV should be considered added in
the model. In addition the current can have varying direction proﬁles. Also di
rect wave loads and the ﬁrst order ﬂoater motions should be investigated. These
eﬀects are not included in the implemented wake model, and some modiﬁcations
might be needed to make the wake model valid in this regime. This modeling,
implementation and simulation study should be performed before full scale tests
can be carried out.
Within the ﬁeld of control several interesting challenges are found. These
could be summarized as:
• Formulating the tension reference trajectory as an optimization problem
with relative horizontal distance between the risers and energy consump
tions as possible parameters in an objective function.
• Synchronization of risers in an array or matrix, choosing one riser as a
leader and the others as followers.
• Preventing new collisions between risers in an array or matrix during tran
sition from an error situation to reach their desired positions and tension
trajectories. The sequence of the risers and how fast each riser should reach
its optimal tension trajectory is of importance. This could be formulated
as a maneuvering problem and called error state maneuvering here.
The proposed tasks could be solved for the twodimensional case ﬁrst and there
after extended to the threedimensional case. Lastly, both model tests and full
151
9. Concluding Remarks
scale experiments would be of importance for the actual implementation of top
tension control of risers to prevent collision aboard installations oﬀshore.
152
Bibliography
Aamo, O. M. (2002). Modeling and Control of Fluid Flows and Marine Structures.
PhD thesis. Departement of Engineering Cybernetics, Norwegian University of
Science and Technology (NTNU). Trondheim, Norway.
Aker Kværner (2007). Compensator and tensioners. [Online; accessed May 18].
URL http://www.akerkvaerner.com.
Aker Maritime (2002). Ormen Lange TLP with full processing facilities. Technical
Report 3700NNX1500117.
Allen, D. W., D. L. Henning and L. Lee (2005). Riser interferense tests on ﬂexible
tubulars at prototype reynolds number. In: Proc. of the 37th Annual Oﬀshore
Technology Conference. Houston, TX. OTC17290.
American Petroleum Institute (1987). Recommended practice for planning, de
signing and constructing tension leg platforms. Technical Report API RP 2T.
Washinton D.C., USA.
˚
Astr¨om, K. J. and B. Wittenmark (1997). ComputerControlled Systems  Theory
and Design. 3rd ed.. Prentice Hall.
Baarholm, R., T. Kristiansen, H. Lie and K. Herfjord (2005). Experimental in
vestigation of dual riser interaction. In: Proc. of 24th International Conference
on Oﬀshore Mechanics and Arctic Engineering. OMAE200567100.
Berge, S., A. Engseth, I. Fylling, C. M. Larsen, B. J. Leira and A. Olufsen (1992).
Handbook on design and operation of ﬂexible pipes. Handbook STF70 A92006.
Structural Engineering, SINTEF. Trondheim, Norway.
Blevins, R. D. (1994). Flow Induced Vibrations. Krieger publishing company.
Blevins, R. D. (2005). Forces on the stability of a cylinder in a wake. Journal of
Oﬀshore Mechanics and Arctic Engineering 127, 39–45.
153
Bibliography
Bokaian, A. and F. Geoola (1985). Hydrodynamic forces on a pair of cylinders.
In: Proc. of the 17th Annual Oﬀshore Technology Conference. Vol. 3. Houston,
TX. pp. 413–420. OTC5007.
Chaudhury, G. and C.Y. Ho (2000). Coupled dynamic analysis of platforms,
risers, and moorings. In: Proc. of the 32nd Annual Oﬀshore Technology Con
ference. Houston, TX. OTC12084.
Chen, X., J. Zhang, P. Y. F. Liagre, J.M. Niedzwecki and P. Teigen (2002).
Coupled dynamic analysis of a mini TLP: Comparison with measurements. In:
Proc. of the 21st International Conference on Oﬀshore Mechanics and Arctic
Engineering. Oslo, Norway. OMAE200228536.
Dean, Q. W. (1980). A riser angle positioning system (raps). In: Proc. of the
12th Annual Oﬀshore Technology Conference. Vol. 2. Houston, TX. pp. 169–
179. OTC3755.
Demirbilek, Z. (1989a). Tension leg platform  an overview of the concept, analysis
and design. In: Tension Leg Platform  A state of the art review (Z. Demirbilek,
Ed.). American Society of Civil Engineers.
Demirbilek, Z., Ed. (1989b). Tension Leg Platform  A state of the art review.
American Society of Civil Engineers.
DNV (2003). Global analysis procedure for riser interaction. Revision No. 3 issued
to NDP Report No. 20021539. Det Norske Veritas.
DNV (2005). Recommended practice for riser interference. Industry version issued
DNVRPF203. Det Norske Veritas.
Duggal, A. S. and J. M. Niedzwecki (1993). An experimental study of tendon/riser
pairs in waves. In: Proc. of the 25th Annual Oﬀshore Technology Conference.
Houston, TX, USA. pp. 323–333. OTC7239.
Egeland, O. (1993). Servoteknikk. Tapir Forlag.
Ersdal, S. (2004). An Experimental Study of Hydrodynamic Forces on Cylinders
and Cables in Near Axial Flow. PhD thesis. Departement of Marine Technology,
Norwegian University of Science and Technology (NTNU).
European Commission (2007). GALILEO  European Satel
lite Navigation System. [Online; accessed February 5]. URL
http://ec.europa.eu/dgs/energy transport/galileo/index en.htm.
154
Faltinsen, O. M. (1990). Sea Loads on Ships and Oﬀshore Structures. Cambridge
University Press. Cambridge, UK.
Fard, M. P. (2000). Modelling and Control of Mechanical Flexible Systems. PhD
thesis. Departement of Engineering Cybernetics, Norwegian University of Sci
ence and Technology (NTNU). Trondheim, Norway.
Fontaine, E., J. Capul, T. Rippol and P. Lespinasse (2007). Experimental and
numerical study on wake interference and clashing between steel catenary ris
ers. In: Proc. of the 26th Internaltional Conference on Oﬀshore Mechanics and
Arctic Engineering. San Diego, CA. OMAE200729149.
Fossen, T. I. (2002). Marine Control Systems  Guidance, Navigation and Con
trol of Ships, Rigs and Underwater Vehicles. Marine Cybernetics. Trondheim,
Norway.
Fredheim, A. (2005). Current Forces on Net Strucutres. PhD thesis. Departe
ment of Marine Technology, Norwegian University of Science and Technology
(NTNU). Trondheim, Norway.
Fylling, I. J., C. M. Larsen, N. Sødahl, H. Omberg, A. G Engseth, E. Passano
and K. Holthe (2005). Theory Manual RiﬂexTM34rev0. SINTEF Marintek.
Trondheim, Norway.
Gopalkrishnan, R., A. G. C. Ekvall, R. B. McDaniel, T. B. Heckler, J. E. Smith,
J. H. Haws and D. W. McMillan (1994). TLP production riser spacing: Steady
drag and lift on two moderately spaced oscillating cylinders. Technical Progress
Report BRC 4194. Shell Development Company. Houston, TX.
Halse, K. H. (2000). Norwegian deepwater program: Improved predictions of
vortewinduced vibrations. In: Proc. of the 32nd Annual Oﬀshore Technology
Conference. Houston, TX. OTC11996.
Herfjord, K., T. Holm˚as, B. Leira, M. Brydum and T. Hanson (2002). Compu
tation of interaction between deepwater risers, collision statistics and stress
analysis. In: 21th Int. Conf. on Oﬀshore Mech. and Arctic Eng.. Oslo, Norway.
OMAE200228152.
Hespanha, J. P. (2001). Tutorial on Supervisory Control. In: Lecture Notes for
the Workshop Control Using Logic and Switching for the 40th Conference on
Decision and Control. Orlando, FL.
Hespanha, J. P. and A. S. Morse (2002). Switching between stabilizing controllers.
Automatica 38(11), 1905–1917.
155
Bibliography
Hespanha, J. P., D. Liberzon and A. S. Morse (2000). Bounds on the number
of switching with scaleindependent hystersis: Application to supervisory con
trol. In: Proc. of the 39th IEEE Conference on Decision and Control. Sidney,
Australia.
Hespanha, J. P., D. Liberzon and A. S. Morse (2003). Hystersisbased switch
ing algorithms for supervisory control of uncertain systems. Automatica
39(2), 263–272.
Høklie, M., N. A. Jensen and I. Ciatti (2002). Position reference based on in
strumented drilling riser. In: 14th Deep Ocean Technology & Exhibition. New
Orleans, LA.
Huang, S., J. Bolstad and A. Marn (2004). Experimental study of wake shielding
eﬀect of the three cylinders in currents. In: Proc. of the 23rd International
Conference on Oﬀshore Mechanics and Arctic Engineering. Vancouver, British
Columbia, Canada. OMAE200451001.
Huse, E. (1987). Drag in oscillatory ﬂow interpreted from wake considerations.
In: Proc. of the 19th Annual Oﬀshore Technology Conference. Houston, TX.
pp. 135–142. OTC5370.
Huse, E. (1993). Interaction in deepsea riser arrays. In: Proc. of the 25th Annual
Oﬀshore Technology Conference. Houston, TX. pp. 313–322. OTC7237.
Huse, E. (1996). Experimental investigation of deep sea riser interaction. In: Proc.
of the 28th Annual Oﬀshore Technology Conference. Houston, TX. pp. 367–372.
OTC8070.
Huse, E. and G. Kleiven (2000). Impulse and energy in deepsea riser collisions
owing to wake interference. In: Proc. of the 32nd Annual Oﬀshore Technology
Conference. Houston, TX. OTC11993.
Imakita, A., S. Tanaka and Y. Amitani ans S. Takagawa (2000). Intelligent riser
angle control DPS. In: Proc. of ETCE/OMAE Joint Conference. New Orleans,
LA. OMAE2000/OSU OFT3001.
Johansen, T. A., T. I. Fossen, S. I. Sagatun and F. G. Nielsen (2003). Wave syn
chromizimg crane control during water enrty in oﬀshore moonpool operations
 experimental results. IEEE Journal of Oceanic Engingeering 28(4), 720–178.
Johansen, V., S. Ersdal, A. J. Sørensen and B. Leira (2006). Modelling of inex
tensible cable dynamics with experiments. International Journal of NonLinear
Mechanics 41(4), 543–555.
156
Kaasen, K. E., H. Lie, F. Solaas and J. K. Vandiver (2000). Norwegian deep
water program: Analysis of vortexinduced vibrations of marine risers based
on fullscale measurements. In: Proc. of the 32nd Annual Oﬀshore Technology
Conference. Houston, TX. OTC11997.
Kalleklev, A. J., K. J. Mørk, N. Sødahl, M. K. Nyg˚ard and A. M. Horn (2003).
Design guidelines for riser collision. In: Proc. of the 35th Annual Oﬀshore
Technology Conference. Houston, TX. OTC15383.
Kavanagh, W. K., L. Imas, H. Thompson and L. Lee (2000). Genesis spar risers:
Interference assessment and VIV model testing. In: Proc. of the 32nd Annual
Oﬀshore Technology Conference. Houston, TX. OTC11992.
Khalil, H. K. (2000). Nonlinear Systems. 3rd ed.. PrenticeHall Inc.. New Jersey.
Korde, U. A. (1998). Active heave compensation on drillships in irregular waves.
Ocean Engineering 25(7), 541–561.
Kreyszig, E. (1999). Advanced Engineernig Mathematics. 8th ed.. John Wiley &
Sons, Inc.
Langen, I. and R. Sigbj¨ornsson (1979). Dynamisk analyse av konstruksjoner.
Tapir. Trondheim, Norway.
Larsen, C. M. (1993). Analysis of tensioner system stroke for marine risers. Tech
nical Report STF70 F93011. SINTEF Structures and Concrete.
Larsen, C. M. (1995). Tension Leg Platforms, Lecture notes NA550/665. Techni
cal report. Dept. of Naval Architecture and Marine Engineering, University of
Michigan.
Leira, B. J., T. Holm˚as and K. Herfjord (2002). Estimation of extreme re
sponse and fatigue damage for colliding risers. In: Proc. of the 21st Interna
tional Conference Oﬀshore Mechanichs and Arctic Engineering. Oslo, Norway.
OMAE200228435.
Litton, R. W. (1989). TLPs and other deepwater platforms. In: Tension Leg
Platform  A state of the art review (Z. Demirbilek, Ed.). pp. 15–20. American
Society of Civil Engineers.
Ma, W., M.Y. Lee, J. Zou and E. W. Huang (2000). Deepwater nonlinear coupled
analysis tool. In: Proc. of the 32nd Annual Oﬀshore Technology Conference.
Houston, TX. OTC12085.
157
Bibliography
Mekha, B. B., C. P. Johnson and J. M. Roesset (1996). Implications of tendon
modeling on nonlinear response of TLP. Journal of Structural Engineering
122(2), 142–149.
Ministry of Petroleum and Energy and the Norwegian Petroleum Directorate
(2007). Facts  the Norwegian petroleum sector.
Mortazavi, M., R. B. Campbell and C. R. Brinkmann (2001). Modeling and
analysis of toptensioned risers. In: Proc. of the 20th International Confer
ence on Oﬀshore Mechanichs and Arctic Engineering. Rio de Janeiro, Brazil.
OMAE2001/OFT1168.
National Oilwell Varco (2007a). Homepage. [Online; accessed June 8]. URL
http://www.nov.com/.
National Oilwell Varco (2007b). NLine Tensioner Course. Technical report. Kris
tiansand, Norway.
Neto, E. T., C. Ruggieri and J. A. Ferrari (2002). Numerical representation of
collision interaction mass for dual risers systems. In: Proc. of 21st International
Conference on Oﬀshore Mechanics and Arctic Engineering. ASME. Oslo, Nor
way. OMAE200228336.
Nguyen, T. D. (2006). Design of Hybrid Marine Control Systems for Dynamic
Positioning. PhD thesis. National University of Singapore (NUS).
Nguyen, T. D., A. J. Sørensen and S. T. Quek (2007). Design of hybrid con
troller for dynamic positioning from calm to extreme sea conditions. Automat
ica 43(5), 768–785.
Nielsen, F. G. (2004). Lecture Notes TMR4225 Marine Operations. Technical
Report UK200486. Dept. of Marine Technology, NTNU.
NOAA Coastal Services Center (2007). Loop Current. [Online; accessed April
21]. URL http://www.csc.noaa.gov/crs/deﬁnitions/loop current.html.
Norsk Hydro (2001). Environmental Conditions. Ormen Lange Project 3700NH
F5000001.
Nowlin, W. D., A. E. Jochens, S. T. DiMarco, R. O. Reid and M. K. Howard
(2001). Deepwater physical oceanography reanalysis and synthesis of historical
data. Synthesis Report MMS 2001064. Dept. of Oceanography, Texas A & M
Univeristy. New Orleans, USA.
158
Nyg˚ard, M.K., F. R. Botros and H. Hannus (2001). Evalution of eﬀect of contact
between top tensioned risers in deep and ultra deep waters. In: Proc. of the 11th
International Oﬀshore and Polar Engineering Conference. Vol. 2. Stavanger,
Norway. pp. 222–226.
Oﬀshore Tecnology (2007). The website for oﬀshore oil & gas industry. [Online;
accessed August 9].
Often, O. (2000). Dry tree semi  reduced costs for dry well completions in deepwa
ter west africa by application of proven semisubmersible and riser technology.
In: Proc. of the 32nd Annual Oﬀshore Technology Conference. Houston, TX.
OTC11876.
Ogata, K. (1995). DiscreteTime Control Systems. 2 ed.. Prentice Hall.
Ormberg, H. and K. Larsen (1997). Coupled analysis of ﬂoater motion and moor
ing dynamics for a turret moored tanker. In: Proc. of the 8th International
Confeerence on the Behaviour of Oﬀshore Structures (BOSS) (J. H. Vugts,
Ed.). Vol. 2. The Netherlands. pp. 469–483.
Ormberg, H., I. Fylling, K. Larsen and N. Sødahl (1997). Coupled analysis of
vessel motions and mooring and riser system dynamics. In: Proc. of the 16th
International Conference Oﬀshore Mechanics and Arctic Engineering. Vol. 1A.
Yokohama, Japan. pp. 91–99.
Ormberg, H., N. Sødahl and O. Steinkjer (1998). Eﬃcient analysis of mooring
systems using decoupled and coupled analysis. In: Proc. of the 17th Inter
national Confonference Oﬀshore Mechanics and Arctic Engineering. Lisbon,
Portugal. OMAE980351.
Paganie, D. (2006). Shell restarts Mars TLP, includes industry ﬁrsts. Oﬀshore
Magazine 66(6), URL http://www.oﬀshore–mag.com/.
Perez, T., A. J. Sørensen and M. Blanke (2006). Marine vessel models in chang
ing operational conditions  a tutorial. In: 14th IFAC Symposium on System
Identiﬁcation (SYSID). Newcastle, Australia.
Perez, T. and P. Steinmann (2007). Modelling and performance of an active heave
compensation for oﬀshore operations. In: Proc. of IFAC Conference on Control
Applications in Marine Systems (CAMS07). Bol, Croatia.
Pollack, J., L. Poldervaart and M. Naciri (2000). A surface tree riser tensioning
system for FPSOs. In: Proc. of the 32nd Annual Oﬀshore Technology Confer
ence. Houston, TX. OTC11902.
159
Bibliography
Price, S. J. (1975). Wake induced ﬂutter of power transimission conductors. Jour
nal of Sound and Vibration 38(1), 125–147.
RD Instruments (2006). Marine Measurements. [Online]. URL
http://www.rdinstruments.com.
Reite, K.J. (2006). Modeling and Control of Trawl Systems. PhD thesis. Departe
ment of Marine Technology, Norwegian University of Science and Technology
(NTNU).
Rigzone (2007). Field development projects. [Online; accessed August 9]. URL
http://www.rigzone.com/.
Rustad, A. M., A. J. Sørensen and C. M. Larsen (2007a). Supervisory switched
control of marine risers. In: Proc. of the IFAC Conference on Control Applica
tions in Marine Systems (CAMS07). Bol, Croatia.
Rustad, A. M., C. M. Larsen and A. J. Sørensen (2007b). Deep water riser collision
avoidance by top tension control. In: Proc. of the 26th International Conference
on Oﬀshore Mechanics and Arctic Engineering. San Diego, CA. OMAE2007
29172, ISSN 1523651x.
Rustad, A. M., C. M. Larsen and A. J. Sørensen (2007c). FEM modelling and
control for collision prevention of top tensioned risers. Marine Structures. El
sevier, ISSN 09518339, Available online July 5.
Rustad, A. M., C. M. Larsen, M. S. Triantafyllou, F. S. Hover, A. H. Jacobsen
and A. J. Sørensen (2006). Modelling and control of colliding top tensioned
risers in deep waters. In: 7th IFAC Conference on Manoeuvering and Control
of Marine Crafts (MCMC06). Lisbon, Portugal.
Sabri, R., C. Putot, F. Biolley, C. Le Cunﬀ, Y. Creﬀ and J. L´evine (2006).
Automatic control methods for positioning the lower end of a ﬁliform structure,
notably an oil pipe, at sea. United States Patent 7,066,686 B2.
Sabri, R., J. L´evine, F. Biolley, Y. Creﬀ, C. Le Cunﬀ and C. Putot (2003). Con
nection of the riser to the subsea wellhead using active control. In: Proc. of the
2003 IEEE International Conference on Robotics and Automation (ICRA03).
Taipei, Taiwan.
Sagatun, S. I. (2002). Active control of underwater installation. IEEE Transac
tions on control systems technology 10(5), 743–748.
160
Sagatun, S. I., K. Herfjord and T. Holm˚as (2002). Dynamic simulation of marine
risers moving relative to each other due to vortex and wake eﬀects. Journal of
Fluids and Structures 16(3), 375–390.
Sagatun, S. I., K. Herfjord, F. G Nielsen and E. Huse (1999). Participating mass
in colliding risers. Journal of Marine Science and Technology 4(2), 58–67.
Sarpkaya, T. and R. L. Shoaﬀ (1978). A discrete vortex analysis of ﬂow about sta
tionary and transversely oscillating circular cylinders. Technical Report NPS
069SL790011. Naval Postgraduate School.
Schlichting, H. (1968). Boundary Layer Theory. McGraw Hill Book Company
Inc. New York.
Simpson, A. (1971). On ﬂutter of a smooth circular cylinder in a wake. The
Aeronautical Quarterly 22(1), 25–41.
Skjetne, R. (2005). The Maneuvering Problem. PhD thesis. Departement of Engi
neering Cybernetics, Norwegian University of Science and Technology (NTNU).
Trondheim, Norway.
Sørensen, A. J. (2005). Structural issues in the design and operations of marine
control systems. Annual Reviews in Control 29, 125–149.
Sørensen, A. J., B. J. Leira, C. M. Larsen and P. E. Høiskar (2002). Riser angle
based position reference system. In: Proc. of the 21st International Conference
on Oﬀshore Mechanics and Arctic Engineering. Oslo, Norway. OMAE2002
28077.
Sørensen, A. J., B. Leira, J. P. Strand and C. M. Larsen (2001). Optimal setpoint
chasing in dynamic positioning of deepwater drilling and intervention vessels.
Int. J. Robust Nonlinear Control 11, 1187–1205.
Sparks, C. P. (1984). The inﬂuence of tension, pressure and weigth on pipe
and riser deformations and stresses. Journal of Energy Resources Technology
106, 46–54. Transactions of ASME.
Stølen, S. F. (2007). Heave compensating systems for deep water drilling. Master’s
thesis. NTNU. Trondheim, Norway.
Sumer, B. M. and J. Fredsøe (1997). Hydrodynamics Around Cylindrical Struc
tures. Vol. 12 of Advanced Series on Ocean Engineering. World Scientiﬁc.
161
Bibliography
Suzuki, H., K. Watanabe and K. Yoshida (1995). Operability improvment of
deepsea riser by active control. In: Proc. of the 14th International Confon
ference on Oﬀshore Mechanics and Arctic Engineering. Vol. IB. Copenhagen,
Denmark. pp. 289–297.
Tsahalis, D. T. (1984). Experimental study and modelling of the vortexinduced
vibrations of a pair of risers/tethers of a compliant structure. Technical
Progress Report WRC 5783. Shell Development Company. Houston, TX.
Tsui, Y. T. (1977). On wakeinduced ﬂutter of a circular cylinder in the wake of
another. Journal of Applied Mechanics 44, 194–200.
T¨ urkyilmaz, Y. (2004). Modelling and Control of Towed Seismic Cables. PhD the
sis. Departement of Engineering Cybernetics, Norwegian University of Science
and Technology (NTNU). Trondheim, Norway.
van Smirren, J. R., J. Romeo and M. Vogel (1999). Developements in acoustic
doppler current proﬁling and its technical application in deepwater drilling. In:
Proc. of the 31st Annual Oﬀshore Technology Conference. Houston, TX. OTC
10752.
Vandiver, K. (1983). Drag coeﬃcients of longﬂexible cylinders. In: Proc. of the
Annual Oﬀshore Technology Conference. Houston, TX. pp. 405–410. OTC4490.
Wanvik, L. and J. Koos (2000). Two tier well top tensioning system. In: Proc. of
the 32nd Annual Oﬀshore Technology Conference. Houston, TX. OTC11904.
Wikipedia (2007a). Loop current  Wikipedia, the free encyclopedia. [Online;
accessed April 21]. URL http://en.wikipedia.org/wiki/Loop Current.
Wikipedia (2007b). Sonars  Wikipedia  the free encyclopedia. [Online; accessed
July 2]. URL http://en.wikipedia.org/wiki/Sonar.
Wroldsen, A. (2007). Modeling and Control of Tensegrity Structures. PhD the
sis. Departement of Marine Technology, Norwegian University of Science and
Technology (NTNU). Trondheim, Norway.
Wu, W., S. Huang and N. Barltrop (1999). Lift and drag forces on a cylin
der in the wake of an upstream cylinder. In: Proc. of the 18th International
Conference on Oﬀshore Mechanics and Arctic Engineering. St. John, Canada.
OMAE99/OFT4211.
Wu, W., S. Huang and N. Barltrop (2001a). Multiple stable/unstable equilibria
of a cylinder in the wake of an upstream cylinder. In: Proc. of the 20th In
ternational Conference on Oﬀshore Mechanics and Arctic Engineering. Rio de
Janeiro, Brazil. OMAE2001/OFT1170.
162
Wu, W., S. Huang and N. Barltrop (2001b). Stationary and hopf bifurcations of
equilibrium positions of a cylinder situated in the near and far wake ﬁelds of
an upstream cylinder. In: Proc. of the 11th International Oﬀshore and Polar
Engineering Conference. Stavanger, Norway. pp. 214–221.
Wu, W., S. Huang and N. Barltrop (2002). Wakeinduced large amplitude low
frequency motions of a vertical riser in the wake of an upstream riser. In:
Proc. of the 21st International Conference on Oﬀshore Mechanics and Arctic
Engineering. Oslo, Norway. OMAE200228035.
Zdravkovich, M. M. (1977). Review of ﬂow interference between two circular
cylinders in various arrangements. Journal of Fluids Engineering 99(4), 618–
633.
Zdravkovich, M. M. (1985). Forces on pipe clusters. In: Proc. of the Conference
Separated Flow around Marine Structures. Norwegian Institute of Technology.
Trondheim, Norway. pp. 201–226.
Zdravkovich, M. M. (2003). Flow around circular cylinders. Vol. 2. Oxford Science
Publications. Oxford.
163
Appendix A
Mathematics and Algorithm
for the FEM Model
This appendix includes a more detailed explanation of the wake ﬁeld model ap
plied in the system. The mass and stiﬀness matrices for an element are derived,
and the damping model discussed. The load on the risers and the quasistatic
algorithm are carefully explained.
A.1 Wake Field of a Single Cylinder
The downstream riser (R2) will experience shielding eﬀects due to the upstream
riser (R1). The prediction of the current on the downstream riser will be based on
wake and momentum considerations given by Huse (1993) and references therein
such as Schlichting (1968). Schlichting (1968) solved the equations of motion
in a wake. By assuming twodimensional motions, neglecting viscous stress and
holding the pressure constant through the ﬂuid, the turbulent wake ﬁeld can be
expressed as
b = 0.25
_
C
D1
D
1
x
s
, (A.1)
U
0
= V
c
_
C
D1
D
1
x
s
, (A.2)
u(y) = U
0
exp
_
−0.693
_
y
b
_
2
_
, (A.3)
where b is the half width of the wake, D
1
is the diameter of the riser generating
the wake, x
s
is the distance from the wake source, U
0
is the maximum velocity
in the velocity proﬁle, V
c
is the undisturbed free stream velocity, u is the wake
velocity proﬁle and C
D1
is the drag coeﬃcient of the riser generating the wake.
165
A. Mathematics and Algorithm for the FEM Model
Equation (A.1) is only expected to be valid some distance downstream of R1.
Close to R1, this will give a wake peak which is too high and narrow, and hence
lead to erroneous results when calculating the force on R2 placed in the wake.
In order to correct for this, a virtual source upstream of R1 is introduced. This
means that the distance x
s
in the equation above is substituted by
x
s
= x + x
v
, (A.4)
where x
v
is the distance from the virtual source to the upstream riser, and x is the
distance between the centers of the wake generating riser R1, and the downstream
riser R2, on which the force is calculated. By requiring that
b =
D
1
2
, (A.5)
at R1, where
x
s
= x
v
, (A.6)
we can ﬁnd x
v
by
1
4
_
C
D1
D
1
x
v
=
D
1
2
, (A.7)
1
4
C
D1
D
1
x
v
= D
2
1
, (A.8)
x
v
=
4D
1
C
D1
. (A.9)
At large distances this correction does not make much contribution, whereas it
makes a large diﬀerence in the wake ﬁeld close to R1. The drag force on each
element of R2 can now be found by Morison’s equation
f
drag,R2
= 0.5ρD
2
C
D2
[V
c
−u
2
[ (V
c
−u
2
) . (A.10)
A problem is though that u
2
varies over the space occupied by the riser R2, such
that this velocity is not well deﬁned. This can be solved by using the root mean
square (RMS) average over the riser diameter. The analytic RMS value is given
by
u
RMS
(x
s
) =
¸
¸
¸
¸
_
_
D/2
−D/2
u
2
(x
s
, y) y
2
dy
_
D/2
−D/2
u
2
(x
s
, y) dy
. (A.11)
The discrete RMS value (in general) is given by
R(x) =
_
Σ
n
i=1
x
2
i
n
, (A.12)
166
System Matrices
such that the numeric RMS value for the current on the riser diameter is
u
nRMS
(x
s
) =
¸
Σ
m
i=1
u
n
(x
s
, y)
2
i
m
. (A.13)
Since it is a symmetric Gaussian function, ﬁve halfside integrating points should
be appropriate.
A.2 System Matrices
Dynamic analysis by the ﬁnite element method consists of four steps (Langen and
Sigbj¨ornsson, 1979). The ﬁrst step, discretization, is to subdivide the structure
into elements. The second step, element analysis, is to establish the mass and
stiﬀness matrices and the load vector for each element. The third step, system
analysis, is to compute the mass and stiﬀness matrices and the load vector for
the whole structure by concatenating the corresponding element matrices. The
fourth and last step is the dynamic equilibrium equation. The second step for
the mass and stiﬀness matrices are derived here. The steps three and four are
given in Chapter 3. The load vector is derived in Appendix A.3.
A.2.1 Mass
The mass matrix is denoted consistent because the same interpolation polynomial
is used for derivation of the displacement for both the mass and the stiﬀness
matrices. From Langen and Sigbj¨ornsson (1979) the consistent mass matrix for
each element is written
m =
_
V
ρN
T
NdV = ρA
_
l
0
N
T
Ndx, (A.14)
where the interpolation polynomial for each pair of coordinates (x
1
, x
2
) and
(z
1
, z
2
) is given by
N =
_
1 −
x
l
,
x
l
¸
. (A.15)
l is the element length, such that
m
2×2
=
_
V
ρN
T
NdV = ρA
_
l
0
N
T
Ndx (A.16)
= ρA
_
l
0
_ _
1 −
1
l
x
_
2
1
l
x
_
1 −
1
l
x
_
1
l
x
_
1 −
1
l
x
_
1
l
2
x
2
_
dx (A.17)
=
ρAl
6
_
2 1
1 2
_
. (A.18)
167
A. Mathematics and Algorithm for the FEM Model
The total structural mass matrix for one element is then
m
si
=
ρ
s
Al
6
_
¸
¸
_
2 0 1 0
0 2 0 1
1 0 2 0
0 1 0 2
_
¸
¸
_
, (A.19)
where A is the cross sectional area of the riser and ρ
s
is the density of the riser
material steel. The shape of the mass for the internal ﬂuid is similar
m
fi
=
ρ
f
A
i
l
6
_
¸
¸
_
2 0 1 0
0 2 0 1
1 0 2 0
0 1 0 2
_
¸
¸
_
, (A.20)
where A
i
is the internal area of the riser and ρ
f
is the density of the internal ﬂuid.
The added mass does only have components in the xdirection. The matrix for
the added mass is given below as
m
ai
=
ρ
w
A
e
l
6
_
¸
¸
_
2 0 1 0
0 0 0 0
1 0 2 0
0 0 0 0
_
¸
¸
_
, (A.21)
where A
e
is the external area and ρ
w
the density of water.
A.2.2 Stiﬀness
The stiﬀness matrix for one element is given in Langen and Sigbj¨ornsson (1979)
as
k =
_
l
0
EAB
T
Bdx, (A.22)
where E is Young’s modulus of elasticity and the interpolation polynomial is
given by
B =
d
dx
N =
_
1
l
,
1
l
¸
, (A.23)
such that
k
2×2
=
_
l
0
EAB
T
Bdx (A.24)
=
EA
l
_
1 −1
−1 1
_
. (A.25)
168
System Matrices
The elastic stiﬀness is acting in the axial direction, and gives the local elastic
stiﬀness matrix for each element i
k
EA
i
=
EA
l
_
¸
¸
_
0 0 0 0
0 1 0 −1
0 0 0 0
0 −1 0 1
_
¸
¸
_
. (A.26)
The geometric stiﬀness acts in the lateral direction, and the geometric stiﬀness
matrix for one element is given by
k
G
i
=
P
i
l
i
_
¸
¸
_
1 0 −1 0
0 0 0 0
−1 0 1 0
0 0 0 0
_
¸
¸
_
. (A.27)
A.2.3 Damping
The damping matrix describes the structure’s ability to dissipate energy. When
damping is present in a system, the kinetic energy in this system will decrease. If
it does not, an external energy supply is added. To model the damping correctly
is diﬃcult, but simpliﬁed models do usually give representative solutions; in this
case, a linear viscous (velocity proportional) damping in the structure itself is
used.
When a structure oscillates in a ﬂuid, a part of the dynamic ﬂuid pressure
on the structure will be in phase with its velocity. This pressure is referred to
as the hydrodynamic damping. This hydrodynamic damping usually consists of
two terms. The ﬁrst term is proportional to velocity and referred to as potential
damping caused by wave generation. The second term is the drag force due to
formation of vortices and is described by Morison’s equation, see (3.25). When
the cross section of the structure is small compared to the wavelength, i.e. a
slender structure, the nonlinear drag term will dominate and the linear term be
negligible. As a rule of thumb, the structure is slender if
λ
w
D
> 5, (A.28)
where λ
w
is the wave length, and D is the diameter of the slender structure. Drag
damping is taken into account by direct use of Morison’s equation.
Structural damping should also be accounted for. A simple way of modeling
this eﬀect is to apply the Rayleigh damping given by
C = α
1
M + α
2
K. (A.29)
169
A. Mathematics and Algorithm for the FEM Model
The inertia proportional term will damp out the low frequencies, while the stiﬀ
ness proportional term has the opposite eﬀect and will tend to damp out the
higher frequencies. Using orthogonality properties from Langen and Sigbj¨ornsson
(1979), we see that C will have the same properties as M and K
φ
T
i
Cφ
j
= α
1
φ
T
i
Mφ
j
+ α
2
φ
T
i
Kφ
j
= 0 for i ,= j. (A.30)
The modal damping coeﬃcients are given by
c
i
= φ
T
i
Cφ
i
= α
1
m
i
+ α
2
k
i
, (A.31)
where φ
i
is the mode shapes and λ
i
the damping relation given by
λ
i
=
c
i
2m
i
ω
i
=
1
2
_
α
1
ω
i
+ α
2
ω
i
_
. (A.32)
Hence, for α
1
= 0, λ
i
is proportional to ω
i
and for α
2
= 0, λ
i
is inverted pro
portional to ω
i
. If we know the overall damping level for two frequencies we can
determine α
1
and α
2
by
α
1
=
2ω
1
ω
2
ω
2
2
−ω
2
1
(λ
1
ω
2
−λ
2
ω
1
) , (A.33a)
α
2
=
2 (λ
2
ω
2
−λ
1
ω
1
)
ω
2
2
−ω
2
1
. (A.33b)
The inertia term leads to damping from rigid body motions in the same way as
these motions give inertia forces. This damping contribution is not wanted since
structure damping is mainly caused by strains in the material of the structure.
Hence, the inertia term in Rayleigh damping should not be included. The stiﬀness
term α
2
K is directly related to strains in the structure and can model interior
structural damping adequately. The damping ratio will, however, increase with
increasing frequency. The damping level, λ
i
, must therefore be correct at the
actual response frequency. The inﬂuence from damping at higher frequencies is
without signiﬁcance for the solution, and may in fact contribute to a more smooth
time history since false oscillations caused by the numerical method might be
suppressed.
A.3 Load on the Risers
The risers will experience load in the global horizontal direction due to currents.
The current loads will induce forces on the riser, which we will ﬁnd by using
Morison’s equation. The algorithm for loads, forces and displacements for the
static initialization procedure could be summarized as follows:
170
Load on the Risers
1. Decompose the current from the global frame to the local frame.
2. Calculate the load at each node for each element locally.
3. Calculate the forces at each node for each element locally.
4. Transform the forces back to the global system.
5. Add the forces for each node to ﬁnd the total force in the global frame.
6. Introduce load terms from for the speciﬁed motions from the TLP.
7. Find the displacement vector by r = K
−1
R.
Below each point is explained in detail.
Decompose the current from the global frame to the local frame. First
the current, given in the global frame, has to be decomposed into the local frame
of the node at which the current attacks. It is assumed that the global current
velocity vector in each node i, v
f
cur,i
, only has a horizontal component, such that
the v
f
z
component is zero. The ﬂow is linear, i.e. there are no vortices in the ﬂow.
The decomposed current in iframe is then
v
f
cur,i
=
_
v
f
x
v
f
z
_
i
=
_
v
f
x
0
_
i
, (A.34)
v
i
cur,i
= T
i
0,f
v
f
cur,i
, (A.35)
_
v
i
x
v
i
z
_
i
=
_
cos θ
i
−sin θ
i
sin θ
i
cos θ
i
_ _
v
f
x
0
_
i
(A.36)
=
_
v
f
x
cos θ
i
v
f
x
sin θ
i
_
i
=
_
v
f
x,i
cos θ
i
v
f
x,i
sin θ
i
_
. (A.37)
Calculate the load at each node for each element locally. The loads
in each element’s two nodes are found by Morison’s equation for the axial and
transverse direction
f
i
ext,x
=
1
2
ρ
w
D
i
C
Dn
l
i
v
i
x
¸
¸
v
i
x
¸
¸
, (A.38)
f
i
ext,z
=
1
2
ρ
w
D
i
C
Dt
l
i
v
i
z
¸
¸
v
i
z
¸
¸
, (A.39)
where f
i
ext,x
is the external force in x
i
direction for element i. When a force,
stiﬀness matrix etc. is described in the iframe, it is implicitly for element i only.
171
A. Mathematics and Algorithm for the FEM Model
ρ
w
is the density of the displaced ﬂuid, D
i
the diameter of element i, and C
Dn
and C
Dt
the drag force coeﬃcients in normal and tangential direction. v
i
x
and v
i
z
are the current velocities in the local iframe. The load in each direction is given
by
q
i
x,1
=
1
2
ρ
w
C
dn
D
i
v
i
x,1
¸
¸
v
i
x,1
¸
¸
, (A.40)
q
i
z,1
=
1
2
ρ
w
C
dt
D
i
v
i
z,1
¸
¸
v
i
z,1
¸
¸
. (A.41)
This gives the load in each of the two nodes of an element to be
q
xi,1
=
1
2
ρ
w
C
dn
D
i
v
f
x,i
cos θ
i
¸
¸
¸v
f
x,i
cos θ
i
¸
¸
¸ , (A.42)
q
zi,1
=
1
2
ρ
w
C
dt
D
i
v
f
x,i
sin θ
i
¸
¸
¸v
f
x,i
sinθ
i
¸
¸
¸ , (A.43)
q
xi,2
=
1
2
ρ
w
C
dt
D
i
v
f
x,i+1
cos θ
i
¸
¸
¸v
f
x,i+1
cos θ
i
¸
¸
¸ , (A.44)
q
zi,2
=
1
2
ρ
w
C
dn
D
i
v
f
x,i+1
sin θ
i
¸
¸
¸v
f
x,i+1
sin θ
i
¸
¸
¸ . (A.45)
Calculate the forces at each node for each element locally. Assuming
linearly varying transverse and axial load, the load on a bar is found in Fylling
et al. (2005) and is given as
f
i
n
=
l
6
_
¸
¸
_
2q
x1
+ q
x2
0
q
x1
+ 2q
x2
0
_
¸
¸
_
, (A.46)
f
i
t
=
l
6
_
¸
¸
_
0
2q
z1
+ q
z2
0
q
z1
+ 2q
z2
_
¸
¸
_
, (A.47)
where f
i
n
is the normal and f
i
t
the tangential forces. The sum of forces acting on
each local element in its frame is then given by
f
i
= f
i
t
+f
i
n
. (A.48)
This could be simpliﬁed to include forces in the x
i
direction only since these are
far larger than the forces in the z
i
direction. The forces in z
i
directions will not
have major impact on the dynamics, hence they could be neglected. The resulting
172
Load on the Risers
external force in each element is then given in the same direction as the load as
f
i
=
l
6
_
¸
¸
_
2q
x1
+ q
x2
0
q
x1
+ 2q
x2
0
_
¸
¸
_
. (A.49)
Transform the forces back to the global system. Before adding the forces
acting at each node, they have to be represented in the same frame. The forces
are transformed from its local frame to the global frame the same way as the
displacements
f
f
i
=T
f
i
f
i
, (A.50)
where f
f
i
is the external force in the two nodes in element i given in the global
fframe.
Add the forces for each node to ﬁnd the total force in the global frame.
After the forces at all nodes are transformed back to the global coordinate system,
the forces and moments are summarized in each node
f
f
x,i
= f
f
x
1
,i
+ f
f
x
2
,i−1
, (A.51)
f
f
z,i
= f
f
z1,i
+ f
f
z2,i−1
. (A.52)
After summarizing the forces at each node, the total force vector in fframe due
to the current forces is called f
drag
.
Introduce load terms for the speciﬁed motions from the TLP. The
speciﬁed motions for the quasistatic case are accounted for by correction terms
in the load vector. These are found from
f
corr
= f
drag
−k
spe
x
TLP
, (A.53)
where f
corr
is the corrected force vector, k
spe
is the speciﬁed stiﬀness vector
and x
TLP
is the scalar TLP motion in surge. f
corr
is a vector due to the two
dimensional model, and k
spe
is a vector due to the speciﬁed motion at the top
node in surge only. All vectors are described in the global fframe.
Find the displacement vector by r = K
−1
R. The displacements of the riser
for the quasistatic case is then found to be
r = K
−1
R, (A.54)
where R is the force vector. Upper case is used here since this is most com
mon within the ﬁeld of structural mechanics for load and force vectors. See the
following iteration algorithm for equilibrium between internal and external forces.
173
A. Mathematics and Algorithm for the FEM Model
A.4 QuasiStatic Algorithm
This algorithm seeks to ﬁnd the equilibrium between the internal and external
forces at each node. This includes corrected element lengths due to the elongation
and top tension, and the setdown correction at each node due to current loads
and deﬂections. The internal forces are the axial forces and are found from the
position of each node, the Pythagorean theorem, ∆l
i
and elastic stiﬀness
EA
l
.
The external forces are given by the top tension in the upper node, weight and
current loads.
Step 1, Initialization
1. l
0
is the length of each element i for a stressfree conﬁguration.
2. The initial top tension P
top
of the vertical, tensioned riser is applied.
3. Find the axial eﬀective tension in the middle of each element given as the
tension in the element above, minus half the weight of the above and the
current element. The eﬀective tension in each element is found by
P
n
= P
top
−
1
2
w
eff,n
, (A.55)
P
i
= P
i+1
−
1
2
(w
eff,i
+ w
eff,i+1
) , i = ¦1, . . . , n −1¦ , (A.56)
w
eff,i
= (w
t
−ρ
w
g A
e
+ ρ
i
g A
i
)
i
l
0,i
, (A.57)
where P
i
is the eﬀective tension in element i positive vertical upwards, and
w
eff,i
is the eﬀective weight of element i. w
t
is the true weight in air and
n is the number of elements.
4. Find the new length of each element given by
l
i
= l
0
+
P
i
EA
l
0
+ pressure terms. (A.58)
The pressure terms take care of the contribution from external and inter
nal pressure in the riser, but are constant in time as long as the pressure
components are constant. It is hence not necessary to include these terms
in the present analysis, see Sparks (1984) and Larsen (1993).
The element lengths will not be equally long and an updated stiﬀness matrix
for the new element lengths is needed.
174
QuasiStatic Algorithm
5. Correct for the new element lengths and ﬁnding the new vertical positions
of the nodes. Starting from the bottom and building the vector upwards,
we have
z
1
= 0, (A.59)
z
i
= z
i−1
+ l
i
, (A.60)
z
n+1
= z
n
+ l
n
= z
top
. (A.61)
The x
i
positions are all zero, due to the vertical riser. The updated ver
tical z
i
positions are together with the x
i
positions stored in the vertical
displacement vector r
0
.
