Marine Control Systems

Propulsion and Motion Control of Ships and Ocean Structures

Lecture Notes

Asgeir J. Sørensen
Department of Marine Technology
Norwegian University of Science and Technology

Report UK-13-76

c
Copyright °2013 Department of Marine Technology, NTNU.
Preface

Marine Cybernetics is a multidisciplinary education and research program offered by the De-
partment of Engineering Cybernetics and the Department of Marine Technology, the Norwegian
University of Science and Technology (NTNU).

The material presented is intended for use as Lecture notes in the course entitled TMR4240
Marine Control Systems and partly in TMR4243 Marine Control Systems II. The courses are
given on graduate level for MSc degree for specialization in Marine Cybernetics. It is believed
that these courses in addition to the course TTK4190 Guidance and Control will give the
students good insight into the design of marine control systems. Besides several dedicated
courses on dynamics, hydrodynamics, machinery systems, nonlinear control, stochastic control,
instrumentation systems and computer science are given.

It is assumed that the students already have acquired themselves basic background in auto-
matic control theory and mathematical modelling of mechanical systems.

This is the third version of the lecture notes and contains minor updates from the previous
one issued in 2011.

The students are greatly acknowledged for comments providing continuously improvements.

Asgeir J. Sørensen
Trondheim, Norway
January 2013.

i
Acknowledgements

Several of the chapters are results from joint research between the author and colleagues from
the Norwegian University of Science and Technology (NTNU), former colleges in ABB Marine
and new partners in the industry. In particular, the author is grateful for years of joint work
with Professor Thor I. Fossen, Department of Engineering Cybernetics, NTNU.
In the period 2005-2009 during my leave from university managing the NTNU spin-off com-
pany Marine Cybernetics, Dr. Tristan Perez, Dr. Trong Dong Nguyen and Dr. Morten Breivik
were acting as stand-in for me running the course Marine Control Systems. All of you did a
great job improving the course and the teaching material.
The following coauthors are acknowledged for cooperation and contributions to the Lecture
Notes:

• Professor Thor I. Fossen, Dr. Jann Peter Strand and Dr. Trong Dong Nguyen for contri-
butions to the Chapter on Dynamic Positioning Systems.

• Dr. Øyvind Smogeli, Dr. Marit Ronæss, Dr. Vegar Johansen, Dr. Jon Refsnes, Dr.
Tristan Perez and Dr. Svein Ersdal for contributions to the Chapter on Mathematical
Modelling of Dynamically Positioned Marine Vessels.

• Dr. Øyvind Smogeli and Dr. Eivind Ruth for contributions to the Chapter on Propulsion
Control.

• Professor Bernt Leira and Professor Carl M. Larsen for contributions to the Section on
Control of Drilling Riser Angles by Dynamic Positioning of Surface Vessels.

• Dr. Anne Marthine Rustad for writing the Section on Modelling and Control Top Ten-
sioned. Professor Carl M. Larsen is also acknowledged for his contributions here.

• Dr. Alf Kåre Ådnanes for writing the Chapter on Maritime Electrical Installations and
Diesel Electric Propulsion.

Finally, the author is grateful for the contributions to all the PhD candidates and MSc
students I have had the pleasure to supervise and being enlightened tenfold in return.

ii
Contents

Preface i

Acknowledgements ii

1 Introduction 1
1.1 Marine Cybernetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Main Topics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

2 Marine Control Systems 5

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Power System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.4 Propulsion System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.5 Marine Automation System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5.1 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.5.2 Data Network and Process Stations . . . . . . . . . . . . . . . . . . . . . 12
2.5.3 Operator Stations and HMI . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5.4 Integration Aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.6 Dynamic Positioning System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.6.1 Position Reference Systems and Sensors . . . . . . . . . . . . . . . . . . . 14
2.6.2 Modes of Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.6.3 Functionality and Modules . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.6.4 Advisory and Surveillance Systems . . . . . . . . . . . . . . . . . . . . . . 20
2.6.5 DP Capability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.7 Power and Energy Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.7.1 Blackout Restoration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
2.7.2 Load Reduction and Blackout Prevention . . . . . . . . . . . . . . . . . . 23
2.7.3 Diesel Engine Governor and AVR Fault Tolerance . . . . . . . . . . . . . 25
2.8 Maritime Industrial IT . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
2.9 Rules and Regulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9.1 Class Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.9.2 Reliability and Redundancy . . . . . . . . . . . . . . . . . . . . . . . . . . 33
2.9.3 Failure Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.10 Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
2.10.1 Simulator Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
2.10.2 Module Hierarchy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37

iii
 2.10.3 Hardware and Software Platform . . . . . . . . . . . . . . . . . . . . . . . 38
2.10.4 Hardware-In-the-Loop Testing . . . . . . . . . . . . . . . . . . . . . . . . 38

3 Maritime Electrical Installations and Diesel Electric Propulsion 41

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.1.1 Scope and Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.1.2 Motivations for Electric Propulsion . . . . . . . . . . . . . . . . . . . . . . 42
3.1.3 Power Flow and Power Efficiency . . . . . . . . . . . . . . . . . . . . . . . 43
3.1.4 Historical Overview of Electric Propulsion . . . . . . . . . . . . . . . . . . 45
3.2 Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.1 Passenger Vessels — Cruise Ships and Ferries . . . . . . . . . . . . . . . . . 49
3.2.2 Oil and Gas Exploitation and Exploration: Drilling Units, Production
Vessels and Tankers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.2.3 Field Support Vessels and Construction Vessels . . . . . . . . . . . . . . . 50
3.2.4 Icebreakers and Ice Going Vessels . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.5 War Ships . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
3.2.6 Research Vessels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.2.7 Trends and New Applications . . . . . . . . . . . . . . . . . . . . . . . . . 54
3.3 Overview of Electric Power System . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.2 Electric Power Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
3.3.3 Electric Power Distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.3.4 Motor Drives for Propulsion and Thrusters . . . . . . . . . . . . . . . . . 62
3.3.5 Propulsion Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.3.6 Trends and New Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
3.4 Power and Propulsion Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
3.4.1 Introduction - Control Hierarchy . . . . . . . . . . . . . . . . . . . . . . . 73
3.4.2 User Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
3.4.3 High Level Control Functionality . . . . . . . . . . . . . . . . . . . . . . . 75
3.4.4 Low Level Control Functionality . . . . . . . . . . . . . . . . . . . . . . . 78
3.5 Electric Propulsion Drives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.5.2 Variable Speed Motor Drives . . . . . . . . . . . . . . . . . . . . . . . . . 81
3.6 System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
3.6.2 Life Cycle Cost Assessment of Conceptual Design . . . . . . . . . . . . . . 98
3.6.3 Standard Network Analysis and Electrical Power System Studies . . . . . 98
3.6.4 Extended Analysis and Studies . . . . . . . . . . . . . . . . . . . . . . . . 103
3.7 Harmonic Distortion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.7.1 Harmonics of VSI Converters . . . . . . . . . . . . . . . . . . . . . . . . . 105
3.7.2 Harmonics of CSI Converters . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.7.3 Harmonics of Cycloconverters . . . . . . . . . . . . . . . . . . . . . . . . . 107
3.7.4 Limitations by Classification Societies . . . . . . . . . . . . . . . . . . . . 108
3.7.5 Harmonics of Ideal 6- and 12-pulse Current Waveforms . . . . . . . . . . 108
3.7.6 Calculating Harmonic Distortion . . . . . . . . . . . . . . . . . . . . . . . 110
3.7.7 Managing Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

iv
 3.8 Example Configurations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117

4 Computer-Controlled Systems 122

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
4.2 Basics in Linear System Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
4.2.1 Causality and State Definition . . . . . . . . . . . . . . . . . . . . . . . . 124
4.2.2 Continuous-time State Space Model . . . . . . . . . . . . . . . . . . . . . 125
4.2.3 Basic Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
4.2.4 Laplace Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
4.2.5 Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129
4.3 Sampler and Zero-Order Hold . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.4 Discrete-time State Space Model . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
4.4.1 ZOH Equivalent of Continuous-time State Space Model . . . . . . . . . . 130
4.4.2 Discrete-time Approximation . . . . . . . . . . . . . . . . . . . . . . . . . 132
4.4.3 Solution of Discrete-time System Equation . . . . . . . . . . . . . . . . . 133
4.4.4 Controllability and Observability of Discrete-time Systems . . . . . . . . . 134
4.5 Discrete-time Approximation Methods . . . . . . . . . . . . . . . . . . . . . . . . 135
4.5.1 Euler’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.5.2 Backward Euler’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
4.5.3 Combined Backward and Forward Euler’s Method . . . . . . . . . . . . . 136
4.5.4 Trapezoidal (Tustin’s) Method . . . . . . . . . . . . . . . . . . . . . . . . 137
4.5.5 Second Order System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.6 Nyquist Frequency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
4.7 The -Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.7.1 Definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141
4.7.2 Stability Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.8 The Pulse-Transfer Function and the Pulse Response . . . . . . . . . . . . . . . . 142
4.8.1 Pulse-transfer Function . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142
4.8.2 Pulse Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.9 Shift-operator Calculus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.9.1 Shift Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
4.9.2 Pulse-transfer Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
4.10 Stability Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 146
4.11 Order of the System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
4.12 Relation Between Shift-Operator Calculus and -Transform . . . . . . . . . . . . 150

5 Signal Quality Checking and Fault Detection 151

5.1 Testing of Individual Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.1.1 Windowing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
5.1.2 Signal Range Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.1.3 Variance Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153
5.1.4 Wild Point Testing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.2 Handling of Redundant Measurements . . . . . . . . . . . . . . . . . . . . . . . . 154
5.2.1 Voting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154
5.2.2 Weighting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.2.3 Enabling and Disabling of Sensors . . . . . . . . . . . . . . . . . . . . . . 156

v
6 Filtering and State Estimation 158
6.1 Analog and Digital Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 159
6.1.1 Nonideal Lowpass Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160
6.1.2 Nonideal Highpass Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.1.3 Notch Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.1.4 Digital Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161
6.2 State Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.2.1 Deterministic Estimators . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
6.2.2 Least Squares Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
6.2.3 Discrete Kalman Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
6.2.4 Extended Kalman Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . 173

7 Mathematical Modeling of Dynamically Positioned Marine Vessels 175

7.1 Environmental Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
7.1.1 Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
7.1.2 Wind . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
7.1.3 Water Current Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
7.2 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
7.2.1 Reference Frames . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 194
7.2.2 The Euler Angle Transformation . . . . . . . . . . . . . . . . . . . . . . . 195
7.3 Vessel Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
7.3.1 Nonlinear Low-Frequency Vessel Model . . . . . . . . . . . . . . . . . . . 201
7.3.2 Environmental Loads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 207
7.3.3 Linear Wave-Frequency Model . . . . . . . . . . . . . . . . . . . . . . . . 209
7.3.4 Thrust Servo Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
7.4 Static Analysis of Cable Segments . . . . . . . . . . . . . . . . . . . . . . . . . . 211
7.4.1 Basic Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 211
7.4.2 Catenary Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
7.4.3 Catenaries as Boundary Value Problems (BVP) . . . . . . . . . . . . . . . 222
7.4.4 Hydrodynamic Drag Loads . . . . . . . . . . . . . . . . . . . . . . . . . . 224
7.5 Mooring System for FPSOs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

8 Dynamic Positioning Control System 228

8.1 Survey of Scientific Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
8.2 Observer Design for Dynamic Positioning . . . . . . . . . . . . . . . . . . . . . . 231
8.2.1 Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
8.2.2 Control Plant Model: Vessel Model . . . . . . . . . . . . . . . . . . . . . . 233
8.2.3 Extended Kalman Filter Design . . . . . . . . . . . . . . . . . . . . . . . . 235
8.2.4 Nonlinear Observer Design . . . . . . . . . . . . . . . . . . . . . . . . . . 236
8.2.5 Adaptive Observer Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
8.2.6 Nonlinear Observer Design for Extreme Seas . . . . . . . . . . . . . . . . 244
8.2.7 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
8.3 Controller Design for Dynamic Positioning . . . . . . . . . . . . . . . . . . . . . . 252
8.3.1 Control Plant Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
8.3.2 Horizontal-plane Controller . . . . . . . . . . . . . . . . . . . . . . . . . . 254
8.3.3 Horizontal-plane Controller with Roll-Pitch Damping . . . . . . . . . . . 257
8.3.4 Controller Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258

vi
 8.3.5 Thrust Allocation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 260
8.3.6 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
8.4 Hybrid Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 268
8.4.1 Control Plant Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
8.4.2 Observer Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 270
8.4.3 Vessel Operational Condition . . . . . . . . . . . . . . . . . . . . . . . . . 270
8.4.4 Concept of Hybrid Control . . . . . . . . . . . . . . . . . . . . . . . . . . 271
8.4.5 Example: Switching from DP to PM . . . . . . . . . . . . . . . . . . . . . 273
8.5 Weather Optimal Positioning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
8.5.1 Mathematical Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
8.5.2 Weather Optimal Control Objectives . . . . . . . . . . . . . . . . . . . . . 283
8.5.3 Nonlinear and Adaptive Control Design . . . . . . . . . . . . . . . . . . . 284
8.5.4 Experimental Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290

9 Propulsion Control 297

9.1 Propellers and Thrusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
9.1.1 Shaft Propulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
9.1.2 Thrusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
9.1.3 Podded Propulsion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
9.1.4 Mechanical Pod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 299
9.1.5 Water Jets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
9.2 Control Problem Formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
9.3 Propeller Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 302
9.3.1 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304
9.4 Propulsion Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307
9.5 Propeller and Thruster Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
9.5.1 In-line Velocity Fluctuations . . . . . . . . . . . . . . . . . . . . . . . . . 309
9.5.2 Transverse Velocity Fluctuations . . . . . . . . . . . . . . . . . . . . . . . 309
9.5.3 Ventilation and In-and-out-of Water Effects . . . . . . . . . . . . . . . . . 311
9.6 Propeller Shaft Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 314
9.6.1 Torque Loop in an Electrical Motor Drive . . . . . . . . . . . . . . . . . . 317
9.6.2 Resulting Thruster Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . 317
9.6.3 Thrust, Torque, Power, and Shaft Speed Relations . . . . . . . . . . . . . 319
9.7 Low-Level Thruster Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
9.7.1 Control Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
9.8 Thruster Control in Normal Conditions . . . . . . . . . . . . . . . . . . . . . . . 320
9.8.1 Controller Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
9.8.2 Control Plant Model Parameters . . . . . . . . . . . . . . . . . . . . . . . 321
9.8.3 Reference Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
9.8.4 Inertia Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
9.8.5 Friction Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
9.8.6 Torque and Power Limiting . . . . . . . . . . . . . . . . . . . . . . . . . . 323
9.8.7 Shaft Speed Feedback Control . . . . . . . . . . . . . . . . . . . . . . . . . 324
9.8.8 Torque Feedforward Control . . . . . . . . . . . . . . . . . . . . . . . . . . 324
9.8.9 Power Feedback Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
9.8.10 Combined Torque and Power Control . . . . . . . . . . . . . . . . . . . . 325

vii
 9.9 Thruster Control in Extreme Conditions . . . . . . . . . . . . . . . . . . . . . . . 326
9.9.1 Loss Estimation and Ventilation Detection . . . . . . . . . . . . . . . . . . 326
9.9.2 Anti-spin Thruster Control . . . . . . . . . . . . . . . . . . . . . . . . . . 328
9.9.3 Implementation aspects . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
9.10 Sensitivity to Thrust Losses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
9.10.1 Shaft Speed Feedback Control . . . . . . . . . . . . . . . . . . . . . . . . . 332
9.10.2 Torque Feedforward Control . . . . . . . . . . . . . . . . . . . . . . . . . . 332
9.10.3 Power Feedback Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . 332
9.10.4 Combined Power and Torque Control . . . . . . . . . . . . . . . . . . . . 333
9.10.5 Positioning Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
9.10.6 Sensitivity Functions Summary . . . . . . . . . . . . . . . . . . . . . . . . 333
9.11 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
9.11.1 Experimental Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 336
9.11.2 Nominal Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 339
9.11.3 Sensitivity Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
9.11.4 Dynamic Tests in Waves . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
9.11.5 Anti-spin Tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 345
9.11.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349

10 Modeling and Control of Ocean Structures 350

10.1 Mathematical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 351
10.1.1 Ordinary Differential Equations (ODE) . . . . . . . . . . . . . . . . . . . 351
10.1.2 Partial Differential Equations (PDE) . . . . . . . . . . . . . . . . . . . . . 357
10.1.3 Methods for Numerical Solution of the Relevant Systems . . . . . . . . . 358
10.2 Modelling and Control Top Tensioned Risers . . . . . . . . . . . . . . . . . . . . 371
10.2.1 Motivation and Problem Description . . . . . . . . . . . . . . . . . . . . . 371
10.2.2 Tension Leg Platform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
10.2.3 Mathematical Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 376
10.2.4 Model Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
10.2.5 Control plant model analysis . . . . . . . . . . . . . . . . . . . . . . . . . 397
10.2.6 Controller System Design . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
10.2.7 Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
10.2.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
10.3 Control of Drilling Riser Angles by Dynamic Positioning of Surface Vessels . . . 422
10.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
10.3.2 Mathematical Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 422
10.3.3 High Level Positioning and Riser Angle Controller . . . . . . . . . . . . . 432
10.3.4 Local Optimization: Optimal Setpoint Chasing . . . . . . . . . . . . . . . 433
10.3.5 Numerical Simulations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 435

11 Modelling and Control of High Speed Craft 443

11.1 Ride Control of Surface Effect Ships . . . . . . . . . . . . . . . . . . . . . . . . . 444
11.1.1 Mathematical Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
11.1.2 Robust Dissipative Controller Design . . . . . . . . . . . . . . . . . . . . . 456
11.1.3 Simulation and Full Scale Results . . . . . . . . . . . . . . . . . . . . . . . 466

A Linear Algebra 496

viii
B Digital PID-Controllers 499
B.1 Continuous-Time PID-Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . 499
B.2 Discrete-Time PID-Controllers . . . . . . . . . . . . . . . . . . . . . . . . . . . . 500

C Definition of Electro-technical Terms 502

C.1 DC — Direct Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 502
C.2 AC — Alternating Current . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 504
C.2.1 One phase AC sources and resistive loads . . . . . . . . . . . . . . . . . . 504
C.2.2 Three phase AC sources and resistive loads . . . . . . . . . . . . . . . . . 506
C.2.3 Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509
C.3 Modelling of Components in Electric Power Generation and Distribution . . . . . 513
C.3.1 Generator model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513
C.3.2 Distribution switchboards . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
C.3.3 Cables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
C.3.4 Synchronous motors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
C.3.5 Asynchronous motors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516

D Standards, rules, and regulations on electric propulsion 518

D.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 518
D.2 International regulations, applicable standards or code of practice . . . . . . . . . 519
D.3 Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 521

ix
Chapter 1

Introduction

1.1 Marine Cybernetics

Marine control systems or Marine cybernetics is defined to be the science about techniques and
methods for analysis, monitoring and control of marine systems.
The main application fields for marine control systems are the three big marine industries:
Sea transportation (shipping), offshore oil and gas exploration and exploitation and fisheries and
aquaculture. So far most of the examples are collected from the mature industries like shipping
and offshore oil and gas exploration and exploitation. It is believed that these industries will
be even more technology demanding with focus on safe and cost effective solutions in the years
to come. More of the offshore activities is assumed to take place in deeper water based on
floating solutions in combination with subsea installations. The need for conducting all-year
marine operations in the ocean space with varying complexity will increase and motivate to the
development of underwater robotics with increased autonomity. Due to increased industrializa-
tion within the fisheries and aquaculture, the industrial content and thereby the introduction of
advanced technology are expected to increase. It is foreseen that there is a huge potential for
technology transfer among all the marine industries.
Concerning marine control systems it is suggested to divide the control structure into two
main areas: real-time control and monitoring and operational and business enterprise manage-
ment, see Figure 1.1. We will here in particular focus on the various aspects related to the design
of real-time control and systems. The integration of real-time systems with operational man-
agement and business transactional systems is by the automation industry denoted as Industrial
IT.
The real-time control structure is as shown in Figure 1.1 divided into low-level actuator
control, high-level plant control and local optimization. We will in the text use demonstrating
examples from the offshore oil and gas industry. In particular, examples with dynamically
positioned (DP) offshore vessels will be used. A DP vessel maintains its position (fixed location
or pre-determined track) exclusively by means of active thrusters. Position keeping means
maintaining a desired position in the horizontal-plane within the normal excursions from the
desired position and heading. The real-time control structure for a DP system may then consist
of (Sørensen [299] and [300]):

• Actuator control : The actuators for DP systems are normally thrusters, propellers, and
rudders. Local control of propellers and thrusters may be done by controlling e.g. the speed

1
 Office Systems
Business enterprise/

Office Network
Fleet management

Ship 3:
Ship 2:
Ship 1:
Operational management

Real-Time Control

Local optimization (min-hour)

Fault-Tolerant Control
Control layers
Real-Time Network

Plant control High level

(0.1-5 s)

Actuator control Low level

(0.001-1 s)

Figure 1.1: Control structure.

(rpm), pitch, torque, and power or combinations of these. Dependent on the actuators are
mechanically, hydraulically and/or electrically driven controllers with different properties
will be used.

• Plant control : In station keeping operations the DP system is supposed to counteract the
disturbances like wave (mean and slowly varying), wind and currents loads acting on the
vessel. The plant controller calculates the commanded surge and sway forces and yaw
moment needed to compensate the disturbances. A multivariable output controller often
of PID type using linear observers e.g. Kalman filter or nonlinear passive observers may
be used. Setpoints to the thrusters are provided by the thruster allocation scheme.

• Local optimization: Depending of the actual marine operation the DP vessel is involved in
optimization of desired setpoint in conjunction with appropriate reference models for e.g.
drilling operations, weather vaning, pipe laying, tracking operations, etc. are used. Notice
that in the guidance, navigation and control literature local optimization corresponds to
the guidance block.

As the controllers rely on proper measurements to work on, signal processing is of vital
importance for the stability and robustness of the control system. This will be the first topic of
the text.

1.2 Main Topics

The topics covered here are:

2
• Chapter 2 chapter gives an overview of marine control systems exemplified by DP vessels
with electrical propulsion systems. Design aspects related to high-level vessel control such
as power management and DP are shown. Rules and regulations including testing and
verification of marine control systems based on Hardware-In-the-Loop (HIL) testing will
also be introduced.
• Chapter 3 gives an introduction to electrical installations on marine vessels. Electric instal-
lations are present in any ship from powering of communication and navigation equipment,
alarm and monitoring system, running of motors for pumps, fans or winches, to high power
installation for electric propulsion. Successful design of control systems for such vessels
rely on understanding of the electrical equipment and components used and the electrical
systems for power generation, distribution and power consumption. Furthermore, success-
ful solutions for vessels with electric propulsion are found in environments where naval
architects, hydrodynamic and propulsion engineers, and electrical engineering expertise
cooperates under constructional, operational, and economical considerations.
• Chapter 4 introduces the reader to the basics in linear system theory and computer-
controlled systems. Today most of the control systems are implemented in computers run
by digital processors. Computers will operate on continuous-time processes at discrete-time
sampling instants. Therefore, computer-controlled systems can be regarded as discrete
control systems. Discrete systems could in many cases be viewed as approximations of
analog systems. Computer-controlled systems may introduce deteriorations of the control
objective, if not accounting for the fact that they operate on discrete instants of processes
which are time-continuous by nature.
• Chapter 5 is about signal quality control and fault detection. All industrial control systems
contains several functions for signal quality checking. The performance of the control
system depends on these functions operate properly. In this chapter the most important
signal processing properties for the purpose of control will be treated.
• In Chapter 6 the most common analog and digital filters in addition to observer theory
are presented. The Kalman filter is also derived. Kalman filters have successfully been
used in marine control systems since the late 1970s.
• In Chapter 7 the mathematical models used in the design of dynamic positioning systems
and thruster assisted position mooring systems are formulated. To readers unfamiliar with
hydrodynamics, the models may look difficult. However, the intention here is to focus on
the structural properties of the models important for control, rather than going into details
about calculation of the hydrodynamic coefficients. If the reader is inspired to enter this
field any further, the author is convinced that this will improve the control system designs
even more.
• Chapter 8 contains a comprehensive introduction to dynamic positioning (DP) and thruster
assisted position mooring (PM) including design of observers and controllers. In addition
an introduction to supervisory-switched control is given for multi-objective control and
expanding the operational window handling extreme seas
• Chapter 9 on propulsion and thruster controls show the importances of considering low-
level control designs of the actuators ensuring that the high-level control demands are
fulfilled according to the control objective.

3
• Chapter 10 presents a new field of research concerning marine control systems applied
on ocean structures. It is expected that control of so-called flexible systems will increase
in the future due to increased water depth for offshore operations, new materials and
applications.

• Chapter 11 is about motion control of high speed craft. In particular the chapter presents
how flexibility in the air cushion pressure dynamics will have impact on the design of a
ride control system for motion damping of a high speed craft called Surface Effect Ships
(SES).

4
Chapter 2

Marine Control Systems

2.1 Introduction
The history of automated closed-loop ship control started with Elmer Sperry (1860-1930), who
constructed the first automatic ship steering mechanism in 1911 for course keeping (Allensworth
[5] and Bennet [22]). This device is referred to as the ”Metal Mike”, and it was capturing much
of the behavior of a skilled pilot or a helmsman. ”Metal Mike” did compensate for varying
sea states by using feedback control and automatic gain adjustments. Later in 1922, Nicholas
Minorsky (1885-1970), presented a detailed analysis of a position feedback control system where
he formulated a three-term control law which today is refereed to as Proportional-Integral-
Derivative (PID) control (Minorsky [190]). These three different behaviors were motivated by
observing the way in which a helmsman steered a ship.
This chapter gives an overview of marine control systems exemplified by dynamically po-
sitioned (DP) vessels with electrical propulsion systems. Design aspects related to high-level
vessel control such as power management and DP are shown. Design requirements and ability
for reconfiguration accounting for physical segregation and redundancy of power, propulsion and
automation systems will be presented. Rules and regulations including testing and verification
of marine control systems based on Hardware-In-the-Loop (HIL) testing will also be introduced.
Other marine control systems; propulsion control systems, control of slender ocean structures,
motion control systems for high speed crafts and details on DP systems will be presented in
separate chapters later in the text. The design steps combining software (SW) and hardware
(HW) architecture, modeling, sensor processing, controller design, simulation and testing will
more or less follow the same principles. However, requirements to performance and redundancy,
and thereby complexity, may differ for the various marine control system applications.
Offshore exploration and exploitation of hydrocarbons have opened up an era of DP vessels.
Currently, there are more than 2000 DP vessels of various kind operating worldwide. DP systems
are used for a wide range of vessel types and marine operations:

• Offshore oil and gas industry: Typical applications in the offshore market are offshore
service vessels, drilling rigs and drilling ships, shuttle tankers, cable and pipe layers, floating
production off-loading and storage units (FPSOs), crane and heavy lift vessels, geological
survey vessels and multi-purpose vessels. Cable and pipe laying are typical operations
which also need tracking functionality.
• Shipping: Currently there is a trend towards more automatic control of marine/merchant

5
 vessels, beyond the conventional autopilot. This involves guidance systems coupled to
automatic tracking control systems, either at high or low speed. In addition, more sophis-
ticated weather routing and weather planning systems are expected. Automatic docking
systems, and a need for precise positioning using DP systems when operating in confined
waters will become more used.

• Cruise ships and yachts: The cruise and yacht market also make use of more automatic
positioning control. In areas where anchors are not allowed due to vulnerable coral reefs,
DP systems are used for station keeping. Precise positioning is also required for operating
in harbors and confined waters.

• Fisheries: Application of more sophisticated guidance, navigation and control systems

for ships during fishing are motivated by the need for precise positioning, reduced fuel
consumption and intelligent selective fishing.

Electric propulsion is not a very new concept. It has been used as early as in the late 19th
century. However, only in few vessels until the 1920s where the electric shaft line concept enabled
the design of the largest Trans-Atlantic passenger liners. Variable speed propulsion was used in
some few applications during the 1950s and 1960s, while, first when the semiconductor technol-
ogy became available in large scale commercial applications, this technology became acceptable
for a wide range of applications. The introduction of AC drives and podded propulsors was
another shift in technology that led to a rapid increase in the use of electric propulsion through
the last 15-20 years. Typically, ships with electric propulsion tend to have more system function-
ality implemented in integrated automation systems, partially because such functionalities are
regarded to be necessary for safe and optimal operation, but also because the electric propulsion
plant enables the use of such functions. In the commercial market the offshore vessels in addition
to cruise ships and ice breakers have been technology drivers concerning automation, power and
propulsion systems. They are characterized by the required ability to conduct complex marine
operations, operational availability, safety focus, cost effectiveness and flexibility in operational
profile concerning transit, station keeping, maneuverability and to some extent also a significant
vessel or process load system. These rather complex power plants opened up for an increasing
use of fully all-electric ships and the introduction of fully integrated computer-controlled systems
in order to operate safely and cost efficiently. Such concepts are today applied in an increasing
number of ship applications.
Consequently, the complexity has also increased with a variety of solutions consisting of
stand-alone systems, partly integrated systems to fully physical and functional integrated sys-
tems. Up to now integrated automation systems have been proprietary with a limited number of
vendors. However, in the automation industry it is a trend towards openness in communication
protocols and network. How this will influence on the technology solutions and responsibili-
ties for multi-vendor integration systems is still a subject for discussion. During the late 1990s
the introduction of low-cost off-shelf computers not originally designed for automation purposes
have also been taken more into use. This development is driven by the need to make more cost
efficient solutions. Nevertheless it also creates new issues of concern:

• New procedures for design and specification that considers compatibility and integration
aspects.

• Failure analysis and test methods, adequate to ensure fault-tolerance in the overall system.

6
2.2 System Overview
A DP offshore vessel, in this case a drilling rig as illustrated in Figure 2.1, may comprise the
following sub-systems:

• Power system.

• Marine control or automation system.

• Positioning systems comprise dynamic positioning (DP) system for automatic station keep-
ing and low speed manoeuvring of free floating vessels or thruster assisted position mooring
(PM) system for anchored vessels. Sensors and position reference systems. Hardware, soft-
ware and sensors to supply information and/or corrections necessary to give e.g. accurate
vessel motion, position and heading measurements.

• Automatic sailing system and nautical system including autopilot, radar, automatic iden-
tification system (AIS), and GPS for automatic course keeping and course changing for
transit operations.

• Joystick system for manual control of the thruster system for positioning of the vessel in
surge, sway and yaw/heading. Often automatic heading control function is included. One
should notice that the operator does not control each thruster individually. The operator
command a desired thrust in surge, sway and yaw. Then the thruster allocation algorithm
map the desired thrust in surge, sway and yaw to individual thrust set-points of the enabled
thruster.

• Manual thruster control system includes individual control of each thrusters.

• Propulsion and thruster system involves all components and systems necessary to supply
a vessel with thrust for dynamic positioning, maneuvering and transit. For details the
reader is referred to Chapter 9.

• Equipment packages and systems for e.g. drilling, off-loading, crane operations, oil and
gas production.

• Mooring system (applicable for moored vessels only). Generally, a mooring system consists
of  lines connected to the structure and horizontally spread out in a certain pattern. At
the seabed the lines are connected to anchors. Spread mooring systems are used both
for semi-submersibles and turret-anchored ships. In the latter case the anchor lines are
connected to a turret on the ship, which can be rotated relative to the ship. The number of
anchor lines may vary, typically from 6 to 12. The length of the anchor lines are adjusted
by winches and determines the pre-tension and thus the stiffness of the mooring system.

• Safety systems.

• Auxiliary systems, such at heating, ventilation and air condition (HVAC), water cooling,
hydraulic systems, etc.

7
 BACK-UP PLANT
SYSTEM NETWORK
SAFETY SYSTEM WIND SENSORS
EMERGENCY SHUTDOWN PROCESS CONTROL
FIRE & GAS STATION
VRU

GYRO
INFORMATION MANAGEMENT
REMOTE DIAGNOSTIC INTEGRATED THRUSTER
CONTROL SYSTEM
- DYNAMIC POSITIONING
- POSMOOR
- AUTOSAIL
CONTROL - OPERATOR CONTROL
NETWORK SYSTEM
INTEGRATED MONITORING &
CONTROL SYSTEM
- EXTENSION ALARM
- PROCESS CONTROL

PROPULSION

AZIPOD
DRILLING DRIVE FIELDBUS
SYSTEM NETWORK
ENERGY MANAGEMENT
SYSTEM
POWER GENERATION
& DISTRIBUTION

Figure 2.1: Illustration of different systems and components in a large, integrated plant (Cour-
tesy ABB Marine).

2.3 Power System

Power system comprises all units necessary to supply the vessel with power. The power genera-
tion and distribution systems are divided into the following main parts, for details see Chapter
3. The power generation and distribution systems are divided into the following main parts
(Hansen [105]):

• A power generation plant with prime mover and generators. In commercial vessels, the
prime movers are typically medium speed diesel engines, due to high efficiency and cost
efficiency. Gas turbines, and also steam turbines, are used in some ship applications due
to their advantages in weight and size. There are intensive research and development in
the field of LNG and feasible fuel cells, which in some future also can be applied in a wider
scale than of today. In addition, combined power plants are also looked more into.

• An electric power distribution system with main and distribution switchboards. The dis-
tribution system is typically split in two, three or four sections, with the possibility to
connect and split the sections by use of bus ties or bus feeders. With a proper protection
system and operational philosophy, such systems may have the flexibility of operating in
a common, connected mode for optimizing the energy production in normal modes, while

8
 isolating faulty parts of the system and obtain the intended redundancy without loss of
maneuverability or station keeping ability by automatic splitting and segregation of the
system.

• Transformers for feeding of alternate voltage levels in main or low voltage switchboards
and motor control centers.

• Uninterruptible power supply of sensitive equipment and automation systems.

• Rotating converters for frequency conversion and clean power supply.

• Cabling, including cable routing and segregation.

In many ships there is a vessel or process plant system, which is critically dependent on the
electrical power system. Such systems can be:

• Hotel loads, e.g. for cruise and passenger vessels.

• Crane and winch systems, e.g. for support vessels.

• Drilling systems, e.g. for drilling rigs.

• Oil and gas production, for FPSOs.

• Load/cargo handling systems, for cargo vessels and tankers.

Even though these systems may not be directly critical for the safe maneuverability of the
vessel, there may be serious consequences by loss of controllability, both with concern to safety,
environmental or economical consequences. Hence, redundancy and segregation in the process
and automation systems may be equally important as in the electrical power and propulsion
system. This can be explained by the redundant configuration of a drilling drive system, with
dual independent power feeding and possibility to segregate into independent sections either in
normal or faulty conditions, as shown in Figure 2.2

2.4 Propulsion System

The propulsion system consists typically of prime movers as diesel engines, generators, trans-
missions and thrusters. A thruster is here defined as the general expression for a propeller unit.
A ship can be equipped with several types of thrusters. Conventional ships typically have a
main propulsion unit located aft. Traditionally, shaft line propulsion with controllable pitch or
fixed pitch propellers is applied, with rudders to direct the thrust. Another common type is
the tunnel thruster which is a propeller inside a tunnel that goes through the hull and produces
a fixed-direction transverse force. A third type is the azimuth (rotatable) thruster, which can
produce thrust in any direction. While the main and the tunnel thrusters have fixed force direc-
tions, the direction of the azimuth thruster can be changed, either manually by the operator or
automatically by the positioning control system. Rudders in combination with main propellers
can also be used actively in the positioning system to produce a transverse force acting on the
stern. In positioning systems the main objective is to control the thruster’s force. However, since
this resulting force cannot be measured directly, it is common to control the thruster revolution

9
 G G G G G G G G

AC AC AC AC AC AC
AC AC AC AC AC AC

M M M M M M

Drilling package
AC Multidrive
AC AC
DC DC

DC DC DC DC DC DC
AC AC AC AC AC AC

M M M M M M

Figure 2.2: Single line diagram of a drilling rig.

speed (RPM control), the pitching of the propeller blades (pitch control), or by these in the
combination (consolidated control). The servo mechanisms for the propulsion devices must be
designed to give accurate and fast response. This is often referred to as low-level thrust control,
which will be treated in more detail in the Section 9.

2.5 Marine Automation System

2.5.1 Overview
In marine industry, there are two main directions of solutions for automation systems:

• Integrated automation systems.

• Stand-alone automation systems.

The two different classes of solutions are found in most kind of vessels. However, since com-
plex functionalities and tasks are handled better by the integrated solutions, this is more often
found in the vessels with complex and advanced propulsion and station keeping installations. In
the following integrated automation system is treated.
Marine automation or vessel control system comprises:

• Control functions for e.g. HVAC control, cargo and ballast control, emergency shutdown
and fire and gas detection, off-loading control, engine control, etc.

10
 Office Plant Network

Control Network

Fieldbus Network

Figure 2.3: Automation plant.

• Power management system (PMS) for handling of generators, black-out prevention, power
limitation, load sharing and load shedding. For advanced vessels more sophisticated energy
management systems may be used for intelligent power planning and allocation.

• The operator’s user interface to the automation system is through the Human-Machine
Interface (HMI) with display systems and operator panels, often denoted as operator
stations.

• Centralized computers with scalable CPU processing and I/O capacities, often denoted as
controllers or process control stations (Figure 2.3).

• Distributed computers or PLCs, typically with local control and interfaces to process and
to centralized computers.

• Communication buses at the different levels of control.

• Associated cabling, segregation and cable routing.

With today’s standardization on communication in automation systems, it is relatively easy

to connect various sub systems into a network communication system. The evolution of inter-
connecting systems in such networks has, today, developed to be an international standard of
many new buildings. It is a challenge of designers, to distinguish between the essential and
important communication, and to define the right level of redundancy and inter-dependency of
sub-systems.

11
2.5.2 Data Network and Process Stations
On larger systems the power, automation and the various parts of the positioning system are
connected by data networks, and the positioning control systems are implemented in the local
process stations. It is important that these systems meet the necessary redundancy level and
communication real-time requirements, such that the positioning system is reliable. Today it
has been common to integrate all automation system into a common plant network (Figure 2.3).

2.5.3 Operator Stations and HMI

The operator’s link to the positioning system is through the Human-Machine Interface (HMI)
on the operator stations. On the operating stations, it is common to arrange a joystick system,
giving the operator the opportunity to direct control the thruster forces in surge and sway and
moment in yaw. The HMI should be user-friendly and sufficient information should be presented
for the operator so that correct decisions can be made.

2.5.4 Integration Aspects

Physical integration based on standardized communication protocols ensure connectivity of de-
vices and integration of controllers and operator stations into three network levels (Figure 2.3):
Real time field bus network communication on low level between devices and controllers, real
time control network connecting controllers and operator stations, and office plant network to
various office systems and information management systems. The last level opens up for satel-
lite communication to land offices at ship operators or vendors. It is common to characterize
the automation system by the number of I/Os and dividing the applications into low-end or
high-end segments. For a drilling rig as shown in Figure 2.1 the number of I/O signals may be
up to 10-15000. For a conventional supply vessel the I/O number is typically lower than 1.500
- 2.000 signals.

2.6 Dynamic Positioning System

In the 1960s systems for automatic control of horizontal position in addition to the course were
developed. Such systems for simultaneously control of three horizontal motions (surge, sway and
yaw motion) are today known as dynamic positioning (DP) systems. Description of DP systems
including the early history can be found in Fay [74]. In the 1960s the first DP system was using
single-input single-output PID control algorithms in combination with low-pass and/or notch
filter. In the 1970s more advanced output control methods based on multivariable optimal
control and Kalman filter theory was proposed. Later on nonlinear control was also applied, see
Chapter 8 for details.

Definition 2.1 A DP vessel is by the class societies e.g. Det Norske Veritas (DNV) [54],
American Bureau of Shipping (ABS) [3] and Lloyd’s Register (LRS or Lloyd’s) [174], defined as a
vessel that maintains its position and heading (fixed location or pre-determined track) exclusively
by means of active thrusters. This is obtained either by installing tunnel thrusters in addition
to the main screw(s), or by using azimuthing thrusters, which can produce thrust in different
directions.

12
 While in DP operated ships the thrusters are the sole source of station keeping, the assistance
of thrusters are only complementary to the mooring system in the case of thruster assisted
position mooring (PM) systems. Here, most of the position keeping is provided by a deployed
anchor system. In severe environmental conditions the thrust assistance is used to minimize
the vessel excursions and line tension by mainly increasing the damping in terms of velocity
feedback control. For turret anchored ships, see Figures 2.4 and 2.5, without natural weather-
vaning properties the thrusters are also used to automatic control of the heading, similarly to
DP operated vessels. In this text, a marine positioning system is either defined as a dynamic
positioning (DP) system or a thruster assisted position mooring (PM) system.

Figure 2.4: Marine operations in offshore oil and gas exploration.

13
 Turret

Anchor lines

Seabed

Figure 2.5: Turret anchored FPSO.

DP systems have traditionally been a low-speed application, where the basic DP functionality
is either to keep a fixed position and heading or to move slowly from one location to another.
In addition specialized tracking functions for cable and pipe-layers, and remote operated vehicle
(ROV) operations have been available. The traditional autopilot functionality has over the
years become more sophisticated. Often a course correction function is available for correction
of course set-point due to environmental disturbances and drifting, such that the vessel follows
a straight line. Way-point tracking is used when a vessel is supposed to follow a pre-defined
track e.g. defined by several way-point coordinates. The trend today is that typical high-speed
operation functionality merges with the DP functionality, giving one unified system for all speed
ranges and types of operations. Further research on hybrid control will enable this.

2.6.1 Position Reference Systems and Sensors

It is common to divide the different measurements used by a positioning system into position
reference systems and sensor systems. The most commonly used position reference systems are:

• Global Navigation Satellite Systems (GNSS): The most commonly used navigation
system for marine vessels is Navstar GPS, which is a US satellite navigation system with
world coverage. An alternative system is the Russian system Glonass, which only covers
certain regions of the globe, see Parkinson and Spiker [220]. An European system, Galileo,
is currently under construction. For local area operations it is now possible to achieve
meter accuracy by using Differential GPS (DGPS), and sub-decimeter accuracy by using
Carrier Differential GPS (CDGPS). The development of wide area augmentation systems
(WAAS) is expected to give meter accuracy across entire continents. When using a satellite
navigation system, at least four satellites must be visible in order to compute a reliable
position estimate (three for sea level navigation). If the ship is entering a shadowed region,
and redundant signals are not available, there will be a loss of position measurement. Other
causes for degraded position measurements (or in the worst case loss or measurement) are
ionospheric disturbances, reflections in the water surface, etc. For DGPS the accuracy
is typically within 1 meter radius with 95 % probability. Recently, the forced signal

14
 distortion (SA) on the GPS signal has been removed, such that the accuracy of the GPS
itself without differential correction is dramatically improved. Depending of antennas
and signal receivers, typical accuracy is within ±10  This opens up for even more cost
effective navigation equipment within the field of marine positioning.

• Hydroacoustic Position Reference (HPR) Systems: By using one or several transpon-

ders located on fixed position on the seabed and one or several transducer mounted under
the hull, the position of the vessel is measured. The accuracy of such systems depends on
the water depth and the horizontal distance between the transponder and the transducer.
There are different principles for performing measurements, where the most common used
are super short baseline (SBL), super/ultra short baseline (S/USBL) or long baseline
(LBL).

• Taut Wire: A taut wire system is used to measure the relative position of a floating vessel
at rest. This system consists of a heavy load located at the sea floor. The load is connected
to the vessel by a wire cord or heavy metal chain which is hold in constant tension by using
a winch onboard the vessel. Furthermore the angles at the top and bottom of the wire
are measured, and the length of the wire. Hence, the relative position (  ) can be
computed by solving three geometric equations with three unknowns.

Other position reference systems can be microwave systems (ARTEMIS, MICRORAGER,

MICRO FIX), Radiowave systems (SYLEDIS), optical systems (laser beams), and riser angle
position measurement systems. The latter system is based on instrumentation on a marine riser
attached to the vessel when performing drilling operations.
The sensor systems may comprise:

• Gyrocompass and/or magnetic compass, which measures the heading of the vessel.

• Vertical reference unit (VRU), which at minimum measures the vessel heave, roll and pitch
motions. Angular velocities are also often available. One of the main functions for the
VRU is to adjust the position measurements provided by GPS, hydroacoustic position
reference (HPR) systems, etc. for roll and pitch motions. For deepwater DP operations
the accuracy of the roll and pitch signals must be high providing accurate HPR position
measurements.

• An IMU typically contains gyros and accelerometers in 3-axes that can be used to measure
the body-fixed accelerations in surge, sway and heave, the angular rates in roll, pitch
and yaw and the corresponding Euler angles (Britting [35], Titterton and Weston [313]).
The IMU can be integrated in a filter (observer) with DGPS and HPR measurements for
instance in order to produce accurate velocity estimates. In most cases only the IMU Euler
angles are used in conjunction with DGPS. This is a minimum configuration since the Euler
angles are needed in order to transform the measured GPS position corresponding to the
GPS antenna down to the vessel fixed coordinate system usually located in the centerline
of the ship.

• Wind sensors, which measures the wind velocity and direction relative to the vessel.

• Draft sensors (used for vessels which is operated over a wide range of drafts).

15
 • Environmental sensors: wave sensors (significant wave height , direction, peak frequency),
current sensors (velocity and direction at sea surface and at different depths). Environ-
mental sensors are not a class requirement. These sensors are quite common on the most
sophisticated offshore installation.

• Other sensors dependent on type of operation, e.g. for pipe layers pipe tension is also
measured an utilized by the DP system.

In many installations redundant measurements are available, and the number and types of
measurements required are specified by certain class rules, see for example DNV [54]. Redundant
measurements increases the safety and availability of the positioning system.
The heading of the vessel is usually measured by one or several gyrocompasses, which dy-
namically are accurate. During rapid turning operations, however, the gyros will produce a
steady-state offset which gradually decreases to zero when the course is constant again. By
using two GPS antennas, the heading can be measured even more accurately, without the drift-
ing effect. In the case of anchored vessels, measurements of line length and tension are usually
interfaced to the PM control system. Any positioning system requires measurement of the vessel
position and heading. Wind sensors, measuring the wind speed and relative direction, are com-
monly used for wind feedforward control. In most commercial systems measurements of surge
and sway velocities are not available with sufficient accuracy and must be estimated on-line. A
guidance rule by DNV [54] is that the accuracy of the position reference data is generally to be
within a radius of 2% of the water depth for bottom-based systems, and within a radius of 3
meters for surface-based systems.
A minimum sensor and navigation configuration for a DP system typically consist of at
least one position reference system, one gyro compass, one VRU for roll and pitch measurement
and one wind sensor. The redundancy of the DP system can of course be increased by multi-
ple measurement devices. However, one should also consider using systems based on different
measurement principles, giving full redundancy not only in hardware configuration.
The trend today is integration between sensor systems and position reference systems. Over
the last years the price of accurate inertial motion units (IMU’s) have been reduced. This trend
opens for integration between inertial navigation systems (which are based on the IMU sensor
units) and other position reference systems e.g. GPS.

2.6.2 Modes of Operation

The DP and PM control functionality are closely connected to the surge, sway and yaw degrees
of freedom (DOF), which can be viewed as independent of the actual thruster configuration, as
long as there are enough thrust capacity to fulfill the force and moment demands. The various
DOFs can individually be controlled in the following modes of operation:

• Manual Control: By using a joystick and a rotation knob the operator of the ship can
generate force set-points in surge and sway and a moment set-point in yaw for manual
control of the ship.

• Damping Control: Feedback is produced from estimated low-frequency vessel velocities

and the objective is to regulate the vessel in the specific axis towards zero. This mode
is especially applicable for anchored vessels, where effective damping will reduce possible
large oscillatory motions, experienced about the resonance period of the anchored vessel,

16
 and thus reduce the stress on the mooring system. Damping is also useful in DP for
obtaining a smooth transition between transit speed and fixed position operations.

• Set-Point Control: Feedback is produced from both estimated low-frequency velocities

and position/yaw angle. The objective is to keep the actual axis at the specified set-point
position or heading.

• Tracking Control: The vessel tracks a reference trajectory which is computed from the
old to the new position or heading set-point.

The most common DP operation is naturally set-point control in all three axes, often referred
to as station keeping. Other fully automatic modes of operation can be station keeping with
weather optimal positioning, and roll- pitch damping. In the first case the vessel automatically
tends to the heading where the effect of the environmental loads are minimized. In the latter
case the roll and pitch motion is suppressed by proper action of the horizontal thrust compo-
nents. A combination between different modes are common, e.g. semi-automatic mode where a
joystick is used for manual surge and sway control, while the heading is automatically controlled.
Other tailor made DP functions exist for tracking operations like cable and pipe laying, ROV
operations, etc.
The combination of damping control in surge and sway and set-point control of heading
is often used in PM systems, especially in bad weather. Moreover, the vessel will tend to an
equilibrium position where the mean environmental forces are balanced by the mooring forces.
Assuming a proper heading set-point, this equilibrium position will be optimal with respect to
the thrust usage and the fuel consumption. For turret-moored ships automatic heading control
is often the most important function. By keeping the heading optimally against the weather
the effect of the environmental loads, and thereby the stress on the mooring system, will be
minimized.

2.6.3 Functionality and Modules

The various DP vendors may differ in design methods. However, the basic DP functionality are
more or less based on the same principles, as outlined in Figure 2.6.

17
 Process Plant
THRUSTER MEASUREMENTS
SETPOINTS

SIGNAL
THRUST PROCESSING
ALLOCATION
POWER
MANAGEMENT
SYSTEM POWER ADAPTIVE VESSEL
LIMITS LAW OBSERVER

CONTROLLER
COMMANDED VESSEL MOTIONS
THRUSTER FORCES

OPERATOR

Figure 2.6: DP SW modules integrated with the power management system.

Signal processing
All signals from external sensors should be thoroughly analyzed and checked in a separate signal
processing module. This comprises testing of the individual signals and sensor signal voting and
weighting when redundant measurements are available. The individual signal quality verification
should include tests for signal range and variance, frozen signals and signal wild points. If an
erroneous signal is detected, the measurement is rejected and not used by the positioning system.
The weighted signals from each sensor group should not contain any steps or discontinuities when
utilized further in the system ensuring a safe operation.

Vessel observer
Filtering and state estimation are important features of a positioning system. In most cases
today, accurate measurements of the vessel velocities are not available. Hence, estimates of the
velocities must be computed from noisy position and heading measurements through a state
observer. The position and heading measurements are corrupted with colored noise, mainly
caused by wind, waves and ocean currents. However, only the slowly-varying disturbances should
be counteracted by the propulsion system, whereas the oscillatory motion due to the waves (1st-
order wave-loads) should not enter the feedback loop. The so-called wave-frequency modulation
of the thrusters will cause unnecessary wear and tear of the propulsion equipment. In the
observer so-called wave filtering techniques are used, which separates the position and heading
measurements into a low-frequency (LF) and a wave-frequency (WF) part. The estimated LF
position, heading and velocities are utilized by the feedback controller. The observer is also
needed when the position or heading measurements temporarily are unavailable. This situation
is called dead reckoning, and in this case the predicted estimates from the observer are used
in the control loop. Another feature of the observer is that it estimates the unmodeled and
unmeasured slowly-varying forces and moments, mainly due to second-order wave loads and
ocean current.

18
Controller logics
The positioning system can be operated in different modes of operation, see Section 2.6.2.
All kind of internal system status handling and mode transitions, model adaptation etc. are
governed by the controller logic. This includes smooth transitions between the different modes
of operation, issue alarm and warnings, and operator interactions.

Feedback control law

The positioning controllers are often of the PD type (multivariable or decoupled in surge, sway
and yaw), where feedback is produced from estimated LF position and heading deviations and
estimated LF velocities. The underlying control methods may, however, vary. Traditionally,
decoupled controllers and linear quadratic controllers have been popular. In addition to the
PD part, integral action is needed to compensate for the static (or slowly-varying) part of the
environmental loads. The controller should be optimized with respect to positioning accuracy,
fuel consumption, and wear and tear of the propulsion system. In the design of positioning
controller, the control inputs are forces in surge, sway and moment in yaw. In the context of
positioning systems, this may be regarded as high-level control, since the actual control inputs
are shaft speed (RPM control) or the pitching of the propeller blades (pitch control), which
indirectly controls the developed force. In the case of azimuth thrusters, the direction of each
thrust device are additional control inputs.

Guidance system and reference trajectories

In tracking operations, where the ship moves from one position and heading to another, a
reference model is needed for achieving a smooth transition. In the most basic case the operator
specifies a new desired position and heading, and a reference model generates smooth reference
trajectories/paths for the vessel to follow. A more advanced guidance system involves way-point
tracking functionality, optimal path planning and weather routing for long distance sailing. The
guidance system could be interfaced to electric map systems.

Feedforward control laws

The most common feedforward control term is wind feedforward. Based on measurements of
wind speed and direction, estimates of wind forces and moment acting on the ship are com-
puted. As a consequence, a fast disturbance rejection with respect to varying wind loads can
be obtained. In order to improve the performance of the system during tracking operations, a
reference feedforward is computed. This is done by using a model of the ship dynamics, reference
accelerations and velocities, given by the reference model. In PM systems wind feedforward is
normally enabled in yaw only, since stationary wind loads in surge and sway should be compen-
sated by the mooring system. In addition, a line break detection algorithm is monitoring the
line tension signals and the corresponding vessel motions, in order to automatically detect break
in the anchor lines. When a line break is detected, the line break feedforward controller will be
activated in order to have the thrusters compensating for the lost forces and moment produced
by the broken line. This will ease the load in the surrounding lines and thus preventing yet
another line break.

19
Thrust allocation
The high-level feedback and feedforward controllers compute commanded forces in surge and
sway and moment in yaw. The thrust allocation module computes the corresponding force and
direction commands to each thrust device. The low-level thruster controllers will then control
the propeller pitch, speed, torque, and power satisfying the desired thrust demands. The thrust
allocation algorithm should be optimized for fuel consumption, wear and tear of the thruster
devices and for obtaining the commanded thrust in surge, sway and yaw. In addition, the
function should take into account saturation of the /pitch inputs and forbidden directional
sectors. The thrust allocation module is also the main link between the positioning system
and the power management system (PMS). The positioning system has a very high priority
as a power consumer. In any case, the thrust allocation must handle power limitation of the
thrusters in order to avoid power system overload or blackout. The thrust allocation module
receives continuously updated inputs from the PMS about available power and prevailing power
plant configuration with status on the bus ties and generators. This should prevent power
black-out and undesired load shedding of other important power consumers.

Model adaptation
The parameters in the mathematical model describing the vessel dynamics will vary with differ-
ent operational and environmental conditions. In a model-based observer and controller design,
the positioning system should automatically provide the necessary corrections of the vessel model
and controller gains subject to changes in vessel draught, wind area and variations in the sea
state. This can be obtained either by gain-scheduling techniques or continuously by using non-
linear and adaptive formulations. In addition other adaptive control and estimation methods
may be applied, either run in batches or processing on-line.

2.6.4 Advisory and Surveillance Systems

Use of advisory systems for diagnostics, simulation and analysis of future operational require-
ments, subject to varying environmental and operational conditions, becomes of increasing im-
portance for optimal operational planning. Such systems are integrated with the positioning
system and typical features are described below:

• DP and PM vessel motion simulators. Such simulators include mathematical models

of the environment and the vessel motion, and the operator can simulate the performance of
the positioning system, either using the prevailing environmental conditions or specify any
environmental condition. In addition, different types of failure scenarios can be simulated,
such as power or thruster failure, drive- and drift off. In PM systems failures in one or
several lines can be simulated as well.