6. Find the local stiﬀness matrices for each element. This gives
k
i
= k
EA
i
(l
i
, P
i
) +k
G
i
(l
i
, P
i
) . (A.62)
In this ﬁrst iteration with the vertical riser, all the inclinations θ
i
will be
zero, such that the global and local frames are parallel. Hence,
θ
i
= 0, (A.63)
T
f
i
=
_
1 0
0 1
_
. (A.64)
7. Concatenate the local stiﬀness matrices to make the global stiﬀness matrix.
This gives
K =
_
¸
¸
¸
¸
¸
¸
¸
_
¯
k
1
11
¯
k
1
12
¯
k
1
21
¯
k
1
22
+
¯
k
2
11
¯
k
2
12
¯
k
2
21
¯
k
2
22
+
¯
k
3
11
.
.
.
¯
k
n−1
22
+
¯
k
n
11
¯
k
n
12
¯
k
n
21
¯
k
n
22
_
¸
¸
¸
¸
¸
¸
¸
_
. (A.65)
This matrix K is the incremental stiﬀness matrix for the vertical riser.
Incremental means stiﬀness against further displacements from the de
formed state. The matrix with rows and columns corresponding to the
nonprescribed degrees of freedom is named K
incr
, whereas prescribed mo
tions are stored in a matrix K
pre
.
Step 2, Move the TLP to an oﬀset position and add current forces to
the system. The top tension and eﬀective weight are now included in K
incr,k
.
175
A. Mathematics and Algorithm for the FEM Model
k is the number of iterations within the time step, ∆r
k
is the displacement correc
tion term and ∆R
k
is the forces from current loads and TLP displacement. The
forces from top tension and eﬀective weight are already included in the stiﬀness
matrix.
1. Solve the ﬁrst iteration
K
incr,1
∆r
1
= ∆R
1
, (A.66)
by following the list below.
a) Find the external load. In this ﬁrst iteration, the inclination is zero,
such that the global and local frames are parallel. f
drag
is found from
the current load and Morison’s equation. Calculate the external load
q in each element i, and the corresponding forces. Add the forces to
ﬁnd the global vector. This is described in detail in Appendix A.3.
b) The force correction due to the boundary conditions in TLP is found
from
∆R
corr,1
= f
drag
−k
spe
r
TLP,1
. (A.67)
The initial value of the current velocity on the downstream riser is
assumed to be equal to the undisturbed free stream velocity in this
initialization, such that the risers are standing in parallel.
c) Here the new positions for the riser elements are found after the TLP
position is included
∆r
1
= K
−1
incr,1
∆R
corr,1
, (A.68)
r
update
= r
0
+ ∆r
1
, (A.69)
where r
update
is the updated riser position vector and ∆r
1
is the contri
bution from the iteration. In the vector r
0
, x
i
= 0 and the z
i
positions
are given from the updated length elements.
2. Geometric correction of the vertical position due to deﬂection. The ele
ment lengths l
i
found in Step 1.4 are assumed correct, and the horizontal
displacements are accepted. This is used to correct the displacements in
the vertical direction and identify the vertical position. This will give the
setdown of the top node of the riser. The new, corrected vertical position
for node i is found by the Pythagorean theorem
z
i,pyth
=
_
l
2
i
−∆x
2
i
, (A.70)
∆x
i
= x
i+1
−x
i
, (A.71)
176
QuasiStatic Algorithm
which means that ∆x
i
is positive if the upper node is to the right of the
node below. Starting from the bottom node, the vertical positions are given
by
z
1,corr
= 0, (A.72)
z
i,corr
= z
i−1,cor
+ z
i−1,pyth
, i = ¦2, . . . , n + 1¦ , (A.73)
with origin of the global coordinate system in the bottom node, and z
i
positive upwards. The vertical position of the next node is found by taking
the previous node as the starting point for the next one. The corrected
vertical position is the sum of the starting point and the distance found
by (A.70). The corrected displacement vector r
corr,1
is now found as the
sum of the corrected vertical positions, whereas the previous values from
the r
update
vector are kept for the x
i
positions
r
corr,1
= r
update,x
+r
corr,z
. (A.74)
The sin θ
i
and cos θ
i
values to be used in transformations between local and
global coordinate systems are found from
cos θ
i
=
z
i,pyth
l
i
, (A.75)
sinθ
i
=
∆x
i
l
i
. (A.76)
Step 3, Equilibrium iterations. In this position, we start the equilibrium
iterations in each node. In the initialization procedure only the static situation
is considered, whereas in the time simulation, the solution is for the dynamic
system.
The resulting internal force vector f
f
int,i
at each node should be balanced with
the external load f
f
ext,i
at each node, see Fig. 3.9. The eﬀective weight should
be included in the external force when ﬁnding the equilibrium balance with the
internal forces. The nodes and freedoms considered are the same as found in
the stiﬀness matrix for the nonprescribed degrees of freedom. Hence, the two
degrees of freedom in the bottom node and the horizontal degree of freedom in
the top node, will not be considered in the equilibrium iterations.
1. To ﬁnd the internal forces in the global frame in this equilibrium iteration,
we use the inclination values found in Step 2.2, and the axial tension from
Step 1.3, assuming that the internal axial force is the same along each
element.
177
A. Mathematics and Algorithm for the FEM Model
Decomposing the forces to ﬁnd f
int,i
at each node, we use the inclination of
this element, and transform the two axial tensions to this frame. P
i
i
is the
local internal axial force vector of element i in frame i. In the ﬁrst iteration,
it is the eﬀective tension found in previous steps, since this is the one we
believe most in at this point of the algorithm. P
i−1
is downwards, whereas
P
i
is upwards, seen from node i
P
i
i
=
_
0
P
i
_
, (A.77)
P
f
i
= T
f
i
P
i
i
, (A.78)
P
f
i−1
= T
f
i−1
P
i−1
i−1
, (A.79)
f
f
int,i
= P
f
i
−P
f
i−1
. (A.80)
The resulting force from the axial forces at the nodes are calculated from
secants for each element.
2. The secant angle over two elements between nodes i − 1 and i + 1 is used
as an approximation for the tangent at node i
sin σ
i
=
∆x
i
√
∆x
i
+ ∆z
i
, (A.81)
cos σ
i
=
∆z
i
√
∆x
i
+ ∆z
i
, (A.82)
∆x
i
= x
i+1
−x
i−1
, (A.83)
∆z
i
= z
i+1
−z
i−1
. (A.84)
This inclination is found in the global system, and is used to calculate
external force due to the current, f
drag,k
normal to the element at each
node. For the upper element we have
σ
n
= θ
n
. (A.85)
The drag forces are found using Morison’s equation locally at each node.
The force is thereafter transformed back to the global system. For the
downstream risers, the current is a function of the distance from the up
stream risers. At the upper node, we ﬁnd the reaction force which gives
equilibrium in the horizontal direction. The unbalanced force in the vertical
direction is included in the equilibrium iterations.
3. To ﬁnd the total external forces f
ext
, we subtract the eﬀective weight, w
f
eff
,
at each node (i.e. half element over and under, assumed constant) from the
drag forces. The eﬀective weight is positive downwards in the global system.
f
f
ext,k
= f
f
drag,k
−w
f
eff
. (A.86)
178
QuasiStatic Algorithm
4. The unbalanced force vector ∆R
f
unb,k
is the sum of internal and external
forces as these are pointing in opposite directions
f
f
ext,k
−f
f
int,k
= ∆R
f
unb,k
. (A.87)
The resultants should be equal, but in opposite directions.
5. The criteria for ending the iteration is given by the pnorm of the unbalanced
force for each degree of freedom at each node. Here we have used the l
2

norm.
[x[
p
=
_
i
[x
i
[
p
_
1/p
, (A.88)
¸
¸
¸∆R
f
unb,k
¸
¸
¸
p
=
_
i
¸
¸
¸∆R
f
unb,k,i
¸
¸
¸
p
_
1/p
, (A.89)
¸
¸
¸
_
n+1
i=1
_
(f
ext,x,i
−f
int,x,i
)
2
+ (f
ext,z,i
−f
int,z,i
)
2
_
< 0.1%P
top
. (A.90)
If this diﬀerence is larger than the given acceptance level, for instance 0.1%
of the applied top tension, we have to start an iteration process to ﬁnd the
equilibrium.
6. The iteration procedure: The element lengths are still assumed to be cor
rect. The stiﬀness matrix K
k
is updated with the new positions, r
corr,1
and
the local matrices are the same as in Step 1.6. The transformation matrix
T
f
i
is used to transform the local matrices to the global frame calculations
k
ik
= T
f
i
(k
EA
i
(l
i
, P
i
) +k
G
i
(l
i
, P
i
)) T
i
f
, (A.91)
K
k
=
_
¸
¸
¸
¸
¸
¸
¸
_
¯
k
1
11
¯
k
1
12
¯
k
1
21
¯
k
1
22
+
¯
k
2
11
¯
k
2
12
¯
k
2
21
¯
k
2
22
+
¯
k
3
11
.
.
.
¯
k
n−1
22
+
¯
k
n
11
¯
k
n
12
¯
k
n
21
¯
k
n
22
_
¸
¸
¸
¸
¸
¸
¸
_
. (A.92)
The full stiﬀness matrix is found, but only the nonprescribed degrees of
freedom are included in the matrix used in the equilibrium equations.
7. The incremental riser displacement is found from
K
incr,k
∆r
k
= ∆R
unb,k
, (A.93)
179
A. Mathematics and Algorithm for the FEM Model
where K
incr,k
is the updated stiﬀness matrix, ∆r
k
is the incremental riser
displacement vector and ∆R
unb,k
is the unbalanced force for iteration k.
8. The incremental riser vector is added to the previous corrected riser vector
r
corr,k
r
corr,k+1
= r
corr,k
+ ∆r
k+1
. (A.94)
9. New axial force in each element in the new, corrected position is calculated
locally by the Pythagorean theorem. The new positions are calculated
locally by
l
new,i
=
_
∆x
2
i
+ ∆z
2
i
, (A.95)
∆x
i
= x
i+1
−x
i
, (A.96)
∆z
i
= z
i+1
−z
i
, (A.97)
∆l
i
= l
new,i
−l
0
, (A.98)
P
i
=
EA
l
∆l
i
. (A.99)
The new inclinations θ
i
and σ
i
are found in the same step as before.
10. We can then ﬁnd the internal and external forces and hence the unbalanced
force. For the downstream riser, the current is updated for the new position
and hence the new relative distance.
A.5 Metrical Norms
The vector norms are deﬁned in Kreyszig (1999). The norm x of a vector x is
a realvalued function with the properties:
• x is a nonnegative real number.
• x = 0 if and only if x = 0.
• kx = kx for all k.
• x +y ≤ x +y.
The norms are labeled with subscripts. The most important is the pnorm,
deﬁned by
x
p
= ([x
1
[
p
+[x
2
[
p
+ +[x
n
[
p
)
1/p
, (A.100)
where p is a ﬁxed number. In practice, the norms most used are is p = ¦1, 2 ∞¦,
that is
180
Metrical Norms
1. l
1
norm: x
1
= ([x
1
[ +[x
2
[ + +[x
n
[).
2. l
2
norm: x
2
=
_
(x
2
1
+ x
2
2
+ + x
2
n
).
3. l
∞
norm: x
∞
= max
j
[x
j
[.
181
Appendix B
Simulation Data
The data used in the veriﬁcations in Chapter 4 and the simulations in Chapters
7 and 8 are given here. This includes the environmental data, the riser charac
teristics and controller gains.
B.1 Environmental Data
The environmental data presented here include the current velocity proﬁles in
troduced in Section 3.1. Furthermore, data for tide and TLP oﬀset and dynamics
are included for fulﬁllment and validation of the values used in the simulations
in this thesis. These data are from the Ormen Lange Field in the North Sea and
data are found in Norsk Hydro (2001) and Aker Maritime (2002).
B.1.1 Current
The data for the current velocity proﬁles presented in Section 3.1 and illustrated
in Fig. 3.2 are given here. The theoretical proﬁles, i.e. uniform, linearly sheared
and bidirectional proﬁles are given in Table B.1. All current velocities are given
in m/s. The geographically based proﬁles from GoM are found in Table B.2.
They are based on Nowlin et al. (2001) and scaled to 1200m water depth and
surface velocity of 1m/s. Recall that GoM1 is a wind driven proﬁle and GoM2
has a loop eddy in the top layer. “” means that the current velocity is linear
between the values at two given water depths. The proﬁle OL1 from the Ormen
Lange ﬁeld is based on Herfjord et al. (2002) and scaled to 1200m waters. It is
also found in Table B.2. The Ormen Lange design current proﬁle, OL2, which is
used in most simulations and ﬁgures in this thesis is found in Norsk Hydro (2001)
and Aker Maritime (2002), and reproduced here in Table B.3 for 1, 10 and 100
years return periods.
183
B. Simulation Data
Depth [m]
Velocity[m/s]
Uniform Linearly Sheared Bidirectional
0 1.0 1.0 1.0
1200 1.0 0 1.0
Table B.1: The theoretical current proﬁles.
Depth [m]
Velocity[m/s]
GoM1 GoM2 OL1
0 1.0 1.0 1.0
120 0.8 0.4 
240 0.5 0.18 
360 0.5 0.15 
480 0.25 0.3 0.65
600 0.2 0.4 
720  0.45 
840  0.55 0.65
960  0.65 
1200 0.2 0.65 0
Table B.2: The geographically based design current proﬁles.
B.1.2 Tide
The tidal range with 100 year return period is according to Norsk Hydro (2001)
equal to 2.2m gives a tidal amplitude of 1.1m. The tide will have a period around
6 hours.
B.1.3 TLP Oﬀset and Dynamics
The extreme case vessel oﬀset is based on the 100 year wave, 100 year wind
plus 10 year current. This gives a mean oﬀset of 50m and a dynamic oﬀset of
184
Riser Data
Depth below sea level [m]
Velocity[m/s]
1 year 10 year 100 year
0 (surface) 1.15 1.30 1.40
20 1.15 1.30 1.40
50 1.15 1.30 1.40
100 1.10 1.25 1.35
200 1.05 1.20 1.25
300 1.00 1.05 1.10
400 0.95 1.05 1.10
600 0.75 0.80 0.95
750 0.55 0.60 0.70
850 0.55 0.60 0.70
Table B.3: Ormen Lange design current velocity proﬁle for various return periods.
17m. The maximum oﬀset is then said to be 67m (Aker Maritime, 2002). We
mainly used 30m static oﬀset, 20m dynamic LF oﬀset and 12m WF oﬀset in the
simulations.
B.2 Riser Data
The riser data decide the static and dynamic riser characteristics. The riser data
for 1200m and 300m water depth are found in Tables B.4 and B.5, respectively.
B.3 Controller Gains
The controller gains used in the simulations in Chapter 8 are given in this section.
Table B.6 gives the controller gains for both risers used in the investigation of
the ﬁrst control objective principle. Table B.7 gives the controller gains for the
second control objective principle. In Table B.8 the controller gains used in
shallow water simulations are found, and Table B.9 gives the control parameters
for the supervisory switched control in Section 8.5.3.
185
B. Simulation Data
Parameter Description Value Dimension
α
2
Damping coeﬃcient 0.0477 []
D
e
Diameter 0.3 [m]
d Water depth 1200 [m]
t
h
Wall thickness 0.015 [m]
C
D
Drag coeﬃcient 1.0 
C
M
Mass coeﬃcient 2.0 
E Modulus of elasticity 206 [GPa]
f
u
Yield stress steel 500 [MPa]
l
r
Riser length 1212 [m]
l
t
Tendon length 1166 [m]
T
max
Upper tension limit 2700 [kN]
T
min
Lower tension limit 1200 [kN]
ξ
0
Initial payout 0.5 [m]
ρ
s
Speciﬁc weight for steel 7850 [kg/m
3
]
ρ
f
Speciﬁc weight for ﬁlling 800 [kg/m
3
]
ρ
w
Speciﬁc weight for sea water 1026 [kg/m
3
]
∆x
d
Initial riser distance 15 [D]
Table B.4: Riser data for 1200m water depth.
186
Controller Gains
Parameter Description Value Dimension
α
2
Damping coeﬃcient 0.0477 []
D
e
Diameter 0.3 [m]
d Water depth 300 [m]
t
h
Wall thickness 0.015 [m]
C
D
Drag coeﬃcient 1.0 
C
M
Mass coeﬃcient 2.0 
E Modulus of elasticity 206 [GPa]
f
u
Yield stress steel 500 [MPa]
l
r
Riser length 312 [m]
l
t
Tendon length 266 [m]
T
max
Upper tension limit 2700 [kN]
T
min
Lower tension limit 350 [kN]
ξ
0
Initial payout 0.5 [m]
ρ
s
Speciﬁc weight for steel 7850 [kg/m
3
]
ρ
f
Speciﬁc weight for ﬁlling 800 [kg/m
3
]
ρ
w
Speciﬁc weight for sea water 1026 [kg/m
3
]
∆x
d
Initial riser distance 8 [D]
Table B.5: Riser data for 300m water depth.
Parameter Description Value Dimension
K
P
Proportional gain 607500 
T
I
Integration time 8 [s]
Table B.6: Controller gains used for both risers in Sections 8.2.2, 8.2.3 and 8.4.
187
B. Simulation Data
Parameter Description Value Dimension
K
P
Proportional gain 20250 
T
I
Integration time 30 [s]
Table B.7: Controller gains for the second control objective principle in Section
8.2.4.
Parameter Description Value Dimension
K
P
Proportional gain 1000000 
T
I
Integration time 8.5 [s]
Table B.8: Controller gains for shallow water in Section 8.3.
Parameter Description Value Dimension
K
P2
Proportional gain 2 81000 
K
P3
Proportional gain 3 27000 
T
D3
Derivation time 0.2 [s]
T
I
Integration time 20 [s]
h Switching hysteresis 0.3 
∆x
d
Desired distance 15 [D]
Table B.9: Controller gains for supervisory switched control in Section 8.5.3.
188
NTNU Norwegian University of Science and Technology Thesis for the degree of philosophiae doctor Faculty of Engineering Science & Technology Department of Marine Technology c Anne Marthine Rustad ISBN 9788247140116 (printed ver.) ISBN 9788247140253 (electronic ver.) ISSN 15038181 Theses at NTNU, 2007:183 Printed at Tapir Uttrykk
Abstract
This doctoral thesis presents new research results on the control of top tensioned risers in deep waters. The main motivation for this work is the development of a riser control system which control the top tension of the individual risers in an array, such that collisions between adjacent risers are prevented. The risk of collision increases with increasing water depth, and as the oil and gas industry is moving to ever deeper waters, riser interaction has become an issue of considerable concern. The two main design parameters to avoid collision are riser spacing and top tension, which both are expensive alternatives. Hence, new solutions are needed and dynamic control of the top tension is the solution proposed here. A mathematical model of the riser system is developed. It includes two risers in a tandem arrangement which are connected to a tension leg platform (TLP) through their top nodes, forcing the risers to follow the prescribed motions of the TLP in the horizontal direction. In the vertical direction the riser motion is decided by the actuator which is a tensioner system. The risers are also exposed to current forces which are found by considering hydrodynamic interaction. The risers are modeled using the ﬁnite element method (FEM). This model is veriﬁed by the commercial software RIFLEX. For the purpose of control applications, the model needs to be computationally fast, but still be able to describe the main physics of the real system. The number of elements needed to keep a desired level of accuracy is therefore investigated. Diﬀerent means to measure the performance of the model are considered, and among these, payout measurements reﬂected the displacement of the riser in the best manner. In addition measurements at the wellhead, like payout and tension, are both accurate and available at most installations today. Hence, the proposed control objectives are mainly based on keeping the top tension, payout or total riser length equal for the risers of concern. A simulation study showed that the dynamic variation of the riser elongation needed to be taken into account in the control algorithm for deep water risers. The reason was that the large diﬀerence between tension in the two risers resulted in a signiﬁcant length diﬀerence that had to be compensated for by diﬀerent payout. In shallow waters the diﬀerence between the riser elongations
i
Abstract
is smaller, and equal payout is shown to be appropriate. Equal tension, which is used by the industry today, could be applied for small current velocities. A second control objective principle was to measure the relative horizontal distance between the risers keeping it at a desired distance. This method showed promising results, but is dependent on measurements that are not easily available today. The best control objective used in the controller design was found to be equal eﬀective length, meaning that the sum of the payout and the actual riser length should be equal for all risers. Introducing this method may reduce the needed spacing between the risers and thereby reduce the wellbay area on the TLP. A simulation study with model based supervisory switched control showed how the diﬀerent controller structures and parameters could be included in the feedback loop depending on the operational conditions and riser behavior. The main contributions in this work are the mathematical modeling of the riser/TLP system, the model analysis, and the proposed controller architecture, including the control objectives and the supervisory switched control concept.
ii
Jon Refsnes and Per Ivar Barth Berntsen for fruitful discussions. Franz S. I deeply appreciate the guidance of and collaboration with Carl Martin. and Kari Unneland. Dr. Stølen for cooperation on the riser veriﬁcation in shallow waters which is included in this thesis. My funding has been provided by the Research Council of Norway through the Center for Ships and Ocean Structures (CeSOS) at NTNU and the Leiv Eiriksson mobility programme. I would like to thank Asgeir for his guidance and enthusiastic encouragement. Jamison Szwalek for proofreading. e and Simen F. The work has primarily been carried out at the Norwegian University of Science and Technology (NTNU) and partly at Massachusetts Institute of Technology (MIT). Another thanks to the MSc students who have suﬀered under my supervision. Andr´ H. August 2007 iii Anne Marthine Rustad . for always backing me up and providing advices both within research and life in general when needed. I am grateful to Professor Michael S. A special thanks to my parents and my sister Lene Annette for their love and support during these years with research.Acknowledgements This thesis represents the main results of my doctoral studies from August 2004 through August 2007. I would like to express my gratitude to my friends and family for their support and continuous motivation. Trondheim. Aronsen at MIT. Dr. Sørensen and Professor Carl Martin Larsen. Hover for their hospitality and advices during my research stay at MIT from October 2005 through April 2006. supervised by Professor Asgeir J. I would like to thank him for always having his door open for discussions throughout this project. Finally. His expertise on modeling and understanding of the physical system has been of invaluable help and inspiration. Jacobsen for working on TLP modeling. The work of this thesis has been a true joint project. Susan Swithenbank and Dr. I would also like to thank my colleges at CeSOS and the Department of Marine Technology in particular the administration by Sigrid B. Wroldsen at NTNU and Kristoﬀer H. Wold and Marianne Kjøl˚ for organias zation. I appreciate the company of my oﬃcemates Anders S. Triantafyllou and Dr. Elizabeth Passano should be thanked for her patience in questions concerning RIFLEX.
.
Notation Abbreviations ACPM CeSOS CFD CPM DNV DOF DP ESM FCPM FEM FPSO GoM LF PDE PPM R1. Bold types are used exclusively to denote vectors and matrices. R2 RC ROC SCR TLP VIV WF WIO Accurate control plant model Center for Ships and Ocean Structures. Computational ﬂuid dynamics Control plant model Det Norske Veritas Degree of freedom Dynamic positioning Error state maneuvering Fast control plant model Finite element method Floating production storage oﬄoading Gulf of Mexico Low frequency Partial diﬀerential equation Process plant model Riser 1. riser 2 Riser characteristics Riser operational conditions Steel catenary riser Tension leg platform Vortex induced vibrations Wave frequency Wake induced oscillations Characters The Roman and Greek letters most frequently used throughout the thesis are given here. Bold uppercase denotes matrices and bold lowercase denotes matrices for one riser v .
Roman A. fint K.to f frame Top tension of riser j Period of the harmonic TLP motion Current velocity in node i Horizontal riser position of node i TLP surge position Vertical riser position of node i Eﬀective weight vi . Ptop r Rf . De . lr li lt M. KP l. external and internal riser area Amplitude of the harmonic TLP motion System and element damping matrices Drag coeﬃcient Inertia and added mass coeﬃcients Diameter. ci CD CM .Notation element or vectors. Cm D. Ai AT LP C. N Pi . KI .x g f Ti Tj TT LP vi xi xT LP zi wef f Crosssectional. internal diameter Young’s modulus of elasticity External and internal force vectors System and element stiﬀness matrices Controller gains Riser length Riser element length Tendon length System and element mass matrices Number of elements Eﬀective axial tension in element i and the applied top tension Riser position vector Rotation matrices Transformation matrix from i. Ry. mi n. ki KD . external diameter. Ae . Di E fext . All symbols used in the speciﬁc contexts are explained when ﬁrst introduced.
ρw σ τ θi ξ Diﬀerence Norm function Monitoring error Process switching signal Density of the internal ﬂuid. steel and water Controller switching signal Controller contribution Inclination of element i Payout vii . ρs .Greek ∆ γ µ ρ ρf .
.
. 13 2. . .5 Riser FEM Modeling . . . . . 3. . . . . . . . . . 1. . . 10 . . . . . . . . . . . . . . . . . . . . . . 2 Tension Leg Platform 13 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4. . . . . . . . . . . . 3. . . . . 3.4.3 Vortex Induced Vibrations . 1 . . .1 TLP Surge Modeling . . . . . . . . 3. . . 1. . 16 3 Mathematical Modeling 3. . . . . . . . . . . . .3 Riser Stroke Calculations . . . . . . . . . . . . . . . . . . . . . . . . .3 Main Contributions . . . . . . . . . . . . . . .3 Existing Platforms . . . . 3. . . . . . . . . . . . . . . . . . 1. . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Organization of the Thesis i iii v ix 1 . . . . . . . . . . . . . .1 Background for the TLP Solution . . . . . . . . . .1 Motivation . . . . . . . . . . . . . . . .1 Shielding Eﬀects . . . . 3. . .3 Kinematics and Coordinate Systems 3. . . . . . . . 9 . . . . . . . . . . . . . . . . . . .2 Hydrodynamic Interaction . . . . . . .2 Wake Induced Oscillation . . . . . . . . . . . . .1 Current Models . .2 Previous Work . . .2 The TLP Concept . . . . . . . . . . . . . . . . . . . . .2. ix . . . . . 3. . . . . . . . . . . . . 4 . . . . . . . . 3. . . . 15 2. . . . . . . . . . .2. . . . . . . . . .Contents Abstract Acknowledgements Notation Contents 1 Introduction 1. . . 23 24 27 28 28 29 30 31 32 33 33 35 . . . . 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . .2 Kinematics . . . . . . . . .4. .4 Tension Leg Platform . . . . . . . . . . . .
. .5. . . . . . 4. . . 3. . . . . . . .1 SetUp . . . . . .3 Riser Control System Overview Implementation Overview . . 5. . . . . . . . . . . . . . . . . . . 3. . . . . . . . 4. . . . . . . . . . . 6 The Controller Architecture 6. . . .5. . . .1 Increasing TLP Oﬀset . . . . . . .7. . . . . . . . . . . . . . .3. . . 3. . . .1 Transformations for the Riser Elements 3. .1. . . . . .3 System Stiﬀness Matrix . . .6 Load and Equilibrium Iteration . . .1. . . . . . . . . . . . . . . . . . . . . .2 Deep Waters . . . . . . . . . . .3. .5. . . Actuator . . . . 4. . . 4. .3. . . . . . .5. . . . 6. . . . . . . . . . . . 4. . . . . . . . . . .5 Hydrodynamic Forces . 5. . . . . . . 5. . . . .6 3. . .2 Control Objectives Based in Measurements Along the Riser 6. . . . . .5.6. . . . . . . . . .4 . .4 Veriﬁcation with Dynamically Moving TLP . . .6 QuasiStatic Veriﬁcation at Shallow Waters . . . . 4.5. . . . . . . . . . . . . . . . . . . .5 Relative Horizontal Distance Controller . . . . . . . . . . .1 Current . .1 5. . . . .2 Increasing Tension . . . . . . . . . 5. . . . . . . . . . . . .4 Top and Bottom Angles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Measurements . . . .7. . . . . 5. . . .3 Discussion . . . . . . . . . . . . . . 5. . . . . . . . .8 Dynamic Equation of Motion . . . . . . . . . . . . .9 QuasiStatic Equation of Motion . . . . . . . . . . . . . .5. . . . . . . 3. . . . . . . . 4. . . . . . . . . . . .1 Control Objectives . .8 Discussion . . . . . .5 Veriﬁcation with Dynamically Varying Tension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . .2 QuasiStatic Veriﬁcation with Increasing TLP Oﬀset . . . . . . . . . . . . . . . . . . .3 Tension and Payout .2 TLP Motions . . . . .1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3. . . . .5. . . . . . . . . .3. . . . . . . .6. . . . .Contents 3. . . . . . . . . . . . . . . . . . . .7 QuasiStatic Veriﬁcation with Stress Joint . . . . . 4. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .7 TLP Prescribed Forces . . . . . . 3. 5 The 5. . . . . . . . . . . . . . . . . . . . Actuator and Constraints . . 4. . . . x . . . 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 QuasiStatic Veriﬁcation with Increasing Top Tension 4. .5. . 36 37 39 39 40 41 42 42 43 43 47 48 49 49 54 55 58 58 59 60 61 63 63 67 67 68 69 70 72 72 72 73 73 77 77 77 80 80 4 Model Veriﬁcation 4.1 Shallow Water . . . . . . . . . . . 3. . . . . . . . .2 5. . .2 System Mass Matrix . .4 Structural Damping . . . . . . . . . . . . 4. . .1 Control Objectives Based on Measurements at the Top . . . . . . . . . . . . . . . . .
2. .5 Integrator AntiWindup . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Dynamically Moving TLP . . . . . . . . . . . . . . . . . . . . . . .1 System Properties . 7.2 Reference Model . . . . . . 6. . . . .3 ScaleIndependent Hysteresis Switching 6. . . . . . . . . . . . . . . . . . . . .1 Constant Equal Tension . . . .4. . . . . 8 Simulation Results 8. . . . . 6. . 6. . . . . . . . . . . . .2.3 Analysis of QuasiStatic Tension Variations . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . 6. . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4 Model Concept Deﬁnitions . . . 6. 81 81 83 84 85 85 86 87 88 89 90 90 94 95 95 96 97 99 99 101 101 102 103 106 107 110 111 112 118 119 122 125 125 126 126 127 129 133 135 xi 7 Control Plant Model Analysis 7. . . . . . . . . . . 7. . . Switched Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .6. . . . . . . . . . . . . . . . . . .2 Switching Logic . . . . . . .3 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . 7. . . . . . . . . . .1 Accurate Control Plant Model . . . . . 8. . . . . . . . . . . .4 Desired Horizontal Distance 8. . .5. . . . . . . . .2. . . . . . . . . . Controller Set . . . . . . . . . 6. 8. . . . . . . . . . . 6. . . . . . . . . . . . . . . . . . . . . 6. . . .1 Riser Characteristics . . . . . . . . . . . . .2.6 Discussion . 7.2 6. . . . . . . . .5 Riser Operational Conditions . . . . . . . . . .2 QuasiStatic Analysis with Increasing TLP Position 7. . . .3. . . . 8. .1 Analysis Input Data and SetUp . .2. . . . . . . . . . . . . .2. . . 7. . . . . . . . . . . . . . . . . . .3 Equal Eﬀective Length . . . . . .4. . . . . .2. .5 Switched System Stability Analysis . . . . . . . . . . . . . . .5 Discussion . . . . . . . . . . . .1 Set Up . . . . . . . . 6. .5.Concept and Properties . . .2 Control Objectives . . Supervisor . .5 Analysis of Dynamic Variation in Top Tension . .3. . . . . 7. . . . . .5. . . . . . . . .3 Eﬀect of Shallow Water . . . . . . 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3.2. . . . . . . . . . .1 Error Norms for Horizontal Positions . . . . . . . . . . . . . . . . . . . . .3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5. . . . . . . . . . . . . . . . . . .4 Payout . 8.5. . . . . . . . .3 Area Under Curve .3 Switching Logic . .2 Riser Operational Conditions . . . . . . . . . 7. . . . . .1 Guidance . .2 Equal Payout . . . . . . . . . . . . . . . . . . 6. . . . . . . . . . . . . . . . 7. 6. . 7. . . . .4 6. .2. . . . . . . . . 6. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. . . . . . . . . . . .4. .3 Controllers . . . . . . . . . . . . . . . . 8. . .4 Transition . . . . . . .2 Riser Top Angle . . . . . . . . . .2. . . . . . . . . . .2 Monitor . . . . . . . . . . . .
. . . A. . . .2 System Matrices . . . . . . . . . . . . . . . . . . . . . . . . . . .2. . . . . . . . . . . . . . .2 Tide . . . . . . . . . . . . . . A. . . . . . . A. . . . . . . . .1 Current . . . . . . . . . . . . 137 140 140 141 142 146 9 Concluding Remarks 149 9. . . . . . . . . .5 Metrical Norms . . . . . . . . . . . . . .2. . . . . . . . . . . . . . . . . . . . .1 Case SetUp . . . . . . . . . . . . . . . . . . . . . . . Supervisory Switched Controller . . . . . . . .1 Wake Field of a Single Cylinder . . . . . . . . . . . . . . . . . . . . . . . . . . B. . . .2. . . . . . . . . . 153 165 165 167 167 168 169 170 174 180 183 183 183 184 184 185 185 xii . . . . . . . . . . . . . . . . . . . . . .4 8. . . . . . .1. . . . . . . . . . . . . .5. . . . . . . . . . . . . . . .2 Riser Data . . . . . . . . . . . . A. . . . . . . . . . . . . 8. . . . .1. . . . . . . . . . . .3 Damping . . . . . . B. . . . . . . . . . . . . A. . . . . . . .1 Mass . . . . .4 QuasiStatic Algorithm . . . . . . . B Simulation Data B. .2 Proposals for Further Work . . . . . . . . . . . . . . . . . . . . . . .1 Conclusion . . . . . . . . 149 9. . . .5. . . . . .1. . . . . . . . . . . . . . . . . . . . .3 Load on the Risers . . . . . . . . . . . . . .3 TLP Oﬀset and B. . .6 TLP Dynamics . .5 8. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . A. . . . . . . . . . . . . . . . . . . . . B. . B. . . .5. . . . . . . . . . . . . .3 Simulations with Control . A. . . . . . . . . . . .2 Stiﬀness . . . Dynamics . . . . . . . . . . . . . 151 Bibliography A Mathematics and Algorithm for the FEM Model A. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .3 Controller Gains . . . . . . . . . . . . .2 Simulations without Control 8. . . . Discussion . .1 Environmental Data . . . . .Contents 8. 8. . .
knowledge and experience have developed. 2007). New contributions are searched for within each discipline. Since then. being more than 50% of the export value in 2006 (Ministry of Petroleum and Energy and the Norwegian Petroleum Directorate. However. the disciplines of hydrodynamics. New approaches will hopefully result in new solutions to currently unsolved problems. but also opens for new possibilities. structural mechanics.1 with a cross marking where this thesis can be found disciplinewise. automatic control. As these reservoirs are exploited together with an increasing demand for oil and gas.Chapter 1 Introduction Oil and gas are by far Norway’s largest export industries. material science. and the petroleum industry in Norway presently has expertise on oﬀshore installations in harsh environment and deep water. 1. further progress can be made. It is believed that by integrating the diﬀerent disciplines. etc. the petroleum industry has been moving to ever deeper waters. but also some hydrodynamics are included. At the Center for Ships and Ocean Structures (CeSOS) at the Norwegian University of Science and Technology (NTNU). Riser technology is an 1 . as new oilﬁelds are explored and developed. This thesis is situated at the intersection between structural mechanics and automatic control. oceanography.1 Motivation Oﬀshore petroleum production began in the most shallow waters. Working in a multidiscipline ﬁeld creates new problems. structural mechanics and automatic control are integrated as a strategy to contribute to the innovation of ships and ocean structures. The ﬁrst oil in Norwegian territory was found in 1969 at the Ekoﬁsk ﬁeld. located at 7075m water depth. which is based on disciplines like hydrodynamics. new challenges arise and new knowledge and research are needed. 1. The ﬁrst production started there in 1972. This is illustrated in Fig.
In deeper waters interference between adjacent top tensioned risers in an array is an issue of considerable concern (DNV. 2005). For deep water production systems. This has inﬂuence on the global platform parameters like deck space and load carrying capacity. the number of collisions by applying control of the riser top tension. Introduction Automatic Control Structural Mechanics Hydro dynamics Figure 1. The main objective of this work is to prevent. Collision may lead to dents in the riser pipe and also damage in the coating. (1) subsea wellheads (wet trees) with ﬂexible risers up to a ﬂoater like a semisubmersible or a production ship (ﬂoating production 2 . In deep waters this means that even a relatively small diﬀerence in static forces may lead to mechanical contact. Increasing the riser spacing. There are mainly two design parameters that will prevent collision between risers: 1. In addition ﬂow separation and shielding eﬀects between risers in an array can change the local ﬂow velocity. 1993). If the riser spacing and properties are kept constant. Even a single collision event may be damaging if the collision takes place with suﬃciently high impact. with fatigue and corrosion as possible consequences. The other option is increasing the top tension to a high and constant level.1: Intersection of the diﬀerent disciplines found in this work. This diﬀerence in current forces may cause large relative motions and lead to contact between neighboring risers. 1. Increasing the top tension. Increasing the riser spacing means increasing the size of the wellbay (Fig. the risk of collision will increase with increasing water depth.1. riser solutions are traditionally divided into two main groups. which will increase the wear and tear on the cylinders in a heave compensation system. since the static deﬂection due to the uniform current drag is proportional to the square of the length (Huse. or at least reduce. Both may result in signiﬁcant cost penalties.2). 2. important issue both when considering ﬁeld development costs and technological feasibility.
Top tensioned risers operated from spars and TLPs are arranged in clusters of (near) vertical riser arrays. cylindrical buoyant platform. like tension leg platform (TLP).FPSO). usually manned.Motivation Figure 1. intervention and workover. (2001) for deep draft caisson vessel with buoyancy cans. Wanvik and Koos (2000) present a two tier well riser tension system. This system separates the low frequency stroke motion from the wave induced stroke motion. export. etc. Dry tree systems are often the preferred solution for production as they provide easy access to the well for maintenance. Pollack et al. which may consist of diﬀerent risers applied for production. This solution allows long stroke for spread moored shipshaped vessels. workover. Alternative platform solutions. The number of individual risers in an array may be 20 or more. while still using dry tree systems. are proposed by Often (2000) for a semisubmersible with heave compensation system. spar1 or deep draft ﬂoater (DDF). These are the dry tree solutions for ﬂoating production systems. drilling. and (2) tensioned risers with wellhead on a compliant platform. The problem of riser collision can be restricted to tensioned risers operated from a TLP or a spar located in deep waters. This is due to the small space available for the high number of risers. in addition to the increased 1 A spar is a vertical. (2000) suggest a weight based tension leg deck (TLD) to which the risers are locked for a dry tree solution on FPSOs. and by Mortazavi et al. storage oﬄoading . 3 . which is a mechanical alternative to hydraulic cylinders working in series.com).statoil.2: The wellbay on the TLP Snorre (www.
Little or nothing has been done to actively control the riser motions. and forced ﬂoater motions. Direct wave loads on the upper part of the riser and ﬁrst order ﬂoater motions are assumed insigniﬁcant for such heave compensation systems. For a reservoir with heavier hydrocarbons this solution may not be possible. but there are also some west of Africa and oﬀ the coast of Brazil. but it is unknown if the ﬁeld is commercial viable. Each top tensioned riser in an array is exposed to environmental loads from waves. if it is decided to develop the ﬁeld. 1. and the deﬁnitions given in the recommended practice will be followed here. Ellida is a large oilﬁeld 60km north of Ormen Lange and located at 1200m water depth. Little information regarding interaction eﬀects due to wave loading is found 4 . environmental loads. because the collision is most likely to occur in the midsection of the riser. Hence. A TLP solution was considered at the Ormen Lange ﬁeld 100km west of midNorway at 850m water depth. ﬂoater oﬀset. Introduction deﬂection for long risers. both at wave frequency (WF) and low frequency (LF). The ﬁeld was discovered in 1997. DNV (2005) gives a thorough introduction to riser interference. the demand for stroke capacity will be much higher. which means that other riser solutions like steel catenary risers (SCR) or ﬂexible risers are preferred. The main reason is that tensioned risers can be applied at a TLP with a relatively small requirement for stroke capacity even at large TLP oﬀsets. and riser properties to mention some. For other ﬂoaters like spars. currents. a TLP could be a possible solution here. Deep water TLPs are mainly found in the Gulf of Mexico (GoM). hydrodynamic interaction. In addition some work has been done on the actual collision forces and the damage it may cause. where the deﬂection is largest due to current loads. Dynamic behavior and collision of risers have been studied by several scientists the last decade.1. This solution is possible for a reservoir mainly consisting of gas. Whether or not collision between two neighboring risers occurs depends on factors like top tension. the control system presented in this thesis is most likely to be applied on a TLP. FPSOs and semisubmersibles.2 Previous Work Most research within the ﬁeld of collision between risers focuses on the hydrodynamic interaction between them. but a system with subsea wells and pipelines to the onshore plant was chosen. This is a consequence of the geometric restrictions for heave motions caused by the axially rigid tendons. The assessment of riser interference is therefore mainly based upon the assumption of a steady state current proﬁle. riser spacing. However. A TLP is chosen as the platform concept for this work.
These eﬀects are (DNV. (2003) have shown that interaction between neighboring risers will not have any hydrodynamic inﬂuence on the upstream riser beyond a certain distance. Bokaian and Geoola (1985).Previous Work (Duggal and Niedzwecki. Furthermore. The mean lift force is directed towards the wake centerline. The parametric wake ﬁeld model is a semiempirical static wake formulation which accounts for the interaction between stationary individual cylinders in steady current. Note that within the ﬁeld of hydrodynamics. and is based on the analytic expression of a turbulent wake by Schlichting (1968). the term riser is often used. etc. Both can be applied in ﬁnite element method (FEM) models. cables. In this method the hydrodynamic forces are accounted for by introducing mean drag and lift forces on the downstream cylinder as a function of the distance from the upstream cylinder. The inﬂow of a cylinder situated in the wake of an upstream cylinder can be computed at any location. the literature distinguishes between three diﬀerent kinds of forces acting on the downstream riser in the wake of an upstream riser. the parametric wake ﬁeld model and the parametric mean force model. and an extensive list of references may be found here. For more industrial studies. 1993). Zdravkovich (1977) gives a careful review on ﬂow between two circular cylinders in various arrangements in steady ﬂow. tendons. • Wake induced oscillation (WIO) on the downstream riser. while signiﬁcant eﬀort has been applied to investigate hydrodynamic interaction in steady current. In Huse (1993). We will strive to use the terms in the same way as in each referred paper. 1993. cylinders is the general term used to denote marine risers. The drag and lift coeﬃcients in 5 . where the current velocity is smallest. A reference should also be made to Blevins (1994) and Zdravkovich (2003) for numerous experiments on hydrodynamic interaction between cylinders. Zdravkovich (2003) presents arrangements with three or more cylinders. • Vortex induced vibrations (VIV) resulting in ampliﬁed drag coeﬃcients on both risers. a scheme for calculating the inﬂow of an array of cylinders is proposed. 1993. but keeping the drag coeﬃcient constant. see Huse (1987. A mean transverse force is formulated for cylinders more arbitrarily placed in the wake. The parametric mean force model was derived by Blevins (1994). This was studied by Huse (1987. 1996). The mean forces are modeled by two diﬀerent methods. 1996). and recent studies by DNV (2003) and Kalleklev et al. (2000). 2005): • Mean force and shielding eﬀects bringing the risers closer together. Experiments by Tsahalis (1984). Huse and Kleiven (2000) and Kavanagh et al. The drag force is computed by taking the actual velocity into account.
which means that they are quasistatic and caused by unstable hydrodynamic forces. occurring about the ﬁrst natural mode of motion. 2003). large amplitude motions of very irregular behavior. unstable and critical). Blevins (2005) developed a model for steady lift and drag forces on a cylinder in the wake of an upstream one by using theoretically based equations ﬁtted to data from experiments.b) investigated the onset criteria for wake induced instability. The objective was to assess whether or not adjacent risers move in the wake of an upstream riser. These may cause slowly varying. Such motions may also occur at signiﬁcantly lower frequencies. Fontaine et al. In addition to mean forces. (2002). while VIV with higher frequencies and velocities accounted for most of the energy in a riser collision. DNV. References could be made to for instance Simpson (1971). The study of WIO dates back to the early 1970’s and the study of power transmission lines. Tsahalis (1984) also found that the VIV responses are diﬀerent for the upstream and downstream 6 . Introduction the parametric mean force model need to be established by model tests or by two dimensional numerical methods like computational ﬂuid dynamics (CFD). Most of these results are published in conﬁdential reports. 1984. and the ﬂow around the risers will cause large hydrodynamic interaction forces. Lately. The parametric mean force model is also able to capture WIO. these motions were ﬁrst observed and described by Huse (1996) and later by Wu et al. The purpose of both studies was to gain necessary information in order to assess the onset of potentially interference problems and collision. (1999) have investigated the drag and lift forces in steady current. and have not been accessible to the public. (1994) is related to Shell’s activity on deep water TLPs like Auger and Mars. Wu et al. The interest for tendon and riser interference in the work of Tsahalis (1984) and Gopalkrishnan et al. dynamic forces are described in the time domain using position dependent force parameters. The basis in this model was a parametric force representation for both upstream and downstream riser based on coeﬃcients found from CFD. WIO are in any case found at frequencies substantially lower than the vortex shedding frequency. (1994) and Wu et al. assuming two identical cylinders in steady uniform ﬂow. An ongoing research program has been in progress since 1989 (Allen et al. For marine applications. (2002) postulated that the slowly varying WIO controlled the relative position of the risers. note that the large riser displacement governed by mean force and WIO are found to be unaﬀected by VIV. (2002). 2005). The onset criteria for WIO and clashing between SCRs were proposed. (2007) reproduced WIO in model scale and identiﬁed the diﬀerent regimes (stable. Sagatun et al. Gopalkrishnan et al. (2001a.1. A slightly diﬀerent approach on the parametric wake force model was proposed by Sagatun et al. The risers are exposed to current over a very long length relative to the diameter. Price (1975) and Tsui (1977). except from the magniﬁed drag force (Tsahalis.. However.
The work includes methodology for contact dynamics of interacting risers and recorded riser impacts. (2002) have used the method from Sagatun et al. Fard (2000) investigated vibration control of ﬂexible mechanical systems (i. Signiﬁcant drag reduction on the downstream cylinders was found. see Sarpkaya and Shoaﬀ (1978) and Vandiver (1983). (2000) did VIV model tests and outlined how riser interference for deep water risers was assessed using a combination of wake ﬂow modeling. (1999) and Neto et al. the industry is considering accepting that riser collision occurs. As keeping the risers clustered in an array with small riser spacing is an advantage with respect to both economy and operation. Kalleklev et al. Nyg˚ ard et al. The VIV response for the downstream riser is given by the mean water velocity of the riser. 2005). Kavanagh et al. The result indicated acceptable stress levels and served as a ﬁrst step to allow some impacts to occur. For the ampliﬁed drag coeﬃcient. (2004) examined the wake shielding for three risers in a towing tank. Little has been done on experiments for more than two risers. The impact of collision and participating mass for colliding risers is also treated in Sagatun et al. Both bare risers and risers with VIV suppression (strakes) or bumpers along the collision zone were tested. Herfjord et al. (2002). (2000) and Halse (2000). (2002) presented a numerical model for the simulation of the interaction and clashing of ﬂexible risers. During the last decade. However. Leira et al.e. including relative velocity at the time of collision which is a measure on its intensity. (2001) investigated the stress levels generated by the impact between two neighboring risers. which is diﬀerent from the velocity on the upstream riser due to the wake eﬀect. global interference analysis and VIV prediction. (2005) did model tests to better understand the mechanics that drive the risers to collision. and found that passive controllers could be applied for dynamic line tensioning in a mooring system. VIV on marine risers) with focus on passivity of nonlinear beams. Huang et al.Previous Work risers. These measured results provided benchmarks for code validation of riser collisions. collision statistics and stress analysis. Allen et al. (2003) proposed a design practice that properly accounts for collision between risers in operating conditions typical for major deep water locations. Baarholm et al. with focus on VIV and Reynolds number. some doctoral theses on modeling and control of cables have been written at NTNU. Lately. T¨rkyilmaz (2004) modeled a towed u 7 . The design practice of today does not allow collision neither under normal nor extreme conditions (DNV. (2002) in a case study to estimate extreme response and fatigue damage for colliding risers. (2005) investigated the riser interference on ﬂexible risers. Aamo (2002) applied FEM to model and analyze passivity properties of mooring lines. see for instance Kaasen et al. VIV is in itself subject to research as it is important for fatigue as well as to calculate drag forces on the riser.
Johansen et al.1. (2000). Suzuki et al. Only position measurements at the bottom riser end and the control action at the top are needed to follow the reference trajectory. and give a fast payback of the investment. (2002). The main objectives were to minimize the bending stiﬀness along the riser and the riser angles at the top and bottom joints. and should not exceed an upper limit of 24 degrees during drilling. The requirement for redundant positioning references motivated also Høklie et al. including experiments on drag and lift forces on an array of risers. by separating axial and transverse motion. actuators like thrusters or remote operated vehicles (ROV) were suggested to be used along the riser. Fredheim (2005) worked on current forces on net structures. They estimated vessel positioning by measuring the riser angles and 8 . This method is patented. Improving the quality of position measurements will increase the availability of the drilling vessel in bad weather. To control the riser angles. Ersdal (2004) developed a FEM model for towed ﬂexible cylinders and investigated the hydrodynamic forces on cylinders in axial ﬂow through experiments. Introduction seismic cable and investigated its passivity properties. The objectives were (1) to minimize the REA by vessel positioning and (2) to perform ship position based on REA measurements with riser angle sensor and neural network. either satellite based or hydroacoustics with ﬁxed reference systems on the seabed. Lately. To achieve this. Ship positioning sensors were not used as the proposed system should act as a backup for commercial available positioning systems. A reduced model for the behavior of the bottom riser end and a control law were developed. The design principle of a riser angle positioning reference system (RAPR) with an observer model is proposed. (2002) were also motivated to use riser angle as a reference system for drilling risers. The riser model is a quasistatic FEM model and accounts for varying current loads and top tension. (2006). Control of risers is in its infancy and little is published. (2006) presented a new model for dynamics of inextensible cables. The use of riser angle positioning system (RAPS) was ﬁrst investigated by Dean (1980). the top and bottom riser angles are of concern. Wroldsen (2007) has been working on cable models based on diﬀerential algebraic equations for eﬃcient computation and realtime purposes. see Sabri et al. in addition to the existing systems. (1995) used modal expressions for a riser at 4000m water depth with current loads in the upper 1000m. (2001) proposed an optimal setpoint chasing algorithm for drilling vessels in deep waters. The DP system for drilling vessels with automatic control of the riser end angle (REA) was proposed by Imakita et al. An exception is the work of Sabri et al. For drilling operations. the position of the drilling vessel is optimized by including the inclination measurements in the feedback loop for the dynamic positioning (DP) system. Sørensen et al. Sørensen et al. The system was successfully tested aboard the drillship Discoverer Seven Seas in 1979. (2003) who use active control to make it possible to connect a riser to a wellhead in harsh weather.
All risers were connected to a common frame in the top end. for instance by Aker Kværner (2007) and National Oilwell Varco (2007a). A dynamic mathematical model is tuned by combining the vessel position measurement and riser inclinations and use this to estimate the current proﬁle. The top tension is today kept constant by passive systems. Control of risers by changing the top tension has not previously been published. But instead of keeping the risers ﬁxed to a plate.3 Main Contributions The main contributions found in this thesis are summarized here. and no collision could be observed in steady current. This was ﬁrst shown in Rustad et al. References to these publications are given below. • A twodimensional model for the TLP and riser system exposed to current is developed. the top tension of each riser is controlled individually using the payout of the heave compensator as the measured input. Recently Perez and Steinmann (2007) performed an analysis of operability and constraints in terms of vessel motions and heave compensator stroke capacity. The system was initially tested with success aboard the drill ship Saipem 1000 west of Africa at 1100m and 2300m water depth in June 2002. The ﬁrst is commercially available. the work in this thesis has shown that it is possible to prevent collision between neighboring risers by use of top tension control and measurements at the riser end and wellhead onboard the TLP. (2003) made model experiments where the strategy was to synchronize the load with the waves in the moon pool. the model is able to estimate the vessel position based on the riser inclination measurements and the estimated current proﬁle. (2006). Active control of a heave compensated crane during the water entry phase for the equipment is addressed by Sagatun (2002). The work in this thesis is motivated by the work of Huse and Kleiven (2000). and later in Rustad et al. Johansen et al.c). After the initial tuning. heave compensation of top tensioned production risers has not earlier been investigated. 1. 9 . Korde (1998) investigated an actively controlled heave compensation aboard deep water drillships subjected to irregular waves. However. • Firstly. (2007b. How the speciﬁed motion from the TLP acts on the risers is not earlier published.Main Contributions using Kalman ﬁltering technics. (2007c). In order to avoid collisions Huse and Kleiven (2000) proposed a strategy based on equal payout for risers in an array. Parts of this thesis have earlier been published. Active heave compensation is used today under drilling and in crane operations for oﬀshore equipment. This modeling is found in Chapter 3 and is published in Rustad et al.
5. This case study is also presented in Rustad et al. exposed to TLP motions in the top end and current forces. respectively.1. The most famous existing platforms are presented. These include two existing objectives. Chapter 7 analyses the control plant model thoroughly. and could to a large extent also be found in Rustad et al. Chapter 3 presents the mathematical modeling of the system. 1. Chapter 6 presents the controller architecture. but can be expanded to three dimensions and a matrix of risers.c). The validation of the model for diﬀerent water depths and bottom end connections is investigated. • A model based.1 and 8. In addition. This includes the control objectives. (2007b.. 2005) in Chapter 4. Chapter 5 gives an overview of the physical implementation of the system being the actuator.4 and 8. Introduction • The mathematical system model is veriﬁed by the commercial software RIFLEX (Fylling et al. riser operational conditions and the diﬀerent controllers. supervisory switched control system is proposed and simulated in Sections 6. These are presented and investigated in Chapters 6. The control objectives are presented for two risers in a tandem arrangement. Parts of this are found in Rustad et al. • Four diﬀerent riser control objectives are formulated. This has not been published earlier. being the actuator in the system is also modeled. one used in the industry today and one earlier proposed by Huse and Kleiven (2000).5. respectively. An extensive analysis on how many elements are needed for the various applications and purposes is found in Chapter 7.4 Organization of the Thesis The outline of the thesis is as follows: Chapter 2 introduces the TLP concept and the background for the TLP solution. 10 . Chapter 4 veriﬁes the riser model from Section 3. The riser tensioner system. (2007c). Both the quasistatic and dynamic models are simulated with varying TLP position and top tension. two new control objectives are proposed here. (2007a). This consists of two risers in a tandem arrangement. measurement and controller system design.
These includes investigation of the different control objectives and a case study with supervisory switched control. not included in Chapter 3. riser data and controller gains used in the simulations in Chapter 8.Organization of the Thesis Chapter 8 shows the simulation results. Chapter 9 summarizes the main conclusions of this thesis and proposes further work. 11 . Appendix B gives the environmental data. Appendix A includes additional details of the mathematical modeling.
.
This was done for the case of simplicity.and is neglected in this work. Faltinsen (1990) and Larsen (1995). Heave motions of a TLP have two components. while heave relative to vertical motions of the upper riser end is subjected to active control. wave and current by generating large internal reaction forces. The ﬁrst oﬀshore oil and gas ﬁelds were found in shallow waters. The other component is referred to as surge induced heave. Fixed platforms will normally have natural periods shorter than 5s and their responses which are caused by extreme waves will therefore be quasistatic since such waves 13 . Little or no motions are observed for such structures. and mainly ﬁxed structures were build in the ﬁrst decades of oﬀshore oil production. This chapter is mainly based on Demirbilek (1989b). Conclusions from this study are hence valid even if this simpliﬁcation was made. One is cause by dynamic elastic strain and deﬂections of the tendons.in particular in extreme wave conditions . but could have been included without any conceptual changes in the approach. The motions caused by local tendon dynamics are signiﬁcantly smaller than the surge induced component .1 Background for the TLP Solution Oﬀshore platform concepts are usually classiﬁed into two major categories. The ﬁxed platforms resist the environmental forces like wind. 2. The surge motions will act as a prescribed dynamic boundary condition in the riser analysis.Chapter 2 Tension Leg Platform The main purpose of the TLP model in this thesis is to represent the platform motions at the wellhead area where the risers are linked to the platform. Fixed platforms stand at the seabed and remain in place by a combination of their weight and/or piles driven into the soil. ﬁxed and compliant. and is easily understood by realizing that the inclined tendon will have another vertical distance from the seaﬂoor than its true length. which will inﬂuence the true distance between its ends.
the choice of platform system depends upon other considerations. modulus of elasticity and length of the tendons decides outofplane stiﬀness. In addition to considering the sensitivity of the platform to external environmental forces. especially that of the TLP. 1989). With increasing water depths. The design premises for the tendon system are hence to determine these parameters so that desired values of the eigenfrequencies can be obtained. The cost of the ﬂoating system. ﬁxed structures had to be designed to be stiﬀer. Any anchored vessel. ﬁxed platforms become more ﬂexible. The governing parameters for the inplane stiﬀness are the tension and length of the tendons. Hence. However. requiring more steel and exponentially increasing costs. Horizontal motions of the deck are compliant. the application of conventional ﬁxed jackets approached its limits principally imposed by the dynamic behavior of the structures (Litton. but resonance may still occur due to wind and higher order wave forces. including water depth. A TLP acts like a ﬂoater with regard to inplane motion components (surge. reservoir size. dynamics of rigid body motions for ﬂoaters are of concern. A ﬂoating platform will have all its structural eigenperiods well below 5s. Towers are ﬁxed to the seaﬂoor by an arrangement that eliminates. which means that ordinary wave loads will not give any structural dynamic response. roll and pitch). production rate. or at least reduces. However. semisubmersible. 14 . Amongst these are the technical and economic factors. service life and removal requirements. Eigenperiods are controlled by the geometry of the waterplane (heave. but vertical motion components are equivalent to a ﬁxed platform. the bending moment at the bottom end. Hydrodynamic damping is the key to reduce these types of response. while the cross section area. but like a ﬁxed platform for the outofplane components (heave. towers and TLPs. Compliant platforms may be divided into three types. To keep the eigenperiods away from this damaging range. sway and yaw). economic factors make the compliant platforms and the TLPs in particular one of the leading candidates for major deep water developments. are relatively insensitive to changes in the water depths compared to the cost of the compliant towers and ﬁxed structures (Litton. The basic idea is to allow for rigid body motions with eigenperiods longer than wave periods (T > 30s). and their natural periods started to enter the high energy levels of the ocean waves. spar and FPSO belong to this category. These eigenperiods are normally above wave periods. The obvious alternative to ﬁxed platforms in deep water is to use compliant platforms. in deep waters the major factor in selecting the platform for a ﬁeld development plan is the cost which is correlated to its weight.2. Tension Leg Platform typically will have periods above 10s. 1989). pitch and roll) and design of the anchor system (surge. ﬂoaters. sway and yaw). As oﬀshore development moved towards deeper waters.
It is a multilevel facility consisting of trusses. the largest environmental forces can be balanced by inertia forces instead of by forces in rigid structural members. but might also be classiﬁed as a moored structure. heave. 15 . Drilling and production risers connecting the platform to the wellhead template on the seaﬂoor are in general not a part of the TLP mooring system (Demirbilek. • The foundation is found at the seabed and consists of templates and piles. wellhead and the subsea well templates. the TLP is similar to other column stabilized moored platforms with one exception. 2.1. • The deck structure supports operational loads. deep girders and deck beams. • The mooring system consists of the tendons and the foundation. or even a gravity system. 1995).2 The TLP Concept The basic idea behind the TLP concept was to make a platform that is partly compliant and partly rigid. The horizontal forces due to waves on the vertical cylinders will always be larger in the horizontal plane than in the vertical direction. In order to maintain a steel riser connection between the seaﬂoor and the production equipment on the platform. In general. • The platform consists of the hull and deck structure. These taut mooring are called tension legs. • The well system includes ﬂowlines. The TLP is deﬁned as a compliant structure. • The tendons connects the platform to the foundation at the seabed. and thus the vertical equilibrium of the platforms requires taut moorings connecting the upper structure to a foundation at the seabed. tethers or tendons. risers. by making inplane motions compliant. production and export risers. This idea can be realized by a pendulum using buoyancy to reverse gravity forces (Larsen. the buoyancy of a TLP exceeds its weight. pontoons and the intermediate structure bracings. 1989a). Hence. 2. • The risers include drilling.The TLP Concept especially when a dry tree solution is preferred. riser tensioners. and deﬁned by the American Petroleum Institute (1987) as follows: • The hull consists of the buoyant columns. roll and pitch motions had to be minimized and are rigid degrees of freedom. The diﬀerent parts of the TLP are illustrated in Fig.
The others are found in the Gulf of Mexico. in addition to all deck equipment and the hull system. on the west coast of Africa and oﬀ the coast of Brazil.1: The layout of the Brutus TLP in the GoM (www. Below follows a list with short descriptions of the some of the most famous platforms. During the next two decades.3 Existing Platforms Interest in TLPs dates back to 1960 and many studies have examined the applicability of this concept for deep water developments (Demirbilek. Tension Leg Platform Figure 2. 2. The increase in water depth for TLPs is illustrated in Fig. There are at the moment installed approximately ﬁfteen TLPs. the TLP concept began attracting more attention from the oﬀshore industry as an appropriate structure for deep water applications.com). • The tension leg platform includes all the above.2. 2.oﬀshoretechnology. 1989a). The deepest installed TLP today is the Magnolia platform. of which three are found in the North Sea. 16 . and especially after the installation of the Hutton TLP in 1984.2.
and the production started in 1992 (Ministry of Petroleum and Energy and the Norwegian Petroleum Directorate.Existing Platforms Hutton Jolliet Auger Mars RamPowell Ursa Magnolia 0 Water depth [m] 500 1000 1500 1984 1989 1994 1996 1997 1999 2003 Installation year Figure 2. Hutton (UK) The Hutton TLP was the ﬁrst of its kind and was installed by Conoco in the British sector of the North Sea in 1984. These tendons have large submerged weight which will contribute to unwanted deﬂections when the tendons are inclined (Larsen. but the TLP solution was chosen to gain experience with this concept for future applications on larger water depths. production. The installation serves as a combined drilling/wellhead. The tendons are made of thickwalled tubes with conical threaded couplings. The platform is an in17 . A conventional steel jacket would be a less costly alternative. 2007).8m in diameter. The Snorre TLP is moored to the seabed by sixteen steel tendons 0. Snorre (Norway) The Snorre ﬁeld lies in the central North Sea approximately 200km west of Florø in Norway at 310m water depth. 2007).2: The development of TLPs and riser depths. Snorre was developed in two phases and the production facilities comprise a steel TLP on the southern part of the ﬁeld and a subsea production system tied back to the TLP (Oﬀshore Tecnology. It was installed in 1988 by Saga Petroleum. and quarter platform on a medium sized oil and gas ﬁeld at 150m water depth. 36 wells are drilled from the TLP. 1995).
Auger (USA) By the Auger platform. Jolliet (USA) The Jolliet platform was the ﬁrst real deep water TLP and was installed by Conoco on the Green Canyon ﬁeld in the GoM at 536m water depth in 1989.com).3 shows a picture of the Snorre TLP. production and living quarter installation. tegrated drilling. 1995). and the tendons have larger diameter and smaller wall thickness (Larsen. Tension Leg Platform Figure 2. Fig. 1995). Auger has catenary anchor lines in addition to the tendons. but the water depth is even larger than for Jolliet with 872m. 1995). These serve two main purposes. The main diﬀerence between Snorre and Hutton is that Snorre is much larger. the TLP concept was brought to a new frontier. 2.statoil.3: The Snorre TLP (www. and is often referred to as a tension leg wellhead platform (TLWP) (Larsen. The platform is of the same size as Hutton. 18 . Many of the design principles from Jolliet are found at Auger. (1) to increase lateral stiﬀness and thereby reduce static oﬀset due to wind and current. and (2) to enable the platform to have a wanted oﬀset position during well drilling and maintenance (Larsen. This platform is signiﬁcantly smaller than both Hutton and Snorre. it has four columns instead of six. Note that the tensioned risers on Auger are inclined in order to decrease the spacing distance between the wellheads on the deck compared to the needed spacing at the seaﬂoor.2.
1995). and has signiﬁcantly larger displacement than the earlier TLPs. Heidrun (Norway) The Heidrun platform was installed in 1995 at 345m water depth. power.4).4) became a bigger problem for this TLP than for the others (Larsen. There are twelve well slots. after undergoing a series of repairs to its damaged platform rig and export pipelines. and the well layout on the seaﬂoor is arranged in a rectangular pattern (30x18m). It was shut down. The Mars platform’s drilling rig was heavily damaged during Hurricane Katrina in August 2005 (see Fig. The deck modules are an open truss frame design (90x90x15m) with a total steel weight of 12. The production at Mars resumed on May 22 2006. at approximately 1200m water depth. as it was exposed to over four hours of sustained winds of 270km/h. Ursa (USA) Shell began the ﬁeld production at Ursa in 1999. It is also the largest producing TLP in the GoM. The hull and tendons system have been tuned to minimize ﬁrst order wave motions. quarters.worldoil.com). This TLP is designed to simultaneously withstand hurricane force waves and winds. The TLP supports a single modiﬁed API platformtype 19 . but the result was that ringing (see Sec.4: The Mars TLP after the hurricane Katrina (www.500 tons. with gusts over 320km/h (Paganie. Mars (USA) Mars was installed in 1996 by Shell at 950m water depth and had the water depth record when installed.Existing Platforms Figure 2. 2. drilling and two process modules. 3. It is the ﬁrst and only concrete TLP. 2006). The deck is composed of six modules: wellbay.
Brutus is Shell’s ﬁfth TLP in the GoM and the seventeenth deep water project in the GoM in which Shell was involved (Rigzone. It was ﬂoating upside down when found after the hurricane. 2007).upstreamonline. The batch setting of the eight wells was completed on January 3 2000. At 1350m. It severed from its mooring and capsized following the hurricane Rita fall 2005.5: The Typhoon TLP of SeaStar design (www. Four of the planned development wells for the eightslot TLP were subsequently predrilled. Fig. the Matterhorn SeaStar was the largest when built in November 2003. Tension Leg Platform (a) Before the hurricane Rita. Brutus (USA) Brutus is a Shell installation at 910m water depth with production start in 2001. (b) Upside down following the hurricane Rita. Morpeth (USA) The threelegged Morpeth SeaStar is a miniTLP and the ﬁrst TLP without surface completions. 2007). Figure 2. Typhoon (USA) Chevron’s miniTLP Typhoon of SeaStar design was installed in July 2001 at 700m water depth. Matterhorn (USA) This miniTLP is an Atlantia SeaStar design of the type previously installed on such deep water projects as Chevron’s Typhoon and BritishBorneo’s Morpeth and Allegheny ﬁelds.5 shows the platform before and after the hurricane Rita struck. however.com).2. with the well layout on the seaﬂoor arranged in a rectangular pattern. It was successfully installed in 1998 in 518m water depth at a cost of less than $100 million (Rigzone. 2. drilling rig (leased) equipped with a surface blow out preventer (BOP) and high pressure drilling riser. and doubled the 20 .
2007). The ﬁeld was installed in August 2005 at 1260m water depth. The TLP contract was awarded to ABB in partnership with Heerema of Holland. which will enable Magnolia to be a regional oﬀtake point for future developments or third party tieins located in Southeastern Garden Banks area (Oﬀshore Tecnology.Existing Platforms size of the previous units. process it and store it until the oil can be oﬄoaded onto waiting tankers (Rigzone. The ﬁeld was discovered in April 2000. The drilling center is a TLP that includes a rig and 36 slots for drilling the project’s oil and gas wells. Magnolia (USA) The Magnolia ﬁeld by Conoco Phillips was installed at 1433m water depth. The ﬁeld was developed with a dry tree TLP installed in January 2004. It came online in July 2004 and is expected to reach peak production of 50. 21 . a new record depth for this type of ﬂoating structure in the GoM.000 barrels of oil equivalent per day (Rigzone. Marco Polo (USA) The Marco Polo ﬁeld is located at 1300m water depth in the GoM. Kizomba (Angola) The Kizomba deep water project is located oﬀ the west coast of Angola. The ﬁeld is located approximately 50 km from the existing infrastructure. Kizomba A consists of three main components. Moored nearby is a FPSO. The production started in December 2004. the last two development wells were drilled. In 2003. It was the ﬁrst unit of this design to incorporate dry trees. 2007). designed to take all of the oil produced from the platform. 2007).
.
only aﬀected by the top tension acting as a vertical force. and wake interference when one cylinder is in the wake of an upstream cylinder (Zdravkovich. Interaction between two cylinders is often classiﬁed into two categories according to the space between them. 3. The system modeled here consists of two risers connected to a TLP through the top nodes. although proximity interference also can take place. were the centers of the two risers are aligned parallel to the free stream (Fig. A successful numerical simulation for design and veriﬁcation of our control system require a suﬃciently detailed mathematical model of the actual process. We usually distinguish between process plant models and control plant models (Sørensen. It may be a part of the modelbased controller. The main purpose of this model is to simulate the real plant dynamics. the proximity interference when the two cylinders are close to each other. We will focus on two risers in a tandem arrangement.Chapter 3 Mathematical Modeling Mathematical models may be formulated in various levels of complexity. It is also found that the inline motions are much larger than the transverse motions. Analysis of the threedimensional eﬀects are subject 23 . control inputs and sensor outputs. forcing the top nodes to follow the prescribed motion from the TLP in the horizontal direction. The CPM is often formulated such that the analytical stability analysis becomes feasible. The top nodes are free in the vertical direction. The control plant model (CPM) is simpliﬁed from the process plant model containing only the main physical properties. It is assumed that a twodimensional model will capture the most important dynamics of the riser array system. 2005). This chapter will focus on the process plant model. In addition the risers are exposed to current forces that are found by considering hydrodynamic interactions between the risers. including environmental disturbances.1). Top tensioned marine riser systems normally fall into the latter category. 1985). The process plant model (PPM) is a comprehensive model of the actual physical process.
1 Current Models Current models could be divided in two: (1) Surface current. In addition. The proﬁles from the North Sea and GoM are design proﬁles. There are three main components in the resulting current proﬁle: • Wind generated currents. • Tide generated currents. • The Gulf of Mexico (GoM). Mathematical Modeling R1 Incoming ﬂow Upstream riser R2 Downstream riser Figure 3. Common to these selected areas are that they are: • Located in deep waters. anchor lines.3. The formulas for these current components can be found in Faltinsen (1990). These are: • The Ormen Lange ﬁeld in the North Sea. tendons. etc. The current proﬁles from these areas will be explained in more detail.1: The risers in a tandem arrangement. and (2) full current proﬁle. • TLPs are already situated here or considered situated here. for use in load models for risers. we have used inplane proﬁles which may vary in velocity through the water column. 24 . • Tend to have strong current velocities. needed for ships and ﬂoaters. Two diﬀerent geographical areas are chosen to investigate the risers behavior under various current velocity conditions. Here. 3. for further research. theoretical proﬁles will be investigated. with diﬀerent proﬁles. • Major ocean currents. some linear.
3. The current velocity data are given in Appendix B. • Ormen Lange 1 which is a one year return period current proﬁle for the Ormen Lange ﬁeld based on Herfjord et al. with a given velocity at the sea surface and 0m/s at the seaﬂoor. based on measurements and dimensioned for 1200m water depth. • Ormen Lange 2 which is a one year return period current proﬁle found in Aker Maritime (2002). • Linearly sheared current (theoretical) which denotes a linearly decreasing current velocity throughout the water column.2. originally given in Norsk Hydro (2001).65m/s at 850m down to 0m/s at the seabed. (2002) and extended to 1200m water depth by linearly decreasing the velocity from 0.1. and can be summarized as: • Uniform current (theoretical) meaning constant current velocity for the entire water column. All current velocity proﬁles are illustrated in Fig.Current Models Uniform 0 200 Waterdepth [m] 400 600 800 1000 1200 −1 0 1 −1 Linearly Sheared Ormen Lange 1 Ormen Lange 2 0 1 −1 0 1 −1 0 1 GoM1 0 200 Waterdepth [m] 400 600 800 1000 1200 −1 0 1 −1 GoM2 Bidirectional 0 1 −1 0 1 Figure 3. It is 25 .2: The simulated current velocity proﬁles in m/s.1.
This gives often strong and variable current. 1999). The Ormen Lange ﬁeld is close to the shelf edge. and is due to a residual warm ﬂow northeastward usually in the upper layer. 2007a. Some a of the current moves straight north into the GoM. and then loops east and south before it exits to the east through the Florida Straits. Mathematical Modeling extended from 850m to 1200m water depth. bidirectional shear current is introduced to investigate the riser behavior and verify the model for a wider range of environmental conditions. For GoM2 a loop eddy is seen to reduce the current at 200400m water depth. • Bidirectional current (theoretical) which represents a linear shear current with opposite current directions at the sea surface and the seabed. Mexico. a bidirectional shear current is presented. Here a simple linear. a further explanation is given.2 the current velocities are seen to be large at the sea surface due to wind generated current. but also large all the way down to the seabed since the ﬁeld is close to the shelf edge. A bidirectional current proﬁle can be found west of Shetland. • Gulf of Mexico 1 (GoM1) which is a wind driven current velocity proﬁle in the GoM.3. 2007). Last. This is illustrated in Fig. and a southwestern cold ﬂow in the lower layer. 3. NOAA Coastal Services Center. In Fig.3. For the rest of the analysis and simulations in this thesis. In addition to the residual ﬂow there are tidal currents. It will be referred to as the Ormen Lange design current proﬁle or just the Ormen Lange proﬁle. 3. the Ormen Lange 2 is used. The loop currents found in the GoM are due to a large ﬂow of warm water that dominates the circulation within the eastern part of the GoM. The current between these depths are considered constant at 0. based on a nondimensional proﬁle in Nowlin et al. (2001). 26 . whereas the velocity increases again for even larger water depths. and might potentially cause operational challenges (van Smirren et al. Some loop currents tear oﬀ the main stream and ﬂow clockwise westward into the GoM (Wikipedia. also based on Nowlin et al. (2001) and scaled the same way as GoM1. It is due to the Gulf Stream which ﬂows northwards between Cuba and the Yuc´tan peninsula. For the current proﬁles related to a geographical area.. It is dimensioned for 1200m water depth and 1m/s current velocity at the sea surface. • Gulf of Mexico 2 (GoM2) which is a current proﬁle with a loop eddy in the top layer.5m/s.
In this thesis we will focus on shielding eﬀects. Also remember the three most important eﬀects for assessment of riser interaction (DNV. but are brieﬂy described here for completeness. Mean force and shielding eﬀects. tending to bring the risers closer. 2005): 1. 2003. The attention will therefore be given to the hydrodynamic inﬂuence on the downstream riser (R2). which is of main importance when calculating the mean current force.2 Hydrodynamic Interaction Recall form Section 1. WIO on the downstream riser. and that R1 can be treated as an isolated riser (Kalleklev et al. DNV. VIV leading to ampliﬁed drag coeﬃcients for both risers.Hydrodynamic Interaction Figure 3. 3. 2.. and hence the position of the second riser in the wake.oceanexplorer.3: Loop current in the Golf of Mexico (www. The two latter eﬀects will not be included in the simulations.noaa. 3.2 that interaction between two neighboring risers will not have any hydrodynamic inﬂuence on the upstream riser (R1) beyond a certain distance. 27 . but to a lesser extent on the control strategy.gov). Inclusion of these eﬀects may have an inﬂuence on the relative riser displacements. 2003).
largeamplitude motion of the downstream riser which wander around in the wake 28 . (3.1) is derived.2 Wake Induced Oscillation The WIO is described as a broad band buﬀeting force due to oncoming turbulent ﬂow and vortices shed from the upstream riser. (3. 3. but this eﬀect is considered insigniﬁcant for the present use of the wake model. CD b = k1 CD De xs . y) = k2 Vc CD De −0. 3.2) where k1 = 0. De is the diameter of the upstream riser. the lift force is not included. The parametric wake model is only applicable for the far wake region larger than two diameters (2D) behind R1.3) It is assumed that mechanical contact occurs when the distance between the riser centers is equal to one riser diameter (1D). xs (3. This may result in a LF. and y is the distance away from the centerline of the incoming velocity proﬁle. and Huse’s model for wake shielding eﬀects is an appropriate model choice.0 for a smooth circular cylinder.1 Shielding Eﬀects R2 experiences reduced mean drag force due to shielding eﬀects. depending on the location in the wake. 2000). 3. is given by Vr (x. x is the distance behind the upstream riser R1. The mean inﬂow on the downstream riser is hence given by Vmean = Vc − Vr .693( y )2 b e . A semiempirical static wake formulation to account for the hydrodynamic interaction between individual risers in steady current was proposed by Huse (1993). The behavior of the ﬂow in the near region is not adequately described as this is a highly nonlinear phenomenon.1. see Fig. xs and b are deﬁned as xs = x + 4De . and xs is the distance between the downstream riser R2 and a virtual wake source upstream of R1. Mathematical Modeling Since the riser model is twodimensional in a tandem arrangement. CD is the drag coeﬃcient. The wake may be inﬂuenced by VIV of the upstream riser. where R2 might experience negative drag forces (Kavanagh et al.2. The reduced velocity ﬁeld in the wake of the upstream cylinder.4.2.. A more extended presentation of how (3.25 and k2 = 1. is found in Appendix A. The centertocenter distance should therefore preferably be kept larger than 2D.1) where Vc is the incoming current velocity on R1.3.
However. It will therefore not be implemented in this work.4) A conservative estimate is to select a high value for the upstream riser and a low value for the downstream riser. (3. 29 . 2002).3 Vortex Induced Vibrations Recall that the large displacements due to shielding eﬀects and WIO are found to be independent of VIV. A more detailed review on existing models are found in DNV (2005). 2005). and is subject for further research. This will bring the risers closer together in their static positions. In this study we have used equal and constant drag coeﬃcients for both risers. It was observed that in addition to the high frequency VIV response of amplitudes up to one half of the diameter. very little information is available for VIV behavior for a downstream riser. and is likely to cause collision (Huse. 1984. 1996.65 . an ampliﬁed drag coeﬃcient due to VIV needs to be included.Hydrodynamic Interaction y Vc Vr (y) x Figure 3. 2003. this kind of oscillations and the physics behind it is so far not well understood. DNV. and will therefore not have impact on the conclusions as such. from Huse (1993). the downstream cylinder also had LF inline oscillations of an apparently very irregular nature with the peaktopeak stroke of these oscillations up to 30 to 40 diameters or more.2. except from the magniﬁed drag coeﬃcient (Tsahalis. Later Wu et al.043 2 D √ 0. The drag coeﬃcients are not used explicitly in the controller algorithm.. These motions occur at the ﬁrst mode of the riser. Anyhow. The ampliﬁed drag is modiﬁed with VIV amplitude A and diameter D. In one experiment Huse (1996) investigated the interaction between cylinders. One of the used relations presented by Vandiver (1983) is CD = CD ′ 2A 1 + 1.4: The decrease in the inline water particle velocity in the wake region. (2002) found the peakto peak range to be 20D. Wu et al. 3. Therefore.
g . f1 denotes the unit vector along the xaxis for the seaﬂoor ﬁxed frame. f1 points to the right. and g3 points downwards normal to the 30 .3. see Fig. 2. and b3 is the normal axis pointing downwards. f2 points into the plane and f3 points upwards normal to the earth surface. b2 is the transversal axis pointing starboard. A bold letter with subscript i = {1. 3.global frame at the surface. 3. i. b1 is the longitudinal axis pointing forward. g2 points to starboard. b .element frames.3 Kinematics and Coordinate Systems Four orthogonal coordinate systems are used to describe the riser and TLP motions.e.ﬁxed frame for ﬁrst riser. seabed ﬁxed frame (fframe): The f frame (of f1 f2 f3 ) is considered inertial and is ﬁxed to the seaﬂoor. i .5. g1 points to the front side of the TLP. 3} denotes the unit vector along the x.body frame of TLP.5: The coordinate systems: f . Bodyﬁxed frame (bframe): The bframe (ob b1 b2 b3 ) is ﬁxed to the body of the TLP with axes chosen to coincide with the principal axes of inertia for the body with origin in the center of gravity. Mathematical Modeling g1 g3 g2 b3 b2 b1 i3 i2 i1 f3 f2 f1 Figure 3. The positions of all the riser nodes in the global system are described relative to this frame. The coordinate systems are all righthand systems. y and z axes in the frame respectively. Global frame (gframe): The gframe (og g1 g2 g3 ) is ﬁxed to the sea surface right above the f frame with a distance in heave direction equal to the water depth and is also considered inertial.
Tension Leg Platform
g g g Earth surface. The position vector rT LP,bg = [xg , ybg , zbg ]T describes the bg position of the bframe relative to the gframe expressed in the gframe. The origin, og , is located on the mean water freefree surface so that zg passes through the center of the TLP when it is in its static undisturbed equilibrium.
Local riser frames (iframe): The iframes (oi i1 i2 i3 ) used in the ﬁnite element modeling (FEM) are located in the ith node of the riser and is a local frame for each element i. i1 is to the right of the element, i2 is pointing into the plane, and i3 is along the axial direction of the element pointing upwards. The forces acting on the element, such as tension, eﬀective weight and current due to drag are computed in this frame for each element and thereafter transformed to the f frame. There are as many iframes as there are number of elements in the FEM model.
3.4
Tension Leg Platform
Recall that the TLP is a partly rigid and partly compliant structure. The eigenfrequencies for inplane motions must be below frequencies for wave energy and the frequencies for the outofplane motions must be above. With respect to the horizontal degrees of freedom, the TLP is compliant and behaves similar to other ﬂoating structures. The horizontal degrees of freedom surge, sway and yaw are inertia dominated with eigenperiods around 12 minutes, well above the range of ﬁrst order waves with periods of 520s. With respect to the vertical degrees of freedom, it is stiﬀ and resembles a ﬁxed structure. The vertical degrees of freedom heave, roll and pitch are stiﬀness dominated with eigenperiods in the order 24s, and well below the period of the exciting ﬁrst order waves (Larsen, 1995). The eigenfrequencies in all six degrees of freedom (DOFs) are tuned relative to the ﬁrst order wave loads. Thus, loads at the wave frequencies do not excite the TLP at its natural frequencies. On the other hand, second or higher order loads at the sum and diﬀerence frequencies can produce signiﬁcant resonant excitations at the TLP natural frequencies because of the small amount of damping available at these frequencies (Faltinsen, 1990, Mekha et al., 1996). • Higher order components and sum frequencies (2ωi , 2ωj , ωi + ωj ) in waves may give signiﬁcant resonance oscillations for the TLP in heave, roll and pitch known as ringing and springing. The restoring forces are due to the tendons and the mass forces due to the TLP, and they are excited by the nonlinear wave eﬀects. Ringing is associated with transients eﬀects, while springing is steadystate oscillations (Faltinsen, 1990).
31
3. Mathematical Modeling
• Loads on diﬀerence frequencies (ωi − ωj ) will give slowly varying wave loads that may give rise to resonant inplane motions. Excitation from wind gusts may appear in the same frequency range and contribute signiﬁcantly to such motions (Faltinsen, 1990, Mekha et al., 1996).
Both these nonlinear phenomena should be considered when designing a TLP. The ringing and springing phenomena have impact on the upper and lower limits of the tendon force. The mean wave drift and current loads on the hull will induce a mean oﬀset force. Assuming linear analysis as the water depth increases, the eigenperiod of the riser system increases as well. The ﬁrst eigenperiod for a riser at 1200m water depth is approximately 3040s depending of the top tension and cross section properties. As the riser eigenperiod is approaching the eigenperiods in surge and sway for the TLP, induced motions from these components may have more inﬂuence on the riser dynamics than in more shallow waters.
3.4.1
TLP Surge Modeling
Floater models and analysis are usually divided into two groups; separated or decoupled analysis and coupled analysis. For separated analysis the vessel motions are found ﬁrst. The eﬀects from moorings and risers are included as nonlinear position dependent forces or stiﬀness. The damping or velocity dependent forces, which are important for estimation of the LF motion may be neglected. In the second step, the dynamic response on the risers and moorings are analyzed, using the vessel response from step one as a forced displacement on the top node. The main problem with this method is that inertia and drag forces on the risers and tendons are not accounted for. These eﬀects could be large in deep waters (Ormberg et al., 1998, Chen et al., 2002). The term coupled analysis means simultaneous analysis of vessel motions, mooring systems and riser dynamics where the full interaction is taken into account. The main drawback with this method is that it is very time consuming since a nonlinear time domain simulation is required. Much eﬀort has been put into the investigation of coupled and separated analysis and solutions in between, see Mekha et al. (1996), Ormberg and Larsen (1997), Ormberg et al. (1997, 1998), Ma et al. (2000), Chaudhury and Ho (2000), Chen et al. (2002). For an extensive description of TLP, tendons and mooring, see Demirbilek (1989b). In this study we assume that the TLP motions inﬂuence the riser behavior. However, the risers do not aﬀect the TLP motion. The TLP represents the speciﬁed motion in surge for the top node of each riser. Hence, a decoupled analysis model is applied. Since the system model is twodimensional, only the
32
Tension Leg Platform
surge motion for the TLP is needed. It is modeled as a LF harmonic motion xT LP = AT LP sin 2π TT LP t + xof f , (3.5)
where xof f is the static TLP oﬀset, AT LP is the amplitude, and TT LP is the period of the TLP motion.
3.4.2
Kinematics
The motions of the TLP are given as the motion of the bframe relative to g the gframe in the gframe with the position vector position vector rT LP,bg = g g [xg , ybg , zbg ]T . As the riser motions are expressed in the f frame, the rotation bg matrix between the f  and gframes is needed. The transformation from g to f frame is given by the diagonal matrix Rf = diag (1, −1, −1) . g (3.6)
Since both frames are considered inertial and ﬁxed with a translation in heave direction only, all axes are parallel, and the xaxes are always pointing in the same direction. In a two dimensional system, only the surge motion of the TLP is needed to deﬁne the prescribed motion of the riser. Hence, the motion in xdirection for the TLP described in the gframe is equal to the xposition given in the f frame. Thus the frames are omitted for the surge position of the TLP, xT LP .
3.4.3
Riser Stroke Calculations
Marine risers made of steel have very low structural strength against lateral loading unless they are tensioned. It is therefore important to maintain the upper end tension at a certain level under all realistic conditions, irrespective of platform motions, dynamic riser response and internal ﬂow parameters. The tensioner system will therefore act as a heave compensating system with an adequate stroke capacity and ability to maintain a near constant tension (Larsen, 1993). If the relative platform/riser motion exceeds the stroke capacity, unwanted loss of tension or tension increase will occur. Such situations may result in excessive bending stresses in the riser, excessive rotations of the ball joint or excessive riser tension. Such loads may cause damage to the well template or the riser tensioning system. Other unwanted eﬀects might be mechanical interaction between neighboring risers or between the riser and the platform. When designing the riser tensioner system, consequences of such events must be taken into considerations (Larsen, 1993).
33
3. Mathematical Modeling
xT LP ∆sr
∆st lr lt
Figure 3.6: Setdown due to tendon/riser geometry.