• Consequence Analysis. This is an advanced version of a position capability analysis,

which continuously verifies that the vessel is capable of keeping position and heading for
different failure scenarios during the prevailing conditions. This can be loss of one or several
thrusters, one engine room or mooring lines (if applicable). In many cases a similar off-line
version is also available, where any environmental condition, operation or failure situation
can be simulated, by request from the operator.

20
2.6.5 DP Capability
The main purpose of a DP system is to keep position (and heading) within a certain excursion
limits within a specified weather window, or so-called design environment. In order to meet
the designed positioning capability, the system components should be reliable and the necessary
redundancy requirements should be met. During the design phase it is important to verify that
the amount of power and thrust capacity installed on a vessel will provide the necessary holding
capacity. This can be done either by static or dynamic analysis. In a static analysis only the
mean slowly varying forces due to wind, current and waves are considered. The items of data
required in such a study include:

• Main vessel particulars, such as displacement, length, breadth and operating draught.

• Directional dependent wind, current and wave-drift coefficients, from which the corre-
sponding forces and moments can be computed.

• The maximum environmental conditions in which the vessel should operate in dynamic
positioning (wind speed, significant wave height and current speed). Important parameters
are dominating wave period and the statistical description of the waves, usually described
by wave spectrum formulations such as the Bretscheider spectrum, the Pierson-Moskowitz
spectrum or the Joint North Sea Wave Project (JONSWAP) spectrum.

A rule of thumb is that the most loaded thruster should not use more than 80% of the
maximum thrust in the design environment to compensate for the static loads, (API). The
20% margin is then left for compensation of dynamic variations. The results of such a static
analysis can be presented as a capability plot (Figure 2.7). The sectors about 120◦ − 133◦ and
240◦ −250◦ where the ”butterfly-shape” curvature crosses the unity circle, indicate lack of thrust
capability to maintain the position and the heading for the environmental loads acting in these
directions. The environmental load data is given to the right on Figure 2.7. The butterfly-
shape curvature results from rotating the environmental loads 360◦ about the vessel. The load
directions given to the right are the prevailing directions from a real DP operation. The static
analysis with 20% margin is often too conservative. This suggests that the capability analysis
should be complemented with dynamic (time domain) simulations, where the thrust allocation,
the inherent thruster dynamics and dynamic thruster losses, forbidden azimuthing sectors and
the whole control loop is taken into account in the final verification. In the traditional capability
simulations the effect of power limitation is often neglected, which in fact often is the most
limiting factor in failure situations. A capability analysis with power limitation is also important
in the design of the power plant; how many power busses, amount of power generation on each
bus, which thruster to be connected to the different busses, etc.

2.7 Power and Energy Management

In a system of electrical power installations, vessel and process automation system, DP system,
and the other parts of the automation system controls their parts of the power system, e.g. the
DP system controls the thruster drives, the off-loading control system use cargo pump drives,
the process control system interacts with the compressors and cooling/heating systems etc. The
interconnecting point for all the installed power equipment is the power distribution system. By
starting and inrush transients, load variations, and network disturbances from harmonic effects,

21
 Figure 2.7: Typical DP capability plot for a supply ship - DP consequence analysis.

the load and generators are interacting and influencing each other. Optimum operation and
control of the power system is essential for safe operation with a minimum of fuel consumption.
As it is the energy control system (power/energy management system — PMS/EMS), which
monitors and has the overall control functionality of the power system, it will be the integrating
element in a totally integrated power, automation and DP system.
The purpose of the PMS is to ensure that there is sufficient available power for the actual
operating condition. This is obtained by monitoring the load and status of the generator sets
and the power system. If the available power becomes too small, either due to increased load
or fault in a running generator set, the PMS will automatically start the next generator set
in the start sequence. A PMS can also have extended functionality by monitoring and control
of the energy flow in a way that utilizes the installed and running equipment with optimum
fuel efficiency. Such systems can be EMS. Energy management is a new approach to control
and monitor the energy flow in marine, oil and gas systems. The EMS extends the concept of
power management in the direction of controlling and coordinating the energy generation and
consumption. In addition to optimize the instantaneous power flow, the historical energy usage
and future energy demands are considered. EMS will then be the integrating element in a totally
integrated power, automation and DP system. For PMS and EMS, the main functions can be
grouped in Sørensen and Ådnanes [301]:

• Power generation management: Overall control with frequency and voltage monitoring
with active and passive load sharing monitoring and possibly control, and load dependent
start and stop of generator sets. Since control logic and interlocking functions are a
significant part of the power system switchboard design, the functionally of these systems
must be coordinated.
• Load management: Load power monitoring and coordination of power limitation functions
in other systems, load shedding and start interlock of heavy consumers based on available

22
 power monitoring.

• Distribution management: Configuration and sequence control for reconfiguration of the

power distribution system. The distribution system should be configured to fit the re-
quirements in the actual operational mode for the vessel.

The new generation production vessels and also drill ships/rigs have a complex power system
configuration with advanced protection and relaying philosophies. There are close connections
between the functional design and performance of the PMS and the power protection system
functions. It is a challenge for involved parties to obtain an optimal and functional solution with
several suppliers involved and a yard being responsible for all the coordination.

2.7.1 Blackout Restoration

Blackout of the power generating system is the most severe fault that can happen in an electric
propulsion system. Should a blackout occur, and it does unfortunately happen from time to
time, there will normally be required to have a system for sequence control of start-up and re-
configuration of the power system. This is implemented at the system control level, and includes
sequences for starting and synchronizing generator sets and loads. There will normally also be a
set of predefined operation modes, e.g. transit mode, station keeping mode, maneuvering mode,
etc. with automatic sequence control for power system reconfiguration. A typical sequence for:
Automatic start and connection of emergency generator control. Separate control system -
not part of the overall vessel control system.

• Automatic start/reconnection of standby generator(s).

• Reconnection of transformer breakers, distribution breakers.

• Tie breakers remain open if tripped.

• Restart of fuel and cooling pumps.

• Restart of other pumps and fans in sequence.

• Restart of thrusters, if required.

2.7.2 Load Reduction and Blackout Prevention

The diesel-electrical power system has better performance and fuel economy if the PMS controls
the number of running engines in the power plant to match the load demand with high loading
of the diesel engine. The optimal load with respect to fuel consumption, tear and wear, and
maintenance is typically about 85% of maximum continuous rating (MCR). With a high load
at the running engines, the system gets more vulnerable to faults in the system, such as a
sudden trip of one diesel engine generator. The remaining, healthy engines will experience a
step load increase and possible overloading, and in worst case, under frequency trip unless the
functionality in the system reduces the load power in accordance with the generating capacity.
In a modern drilling vessel, this load reduction and blackout prevention function is distributed
and handled by several subsystems like:

• Power management’s load management and blackout prevention functions.

23
 10 sec
Fixed speed CPP load red
5 sec
Risk for
DP power limitation
1 sec Blackout
PMS blackout prevention
0,5 sec

0,1 sec VSI Drives FPP load red

Worst case overload Load

100% 200% 300% x MCR

Figure 2.8: Regulation time constants for power reduction, with maximum response time in the
order of 500ms (Illustrative only).

• DP system’s power limitation functions.

• Thruster drive’s load reduction and load phase back functions.

• Drilling drive’s load reduction and load phase back functions.

The ability to withstand such faults is also highly depending on the design and engineering
methods and solutions, such as:

• Load capability and dynamics of the diesel engine.

• Governor and AVR configuration and settings.

• Generator and switchboard protection relay settings.

• Critical equipment should be fault-tolerant with loss of power ride-through functionality.

In DP vessels with high efficient speed controlled fixed pitch thrusters, the total load under
normal operation is so low that the power plant runs optimal with few, often only two running
diesel engines. In order to utilize this saving potential, the challenge is to design and tune the
system to be capable of handling fault scenarios within a time frame not compromising the
stability of the power supply.
Figure 2.8 shows, for illustration, the diesel engines capability to maintain the frequency for
the load step associated with the loss of a parallel run engine. In typical installations, it has
been seen that the actions of load reduction and blackout prevention must be effective within
less than 500ms in order to not compromise the power system stability and limit the flexibility of
operation. Several solutions are in use and the operators and owners have different preferences.
However, some common conclusions can be made on what is typically required for the blackout
prevention functionality:

24
 • Thruster and thruster drives: Variable speed FPP thrusters must have a load reduction
scheme, either monitoring the network frequency and/or receiving a fast load reduction
signal from the PMS, either as a power phase-back signal, maximum power limitation sig-
nal, or — if well coordinated — fast RPM reference reduction. In order to avoid instabilities
in the network frequency, the load reduction should be as precise as possible in order to
dampen potential oscillations. Fixed speed CPP thrusters do not have fast enough re-
sponse time for blackout prevention. These must be included in the power management’s
load shedding scheme.

• Drilling drives: Similar to the requirements of the thruster drives, with built-in priorities
for the individual drilling drives.

• PMS: By class requirements, the PMS must include blackout prevention with load reduc-
tion/load shedding functionality. It was observed earlier, that the response time in this
system was too long to obtain the desired level of fault-tolerance without a fast acting,
stand-alone load reduction scheme in the thruster drives. With the knowledge of today,
this has been claimed solved by use of fast acting, and possibly event-trigged load reduction
algorithms.

• DP system: The DP system is also equipped with a power limitation function, normally
based on a permitted maximum power consumption signal from the PMS. Generally, this
has shown to be effective in avoiding overloading of the running plant, but not fast enough
to handle faults and loss of diesel-generator sets. Of importance is also that the power
limitation in manual and joystick control of the thrusters.

Based on experience, it is recommended that all load reduction and blackout prevention
functions described above are installed and well coordinated, tuned and tested during commis-
sioning and sea trial. Also, the need for retuning and testing must also be considered after
modifications in the installation that may affect the coordination.
A typical coordination diagram is shown in Figure 2.9. Auto start and auto stop limits
shows the level and time settings for load dependent automatic start and shutdown of the
engines. Available power will under normal operations be within these limits. Upon faults, and
sudden loss of engine, the available power is being reduced. The power reduction functions of
the DP can be distinguished between critical or non-critical situations, allowing the DP to take
all available power after possible load reduction and load shedding of non-essential consumers
or consumers with lower priority.

2.7.3 Diesel Engine Governor and AVR Fault Tolerance

Although the rules and regulations have allowed for DP operations with closed bus ties in the
electrical power system and one commonly connected power plant, the practices were until mid
1990s to split the network into two or more separated sections in DP 2 or 3 operations. With
the developments on switchboard protection relays, and thruster drives together with new PMS
with more precise and faster load reduction and blackout prevention functions, this has changed
and it is now more common to operate with normally closed bus tie breakers and also with
closed ring main bus. This development is motivated by the benefits of better and more flexible
utilization of the installed generating capacity and to gain an improved fuel economy. The fuel
savings for optimized loading of running engines have shown to be significant.

25
 t

PA,normal

Load
Reduction/
Shedding

Figure 2.9: Coordination of auto start/stop and blackout prevention functions (example).

In such systems, it has been observed that under some load and operating conditions, certain
faults in the governors and automatic voltage regulators, AVRs, can be difficult to identify
with a regular protection scheme. In worst case, faults have shown to interfere with healthy
equipment causing undesired shutdowns, and in some cases, even blackout. Developments in
digital generator protection relays with multifunction and programmable protection logics have
enabled a possibility to combine protection functions in new manners, and to include logic and
algebraic functionality into the protection scheme. This has significantly improved the protection
relay’s ability to detect governor and AVR faults, and improved the system’s fault tolerance to
such faults. Figure 2.10 shows a possible monitoring scheme for AVR, including correlation
functions of multiple variable and voting functions for multiple gensets. Such functionalities
can be installed in the modern, programmable multifunction generator protection relays, or in
separate logic controllers for retrofit and upgrades.

2.8 Maritime Industrial IT

We will in this chapter also shortly address aspects related to the information flow between
the real time systems ensuring on- line control and the management systems optimizing the
operations and the business processes. This includes fleet allocation related to transport logistics,
supply chain management, monitoring and diagnostics of technical condition of equipment and
systems.
As reported in Rensvik et al. [228] systems for operational management such as condition
monitoring and diagnostics systems, supply chain management systems, enterprise management
systems, etc. have increased the possibility to improve operational performance, productivity
and life cycle optimization of the assets related to operation of the installations.
Lately, the automation vendors, mainly in land-based industry sectors, have started the next

26
 I1-I4
U1-U4 Voting
algorithm Demagnetize
Q1-Q4 Trig limit
and Disconnect

Ii time delay Excitation Alarm

settings
Ui Correlation Excitation Fault
Qi algorithm

Figure 2.10: A monitoring scheme for AVR fault detection.

step to physically and functionally integrate the real time control systems with the operational
management systems. This has been denoted as industrial IT. The introduction of industrial IT
into marine applications has yet only started, and is still an area of research and development.
For vendors and ship operators it is a challenge to take out the potential this shift of technology
gives.
Maritime industrial IT solutions for the various marine market segments will be dependent
on type of trade and charter, vessel complexity, safety and availability requirements, size of
fleet, etc. Concerning vessel automation we will here focus on ships and vessels characterized
as advanced and specialized with high number of input/outputs (I/O) and rather complicated
operational functions. This is denoted as the high-end market; see also Figure 2.11. Examples
are ships (Figure 2.4) and rigs (Figure 2.1) for oil and gas exploration and exploitation, passenger
and cruise vessels. Typical applications in the offshore market are service vessels, drilling rigs
and ships, shuttle tankers, cable and pipe layers, floating production off-loading and storage
units (FPSOs), crane and heavy lift vessels, geological survey vessels and multi-purpose vessels.

As a part of enabling industrial IT solutions some vendors have installed condition monitoring
and control functions locally on the power equipment and field devices as indicated in Figure
2.3 with the possibility for remote monitoring and diagnostics.
In many installations redundant systems are available, and the number and types of measure-
ments required are specified by certain class rules. Physical integration based on standardized
communication protocols ensure connectivity of devices and integration of controllers and oper-
ator stations into three network levels (Figure 2.3) real time field bus network communication
on low level between devices and controllers, real time control network connecting controllers
and operator stations, and office plant network to various office systems and information man-
agement systems. The last level opens up for satellite communication to land offices at ship
operators or vendors, see also Figure 2.12. Industrial IT is supposed to increase the integration
of vessel plant data with the business management systems ensuring optimized asset manage-
ment and operation of each vessel in particular and the whole fleet on corporate level, see Fig.

27
 Automation
Level
High
Cruise Vessels,
end
FPSO, Drilling Ships/Rigs,
Shuttle tankers,
LNG Ships

Offshore service
vessels Chemical
tankers
Fishing
vessels RO-RO Car
ships
Tankers
Container
Low Bulk ships ships
end

Low end High end Logistics: Ship

Management Level

Figure 2.11: Low-end and high-end market segments within automation and ship management.

Maintenance Maintenance
management management
system system
Operation Operation
system system
Mail
Mail
system
system

Office Plant Network External parties:

Control Network -Suppliers
-Class
-Authorities
T A R O NG A -..
T A R O NG A
T Ø NS B E R G

Figure 2.12: Distributed information systems ships — shore for flexible information flow and
sharing.

28
 Enterprise business:
- Logistics chains (D2D, ..)
Business - Fleet management/allocation
Systems - Corporate management

Real Time Flow of Plant and

Operational management:
- Maintenance philosophies
Operational Systems - Enterprise resource planning

Transactional Data
- Supply chain management
- Diagnostics
- Condition monitoring

Physical and functional

Integration:
Automation Systems and Devices - Connectivity
- Modularity
- Openness
- Optimization

Figure 2.13: Industrial IT architecture.

Figure 2.14: Fleet allocation and operations managed on cooperate level.

2.13
As cost on the vessel-to-land satellite communication is reduced (Figure 2.14) and the mar-
itime information technology architecture is improved this kind of information flow is expected to
be working seamless in real time, as opposed today, where a limited amount of date is transferred
at discrete events.
Up to now integrated automation systems have been proprietary with a limited number of
vendors. However, in the automation industry it is a trend towards openness in communication
protocols and network. How this will influence on the technology solutions and responsibilities
for multi-vendor integration systems is still a subject for discussion.
Condition monitoring and RCM (Reliability Centered Maintenance) are well known concepts
in many industries. These concepts have now also been incorporated in ship maintenance man-
agement systems. Today, ships in many trades have only very short loading and discharging
periods in port. This means that maintenance that only can be performed in port including

29
 Internal: Fleet Manager, superintendent, risk management personnel, ….

External: Class, authorities, customers, yard, assurance, suppliers, ….

Condition
of vessels

Condition
of system
Hull
Hull
Propellers/
Propellers/
thrusters
Powerplant
Power plant
Prime
Prime
movers
.. Cargo
Cargo
system
thrusters movers system

Main Auxiliary
engine systems Condition of
components
and equipment

Monitoring – diagnostics – control – maintenance – spare part – survey data, ….

Figure 2.15: Technical condition philosophy.

flag state and class surveys have to be carried out during these short periods to avoid down-
time. To improve such a harmonization, analysis and planning methods have to be available to
monitor the status of the vessel, to schedule surveys and to predict the future state of the ship
based on frequent reporting and continues monitoring of technical condition for structures and
equipment (Figure 2.15). This concept can further be developed into a “continues” risk analysis
based on this technical condition information, and adapted to highlight risk equipment related
to the different operational mode of the ship, i.e. a way to enhance safety measures. Frequent
tools in technical monitoring is the use of signal processing techniques such as spectrum analysis
and other statistical methods. Examples are acceleration measurements for vibration analysis,
temperature measurements for overheating, etc. Lately, the use of observers for state estimation
may also be used.

2.9 Rules and Regulations

2.9.1 Class Requirements
In many cases it is up to the owners to decide which type of classification (or to seek for class
approval at all) for the positioning system installed. Three examples of class societies that have
rules for classification of DP systems are DNV [54], ABS [3] and LRS [174]. The International
Marine Organization (IMO) has also developed Guidelines for Dynamic Positioning, in order to
provide an international standard for DP systems on all types of new vessels. The purpose of
the Guidelines and class rules is to recommend design criteria, necessary equipment, operating
requirements, and a test and documentation system for DP systems to reduce the risk to per-
sonnel, the vessel itself, other vessels and structures, sub-sea installations and the environment

30
while operating under dynamic positioning control. Taking into consideration that DP operated
vessels often operate in different parts of the world, such a standardization provides a useful tool
for the different Costal states to specify the local rules and regulations, defining levels of safety
requirements, requirements for redundancy and operations for DP vessels. For moored vessels
using automatic thruster assistance, IMO does not have any guidelines. DNV has the class
notation called POSMOOR ATA, DNV [55] and Lloyd’s Register of Shipping has the notation
PM or PM T1, LRS . More general document for analysis, design and evaluation of moored,
floating units is published by the American Petroleum Institute [11]. Here some guidelines for
DP systems are also included.
The requirements for hardware and software on DP systems are closely connected to the
level of redundancy, which is defined as:

Definition 2.2 (Redundancy) Redundancy means ability of a component or system to main-

tain or restore its function when a single fault has occurred. This property can be obtained by
installation of multiple components, systems, or alternative means of performing a function.

A DP system consists of components and systems acting together to achieve sufficiently

reliable positioning keeping capability. The necessary reliability of such systems is determined
by the consequence of a loss of position keeping capability. The larger the consequence, the
more reliable the DP system should be. To achieve this philosophy the requirements have been
grouped into three different equipment classes. The equipment class depends on the specific DP
operation, which may be governed by Costal state rules and regulations or in agreement between
the DP operator company and their customers. A short description of the different classes is
given below.

Class 1 For equipment class 1, loss of position may occur in the event of a single fault, e.g. the
DP control system need not to be redundant.

Class 2 For equipment class 2, loss of position is not to occur in the event of a single fault
in any active component or system. The DP control system must have redundancy in all
active components, e.g. the hardware must consist of at least two independent computer
systems with self-checking routines and redundant data transfer arrangements and plant
interfaces. At least three independent position reference systems and three sensor systems
for vertical motion measurement, three gyrocompasses and three wind sensors.

Class 3 Same as DP class 2, with additional requirements on redundancy in technical design

and physical arrangement.

Class 2 or 3 system should include the function ”Consequence analysis”, which continuously
verifies that the vessel will remain in position even if the worst single failure occurs. The IMO
Guidelines also specifies relationship between equipment class and type of operation. DP drilling
operations and production of hydrocarbons, for instance, requires equipment Class 3, according
to IMO. In Table 2.1 the coherence for DP class notations by DNV, LRS, ABS and IMO are
shown.
In Figure 2.16 DNV [54] has listed up the minimum requirements for DP class notation.

31
 IMO Equipment Class DNV Lloyds ABS
Not applicable Dynpos AUTS DP CM DPS-0
Class 1 Dynpos AUT DP AM DPS-1
Class 2 Dynpos AUTR DP AA DPS-2
Class 3 Dynpos AUTRO DP AAA DPS-3

Table 2.1: Mapping table of DP class notation by IMO, DnV, Lloyds and ABS.

Figure 2.16: Minimum DNV requirements (2010) for DP class notation, courtesy by DNV.

32
2.9.2 Reliability and Redundancy
From a safety point of view a DP system can be viewed as four different sub-systems. Each
sub-system can then be further split recursively into sub-sub systems. For instance:

• Level 1: Power system. Level 2: Power generation, power distribution, drives, etc.

• Level 1: Propulsion system. Level 2: Main screw, tunnel thrusters, azimuth thrusters.

• Level 1: Positioning control system. Level 2: Computer and I/O, Operator HMI, UPS,
Operator interaction.

• Level 1: Sensor system. Level 2: Gyros, Position reference systems, Wind sensors.

Starting from the bottom, the reliability of each sub-component can be specified and de-
pending on the level of redundancy, the reliability and availability of the total system can be
computed using statistical methods. Each component can be characterized by:

• Failure rate , defined Maximum number of failures per million hours.

• Mean time between failures (MTBF), for one component given by

1
   =


• Total down time,  , of a component, including mean time to repair (MTTR).

• Availability, , defined as

  
=
   + 
The above characteristics are specified for each component or each level in a reliability
analysis and will at the top level characterize the whole plant.
The redundancy concept of the electrical power and propulsion plant will be based on the
required ability for maneuvering and propulsion after faults in the system. In commercial vessels,
these requirements are determined by national and international legislations, and specified by
the classification societies by the different class notations. A typical redundancy diagram for a
DP class 2/3 drilling rig is indicated in Figure 2.17. Each block represents a part of the system
which is susceptible to a single failure.
At higher level of controls the redundancy concept is different and achieved by the duplication
of the control systems in hot back-up configuration, as seen in Figure 2.18 for a typical DP 3
drilling rig.
The vessel control system must follow the same redundancy and segregation principles as
the electric system. Principally, this is achieved by designing the vessel control structure as a
mirror image of the electrical power plant, as shown in Figure 2.19.

33
 Figure 2.17: Each block represents a part of system which is sensitive to a single failure.

2.9.3 Failure Analysis

The cost of making design changes during the initial project phase is small, but as the project
progresses the cost of design changes increases significantly. Minor changes in a system can
cause project delays and huge additional costs during the commissioning and seatrials. During
the whole design phase of newbuildings, the reliability of the total system can be thoroughly
investigated by different reliability methodologies, in order to detect system design errors and
thus minimize the risk. Such analysis of DP system and it’s subsystems is not required by the
class societies or the IMO Guidelines. However, the Costal states, oil companies or customers
may require it, in addition to classification approval. There are different methods available for
assessing the reliability of such complex systems.
One common methodology is the Failure Mode and Effect Analysis (FMEA). This is a
qualitative reliability technique for systematically analyzing each possible failure mode within
a system and identifying the resulting effect on that system, the mission and personnel. This
analysis can be extended by a criticality analysis (CA), a quantitative procedure which ranks
failure modes by their probability and consequence.

2.10 Simulation
For years numerical simulators have been used as tools in system design and analysis, both in
academia and in the industry. The simulator may use models of various fidelity to reconstruct
the real physical properties of a dynamic system, Sørensen et al. [298]. For control system

34
Figure 2.18: A total integrated vessel control system for drilling vessels (Courtesy of Kongsberg
Maritime).
35
 G G G G

AC AC AC AC
AC AC AC AC

M M M M

G G G G

AC AC AC AC
AC AC AC AC

M M M M

Figure 2.19: Mirror image design of the vessel control system ensures to achieve a unified
redundancy and segregation philosophy.

36
 Simulator

Figure 2.20: Mathematical software models of ship, equipment and its systems are implemented
and analyzed subject to varying environmental loads, operational requirements and scenarios.

design and testing purposes it is convenient to develop a real-time system simulator, Figure 2.20
In order to operate in real-time often simplified or equivalent models of fast dynamic systems
like power electronics, complex systems like multi-dimensional finite element method (FEM)
models of structures and panel methods of hydrodynamics must be used. Model reduction and
simplifications must be done with care such that important structural information and properties
of the dynamic system are not lost
At NTNU a marine system simulator illustrated in Figure 2.21, Marine Systems Simulator
(MSS, [195]), based on MATLAB/SIMULINK. MSS integrates the disciplines hydrodynamics,
structural mechanics, marine machinery, electric power generation and distribution, navigation
and automatic control of marine vessels. The main purpose of MSS is to improve the accu-
mulation and reuse of knowledge and thereby the quality of the education and research. The
simulator will be continuously developed by students and researchers, and will serve a diversity
of applications. This necessitates a modular structure, in which each module is a self-contained
unit with a well-defined interface and functionality.

2.10.1 Simulator Structure

The core of the simulator is the process plant models or simulation models, which give the neces-
sary detailed description of the vessel dynamics, systems and components and its surroundings,
see Figure 2.21. The other main parts of the simulator are the control systems interfaced with
the sensor and actuator modules. The control systems may for instance be DP system, thruster
assisted position-mooring system, tracking controllers and autopilots, local thrust controllers,
PMS, crane control systems, etc. MSS is made partly as open-source SW.

2.10.2 Module Hierarchy

Depending on the development stage and required accuracy of the operation and application
studied, a hierarchy in module complexity is allowed. For any application, several modules of
varying technical complexity, built-in functionality and release levels may exist. These should
all cover the same basic functionality and have the same interfaces, and may therefore easily

37
 Simulation Environment
Scenario FMEA User Visualization Analysis DP
Management Interfaces Toolbox control

Power
Run Time Infrastructure management

Power System: Other Models: Sensor and Crane

Vessel Models: Propulsion
- Current, wind, System:
- Diesel engine - Crane Instrumentation: control
waves, Ice - Gas turbines - Balla st -DGPS
- Thrus ter
- Stationkeeping - Generators - Towing / cables -HPR
- Rudde r
- Maneuvering - Transformators - Trenc hing -Gyros
- Pod Ballast
- Converters - Risers -MRU/VRS
- e tc.
- Switchboards - Mooring -Wind sensors control
- etc. -…

Environment data Hulls Propulsion units Power components Cranes

Figure 2.21: Marine Systems Simulator, NTNU.

be interchanged. Figure 2.22 shows a module hierarchy example, where module complexity is
plotted versus release level and application.

2.10.3 Hardware and Software Platform

The simulator is currently being developed in a MATLAB/SIMULINK environment on the
Windows PC platform. SIMULINK was chosen for running the main simulation loop because
of its flexibility towards several programming languages. Applications written in MATLAB,
C, C++, and Fortran may easily be linked to the simulation by use of S-functions. This is
convenient for generating simulator modules from existing code, and makes development of new
modules more user-friendly.

2.10.4 Hardware-In-the-Loop Testing

As the DP vessels become more demanding and complex, safety, reliability and integration
aspects with the navigation system, power plant, vessel automation, propulsion system and
other consumers become more important. In order to reduce these risks, regulatory bodies,
class societies and independent consultants have been continuously addressing advances in rules
and regulations and testing and verification methodologies. In this context the safety and
verification regime for DP systems may be seen as an example to be followed for other mission
critical control systems as well.
The successful operation of DP vessels such depends more and more on advanced integrated
functionality of software-based control systems. Consequently, software related problems, of-
ten in conjunction with hardware and/or human errors, may lead to vessel construction delays,
downtime during operation, reduced income for clients, increased cost, and reduced safety. In

38
 Figure 2.22: Module hierarchy example

order to reduce these risks, independent third party Hardware-In-the-loop (HIL) simulator test-
ing has recently been applied for extensive software testing and verification of DP systems on
several offshore vessels. In the work of Johansen et al. ([135], [136]), Johansen and Sørensen
[137] and Smogeli, [267] the concept of HIL testing is described, and the experiences and findings
statistics are reported from HIL testing of DP computer systems, power management systems
and steering, propulsion and thruster control systems on drilling vessels, offshore service and
construction vessels, and shuttle tankers. The main idea is testing and verification of the com-
puter software using a vessel specific simulator (Figure 2.23) capable of simulating the dynamic
response of the vessel, thruster and propulsion system, sensors, position reference systems, power
generation, distribution, main consumers, and other relevant equipment (Sørensen et al., [298]).
The simulator is connected via network or bus interfaces to the targeted control system such that
all relevant feedback and command signals are simulated. In order to achieve the test objective,
the simulator is capable of simulating a wide range of realistic scenarios defined by operational
modes, operational tasks and single, common mode and multiple failure modes in order to verify
correct functionality and performance during normal, abnormal and faulty conditions.
HIL testing may be conducted in several phases of a new-building or retrofit, where the
first phase is usually an extensive software test conducted at factory or a lab facility. By
using HIL simulator technology a virtual sea trial with thorough testing is conducted before
the vessel is built. The objective is fully functional and failure testing of the software before
the commissioning and integration starts, ensuring that the software will be more finalized and
ready for commissioning. Follow-up system and integration testing is normally conducted during
commissioning, and a final verification of the integrated functionality is conduced onboard the
vessel at the end of commissioning.

39
 Vessel Simulator
Real time
interface

DP system

Figure 2.23: HIL Testing.

40
Chapter 3

Maritime Electrical Installations and

Diesel Electric Propulsion

Alf Kåre Ådnanes

ABB Marine, Pb. 6540 Rodeløkka, N-0501Oslo, Norway.

3.1 Introduction
3.1.1 Scope and Objectives
Electrical installations are present in any ship, from powering of communication and navigation
equipment, alarm and monitoring system, running of motors for pumps, fans or winches, to high
power installation for electric propulsion.
Electric propulsion is an emerging area where various competence areas meet. Successful
solutions for vessels with electric propulsion are found in environments where naval architects,
hydrodynamic and propulsion engineers, and electrical engineering expertise cooperate under
constructional, operational, and economical considerations. Optimized design and compromises
can only be achieved with a common concept language and mutual understanding of the different
subjects.
The objective of this section is to give an introduction to electro-technology in general, and
put special emphasis on installations for electric propulsion. It is the aim to give engineers
with marine competence and background the necessary understanding of the most important
electro-technical subjects used in design and configuration of ships with electric propulsion.
After an introductory review of the history of electric propulsion in Section 3.1 and ap-
plication areas of electric propulsion in Section 3.2, an overview of the electric power system
in Section 3.3 and its associated control systems in Section 3.4 follows before the main char-
acteristics of the electric propulsion drives are presented in Section 3.5. Important design and
engineering considerations are discussed in Sections 3.6 and 3.7 before ending by showing typical
arrangements in Section 3.8 by use of single line drawings of the electrical installations in some
important applications.

41
Figure 3.1: Three comparative concepts of a Ropax vessel showing how space can be better
utilized with electric propulsion and podded propulsion.

3.1.2 Motivations for Electric Propulsion

The concept of electric propulsion is not new, the idea originated more than 100 years ago.
However, with the possibility to control electrical motors with variable speed in a large power
range with compact, reliable and cost-competitive solutions, the use of electrical propulsion has
emerged in new application areas during the 80’s and 90’s.
Electric propulsion with gas turbine or diesel engine driven power generation is used in
hundreds of ships of various types and in a large variety of configurations. Installed electric
propulsion power in merchant marine vessels was in 2002 in the range of 6-7  (Gigawatt),
in addition to a substantial installation in both submarine and surface war ship applications.
By introduction of azimuthing thrusters and podded thrust units, propulsion configurations
for transit, maneuvering and station keeping have in several types of vessels merged in order
to utilize installed thrust units optimally for transit, maneuvering and dynamically positioning
(dynamic positioning - DP).
At present, electric propulsion is applied mainly in following type of ships: Cruise vessels,
ferries, DP drilling vessels, thruster assisted moored floating production facilities, shuttle tankers,
cable layers, pipe layers, icebreakers and other ice going vessels, supply vessels, and war ships.
There is also a significant on-going research and evaluation of using electric propulsion in new
vessel designs for existing and new application areas.
The following characteristics summarize the main advantages of electric propulsion in these
types of vessels:

• Improved life cycle cost by reduced fuel consumption and maintenance, especially where

42
 there is a large variation in load demand. E.g. for many DP vessels a typically operational
profile is equally divided between transit and station keeping/maneuvering operations.

• Reduced vulnerability to single failure in the system and possibility to optimize loading of
prime movers (diesel engine or gas turbine).

• Light high/medium speed diesel engines.

• Less space consuming and more flexible utilization of the on-board space increase the
payload of the vessel, see Figure 3.1.

• Flexibility in location of thruster devices because the thruster is supplied with electric
power through cables, and can be located very independent on the location of the prime
mover.

• Improved maneuverability by utilizing azimuthing thrusters or podded propulsion.

• Less propulsion noise and vibrations since rotating shaft lines are shorter, prime movers
are running on fixed speed, and using pulling type propellers gives less cavitation due to
a more uniform water flow.

These advantages should be weighted up against the present penalties, such as:

• Increased investment costs. However, this is continuously subject for revisions, as the cost
tends to decrease with increasing number of units manufactured.

• Additional components (electrical equipment — generators, transformers, drives and mo-

tors/machines) between prime mover and propeller increases the transmission losses at full
load.

• For newcomers a higher number and new type of equipment requires different operation,
manning, and maintenance strategy.

High availability of power, propulsion and thruster installations, as well as safety and automa-
tion systems, are key factors in obtaining maximum operation time for the vessel. The safety
and automation system required to monitor, protect, and control the power plant, propulsion
and thruster system, becomes of increasing importance for a reliable and optimum use of the
installation (Blokland and van der Ploeg [33]).

3.1.3 Power Flow and Power Efficiency

In any isolated power system, the amount of generated power must be equal to the consumed
power including losses. For an electric system consisting of an electric power generation plant,
a distribution system, including distribution transformers and a variable speed drive, the power
flow can be illustrated in Figure 3.2.
The prime movers, e.g. diesel engines or gas turbines, supply a power to the electric generator
shaft. The electric motor, which could be the propulsion motor, is loaded by a power from its
connected load. The power lost in the components between the shaft of the diesel engine and the
shaft of the electric motor is mechanical and electrical losses which gives heat and temperature
increase in equipment and ambient. The electrical efficiency of the system in Figure 3.2 is

43
 POWER FLOW Pout
Pin

Power losses

Pin Pout

Generator Switchboard Trans- Frequency Electric

former converter Motor

Figure 3.2: Power flow in a simplified electric power system.

 
= =  (3.1)
  + 
where  is the generated power,  is the delivered power to the load (thruster), and 
is the losses. For each of the components, the electrical efficiency can be calculated, and typical
values at full (rated) power are:
- Generator:  = 095 − 097
- Switchboard:  = 0999
- Transformer:  = 099 − 0995
- Frequency converter:  = 098 − 099
- Electric motor:  = 095 − 097
Hence, the efficiency of a diesel electric system, from diesel engine shaft, to electric propulsion
motor shaft, is normally between 0.88 and 0.92 at full load. It must be noted that efficiency is
strongly depending on the loading of the system.
Since the additional components between the prime mover and the propeller shaft in a
diesel electric propulsion system contributes to a total of approximately 10% losses, the fuel
savings potential is not due to the electrical component. One must regard the hydrodynamic
efficiency of a speed controlled propeller compared to a fixed speed controllable pitch propeller
(CPP), and the fuel efficiency of the prime mover when installed in a diesel electric system with
constant speed and high loading, compared to in a mechanical propulsion system with strongly
varying load. The differences may be significant, especially on low thrust operations as DP and
maneuvering.
Fig. 3.3 shows the fuel efficiency of a typical medium speed diesel engine, and Fig. 3.4 a
power vs. thrust comparison of a variable speed and a controllable pitch propeller (CPP).
The hydrodynamic losses will vary significantly dependent on the operational condition for
a CPP used in direct driven diesel solutions compared to variable speed fixed pitch propellers
(FPP), which normally are used in electrical propulsion. In low load condition it is a rule of
thumb that the zero-load hydrodynamic losses for a CPP is about 15%, while it is close to 0
for a speed controlled FPP, see Fig. 3.4. Notice that in most CPP configurations the propeller

44
 350

300

250
[g/kWh]

200

150

100

50

0
0% 20 % 40 % 60 % 80 % 100 % 120 %
[%MCR]

Figure 3.3: Diesel engine specific fuel consumption (typical).

speed has to be kept constant on quite high rotations per minute (  ) even though the
thrust demand is zero. For FPP the variable speed drive will allow zero   at zero thrust
demand. The advantage with CPP is that the propeller pitch ratio will be hydro-dynamically
optimized for a wider speed range. A propeller designed for high transit speed, will have reduced
efficiency at low speed and vice versa. Hence, the operational profile is of major importance
while designing the propulsion system.
The fuel efficiency characteristics of the diesel engine, with maximum fuel efficiency in the
load range of 60 to 100% load, strongly contribute to the difference in power consumption for
a traditional mechanical propulsion system, and a diesel electric propulsion system. In a power
plant for diesel electric propulsion, the power generation will consist of multiple smaller diesel
engines, where the number of running aggregates can be selected to have an optimum loading
of each engine. The rating of the engines can also be adapted to fit the intended operational
profile of the vessel, ensuring that it is possible to find an optimal configuration for most of the
operational modes and time.
For a field support vessel, with operational profile as shown in Fig 3.5, it was found that
the fuel savings by using diesel electric propulsion was in the range of 700 tons of diesel per
year. With a price of approximately 3 NOK per liter, this gives annual savings of in the order of
2.1 million NOK (280 thousand USD). As shown, the savings will strongly be dependent on the
operational profile, as shown in Fig 3.6. Here, the operational profile is split in DP/maneuvering
and in Transit, showing how an increase portion of DP operations will increase savings, and vice
versa.

3.1.4 Historical Overview of Electric Propulsion

After the rather experimental applications of battery driven electric propulsion at the end of
the 19th century took place in Russia and Germany, the first generation electric propulsion was
taken into use in the 1920’s as a result of the strong competition for reducing transatlantic

45
 3000 Pow er, kW

2500

2000
Fixed speed
1500 C P propeller
1000
Variable speed
500 FP propeller

100 200 300 400 500

100 Thrust, kN
200

300 C um ulative
thrust dem and
Days

Figure 3.4: Propeller bollard pull characteristics (example).

At quay, 6 % By platform / ROV, 11 %

Standby
40 % Transit, 11-12 knots
40 %

Anchor handling, 6 %

Figure 3.5: Operational profile for a field support vessel

46
Figure 3.6: Fuel consumption, compared for electric propulsion (Azipod) and conventional me-
chanical propulsion

crossing times for passenger liners. At that time, the high propulsion power demand could
only be achieved by turbo-electric machinery. “S/S Normandie” was one of the most renowned.
Steam turbine generators provided electric power that was used to drive the 29  synchronous
electrical motors on each of the four screw shafts. The rotational speed was given by the electrical
frequency of the generators. The generators would normally run one propulsion motor each, but
there were also possibility for feeding two propulsion motors from each generator for cruising at
lower speeds.
With the introduction of high efficient and economically favorable diesel engines in the middle
of the 20th century, steam turbine technology and electric propulsion more or less disappeared
from merchant marine vessels until the 1980’s.
The development of variable speed electric drives, first by the / rectifier (Silicon Con-
trolled Rectifier — SCR) in the 1970’s and the / converters in the early 1980’s enabled the
power plant based electric propulsion system, which is typical for the second generation electric
propulsion. A fixed voltage and frequency power plant consisting of a number of generator-sets
feeding to the same network was supplying the propulsion as well as the hotel and auxiliary
power. The propulsion control was done by speed control of the fixed pitch propellers (FPP).
These solutions were firstly used in special vessels like survey ships and icebreakers, but also
in cruise vessels. “S/S Queen Elizabeth II” was converted to electric propulsion in the mid
1980’s, and later followed the Fantasy and Princess class cruise vessels, several DP vessels, and
shuttle tankers. Notice that in direct driven diesel propulsion the thrust is normally controlled
by a hydraulic system varying the propeller pitch angle. This is denoted as controllable pitch
propellers (CPP).
Podded propulsion was introduced in early 1990’s where the electric motor is installed di-
rectly on the fixed pitch propeller shaft in a submerged, rotatable pod. While this concept was
originally developed to enhance the performance of icebreakers, it was early found to have addi-
tional benefits on hydrodynamic efficiency and maneuverability. After the fist application in a
cruise liner, “M/S Elation”, the advantages were so convincing that podded propulsion almost
over night became a standard on new cruise liners, Figure 3.7.

47
Figure 3.7: Cruise vessel “M/S Elation” (lower right) equipped with Azipod propulsion frees
up space compared to sisterships (upper left) that can be utilized for other purposes, e.g. grey
water treatement.

48
3.2 Applications
3.2.1 Passenger Vessels — Cruise Ships and Ferries
Passenger vessels, cruise ships, and ferries have very high requirement for on-board comfort
regarding noise and vibration. In addition, the reliability and availability is very critical for the
safety of the passengers and the vessel. Consequentially, electric propulsion was early evaluated
to be beneficial and taken into use.
The list of cruise vessels with electric propulsion is today long and increasing. As the podded
propulsion is shown to give significant improvements in maneuverability and fuel costs, with an
increase in propulsion efficiency of up to 10% (Kurimo [157]), a large and increasing portion of
new-buildings are specified with electrical podded propulsion.
As the environmental concern is increasing, the requirements of reduced emission, spill, and
damages on coral reefs by anchoring of the cruise vessels are increasing. Hence, the vessel must
maintain its position solely by thrusters controlled by a DP system. This will increase the need
for electrical propulsion and podded propulsion in the cruise market even more.
The same restrictions and tax penalties for gas emissions (COx, Nox and Sox) have resulted in
that several recent new buildings of ferryboats for fjord and strait crossing have been equipped
with electric propulsion. With frequent crossing schedules and quay docking, the improved
maneuverability by podded propulsion has significantly reduced the fuel consumption. The
propulsion power varies with the size of the vessel, from some few   for smaller ferries up
to 30-40  for large cruise liners. The hotel load can be a significant part of the total power
installation, for a large cruise liner typically in order of 10-15   .
Fig. 3.8 shows a schematic overview of the main electrical and automation components in a
typical cruise vessel with diesel-electric podded propulsion.

3.2.2 Oil and Gas Exploitation and Exploration: Drilling Units, Production
Vessels and Tankers
Still a few years ago extensive oil and gas resources were accessible in shallow waters and
could be exploited by fixed drilling and production units. In the North Sea, Gulf of Mexico
(GoM), and Brazil as in several other areas, those new resources that remain are found in
smaller and/or less available fields in deeper waters. These fields require new cost-effective
methods to obtain acceptable economy and profit. Deep-water drilling and floating production
have become possible with dynamic positioning or thruster-assisted position mooring. Thruster
assisted positioning is applied in the North Sea, Canada, and areas with harsh environment. In
Brazil, West Africa, and the planned USGOM (US Gulf of Mexico) installations, the trend has
been to rely on mooring without thruster assistance for oil production and dynamic positioning
for deepwater drilling.
The thrusters used for station keeping (DP operation) typically also constitutes the main
propulsion in transit and maneuvering of the vessel, either all or selected units only.
Typical of these vessels is their large installed thruster power, typically 20-50   . Together
with the production, drilling, utilities, and hotel loads, the installed power is typically 25-55
  . The typical installation has a common power plant for all these loads, enabling flexibility
to operation with high energy-efficiency and high availability. Fig. 3.9 shows a schematic
overview of a semi-submersible drilling rig. See Farmer [72], Ådnanes et al. [337] and Ådnanes
[335].

49
 INTEGRATED VESSEL AUTOMATION
DYNAMIC POSITIONING

COMMUNICATION LINE TO SHORE

POS. REF. SYSTEMS

X TERMINALS

PLANT NETWORK

ENGINE
CONTROL

TUNNEL
THRUSTERS

AUX POWER GENERATION

& DISTRIBUTION

FIELDBUS
NETWORK

CONTROL
NETWORK

MAIN POWER GENERATION

& DISTRIBUTION

DRIVES

PROPULSION/
POD

Figure 3.8: Example of propulsion and control system layout for a cruise vessel.

Shuttle tankers are used for transport of oil from an offshore facility (platforms, buoys, towers
or FPSOs) to a processing or storage terminal onshore. There are numerous different off-loading
methods in use. For most of them the shuttle tankers need to maintain a fixed position (station
keeping) with high accuracy subject to varying environmental conditions. Therefore, most of
the shuttle tankers are equipped with a DP system. Most of the ships have installed electrical
tunnel or azimuthing thrusters, whereof some are also having diesel-electrical systems for main
propulsion, Hansa-Schiffart [104].
For many applications there is a high degree of redundancy in the propulsion for transit and
station keeping. The solutions have normally a redundant power generation and distribution
system, with redundant propulsion converters, and a tandem or redundant propulsion motor.
The introduction of podded propulsion may influence the design of the diesel-electric shuttle
tankers, since it may be a more cost efficient solution to obtain redundant propulsion with two
pod units than with two conventional shaft lines

3.2.3 Field Support Vessels and Construction Vessels

For vessels with dynamic positioning (DP) as the main operating mode, such as diving support
vessels, crane ships, and pipe layers (Fig. 3.10 and 3.11), electric propulsion was early taken
into use, first with fixed speed CP propellers and later with variable speed thrusters.
The reduction in fuel consumption and environmental emission from diesel electric propul-
sion compared to conventional mechanical propulsion is significant for vessels with a diversified
operational profile. Savings of 30-40% in fuel consumption annually has been reported from ship
owners, and with the increased focus on operation costs and environmental impact from the oil

50
 BACK-UP
PLANT
SYSTEM
NETWORK
SAFETY SYSTEM
EMERGENCY SHUTDOWN WIND SENSORS PROCESS CONTROL
FIRE & GAS STATION

MRU

GYRO

- POSMOOR
- AUTOSAIL

- EXTENSION ALARM
- PROCESS CONTROL

DRILLING DRIVE
SYSTEM FIELDBUS
NETWORK
ENERGY MANAGEMENT
SYSTEM

POWER GENERATION
& DISTRIBUTION

Figure 3.9: Example of electrical system layout for a semi-submersible drilling unit.

51
 Figure 3.10: Offshore supply vessel with electric propulsion.

industry, has given a large growth in number of field support vessels, first in the North Sea, and
later in other geographical areas.
With the rapidly increasing need for high-speed communication system and a global fiber
optic cable network, there has been established a large fleet of cable laying vessels with electric
propulsion and dynamic positioning.
These vessels will be configured as DP vessels, class 2 or 3 (DnV [54], LRS [174] and ABS
[3]), and most will have electric propulsion with a total power demand of 8-30   , depending
on size and drilling/lifting capability.

3.2.4 Icebreakers and Ice Going Vessels

The dynamic requirements for the frequency converters in propulsion application are low, com-
pared to many other industrial applications. But in ice-going vessels and icebreakers (Figure 3.12
and 3.13), the load variations may be significant and rapid, and this implies that the propulsion
system must have high dynamic performance in order to avoid over-loading of components and
undesired tripping. Electric propulsion has been used in a majority of new-buildings since the
80’s. The basic configuration can be similar as for service vessels, with a redundant power gen-
eration and distribution system, although there will normally not be any DP requirement for
the icebreakers.
The installed propulsion power may be in the range of 5-15  , depending on ice breaking
capability, Cegelec [44], Hill et al. [117] and MacDonald [181].

3.2.5 War Ships

Despite the great interest in the application of electric propulsion to warships, there are quite
few conventional surface warships with pure electric propulsion, but more are being projected.
For sub-marines, electric propulsion with diesel engine generation and battery storage, fuel cell
or nuclear power plant is applied.

52
 Figure 3.11: Some Offshore and Construction Vessels

Figure 3.12: “M/S Botnica”, Icebreaker which serves as a supply vessel in summer season,
equipped with Azipod propulsion.

53
 Figure 3.13: Azipods installed on Icebreakers.

Electric propulsion for war ships does not conceptually differ much from the merchandise
vessels, but the solutions may differ since the requirements to availability and redundancy are
normally stricter. Also, the ability to withstand shock and provide low noise signatures are
prerequisites for electric drive when applied to a warship.
Fig. 3.14 shows the K/V Svalbard, a coast guard vessel in service since 2002 for the Nor-
wegian Navy, equipped with dual Azipod propulsion system, and partially fulfilling military
requirements.

3.2.6 Research Vessels

Geo technical research vessels, oceanographic vessels, and fishing research vessels have in com-
mon very strict underwater noise requirements, typically several decades dB below normal levels
for other applications.
This has traditionally been achieved by use of direct propulsion with  motors, special
considerations for filtering and reduction of vibrations and torque variations.
By use of modern frequency converters and filtering techniques, it is likely that  motors
will be feasible for such high demanding applications as well.

3.2.7 Trends and New Applications

Electric propulsion is continuously being investigated and evaluated for new applications. LNG
and chemical tankers, Ro-Ro vessels, container vessels, fishing vessels are typical examples of
large volume markets where electric propulsion yet is not taken into use because of the increased
investment costs.
However, only small changes in operation and design criteria, such as increased fuel or
emission costs, regulatory restrictions, and equipment cost reduction, may give a tremendous
shift in technology application for several of new areas. Figure 3.15 shows the world new build
market of ships.

54
Figure 3.14: "K/V Svalbard", Ice breaking coast guard vessel with podded propulsion for the
Norwegian Navy

Figure 3.15: The world new build market of ships. Electric propulsion dominates in the sectors
of construction, cable and pipe layers, offshore, icebreakers, and passenger vessels. In other
segments, electric propulsion is less in use although there is a significant increase in interest for
conceptual studies and designs.

55
3.3 Overview of Electric Power System
3.3.1 Introduction
The main difference between the marine and a land-based electrical power system is the fact
that the marine power system is an isolated system with short distances from the generated
power to the consumers, in contrast to what is normal in land-based systems where there can be
hundreds of kilometers between the power generation and the load, with long transmission lines
and several voltage transformations between them. The amount of installed power in vessels
may be high and this gives special challenges for the engineering of such systems. High short
circuit levels and forces must be dealt with in a safe manner. The control system in a land-based
electrical power system is divided in several separated sub-systems, while in a vessel; there are
possibilities for much tighter integration and coordination.
The design of power, propulsion and control systems for a vessel have undergone significant
changes and advances over a relatively recent period of time. Because of the rapidly expanding
capabilities of computers, microprocessors and communications networks, the integration of
systems which were traditionally separate, stand alone systems is now not only feasible, but fast
becoming industry standards. The increasing demand for redundant propulsion and DP class 2
and class 3 vessels, requires system redundancy with physical separation. The interconnections of
the diverse systems on a vessel have become increasingly complex, making the design, engineering
and building of a vessel a more integrated effort.
Fig. 3.16 shows the schematics of the main power installations in a vessel with electric
propulsion in a Single Line Diagram (SLD). This chapter describes the main components as
they are applied in a marine electric installation:

• Electric Power Generation

• Electric Power Distribution
• Variable Speed
• Propulsion / Thruster units

3.3.2 Electric Power Generation

Prime Mover
The source for power is most often a generator set driven by a combustion engine which is
fueled with diesel or heavy fuel oil. Occasionally one can find gas engines, and also gas turbines,
steam turbines or combined cycle turbines, especially for higher power levels, in light high-speed
vessels, or where gas is a cheap alternative (e.g. waste product in oil production, boil-off in LNG
carriers, etc.).
In a diesel-electric propulsion system, the diesel engines are normally medium to high-speed
engines, with lower weight and costs than similar rated low speed engines that are used for direct
mechanical propulsion. Availability to the power plant is of high concern, and in a diesel electric
system with a number of diesel engines in a redundant network; this means high reliability but
also sophisticated diagnostics and short repair times.
The combustion engines are continuously being developed for higher efficiency and reduced
emissions, and at present, a medium speed diesel engine has a fuel consumption of less than

56
Figure 3.16: Single line diagram for a shuttle tanker with podded electric propulsion. G1-
G4: Generators, SWBD: Switchboard, TRANSF: Transformer, BT: Bow Thruster, AZ THR:
Azimuthing thruster, AZIPOD° R
: Podded propulsion.

200g per produced   at the optimum operation point as seen in Fig. 3.17a). Even though
this is regarded to be a high utilization factor of fuel, it represents only about 40% of the energy
in the fuel, the rest of the energy being removed by the exhaust or heat dissipation.
Moreover, the efficiency drops fast as the load becomes lower than 50% of MCR (Main
Continuous Rating). At this working condition, the combustion is inefficient, with high NOx
and SOx content, and with a high degree of sooting (carbon deposits) which increases the need
for maintenance. In a diesel electric system with several diesel engines it is hence an aim to
keep the diesel engines loaded at their optimum operating conditions by starting and stopping
generator sets dependent on the load, as seen in Fig 3.17b), with an aim to keep the average
loading of each running diesel engine closest possible to its optimum load point.
For detailed description of design and functionality of diesel combustion engines, see Mahon
[178].

Generators
The majority of new buildings and all commercial vessels have an  power generation plant
with  distribution. The generators are synchronous machines, with a magnetizing winding
on the rotor carrying a  current, and a three-phase stator winding where the magnetic field
from the rotor current induces a three-phase sinusoidal voltage when the rotor is rotated by the
prime mover. The frequency  [] of the induced voltages is proportional to the rotational
speed  [] and the pole number  in the synchronous machine
 
=  (3.2)
2 60

57
 350  [%]
50
300 45
Optimal operation area

Efficiency from diesel to propeller

250 diesel electric propulsion
40
Optimal operation area
200 mechanical drive
[g/kWh]

150 Diesel-electric propulsion

four prime movers
25
100 Mechanical propulsion
one prime mover
50

0
0% 20 % 40 % 60 % 80 % 100 % 120 %
10 P [%]
[%MCR]
15 50 60 100 [MCR]

Figure 3.17: a) (left) Example fuel consumption for a medium speed diesel engine. b) (right)
Total efficiency from engine to propeller shaft, in a single machine direct mechanical propulsion
system and a four machine diesel electric propulsion system.