The platform vertical motion is the most important parameter that deﬁnes stroke. For a TLP with vertical and parallel tendons in the initial zero oﬀset position, motions in surge direction will not induce any pitch motion, but will be coupled to the vertical motion referred to as surge induced heave or setdown. Using the Pythagorean theorem and assuming both risers and tendons to be straightlined, the relative setdown between the riser and the platform can be found. The position of the platform is controlled by the tendons. The oﬀset in surge, xT LP , is equal for the risers and the tendons, but as the risers are longer and hence have larger radius, their setdown is smallest. The relative setdown, ∆s, is found as ∆s = ∆st − ∆sr = lt −
2 lt − x2 LP T
− lr −
2 lr − x2 LP T
,
(3.7)
where ∆st and ∆sr are the tendon and riser setdown respectively, and lt and lr are the tendon and riser lengths. This is illustrated in Fig. 3.6. For given lt and lr , we can calculate the setdowns for the upper end of the risers and tendons. The relative setdown is for normal TLP designs small, even for large water depths and oﬀsets, due to the tendon and riser geometry. Hence, the requirement for heave compensation due to surge induced oﬀset is small.
34
Hence. (3. z2 3. however. and z is along the element. 3. and the top node displacement is prescribed in the horizontal direction. that is two translational DOFs in both ends of the element. In cases where the global geometry is of major importance this will only introduce a small error. x is transverse of the element. 3. The stiﬀness matrix in the FEM model will have an elastic and a geometric component. and the geometric stiﬀness will become more important than the elastic stiﬀness. Node 1 to the left with the two ﬁrst DOFs.7: The bar element with four DOFs.5 Riser FEM Modeling The partial diﬀerential equation (PDE) governing the static and dynamic behavior of a riser can not be solved exactly for arbitrary riser problems and load patterns. Hence. such as the ﬁnite element method (FEM). The elastic stiﬀness matrix accounts for the axial and bending stiﬀness as present in any beam. whereas it is free to move vertically. which prevents the riser from collapsing. the geometric stiﬀness gives the main contribution to lateral resistance against the static and dynamic forces acting perpendicular to the longitudinal riser axis. that the top tension. Each bar element in a two dimensional model can be described with four DOFs.Riser FEM Modeling 2. At larger water depths a simpliﬁcation of the riser model can be made by neglecting the bending stiﬀness and assuming free rotations at the ends. while Ptop is used in the modeling 35 . the riser will behave more and more like a cable. The tension vector. x1 4. Hence. As the depth is increasing. ftop . see Fig. x2 Figure 3. Note. Ptop .8) In later chapters top tension is referred to as T . a numerical method is required. is introduced at the upper end. is an external force and will be found on the right hand side of the equation of motion according to ftop = 01×(2n+1) Ptop T . positive upwards along i3 . Node 2 to the right with the third and fourth DOFs. while the geometric stiﬀness matrix will take into account the changes of the global geometry and the stiﬀening eﬀect from the tension. positive to the right along i1 . z1 i3 i1 1. a model consisting of bar elements is suﬃcient.7. The riser is ﬁxed to the seabed.
which is situated at the seabed in the ﬁrst node of the upstream riser.8: Numbering of the elements (encircled).9) (3. nodes and the corresponding inclination. relative to the global frame. The positions xi and zi of the nodes along the riser are found through equilibrium iterations. li 2 ∆x2 + ∆zi . needed for use in the transformations between the global and local coordinate systems. Note that i is used as the numbering of the elements. li 36 .3. f nodes and the inclination of each local element are illustrated in Fig.1 Transformations for the Riser Elements The risers are modeled with n elements and n+1 nodes.8. θi is the inclination of element i relative to the global coordinate system f frame. i (3. 3. The length of each element i is found by use of the Pythagorean theorem ∆xi = xi+1 − xi .5. The numbering of elements. Mathematical Modeling vn+1 vn n n+1 θn i n i+1 vi θi i v2 v1 1 2 θ2 2 1 θ1 Figure 3. with node number 1 at the seabed and node n + 1 at the wellhead area on the platform. as tension applied at the top node.10) sin θi = ∆xi . vi is the current in node i expressed in the inertial f frame. and are used to calculate the sine and cosine of the inclination θi of each element. nodes (boxes). current vi in each node i and the inclination θi of each local element i. cos θi = li = ∆zi . ∆zi = zi+1 − zi . 3.
and li is the length of element i found from (3.2 System Mass Matrix where ρs is the density of the riser material steel. 3.14) as the inclination of each element is a function of the positions of its nodes.12) where r is the displacement vector.i 02×2 Tf (r) 0. The internal ﬂuid is included as well 2 0 1 0 ρf Aint li 0 2 0 1 mf i = (3.11) − sin θ 0 cos θ Since we are only considering a twodimensional system with two DOFs in each node. the transformation matrix from i to f is written as Tf (r) = 0. The structural mass is given by 2 0 1 0 ρs Ali 0 2 0 1 .15) 6 1 0 2 0 0 1 0 2 37 .Riser FEM Modeling The transformation between the local iframe and the global. msi = (3. For notational simplicity the superscript f is omitted as r is always given in the f frame. and the hydrodynamic added mass ma . Ry. (3. (3. 2002) cos θ 0 sin θ 0 1 0 .5. 6 0 1 0 2 For each element a mass matrix is deﬁned based on the local coordinate system. The full displacement vector is given as r= x1 z1 x2 z2 · · · xi zi · · · xn zn xT LP Tf (r) 02×2 0. A is the cross sectional area of the riser.i cos θi sin θi − sin θi cos θi . the internal ﬂuid mf . as the inclination in each element is given as a function of the positions.9).16) 1 0 2 0 .θ = (3. ﬁxed f frame is described by a rotation about the yaxis (Fossen.13) For all four DOFs related to an element we have Tf (r) = i . The local mass matrix consists of three terms. the structural mass of the riser ms .i zn+1 T . (3. given in the seaﬂoor ﬁxed f frame.
needed for the concatenation to the global matrix. The matrix for the added mass is then given as 2 ρw Cm Ae li 0 mai = 1 6 0 0 0 0 0 1 0 2 0 0 0 . 1997). The coeﬃcient Cm is called the hydrodynamic added mass coeﬃcient.1. Ti is the transformation matrix from the f frame to the f iframe. global mass matrix in the f frame for n elements is then m1 ¯ 11 m1 ¯ 12 m1 m1 + m2 m2 ¯ 12 ¯ 21 ¯ 22 2 ¯ 11 2 + m3 m21 ¯ m22 ¯ 11 ¯ M= . and the subscript is used to identify the diﬀerent parts of the mass matrix. mn−1 + mn mn ¯ 22 ¯ 11 ¯ 12 n m21 ¯ mn ¯ 22 .18) Each local matrix mi is transformed to the global f frame ﬁxed to the seabed.3. (3. The local added mass in the axial zdirection is assumed to be zero.. For mi the ¯ superscript is used to recognize the element number. (3. and the superscript f is hence omitted. Mathematical Modeling where Aint is the internal area of the riser. and all the local mass matrices are assembled to form the global mass matrix M for the riser.20) The global mass matrix M is always given in the f frame. while the added mass for a circular cylinder in the lateral xdirection is equal to the displaced water volume (Sumer and Fredsøe.2. 0 0 (3. 38 . The size of the global matrix is 2(n + 1) × 2(n + 1). The full. The total local mass matrix mi for each element i in its own frame is then the sum of the three terms written as mi = msi + mf i + mai . Each local mass matrix is transformed to f frame by mi = Tf mi Tf T = ¯ i i mi mi ¯ 11 ¯ 12 mi mi ¯ 21 ¯ 22 . . (3.17) where Ae is the area found from the exterior diameter De . and ρw is the density of water.19) ¯ 11 where mi is the mass matrix for element i in the global f frame. and ρf is the density of the internal ﬂuid. The derivation of the consistent mass matrix structure is found in Appendix A.
4 related to the elongation of the element by Pi = EA ∆li . . and the geometric stiﬀness. The derivation of the consistent stiﬀness matrix is found in Appendix A. To calculate the global stiﬀness matrix for the riser. li 0 −1 0 1 0 0 0 0 (3. the elastic stiﬀness. The resulting stiﬀness matrix ki for the element i in its own frame becomes 0 0 0 0 1 0 −1 0 0 1 0 −1 0 0 0 0 1 ki = kEi +kGi = EA 0 0 0 0 + Pi −1 0 1 0 . global stiﬀness matrix in the f frame for n elements is then ¯1 ¯ k11 k1 12 ¯ ¯ ¯ ¯ k1 k1 + k2 k2 22 11 12 21 2 2 + k3 ¯ ¯ k21 k22 ¯11 (3. The full.5.3 System Stiﬀness Matrix The local stiﬀness matrix consists of two terms. kE .24) 39 . 1992) C = α2 K.23) K= ... we transform each local stiﬀness matrix from its local iframe to the global f frame. see Faltinsen (1990) and Mekha et al. . similar to what was done with the mass matrix. These are concatenated to the global stiﬀness matrix K in the same way as the mass matrix. The structural damping is due to the strain and elasticity properties of steel. The geometric stiﬀness matrix includes the eﬀective axial tension Pi acting on the particular element i.2. (1996).4 Structural Damping The damping experienced by a riser is a combination of the structural damping and hydrodynamic damping. whereas the geometric stiﬀness in the lateral direction.22) where l0 is the initial length of an element in a stressfree riser.Riser FEM Modeling 3. l0 ∆li = li − l0 . The elastic stiﬀness matrix works in the axial direction. kG . The damping matrix for a tensioned steel riser is based on proportional Rayleigh damping and assumed proportional to the global stiﬀness matrix (Berge et al. resulting from both radiation and viscous dissipation of energy.21) where E is Young’s modulus of elasticity. The eﬀective axial tension Pi in each element is according to Appendix A.2. (3.5. (3. ¯ ¯ ¯ kn−1 + kn kn 11 12 22 ¯n ¯ k21 kn 22 3.
The force vector fdrag is found at the right hand side in the equation of motion given in (3. Mathematical Modeling The coeﬃcient α2 can be found from simple equations if the total damping level is known at a speciﬁed frequency. However. means that the ﬁrst term of (3. r is the ˙ response velocity. such that λ = 0. For this application such eﬀects are considered to be insigniﬁcant. and the damping is assumed linearly increasing with logarithmic increasing ω.27) later in the text. The acceleration of the water has a material derivative. but Dt this is assumed to be small.3. it is not easy to describe such variations. and r is the structure’s acceleration. the only hydrodynamic force included on the right hand side is the drag force dfdrag . This and the fact that the wave induced water motion is assumed negligible for a riser in deep waters. (3. a = Dv . Hence. and a value of CD = 1.25) ¨ 4 4 2 where v and a are the undisturbed water velocity and acceleration. The term proportional to mass is neglected in order to avoid damping from rigid body motions. which is the last term of (3. and it is therefore common practice for engineering purposes to assume a constant value. The added mass term is already included in the mass term on the left hand side of the equation of motion. this value may for the downstream cylinder be inﬂuenced by the wake.25) can be neglected.2. This approach is taken in the present study. The horizontal hydrodynamic force on a strip of the cylinder can be written dfhyd = ρw CM 2 2 πDe πDe 1 ˙ ˙ adz − ρw Cm rdz + ρw CD De (v − r) v − r dz. the material damping is small relative to the hydrodynamic damping. as the water velocity ﬁeld is varying in space due to shielding eﬀects.5 Hydrodynamic Forces The hydrodynamic forces on the risers are computed using Morison’s equation modiﬁed for a moving circular cylinder.5. In this work we have assumed that the structural damping is 1. CM = Cm + 1 is the ¨ inertia force coeﬃcient. VIV for both risers may also cause some variations of Cm . and CD is known to depend on the ﬂow velocity for a speciﬁc cross section. Frequencies lower than this will have smaller damping. Thus. It is also important to note that variations of CD will not alter any conclusions regarding 40 . This drag force is calculated for each element and summarized as concentrated forces in each node.0. A further discussion on use of the Rayleigh damping can be found in Appendix A.5% of the critical damping at the eigenperiod of T = 10s. However. and a constant value is therefore applied in this study. The added mass Cm for a circular cylinder is known to be 1.15% for T = 100s. The drag coeﬃcient is inﬂuenced by similar eﬀects as added mass.25). 3.3.0 for both risers is applied.
9: Balanced internal and external force. Iteration continues until an equilibrium solution is found with desired accuracy. A detailed description is given in Appendix A. As in traditional FEM. and the lateral stiﬀness of the riser will vary due to ˙ dynamic tension variation. 3. as illustrated in Fig. 41 . For each time step the force equilibrium between the internal force fint . and the external force fext . The top tension deﬁnes geometric stiﬀness according to (3. due to drag forces fdrag and eﬀective weight wef f . The forces acting on each element. is found.3.5.6 Load and Equilibrium Iteration The modeled system contains nonlinearities in the mass. eﬀective weight and drag are computed in the iframe for each element and thereafter transformed to the f frame.Riser FEM Modeling Pi fint fdrag wef f Pi−1 Figure 3. A combined load and equilibrium iteration is needed since drag forces depend on the relative velocity v − r. due to tension.9. stiﬀness and damping matrices in addition to the nonlinearities due to response dependent drag forces. fext the control system design for this study since the estimates of drag forces are not needed for control purposes. 3. such as tension.21) and will vary in time as it is the control input. the nonlinearities are solved numerically by incremental formulation with the Newmarkβ time integration method and NewtonRaphson equilibrium iteration.
The ﬁxed DOFs are the positions at the bottom (both xi=1 and zi=1 ). (3. and fdrag is the drag forces from current and riser motions. The riser position vector is then r= x2 z2 · · · xi zi · · · xn zn zn+1 T . 3.26) where fT LP is the force vector that originates from the TLP motions.28) Note that this vector corresponds to the entire position vector found in (3.5. damping and mass matrices found for the riser can be divided into submatrices that contain the free and prescribed (ﬁxed or with speciﬁed motion) DOFs respectively. the massdamperspring system for the free DOFs in the riser system are found on the left hand side.13).3.27) where r is the riser position vector. and the inﬂuence from speciﬁed DOFs is included on the right hand side as a force acting on the system. In this equation the ﬁxed and prescribed DOFs are removed from the equation. These contributions are found from elements in the original riser system matrices that links the free DOFs to the DOFs with speciﬁed motions (spe).27). whereas xn+1 = xT LP at the top node is the only node with speciﬁed motion.7 TLP Prescribed Forces The complete stiﬀness. but the prescribed DOFs are left out. and hence are removed from the riser system matrices used in the dynamic equation of motion. On the right hand side the external forces from top tension. ¨ ˙ (3. 42 . The superscript f is omitted as the equation of motion is always given in the f frame. For the present case this contribution will be given by fT LP = mspe xT LP + cspe xT LP + kspe xT LP. mspe . Mathematical Modeling 3. The speciﬁed motion will give contribution to the vector with dynamic loads found on the right hand side of the equation of motion. r ˙ (3. meaning that only the free DOFs are present in the dynamic equation of motion. ftop is the top tension. see (3.8 Dynamic Equation of Motion In the dynamic equation of motion. current and drag forces. The columns and rows corresponding to the prescribed DOFs are removed from the original system matrices. and the speciﬁed motion from the TLP is included M (r) ¨ + C (r) r + K (r) r = ftop + fdrag − fT LP . The dynamic simulation algorithm could be found in Fylling et al. (2005).5. cspe and kspe are columns in the original riser system matrices that correspond to xT LP .
i. position dependent force from the TLP. and the speciﬁed. This means that only the nonlinear stiﬀness for the riser system is found on the left hand side. (1) constraints due to stroke and (2) constraints due to tension.e.10.Actuator and Constraints 3.6 Actuator and Constraints The riser tensioner system or heave compensator can be implemented as a hydraulic cylinder with a piston. The dynamic stroke is the length variation needed to tension the riser.qs . The stroke variation is the maximum length variation the tensioner system can provide. Payout is the distance between the bottom of the cylinder and the top of the riser. These could be divided into two groups.29) The algorithm for simulation of the quasistatic model is found in Appendix A. The stroke parameters are based on the deﬁnitions by Larsen (1993). 3. positive downwards. this setup strives to keep the tension close to constant. no active control is needed.27). where dfcur = fT LP. The static position is a particular case with deﬁned environmental and operational conditions. This gives K (r) r = ftop + fcur − fT LP. the riser will experience buckling. less than the eﬀective weight of the riser. the nonlinear viscous drag forces due to the current. the top tension forces on the riser is physically constrained with upper and lower boundaries.4. The initial position refers to a riser and platform condition without oﬀset or environmental forces. The dynamic stroke must compensate for the relative motion between the platform and the riser subjected to all environmental conditions.9 QuasiStatic Equation of Motion The quasistatic equation of motion includes the position dependent terms of (3.5. If this tension is too low. The upper tension limit is restricted 43 . Designing the heave compensator such that the payout is controlled. 2 = kspe xT LP . will still give the same physical constraints. the lower limit for tension should be the eﬀective weight plus a safety margin. 3. On the right hand side the static external forces are given.31) (3. In addition to the boundaries given by the limitations for payout and stroke. Today. Hence. (3. and a desired level of top tension.30) (3.qs 1 ρw CD De vvdz. This is obtained by using a compressed air volume as a soft spring in the hydraulic system. These include the top tension. Hence. while the relative velocity is left out. slightly modiﬁed. and illustrated in Fig.
Maximum 44 . When the TLP is in an oﬀset position. (3. ∆sT is found from zinit (t) = z0 − ∆sT (t). is deﬁned with the design tension and no environmental forces or oﬀsets. ξmin . Mathematical Modeling Follows TLP motion Cylinder Initial payout Initial position Piston Minimum payout zinit ξmin Payout variation Maximum payout + Static position Dynamic stroke ξmax Follows the riser motion Figure 3. means that the piston is as far into the cylinder as possible. The noncompensated initial position. A given tensioning system will also have limitations regarding maximum tension due to limitations of pressure in the hydraulic system. by the yield stress for the riser material and chosen such that the maximum axial stress is less than 40% of the yield stress for steel. for simplicity assumed zero here. such that the initial position compensated for setdown.10: Stroke parameters of one individual cylinder. denoted ξj .32) The measured parameter used in the control loop is the payout for each piston. the setdown of the initial position can be found from the tendon geometry. and a tension rate limit on how fast tension may be varied. and it will be equal for all risers in the present study. z0 .3. Minimum payout. Payout is deﬁned positive when it adds elongation to the riser and negative in the opposite direction.
n+1 is the vertical top position of riser j. refers to the position with maximum free piston length. 2.33) 45 . The initial payout. The total dynamic payout is given as ξj (t) = ξ0 + zinit (t) − zj.n+1 (t).Actuator and Constraints payout. is the distance between the lower end of the cylinder and the initial position. (3. ξmax . j = 1. ξ0 . where zj.
.
The motivation behind this chapter is to validate the riser model from Chapter 3 and verify that it is appropriate as a model of the real world used in the simulations and for the purpose of control system design. • Stress. the simpliﬁcations can be summarized as: • No bending stiﬀness included. with focus on the maximum horizontal displacement and the vertical riser top position. • Free rotations in the ends. bending stiﬀeners in the top and bottom end points. Features often included in riser analysis. in the presented Simulink model. Hence. i. which are of less importance in this case are: • Bending stiﬀness EI in the riser model. The ﬁrst two of these items are assumed to be insigniﬁcant at large water depths (large length to diameter ratio). a bending stiﬀener is implemented at the bottom end node for 47 .e. For the purpose of control system design. • End conditions. 220 versus 400.Chapter 4 Model Veriﬁcation RIFLEX is a commercial FEM program for static and dynamic analysis of slender marine structures (Fylling et al. The model should give a good picture of the global geometry of the riser. In Section 4. despite the simpliﬁcations made in the modeling. as long as the global geometry is of importance.7.. • Few elements compared to typical structural analysis. a more convenient numerical procedure and software code has been implemented in Matlab/Simulink. 2005). exposed to varying TLP positions and tensions.
4. a relatively large number of elements (N = 20) are chosen to increase the resolution and accuracy. The NewtonRaphson method is used for equilibrium iterations within each time step.1 SetUp The equilibrium solution in each time step is found numerically by incremental equilibrium correction. and compared to the Simulink model with free rotation to investigate the eﬀect of such a simpliﬁcation.e.4. the riser top is assumed to be at the level of the free surface. the analyses were run with seven diﬀerent current velocity proﬁles. i. (2005). this model with the number of elements varying between 2 and 20 are analyzed and compared to investigate the size of the error introduced. In the veriﬁcation of the code. For simplicity. The implemented algorithm is found in Fylling et al. Only the terms proportional to the position r are included.4. The integration method for the dynamic riser model is the Newmarkβ method with ﬁxed step size and constant acceleration in each time step. 3. see Appendix A. Model Verification the RIFLEX model. The static riser con˙ r ﬁguration is calculated for each increment in TLP position or top tension.27). The dynamic analysis are run with the full dynamic model from (3. while still maintaining a suﬃcient level of accuracy.29). In Chapter 7. The TLP position and the top tension are harmonically varied. Quasistatic veriﬁcation with increasing TLP oﬀset. For the quasistatic algorithm. Dynamic veriﬁcation with harmonic top tension variations. 2. such that the implemented model can represent the geometry of the real world. The quasistatic analyses are run with the static model given in (3. Dynamic veriﬁcation with harmonic TLP motions. Most of the tests are run at 1200m water depth. The model implemented in Simulink is run with 20 elements which gives an element length of 60m. it is of interest to minimize the number of elements applied in the riser model. while the dynamic terms proportional to velocity r and acceleration ¨ are ignored. A wide range of cases were analyzed to better investigate the robustness of the model. Quasistatic veriﬁcation with increasing top tension. 4. In order to limit the computation time for realtime control applications. Four tests are run: 1. To verify the riser model. 1200m above the seaﬂoor. Note that bending stiﬀness has been included in the RIFLEX model in all analyses presented herein. The current 48 . each of length 3m. The model run in RIFLEX consists of 400 elements. representing diﬀerent geographical areas.
4. The models were run for all the diﬀerent current proﬁles and with various current velocity amplitudes. The deviation in the vertical position of the node with max displacement is due to the relatively low number of elements in the Simulink model. which also is used to describe the TLPs motion on a sphere surface. 4.2.0m/s for the bidirectional current proﬁle to 1.2. The TLP oﬀset was increased in steps of 5m from 0m to 70m. Current proﬁle data are found in Appendix B. with surface velocity ranging from 0. However. We achieved similar correspondence in snapshots and setdown for the other current proﬁles. 4. The static equilibrium solution is found for each TLP position for both the RIFLEX and the Simulink models.2 QuasiStatic Veriﬁcation with Increasing TLP Oﬀset The quasistatic riser model was veriﬁed ﬁrst. 4.1s is seen to be stable in all DOFs for all simulations.2 and described in Section 3. Most cases were run with between 0m/s and 0. The setdown curve reﬂects the setdown due to surge. A step length of 0. with most cases about 1m/s. The deviation for the riser conﬁguration is seen to be small.1. 4. for nonlinear systems the step length the simulations must be small enough to capture system dynamics also for the stiﬀ vertical DOFs. Deﬂections of the riser with the design current from the Ormen Lange ﬁeld with surface velocity 1.1.1a). The riser models match close to perfect.1. The setdown is given in Fig. corresponding to 0 to 5.3m/s to 2.5m/s at the seabed. respectively. the tension was increased from 1200kN to 2700kN in steps of 50kN.8% of the water depth.1b). The current velocity at the seabed varied from 1. see Section 7. gave as expected a better correspondence between the RIFLEX and Simulink calculations than fewer elements. These current proﬁles were run with various current velocity amplitude. x and o mark the position along the riser with the maximum horizontal displacement for RIFLEX and Simulink. like 5 or 10.QuasiStatic Verification with Increasing TLP Offset velocity proﬁles are illustrated in Fig. 20 elements as seen in the Fig.3 QuasiStatic Veriﬁcation with Increasing Top Tension In the second veriﬁcation of the quasistatic riser model. 3.15m/s are seen in Fig. It is seen to correspond nicely for the two models. All current proﬁles were tested with 49 .2m/s for the ampliﬁed GoM2.5m/s.1. Physical riser data are found in Appendix B. The Newmarkβ method is unconditionally stable for linear systems.
Fig.1: Deﬂections (a) and setdown (b) for the quasistatic riser model with TLP oﬀsets from 0m to 70m and Ormen Lange design current.4. Eﬀects of the reduced tension along the riser are clearly seen at low tensions. and is clearly seen from the ﬁgure. As before. Fig. Model Verification Snapshots 1200 1000 Vertical position [m] Top position [m] 800 600 400 200 0 1200.3 shows the deﬂections of the quasistatic riser model exposed to the bidirectional current velocity proﬁle. especially about 200m above the seabed. xof f = {0. For each curve the tension increases with 50kN and the deﬂection decreases correspondingly. Also the deviation between the models in RIFLEX and Simulink is larger than for the other current proﬁles. The tension along the riser is decreased from top to bottom due to the eﬀective weight of the riser. This will lead to a more symmetric deﬂection shape for increasing tension in uniform current. The reason is that the relative diﬀerence between top and bottom tension will decrease with increasing tension.2 shows the deﬂections of a riser exposed to the Ormen Lange current and increasing tension.5 1200 1199. The curve to the right with horizontal displacement of approximately 20m corresponds to the lowest tension of 1200kN. x and o mark the position along the riser with the maximum horizontal displacement for the RIFLEX and Simulink model. TLP positioned in zero and 30m oﬀset. even with a symmetric current proﬁle. The lower tension limit was the eﬀective weight plus a safety margin for the structure connection at the seabed. the deﬂection is much larger in the lower half than the upper half. respectively.5 1198 1197. RIFLEX(– x) and Simulink (. This is particularly clear at low tensions.5 Setdown RIFLEX N = 400 Simulink N = 20 0 20 40 TLP position [m] 60 0 20 40 Horizontal position [m] 60 Figure 4. The curvature is large. 4. The upper tension limit was given by a percentage of the yield stress for steel.5 1199 1198.o). 30}m. Note that the vertical position of the max horizontal displacement increases when the tension increases. 4. 20 elements seems to 50 .
QuasiStatic Verification with Increasing Top Tension
1200
1000
800 Vertical position [m]
600
400
200
0
0
2
4
6
8 10 12 Horizontal position [m]
14
16
18
20
Figure 4.2: Deﬂections of the quasistatic riser model with increasing top tensions from 1200kN to 2700kN in steps of 50kN, and current proﬁle from the Ormen Lange ﬁeld. RIFLEX(– x) and Simulink ( o).
be too few to describe the riser conﬁguration accurately, i.e. more elements are needed to describe the conﬁguration satisfactory, if the curvature is large. The model is seen to be closer to the RIFLEX model for large tensions giving less deﬂections. However, the important characteristics are kept and the model can picture the main riser conﬁguration. The relations between top tension, vertical top position and maximum horizontal displacement are investigated. The relations for the Ormen Lange design current is illustrated in Fig. 4.4. The veriﬁcation showed good agreement between the models within the tension limits. For lower tensions, nonlinear eﬀects like buckling may appear. These eﬀects are not implemented in the Simulink model. Top tension less than the lower saturation limit is therefore not simulated. The physical correspondence between the vertical top position, maximum horizontal displacement, top tension and geometric stiﬀness should be explained more thoroughly. This is best done by considering the numerical derivatives
51
4. Model Verification
1200
1000
800 Vertical position [m]
600
400
200
0
−5
−4
−3 −2 −1 Horizontal position [m]
0
1
Figure 4.3: Deﬂections of the quasistatic model with increasing tension exposed to the bidirectional shear current. RIFLEX(– x) and Simulink ( o).
Max displacement [m]
Displacement vs top position 25 20 15 10 5 1199 1200 1201 Top position [m]
Displacement vs tension 25 20 15 10 5 1500 2000 2500 Top tension [kN] Top position [m]
Top position vs tension 1201 1200.5 1200 1199.5 1199 1500 2000 2500 Top tension [kN]
Figure 4.4: From left: a) Maximum horizontal displacement vs vertical top position, xmax (zmax ), b) Maximum horizontal displacement vs top tension, xmax (T ), c) Vertical top position vs top tension, zmax (T ). All plots are for the quasistatic riser model exposed to the Ormen Lange proﬁle. RIFLEX(– x) and Simulink ( ).
52
QuasiStatic Verification with Increasing Top Tension
−4 −6 ∆x/∆T ∆x/∆z −8 −10 −12 1199.5 1200 1200.5 1201 Vertical top position [m]
0
4 3 ∆z/∆T 2 1 0
x 10
−3
−0.01
−0.02
−0.03
1500 2000 2500 Top tension [kN]
1500 2000 2500 Top tension [kN]
Figure 4.5: From left: a) The derivative of x w.r.t. z, ∆x (z) b) The derivative of ∆z ∆x ∆z x w.r.t. tension, ∆T (T ), c) The derivative of z w.r.t. tension, ∆T (T ). All for the design current at the Ormen Lange ﬁeld, RIFLEX(– x) and Simulink ( ).
∆z and ∆T (T ) in Fig. 4.5. The eﬀect of the increase in tension could be divided into three diﬀerent stages depending on the tension level:
∆x ∆x ∆z (z), ∆T (T )
1. Low tension, large deﬂection. 2. Medium tension, medium deﬂection. 3. Large tension, small deﬂection. The ﬁrst of these may or may not be clearly seen in the analysis, depending on the current proﬁle and the top tension. In the exploration of all stages, the top tension is increased with a ﬁxed magnitude ∆T . For stage 13 above this results in: Stage 1: Small tension, large deﬂection • Large decrease in deﬂection. • Large increase in top position, due to geometric ﬂexibility. • A ﬁxed increase of z by ∆z will give a medium decrease in displacement, ∆x. Stage 2: Medium tension, medium deﬂection • Medium decrease in deﬂection. • Medium increase in top position, reduced geometric ﬂexibility; some contributions from elastic ﬂexibility.
53
4. Model Verification
• A ﬁxed increase in z gives a large decrease in displacement. Stage 3: Large tension, small deﬂection • Small decrease in the deﬂection. • Small increase in top position, controlled by elastic ﬂexibility, negligible geometric ﬂexibility. • A ﬁxed increase in z gives a small decrease in displacement. The geometric stiﬀness is proportional to the increase in tension, ∆KG ∝ ∆T , meaning that a ﬁxed increase in tension gives a proportional change in the geometric stiﬀness. For low tensions, a ﬁxed increase has large eﬀects on the geometry and the displacements. For large tensions, and already large geometric stiﬀness, an additional increase in tension, does not have the same eﬀect. There were also run tests with decreasing top position in RIFLEX. The top tension started at 2900kN and decreased corresponding to steps in the vertical top position of 5cm. These result are similar to what is presented here, and can be found in Rustad et al. (2007c).
4.4
Veriﬁcation with Dynamically Moving TLP
The dynamics of the riser model and the inﬂuence from the drag forces were tested by harmonic motions in surge for the TLP, with amplitude AT LP = 20m, and periods in the range TT LP = 60 − 300s. The static oﬀset was xof f = 30m, and the riser tension was kept constant at 1800kN. Fig. 4.6 shows snapshots of the dynamic riser motion at the Ormen Lange ﬁeld simulated in Simulink with TT LP = 60s. The dynamic envelope curves are found from RIFLEX. The snapshots ﬁt nicely between the envelope curves given by RIFLEX. The top position versus time is plotted in Fig. 4.7a), while Fig. 4.7b) shows the correspondence between the top position and the TLP surge position. The change in top position is bigger for larger curvature of the riser. This was best seen when the TLP was moving from its maximum to its minimum oﬀset, with the smallest surge period, TT LP = 60s, seen in Fig. 4.6a). Let us for simplicity assume modal representation of the riser. Then the riser top position could be said to be a function of time and space, i.e. TLP motion and deﬂection z(t, x) = Σi φi (x) · qi (t), (4.1) where qi (t) is dependent of time, being the TLP motion and the ﬁrst mode shape. φi (x) is higher modes due to the deﬂection and resonance for the riser at its
54
8b). like TT LP = 300s. 4. When the TLP is moved from left to right. the TLP velocity has a large inﬂuence on the riser deﬂection. and is even more clearly seen in the simulations with small current velocities in the top layer. causing resonance in the higher order modes. and from left to right (b) for the Ormen Lange design current. including the second mode shape. To summarize. such that the second order mode shape is more visible. This second mode can also be found in the top position versus time. For longer surge periods. Fig. with an amplitude of 750kN 55 . like the GoM2 velocity proﬁle. 4.6b). more resonance frequencies can be excited at higher modes.5 Veriﬁcation with Dynamically Varying Tension The fourth and last of the riser model veriﬁcation tests included harmonic tension variations.7a). 4. Depending on the period of the TLP. not shown here. giving less need for stroke capacity than with faster TLP motions. and hence the need for payout and stroke. The second order mode shape is due to the TLP motion. The snapshots are from Simulink and the thick. in Fig. This will give smaller setdown of the top node. The riser setdown can be said to be dependent on the TLP motions. The ﬁrst order mode shape is less dominating due to the smaller current velocities and smaller deﬂection. The top node position in time for GoM2 is seen in Fig.6: Snapshots of the TLP moving from right to left (a). The initial tension was 1950kN. Also the setdown due to deﬂection was smaller. 4. blue lines are envelope displacement curves from RIFLEX. eigenperiods. the motions were close to quasistatic giving a smaller curvature and only the ﬁrst mode shape was observed. the second mode is seen. 4.Verification with Dynamically Varying Tension Snapshots of TLP moving from right til left 1200 1000 Vertical position [m] 800 600 400 200 0 Vertical position [m] 1200 1000 800 600 400 200 0 Snapshots of TLP moving from left til right 0 10 20 30 40 Horizontal position [m] 50 0 10 20 30 40 Horizontal position [m] 50 Figure 4.8a) and the top node position versus TLP surge in Fig.
2 0 100 Time [s] 200 300 1199.4 1199.4 1198.2 1199 1198.2 10 Top node xz−trajectory 20 30 40 TLP position [m] 50 Figure 4.6 1199.4 1199.6 1199.7: Top position as a function of time (a) and TLP motion (b) with period of 60s.) for the Ormen Lange design current.6 1198.6 1198.2 1200.8 1199.8 1199.4 1199.8: Top position as a function of time (a) and TLP motion (b) with period of 60s. RIFLEX(–) and Simulink (.2 1199 1198.2 10 20 30 40 TLP position [m] 50 Figure 4.) for GoM2 current and TLP dynamic period of 60s.4.8 1199.4 1198.2 0 100 Time [s] 200 300 Top position [m] 1200 1199.4 1199.6 Top position [m] 1199. 56 . RIFLEX(–) and Simulink (.8 1199. Top node position 1200.2 Top node xz−trajectory 1200 Top position [m] Top position [m] 1200 1199. Model Verification Top node position 1200 1199.8 1198.6 1199.8 1198.
4. Fig. and periods of 60s and 120s. The thick lines are the static conﬁguration for the initial tension and the envelope displacement curves from RIFLEX.10 c). The snapshots from Simulink are seen to be within these envelope curves.10 shows the dynamic variation of the top tension (a) and the top position (b). GoM2 and the bidirectional current proﬁle.Verification with Dynamically Varying Tension Snapshot 1200 1000 800 Vertical position [m] 600 400 200 0 0 5 Horizontal position [m] 10 15 Figure 4.9: Snapshots with harmonically changing top tension for the Ormen Lange design current (black). There were also run tests where the motion of the top node in RIFLEX was 57 . respectively. 4. Fig. 4. Simulations were run with currents corresponding to the Ormen Lange ﬁeld. Hence. The motion trajectory for vertical position versus tension is seen in Fig. The results from simulations with 60s are similar.9 shows the snapshots with dynamical tension variation of period 120s from Simulink. The blue lines are the static conﬁguration and envelope displacement curves from RIFLEX.5kN/s. This gives maximum tension rates of 75kN/s and 37. A transient period from the static initial condition is seen. This is far more than the limit of a conventional tension system. if the riser model is valid for these large and fast changes in tension. it will also be valid for slower limit rates.
This section is written in cooperation with Stølen (2007). The correlation between tension and vertical top position is nonlinear and asymmetric.5m.10: Top tension (a) and position (b) as functions of time.5m than the reduction in tension when lowering the top with 0. To investigate the area of application for the riser model with free end rotations. the results presented in this section are better suited for the purpose of veriﬁcation.4. In shallow waters.6. TLP and tensioned risers are not likely to be applied. the bending stiﬀness is assumed to be of considerable importance. but can be found in Rustad et al. and the test setup is similar to the corresponding veriﬁcations for 1200m water depth.1 Increasing TLP Oﬀset At 300m water depth. The 58 . (2007c). it is not necessary to investigate the validation at even smaller water depths.5 1200 0 100 200 300 Time [s] 1200 1000 2000 3000 Top tension [kN] Figure 4. being 300m and 600m. sinusoidal tension was used. In even shallower waters. and the riser behaves more like a cable. The results with prescribed motions are not presented here. A linearly sheared current is used in all tests. and top position vs top tension (c).5m.6 QuasiStatic Veriﬁcation at Shallow Waters In deep waters. the TLP oﬀset is varied from 0m to 30m in steps of 2. RIFLEX(–) and Simulink (.5 1200. This is caused by hydrodynamic drag forces that will increase needed tension for an upwards motion. it is tested in shallow water. the lateral stiﬀness is assumed to be dominated by the geometric stiﬀness. Model Verification Top tension vs time 1201 Top tension [kN] Top position [m] 2500 2000 1500 0 100 200 300 Time [s] Top position vs time Top position vs top tension 1201 1200. In the corresponding simulations in Simulink. At 600m water depth the oﬀset is varied from 0m to 40m in steps of 5m. meaning that more tension is needed to lift the riser 0. However. 4. a prescribed vertical sinusoidal. Hence. but decrease the tension reduction for a downwards motion.) for Ormen Lange design current. 4.
and the agreement is better.5 299 298. This is the same top tension factor compared to the eﬀective weight as for 1200m.2 Increasing Tension The quasistatic analysis is now run with increasing tension at 300m and 600m water depth. The deviations in Fig. Similar results are found for 600m.5 RIFLEX N = 100 Simulink N = 20 0 10 20 Horizontal position [m] 30 298 0 10 20 TLP position [m] 30 Figure 4. For higher tensions. The eﬀect of bending stiﬀness is not included in the Simulink model such that the total stiﬀness is too small in particular at low tensions.11: Deﬂections (a) and setdown (b) for the quasistatic riser model with TLP oﬀsets from 0m to 30m. and decreased for increased tensions. 300m water depth and linearly sheared current. and the boundaries for the validation of the Simulink model.2%) for the lowest tension. This is less than the safety margin. The results for 300m and 600m are found in Fig. The eﬀective weight is 230kN which results in a minimum residual tension of 20kN at the bottom. 4.12a) for the rightmost curves are seen to be large (0. The lower tension limit should be set to give a minimum tension at the bottom connection of the riser. 4. the relative error was less than 1%. respectively.o). similar to what is found at 1200m water depth.11.6. with Tf = 1.5 Setdown 300 299. The deviation is small. both for the lateral deﬂections and the setdowns.94. 59 . Except for the two lowest tensions. RIFLEX(– x) and Simulink (. top tensions are 450kN and 900kN at 300m and 600m. For the 300m case the tension is increased from 250kN in steps of 25kN up to 700kN. and hence too large deﬂections must be expected. which are out of range of the validity of the Simulink model.QuasiStatic Verification at Shallow Waters Snapshots 300 250 Vertical position [m] 200 150 100 50 0 Top position [m] 300. the bending stiﬀness is of less relative importance.12m and 3. 4.
in most cases this is not a feasible solution. At 600m water depth the tension is varied from 600kN to 1200kN in steps of 50kN. In this section the eﬀect of a bending stiﬀener at the bottom connection will be analyzed. The eﬀective weight is 450kN. is to tie down a marine riser to the seaﬂoor through a ball joint. However.o).7 QuasiStatic Veriﬁcation with Stress Joint The Simulink riser model has free rotation at the ends. 600m and 1200m water depths 60 .4. giving a safety margin of 150kN. is implement in RIFLEX. This has zero rotation stiﬀness and will eliminate bending stresses at the riser end. The snapshots in Fig. and the end conditions are assumed to be insigniﬁcant for the global geometry in deep waters. 4.1%. The main purpose of the stress joint is to provide a gradual transition between the relatively ﬂexible riser and the rigid wellhead. This section is written in cooperation with Stølen (2007). but also suﬃciently strong to resist the forces and moments introduced by the riser at the top of the wellhead. The relative error was less than 0. The risers will be located at 300m. The joint must have suﬃciently ﬂexibility to keep the bending stresses in the lower part of the riser at an acceptable level. 4. Model Verification 300m 300 250 Vertical position [m] 200 150 100 50 0 Vertical position [m] 600 500 400 300 200 100 0 600m 0 1 2 3 Horizontal position [m] 4 0 1 2 3 4 Horizontal position [m] 5 Figure 4.12: Deﬂections for the quasistatic riser model with increasing tension in (a) 300m and (b) 600m water depth and shear current.12b) show satisfactory performance. A stress joint. with respect to stresses. The solution. RIFLEX(– x) and Simulink (. with the same material speciﬁcations as the riser.
For higher 61 .6. The stress joint is seen to have a large impact on the curvature of the riser and also the maximum horizontal displacements. The internal diameter is the same for all segments and equal to the riser. which does not include the eﬀects from bending stiﬀness. Only one element is applied for each segment. Riser hSi Segmenti t Si DSi Figure 4.1%. The external diameter is increased from the riser thickness at the top to 4t at the bottom.1 Shallow Water The stress joint conﬁguration was ﬁrst tested at 300m water depth. and if the assumption of insigniﬁcance of the end connections is valid for the various water depths.13 shows the stress joint model.2 that the lowest tensions (and rightmost curves) are not valid for the Simulink model in 300m. and that a ball joint can be used at the bottom connection.QuasiStatic Verification with Stress Joint and tested with increasing tensions like in Sections 4. 4. These tests are needed since it has been assumed that the maximum horizontal displacement and the vertical riser top position are unaﬀected by the presence of a bending stiﬀener in deep waters. see Fig. No stress joint is implemented in the Simulink model.6.3% and 10.13: Stress joint conﬁguration.3 and 4. Fig. and Table 4. The two lowest tension cases give a relative error of 17. Recall from Section 4.2. The stress joint model is seen to give the riser a vertical orientation close to the bottom.7. 4. 4.1 summarizes the geometric parameters.14. The objective is to test what eﬀect the stress joint will have on the global geometry.
The horizontal displacements are also less aﬀected by the stress joint.14b).0 t 4.0 0.o).0 t 3. RIFLEX(– x) and Simulink (. As expected the results are now in better agreement than for the 300 meter case.1: Stress joint data. 300m 300 250 Vertical position [m] 200 150 100 50 0 Vertical position [m] 600 500 400 300 200 100 0 600m 0 1 2 3 Horizontal position [m] 4 0 1 2 3 4 Horizontal position [m] 5 Figure 4. Model Verification Riser Segment 1 Segment 2 Segment 3 Segment 4 Total stress joint hsi [m] 2. the relative error in maximum horizontal displacement is 3. increasing top tension and linearly sheared current at (a) 300m and (b) 600m water depth.4. For the lowest tension. 4.7 1. The stress joint imposed at the lower end has less inﬂuence on the curvature due to the increased riser length. tensions the deviations are much smaller decreasing from 6.8 1.5 6 tsi [m] t 1.5 t 2.0 t  Dsi [m] D D+t D+2t D+4t D+6t  Table 4.14: Snapshots for quasistatic riser model with stress joint of 6m.2% and less than 1% for tension higher than 850kN.2% down to less than 1% for the highest tensions. The snapshots for 600m are shown in Fig. 62 .
However.8 Discussion In this chapter the mathematical model developed in Chapter 3 has been veriﬁed. 4.o). see Fig. This gives larger deﬂection and larger horizontal deviations in meters (see Fig.5%. the deviations with the stress joint are still seen to be small. and several diﬀerent 63 . even for the lowest tensions. The RIFLEX model with 6m stress joint was also exposed to the Ormen Lange current. For the 12m stress joint. even for the longest stress joint. The correspondence between the RIFLEX and the Simulink model was better with the ball joint conﬁguration. 4. The importance of a stress joint and inclusion of bending stiﬀness is seen to give largest error for tensions less than 1600kN.1 are doubled. Both models were exposed to TLP motions and tension variations. all dimensions from Table 4.Discussion 6m joint 1200 1000 Vertical position [m] 800 600 400 200 0 Vertical position [m] 1200 1000 800 600 400 200 0 12m joint 0 2 4 6 8 Horizontal position [m] 10 0 2 4 6 8 Horizontal position [m] 10 Figure 4. RIFLEX(– x) and Simulink (. For tension higher than 1600kN.15.2 Deep Waters The riser at 1200m water depth is tested in linearly sheared current with stress joints of 6m and 12m.15: Snapshots for quasistatic riser model with increasing top tension at 1200m water depth including stress joint of (a) 6m and (b) 12m exposed to a linearly sheared current. 4. 4.17 the relative errors in horizontal maximum deﬂection for each of the three cases are shown. even for the 12m stress joint.7. The veriﬁcation was performed for both quasistatic and dynamic models.16). The importance of the stress joint is seen to be less signiﬁcant at this water depth. but the relative deviations are smaller. the error is less than 0. 4. In Fig.
RIFLEX (–x) and Simulink (.o).16: Snapshots for quasistatic riser model with a 6m stress joint and increasing top tension in 1200m waters exposed to the Ormen Lange design current. Model Verification 1200 1000 800 Vertical position [m] 600 400 200 0 0 2 4 6 8 10 12 Horizontal position [m] 14 16 18 20 Figure 4. 64 .4.
2 1 [%] 0. when applying this model.6 6m stress joint shear 12m stress joint shear 6m stress joint OL 1.*) stress joint in Ormen Lange current. current proﬁles with a wide range of current velocities have been analyzed. especially at the lowest tensions. The quasistatic model was tested in shallow waters.6 0.2 0 1200 1400 1600 1800 2000 Top tension [kN] 2200 2400 2600 Figure 4. where the model with free rotations could be used for drilling purposes.8 0. The simulations have shown that there is a good agreement between the model from Chapter 3 and the RIFLEX model. the eﬀect of the neglected bending stiﬀness was seen. The eﬀect was large in shallow waters. acting together with the bending stiﬀness. In deep waters. For tension levels lower than the limit for the Simulink model.4 0. one should be aware of this eﬀect.o) in linearly sheared current and 6m (.4 1.17: Relative error in horizontal maximum displacement for stress joints of 6m (–x) and 12m (. It showed good performance at suﬃciently high tension levels. Hence. as this is the 65 . A stress joint was implemented in the RIFLEX model. the eﬀect of the stress joint on the global geometry is small and the assumption of free rotations at the ends is acceptable.Discussion Relative error in horizontal displacement 1. and avoid use of the model for low tension values. The main focus has been deep waters and second order LF induced motions.
2). in the riser analysis in Chapter 7. no need for control. 66 .3. The small deﬂections are due to the second order mode shape. Note that WF motions have not been included in this veriﬁcation. but could still draw the main conﬁguration. 20 elements did not describe the riser accurately. Anyhow. the risk of collision is limited and hence. in fully developed. Model Verification main motivation for the thesis. There is a good agreement with the RIFLEX model in all the relevant veriﬁcation tests for this proﬁle. some additional current proﬁles will also be used. on the other hand. However. undisturbed bidirectional current and the lowest top tension. As the deﬂections are small. It will therefore be used as the current proﬁle in the simulations with control. gives a large deﬂection. which will increase the risk of collision. see Fig. However. with zero oﬀset and low tension. 20 elements have shown to be suﬃcient to describe the geometry of the riser in most cases. the largest deﬂection is less than 2m in the upper half and less than 6m in the lower half.4. 4. This is far less than the maximum deﬂection of approximately 20m with the lowest tension for the Ormen Lange current (Fig. 4. in cases with large curvature. The Ormen Lange proﬁle. like the bidirectional current proﬁle.
This includes identiﬁcation of the various stages of riser behavior depending on the environmental conditions. The TLP is described in Chapter 2 and modeled in Section 3. including actuators and measurements. and the lateral deﬂections are thereby aﬀected. The actuator is the riser tensioner system and a part of the implementation. 5. the TLP motions are assumed independent of the risers. The TLP motions depend on the environmental conditions current.5.4. the risers are tightened or loosened.1.1 Implementation Overview The implementation of the system is shown in Fig. waves and wind. 67 . waves and current. 5. the control architecture.e. starting with the physical components. The environment consists of wind. The main goal of this system is to prevent the risers from colliding subjected to the given constraints. The current proﬁles are described in Section 3. The behavior of the two risers depends on the TLP motions as well as the current velocity proﬁle. On the other hand. The diﬀerent parts of the system. is brieﬂy described below. how the various parts are working together and the controller system design. By increasing or decreasing the tension and payout. The risers are modeled in Section 3. i.1. The entire current proﬁle has inﬂuence on the riser conﬁguration.Chapter 5 The Riser Control System Overview This chapter aims to develop the control architecture and system design for the riser tensioning system.
In oﬀshore operations the tension must be maintained independent of the movements of the platform. The aim of the heave compensation system is to keep the riser tension unaﬀected by the vertical TLP motions. introducing setdown in an oﬀset positions. This chapter focuses on the actuator. The controller takes input from an operator or computes new optimal setpoints based on the measured inputs. relative horizontal distance. Most passive heave compensation systems are pneumatichydraulic systems with glycol and air in compression. Heave compensators are usually divided into three groups. 5. wind gusts and second order LF wave forces cause the horizontal motions of the platform. The TLP motions will induce motions on the risers and impact the tension. The measurements used by the controller could be the current proﬁle measurements.5. The WF oscillations of the riser are caused by ﬁrst order wave loads. riser inclinations. For riser systems today. Current. The Riser Control System Overview TLP Controller Actuator Risers Environment Measurements Figure 5. this means to keep the tension close to constant. The environment. The output is the control input for the actuator.1: Outline of the implementation of the system. Passive heave compensation systems can be described as springdamper systems. TLP motions. measurements and controller system.2 Actuator The actuator in this system is the riser tensioner system. The compensation for the relative vertical motion between the TLP and the risers is called heave compensation. passive. TLP and risers are described in Chapter 3. active and a combination of these. and the load is balanced by the pressure of ﬂuid volume act68 . which is designed to carry the load of the risers and transfer it to the platform structure. and do not require any input of energy under operation. tensions and payouts for the risers.
A tensioner system for production risers at Jolliet is shown in Fig. Both riser tensioner systems give a nearly constant tension to the marine risers and compensate for rig motions. The Nline system has six cylinders in a ring. and the remaining four keep on working.Measurements (a) Wireline. In our application 23m could be suﬃcient. 5.3. In drilling operations. Figure 5. See Nielsen (2004) for more details. 5. (b) Nline. the TLP motions and the top tension for each riser. 69 . we could measure the payout and the top and bottom riser angles.3 Measurements To be able to run a supervisory. the opposite cylinder is set free. The spring is nonlinear due to the compressed gas volume.2: The diﬀerent riser tensioner systems (National Oilwell Varco. see Fig. The system is symmetrical. the stroke length could be as much as 15m. The measured payout and the relative horizontal positions are used by the controller to calculate feedback. iterative FEM model we need to know the incoming current. There are two types of heave active compensators: (1) Wireline and (2) Nline riser tensioner systems.2. so if one cylinder fails. 2007b). The cylinders are installed in frames on the cellar deck and connected to a tensioner ring on the riser. 5. ing on a piston. To ensure that this monitoring model is giving a correct result.
5. working frequency for the unit and cell size.3. Fig. With increased operating depths. 2006). depending on the conﬁguration including the placement of the sensors. 5.1 Current The incoming undisturbed current proﬁle can be measured using an acoustic Doppler current proﬁler (ADCP). near surface current can be 70 .4 shows how current proﬁles can be obtained in water depth larger than 1000m.3: Production riser tensioner system at the TLP Jolliet (National Oilwell Varco. the midcolumn current and the bottom current. By using a horizontally directed ADCP the undisturbed. Hence. Three diﬀerent placements of the ADCPs may be used to cover the surface current. the platform itself introduces changes in the current and wave ﬁeld. oﬀshore oil rigs and polar research moorings (RD Instruments. a single downward looking ADCP may not provide enough coverage. Near the surface.5. The Riser Control System Overview Figure 5. 2007b). ADCP measurement units are currently deployed on environmental monitoring buoys. The current proﬁle can be measured for water depths down to 2000m. due to its size and the use of thrusters for station keeping for ﬂoaters like the semisubmersible. the near surface current can not be reliably accomplished by using a current meter that measures the current in the direct vicinity of the rig.
com). While the near surface current is of main importance for the oﬀset of the rig or platform. an upwardlooking ADCP is attached to the bottom structure or moored to the seabed. the midcolumn velocity is aﬀecting the riser deﬂections.Measurements Figure 5. This bottommounted ADCP system collects current data and transmits it to the surface in realtime via an acoustic modem.4: Using ADCP to cover the full current proﬁle. In this way. measured. To measure the bottom current. fullwatercolumn current proﬁle. The ADCP measuring the midcolumn velocity is mounted on the rig or near the surface and directed downwards. the data of the upwardlooking ADCP is combined with the downlooking and near surface data. to form the a continuous. midcolumn and bottom current (www.rdinstruments. 71 . including near surface.
The RPR is calibrated with other available position reference measurements (Høklie et al. the tension force can be calculated. which computes the position..3 Tension and Payout The top tension can be found by measuring the pressure inside the pneumatichydraulic pistons. either electronic riser angle (ERA) or acoustic riser angle (ARA) measurements. such as the local hydroacoustic position reference (HPR) system and the global navigation satellite system (GNSS). 2007). The GPS system provides position measurements with an accuracy less than 1m in the horizontal direction. Galileo. The position estimation is based on measured top tension and the top and bottom riser inclinations. We may therefore calculate larger end inclinations in this study than what is actually experienced in the industry. see Fossen (2002) and references therein. 5. estimated to begin operational services in 2010 (European Commission.2 TLP Motions Several position measurement systems are commercially available. is under construction. The two commercial satellite systems available today are the US system NAVSTAR GPS (NAVigation Satellite Timing And Ranging Global Positioning System) and the Russian GLONASS (GLObal NAvigation Satellite System). In addition a European satellite navigation system. and the resulting system is known as a strapdown inertial navigation system (Fossen.5. velocity and attitude from the measurements. a GNSS is used as a reference for the position. To prevent the system from drifting. An inertial navigation system (INS) consists of a measurement part and a software part. The software part is a stateobserver. Recall that it in the industry is common to use bending stiﬀeners to prevent large rotations in the connections. The measurement part is called an inertial measurement unit (IMU). This is an accurate method down to the mmlevel. measuring the linear and angular accelerations. To ﬁnd the acceleration and velocity of the TLP we can use available commercial inertial measurement technology.3. It should also be noted that the company SeaFlex in cooperation with Kongsberg Maritime has developed a riser position reference system (RPR) which is an independent riser position reference system for a dynamic positioned drilling vessel. The riser end connections are in this study modeled with free rotations. 2002).3.4 Top and Bottom Angles The top and bottom angles for each riser could be measured by inclinometers. The payout is found by counting magnetic coils along the piston. 2002). 5. which is the typical accuracy for ship positioning systems today. This simpliﬁcation is 72 . When we know the area of the piston stamp.3. The Riser Control System Overview 5.
spatially separated. We will later on denote this as regimes. which is satisfactory for the sampling time here. When more transmitters are used. there could be large variations in the requirements for the controllers depending on both the operational conditions. riser properties and the water depth. 5.5 Relative Horizontal Distance To determine the relative horizontal distance between two risers. In addition. When installed in deep waters. which listens without transmitting. like the TLP. The time from transmission of a pulse to the reception is measured and converted to distance when we know the speed of sound in water (approximately 1500m/s). it is called a multistatic operation (Wikipedia. passive sonars. one nonlinear controller covering all possible regimes or a combination of various controllers into 73 . which use measurements as inputs to decide which action to take and calculate the feedback. In water this technique is known as sonar (SOund Navigation And Ranging). For a travel distance of 15m. and the TLP will experience large LF and WF motions. 5.7. Here. the time will be 10ms. diﬀerent static and/or dynamic properties of the risers in the array should be taken into consideration in the controller design. and active sonars which creates pulses (pings) in order to produce echo. For permanent installations. also when designing the control system. Hence. see Section 4.3. control of the top tension is introduced. 2007b). The active sonar consists of a transmitter and receiver. or there could be a model of the system (CPM) running in parallel. In the North Sea the weather conditions can be harsh. while the second approach is used in a more realistic case scenario. one could use acoustic location. This dynamically control output could either be directly based on one or two of the measurements presented above. and in deep waters this simpliﬁcation will only introduce small errors in the end point position. the allyear weather changes must be taken into consideration. Marine systems of today are required to operate in a wide range of environmental conditions while still giving acceptable safety and performance. the ﬁrst approach is used in investigation of the various control strategies. which again aﬀect the risers.4 Controller To prevent collision between risers. These could be placed at the same location (monostatic operation) or at separate locations (bistatic).Controller made as the main objective is to calculate the global geometry. There are two types. There may be two implementations of controller designs. the risers are more ﬂexible than in shallow waters and behave more like cables.
5. σ. (2007) have proposed a hybrid control structure for dynamic positioning systems of marine vessels. w. The work on supervisory control in this thesis is motivated by the work of Nguyen (2006).5). forcing the risers to behave as desired. are used as inputs in the supervisor and the controller. (2003) and references therein. one. u. the riser array. The two main blocks in the supervisory switched control system are the supervisor and the controller set (Fig. the drawback could be chattering between controllers. Nguyen (2006) and Nguyen et al. is then sent to the actuator. The correct controller is activated. The second solutions is a concept of supervisory control switching between a set of controllers.5: The hybrid control system. The supervisor monitors the conﬁguration of the diﬀerent risers and decides which controller set to use. More general solutions are therefore developed to secure stability. The ﬁrst solution may give a complicated controller.5. aﬀects the physical marine system. 74 . This switching between controllers may lead to instability. consisting of the TLP. Hespanha et al. A hybrid system often consists of continuous controllers and discrete switching between them. The measured values. On the other hand. Designing a controller for each regime is easier than one for all regimes. which could be almost impossible to design if the dynamics changes considerably. is sent to the controller. y. This strategy is called supervisory control. Hespanha and Morse (2002). Each controller is made for a prespeciﬁed operational regime. the actuator and the measurements. and complex logics and synchronizing functions handling the switching. and the output. A systematic approach can be found in Hespanha (2001). and makes it possible to switch between the diﬀerent controllers corresponding to the current operational condition. The block representing the environment. The controller switching signal. The Riser Control System Overview y Supervisor w Environment Controller Set u System y Figure 5.
Controller The supervisor and the controllers are described in detail in Chapter 6. 75 .
.
R2 is in the wake of R1. keeping it at a desired distance. (6. at one or more predeﬁned depths. 6. All in all this describes the controller architecture and its components used in the simulations. 2. whereas the second principle uses measurements of the relative horizontal distance between the risers.1 Control Objectives The control objectives may be formulated based on two diﬀerent principles subjected to available measurements. R1 will 77 . such as payout and top tension. The concepts and properties for a switched system in general are presented. This includes speciﬁcations of the diﬀerent control objectives and a description of the riser operational conditions. Keeping the tensions constant and equal in both risers may lead to the following scenario: For the two risers in a tandem arrangement.1) where Tj is the tension of riser j = 1. the controller and their components are given. This can be measurements at the TLP wellhead. The ﬁrst control objective principle is based on wellhead measurements. A more in depth deﬁnition of terms and description of the supervisor. 6. but they could easily be extended to an array of risers. Due to the shielding eﬀects on R2. For simplicity the objectives are presented for two risers. or available measurements along the riser.1 Control Objectives Based on Measurements at the Top Today the top tension is kept close to constant by a passive heave compensation system for each riser T1 = T2 .Chapter 6 The Controller Architecture This chapter aims to give a more thorough presentation of the controller architecture.1.
1 a). proposed by Huse and Kleiven (2000). (6.c) has shown that due to this tension variation. (2007b. Another strategy is equal payout by connecting all risers to a common frame.6. equal payout (b). Equal payout by adjusting the top tension through automatic control is studied in this work.1: Eﬀect of equal tension (a). (6. the deﬂection of R1 will exceed the deﬂection of R2.3) . This will give varying top tension on the risers depending on the drag forces and the position in the riser array. see Fig. and the two risers may collide.2) where ξj is the payout of riser j. experience larger drag force than its downstream neighbor. two equal risers will experience diﬀerent length due to axial strain according to ∆lR = 78 ∆T lR0 . 6.1b). The Controller Architecture T1 = T2 ξ1 = ξ2 ξ1 + lR1 = ξ2 + lR2 Figure 6. 6. If both risers have the same top tension. and equal eﬀective length (c). with the control objective ξ1 = ξ2 . EA ∆T = T1 − T2 . Further work proposed in Rustad et al. shown in Fig.
By introducing this way of controlling the top tension. To summarize. 6. In addition to the eﬀect of axial strain increases for longer riser lengths and with larger tension variations.5) where lRj (T0j ) is the initial length of riser j with the initial tension T0j . However. the sum of payout and riser length should be equal such that the controller objective can be formulated as ξ1 + lR1 = ξ2 + lR2 . We therefore propose a new control strategy letting the risers have equal eﬀective length. Hence. riser material or ﬁlling.Control Objectives where ∆lR is the length variation due to the diﬀerence in top tension ∆T of the two risers. The riser length can be formulated as lRj = lRj (T0j ) + ∆lRj. the risers may be placed with closer spacing without increasing the risk of collision. Equation (6. EA (6. (6.5) is a good estimate on the riser elongation due to tension. For two or more risers with diﬀerent characteristics with respect to diameter. an increase ∆T in top tension will give the same increase for all elements along the riser. see Fig. This means that in contrast to the strategy of equal payout we also compensate for the axial elasticity due to the tension variation. we propose three diﬀerent control objectives based on top measurements: 1. Equal tension: T1 = T2 . The riser material is assumed linear as long as the tension is much smaller than the yield stress.4) could be then be rewritten as ξ1 + lR1 (T0j ) + ∆lR1 = ξ2 + lR2 (T0j ) + ∆lR2 . ∆lRj = Tj − T0j lR0 . By using automatic control of the heave compensators and top tension. ∆lRj is the elongation of riser j relative to its initial tension. 79 . 2. Equation (6. otherwise this diﬀerence needs to be included in the equation.1c). Equal payout: ξ1 = ξ2 . and (6. The risk of collision increases with increasing depth. 3.3) is a simpliﬁcation valid for equal risers. (6. lR0 is the untensioned initial riser length.6) The initial length and static payout can be found individually.4) where lRj is the length of riser j. a more general expression is needed. Applying equal payout collision can still occur. Equal eﬀective length: ξ1 + lR1 = ξ2 + lR2 . but less frequent and in a smaller riser segment than for the equal tension strategy under the same environmental conditions. note that the payouts need to have the same initial positions for the equation to be valid.
∆xR12. Also. a system model is required and additional measurements of TLP motions. the distance between the risers ∆xR12. 6. This may introduce some scattering in the vertical direction if the reference is not slow enough. we do not need the additional measurements or a good process model for the linear controllers. reliable and accurate measurements of payout and tension are available today.1. 80 . However. with predeﬁned location along the riser.1.7) The measurement should be placed where the risers are likely to be closest. which is determined by the current proﬁle. the horizontal distance between them should be constant and equal along the entire length of the risers. We will in this work focus on the ﬁrst principle of control objectives. In order to achieve near parallel risers. such that this may not be a problem after all. Hence. the measurements and the actuator (top tension force) are found at the same location.2 Control Objectives Based in Measurements Along the Riser The second control objective principle is based on available measurements of the relative horizontal distance between the risers. (6. thus preserving passivity properties of the closed loop system more easily due to collocated control. Furthermore.m (z) = ∆xd . These are used to calculate the smallest relative distance between the risers.1. Note that we in this control objective do not need to consider the elasticity of the riser material directly. If supervisory switched control is used. For the second principle of control objectives. but the horizontal direction will still be stable. as described in Section 6. To calculate the total riser length as a function of time and tension. With only one measurement along the riser length. we do not know if these are placed at the water depth where the risers are closest. Hence.1. this setup is not collocated.m (z) should be equal to the distance at the top and bottom being the desired distance ∆xd .6. the relative distance between the risers is measured directly.3 Discussion For the ﬁrst principle of control objectives. the system is much slower in the horizontal than the vertical direction. The Controller Architecture 6. the initial riser length corresponding to the initial tension needs to be known. The drawback of this method is that we are limited to a ﬁnite number of measurements. undisturbed current velocity proﬁle and a good model for the hydrodynamic interaction are needed. As the measurements will be along the riser and the actuator is located at the top end.
The most important parameters aﬀecting the dynamics of a steel riser are: 81 . The risers in an array may often have diﬀerent physical properties. • Import. • Workover/Maintenance.2 Riser Operational Conditions The changes in the environmental conditions. Some typical riser applications are: • Drilling. The external diameter. A speciﬁcation of the different ROCs and the controllers based on these leads to a supervisory system where the time parameters and the necessary controller components are included according to the prevaling ROC. All risers referred to in this work are vertical steel risers. The applications and properties of the riser. may require diﬀerent control algorithms during the lifetime of a TLP/riser system. depending on whether the riser is used for drilling.1 Riser Characteristics During drilling and production on oﬀshore ﬁelds diﬀerent riser types are used for diﬀerent operations and purposes. Together these factors decide the dynamics of the riser. • Injection. export or workover to mention some. • Export.Riser Operational Conditions 6. The risers are specially made for each purpose. • Production. giving diﬀerent properties which decides the risers physical behavior. (2006). production. also aﬀect the controller gains.2. wall thickness. The purpose of the control system is to prevent collision for all environmental conditions. This concept of ROC is motivated by the work on vessel operational conditions (VOC) by Perez et al. Flexible risers or SCRs are not considered here. the riser characteristics (RC). riser types and water depth. The aim of this section is to highlight the essential characteristics of the conditions that aﬀect the dynamics of the riser system and use this information to decide which control action to perform. in addition to the various riser types during operation and production. connected to a TLP. denoted as regimes or riser operational condition (ROC). 6. A classiﬁcation of the various regimes could help the design of appropriate controllers and smooth switching between them. both statically and dynamically. material and density of the internal ﬂuid could vary.
the cross sectional area. which together with the riser length is of major importance for the eigenperiods of the risers. a diﬀerence in the variation of the contents aﬀectes the total eﬀective weight and the eﬀective weight gradient. For two risers with diﬀerent external diameter. • Cross sectional area and wall thickness. • Top tension. depending on the diﬀerent dynamic properties and corresponding eigenperiods. The eﬀects of the riser properties are not further investigated here. Hence. In extreme cases the longest riser eigenperiod can be close to the typical LF motions. This could be seen statically as at which water depth the deﬂection is largest. diﬀerent cross sectional area will give diﬀerent elongation according to the stiﬀness EA. • Density of the riser contents. whereas the lower limit is given by the eﬀective weight plus a safety margin. Some of these parameters are closely related. but should be taken into consideration when deciding the control strategy and the controller gains. The top tension level is dependent on the weight of the riser. For another riser case. Recall that the drag forces on the riser are proportional to the external diameter. the ﬁrst eigenperiod could be close to the slowest WF motions. The eﬀective weight decides the tension level along the riser. but the lower limit is smaller for the riser with lighter contents. increased or decreased by the diﬀerence in the density of the internal ﬂuid between the risers.6. The upper tension limit is decided by the yield stress for steel. Hence. For two otherwise equal risers. 82 . • Elasticity.e. but equal weight per unit length. the upper tension limit is the same. The Controller Architecture • External diameter. the length of the riser and the density of the internal ﬂuid. Hence. A larger eﬀective weight will have its maximum deﬂection at larger water depths than for otherwise equal risers. In addition it should be noted that the current ﬁeld behind a riser and the shielding eﬀect. is depending on the diameter of the upstream riser. resonance may occur at diﬀerent frequencies. i. The cross sectional parameters also contribute to a weight diﬀerence. for the otherwise equal risers with diﬀerent contents. The riser elongation is proportional to the modulus of elasticity times the cross sectional area. the riser with the largest diameter needs a larger top tension to compensate for the horizontal displacement due to the drag forces. • Riser length.
6. T LP. waves and current. 83 (6. In harsh weather conditions the TLP will experience larger forces and motions and will induce more motions on the risers than in calm weather.Riser Operational Conditions Current Profile and velocity variation Strong Control Calm No Control Stationary TLP Dependening on Dynamic environmental conditions Initial Conditions Riser Characteristics Figure 6. acting on the riser array.2.2 shows how these attributes are linked to the demand for control. wind. Riser Characteristics (RC) refers to the static properties like riser length. Riser Characteristics where Current refers to the current proﬁle and velocity. diameters and crosssectional area as well as the density of the internal ﬂuid which may vary during the risers lifetime depending on the operation purpose. The diﬀerent RC along the outofplane axis decides which control objective to apply and an appropriate set of control gains.8) . Figure 6. Going to the right along the TLP axis will result in more motions and demand for a controller. The motion of the TLP due to the environmental conditions is the main parameter to compensate for. TLP refers to the motion of the ﬂoater which is aﬀected by the environment.2: Riser operational conditions and inﬂuence on the controller choice.2 Riser Operational Conditions The ROC may be described by a triplet of attributes for a given riser ROC := Current.
(6. The set of admissible process models considered in estimatorbased supervision is M := Mp . u. Hence. Detailed behavior of the supervisor or the controller set can be abstracted and we can concentrate on a small set of properties for the system. the further we move up and to the right. see Section 6. between switching times.6. This simpliﬁes the stability analysis of the system. The decision is made in the switching logic block. xE . (6. The Controller Architecture Large ocean currents give larger riser displacements. u. Estimatorbased supervisors continuously compare the behavior of the process with the behavior of the estimators or admissible process models to determine which model is best describing the actual process.10) q∈Q p and q are the parameters taking values on the set P and Q. In addition. which provides satisfactory performance for each model.9) p∈P Each process model maps to a controller.12) C := {xq = Fq (zq . A controller selection function χ : P → Q maps each parameter value of p ∈ P with the corresponding index q = χ(p) ∈ Q of controller Cq which provides satisfactory performance when connected to model Mp . respectively. the process is connected to one of the candidate controllers only. y) : q ∈ Q}. the larger is the demand for a controller. the characteristics of the current proﬁle decide the riser conﬁguration and where the deﬂection is largest. y). y). ˙ 84 ep = yp − y : p ∈ P}.3 Switched Systems . u = Gq (zq .4. Hespanha and Morse (2002) and Hespanha et al. y). Going upward with increasing current velocity and possibly large changes in the velocity ﬁeld may also require a controller. yp = CE (p.Concept and Properties This section is based on Hespanha (2001). Hence. The main idea with the supervisory switched controller is to automatically switch between diﬀerent controllers depending on the situation. Note that the supervisor and the controller are decoupled. 6.11) (6. The estimator and controller may be described by linear or nonlinear systems according to M := {xE ˙ = AE (xE .3. The controller set is denoted C := Cq . . and the dynamics of the supervisor is not visible in the closedloop system. (6. (2003).
The decision logic compares all the estimation errors and ep can be regarded as a measure of the likelihood that the actual process is inside the ball Mp . ep is the estimation error vector. This means that ep is small whenever the actual process is inside the corresponding Mp .13) (6. Matching Property means that the set of estimators should be designed such that each particular yp provides a good approximation to the process output y. The mapping from the process switching signal to the controller switching signal is written σ = χ(ρ). p ∈ P. the following definition of a switched system (Hespanha. the switched system must be detectable with respect to the estimation error ep when the value of the switching is frozen at σ = χ(p) ∈ Q. 6. the corresponding p ∈ P describes the ongoing process best and is mapped to the controller switching signal σ = χ(p) ∈ Q. According to the certainty equivalence principle. 2001) is used ˙ x = Aσ (x. The small error 85 . w). yp is the output estimate vector. w). and w is the environmental disturbance. (6. This could for instance be dwelltime switching or hysteresis switching.3. the index p ∈ P is the likely value of the parameter.3. To prevent chattering. ∞) → P is introduced to indicate the current estimate ρ(t) ∈ P of the index p used in the feedback loop. The two basic properties for the switched system are the matching and detectability properties. a process switching signal ρ : [0. 6. and zq is the state vector of the controller. ep = Cp (x. u is the control input vector. a delay is introduced in the switching process. the multicontroller and the multiestimate.2 Switching Logic The index σ of the controller in the feedback loop is determined by the switching logic which takes the estimation error vector ep as an input.1 System Properties For the formal stability proofs of supervisory switched system. Note that this output for the switching logic is the one that determines which controller should be used. Detectability Property means that for every ﬁxed estimator. when a particular estimation error ep is the smallest element in the error vector ep . Since the value of p which corresponds to the smallest ep varies.Concept and Properties where xE is the estimation of the state vector.Switched Systems . Hence. y is the process output vector.14) where x denotes the state vectors of the process.
is compared to the other monitoring signals. the system switches.3.6. and ρ = arg min µp returns the index of the smallest monitoring signal µp . 6. and the concern of too much switching which may violate the nondestabilization property. by the switching logic. This property is satisﬁed if: • The switching is slow on the average by using dwelltime switching logic which strictly is used among linear models and controllers. The Controller Architecture and nondestabilization properties need to be satisﬁed by the monitoring signal generator and the switching logic. the actual value is held. i. When a new signal µp times the contribution from the hysteresis is smaller than the prevailing signal µρ . to be smaller than a constant times the norm of the smallest ep . 2000). h is a positive hysteresis constant. NonDestabilization Property The switching signal σ is said to have the nondestabilization property if it preserves the detectability in a timevarying sense.3. p ∈ P. • The switching stops in ﬁnite time using scaleindependent switching logic which can be used in switching between both linear and nonlinear models and controllers. µ(0) > 0. The hysteresis switching logic in general slows down the switching based on the observed growth in estimation error vector ep instead of for a ﬁxed dwelltime. a) → [0.3 ScaleIndependent Hysteresis Switching Since the modeling of the riser system is nonlinear. 1 86 . µρ . A continuous function α : [0. For the switching logic there is a conﬂict between the desire to switch to the smallest estimation error to satisfy the small error property. ∞) is said to belong to a class K if it is strictly increasing and α(0) = 0 (Khalil. ˙ (6. 6. Otherwise. Small Error Property means that there is a bound on eρ in the terms of the smallest signal ep for a process switching signal for which σ = χ(ρ).15) where λ is a nonnegative forgetting factor. The norm of eρ should be guaranteed. the scaleindependent hysteresis switching based on Hespanha (2001) will be used. γ is a class Kfunction1 and · is a norm. The monitoring signal for the prevailing model. if the switched system is detectable with respect to the switched output eρ for ρ and σ = χ(ρ). The monitoring signal µp for each process is based on the norm of the estimation error and is deﬁned by Hespanha (2001) as µp = −λµp + γ( ep ). The switching procedure is illustrated in Fig.e.
The model which best describes the ongoing process is used.Switched Systems .Concept and Properties Start ρ := arg min µp σ := χ(ρ) Do switch Do not switch Yes µρ ≤ (1 + h)µp . Observers or state estimators are aiming to obtain an estimate of the current state of the dynamic system by using available measurements online (each time step). the matching and detectability properties are important for the multiestimator and the multicontroller. ∀p No Figure 6. consists of a set of admissible models which are compared to the actual process. The smallerror property makes sure that the selected controller is the best.4 Model Concept Deﬁnitions The supervisor described in Hespanha (2001). 2001). To summarize. Nondestabilization will prevent chattering by providing switching logic such as hysteresis switching. can not be directly transferred to the setup and mathematical models used in this work.3. 6. An observer often copies the dynamics of the system and add 87 . the diﬀerence between an observer and an independent model needs to be clariﬁed. Furthermore. fast switching may aﬀect the stability of the system.3: Scaleindependent hysteresis switching logic (Hespanha. The terms admissible process model and actual process used in their works. Hespanha and Morse (2002) and Hespanha et al.3. respectively. When the process is between two regimes. and also given here in Section 6. (2003).
For any p ∈ P we have that Nσ (τ. Notice that also here the model will be updated based on available information and measurements. t). This follows a similar approach as in Reite (2006). The second part with the small error and nondestabilization properties are of importance. The matching and detectability properties are deﬁned for observers.5 Switched System Stability Analysis The scaleindependent hysteresis switching logic guarantees the nondestabilization and the small error properties by the following theorem: Theorem 6. corresponds to what we denote the physical process or the process plant model (PPM). What Hespanha (2001). The main diﬀerence is found for the supervisor. Hence. the matching property with respect to the accuracy of the model and the boundaries for the diﬀerent regimes are meaningful. log(1 + h) (6.1 (ScaleIndependent HysteresisSwitching. Hespanha (2001)) Let σ be the switching signal and Nσ (τ. is deﬁned and calculated for each regime p ∈ P. this is not done by an injection term. Let P be a ﬁnite set with m elements. t > τ ≥ 0 be the number of discontinuous of σ in an open interval (t. 6. covering all the operational regimes.16) . in sense of global exponential observers.1. Independent models are also used to estimate the states of dynamic systems. They are therefore dependent on accurate system models and inputs to calculate good results. t) ≤ 1 + m + 88 m log( µp (t) ǫ+e−λtǫ0 ) log(1 + h) + mλ(t − τ ) . denoted ACPM (accurate control plant model). and are not directly applicable for the independent model. However. the desired distance between the risers is deﬁned according to the operational regimes. However. The Controller Architecture an injection term constructed such that the state estimate should be a reconstruction of the unmeasured state. corresponding to the estimation error. even though the observer properties are not relevant.3. The switching logic is used. These models do not use online measurements to correct the states of the system. The states are estimated based on inputs and the system model. τ ).3. Hespanha and Morse (2002). (2003) refer to as the actual or ongoing process. In this work we have used only one nonlinear model type. Depending on the deviation from the desired riser conﬁguration.6. this independent process model (ACPM) and the resulting error models ep correspond to the set of admissible process models or multiestimators in the deﬁnitions in Section 6. a model error ep . Hespanha et al. However.
see Hespanha et al. See Hespanha et al. 6.17) guarantees the small error property. the monitor and the switching logic. 6. For details on the stability analysis.Supervisor y ACPM Monitor Switching logic Supervisor Model Controller Set No control Controller 1 Environment w u Marine System Stationary Controller Controller 2 Dynamical Controller Controller 3 y Error State Manuevering Controller 4 Figure 6. (2000).4 Supervisor The supervisor’s main task is monitoring of the riser array. (6.16) guarantees the nondestabilization of switching. ǫ0 are nonnegative factors. deciding which control action to perform and trigging the correct controller. (2000).17) where ǫ. As inputs it takes the measured 89 . while (6. The supervisor consists of three parts.4. The ACPM calculates the conﬁguration of the risers and is the most accurate estimate of the riser system behavior. see Fig. Proof.4: Structure of the switching control concept. the riser model. with at least one of them strictly positive. and 0 t e−λ(t−τ ) γp ( eρ (τ ) )dτ ≤ (1 + h)mµp (t) (6.
The switching decides which controller to use and which action to perform based on the inputs from the monitor. velocity and acceleration. iterative FEM model of the riser system based on the PPM from Chapter 3.4. (4) the risers are subject to fast and large disturbances due to TLP motions or currents. the TLP motion. When we know these inputs.1 Accurate Control Plant Model The ACPM is a nonlinear.2 Monitor The diﬀerent regimes can be divided into ﬁve: (12) normal regimes where all riser nodes are close to their desired positions. It is characterized as an independent model. (5) the risers are in a (near) collision situation called an error state. 90 . the TLP motions and the top tension for each riser.6. However. The ACPMs are simulated with fewer elements (812) than the PPMs (1520). the position of R1 and the undisturbed current. For R2 we know its top tension. the inputs and procedures are similar. in terms of an injection term. For the risers further down the array. The current acting on R2 is a function of the incoming current on R1 and the distance between the two risers. This is further analyzed and discussed in Chapter 7. This allows the ACPMs to run faster. The monitor is continuously comparing the calculated riser conﬁguration data from the ACPM and the measured tension with a set of values describing the diﬀerent regimes. a monitoring signal µp . The Controller Architecture undisturbed incoming current. 6. It is used to accurately estimate the state of the riser system and its behavior. and the top tension for each riser. 6. to update its estimated states. Based on the regimes and the deviation from the desired riser conﬁguration. as it is not taking any measurements. or the deviations are small and the tension variations have been small for a given time period. As inputs it takes the measured undisturbed incoming current. One model is running for each riser. the number of elements used for the ACPM is depending on the accuracy needed in the monitor. we can easily calculate the conﬁguration of R1. the TLP position. is calculated.4. (3) the risers have a medium deviation from the desired conﬁguration due to slow variations from LF TLP motions and tide. but some of the accuracy is lost. We will therefore need iterations to ﬁnd the right position of R2. p ∈ P is the regime index number.
The switching conditions for the error calculations are found in Table 6. depending on the regime. In some cases additional conditions on the horizontal R2 velocity. The model error vector ep used in the calculations of the monitoring signals is mainly found from the maximum deviation from the desired riser distance. (6. R2d . t0 )max are included. where ∆xd is the vector of the desired distance between the risers in each node.Supervisor R1 R2 R2d ∆xR12 ∆xd ∆erel Figure 6. D. where how fast they are moving towards each other is of main importance. The maximum deviation along the riser is deﬁned as ∆erel ∞ .18) (6. ∆xR12 is the horizontal distance between two neighboring risers. or the maximum tension ˙ variations for a given period of time ∆T (t.1 is based on a desired relative riser distance ∆xd = 15D. The monitor comparing values in Table 6. For a riser segment. and xRj is the horizontal position vector of riser j. in the riser segment shown here. The switching conditions are mainly deﬁned by the deviation from the desired horizontal distance between the risers ∆erel = ∆xd − ∆xR12 .1. This number will be small in most cases since the risers move 91 . xR2 . which is the distance at the top and bottom end points used in the simulations here at 1200m water depths. 6.5. The relative velocity vector ∆erel tells how fast the risers are moving towards ˙ or away from each other. The max deviation is given a value in diameters. the scalar values are found in Fig.19) ∆xR12 = xR2 − xR1 .5: Actual and desired riser positions. ∆erel is the (scalar) deviation from the desired R2 position.
p = 1 This regime is deﬁned as when the risers are parallel. p = 2 In addition to the deviation for the desired relative horizontal distance shown above. it may not e lead to an instant collision. such that the variance in position will be captured by considering the relative distance. the maximum deviation from the desired position determines the error and the monitoring signal. t0 )max ≤ 20kN ∆erel ∞ = 2. By considering only xR2 . Normal 1. and t0 is the oldest time for which information is kept. The Controller Architecture Regimes Normal 1 Normal 2 Slowly varying Fast changing Error state Monitor comparing values ∆erel ∞ = 0D ∆erel ∞ = 1. this could be captured both by considering ∆˙ rel and xR2 . since the ˙ relative motion might be small. 6. t is the current time.20) ∞. On the other hand. The model error closest to the deﬁnition of each regime is the smallest. the fast TLP motions are ˙ captured.5D & ∆T (t.1 gives the following model error and γfunction: e1 = ∆erel − 0D. if both risers are moving fast.6m/s ˙ ∆erel ≥ 13D µp µ1 µ2 µ3 µ4 µ5 ρ 1 2 3 4 5 σ 1 1 2 3 2 Table 6.1: The switching conditions. Normal 2. the variation in tension is also included. slowly.5D ∆erel ∞ = 2. ses Fig. ∆erel would not necessarily capture these motions. t0 ) is the tension for a given time window.6. for instance t0 = t − 100s. The function T (t. we will consider xR2 . if ∆˙ rel is small. If R2 is exposed to varying drag forces due hydrodynamic interaction. If the 92 . γ1 = γ( ∆erel ) = ∆erel (6.6. ˙ The scalar model error ep for each regime p ∈ P is given below. for instance due to WF TLP motions. Furthermore.5D & xR2 max ≥ 0. For activation of the e ˙ dynamical controller. (6. and its index number is used to decide the controller. Using the monitoring comparing values from Table 6. with the desired horizontal distance between the nodes at the each water depth.21) Hence.
this regime does 93 . the error from the relative horizontal distance is the only contribution. t0 )max Tmin t0 t Time Figure 6.5D gives e2.15. t0 )max .∆x = 0.∆x ] .T )2 + (e2.T γ2 = γ2 ( e2 ) = where kT = 1e−5 . tension diﬀerence between maximum and minimum tension in the time window is smaller than the boundary. t0 )max − T (t.T > 0. t0 )max ≤ 20kN.27) = 0 for ∆T(t. the error contribution for tension will be e2. the errors are weighted and summarized. ∆erel − 1.22) (6. e2. ∆T (t.5D e2.6: Time window for the top tension. For a smaller tension deviation.23) (6.T e2.2.T = kT ∆T (t. such that for ∆T (t. Slowly varying. whereas the contribution from the relative horizontal distance with deviation 0. the error contribution from tension is set to zero. t0 )max > 20kN.26) (6. T (6. (e2. t0 )max = T (t. p = 3 The slowly varying regime is for LF motions and tide giving medium deviation from the desired relative distance. e2.∆x = e2 = [ e2. t0 )min . Otherwise.Supervisor Top tension T (t) Time window Tmax ∆T (t. else ∞.∆x )2 . Like for the regime normal 1.25) (6. here 20kN.24) (6.
The Controller Architecture only consider the relative horizontal distance.32) (6. p = 4 For the fast changing regime. 6.15). xR2 . The contribution from relative distance is weighted to be of less importance. Hence.3 Switching Logic The switching logic decides which controller to use through the controller switching signal σ.5D 2 = [ e4.max .max = xR2 ∞ . Hence. This could be written e5 = ∆erel − 13D.33) = e4. γ3 = γ3 ( e3 ) = e3 (6. the resulting error will be chattering between zero and the calculated value. This can be monitored through the horizontal velocity vector of R2. Since we want to capture the fast motions. we here investigate if the closest node is in a (near) collision situation. ˙ If xR2. p = 5 Instead of ﬁnding the maxdeviation from a desired conﬁguration.6.∆x e4. we have e4.35) γ5 = γ5 (e5 ) = e5 min .x = 0.6m/s − xR2. (6.x ]. 1 ∆erel − 2. Fast changing.max is larger than the boundary value given in Table 6. A lowpass ﬁlter is therefore introduced to give a mean error contribution from the velocity. ˙ ˙ Otherwise the error is given by e4. e3 = ∆erel − 2.1. which is included in this error model in addition to the relative riser ˙ distance. The largest horizontal velocity is found by ˙ (6. corresponding to the WF motions with periods T = 5 − 20s. Since the risers are ˙ ˙ changing the direction according to the speciﬁed motion from the TLP.31) (6. we use a highpass ﬁlter. The 94 . the risers experience faster motions due to the TLP motions or current.28) ∞. e4. based on the value of the monitoring signal µp in (6.∆x = e4 γ4 = Error State.x . ˙ e4 1 ∞.34) (6.5D.29) (6.x = 0. ˙ This regime will also be activated if the risers are far apart. We are interested in the largest velocity and how fast it changes.∆x  + e4.30) xR2.4. (6. for instance if the second riser has a large deﬂection.