A two-pole generator will give 60  at 3600 , a four-pole at 1800 , and a six-pole at
1200 , etc. 50  is obtained at 3000 , 1500 , 1000  for two-, four-, and six-pole
machines. A large medium speed engine will normally work at 720  for 60  network (10
pole generator) or 750  for 50  networks (8 pole generator).
The  current was earlier transferred to the magnetizing windings on the rotor by brushes
and slip rings. Modern generators are equipped with brushless excitation for reduced mainte-
nance and downtime. The brush-less excitation machine is an inverse synchronous machine with
 magnetization of the stator and rotating three-phase windings and a rotating diode rectifier.
The rectified current is then feeding the magnetization windings.
The excitation is controlled by an automatic voltage regulator (AVR), which senses the ter-
minal voltage of the generator and compares it with a reference value. Simplified, the controller
has PID characteristics, with stationary limited integration effect that gives a voltage drop de-
pending on the load of the generator. The voltage drop ensures equal distribution of reactive
power in parallel-connected generators. According to most applicable regulations, the stationary
voltage variation on the generator terminals shall not exceed ±25% of nominal voltage. Also,
the largest transient load variation shall not give voltage variation exceeding -15% or +20%
of the nominal voltage unless other has been specified and accounted for in the overall system
design. In order to obtain this transient requirement, the AVR is normally also equipped with
a feed-forward control function based on measuring the stator current.
In addition to the magnetizing winding, the rotor is also equipped with a damper winding
which consists of axial copper bars threaded through the outer periphery of the rotor poles, and
short circuited by a copper ring in both ends. The main purpose of this winding is to introduce
an electromagnetic damping to the stator and rotor dynamics. A synchronous machine without
damper winding is inherently without damping and would give large oscillations in frequency
and load sharing for any variation in the load.
In section 3.5, stationary, transient and sub-transient models will be introduced. Simplified

58
one could say that the flux linkages in the damper winding, which are “trapped” and resist
changes due to being short-circuited, characterize the sub-transient interval. This is observed
as an apparent lower inductance in the generator, which gives a stiffer electric performance
during quick load variations, and helps to reduce transient voltage variations and the voltage
variations due to harmonic distortion in load currents. This effect is only contributing for
dynamic variations faster than characterized by the sub-transient time constant such as the
first period of motor start transients and transformer inrush, and for harmonic distorted load
currents.
Often, the generators are connected to a propulsion engine’s shaft, i.e. a shaft generator.
The shaft generators are in some applications made for two-directional power flow, which means
that it can be run as motor. This principle may be called a PTI-PTO concept (Power take-in
— Power take out). Shaft generators have the disadvantage of forcing the main propeller to
work at fixed speed if the generator output shall have constant frequency. This will reduce the
efficiency of the propeller in low load applications. Static converters may be installed to keep
fixed frequency for variable speed.

3.3.3 Electric Power Distribution

Switchboards
The main (or generator) switchboards are usually distributed or split in two, three, or four
sections, in order to obtain the redundancy requirements of the vessel. According to rules and
regulations for electric propulsion, one shall tolerate the consequences of one section failing, e.g.
due to a short circuit. For strictest redundancy requirements, one shall also tolerate failure due
to fire or flooding, meaning that water and fireproof dividers must be used to segregate the
sections.
In a two-split configuration, with equally shared generator capacity and load on both sides,
the maximum single failure scenario will hence be to loose 50% of generator capacity and loads.
In order to avoid high installation costs, the system will often be split in three of four, which
reduces the required additional installations. Also, change-over switches which ensures that a
generator or a load can be connected to two switchboard sections will have similar cost reducing
effects, e.g. for azimuth thruster in Fig.3.16.
In propulsion mode, the switchboards are normally connected together, which gives the best
flexibility in configuration of the power generation plant. The load transients are distributed on
a large number of diesel-generators, and the most optimal number of units can be connected to
the network.
Another possibility is to sail with independent switchboard parts supplying two or more
independent networks. In this case the ship is often assumed to be virtually blackout proof,
which could be attractive in congested waters. In this operating mode one network including its
connected propulsion units is lost if one switchboard section fails, the other, however, remaining
operable. I practice, there are also other considerations to be made in order to obtain such
independence, especially all auxiliaries, such as lubrication, cooling, and ventilation must be
made independent. Also, loss of propulsion or station keeping power on one part of the system,
will through control systems also have impact on the remaining parts, as the total power or
thrust tends to be kept the same, e.g. for dynamic positioning.
The normal operation in DP vessels, in particular for class 3 operations, is to split the
network in order to be tolerant to failure of one section. However, rules and regulations now

59
allows for operation with closed tie breakers, if the protection circuits are designed to detect and
isolate faulty parts without tripping the healthy parts. The NMD rules (Norwegian Maritime
Directorate) has one of the more stricter practicing of these rules and will normally not accept
connected networks in class 3 operations.
As the installed power increases, the normal load currents and the short circuit currents
will increase. With the physical limitations on handling the thermal and mechanical stresses in
bus bars and the switching capacity of the switchgear, it will be advantageous or necessary to
increase the system voltage and hence reduce the current levels. Medium voltage has become a
necessity to handle the increasing power demand in many applications. Using the IEC voltage
levels the following alternatives are most common selected for the main distribution system,
with application guidelines from NORSOK [207]:

• 11 : Medium voltage generation and distribution. Should be used when total installed
generator capacity exceeds 20  . Should be used for motors from 400 and above.

• 66 : Medium voltage generation and distribution. Should be used when total installed
generator capacity is between 4 − 20  . Should be used for motors from 300 and
above.

• 690 : Low voltage generation and distribution. Should be used when total installed
generator capacity is below 4  . Should be used for consumers below 400 and as
primary voltage for converters for drilling motors.

• For utility distribution lower voltage is used, e.g. 400230 .

In US, or where the ANSI standard applies, several additional voltage levels are recognized,
such as in IEEE [125]: 120 , 208 , 230 , 240 , 380 , 450 , 480 , 600 , 690 , 2400 ,
3300 , 4160 , 6600 , 11000 , and 13800 . The 3300 is also a commonly used system
voltage in IEC applications, even though not recognized in NORSOK [207].
Since the load current and fault current determine the limitation of the equipment, the actual
power limits for each system voltage may deviate from these recommendations. This particularly
applies to systems where a major part of the load is converter loads and does not contribute to
short circuit power. Since these do not contribute to short circuit currents in the distribution
system, it often allows increasing the power limits for the different voltage levels.
Also, the selection of system voltage level can be influenced by other criteria, such as avail-
ability of equipment. A lot of ship equipment is available only in 440V, which means that it
might be difficult to avoid this voltage level in ship applications.
Safety is an issue of concern when yards and ship owners changes from low to higher voltages,
often leading to a misunderstanding effort to keep voltages as low as possible. In the context of
safety, it should be regarded that medium voltage switchboards is designed to prevent personnel
to get contact with conductors, even in maintenance of the switchgears. The normal and fault
currents are similarly smaller, giving less forces on the conductors and cables during e.g. short
circuit. Although short circuits inside the switchboards are extremely rare, arc-proof design
(IEC 298-3) is available and will prevent person injury and limit the equipment damages if
worst case should occur.
Circuit breakers are used for connecting and disconnecting generator or load units to the
switchboards, or different parts of the switchboards together. Various circuit breaker technolo-
gies are applied. Air insulated units are the traditional solution, but today rarely applied except

60
at low voltage levels. In the commonly used SF6 and vacuum breaker technologies, the current
interruption takes place in an enclosed chamber, where the first one is filled with SF6 gas, which
has higher insulation strength than air, and the vacuum breaker is evacuated by air. These
designs give compact and long term reliable solutions for medium voltages. One should consider
that vacuum breakers may chop the current and can cause overvoltage spikes when breaking an
inductive loads with high di/dt that may require installation of overvoltage limiters.
For smaller powers, fused contactors are a cost and space beneficial alternative to the circuit
breaker, and are available in air (low voltage), SF6 or vacuum insulated types. The problem
with switching spikes is less with fused contactors since current interruption is softer (lower
di/dt).

Transformers
The purpose of the transformer is to isolate the different parts of the electric power distribution
system into several partitions, normally in order to obtain different voltage levels and sometimes
also for phase shift. Phase shifting transformers can be used to feed frequency converters, e.g.
for variable speed propulsion drives, in order to reduce the injection of distorted currents into
the electric power network by canceling the most dominant harmonic currents. This reduces the
voltage distortion for generators and other consumers. The transformers also have a damping
effect of high frequency conductor emitted noise, especially if the transformer is equipped with
a grounded copper shield between primary and secondary windings.
There are numerous different transformer designs in use, and the most common types are; air
insulated dry type, resin insulated (cast or wound), or oil/fluid insulated. Regulations, ambient
conditions, and user’s, yard’s, or supplier’s preferences govern the selection of type, material,
and design of the transformer.
Physically, the transformer is normally built as three-phase units, with three-phase primary
coils and three-phase secondary coils around a common magnetic core. The magnetic iron
core constitutes a closed path for magnetic flux, normally with three vertical legs and two
horizontal yokes; one in bottom and one at top. The inner winding constitutes the low voltage
or secondary windings, and the outer is the primary or high voltage winding. The ratio of
primary to secondary windings gives the transformation ratio. The coils may be connected as a
Y-connection or ∆-connection (also called D-connection). The connection may be different on
primary and secondary sides, and in such transformers, not only the voltage amplitude will be
converted, but there will also be introduced a phase shift between the primary and secondary
voltages.
A transformer with ∆-connected primary and Y-connected secondary is called a Dy type
transformer. The first and capital letter describes the primary winding, and the second and
small letter describe the secondary winding. The letter n is used to described if the common
point in a Y-connection is grounded, e.g. Dyn or Ynyn.
Transformers may be designed according to IEC standards. For converter transformers, it
is essential that the design accounts for the additional thermal losses due to the high content of
harmonic currents. IEC also gives design rules and guidelines for such applications.

61
3.3.4 Motor Drives for Propulsion and Thrusters
Introduction
The electrical motor is the most commonly used device for conversion from electrical to mechan-
ical power and is used for electric propulsion, thrusters for propulsion or station keeping, and
other on-board loads such as winches, pumps, fans, etc. Typically, 80-90 % of the loads in ship
installations will be some electrical motors.
In this chapter, a brief overview of different motors and their applications in ship installations
is given and for more detailed description of design, performance, and characteristics, references
to other books are made.
The electrical motors in use are:

•  motors: The  motor must be fed from a  supply, and since the power generation
and distribution system normally is a three-phase system, this means that a  motor
must be fed from a thyristor rectifier. This gives also a speed control of the motor. For
detailed description of the various construction of the  motor, see Fitzgerald et al. [77].

• Asynchronous (induction) motors: The asynchronous or induction motor is the workhorse

of the industry. Its rugged and simple design ensures in most cases a long lifetime with
a minimum of breakdown and maintenance. The asynchronous motor is used in any
applications, either as a constant speed motor directly connected to the network, or as a
variable speed motor fed from a static frequency converter. Fitzgerald et al. [77] gives a
good explanation on design and operating performance of asynchronous motors.

• Synchronous motors: The synchronous machine is normally not used as a motor in ship
applications, with exception of large propulsion drives, typically 5  directly connected
to propeller shaft, or 8-10  with a gear connection. In power range smaller than this,
the asynchronous motor is normally cost-competitive. The design of a synchronous motor
is similar to that of an synchronous generator. It is normally not used without a frequency
converter supply for variable speed control in ship applications. See Fitzgerald et al. [77]
for design and operating performance.

• Permanent magnet synchronous motors: Permanent magnet synchronous motors are used
in industrial drives for some few  drives, also for direct on-line applications. In recent
years, it has been introduced also for large power applications; in several   propulsion
drives, firstly in navy applications but now also in podded propulsion applications. The
benefit of this design is high efficiency with compact design, making it especially interesting
for podded propulsion where the dimensions should be as small as possible, and direct
water cooling would eliminate the need for air cooling of the pod motor and simplify
the construction and installation work. See Ådnanes [334] for description of design and
performance.

• Other motors: A range of other motors is used in commercial or experimental applications.

Few of them have gained a high market share, and especially not in marine applications. It
might in future be seen some new concepts for variable speed drives, based on motor designs
with higher efficiency, small dimensions, or specially designed for certain applications.
They will most likely though be based on the principles described above or derivatives of
these.

62
The constant speed, Direct-on-Line motor
An electric motor can be directly connected to the network, and such direct-on-line (DOL)
motors are normally three-phase asynchronous, or induction motors. The asynchronous motor
has a rugged and simple design, where the three-phase stator windings are similar to a generator
stator winding. The rotor is cylindrical, with a laminated iron core and a short circuited winding
similar to the damper winding in a synchronous machine. At no-load, the voltages imposed to
the stator winding will set up a magnetic field in the motor, which crosses the air gap and rotates
with a speed given by the frequency of the imposed voltages, called synchronous frequency,  .
The synchronous speed  in  is hence
 60
 =   (3.3)
2

As the shaft gets loaded, the rotor speed will decrease, and there will be induced currents in the
rotor winding since they are rotating relatively to the synchronous rotating magnetic field from
the stator windings. One defines the slip, , as the relative lag of motor speed to the synchronous
speed 
 − 
=  (3.4)

Hence the slip varies from 0 (no-load) to 1 (blocked rotor). The slip at rated load is normally
below 005 (5%) for most motor designs, and even lower (2 − 3%) for large motors.
>From the electrical model of the asynchronous motor, a mathematical formula for the
rotor and stator current, shaft torque, and power can be developed as a function of slip. A
complicating factor is that the parameters, especially the rotor parameters, are very dependent
on the slip, i.e. frequency of the rotor currents, and these frequency dependencies must be
regarded in order to obtain accurate results.
Figure 3.18 shows the stator currents and shaft torque for an asynchronous motor which
is connected to a fixed frequency and stiff network, as function of rotor speed or slip. It also
indicates the load curve for a typical CPP thruster application at zero pitch and full pitch.
Start-up of a thruster motor should always be done with zero pitch in order to make sure that
sufficient torque margin is available to secure start-up, and to minimize the starting time.
Under stationary conditions, the motor speed is close to synchronous speed, and the induced
rotor currents are nearly proportional to the slip, and also the shaft torque. From the electrical
model in Figure C.14, one can then derive a simplified expression of the resulting  stator
current,  r
p
 =  2 + 2 = 2
 2  
+  (3.5)


where  is the magnetizing current flowing through the magnetizing inductance  , neglecting
magnetizing losses in  .  is the rotor current referred to stator side, and  is the torque.
A subscript  annotates quantities under nominal (rated) conditions, and simplified one may
express, if neglecting effects of leakage inductance in rotor and stator

 =  cos   (3.6)

 =  sin   (3.7)

When the slip approaches the peak torque and higher, the assumptions are not valid any longer,
since the effects from neglecting leakage inductances becomes considerable, and the load stator

63
 Stato r cu rren t

T orqu e
L oad cu rves
- fu ll pitch
- zero p itch
n
S peed
ns
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
ns  n
S lip
ns
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Figure 3.18: Load characteristics for a direct on line asynchronous motor with load curves for a
CPP propeller.

current will typically follow a characteristics as shown in Figure 3.18, typically with a quite flat
current amplitude of approximately five times nominal current (locked rotor current).
Due to the high starting current of asynchronous machines, it will often be necessary to
install devices for soft starting. Soft starters typically reduce the locked rotor currents from 5
times to 2-3 times nominal current, and thereby also reduce the voltage drop. Soft starters must
always be adapted to the load characteristics, as their principle is based on reducing the motor
voltage at start-up, and hence reducing the torque capability of the motor. The most commonly
used are:

• Star-delta ( − ∆) coupling
• Autotransformer
• Semiconductor (thyristor) soft starter

Variable speed motor drives and control strategies

The direct on line motor will rotate with a speed directly determined by the network frequency.
For propulsion, thrusters, pumps, winches, etc., there might be significant savings in power or
fuel consumption by reducing the no-load dependent losses in operations. Also, controllability
of the driven load will be greatly enhanced by controlling the speed of the motor. The penalty
is primarily economically, by introducing additional investment costs, and also components that
require maintenance. The additional investment should paid back by reduced operating costs
or increased earnings if the investment shall be justified. With an energy cost of approximately
1  per   for generated power onboard a vessel, one will briefly save 8 − 9  
annually for a 1  average power reduction.

64
 Interface to control systems, commands and monitoring

Controller

Controls signals Measurements

Rectifier Inverter
Motor
(AC to DC) (DC to AC)

Network DC Link Motor supply

Figure 3.19: Schematics of a variable speed drive, showing a frequency converter with DC Link,
typically for VSI and CSI type converters

The most commonly used motor drives are:

• Voltage source inverter (VSI) type converters for  motors, normally asynchronous mo-
tors

• Current source inverter type (CSI) converters for  motors, normally synchronous motors

• Cycloconverters (Cyclo) for  motors, normally for synchronous motors

•  converters, or SCR (Silicon Controlled Rectifier) for  motors

In ships, the most used variable speed drives uses  motors. Most drives, except the cy-
cloconverter, will consist of a rectifier which rectifies the line voltage, and an inverter, which
generates the variable frequency and variable voltage source for the motor. More detailed de-
scription of these concepts will follow in a later section.
A motor controller contains the speed control, and the control of motor currents by con-
trolling the switching elements of the rectifier and/or inverter. An interface to an overriding
control system, vessel management system, maneuvering control, or dynamic position control
is normally required. The motor controller acquires measurement signals and feedback signals
from sensors in the drive, and motor. Typically motor currents, motor speed, and in some cases
temperatures and voltages are measured.
The semiconductor components of the power electronic circuits are either uncontrollable
(diodes) or controllable (thyristors, IGBTs, IGCTs). Fig 3.20 shows a low voltage IGBT module,
containing all switching elements for a 690V inverter module, and Fig. 3.21 a discrete medium
voltage IGCT used in 3300V inverters.
Power electronics is regarded as a separate field of science, and for further studies, Bose [36]
and Mohan et al. [191] can be recommended.

65
Figure 3.20: An IGBT with the encapsulation removed. This consists of several integrated
components in one module.

Figure 3.21: IGCT “hockey puck” opened, showing its building components. The silicon wafer
is on top (middle).

66
 Torque
Quadrant II Quadrant I
Breaking Motoring
Speed<0, Torque >0 Speed>0, Torque >0

P<0 P>0

Speed

P>0 P<0

Quadrant III Quadrant IV

Motoring Breaking
Speed<0, Torque <0 Speed>0, Torque <0

Figure 3.22: The motor drives are often categorized by which quadrant they are designed to
operate in, e.g. whether they are designed to absorb regenerated power from its load. The
quadrants refer to the torque-speed diagram.

A motor can, if designed for it, run in both direction, with either a driving (motoring) or
breaking shaft torque. In order to categorize what conditions the motor drive is designed for,
the quadrant terms are often applied. The quadrants refer the four quadrants of a speed-torque
diagram, as shown in Figure 3.22. The motor is motoring, i.e. running the load with power
input to the load shaft in quadrants I and III. Oppositely, the motor is breaking, i.e. mechanical
power is transferred from the load to the drive, when operating in quadrants II and IV.
The motor drive normally comprises a speed control function, and the output from this
control function can be interpreted as a torque command or reference which is the input to the
motor control algorithms. These algorithms use a more or less advanced motor model to control
the motor currents and voltages by turning on or off the switching elements of the rectifier (if
controllable) and the inverter.
In principle, the controllers will normally have a control block diagram as shown in Figure
3.23. Torque control is achieved by removing the speed control loop, and give torque reference
as a direct input to the motor drive as shown with dashed lines in Figure 3.24. Motor speed is
normally measured, but new motor controllers are equipped with motor speed estimator which
eliminate the need for a dedicated speed sensor for most ship applications.
For most practical reasons, the speed control loop of a motor drive can be regarded as a
PI (or PID) controlled closed loop with an inner closed torque control loop, which for control
purposes be regarded as a first order time lag. For simulations and synthesis of overriding control
loops, the simplified block diagram in Fig 3.24 should be applicable.

67
 T orq ue
reference

Sp eed Spee d T orqu e Motor Frequen cy

limitation contro l lim itation controller Converter
To rqu e Control
Speed
referen ce signals
reference

Flu x
referen ce
PI

Flux Flux Torque

Co ntrol estim ate estimate

Measu rement
and
motor m odel Currents

Speed
feedback Sp eed
M otor

Figure 3.23: A generic and typical control block diagram for a motor drive controller.

Torque
reference

Speed Speed Torque Torque Motor

limitation control limitation control loop inertia
Torque
Speed PI Torque
reference Speed
reference

(1 + Tis) 1 1
Kp
TiS 1 + Ts Js

Load
curve

Figure 3.24: Simplified block diagram for simulations and synthesis of overriding control loops.

68
3.3.5 Propulsion Units
Introduction
This section presents the most commonly used principles of propulsion units in vessels with
electric propulsion. The overview is not complete, since there are also other alternatives, e.g.
water jets, also in use, however only for special and limited applications.

Shaft propulsion
In a diesel-electric power and propulsion system wit shaft propeller, the propellers are normally
driven by variable-speed electric motors. The horizontal motors may be directly connected to
the shaft, which results in a simple and mechanically robust solution, or via a gear coupling,
which allows for increased rotational speed of the motor and results in a much more compact
motor. The disadvantage is increased mechanical complexity and increased mechanical power
losses.
In diesel electric vessels, shaft lines are used in applications typically where the propulsion
power is higher than available for azimuthing thrusters, or where the ability to produce transverse
thrust, e.g. in station keeping and maneuvering is not needed — or can be produced cheaper
by tunnel thrusters. Typically this applies for shuttle tankers, research vessels, larger anchor
handler vessels, cable layers, etc.
The shaft line propulsion will always be combined with rudders, one rudder per propeller.
By use of high-lift rudders, shaft propellers may also be used to provide a certain degree of
transverse thrust. If additional transversal thrust is needed for maneuvering or station keeping,
there will normally be required to install additional tunnel thrusters also in the aft of the vessel.
The propeller is normally speed controlled FPP (Fixed Pitch Propeller) type, which gives a
simple and robust propeller design. In some applications, the propeller may be CPP (Control-
lable Pitch Propeller) type, even if it is speed controlled. To a certain degree, speed and pitch
can be optimized for higher efficiency, and faster response than with only one control parameter.
These benefits normally do not justify the additional investments in order to obtain combined
speed and pitch control.
Fig 3.25 shows some typical drive configurations for shaft line propulsion system. These can
be installed in single shaft propeller designs, or dual shaft designs.

Azimuth thrusters
Azimuth thrusters are thrusters that can be rotated in order to produce thrust in any direction.
The thrust is controlled either by constant speed and CPP design, variable speed FPP design,
or in rare cases with a combination of speed and pitch control. Variable-speed FPP designs
has a significantly simpler mechanical underwater construction with reduced low-thrust losses
compared to constant speed, CPP propellers.
In vessels with strict limitation of in-board height of the thruster room, the electric motor
will normally be horizontal, and the azimuthing thruster will then consist of a Z-type gear
transmission. Due to a simpler construction with less power transmission losses, vertically
mounted motors and L-shaped gear transmission will normally be selected when the height in
the thruster room allows for it.
A limitation of azimuth thrusters is their limited ability for producing thrust at negative
pitch or   , because they are designed and optimized for unidirectional thrust. If they have

69
 PM
PM PM
PM PM
PM

Figure 3.25: Some examples of shaft line drive configurations

a certain degree of negative thrust capability this should be utilized in order to maintain dynamic
thrust capacity without performing continuous azimuth rotation.
The conventional azimuth thruster was earlier used for station keeping and maneuvering,
but has recently also been taken in use as the main propulsion device in vessels with electric
propulsion. In order to improve the hydrodynamics and steering capability that is required for
propulsion, the shape of the thruster has been adapted, such as the “mechanical pod”. This is
an azimuthing thruster, which is powered from an in-board, typically a horizontal motor, and
the mechanical power is then transferred to the propeller with a Z-shaped gear. The underwater
shape is optimized for low hydrodynamic resistance at higher ship velocity, for higher propulsion
efficiency.
Some vendors can supply thruster devices with dual propeller, either on the same shaft, or
with contra-rotating propellers. Contra rotating propeller increases the hydrodynamic efficiency
by utilizing the rotational energy of the jet stream from one propeller, to create thrust from the
other that rotates the opposite direction. Conventional azimuth thrusters are at present (2002)
in use with power ratings up to 6-7   .

Podded propulsion
Like the conventional azimuth thruster, the podded propulsion unit is freely rotate-able and
may produce thrust any direction. The main difference is the integration of the electrical motor
directly to the propeller shaft, inside a sealed pod unit that is submerged under the vessel hull.
The high power pod schematically drawn in Fig 3.26 shows the variable-speed electric motor,
which is located in the sealed and compact pod. The fixed-pitch propeller is mounted directly
on the motor shaft. Since a mechanical gear is avoided, the transmission efficiency is higher
than in an azimuth thruster. The electrical power is transferred to the motor via flexible cabling
or slip rings for 360-degree operation. Since the propeller pitch is fixed and there is no gear
transmission, the mechanical construction has a lower mechanical complexity.

70
 Slipring Unit (Power/Data Transmission)
Hydraulic Steering Unit
Ventilation Unit

Air Cooling

Installation Block

Bearing, Shaft Seals Bearing

FP Propeller Shaft Line Electric Motor

Figure 3.26: Podded propulsion.

The pod can be designed for pushing or pulling operation. Especially the pulling type pod
gives the propeller a near optimum and uniform wake field, which increases the hydrodynamic
efficiency of the propeller and reduces the risk for cavitation, and hence give reduced noise and
vibrations. A podded unit can rotate in both forward and aft directions if the thrust bearings
are designed for it. The propeller is normally optimized for one main thrust direction, giving
some reduced negative thrust capacity, but without the mechanical limitations of the mechanical
thruster.
Podded propulsion units have been in operation in a decade in cruise vessels, icebreakers, ser-
vice vessels and tankers. Recent new-built field support vessels, Fig 3.10, and semi-submersible
drilling units are now also utilizing podded propulsion as station keeping/transit propulsion
thrusters. The system is today available in power ranges from approximately 1   up to at
least 25   . The larger units provide access into the pod for visual inspection.

3.3.6 Trends and New Concepts

Electric power generation
Electric power generation based on rotating prime movers and electric generators, is a mature
technology, but there is continuously being evaluated alternatives to the traditional synchronous
generator. The fuel cell technology is an area of great interest and research effort, mainly in the
automotive industries.
A fuel cell is an electrochemical device that combines a fuel, e.g. hydrogen, and oxygen from
the air to produce electricity, heat and water. Fuel cells operate without combustion, see Fig.
3.27; hence, a hydrogen fuel cell is virtually pollution free. Since the fuel is converted directly to
electricity, a fuel cell can operate at much higher efficiencies than internal combustion engines,

71
 Figure 3.27: Principles of a Hydrogen fuel cell

extracting more electricity from the same amount of fuel. The fuel cell itself has no moving
parts - making it a quiet and reliable source of power. The fuel cell is composed of an anode (a
negative electrode that repels electrons), an electrolyte membrane in the center, and a cathode
(a positive electrode that attracts electrons).
There is a range of various concepts under evaluation, such as; Phosphoric Acid, Proton Ex-
change Membrane or Solid Polymer, Molten Carbonate, Solid Oxide, Alkaline, Direct Methanol
Fuel Cells, Zinc Air Fuel Cells, and Protonic Ceramic Fuel Cells.
Common for any of these alternatives is that the technology at present has a high production
and maintenance cost — with high energy price. The dynamic capability should still be improved,
and with the present low power density of the unit the use is in practice limited to smaller power
ratings. There are reported some experimental vessel designs using fuel cell power generation,
and is commonly assumed and expected that fuel cell will become commercially feasible within
some time, but not in the very close future.
The fuel cell generates  voltage and one could imagine two alternatives for distribution:

•  distribution and  load — most of the converter principles for propulsion drives are
based on rectifying the  supply before converting to variable frequency. The problem
is mainly related to switching and protection of the distribution system, since interruption
of high  currents is difficult. The conversion to different voltage levels and supply of
consumers that need  supply will also become expensive.  distribution has mainly
been evaluated, and to some extent been used, in military applications.

• - conversion of the power. With this solution, the problems with voltage distribu-
tion and supply will be reduced, but for higher power levels, the solution is yet expensive.

 distribution could also enable the use of compact and lightweight high-speed permanent
magnet exited generators. As the power density of a rotating machine increase by its nominal
rotational speed, there would be significant savings in weight and volume if speed could be

72
increased from typically 500-900   to 15000   . The disadvantage will be a high fre-
quency output with uncontrolled and varying voltage that would require large power conversion
installations.
The fuel cell and power electronics developments open for large shifts in applied technology.
The electric power generation, distribution, and propulsion technology, which has been applied
until recently are lagging state-of-the-art industrial and automotive research and application,
but these technologies will make progress step by step just similar to development of other new
technologies.

Electric power distribution

It is not likely that there will be a dramatic shift in technology in electric power distribution
for  systems. A gradually introduction of new protection relays, programmable and with
field-bus communication is already started and will continue. This means that as flexibility will
increase, the total project costs will be reduced.
There is research and development of power-formers, static power converters that transforms
voltages by use of power electronics. A high frequency transformer can give galvanic insulation,
and the benefits will be reduced dimensions and weights, and elimination of transformer inrush
problems. It is yet work to be done before power-formers will be applied in commercial vessels,
but they will be essential components in order to enable future development of  distribution
systems.

Propulsion
The podded propulsion system is the largest step forward as propulsion unit in the recent years.
The concept is not yet fully developed for all market and power applications, and there will be
a continuous development to be seen. This may have significant impact on vessel design and
allow for total new ship concepts for many areas.
In the CRP concept (Contra Rotating Pod), the podded propulsion is combined with a
traditional shaft line driven propeller is shown in Fig. 3.14. The Pod is azimuthing and controlled
with a variable speed electrical motor, and the shaft line propeller can be either speed controlled
with electrical motor, or a conventional pitch controlled direct diesel driven propeller. The CRP
concept has shown to give a large improvement in propulsion efficiency, as well as increased
redundancy and propulsion power for a several ship types.

3.4 Power and Propulsion Control

3.4.1 Introduction - Control Hierarchy
Fig. 3.29 may illustrate a modern, integrated control, monitoring, and protection system of
a power and propulsion plant system, and the implemented functionality can be described in
control hierarchy of Fig. 3.30.
The user interface with presentation of status and measurements, input of operator com-
mands, alarm handling, etc., is often implemented in operator stations, with a graphical user
interface, push buttons etc. Operator stations are placed on bridge, engine control rooms, etc.
The system level controllers are implemented in control stations or PLCs. They can be
centralized or distributed computers, depending on design philosophy for the vessel. In these

73
 Figure 3.28: The Contra Rotating Pod (CRP) concept.

Figure 3.29: Integrated control system for a vessel, generic configuration.

74
 User Interface
- Op erato r Statio ns

System Level Con trol

- Co ntro llers / P LCs

Low L evel Co ntro l

- Go vernors, A VR, Protection

Figure 3.30: Control hierarchy for power plant.

one will find the energy management functions, such as power management, blackout prevention
functions, start-up and reconfiguration sequence control.
Due to the need for separate testing, response time requirements, and vendor’s responsibility,
there will be a need for low-level fast-response control, monitoring, and protection of devices.
Here are the fast control functions and most basic safety functions implemented. These are
linked to the system control level by hard-wired or field-bus signal interface.

3.4.2 User Interface

The user interface, represented by the bridge consoles and monitors, are the crew’s contact for
monitoring and control of the power and propulsion installations.
It has become more common to utilize GUI (Graphical User Interface) in ship applications,
where light bulbs and push buttons were common few years ago. This allows for much more
flexible and cost-efficient solutions, but also introduce a major challenge for vendors and users
to specify and design a good user interface which combines safety issues with user friendliness,
logical design and easy access to essential and desired information.

3.4.3 High Level Control Functionality

Power management — Energy Management
In a system of electrical power installations, vessel and process automation system, and posi-
tioning system, the various parts of the automation system controls their parts of the power
system, e.g. the dynamic positioning system controls the thruster drives, the off-loading con-
trol system use cargo pump drives, the process control system interacts with compressors and
cooling/heating systems etc. The interconnecting point for all installed power equipment is
the power distribution system. By starting and inrush transients, load variations, and network
disturbances from harmonic effects the load and generators are interacting and influencing each
other. Optimum operation and control of the power system is essential for safe operation with

75
a minimum of fuel consumption. As it is the energy control system (energy, and power manage-
ment system), which monitors and has the overall control functionality of the power system, it
will be the integrating element in a totally integrated power, automation and positioning system.
The purpose of the Power Management System (PMS) is to ensure that there is sufficient
available power for the actual operating condition. This is obtained by monitoring the load and
status of the generator sets and the power system. If the available power becomes too small,
either due to increased load or fault in a running generator set, the PMS will automatically
start the next generator set in the start sequence. A power management system can also have
extended functionality by monitoring and control of the energy flow in a way that utilizes the
installed and running equipment with optimum fuel efficiency. Such systems can be called
Energy Management System (EMS).
For PMS and EMS systems, the main functions can be grouped in:

• Power generation management: Overall control with frequency and voltage monitoring
with active and passive load sharing monitoring and possibly control, and load dependent
start and stop of generator sets. Since control logic and interlocking functions are a
significant part of the power system switchboard design, the functionally of these systems
must be coordinated.
• Load management: Load power monitoring and coordinator of power limitation functions
in other systems, load shedding and start interlock of heavy consumers based on available
power monitoring.
• Distribution management: Configuration and sequence control of reconfiguring the power
distribution system. The distribution system should be configured to fit the requirements
in the actual operational mode for the vessel.

The new generation production vessels and also drill ships/rigs have a complex power system
configuration with advanced protection and relaying philosophies. There are close connections
between the functional design and performance of the energy control system (power management
system) and the power protection system functions. It is a challenge for involved parties to obtain
an optimal and functional solution with several suppliers involved and a yard being responsible
for all coordination.
Blackout of the power generating system is the most severe fault that can happen in an
electric propulsion system. Various mechanisms to avoid blackout are linked to the power man-
agement system, such as the auto start/stop functions, reduction of propulsion and other loads,
or shedding of non-critical loads. Figure 3.31 illustrates the coordination diagram for a typical
installation. Normally, the available power will be controlled within the boundaries for auto
start/stop, but if a sudden increase in load, or tripping of a generator set should occur, the
available power can be reduced. By monitoring load balance and/or network frequency, the load
reduction and load shedding functions will then be activated to reduce the loading and safeguard
the power generation until a new generator set is started and connected.
Should a blackout occur, and it does unfortunately happen from time to time, there will
normally be required to have a system for sequence control of start-up and reconfiguration of
the power system. This is implemented at the system control level, and includes sequences
for starting and synchronizing generator sets and loads. There will normally also be a set of
predefined operation modes, e.g. transit mode, station keeping mode, maneuvering mode, etc.
with automatic sequence control for power system reconfiguration.

76
 t

PA,normal
Load
Reduction/
Shedding

Figure 3.31: Coordination diagram of black-out prevention functions.

Vessel management
In the Vessel Management term, it is common to include manual, automatic, and semi-automatic
control of the vessels auxiliaries and helping systems, such as valves, HVAC (Heat, Ventilation,
and Air Conditioning) system, ballast control, cargo control, etc. Also the alarm systems, watch
call system, and safety systems, may be integrated in the Vessel Management system.
Most of the functions are normally monitored and possible to control from the bridge, or
locally from control stations close to the system or engine control room.

Propulsion control and dynamic positioning

The propulsion control systems normally consist of

• Manual Thruster Control (MTC) system providing individual control of the thrusters and
propellers.

• Autosail or auto pilot system performing automatic course keeping and course changing
during transit operations, often in combination with tracking functionality.

And if the vessel is intended for use in station keeping operations, either

• Dynamic Positioning (Dynpos) System providing manually or automatically positioning

by means of proper action of the thruster system, or

• Thruster assisted position mooring (Posmoor ATA) system providing manually (Posmoor
TA) or automatically thruster assistance (Posmoor ATA) for position and heading control
of anchored vessels.

It is not the intention of this system to describe these control systems in detail, for further
studies, Fossen [78] can be recommended.

77
 It should, however, be emphasized, that the propulsion and station keeping control functions
are in most cases very critical for the safety of operations of the vessel. The need for careful con-
sideration of their design is obvious, but one should never underestimate the need for matching
and testing these control functions together with the power / energy management system, and
low level controllers in switchboards, and propulsion drives.

3.4.4 Low Level Control Functionality

One may divide the low-level controllers in protection and control functions. The protection
devices shall monitor the units for faults and from exceeding design constraints.
The low-level controllers are dedicated controllers for the purpose, often integrated with the
equipment.

Engine protection and governor

The engine protection devices prevent and shut down the engine at over-speed, excessive tem-
peratures, loss of lubrication, etc. The engine protection is usually delivered as an integral part
of the engine, by the engine vendor, or partially integrated with vessel management system.
The governor controls the generated frequency by commanding fuel input to the prime mover,
Fig. 3.32. It may be a so-called “speed droop” type, which implies that the steady-state
frequency will drop proportionally with the active load ( ). The speed droop mode is a simple
and robust method for obtaining load sharing between parallel-connected generators. However
the load dependent frequency variations may cause difficulties in synchronizing generators or
different bus sections. The frequency variation may also be undesired from the operation of loads.
The isochronous governor contains a regulator with an integral effect and keeps the frequency
equals to the set point. A signal, either hardwired or by high-speed bus-communication, between
the governors, ensures a proper load sharing between the prime movers. Most governors applied
in electric propulsion plants have both control modes available.

Automatic voltage regulator

The Automatic Voltage Regulator, AVR (see Fig. 3.33), controls the voltage by commanding
magnetizing current to the field winding of the generator. As the governor, the voltage regulator
can be controlled in droop mode, meaning that voltage varies a few percents (±2.5%) with the
load. This gives a robust and simple sharing of reactive load (kVAr), which is a prerequisite
for equal loading of the generators and the voltage variations normally gives no impact on
synchronization and functionality. In some applications, where the voltage variations are not
accepted, an integrating controller will adjust the voltage set point in order to keep the output
voltage constant, but the droop mode control is normally preferred.

Protection relays
The generator protection prevents and disconnects the generator from the switchboard at ex-
cessive currents in the winding, short circuits, ground faults, faulty synchronization, etc. (See
figure 3.34). The generator protection relay is normally an integral part of a generator control
panel in the switchboard. Similarly, all sections and cables of the power system and consumers
are protected in order to avoid over-loading and to reduce the consequences of a failure.

78
Figure 3.32: Governor for a diesel engine, schematics of speed droop and isochronous control
modes.

Automatic Voltage Regulator - AVR Line-voltage

Set-point
6600V PID +2.5% Generator 1
Droop

-2.5%

Paralleling
Generator 2
Magnetizing - Equal magnetization
equipment - Reactive load sharing

Load

Figure 3.33: in order to obtain reactive load sharing.

79
 Amp
max
short circuit
min

overload
normal load

Milliseconds

Figure 3.34: Simplified setting diagram for a short circuit / overload relay, showing amplitude
and time settings

Typically, the protection functions for a feeder to e.g. a motor will contain functions for
overload prevention, isolate or alarm at earth fault, and isolate at short-circuit. For short circuit
and overload protection, the protection should be adjusted for the equipment’s maximum fault
ratings, as well as adapted to protection relays further up or down in the system in order to
secure the selectivity that is required by class society.

Propulsion controller
As earlier explained, the propulsion is normally speed controlled. The propulsion controller
will then keep the reference speed as far possible within the speed and torque limitations, and
dynamic capability.
The propulsion controller will normally be interfaced with the thruster / propulsion system,
the power generation and distribution system and/or power management system, and bridge
control systems including remote control joystick, autosail systems, dynamic positioning, etc.
Since the propulsion power normally constitutes the major part of the total load for the power
plant, it is also essential that the load reduction functions and blackout prevention functions are
highly coordinated with the power plant design and power management functionality. Based on
an overall blackout prevention philosophy, and prioritizing of the different consumers, a complete
coordination program as shown in Fig. 3.31 can be made. Since response time is critical for
blackout prevention, timing of load reduction is essential. Typically, the propulsion controller
will contain three levels of load reduction in the blackout prevention philosophy:

• Maximum load limitation from available power calculation is normally received from the
power management system. This gives a max  loading for the motor drive, depending
on the power allocation and priority used for the design of the power management system.
• A fast acting, event-triggered load reduction, typically a digital signal forcing the motor
drive power to be reduced to a preset reduction ratio, or absolute value. The signal may
come from the power management system.

80
Figure 3.35: Torque, or power control of thrusters and main propulsion not only improve the
dynamic performance of the positioning or sailing control functions, but also stabilizes the power
taken from the network, and reduces the power disturbances compared to pure speed control
and even more compared to pitch control

• A fast acting, frequency triggered load reduction, only depending on a local frequency
measurement at the drive’s line supply. Normally regarded as a last protection against
under frequency trip due to generator overload in the blackout prevention philosophy.
Safety and monitoring functions of propulsion related equipment and auxiliaries are to some
extent required by classification society. Depending on the overall system design, monitoring
and shutdown functions may be included as part of the propulsion control system, or in the
integrated automation system, or in a combination.
There has been shown that utilizing torque control for propulsion and thrusters (Lauvdal
et al. [162]) will have a stabilizing impact on performance and network disturbances in sailing
and station keeping conditions. Figure 3.35 shows how the power drawn from the network will
stabilize with a torque control approach compared to speed control, by reducing the effects from
sea current and wave impact on the propeller’s thrust characteristics.

3.5 Electric Propulsion Drives

3.5.1 Introduction
Variable speed drives has been in industrial use since in many decades, but first at the end
of the 1960’s by use of power semiconductors. At the beginning,  motors where the most
feasible alternative for propulsion control, but during the 1980’s,  motor drives became
industrially available, and commercially competitive. Since then, almost all new deliveries of
electric propulsion are based on one of the  drive topologies.

3.5.2 Variable Speed Motor Drives

The most commonly used converters for motor drives are, in the following order:

81
 • Voltage source inverter (VSI) type converters for  motors, normally asynchronous mo-
tors

• Cycloconverters (Cyclo) for  motors, normally for synchronous motors

• Current source inverter type (CSI) converters for  motors, normally synchronous motors

•  converters, or SCR (Silicon Controlled Rectifier) for  motors

The topic will be approached in opposite order, due to the fact that the  converter is the
simplest and easiest to understand, while the other ones have a more complex configuration, but
building on much of the same building blocks as the  converter,

Full-bridge thyristor rectifiers for  motor drives (SCR)

The most commonly used  motor is the shunt motor, which has separately supplied field
winding and armature (rotor) winding. The armature current is transferred from the stationary
terminals to the rotor by use of brushes connected to the rotating commutator. In practice,
the armature current also flows through some additional stationary windings, which aids the
commutation of current between the segments of the commutator, but this effect is not regarded
here.
In a shunt  motor the induced armature voltage is proportional to the magnetic field and
rotational speed. The magnetic field is a function of the field current, and because of saturation
effects, they are in practice not proportional. However, if neglecting the saturation, the armature
voltage is:
 = Φ( ) ≈ Φ   =    (3.8)
where  is the induced voltage constant,  the magnetization (field) current,  is the rotational
speed,  and  proportional constants and  is the motor flux. The developed torque is
proportional to armature current and magnetic field

 =   Φ( ) ≈   Φ  =     (3.9)

where  is the torque constant and  the armature current.

Since the  motor must be supplied from a  source with limited voltage, field, and
armature currents, the characteristic boundary of operations will be as shown in Fig. 3.36.
The operation is divided in a constant torque region, characterized with a constant field
current, and a field weakening region where the field current is reduced to maintain the maximum
armature voltage level as speed increases. Hence, the maximum torque boundary is in principle
constant in the constant torque region, and inverse proportional to speed in the field-weakening
region.
However, in the lower speed region, the armature current normally must be limited to avoid
burning of the commutator, and in the higher speed region, it must be reduced to avoid flashing
between the segments of the commutator. These limitations are indicated in the diagram.
In the most common high-power applications, a full-bridge thyristor rectifier (Fig 3.37) feeds
the  motor with a controlled armature (rotor winding) current. Similarly, the field winding is
excited with a regulated field current. The torque is controlled accurately and with low ripple if
the armature inductance is high, but this, on the other hand, reduces the dynamic performance
since the time constant of the armature increases. In this topology, the  voltage on the motor

82
 Constant torque Field weakening
region region

Maximum armature current boundary

Maximum torque boundary

Reduced armature
current due to burning of Reduced armature
commutator current due to flashing
commutator
Field (magnetizing) current

Armature voltage

Speed

Figure 3.36: Characteristics maximum operating boundaries for a shunt DC motor

armature windings is controlled by phase shifting the thyristors’ conduction interval by the gate
firing angle . The gate firing angle can in principle be controlled from 0 to 180 degrees, and
the voltage on the armature windings can hence be regulated from +1,35 to —1,35 times the line
voltage. In practice, however, the gate firing angle will not be lower than 15degrees, in order to
ensure controllability of the motor drive also with voltage drops in the network, and limited to
150 degrees to have a so-called commutation margin.
Since the armature current is controlled by use of the firing angle of the thyristor devices,
the  currents will be phase-shifted with respect to fundamental voltage. The phase angle of
the current is almost equal to the gate-firing angle. Since the armature voltage is proportional
to the rotating speed, one see that the phase angle, which is approximately equal to the gate
firing angle, also will be approximately proportional to voltage and hence rotational speed. In
a  motor drive, where the speed is varying from 0 to 100%, the power factor will hence vary
also from 0 to 0.96 (=15degrees). A low power factor increases losses in the generation and
distribution system and more generators may have to run than the active power of the load
apparently would require.
Wear and tear on brushes and commutator is a source of failure and maintenance and also
limits the standstill torque performance. When accounting for this and also for the fact that
the practical limit for  motor drives is 2 − 3  , the application of  thruster drives is
limited, with the exception of retrofits, in which existing installations are reused.

83
 400
t
300
t
cosphi= 0 ... 0.96 constant current 200

100 phas eA
0 phas eB
0 5 10 15 20 phas eC 25
-1 00
PM -2 00

-3 00

-4 00

Figure 3.37: (Left) Full bridge thyristor DC drive (SCR). (Right) DC voltage to motor is
controlled by delaying conduction of the thyristors with the gate firing angle  ( = 30 ).
Average motor voltage is  = 135 cos 

Current source converters

A  current link fed by a thyristor-controlled rectifier and smoothed by an inductor charac-
terizes the current source inverter (CSI), occasionally referred to as a load-commutated inverter
(LCI) or Synchro. This converter is normally used together with a synchronous motor, but
can also, with some modifications, be used to drive an asynchronous motor. The asynchronous
motor variant was more common in the past, but is rarely seen in new installations.
The synchronous motor is similar to the synchronous generator, with rotating field (mag-
netizing) windings and three- or six-phase stator windings. Six-phase stator windings must be
supplied from a double CSI inverter and is used to reduce the torque harmonics on the shaft.
From the network side, the current source inverter is identical to a full bridge thyristor
converter used for  motor drives, and the characteristics towards the network can very much
be considered to be the same. The inverter side, which feeds the motor, has the same topology
as the rectifier, and uses the induced voltages from the motor instead of the network voltage.
The thyristor rectifier results in a speed-dependent varying power factor, which is high (0.9)
at nominal motor speed, and decreasing toward zero for low speeds. The supply current contains
harmonics that must be regarded during the system design and should normally be reduced by
use of a 12 pulse, 6-phase configuration.
The  link current is directed through the motor phases by controlling the thyristors of
the inverter stage. A 6-step current waveform is obtained, resulting in motor harmonics and
torque ripples. The CSI requires a certain counter-induced voltage (EMF) from the motor to
perform commutation. Hence, it is mainly used in synchronous motor drives in which the motor
can be run with capacitive power factor.
At lower speeds, typically below 5-10% of rated speed, the EMF is too low to perform a
natural commutation. In this speed range, the CSI is run in pulsed mode in which the current
is controlled at zero level during commutation of the inverter output stage. Since the current
and hence the torque are forced to zero level, the torque pulsation at the motor shaft is large in
this operational area. The torque ripple and hence shaft vibrations should be carefully regarded
in the propulsion system design to reduce vibrations and acoustic noise. These may have a
detrimental effect on geared thrusters operating in DP mode.
The characteristic operating boundaries are shown in Figure 3.39 for the topology in Figure

84
 t
constant current
t
t
cosphi= 0 ... 0.9
six-step waveform

PM

Figure 3.38: CSI (LCI) Drive

3.38
The CSI is used in large synchronous motor drives; the biggest one supplied is approximately
100  .

Cycloconverters
The cycloconverter (Cyclo) is a direct converter without a  link (see Fig. 3.40). The motor
 voltage is constructed by selecting phase segments of the supply voltage by controlling the
anti-parallel thyristor bridge. A 12-pulse configuration with reduced line harmonic is drawn, but
the cyclo can also be supplied in a 6-pulse configuration. In 6- pulse configuration, the feeding
transformers can be substituted with reactors when the supply voltage matches the inverter
voltage.
The motor voltage is controllable up to about one third of the supply frequency (about
20); thus it is most applicable in direct shaft drives without gear. It has been used for main
propulsion systems, including podded propulsion.
The motor voltage contains a lower level of harmonics than the CSI, and the motor power
factor may be kept high (unity in synchronous motor drives).
The supply power factor is motor voltage-dependent and is about 0.76 in the field weakening
range. The content of line harmonics is speed-dependent and must be carefully regarded in
system design when the motor drive is large compared with the installed power.
The operation boundaries are similar to those found in the CSI type of synchronous motor
drives, except that the low speed limitations are not present, since the commutation takes place
towards the network voltages and not the motor voltages. The cyclo converter has hence been
preferred in applications where low speed operation and performance is essential, especially in
ice breaking or ice going systems, but also in DP and passenger vessel applications where low
speed / maneuvering performance is essential.
The Cyclo is available in a power range of 2 − 30  per drive motor.

85
 Constant torque Field weakening
region region

Maximum stator current boundary

Maximum torque boundary

Restricted operation
start-up only

Armature reaction
limitation
Field (rotor) current

Stator voltage

Speed

Figure 3.39: Characteristics operating boundaries for a CSI (LCI) converter fed synchronous
motor

PM

Figure 3.40: Cycloconverter drive with input and fundamental output waveforms. The output
voltage is constructed by selecting phase segments of the supply voltage.

86
 t t
constant voltage t
cosphi = 0.95 (constant)
near sinusoidal
currents

PM

Figure 3.41: Two-level PWM drive.

Voltage source inverters

The VSI (Voltage Source Inverter) converter is by far the most used frequency converter in
industrial applications. It gives the most flexible, accurate and high performance drive, and
can be used with an asynchronous motor. It can also be used for synchronous and permanent
magnet synchronous machines with much better performance than other alternatives. The
main limitation of this drive topology has been the availability to high power components, and
its competitiveness towards other drive topologies in the high power range. Until recently, the
practical limit for these drives was around 8 −10  , but as new components become available,
this limit tends to be higher, today up to about 25  .
The VSI is characterized by a rectifier, normally an uncontrolled diode rectifier connected
to the network. This rectifies the network voltage, which hence gives a relatively constant 
voltage, which is further smoothed by a capacitor bank in the  link. The capacitor in the
 link also ensures that high frequency switching ripple from the inverter module does not
enter the network. A six-pulse VSI converter fed induction motor drive is drawn in Fig 3.41 and
its characteristic operation boundaries drawn in Fig. 3.42.
The rectifier in Fig. 3.41 represents a six-pulse configuration used where the converter is
directly connected to the network. The dominant harmonic currents are of the 5th, 7th, 11th,
and 13th harmonic order. The harmonic distortion can be further decreased when using 12-pulse
configuration with a dual feeding via a three-coil transformer, hence canceling the 5th and 7th
harmonics. Where a transformer is necessary for voltage adaptation, the 12-pulse configuration

87
Figure 3.42: Characteristic operation boundaries for a six-pulse VSI converter fed induction
motor.

should normally be used. Using PWM drive and 12-pulse configurations, the resulting harmonic
distortion will normally be close to the limits defined by rules and guidelines, but additional
means may be required, e.g. filtering.
The topology in Fig. 3.41 is capable of running the motor in both directions. Due to the
diode rectified power supply, power can however only be taken from the network, not be fed
back to the network during regenerative breaking. The inverter part, feeding the motor, has
the capability to also perform regenerative network. If this should happen in the topology
shown, the  link voltage would increase and the components might suffer from damages due
to overvoltage. All converters have a built in overvoltage protection, that limits the breaking
power if the  link voltage increases above a safety limit.
In order to be able to regenerate power, e.g. due to a crash-stop maneuver by reversing
the propeller speed, as , it is common to build in a transistor controlled resistor bank to the
 link, which is activated before the safety limit for  link overvoltage. The regenerated
power will then be dumped in this resistor. Alternatively, the rectifier can be equipped with
a full bridge thyristor rectifier in anti-parallel with the diode rectifier (see next section) or an
active front end (similar to the inverter module) can be applied as a rectifier unit in order to
feed power into the network.
There are several ways to control the switching elements in order to obtain the desired
voltage output to the motor. The most common methodology is to use the PWM (Pulse Width
Modulation) in some variant. In its most basic version, a three phase PWM voltage is generated
by comparing three sinusoidal reference values to a high frequency triangular signal, as shown
in Fig. 3.43.. While the sinusoidal reference is higher than the triangular signal, the upper

88
 On
Off 0 5 10 15 20 5 10 15 20 5 10 15 20

Upper and lower switching

elements are switched in opposite orders:
- ON: Upper = on, Lower = off
- OFF: Upper = off, Lower = on

PM

b-a

Figure 3.43: Generation of switching pulses to the PWM modulated inverter, and resulting
output voltages to the motor phase and between two lines.

switching element in the inverter leg gets a firing signal; the lower is turned off, and opposite
when the sinusoidal reference signal is lower than the triangular signal. The voltages from
the inverter to the motor terminals follow the same pattern; with instantaneous values equal
positive and negative voltage levels of the  link for positive and negative gate control signal
respectively. The line voltage, which is what influences the motor, is then the difference between
two phase voltages as shown in Fig. 3.43.
As an alternative to such PWM methods, there are vector modulation techniques, and direct
modulation techniques as in the direct torque control (DTC), where the gate firing signals are
generated directly by calculating which of the 8 possible voltage vectors (including two zero
vectors) to be applied to the stator windings.
There are various methods to implement the motor controller, which attempts to create
input to the motor which gives the desired torque:

• Scalar control: Scalar control is the simplest and first applied technology for control of
asynchronous motors. It was possible to implement in analog electronics, which was the

89
 only feasible in the earlier days of motor control. The scalar control is based on the
stationary model of the asynchronous motor, and from this the corresponding voltage
and frequency was calculated which would give the desired torque or speed in the motor.
The disadvantage is that the model is only valid in stationary condition, and that the
model parameters are highly dependent on temperature, frequency, etc. Hence, the scalar
methods have a poor dynamic performance, with a poor utilization of the motor capacity.

• Rotor flux vector control: This methodology was developed in the late 60’s by the German
scientist Blaschke. The method is based on a model of the motor voltage, fluxes, and
currents referred to vectors in a rotating coordinate system. With the coordinates oriented
in synchronism with the rotating flux in the rotor winding, the current vector’s components
are de-coupled in a flux component and a torque component, similar to the  motor’s
field current and armature current. The method requires computer capacity far above
what was available when the control methodology was developed, and this method found
its commercial application in the early to mid 80’s. A disadvantage is still that the model
required to do the vector transformation, contains parameters highly varying, especially
the rotor resistance which depends on temperature. In order to obtain a good dynamic
performance, the rotor resistance should be adapted on-line or the temperature should be
measured. Figure 3.44 shows a schematic diagram for this control scheme.

• Advanced stator vector control: The same de-coupling of flux and torque control can be
achieved by using a model of stator fluxes and currents in a stator oriented coordinate
system. This model can be made independent on the highly varying rotor parameters, but
requires a much higher computer capacity of the controller. By simulation, this methodol-
ogy was known as early as the late 80’s. Then in the mid 90’s the method, also known as
direct torque control (DTC) was made commercially available. The mathematical model
of the asynchronous motor must be solved with typically 40kHz sampling frequency for
accurate control, and will then not be able to estimate the electrical quantities of the mo-
tor, but also the mechanical speed of the motor. This enables the use of tacho-less drives
in most applications, which is regarded as a great enhancement of the reliability of the
system.

Fig 3.45 shows the limitations of the VSI type asynchronous motor drive. The voltage
limitation on stator voltage is given by the maximum output voltage from the inverter by the
given, constant  link voltage. The inverter unit or motor max current gives the limitations
in stator current and torque. Normally, this means that the motor is limiting the current and
torque in continuous load operations, and the inverter current at intermittent load. There is
also another limitation; labeled with “pitching moment limitation” that is a characteristic of the
induction motor itself. It normally occurs at 150 to 200% of nominal speed, which normally is
outside of the operational speed range of a propulsion motor.
The topology in Figure 3.41 is capable of running the motor in both directions. Due to
the diode rectified power supply, power can however only be taken from the network, not be
fed back to the network during regenerative breaking. The inverter part, feeding the motor,
has the capability to also perform regenerative network. If this should happen in the topology
shown, the  link voltage would increase and the components might suffer from damages due
to overvoltage. All converters have a built in overvoltage protection, that limits the breaking
power if the  link voltage increases above a safety limit.

90
 SPEED/ FREQ
CONTROL
TORQUEREF CURRENT REGULATOR

CONTROL
SPEED_REF VOLTAGE REF’S

TORQ_USED_REF CURRENT REGULATOR

IN DQ-FAME AND PWM Sa,Sb,Sc
TORQUE_REF TRANSFORMATION OF SWITCHING
FLUX_USED_REF VOLTAGES TO LOGIC
SPEED_ACT STATOR VOLTAGE
REFERENCES
PI SWFREQ_REF

FLUX_ACT TORQ_ACT
CONTROL_MODE
FLUXREFERENCE
CONTROL

FLUX_REF
MOTORMODEL

CALCULATE
ACTUALVALUES

CURRENT
ESTIMATE
ORPARAMETERS MEASUREMENT

M
SPEED MEASUREMENT

Figure 3.44:

Constant torque Field weakening

region region

Maximum stator current boundary

Maximum torque boundary

Magnetic flux level

Pitching moment
limitation
Stator voltage

Stator frequency
RPM

Figure 3.45: Characteristic maximum operating boundaries for a VSI asynchronous motor drive.

91
 3000 100

2500 80
60
2000
40
1500 20
Thrust Thrust
1000 0
Power Power
-150 % -100 % -50 % -200 % 50 % 100 % 150 %
500
-40
0
-60
-150 % -100 % -50 % 0% 50 % 100 % 150 %
-500 -80
-1000 -100

Figure 3.46: Breaking capability (4-quadrant operation) is required for propulsion drives, where
the crash stop maneuver is accomplished by speed reversal of the propeller.

In order to be able to regenerate power, e.g. due to a crash-stop maneuver by reversing

the propeller speed, as seen in Figure 3.46, it is common to build in a transistor controlled
resistor bank to the  link, which is activated before the safety limit for  link overvoltage.
The regenerated power will then be dumped in this resistor. Alternatively, the rectifier can be
equipped with a full bridge thyristor rectifier in anti-parallel with the diode rectifier (see next
section) or an active front end (similar to the inverter module) can be applied as a rectifier unit
in order to feed power into the network.
Under bollard pull condition the power flow will be positive, i.e. from network to motor,
in stationary. Dynamically, there might be a dynamic breaking moment to stop or reduce the
speed of the propeller, if the speed reduction is faster than the load from the water resistance
may
Figure 3.47 illustrates the four quadrants of operation in a torque-speed diagram. The
topology in Figure 3.41 may only operate in quadrants I and III. With resistor breaking or active
front end, the VSI converter may operate in all four quadrants. This is typically needed for main
propulsion, winches, elevators, etc., while thrusters normally use a two quadrant converter.
The six-pulse converter in Fig.3.41 does not draw a sinusoidal current from the network. In
order to reduce the distortion, it is common to use a 12 pulse configuration as is shown in Fig
3.48. This is fed from a transformer with 30 degrees phase shifted secondaries ( transformer)
and with series or parallel connection of two six-pulse diode rectifiers, the resulting distortion
of the primary currents will be significantly reduced. Similarly, one can also use 18-pulse (three
diode rectifiers and four winding transformer) or 24-pulse (four diode rectifiers and a five-winding
transformer) to further reduce the distortion. A 12-pulse configuration will normally be enough
to bring the distortion down to an acceptable level. Harmonic distortion will be described more
in detail in a later section.
Medium voltage converters are usually a modified version of the one shown in 3.41. Due to
the increased voltages, the  link voltage must be divided on more components in series in
the inverter part, as shown in Fig. 3.49. This shows a three-level converter, since the output
voltage now can be varied between three levels, +, 0, and -, while the two-level converter in Fig
3.41 can only be varied between + and -. The three-level converter will also give lower current
distortion in the feeding currents to the motor for the same switching frequency as a two-level
converter. This means that switching frequency can be kept lower, given lower losses in the
component, with an acceptable current, and hence torque ripple, as shown in Figure 3.50. A

92
 Torque
Quadrant II Quadrant I
Breaking Motoring
Speed<0, Torque >0 Speed>0, Torque >0

P<0 P>0

Speed

P>0 P<0

Quadrant III Quadrant IV

Motoring Breaking
Speed<0, Torque <0 Speed>0, Torque <0

Figure 3.47: The four quadrants of operations with operational limits for VSI converter.

Vdc =
1.35xVll
Vdc =
2x1.35xVll

Ddy Ddy
transformer transformer
Series connection Parallel connection

Figure 3.48: 12 pulse diode rectifiers. Series connection is often used in medium voltage con-
verters, parallel in low voltage converters.

93
 a) b)
Positive Current:
+ + + +

Vdc/2 0 Phase 0 Phase 0 Phase

_ _ _
0
Negative Current:
Vdc/2 + + +

_ 0 Phase 0 Phase 0 Phase

_ _ _

Phase 1 Phase 2 Phase 3

Figure 3.49: a) Three-level inverters are often applied in medium voltage converters, since the
voltage stress on each component will be lower and the harmonic distortion of the motor current
will be lower at the same switching frequency, see Fig 5.13. b) Shows how current may flow to
obtain, from left; + voltage, 0 voltage, and — voltage to the motor.

medium voltage, 12 pulse, three-level VSI drive for induction motor drives with water cooling is
shown in Figure 3.51.

Other converters
In addition to the most used topologies mentioned here, other variants are occasionally also seen.
Examples are CSI with PWM current output, step-wave with multiple transformer output and
extremely low motor voltage distortion, and VSI Pulse Amplitude Modulated (PAM) converters
with a thyristor-controlled  link voltage and a 6-pulse voltage output. The use of these
technologies is limited and normally only seen in special applications.

Comparison of electric motor drive alternatives

The characteristics of a DOL-started fixed-speed CPP unit and a variable speed FPP unit
are compared in Fig. 3.52. Because of a lower power factor and higher starting transients
more diesel-generators should, in general, be connected to the power network with fixed-speed
thrusters than with variable- speed thrusters, and hence with variable speed thruster and propul-
sion drives:

• Average loading will be higher with less running hours.

• Fuel costs, wear and tear, and maintenance will decrease.

• Smaller dimensioning of the power plant may be achieved.

A thruster drive has two prices: one before and one after installation. The equipment cost
of the variable-speed drives will often be higher than that of the constant-speed controllable-
pitch propeller. Maintenance cost and fuel consumption will on the other hand be reduced since
in DP operation the thruster power is usually only partially utilized.

94
 Torque

Line to line voltage

Current

Figure 3.50:

Figure 3.51: A medium voltage, 3300V IGCT VSI drive for induction motor drives (ABB ACC
6000). Cabinets (from left):1: Diode rectifers, 2: Terminal cabinet and control modules, 3:
IGCT inverter module, 4: DC link capacitors, 5: Water cooling unit with heat exchanger
and circulation pumps. The internal cooling water must be de-ionized (non-conducting) since
the components are directly mounted without insulation to the water-cooled heat sinks. In low
voltage drives, the power modules are normally isolated from the cooling heat sink, hence normal
fresh water can be applied.

95
 DOL SCR DC Cyclo- 1 CSI (LCI) 2 VSI PWM 3
asynchrono motor drive converter
us motor +
CPP
Start-up amps Typ. 5 x 0 0 0 0
rated current (transformer (transformer (transformer (transformer
inrush) inrush) inrush) inrush)
Start-up torque transients Typ. 2-3 x 0 0 Up to 50% 0
rated torque of rated
torque
Power consumption, low  15% of 0 0 0 0
thrust nominal
power
Amps at low thrust 45-55% F(torque) F(torque) F(torque) 0
of nominal
Power Factor - full load  0.85 > 0.9 > 0.76 > 0.9 > 0.95
Power factor variation 0.15 .. 0.85 0 .. 0.9 0 .. 0.76 0 .. 0.9 > 0.95
with load (cos (non-linear) (prop. (prop. (prop. ( constant)
speed) speed) speed)
Dynamic response 3-5 sec < 100 ms < 100 ms Slower < 50 ms
(power, torque) (pitch
control)
Torque ripple None Smooth Smooth Pulsating Smooth
Zero-thrust crossing Smooth if Discontinuo Smooth Pulsating Smooth
negative us
thrust
allowed
Efficiency at full load High Lower High High High
Harmonic distortion:
- at low speed /thrust None F(torque) F(torque) F(torque) 0
- at full speed /thrust None F(torque) F(torque) F(torque) F(power)
Short circuit contribution Typ. 5 x No No No No
nominal
power
Motor matching required - Some Some Yes No
Commutator No Yes No No No
(sliprings) (sliprings)
1 With brushed synchronous motor
2 With brushed synchronous motor
3 With cage induction motor

Figure 3.52: Comparisons of drive alternatives. Low thrust equals low speed in variable speed
drives and small pitch in fixed speed drive for CPP propeller.

96
 Market Require- Detailed System
Study ments design eng.

Concept Specifi- Detailed

Purpose
design cations eng.

Figure 3.53: Stages in design and engineering work

3.6 System Design

3.6.1 Introduction
One of the more experienced class surveyor has stated: “Nobody should make the mistake of
assuming that diesel electric propulsion is easy”. On the other side, diesel-electric propulsion
has proven to be reliable and robust when all aspects of design and engineering are sufficiently
carefully examined.
The main difference between the marine power system and a land-based system is the fact
that the marine power system is an isolated system with short distances from the generated power
to the consumers. The amount of installed power gives special challenges for the engineering of
such systems.
The design and engineering phase can simplified be illustrated by Fig. 3.53. Even before the
vessel concept design starts, a market assessment based purpose and requirement specification
for the vessel should be the made as the basis for the design work.
The system design and engineering can be divided in two phases, however, the first phase is
not fully independent on the possibilities and constraints of the second;

1. Design, conceptual and detailed: In these phases, the vessel capability must be specified
and analyzed in order to define the amount of thrust required for sailing, maneuvering,
and station keeping, as applicable for the intended operations. Based on the goals and
objectives for the vessel operations, the type of propulsion and propulsion / thruster units,
their rating and location on the vessel should be determined, as well as the most optimal
configuration and splitting of the power generation and distribution system. The design
phase will result in a set of technical specifications for the vessel, which is the basis for the
further engineering work.
2. Engineering, system and detailed: During these engineering phases, several analytical and
numerical calculations have to be performed in order to achieve safe and reliable operation,
in common described as standard network analysis or electrical power system studies:

• Load flow calculation.

• Short circuit calculations.

97
 • Ground fault calculations.
• Relay coordination study.
• Harmonic analysis.
• Voltage drop calculation of inrush of transformers and starting of motors.

Dependent on system configuration and vessel application the following extended analysis
can also be required, or necessary:
• Transient analysis of network behavior after disturbance, e.g. short circuit.
• Reliability or failure mode analysis.
A thorough and precise work in this phase is essential for safe, reliable, and cost efficient
operations, and flexibility for future upgrades and modifications of the system later during the
life time of the vessel. Such work has been made much more simple and accurate by use of
computer aided engineering and design tools, and there is a whole range of such available and
in daily use, such as (e.g. EDSA, POWER TOOLS, SIMPOW, SIMSEN, etc.).

3.6.2 Life Cycle Cost Assessment of Conceptual Design

The starting point for the conceptual design of the electric power generation and propulsion
system is based on the intended operation and operating profile of the vessel which is the results
of the vessel and hull design work.
If the vessel has a relatively flat operating profile, meaning that propulsion power is close
to constant for most of the life time, e.g. a VLCC tanker in fixed charters, electric propulsion
will normally not be economically feasible unless there are other requirements which makes it
beneficial. Such can be; large power requirements for other processes, high maneuverability,
redundancy, low noise and vibration, etc.
For vessels with more variable operation profile, as shown in Figures 3.54 and 3.55, diesel
electric propulsion may be profitable, by pure fuel and maintenance savings, or in combination
with increased income.
An accurate comparison of different concepts should be accomplished by use of a Life Cycle
Cost assessment (LCC). There are several ways to do this, and NORSOK has suggested a
method, intended for use in offshore industry, but it can also be applied for ships. The main
elements of the LCC are divided in CAPEX (Capital Expenditures) and OPEX (Operation
Expenditures), and are listed in Figure 3.56.
Although a life cycle cost of the complete vessel and its operations ought to be done for a fair
and precise comparison of different concepts, this is too seldom done, mainly because of lack of
competence, lack of reliable data or lack of time and resources at this stage, or a combination. It
is likely to believe that conventional propulsion system is more often selected as a consequence
of this, than opposite.

3.6.3 Standard Network Analysis and Electrical Power System Studies

Load flow calculations
The aim of the load flow and short circuit calculations is to determine whether the thermal and
mechanical stresses on equipment, such as generators, cables, switchboards, and transformers,
is below the maximum design values, under normal as well as contingency conditions.

98
 OPERATING PROFILE OF PANAMAX CRUISE SHIP OPERATION PROFILE

25000
1800

1600
20000
1400

1200 15000
Annual hours

kW
1000
10000
800

600
5000
400

200 0
0 50 100 150 200 250 300 350
0
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 DAYS 1990
knots

Figure 3.54: (Left) Operating profile for a cruise vessel (hours per year) and (Right) for a shuttle
tanker in the spot market (per day over a year).

At quay, 6 % By platform / ROV, 11 %

Standby
40 % Transit, 11-12 knots
40 %

Anchor handling, 6 %

Figure 3.55: Operating profile for a field support vessel.

99
Life Cycle Cost, LCC = CapEx + OpEx
Capital Expenditures Operational Expenditures

Figure 3.56: Elements in NORSOK’s Life Cycle Cost assessment.

100
 Subtransient Transient Stationary
Amp 10 - 50ms 100 - 500ms >1000ms

10 15 DC
x In <200ms

3 x In

Time

Figure 3.57: Typical short circuit current.

It also gives information about the setting of transformer tappings and voltage regulators
that ensures that the voltage levels on the different distribution buses and load terminals are
within permitted stationary deviation limits.
Load flow calculations are performed to find stationary values of loads in the generation and
distribution network, and since the network in ships normally are radial, it is a fairly straight-
forward exercise. It is normally done prior to the short circuit calculations in order to obtain its
starting values.

Short circuit calculations

Short circuit calculations are done to ensure that fault current in short circuits does not exceed
breaker’s and other equipment’s maximum ratings. Under short circuit, the mechanical stresses
on bus-bars and cabling becomes much higher than under normal operation, and class rules and
standards defines certain limits that the equipment shall be designed for.
A typical short circuit current waveform for a generator is shown in Figure 3.57. As seen,
the initial current has a significant  component, which together with the sub-transient short
circuit current may give a high peak value typically in the order of 10 times nominal current
for the generators. The short circuit currents are being reduced, as the  component decays
rapidly, typically with a time constant of 20 − 100. After 300 − 500, the transient short
circuit current is typically reduced to 3 − 5 times nominal generator current, and dependent on
the system design, this is normally the breaking current for a circuit breaker for a branch.
A sustained short circuit current will after a second or more, reach a stationary value, which
according to rules and regulations shall be at least three times nominal current for a generator
in order to detect faults reliably, for clearing faulty branches. The short circuit current is found
by numerical simulation or by analytical methods as defined in IEC.

Ground fault calculations

If one phase is short-circuited to ground, there will flow a ground fault current. Its magnitude
depends heavily on the method of system grounding, which are:

• Ungrounded: Ground fault currents will still flow due to capacitive coupling between

101
 healthy phases and ground, typically in the order of a few amps in a ship installation.
Required for tankers by SOLAS.

• Low resistance grounding: Ground fault currents will flow through the ground fault, with
a low resistance ground resistor limiting the fault current to not less than 100.

• High resistance grounding: Ground fault currents will flow through the ground fault, with
a high resistance ground resistor, limiting the fault current typically less than 20.

• Bolted ground: Create a high ground fault current, in the order of a short circuit current.

• Coil grounding: With a proper tuned coil grounding, the fault current is theoretically
very low. Not commonly used in ships, mainly since the network configuration varies, and
tuning of the coil is impractical.

System grounding in ships are normally either low or high resistance grounding of system
neutral point or, ungrounded. It might also be a combination of these in different parts of the
distribution system.
Ground fault calculations are done to ensure that the fault current is low enough to reduce
the risk of damaging equipment, and to determine the levels needed to adjust ground fault
detection relays.
For some ships (especially tankers), there might be desired or required to operate with
isolated neutral point in order to reduce fault currents to a minimum. Due to system capacitance,
a fault current with a typical magnitude of some Ampere will still flow through the ground fault.
The power system may be allowed to continue operation with such low fault currents until it
is possible to disconnect and repair faulty parts without large disturbances in operations. The
disadvantages with this grounding method is that one can experience high overvoltages due to
resonance in the fault current circuits, and it might also be difficult to identify in which branch
the fault current occurred.
A high impedance neutral point grounding limits the ground fault currents typically to less
than 20. Also with this fault current it might be allowed to continue operation with a fault
for a limited period of time. The grounding resistor will reduce the risk of resonance oscillations
and it is easier to detect and disconnect the faulty branch of the power system. High impedance
neutral point grounding is normally the preferred method in medium voltage systems.
Low impedance neutral point grounding gives high ground fault currents and a ground fault
must be cleared by disconnecting faulty parts immediately (typically  200). Compared to
isolated and high impedance grounded neutral point, this method will reduce the voltage stress
on the healthy phases during the fault.

Relay coordination/selectivity study

Each feeder in and out from a switchboard is equipped with protection relays or fuses for
detection and disconnection at short circuit, sustained overload conditions, and ground fault.
Ground faults with low fault currents may be accepted for continuous operation.
An over-current / short circuit relay is typically adjustable by level of fault currents and
time to disconnect. Fuses have correspondingly a certain current-time characteristic selected for
the application. A normal load condition shall not initiate a disconnection. If the load current
is higher than the defined normal condition, i.e. in overload, the relay starts a time counter, and

102
disconnects after a certain, and preset time delay. If load current is even higher, the relay starts
another time counter, that disconnects the branch after a shorter time period, defined by the
selectivity study in a way that protection devices in lower parts of the system shall be activated
first, and protection in higher parts later.
Adjusting these settings or selecting of fuses shall according to rules and regulations clear
any fault selectively, by disconnecting a minimum of the distribution system limited to the parts
that are directly affected by the fault. Deviations might be necessary, if the consequences are
regarded not critical.

Harmonic distortion
The harmonic distortion level may be significant in electric propulsion systems, as the main loads
usually are variable speed propulsion/thruster drives with frequency converters. Harmonics and
harmonic analysis are described in a separate section.

Voltage drop calculations

At start-up of heavy motor consumers or energizing large transformers, the start-up transient
current may be several times larger than the nominal rated current. For a motor, typically 5-8
times higher, and for a transformer, up to 10-12 times higher.
A typical motor start transient with a direct on-line (DOL) started motor is shown in Figure
3.58. The motor starting current, is similarly to a short circuit, a complex time varying curve
containing a  term, and transient as well as stationary terms during the acceleration time.
The time for acceleration to full speed is determined by the motor rating and the load curve of
the motor.
The associated voltage transient is shown as a nearly instantaneous maximum drop imme-
diately after the motor is connected. Then, the AVR starts to increase magnetization to com-
pensate for the increased load currents of the generators. When the motor reaches the pitching
moment, the stator currents quickly reduce and since the generators now are over-magnetized for
the new load, a certain voltage overshoot occur. High start-up currents may cause a significant
voltage disturbance in the network, and class requirements set a limit for acceptable transient
voltage variations. This limit is typically−15% and +20% (DnV).
There are analytical methods for calculating such voltage drops, though the most accurate
result is found by numerical simulations.
In order to not exceed the required voltage variations, there might be a need to adjust the
characteristics of the generator or the large consumers, or introduce means to reduce the start-
up transients. For motors, such means could be soft-starting devices, star-delta starters, or
auto-transformer start. For transformers, pre-magnetizing could be evaluated. Generators with
low transient reactance will also give a lower transient voltage drop.

3.6.4 Extended Analysis and Studies

Transient analysis
In addition to the fault calculations and voltage drop calculations mentioned, there are occa-
sionally a need for more thorough analysis of the transient behavior of the network during and
after clearing a fault.

103
 Stator current

Amp <200ms 2-10 s

Torque
Load curves
- full pitch
- zero pitch

Speed
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Slip
1 0.9 0.8 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0

Time
V

Time

Figure 3.58: DOL (direct on-line) motor start current and voltage transient.

Typically, such analysis includes voltage and frequency stability (i.e. will voltage and fre-
quency of the generators be re-established after the fault is cleared?) and re-acceleration of motor
loads (will essential motors be able to accelerate without tripping after the fault is cleared?).
Such analysis are extensive and require accurate modeling of the network and regulators for
voltage and frequency in order to be reliable. It is normally only done when regarded necessary,
and in systems with large motors.

Reliability analysis and FMEA

Especially for ships where availability of the power system is essential and with redundancy
requirements, there is normally a request for reliability analysis or Failure Mode and Effects
Analysis (FMEA). These shall identify the consequences of faults in components or system,
their criticality and assess the probability for such events. Its objective is to identify critical
characteristics that can be improved in the design or specially considered for operations.
The reliability analysis is a quantitative approach to find the likelihood for certain fault
scenarios, such as loss of parts of propulsion power, blackout, loss of positioning capability, etc.
It is normally accomplished by use of a fault three analysis and calculation, known from statistics
theories.
The FMEA is a more qualitative approach, focusing on identifying the consequences of certain
fault scenarios, with a qualified description of how such scenarios are detected, and avoided /
compensated for.
Typical elements for each scenario are description of:

1. Initial conditions.
2. Failure mode of components.

104
 3. Effects on system.

4. Consequences (No, low, high, severe, catastrophic).

5. Occurrence (Rarely, seldom, normal, frequent).

6. Detection and corrective actions.

7. Criticality (acceptable, not acceptable).

The result is normally input to an overall FMEA analysis for the plant, documenting the
need for design improvements, operational instructions, etc.

3.7 Harmonic Distortion

When the connected load in a network is not linear, i.e. not draws sinusoidal currents, the load
currents will distort the sinusoidal voltages. This deviation from a sinusoidal voltage or current
wave form is called harmonic distortion.
Distortion of the voltage wave form may lead to:

• Accelerated aging of insulation material: Increased power dissipation (losses) in equipment

connected to the network, such as generators, motors, transformers, cables, etc., from the
harmonic currents, may cause overheating and deterioration of the insulation, and reduced
life time of the equipment.

• Overloading of electronic equipment: Increased load current of electronic equipment that

has been designed for sinusoidal voltage supply, may cause overheating and malfunction
of this equipment.

• Malfunction: The distorted wave form may cause electromagnetic interference or erroneous
measurement signals if the equipment is not designed for the actual distortion. It is
particularly necessary that measurement systems of monitoring and protection devices are
made for true rms measurements in order to function properly.

The harmonic distortion level may be significant in electric propulsion systems, as the main
loads usually are variable speed propulsion/thruster drives with frequency converters.
Rules and regulations normally give guidelines or requirements that limit the harmonic dis-
tortion in a ship network. However, these limitations are not a guarantee for proper functionality.
It is therefore necessary to be able to predict harmonic distortion, evaluate the effects, and per-
form the proper means to manage the voltage distortion, without functional faults over the life
time of the installation.

3.7.1 Harmonics of VSI Converters

Any periodic wave form can, in general, be expressed by a Fourier series, as the infinite series
of sinusoidal components and a  term

() =  + 1 sin  1  + 2 sin (2 1  + 2 ) + 3 sin (31  + 3 ) + 

+ sin (1  +  ) +  (3.10)

105
 6-pulse 12-pulse

Vdc =
1.35xVll

(a) (b)

Figure 3.59: Ideal waveforms in a 6- and 12-pulse VSI converter.

Some of the terms can be zero, such as the  terms in most  applications and the triple
harmonics in symmetric three-phase systems which are isolated from ground.
The frequency converters are inherently non-linear and the currents to a motor drive are
not sinusoidal but distorted by harmonic components of generally any order, but as we will see,
most of the frequency components are zero under ideal conditions.
When analyzing the harmonic distortion of the network supply, one can normally, at least
initially, disregard the motor side behavior, by assuming ideal de-coupling between the network
and motor sides by the  link. In Figure 3.41, a VSI converter with diode rectifier and
smoothing  inductor and capacitor has been shown in a 6-pulse configuration, and a 12-
pulse configuration in Figure 3.48. If the smoothing  components are large, the current wave
forms will approach the ideal shapes as shown in Figure 3.59.
By observation, one can see that the current into the 12 pulse rectifier is equal to the 6-pulse
rectifier, but the phase shift of the  −connected transformer secondary will shift all voltages
and currents by 30 , compared to the −connected secondary. The current waveforms shown,
are those of the transformer windings.
Assuming that converter and transformer are symmetrically designed and output stage of
converter is assumed to be de-coupled from rectifier current, only the characteristic harmonic
components is present in the input currents to the line supply of the frequency converter. For a
6-pulse converter these are

 = 6 ± 1  = 1 2  (3.11)

⇓
 = 5 7 11 13  (3.12)

In a 12-pulse converter, multiples of sixth (±1) harmonics, which are present in the secondary
and tertiary windings of the feeding transformer, will due to the 30 shift be cancelled in the
primary windings and thus the remaining harmonic current components will be of order

 = 12 ± 1  = 1 2  (3.13)

⇓
 = 11 13 23 25  (3.14)

106
 The Total Harmonic Distortion ( ) is a measure of the total content of harmonic com-
ponents in a measured current,  (), or voltage,  ()
s
P
∞
2
=2
 () = 100% ×  (3.15)
1
and s
P
∞
2
=2
 () = 100% ×  (3.16)
1
where 1 , 1 are the fundamental  value of the voltage and current, and  ,  are the 
value of the  harmonic of the voltage (or current).

3.7.2 Harmonics of CSI Converters

For a CSI converter, the characteristic harmonics will be similar to a VSI converter. However,
the de-coupling between the line supply and the motor sides are not as ideal as for the VSI, and
the harmonics of the line side currents are strongly influenced by the motor side harmonics. In
addition to the pure harmonics, a CSI drive also generates non-integer harmonics to the power
network. Non-integer harmonics are interfering components at frequencies that are not exact
multiples of the system frequency.
In an CSI drive these non-integer harmonics are due to the  pulsation frequencies caused
by the machine converter and are therefore synchronous with the motor frequency according to
the following formula
 =  ±   (3.17)
where
 Non-integer harmonic component.
 Characteristic harmonic component from drives (1, 5, 7, 11, 13 etc.).
 Network frequency.
 Pulse-number of the drive.
 Machine frequency.
The amplitude of the non-integer harmonic components are mainly determined by the size
of the  inductor, i.e. the larger inductor the lower amplitudes. Secondly, the amplitudes are
in general much smaller than the integer harmonic components.

3.7.3 Harmonics of Cycloconverters

For Cycloconverters the harmonic component content of the input current will be a function of
both input frequency and output frequency, for a 6-pulse cycloconverter
 = (6 ×  ± 1)  ± (6 × )    = 1 2   = 0 1  (3.18)
where  is input fundamental frequency and  is output fundamental frequency. As seen, also
here there is a rich content of both harmonics and no-integer harmonics in the current, and thus
voltage waveforms.
The amplitude of the non-integer harmonics are normally significantly high, and it is normally
regarded difficult to establish an efficient tuning of a passive filter to reduce the harmonic level
with Cycloconverter loads.

107
3.7.4 Limitations by Classification Societies
The classification societies have quite recently started to define limitation on allowable THD for
voltage waveforms on the switchboards. For example DnV says that for distribution systems
the THDv shall normally not exceed 5 %, unless being documented that affected equipment is
designed and tested to the actual conditions. It is not defined how documentation and testing
shall be done.
A harmonic analysis study is usually required for documenting the harmonic distortion level
and to find dimensioning criteria for generators, transformers, and if necessary, filters for reduc-
tion of harmonic distortion

3.7.5 Harmonics of Ideal 6- and 12-pulse Current Waveforms

For the idealized current waveforms in Figure 3.59 (a), one can establish the harmonic spectrum
by the following relation (since the wave-form is an odd function with average zero)

Z 2
2 2
 = () sin   (3.19)
 
− 2

Using this relation, one find the following spectrum, where ˆ is the amplitude of the current
ˆ
1 = 
2 = 3 = 4 = 0
1
5 = 
5
6 = 0
1
7 = 
7
8 = 9 = 10 = 0
1
11 = 
11
12 = 0
:

That is X1
() = ˆ   = 6 ± 1  = 1 2 3 4  (3.20)



Figure 3.60 shows the result of this series, where the terms up to 37 harmonics are included,
showing how the resulting waveform converges towards the original six-pulse shape.
In the 12 pulse current waveform, the harmonics of order 5, 7, 17, 19, etc., cancel due to
the 30 degree phase shift of the three-winding transformer. These harmonics will flow in the
transformer windings, but with opposite phase, in the secondary windings of the transformer, and
by the summing, they will circulate inside the transformer only, and not flow into the network.
The total harmonic distortion of these current wave forms can be found by the relation given
in (3.15), which yields  () about 30% for the 6-pulse current, and 15% for the 12-pulse
current.

108
 15

h=1
10
h=5
h=7
5 h=11

h=13
0
0
0

8
16

24

32

48

56

64

72

80

88

96

84

60
4

2
4

28
10

11

12

12

13

14

15

16

16

17

19

20

20

21

22

23

24

24

25

26

27

28

29

30

31

32

32

33

34

35
1

3
u (t )  10 sin(1t )
-5
10 10
 sin(51t )  sin( 71t  3 )
5 7
-10
 ...
10 10
 sin(351t )  ...  sin( 371t )
35 37
-15

Figure 3.60: Harmonics up to 37 of a six-pulse current waveform.

25 %

20 %

15 %
Ih(6-p)
Ih(12p)
10 %
6-Pulse 12-Pulse
5%
h  12 xn  1, n  1, 2,...
h  6 xn  1,n  1,2,..
0%  h  11,13, 23,25,...
 h  5,7,11,13,... 51 72 11
3 13
4 17
5 19
6 23
7 25
8 29
9 31 35 12
10 11 37

Figure 3.61: Characteristic harmonics of a six and twelve pulse current wave-form.

109
 10

8 Real
7

6
Ideal
(1/h)
5

0
10

16
18
20

26
28

32

36
38

42
44
46
48
12
14

22
24

30

34

40

50
2
4
6
8

Figure 3.62: Characteristic harmonics of a 12 pulse current wave-form, comparing the real values
of a practical installation with the ideal amplitudes of Figure 3.61.

These are ideal current waveforms. In practice, impedance due to inductance, resistance,
and capacitance alters the current shape, as seen in Figure 3.63. The corresponding harmonic
spectrum that can be measured in a typical installation with VSI converters are shown in Figure
3.62.
It is obvious that the characteristic harmonics are lower than could be expected from ideal
curves, and that non-characteristic harmonics, here 5 , 7 , etc., occur due to non-ideal trans-
formers and uneven load distribution of the parallel connected rectifiers in the 12 pulse converter.

3.7.6 Calculating Harmonic Distortion

Basics
The harmonic currents drawn by a non-linear load from the network will be distributed in
the network and flow through the other equipment in the power network. If regarded to be a
current source of harmonic current components, it is obvious that the harmonic currents will flow
through the paths with lowest impedance for the harmonics. These are normally the running
generators, large motors, or large distribution transformers to other (higher or lower) voltage
levels.
There are two types of simulation tools available: time domain simulation and the more
commonly applied, which calculates in frequency domain. The benefit of the frequency domain
calculation tools is that the time and work for modeling and calculation of large systems is much
shorter than for a time domain simulation. However, the accuracy will normally be lower, since
one has to decide the harmonic content of the load current, which in reality is dependent on the
network configuration and can only be determined by time domain simulation or by equivalent

110
 (a) (b)

Figure 3.63: (a): Example of switchboard voltage waveforms for a system with converters with
diode bridge rectifiers.   is approximately 7%. (b): Example of switchboard voltage wave-
forms for a system with converters with both diode and thyristor bridge rectifiers.   is
approximately 9%.

figures from similar systems. Special considerations should be made for PWM type of controllers
and use of passive filters, where time domain simulations are strongly recommended in order to
obtain results that are necessary for correct design and dimensioning.
The simulation circuits can normally be assumed ideal, with symmetric supply and ne-
glected impedance in switchboards and cables. In practice transformers and converters are not
ideally symmetrical, nor is the network impedance. Further, there must be expected that non-
characteristic current components are present. These effects will usually have a negligible effect
unless resonance frequencies become excited.
Cable and load impedance’s, especially capacitive components may increase the distortion
on low voltage distributions. In network with high voltage distortion one should avoid the use
of tube lighting with capacitive compensators.
An example of waveforms of switchboard voltage and currents to the thruster is calculated
by the time domain analysis program KREAN and shown in the two figures below. Fig. 3.63a
shows a simulation with only thruster drives with diode bridge converters. Figure 3.63b shows
a simulation where thruster drives with diode bridge and drilling drives with thyristor bridge
are running simultaneous.

Frequency domain - harmonic injection

In this method, the nonlinear load is represented by a harmonic current source, injecting har-
monic currents to the network. The network in terms are modeled as a system where its various
parts, generator, cable, transformer, motors, etc., are modeled with an appropriate impedance
model, representing the impedance for the harmonic frequency currents injected by the harmonic
current source.
An example of such models is shown in Figure 3.64, with a harmonic current source repre-
senting the frequency converter, and impedance models for generator, cable, transformers, and
loads, e.g. motors.
By calculating the resulting voltages from the harmonic currents, the harmonic voltages are

111
 Bus 1

Harm load 1
v_sin

Generator 1 v_sin 1m 3m 0.5m Load 1

2m 1m
2m 10m

Cable 1

2m Load 2
10 1m

Bus 2

3m 1m
Load 3
2m 2m 10
1m

Trafo 1

Load 4
2m
10
1m

Figure 3.64: Impedance model of a network used in frequency domain calculation of harmonic
distortion.

found in the branches or points of interest. Summing up these, the harmonic voltage distortion
is finally found.
There are several calculation software programs assisting in building up and calculation of
harmonic distortion in the frequency domain. Building up large networks are quite simple by
library models, and calculation times are short.
The main challenge is to find a good harmonic representation of the converter, especially
when using converter types where the harmonic spectrum is highly dependent on the network,
such as with VSI converters. Library models are not always to be trusted.

Time domain - network simulation

By building up a circuit model of the network, with discrete impedance models, one can perform
a time domain simulation of the system. Initial values of voltages and currents are chosen,
and after some simulation time, the system has stabilized sufficiently to represent stationary
conditions.
By taking one fundamental period of the voltage or current waveform of interest, one can
then perform a Fourier transformation and find the harmonic spectrum at any point or branch
of the system.
A simplified circuit model for the same system as in Figure 3.64 is shown in Figure 3.65. It is
quite obvious that a complex network is cumbersome to model and time consuming to simulate.
Time step in the simulation must also be relatively short in order to give accurate results.
The great benefit is, that this model gives an accurate calculation of the voltages and currents,
and also the harmonic spectrum of the nonlinear loads.

112
 L R
E

L R
E
L R
E

L R L R
E E L R
E

L R
E
L R
E
L R

L R
E L R

L R

R
1

Ddy

Figure 3.65: Simplified circuit model of a network used in time domain simulation of harmonic
distortion.

Comparison of the frequency and time domain simulation

Frequency domain calculations are widely used because of the simple modeling and short calcu-
lation times. If the harmonic representation of the converter currents is accurate, the results are
also accurate. It is not always straightforward to find the harmonic representation, which may
strongly be influenced by the network characteristics. Then, a time domain simulation can be
used, either for a complete calculation, or for a part of the system that is representative enough
to give a good harmonic model of the converter, and feed this results into a frequency domain
calculation for the complete system.
To illustrate the potential faults, if comparing the results of the time domain simulation in
Figure 3.63 (a) with a corresponding frequency domain calculation of the same system with the
ideal harmonic currents of a 12-pulse rectifier in Figure 3.61, the real voltage distortion gives
  = 8%, while the frequency domain calculation with ideal current waveforms results in
20%. This difference is of course so large that the result from the latter calculation is use-less
for any engineering aspect.

3.7.7 Managing Harmonics

In a vessel with diesel electric propulsion, the frequency converters may constitute up to 80-
90% of the actual load of the generators. Harmonic effects must be considered, and managed, in
order to avoid deterioration and malfunction of equipment, and to meet the rules and regulations’
requirement for harmonic distortion levels.
There are certain engineering aspects that may be used to obtain these objectives, discussed
in the following.

113
Generator impedance
The harmonic currents injected to the power distribution system will mainly follow the lowest
impedance routes, which normally are the generators.
For the frequencies of interest in the harmonic analysis, the generator sub-transient im-
pedance is used. The - and -axis sub-transient inductance are normally different, especially
in a generator with salient poles, and the average of these,  ” and  ” are normally used,
alternatively the negative sequence impedance, − .
A generator with low sub-transient inductance is normally larger than one with larger sub-
transients. Normally, the value is about 20%, while it is relatively achievable to reduce this to
15% or even lower.
Another effect of lowering the sub-transient inductance is that the short circuit current level
increases. One must therefore always make a trade-off with what is desired from harmonic
distortion point of view, the equipment rating for short circuit currents, and the associated
overall costs.

Converter topology
The different converter topologies give different harmonic distortion. Normally the power rating
and application determine the selection, but when possible, a converter with lower harmonic
distortion should be selected to manage the overall distortion levels. Normally, a VSI type
converter gives the lowest distortion in electric propulsion applications.
Increasing the rectifier’s pulse numbers also give a lowering of the harmonic distortion, but
must be trade-off by the associated costs of transformers and converter.
There are also converters with an active front end, constituted by switching elements instead
of diodes. This kind of converters give a much more sinusoidal current shape towards the
network, similar as to the motor. However, the costs of these products are much higher than
with diode rectifiers.

Design of supply transformer

When a transformer feeds the converter, the transformer’s short circuit impedance should be
selected high to smoothen the load current and reduce the harmonic content. It must not
be selected so high that the voltage drop over the transformer at full load reduces the power
capability of the converter below the specified rating. Normally, the short circuit impedance will
be selected between 5 and 8%. A typical distribution transformer will rarely exceed 4%.
Also, the transformer should be equipped with a conductive sheet between the primary
and secondary windings, grounded with an efficient high-frequency ground strap. This will
not influence the lower harmonic transfer from secondary to primary windings, since these are
magnetically coupled. For the very high frequencies, typically above MHz range, the coupling
is more capacitive, and the grounded sheet will act as a screen for capacitive currents, leading
them to ground instead of to the primary. Such screens are normally required to fulfill EMC
regulations (Electro-Magnetic Compatibility), and will also aid as a protection against flashover
from a high voltage primary to the low voltage secondary if the insulation should fail.

114
 Z()
Z
Filter

Lg L
Generator
2 . . f . Lg

C
Resulting

Aggregated First order

Generator / motor model undamped LC filter
5 7 11 13 f / 1f

(a) (b)

Figure 3.66: Passive filter in a network with a generator. () represents the resulting frequency
between two lines, as experienced by a frequency converter connected to the network. (a) circuit
diagram. (b) frequency response.

Passive filters
A passive filter consists of inductances and capacitances, and sometimes also a resistance. Figure
3.66 shows the circuit diagram for a first order LC filter. The impedance in one of these branches
for a certain frequency  is (where  = 2 )
µ ¶
1 1
  () =  + =  1 − 2 (3.21)
  

As seen, the impedance has a series resonance, i.e. a zero impedance frequency, for
r
1
=  (3.22)

For currents with this frequency the impedance through the passive filter approaches zero,
and the filter will ideally draw all currents of this frequency from the network without distortion.
When tuned to the most significant harmonic frequency, the voltage distortion will thereby be
reduced.
If there are several harmonics with significant value, it might be used several parallel con-
nected filters, each tuned for one harmonic frequency.
Connecting the filter to the network, its impedance will come in parallel to the generator im-
pedance. The resulting network impedance will then be a paralleling of the generator impedance
and filter impedance
¡ ¢
  1 − 21
() =  ()| |  () = ¡ ¢ (3.23)
 +  1 − 21
q
1
In addition to the series resonance with a zero impedance at  =  , this also have a
q
1
parallel resonance at  = (1+ ) , meaning that the impedance of harmonics with this

115
 Z
Filter

Z()

Generator
2 . . f . Lg
Lg L L

Resulting
C C

Aggregated First order 5th harm First order 3rd harm

Generator / motor model undamped LC filter undamped LC filter

5 7 11 13 f / 1f

Figure 3.67: Passive filter in a network with a generator. () represents the resulting frequency
between two lines, as experienced by a frequency converter connected to the network. (a) circuit
diagram. (b) frequency response.

frequency approaches infinite high values. If this network is injected by harmonic currents of
this particular frequency, the result can be excessive harmonic distortion and deterioration of
equipment. Parallel resonance will always occur when passive filters are applied, the objective
is to ensure that it will not be excited by any harmonic current.
In practice, both the series and the parallel resonance approach finite values due to damping
effects from resistive components in the network impedance. Figure 3.66 (b) shows a typical
impedance curve for a network with generators and a passive filter tuned to the 5 harmonic.
As seen, the zero impedance coincides with the 5 harmonic, but also have some reducing effects
on the 7 and higher harmonics due to its paralleling of the inductance with the generator.
Also, it is seen that a series resonance occur at about 3 harmonic. This may cause problems
if one expects that the network is subject to third harmonic currents, e.g. from transformer
inrush. This resonance frequency can be shifted by adding a 3 harmonic filter to the network
in parallel to the 5 harmonic network as shown in Figure 3.67, with resulting frequency as
shown.
Passive filters may be an efficient way to reduce harmonic distortion. The design may be
difficult, especially if the network is complicated, in the meaning of many possible configura-
tion alternatives. Paralleling of filter units, minimum and maximum generator configurations,
maximum capacitive loading of the generators, etc. are important aspects in the design. Also,
adding filters to the network will alter the load current wave forms of the converters, and the
filter design will always be an iterative approach before finding the final design.

Active filters
An active filter is a power electronic unit connected to the power distribution with switching
components, such as IGBTs, similar to the inverter stage of a motor drive. This feeds a capacitor
bank, as shown in Figure 3.68.
By use of the switching elements, one can define a shape of the currents which shall flow
from the active filter towards the network. If measuring the load current of a nonlinear load,
e.g. a motor drive, the active filter can then be used to compensate for the harmonics of the
nonlinear load, such that the resulting current of the nonlinear load and the active filter becomes

116
 15

10

0
11

17

23

27

39

79

85
13
15

19
21

25

29
31
33
35
37

41
43
45

65
67
69
71
73
75
77

81
83

87
89
5

51

57

63
3

47
49

53
55

59
61
1

i2
-5

i1
-10
ig
G

-15

ig = i1+i2
HF filter
i1 i2

Active
Non- filter
M linear load

Figure 3.68: An active filter connected to a six-pulse converter, generating the harmonic currents
that compensates for all harmonics of the non-linear converter load. The resulting current flowing
to the network and generator ideally becomes sinusoidal.

sinusoidal. Due to the switching of the power semiconductors, a high frequency filter is necessary
to remove high frequent noise.
Active filtering is an efficient way of removing harmonic distortion. However, the rating of
the filter is relatively high compared to the nonlinear load it is supposed to filter, and the cost
tend to be higher than many other alternatives.

3.8 Example Configurations

Example configurations are found in figures:

• 3.69: Offshore supply/Anchor handling vessel.

• 3.70: Ferry.

• 3.71: Shuttle tanker. The configuration shown here has been installed on several shuttle
tankers and multipurpose shuttle tankers operating in accordance to requirements of the
North Sea, delivered in mid 90’s.Today, a solution with podded propulsion like in Fig. 3.16
would be more likely to build.

• 3.72: Semi-submersible drilling rig.

• 3.73: Floating production vessel.

117
 2500kW 2500kW 2500kW 2500kW

G G G G

3.3kV / 60Hz

M M
f1 f1
f2 f2
G G
f1 f1
f2 f2

400V M M 400V
M M
5000 kW 5000 kW
120 RPM 120 RPM

880kW 740kW
750RPM 750RPM

Figure 3.69: Example of single line diagram of a two-split power plant for a offshore supply vessel
with four diesel-generator sets, two Azipods, one azimuthing thruster and one tunnel thruster.

Vessel Loads

Propulsion Auxilliaries

Figure 3.70: Example of a single line diagram for a ferry.

118
 4000kW 4000kW 4000kW 4000kW

G G G G

11kV / 60Hz

f1 f1 f1 f1 f1 f1
f2 f2 f2 Cycloconverter Cycloconverter f2 f2 f2 11kV/690V
11kV/690V

Pumps M M M M Pumps

PM

1700kW 3000kW 3000kW 1700kW

750RPM 750RPM PM 750RPM 750RPM

12000kW
90RPM

Figure 3.71: Example of a single line diagram of a two-split power plant for a shuttle tanker with
four diesel-generator sets, one main propulsion unit, two azimuthing thrusters and two tunnel
thrusters.

119
 5200kW 5200kW 5200kW 5200kW

G G G G

11kV / 60Hz

M
f1 f1 f1 11kV/690V f1
f2 11kV/690V f2 f2 f2
5200kW 5200kW 5200kW 5200kW

M M M M

3200kW 3200kW 3200kW 3200kW

740RPM G G 740RPM 740RPM G G 740RPM

5000kVA
11kV/660V

f1 11kV/690V f1 f1 f1
f2 f2 f2 11kV/690V f2

M M M M
M M M M M M

Drilling package -
Mud pumps, draw works, top drive
3200kW 3200kW 3200kW 3200kW
740RPM 740RPM 740RPM 740RPM

Figure 3.72: Example of single line diagram of a four-split power plant for a drilling rig with
eight diesel-generator sets and eight azimuthing thrusters.

120
 6750kW 6750kW 6750kW 6750kW

G G G G

11000V / 50Hz

M M M M M M M

3150kVA 3150kVA f1 f1 3150kVA 3150kVA

6750kW f2 f2
11kV/690V 11kV/690V 11kV/690V 11kV/690V

M 2800kW M 2800kW
800RPM 800RPM

11000V / 50Hz

3150kVA 3150kVA
11kV/690V 11kV/690V

f1
f2

M 2800kW
800RPM

Figure 3.73: Example of a single line diagram of a two-split power plant for a FPSO with five
diesel-generator sets and three azimuthing thrusters.

121
Chapter 4

Computer-Controlled Systems

4.1 Introduction
Today most of the control systems are implemented in computers run by digital processors. For
the applications studied here, computers will operate on continuous-time processes at discrete-
time sampling instants. Therefore, computer-controlled systems can be regarded as discrete
control systems. Discrete systems could in many cases be viewed as approximations of analog
systems. However, it is important to be aware of that computer-controlled systems may also
open up new possibilities that were not possible in analog systems. On the other hand computer-
controlled systems may also introduce deteriorations of the control objective, if not accounting
for the fact that they operate on discrete instants of processes which are time-continuous by
nature. Some important aspects to consider in computer-controlled systems are:

• Signal sampling and reconstruction of time-continuous signals. In computer based control

systems discrete-time approximations of continuous signals have to be done before they can
be processed by the computer. The discrete-time approximation is based on the division
of the time axis into sufficient small time increments. The effect of the discrete-time
approximations on the control system design will be discussed.

• Signal detection and quality checking and handling of multiple signals Each sensor signal
has to be checked for errors subject to certain criteria before processed by the control
system. If several sensors provide measurements of the same state variable, weighting and
voting mechanism must be introduced. Signal detection and handling of multiple signals
will be treated in Chapter 5.

This chapter will present the basic properties of linear system theory and computer-controlled
systems. Most of the theory presented is general and valid for almost all kind of industrial control
problems.
The fundamental theory presented in this chapter is a short extract from well recognized
text books in this field. Fore more thoroughly presentation the reader is referred to Johansson
[138], Egeland [63], Electronics Engineer’s Handbook [61], Oppenheim et al. [215], The Control
Handbook [170], Åstrøm and Wittenmark [333], Dorf and Bishop [59], Balchen et al. [21] and
Chen [47] and [48].
The class of systems studied in this text will have input and output terminals as shown in
Figure 4.1.

122
 u(t) y(t)

u(t) Continuos y(t)

t System t

p(t) p(t)

Sampling
t t

u[k] y[k]

u[k] Discrete y[k]

System
k k

Figure 4.1: Continuous-time and discrete-time systems.

It is assumed that the relation between excitation/cause and response/effect can be formu-
lated by either single-input single-output (SISO or monovariable) systems or by multi-input
multi-output (MIMO or multivariable) systems. The input and output signals of monovariable
systems are denoted by () and (), while multivariable systems will be defined by the input
and output vectors u() and y(). A system is called a continuous-time system, if it accepts
continuous-time signals as its input and generates continuous-time signals as its output. Simi-
larly, a system is a discrete-time system, if the input and output relation is described by discrete
inputs and outputs.
Today computer-controlled systems as shown in Figure 4.2 are widely used in many in-
dustrial control applications including marine control systems. Sensor measurements are often
provided by digital signals in form of serial signals and are thereby discrete. Analog (continuous-
time) measurements will be made discrete by a so-called analog-to-digital (AD) converter before
processed by the computer. The calculated outputs of the controller will be made analog by a
digital-to-analog (D/A) converter before acting the process plant. A computer-controlled system
consists of:

• The real process denoted as process plant.

• Sampler with analog-to-digital (AD) converter is converting the analog output signal of
the real process into finite digital numbers. The resolution is dependent on how many bits
that are used. The sampling process is quantized by the clock.

• Digital-to-analog (DA) converter with hold circuit. In Figure 4.2 a DA converter with
Zero-Order-Hold (ZOH) circuit is used to convert and translate the digital signal into a
continuous-time signal that is applied to the real process.

• Computer with clock and software (SW) for real-time control applications and interrupt
handling. The control algorithms are implemented in a SW program and work with input
data that are quantized in time and in level.

123
 Operator

Output
Controller D/A
converter Process
Plant

A/D
converter

Figure 4.2: Computer-controlled system.

• Communication network. One should notice that in a distributed automation system as

shown in Chapter 2 there are several computers with clocks that need to be synchronized
with hard constraints achieving a real-time control system.
• Operator station with graphical user interface to the operator.
The process of converting a sequence of numbers into a continuous-time signal is called
signal reconstruction. Hence, the calculations in the computer will be on the discretized signals.
The discrete systems treated here will be discretized version of physically continuous systems as
shown in the Figure 4.2. Sampling means that a continuous-time signal is replaced by a sequence
of numbers representing the signal at certain times. Let  = { −2 −1 0 1 2 } define the
positive and negative integers, and let { :  ∈ } be a subset of the real numbers called the
sampling instants. Then, the sampled version of the signal () is the sequence
{( ) :  ∈ }  (4.1)
For the systems studied here negative integers of  will not be considered. Sampling is a linear
operation and the sampling instants are often equally spaced in time such that
 =  (4.2)
where  is the sampling period or the sampling time. The discrete-time model between the
input sequence {( )} and the output sequence {( )} is called a stroboscopic model since it
gives the relationship between the system variables at the sampling instant only.

4.2 Basics in Linear System Theory

4.2.1 Causality and State Definition
We will in this text only consider systems that current output depends on past and current
inputs, hence, not future inputs. Then the system is said to be causal or nonanticipatory

124
system. A noncausal system can predict what will be applied in the future by the fact that
current output will depend on future input. This capability is not possible for physical systems,
which then are said to be causal.
By the definition of state vector denoted by x(), a causal system’s output can be properly
defined for inputs going back to −∞ to the current time .

Definition 4.1 Chen [47]. The state x( ) of a system at time  is the information at  that,
together with the input u(), for  ≥  , determines uniquely the output y() for all  ≥  

By definition, by knowing the state x() at  there is no need to know the input u() applied
before  in determining the output y() after  . The state summarize the effect of the past
input on the future output or represents the memory that the dynamic system has of its past.
Furthermore, a dynamic system is said to be lumped, if the state vector has finite number of
state variables. A dynamic system is called a distributed system, if its state vector has infinitely
many state variables.

4.2.2 Continuous-time State Space Model

A general dynamic continuous-time nonlinear process plant may be described by the state space
model consisting of a finite number  of coupled first-order differential equations according to

ẋ = f ( x u w) (4.3)

y = h( x u v) (4.4)

where ẋ denotes the time derivatives of x with respect to the time variable . The functions
f ∈ R and h ∈ R are assumed to be smooth. The states variables are defined by the -
dimensional vector x, u is the control input vector of dimension , w is the -dimensional process
disturbances vector, y the output vector of dimension  comprises variables of particular interest
in the analyses of the dynamic system, and v is the -dimensional noise vector. Here, the output
vector y is regarded as physically measured variables.
Eq. (4.3) and (4.4) may be further linearized about the vector (operation points) x . Then,
for linear time-variant (LTV) systems (4.3) and (4.4) takes the form

ẋ = A()x + B()u + E()w (4.5)

y = C()x + D()u + v (4.6)
f f
where A = x |x is the  ×  dimensional system matrix, B = u |x is the  ×  dimensional
input matrix, E = w |x is the  ×  dimensional disturbance matrix, C = h
f
x |x is the  × 
h
dimensional measurement matrix, and D = u |x is an × dimensional matrix. If these matrices
have constant coefficients, the system described in (4.5) and (4.6) is defined to be linear time-
invariant (LTI) and will be written

ẋ = Ax + Bu + Ew (4.7)
y = Cx + Du + v (4.8)

Consider the simplified LTI system where process disturbance and sensor noise are disregarded

ẋ = Ax + Bu (4.9)
y = Cx + Du (4.10)

125
From linear control theory it is well known that the solution of (4.9) is
Z 
x() = A(−0 ) x(0 ) + A(− ) Bu( )  (4.11)
0

Let us define the transition matrix to be Φ()=A . Furthermore, the transition matrix can be
represented by a Taylor series according to
1 2 2 1
Φ()=A = I + A + A  +  + A  +  (4.12)
2! !
This representation will be used later in the text when making an approximated time-discrete
formulation of (4.7) and (4.8).