λp is the forgetting factor for µp .15) on vector form we have µ = −λµ + γ( ep ).3) shows how the process switching signal ρ is found based on the smallest monitoring signal µp .3. 6.5 Controller Set The controller set’s main task is to prevent riser collision. The reference model generates the trajectory based on the guidance model. 4. Which controller to use is decided by the supervisor which sends the controller switching signal σ to the controller block such that the most appropriate controller is used in the feedback loop. To decrease the detection time when active control is needed.5.4 consists of a set of controllers designed for each of the regimes described in Section 6. a weighting function is introduced. To provide a smooth transition between the controllers.2.5 2 1.Controller Set scaleindependent hysteresis switching logic from Section 6. The active controllers comprise of the following components: The guidance block calculates the guidance trajectory for each riser in the array based on the control objectives. and the controller best suited for the current operational conditions is determined by the switching logic.1. the memory is slightly decreased by choosing λp slightly bigger for p ∈ {3. One controller is designed specially for each operational regime. 2005).1 Guidance The guidance trajectory determined by the actual control objective.5 ). Rewriting (6. The initial monitoring value is set to T µ(0) = 0. 6. This could be formulated as a maneuvering problem (Skjetne. 6. Depending on the control objective it takes relative distance. ˙ (6. λ is the diagonal matrix of non˙ T negative constants (here λ = diag 1 1 1.3 is used.4.1. the sequence and how fast the risers are tensioned are of importance. 5}. Other inputs in the controller block are the measured top tension and payout. To prevent collision in an array of risers during transit to reach the desired tension. 6. and is a subject for further research. The mapping σ = χ(ρ) is found in Table 6. The controller block in Fig.36) where µ is the vector of monitoring errors.2 based on the ROCs from Section 6. A larger factor λ gives faster forgetting. while a smaller λ gives longer memory. and γ( ep ) is the resulting vector of the class Kfunctions.9 1 1 1 1 > 0. described in Section 6. The statediagram (Fig. payout 95 .
2 (T0. the guidance is written ξr.r (z) = ∆xd .51) later. The desired distance is ∆xd . The Controller Architecture and/or tension as inputs. In the guidance formulations below. indicates that when R1 approaches R2. we need to know the prevailing top tensions. a reference model is designed for the diﬀerent guidance models.5.6.1 (T0. T ) = ξ1 (t) + lR.2 Reference Model To provide a smooth trajectory and high performance.2 (t) = ξ1 (t). these principles are easily expanded to an array of risers.39) The error (6. the relative distance is reduced. The reference is introduced to calculate 96 . with R1 being the reference of R2. However. Relative Horizontal Distance Here we measure the relative horizontal distance between the risers at one predeﬁned water depth z.2 ) + ∆lR.2 (t. For the payout to be equal for both risers.38) (6.37) If R2 is used as a reference. one could be chosen as the leader. and the guidance is written ∆xR12. Equal Eﬀective Length When the riser elongation due to changes in top tension is included. in addition to the measured payout to the reference riser. the indexes are switched.1 (T1 ) − [lR. where ξ1 is the measured payout of R1 and ξr.2 (T2 )] . Using R1 as a reference for R2 we rewrite (6. (6.6) to ξr. If only R2 is controlled.2 is the guidance for R2. Equal Payout The equal payout guidance block needs the payout measured for the reference risers. and the payout for all the other risers could be calculated relative to this one. the initial top tensions and the initial riser length for both risers. (6.1 ) + ∆lR. 6. For an array of risers. the top tension of R2 needs to be decreased to maintain the same relative distance. we have assumed two risers.
j + τc.41) where ξd and its derivatives are the desired payout position. ∆erel. with a geometric and a dynamic task. This provides a smooth transfer between diﬀerent setpoints. For an array of risers.d + 2ζd ωd ∆erel. The geometric task is the path to follow.ref + ∆erel.r is the new reference coordinates in the same frame.d and its derivatives are the desired relative horizontal distance. Controller 1. σ1 is for normal conditions and low current velocities.5. To avoid this problem. (6. ζd is the relative damping ratio. The main objective of this block is to generate a safe and legal reference trajectory for the top tension to follow. while the dynamic task is related to how fast we follow this path. 1 1 ˙ ξref = − ξref + ξr . This is a subject to further study.41). velocity and acceleration trajectories. The following third order ﬁlter is demonstrated appropriate 2 2 ¨ ˙ ξd + 2ζd ωd ξd + ωd ξd = ωd ξref .j . but possibly at diﬀerent values for the two risers.40) (6.ref . The initial tension is equal to the pretension.2. to reduce unnecessary wear and tear. changing the tension too fast for one riser and too slow for another might lead to a near collision situation between risers not originally conﬂicting.ref = − ∆erel. ξr is the new reference coordinates in the same frame. ˙ td td (6.Controller Set a feasible trajectory for the payout or the relative horizontal distance decided in the guidance block. The total top tension in each riser is equal to the pretension. where the top tensions are kept constant. For the relative horizontal distance the corresponding third order ﬁlter is written 2 2 ∆¨rel.ref is the low pass ﬁltered coordinate.j . and ∆erel. The deﬁnition of the maneuvering problem is found in Fossen (2002) and Skjetne (2005). td td (6. This may be expressed as a maneuvering problem. velocity and acceleration trajectories. 97 . and td is the cutoﬀ period period of the low pass ﬁlter in (6. The tensions are equal to the previous controller in time. plus the contribution from the controller Tj = T0.3 Controllers Each riser has a pretension T0.r .d = ωd ∆erel. e ˙ 1 1 ∆erel.d + ωd ∆erel. we have to ﬁgure out how fast each riser should follow its trajectory. ωd is natural frequency. and ξref is the low pass ﬁltered coordinate.44) The controllers are designed for each of the regimes deﬁned in Section 6.43) where ∆erel.42) (6. 6.
j τc3. Controller 4. Hence.d − ∆xR12. ˙ (6. τc2.j = −KP.j − ξj .j are tuned for this control objective. For two risers only.j is the I.48) (6. The Controller Architecture Controller 2.j ej .j is the integrator gain. the controller algorithm and the calculation of the reference trajectory may be more complicated to avoid collisions.50) (6. ej dt − KD3.45) (6. and the derivative properties should be included.j τc4.j ej − KI. and KD3.j = const. j = 1.j and the actual payout ξj K for riser j.j ∆erel (z) = ∆erel.m (z). ej dt.j = TP 2.6.47) (6. KI.j and KI. 2.j = −KP 2.j TD3.j = τc2.46) (6.j ej − KI. the ESM controller can be the same PIcontroller as for controller 2. σ2 is for slow changes and stationary conditions. Controller 3. ej = ξd. calculating the new top tensions based on the payout.j . The integrator is the same in both controllers 2 and 3.j derivation gain. a PIDcontroller could be used.49) where ej is the error between the desired payout ξd. A PIcontroller could be appropriate. (6. For the ﬁrst control objective principle the controllers can be written τc1. σ3 is for faster changing conditions. It is used when we experience fault conditions and is structurally diﬀerent from the other controllers. For an array of risers.51) . σ4 is called error state maneuvering (ESM).j = KP 3. and integrator antiwindup is included to avoid saturation of the actuator. 98 ∆erel dt. For the second control objective principle based on the relative horizontal distance we propose a PIcontroller τc2. where KP.j = −KP 3. top tension and riser length.j ∆erel − KI.
In real life the switching may take longer time and is dependent on how fast the environmental conditions change.2 0. 6.j is the resulting controller.4 0. The switching between the controllers is tuned to approximately 20s here.8 0.5 0.6 0.j = α1 (t)τold + α2 (t)τnew .54) . Their sum is always 1.7: Weighting functions α1 and α2 .7 Weighting value 0. the feedback path will 99 .4 Transition To ensure smooth transition when switching between the controllers.53) (6.9 α (t) 0. 6. If the error is so large over a period of time that the integrator saturates the actuator.52) (6.3 0.1 0 0 2 4 6 8 10 Time [s] 12 14 16 18 20 Figure 6. where τc. The functions are seen in Fig. α2 = 1 − exp −10 ((t/20))2 . respectively.7.Controller Set 1 α (t) 1 2 0. α1 = exp −10 ((t/20)) 2 (6. a weighting function is used such that τc.5. α1 and α2 are weighting functions dependent on the time from the switching moment.5 Integrator AntiWindup A controller with integral action in combination with an actuator which may become saturated might give undesirable eﬀects. τold the previous and τnew the new controller output in time. 6.5.
56) where u is the controller output and v is the actuator output (Egeland. we can stop updating the integrator when the actuator saturates. 1997). The Controller Architecture be broken. The integrator may continue to integrate up to a large value. the error is zero. The error diﬀerence between the actuator output and the controller output times a gain is fed back to the system. ˚str¨m and Wittenmark. This could be written Kp 1 e(t) − [v(t) − u(t)]. or we can discharge it. uI ˙ = (6. When the error ﬁnally is reduced. 1993. the integrator value may be so large that it takes time to discharge it. The second method is implemented here. When the actuator is in saturation. A o 100 . this error signal works to discharge the integrator such that the controller output is at the saturation limit. To prevent it. Ti Ti v(t) = sat(u(t)).55) (6.6. When the actuator is not in saturation. This eﬀect is called integrator windup.
A low order model gives small system matrices and keeps the number of numerical operations down. The model had the same physical behavior as RIFLEX in all deep water cases. The model used in the veriﬁcations had 20 elements. 20}. For the purpose of control applications.. and could be said to be close to the real world and a good model of the riser process. both for the PPM and for the CPM. each of length 3m. N = {2. This is motivated by the realtime requirements for control systems. 10. and satisﬁed the desired level of accuracy. while large and fast computers cost more. A simple processor could then be able to run the model online. Generally. Small. This model was implemented in Simulink and veriﬁed by comparing it to the commercial FEM software RIFLEX (Fylling et al. 4. however. The RIFLEX model has 400 elements. 2005) in Chapter 4. ranging from 2 to 20. i. better performance is expected for more elements. and hence more easily 101 . 5. i. still describe the main physics.Chapter 7 Control Plant Model Analysis In Chapter 3. the CPM should be computationally fast. This is done to get the nodes at the same height above the seabed. As deﬁned in Chapter 3. The number of elements for the Simulink riser models are common multiples of 400. Diﬀerent means to measure the performance of a model are considered. 8. the purpose of the PPM is to describe the actual physical process as accurately as possible.1 Analysis Input Data and SetUp The riser top is assumed to be at the level of free sea surface. a mathematical model for a riser is derived.e.e. 1200m above the seabed. How many elements are good enough? And which cases might need more elements to keep the desired level of accuracy? 7. simple computers help to keep the costs to a minimum. In this chapter we investigate how many elements are needed to keep a desired level of accuracy. 16.
With a water depth of 1200m.2 QuasiStatic Analysis with Increasing TLP Position The quasistatic CPMs were run with increasing TLP position in steps of 5m.7. illustrated in Fig. from 0m to 70m. the element lengths were respectively l0 = {600. 60}m for the diﬀerent models and increasing number of elements.5m/s. The test setup was otherwise similar to what was used in Sections 4.24. • One year return period Ormen Lange current proﬁle. Three of these positions (0m.5. • Payout. 102 . especially of interest in the quasistatic analyses. 7. 150. The tension and the TLP motions are speciﬁed in each section.7m/s. Control Plant Model Analysis compare the horizontal displacements. 75. These 7 versions of the model are identical except for the number of elements. These could be summarized as: • Deﬂections/Deformation shapes. The simulation results for the RIFLEX model are the same as in Sections 4.24. The Simulink models are hereafter called CPMs. The measurement of the (absolute) horizontal positions along the riser is expensive and not accurate enough to be a part of the control loop. Two current proﬁles were run in all tests. diﬀerent measurements and methods were used. 120.2. • Riser top angle. 400. and are compared under the same environmental conditions. additional proﬁles were run to investigate and illustrate special phenomena. with surface velocity 1. 30m and 50m) were subject for more detailed investigations.15m/s and velocity close to the seabed 0. 240. These are: • Uniform current with velocity 0. Most of these methods are mainly of theoretical interest. 3. The top tension was kept constant at 1800kN. • Area under the curve. To investigate the error introduced with few elements. For some of the cases.5 and are used in the analysis to illustrate the assumed correct solution. • Error norms for horizontal positions. both top and bottom riser inclinations and payout are already measured today. On the other hand.
N −1 (7.1 Error Norms for Horizontal Positions The idea behind this method is to compare the horizontal displacement for nodes at the same vertical position.and l2 norms are dependent on the number of elements and are therefore divided by the number of nodes.1) where i is the number of elements. the sum of errors and the maximum error. which is clearest seen for the l1 .e. • l∞ norm . and the norms are close to zero when N ≥ 16.2. Both the l1 . which are always located at the same place. and is larger for N = 10 than for N = 8. The three norms used here are: 1.2) This makes it easier to compare the various models. 7. Fig. and 5 elements are all seen to give rather rough estimates of the riser conﬁguration. Two of the norms above could then be more commonly explained as: • l1 norm . accounted for the number of nodes can be written l1 : x 1 . the normalized l1 . the error norm is increasing. All the error norms converge to zero when N is increasing. (7. the deviation is even smaller. j is the height of the actual node.QuasiStatic Analysis with Increasing TLP Position 7. ∞ Here x is replaced by the horizontal error vector given by ∆xN =i (j) = xR (j) − xN =i (j). For more elements. l∞ norm: x 1 2 = (x1  + x2  + · · · + xn ). at N = 10. except for the top and bottom nodes. The same color code as in Fig. see Appendix A.1 is used throughout this chapter. xR is the RIFLEX solution. 7. l2 norm: x 3. which is opposite of what is expected. The models with 2. However. 4. 7. The diﬀerent error norms for this TLP position are given in Fig. Zooming in on the 103 .2.sum of errors in all nodes. n 1 2 = maxj xj . l1 norm: x 2. The maximum deﬂection error were found at 600m above the seabed.and l∞ norms. N −1 l2 : x 2 .1 shows the riser conﬁguration without TLP oﬀsets for all CPMs. Hence.e.5.and l2 norms. i. unless otherwise speciﬁed. and xN the horizontal positions for the CPM with N elements. all nodes for each CPM is compared to the RIFLEX model. = (x2 + x2 + · · · + x2 ). To be able to compare the models better. we apply metrical norms. The 10 elements model give a good impression of the riser curve.maximum displacement error. i.
7. TLP 0m 0.1: Riser conﬁguration without TLP oﬀset for an increasing number of elements exposed to the Ormen Lange design current.7 0.1 0 l2 l∞ 2 4 6 8 10 12 14 Number of elements 16 18 20 Figure 7. Norms for horizontal positions.4 0.2: The diﬀerent error norms for the horizontal nodes without TLP oﬀset. exposed to the Ormen Lange design current.5 0. Control Plant Model Analysis 1200 N=2 N=4 N=5 N=8 N=10 N=16 N=20 N=400 1000 800 Vertical position [m] 600 400 200 0 0 2 4 6 8 Horizontal position [m] 10 12 Figure 7.6 0.2 0.8 l1 0.3 0. 104 .
another number of elements could indicate to be the best. It is dependent on how good each node matches the RIFLEX solution for a given current proﬁle and number of elements.3) we see that the model with 8 elements match the RIFLEX solution better in the node points. To summarize. For other current proﬁles. see Fig 7.4). while N = 10 follow the curve better in general. Also note that each norm method gives approximately the same norm curve independent of the TLP oﬀset.5 11 500 450 400 Figure 7.QuasiStatic Analysis with Increasing TLP Position 850 800 750 700 Vertical position [m] 650 600 550 N=2 N=4 N=5 N=8 N=10 N=16 N=20 N=400 9. 7. using this method with error norms on the horizontal displacement only to quantify the performance of each model is not reliable.5 10 Horizontal position [m] 10. The diﬀerent norms show various sensitivity. riser conﬁguration (Fig. The l2 norm seems to be least sensitive to the single node error (Fig. 7.3: Zoomed riser conﬁguration without TLP oﬀset for an increasing number of elements exposed to the Ormen Lange design current.4. 105 .
4 0.8 0. For a TLP without oﬀsets. while the inclination should have been negative.8 0.8 0. 5} are all seen to give the wrong sign.2 Riser Top Angle Today the top and bottom riser angles are measurements available oﬀshore. which is the case for drilling risers. we would like to investigate the riser top angle as it tells something about the riser conﬁguration. Production risers usually have bending stiﬀeners in their top and bottom end connections. The riser inclination calculated here might therefore be larger than what could be measured for production risers in the industry. However. 4.4: The diﬀerent norms for the TLP oﬀsets. the deﬂection and the accuracy of the riser model.7) three models N = {2. 7.6 l2 norm 0. 7.4 0. 2002). Here we have used the angle given by the straight line between the two upper nodes to represent the top angle.2 0 2 4 6 8 10 12 14 16 18 20 TLP 0m TLP 30m TLP 50m 0.4 0.5 the riser top angles are given for the three TLP oﬀset.6 l1 norm 0.7. while our model assumes free end rotations. 7.2.6). For a TLP oﬀset of 30m (see Fig. These are used to calculate the riser conﬁguration and may be used as inputs to the DP systems of ﬂoaters (Sørensen et al. while for N > 8 the top angle is approaching the correct value. An alternative method would be to assume constant 106 . At a TLP oﬀset of 50m (Fig.6 l∞ norm 0. Høklie et al. Control Plant Model Analysis Norms for horizontal positions 0. two elements are too few to describe the riser conﬁguration and gives a positive sign. 2002.. all inclinations have correct sign. 7.2 0 2 4 6 8 10 12 14 16 18 20 TLP 0m TLP 30m TLP 50m 0.. exposed to the Ormen Lange design current.2 0 2 4 6 8 10 12 Number of elements 14 16 18 20 TLP 0m TLP 30m TLP 50m Figure 7. In Fig.
current proﬁle and velocity. these analyses show the importance of having enough elements in the model. especially when using inclinations in the measurements in the control feedback loop.5: The riser top angles with diﬀerent TLP oﬀsets.1. 7. this example is from the Ormen Lange current. 7.5 2 4 6 8 10 12 Number of elements 14 16 18 20 Figure 7.5 −1 −1.5 0 Top angle [deg] −0.5 TLP 0m TLP 30m TLP 50m 1 0. exposed to the Ormen Lange design current. with an oﬀset at 50m.5 −2 −2.3 Area Under Curve A better way to compare the performance of the models is to consider the area under the riser curves. The phenomenon described here is dependent on the TLP oﬀset. for each CPM can be found by summarizing the area of the trapezoids between x = 0 107 . However. recall Fig. The area under the riser curve. which is a most realistic case. It is more likely to be observed in a large oﬀset and medium to large current velocity. The max deﬂection will then be slightly larger than the TLP oﬀset and near the top. but converge to the same result for more elements. This would give a better angle estimate for few element CPMs.2. curvature in the upper element and apply the tangent at the upper node as the top angle. AN . Anyhow.QuasiStatic Analysis with Increasing TLP Position 1.
108 . 1200 1150 Vertical position [m] 1100 1050 1000 N=2 N=4 N=5 N=8 N=10 N=16 N=20 N=400 950 900 49 49.6: Riser conﬁguration with TLP oﬀset of 30m for an increasing number of elements exposed to the Ormen Lange design current.7.5 50 Horizontal position [m] 50.5 Figure 7. Control Plant Model Analysis 1200 1000 800 Vertical position [m] N=2 N=4 N=5 N=8 N=10 N=16 N=20 N=400 600 400 200 0 0 5 10 15 20 Horizontal position [m] 25 30 35 Figure 7.7: Zoomed riser conﬁguration with 50 TLP oﬀset for an increasing number of elements exposed to the Ormen Lange design current.
5 TLP 0m TLP 30m TLP 50m 2 1. For N ≥ 8.8). the relative error is less than 3% for all TLP positions. since all node values and not only the common multiplier are included in the calculations. 7. and the riser curve for each pair of nodes N AN = i=0 xi + xi+1 l0 .4) ARIF LEX where AR is the RIFLEX area for 400 elements.QuasiStatic Analysis with Increasing TLP Position 4.5 1 0. 2 (7. Shorter riser elements will capture the deﬂection better with more trapezoids. Hence. (7. This method is less dependent on how the selected nodes match the correct solution. The riser areas are seen to converge to the desired value for N ≥ 10 (Fig.5 Area under curve [m ] 3 2 2. and l0 is the element length.5 2 4 6 8 10 12 Number of elements 14 16 18 20 Figure 7. It should be noted that N = 10 gives better results than N = 8. which is expected. 7. The relative error under the curve is illustrated in Fig.9 and given by ARIF LEX − AN =i eA = · 100%.8: The area under the riser curves for TLP oﬀsets. this method gives overall a better evaluation of the convergence of the various riser models than the method 109 .5 x 10 4 4 3.3) where xi is the horizontal position of node i.
7. with horizontal error norms.4 Payout The payout is a direct result of the deﬂection and how well the CPMs correspond to the correct curve. 110 . while the area under the curve is a theoretical calculation. the payout can actually be measured and used in the feedback loop. applying Simpson’s method to ﬁnd the area would give better estimates.2. see Fig.2. The payouts converge to the desired values for increasing number of elements.7.11.10. However. the estimate of the area will converge to the same value for both methods. For CPMs with few elements. Control Plant Model Analysis 35 TLP 0m TLP 30m TLP 50m 30 25 Error area under curve [%] 20 15 10 5 0 2 4 6 8 10 12 Number of elements 14 16 18 20 Figure 7. It is similar to the method with area under the curve presented in Section 7. The static payout for the diﬀerent TLP oﬀsets and various models are shown in Fig. the payout is a good and robust method.3. For more elements. and at N ≥ 8 the payout error is less than 2%.9: The relative error in the area under the riser curves for TLP oﬀsets. Hence. 7. 7.
The horizontal error norms were sensitive to the prevailing current proﬁle such that a lower number of elements appeared to have better accuracy than the true situation. top angle. Too few elements could give the opposite sign on the riser top angle for some environmental conditions. Anyhow. horizontal error norms.95 1 2 4 6 8 10 12 Number of elements 14 16 18 20 Figure 7. one should be careful when deciding the number of elements in the model. only payout will be used. Calculation of the area showed good robustness.7 0. which is easily measured and can be used in feedback. All methods showed.2. payouts proved to be a robust and accurate method.10: The payout error for TLP oﬀsets. area and payout. For the rest of the analysis. 111 . The l2 norm was the least sensitive method. as expected.1.65 TLP 0m TLP 30m TLP 50m 0. When using inclinations in the feedback loop for riser control or DP operations.9 0.85 0. this method could not be used online as it requires accurate measurements of the horizontal positions. 7. This was especially seen for the l∞ norm and the l1 norm.75 Payout [m] 0. Finally.8 0. that increasing the number of elements gave a better model closer to the correct solution.QuasiStatic Analysis with Increasing TLP Position 0. The results are summarized in Table 7.5 Discussion Four methods have been investigated.
1: The pros and cons for the diﬀerent methods to quantify the performance of the CPMs. in addition to the uniform and the Ormen Lange currents. The riser models were exposed to the bidirectional current. The TLP was positioned in zero oﬀset. The physics of the riser behavior and the relations between the top tension. top position and the maximum deﬂection were explained in Sec112 .3 Analysis of QuasiStatic Tension Variations The quasistatic CPMs were analyzed with increasing top tension from 1200kN to 2700kN in steps of 50kN. Method Horizontal error norms Top angle Area Payout Robustness Sensible for single node errors and current proﬁles Wrong sign for low order models Robust Robust Possible to measure Theoretical Measurements Theoretical Measurements Table 7. Control Plant Model Analysis 20 TLP 0m TLP 30m TLP 50m 18 16 14 Relative Payout Error [%] 12 10 8 6 4 2 0 2 4 6 8 10 12 Number of elements 14 16 18 20 Figure 7.11: The relative error in payout for the diﬀerent TLP oﬀsets.7. 7.
Longer discontinuities were seen for longer 113 . the derivatives deviate noticeably from the RIFLEX solution.13. The element length is only 3m. The same is seen for the derivatives in Fig. For each step in tension the node with max deﬂection is found. N = {4.3.12 shows the relation between the maximum horizontal deﬂection as a function of the vertical top position x(z) and its derivative ∆x/∆z. For all these low element order CPMs. This node number will vary. 5} match better.5 Vertical top position [m] 1201 −5 −6 −7 −8 −9 −10 −11 −12 −13 −14 1199 The derivative of x(z) 1199. In this section we mainly focus on the error introduced by decreasing the number of elements. 20} have a sudden discontinuity.5 1200 1200. and its derivative (b) for the CPMs exposed to the Ormen Lange current.12b). depending on the tension. Fig. such that the change ∆z is much smaller. For larger tensions and higher top positions all CPMs match the solution better than for lower tensions. The change in vertical position ∆z is dependent on the length of the element.Analysis of QuasiStatic Tension Variations Displacement vs top position 22 Max horizontal displacement [m] 20 18 ∆ x / ∆ z [−] 16 14 12 10 8 6 1199 N=2 N=4 N=5 N=8 N = 10 N = 16 N = 20 N = 400 1199.5 1200 1200. For the three lowest number of elements. 7. The RIFLEX model does not have any noticeable discontinuities.” For a low top position and small top tension. 7. tion 4. 7. the curves match the RIFLEX solution satisfactorily.12: Maximum deﬂection as a function of top position (a). Fig. 7. This can be explained by looking at the snapshots in Fig. Note that the physical interpolation of ∆x is “How much will the max∆z imum horizontal deﬂection change for a unit variation of vertical position of the upper riser end.251m higher than the correct value.5 Vertical top position [m] 1201 Figure 7. such that the maximum horizontal displacement is too small. and is due to the numerical way of ﬁnding the derivatives. 10. For eight or more elements. This will give a drop in ∆x/∆z. the maximum deﬂections at the lowest tensions are 12m too small such that the riser end is assumed to be 0.12a) shows that 2 elements are too few. For all higher number of elements. the x(z)curves match fairly well with the RIFLEX solution. 16. The values of the derivatives for N = {5.
Maximum deﬂection versus top tension x(T ) and its derivatives are seen in Fig. The error is seen to de114 . The variation in tension from one step to another is constant. 7.13: Riser conﬁguration for 10 (opink). Due to this. Note that these discontinuities are due to numerical errors in calculation of the derivatives. ∆x is slightly larger than when it is kept at the same height. some smaller discontinuities were noticed in Fig. and the lower and upper tension limits. Three tensions are chosen. When the node for which the maximum deﬂection is changed. respectively. The three CPMs with least elements do not have the same maximum deﬂection as for the higher order models. 1200kN and 2700kN. The x(T ) curves all have the same shape. a medium initial tension (1800kN). At high tensions. the derivatives converge to the same value. Control Plant Model Analysis 1200 N = 10 N = 16 N = 20 1000 800 Vertical position [m] 600 400 200 0 0 2 4 6 8 10 12 Horizontal position [m] 14 16 18 20 Figure 7.14. 7. The absolute error in top position as a function of tension relative to the RIFLEX solution.7. (7. 7.14b).5) where zR is the RIFLEX top position and zN is the top position for each CPM. 16 (*green) and 20 (∆–blue) elements and their maximum deﬂection. is found in Fig.15 and given by ez(T )  = zR (T ) − zN =i (T ). element lengths. while the simulation model experience continuous derivatives.
18. In Fig. 7. Approximately the same value for ∆z/∆T is seen for all current proﬁles as it is only dependent on the elasticity of the riser material. crease with increasing number of elements and also increasing tension. For higher tensions.025 −0. z(T )N =2 is an asymptote which the other CPMs are approaching. This can be derived from 115 . For low tensions. 7. 7. ∆z/∆TN =2 is constant and the others are approaching this level. see Fig.005 ∆ x / ∆ T [m/kN] −0. the graphs are spread.015 −0. For high tensions.17.18b). The eﬀect on the top position by increasing the top tension is largest for low tensions.16b). and its derivative (b) for the CPMs exposed to the Ormen Lange current. The two element model does not capture the current forces and is vertical. In Fig. the increase in the top positions is mainly due to the increase in length.02 The derivative of x(T) 1500 2000 Top tension [kN] 2500 Figure 7. all curves approach the same asymptote. Neglected eﬀects in the CPM seem to be more signiﬁcant for low tensions. when the tension increases and the contribution to z from the geometry due to deﬂection decreases. 7. The only increase in top tension is therefore due to elasticity.03 N=2 N=4 N=5 N=8 N = 10 N = 16 N = 20 N = 400 0 −0.Analysis of QuasiStatic Tension Variations Displacement vs top tension 22 Max horizontal displacement [m] 20 18 16 14 12 10 8 6 1500 2000 Top tension [kN] 2500 −0. This is particularly obvious when considering the two element model with bidirectional current. 7. The derivatives are seen in Fig. where the geometric stiﬀness is dominating. For instance bending stiﬀness (EI) is more important for low tensions than for higher tensions. As expected.14: Maximum deﬂection as a function of top tension (a).16a) the relationship between the vertical top position and top tension z(T ) is plotted. For higher tensions the dominance of elastic ﬂexibility over geometric ﬂexibility is noticed.01 −0. The same is seen in Fig.
Vertical top position vs top tension 1201 1200.5 2 1.16: Maximum top position as a function of top tension (a).4 0. Control Plant Model Analysis Absolute Error  ez(T) 0.2 1200 1199.5 1 0.5 3 2.6 1199.8 1199.6 ∆ z / ∆ T [m/kN] 1200. and its derivative (b) for the CPMs exposed to the Ormen Lange current.1 0.8 Vertical top position [m] 1200.15: Error in top position for the various tensions.05 0 2 4 6 8 10 12 Number of elements 14 16 18 20 Figure 7.25 [m] 0.5 0 x 10 −3 The derivative of z(T) 1500 2000 Top tension [kN] 2500 Figure 7.35 0.7.2 0.45 1200kN 1800kN 2700kN 0.4 1200. 116 .15 0.2 1500 2000 Top tension [kN] N=2 N=4 N=5 N=8 N = 10 N = 16 N = 20 N = 400 2500 4 3.3 0.4 1199.
7. T0 + ∆T ⇒ ∆T ∆z ∆T = = = = EA (z0 − l0 + ∆z).0134m2 (7. (3.06e11 N/m2 · 0.7) (7.17 is also of interest from a mode shape point of view.Analysis of QuasiStatic Tension Variations Top tension 1200kN 1200 1000 800 Vertical position [m] N=2 N=4 N=5 N=8 N=10 N=16 N=20 N=400 600 400 200 0 −6 −5 −4 −3 −2 −1 Horizontal position [m] 0 1 2 Figure 7. FEM can be considered as a spatial discretization of the riser conﬁguration. Remark: The riser conﬁguration as shown in Fig.18b). 2. the slowest sampling frequency is decided by Shannon’s sampling theorem and the Nyquist frequency: If the sampling frequency is higher 117 .22).9) This value is the same as in Fig.35e−4 m/kN.6) (7. similar to what can be found for time varying signals.17: Riser conﬁguration for the CPMs exposed to the bidirectional current velocity proﬁle. l0 z0 EA 1200m = 4. The bidirectional current velocity proﬁle excites the second mode shape. For time varying systems. l0 EA ∆z.8) (7. 7.
none of the deﬂections are captured. 1995).4 8 1200. The other few element CPMs (N = {4. while the highest order CPMs (N = {16.4. and its derivative (b) for the CPMs exposed to the bidirectional current proﬁle. Figs.20 118 . and their change in payout is only half as much as for the RIFLEX solution. for TLP periods 60s and the Ormen Lange current proﬁle. 10}) follow the solution nicely. it can be reconstructed completely from the sampled signal (Ogata. For the highest order CPMs.8 Vertical top position [m] ∆ z / ∆ T [m/kN] 12 1200.19 a) shows that the ﬁrst mode corresponding to the TLP motion is dominating. and also the payout.4.4 Dynamically Moving TLP The dynamic CPMs were then exposed to harmonic TLP motions using the same setup as in Section 4. respectively. 7. All CPMs are close to the RIFLEX solution at large oﬀsets where the risers are straight. 7. than twice the frequency of the highest frequency component of the continuous signal. How many elements are needed to represent the main riser conﬁguration? For N = 2. This is the time reconstruction of a signal.2 6 1200 4 1500 2000 Top tension [kN] 2500 Figure 7.19 a) and b) show the payout versus time and TLP position. The TLP motion and setdown is seen in Fig. 20}) are hardly seen as they are on the curve for the RIFLEX solution.18: Maximum top position as a function of top tension (a). 7. The two element model does not capture the large deﬂection. the riser is drawn with fair accuracy. For the other low order CPMs.6a)) the deﬂection is increasing. 5}) capture the main deﬂection. The medium order CPMs (N = {8. 7. A parallel could here be drawn to our spatial model reconstruction. the main conﬁguration is seen. Control Plant Model Analysis Vertical top position vs top tension 1201 14 x 10 −4 The derivative of z(T) 1200.6 N=2 N=4 N=5 N=8 N = 10 N = 16 N = 20 N = 400 1500 2000 Top tension [kN] 2500 10 1200.7. The payout in Fig. When the TLP is moving from right to left (see Fig.
4. The error for N = 2 is up to 45% for the smallest TLP oﬀset with the largest deﬂection. especially for the low order models. At 300s.5 1. At larger TLP periods. The error from ﬁrst mode shape is also dominating for N = {4. the dynamical CPMs were analyzed with harmonically varying tension. The setup was the same as in Section 4. Such a large error on the ﬁrst mode shape dominates the second mode shape seen in Fig.5 2. Note that these results presented here are those with largest relative error. like 120s. the total errors are less than 8% and 5%. Smaller current velocities gave also smaller errors.19: Payout as a function of time (a) and TLP oﬀset(b) exposed to the Ormen Lange current. The payout error eξ  = ξR − ξN =i .Analysis of Dynamic Variation in Top Tension Piston payout 0. as the bottom of the cylinder in the tension system (blue. thick line). For the low order CPMs. the risers were close to quasistatic and only the ﬁrst mode shape was seen. The riser tops are plotted in the same graph. (7. the errors were smaller. For the medium and high order CPMs.5 N=2 N=4 N=5 N=8 N = 10 N = 16 N = 20 N = 400 20 30 40 TLP position [m] 50 2 2 2. the ﬁrst mode dominates. such that the distance between them are the payouts. 7.5. The simulations shown here are 119 . 5}.5 Analysis of Dynamic Variation in Top Tension Finally.21.6b) where the TLP moves from left to right.5 3 0 100 Time [s] 200 300 3 10 Figure 7. The second order riser mode shape is clearly seen in the riser tops and corresponds to the second order mode shape in Fig.10) is plotted in Fig. respectively.5 Payout vs TLP 1 1 Payout [m] 1. 7.5 0. 7. respectively.20. The ﬁrst mode shape errors are small such the second mode shape errors are observed. with errors of approximately 20% and 15%.
Maximum errors are seen for the low order CPMs.7. Control Plant Model Analysis Cylinder bottom and riser tops 1201. a given tension does not correspond to one top position. and 0.02m. The top position versus top tension is seen to be a hysteresis function due to relative velocity and drag.22 b) illustrates how an increasing number of elements result in a better match between the CPM model and RIFLEX.23 shows the relative payout error. Hence.5 1199 1198.3 and 7. Fig. i. but depends on whether the tension is increasing or decreasing as well. with a tension period of 120s and a riser exposed to the Ormen Lange current. 7.5 1201 1200.4.5 Vertical top position [m] 1200 1199. Fig. A small second order nonlinearity and phase shifts are seen. the deviations are smaller. 120 .5 Cylinder N=2 N=4 N=5 N=8 N=10 N=16 N=20 N=400 0 50 100 150 Time [s] 200 250 300 1198 Figure 7. For higher tensions. the maximum error is ± 0. 7.09m for N = 4. 7. Fig. the drag forces and diﬀerence in relative velocity when the tension is decreased and increased. This is expected based on the results from Sections 7.e.22 shows the top position as a function of time (a) and top tension (b). For the medium and high order CPMs. with approximately 0.20: Cylinder bottom (blue line) and riser top positions for TLP motions and Ormen Lange current. The deviations in top position between the models are largest for low tensions and low top positions.22m for N = 2. This is due to the nonlinear viscous forces.
4 N=2 N=4 N=5 N=8 N = 10 N = 16 N = 20 1.6 0.2 0 50 100 150 Time [s] 200 250 300 Figure 7.8 1199.22: Riser tops as a function of time (a) and tension (b) exposed to the Ormen Lange current.4 1200. 121 .6 1200.8 1199.6 1201 1200.8 1200.2 1 0.4 1200.6 1200.8 1200. Top position vs time 1201 1200.Analysis of Dynamic Variation in Top Tension Payout Error 1.4 0.2 1200 1199.2 1200 1199.8 [m] 0.21: Relative payout error for harmonic TLP motions and Ormen Lange current.6 1000 Top position vs top tension Top position [m] N=2 N=4 N=5 N=8 N = 10 N = 16 N = 20 N = 400 1500 2000 2500 Top tension [kN] 3000 0 100 200 Time [s] 300 Figure 7.2 0 −0.
05 0 50 100 150 200 Time [s] 250 300 350 400 Figure 7.3 N=2 N=4 N=5 N=8 N = 10 N = 16 N = 20 0. we investigated what happens if the number of elements is decreased. follow the main conﬁguration.7. This slows down the computation speed considerably for the time consuming matrix operations and iterations. The matrix operations are fewer.6 Discussion In Chapter 4. Control Plant Model Analysis Top Position Error 0. the accurate control plant model (ACPM) and the process plant model (PPM). These correspond to the fast control plant model (FCPM).23: Relative payout error for a riser exposed to harmonic tension variations with periods 120s and the Ormen Lange current. and follow the dynamics with 122 . where n is the number of riser segments. Hence. Recall that the size of the system matrices are 2(n+1)×2(n+1) giving 4(n+1)2 matrix elements. have small quasistatic deviations.05 0 −0. The main dynamic properties are still kept in this model. the possibility of realtime operations are of great importance. In this chapter. doubling the number of riser nodes increases the number of matrix elements by a multiple of four. 7. The medium order CPMs. with N = {4. The analysis in this chapter could be divided into three levels of accuracy. 10}. We therefore seek the smallest and least complex model which still aﬀord the desired level of accuracy for the current operation. The low order CPMs.1 0. which gives shorter simulation times. A small number of elements gives small system matrices.15 [m] 0. we veriﬁed the mathematical model. 5}. with N = {8.2 0. From a control point of view.25 0.
Hence. The proposed limits and relative errors are based on the worst simulated case. For slower TLP dynamics and smaller current velocities. These results are categorized in Table 7. 20}. Note that the number of elements for the FCPMs and ACPMs does not need to be a common multiple of the PPM. The high order CPMs with N = {16. 123 . the results are better with smaller relative errors.2: The categorized results of the CPM analysis. the limit for the diﬀerent model classiﬁcations could be set diﬀerently. such that the needed number of elements N are chosen to correspond to the physics of the environment. errors less than 8%.Discussion Type FCPM ACPM PPM N 46 812 ≥ 15 QS error 510% 25% < 1% Dyn error 1520% < 8% < 5% Description Main physics used in control analysis Good model for monitoring of the risers High accuracy for the riser process model Table 7. Increasing the number of elements gives a more accurate model at the expense of slower simulations due to larger system matrices.2. have very small static deviations and follow the dynamics of the speciﬁed motion of the TLP and tension with errors less than 5%. but can be chosen freely within the given limits.
.
However.5 was veriﬁed with a large variety of current proﬁles in Chapter 4. Some limitations and assumptions are made in relation to the shielding eﬀect and the resulting current on R2 to keep the riser and current models valid. 8. The initial centertocenter distance between the risers is 15D in 1200m waters. the equal payout and equal eﬀective length.1 are investigated and compared in deep waters. and used in a case study with changing environmental conditions and supervisory switched controller.Chapter 8 Simulation Results The TLP/riser system was modeled in Chapter 3. The best working control objective is thereafter simulated with dynamic TLP motions. unless otherwise speciﬁed. with R2 in the wake of R1. and an amplitude within the band of 5% of the current velocity in each node. respectively. 2001).1 Set Up The riser model from Section 3. However. • The current is coming from one direction only. Two of them. In addition a variance in the current is made by ﬁltering white noise through a low pass ﬁlter with a period of 100s. The latter will not be true for the last part of the case simulations with WF TLP motions. keeping R2 in the wake of R1. • Small TLP velocities such that the current is larger than the riser velocities and the relative velocity always positive. in the simulations included in this chapter. such that the risers are always in a tandem position. are also tested in shallow waters. this part of the simulation is included to demonstrate the 125 . and each riser model consists of 10 elements. a current proﬁle with one year return period from the Ormen Lange ﬁeld is used as a basis (Norsk Hydro. and the controller system design and architecture were presented in Chapters 5 and 6. In this chapter the control objectives proposed in Section 6.
seen as relative horizontal distance smaller than 2D (bottom left). As the current increases.2. the top tension of the risers are constant and equal to a pretension. and there is no TLP oﬀset.1. The corresponding payout due to deﬂection is seen bottom right in Fig. The hydrodynamic interactions for the threedimensional case and faster relative motions should be further investigated for industrial implementations. Increasing the top tension to the upper limit of T0. which is clearly seen in the payout plot. Simulation Results eﬀect of control.1) A medium top tension of T0. collision could be avoided for these design 126 .2.1) we have T1 = T2 = T0.j = 1800kN is applied. where it is seen that collision occurs at the ﬁrst node after approximately 400s. Keeping the tension at the upper limit. From (6. The incoming current proﬁle is increased as a second order lowpass ﬁltered step from zero to the design current proﬁle. seen in the snapshots (top left) and relative horizontal distance (bottom left) in Fig. The TLP is then put in an oﬀset position of 30m. 8. The riser data. Collision is seen to occur at the same nodes independent of the oﬀset position. The relative horizontal distance between three selected nodes are shown in bottom left of Fig. 8. The maximum horizontal deﬂection is seen at 600m above the seabed. equal eﬀective length and desired horizontal distance control objectives are run with control of R2.1 (top left). 8. The equal tension control objective is run with two diﬀerent top tension levels and 0m or 30m oﬀset. the risers are seen to slide out to the right in Fig. current proﬁle velocities and control gains are found in Appendix B.1 are tested and compared. no collision occurs at any nodes. bottom right).2. 8.j = 2700kN and keeping the TLP in a nooﬀset position. so oﬀsets are always simulated in the positive direction.2 Control Objectives The four diﬀerent control objectives from Section 6.3. control of R1 with the payout of R2 as a reference is also simulated. (8.8.j . 8. 8. 8. For the equal eﬀective length control objective.1 Constant Equal Tension In the ﬁrst case.2. using measurements from R1 as the reference. The equal payout. The TLP is not very likely to have an oﬀset in the opposite direction of the surface current. The payout of R1 is larger than for R2 due to the larger deﬂection (Fig. 8. with increasing current. Collision occurs along most of the riser (nodes 3 to 9). seen top left in Fig. R1 has the largest deﬂection and hence setdown. but is outside the scope of this thesis.
2 (t) = ξ1 (t).2) 127 .j = 1800kN for both risers.37) ξr. For the rest of the simulations.8 0 200 400 Time [s] 600 800 ξ1 ξ2 600 800 Relative horizontal distance 15 10 5 [D] 0 Figure 8. see (6. Snapshots of the riser conﬁguration (top left).6 0.65 0.2.Control Objectives Snapshots 1200 1000 1800 800 [kN] [m] 2000 Top tension T 1600 600 400 1 T2 1400 200 0 0 2 4 6 [m] 8 10 1200 0 200 400 Time [s] Payout 0. the payout of all risers should be equal. relative horizontal distance (bottom left) and payout (bottom right). 8. currents.7 −5 −10 0.1: Incrementing current velocity from zero to the Ormen Lange design current.55 0. (8. we have from (6.2 Equal Payout For the equal payout controller objective. The payouts (bottom right) are smaller than for the smaller top tensions.75 0 200 400 Time [s] 600 800 0. top tension (top right). operation of the risers at this tension level is not desired due to increased stress in the riser and excessive wear of the tension system. Using the measured payout of R1 as a reference for R2.45 Node 3 Node 6 Node 9 [m] 0.2). the pretension is kept to T0. A fully developed current proﬁle for the Ormen Lange ﬁeld is used.5 0. However.
2 = −KP.8. collision still occurs. T1 = T0.44).49) with parameters from Appendix B.2 e2 − KI.5 0.6) where ξd.41).3. seen bottom right in Fig.40)(6.2 is the guidance trajectory passed through the third order ﬁlter in (6. τc.4 0. top tension (top right). T2 = T0.2 + τc.2 − ξ2 . Simulation Results Snapshots 1200 1000 1800 800 [kN] [m] 2000 Top tension T 1600 600 400 1 T2 1400 200 0 0 10 [m] Relative horizontal distance 20 15 10 [m] [D] 5 0 −5 −10 0 200 400 Time [s] 600 800 Node 3 Node 6 Node 9 0.1 . Snapshots of the riser conﬁguration (top left).2 e2 = ξd. relative horizontal distance (bottom left) and payout (bottom right). Furthermore.3) (8. 8. Top left shows how R2 slides out to the right due to decreased tension.6 0.2 . with the top tension shown in the top right ﬁgure.4. e2 dt. but in a smaller riser segment than with equal 128 . (8.46) and (6.2: Incrementing current and 30m TLP oﬀset.5) (8. However. The results are found in Fig.4) (8. 8.7 0.8 0.9 1 0 200 400 Time [s] 600 800 ξ1 ξ2 20 30 1200 0 200 400 Time [s] Payout 600 800 Figure 8. (6. a PIcontroller for the top tension could be formulated from (6.4. This control algorithm gives equal payout for the risers.
2 = ξ1 + ∆lR . which in turn gives less axial elongation.3: Incrementing current with top tension 2700kN. Snapshots of the riser conﬁguration (top left).18 0.22 0 200 ξ1 ξ2 400 Time [s] 600 800 Figure 8. The collision is caused by the lower tension in R2 compared to R1. collision may still occur for long risers when using the equal payout control objective.2 0.3) and (6.1 0. the guidance trajectory from (6. relative horizontal distance (bottom left) and payout (bottom right).38). due to the elasticity of steel. For equal risers with the same pretension.Control Objectives Snapshots 1200 1000 800 [kN] [m] 600 400 200 0 0 2 4 [m] 6 8 2200 2000 3000 2800 2600 2400 Top tension T 1 T2 0 200 400 Time [s] Payout 600 800 Relative horizotnal distance 16 14 12 [D] 10 8 6 4 2 0 200 400 Time [s] 600 800 Node 3 Node 6 Node 9 [m] 0. using (6.5) is simpliﬁed to ξr. (8.2. Hence.7) 129 .14 0.3 Equal Eﬀective Length In the next control objective. 8. the elasticity of the riser material is included. This could be done by considering the riser length plus the payout.12 0. and a shorter length of R2 than R1. also seen for the relative distance (bottom left).16 0. tension (nodes 5 and 6). top tension (top right).
increasing the horizontal distance between the risers to avoid collision. The snapshots found top left show how the midposition of R2 starts in front of R1 when the pretensions are equal. top tension (top right). To the bottom right in Fig.8. 8. As before R2 slides out to the right with decreasing tension.9 0 200 400 Time [s] 600 800 Figure 8.5 ξ1 5 [m] [D] 0. 8. The payout of R2 approaches the reference trajectory and both risers achieve similar deﬂection. The smaller variation in the horizontal position is due to variance in the current. Snapshots of the riser conﬁguration (top left).4: Equal payout algorithm and control of R2.7 −5 Node 3 Node 6 Node 9 0 200 400 Time [s] 600 800 0.3) through (8. also seen in the relative horizontal distance to the lower left. see Fig.2. The top tension T2 is seen to stabilize about 1400kN for 130 .5 we clearly see that R2 has a larger payout to compensate for smaller riser elongation and achieve equal eﬀective length. (8. relative horizontal distance (bottom left) and payout (bottom right).5. The payouts (bottom right) are larger in the oﬀset case than without oﬀset due to the eﬀects from weight and current.6 ξ 2 10 1200 0 200 400 Time [s] Payout 600 800 0 0. Simulation Results Snapshots 1200 1000 1800 800 [kN] [m] 2000 Top tension T 1600 600 400 1 T2 1400 200 0 0 5 [m] Relative horizontal distance 10 0. 8.2. The controller is the same as in Section 8. R2 then slides out to the right to avoid collision as the tension decreases. The same algorithm is applied for 30m TLP oﬀset (Fig.6). The results are seen in Fig.8 −10 0.6).6 top left. 8.
1 + τc. Hence.2 0 200 400 Time [s] 600 800 ξ1 ξ 2 10 15 1200 0 200 400 Time [s] Payout 600 800 Figure 8. compensated for axial elasticity. payout and horizontal deﬂection.11) T1 = T0.6. and the relation between tension. Hence.Control Objectives Snapshots 1200 1000 1800 800 [kN] [m] 2000 Top tension T 1600 600 400 1 T2 1400 200 0 0 5 [m] Relative horizontal distance 20 15 10 5 0 −5 −10 0 200 400 Time [s] Node 3 Node 6 Node 9 600 800 [m] [D] 0. 8.1 . a smaller tension diﬀerence is needed in the oﬀset position.8) (8. top tension (top right). Now the top tension of R1 is controlled using the payout of R2 as reference. respectively.2 .7 0.6 0.2. Fig.5 0.9) (8. Snapshots of the riser conﬁguration (top left). The controller is the same as for Section 8.5 and 8.5: Equal eﬀective length and control of R2.8 0. relative horizontal distance (bottom left) and payout (bottom right). an increase in tension has larger eﬀect on the lateral deﬂection than if the riser already has a high tension level.2. the oﬀset case and 1300kN for the nooﬀset case.10) (8. e1 = ξd.7 shows the snapshots of 131 .1 = ξ2 − ∆lR . For a long riser with small tension. T2 = T0. R2 is not controlled. (8. This is due to a longer eﬀective length (riser plus payout).1 − ξ1 . to the upper left in Figs.9 1 1. 8.1 1. ξr.
8. relative horizontal distance (bottom left) and payout (bottom right). top tension (top right). Top tension was seen to decrease compared to the nooﬀset case and stabilize about 2650kN and 2400kN with tension diﬀerences of 850kN and 600kN. the risers. This is due to the increased current and reduced shielding eﬀect on R2 as the distance between the risers increases.6: 30m TLP oﬀset. It should be noted that the simulation is valid only when the downstream riser is more than 2D behind the upstream riser. and a need of even more tension to straighten up to prevent 132 . The risers behave similarly as for the case without oﬀset. respectively shown in Figs. In addition. how the piston end of R1 is pulled in.8. Simulation Results Snapshots 1200 1000 1800 800 [kN] [m] 2000 Top tension T 1600 600 400 1 T2 1400 200 0 0 10 20 [m] 30 1200 0 200 400 Time [s] Payout 0.2 0 200 400 Time [s] 600 800 Figure 8.8. see snapshots and payouts in Fig. It should also be noted that the diﬀerence and payout is larger when controlling R1 than R2.8 1 1. the deﬂection decreases. This is caused by the more straightlined conﬁguration at higher tension levels. 8.6 [m] ξ 2 600 800 Relative horizontal distance 25 20 15 [D] 10 5 0 −5 −10 0 200 400 Time [s] Node 3 Node 6 Node 9 600 800 0. equal eﬀective length and control of R2. The payouts. The TLP is placed at 30m oﬀset. and how the payout of R2 is slightly increased when the deﬂection increases.7 and 8. As the tension in R1 increases and it is tightened up. R2 moves behind since R1 comes in front. are seen (bottom right). Snapshots of the riser conﬁguration (top left).4 ξ1 0. 8.
∆x ∆T p. 53) between the tension 8.2.2 0.4 Desired Horizontal Distance The desired horizontal distance control objective requires one or more measurements of the relative distance between the riser at predeﬁned water depths. Snapshots of the riser conﬁguration (top left). collisions.12) 133 .6 0.9 0 200 10 0 200 Top tension T 600 400 1 T2 400 Time [s] Payout 600 800 ξ1 ξ2 400 Time [s] 600 800 Figure 8.r (z) = ∆xd .Control Objectives Snapshots 1200 2600 1000 2400 800 [kN] [m] 2200 2000 1800 200 1600 0 0 5 [m] Relative horizontal distance 20 15 10 5 0 −5 −10 0 200 400 Time [s] Node 3 Node 6 Node 9 600 800 [m] [D] 0. and the deﬂection. (8. relative horizontal distance (bottom left) and payout (bottom right).4 0.3 0. top tension (top right). Keeping a desired relative distance.8 0.5 0.7: Equal eﬀective length and control of R1.7 0. gives the following guidance trajectory from (6. due to the nonlinear relationship (cf.39) ∆xR12.
134 . τc. However.6 0. The corresponding payout is seen bottom right. top tension (top right). equal eﬀective length and control of R1.m (z). The snapshots (top left) show how R2 slides to the right when the tension decreases and removes the risers from the error state.8 0.2 + τc. it is dependent on measurement(s) along the riser. (6.44). This control objective is seen to eﬀectively remove the risers from a collision situation.9 (top right).13) (8. 8.8. Snapshots of the riser conﬁguration (top left). T2 = T0.d − ∆xR12.7 0.2 ∆erel − KI.2 = −KP.4 0.5 0.2 .14) (8.1 . Controlling R2 with a PIcontroller gives T1 = T0.9 1 0 200 20 30 0 200 Top tension T 600 400 1 T2 400 Time [s] Payout 600 800 ξ1 ξ2 400 Time [s] 600 800 Figure 8. (8. relative horizontal distance (bottom left) and payout (bottom right). located where the risers are most likely to collide.51).15) (8.8: TLP oﬀset of 30m.16) using (6.50) and (6. Simulation Results Snapshots 1200 2600 1000 2400 800 [kN] [m] 2200 2000 1800 200 1600 0 0 10 [m] Relative horizontal distance 20 15 10 5 0 −5 −10 0 200 400 Time [s] Node 3 Node 6 Node 9 600 800 [m] [D] 0. The tension is found in Fig.2 ∆erel (z) = ∆erel. ∆erel dt.
the water depth and also the eﬀective weight is one fourth of the previously simulated depth of 1200m.6 0. 8. whereas the lower tension limit is due to eﬀective weight. Here. the equal payout and equal eﬀective length control objectives are subject for the next investigation in shallow waters. The 135 .9 1 1. top tension (top right). The risers are located at 300m water depth and the current proﬁle from the Ormen Lange ﬁeld is scaled to this depth.3 Eﬀect of Shallow Water The control objectives equal payout and equal eﬀective length. the upper tension limit could be the same due to stress considerations. relative horizontal distance (bottom left) and payout (bottom right).9: Desired relative horizontal distance and control of R1. The centertocenter distance is decreased to 8D.5 0.8 0. Hence.7 0.Effect of Shallow Water Snapshots 1200 1000 1800 800 [kN] [m] 2000 Top tension T 1600 600 400 1 T2 1400 200 0 0 5 [m] Relative horizontal distance 20 15 10 5 0 −5 −10 0 200 400 Time [s] Node 3 Node 6 Node 9 600 800 [m] [D] 0. keeping the same velocity in each node.1 1. are based on payout and tension measurements at the wellhead and available today. and might therefore more likely be implemented by the industry. Keeping the same riser diameter. The risers have the same physical dimensions as before except for the length. Snapshots of the riser conﬁguration (top left).2 0 200 400 Time [s] 600 800 ξ1 ξ 2 10 15 1200 0 200 400 Time [s] Payout 600 800 Figure 8.
it is of far less importance than in 136 . zero oﬀset and control of R2.56 0. top tension (top right).7D in the ﬁrst case and 7. starting after 200s. and hence excepted to be of less importance. R2 slides slightly more to the right. The tension level is much lower and the axial elongation therefore smaller. The tension of R2 is right above 400kN using equal payout and right below 400kN using the equal eﬀective length control objective. The smallest relative horizontal distance between the risers is about 6.10: Equal payout control of R2 at 300m water depth. However.54 0.58 0. a small eﬀect of taking the elasticity into account is seen even at 300m water depth. Fig. Simulation Results Snapshots 650 300 250 200 [kN] [m] 150 100 50 0 0 2 [m] Relative horizontal distance 10 8 6 4 2 0 −2 0 200 400 Time [s] Node 3 Node 6 Node 9 600 800 [m] [D] 0. For the equal eﬀective length control objective in Fig.10 (top left) shows how R2 slides out to the right due to the decreased tension when the control is turned on. relative horizontal distance (bottom left) and payout (bottom right). 8. due to the compensation of the elasticity.66 0 200 4 6 600 550 Top tension T 500 450 400 350 0 200 400 Time [s] Payout ξ 600 1 T2 800 1 ξ2 400 Time [s] 600 800 Figure 8. 8.11.52 0.2D in the latter.64 0.8. Snapshots of the riser conﬁguration (top left). lower tension limit could be as low as 350kN at 300m water depth.6 0.10 (bottom right). Hence. The payout is seen to be equal for the two risers in Fig. The equal payout and the equal eﬀective length control objectives are tested with an initial top tension of 600kN.62 0. 8. and the payout is seen to be 23cm larger for R2 than for R1.
and the horizontal positions for R2 increased at the same time due to reduced shielding. not shown here. 8. The period is 120s. deeper waters. In this case. Snapshots of the riser conﬁguration (top left). 1200m water depth is used.7 0 200 400 Time [s] 600 800 Figure 8. 137 . top tension (top right). In this case where R2 is controlled. It is seen in these plots.4 TLP Dynamics For the simulations with dynamic TLP motions. the horizontal positions of R1 decreased.11: Equal eﬀective length control of R2. relative horizontal distance (bottom left) and payout (bottom right). that the upper nodes are most inﬂuenced by the TLP motion. When R1 was controlled. we see that the collision is avoided by increasing the horizontal position of R2. The controllers are enabled after 400s.5 4 6 600 550 Top tension T 500 450 400 350 0 200 400 Time [s] Payout ξ 600 1 T2 800 1 ξ2 0.6 0. Fig. 8.55 6 4 2 0 −2 0 200 400 Time [s] Node 3 Node 6 Node 9 600 800 0.12 shows the horizontal positions for three selected nodes.TLP Dynamics Snapshots 650 300 250 200 [kN] [m] 150 100 50 0 0 2 [m] Relative horizontal distance 10 8 0. the TLP moves with harmonic motions in surge about a static oﬀset of 30m. as expected. and the peaktopeak amplitude 40m.65 [m] [D] 0.
8. Simulation Results
Node 9 50 40 [m] 30 20 10 Node 6 50 40 [m] 30 20 10 Node 3 50 40 [m] 30 20 10 0 100 200 300 400 500 Time [s] 600 700 800 900 1000 R1 R2
Figure 8.12: Dynamic TLP motions and control of R2 enabled after 400s. Horizontal positions for node 3, 6 and 9 for R1 and R2.
The riser slides out to the right when the payout ξ2 increases, see Fig. 8.13. The relative horizontal position between the same three nodes is seen in Fig. 8.13 (bottom left). The eﬀect of control is clearly seen. Before the controller is turned on, collision occurs twice for each cycle. After 400s, the controller is turned on, and the mean distance is about 15D, equal to the top and bottom distance, for all nodes, giving a similar conﬁguration for the two risers. Also, the variation in distance between corresponding nodes decreases signiﬁcantly. The payout with control of R1 and R2 are given in Figs. 8.13 and 8.14, respectively. For the uncontrolled ﬁrst 400s, the dynamic stroke of R1 is 1.6m and 0.8m for R2. This diﬀerence is caused by larger drag and deﬂection of R1. R2 has a more straightlined conﬁguration due to reduced drag forces. Note that when R2 comes in front, the model is not valid and R2 keeps its straight conﬁguration. When R1 is the reference, the stroke of R2 is increased, and the stroke of R1 is slightly decreased, giving a dynamic stroke about 1.5m for both risers. The mean tension of R2 is decreased to about 1450kN, giving a tension diﬀerence of 350kN
138
TLP Dynamics
Snapshots 1200 1000 1800 800 [kN] [m] 600 400 1400 200 0 0 10 20 30 [m] 40 50 1200 0 200 1600 2000
Top tension T1 T2
400 600 Time [s] Payout
800
1000
Relative horizontal distance 30 20 10 0 −10 −20 Node 3 Node 6 Node 9 0 200 400 600 Time [s] 800 1000 [m] 1.5 2 2.5 [D] 0 0.5 1
ξ1 ξ2 0 200 400 600 Time [s] 800 1000
Figure 8.13: Dynamic TLP motions and control of R2 enabled after 400s. Snapshot of the riser conﬁguration (top left), top tension (top right), relative horizontal distance (bottom left) and payout (bottom right).
between the risers, see Fig. 8.13. The smaller mean tension compared to the static case is due to longer payout and total eﬀective length, such that a smaller tension diﬀerence is needed to avoid collision. The dynamic tension variation has a period of 120s as for the TLP motions, and a peaktopeak amplitude about 120140kN. When R2 is the reference for R1, its dynamic stroke decreases, giving both R1 and R2 a dynamic stroke about 1m, see Fig. 8.14. The mean tension level is increased with 400kN for T1 , seen in Fig. 8.14. The dynamic tension variation was about 120130kN. Increasing the TLP period to 300s, a more quasistatic riser behavior is seen with less dynamic deﬂection. This gives less need for stroke, a smaller payout and eﬀective length and a larger tension diﬀerence closer to the static case.
139
8. Simulation Results
Snapshots 1200 2600 1000 2400 800 600 400 200 0 0 10 20 30 [m] 40 50 [kN] [m] 2200 2000 1800 1600
Top tension T1 T2
0
200
400 600 Time [s] Payout
800
1000
Relative horizontal distance 20 0 0.5 1 0 [m] 1.5 −10 Node 3 Node 6 Node 9 0 200 400 600 Time [s] 800 1000 2 2.5 [D]
10
ξ1 ξ2 0 200 400 600 Time [s] 800 1000
−20
Figure 8.14: Dynamic TLP motions and control of R1 enabled after 400s. Snapshot of the riser conﬁguration (top left), top tension (top right), relative horizontal distance (bottom left) and payout (bottom right).
8.5
Supervisory Switched Controller
For permanent installations, like the TLP, the allyear weather changes must be taken into consideration, also when designing the control system. The diﬀerent ROCs should be identiﬁed and the control actions adjusted to each regime. This includes controlling the top tension of the risers depending on the prevailing situation, and also turning oﬀ the controller in calm weather to avoid unnecessary wear and tear of the cylinders. Switching between the diﬀerent regimes and control actions are illustrated in this section.
8.5.1
Case SetUp
The initial current is of factor 0.8 of the Ormen Lange design current with the same variation as used before. The TLP is in a nooﬀset position, and the centertocenter distance between the risers is 15D at the top and bottom end points.
140
Supervisory Switched Controller
At 400s, an additional, uniform tide current increases from 0m/s to 0.3m/s over 300s. This total current velocity is used for the rest of the simulations. At 500s, the TLP starts moving to an oﬀset position of 30m, which it uses approximately 300s to reach. When the TLP has almost arrived the oﬀset position, a LF motion in surge with a period of 120s and a peaktopeak amplitude of 40m starts (800s). At 1200s, an additional extreme WF motion is superimposed on the LF TLP motion. This extreme wave has a 12s period and an amplitude of 20m. The linear waveinduced surge motion on the TLP can be found from the response amplitude operator (RAO) for the TLP ASM600, see Faltinsen (1990). This gives a WF TLP surge motion with the same period of 12s and a peaktopeak amplitude of 10m. The changes in current and TLP motions are given in Table 8.1 and illustrated in Fig. 8.15. It should be noted that the changes in mean current velocity and the diﬀerent environmental conditions happen faster than in real life, but the magnitudes themselves are representative. Disturbance Tide current TLP oﬀset TLP LF TLP WF Start time [s] 400 500 800 1200
Table 8.1: Case simulation. The PPMs consist of 16 elements, while the ACPMs consist of 8 elements. Only R2 is controlled, using the payout from R1, ξ1 , as a reference. Due to the deep water, the riser elongation ∆lR is included, giving the payout reference ξr,2 = ξ1 + ∆lR , ∆lR = ∆T l0 . EA (8.17)
8.5.2
Simulations without Control
First the simulation case is run without control for two diﬀerent pretensions, see Fig. 8.16. To the left simulations with a pretension of 1800kN for both risers are seen. The risers are close to collision from the start, but actual collision is seen for node 9 after 200s, due to the small variations in current velocity. The relative distance is smaller than 2D (bottom left). The collision is present until 400s. A new collision is seen at 450s, and is lasting for the rest of the simulation. Recall that the model is not valid for riser distances smaller than 2D, and we can only say that for smaller relative distance than this, collision is likely to occur. With a pretension of 2700kN for both risers (Fig. 8.16, right), collision ﬁrst happens after 600s, when the current velocity is increased due to tide. For LF
141
8. Simulation Results
Current velocities 1.5 1 0.5 Node 3 0 0 500 TLP position 60 40 [m] 20 0 1000 Node 9 Node 15 1500
[m/s]
0
500 Time [s]
1000
1500
Figure 8.15: Disturbances from current (top) and TLP (bottom), acting on the risers.
TLP motions, the risers are in a no collision situation in the rightmost TLP oﬀset, where the risers are most straightlined. When the TLP moves to the left, the risers experience larger relative velocities. R1 experiences larger drag forces than R2, and hence a larger deﬂection and increased risk of collision. Notice the diﬀerence in payout values in the top plots. The medium pretension of 1800kN results in a payout approximately twice as big as the payout for the pretension of 2700kN. The relative distance between the risers are similar.
8.5.3
Simulations with Control
The risers are exposed to the same environmental setup, but now control of top tension is introduced. R2 is controlled by switching between the controllers described in (6.45) to (6.49) as described in Section 6.4. The initial top tension is 1800kN for both risers. Due to the shielding eﬀect which gives diﬀerences in drag forces, the risers are starting in a state close to collision, with 3D being the smallest distance between the risers (see Fig. 8.18). The monitoring signal is smallest for process µ5 , leading to the process switching signal ρ = 5 which maps to the controller switching signal σ = 2 and the PIcontroller (see Fig. 8.17). The tension for R2, T2 , is decreasing such that the distance between the risers is increasing to a level about 15D (Fig 8.18). The payout for the two risers are also found in Fig. 8.18, showing a diﬀerence of 0.2m, corresponding to the riser elongation due to tension.
142
Supervisory Switched Controller
Payout for T = 1800kN 0 0
Payout for T = 2700kN
1 [m] [m] R1 R2 4 0 500 1000 1500
1
2
2
3
3 R1 R2 4 0 500 1000 1500
Relative horizontal distance 10 0 [D] −10 −20 −30 Node 3 Node 9 Node 15 0 500 Time [s] 1000 1500 [D] 10 0 −10 −20 −30
Relative horizontal distance
Node 3 Node 9 Node 15 0 500 Time [s] 1000 1500
Figure 8.16: Eﬀects of the constant top tension, 1800kN (left) and 2700kN (right), payouts (top) and relative horizontal distance between the risers (bottom).
As the riser distance increases, the monitoring signal for the slowly varying process, µ3 , is the smallest, and the process switching signal switches to ρ = 3. The same controller is kept, as it still maps to the same controller switching signal σ = 2. After approximately 310s, R2 has reached a stable position near the desired horizontal riser distance and a small variation in tension the last 100s. The smallest monitoring signal is found for normal state µ2 , giving ρ = 2 and σ = 1 (Fig. 8.17). The top tension T2 is set constant and equal the prevailing value at the switching moment (Fig. 8.18). From 550s, the smallest distance between the risers decreases, due to increasing current velocity from tide. The PIcontroller for T2 is turned on when µ3 is the least monitoring signal, giving ρ = 3 and σ = 2 (Fig. 8.17). The same controller is kept when the TLP starts moving with LF motions. An additional WF TLP motion is superimposed, starting at 1200s, see Fig. 8.19. At 1250s, when the WF motions have achieved almost full amplitude, µ4 is the smallest monitoring signal, giving ρ = 4 and σ = 3 (Fig. 8.17), and the PIDcontroller is
143
8. Simulation Results
Controller switching signal 4 3 σ 2 1 0 0 500 Process switcing signal 6 5 4 ρ 3 2 1 Monitoring signal 4 µp 2 0 0 500 Time [s] 1000 1500 µ1 µ2 µ3 µ4 µ5 1000 1500
Figure 8.17: From the top: Controller switching signal σ, process switching signal ρ and switching error µp .
phased in. The eﬀects of the TLP WF motions on the top tension and payout are seen in Fig. 8.19. The top tension desired by the controller (–) and the saturated tension from the actuator ( ) are seen in the middle graph. The switching from the PI to the PIDcontroller happens at 1250s, with a smooth transition lasting 20s, such that at 270s the PIDcontroller is working alone. T2 is in rate saturation with the PIcontroller just before the PIDcontroller is phased in. The tension variations are larger for the PIcontroller than the PIDcontroller due to a larger proportional gain. Anyhow, the PIDcontroller is also seen to be in rate saturation every 120s, corresponding to the LF TLP motions, where the TLP moves from right to left, giving the largest deﬂections. When the TLP moves against the current velocity, the relative velocity increases, and hence the drag forces increase with the square of the velocity. For increasing relative velocity, the riser feels stiﬀer, and to straighten the riser or even maintain the same curvature, a higher tension is needed. For a pure LF motion the tension rates are not exceeded, but for an additional WF motion,
144
Supervisory Switched Controller
Top tension 2000 T1 1800 [kN] 1600 1400 1200 0 500 Payout 0 1 [m] 2 3 4 0 500 Relative horizontal distance 20 15 [D] 10 5 0 0 500 Time [s] Node 3 1000 Node 9 Node 15 1500 R1 R2 1000 1500 1000 1500 T2
Figure 8.18: Eﬀects of the controller. From the top: Top tension, payout and relative riser distance.
large and fast changes in tension are requested, and rate saturation occurs. This rate saturation is seen to aﬀect the payout of R2. The diﬀerence in payouts for R1 and R2 corresponds to the riser elongation due to the tension diﬀerence in the unsaturated periods. While the tension is in rate saturation, the payout for R2 is not large enough, such that the payouts for R1 and R2 are close to equal. However, this is not a problem with respect to collision for such fast motions here. Regardless of this, we can not neglect that this could be a problem for risers with diﬀerent physical properties and dynamics. The relative distance between the upper nodes is naturally more aﬀected by the WF motions than the nodes further down the riser (Fig. 8.18). For instance, node 15 experiences more dynamics, with larger motions over a longer period of time. It takes some time for the motions to travel down the riser. The riser top seems most sensitive to the WF motions when it is in a nearly Straightlined conﬁguration, corresponding to a TLP movement from left to right. The least variation in relative distance is seen when the TLP moves in the opposite
145
This means that when the ﬁrst mode is the dominating mode shape. higher order mode shapes are less pronounced. and used in the monitor to trigger switching between the diﬀerent controllers. For low current velocities. However.19: The WF motion. Equal payout is shown to be appropriate in shallow waters (∼300m).8. 8. a model is used to estimate the distance between the risers. The desired relative horizontal distance control objective indirectly includes the elasticity and showed promising results.6 Discussion The controller objective may change according to the ROC. top tension and saturation for T2 and payout. If measurements were easily available. direction and the deﬂection is largest. the eﬀects of elasticity and riser elongation need to be included in the control strategy in order to prevent riser collision in deep waters when top measurements are used. but the measurements along the riser used in this algorithm are not easily available oﬀshore. Simulation Results TLP position 60 40 [m] 20 0 1200 1600 1500 [kN] 1400 1300 1200 1200 0 1 [m] 2 3 4 1200 1250 1300 1350 Time [s] 1400 R1 1450 R2 1500 1250 1300 1350 Payout Actuator tension 1400 Controller tension 1450 1500 1250 1300 1350 Top tension saturation T2 1400 1450 1500 Figure 8. From the top: TLP position. they could be used 146 . Instead. constant tension may be a suﬃcient solution.
147 . For large riser or TLP velocities. the wake model is no longer valid.Discussion instead. However. WF TLP motions are included in the simulation study to illustrate the eﬀect of control also in this regime. Note that the model used here is veriﬁed for LF TLP motions.
.
The main conclusion from this study is that it may be possible to prevent riser collision by use of top tension control. The risers were modeled with FEM and integrated load and equilibrium iterations. we have considered how we can prevent collision between risers in an array. TLP motions. current and hydrodynamic interaction forces by controlling the top tension. and compared to RIFLEX models with and without a stress joint at the seabed. Simpliﬁcations were made by neglecting the bending stiﬀness and assuming free end rotations. hydrodynamic interaction between risers. More speciﬁcally.1 Conclusion This thesis has focused on modeling. the riser model with 20 elements was compared to a RIFLEX model with 400 elements. 2005). the riser model and the actuator. The hydrodynamic interaction could be described by three diﬀerent components. The stress joint was seen to be of importance in shallow waters. The riser model was veriﬁed with the commercial software RIFLEX (Fylling et al. A twodimensional mathematical model of a riser/TLP system was developed. whereas it had less impact on the global geometry in deep waters and could be considered insigniﬁcant. A quasistatic model was also veriﬁed for shallow waters. exposed to platform motions. as this was assumed to be of major importance for the risers’ relative position. WIO and VIV. Hence. 149 . both with respect to payout and horizontal deﬂections. At 1200m water depth. This included the current velocity proﬁles.Chapter 9 Concluding Remarks 9.. For a larger curvature. The agreement was better for high tensions and when the curvature was small. These veriﬁcations showed good agreement. when simulating production risers in shallow waters. Here. the mean shielding eﬀects. the stress joint should be included. simulation and control of top tensioned marine risers. more elements were needed to maintain the same accuracy. only the ﬁrst one was considered.
area under the riser curve and payout. Equal top tension is used by the industry today. This model could be used in control analysis or fast simulations (FCPM). With more than 15 elements. However. The models with 812 elements showed good agreement and rather accurate estimates of the process. collision was prevented. equal eﬀective length and desired horizontal distance. Here equal payout showed to be satisfactory. while still being able to describe the main physics with desired accuracy. equal payout. Equal payout gave collision in a smaller riser segment. for very low current velocities. where both fast simulations and high accuracy are required. However. Anyhow. as the relative riser elongation was observed to be much smaller and hence of less importance. the errors were larger. The diﬀerent control objectives to prevent riser collision were proposed. we achieved high accuracy and an accurate description of the physical process (PPM). The diﬀerent regimes for riser operations were deﬁned. This method was therefore used to compare the performance of the dynamic model with diﬀerent number of elements. equal tension can be appropriate. the various element models were classiﬁed for diﬀerent applications. In deep waters equal top tension resulted in collision for medium current velocities. By taking the riser elongation into account by introducing equal effective length. The performance of the quasistatic model with increasing TLP oﬀset was investigated through error norms of the horizontal position. but collision still occurred. For few elements models (46). the measurements used in this method is not easily available oﬀshore today. in deeper waters the eﬀect of elasticity and riser elongation needs to be included in the control strategy in order to avoid riser collision. These two latter control objectives were also simulated in shallow waters (300m). The riser model was then run with 2 to 20 elements and compared to the RIFLEX model with 400 elements. but the main physics were still captured. A control system design for control of top tensioned marine risers has been proposed and needed instrumentation has been suggested. We have in this thesis extended these results to also include equal eﬀective length and desired horizontal distance as new control objectives. This was motivated by the purpose of realtime control applications which require computationally fast models. Simulations showed that payout best reﬂected the models’ performance. Desired horizontal distance included the elasticity indirectly and showed promising results. They are appropriate for ACPM purposes. Concluding Remarks For a drilling riser which actually has a balljoint at the lower end. Based on the payout error. and a relation between 150 .9. free end rotation was a more correct way of modeling. whereas equal payout was proposed by Huse and Kleiven (2000). riser top inclination. Instrumentation includes payout and top tension at the TLP wellhead and the relative horizontal distance along the riser. These included equal top tension.
2 Proposals for Further Work This work has focused on a twodimensional model of risers in a tandem arrangement. In these analyses we have focused on two equal risers. etc. and further studies should be carried out to investigate the possibility of physical implementation oﬀshore. These could be summarized as: • Formulating the tension reference trajectory as an optimization problem with relative horizontal distance between the risers and energy consumptions as possible parameters in an objective function. The largest challenge will be to make a good threedimensional wake model.) and thereby physical behavior. etc. A simulation study with supervisory switched control showed a case. The proposed tasks could be solved for the twodimensional case ﬁrst and thereafter extended to the threedimensional case. and some modiﬁcations might be needed to make the wake model valid in this regime. The sequence of the risers and how fast each riser should reach its optimal tension trajectory is of importance. This could be formulated as a maneuvering problem and called error state maneuvering here. • Preventing new collisions between risers in an array or matrix during transition from an error situation to reach their desired positions and tension trajectories. more than two risers should be included in an array or matrix. meaning that lift forces should be included. Further analyses using the same model should include riser for diﬀerent purposes (drilling. Lastly. Also direct wave loads and the ﬁrst order ﬂoater motions should be investigated. As a next step. To simulate risers placed in a matrix structure a threedimensional model is needed.Proposals for Further Work regime and control action was proposed. This modeling. with switching between the diﬀerent controllers depending on the prevailing regime. 9. choosing one riser as a leader and the others as followers. ﬂuid density. Within the ﬁeld of control several interesting challenges are found. A simple twodimensional wake model was applied. pretension. but also other hydrodynamic interaction eﬀects such as WIO and VIV should be considered added in the model. In addition the current can have varying direction proﬁles. implementation and simulation study should be performed before full scale tests can be carried out. production. These results may be of industrial interest. both model tests and full 151 . These eﬀects are not included in the implemented wake model. • Synchronization of risers in an array or matrix.) giving diﬀerent riser characteristics (diameter.
152 .9. Concluding Remarks scale experiments would be of importance for the actual implementation of top tension control of risers to prevent collision aboard installations oﬀshore.
Lie and K. OMAE200567100. H.C. Berge. R. (2002). Modeling and Control of Fluid Flows and Marine Structures. Experimental investigation of dual riser interaction. (2005). M.Bibliography Aamo. American Petroleum Institute (1987). Ormen Lange TLP with full processing facilities. Aker Maritime (2002). D.com. Norwegian University of Science and Technology (NTNU). Washinton D.akerkvaerner.. Prentice Hall. In: Proc. R. and B. designing and constructing tension leg platforms. D. of the 37th Annual Oﬀshore Technology Conference. Structural Engineering. Trondheim. J. of 24th International Conference on Oﬀshore Mechanics and Arctic Engineering. [Online. URL http://www. Norway. SINTEF. J. O. K. ComputerControlled Systems . Engseth. Norway. R. Aker Kværner (2007). Olufsen (1992). 153 .. Forces on the stability of a cylinder in a wake. Wittenmark (1997). Trondheim. Recommended practice for planning. Handbook STF70 A92006. ˚str¨m. Kristiansen. M. (1994). A. Baarholm. C. PhD thesis. W. L. Flow Induced Vibrations. OTC17290. Departement of Engineering Cybernetics. TX. D.. B. Larsen. S. Journal of Oﬀshore Mechanics and Arctic Engineering 127. Compensator and tensioners. Krieger publishing company. 3rd ed. 39–45. Blevins. Houston. Handbook on design and operation of ﬂexible pipes. Riser interferense tests on ﬂexible tubulars at prototype reynolds number. D. Allen.. T. Technical Report 3700NNX1500117.. Blevins.Theory A o and Design. Technical Report API RP 2T. Henning and L. Fylling. In: Proc. USA. accessed May 18]. Herfjord (2005). Leira and A. Lee (2005). I.
Hydrodynamic forces on a pair of cylinders. Demirbilek. and C. of the 21st International Conference on Oﬀshore Mechanics and Arctic Engineering. (1989a). Tension leg platform . American Society of Civil Engineers. of the 17th Annual Oﬀshore Technology Conference. P. Recommended practice for riser interference. 20021539. Det Norske Veritas. X. of the 12th Annual Oﬀshore Technology Conference. Chaudhury. Coupled dynamic analysis of platforms. pp. Coupled dynamic analysis of a mini TLP: Comparison with measurements. Tapir Forlag. In: Tension Leg Platform . Revision No. F. (2004). Det Norske Veritas. and J.europa. Z. of the 32nd Annual Oﬀshore Technology Conference.an overview of the concept. In: Proc. http://ec. OTC7239. (1993).Y.Bibliography Bokaian. European Commission (2007). PhD thesis. OTC5007. Q.M. OTC3755. Departement of Marine Technology.A state of the art review. Houston. Ed. Duggal.htm. Zhang. pp. Oslo. 413–420. and F. Y. 169– 179. Chen. M. American Society of Civil Engineers. Demirbilek. Houston. Liagre. (1989b). J. Egeland. Vol. An experimental study of tendon/riser pairs in waves. (1980). TX. A. Teigen (2002). OTC12084. Industry version issued DNVRPF203. An Experimental Study of Hydrodynamic Forces on Cylinders and Cables in Near Axial Flow. Servoteknikk. TX. 2. risers. accessed February 5]. 323–333. 3 issued to NDP Report No.A state of the art review (Z. Niedzwecki and P. In: Proc. OMAE200228536. In: Proc. J. TX. TX. S. analysis and design. DNV (2003). A. USA. Geoola (1985). A riser angle positioning system (raps).. G. Ed. Norway. Houston.). DNV (2005). pp. Houston. Norwegian University of Science and Technology (NTNU). [Online. In: Proc. Tension Leg Platform . In: Proc. Dean. Global analysis procedure for riser interaction. O. 3.. GALILEO European lite Navigation System. of the 25th Annual Oﬀshore Technology Conference. and moorings. Vol. W.eu/dgs/energy transport/galileo/index en. Ho (2000). Z. Ersdal. Demirbilek. S. 154 SatelURL . Niedzwecki (1993).
M. Marine Control Systems .. on Oﬀshore Mech. C. Hanson (2002). Haws and D. Experimental and numerical study on wake interference and clashing between steel catenary risers. Smith. Cambridge. Norwegian University of Science and Technology (NTNU). Morse (2002). Ekvall. Larsen. Hespanha. (1990). A. K. Rigs and Underwater Vehicles. C. Trondheim.Faltinsen. SINTEF Marintek. TX. G. Fredheim. K.. O. Fontaine. of the 32nd Annual Oﬀshore Technology Conference. E. (2005). (2002). E. San Diego. Heckler. Trondheim. and Arctic Eng. J. Fossen. P. G Engseth. McDaniel. J. Oslo. Brydum and T. Trondheim. Orlando. Norway. J. R. OTC11996. Norway. Marine Cybernetics..Guidance. In: Proc. Automatica 38(11). Conf. Tutorial on Supervisory Control. R. M. Theory Manual RiﬂexTM34rev0. Trondheim. I. Navigation and Control of Ships. In: Lecture Notes for the Workshop Control Using Logic and Switching for the 40th Conference on Decision and Control. J. 1905–1917. W. Capul. tation of interaction between deepwater risers. Leira.. M. A. T. I.. (2000). T. Switching between stabilizing controllers. In: Proc. Rippol and P. of the 26th Internaltional Conference on Oﬀshore Mechanics and Arctic Engineering. Current Forces on Net Strucutres. Compuas. Herfjord. TX. Departement of Engineering Cybernetics. J. Norway. E. Hespanha. OMAE200228152. (2001). PhD thesis. and A. Fard. Sea Loads on Ships and Oﬀshore Structures. PhD thesis. (2000). H. Modelling and Control of Mechanical Flexible Systems. J. B. Lespinasse (2007). N. T. M. OMAE200729149. Norway. Gopalkrishnan. Cambridge University Press. TLP production riser spacing: Steady drag and lift on two moderately spaced oscillating cylinders. T. Norwegian University of Science and Technology (NTNU). In: 21th Int. Norwegian deepwater program: Improved predictions of vortewinduced vibrations. Holthe (2005). H. Houston. A. H. Omberg. Passano and K. Departement of Marine Technology. S. Holm˚ B. Fylling. B. Technical Progress Report BRC 4194. McMillan (1994). Halse. FL. 155 . Houston. P. collision statistics and stress analysis. P. UK. Norway. Sødahl. CA. Shell Development Company.
S. T. M. Position reference based on instrumented drilling riser. TX.. In: Proc. 263–272. P. Jensen and I. British Columbia. New Orleans. (1993). IEEE Journal of Oceanic Engingeering 28(4). of the 39th IEEE Conference on Decision and Control. Marn (2004). Johansen. Canada. OTC11993. Morse (2000). 135–142. OTC8070. J. Hespanha. pp. Drag in oscillatory ﬂow interpreted from wake considerations. In: Proc. Tanaka and Y. I. I. Imakita. 543–555. of the 25th Annual Oﬀshore Technology Conference. S. E. TX. In: Proc. Nielsen (2003). Experimental study of wake shielding eﬀect of the three cylinders in currents. Liberzon and A. Bolstad and A. and G. Huse. S. Amitani ans S. 156 . In: Proc. pp. Interaction in deepsea riser arrays. Hystersisbased switching algorithms for supervisory control of uncertain systems. E. In: Proc. Houston. TX. 367–372. of ETCE/OMAE Joint Conference. E. Automatica 39(2). LA. Australia. Wave synchromizimg crane control during water enrty in oﬀshore moonpool operations .. OTC5370. of the 19th Annual Oﬀshore Technology Conference. 313–322. S. J. Experimental investigation of deep sea riser interaction. Huse. A. International Journal of NonLinear Mechanics 41(4). Ciatti (2002). Houston. Vancouver. Huang. pp. of the 23rd International Conference on Oﬀshore Mechanics and Arctic Engineering. Leira (2006).. TX. Modelling of inextensible cable dynamics with experiments. E.. J.experimental results. A. OMAE2000/OSU OFT3001. 720–178. S. Morse (2003).. LA. Liberzon and A.. Huse. In: Proc. (1996). N. V. of the 32nd Annual Oﬀshore Technology Conference. In: 14th Deep Ocean Technology & Exhibition. T. In: Proc. Fossen. Impulse and energy in deepsea riser collisions owing to wake interference. Houston. Sagatun and F. of the 28th Annual Oﬀshore Technology Conference. Sørensen and B. P. Høklie. Houston.. New Orleans. OTC7237. Takagawa (2000). Huse. J. Johansen.Bibliography Hespanha. G. A. D. Bounds on the number of switching with scaleindependent hystersis: Application to supervisory control. D. A. Sidney. Ersdal. OMAE200451001. Kleiven (2000). Intelligent riser angle control DPS. S. (1987).