Definition 4.2 (Chen [47]). The state equations (4.9) and (4.10) are said to be observable if
for any unknown initial state x(0), there exists a finite time 1  0 such that the knowledge of
the input u and the output y over [0 1 ] suffices to determine uniquely the initial state x(0).
Otherwise (4.9) and (4.10) are said to be unobservable.

Theorem 4.1 (Chen [47]) The -dimensional pair (A C) is observable if the  ×  observ-
ability matrix Q has full column rank , where

Q = [C  A C    (A )−1 C ]  (4.13)

Definition 4.3 (Chen [47]). The state equation (4.9) or the pair (A B) is said to be control-
lable if for any initial state x(0) = x0 and any final state x1 , there exists an input that transfers
x0 to x1 in a finite time. Otherwise (4.9) or the pair (A B) is said to be uncontrollable.

Theorem 4.2 (Chen [47]) The -dimensional pair (A B) is controllable if the  ×  control-
lability matrix Q has full row rank , where

Q = [B AB   (A)−1 B] (4.14)

Remark 4.1 In Matlab the controllability and observability matrices can be generated by calling
ctrb and obsv respectively.

Remark 4.2 A sufficient and necessary condition for that the  ×  matrix Q has full rank is
that Q−1 exists.

Remark 4.3 When the vector x consists of  real numbers, it is said to be -dimensional.
Mathematically we can also write x ∈ R . When the matrix A consists of  rows and  columns
of elements of real numbers, it is said to be  ×  dimensional, or simply  × . Mathematically
we can write A ∈ R× . In this text both notations will be used.

Example 4.1 Consider the state equations

⎡ ⎤ ⎡ ⎤
2 1 0 0 1
ẋ = ⎣ 2 0 1 ⎦ x+ ⎣ 1 0 ⎦ u
0 2 −1 0 0
£ ¤
 = 1 0 1 x

126
 1
zn(t), n 
z2(t)

z1(t)

Figure 4.3: Step function.

Controllability and observability can be evaluated by the following Matlab script:

A=[2 1 0; 2 0 1; 0 2 -1]; B=[0 1; 1 0; 0 0]; C=[1 0 1];

CO=CTRB(A,B); OB=OBSV(A,C);
rank(CO); rank(OB);

The state equations are both observable and controllable as Matlab returns:
⎡ ⎤ ⎡ ⎤
0 1 1 2 2 6 1 0 1
 = ⎣ 1 0 0 2 4 4 ⎦   = ⎣ 2 3 −1 ⎦ 
0 0 2 0 −2 4 10 0 4
() = 3 () = 3

4.2.3 Basic Functions

We will here define some of the most used basic functions applied in control system design and
analyses.

Step function
Consider the sequence of functions { ()}where
⎧ 1
⎨ 0   − 2
1 1
 () =  + 12 − 2 ≤  ≤ 2  (4.15)
⎩ 1
1 2  

where  is an integer. As seen in Figure 4.3 for  → ∞ the sequence { ()} converges to the
step function, also called Heaviside function, by
⎧
⎨ 0 0
 () = 12 =0  (4.16)
⎩
1 0

127
 sn(t), n 

s2(t)

s 1(t)

Figure 4.4: Impulse function.

Impulse function
The impulse function can be thought as the derivative of the step function such that

 ()
 = → () (4.17)
 →∞
The unit impulse (), see Figure 4.4, also known as the Dirac function or delta function is
defined to have zero duration and infinite amplitude. The Dirac function is assumed to have the
property
Z =0+
() = 1 (4.18)
=0−
The Dirac impulse function is a mathematical abstraction which cannot be realized in practice
since no physical signal can have infinite amplitude with zero duration.

Ramp and parabolic functions

The unity ramp function, see Figure 4.5, can be regarded as the integral of the step function
½ R
−∞  () = 
0
() =  (4.19)
0 ≤0

The unity parabolic function, see Figure 4.5, can be regarded as the integral of the ramp function
according to ½ R 2 0
() = −∞ () =    (4.20)
0 ≤0

4.2.4 Laplace Transform

The two-sided Laplace transformation  () of a continuous signal  () defined for all time  is
defined as Z ∞
 () = L{ ()} =  ()−  (4.21)
−∞

128
 a) b)

Figure 4.5: (a) Ramp function and (b) parabolic function.

where the argument  =  +  is the complex frequency. The inverse Laplace transform is
defined to be Z +∞
−1 1
 () = L { ()} =  ()  (4.22)
2 −∞
For the case  () only takes values for   0, it is customary to restrict the Laplace transform
to the be one-sided according to
Z ∞
L{ ()} =  () =  ()−  (4.23)
0

The one-sided Laplace transform will be identical to the two-sided only when  () = 0 for  ≤ 0.
In control problems we normally deal with one-sided Laplace transforms.
Tables of the most used Laplace transforms are found in Dorf and Bishop [59], Balchen et
al. [21] or in mathematical text books covering Laplace transforms.

4.2.5 Fourier Transform

A spectrum is defined as the Fourier transform of a signal  () and is given
Z ∞
 () = F{ ()} =  ()−  (4.24)
−∞

The inverse Fourier transform is defined

Z ∞
−1 1
 () = F { ()} =  ()  (4.25)
2 −∞

Remark 4.4 Assume that the Fourier and Laplace transform exists, that is, take finite values
(remain bounded), then the Fourier transform and the Laplace transform coincide for the choice
 = , where  can be regarded as the angular frequency [rad/s]. The major difference between
the Laplace and Fourier transforms is that the Laplace transform is valuable for analysis of
transient behavior, whereas the Fourier transform is mainly applicable to periodic signals.
2 3
Remark 4.5 Alternatively  ,  
 2 , 

2   , 3   could also be considered to be the time
3 

differential operator which is convenient to apply in manipulation of linear differential equations

with constant coefficients. Notice that an operator should not be confused with the complex
variable in the Laplace transformation theory.

129
4.3 Sampler and Zero-Order Hold
A continuous-time signal may be discretized by a sampler. A sampler is basically a switch that
closes every  second for one instant of time. Let () be a continuous signal, where () = 0
for   0. Let the input to the sampler be (), then the output of the sampler is denoted
∗ (). We define  to be the current sample time, where the current value of ∗ () is ( ) By
introducing a periodic impulse train  () and multiply it with the continuous signal, the series
of impulses can be represented by
∗ () = () () (4.26)
where  () is often referred to as the sampling function and is defined to be
∞
X
 () = ( −  )   0 (4.27)
=0

(− ) is the Dirac impulse function at  − . This mechanism is also known as impulse-train
sampling.

Remark 4.6 Notice that multiplying () by a unit impulse samples the value of the signal at
the point at which the impulse is located. Hence, ()( −  ) = ( )( −  )

Remark 4.7  is denoted as the sampling period and   = 2  is the fundamental frequency of
 () denoted as the sampling frequency in .  = 1 is the sampling frequency in 

We can now rewrite (4.26) such that

∞
X
∗
 () = ( ) ( −  ) (4.28)
=0

In control application it is required that the sampled signal will be kept constant over the
sampling interval. This may be done by a ZOH circuit in the digital-to-analog (D/A) converter.
The D/A converter is often installed on separate Input/Output (I/O) boards. Thus a sampler
together with a ZOH is introduced, see Figure 4.6. This will convert the sampled signal ∗ () to
a continuous signal ()The transfer function of the ZOH circuit is given by

1 1 −  (1 − − )
 () = −  =  (4.29)
  
where  is the complex variable known from Laplace transformation theory. The ZOH circuit
holds the value of ( ) constant for  ≤   ( + 1) In Figure 4.6 (b) this is shown for the
case the impulse input ( ) = 1 for  = 0 and ( ) = 0 for  6= 0, so that ∗ () = (0)().

4.4 Discrete-time State Space Model

4.4.1 ZOH Equivalent of Continuous-time State Space Model
Let us define the sampling instants,  =  , as the times when the control input u ∈ R changes.
The zero-order-hold will result in a discontinuous control signal. We will adopt the convention

130
 a)
Zero-order
Sampler hold
r*(t)
r(t) G0(s) p(t)

b) c)
p(t) p(t),
r(t)
6

r(t)
1 4
p(t)
2

T t 2T 4T 6T t

Figure 4.6: (a) Sampler and ZOH circuit, (b) response of a ZOH to an impulse input ( )
which equals 1 when  = 0 and equals 0 when  6= 0, so that ∗ () = (0)(), and (c) response
of a sampler and ZOH for a ramp input () = .

that the control signal is kept constant in the time interval  ≤  ≤ +1 and continuous from
the right. The state x ∈ R at time  of (4.9) is thus given by (Åstrøm and Wittenmark [333])
Z 
A(− ) 0
x() =  x( ) + A(− ) Bu(0 )0 

Z 
0
= A(− ) x( ) + A(− ) 0 Bu( )

Z −
= A(− ) x( ) + A Bu( )  =  − 0
0
= Φ(  )x( ) + ∆(  )u( ) (4.30)
The state vector x() at time  is thus a linear function of x( ) and u( ). If the AD and
DA converters are perfectly synchronized and if the conversion times are negligible, the input
u( ) and output y( ) ∈ R can be regarded as being at the same instants such that the system
equation of the sampled system at the sampling instants is given by the linear difference equation

x(+1 ) = Φ(+1   )x( ) + ∆(+1   )u( ) (4.31)

y( ) = Cx( ) + Du( ) (4.32)
where
Φ(+1   ) = A(+1 − )  (4.33)
Z +1 −
∆(+1   ) = A B (4.34)
0

131
and Φ ∈ R×  ∆ ∈ R×  C ∈ R× and D ∈ R× . The zero-order-hold solution (4.31) and
(4.32) gives the exact values of the state variables because the control input is constant between
the sampling instants. The intersample behavior of the system, that is the state vector between
the sampling instants, can be investigated by using (4.30). For periodic sampling interval,
 =  where  = 0 1 2 3  the model described by (4.31) and (4.32) simplifies to a linear
time-invariant system according to

x[( + 1) ] = Φx[ ] + ∆u[ ] (4.35)

y[ ] = Cx[ ] + Du[ ] (4.36)

where

Φ = A  (4.37)
Z 
∆ = A B (4.38)
0

4.4.2 Discrete-time Approximation

A discrete-time approximation based on Euler’s method for integration of the state space equa-
tions may be found by approximating the derivative with a difference

x( +  ) − x( )
ẋ =  (4.39)

As above, let  be an integer index so that  = 0 1 2 3  If we assume small sampling periods,
a discrete-time approximation of (4.7) may be written

x( +  ) ≈ ( A + I)x()+ Bu()+ Ew() (4.40)

Introducing discrete notation (4.40) is written as

x[( + 1) ] ≈ (I+ A)x[ ]+ Bu[ ]+ Ew[ ] (4.41)

We can now recognize that (I+ A) is the same as the first two terms of (4.12). If we assume
small  , the transition matrix can be approximated by Φ( ) ≈ (I+ A). Similarly, the control
input and disturbance matrices become ∆ =  B and Γ =  E By omitting  in the arguments
such that x() = x( ) , x[], (4.41) can be rewritten

x[ + 1] = Φx[] + ∆u[] + Γw[] (4.42)

and the corresponding discrete-time formulation of (4.8) becomes

y[] = Cx[] + Du[]+v[] (4.43)

Remark 4.8 The control input matrix is often written

∆ = A−1 (A − I)B (4.44)

132
Remark 4.9 In several textbooks and Matlab, the discrete-time state space formulation is writ-
ten using the same matrix symbols as for time-continuous systems. This means that for LTI
systems (4.42) and (4.43) are reformulated according to

x[ + 1] = Ax[] + Bu[] + Ew[] (4.45)

y[] = Cx[] + Du[] + v[] (4.46)

The corresponding discrete-time version for a LTV system is given by

x[ + 1] = A[]x[] + B[]u[] + E[]w[] (4.47)

y[] = C[]x[] + D[]u[] + v[] (4.48)

4.4.3 Solution of Discrete-time System Equation

Consider the following difference equation

x[ + 1] = Φx[] + ∆u[] (4.49)

y[] = Cx[] (4.50)

A solution of (4.49) can be found by assuming that the initial condition x[0 ] and the input
sequence u[] u[0 + 1]  are known. Thus by iteration we have

x[0 + 1] = Φx[0 ] + ∆u[0 ]

x[0 + 2] = Φx[0 + 1] + ∆u[0 + 1]
= Φ2 x[0 ] + Φ∆u[0 ] + ∆u[0 + 1]
x[0 + 3] = Φx[0 + 2] + ∆u[0 + 2]
¡ ¢
= Φ Φ2 x[0 ] + Φ∆u[0 ] + ∆u[0 + 1] +∆u[0 + 2]
= Φ3 x[0 ] + Φ2 ∆u[0 ] + Φ∆u[0 + 1] + ∆u[0 + 2]
·
·
0 −1
X
0 0 0
x[0 +  ] = Φ x[0 ]+ Φ −−1 ∆u[ + 0 ] (4.51)
=0

Let us define  = 0 + 0 and  =  + 0 , then (4.51) can be reformulated

−
X 0 −1

x[] = Φ−0 x[0 ]+ Φ−(+0 )−1 ∆u[ + 0 ]

=0
−1
X
x[] = Φ−0 x[0 ]+ Φ−−1 ∆u[] (4.52)
=0

We can observe that the first part depends on the initial condition, while the second part is a
weighted sum of the input signals. It is also seen that the eigenvalues of Φ obtained from the
characteristic equation
det(I − Φ) =0 (4.53)
will determine the properties of the solution (4.52).

133
Theorem 4.3 Asymptotic stability of linear systems (Åstrøm and Wittenmark [333]). A discrete-
time linear time-invariant system on the form x[ + 1] = Φx[] is asymptotically stable if and
only if all eigenvalues of Φ are strictly inside the unit disk, that is | |  1 for  = 1 2  

Remark 4.10 In general for discrete-time systems the stability is defined for a particular so-
lution and not for the system, and is thus a local concept. However, for linear time invariant
discrete-time systems, stability is also a property of the system.

4.4.4 Controllability and Observability of Discrete-time Systems

Let us define 0 = 0 and  =  in (4.52) such that
−1
X

x[] = Φ x[0]+ Φ−−1 ∆u[]
=0
−1

= Φ x[0] + Φ ∆u[0] + Φ−2 ∆u[1] + Φ−3 ∆u[2] +  + ∆u[ − 1]
= Φ x[0] + Q U (4.54)

where

Q = [∆ Φ∆   (Φ)−1 ∆] (4.55)

  
U = [u [ − 1]   u [0]]  (4.56)

If Q has full row rank , it is possible to find  equations from which the control signal can
be found such that the initial state is transferred to the desired final state [].

Definition 4.4 (Chen [47]) The discrete LTI equations (4.49) and (4.50) with × state Φ and
 ×  control ∆ matrices described by the pair (Φ ∆) are defined to be be controllable if for any
initial state x[0] = 0 and any final state x[] =   there exists an input sequence of finite
length that transfers x[0] to x[] Otherwise the discrete LTI equations (4.49) and (4.50) or the
pair (Φ ∆) are said to be uncontrollable.

Theorem 4.4 (Chen [47]) The -dimensional pair (Φ ∆) is controllable if the  ×  control-
lability matrix Q has full row rank , where

Q = [∆ Φ∆   (Φ)−1 ∆] (4.57)

Chen [47] defines three different controllability definitions:

1. Transfer any state to any other state in finite time, as adopted here.

2. Transfer any state to the zero state in finite time, called controllability to the origin.

3. Transfer the zero state to any state in finite time, called controllability from the origin or,
more often, reachability.

Remark 4.11 In the literature, there are different definitions of controllability and reachability.
In Åstrøm and Wittenmark [333] definition 1 is called reachability, while definition 2 is called
controllability.

134
Remark 4.12 Controllability definitions 1 and 2 are equivalent if Φ is invertible. Definition
2 does not imply definition 1. If Φ x[0] = 0, then the origin of (4.54)will be reached with zero
input, but the system is not necessarily controllable.

Consider (4.49) and (4.50), and for simplicity assume that the control input is equal to zero,
u[] = 0 for all . This gives the following set of equations

y[0] = Cx[0]
y[1] = Cx[1] = CΦx[0]
:
y[ − 1] = CΦ−1 x[0] (4.58)

or on vector form
h i £ ¤
C  (CΦ)   (CΦ−1 ) x[0] = y [0] y [1]  y [ − 1] 
£ ¤
Q x[0] = y [0] y [1]  y [ − 1]  (4.59)

Definition 4.5 (Chen [47]) The discrete LTI equations (4.49) and (4.50) with × state matrix
Φ and  ×  output matrix C described by the pair (Φ C) are defined to be be observable if for
any unknown initial state x[0] there exists a finite integer 1  0 such that the knowledge of the
input sequence u[] and output sequence y[] from  = 0 to 1 suffices to determine uniquely the
initial state x[0]. Otherwise the discrete LTI equations (4.49) and (4.50) or the pair (Φ C) are
said to be unobservable.

Theorem 4.5 (Chen [47])The -dimensional pair (Φ C) is observable if the × observability
matrix Q has full column rank , where

Q = [C  Φ C    (Φ )−1 C ]  (4.60)

Definition 4.6 (Åstrøm and Wittenmark [333]). The discrete LTI equations (4.49) and (4.50)
is detectable if the only unobservable states are such that they decay to the origin. That is, the
corresponding eigenvalues are stable.

4.5 Discrete-time Approximation Methods

A solution of (4.49) was found by iteration and was given in (4.52). An advantage of this method
is that stability problems due to discretization may be avoided providing accurate calculation
of Φ However, the main limitation is that this method in practice is only applicable for linear
systems with constant coefficients. We will here briefly present three alternative numerical
integration schemes often used in control applications. Numerous of other numerical methods
exist, whereof the so-called Runge Kutta methods often are favorable with respect to accuracy,
see Egeland and Gravdal [64] and Cheney and Kincaid [49] for details.
We will here consider a system on the form

ẋ = f ( x u) (4.61)

135
4.5.1 Euler’s Method
The Euler’s method is based on the assumption that the derivative is approximated by the
forward difference according to

x( +  ) − x()
ẋ = = f ( x u) (4.62)

We introduce the discrete notation  =  , then the integration method is written

x[ + 1] = x[] +  f (x[] u[]) (4.63)

Euler’s method is an explicit 1-step method. An explicit difference equation means that x[ + ]
can be recursively calculated, since all terms on the right hand side are calculated in previous
sampling instants, that is for +−1 +−2 .. The method is denoted to be 1-step, since only
values of the previous sampling is included. This method has local interrupt error proportional
to  2  often written ( 2 ) Euler’s method is suitable in simulation of simple systems, where the
requirements to the accuracy is not strict. The accuracy is of course sensitive to the sampling
interval. It can be shown that Euler’s method is stable for systems with its poles  within
the circle |1 +   | = 1 For systems with real poles in the left-hand plane, a requirement for
stability is
2
   (4.64)
|max |

4.5.2 Backward Euler’s Method

The backward Euler’s method is based on the assumption that the derivative is approximated
by the backward difference according to

x() − x( −  )
ẋ = = f ( x u) (4.65)

Using discrete notation
x[] = x[ − 1] +  f (x[] u[]) (4.66)
or
x[ + 1] = x[] +  f (x[ + 1] u[ + 1]) (4.67)
Also this method is having local interrupt error of order ( 2 ). Euler’s method is an implicit
1-step method. An implicit difference equation means that x[ + ] cannot be recursively
calculated, since terms on the right hand side are also given at the time instant  + . The
region of stability for backward Euler’s method is |1−1  |  1

4.5.3 Combined Backward and Forward Euler’s Method

An appropriate method for integration of second order systems with complex conjugated poles
is the combined backward and forward Euler’s method. Consider the following second order
system on phase-variable form

ẋ1 = x2  (4.68)
ẋ2 = f ( x u) (4.69)

136
Let us apply Euler’s method to integrate (4.69) and backwards Euler’s method on (4.68) ac-
cording to

x2 [ + 1] = x2 [] +  f (x1 [] x2 [] u[]) (4.70)

x1 [ + 1] = x1 [] +  x2 [ + 1] (4.71)

We avoid the problem with the implicit equation (4.71) since of x2 [ + 1] is known from the
previous calculation given in (4.70). Hence, we can solve the integration recursively.

4.5.4 Trapezoidal (Tustin’s) Method

The trapezoidal method is an implicit 2-step method and is written

x[ + 1] = x[] + [f (x[] u[]) + f (x[ + 1] u[ + 1] + 1)]  (4.72)
2
Condition for stability is ¯ ¯
¯1 +   ¯
¯ 2 ¯
¯ ¯  1  = 1   (4.73)
¯1 −   ¯
2
where  is the poles of the system. If Re( )  0 then Re(  )  0 provided   0. The
trapezoidal method is stable for all   0 and is having local interrupt error of order ( 3 )

4.5.5 Second Order System

In the following examples we will show how Tustin’s method subject to smaller sampling periods
converge to the real solution, and then a comparison is done for backward Eulers, forward Euler’s
and Tusin’s methods for discretization. Let us consider the following second order system
1
̈ + 2 0 ̇ +  20  =  (4.74)

2 2
where m=10 kg, =0,2,  0 =  = 2 and F=10 N. The eigenvalues are  = -0.6283 ± 3.0781i.

Example 4.2 Let us consider Tustin’s method for discretization of (4.74) with increasing sam-
pling periods from T=0.01s, 0.05s, 0.1s, 0.2s, and 0.5s. As seen in Figure 4.7 for T ≤ 0.1s the
system becomes well approximated. For T ≥ 02 the approximation becomes oscillatory.

Example 4.3 Let us compare backward Eulers, forward Euler’s and Tusin’s methods for dis-
cretization of (4.74) with T= 0.1s. As seen in Figure 4.8 forward Euler is as expected less
accurate.

4.6 Nyquist Frequency

Assume that a continuous signal is band limited, that is the Fourier transform is zero outside a
finite band of frequencies. If the samples of the continuous signal are taken sufficiently close to-
gether in relation to the highest frequency of the band, then the samples will uniquely reconstruct
the signal. Stable continuous LTI systems have the property that the steady-state response to

137
Figure 4.7: Step response of second order system using Tustin’s method with increasing sampling
period.

Figure 4.8: Step response of second order system using backward Euler’s,forward Euler’s and
Tustin’s methods.

138
sinusoidal excitations is sinusoidal with the frequency of the excitation signal (usually with some
phase lag). However, discrete systems will create signals with new frequencies. A discrete control
system operates on process variables at discrete times only. Achieving satisfactory performance
and stability of the discrete control system, appropriate choice of the sampling period is crucial.
The main concern is under what condition will the discrete value recover the continuous signal.
In order to have a minimum fair representation of sinusoidal signals with frequency   , Nyquist
showed that it was necessary to sample at least twice per period,    
Consider the spectrum of the sampled signal ∗ () obtained as the discrete Fourier transform

 () = F{∗ ()} = F {()} ∗ F { ()} = () ∗  () (4.75)

where (∞ )
X ∞
2 X 2
 () = F ( −  ) = ( − ) (4.76)
 
=0 =0
Hence, since convolution with impulses shifts a signal, we have
∞
1X 2
 () = (( − )) (4.77)
 
=0

Hence,  () is a periodic function of the sampling frequency  , see Figure 4.9 (a) and (b). In
Figure 4.9 (c) we can see that if   (  −  ) or   2  , there is no overlap between the
shifted replicas of () and is thus properly recovered by  (). However, if the signal to be
sampled contains frequency components larger than half the sampling frequency    2 , the
sampling process itself will create low frequency components in the sampled signal Sampling of
a signal with frequency  creates signal components with frequencies

 =   ± 

where  is an arbritary integer. The phenomenon that the frequency process creates new fre-
quency components is called aliasing, see Figure 4.9 (d). The frequency that is half of the
sampling frequency   is called the Nyquist frequency
1 
 =  =  (4.78)
2 
To avoid aliasing it is important that all frequency components of signal higher than the
Nyquist frequency are removed before the sampling. This could be done by introducing a
presampling filter or antialiasing filter, which will be of lowpass characteristics. This is often
done by an analog low pass filter in the front of the sampler. The lowpass filter must have a
cut-off frequency equal or less the Nyquist frequency,  ≤   . In Figure 4.10 we can see the
recovery of a time-continuous signal from its samples using ideal lowpass filter.

Theorem 4.6 Shannon’s sampling theorem. A continuous-time signal with a Fourier transform
that is zero outside the interval (−    ) is given uniquely by its value in equidistant points
if the sampling frequency is higher than 2  

Remark 4.13 The Fourier transform  () of a sampled signal is a periodic function of the
original spectrum () along the frequency axis with the period equal to the sampling frequency
 .

139
 a) b)
R(j) P(j)

1 2
T

   s s s 0 s s s 

c) d)
Rd(j)
1 1
T T

  s  s 
s s

Figure 4.9: (a) Power spectrum of continuous-time signal, (b) power spectrum of sampling
function, (c) power spectrum of sampled signal with    2  , and (d) power spectrum of
sampled signal with    2  resulting in aliasing.

P(j)

X(j) Xp(j) H(j) Xr(j)


c c
 M<c<(s-M)  

Retrived
Spectrum Lowpass filter Spectrum

Figure 4.10: Recovery of a time-continuous signal from its samples using ideal lowpass filter.

140
4.7 The -Transform
4.7.1 Definition
Let us define
 =    (4.79)
m
−1
 = −   (4.80)
The -transform maps a semi-infinite time sequence (only a sequence of positive integers) into
a function of a complex variable . Applying Laplace transform on (4.28) we have
∞
X
∗
L{ ()} = ( )−   (4.81)
=0

A new -transform is then defined, so that for a general function () the -transform is found
to be
X∞
∗
{()} = { ()} = () = ( ) −  (4.82)
=0

Tables of the most used -transforms are found in Åstrøm and Wittenmark [333], Dorf and
Bishop [59] and Balchen et al. [21].

Definition 4.7 For a discrete signal  [] a similar transformation as the Laplace transforma-
tion, which is defined as the -transform  () can be found by
∞
X
 () = { []} =  [] −  (4.83)
=0

Example 4.4 Consider the exponential function

 () = −   ≥ 0 (4.84)
The -transform is determined to be
∞
X ∞
X
− − −
¡  ¢−
{ } =  () =   =   (4.85)
=0 =0

which can be rewritten in closed form as

1 
 () = =  (4.86)
1 − ( )−1  − −

Example 4.5 Consider the step function  () given in (4.16). The -transform is determined
to be
X∞
{ ()} =  () =  −  (4.87)
=0
which can be rewritten in closed form as
1 
 () = =  (4.88)
1 −  −1 −1

141
 Stable system in
s-domain

Figure 4.11: Correspondance between stability regions in -domain and -domain.

4.7.2 Stability Properties

Consider  =   , where  =  + . We then find
 =   = (+)  (4.89)
where
|| =   (4.90)
] =  (4.91)
We observe that the left-half plane in the -domain with  ≤ 0 will in the -domain correspond
to
0  || =  ≤ 1 (4.92)
Hence the left-half plane in the the -domain corresponds to the unit circle in the -domain.
Stable roots in the -domain will in the -domain be within the unit circle, see Figure 4.11.

4.8 The Pulse-Transfer Function and the Pulse Response

4.8.1 Pulse-transfer Function
Consider the LTI system on the form
[ + 1] = Φ[] + ∆[] (4.93)
[] = C[] + D[] (4.94)
The -transform gives
∞
Ã∞ ! ∞ ∞
X X X X
 −  [ + 1] =   −  [] −  [0] = Φ −  [] + ∆ −  []  (4.95)
=0 =0 =0 =0

Hence,
 (() −  [0]) = Φ() + ∆ () (4.96)
m
() = (I − Φ)−1 ((0) + ∆ ())  (4.97)

142
and ³ ´
 () = C (I − Φ)−1 (0) + C (I − Φ)−1 ∆ + D  () (4.98)
The pulse-transfer function is defined as

() = C (I − Φ)−1 ∆ + D (4.99)

4.8.2 Pulse Response

Let [ − ] describe the discrete impulse sequence defined by
½
1  = 
[ − ] ,  (4.100)
0  6= 
where  and  are both integers. [ − ] is the discrete counterpart of the Dirac function (4.17).
When the input [] = [0] acts on a LTI system, the discrete impulse response [] appears.
When a input signal [] act on a LTI system with impulse response [], then the discrete
output [] is defined as the discrete convolution integral according to

[] = [][0] + [ − 1][1] +  + [1][ − 1] + [0][]


X 
X
= [ − ][] = [][ − ] = [] ∗ [] (4.101)
=0 =0

If we substitute (4.101) into (4.82) and let  → ∞ we can find the discrete pulse response
function
∞
Ã∞ ! ∞
̰ !
X X X X
 [] = [ − ][]  − = [ − ][]  −(−)  − 
=0 =0 =0 =0
∞
Ã∞ ! Ã∞ !à ∞
!
X X X X
= [ − ] −(−) [] − = [] − [] − 
=0 =0 =0 =0
, () () (4.102)

where  =  −  and [] = 0 for   0. As for continuous systems, we can see that convolution in
the time domain (-continuous/-discrete) corresponds to multiplication in the frequency domain
(-continuous/-discrete).

Theorem 4.7 The pulse response (4.102) and the pulse-transfer function (4.99) are a -transform
pair, that is
 {[]} = () (4.103)

Example 4.6 For a continuous SISO system

̇ = − +  (4.104)

where () = 0 for  ≤ 0 the corresponding convolution integral will be

Z 
() = ( −  )( ) = () ∗ () (4.105)
0

where ( −  ) = −(− ) 

143
Example 4.7 Consider the unit-sampling-time delay system defined by

[] = [ − 1] (4.106)

The output equals the input delayed by one sampling period. Its impulse response is [] = [−1]
and its discrete transfer function is
1
() = {[ − 1]} =  −1 =  (4.107)

One should notice that every continuous system involving time delay is a distributed system.
This is not the so in the discrete case, where () is a rational function in . Similarly the
discrete transfer function of [] = [ − ] is

() = {[ − ]} =  −  (4.108)

Remark 4.14 Notice that if [] 6= 0 for only a finite number of , the system is called a finite
impulse-response (FIR) system. This means that the output only will be influence by a finite
number of inputs.

4.9 Shift-operator Calculus

2 3
For continuous systems differential-operator calculus, where  ,  
 2 , 

2   , 3   is a
3 

helpful tool for manipulating linear differential equations with constant coefficients. For discrete
systems a similar equivalent shift-operator calculus can be defined for linear difference equations
with constant coefficients. All signals to be considered has to be defined by double sequences
according to
{[] :  =  − 2 −1 0 1 2 } (4.109)

4.9.1 Shift Operator

The forward-shift operator is here denoted by  such that

[] = [ + 1] (4.110)

Remark 4.15 Notice that in some literature such as Åstrøm and Wittenmark [333] the letter
 is used in order to have a clear distinguish from the -transform which is a complex variable
which also take initial conditions into consideration. However, here we will use  as long as it
is clear from the text which purpose  is used. Forward-shift operator is often used in relation
to stability analysis related to the characteristic equation.

Example 4.8 Åstrøm and Wittenmark [333]. Consider the difference equation

[ + 1] − [] = [] (4.111)

where ||  1. In operator notation the equation can be written as

( − )[] = [] (4.112)

144
A solution is
1  −1
[] = [] = [] (4.113)
− 1 −  −1
Since  −1 has unit norm, the right-hand side can be represented as a convergent series according
to

[] =  −1 (1 +  −1 + 2  −2 + )[]
X∞
= −1 [ − ] (4.114)
=1

If [0 ] = 0 it follows from (4.52) that the solution is

−1
X
[] = −0 0 + −−1 []
=0
−
X0
= −0 0 + −1 [ − ] (4.115)
=1

It is clear that (4.114) is equal to (4.115) only if it is assumed that 0 = 0 or that  − 0 → ∞

The inverse of the forward-shift operator is called the backward-shift operator or the delay
operator, and is denoted by  −1 such that

 −1 [] = [ − 1] (4.116)

If the range is not double infinite sequences, the inverse of the the forward-shift operator may
not exist. The backward-shift operator is often used in relation to causality considerations.

4.9.2 Pulse-transfer Operator

Consider the LTI system on the form

[ + 1] = Φ[] + ∆[] (4.117)

[] = C[] + D[] (4.118)

An input-output relationship is obtained by

 [] = Φ [] + ∆ []  (4.119)

Hence,
[] = (I − Φ)−1 ∆ []  (4.120)
and ³ ´
 [] = C (I − Φ)−1 ∆ + D  []  (4.121)
The pulse-transfer operator is defined as
B()
() = C (I − Φ)−1 ∆ + D =  (4.122)
A()
The poles of (4.122) is given by the zeros of the characteristics polynomial A().

145
4.10 Stability Regions
Some important aspects concerning stability, when applying the different approximation meth-
ods for integration as presented in Section 4.5, appears nicely by applying the shift-operator
calculus. This will be demonstrated by examples.

Example 4.9 Egeland [63]. Consider the LTI on the form

ẋ = Ax (4.123)

Let us then apply the trapezoidal method given by (4.72) according to


x[ + 1] = x[] + [Ax[] + Ax[ + 1])]  (4.124)
2
If we apply the shift-operator calculus, we have

x[] = x[] + [Ax[] + Ax[])]  (4.125)
2
m
µ ¶
2 −1
x[] = Ax[] (4.126)
 +1

Alternatively (4.126) can be rewritten

µ ¶
2 1 −  −1
x[] = Ax[] (4.127)
 1 +  −1

Example 4.10 Let us then apply Euler’s method based on forward difference (4.63) on (4.123)
according to
x[ + 1] = x[] +  Ax[] (4.128)
If we apply the shift-operator calculus, we have

x[] = x[] +  Ax[] (4.129)

m
µ ¶
−1
x[] = Ax[] (4.130)


Example 4.11 Let us then apply Euler backward method (4.66) on (4.123) according to

x[ + 1] = x[] +  Ax[ + 1] (4.131)

If we apply the shift-operator calculus, we have

x[] = x[] +  Ax[] (4.132)

m
µ ¶
−1
x[] = Ax[] (4.133)


146
Remark 4.16 Transformation from the -domain to the -domain may appear by using one of
the following approximations

2(1 −  −1 )
 =  (4.134)
 (1 +  −1 )
−1
 =  (4.135)

−1
 =  (4.136)

Eq. (4.134) is known as the bilinear transformation or Tustin’s approximation often used in
stability analysis of discrete controllers in the so-called -domain, see Egeland [63] and Balchen
et al. [21]. This transformation maps the imaginary axis in the -domain to the unit circle
in the -domain. Thus, stable poles in the left-hand plane in the -domain correspond to poles
within the unit circle in the -domain.

Example 4.12 Egeland [63]. Consider the second order system given by

̇1 = 2  (4.137)
̇2 = −20 1 − 2 0 2 +  (4.138)

where  is the damping ratio,  0 is the resonance frequency and  is the disturbance. Applying
Euler’s method of integration gives

1 [ + 1] = 1 [] +  2 [] (4.139)

¡ ¢
2 [ + 1] = 2 [] +  − 20 1 [] − 20 2 [] + []  (4.140)

By using shift-operator calculus we have the -domain formulation

( − 1)
1 [] = 2 [] (4.141)

( − 1) ¡ ¢
2 [] = −20 1 [] − 20 2 [] + []  (4.142)

Assume  = 0 by setting the expression for 2 [] given by (4.141) into (4.142), the the
-domain formulation appears to be

1 2
() = 2  (4.143)
  − 2 + 1 +  2  20

Notice that this expression has poles for 12 = 1 ±   0  Thus, the poles is outside the unit
circle, such that the discretized system is unstable for all  .

Example 4.13 Egeland [63]. Consider the second order system given by (4.137) and (4.138).
Applying combined forward and backwards Euler’s method of integration gives
¡ ¢
2 [ + 1] = 2 [] +  −20 1 [] − 20 2 [] + []  (4.144)
1 [ + 1] = 1 [] +  2 [ + 1] (4.145)

147
 Euler’s Method

Figure 4.12: Mapping of stability regions in the -domain on the -domain using Euler’s method,
backward Euler and trapezoidal (Tustin’s) method for difference approximation.

The shift-operator calculus gives

 ¡ 2 ¢
2 [] = −0 1 [] − 20 2 [] + []  (4.146)
( − 1)

1 [] = 2 [] (4.147)
( − 1)

Assume  = 0 by setting (4.147) into (4.146) yields the -domain formulation

1  2 − 2 + 1
() = 2  (4.148)
  + ( 2 20 − 2) + 1
q
 2 2  2  20
Notice that this expression has poles for 12 = 1 − 2 0 ±   0 4 − 1 The poles are complex
2
conjugated for   0 such that
r
 2  20  2  20
12 = 1 − ±  0 1 −  =  ±  (4.149)
2 4
q
 2  20  2 2
where  = 1− 2 and  =   0 1 − 4 0 . Since 1 2 = 2 +2 = 1 this indicates that the poles
are on the unit circle, which corresponds to the poles on the imaginary axis in the -domain.
Hence, stability is ensured.

In Figure 4.12 it is shown how the stability region for Re()  0 is mapped on the -
plane for Euler’s method (4.135), backwards Euler (4.136) and trapezoidal method (4.134). We
observe that Euler’s method may map the time-continuous system into an unstable region in the
discrete-time system, that is outside the unit circle. For systems with complex conjugated poles
with small damping Euler’s method may cause stability problems as we noticed in the example
above. When the backward approximation is used a stable continuous-time system always give
a stable discrete-time system. Unfortunately, some unstable continuous-time systems may be
transformed into stable discrete systems. Tustin’s approximation or the trapezoidal method
has the advantage that the left-hand of the -plane is transformed into the whole unit circle.
Hence, stable (unstable) continuous-time systems are transformed into stable (unstable) discrete
systems.

148
4.11 Order of the System
Let a difference equation be written as
[ +  ] + 1 [ +  − 1] +  +  [] = 0 [ +  ] + 1 [ +  − 1] +  [] (4.150)
where the order of the (4.150) is defined to be equal to the poles, that is   The zeros is defined
to be  . We also assume that  ≥   The pole excess is defined to be  =  −   Applying
forward-shift operator on (4.150) gives
¡  ¢ ¡ ¢
 + 1   −1 + 2   −2 +  +  [] = 0   + 1   −1 + 2   −2 +  +  []
(4.151)
Let us define the polynomials
() =   + 1   −1 + 2   −2 +  +   (4.152)
and
() = 0   + 1   −1 + 2   −2 +  +   (4.153)
Hence, (4.150) can be reformulated as
()[] = ()[] (4.154)
It is also possible to reformulate 4.150 such that
[] + 1 [ − 1] +  +  [ −  ] = 0 [ +  −  ] + (4.155)
1 [ +  −  − 1] +  +  [ −  ](4.156)
By reversing the order of the coefficients of (), the reciprocal polynomial ∗ () is defined
∗ () = 1 + 1  + 2  2 +  +     (4.157)
 − −( −1) −( −2)
=  ( + 1  + 2  +  +  )
¡ −1 ¢

=    
Notice that
∗ ( −1 ) = 1 + 1  −1 + 2  −2 +  +   −  (4.158)
=  − (  + 1   −1 + 2   −2 +  +  )
=  −  () 
and
() =   ∗ ( −1 ) (4.159)
For () we have similarly
() =    ∗ ( −1 ) (4.160)
Using reciprocal polynomials (4.154) can be written
()[] = ()[]
m
 ∗ −1
  ( )[] =    ∗ ( −1 )[]
m
∗ −1
 ( )[] =   −  ∗ ( −1 )[]
m
∗ −1
 ( )[] =  ∗ ( −1 )[ +  −  ] (4.161)

149
Notice that ∗∗ () is not necessarily the same as () The polynomial () is said to be
self-reciprocal if
∗ () = () (4.162)

Example 4.14 Let () =  Then the reciprocal is ∗ () =  ·  −1 = 1 The reciprocal of ∗ ()
is ∗∗ () = 1 which of course is different from ()

4.12 Relation Between Shift-Operator Calculus and -Transform

Calculation with -transform and shift-operator calculus are related to each other. However,
it is important to be aware of that in shift-operator calculus  denotes an operator that acts
on a double sequences, while in -transform  denotes a complex variable. Due to this several
authors make a distinction between by using  instead of  in shift-operator calculus. Consider
the following example from Åstrøm and Wittenmark [333] about pole-zero cancellations:

Example 4.15 Consider the difference equation

[ + 1] + [] = [ + 1] + [] (4.163)

where  can be an arbitrary real number. The difference equation of (4.163) has the solution

[] = (−) [0] + [] (4.164)

The pulse-transfer function, see (4.99), of (4.163) is

+
() = = 1 (4.165)
+
The last equality is obtained because  is a complex variable. We may misled to believe that
(4.163) is identical to
[] = [] (4.166)
This is only true if the initial condition [0] = 0 If shift-operator calculus is used in solving
(4.163), we achieve
( + )[] = ( + )[] (4.167)
Since  here is an operator we can not formally divide by ( + )

Remark 4.17 Hence, in -transform calculus it is allowed to divide with an arbitrary expres-
sion. However, this is not allowed in shift-operator calculus unless special assumptions are made
i.e. initial condition is equal to zero.

150
Chapter 5

Signal Quality Checking and Fault

Detection

In this chapter the most important signal processing properties for the purpose of control will
be treated. Accurate control performance depends on reliable sensor signals. Unreliable signals
from the sensors will influence the control systems capability for accurate and safe performance.
Therefore poor signal quality should be detected, and the available information should be utilized
in an optimal way, see Figure 5.1. Each signal should be checked for errors and bad signals should
be rejected in a signal processing module. For many safety critical marine operations redundancy
in instrumentation is required by the authorities. This means that two or more sensors measuring
the same state are installed and interfaced to the control system. The redundant sensors could be
of same type, or they could be different. For instance in dynamic positioning having redundancy
in measuring the vessel position in North-East coordinates, two Differential Global Positioning
Systems (DGPS) may be installed. The GPS is a satellite based positioning system. Even better,
one DGPS and one hydroacoustic positioning reference (HPR) system could be installed. The
HPR is an underwater based positioning system. In the last case redundancy is also achieved
both in hardware, software and in the measuring principle. As we will see in Section 2.9 the
authorities requires three or more number of sensors for conducting dynamic positioning in the
most safety critical operations where loss of position could cause fatal accidents, severe pollution
or damages with major economic consequences. This is defined as equipment or consequence class
3 operation. This means that if one of the three sensors fails, the dynamic positioning operation
must be stopped. In order to increase the overall availability and reduce the days of off-hire,
the ship owner then may decide to install a fourth and even a fifth system. Having redundant
sensor signals available put additional requirements to the signal processing concerning signal
weighting (which one should be trusted most), and signal drifting (signal voting may be used
to reject a sensor with large deviation). Hence, handling of multiple signals in an consistent
manner is crucial ensuring a safe and optimal operation.

5.1 Testing of Individual Signals

A signal processing module (Figure 5.1) should perform quality checking of all sensor signals, po-
sition reference system signals and thrust device feedback signals. Single signal quality checking
will include different tests such as (Grøvlen et al. [101]):

151
 SATELLITE

THRUSTER
SETPOINTS MEASUREMENTS

SIGNAL
PROCESSING
THRUSTER
ALLOCATION

REAL WORLD VESSEL

OBSERVER

CONTROLLER
COMMAND VESSEL
THRUSTER MOTIONS
VESSEL
FORCES
SETPOINTS

Figure 5.1: Controller architecture.

• Signal range testing.

• Variance testing.

• Wild point testing.

The set of the tests above may be performed for every signal. If a failure situation is detected,
the signal is unusable and the signal should be rejected.

5.1.1 Windowing
Different formulas based on different techniques for windowing for calculating a signal variance
can be used (Oppenheim et al. [215]). Windowing is the operation of taking a signal [] and
multiplying it by a finite-duration window signal []. That is,

[] = [][] (5.1)

Then [] is also of finite duration. In practice we are only able to measure a signal [] over a
finite time interval, time window. The actual signal available for analysis is then
½
[] − ≤  ≤ 
[] = (5.2)
0 otherwise,

where − ≤  ≤  is the time window.

Using a rectangular window the finite-duration window signal is written
½
1 − ≤  ≤ 
[] = (5.3)
0 otherwise.

152
One of the problems using rectangular window is that it may introduce ripples on the Fourier
transform. This is related to the so-called Gibbs phenomenon. Other techniques using expo-
nential window and Hanning window may improve this situation. For details see Oppenheim et
al. [215] and Electronics Engineer’s Handbook [61].

5.1.2 Signal Range Testing

Most of the signals available have a defined range. An example of this is the gyro sensor, whose
heading output is in the range of 0 − 360o . If the signal processing module receives a gyro signal
outside this range, it assumes that the sensor signal is faulty and will therefore reject it. Thus
it is required that the signal is within a set of allowed values defined by a minimum value min
and a maximum value max according to

[] ∈ [min  max ] (5.4)

5.1.3 Variance Testing

The signal variance gives an indication about the variations in amplitude and frequency of a
signal. For instance, a high level of measurement noise gives high variance and vice versa.
However, the signal variance may also be high due to a high level of process noise, i.e. heavy
seas. This is not a failure situation and should therefore be reflected in the calculations.
Consider the sequence {[]} at  =  consisting of the signal itself and −1 historical values
according to
{[] :  =  − ( − 1)   − 1 } (5.5)
The average value of the sequence ̄ can be calculated to be

X
1
̄ = [] (5.6)

=−(−1)

The corresponding variance is then found to be

⎛ ⎞
X
1 ⎝
 2 = []2 − ̄2 ⎠  (5.7)
−1
=−(−1)

Let us define

X
1
[] = []2  (5.8)

=+1−(−1)

On recursive form (5.7) can be written as

 ¡ ¢
 2+1 = [ + 1] − (̄[ + 1])2  (5.9)
−1
where
1¡ ¢
[ + 1] = [] + ([ + 1])2 − ([ − ( − 1)])2  (5.10)

High variance may be a symptom of a sensor failure or inaccurate measurement. A frozen
signal can indicate a failure in a sensor, leading to zero variance. Hence, both an upper and a
lower limit is considered in the variance test, see Figure 5.2.

153
 High variance limit

Calculated signal variance

Low variance limit

sec

Figure 5.2: Signal variance test.

5.1.4 Wild Point Testing

If a wild point is detected, this measurement should be rejected for one sample. A wild point is
a measurement that deviates considerably from the previous measurements, that is, it is outside
a certain band about the estimated mean signal, see Figure 5.3. The sampled value [] may
be accepted if
[] ∈ [̄ −  ̄ + ] (5.11)
where  often is set to be in the interval 3 − 9 Normally, the wild point is replaced by a
calculated value. This could be the value from the previous sample, the mean value or an
estimated minimum variance value.

5.2 Handling of Redundant Measurements

For redundant sensor or position reference system configurations, the possibilities for fault de-
tection, weighting and voting are improved. Signals from different sensors or position reference
systems are compared and the difference between them will indicate a failure situation. Weight-
ing principles are used for calculation of optimal input data to the controller algorithm.

5.2.1 Voting
The signal processing module provides two levels of voting for detection of drifting of sensors
and position reference systems.
If two sensors or position reference systems are available, the signal processing module can
detect drifting (bias) between the two sources. However, it is not able to detect which sensor
or position reference system that has failed unless an individual signal erroneous situation is
detected. Anyway, the operator receives a warning that a signal is detected.
If three or more sensors are available, the signal processing module is also able to carry out
voting in order to detect whether one of the signals are truly drifting. Figure 5.4 illustrates this
situation. The three measurements are different, but it seems that sensor no 2 deviates from
sensor no 1 and no 3. The voting algorithm will reject sensor no 2 when the deviation exceeds
a defined limit.

154
 Wild point

Wild point limit

Measurement

Wild point limit

sec

Figure 5.3: Wild point test.

Value

1 2 3 Sensor No

Figure 5.4: Voting of redundant measurements.

155
5.2.2 Weighting
For redundant sensor or position reference system configurations the signal processing module
performs a weighting of the individual signal before in order to calculate the optimal average
signals. The individual signal weighting factors are either calculated automatically based on the
signal variance or set manually. A sensor which has high variance compared to the other sensors
will, in the automatic case, have a correspondingly low weighting factor and influence on the
weighted signal. However, if the variance is too low (frozen signal) the signal is discarded from
the weighting calculation.
The equation below illustrates calculation of a weighted sensor measurement,  . Assume
that three sensor measurements are available, 1 , 2 and 3 , with weighting factors 1 , 2 and
3 , then the weighted measurement becomes
1 1 + 2 2 + 3 3
 =  (5.12)
1 + 2 + 3
Consider an unbiased estimated signal consisting of a sum of  independent weighted signal
according to
X
̂ =    (5.13)
=1
where

X
 = 1 (5.14)
=1
The weighting factors can be set manually by an operator or automatically calculated based on
the principle of minimum variance. For manual weighting we have

 = P  (5.15)
=1 

For automatic weighting consisting of two measurements the weights become

 22
1 =  (5.16)
 21 +  22
 21
2 =  (5.17)
 1 +  22
2

Notice that a higher variance in one of the signals results in higher weight in the remaining
signal. The general automatic weighting algorithm for  signals can be written
Q 2

6=
 = P Q  (5.18)
=1  2
6=

5.2.3 Enabling and Disabling of Sensors

Operator initiated disabling or an abrupt loss of a sensor can lead to the signal deviation effect
shown in Figure 5.5. This can be avoided by filtering the signal average a specific time period
after the loss of signal has occurred, illustrated by Figure 5.6. The filter should be activated

156
 response

deviation in position

Figure 5.5: Effect of step in position measurement.

filtering

Sensor 1

position

Sensor 2

Figure 5.6: Filtering of position average.

subject to an event indicating the loss of the sensor signal. An appropriate time period 
depending on a signal difference and a maximum change rate should be specified. A change
in the average value is inevitable after a signal loss but the filtering will prevent the unwanted
response.
Let  denote the new calculated weighted signal after the failure situation. Then, the
filtered weighted signal   entering the control system could be found by solving the lowpass
equation below
1 1
̇  = −   +   (5.19)
 
The filtering should not inflict a phase change of the measurement too much. When enabling a
sensor, the signal will remain smooth and no filtering is necessary.

157
Chapter 6

Filtering and State Estimation

In order to be successful in the design of industrial control system in addition to appropriate

signal processing, filtering of signals are essential. Topics considered in this chapter are basic
introduction to:

• Filtering of measurement noise. Most sensor signals are contaminated by some noise
caused by external disturbances and internal properties of the sensor itself. The noise may
have negative impact on the controller performance, if no precaution is taken. By filtering
we achieve a change in the relative amplitudes of the frequency components in a signal or
even elimination of some frequency components entirely.
• Reconstruction of non-measured data. For many applications important process states are
not measured. Typical reasons for this could be that no convenient sensors exists, or simply
that cost reasons motivate to not to install the sensor. In such cases sophisticated model
based filtering techniques - state estimation - can be applied. The main purpose of the
state estimator (observer) is to reconstruct unmeasured signals and perform filtering before
the signals are used in a feedback control system. The input to the state estimator in a
dynamic positioning system is sensor data e.g. from an inertial measurement unit (IMU)
measuring the vessel’s heave, roll and pitch motions, a compass or gyro measuring the
vessel’s heading and a position reference system like the satellite navigation system DGPS
(differential global positioning system) measuring the vessel’s North and East position
coordinates, see Chapter 2 for more description of these sensors.
• Dead reckoning. All kind of equipment will fail according to some failure rate. Experience
from industrial applications has shown that the most frequent control system failure are
caused by sensor failures. In safety critical marine applications a sudden drop-out of the
control system may lead to dangerous situations, if not an adequate signal substitution
will take place. Applying model based filters the predicted sensor signal may, at least for
some period of time, replace the measured signal. Then only the mathematical model of
the process and the sensor device itself is used to predict the signal to be used by the
control system. Using only model prediction for calculation of state signal is called dead
reckoning. Use of observers for the purpose of diagnostics and in the design of so-called
virtual sensors for fault-tolerant control is an increasing field of research

This chapter will present the basic properties of the most used conventional analog and digital
filters, and model based filters (also called observers) used in control system design. Most of the

158
theory presented is general and valid for almost all kind of industrial control problems. However,
certain filtering and state estimation problems a more characteristics for marine applications.
These will be paid more attention to later in the text in Chapter 8. The chapter is organized as
follows: Section 6.1 will handle conventional analog and digital filtering techniques. In Section
6.2 model based state space filtering techniques also called observers will be presented.
The theory presented in this chapter is solely based on earlier text books. For more thor-
oughly presentation the reader is referred to Johansson [138], Egeland [63], Fossen [78], Elec-
tronics Engineer’s Handbook [61], Oppenheim et al. [215] and Åstrøm and Wittenmark [333].
Concerning diagnostics and fault-tolerant control the reader is referred to Blanke et al. [31].

6.1 Analog and Digital Filtering

According to Oppenheim et al. [215] Linear time-invariant (LTI) systems that change the shape
of the systems power spectrum are referred to as frequency-shaping filters. Filters that are
designed to pass some frequencies essentially undistorted and significantly attenuate or eliminate
others are called frequency-selective filters. Due to the broad field of applications we will below
first concentrate on frequency-selective filters. Several basic types of filters has been designed,
and they are given name indicating their main function. The most common ideal filters are:

• Lowpass filter passes low frequencies of signals from zero to cut-off frequency and attenuate
or reject all higher frequencies.

• Highpass filter passes the high frequencies of signals and attenuate or reject all low fre-
quencies from zero to cut-off frequency.

• Bandpass filter passes a band of frequencies between lower and upper cut-off frequencies
and attenuate frequencies higher and lower than those defined by the band.

• Bandstop filter (or reject filter) stops signal frequencies between its lower and upper cut-off
frequencies, and transmits all other signals.

• All-pass filter transmits all signal frequencies and produces a predictable phase shift.

Example 6.1 A continuous-time ideal lowpass filter is a LTI system that passes complex ex-
ponential  for values of  in the range of − ≤  ≤   and rejects all other frequencies,
where  is the cut-off frequency. Hence, the frequency response is given on the form:
½
1 || ≤ 
 () =  (6.1)
0 ||  

Adopting the complex exponential signal  , ideal filters are usually defined to have sym-
metrical frequency response about  = 0 where | ()| = 1.

Ideal filters is useful to describe idealized system configurations. However, they are not re-
alizable and must be approximated in real applications as nonideal filters. Nonideal filters do
have a gradual transition band from passband to stopband. The acceptable deviation limits in
the passband and the stopband are often described by  1 and  2 , see Figure 6.1. The amount by
which the frequency response differs from unity in the passband is referred to as passband ripple,

159
 |H(j)|





Passband  Transition  Stopband 
p s

Figure 6.1: Tolerances for magnitude characteristics of a lowpass filter.

and corresponding deviation from zero in the stopband is referred to as stopband ripple. In ad-
dition, especially for control purposes the phase characteristics of the filter is of importance. For
nonideal lowpass filters, there is a trade-off between the width of the transition band (frequency
domain) and the rise time, delay time and overshoot (time domain) of the step response.

6.1.1 Nonideal Lowpass Filter

A widely-used class of LTI systems for lowpass filtering (Oppenheim et al. [215]) is the so-called
n -order Butterworth filters. In the frequency domain for  = 1 2 and 3


 = 1 : () =  (6.2)
 + 
2
 = 2 : () = ¡ √  ¢ (6.3)
2 + 2   +  2
3
 = 3 : () =  (6.4)
(2 +    +  2 ) ( +  )

The corresponding filters may also be written in time domain according to

 = 1 : ̇ +   =   (6.5)
√
 = 2 : ̈ + 2  ̇ +  2  = 2  (6.6)
...
 = 3 :   + 2  ̈ + 2 2 ̇ +  3 =  3  (6.7)

where  is the signal to be filtered (input),  is the filtered signal (output) and  is the cut-off
frequency. The Butterworth filters have a frequency response which is optimized to give as flat
amplitude as possible in the passband, see Figure 6.2. Similarly other classes of lowpass filter
exist, where the most known are Chebyshev filter, Bessel filter, Legendre-Papoulis filter and
Elliptic filter, see Part 16 in Electronics Engineer’s Handbook [61] for more details. The Elliptic
filter does also have zeros that may differ from infinite.

160
 Figure 6.2: Bode plot of Butterworth filter.

6.1.2 Nonideal Highpass Filter

A highpass filter may be designed by substituting
1
→  (6.8)

in the equations describing the lowpass filter.

6.1.3 Notch Filter

Cascaded notch filter is an often used bandstop filter and is written

Y 2 + 2     +  2 
 =  (6.9)
=1
2 + 2     +  2

where   is the center frequencies of the filter,  is the number of center frequencies, and   ¿ 1
and 0    are the damping ratios in the numerator and the denominator, respectively. A
typical choice could be   = 01 and   = 1. Notch filter in cascade with lowpass filter have
often in the past been used for wave filtering. An appropriate number of center frequencies,
typically 3, with a distribution about the domination wave frequency is chosen, see Figure 6.3.

6.1.4 Digital Filtering

Digital filters provide many of the same frequency selective services as analog filters. Digital
filters are often defined in terms of equivalent analog filters. However, many digital filters are

161
 0

−5

−10
Magnitude [dB]
−15

−20

−25

−30

−35

−40

200

150

100
Phase [deg]

50

−50

−100

−150

−200
−1 0 1
10 10 10
Frequency (rad/sec)

Figure 6.3: Notch filter where   = 01 and   = 1  1 = 04 [] 2 = 063 [] and
 3 = 10 [].

designed using properties that are unique to this technology. Compared to analog filters, digital
filters generally have the advantages that the stability of certain classes can be guaranteed,
coefficients can easily be altered, they can operate over a wide range of frequencies with wide
dynamic range and with high precision.
Consider the equation
̇ =  (6.10)
Approximation of this equation could be
[ + 1] − [] 1
= ([ + 1] + [])  (6.11)
 2
Applying -transform gives
( − 1)[] 1
= ( + 1)[]
 2
m
 ( + 1)
[] = [] (6.12)
2( − 1)

Hence, once an analog filter () is defined, the discrete filter () based on trapezoidal approx-
imation method could be constructed by replacing  in () by

2( − 1)
=  (6.13)
 ( + 1)

162
 Setting  =  in (6.13) gives
µ ¶
2( − 1) 2 
= 
=  tan =  (6.14)
 ( + 1)  2
where  is the discrete frequency corresponding to . When  
2 → 2 ,  → ∞ As noticed

in (4.78)  =  is known as the Nyquist frequency, which is the absolute upper limit for the
discrete filter to approximate the analog filter.
Linear constant coefficients filters can be categorized into two main classes known as finite
impulse-response (FIR) filter or infinite impulse-response (IIR) filters.

Finite Impulse-Response (FIR) filter

FIR filters can be expressed as
[] = 0 [] + 1 [ − 1] +  +  [ − ]
X
=  [ − ] (6.15)
=0

where the coefficients { } are called filter tap weights. In terms of -transform the filter discrete
transfer function is written
X
() =   −  (6.16)
=0
The filter’s transfer function consists of zeroes only (i.e. no poles). Because of this the FIR filter
is referred to as an all-zero or transversal filter. P
If the filter input is bounded (i.e. |()| ≤ 1 for
all ), the maximum value of the output () is | |. If all the tap weights { } are bounded,
the filters output is likewise bounded and, as result, stability is guaranteed. Furthermore, the
phase of (), when plotted with respect to frequency, is linear with constant slope.
Example 6.2 For moving average filter it is convenient to apply a FIR formulation, where the
output [] for any , i.e. 0 , is an average of values [] in the vicinity of 0  An example of
a two-point average filter is
1
[] = ([ − 1] + [])  (6.17)
2
Three-point moving average filter may be
1
[] = [ − 1] + [] + [ + 1] (6.18)
3

Infinite Impulse-Response (IIR) filter

IIR filters can be expressed as
[] = −1 [ − 1] − 2 [ − 2] −  −  [ − ] + 0 [] + 1 [ − 1] +  +  [ − ] (6.19)
where the the coefficients { } and { } are the filter tap weights. IIR filter generally satisfies
a given magnitude frequency response design objective with a lower order filter compared to
a corresponding FIR filter. However the phase characteristics does not generally exhibit linear
phase. FIR filter can only be realized in digital form, while IIR filter may be realized analog.
The most known IIR filters are Tsjebyscheff, Elliptic and Butterworth. The discrete IIR
version of a Butterworth filter may be derived putting (6.13) into (6.2)-(6.4).

163
6.2 State Estimation
Analog filters are not always suitable for all control applications. In many cases the signals to be
separated do not appear in totally disjoint frequency bands. A disadvantage concerning control
is that undesired phase lag often is introduced by filters. This consequently reduce the control
performance and stability margins in terms of bandwidth reduction, and by less gain and phase
margin. Thus for the purpose of control especially higher order lowpass filter or cascades of filters
should be used with care. Very often the signals of interest consists of frequency components
in different frequency band slightly overlapping each others. This is in particular the case for
many marine applications.
In industrial control systems many of the states of interest concerning control are impossible
to measure, or if possible, the cost for installing the sensors may be too high. It is therefore of
importance to determine or reconstruct those states from the available measurements. This is
called state estimation and is provided by so-called observers.

Definition 6.1 An observer or state estimator produces the state of a system from measure-
ments of inputs and outputs.

In positioning of marine vessels, the main purpose of the observer (Figure 6.4) is to estimate
velocities and current and wave drift forces from position measurements. Later in the text in
Section 8.2 we will see that, in addition, 1st-order wave disturbances should be filtered out e.g.
by using a notch filter suppressing wave-induced disturbances close to the peak frequency of the
wave spectrum. The position and heading measurements are corrupted with colored noise caused
by wind, waves and ocean currents. However, only the slowly-varying disturbances should be
counteracted by the propulsion system whereas the oscillatory motion due to the waves (1st-
order wave-induced disturbances) should not enter the feedback loop. This is done by using
so-called wave filtering techniques, which separates the position and heading measurements
into a low-frequency (LF) and a wave-frequency (WF) position and heading part, see Figure
7.20. Even though accurate measurements of the vessel velocities are available when using
differential GPS systems or Doppler log, a state estimator must be designed in order to satisfy
the classification rules for the positioning systems. In the case of temporarily loss of position and
heading measurements, the observer must be able to operate in a dead reckoning mode implying
that predicted observer velocity, position and heading is used for feedback. A temporarily loss
of these measurements will not affect the positioning accuracy. When the necessary signals
reappear, the estimated values will give a smooth transition back to the true position and
heading.
Before we address the marine applications any further, basic introduction to the most used
linear observers will be given in this chapter.

6.2.1 Deterministic Estimators

In the ideal situation when there are no uncertainties in the system with regards to modelling
parameters and noise, the system is called deterministic. For this simplified case, a deterministic
estimator may be constructed to illustrate the concept of state estimation. In the realistic case
with modelling uncertainties and noise on measurements and control signals, stochastic theory
may be applied in the observer design. The best known stochastic estimator is the Kalman
filter, which will be treated subsequently.

164
 measurement
disturbances noise

control inputs measurements

Ship
y

Observer estimated states

(state estimator)
x

Figure 6.4: Block diagram showing the principle of state estimation.

Continuous-time deterministic estimator

Consider the following control plant model of the linear time-invariant (LTI) system of the form

ẋ = Ax + Bu (6.20)
y = Dx (6.21)

This system is completely deterministic, and it is assumed that the pair [A D ] is observable
(actually, detectability is sufficient). This means that it is possible to estimate the system states
from its inputs and outputs. A first attempt at constructing an estimator might be to require
that the estimator copies the control plant model exactly

x̂˙ = Ax̂ + Bu (6.22)

ŷ = Dx̂ (6.23)

where x̂ is the estimated state vector. But if the initial conditions x( ) and x̂( ) differ, the
estimator will not reproduce the correct states. To correct this, an injection term using the
difference between the measured output y and the estimated output ŷ is added to the estimator

x̂˙ = Ax̂ + Bu + K (y − ŷ) = Ax̂ + Bu + K D(x − x̂) (6.24)

The state-feedback gain matrix K may then be used for tuning the estimator. The error
dynamics is found by subtracting (6.24) from (6.20)

x̃ = ẋ − x̂˙ = A(x − x̂) − K D(x − x̂) = (A − K D)(x − x̂) = (A − K D)x̃ (6.25)

Hence, the error x̃ will exponentially approach zero if the eigenvalues of (A − K D) have negative
real parts. Figure 6.5 shows a block diagram of this estimator.
If K is chosen only with respect to obtaining fast convergence of x̂ to x, the poles will be
moved far away into the left half-plane. This will increase the bandwidth of the estimator, which
means that x̂ will be more sensitive to noise in u and y. The choice of K is thus a trade-off
between fast estimation and noise-rejection properties, where the best compromise is achieved
by an optimal estimator (Anderson and Moore [7]).

165
 Plant

u + x x y +
B D
+ -

A
Plant model
+ x x y
B +
D
+

Ke

Figure 6.5: Deterministic estimator illustrating the plant model concept.

Discrete-time deterministic estimator

Consider the control plant model of the discrete system
x[ + 1] = Φx[] + ∆u[] (6.26)
y[] = Cx[] (6.27)
The estimator can found by copying the control plant model such that
x̂[ + 1] = Φx̂[] + ∆u[] + L(y[] − ŷ[]) (6.28)
ŷ[] = Cx̂[] (6.29)
where L is the state-feedback gain matrix, and x̂ is the estimate of x. The error dynamics
become (with x̃ = x − x̂)
x̃[ + 1] = Φx̃[] − L(y[] − ŷ[]) = Φx̃[] − LC(x[] − x̂[]) =(Φ − LC)x̃[] (6.30)
Now L must be chosen such that (6.30) is asymptotically stable, which means that the magnitude
of the eigenvalues of (Φ − LC) must be within the unit circle. This is again only possible if the
system is observable.

Certainty equivalence principle

If a dynamic system is defined by (6.26) and (6.27), a feedback control law may be defined by
using a complete state information and the positive gain matrix G  0 such that
u[] = −Gx[] (6.31)
x[ + 1] = Φx[] − ∆Gx[] = (Φ − ∆G)x[] (6.32)

166
If the complete state information is unavailable, the observer defined in (6.28) may be used,
and the estimated state vector x̂ replaces x in the calculation of the control input according to
u =−Gx̂[] This is in the literature denoted as the certainty equivalence principle or separation
theorem. The closed loop error dynamics of the complete feedback system becomes

x[ + 1] − x̂[ + 1] = (Φ − ∆G)x[] − (Φ − ∆G)x̂[] − LCx̃[]

= (Φ − LC − ∆G)x̃[] (6.33)

This system is of order 2 (with  as the order of the control plant model). The eigenvalues of
the matrix (Φ − ∆G) is the eigenvalues of the closed-loop pole-placement problem with full
state information. The eigenvalues of the matrix (Φ − LC) of the observer is assumed to be
faster than those of the controller, typical one decade. The solution of these two problems are
dual, and the stability of the two systems may be analyzed separately (Aaström and Wittenmark
[333]). This principle of duality may be extended to more complex combinations of controllers
and observers, and is a subject for research for nonlinear control and observer designs.

6.2.2 Least Squares Estimation

This section is partly based on Henriksen [112]. The state vector x of a static, linear stochastic
process may be estimated from the measurement vector y and an a priori estimate of the state
vector x̄, made prior to the time of the measurement. Let the measurement be described by

y = Cx + v (6.34)

where C is a known matrix, and v is the measurement noise, which is assumed statistically
independent of x with a known, positive definite covariance matrix V and expectation value
equal to 0 such that

[v] = 0 [v] = [vv ] = V  0 [xv ] = 0 (6.35)

The uncertainty of the a priori estimate x̄, which also is the expectation value of the corrected
state vector x, is given by the covariance matrix X̄

[x] = x̄ [x] = [(x − x̄)(x − x̄) ] = X̄  0 (6.36)

A weighted least squares method for finding x from x̄ and y is given by

1 
 = [(x − x̄)T P(x − x̄) + (y − Cx) Q(y − Cx)] (6.37)
2
where P and Q are appropriately chosen positive semidefinite and positive definite weighting
matrices, respectively. By requiring the derivative of  to be zero, the value x̂ of x which
minimizes  may be found
¯
 ¯¯
= P(x̂ − x̄) − C Q(y − Cx̂) = 0 (6.38)
x ¯(x=x̂)
(P + C QC)x̂ = Px̄ + C Qy =(P + C QC)x̄ + C Q(y − Cx̄) (6.39)
T −1 
x̂ = x̄ + (P + C QC) C Q(y − Cx̄) = x̄ + K(y − Cx̄) (6.40)

167
x̂ is the improved estimate of x after the measurement has been made, is termed the corrected
or aposteriori estimate of x. The uncertainty of x̂ is given by the covariance of the estimation
error x̃ = x − x̂, which is termed X̂ according to

X̂ = [x̃x̃ ] = [(x − x̂)(x − x̂) ]

= [(x − x̄ − K(y − Cx̄))(x − x̄ − K(y − Cx̄)) ]
= [(x − x̄ − K(Cx + v − Cx̄))(x − x̄ − K(Cx + v − Cx̄)) ]
= [(x − x̄ − KC(x − x̄) − Kv))(x − x̄ − KC(x − x̄) − Kv)) ]
= [((I − KC)(x − x̄) − Kv))((I − KC)(x − x̄) − Kv)) ]
= [(I − KC)(x − x̄)(x − x̄) (I − KC) − Kv(x − x̄) (I − KC) 
−(I − KC)(x − x̄)v K + Kvv K ]
= (I − KC)[(x − x̄)(x − x̄) ](I − KC) + K[vv ]K 
−K[v(x − x̄) ](I − KC) − (I − KC)[(x − x̄)v ]K
= (I − KC)X̄(I − KC) + KVK  (6.41)

because [v] = [xv ] = 0. One (important) way of choosing P and Q is to minimize the
sum of diagonal terms in the covariance matrix X̂. It can be shown that this is achieved by the
choice (Henriksen [112])
−1
P = X̄  Q = V−1  (6.42)
The gain matrix K is then given by
−1 −1  
K = (X̄ +CT V C)−1 C V−1 = X̄C (CX̄C + V)−1  (6.43)

It may be shown that the estimator x̂ is unbiased, which means that the expectation value of
the a-posteriori estimate equals the a priori expectation value

[x̂] = [x] = x̄

These results may now be used to develop the discrete Kalman Filter.

6.2.3 Discrete Kalman Filter

Consider the following discrete LTI system on the form

x[ + 1] = Φx[] + ∆u[] + Γw[] (6.44)

y[] = Cx[] + v[] (6.45)

where it is assumed that the system is observable.

Let us assume that the process noise w[] and the sensor noise v[] are discrete Gaussian
white noise processes with known covariance matrices.
That is, w[] ∼ (w̄ W[]) and v[] ∼ (v̄ V[]) The mean value of the process disturbance
is defined as
E[w[]] = w̄ (6.46)
The covariance matrix of the process disturbance is positive definite (Appendix A) and is defined
as
E[(w[] − w̄)(w[] − w̄) ] = W[]   (6.47)

168
where   = 1 if  =  and   = 0 if  6=  The mean value of the sensor noise is defined as

E[v[]] = v̄ = 0 (6.48)

The covariance matrix of the sensor noise is positive definite and defined as

E[(v[] − v̄)(v[] − v̄) ] = V[]   (6.49)

Assume that the process disturbance and measurement noise are uncorrelated:

E[v[]w[] ] = 0 (6.50)

The expectation value of the initial condition of the state is defined as

E[x[0]] = x̄(0) (6.51)

The corresponding covariance matrix of the initial condition of the state is positive definite and
is defined as
E[(x[0] − x̄(0))(x[0] − x̄(0)) ] = X̄0  (6.52)
Assume that the process disturbance and the sensor noise is uncorrelated with the initial con-
dition according to

E[(x[0] − x̄(0))w[] ] = 0 (6.53)


E[(x[0] − x̄(0))v[] ] = 0 (6.54)

Remark 6.1 Notice that if the real disturbance is not a Gaussian process, it can be reformulated
in a disturbance model (approximation) which is driven by white noise. The disturbance model
is then included in an augmented state vector.

Assume that the data

Y = {y[] u[] |  ≤ } (6.55)
is known. Using Y it is possible to have three cases for estimation (Figure 6.6 ) of x[ + ]:

• Smoothing for   0

• Filtering for  = 0

• Prediction for   0

We will first consider the case when  = 1This is a so-called one-step prediction. Then the
filtering case with  = 0 will be shown.

169
 Signal Smoothing Filtering Prediction

k k k
Estimate

k-1 k k+1

Figure 6.6: Smoothing, filtering and prediction.

Kalman filter - prediction case

Let us denote the vector x̂ to be the state estimate of the state vector x. Consider the system
given by (6.44) and (6.45) and assume that the estimated state vector is a function of past inputs
according to

x̂[ + 1|] = Φ x̂[| − 1] + ∆u[] + Γ w̄[]+K[] (y[] − ŷ[])  (6.56)

ŷ[] = C x̂[| − 1] (6.57)

A new feedback term K[] (y[] − ŷ[]) is introduced. K[] is a  ×  gain matrix and is also
called Kalman filter gain matrix if it is chosen subject to some principles as we will see later in
the text. The notation x̂[ + 1|] is used to indicate that is is an estimate of x[ + 1] based on
measurements available at time . As long as the predicted measurement vector ŷ[] coincides
exactly with the measurement vector y[] the feedback term will give no contribution. We often
call the difference y[] − ŷ[] to be the innovation. Eq. (6.56) and (6.57) is an observer of the
LTI system (6.44) and (6.45).
The error dynamics of the reconstruction (estimation) can be found by subtracting (6.44)
from (6.56) according to

x̃[ + 1|] = x[ + 1]−x̂[ + 1|] (6.58)

m
x̃[ + 1|] = Φ x̃[| − 1] + Γ(w[] − w̄[]) − K[] (y[] − ŷ[])  (6.59)
m
x̃[ + 1|] = (Φ − K[]C) x̃[| − 1] + Γ(w[] − w̄[]) − K[]v[] (6.60)
m
µµ ¶ µ ¶¶
Φ Γ(w[] − w̄[])
x̃[ + 1|] = (I − K[]) x̃[| − 1]+  (6.61)
C v[]

Remark 6.2 If we disregard the effect of Γ (w[] − w̄[]) −K[]v[] we observe that the choice
of K[] can be made such that (6.61) is asymptotically stable. Thus, the error dynamics converges

170
to zero even if the system itself is unstable providing that the pair (Φ C) the criterion for
observability is fulfilled. Hence, the  ×  observability matrix Q has full rank, see (4.60) in
Section 4.4.

However, we will here take the properties of the noise into account in the design of K[]
minimizing the variance of the error dynamics, which is denoted as

X̄[] = E[(x̃[] − E(x̃[]))(x̃[]−E(x̃[])) ] (6.62)

The expectation value of the error x̃ is

E[x̃[ + 1]] = (Φ − K[]C) E[x̃[]] (6.63)

Let E[x[0]] = x̄(0) If x̂[0] = x̄(0) then E[x̃[0]] = 0 Thus the mean value of the error dynamics
is zero for all  ≥ 0 independent of K[] Since E[x̃[]] is assumed to be independent of w[]
and v[](6.61) gives

X̄[ + 1] = E[(x̃[ + 1] − E(x̃[ + 1]))(x̃[ + 1]−E(x̃[ + 1])) ] (6.64)


= E[x̃[ + 1]x̃[ + 1] ]
õ ¶ µ ¶ µ ¶! µ ¶
Φ Φ ΓW[]Γ 0× I×
= (I − K[]) X̄[] + 
C C 0× V[] −K[]

Further X̄[0] = X̄0  From (6.64) it follows that if X̄[] is positive semidefinite, then X̄[ + 1] is
also positive semidefinite.
Let us first define a general principle called the completion of squares (Åstrøm and Witten-
mark [333]), where the quadratic loss function is defined as
³ ´µ Q Q
¶µ
x − x̂
¶
  
(x − x̂ y − ŷ) = (x − x̂)  (y − ŷ)  (6.65)
Q Q y − ŷ

where Q  0 is a symmetric positive definite matrix and Q ≥ 0 is a symmetric positive

semidefinite matrix. The minimum of (6.65) with respect to x is found by
¯
(x − x̂ y − ŷ) ¯¯
¯ = 0 (6.66)
x x=x̂

Then there exists an K satisfying

Q KC = Q  (6.67)
such that (6.65) can be rewritten
³ ´
(x − x̂ y − ŷ) = (x − x̂) Q − (KC) Q KC (x − x̂)+ (6.68)
(y − ŷ + KC(x − x̂)) Q (y − ŷ + KC(x − x̂)) (6.69)

Because (6.68) is quadratic in y − ŷ and both terms are greater or equal to zero, it is seen that
(6.68) is minimized for
y − ŷ = −KC(x − x̂) (6.70)

171
Thus, it can be shown that K is unique since Q  0 The minimum is therefore
³ ´
(x − x̂ y − ŷ) = (x − x̂) Q − (KC) Q KC (x − x̂) (6.71)

By using the idea of completion of squares, it follows that α X̄[ + 1]α is minimized by K[]
satisfying ³ ´
K[] V[] + CX̄[]C = ΦX̄[]C  (6.72)

for any α If V[] + CX̄[]C is positive definite then

³ ´−1
K[] = ΦX̄[]C V[] + CX̄[]C  (6.73)

This inserted into (6.64) gives

³ ´−1
X̄[ + 1] = ΦX̄[]Φ + ΓW[]Γ −ΦX̄[]C V[] + CX̄[]C CX̄[]Φ  (6.74)

Theorem 6.1 Åstrøm and Wittenmark [333]. The Kalman filter predictor case is defined by
(6.56)
³ and (6.57).´ It is optimal with respect to minimum variance of the error dynamics if
V[] + CX̄[]C is positive definite, the disturbance and measurement error are Gaussian,
and if the Kalman filter gain matrix is chosen as (6.73), where the covariance matrix of the
error dynamics X̄[ + 1] is found by (6.74).

Remark 6.3 As long as E[v[]] = 0 it is possible to show that the an unbiased estimate is
possible to achieve.

Remark 6.4 Another notation of (6.74) is X̄[| − 1], indicating that the measurements up to
and including  − 1 are used. The term ΦX̄[ − 1]Φ shows the influence of system dynamics
on the covariance matrix of the error dynamics. ΓW[]Γ shows that the covariance matrix
will increase for increasing process disturbance. The term
³ ´−1
−ΦX̄[ − 1]C V[ − 1] + CX̄[ − 1]C CX̄[ − 1]Φ  (6.75)

shows how the variance is decreased accounting for information obtained through the measure-
ments.

Remark 6.5 Since X̄[| − 1] does not depend on any measurements and all other terms are
assumed to be known, it is possible to precompute (6.73) and X̄[| − 1]

Remark 6.6 For LTI where Φ , ∆ Γ and C are constant stationary solution of (6.73) can be
found, where X̄[ + 1] → X̄∞ such that K[] → K∞ 

172
Kalman filter - filter case
³ ´
Consider the LTI process given by (6.56) and (6.57). If the matrix V[] + CX̄[| − 1]C
is positive definite, then the optimal filter can be found. The so-called corrector or aposteriori
estimate is given by

x̂[|] = x̄[| − 1] + K [] (y[] − C x̄[| − 1])  (6.76)

The predictor or apriori estimate is given by

x̄[ + 1|] = Φ x̂[|] + ∆u[] + Γw̄[|]

= Φ x̄[| − 1] + ∆u[] + Γw̄[|] + K[] (y[] − C x̄[| − 1])  (6.77)

where
³ ´−1
K [] = X̄[| − 1]C V[] + CX̄[| − 1]C  (6.78)
K[] = ΦK [] (6.79)

The covariance matrix of the estimation error dynamics is given by a so-called Ricatti equation
according to
³ ´
X̄[ + 1|] = ΦX̄[| − 1]Φ + ΓW[]Γ −K[] V[] + CX̄[| − 1]C K[] (6.80)
³ ´−1
X̄[|] = X̄[| − 1] − X̄[| − 1]C V[] + CX̄[| − 1]C CX̄[| − 1](6.81)
X̄[0| − 1] = X̄0  (6.82)

In Figure 6.7 the Kalman filter is illustrated graphically.

Remark 6.7 The notation X̄[| − 1] is used to specify the available data. Notice that X̄[|]
is the variance of the estimation error dynamics at time  given Y .

Remark 6.8 The innovation given by ²[] = y[] − ŷ[] = y[] − Cx̄[| − 1] will be white
noise in a Kalman filter given as above.

Remark 6.9 In dead reckogning (with a failure in sensor  and thus loss of  []) the corre-
sponding innovation will be set to  [] = 0 Hence, only the predictor (6.77) will be active.
Usually the predicted state ̄ [] becomes less noisy and sluggish without the corrector. Depend-
ing on the external disturbances and the accuracy of the model, the control performance may be
satisfactory for quite long time. In dynamic positioning dead reckogning with satisfactory control
performance up to several minutes has been experienced.

6.2.4 Extended Kalman Filtering

In the previous section we assumed that the process dynamics could be described by a LTI
system. However, in many cases the process will be nonlinear. In marine applications conduction
dynamic positioning the kinematics relating Earth-fixed coordinates to vessel-fixed coordinates
represents such an nonlinearity. This will be explained more in detail later on. A common way
to circumvent the problem of nonlinearities is to introduce a so-called extended Kalman filter,

173
 Disturbance Measurement
noise
w[k]
v[k]
u[k] Real y[k] [k]
K[k]
Process

[k] Estimated
measurement

w[k] x[k+1] x[k] y[k]

[k] C

x[k]
[k]

Figure 6.7: Discrete Kalman filter.

which is briefly described in the following. Consider a nonlinear time-variant state-space model
on discrete form

x[ + 1] = f (x[] u[]) + Γw[]

y[] = h (x[]) + v[] (6.83)

In the extended Kalman filter, linearizations of the nonlinear dynamics f [] and measurement
h[] are made about the current state estimate according to
¯
f (x[] u[]) ¯¯
Φ[] = ¯ (6.84)
x[] x[]=x̂[]
¯
h (x[]) ¯¯
C[] = (6.85)
x[] ¯ x[]=x̂[]

Φ[] and C[] may then be substituted for Φ and C in the equations for the discrete Kalman filter
(see above) such that the Kalman filter gain matrix K must be updated for each linearization,
see above. In practice, we often make local linearized models about a set of predefined operating
points and then apply individual precalculated Kalman filters for each model. By using proper
gain scheduling techniques the correct filter model will be applied. The reader is referred to
Bagchi [13] for details.

174
Chapter 7

Mathematical Modeling of
Dynamically Positioned Marine
Vessels

Depending on the operational conditions the vessel models may briefly be classified into station
keeping, low velocity and high velocity models (Figure 7.1). Considering vessel motions in waves
is defined as sea keeping and will here apply both for station keeping (zero speed) and forward
speed. Dynamically positioning (DP) of vessels or thruster assisted position mooring (PM)
of anchored marine vessels can in general be regarded as station keeping and low velocity or
low Froude number applications. This assumption will particularly be used in the formulation
of mathematical models used in conjunction with the controller design. Notice that the ma-
noeuvring models often assumes low-frequency motions only, where the effect of wave-frequency
motions is included afterwards by linear superposition.
Easy access to computer capacity, and the presence of efficient control system design tool-
boxes, such as Matlab, Simulink and others, have motivated more extensive use of numerical
simulations for design and verification of control systems. Essential in being successful in this
is the ability to make sufficiently detailed mathematical models of the actual plant or process.
From an industrial point of view, the same tendency is driven by the fact that control system
safety and performance testing by hardware-in-the-loop (HIL) simulations contribute to reduced
time for tuning during commissioning and sea trials and, not at least, reduced risk for incidents
during operation caused by software bugs and erroneous control system configurations.
In controller design it is important to consider any coupling effects between the different
degrees of freedom and subsystems, in addition to the natural periods (resonances) of the dy-
namic system, e.g. vessel, riser and mooring in open-loop (without control) and closed-loop
(with control). The coupling effects and natural periods may be excited by the environmental
loads or unintentionally by the control system itself if not properly accounted for in the design
process. In Figure 7.2, typical natural periods for ships, semisubmersibles, risers and mooring
system are given in addition to periods of wave loads. The working range or the bandwidth of
a DP system is also indicated.
For the purpose of model-based observer and controller design, it is sufficient to derive
a simplified mathematical model, which nevertheless is detailed enough to describe the main
physical characteristics of the dynamic system. Then, structural information of the physical
properties of the dynamic system is incorporated in the controller design in order to achieve a

175
 • Station keeping models • Manoeuvring models
• Marine operation models • Linearized about some Uo
• Slender structures • Sea keeping
• Multibody operations • Motion damping

High speed tracking/Transit

Low speed tracking

Marked position

Station keeping
Speed [knots]

0 1 2 3 4 5 6 7 …..

Figure 7.1: Modelling properties.

Bandwidth DP System, 15 s ‐>
Closed loop natural periods in surge, sway and yaw: 70‐200 s 

Natural periods in surge and sway for 
moored vessels with water depth 100‐500 m

Riser natural periods, 2000 m
Riser natural periods, 300 m

Ship              Semisubmersible: 
natural periods in heave, roll and pitch

10‐15 s 25‐80 s

15‐40 s
Swell
Wind generated seas Periods [s]
5‐20 s

0 20 40 60 80 100 120 200

Figure 7.2: Periods of interest.

176
better performance and robustness compared to the conventional PID control design methods.
For processes dominated by nonlinearities, it has also been experienced that nonlinear model-
based control design will simplify the overall control algorithm since linearizations about different
working points is avoided.
Hence, the mathematical models may be formulated in two complexity levels:

• Control plant model s a simplified mathematical description containing only the main
physical properties of the process or plant. This model may constitute a part of the
controller. The control plant model is also used in analytical stability analysis based on
e.g. Lyapunov stability and passivity.

• Process plant model or simulation model is a comprehensive description of the actual

process and should be as detailed as needed using high fidelity models. The main purpose
of this model is to simulate the real plant dynamics. The process plant model is used in
numerical performance and robustness analysis and testing of the control systems.

Due to the lack of process knowledge and thereby proper models, control plant models are
often used as process plant models. This may result in bad controller designs, since "perfect"
models are assumed. In Goodwin et al. [95] calibration model is used to describe the process
plant model and nominal model is used to describe the control plant model.
This section will focus on the formulation of process plant models. In this context it means
mathematical models of the vessel dynamics and external forces and moments, in terms of
environmental loads, thruster/propeller forces and mooring forces (if any), acting on the vessel.
Description of the control plant models will be presented in the controller design sections.

7.1 Environmental Models

Environmental models are essential for achieving realistic simulations of a marine vessel. Section
7.3 describes the environmental loads on the vessel, whereas this section describes environmental
models for waves, wind and current.

7.1.1 Waves
Irregular waves are commonly described by a wave spectrum ( ), which in general is a
function of both frequency  and direction  (Myrhaug [194]). The wave spectrum is often
divided in a frequency spectrum () and a directional spreading function ( ), such that

( ) = ()( ) (7.1)

The integral over all frequencies and directions represents the total energy in the sea state 
Z ∞ Z 2
 = ( ) (7.2)
0 0

Linearity is assumed, so that the harmonic wave components extracted from the spectrum may
be superposed.

177
 S(w)

Dwp

wp w

Figure 7.3: Wave spectrum as function of frequency.

Frequency spectrum
The frequency spectrum () describes the energy distribution of the sea state over different
frequencies, with the integral over all frequencies representing the total energy of the sea state.
From a given spectrum, a set of harmonic wave components representing the total energy may
be extracted by choosing a set of frequencies ( ) and frequency intervals (∆  ) such that the
whole area below the spectral curve is covered, see Figure 7.3. In the limiting case, where the
number of wave components  → ∞, and such ∆ → 0, this approaches the Riemann integral
over (). The amplitude   of one wave is determined from the spectral value (  ) by the
formula q
  = 2( )∆  (7.3)
In the following we will use a constant frequency interval ∆. The harmonic waves are assumed
to have phase angles ( ) that are evenly distributed between 0 and 2. Each component is
then described by
  () =   cos(   +  ) (7.4)
Note that these harmonic waves represent the total wave energy with waves in only one direc-
tion. To include different wave directions, a spreading function must be added. The maximum
frequency included in the realization of the wave spectra is denoted  max . The frequency spec-
trum is commonly defined in terms of the significant wave height  and the peak frequency
  .  is the average height of the 1/3 largest waves in the sea state, and   is the frequency
that contains the most energy in the sea state (i.e. defines the peak of the frequency spectrum).
Notice that the wave height is defined as the distance from the bottom of a wave trough to the
top of a wave peak, and hence is twice the wave amplitude.

178
 D(0)

r

(0)[rad]
r

Figure 7.4: Spreading function.

Spreading function
The spreading function ( ) gives the directional distribution of wave energy, spanned about
the mean wave direction   . It is in general a function of both direction and frequency, but is
often simplified to be a function of only direction. The integral over all directions (− ) must
be unity, so that the total wave energy in the sea state is kept unchanged. To obtain ”scaling
factors” for the wave amplitudes in different directions, a set of directions (  ) and direction
intervals (∆  ) must be chosen such that the whole area below the curve is covered, in the same
way as with the frequency spectrum, see Figure 7.4. In the following we will use a constant
direction interval ∆.

Wave spectrum
The wave spectrum ( ), composed of the frequency spectrum and spreading function as in
(7.1), now represents the energy distribution over both frequencies and directions. The integral
over all directions equals the frequency spectrum (), and the double integral over all directions
and frequencies equals the total wave energy
Z ∞ Z 0 + Z ∞
( ) = () (7.5)
0 0 − 0

Figure 7.5 shows a wave spectrum , where each square composed of a ∆ and a ∆ represents
one harmonic wave amplitude. The wave amplitude   is found by a modification of (7.3)
q
  = 2(    )∆∆ (7.6)

One harmonic wave component is now represented by four parameters: Direction   , fre-
quency   , amplitude   and phase angle  . Note that the phase angles of all wave compo-
nents are evenly distributed between 0 and 2. In an Earth-fixed coordinate system, the surface
elevation in the coordinate ( ) will be given as

  (  ) =   sin[   +  −  ( cos   +  sin   )] (7.7)

 is the wave number, which equals 2 , where  is the wave length. For deep water, the
dispersion relation  2 = , where  is the acceleration of gravity, gives the wave number from

179
 n
t io
Fre r ec
q Di
ue
nc
y

Figure 7.5: Wave spectrum as function of frequency and direction.

180
the wave frequency. The coordinate system is defined so that waves with zero heading (  = 0)
are propagating towards north. The total surface elevation of all wave components at the point
( ) at time  for  frequencies and  directions will then be
 q
 X
X
(  ) = 2(     )∆∆ sin(   +  −  ( cos   +  sin   )) (7.8)
=1 =1

Observe that similar equations in e.g. Faltinsen [66] are given with different coordinate system
conventions, and thus differ from the equation above.

Implementation considerations
Several modifications of the wave spectrum defined by (7.8) can be made in order to facilitate
more realistic simulations when it is desired to keep the number of wave components as low as
possible (and hence reduce computational effort).

Random frequencies and directions The expression in (7.8) will repeat itself after the time
 = 2∆. This means that a large number of wave components should be used, but as this
is computationally demanding, a better solution is to choose a random frequency within each
frequency interval (  − ∆2   + ∆2) (Faltinsen [66]). To further increase realism, the
direction may also be chosen at random within each direction interval (  − ∆2   + ∆2).
The number of wave components is according to Faltinsen still recommended to be as high as
1000.

Removing insignificant components If implementing (7.8), with or without random fre-

quencies and directions, the total number of harmonic wave components  is the product of
the number of directions in the grid and the number of frequencies in the grid,  =  ·  .
However, both for high and low frequencies, as well as for directions far from the mean wave
directions, the wave spectrum contains little energy. This means that the wave components
generated from the these areas have small amplitudes, and hence contribute little to the total
sea state.
A way of reducing computational effort is therefore to avoid having too many wave com-
ponents with close to zero amplitude. A first step at solving this is to choose the maximum
frequency of the realization of the spectrum,  max , as low as possible, see Figure 7.3, and thereby
avoid including insignificant wave components from the tail of the frequency spectrum. An easy
way of implementing this for varying peak frequencies in the spectrum is to define a frequency
cutoff factor  such that
max =   (7.9)
 would typically be in the interval (2 3). A second step is to replace the outer limits of the
direction spectrum ( 0 −  0 +) with user-defined limits ( 0 − lim   0 + lim ), see Figure 7.4,
and thereby avoid including insignificant wave components from the outer limits of the direction
spectrum.  lim can be chosen larger for spreading functions with most energy centered about
the mean direction than for spreading functions with much energy far from the mean direction.
A typical maximum value for  lim would be approximately 65◦ .
A more effective way of removing the insignificant wave components is to do a test of the
energy content of each wave component, and then discard the components with little energy.
Two ways of doing this is:

181
 1. Choose a fixed number of waves,  , which naturally must satisfy  ≤  . Find
the energy (     ) of each wave component and sort them in descending order. Then
choose the  first components for the simulation. The upside to this approach is that
you get a user-defined number of waves. The downside is that you have no guarantee of
not loosing any significant wave components. If  =  , all wave components are
used.

2. Choose a relative wave component energy limit  ∈ (0 1) and discard all wave components
which contain less than this ratio of the total sea state energy  :
(     )
  ⇒ component discarded. (7.10)

The upside to this approach is that you are guaranteed that all components with energy
above the user-defined limit is kept. The downside is that you don’t have control over the
number of waves. If  = 0, all wave components are used.

Common wave spectra

Common frequency spectra are the Pierson-Moskowitz (PM) spectrum, the ITTC/ISSC spec-
trum , the JONSWAP spectrum. and the more recent doubly peaked spectrum by Torsethaugen.

PM/ITTC/ISSC spectra The PM type spectra have the general form

 
() = 5
exp(− 4 ) (7.11)
 
They apply to fully developed sea states in open seas. The PM spectrum is given from wind
speed  at 19.5 meters height

 = 00081 2 ;  = 074( )4  (7.12)

The peak frequency of the spectrum,  , may be used in place of the wind speed, giving
5
 = 00081 2 ;  = 4  (7.13)
4
The ITTC and ISSC spectra are identical for open seas, and given by the significant wave height
 and peak wave frequency  

 = 0312  4 ;  = 1254  (7.14)

 and   are interdependent and must be chosen with care.

JONSWAP spectrum The JONSWAP spectrum may according to Myrhaug [194] be for-
mulated as
2 5  −
exp[− 12 (   )2 ]
() =  5 exp[− ( )4 ]   (7.15)
 4 
  and  are parameters to be determined, and  is the peak frequency of the spectrum.  is
a peak parameter, and may vary between 1 and 7. For  = 00081 and  = 1 the JONSWAP
spectrum is identical to the PM spectrum. The JONSWAP spectrum was developed after

182
measurements in an area of the North Sea which is relatively shallow and close to land. It is
a spectrum for not fully developed seas, with a much sharper peak than the PM type spectra.
Increasing  gives a sharper peak of the spectrum. Faltinsen [66] recommends the following
values for the JONSWAP parameters
½
007  ≤   2  4
 = 33  = ;  = 02 2  (7.16)
009  ≥   

To keep within the validity area of the spectrum, Myrhaug [194] suggests the following constric-
tion on  and   p p
125      175   (7.17)
If no information on  is available, DNV proposes the following  based on  and   :
2
= √ 
 

 ≤ 36 ⇒  = 5
 ≤ 50 ⇒  = (575−115)  (7.18)
  50 ⇒  = 1

Doubly peaked (Torsethaugen) spectrum The spectra presented above may all be termed
wind generated, and have been criticized for not giving a good representation of the low-
frequency wave energy, or swell. Torsethaugen [314] developed a doubly peaked wave spectrum;
a low frequency peak due to swell, and a high frequency peak due to locally wind generated
waves. The spectrum was developed by curve fitting of experimental data from the North Sea,
and was standardized under Norsok Standard [207]. All parameters are found by the significant
wave height and spectral peak period. The total wave spectrum is a sum of two spectral peaks,
called primary and secondary. For fully developed seas, the secondary peak vanishes, and the
wave energy is centered in a narrow peak about the spectral peak period  = 2  . For the
same significant wave height  , primary peak periods higher than the value for fully developed
seas can only be due to swell components, with a secondary peak of locally wind generated
waves. For lower primary peak periods, the total energy cannot be set up by local winds alone,
and an additional secondary swell peak is needed. For   66 , the primary peak is in
the swell generated, and for   66 , the primary peak is in the wind generated. When
the sea state is dominated by swell, the shape of the wave specter will differ significantly from
the JONSWAP and PM wave spectra The formulation of the doubly peaked spectrum is not
presented here, but it is implemented in the Marine Systems Simulator [195] GNC Toolbox as
[ ] = (     max   ), where  max is the maximum frequency and  the number of
wave components.

Comparison of the spectra Figure√7.6 shows a comparison of the PM, ITTC and JONSWAP
spectra, all for  = 6 and   = 15  ≈ 061. The PM spectrum does not account for the
actual wave height, and contains less energy than the two others. The ITTC and JONSWAP
spectra contains approximately the same amount of energy, but the JONSWAP spectrum has
more of this energy close to   . Figure 7.7 shows how the doubly peaked spectrum varies with
the peak frequency for constant  . The two spectra with thick lines are examples of a primary
swell spectrum (largest peak for low frequencies) and a primary wind-generated spectrum.

183
 12
PM
ITTC
10 JONSWAP

8 S [m s]
2

2
ω [rad/s]
0
0 0.5 1 1.5 2

Figure 7.6: Comparison of PM, ITTC and JONSWAP spectra.

wp

Figure 7.7: Doubly peaked wave spectrum for varying   and constant  

184
Linear wave theory limitations Linear wave theory only considers non-breaking waves. A
rule-of-thumb for modelling of non-breaking waves is

 ≤ 17
⇓
 ≥ 7 (7.19)

where  is the wave height (peak-to-peak) and  the wave length.  is related to the wave
number  by
1
=  (7.20)

In deep water, the dispersion relation is

2
=  (7.21)

where  is the wave frequency and  the acceleration of gravity. This gives
r r r
p   14
 =  = ≤ ≈ 
 7 

In terms of the wave amplitude  = 2, we get the maximum valid wave frequency from the
wave amplitude r r
 07
≤ ≈ 
14 
or equivalently the maximum valid wave amplitude from the wave frequency
 07
≤ ≈ 2
142 
Defining the wave length to height ratio as , the corresponding wave frequency can be calculated
from  and 

 = 

r

 = 


Common spreading functions

The most commonly used spreading function is according to Myrhaug [194] formulated as
( 2−1
2 !(−1)!
(2−1)! cos2 ( −  0 ) for − 2  ( −  0 )  2
( −  0 ) =  (7.22)
0 elsewhere

 is here an integer, with  = 1 recommended by ITTC and  = 2 recommended by ISSC. Figure

7.8 shows the spreading function for varying values of . Increasing  centers the wave energy
more about the mean direction  0 .

185
 2

D(y-y0)
1.8

1.6

s=10
1.4

1.2

0.8

0.6

s=6
0.4
s=2
0.2
s=1
(Y-y0)[rad]
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

Figure 7.8: Example of spreading function realizations.

Implementation examples
The following figures show how the number of waves can be decreased by the methods described
above. The wave spectrum is composed of a Torsethaugen frequency spectrum with significant
wave height  = 4 and peak frequency  = 1, and a spreading function with mean
wave direction  0 = −30◦ and spreading factor  = 4. The number of frequencies  and number
of directions  were kept constant at  = 20 and  = 10 for all realizations. Figure 7.9 shows
the wave spectrum for frequency cutoff factor  = 4 and the direction limit  lim = 0. Clearly,
a lot of insignificant components are included for high frequencies and far from the mean wave
direction. If instead choosing  = 25 and the direction limit  lim = 20◦ , a much better resolution
in the interesting area (i.e. about the peak frequency and mean direction) is obtained, as can
be seen in Figure 7.10. The next step is to reduce the number of wave components. Figure 7.11
shows the chosen wave components as red stars when choosing the wave component energy limit
 = 0005. A total of 96 wave components is included and 104 discarded. If in addition choosing
the frequencies and directions at random within each interval the result is shown in Figure
7.12. To justify the exclusion of insignificant wave components, examples of surface realizations
with varying number of wave components are shown in Figures 7.13 to 7.15. Figure 7.13 shows
a realization of the spectrum defined above with all 200 wave components included, Figure
7.14 shows a realization of the same spectrum with the 100 most important wave components
and Figure 7.15 shows a realization of the same spectrum with the 50 most important wave
components. The difference between the realizations with 200 and 100 components is minimal,
and yet the computational effort for the latter is only half. The difference is noticeable for the
realization with 50 components, but the major waves are clearly unaltered; the difference lies in
the smaller ripples in the surface due to the smaller wave components.

186
Figure 7.9: Torsethaugen frequency spectrum with  = 4 and  = 1, spreading
function with  0 = −30◦ and  = 4, number of frequencies and directions  = 20 and  = 10
frequency cutoff factor  = 4 and wave direction limit  lim = 0.

Figure 7.10: Torsethaugen frequency spectrum with  = 4 and   = 1, spreading

function with  0 = −30◦ and  = 4, number of frequencies and directions  = 20 and  = 10
frequency cutoff factor  = 25 and wave direction limit  lim = 20◦ .

187
Figure 7.11: Wave spectrum with wave energy limit  = 0005. The frequencies and directions
of the components, which are shown as red stars, are not chosen at random.

Figure 7.12: Wave spectrum with wave energy limit  = 0005. The frequencies and directions
of the components, which are shown as red stars, are chosen at random.

188
Figure 7.13: Realization of a Torsethaugen spectrum with 200 wave components.

Figure 7.14: Realization of a Torsethaugen spectrum with 100 wave components

189
Joint distribution for wind and waves
The parameters describing the wind and wave state are interdependent. Johannessen et al.
[132] have developed a joint distribution for wind (described by the average wind speed 10
meters above sea level  10 ) and waves (described by significant wave height  and peak period
 = 2  ) in the northern North Sea, based on measurements in the 1973-99 period. Here we
present only the mean peak period and mean wind speed as function of significant wave height,
for use as a rule of thumb in modelling

( ) = 4883 + 2680054  (7.23)

( 10 ) = 1764 + 3426078  (7.24)

7.1.2 Wind
Wind is commonly divided in two components; a mean value and a fluctuating component, or
gust. The mean component decreases with the distance to the ground, whereas the gust is
approximately constant with the distance to the ground. Wind is in reality a three-dimensional
phenomenon, but the descriptions commonly used are restricted to velocities in the horizontal
plane, parameterized by the velocity  and the direction .

Mean wind component

The mean velocity ̄ at elevation  may be written as

̄ () 5√  2
=  ln ; 0 = 10 exp(− √ ) (7.25)
̄10 2 0 5 

̄10 is the 1 hour mean wind speed at 10 elevation and  is the sea surface drag coefficient
(Myrhaug [194]). Slowly-varying variations in the mean wind velocity may be implemented by
a 1st order Gauss-Markov Process (Fossen [80])
·
̄ + ̄ =  (7.26)

 is Gaussian white noise and  ≥ 0 is a constant. For  = 0, this is a random walk process.
The magnitude of the velocity should be restricted by saturation elements

0 ≤ ̄min ≤ ̄ ≤ ̄max  (7.27)

The variation in wind direction may be implemented in a similar way

̇ + 2  = 2 (7.28)
 min ≤  ≤  max  (7.29)

where 2 and 2 are white noise and a positive constant respectively. No saturation elements
are necessary here, but may be implemented if wanted.

190
Wind gust
The wind gust is commonly described by a spectrum, in a similar way as for waves. A widely
used formulation is the Harris wind spectrum
4 10 
( ) = 5  e =  (7.30)
(2 + f2 ) 6  10
 is a scaling length,  is the sea surface drag coefficient and  is the frequency in . Example
values for the parameters are  = 1800 and  = 00026. The Harris spectrum was based on
measurements over land, and more resent studies have given alternative representations. Norsok
Standard [207] recommends a wave spectrum which also varies with 
( 1010 )2 ( 10
 045
)
( ) = 320 5   = 0468; (7.31)
(1 +  ) 3
 10 −34
 = 172 ( )23 ( ) 
10 10
Figures 7.16 shows the Harris and NORSOK wave spectra for  10 = 10. It is clear that
the NORSOK spectrum contains more energy at lower frequencies.
A realization of the wind state may be done by superposing the gust realization and the
mean wind at the desired elevation. The gust realization is done in the same manner as a the
realization of a sea state from a wave spectrum. Harmonic component number  is defined as
p
 () = 2( )∆ cos(2  +  ) (7.32)
where  is the frequency, ∆ the frequency interval and  an evenly distributed phase angle.
The total wind realization with  gust components is then written as

X
 ( ) = ̄ () +  () (7.33)
=1

Implementation example Figure 7.17 shows a wind time series using the models defined
above. The wind direction is slowly-varying with mean 0 degrees, the mean wind speed is 10 m/s
and the wind gust is modelled by a NORSOK wind spectrum with 100 frequency components.

7.1.3 Water Current Model

We may divide current modelling in two levels of detail:

• Surface current, for use in modelling of surface vessel response

• Full current profile, for use in modelling of risers, anchor lines etc.

Surface current
For modelling of surface vessels or other applications in the proximity of the surface, a 2-
dimensional current model is sufficient. If the current is given by magnitude  and direction in
the NED frame   , the current velocity vector   may be written as
  = [ cos(  )  sin(  ) 0] (7.34)

191
Figure 7.15: Realization of a Torsethaugen spectrum with 50 wave components

200
Harris
NORSOK, z=3
150 NORSOK, z=6
NORSOK, z=10
S [m /s /Hz]
2

100
2

50

f [Hz]
0
−4 −3 −2 −1 0
10 10 10 10 10

Figure 7.16: Harris and NORSOK wind spectra.

192
 4

angle [deg]
0

−2

−4
0 50 100 150 200 250 300 350 400

14
speed [m/s]

12

10

8
0 50 100 150 200 250 300 350 400
time [s]

Figure 7.17: Wind realization of a slowly-varying direction with mean 0 degrees, mean speed of
10 m/s and gust components from the NORSOK wind spectrum.

The variation in current velocity may be implemented by a 1st order Gauss-Markov Process
(Fossen [80])
̇ +  =  (7.35)
 is Gaussian white noise and  ≥ 0 is a constant. For  = 0, this is a random walk process.
The magnitude of the velocity should be restricted by saturation elements

min ≤  ≤ max  (7.36)

The variation in current direction may be implemented in a similar way

̇  + 2   = 2 (7.37)
 min ≤   ≤  max  (7.38)

In addition, the current is often divided in two components: Tidal and wind generated. If actual
measurement data are available, these may be separated, but for most purposes they may be
lumped into one component.

Current profile
In some applications, the actual current profile in which the current varies with the depth, is
needed. If no actual field measurements are available, DNV recommends the following current

193
profile  (), with  the depth (positive downwards)
 () =  () +  () (7.39)
 −  17
 () =  ( ) for  ≥ 0 (7.40)

0 −  17
 () =  ( ) for 0 ≤  ≤ 0 (7.41)
0
 () = 0 for  ≥ 0  (7.42)
where  is the tidal current velocity at surface level,  is the wind generated current
velocity at surface level,  is the water depth, and 0 is the reference depth for wind generated
current, example value 0 = 50
The wind generated current may be taken as  = 0015 10 , where  10 is the mean
wind velocity 10 meters above sea level. In addition, the current profile should be stretched or
compressed vertically to account for the change in water depth with the surface elevation due
to waves. This may however be neglected at large water depths.

7.2 Kinematics
7.2.1 Reference Frames
The different reference frames used are illustrated in the Figures 7.18 - 7.19, and are described
below:
• The Earth-fixed reference frame is denoted as the E E E -frame. Measurement of the
vessel’s position and orientation coordinates are done in this frame relatively to a defined
origin. One should notice that each position reference system has its own local coordinate
system, which has to be transformed into the common Earth-fixed reference frame.
• In sea keeping analysis (vessel motions in waves) the hydrodynamic frame h h h -frame
is generally moving along the path of the vessel with the -axis positive forwards, -axis
positive to the starboard, and -axis positive downwards. The h h -plane is assumed
fixed and parallel to the mean water surface. The vessel is assumed to oscillate with
small amplitudes about this frame such that linear theory may apply for modelling of
the perturbations. Often in forward speed sea keeping analysis the hydrodynamic frame
is moving forward with constant vessel speed  . In station keeping operations (dynamic
positioning) about the coordinates d  d  and  d the hydrodynamic frame is Earth-fixed
and denoted as the reference-parallel frame R R R  It is rotated to the desired heading
angle  d  and the origin is translated to the desired d and d position coordinates for the
particular station keeping operation studied. Assuming small amplitudes of motion it is
convenient to use this frame in the development of the control schemes.
• The body-fixed  -frame is fixed to the vessel body with the -axis positive forwards,
-axis positive to the starboard and -axis positive downwards. For ships it is common to
assume that the centre of gravity is located in the centre line of the vessel, and that the
submerged part of the vessel is symmetric about the - plane (port/starboard). Here it
is assumed that the origin is located in the mean oscillatory position (flotation point) in
the average water plane. Hence, the centre of gravity is then located at (G  0 G ) in body
coordinates. The motion and the loads acting on the vessel are calculated in this frame.

194
 The vectors defining the generalized vessel’s Earth-fixed position and orientation, and the
body-fixed translation and rotation velocities are using SNAME [268] notation given by

η 1 = [  ] , η 2 = [  ] 

(7.43)
ν 1 = [  ] , ν 2 = [  ] 

Here, η 1 denotes the position vector in the Earth-fixed frame, and η 2 is a vector of Euler angles.
ν 1 denotes the body-fixed linear surge, sway and heave velocity vector, and ν 2 denotes the
body-fixed angular roll, pitch and yaw velocity vector. For surface vessels the orientation is
normally represented in terms of Euler angles (Fossen [78]).

Definition 7.1 Degrees-of-Freedom - DOF. For a general body,the DOF is the set of indepen-
dent displacements and rotations that completely specify the displaced position and orientation
of the craft. A body that can move freely in the 3D space has maximum 6 DOFs with three
translational and three rotational components.

Definition 7.2 Configuration Space. The n-dimensional configuration space is the space of
possible positions and orientations that a body may attain subject to external constraints.