Houston. 8th ed. I. B. Houston. Huang (2000). Lee (2000). Norway.A state of the art review (Z. E. PrenticeHall Inc. H. Technical report. Leira. L. J. University of Michigan. F. Kavanagh. of the 35th Annual Oﬀshore Technology Conference. (1993). OTC15383. John Wiley & Sons. H. K. o Tapir. K. Imas. American Society of Civil Engineers. TX.). In: Proc. Kalleklev. OTC11992. M. A. Advanced Engineernig Mathematics. Langen. 15–20. Dept. Ed. Deepwater nonlinear coupled analysis tool. In: Proc. New Jersey. Sødahl.. (1998). In: Tension Leg Platform . Ma. K. Kreyszig. K. J. TLPs and other deepwater platforms.. J.. TX. (1989). W. Solaas and J. Nyg˚ and A. Inc. W. Larsen. Estimation of extreme reas sponse and fatigue damage for colliding risers. Thompson and L. Trondheim. Lie.Y. In: Proc. Mørk. E. OTC12085.. pp. W. Houston. T. Ocean Engineering 25(7). Larsen. (2000). Norway. Demirbilek.. Khalil.. In: Proc. M. (1999). of the 21st International Conference Oﬀshore Mechanichs and Arctic Engineering. In: Proc. M. ard Design guidelines for riser collision. M.. Litton. Norwegian deepwater program: Analysis of vortexinduced vibrations of marine risers based on fullscale measurements. C.Kaasen. and R. Technical Report STF70 F93011. of Naval Architecture and Marine Engineering. C. M. Holm˚ and K. K. Lee. Zou and E. Herfjord (2002). OMAE200228435. Korde. TX. Dynamisk analyse av konstruksjoner. Vandiver (2000). A. Sigbj¨rnsson (1979). Active heave compensation on drillships in irregular waves. N. Analysis of tensioner system stroke for marine risers. U. OTC11997. J. of the 32nd Annual Oﬀshore Technology Conference. of the 32nd Annual Oﬀshore Technology Conference. TX. Lecture notes NA550/665. of the 32nd Annual Oﬀshore Technology Conference. SINTEF Structures and Concrete. Tension Leg Platforms. Oslo. R. Horn (2003). 157 . H. 541–561. Nonlinear Systems.. 3rd ed. K. W. Houston. (1995). Genesis spar risers: Interference assessment and VIV model testing.
of Oceanography. ASME. B. DiMarco. of Marine Technology. Facts . 142–149.the Norwegian petroleum sector. Jochens. Design of hybrid controller for dynamic positioning from calm to extreme sea conditions. [Online. Ministry of Petroleum and Energy and the Norwegian Petroleum Directorate (2007). OMAE2001/OFT1168. Roesset (1996).. National Oilwell Varco (2007b). Numerical representation of collision interaction mass for dual risers systems. Synthesis Report MMS 2001064. Mortazavi. Automatica 43(5).. Ferrari (2002)..gov/crs/deﬁnitions/loop current. Norway. Kristiansand. OMAE200228336. Nowlin. In: Proc. (2006).. M. Campbell and C. PhD thesis. Neto. P. S. T. Norway. A. Nielsen. accessed April 21]. of 21st International Conference on Oﬀshore Mechanics and Arctic Engineering. New Orleans. R. Reid and M. Design of Hybrid Marine Control Systems for Dynamic Positioning. Nguyen. NOAA Coastal Services Center (2007). Brazil. D. accessed June 8]. F. Dept. Texas A & M Univeristy. URL http://www. URL http://www. G.Bibliography Mekha.csc. C. R. NLine Tensioner Course.. C.nov. Dept. Environmental Conditions. B. D. Ruggieri and J. T. Deepwater physical oceanography reanalysis and synthesis of historical data. National University of Singapore (NUS). (2004).html. Lecture Notes TMR4225 Marine Operations. Technical report. 158 . Implications of tendon modeling on nonlinear response of TLP. Rio de Janeiro. T. 768–785.noaa. [Online. K. D. Johnson and J. Technical Report UK200486. Brinkmann (2001). A. R. of the 20th International Conference on Oﬀshore Mechanichs and Arctic Engineering. J. USA. B. Quek (2007). NTNU. Howard (2001). E. Nguyen. Norsk Hydro (2001). National Oilwell Varco (2007a). In: Proc. M. Homepage. W. Modeling and analysis of toptensioned risers. Loop Current. T.com/. Ormen Lange Project 3700NHF5000001. E. T. Oslo. A. Sørensen and S. O. Journal of Structural Engineering 122(2).
of the 8th International Confeerence on the Behaviour of Oﬀshore Structures (BOSS) (J. 91–99. of the 11th International Oﬀshore and Polar Engineering Conference. DiscreteTime Control Systems.. Larsen (1997). Steinkjer (1998). Perez. Vugts. Botros and H.. Vol. URL http://www. H. H. Sødahl (1997). Croatia. Larsen and N. J. of IFAC Conference on Control Applications in Marine Systems (CAMS07). 2 ed. Coupled analysis of vessel motions and mooring and riser system dynamics. pp.reduced costs for dry well completions in deepwater west africa by application of proven semisubmersible and riser technology. Ogata. In: Proc. Hannus (2001). In: Proc. and P.). In: Proc.oﬀshore–mag. Ormberg. In: Proc.. Poldervaart and M. 2. Dry tree semi . (2000). H. 159 . K. Pollack. 469–483. Modelling and performance of an active heave compensation for oﬀshore operations. and K. 2. Australia. A surface tree riser tensioning system for FPSOs. 222–226. Portugal. [Online. Steinmann (2007). Ormberg. Houston.. Norway. of the 32nd Annual Oﬀshore Technology Conference. Oﬀshore Magazine 66(6). In: Proc. Perez. pp. of the 32nd Annual Oﬀshore Technology Conference. Newcastle. T. Ormberg. Fylling. Houston. Japan.com/. A. N. L. Vol. pp. OMAE980351. Naciri (2000). Eﬃcient analysis of mooring systems using decoupled and coupled analysis. Vol.. Often. Sørensen and M. 1A. Ed. OTC11876.K. accessed August 9]. I. of the 16th International Conference Oﬀshore Mechanics and Arctic Engineering. K. Shell restarts Mars TLP. T. Sødahl and O. The Netherlands.. Lisbon. of the 17th International Confonference Oﬀshore Mechanics and Arctic Engineering. Paganie. between top tensioned risers in deep and ultra deep waters. O.Nyg˚ M. Evalution of eﬀect of contact ard. In: 14th IFAC Symposium on System Identiﬁcation (SYSID). R. includes industry ﬁrsts. OTC11902. In: Proc. Coupled analysis of ﬂoater motion and mooring dynamics for a turret moored tanker. H. F. (1995).a tutorial. J. D. (2006). TX. Yokohama. In: Proc. Oﬀshore Tecnology (2007). TX. The website for oﬀshore oil & gas industry. Blanke (2006). Marine vessel models in changing operational conditions . Bol. Prentice Hall. Stavanger.
Rigzone (2007). In: 7th IFAC Conference on Manoeuvering and Control of Marine Crafts (MCMC06). Field development projects. Journal of Sound and Vibration 38(1). 125–147. Le Cunﬀ and C. Portugal. Sabri. Available online July 5. S. Marine Structures. ISSN 1523651x.. [Online. J. In: Proc. United States Patent 7.com. at sea. F. Biolley. M. Triantafyllou. ISSN 09518339. Putot (2003). L´vine. M. L´vine (2006). Y. Biolley. Rustad. R.Bibliography Price. J. (2006). Sabri. J. Sørensen (2007b). A. URL Reite. Putot. M. notably an oil pipe. of the 26th International Conference on Oﬀshore Mechanics and Arctic Engineering. H. C. Jacobsen and A. Modelling and control of colliding top tensioned risers in deep waters. CA. M. S. M. Measurements. A.J.rigzone. 743–748. RD Instruments (2006). Rustad. In: Proc. URL http://www. Larsen. (1975). M. Larsen and A. C. accessed August 9]. OMAE200729172.686 B2. A. Wake induced ﬂutter of power transimission conductors... C. M. A. F. In: Proc. F. A. Rustad. Le Cunﬀ.. Departement of Marine Technology. J. M. Sagatun. Sørensen (2007c). Modeling and Control of Trawl Systems.. C. Taiwan. Norwegian University of Science and Technology (NTNU). Sørensen (2006). Croatia. I. Taipei. Rustad. Bol. J. Sørensen and C. Hover. of the IFAC Conference on Control Applications in Marine Systems (CAMS07). Creﬀ and J. (2002). PhD thesis. Lisbon.066. 160 . Active control of underwater installation. S.com/.. Marine http://www. e Automatic control methods for positioning the lower end of a ﬁliform structure. Deep water riser collision avoidance by top tension control.rdinstruments. R. C. San Diego. Y. J. Larsen and A. of the 2003 IEEE International Conference on Robotics and Automation (ICRA03). Cone nection of the riser to the subsea wellhead using active control. A. C. M. FEM modelling and control for collision prevention of top tensioned risers. [Online]. IEEE Transactions on control systems technology 10(5). K. Creﬀ. Supervisory switched control of marine risers. S. Larsen (2007a). Elsevier.
A. Transactions of ASME. Technical Report NPS069SL790011. Departement of Engineering Cybernetics. Sørensen. 58–67. J. Strand and C. C. Vol. C. Master’s thesis. Oslo. Heave compensating systems for deep water drilling. I. M. B. and R. (2005). Sørensen. World Scientiﬁc. of the 21st International Conference on Oﬀshore Mechanics and Arctic Engineering. Sparks. H. 1187–1205.. B. Høiskar (2002). Leira. OMAE200228077. Hydrodynamics Around Cylindrical Structures. R. Fredsøe (1997). 125–149. Int. Robust Nonlinear Control 11. Simpson. Norway. pressure and weigth on pipe and riser deformations and stresses. Boundary Layer Theory. S. New York. The Aeronautical Quarterly 22(1). Sagatun. I. In: Proc. The inﬂuence of tension. 375–390. S. Naval Postgraduate School. Sarpkaya. L. A. Skjetne. Norway. Norway. Herfjord. F. Participating mass in colliding risers. B. (1968). J. Shoaﬀ (1978). Annual Reviews in Control 29. (1971). 161 . Huse (1999). Stølen. 25–41. M. Norwegian University of Science and Technology (NTNU). Trondheim. (1984). Holm˚ (2002). A. (2007). K. P. Dynamic simulation of marine as risers moving relative to each other due to vortex and wake eﬀects. K. Journal of Marine Science and Technology 4(2). J. J. F. 46–54. Sørensen. Schlichting. PhD thesis. P. McGraw Hill Book Company Inc.. T. E. 12 of Advanced Series on Ocean Engineering. Trondheim. Larsen (2001). Sumer. NTNU.. A. Leira.Sagatun. M. A discrete vortex analysis of ﬂow about stationary and transversely oscillating circular cylinders. Optimal setpoint chasing in dynamic positioning of deepwater drilling and intervention vessels. G Nielsen and E. and J. Structural issues in the design and operations of marine control systems.. Journal of Fluids and Structures 16(3). On ﬂutter of a smooth circular cylinder in a wake. (2005). Journal of Energy Resources Technology 106. Herfjord and T. Larsen and P. J. S. J. The Maneuvering Problem. Riser angle based position reference system.
In: Proc. TX. John. J. Huang and N.org/wiki/Sonar. Tsui. of the Annual Oﬀshore Technology Conference. (1983). Wroldsen. Koos (2000). Houston. OMAE2001/OFT1170. OMAE99/OFT4211. TX. 162 . In: Proc. URL http://en. Departement of Marine Technology. Operability improvment of deepsea riser by active control. of the 32nd Annual Oﬀshore Technology Conference. W. Huang and N. Technical Progress Report WRC 5783. of the 31st Annual Oﬀshore Technology Conference. Developements in acoustic doppler current proﬁling and its technical application in deepwater drilling. Norway. [Online.the free encyclopedia.wikipedia. Norwegian University of Science and Technology (NTNU). Yoshida (1995). In: Proc. T¨rkyilmaz. Wikipedia (2007a). the free encyclopedia. R. (2007). Wu. Y. TX. St. van Smirren. Wikipedia (2007b). Tsahalis. OTC11904. Shell Development Company. OTC4490. H. (2004). (1977). Houston. T. PhD theu sis. Romeo and M. Copenhagen.org/wiki/Loop Current. accessed April 21]. Barltrop (2001a). In: Proc.. Modeling and Control of Tensegrity Structures. of the 18th International Conference on Oﬀshore Mechanics and Arctic Engineering. pp. Journal of Applied Mechanics 44. PhD thesis. 194–200. [Online. Drag coeﬃcients of longﬂexible cylinders. T. K. Vandiver. Loop current . Sonars . Houston. Canada. Norwegian University of Science and Technology (NTNU). Norway. In: Proc. Rio de Janeiro. of the 20th International Conference on Oﬀshore Mechanics and Arctic Engineering.. Denmark. Houston. Wu.Bibliography Suzuki. Barltrop (1999). of the 14th International Confonference on Oﬀshore Mechanics and Arctic Engineering. On wakeinduced ﬂutter of a circular cylinder in the wake of another. URL http://en. pp. Multiple stable/unstable equilibria of a cylinder in the wake of an upstream cylinder. S. In: Proc.wikipedia. 405–410. TX. Wanvik. Two tier well top tensioning system. Trondheim. Watanabe and K. Brazil. Trondheim. A. (1984). Vol. K. Lift and drag forces on a cylinder in the wake of an upstream cylinder.. 289–297.Wikipedia . Y. accessed July 2]. IB. J. and J. Vogel (1999). L. W.. Modelling and Control of Towed Seismic Cables. D. Experimental study and modelling of the vortexinduced vibrations of a pair of risers/tethers of a compliant structure.Wikipedia. S. OTC 10752. Departement of Engineering Cybernetics.
pp. Review of ﬂow interference between two circular cylinders in various arrangements. M. Flow around circular cylinders. S. 201–226. Trondheim. In: Proc. M. pp. Stavanger. Barltrop (2001b). Wakeinduced large amplitude lowfrequency motions of a vertical riser in the wake of an upstream riser. Norwegian Institute of Technology. Norway. In: Proc. (1977).. (1985). 214–221. Vol. Zdravkovich. (2003). M. Zdravkovich. W. W. of the Conference Separated Flow around Marine Structures. Huang and N. Oslo. Oxford Science Publications. Journal of Fluids Engineering 99(4). of the 11th International Oﬀshore and Polar Engineering Conference. Norway..Wu. OMAE200228035. In: Proc. of the 21st International Conference on Oﬀshore Mechanics and Arctic Engineering. Oxford. Wu. M. Barltrop (2002). Forces on pipe clusters. 163 . 2. Norway. M. S. Stationary and hopf bifurcations of equilibrium positions of a cylinder situated in the near and far wake ﬁelds of an upstream cylinder. M. Zdravkovich. Huang and N. 618– 633.
.
By assuming twodimensional motions. Vc is the undisturbed free stream velocity. U0 is the maximum velocity in the velocity proﬁle. u is the wake velocity proﬁle and CD1 is the drag coeﬃcient of the riser generating the wake.1) (A. A. xs y b 2 (A.3) u (y) = U0 exp −0. The load on the risers and the quasistatic algorithm are carefully explained. CD1 D1 .25 U0 = Vc CD1 D1 xs . xs is the distance from the wake source.Appendix A Mathematics and Algorithm for the FEM Model This appendix includes a more detailed explanation of the wake ﬁeld model applied in the system. and the damping model discussed. D1 is the diameter of the riser generating the wake. (A. neglecting viscous stress and holding the pressure constant through the ﬂuid.1 Wake Field of a Single Cylinder The downstream riser (R2) will experience shielding eﬀects due to the upstream riser (R1). the turbulent wake ﬁeld can be expressed as b = 0. The mass and stiﬀness matrices for an element are derived.693 where b is the half width of the wake. Schlichting (1968) solved the equations of motion in a wake.2) . 165 . The prediction of the current on the downstream riser will be based on wake and momentum considerations given by Huse (1993) and references therein such as Schlichting (1968).
1) is only expected to be valid some distance downstream of R1. where xs = xv . This can be solved by using the root mean square (RMS) average over the riser diameter. Close to R1. y) dy −D/2 (A.R2 = 0.7) (A. 2 (A. D/2 u2 (xs .11) The discrete RMS value (in general) is given by R (x) = Σn x2 i=1 i . and hence lead to erroneous results when calculating the force on R2 placed in the wake. on which the force is calculated. such that this velocity is not well deﬁned.6) D1 . y) y dy . we can ﬁnd xv by 1 4 CD1 D1 xv = D1 . whereas it makes a large diﬀerence in the wake ﬁeld close to R1. 2 (A.8) (A. a virtual source upstream of R1 is introduced. 4 4D1 xv = .12) 166 .5ρD2 CD2 Vc − u2  (Vc − u2 ) . This means that the distance xs in the equation above is substituted by xs = x + xv . The drag force on each element of R2 can now be found by Morison’s equation fdrag.4) where xv is the distance from the virtual source to the upstream riser. this will give a wake peak which is too high and narrow.A. In order to correct for this.5) 1 2 CD1 D1 xv = D1 . and x is the distance between the centers of the wake generating riser R1. By requiring that b= at R1. n (A.10) A problem is though that u2 varies over the space occupied by the riser R2. and the downstream riser R2. (A. CD1 At large distances this correction does not make much contribution. (A.9) (A. Mathematics and Algorithm for the FEM Model Equation (A. The analytic RMS value is given by uRM S (xs ) = D/2 2 −D/2 u2 (xs .
z2 ) is given by N = 1 − x. 1979). ﬁve halfside integrating points should be appropriate.2 System Matrices Dynamic analysis by the ﬁnite element method consists of four steps (Langen and Sigbj¨rnsson.System Matrices such that the numeric RMS value for the current on the riser diameter is unRM S (xs ) = Σm un (xs .1 Mass The mass matrix is denoted consistent because the same interpolation polynomial is used for derivation of the displacement for both the mass and the stiﬀness matrices. (A. 2 NT Ndx 1 lx (A. A. such that m2×2 = V l l 0 ρNT NdV = ρA 1 − 1x l 1 1 lx 1− lx . A. The steps three and four are given in Chapter 3. system analysis.15) l l l is the element length.18) = ρA 0 1 − 1x l 1 2 x l2 = ρAl 6 2 1 1 2 167 . The load vector is derived in Appendix A. is to subdivide the structure o into elements.16) dx (A.2. is to compute the mass and stiﬀness matrices and the load vector for the whole structure by concatenating the corresponding element matrices. discretization. x2 ) and (z1 .13) Since it is a symmetric Gaussian function. y)2 i i=1 . element analysis. The fourth and last step is the dynamic equilibrium equation. x . (A. is to establish the mass and stiﬀness matrices and the load vector for each element.14) where the interpolation polynomial for each pair of coordinates (x1 . The ﬁrst step. The second step.3. The third step. From Langen and Sigbj¨rnsson (1979) the consistent mass matrix for o each element is written m= V ρNT NdV = ρA 0 l NT Ndx. The second step for the mass and stiﬀness matrices are derived here. m (A.17) (A.
0 0 (A. The matrix for the added mass is given below as 2 ρw Ae l 0 mai = 6 1 0 0 0 0 0 1 0 2 0 0 0 .24) (A. 1 .A.22) where E is Young’s modulus of elasticity and the interpolation polynomial is given by d B= N = 1.21) where Ae is the external area and ρw the density of water. 0 2 (A. (A.23) l l dx such that l k2×2 = 0 EABT Bdx 1 −1 −1 1 .2 Stiﬀness The stiﬀness matrix for one element is given in Langen and Sigbj¨rnsson (1979) o as l k= 0 EABT Bdx. The added mass does only have components in the xdirection. Mathematics and Algorithm for the FEM Model The total structural mass matrix for one element 2 0 1 ρs Al 0 2 0 msi = 6 1 0 2 0 1 0 is then 0 1 . The shape of the mass for the internal ﬂuid is similar 2 ρf Ai l 0 = 6 1 0 0 2 0 1 1 0 2 0 0 1 . (A.19) where A is the cross sectional area of the riser and ρs is the density of the riser material steel.2. 0 2 mf i (A.20) where Ai is the internal area of the riser and ρf is the density of the internal ﬂuid.25) = EA l 168 . A. (A.
the structure is slender if λw > 5.26) A. D (A. a linear viscous (velocity proportional) damping in the structure itself is used. a slender structure.25). The ﬁrst term is proportional to velocity and referred to as potential damping caused by wave generation.27) li −1 0 1 0 0 0 0 0 .2. As a rule of thumb.e. the kinetic energy in this system will decrease. A simple way of modeling this eﬀect is to apply the Rayleigh damping given by C = α1 M + α2 K. When a structure oscillates in a ﬂuid. i. kGi = (A.3 Damping The damping matrix describes the structure’s ability to dissipate energy. and D is the diameter of the slender structure. To model the damping correctly is diﬃcult. an external energy supply is added. and the geometric stiﬀness matrix for one element is given by 1 0 −1 0 Pi 0 0 0 0 . This hydrodynamic damping usually consists of two terms. a part of the dynamic ﬂuid pressure on the structure will be in phase with its velocity. (A.28) where λw is the wave length. but simpliﬁed models do usually give representative solutions. the nonlinear drag term will dominate and the linear term be negligible. (A. Drag damping is taken into account by direct use of Morison’s equation.System Matrices The elastic stiﬀness is acting in the axial direction. When the cross section of the structure is small compared to the wavelength. The second term is the drag force due to formation of vortices and is described by Morison’s equation. in this case. When damping is present in a system. stiﬀness matrix for each element i 0 0 0 0 EA 0 1 0 −1 kEAi = l 0 0 0 0 0 −1 0 1 and gives the local elastic The geometric stiﬀness acts in the lateral direction.29) 169 . If it does not. see (3. Structural damping should also be accounted for. This pressure is referred to as the hydrodynamic damping.
The damping level. ω2 − ω1 2 (λ2 ω2 − λ1 ω1 ) . i where φi is the mode shapes and λi the damping relation given by λi = ci 1 = 2mi ωi 2 α1 + α2 ωi .33a) (A. If we know the overall damping level for two frequencies we can determine α1 and α2 by α1 = α2 = 2ω1 ω2 2 2 (λ1 ω2 − λ2 ω1 ) .30) Hence. which we will ﬁnd by using Morison’s equation. 2 2 ω2 − ω1 (A. ωi (A. The current loads will induce forces on the riser. This damping contribution is not wanted since structure damping is mainly caused by strains in the material of the structure.33b) The inertia term leads to damping from rigid body motions in the same way as these motions give inertia forces. Mathematics and Algorithm for the FEM Model The inertia proportional term will damp out the low frequencies. The damping ratio will.A. Using orthogonality properties from Langen and Sigbj¨rnsson o (1979). (A.32) (A. The algorithm for loads. while the stiﬀness proportional term has the opposite eﬀect and will tend to damp out the higher frequencies.31) for i = j. the inertia term in Rayleigh damping should not be included. A. The stiﬀness term α2 K is directly related to strains in the structure and can model interior structural damping adequately.3 Load on the Risers The risers will experience load in the global horizontal direction due to currents. must therefore be correct at the actual response frequency. The inﬂuence from damping at higher frequencies is without signiﬁcance for the solution. λi is proportional to ωi and for α2 = 0. λi is inverted proportional to ωi . we see that C will have the same properties as M and K φT Cφj = α1 φT Mφj + α2 φT Kφj = 0 i i i The modal damping coeﬃcients are given by ci = φT Cφi = α1 mi + α2 ki . for α1 = 0. Hence. and may in fact contribute to a more smooth time history since false oscillations caused by the numerical method might be suppressed. λi . forces and displacements for the static initialization procedure could be summarized as follows: 170 . however. increase with increasing frequency.
i (A. has to be decomposed into the local frame of the node at which the current attacks. The ﬂow is linear.i . First the current.i cos θi f vx. given in the global frame. The decomposed current in iframe is then f vcur. 4.36) i = = i f vx. there are no vortices in the ﬂow.Load on the Risers 1.i i vx i vz f vx f vz f vx 0 = .x = i fext. is described in the iframe. When a force. it is implicitly for element i only. Add the forces for each node to ﬁnd the total force in the global frame.x is the external force in xi direction for element i. 2. Introduce load terms from for the speciﬁed motions from the TLP. 2 (A.i sin θi . 6.z = 1 i i ρw Di CDn li vx vx .35) f vx = = i f i T0. vcur. Transform the forces back to the global system. (A. i cos θi − sin θi sin θi cos θi f vx cos θi f vx sin θi 0 (A.i = i vcur. Below each point is explained in detail. only has a horizontal component. 7. 2 1 i i ρw Di CDt li vz vz . 5. such that f the vz component is zero. Decompose the current from the global frame to the local frame.34) (A.e. 171 . stiﬀness matrix etc.38) (A. i. Calculate the load at each node for each element locally.39) i where fext.37) Calculate the load at each node for each element locally. It is assumed that the global current f velocity vector in each node i.i . 3. Find the displacement vector by r = K−1 R.f vcur. The loads in each element’s two nodes are found by Morison’s equation for the axial and transverse direction i fext. Calculate the forces at each node for each element locally. Decompose the current from the global frame to the local frame.
i+1 sin θi .45) Calculate the forces at each node for each element locally.A.1 vz. 0 6 qz1 + 2qz2 i fn = (A.2 = 1 f f ρw Cdn Di vx.1 vx.i cos θi .40) (A.47) i where fn is the normal and fti the tangential forces.i sin θi vx. Assuming linearly varying transverse and axial load.i sin θi .1 . The forces in zi directions will not have major impact on the dynamics. The resulting 172 . 2 1 f f ρw Cdt Di vx. The sum of forces acting on each local element in its frame is then given by i f i = fti + fn .i+1 cos θi vx. 2 1 f f ρw Cdt Di vx.1 = qxi. The load in each direction is given by i qx. 2 1 f f ρw Cdn Di vx.1 .41) This gives the load in each of the two nodes of an element to be qxi. 6 0 0 l 2qz1 + qz2 .i+1 sin θi vx. 2 1 i i ρw Cdt Di vz. Mathematics and Algorithm for the FEM Model ρw is the density of the displaced ﬂuid.44) (A. (2005) and is given as 2qx1 + qx2 l 0 qx1 + 2qx2 .43) (A. 2 (A.1 = i qz.2 = qzi.1 = qzi. vx and vz are the current velocities in the local iframe.42) (A. hence they could be neglected. 2 (A.i+1 cos θi . Di the diameter of element i.46) fti = (A.i cos θi vx. and CDn i i and CDt the drag force coeﬃcients in normal and tangential direction. (A. the load on a bar is found in Fylling et al.48) This could be simpliﬁed to include forces in the xi direction only since these are far larger than the forces in the zi direction.1 = 1 i i ρw Cdn Di vx.
52) After summarizing the forces at each node. (A. for the quasistatic case is then found to be r = K−1 R. and kspe is a vector due to the speciﬁed motion at the top node in surge only. (A. f f f fz. Find the displacement vector by r = K−1 R.Load on the Risers external force in each element is then given in the same direction as the load as 2qx1 + qx2 l 0 fi = (A. All vectors are described in the global f frame. Before adding the forces acting at each node.i−1 .i = fx1 .51) (A.53) where fcorr is the corrected force vector. kspe is the speciﬁed stiﬀness vector and xT LP is the scalar TLP motion in surge. After the forces at all nodes are transformed back to the global coordinate system. the total force vector in f frame due to the current forces is called fdrag . Upper case is used here since this is most common within the ﬁeld of structural mechanics for load and force vectors. Add the forces for each node to ﬁnd the total force in the global frame. The forces are transformed from its local frame to the global frame the same way as the displacements fif =Tf f i .50) i where fif is the external force in the two nodes in element i given in the global f frame. 173 . Introduce load terms for the speciﬁed motions from the TLP. 6 0 Transform the forces back to the global system. The speciﬁed motions for the quasistatic case are accounted for by correction terms in the load vector.49) qx1 + 2qx2 .i−1 . they have to be represented in the same frame. See the following iteration algorithm for equilibrium between internal and external forces.i = fz1.i + fx2. fcorr is a vector due to the twodimensional model. (A.i + fz2.54) where R is the force vector. These are found from fcorr = fdrag − kspe xT LP . the forces and moments are summarized in each node f f f fx. The displacements of the riser (A.
Initialization 1. Find the axial eﬀective tension in the middle of each element given as the tension in the element above.57) where Pi is the eﬀective tension in element i positive vertical upwards. and the setdown correction at each node due to current loads and deﬂections. This includes corrected element lengths due to the elongation and top tension. the Pythagorean theorem. wt is the true weight in air and n is the number of elements. minus half the weight of the above and the current element. weight and current loads. .i+1 ) . see Sparks (1984) and Larsen (1993). 174 .58) The pressure terms take care of the contribution from external and internal pressure in the riser. ∆li and elastic stiﬀness EA . The eﬀective tension in each element is found by 1 Pn = Ptop − wef f. .n . 2.i . (A. tensioned riser is applied. but are constant in time as long as the pressure components are constant. Mathematics and Algorithm for the FEM Model A.i + wef f. (A. (A.i = (wt − ρw · g · Ae + ρi · g · Ai )i l0.4 QuasiStatic Algorithm This algorithm seeks to ﬁnd the equilibrium between the internal and external forces at each node.A.56) 2 wef f. Find the new length of each element given by li = l0 + Pi l0 + pressure terms. The internal forces are the axial forces and are found from the position of each node.i is the eﬀective weight of element i. It is hence not necessary to include these terms in the present analysis. Step 1. The element lengths will not be equally long and an updated stiﬀness matrix for the new element lengths is needed. 3. l The external forces are given by the top tension in the upper node.55) 2 1 Pi = Pi+1 − (wef f. The initial top tension Ptop of the vertical. EA (A. i = {1. n − 1} . . l0 is the length of each element i for a stressfree conﬁguration. . 4. and wef f.
Step 2. 175 . Pi ) . Incremental means stiﬀness against further displacements from the deformed state. .61) The xi positions are all zero.63) (A. Move the TLP to an oﬀset position and add current forces to the system.60) (A.. Correct for the new element lengths and ﬁnding the new vertical positions of the nodes.59) (A. zi = zi−1 + li . such that the global and local frames are parallel. Tf i = 1 0 0 1 . The matrix with rows and columns corresponding to the nonprescribed degrees of freedom is named Kincr .62) In this ﬁrst iteration with the vertical riser. ¯ ¯ ¯ kn−1 + kn kn 11 12 22 ¯ ¯ kn kn 21 22 This matrix K is the incremental stiﬀness matrix for the vertical riser. The top tension and eﬀective weight are now included in Kincr.65) . This gives ¯1 ¯ k1 k11 12 ¯ ¯ ¯ ¯ k1 k1 + k2 k2 22 11 12 21 ¯2 ¯2 + k3 ¯ k21 k22 11 K= (A. Hence. 6.QuasiStatic Algorithm 5. . (A. θi = 0. The updated vertical zi positions are together with the xi positions stored in the vertical displacement vector r0 . zn+1 = zn + ln = ztop . Concatenate the local stiﬀness matrices to make the global stiﬀness matrix. we have z1 = 0. all the inclinations θi will be zero. Starting from the bottom and building the vector upwards. Find the local stiﬀness matrices for each element.k .64) 7. (A. Pi ) +kGi (li . This gives ki = kEAi (li . (A. whereas prescribed motions are stored in a matrix Kpre . due to the vertical riser.
Add the forces to ﬁnd the global vector. The new.4 are assumed correct. incr. ∆rk is the displacement correction term and ∆Rk is the forces from current loads and TLP displacement. by following the list below.1 rupdate = r0 + ∆r1 .67) The initial value of the current velocity on the downstream riser is assumed to be equal to the undisturbed free stream velocity in this initialization. (A.pyth = 2 li − ∆x2 .66) where rupdate is the updated riser position vector and ∆r1 is the contribution from the iteration. The forces from top tension and eﬀective weight are already included in the stiﬀness matrix. Solve the ﬁrst iteration Kincr. i (A. Geometric correction of the vertical position due to deﬂection.3. corrected vertical position for node i is found by the Pythagorean theorem zi.69) (A. This is used to correct the displacements in the vertical direction and identify the vertical position.1 = fdrag − kspe · rT LP. fdrag is found from the current load and Morison’s equation. the inclination is zero.1 . (A. x i = 0 and the zi positions are given from the updated length elements. This is described in detail in Appendix A. such that the risers are standing in parallel. b) The force correction due to the boundary conditions in TLP is found from ∆R corr. 2. 1. This will give the setdown of the top node of the riser. 176 . such that the global and local frames are parallel. In this ﬁrst iteration. c) Here the new positions for the riser elements are found after the TLPposition is included ∆r1 = K−1 ∆Rcorr.70) (A.1 . The element lengths li found in Step 1.71) ∆xi = xi+1 − xi . Calculate the external load q in each element i.1 ∆r1 = ∆R1 . In the vector r0 . and the corresponding forces. and the horizontal displacements are accepted.A. a) Find the external load. Mathematics and Algorithm for the FEM Model k is the number of iterations within the time step.68) (A.
70). 3. The corrected vertical position is the sum of the starting point and the distance found by (A. In this position. the vertical positions are given by z1. Starting from the bottom node.73) with origin of the global coordinate system in the bottom node.i at each node.pyth . we start the equilibrium iterations in each node. . zi.cor + zi−1. and zi positive upwards.1 = rupdate. To ﬁnd the internal forces in the global frame in this equilibrium iteration. . The corrected displacement vector rcorr. . . will not be considered in the equilibrium iterations. 1.corr = zi−1. The nodes and freedoms considered are the same as found in the stiﬀness matrix for the nonprescribed degrees of freedom. assuming that the internal axial force is the same along each element.72) (A.z . The eﬀective weight should be included in the external force when ﬁnding the equilibrium balance with the internal forces. and the axial tension from Step 1.i at each node should be balanced with f the external load fext.corr = 0. the two degrees of freedom in the bottom node and the horizontal degree of freedom in the top node. i = {2. Hence. (A.3. li (A. (A. see Fig. n + 1} . f The resulting internal force vector fint. 177 .9.pyth .74) The sin θi and cos θi values to be used in transformations between local and global coordinate systems are found from cos θi = sin θi = zi. whereas the previous values from the rupdate vector are kept for the xi positions rcorr.75) (A. In the initialization procedure only the static situation is considered. li ∆xi .2. whereas in the time simulation. The vertical position of the next node is found by taking the previous node as the starting point for the next one.76) Step 3.x + rcorr.1 is now found as the sum of the corrected vertical positions.QuasiStatic Algorithm which means that ∆xi is positive if the upper node is to the right of the node below. the solution is for the dynamic system. Equilibrium iterations. we use the inclination values found in Step 2.
it is the eﬀective tension found in previous steps. The eﬀective weight is positive downwards in the global system. 2. To ﬁnd the total external forces fext . at each node (i. (A. Pi−1 is downwards. the current is a function of the distance from the upstream risers. fdrag. and transform the two axial tensions to this frame. For the downstream risers. we subtract the eﬀective weight.79) (A. assumed constant) from the drag forces.83) (A. f 3. f f f fext. whereas Pi is upwards. since this is the one we believe most in at this point of the algorithm. (A. The secant angle over two elements between nodes i − 1 and i + 1 is used as an approximation for the tangent at node i sin σi = cos σi ∆xi ∆xi √ . In the ﬁrst iteration. The unbalanced force in the vertical direction is included in the equilibrium iterations. ∆xi + ∆zi = xi+1 − xi−1 .78) (A.84) ∆zi = zi+1 − zi−1 . i−1 i−1 f Pi − Pf .k = fdrag. (A. The force is thereafter transformed back to the global system. half element over and under. (A. Pi is the i local internal axial force vector of element i in frame i. we use the inclination of this element.85) The drag forces are found using Morison’s equation locally at each node.77) (A.i at each node.86) 178 . Mathematics and Algorithm for the FEM Model Decomposing the forces to ﬁnd fint. and is used to calculate external force due to the current.81) (A.A. i i = = Tf Pi−1 .k normal to the element at each node.82) (A. i−1 The resulting force from the axial forces at the nodes are calculated from secants for each element. ∆xi + ∆zi ∆zi = √ .k − wef f .80) = Tf Pi .e. we ﬁnd the reaction force which gives equilibrium in the horizontal direction. seen from node i Pi = i Pf i Pf i−1 f fint. For the upper element we have σn = θ n . This inclination is found in the global system. wef f .i 0 Pi . At the upper node.
k n+1 i=1 i xi  p . Pi )) Ti .k is the sum of internal and external forces as these are pointing in opposite directions f f fext. (A. 6.QuasiStatic Algorithm 4.z.88) p = i ∆Rf unb.i − fint. 5. Pi ) +kGi (li . p 1/p (A. The unbalanced force vector ∆Rf unb.92) . but in opposite directions. Here we have used the l2 norm.i .k ∆rk = ∆Runb.k . The iteration procedure: The element lengths are still assumed to be correct.1 and the local matrices are the same as in Step 1.x. rcorr.i)2 < 0. (A.1%Ptop . . 1/p xp = ∆Rf unb.k. The stiﬀness matrix Kk is updated with the new positions. (A.z.90) If this diﬀerence is larger than the given acceptance level. 7.. The incremental riser displacement is found from Kincr.i )2 + (fext.k = ∆Rf unb. for instance 0.k − fint. (A.91) The full stiﬀness matrix is found.k .6. The criteria for ending the iteration is given by the pnorm of the unbalanced force for each degree of freedom at each node.93) 179 .i − fint. The transformation matrix Tf is used to transform the local matrices to the global frame calculations i kik = Tf (kEAi (li . we have to start an iteration process to ﬁnd the equilibrium. (A.x.1% of the applied top tension. but only the nonprescribed degrees of freedom are included in the matrix used in the equilibrium equations.87) The resultants should be equal. f i ¯1 ¯ k1 k11 12 1 1 + k2 ¯ ¯ ¯ k21 k22 ¯11 k2 12 ¯2 ¯2 + k3 k21 k22 ¯11 Kk = . ¯ ¯ ¯ kn−1 + kn kn 11 12 22 n ¯ ¯ k21 kn 22 (A.89) (fext.
98) (A. deﬁned by x p = (x1 p + x2 p + · · · + xn p )1/p . the current is updated for the new position and hence the new relative distance. New axial force in each element in the new. ∆zi = zi+1 − zi . that is 180 . The norms are labeled with subscripts. the norms most used are is p = {1. l The new inclinations θi and σi are found in the same step as before.k + ∆rk+1 . The incremental riser vector is added to the previous corrected riser vector rcorr. We can then ﬁnd the internal and external forces and hence the unbalanced force.97) (A. • x = 0 if and only if x = 0. (A.96) (A.99) ∆xi = xi+1 − xi .i = 2 ∆x2 + ∆zi .i − l0 . The norm x of a vector x is a realvalued function with the properties: • x is a nonnegative real number. 2 ∞}.100) where p is a ﬁxed number.k rcorr. 10. ∆li = lnew.5 Metrical Norms The vector norms are deﬁned in Kreyszig (1999). The most important is the pnorm.A. The new positions are calculated locally by lnew. In practice. 8. corrected position is calculated locally by the Pythagorean theorem.k is the updated stiﬀness matrix. i (A. • kx = k x for all k. Mathematics and Algorithm for the FEM Model where Kincr. EA Pi = ∆li .95) (A. (A.k+1 = rcorr. ∆rk is the incremental riser displacement vector and ∆Runb. • x+y ≤ x + y . A.94) 9.k is the unbalanced force for iteration k. For the downstream riser.
∞ 181 . l2 norm: x 3. n 1 2 = maxj xj . l1 norm: x 2. = (x2 + x2 + · · · + x2 ). l∞ norm: x 1 2 = (x1  + x2  + · · · + xn ).Metrical Norms 1.
.
1 and illustrated in Fig. Recall that GoM1 is a wind driven proﬁle and GoM2 has a loop eddy in the top layer. Furthermore.2.1.1.2 are given here. (2001) and scaled to 1200m water depth and surface velocity of 1m/s. 10 and 100 years return periods. The geographically based proﬁles from GoM are found in Table B.1 Current The data for the current velocity proﬁles presented in Section 3. This includes the environmental data. The proﬁle OL1 from the Ormen Lange ﬁeld is based on Herfjord et al. 3. the riser characteristics and controller gains. All current velocities are given in m/s.1 Environmental Data The environmental data presented here include the current velocity proﬁles introduced in Section 3. It is also found in Table B. The Ormen Lange design current proﬁle. uniform. “” means that the current velocity is linear between the values at two given water depths. OL2. i.2.e. (2002) and scaled to 1200m waters.Appendix B Simulation Data The data used in the veriﬁcations in Chapter 4 and the simulations in Chapters 7 and 8 are given here. They are based on Nowlin et al. linearly sheared and bidirectional proﬁles are given in Table B.1. data for tide and TLP oﬀset and dynamics are included for fulﬁllment and validation of the values used in the simulations in this thesis. B. The theoretical proﬁles. These data are from the Ormen Lange Field in the North Sea and data are found in Norsk Hydro (2001) and Aker Maritime (2002). which is used in most simulations and ﬁgures in this thesis is found in Norsk Hydro (2001) and Aker Maritime (2002). B. and reproduced here in Table B.3 for 1. 183 .
0 Table B.18 0.0 1.45 0. Simulation Data Depth [m] 0 1200 Uniform 1.8 0. B.5 0.4 0.65 1.2m gives a tidal amplitude of 1.0 0 1.3 0.0 1.65 0.25 0. Depth [m] 0 120 240 360 480 600 720 840 960 1200 Velocity[m/s] GoM1 GoM2 OL1 1.1.0 0.2 0.0 0. The tide will have a period around 6 hours.2: The geographically based design current proﬁles.3 TLP Oﬀset and Dynamics The extreme case vessel oﬀset is based on the 100 year wave.0 0.1m.2 Tide The tidal range with 100 year return period is according to Norsk Hydro (2001) equal to 2.5 0.2 1.1: The theoretical current proﬁles.65 0 Table B.55 0. 100 year wind plus 10 year current.1.0 Velocity[m/s] Linearly Sheared Bidirectional 1.B.4 0. B. This gives a mean oﬀset of 50m and a dynamic oﬀset of 184 .65 0.15 0.
00 0.25 1.7 gives the controller gains for the second control objective principle. B.60 0.10 0.80 0.05 1.25 1.05 1.3. B. Table B. Table B.95 0.40 1.30 1. The riser data for 1200m and 300m water depth are found in Tables B.Riser Data Depth below sea level [m] 0 (surface) 20 50 100 200 300 400 600 750 850 Velocity[m/s] 1 year 10 year 100 year 1.95 0. We mainly used 30m static oﬀset.8 the controller gains used in shallow water simulations are found.55 1. 185 .10 1.9 gives the control parameters for the supervisory switched control in Section 8.5. 20m dynamic LF oﬀset and 12m WF oﬀset in the simulations.30 1.30 1.70 Table B. 2002).15 1.60 1.75 0. and Table B.3 Controller Gains The controller gains used in the simulations in Chapter 8 are given in this section. The maximum oﬀset is then said to be 67m (Aker Maritime.10 1.15 1.3: Ormen Lange design current velocity proﬁle for various return periods.35 1.70 0.5. 17m. respectively.6 gives the controller gains for both risers used in the investigation of the ﬁrst control objective principle.40 1.40 1.20 1. In Table B.4 and B.05 0.15 1.2 Riser Data The riser data decide the static and dynamic riser characteristics.55 0.
4: Riser data for 1200m water depth.3 1200 0.0477 0.B.5 7850 800 1026 15 Dimension [] [m] [m] [m] [GPa] [MPa] [m] [m] [kN] [kN] [m] [kg/m3 ] [kg/m3 ] [kg/m3 ] [D] Table B.0 2.015 1.0 206 500 1212 1166 2700 1200 0. Simulation Data Parameter α2 De d th CD CM E fu lr lt Tmax Tmin ξ0 ρs ρf ρw ∆xd Description Damping coeﬃcient Diameter Water depth Wall thickness Drag coeﬃcient Mass coeﬃcient Modulus of elasticity Yield stress steel Riser length Tendon length Upper tension limit Lower tension limit Initial payout Speciﬁc weight for steel Speciﬁc weight for ﬁlling Speciﬁc weight for sea water Initial riser distance Value 0. 186 .
2.0 206 500 312 266 2700 350 0.2.0477 0. 187 .5: Riser data for 300m water depth.3 and 8.015 1.Controller Gains Parameter α2 De d th CD CM E fu lr lt Tmax Tmin ξ0 ρs ρf ρw ∆xd Description Damping coeﬃcient Diameter Water depth Wall thickness Drag coeﬃcient Mass coeﬃcient Modulus of elasticity Yield stress steel Riser length Tendon length Upper tension limit Lower tension limit Initial payout Speciﬁc weight for steel Speciﬁc weight for ﬁlling Speciﬁc weight for sea water Initial riser distance Value 0.5 7850 800 1026 8 Dimension [] [m] [m] [m] [GPa] [MPa] [m] [m] [kN] [kN] [m] [kg/m3 ] [kg/m3 ] [kg/m3 ] [D] Table B.6: Controller gains used for both risers in Sections 8. 8. Parameter KP TI Description Proportional gain Integration time Value 607500 8 Dimension [s] Table B.3 300 0.0 2.4.2.
5 Dimension [s] Table B. Simulation Data Parameter KP TI Description Proportional gain Integration time Value 20250 30 Dimension [s] Table B.B.2.3.2 20 0. 188 . Parameter KP TI Description Proportional gain Integration time Value 1000000 8.3 15 Dimension [s] [s] [D] Table B.5.3. Parameter KP 2 KP 3 TD3 TI h ∆xd Description Proportional gain 2 Proportional gain 3 Derivation time Integration time Switching hysteresis Desired distance Value 81000 27000 0.9: Controller gains for supervisory switched control in Section 8.7: Controller gains for the second control objective principle in Section 8.8: Controller gains for shallow water in Section 8.4.