Ships and rigs are often described in the horizontal plane only with surge sway and yaw
(with  = 3 DOFs), while underwater vehicles may be described in n = 6 DOFs including both
the horizontal (surge, sway, yaw) plane and vertical plane (heave, roll, pitch)

Definition 7.3 Working Space The working space is a reduced space of dimension m  n in
which the control objective is defined.

An underactuated vehicle has independent control forces and moments in only some DOFs
such that   . For an underwater vehicle such as a Remotely Operated Vehicle (ROV) that is
self-stabilized in roll and pitch, the working space may be  = 4 If the number of actuators 
are less than  the the control problem is referred to as underactuated control Stabilizing and
tracking controllers for underactuated vehicles are usually designed by considering a working
space of dimension    satisfying  =  (fully actuated in the working space but not in the
configuration space). This may be the case for the ROV. Notice that many ships and underwater
vehicles are overactuated with    making the control allocation to an optimization problem
with infinite many solutions.

7.2.2 The Euler Angle Transformation

Linear velocity
The Euler angles result from a sequence of rotations which is not arbitrary. The Euler angles
appear from the following rotation sequence:

• Let 3 3 3 be a Earth-fixed reference frame that is translated such that the origin is
coinciding with the body-fixed  -frame.

• Rotate 3 3 3 a yaw angle  about the 3 -axis such that a new reference frame denoted
as 2 2 2 appears.

195
 v (sway) q (pitch)
Y

r (yaw)

u (surge) p (roll)
w (heave)
X
Z

Figure 7.18: Definition of surge, sway, heave, roll, pitch and yaw modes of motion in body-fixed
frame.

Vc
c  X
w V
w

d

( x , y)
XE
( xd , yd )

YR

YE

Figure 7.19: Definition of frames: Earth-fixed, reference-parallel and body-fixed.

196
 • Rotate 2 2 2 a pitch angle  about the 2 -axis such that a new reference frame denoted
as 1 1 1 appears.

• Rotate 1 1 1 a roll angle  about the 1 -axis such that the reference frame denoted as
  appears.

The rotation sequence (Fossen [78]) is then given by

J1 (η 2 ) = C C C  (7.44)

where
⎡ ⎤
  0
C = ⎣ −  0 ⎦  (7.45)
0 0 1
⎡ ⎤
 0 −
C = ⎣ 0 1 0 ⎦ (7.46)
 0 
⎡ ⎤
1 0 0
C = ⎣ 0   ⎦  (7.47)
0 − 

Hence, the rotation matrix J1 (η2 ) ∈ SO(3) can be defined as

⎡ ⎤
 − +   + 
J1 (η 2 ) = ⎣   +  − +  ⎦  (7.48)
−  

where  · = cos(·),  · = sin(·) and  · = tan(·). The symbol SO(3) denotes special orthogonal
group of order 3. Thus, the linear velocities of the vessel in the Earth-fixed frame is given by
the transformation
η̇1 = J1 (η 2 )ν 1  (7.49)

Remark 7.1 Two vectors x1  x2 ∈ ×Rn are said to be orthogonal if x1 x2 = x2 x1 = 0. A set
of vectors x ∈ ×Rn   = 1 2  , is said to be orthonormal if
½
 0   6= 
x x = (7.50)
1   = 

Remark 7.2 The reference frame matrix C ∈ SO(3) has the property (Fossen [78])

C C = C C = I (7.51)

where C denotes a rotation angle  about the -axis, and where

det C = 1 (7.52)

£ ¤
implies that C = [c1  c2  c3 ] and C = c1  c2  c3 consist of orthonormal columns with the
consequence that
C−1
 = C 

(7.53)

197
Remark 7.3 Based on (7.51) we can derive that
J−1 
1 (η 2 ) = J1 (η 2 ) (7.54)
Hence,
ν 1 = J−1 
1 (η 2 )η̇ 1 = J1 (η 2 )η̇ 1  (7.55)

Angular velocity rotation

The angular velocities of the vessel in the Earth-fixed frame is given by
η̇2 = J2 (η 2 )ν 2  (7.56)
The angular velocities in the body-fixed frame appear from the relation (Fossen [78])
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
̇ 0 0
ν 2 = ⎣ 0 ⎦ + C ⎣ ̇ ⎦ + C C ⎣ 0 ⎦ = J−1 2 (η 2 ) η̇ 2  (7.57)
0 0 ̇
where we can expand (7.57) and derive that
⎡ ⎤
1 0 −
J−1
2 (η 2 ) =
⎣ 0   ⎦  (7.58)
0 − 
Inverting (7.58) gives J2 (η 2 ) ∈ R3×3 according to
⎡ ⎤
1  
J2 (η 2 ) = ⎣ 0  − ⎦   6= 0 (7.59)
0  
where  · = cos(·),  · = sin(·) and  · = tan(·).

6 DOF kinematics
The linear and angular velocities of the vessel in the Earth-fixed frame is given by
∙ ¸ ∙ ¸∙ ¸
η̇ 1 J1 (η 2 ) 03×3 ν1
η̇ = = = J(η 2 )ν (7.60)
η̇ 2 03×3 J2 (η 2 ) ν2

6 DOF kinematics (small roll and pitch angle representation)

A frequently used kinematic approximation for metacentric stable vessels with small roll and
pitch angles is
η̇ = J(η 2 )ν (7.61)

⇓ η 2 = [0 0 ] 
η̇ = J()ν (7.62)
where ∙ ¸
C 03×3
J() =  (7.63)
03×3 I3×3
Remark 7.4 For small -roll and -pitch angles (less that 10◦ ), we approximate cos  ≈ 1 and
cos  ≈ 1 and sin  ≈ 0 and sin  ≈ 0.

198
3 DOF kinematics (horizontal motion)
In many practical applications only the horizontal modes of motion are of interest. It is then
convenient to find an appropriate kinematic representation only for the horizontal-plane motion.
For surge, sway and yaw the 6 DOF kinematics reduces to

η̇ = R()ν R−1 () = R () (7.64)

where we have redefined the state vectors according to η =[  ] and ν = [  ]  and
⎡ ⎤
 − 0
R() = C = ⎣   0 ⎦ 
0 0 1
In a reference-parallel formulation we can define

η̇  = R( −   )ν (7.65)

Remark 7.5 In hydrodynamic literature several conventions for hydrodynamic coefficients and
reference frames are used. In Faltinsen [66], added mass, wave radiation damping and restoring
coefficients are denoted as      with the -axis positive backwards, -axis positive to the
starboard and -axis positive upwards. In Newman [199] a second convention is used with the
-axis positive forwards, -axis positive upwards and -axis positive to the starboard. Here, the
notation of SNAME [268] is used with the -axis positive forwards, -axis positive to the starboard
and -axis positive downwards. The SNAME notation is more established in hydrodynamic and
control problems related to manoeuvring and positioning of ships.
R
Remark 7.6 Integration of the angular velocities  ν 2 ( ) does not have any physical mean-
ing. In order to have proper generalized coordinates η̇ 2 should be integrated.

Remark 7.7 J2 (η2 ) is undefined for  = ±90◦ . Consequently J−1 

2 (η 2 ) 6= J2 (η 2 ) due to the
◦
singularity in  = ±90 . For surface vessels this is not of any problem. However, for underwater
vehicles special precaution must be taken. Then a so-called quaternion (Euler parameters) could
be used, see Fossen [78] for details.

Remark 7.8 For low speed manoeuvring and change of setpoint in station keeping operations,
the R R R —frame will be time-varying following the trajectories for d () d () and  d ().

Remark 7.9 In the reference-parallel frame R R R (and hydrodynamic frame as well) as-
suming small amplitudes of motion in yaw (| −   | ≈ 0), we have

η̇  = R( −   )ν ≈ I3×3 ν (7.66)

where I3×3 ∈ R3×3 is the identity matrix. Thus, the reference-parallel frame is convenient for
design of linear controllers as we avoid the nonlinear kinematics.

Remark 7.10 When considering moored structures, such as turret-moored tankers or moored
semi-submersibles, it is common to locate the origin of the Earth-fixed frame in the natural
equilibrium point for the mooring system. This position is often referred to as the field zero
point (FZP). The body-fixed frame is located in the geometrical centre of the mooring system on
the structure; for turret-moored ships this will be the centre of turret (COT).

199
 Low-frequency (LF) and wave-frequency (WF) motion

Total motion, LF + WF
LF motion

WF motion

0 50 100 150
time

Figure 7.20: The total motion of a ship is modeled as a LF response with the WF response
added as an output disturbance.

7.3 Vessel Dynamics

In mathematical modeling of the marine vessel dynamics it is common to separate the total model
into a low-frequency (LF) model and a wave-frequency (WF) model by superposition. Hence,
the total motion is a sum of the corresponding LF and the WF components, see Figure 7.20. The
WF motions are assumed to be caused by first-order wave loads. Assuming small amplitudes
these motions will be well represented by a linear model. The LF motions are assumed to be
caused by second-order mean and slowly varying wave loads, current loads, wind loads, mooring
and thrust forces. These motions are generally nonlinear, but linear approximations about
certain operating points can be found.
One should notice that for surface vessels, only the horizontal motions i.e. the surge, sway
and yaw degrees-of-freedom (DOF), are subject for control. In the design of positioning control
systems one should expect that it is sufficient to only consider the horizontal-plane dynamics in
the LF model. This is normally an appropriate assumption, where the effect of vertical-plane
dynamics i.e. the heave, roll and pitch DOF will be of minor practical interest for the control
problem. However, according to Sørensen and Strand [290] under certain circumstances this
assumption may cause unacceptable reduction in the control performance. In their work it was
suggested to characterize the control problem into two categories whether the natural periods
in roll and pitch are within or outside the bandwidth of the positioning controller. Hence, the
modelling problem can either be regarded as a three or a six DOF problem, where the following
prerequisites can be made:

• 3 DOF: For conventional ship and catamaran hulls in the low-frequency model only the
three horizontal-plane surge, sway and yaw DOF are of practical interest for the controller
design. For those vessels it can be assumed that the LF vertical-plane dynamics and the
thruster action will not have any mutual influence on each other.

200
 • 6 DOF: For marine structures with a small-waterplane-area and low metacentric height,
which results in relatively low hydrostatic restoring compared to the inertia forces, an
unintentional coupling phenomenon between the vertical and the horizontal planes through
the thruster action can be invoked. Examples are found in semi-submersibles and SWATHs,
which typically have natural periods in roll and pitch in the range of 35 − 65 seconds. If
the inherent vertical damping properties are small, the amplitudes of roll and pitch may
be emphasized by the thruster’s induction by up to 2◦ -5◦ in the resonance range. These
oscillations have caused discomfort in the vessel’s crew and have in some cases limited the
operation. Hence, both the horizontal and vertical planes DOF should be considered.

Concerning the WF model it is normal to include all 6 DOF no matter how the LF model is
formulated. The vertical-plane WF motions (heave, roll and pitch) must be used to adjust the
acquired position measurements to some defined origin on the vessel. This could be in the plane
of the undisturbed free surface with the -axis through the centre of gravity (CG), the centre of
buoyancy (CB) or the flotation point. Thus, the raw position reference system measurements
will be strongly influenced by the vertical-plane motions depending on the locations of the
installed GPS antennas, hydroacoustic position reference system transducers, Artemis receivers,
Taut wire, etc. relative to the defined origin.

7.3.1 Nonlinear Low-Frequency Vessel Model

The nonlinear 6 DOF body-fixed coupled equations of the LF motions in surge, sway, heave,
roll, pitch and yaw are written as follows

Mν̇ + CRB (ν)ν + CA (ν r )ν r +D( ν r ) + G(η) = τ env + τ moor +  ice + τ thr  (7.67)

The right-hand expression of (7.67) represents generalized external forces acting on the vessel
and is treated later in this section. Forces in surge, sway and heave and moments in roll, pitch and
yaw are referred to as generalized forces. τ env ∈ R6 represents the slowly-varying environmental
loads with the exception of current loads acting on the vessel. The effect of current is already
included on the left hand side of (7.67) by the introduction of the relative velocity vector.
τ thr ∈ R6 represents the generalized forces generated by the propulsion system. Even if only
the horizontal-plane surge, sway and yaw DOF are subject for control, geometrical coupling
to the vertical-plane heave, roll and pitch DOF, will be invoked due to the actual locations of
the thrusters. As discussed above the produced thrust components in the vertical-plane will be
important to consider for marine structures with small-waterplane-area. If the vessel is attached
to a mooring system, the effect of this is represented by τ moor ∈ R6 . For operation in ice the
corresponding loads from level ice, ice floes and ice ridges are modeled by τ ice ∈ R6 . In Nguyen
et al. [204] models of level ice loads are presented.

201
Generalized inertial forces, Mν̇ :
The system inertia matrix M ∈ R6×6 including added mass is defined as
⎡ ⎤
 −  ̇ 0 − ̇ 0   − ̇ 0
⎢ 0  −  ̇ 0 −  − ̇ 0   − ̇ ⎥
⎢ ⎥
⎢ − ̇ 0  −  ̇ 0 − − ̇ 0 ⎥
M=⎢ ⎢
⎥
⎥
⎢ 0 −  − ̇ 0  − ̇ 0 −  − ̇ ⎥
⎣   − ̇ 0 − − ̇ 0  − ̇ 0 ⎦
0  − ̇ 0 −  − ̇ 0  − ̇
(7.68)
where  is the vessel mass,  ,  and  are the moments of inertia about the -, - and
-axes and  =  are the products of inertia. The zero-frequency added mass coefficients
̇  ̇  ̇  ̇  and so on at low speed in surge, sway, heave, roll, pitch and yaw due to accel-
erations along the corresponding and the coupled axes are defined as in Faltinsen [66]. Hence,
it can be shown that the system inertia matrix is symmetrical and positive definite (Appendix
A), i.e. M = M  0 and Ṁ = 0.

Generalized Coriolis and centripetal forces, CRB (ν)ν + CA (ν r )ν r :

The matrix CRB (ν) ∈ R6×6 is the skew-symmetric Coriolis and centripetal matrix of the rigid
body written (Fossen [78])
⎡ ⎤
0 0 0 41 −51 −61
⎢ 0 0 0 −42 52 −62 ⎥
⎢ ⎥
⎢ 0 0 0 −43 −53 63 ⎥
CRB (ν) = ⎢
⎢
⎥
⎥ (7.69)
⎢ −41 42 43 0 −54 −64 ⎥
⎣ 51 −52 53 54 0 −65 ⎦
61 62 −63 64 65 0

where
41 =   42 =  43 =  (  − )
51 =  (  − ) 52 = (  +  ) 53 =  (  + ) 54 =   −  
(7.70)
61 =  ( +  ) 62 = − 63 =   64 =  
65 =   +  

Wichers [326] divided the effect of current into two parts: the potential part and the viscous
part. The Coriolis and centripetal matrix of the added mass including the potential part of the
current load is formulated according to
⎡ ⎤
0 0 0 0 −51 −61
⎢ 0 0 0 −42 0 −62 ⎥
⎢ ⎥
⎢ 0 0 0 −43 −53 0 ⎥
⎢
CA (ν r ) = ⎢ ⎥ (7.71)
⎢ 0 42 43 0 −54 −64 ⎥ ⎥
⎣ 51 0 53 54 0 −65 ⎦
61 62 0 64 65 0

202
where
42 = −̇  − ̇  − ̇  43 = ̇  + ̇  + ̇ 
51 = ̇  + ̇  + ̇  53 = −̇  − ̇  − ̇  54 = ̇  + ̇  + ̇ 
61 = −̇  − ̇  − ̇  62 = ̇  + ̇  + ̇  64 = ̇  + ̇  + ̇ 
65 = ̇  + ̇  + ̇ 
(7.72)
Notice that the so-called Munk moments appear from the expression CA (ν r )ν r , see Faltinsen
[66] and Newman [199].

Generalized damping and current forces, D( ν r ) :

The damping vector may be divided into a nonlinear and a linear component according to

D( ν r )= DL ( ν r )ν r +dNL (ν r   r ) (7.73)

The linear damping is assumed to vanish for increasing speed as the flow becomes turbulent. In
order to incorporate this effect the linear damping is multiplied with an an exponential decaying
functions according to:
⎡ ⎤
 −| |   −||
DL ( ν r )= ⎣    ⎦
  −| |    −||

where  is a positive constant such that  ∈ R+ 

Furthermore, the effect of current load is normally included in the nonlinear damping term
by the definition of the relative velocity vector according to:
£ ¤
νr =  −   −       (7.74)

The horizontal current components in surge and sway are defined as:

 =  cos (  − ) ,  =  sin (  − )  (7.75)

where  and   are the current velocity and direction respectively, see Figure 7.19. Notice that
the current velocity components in heave, roll, pitch and yaw are not considered. The total
relative current velocity is then defined as for  =  −  , and  =  −  according to
p
 = 2 + 2  (7.76)

The relative drag angle is found from the following relation:

 r = atan2(−  − ) (7.77)

where 2 is the four quadrant arctangent function of the real parts of the elements of 
and  , such that − ≤atan2( ) ≤  The nonlinear damping is assumed to be caused by
turbulent skin friction and viscous eddy-making, also denoted as vortex shedding, Faltinsen [66]
and Faltinsen and Sortland [67].

203
Nonlinear damping and current forces Assuming small vertical motions, the 6-dimensional
nonlinear damping vector is often formulated as:
⎡ ⎤
 ( r )| |
⎢  ( r )| | ⎥
⎢ ⎥
⎢  ( r )|| ⎥
⎢
dNL (ν r  r ) = 05  ⎢ ⎥ (7.78)
2 ⎥
⎢   ( r )|| +   ( r )| | ⎥
⎣   ( )|| −   ( )| | ⎦
r r
  ( r )| |

where  ( r ),  ( r ),  ( r ),  ( r )  ( r ) and  ( r ) are the nondimensional drag
coefficients found by model tests for the particular vessel with some defined location of the origin.
 is the breadth,  is the density of water,  is the length between the perpendiculars, and 
is the draft. The second contributions to roll and pitch are the moments caused by the nonlinear
damping and current forces in surge and sway, respectively, attacking in the corresponding
centers of pressure located at  and  .
For relative current angles |  − |  0 the cross flow principle (Faltinsen [66]) may be
applied to calculate the nonlinear current loads in sway 2 and yaw 6 in (7.78). In sway the
cross flow formulation is written
Z
2
2 = 05 () ( r  ) ( + ) | + | (7.79)


where () is the longitudinal varying draft, and  2 (  ) is the 2 dimensional drag coefficient.
r
Similar as for sway, the cross flow expression in yaw is written
Z
2
6 = 05 () ( r  ) ( + ) | + | (7.80)


For simplicity let us assume that () = , and that the drag coefficient is constant over
2 (  ) =  2 . Typical  2 values for ship (varying between
the entire vessel length, that is  r  
045 −14 dependent on Reynolds number), see Figure 6.16 in Faltinsen [66]. As shown in Figure
6.20 on page 197 in Faltinsen [66] for slender bodies the drag coefficient may be adjusted for
3-dimensional effects due to the vertical vortex system at ship ends.
Let the origin be located in the middle of the ship, such that the distance to the bow is
 2 and to the stern is − 2. Hence, a solution of (7.79) may be found to be
Z
2
2 = 05  ( + ) | + | =

⎛ 
⎞
R 2 ¡ ¢
⎜ sgn( )|2 +  | + sgn()|  + 2 2 | + ⎟
2 ⎜ 0 ⎟
05  ⎜ R0 ¡ ¢ ⎟=
⎝ 2 2 2 ⎠
sgn( )| +  | + sgn()|  +   | 
− 2
⎛ h¯ ¯ ¯ ¯i ⎞
¯  2 2 ¯ ¯  2 2 ¯
2 ⎝
sgn( ) ¯ 2  + 8  ¯ + ¯ 2  − 8  ¯ +
05  h¯ 2 ¯ ¯ ¯i ⎠  (7.81)
¯  3 2 ¯ ¯ 2 3 2 ¯
sgn() ¯ 8   + 24  ¯ − ¯− 8   + 24  ¯

204
sgn is the sign function defined such that sgn() = 1 for  ≥ 0 and sgn() = −1 for   0.
Notice that sgn() = sgn() sgn().
Similarly, (7.80) is rewritten
Z
2
6 = 05   ( + ) | + | =

⎛ 
⎞
R 2 ¡ ¢
⎜ sgn( )|2  +  2 | + sgn(2 )|  2
+ ⎟ + 2 3 |
2 ⎜ 0 ⎟
05  ⎜ R0 ¡ ¢ ⎟=
⎝ 2 2 2 2 2 3 ⎠
sgn( )| +   | + sgn( )|  +   | 
− 2
⎛ h¯ 2 ¯ ¯ 2 ¯i ⎞
¯  2 3 ¯ ¯  2 3 ¯
2 ⎝
sgn(  ) ¯ 
8  + 24    ¯ − ¯ − 8  + 24   ¯ +
05  h¯ 3 ¯ ¯ 3 ¯i ⎠ (7.82)
¯  4 2 ¯ ¯  4 2 ¯
sgn() ¯ 24   + 64  ¯ + ¯ 24   − 64  ¯

Notice that sgn(2 ) = sgn() sgn(2 ) = sgn().

Remark 7.11 One should notice that for sway and yaw in (7.78), it is assumed that
Z
2
05 () ( r  ) ( + ) | + | ≈ 05   ( r )| | 


and Z
2
05 () ( r  ) ( + ) | + | ≈ 05 2  ( r )| | 


implying that the effect of yaw angular velocity, , is small relative to the effect of  .

Remark 7.12 Furthermore, if the drag coefficient in sway is given by model tests, an alternative
expression for the 2-dimensional drag coefficient applied in (7.79) and (7.80) may be found to
be
2
 =  (2) (7.83)

Linear damping It is important to notice that for velocities close to zero, linear damping
becomes more significant than the nonlinear damping. The strictly positive linear damping
matrix DL ∈ R6×6 caused by linear wave drift damping and the laminar skin friction is written
as ⎡ ⎤
 0  0  0
⎢ 0  0  0  ⎥
⎢ ⎥
⎢  0  0  0 ⎥
DL = − ⎢⎢ 0 
⎥ (7.84)
⎢ 0  0  ⎥ ⎥
⎣  0  0  0 ⎦
0  0  0 
This kind of damping must not be confused with the frequency-dependent wave radiation
damping used in the wave-frequency model. In general, the effect of wave radiation damping
can be neglected in the LF model due to the low frequency of oscillation, especially for the
horizontal modes of motions. For the vertical-plane modes (heave, roll and pitch), potential
damping may be of interest in the frequency range of interest for control, and should therefore

205
 Dominating
High sea state Low sea state
damping
Nonlinear turbulent skin friction,
Linear wave drift.
when | |  0
Surge Nonlinear turbulent
Linear laminar skin friction,
skin friction.
for low KC number and  → 0
Nonlinear eddy-making,
Nonlinear eddy-making. when | |  0
Sway
Linear wave drift. Linear laminar skin friction,
for low KC number and  → 0
Nonlinear eddy-making,
Nonlinear eddy-making. when |  |  0
Yaw
Linear wave drift. Linear laminar skin friction,
for low KC number and    → 0

Table 7.1: Dominating damping effects.

be considered to be included in the LF model. Wave drift damping in surge can be interpreted
as added resistance for the vessel advancing in waves and is proportional to the square of the
significant wave height.
It is for most bodies hard to calculate the damping coefficients. A combination of empiri-
cal formulas, model tests and computational fluid dynamics (CFD) are normally used to find
the damping coefficients. In Table 7.1 the dominating damping effects are described for the
horizontal-plane DOF. The heave, roll and pitch will normally follow the same tendency as the
sway and yaw.
The Keulegan-Carpenter number in Table 7.1 is defined as  =   , where  is the
free stream velocity,  is the oscillation period and  is the characteristic length of the body.
For sway and yaw the nonlinear damping due to eddy-making will dominate the damping until
very low velocity, where the damping behavior is seen to be linear, suggesting that laminar skin
friction is present. In the literature some authors do not include any linear damping terms in
the models. Instead they let the drag coefficients increase to large values for small velocities
     → 0 However, from a control point of view it is appropriate to include both linear and
nonlinear damping, as this seems to be more in agreement with the physics.
The coefficients can be calculated by special software or found by model tests. In the last
years the importance of wave-drift damping has been more appreciated, and some effort has been
made to include wave-drift damping predictions in hydrodynamic software programs, see Finne
and Grue [75] and the references therein. The cross-flow principle and strip theory can be used
to calculate viscous damping in sway and yaw. The cross-flow principle assumes that the flow
separates due to the cross-flow past the ship, and that the transverse forces on a cross section is
mainly due to separated flow effects on the pressure distribution around the ship. The method is
semi-empirical in the sense that empirical drag coefficients are employed. For conventional ships
data on nonlinear drag coefficients for the horizontal-plane modes are available, see OCIMF
[213]. Instead of applying semi-empirical methods, model tests are often used. When using
model tests it should be kept in mind that scale effects may be important for the viscous forces.
The transition from laminar to turbulent boundary layer is dependent on the Reynolds number,
which is different in model and in full scale. This may also affect the separation point and thus

206
the eddy-making damping. In cases with a clearly defined separation point, such as a bilge
keel, scale effects are not supposed to be significant for the eddy-making damping. Wave drift
damping is considered to be unaffected by scale effects.
If there already exists data from other ships, based on model tests or numerical calculations,
experience has shown that using proper scaling techniques, these coefficients may represent a
reasonable good estimate for ships with similar geometrical hull shapes. For further details about
damping the reader is referred to Faltinsen [66], Faltinsen and Sortland [67], and Newman [199].

Remark 7.13 In implementation of the linear damping terms for ships due to linear skin fric-
tion, it is suggested to ramp down these damping contributions when the relative speed exceeds
025  Notice that the linear damping due to wave drift should be maintained.

Generalized restoring forces, G(η) :

Here it is assumed small roll and pitch angles, such that the restoring vector can be linearized to
Gη, where G ∈ R6×6 is a matrix of linear generalized gravitation and buoyancy force coefficients
and is for -plane symmetry written
⎡ ⎤
0 0 0 0 0 0
⎢ 0 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0  0  0 ⎥
G = −⎢ ⎢ ⎥ (7.85)
0 0 0  0 0 ⎥
⎢  ⎥
⎣ 0 0  0  0 ⎦
0 0 0 0 0 0
where the coefficients are defined as
 , −    (7.86)

ZZ
 =  ,    (7.87)
 
ZZ
 , − ∇( −  ) −    2  = −    (7.88)
 
ZZ
 , − ∇( −  ) −   2  = −    (7.89)
 
Here,  is the acceleration of gravity,   is the waterplane area,  =  ∇ is the displaced
volume of water, and  and  are the transverse and longitudinal metacentric heights,
respectively.

7.3.2 Environmental Loads

The slowly-varying environmental loads acting on the vessel are composed of
 env =  wind +  wave2  (7.90)
Remember that the current load vector already is included by the relative velocity vector in the
nonlinear damping term.

207
Wind load model
The effect of wind may be divided into mean, slowly-varying and rapidly-varying wind loads.
The relative wind velocity vector is defined as
£ ¤
ν rw =  −   −       (7.91)
The components of the wind velocities are defined according to
 =  cos (  − ) ,  =  sin (  − )  (7.92)
where  and   are the wind velocity and direction respectively, see Figure 7.19. The total
relative wind velocity is then defined as for  =  −  , and  =  −  according to
p
 = 2 +  2  (7.93)
The relative wind angle is found from the following relation:
 w = atan2(−  − ) (7.94)
The wind load vector is then formulated
⎡ ⎤
  (  ) | | 
⎢   (  ) | |  ⎥
⎢ ⎥
⎢ 0 ⎥
 wind = 05 ⎢
⎢    ( ) | | 
⎥
⎥ (7.95)
⎢  ⎥
⎣ −   ( ) | |  ⎦

   (  ) | | 
Here,  is the density of air,  is the overall length of the vessel,  and  are the
vertical distances between transverse and longitudinal origin and the wind load point of attack,
 and  are the lateral and longitudinal areas of the non-submerged part of the ship projected
on the -plane and -plane.  (  )  (  )and  (  ) are the non-dimensional wind
coefficients in surge, sway and yaw respectively. These coefficients are often found by model test-
ing or by employing semi-empirical formulas as presented in Isherwood [128] and Blendermann
[32].
Remark 7.14 In station keeping operations the wind velocity is often assumed to be much larger
pvessel velocity, such that the relative wind angle may be simplified to   =   −  and
than the
 = 2 +  2

Wave load model

The wave drift loads contribute to a significant part of the total excitation force in the low-
frequency model. The second-order wave effects are divided into mean, slowly varying (difference
frequencies) and rapidly varying (sum frequencies) wave loads. For the applications considered
here the effect of the rapidly varying second order wave loads can be neglected. The determi-
nation of the second-order wave effects can be done by means of quadratic transfer functions
(Newman [198] and Faltinsen [66])
 2 = ̄  +     = 16 (7.96)
 X
X 
£  
¤
=    cos ((  −   )  +  −  ) +  sin (( −   )  +  −  ) 
=1 =1

208
where   is the wave frequency,  is the wave amplitude and  is a random phase angle.
The superscript  and  denote cos and sin, respectively. The quadratic transfer functions 
are dependent on both the first and second order velocity potentials, which require a nonlinear
panel methodology. In addition, it is time-consuming to calculate the  for all combinations
of   and    This motivates to derive some simplifications. One should notice that the transfer
functions when  = ,   represents the mean wave loads, and can be calculated from the first
order velocity potential only. The most interesting slowly-varying wave loads are those where
  −   is small and the loads are truly slowly-varying. Normally,  will not vary significantly
with the frequency. Then, the following approximation by Newman [198] will give satisfactory
results
  1 ¡  
¢
 =  =  +  (7.97)
2
 
 =  =0 (7.98)

The slowly-varying loads are approximated by the mean drift loads, and hence, the computation
becomes much simpler and less time consuming. This approximation based on frequency de-
pendent wave drift coefficients will then further be applied. By dividing the sea wave spectrum
(usually of Pierson-Moskowitz type) into N equal frequency intervals with corresponding wave
frequency,    and amplitude,  , the wave drift loads are found to be

 2 = ̄  +   (7.99)

⎛ ⎞2

X ¡  ¢12
= 2⎝   (    − ) cos (   +  )⎠ 
=1

where   0 is the frequency-dependent wave drift function and 

 is the mean wave direc-
tion (assumed to follow the same sign convention as wind and current). A disadvantage with
this approximation is the numerical generation of high-frequency components of no physical
meaning. By numerical filtering this can be avoided. Eq. (7.99) can also be extended to include
wave spreading. In general, the second-order wave loads are much smaller than the first-order
wave loads. The second-order wave loads are proportional to the square of the wave amplitude,
whereas the first-order wave loads are proportional to the wave amplitude. This means that the
second-order wave loads have an increased importance for increasing sea states.

7.3.3 Linear Wave-Frequency Model

In linear theory small waves and amplitudes of motion are assumed. The WF motion is in
the literature calculated in the hydrodynamic frame. The hydrodynamic problem of a vessel in
regular waves is solved as two sub-problems wave reaction and wave excitation, which are added
together (Faltinsen, [66]). Potential theory is assumed, neglecting viscous effects.

• Wave Reaction: Forces and moments on the vessel when the vessel is forced to oscillate
with the wave excitation frequency. The hydrodynamic loads are identified as added mass
and wave radiation damping terms.

• Wave Excitation: Forces and moments on the vessel when the vessel is restrained from
oscillating and there are incident waves. This gives the wave excitation loads which are

209
 composed of so-called Froude-Kriloff (forces and moments due to the undisturbed pressure
field as if the vessel was not present) and diffraction forces and moments (forces and
moments because the presence of the vessel changes the pressure field).

In station keeping operations assuming small motions about the coordinates d  d  and  d ,
the coupled equations of WF motions can in the hydrodynamic frame be formulated according
to
M()η̈ + D ()η̇  + Gη  = τ wave1
(7.100)
η̇  = J(η̄ 2 )η̇  
where η  ∈ R6 is the WF motion vector in the hydrodynamic frame, η  ∈ R6 is the WF
motion vector in the Earth-fixed frame, and η̄ 2 = [0 0   ] . τ wave1 ∈ R6 is the first order
wave excitation vector, which is dependent on the vessel heading relative to the incident wave
direction. M() ∈ R6×6 is the system inertia matrix containing frequency dependent added
mass coefficients in addition to the vessel’s mass and moment of inertia. D () ∈ R6×6 is the
wave radiation (potential) damping matrix. G ∈ R6×6 is the linearized restoring coefficient
matrix due to the gravity and buoyancy affecting heave, roll and pitch only. It is assumed that
the mooring lines will not affect the WF motion (Triantafyllou, [316]).
Generally, a time domain equation cannot be expressed with frequency domain coefficient.
However, this is a common used formulation denoted as a pseudo-differential equation. .An im-
portant feature of the added mass terms and the wave radiation damping terms is the memory
effects, which in particular are important to consider for non-stationary cases, e.g. rapid changes
of heading angle. Memory effects can be taken into account by introducing a convolution inte-
gral or a so-called retardation function (Newman, [199]) or state space models as suggested by
Kristiansen and Egeland [155] and Fossen [81].
Results from model tests and computer programs for vessel response analysis often come
in the form of transfer functions or tables of coefficients. This applies to linear wave-induced
motions, 2 -order wave drift and slowly varying motions. To a large extent, linear theory is
sufficient for describing wave-induced motions and loads on vessels. This is especially true for
moderate sea states. In irregular seas, the response of the vessel may be calculated by adding
results for regular waves of different amplitudes, frequencies and directions. Nonlinear effects
become increasingly important in severe sea states, and is still a subject for further research.

7.3.4 Thrust Servo Model

Usually, perfect control action without imposing the thruster dynamics are assumed. Unfortu-
nately, this will not be true in a real system. The thrust response is affected by the dynamics
in the actuators and the drive system. This will cause reduced command following capabilities
such as phase lag and amplitude reduction when the frequency increases. There will also be a
loss of thrust efficiency due to disturbances in the water inflow to the thruster blades, caused
by thrust-to-thrust and thrust-to-hull interactions, current and vessel velocities. In addition
the influence from the free surface will affect the thrust efficiency. Reduced thruster efficiency
caused by disturbances in the water inflow is compensated for in the controller. This will be
addressed in detail in Chapter 9. However, experience obtained from full-scale experiments
indicates that a first-order model is well suited as a first approximation. Hence, the thruster
dynamics is represented by the following simplified model:

τ̇ thr = −Athr (τ thr + τ c )  (7.101)

210
where τ c ∈ R3 is the commanded thrust vector in surge, sway and yaw produced by the con-
troller. Athr = diag{11 12 13 } is the diagonal thruster dynamics matrix. Determination
of the commanded thrust vector will be treated more in detail in Section 8.3.

7.4 Static Analysis of Cable Segments

7.4.1 Basic Assumptions
A basic assumption when working with cables is that we have no bending stiffness and no
torsional stiffness. This means that only axial stiffness will be considered, and this simplifies the
analysis of cables considerable compared to for instance beams.
Another assumption is that the axial tension in the cable is small enough to allow us to
operate in the linear range of stress/strain relationship. This is reasonable for metallic cables
and synthetic cables under normal tension.

T+dT
n d
dp
dz
 t
T dx

Figure 7.21: An infinitesmall cable segment.

Consider Figure 7.21 which shows a cable segment. The cable’s weight per unit length in air
is denoted  = , where  is the earth’s gravity and  is the mass per length unit. When
the cable is submerged a hydrostatic force will appear according to

 =   (7.102)

 is the density of water and  is the cross-sectional area. Given a cable tension,   Tri-
antafyllou [315] shows that the effective tension,  may be written as

 =  +   (7.103)

where  is the hydrostatic pressure at the specific point of the cable. This leads to the following
definition of the stretched cable’s weight in water, 1 :

1 =  −  (7.104)

When working with cables we assume that the material is isotropic. Isotropic materials’ proper-
ties are by definition independent of direction. Such materials have only 2 independent variables
(i.e. elastic constants) in their stiffness and compliance matrices. The generalized Hooke’s law
for axial strain may be written as

 =  =  (7.105)


211
where  is strain,  is Young’s modulus,  is axial strain and  is Poisson’s ratio. Applies this
law on the submerged cable and find
µ ¶
1 
= + 2  (7.106)
 

Notice that a Posisson’s ration of 12 results in a incompressible material. Poisson’s ratio for syn-
thetic cables are close to 12  while metallic materials have a ratio approximately 13 (Triantafyllou
[315]). A substantial simplification of the forthcoming analysis is achieved if we assume that
Poisson’s ratio is close to 12  Using (7.103) in (7.106) a simplified expression for the axial strain
is found to be
1 
= ( +  ) =  (7.107)
 
This leads to the last assumption; the cross-sectional area of the cable, , will not undergo
significant changes due to the axial deformation of the cable.

7.4.2 Catenary Equations

Catenary equations are widely used in mooring analysis of anchored bodies. In mooring analy-
sis the effect of each mooring line is first analyzed separately and thereafter summarized to
constitute the complete mooring system, often consisting of several deployed mooring lines.
For anchored floating vessels it is assumed that each mooring line is fastened in the seabed
and in the vessel. Vertical and horizontal forces caused by the mooring system will act on the
vessel at the terminal points. However, often only the horizontal forces are important for the
vessel response in the horizontal plane. Along the mooring line gravity and buoyancy forces are
assumed to be most important. For this purpose a two-dimensional approach is often considered
to be sufficient. If varying current in space is considered, a three-dimensional approach may be
needed.

Two-dimensional approach
An infinite small element of a cable is shown in Figure 7.21. The tangential and normal forces
to this element may be written as
∙ ¸ ∙ ¸ ∙ ¸
 − − 1 sin   + ( +  ) cos () 0
F= = =  (7.108)
 −1 cos   + ( +  ) sin () 0

Taking the limit " #

 ∙ ¸
 1 sin 
lim F ⇒ =  (7.109)
→0  

1 cos 
because ∙ ¸ ∙ ¸
sin () 
lim =  (7.110)
→0 cos () 1
From the Figure 7.21 we also find that
∙ ¸ ∙ ¸
  cos 
=  (7.111)
  sin 

212
 It is appropriate to formulate the equations for the tension as a function of the cable’s
Lagrangian coordinate, denoted as  along the unstretched length. The cable’s strain is given
in (7.107), but may also be written as

 −  
= = − 1 (7.112)
 
where  is the cable’s stretched length. Triantafyllou [315] shows the relation between mass,
buoyancy and specific weight of the stretched and unstretched cable are

0  =   (7.113a)
0  =   (7.113b)
  = 1  (7.113c)

Combining (7.109) and (7.111) with (7.112) and (7.113) leads to the following two-dimensional
catenary equations

=  sin  (7.114a)


 =  cos  (7.114b)


= (1 + ) cos  (7.114c)


= (1 + ) sin  (7.114d)

Notice that (7.114) is a set of ODEs which may be written in the form (10.1) described in Section
10.1.1. If we know that an axial force is acting at the end of the cable, this is an initial value
problem where  is the independent variable. Numerical routines like MATLAB’s ode45 could
be utilized, but in the next section we will show that there exists an analytical solution.

Solution of the two-dimensional equations

Consider the cable in Figure 7.22. An axial force  acts in point B, and no vertical force is
present in point . This is decomposed into a vertical and a horizontal force

 =  sin  (7.115a)
 =  cos  (7.115b)

tan  =  (7.115c)

Combining (7.114a) and (7.114b) gives


= tan   (7.116)

Integrating and applying the boundary condition in (7.115b) leads to


 =  (7.117)
cos 

213
 VB TB

HB
w B

Figure 7.22: A marine mooring cable.

Introducing (7.117) into (7.114b)

 
= cos2  (7.118)
 
Integration and use of (7.115c) leads to

 
tan  = − ( − ) (7.119)
 
where  is the cable’s unstretched length.
 may now easily be calculated as a function of  in (7.119). First, notice the following
relations
1
cos  = √  (7.120)
± 1 + tan2 
p
sinh−1 () = ln( + 1 + 2 ) (7.121)
tan 
sin  = √  (7.122)
± 1 + tan2 
Use of (7.120) leads to an expression for the tension along the Lagrangian variable 

 p
 () = =  1 + tan2  (7.123)
cos 
s
q
( −  ( − ))2 2 + ( −  ( − ))2 
=  1 + 2 =   


Notice that for  =  this gives the relation 2 = 2 +  2


Sometimes it may be valuable to calculate the Cartesian coordinates of the line. Eqs. (7.114c)
and (7.114d) can be rewritten to:
µ ¶
  ()
= (1 + ) cos  = 1 + cos  (7.124a)
 
µ ¶
  ()
= (1 + ) sin  = 1 + sin  (7.124b)
 

214
Still assuming constant  we find that
µ ¶
  () 
= 1+ cos  = cos  +  (7.125)
  
µ ¶
  ()  ()  −  ( − )
= 1+ sin  = sin  + = sin  +  (7.126)
   

Use of (7.120) and (7.122) yields

Z  Z  Z 
 1 
 = cos  +  = r +  (7.127)
0 0  0 ( − (−))2 
1+ 2

Z  Z 
 −  ( − )
 = sin  +  (7.128)
0 0 
Z   − (−)
  −  ( − )
= r + 
0 ( − (−))2 
1+ 2 

Z  Z 
 
 = q +  (7.129)
0 0 2
 + ( −  ( − ))2 
Z  Z 
 −  ( − )  −  ( − )
 = q +  (7.130)
0 0 2 + ( −  ( − ))2
 
 

Integration of this expression along the variable  leads to

∙ q ¸
 2 2
() = ln  −  ( − ) +  + [ −  ( − )]

∙ q ¸
 2 2
 
− ln  −   +  + ( −  ) + +   (7.131)
 
µq q ¶
1 2 2 2 2
() =  + [ −  ( − )] −  + ( −  )

1 ³  ´
+   + ( − )2 +   (7.132)
 2
We may want our coordinates for the start point  = 0 to be (0 0) Applies this boundary
condition to find  and   This gives

 = 0 (7.133)
1
 = −  2  (7.134)
2

215
 In some literature (e.g. Triantafyllou [315]) it is common to write () as a hyperbolic sine.
If we make use of (7.121) we may write
⎡ s ⎤
2
 ⎣  −  ( − ) [ −  ( − )] ⎦
() = ln + 1+ 2
  
" s #
  −   ( −  )2  
− ln + 1+ 2 +  (7.135)
   
∙ µ ¶ µ ¶¸
 −1  −  ( − ) −1  −    
() = sinh − sinh +  (7.136)
   
Sometimes it may be useful to express the - and -coordinates as a function of each other.
Remember that  (0) = 0 Eq. (7.123) may be written as
q
 () =  2 + ( )2  (7.137)


Eq. (7.119) may be rewritten to

 ()   
tan  = = =  (7.138)
  
Combining the equations above leads to
s µ ¶2 s µ ¶2
2 2  
 =  +  =  1 +  (7.139)
 
From Figure 7.21
2 = 2 +  2  (7.140)
and by use of the relation  = (1 + ), we find
s µ ¶2
 1 
= 1+  (7.141)
 1+ 
s µ ¶2
Z 
1 
⇒ () = 1+  (7.142)
0 1+ 
Combining with (7.138) the following relation appears
s µ ¶2
Z 
  1 
= 1+  (7.143)
  0 1 +  
If we assume inelastic material and substitute

sinh  =  (7.144)

which leads to
Z p

sinh  = 1 + sinh2  (7.145)
 0
Z 

sinh  = cosh   (7.146)
 0

216
Derivation on both sides
 
cosh  = cosh  (7.147)
 

 =  (7.148)


 =  + 1  (7.149)

Considering the boundary condition sinh |=0 = 0 we find that 1 = 0 Integration of (7.144)
 
= cosh  + 2  (7.150)
 

Boundary condition |=0 = 0 ⇒ 2 = − 
 Finally we have for the inelastic case:
  
() = cosh −  (7.151)
  
For the elastic case the substitution in (7.144) does not solve the problem, and the final
expression for () becomes very unhandy.

Catenary in the three-dimensional case

This section is mainly based on Sagatun [239]. Remember the assumption of no significant
reduction of the cable’s cross-sectional area when it is stretched. Considering the cable with
one-dimensional strain along the Lagrangian variable , Hooke’s law is

= =  (7.152)

⇒  =  =  (7.153)

where  is given in (7.112). Combining the equations we find that

µ ¶

 =  −1  (7.154)

r r 
Seeking a solution in Cartesian coordinates of the cable as a function of , the identity  =  
will be useful. r = [  ] contains the Cartesian coordinates as a function of  or . This leads
to the relation µ ¶
r r 
= +1  (7.155)
  

Consider the first segment of a cable with distributed vertical force,   a vertical concen-
trated force, 1 and axial tension  in the end point. Inspection of Figure 7.23 gives
¯
 ¯¯
0 =  ¯ + ¯1 +  (1 − 0 ) (7.156)
 =1

for static equilibrium.  indicates the value of  in node  and ¯ indicates the value of
concentrated force in node  and Cartesian direction 

217
 -f0
wz

s0

s1 T

f1,z

Figure 7.23: A cable segment with concentrated and distributed loads (Sagatun [239]).

Assume that the cable is terminated in point 0 and that concentrated forces, f̄  may act in
discrete points (cable nodes). Combination of this assumption and (7.156) written in vectorial
form gives the following equation for the forces acting in the cable’s terminating point:
¯ X 
X
r ¯¯
f0 =  ¯ + f̄ + w ( − −1 ) (7.157)
 =
=1 =1

where { ∈ [0 ] → R  ∈ [1 ] → N |  ∈ [−1   ]}   is the number of cable segments and 

is the cable’s unstretched length. w = [   ] indicates the constant distributed force
vector in segment 
An investigation of Figure 7.24 may make it easier to keep the indexes of the variables right
.
s0
s1
f2
f1 s2
w2
w4 fn
s3
s4 sn=L

Figure 7.24: A cable of several segments with concentrated forces, f̄  and distributed forces, w 
Notice the indexes.

The cable must be divided into a new segment when one of the following conditions are
fulfilled:

1. A concentrated force occurs

2. A change in distributed force occurs

218
 3. A change in cross-sectional area occurs

4. A change in the cable’s Young’s modulus occurs

A constant distributed force along the cable leads to a simplification of (7.157)

¯ 
X
r ¯¯
f0 =  ¯ + f̄ + w (7.158)
 (=)
=1

where w is the constant distributed force along the cable.

Sagatun [239] gives the following derivation of a differential equation for the cable. It is
assumed that (7.158) holds. Reordering of (7.158) leads to
¯ 
X
r ¯¯
 ¯ = f0 − f̄ − w (7.159)
 (=)
=1

We want to find the tension vector as a function of . Assume that () may be found from
(7.155). Then the following relation must hold

X 
r
 () = f0 − f̄ − w (7.160)

=1

where { ∈ [0 ] → R  ∈ [1 ] → N |  ∈ [−1   ]}  We define

⎡  ⎤

⎢  ⎥ r
T() = ⎣  ⎦=  (7.161)
 


r  r
T T =  2  (7.162)
 
where T ∈ R3  The substitution

X
f = f0 − f̄  (7.163)
=1

where { ∈ [1 ] → N  ∈ [0 ] → R |  ∈ [−1   ]}  gives

q
 () = (f − w) (f − w) (7.164)

Combining (7.163) and (7.160) gives

X 
r
 () = f0 − f̄ − w = f − w (7.165)

=1

Use of (7.155)
µ ¶−1
r 1 1 r
 = + = f − w (7.166)
   

219
Reordering leads to µ ¶
r 1 1
= (f − w) +  (7.167)
  
Compare this equation to (7.124) found in the previous section. Remember that T = (f − w)
represents the tension in vectorial form. The trigonometric terms in (7.124) is a way of decom-
posing T into the -plane. This shows the relation between the two-dimensional case solved
by Triantafyllou [315] and the three-dimensional case solved by Sagatun [239].
Substituting (7.164) gives the final differential equation
⎛ ⎞
r 1 1
= (f − w) ⎝ +q ⎠ (7.168)
  
(f − w) (f − w)
This is an ODE with  as the independent variable. This may be solved directly from the
rewritten form
⎛ ⎞
Z 
1 1
r = (f − w) ⎝ +q ⎠  (7.169)
−1  
(f − w) (f − w)

where  is the actual cable segment. r is an expression for the local solution within segment 
Numerical methods could have been utilized to solve the ODE/IVP, but Sagatun [239] gives
the following solution to the integral
q
() ¡ 2 ¢ 1
r () = f  − f ⊗ w ⊗ w − f  ⊗ (P (f ⊗ w)) − w (f − w) (f − w)
3 2
µ ¶
1 1 2
+ f  − w + C  (7.170)
 2

where ⊗ denotes component wise multiplication and

√
 = w w = kwk2  (7.171)
∙µ ¶ q ¸
1  
 = ln  − f w + (f − w) (f − w)  (7.172)

⎡ ⎤
0 1 1
P = ⎣ 1 0 1 ⎦ (7.173)
1 1 0

Assume that we want segment no. 1 of the global solution r() to start in the origin. We also
want to ensure continuity between the segments. The following conditions should be fulfilled

r(0) = 0 (7.174)
− +
r( ) = r( )  { ∈ [1  − 1] → N}  (7.175)

The integration constants, C   ∈ [0  − 1] may be calculated from

½
−r1 (0)    = 1
C−1 =  (7.176)
r−1 (−1 ) − r (−1 )   ∈ [2 ] → N    ≥ 2

220
If  belongs to a segment with index higher than 1 C−1 must be calculated iteratively before
the solution of r() can be found. The final global solution may now be calculated from
q
() ¡ 2 ¢ 1
r() = 3 f  − f ⊗ w ⊗ w − f  ⊗ (P (f ⊗ w)) − w 2 (f − w) (f − w)
 
µ ¶
1 1 2
+ f  − w + C−1 (7.177)
 2
where { ∈ [0 ] → R  ∈ [1 ] → N |  ∈ [−1   ]} 

Spatial variation in the distributed load

The solution given above is only valid for constant distributed load along the entire cable.
However, an extension to different distributed loads for each segment may be found. Instead of
using (7.158), we will assume that the distributed load is constant only within each segment.
This means that f0 must be calculated from (7.157). Eq. (7.165) must be rewritten to

X X −1
r
 () = f0 − f̄ − w ( − −1 ) − w ( − −1 ) (7.178)

=1 =1

where { ∈ [0 ] → R  ∈ [1 ] → N |  ∈ [−1   ]}  We modify (7.163) to


X −1
X
f = f0 − f̄ − w ( − −1 ) + w −1  (7.179)
=1 =1
Now the derivation goes just like in the previous section, and we end up with
⎛ ⎞
Z 
1 1
r = (f − w ) ⎝ +q ⎠  (7.180)
−1  
(f − w ) (f − w )

Continuing on the boundary condition and claiming continuity between segments, the integration
constants may be calculated as before. The new solution for spatially varying distributed loads
with the assumptions made above, are given from

q
() ¡ 2 ¢ 1
r() = 3 f  − f ⊗ w ⊗w  −f  ⊗ (P (f ⊗ w )) − w 2 (f − w ) (f − w )
 
µ ¶
1 1
+ f  − w 2 + C−1  (7.181)
 2

q
 = w w = kw k2  (7.182)
∙µ ¶ q ¸
1  
 = ln  − f w + (f − w ) (f − w )  (7.183)

⎡ ⎤
0 1 1
P = ⎣ 1 0 1 ⎦ (7.184)
1 1 0

221
7.4.3 Catenaries as Boundary Value Problems (BVP)
To this point we have considered catenaries as a cable with known properties and known loads.
The goal has been to achieve a solution for the end point. Let us revert this and try to find
the force components when the two end points are known. In this section we will assume that
the cable consists of only one segment. The governing differential equations are still known, see
(7.168). The problem has now changed from a ODE/IVP to a ODE/BVP (see Section 10.1.1).
An analytical solution to this problem could be found if f was solved from (7.168). Nobody
has published such analytical solution, and we will have to deal with numerical solutions of the
problem. Both MATLAB’s function bvp4c or e.g. a finite element solution will do if the cable’s
geometry is of particular interest.
Sometimes other problems arises when we are dealing with boundary value problems. We
may for instance know the boundary coordinates, but want to find the end force to yield this
solution. The ODE/BVP has now increased to include an estimation of the parameter f̄1 . A
shooting method based on the accurate analytical solution might be applied. We will now given
an example where we use MATLAB’s routine to estimate this force.

Example 7.1 3D catenary BVP with end force estimation.

A cable of length 10 meters is fastened in (0,0,0). The cable’s weight is 1N/m, and  =
7854·104 . The end force is f̄1 = [10 0 10] N. Eq. (7.177) gives the end point (8.8137,0,4.1421).
MATLAB’s routine bvp4c was applied to these parameters, and an estimate of the end force
was found. The results are shown in Figure 7.25.Even if the initial guess for the estimated force

Cable bvp
4.5

3.5

2.5 Initial guess

z

1.5

Solution
1

0.5

0
0 1 2 3 4 5 6 7 8 9
x

Figure 7.25: Solution of the boundary value problem for a hanging cable. MATLAB’s routine
bvp4c was applied. The dashed line shows the initial guess for a solution.

deviated from the correct value, the routine managed to find very good estimates. The results
were dependent on the numerical option and the mesh in the variable  The MATLAB code for
this solution is shown below.

222
function cable3dbvp
W=[0 0 -1]’; %Constant distributed force
Fguess=[5 0 8]’; %Guess for end force (We know this guess is wrong)
E=1e9; %Modulus of elasticity
D=0.1; %Cable’s diameter
A=pi*(D/2)^2; %Cable’s cross-sectional area
L=10; %Cable’s length
EndP=[8.8137 0 4.1421]’; %Boundary point for second end.

s=0:.1:L; %Initial mesh

%Calculates an initial guess for the cable’s geometry (a straight line)

n=length(s);
Xguess=linspace(0,EndP(1),n);
Yguess=linspace(0,EndP(2),n);
Zguess=linspace(0,EndP(3),n);
sol.x=s;
sol.y=[Xguess; Yguess; Zguess];
sol.parameters=Fguess;

%Plots the initial guess

figure(1)
clf
plot(sol.y(1,:),sol.y(3,:),’k--’)
xlabel(’x’)
ylabel(’z’)
title(’Cable bvp’)

sol = bvp4c(@cableODE,@cableBC,sol,[],W,E,A,L,EndP); %Solves the BVP

%Plots the result

hold on
plot(sol.y(1,:),sol.y(3,:),’k-’)
hold off

disp(sprintf(’Calculated end force is ...

[%1.2f %1.2f %1.2f] N \n’,sol.parameters(1),sol.parameters(2),sol.parameters(3)))

%System’s ODEs
function drds = cableODE(s,r,F,W,E,A,L,EndP)
Fk=F+W*L;
drds = (Fk-W*s)*(1/sqrt((Fk-W*s)’*(Fk-W*s))+1/(E*A));

%System’s boundary values

223
function res = cableBC(ya,yb,F,W,E,A,L,EndP)
res = [ ya(1)
ya(2)
ya(3)
yb(1)-EndP(1)
yb(2)-EndP(2)
yb(3)-EndP(3)];

7.4.4 Hydrodynamic Drag Loads

Consider the current acting on a small cable element  in Figure 7.26. Hydrodynamic loading

Fn
dp
Current
Ft

Figure 7.26: Static hydrodynamic forces acting on an infinitesimall cable element dp.

on mooring lines are usually modelled with the so-called cross-flow principle. This principle
assumes that the flow separates due to cross-flow past the cable, that the longitudinal current
components do not influence on the transverse forces on a cross-section, and that the transverse
forces on a cross-section is mainly due to separated flow effects on the pressure distribution
around the ship Faltinsen [66]. From Morison’s equation we get
1 1
f = −    () |v · t| (v · t) t = −    () |v | v  (7.185)
2 2
1 1
f = −    () |v− (v · t) t| (v− (v · t) t) = −    () |v | v  (7.186)
2 2
where  is the density of water,  is the cable’s diameter,  and  are tangential and
normal drag coefficients, t is tangential vector of the cable, v v  and v are velocity, normal
velocity and tangential velocity on the cable on vectorial form. Notice that it is common to
normalize to the projected area  · (), in this kind of hydrodynamic forces. Some refer to f
as the normal friction force, but this is inaccurate. Friction forces are usually normalized to wet
surface of the body. For a cable in the normal direction this leads to
1 1
f = −  () |v− (v · t) t| (v− (v · t) t) = −   () |v | v  (7.187)
2 2
where  = 1  
Hydrodynamic forces may not be handled in the same way as gravity in the catenary equa-
tions. Gravity acts on each cable segment independently of it’s orientation. This means that the

224
shape and tension may be solved explicitly. Calculation of current forces requires information on
the cable segment’s orientation, and calculation of orientation requires knowledge of the current
forces. Consequently, an iteration procedure is needed.
The calculation of hydrodynamic forces (7.185) and (7.186) assumes that the water’s velocity
is constant over the actual segment. For analytical solutions the hydrodynamic forces have to be
constant over each segment. This means that the density of segments should be higher is areas
with high curvature, and lower in areas with low curvature. For the calculation, it is natural to
apply the vector between the segment boundaries as the vector t

Weight

One segment End force

Two segments Cable

Current

Figure 7.27: Division of the cable into one segment shows that the current will in this case have
no effect when the straight line between the nodal points are used as the tangential vector, t in
the expression for hydrodynamic forces. Two segments yields a slightly better result.

When the analytical solution is applied to a problem which involves high curvature on the
cable, the effect of current forces should be iterated to obtain a good result. First, a geometry
should be calculated with no influence from the current. In the next step the current forces
should be turned on, and a new geometry will be obtained. This will again make the basis for
the next solution, and so on. The iteration can be stopped when the difference between two
consecutive solutions is low.

7.5 Mooring System for FPSOs

Generally, a mooring system consists of  lines connected to the structure and horizontally
spread out in a certain pattern. The anchor lines are composed of chain, wire lines or synthetic
material, often partitioned into several segments of different types and properties. For turret-
moored ships, when the turret is rotatable, the relative angle between the turret and the body-
fixed frame is given by the turret angle tu . The length of each anchor line is adjusted by
winches and determines the pre-tension and thus the stiffness of the mooring system. The
anchor lines enter the turret through fairleads below the hull and the coordinates are defined
as terminal points (TP). Mooring lines are subjected to three types of excitation (Triantafyllou
[316]): Large amplitude LF motions, medium amplitude WF motions and small amplitude, very
high frequency vortex-induced vibrations. For the purpose of PM control system design it is
appropriate to consider the mooring lines’ influence on the low-frequency vessel model.
A horizontal-plane spread mooring model can be formulated as

τ moor = −RT ()gmo (η) − dmo (ν) (7.188)

225
where τ mo ∈ R3 is the vector of generalized mooring forces, dmo ∈ R3 represents the additional
damping in the system due to the mooring system, and gmo ∈ R3 is the Earth-fixed restoring
term ⎡ ⎤
n
X  cos  
gmo = ⎣  sin   ⎦ (7.189)
=1  ̄ sin   −  ̄ cos  
which is the vectorial sum of the force contribution from each line.  is the horizontal force at
the attachment point of the ship along the direction of line  and   is the earth-fixed direction
of the line. ̄ and ̄ are the corresponding moment arms. In a quasi-static approach, by
disregarding the dynamic effects in the mooring lines, the restoring forces gmo are treated as
function of the low-frequency ship position and heading η only according to

gmo ≈ ḡmo (η; tu ) (7.190)

where the horizontal force contributions  in (7.189) are replaced by the static line character-
istics (distance/force relationships) for each line  by

̄ =  ( ) (7.191)

which is a function of the horizontal distance  between TP and the anchor of each line. About
a working point the line characteristics (7.191) can be linearized by
¯
 ¯¯
̄ = ̄ + ∆  (7.192)
 ¯ =
¯
 ¯
where ̄ is the average horizontal force in the working point   and  ¯ = is the slope
 
of the line characteristics (7.191) at  . By assuming fixed anchor line length and neglecting
the influence of the current field along the line profile, the generalized mooring forces (7.188)
in a working point can be approximated by a 1st-order Taylor expansion of the static restoring
mooring forces and the mooring damping about the working points η = η o and ν = 0 according
to

ḡmo (η) = ḡmo (η o ) + Gmo (η − η o ) +  (7.193)

dmo (ν) = Dmo ν +  (7.194)

where h.o.t. denotes higher order terms and

¯ ¯
ḡmo ¯¯ dmo ¯¯
Gmo =  Dmo =  (7.195)
η ¯= ν ¯=0

For simplicity, the Earth-fixed frame is often placed in the natural equilibrium point of the
mooring system, i.e. ḡmo (ηo = 0) = 0. D and G are the linearized mooring damping and
stiffness matrices assumed to only contribute in the horizontal-plane. They can, for symmetrical
mooring patterns about the - and - planes be formulated as
© ª
Gmo = diag 11 22 0 0 0 66  (7.196)
© ª
Dmo = diag 11 22 0 0 0 66  (7.197)

226
Hence, the quasi-static mooring model can be written

τ moor = −RT ()Gmo η − Dmo ν (7.198)

For further details in modelling, see Faltinsen [66], Huse and Matsumoto [122] and Triantafyl-
lou and Yue [317]. For fully dynamically positioned vessels with no anchor system, τ moor is equal
to zero.

227
Chapter 8

Dynamic Positioning Control System

This chapter is an extension of the work summarized in the Ocean Engineering Handbook, [278],
[292], [87] [293], and [88], Sørensen [300] and [305]

8.1 Survey of Scientific Contributions

The real-time control hierarchy, or the control structure, of a marine control system (Sørensen,
[300]) may be divided into three levels: the guidance system, the high-level plant control (e.g.
DP controller including thrust allocation), and the low-level thruster control. Description of DP
systems including the early history can be found in Fay [74]. In the 1960s the first DP system
was introduced for horizontal modes of motion (surge, sway and yaw) using single-input single-
output PID control algorithms in combination with low-pass and/or notch filter. In the 1970s
more advanced output control methods based on multivariable optimal control and Kalman
filter theory was proposed by Balchen et al. [15]. This work was later improved and extended
by Balchen et al. [19], Jenssen [131], Sørheim [306], Sælid et al. [280], Fung and Grimble
[91], Grimble and Johnson [99], Fossen [78], Sørensen et al. [286], Fossen et al. [79], Katebi
et al. ([147], [148]), Mandzuka and Vukic[180], Kijima et al. [151], Tannuri and Donha [307],
Volovodov et al. [320] and Perez and Donaire [222]. The introduction of observers with wave
filtering techniques based on Kalman filter theory (Fossen and Perez, [89]) by Balchen, Jenssen
and Sælid is regarded as a breakthrough in marine control systems in general, and has indeed
been an inspiration for many other marine control applications as well.
In the 1990s nonlinear DP controller designs were proposed by several research groups.
Stephens et al. [271] proposed fuzzy controllers. Aarset et al. [2], Strand and Fossen [273],
Fossen and Grøvlen [82], and Bertin et al. [26] proposed nonlinear feedback linearization and
backstepping for DP. In the work of Fossen and Strand [83], Strand and Fossen [277] and Strand
[276] the important contribution of passive nonlinear observer with adaptive wave filtering is
presented. One of the motivations using nonlinear passivity theory was to reduce the complexity
in the control software getting rid of cumbersome linearizations and the corresponding logics.
Pettersen and Fossen [223], Pettersen et al. [224] and Bertin et al. [26] addressed DP control of
under-actuated vessels. Agostinho et al. [4] and Tannuri et al. [310] proposed to use nonlinear
sliding mode control for DP.
As the DP technology became more mature research efforts were put into the integration
of vessel control systems and the refinement of performance for the various vessel types and
missions by including operational requirements into the design of both the guidance systems

228
and the controllers. Sørensen and Strand [290] proposed a DP control law for small-waterplane-
area marine vessels like semisubmersibles with the inclusion of roll and pitch damping. Sørensen
et al. [295] recommended the concept of optimal setpoint chasing for deep-water drilling and
intervention vessels. Leira et al. [166] extended this work and proposed to use structural
reliability criteria of the drilling risers for the setpoint chasing. Jensen [130] showed how proper
modeling of pipe dynamics can be included in the DP guidance system. Fossen and Strand [84]
presented the nonlinear passive weather optimal positioning control system for ships and rigs
increasing the operational window and reducing the fuel consumption.
Most of the current DP systems have been designed to operate up to a certain limit of weather
condition limited by the thrust and power capacity. Due to the accuracy and availability of the
inertia measurement units (IMU), Lindegaard (2003) proposed acceleration feedback (AFB) to
increase the performance of DP systems in severe seas. AFB denotes here output acceleration
feedback in addition to output PID controller. Sørensen et al. [296] and Sørensen [300] proposed
passive nonlinear observer without wave frequency (WF) filtering for output PID-controller in
extreme seas, especially where swell becomes dominant.
Use of hybrid control theory as proposed by Hespanha [114], Hespanha and Morse [115],
and Hespanha et al. [116] and fault-tolerant control by Blanke et al. [31] enabled the design
of proper control architecture and formalism for the integration of multi-functional controllers
combining discrete events and continuous control. Sørensen et al. [302], Nguyen [201], Nguyen
et al. [202], [203] and Nguyen and Sørensen [206] proposed the design of supervisory-switched
controllers for DP from calm to extreme sea conditions and from transit to station keeping
operations. The main objective of the supervisory-switched control is to integrate an appro-
priate bank of controllers and models at the plant control level into a hybrid DP system being
able to operate in varying environmental and operational conditions. Implementing the hybrid
control concept will increase the so-called weather window making it possible to conduct all-
year marine operations, such as subsea installation and intervention, drilling, and pipe laying in
harsh environment. Concerning large changes in environmental conditions, in particular, when
conducting marine operations in deep-water, the feature of hybrid control is important as the
operations are more time consuming, and hence more sensitive to changes in sea states. Lately,
with increasing interest for hydrocarbons in the arctic DP operations in various ice conditions
like level ice, managed ice and ice ridges have been studied. In Nguyen et al. [204] DP in level
ice is presented. For DP vessels operating partly in ice and open water, see Figure 8.1, switching
between controllers and control settings on both the plant-level and low-level will be necessary.
The number of the safety critical and demanding DP operations is increasing. As a conse-
quence of this the system integrity and requirements to further physical and functional integra-
tion between the DP system, marine automation system, thruster and propulsion system and
power plant will follow accordingly. It is believed that more research efforts will be directed into
diagnostics and fault-tolerant control (Blanke et al. [31]). As a part of this proper testing and
verification of the DP system software are crucial for the safety and profitability (Johansen et
al., [135], [136], Johansen and Sørensen, [137] and Smogeli, [267]).
The importance of the DP control system for the closed-loop performance of the station
keeping operation is clearly demonstrated in several studies. Morishita and Cornet [192], Mor-
ishita et al. [193], Tannuri and Morishita [308], and Tannuri et al. [309] have conducted detailed
performance studies of the DP operations for shuttle tanker and FPSOs.
Thruster assisted position mooring (PM) systems have been commercially available since the
1980s and provide a flexible solution for floating structures for drilling and oil&gas exploitation

229
 Figure 8.1: DP operations in arctic hydrocarbon exploration.

on the smaller and marginal fields. Modeling and control of turret-moored ships are treated in
Strand et al. [274], Strand [276], Sørensen et al. [289], Berntsen et al. [25], Berntsen [24] and
Nguyen and Sørensen [205], [206].
Important areas of guidance and navigation are not covered here. For further references on
these topics the reader is referred to Fossen [81], Skjetne [254], Ihle [126] and Breivik [34].

230
8.2 Observer Design for Dynamic Positioning
This section is a extension and modification of the observer design presented in Fossen and Strand
[87]. Filtering and state estimation are important features for all kind of control systems. A
measured signal very often content noise which may have negative impact on the controller
performance if no precaution is taken in the dynamic positioning (DP) and position mooring
(PM) systems, see Fossen [78], Fossen [80] and Strand [276]. The main purposes of the state
estimators (observers) in the positioning systems are:
• Reconstruction of non-measured data. For many applications important process states are
not measured. Typical reasons for this could be that no convenient sensors exist, or simply
that cost reasons motivate to not to install the sensor. In such cases sophisticated model
based filtering techniques - state estimation - can be applied. The main purpose of the
state estimator (observer) is to reconstruct unmeasured signals and perform filtering before
the signals are used in a feedback control system, see Figure 6.4. The input to the state
estimator is sensor data e.g. from inertial measurement units (IMU/VRS/MRU), gyro
compasses and positioning reference systems (DGPS, HPR, Artemis, Taut wire, Fanbeam,
etc.), see Section 2.2.
• Dead reckoning. All kind of equipment will fail according to some failure rate. Experience
from industrial applications has shown that the one of the most frequent control system
failure are caused by sensor failures. In safety critical marine applications a sudden drop-
out of the control system may lead to dangerous situations, if not an adequate signal
substitution will take place. Applying model based filters, the signal may, at least for
some period of time, replace by model prediction the measured signal.
• Wave filtering. Motions of marine vessels can often be divided into a low-frequency (LF)
part and a wave-frequency (WF) part. For most positioning applications the WF motion
is not subject for control. The reason for this could be that the WF motion does not
matter for the particular operation, or that the vessel does not have enough power and
thrust capacity for doing any noticeable compensation at all. The latter is the most usual
reason. Hence, there is no point to waist fuel and cause additional wear and tear of the
propulsion equipment. In order to avoid this wave filters are designed filtering out the WF
motions, see Figure 7.20.
We will in this section discuss three different methods for position and velocity state estima-
tion:
• Extended Kalman filter design (1976-present): The traditional Kalman filter based estima-
tors are linearized about a set of pre-defined constant yaw angles, typically 36 operating
points in steps of 10 degrees, to cover the whole heading envelope between 0 and 360
degrees (Balchen et al. [15], [18], Balchen et al. [19], Grimble et al. [97], [98] and Sørensen
et al. [286]). When this estimator is used in conjunction with a linear quadratic Gaussian
(LQG), PID or ∞ controller in conjuncture with a separation principle, there is no guar-
antee for global stability of the total nonlinear system. However, these systems have been
used by several DP producers since the 1970s. The price for using linear theory is that
linearization of the kinematic equations may degrade the performance of the system. Be-
sides, the number of filter gains and switching transitions between the various sectors are
large.

231
 • Nonlinear observer design (1998-present): The nonlinear observer is motivated from pas-
sivity arguments (Fossen and Strand [83] and Fossen [80]). Also the nonlinear observer
includes wave filtering, velocity and bias state estimation. In addition it is proven to be
global asymptotic stable (GES), through a passivation design. Compared to the Kalman
filter, the number of tuning parameters is significantly reduced, and the tuning parameters
are coupled more directly to the physics of the system. By using a nonlinear formulation,
the software algorithms are simplified. The observer of Fossen and Strand [83] has been
used by Aarset et al. [2], Strand and Fossen [275] and Loria, Fossen and Panteley [176]
and Sørensen et al. [296] in output-feedback controller design.

• Adaptive and nonlinear observer design (1999-present): The nonlinear and passive observer
is further extended to adaptive wave filtering by an augmentation design technique, see
Strand and Fossen [277]. This implies that the wave frequency model can be estimated
recursively on-line such that accurate filtering is obtained for different sea states. However,
conducting DP operations in extreme seas Sørensen et al. [296] has shown that adaptive
wave filtering may degrade the performance significantly using the same controller strategy
as in normal DP operation in calm and moderate seas.

8.2.1 Objectives
The objective of this section is to present the state-of-the-art and most recent observer designs
for conventional surface ships and rigs operating about zero (station keeping) and low speed
tracking. The vessels can either be both free-floating or anchored. In the latter case we assume
a symmetrical, spread mooring system, with linear response of the mooring system. It is assumed
that the vessels are metacentric stable, which implies that there exist restoring forces in heave,
roll and pitch, such that these motions can be modeled as damped oscillators with zero mean
and limited amplitude. Hence, the horizontal motions (surge, sway and yaw) are considered in
the modelling and observer design. The objectives for the observer are, see Strand [276]:

• Position and velocity estimation. It is assumed that only position and heading measure-
ments are available. Hence, one objective is to produce velocity estimates for feedback
control. In addition, the observer should remove measurement noise from the position and
heading measurements.

• Bias estimation. By estimating a bias term, accounting for slowly-varying environmental

loads and unmodelled effects, there will be no steady-state offsets in the velocity esti-
mates. Moreover, the bias estimates may be used as a feedforward term in the positioning
controller. However, this should be done with care as the bias estimate under certain
circumstances may be rather fluctuating.

• Wave filtering. The position and heading signals used in the feedback controller should
not contain the WF part of the motion. By including a synthetic wave-induced motion
model in the observer, wave filtering is obtained, see below:

Definition 8.1 (wave filtering) Wave filtering can be defined as the reconstruction of the LF
motion components from noisy measurements of position and heading by means of analog or
digital filters. In addition to this, if an observer (state estimator) is used, noise-free estimates of
the nonmeasured LF velocities can be produced. This is crucial in ship motion control systems

232
since the WF part of the motion should not be compensated for by the positioning system. If
the WF part of the motion enters the feedback loop, this will cause unnecessary wear and tear of
the actuators and increase the fuel consumption.

8.2.2 Control Plant Model: Vessel Model

As shown in Chapter 7, it is common to separate the modeling of marine vessels into a LF
model and WF model. The nonlinear LF equations of motion are driven by 2-order mean and
slowly-varying wave, current and wind loads as well as thruster forces. The WF motion of the
ship is due to 1st-order wave-induced loads, see Figure 7.20. For the purpose of model-based
observer and controller design it is sufficient to derive a simplified mathematical model, control
plant model, which nevertheless is detailed enough to describe the main physical characteristics
of the dynamic system.

Low-frequency control plant model

For the purpose of controller design and analysis, it is convenient to simplify (7.67) and derive
a nonlinear LF control plant model in surge, sway and yaw about zero vessel velocity according
to
η̇ = R()ν (8.1)
 
Mν̇ + Dv + R ()Gη = τ + R ()b (8.2)
where v = [  ]  η = [  ]  b ∈ R3 is the bias vector, and τ = [        ] is the
control input vector. Notice the control plant model is nonlinear because of the rotation matrix
⎡ ⎤
 − 0
R() = ⎣   0 ⎦
0 0 1
For low-speed applications, the different matrices are defined according to
⎡ ⎤
 − ̇ 0 0
M = ⎣ 0  − ̇  − ̇ ⎦ 
0  − ̇  − ̇
⎡ ⎤ ⎡ ⎤
− 0 0 − 0 0
D = ⎣ 0 − − ⎦  G = ⎣ 0 − 0 ⎦
0 − − 0 0 −
Remark 8.1 We have here assumed that G is a constant, diagonal matrix. One should notice
that for DP operations (no mooring system present) there is no stiffness such that G = 03×3 .
However, for anchored structures there will be restoring terms due to the mooring lines.

Wave-frequency control plant model

In the controller design synthetic white-noise-driven processes consisting of uncoupled harmonic
oscillators with damping will be used to model the WF motions. The synthetic WF model can
be written in state-space form according to
ξ̇ = A ξ + E w 
(8.3)
η  = C ξ 

233
η  ∈ R3 is the position and orientation measurement vector, w ∈ R3 is a zero-mean Gaussian
white noise vector, and ξ ∈ R6 . A linear 2nd-order WF model is considered to be sufficient
for representing the WF-induced motions, although higher order models may also be used, see
Grimble and Johnson [99]. The system matrix A ∈ R6×6 , the disturbance matrix E ∈ R6×3
and the measurement matrix C ∈ R3×6 may formulated as
∙ ¸
03×3 I3×3
A = , (8.4a)
−Ω2 −2ΛΩ
∙ ¸
£ ¤ 03×3
C = 03×3 I3×3 , E =  (8.4b)
K

where Ω = diag{ 1  2   3 } Λ = diag{ 1   2   3 } and K = diag{1  2  3 } This model

corresponds to
   
() = 2  (8.5)
  + 2    +  2
From a practical point of view, the WF model parameters are slowly-varying quantities depend-
ing on the prevailing sea state. Typically, the wave periods   corresponding to wave frequency
  = 2  are in the range of 5 to 20 seconds in the North Sea for wind generated seas. The
periods of swell components may be even longer than 20 seconds. The relative damping ratio
  will typically be in the range 005 − 010 As suggested by Strand and Fossen [277] adaptive
schemes may be used to update  for the varying sea states. However, this should be done with
care in heavy sea states with long wave lengths.

Bias model
A frequently used bias model b ∈ R3 for marine control applications is the first order Markov
model
ḃ = −T−1
 b + E w  (8.6)
where w ∈ R3 is a zero-mean Gaussian white noise vector, T ∈ R3×3 is a diagonal matrix of
bias time constants, and E ∈ R3×3 is a diagonal scaling matrix. The bias model accounts for
slowly-varying forces and moment due to 2nd-order wave loads, ocean currents and wind. In
addition, the bias model will account for errors in the modeling.
Alternatively, the bias model may also be modelled as random walk, i.e. Wiener process

ḃ = E w  (8.7)

Measurements
The measurement equation is written

y = η + η  + v (8.8)

where v ∈ R3 is the zero-mean Gaussian measurement noise vector.

234
Resulting control plant model
The resulting control plant model is written

ξ̇ = A ξ + E w  (8.9a)
η̇ = R()ν (8.9b)
ḃ = E w  (8.9c)
 
Mν̇ = −Dν − R ()Gη + R ()b + τ  (8.9d)
y = η + C ξ + v (8.9e)

Here, the Wiener bias model is used. The state-space model is of dimension x ∈ R15  τ ∈ R3 and
y ∈ R3 . In addition, it is common to augment two additional states to the state-space model
if wind speed and direction are available as measurements. If not, these are treated as slowly-
varying disturbances to be included in the bias term b. In the next section, we will show how
all these states can be estimated by using only 3 measurements.

Control plant model for extreme seas

As suggested in Sørensen et al. [296] the wave filtering should be avoided for long wave lengths
(low wave frequencies) appearing in extreme seas or in swell dominated seas. For such conditions
the following control plant model is proposed

ḃ = E w  (8.10a)
 
Mν̇ = −Dν − R ()Gη + R ()b + τ  (8.10b)
η̇ = R()ν (8.10c)
y = η + v (8.10d)

8.2.3 Extended Kalman Filter Design

The extended Kalman (EKF) filter design is based on the nonlinear model

ẋ = f (x) + Bu + Ew (8.11a)

y = Hx + v (8.11b)

where f (x) B E and H are given by (8.9a)—(8.9e). Moreover,

⎡ ⎤ ⎡ ⎤
A ξ 06×3
⎢ R()ν ⎥ ⎢ 03×3 ⎥
f (x) = ⎢ ⎣
⎥
⎦ B =⎢ ⎥
⎣ 03×3 ⎦  (8.12)
−T−1
 b
−M−1 Dν − M−1 R ()Gη + M−1 R ()b M−1
⎡ ⎤
E
⎢ 03×3 ⎥ £ ¤
E = ⎢ ⎥
⎣ E ⎦  H = C I3×3 03×3 03×3  (8.13)
03×3

where x = [ξ  η   b  ν  ]  w = [w  w ] and u = τ  Hence, for a commercial system with
 = 15 states the covariance weight matrices Q =(w w) ∈ R× (process noise covariance

235
matrix) and R = (v v) ∈ R3×3 (position and heading measurement noise covariance matrix).
This again implies that  + ( + 1)2 = 135 ODEs must be integrated on-line. In order to
simplify the tuning procedure these two matrices are usually treated as two diagonal design
matrices which can be chosen by applying Bryson’s inverse square method, [37], for instance.

Discrete-time EKF equations

The discrete-time EKF equations are given by, see Gelb [94]:

Initial values:

x̄=0 = x0  (8.14a)
£ ¡ ¢¤
P̄=0 =  (x(0) − x̂(0)) x(0) − x̂(0) = P0  (8.14b)

Corrector:
£ ¤−1
K = P̄ H HP̄ H + R  (8.15a)
P̂ = (I − K H)P̄ (I − K H) + K RK  (8.15b)
x̂ = x̄ + K (y − Hx̄ ) (8.15c)

Predictor:

P̄+1 = Φ P̂ Φ + Γ QΓ  (8.16a)

x̄+1 = f (x̂  u ) (8.16b)
 
where x̂ = [ξ̂  η̂   b̂  ν̂  ]  f (x̂  u ) Φ and Γ can be found by using forward Euler for
instance. Moreover

f (x̂  u ) = x̂ +  [f (x̂ ) + Bu ] (8.17a)

¯
f (x  u ) ¯¯
Φ = I× +  ¯  (8.17b)
x x =x̂
Γ =  E (8.17c)

where   0 is the sampling period and  = 15. The EKF has been used in most industrial
ship control systems. It should, however, be noted that there are no prove of global asymptotic
stability when the system is linearized. In particular, it is difficult to obtain asymptotic con-
vergence of the bias estimates b̂ when using the EKF algorithm in DP and PM. In the next
section, we will demonstrate how a nonlinear observer can be designed to meet the requirement
of global exponential stability (GES) through a passivation design. The nonlinear observer has
excellent convergence properties, and it is easy to tune since the covariance equations are not
needed.

8.2.4 Nonlinear Observer Design

Two different nonlinear observers will be presented. The first one is similar to the observer of
Fossen and Strand [83] for dynamically positioned (free-floating) ships with extension to spread
mooring-vessel systems (Strand [276]). The second representation of the nonlinear observer is

236
an augmented design where the filtered state of the innovation signals are used to obtain better
filtering, see Strand and Fossen [277]. By using feedback from the high-pass filtered innovation in
the WF part of the observer there will be no steady-state offsets in the WF estimates. Another
advantage is that by using the low-pass filtered innovation for bias estimation, these estimates
will be less noisy and, thus, may be used directly as a feedforward term in the control law.
However, this should be done with care.
The adaptive observer proposed in Section 8.2.5 is an extension of the augmented observer.
SPR-Lyapunov analysis is used to prove passivity and stability of the nonlinear observers.

Observer equations in the Earth-fixed frame

When designing the observer, the following assumptions are made in the Lyapunov analysis:
A1 Position and heading sensor noise are neglected, that is v = 0, since this term is negligible
compared to the wave-induced motion.
A2 The amplitude of the wave-induced yaw motion   is assumed to be small, that is less than
1 degree during normal operation of the vessel and less than 5 degrees in extreme weather
conditions. Hence, R() ≈ R( +   ). From A1 this implies that R() ≈ R(  ), where
  ,  +   denotes the measured heading.
This assumptions are only for a matter of convenience such that Lyapunov stability theory
can be used to derive the structure of the observer updating mechanism. It turns out the
SPR-Lyapunov based observer equations are robust for white Gaussian white noise so these
assumptions can be relaxed when implementing the observer.
We will also exploit the following model properties of the inertia and damping matrices in
the passivation design
M = M  0 Ṁ = 0, D  0

Observer equations (representation 1)

A nonlinear observer copying the vessel and environmental models (8.9a)—(8.9e) is

ξ̂˙ = A ξ̂ + K1 ỹ (8.18a)

η̂˙ = R(  )ν̂ + K2 ỹ (8.18b)
b̂˙ = −T−1 b̂ + K ỹ
 3 (8.18c)
Mν̂˙ = −Dν̂ − R (  )Gη̂ + R (  )b̂ + τ + R (  )K4 ỹ ,
  
(8.18d)
ŷ = η̂ + C ξ̂ (8.18e)
where ỹ = y − ŷ ∈ R3 is the estimation error (in the literature also denoted as the innovation or
injection term), K1 ∈ R6×3 , and K2  K3  K4 ∈ R3×3 are observer gain matrices to be determined
later. The nonlinear observer is implemented as 15 ODEs with no covariance updates. Using
Wiener bias model (8.18a) is simply replaced with

b̂˙ = K3 ỹ (8.19)

Vik and Fossen [319] has shown that using the Wiener bias model the estimation error will
converge aymptotically (asymptotic stable), while exponential stability are proven using the
Markov model.

237
Observer estimation errors
The observer estimation errors are defined as ξ̃ = ξ − ξ̂, η̃ = η − η̂, b̃ = b − b̂ and ν̃ = ν − ν̂.
Hence, from (8.9a)—(8.9e) and (8.18a)—(8.18e) the observer error dynamics is

ξ̃˙ = A ξ̃ − K1 ỹ + E w  (8.20a)
η̃˙ = R(  )ν̃ − K2 ỹ (8.20b)
b̃˙ = −T−1
 b̃ − K3 ỹ + E w  (8.20c)
˙
Mν̃ = −Dν̃ − R (  )Gη̃ + R (  )b̃ − R (  )K4 ỹ (8.20d)
ỹ = η̃ + C ξ̃ (8.20e)

By defining a new output

z̃ , K4 ỹ + Gη̃ − b̃ , C x̃  (8.21)
and the vectors ⎡ ⎤
ξ̃ ∙ ¸
w
x̃ , η̃  w ,
⎣ ⎦  (8.22)
w
b̃
the error dynamics (8.20a)—(8.20d) can be written in compact form as

Mν̃˙ = −Dν̃ − R (  )C x̃  (8.23a)

x̃˙  = A x̃ + B R(  )ν̃ + E w (8.23b)

where
⎡ ⎤
A −K1 C −K1 06×3
A = ⎣ −K2 C −K2 03×3 ⎦ 
−K3 C −K3 −T−1
£ ¤
C = K4 C K4 +G −I3×3 
⎡ ⎤ ⎡ ⎤
06×3 E 06×3
B = ⎣ I3×3 ⎦  E = ⎣ 03×3 03×3 ⎦ 
03×3 03×3 E

The observer gain matrices can be chosen such that the error dynamics is passive and GES.
It is convenient to prove passivity and stability by using an SPR-Lyapunov approach. In the
error dynamics in Figure 8.2 two new error terms ε and ε are defined according to

ε , −R (  )z̃  ε , R(  )ν̃ (8.24)

Thus, the observer error system can be viewed as two linear blocks H1 and H2 , interconnected
through the bounded transformation matrix R(  ).
Based on the physical properties of the ship dynamics, we can make the following statement:

Proposition 8.1 The mapping ε 7→ ν̃ is state strictly passive (system H1 in Figure 8.2).
Proof. Let
1
1 = ν̃  Mν̃  0 (8.25)
2

238
 z -1 ˜
- - M

D
H1
RT y R y

z̃o ˜o
x 
Co Bo

Ao
H2

Figure 8.2: Block diagram of the observer error dynamics.

DGPS
Compass
y

Bias estimator: Wave estimator:

˜y ˆy
- wave drift loads - 1st-order wave-induced motion
K4 - currents K3 K2 K1
- wind
ˆb ˆ
R ( y)
T
Cw
-
T
R (y) T -1 Aw
ˆw

 ˆ ˆ
- - M -1 R(y)

D
T
R ( y ) G

Figure 8.3: Block description of the observer.

be a positive definite storage function. From (8.23a) we have:

1
̇1 = − ν̃  (D + D )ν̃ + ν̃  ε (8.26)
2
⇓
ν̃  ε ≥ ̇1 + ν̃  ν̃ (8.27)

where  = 12 min (D + D )  0 and min (·) denotes the minimum eigenvalue. Thus, (8.27)
proves that ε 7→ ν̃ is state strictly passive, see e.g. Khalil [150]. Moreover, since this mapping
is strictly passive, post-multiplication with the bounded transformation matrix R(  ) and pre-
multiplication by it’s transpose will not affect the passivity properties. Hence the block H1 is
strictly passive. ¤

Passivity and stability of the total system will be provided if the observer gain matrices K1   K4
can be chosen such that the mapping ε 7→ z̃ is strictly positive real (SPR). This is obtained

239
if the matrices A , B  C in (8.23a)—(8.23b) satisfy the Kalman-Yakubovich-Popov (KYP)
Lemma:

Lemma 8.1 (Kalman-Yakubovich-Popov) Let Z() = C(I − A)−1 B be an  ×  transfer

function matrix, where A is Hurwitz, (A B) is controllable, and (A C) is observable. Then Z()
is strictly positive real (SPR) if and only if there exist positive definite matrices P = P  0
and Q = Q  0 such that

PA + A P = −Q (8.28)

B P = C (8.29)

Proof. See Yakubovich [329] or Khalil [150].

Theorem 8.1 (Main result: passive observer error dynamics) The nonlinear observer er-
ror dynamics (8.18a)—(8.18d) is passive if the observer gain matrices K ( = 1  4) are chosen
such that (8.23b) satisfies the KYP-Lemma.
Proof. Since it is established that H1 is strictly passive and H2 , which is given by A = A ,
B = B and C = C  can be made SPR by choosing the gain matrices K ( = 1  4) according
to the KYP lemma, the nonlinear observer error (8.18a)—(8.18d) is passive. In addition the
observer error dynamics is GES (see Figure 8.2), see Fossen and Strand [83].¤

In practice it is easy to find a set of gain matrices K ( = 1  4) satisfying the KYP lemma.
Since the mooring stiffness matrix G is assumed to be diagonal, the mapping ε 7→ z̃ will, by
choosing a diagonal structure of the observer gain matrices

∙ ¸
diag{1  2  3 }
K1 =  (8.30a)
diag{4  5  6 }
K2 = diag{7  8  9 } (8.30b)
K3 = diag{10  11  12 } (8.30c)
K4 = diag{13  14  15 } (8.30d)

be described by three decoupled transfer functions

z̃
() = diag{1 () 2 () 3 ()} (8.31)
ε
where a typical transfer function  () is presented in Figure 8.4. The wave filtering properties
is clearly seen by the notch effect in the frequency range of the wave motion. In order to meet the
SPR requirement, one necessary condition is that the transfer functions  (), ( = 1 2 3) have
phase greater than −90◦ and less than +90◦ . Regarding the choice of observer gain matrices, the
tuning procedure can be similar as for the observer for free-floating ships in Fossen and Strand
[83], where loop shaping techniques are used.

240
 Wave drift forces
Current, wind 1st-order wave loads

Notch effect

1
Tb o

Figure 8.4: A typical bode plot of the transfer function  ().

Tuning rules
In order to ensure passivity and to relate the observer gains of (8.30a) and (8.30b) to the
dominating wave response frequencies, it is proposed that
 
 = −2 (  −   )   = 1 2 and 3 (8.32)

 = 2  (  −   )   = 4 5 and 6 (8.33)
 =    = 7 8 and 9 (8.34)
where     is the filter cut-off frequency.      are tuning parameters to be set between
01 − 10 (typic values:   = 10 and   = 01). 10 − 12 should be sufficient high to ensure
proper bias estimation. One should notice that A is also dependent on the actual sea state
through the parameter   . Notice that   is often set equal to the wave peak frequency, that
is   ≈ 2 . This assumption may be appropriate for most cases. However, it is a common
misunderstanding in the literature to mix the wave response estimation with the wave amplitude
estimation. The wave response is the result of a series of signals transformations from the wave
amplitude to the wave load and finally to the wave response.

Augmented observer (representation 2)

The proposed observer can be further refined by augmenting a new state. The augmented
design provides more flexibility, better filtering and it is the basis for the adaptive observer in
Section 8.2.5. We start by adding a new state, x , in the observer, which is the low-pass filtered
innovation ỹ Moreover
ẋ = −T−1 −1
 x + ỹ = −T x + η̃ + C ξ̃ (8.35)

241
where x ∈ R3 , and T =diag{ 1   2   3 } contains positive filter constants. Highpass filtered
innovation signals can be derived from x by

ỹ = −T−1 −1
 x + ỹ = −T x + η̃ + C ξ̃ (8.36)
Thus, both the low-pass and high-pass filtered innovation are available for feedback. Moreover,
{}  )
x () = 1+   ỹ{} ()
{}   , ( = 1 2 3) (8.37)
ỹ () = 1+    ỹ{} ()

The cut-off frequency in the filters should be below the frequencies of the dominating waves in
the WF model (8.3). The augmented observer equations are

ξ̂˙ = A ξ̂ + K1h ỹ  (8.38a)

η̂˙ = R(  )ν̂ + K2 ỹ + K2l x + K2h ỹ  (8.38b)
b̂˙ = −T−1 b̂ + K ỹ + K x 
 3 3l  (8.38c)
Mν̂˙ = −Dν̂ − R (  )Gη̂ + R (  )b̂ + τ 
 

+R (  ) (K4 ỹ + K4l x + K4h ỹ )  (8.38d)

ŷ = η̂ + C ξ̂ (8.38e)
where x is the low-pass filtered innovation vector and ỹ is the high-pass filtered innovation
given by (8.35) and (8.36), respectively. Here K1h ∈ R63 and K2l , K2h , K3l , K4l , K4h ∈ R33
are new observer gain matrices to be determined.

Augmented observer error equations

The augmented observer error dynamics can be written compactly as
Mν̃˙ = −Dν̃ − R (  )C x̃  (8.39a)
x̃˙  = A x̃ + B R(  )ν̃ + E w (8.39b)
where
h i

x̃ , ξ̃ η̃  x b̃  (8.40)
z̃ , K4 ỹ + K4l x + K4h ỹ + Gη̃ − b̃ , C x̃  (8.41)
and
⎡ ⎤
A − K1h C −K1h K1h T−1 06×3
⎢ −1
−(K2 + K2h )C −(K2 + K2h ) K2l T − K2l 03×3 ⎥
⎢ ⎥
A = ⎢ ⎥
⎣ C I3×3 −T−1 03×3 ⎦
−K3 C −K3 −K3l −T−1

⎡ ⎤ ⎡ ⎤
06×3 E 06×3
⎢ I3×3 ⎥ ⎢ 03×3 03×3 ⎥
B = ⎢
⎣
⎥ E = ⎢
⎣ 03×3 03×3 ⎦ 
⎥
03×3 ⎦
03×3 03×3 E
h i
C = (K4 + K4h )C (K4 + K4h ) + G −K4h T−1
 + K4l −I3×3 

242
The signals ỹ and x are extracted from x̃ by ỹ = C x̃ and x = C x̃ where
h i £ ¤
C = C I3×3 −T−1  03×3  C = 06×3 03×3 I3×3 03×3  (8.42)

Again, the gain matrices should be chosen such that (A  B  C ) satisfies the KYP lemma in
order to obtain passivity and GES, see Strand and Fossen [277] for more details.

8.2.5 Adaptive Observer Design

In this section we treat the problem when the parameters of A in the WF model (8.9a) are not
known (Strand and Fossen [277]). The parameters vary with the different sea-states in which the
ship is operating. Gain-scheduling techniques, using off-line batch processing frequency trackers
and external sensors such as wind velocity, wave radars and roll, pitch angle measurements can
also be used to adjust the WF model parameters to varying sea states (Fossen [78]). Additional
sensors units can, however, be avoided by using an adaptive observer design. Since the wave
models are assumed to be decoupled in surge, sway and yaw, Λ and Ω in A are diagonal
matrices given by
∙ ¸ ∙ ¸
03×3 I3×3 03×3 I3×3
A (θ) = ,  (8.43)
−Ω2 −2ΛΩ −diag(θ1 ) −diag(θ2 )

where θ = [θ1  θ2 ] , and θ1  θ2 ∈ R3 contain the unknown wave model parameters to be
estimated. We start with the following assumption:

A3 Constant environmental parameters. It is assumed that the unknown parameters Ω

and Λ in A are constant or at least slowly-varying compared to the states of the system.
In addition, the wave spectrum parameters are limited by
¾
0  min     max
,  = 1 2 3 (8.44)
0   min      max

such that A is Hurwitz. Hence, the unknown wave model parameters are treated as
constants in the analysis, such that:
θ̇ = 0 (8.45)

Adaptive observer equations

The adaptive version of the observer is based on the augmented observer equations (8.38a)—
(8.38e), except from the WF part where we now propose to use the estimated WF parameters,
θ̂, such that
ξ̂˙ = A (θ̂)ξ̂ + K1h ỹ  (8.46a)
The parameter update law is

θ̂˙ = −Γ Φ(ξ̂)C x̃

= −Γ Φ(ξ̂)ỹ  Γ  0 (8.47)

where Φ(ξ̂) ∈ R6×3 is the regressor matrix. The regressor matrix is further investigated by
considering the error dynamics.

243
Adaptive observer error dynamics
The adaptive WF observer error dynamics is

ξ̃˙ = A ξ − A (θ̂)ξ̂ − K1h ỹ + E w  (8.48)

By adding and subtracting A ξ̂, defining
B Φ (ξ̂)θ̃ , (A − A (θ̂))ξ̂ (8.49)
where θ̃ = θ̂ − θ denotes the estimation error,
£ ¤
Φ (ξ̂) , diag(ξ̂1 ) diag(ξ̂2 )  (8.50)
£ ¤
B , 03×3 I3×3  (8.51)
 
where ξ̂ = [ξ̂1  ξ̂2 ] , ξ̂1  ξ̂2 ∈ R3 , and by using (8.36), then (8.48) can be rewritten as

ξ̃˙ = (A − K1f C )ξ̃ − K1f η̃ + B Φ (ξ̂)θ̃

+K1f T−1
 x + E w  (8.52)
The observer error dynamics can be written compactly as
Mν̃˙ = −Dν̃ − R (  )C x̃  (8.53a)
x̃˙  = A x̃ + B R(  )ν̃ + H Φ (ξ̂)θ̃ + E w (8.53b)
where £ ¤
H = B 03×3 03×3 03×3  (8.54)
In the adaptive case, we want the WF adaptive law to be updated by the high-pass filtered
innovations signals. Hence, it is required that the observer gain matrices are chosen such that
A P +P A = −Q  (8.55a)
B P = C  (8.55b)
H P = C  (8.55c)
in order to obtain passivity, see Strand and Fossen [277]. It should be noted that GES cannot be
guaranteed for this case since this requires persistency of excitation. However, global convergence
of all state estimation errors to zero can be guaranteed at the same time as the parameter
estimation error is bounded. We will, however, see from the experimental results that also the
parameters converge to their true values when considering a ship rest but exposed to waves.

8.2.6 Nonlinear Observer Design for Extreme Seas

Frequency domain analysis
In order to analyze the effect of the nonlinear observer assume that the rotation matrix R is equal
 
to the identity matrix I3×3 . Let us define x̂ = [ξ̂  η̂   b̂  ν̂  ]  then a state space formulation
for the linearized observer can be written
x̂˙ = Ax̂ + K (y − ŷ) 
= (A − KH) x̂ + Ky
= A x̂ + Ky (8.56)

244
 Bode Diagram
10

Magnitude (dB)
-10

-20

-30

-40
45
Phase (deg)

-45

-90
-2 -1 0 1
10 10 10 10
rad/d (rad/sec)

Figure 8.5: Transfer functions between measured and estimated LF surge position (left  = 40,
middle  = 20 and right  = 15).

where ŷ = Hx̂ and ∙ ¸

A 06×9
A=  (8.57)
09×6 A
where A is as defined in (8.4a), H is as defined in (8.13), and
⎡ ⎤
03×3 03×3 I3×3
A = ⎣ 03×3 −T−1 03×3 ⎦  (8.58)
−1
−M G M −1 −M−1 D
£  ¤
K = K1 K2 K3 (M−1 K4 )  (8.59)

When performing adaptive wave filtering subject to very high wave periods, the LF feedback
signals supposed to be controlled are filtered away (Sørensen et al. [296]). This happens because
notch effect with  = 2 will be within the bandwidth of the controller for large wave peak
periods  . This is illustrated in Fig. 8.5, that shows the transfer functions between the
estimated and measured surge position using (8.56) for a DP operated shuttle tanker (Figure
8.6) with significant wave height  = 6 and  equal to 15 (right), 20 (middle) and 40
(left). The length between perpendiculars is 256, and mass is 143000 . The passivity
property of the linearized observer is also shown by studying the phase that is within ±90 
Notice that in this example there is no mooring system such that G = 0

245
 Figure 8.6: Shuttle tanker.

Observer equations extreme seas (representation 3)

Operating in extreme seas we will propose to reformulate the observer given in (8.18a)—(8.18e)
as

η̂˙ = R(  )ν̂ + K2 ỹ (8.60a)

b̂˙ = K3 ỹ (8.60b)
Mν̂˙ = −Dν̂ − R (  )Gη̂ + R (  )b̂ + (8.60c)

τ + R (  )K4 ỹ ,
ŷ = η̂ (8.60d)

The notch effect is removed. In order to ensure bumpless transition between the two observers,
proper smooth switching schemes subject to the prevailing sea state must be introduced.

8.2.7 Experimental Results

Both the augmented and the adaptive observer have been implemented and tested at the Guid-
ance, Navigation and Control (GNC) Laboratory at the Department of Engineering Cybernetics,
NTNU. A detailed description of the laboratory is found in Strand [276]. In the experiments
Cybership I was used, see Figure 8.7.
A nonlinear PID controller is used for maintaining the ship at the desired position (   )
and heading  .

An illustration of the experimental setup is given in Figure 8.8. The experimental results are
√
transformed to full scale by requiring that the Froude number  =   =constant. Here 
is the vessel speed,  is the length of the ship and  is the acceleration of gravity.

246
 Figure 8.7: CyberShip I: Model ship scale 1:70.

position: 
p 
linear velocity: p 
angular velocity:  
linear acceleration: 1
angular acceleration:  
force:  
 
moment:
p
 
time:  

Table 8.1: Scaling factors used in the experiments (Bis scaling).

247
 GNC Laboratory

Basin

wind generator
(ducted fan)

( x, y )

Y wave generator

Figure 8.8: Experimental setup in model basin.

The scaling factors are given in Table 8.1 where  is the mass and the subscripts  and 
denote the model and the full-scale ship, respectively. The length of the model ship is  = 119
meters and the mass is  = 176 kg. A full scale ship similar to Cybership I has typically a
length of 70 − 90 meters and mass of 4000 − 5000 tones.
The experiment can be divided in three phases:

• Phase I (No waves). Initially the ship is maintaining the desired position and heading with
no environmental loads acting on the ship (calm water). The reference heading is -140
degrees. When the data acquisition starts, a wind fan is switched on. There is no adaptive
wave filtering and the observer is identical to the augmented design (Representation 2).
The effect of the wind loads are reflected in the bias estimates in Figure 8.10.

• Phase II (Waves, adaptive wave filter is off). After 1700 seconds the wave generator is
started. In this phase we can see the performance of the observer without adaptive wave
filter. In the wave model we are assuming that the dominating wave period is 9.2 seconds
and the relative damping ratio is 0.1, see Figure 8.9.

• Phase III (Waves, adaptive wave filter is on). After 2800 seconds the adaptive wave filter
is activated. The estimates of dominating wave period and relative damping are plotted
in Figure 8.9 for surge, sway and yaw.

A spectrum analysis of the position and heading measurements shows that the estimated wave
periods converge to their true values, that is wave periods of approximately 7.8 seconds and
relative damping ratios of 0.07, see Figure 8.9. In Figure 8.10 the measured position deviation
and heading are plotted together with the corresponding LF estimates. The effect of the adaptive
wave filtering is clearly seen in Figure 8.12, where the innovation signals are significantly reduced
during Phase III, when the adaptation is active and the wave model parameters start converging

248
 I II III I II III

Figure 8.9: Estimated wave periods (left) and estimated relative damping ratios (right) for
surge (solid), sway (dashed) and yaw (dotted). The adaptive wave-filter is activated after 2800
seconds.

to their true values. The effect of bad wave filtering is reflected by noisy control action by the
thrusters during phase II, see Figure 8.12. A zoom-in of the heading measurement together with
the LF estimate is given is Figure 8.11 both for phase II and III. Here we see that the LF estimates
have a significant WF contribution when the adaptive wave filter is off. This is the reason for
the noisy control action in phase II. The other zoom-in shows excellent LF estimation when
the adaptive wave filter is active and the wave model parameters have converged to their true
values. Hence, it can be concluded that adaptive wave filtering yields a significant improvement
in performance compared to filters with fixed WF model parameters operating in varying sea
states.

249
 Measured (gray) and estimated LF (solid) x-position deviation [m] Estimated bias in surge [kN]

Measured (gray) and estimated LF (solid) y-position deviation [m] Estimated bias in sway [kN]

Measured (gray), estimated LF (solid) and desired (dotted) heading [deg] Estimated bias in yaw [kNm]

I II III I II III

Figure 8.10: Left column: Measured position and heading (gray) together with corresponding
LF estimates (solid). Right column: Estimated bias in surge, sway and yaw.

Zoom-In: Measured (dotted) and estimated LF (solid) heading (dashed) [deg]

II: Adaptive Wave-Filter Off III: Adaptive Wave-Filter On

Figure 8.11: Zoom-in of measured and estimated LF heading. Left: Observer without adaptive
wave-filtering. Right: Observer with adaptive wave-filtering.

250
 I II III I II III

Figure 8.12: Left column: Innovation in position and heading. Right column: Commanded thrust
in surge, sway and yaw.

251
8.3 Controller Design for Dynamic Positioning
In the design of dynamic positioning (DP) systems and thruster assisted position mooring (PM)
systems it has been adequate to regard the control objective as a three degrees-of-freedom (DOF)
problem in the horizontal-plane in surge, sway and yaw respectively However as mentioned in
Section 7, for semi-submersibles an unintentional coupling phenomena between the vertical and
the horizontal planes through the thruster action can be invoked. The natural periods in roll
and pitch are in the range of 35-65 seconds and are within the bandwidth of the positioning
controller. In Sørensen and Strand [288] and [290] it was shown that roll and pitch may be
unintentionally excited by the thruster system, which is only supposed to act in the horizontal-
plane. If the inherent vertical damping properties are small, the amplitudes of roll and pitch
may be emphasized by the thruster’s induction by up to 2◦ -5◦ in the resonance range.
In this section different model-based multivariable control strategies accounting for both
horizontal and vertical motions, with the exception of heave, is described. Since it is undesirable
to counteract the wave-frequency (WF) motions caused by first-order wave loads, the control
action of the propulsion system is produced by the low-frequency (LF) part of the vessel’s motion,
which is caused by current, wind and second-order mean and slowly varying wave loads. The
computation of feedback signals to be used in the controller are based on nonlinear observer
theory as presented in the Section 8.2.

8.3.1 Control Plant Model

The control plan model is as in Section 8.2.2.

Linear low-frequency model

For the purpose of controller design and analysis, it is convenient to derive a linear LF control
plant model about zero vessel velocity. By assuming small roll and pitch amplitudes, and that
the yaw angle is defined with respect to the desired heading angle, the rotation matrix J(η 2 ) can
be approximated by the identity matrix, see Section 7.2. The measured position and heading
signals are transformed into the reference-parallel frame before the estimator and the controller
process them.
In the new control strategy for small-waterplane-area marine constructions the conventional
3 DOF multivariable controller in surge, sway and yaw will be extended to account for couplings
to roll and pitch. In order to derive the new controller it is appropriate to define the model

252
reduction matrices ⎡ ⎤
1 0 0 0 0 0
⎢ 0 1 0 0 0 0 ⎥
⎢ ⎥
H5×6 =⎢
⎢ 0 0 0 1 0 0 ⎥,
⎥
⎣ 0 0 0 0 1 0 ⎦
0 0 0 0 0 1
⎡ ⎤
1 0 0
⎢ 0 1 0 ⎥
⎢ ⎥ (8.61)
H5×3 =⎢
⎢ 0 0 0 ⎥
⎥
⎣ 0 0 0 ⎦
0 0 1
⎡ ⎤
1 0 0 0 0 0
H3×6 = ⎣ 0 1 0 0 0 0 ⎦,
0 0 0 0 0 1
H3×3 = I3×3  H6×6 = I6×6 
Hence, the  ×  dimensional mass, damping and restoring matrices can be written in reduced
order form according to
M = H×6 MH×6 ,
D = H×6 (D + D )H×6  (8.62)
G = H×6 (G + G )H×6 
where  = 3 describes the conventional horizontal model matrices in surge, sway and yaw, and
 = 5 will represent the 5 DOF model of surge, sway, roll, pitch and yaw. D  D  G and
G are as specified in Section 7.3. In reference parallel frame the linearized control plant model
can be written
M ν̇  + D ν  + G η  = τ  + w  (8.63)
The corresponding linear LF state-space model can be formulated as
ẋ = A x + B τ  + E w
(8.64)
y = C x + v 
For  = 3 and  = 5 the state-space vectors become

x3 = [     ]  (8.65)

x5 = [         ]  (8.66)

τ  ∈ R3 is the commanded control vector. w , y  v ∈ Ri are the disturbance, measurement

and noise vectors respectively. The system matrix A ∈ R2i×2i is then written
∙ ¸
−M−1 D −M−1  G
A =  (8.67)
I× 0×

where I× ∈ Ri×i is the identity matrix and 0× ∈ Ri×i is the zero matrix. The control input
matrix B ∈ R2i×3 is defined to be
∙ ¸
M−1
 H×3
B =  (8.68)
0× H×3

253
The disturbance matrix E ∈ R2i×i becomes
∙ ¸
M−1

E =  (8.69)
0×

In this text it is assumed that only the position and the orientation are measured. Converting
the position and heading measurements to the reference-parallel frame, the LF measurement
matrix C ∈ Ri×2i becomes £ ¤
C = 0× I×  (8.70)

Linear wave-frequency model

In the controller design synthetic white-noise-driven processes consisting of uncoupled harmonic
oscillators with damping will be used to model the WF motions. Applying model reduction as
for the LF model, the synthetic WF model can be written in state-space form for  = 3 and
 = 5 according to
ξ̇ = A ξ + E w 
(8.71)
η  = C ξ 
η  ∈ Ri is the measurement vector of the wave-frequency motion, w ∈ R is a zero-mean
Gaussian white noise vector, and ξ ∈ R2 . The system matrix A ∈ R2i×2i , the disturbance
matrix E ∈ R2i×i and the measurement matrix C ∈ Ri×2i are formulated as in Section 8.2.2.

8.3.2 Horizontal-plane Controller

The horizontal-plane positioning controller consists of feedback and feedforward controller terms,
which may be linear or nonlinear. Usually, linear controllers have been used. However, extensive
research in this field has contributed to the introduction of nonlinear control theory, see Strand
[276] and the references therein.

Linear horizontal-plane PD control law based on LQG synthesis

A conventional linear multivariable proportional and derivative (PD) type controller based on
the horizontal-plane surge, sway and yaw DOF can be written

τ pd = −Ge = −G e2 − G e1  (8.72)

where the the error vector is decomposed in the reference-parallel frame according to

e2 = R (  ) [η̂ − η  ]  (8.73)
e1 = ė2 (8.74)
h i
with η̂ = ̂ ̂ ̂  η̂  = [     ] and G  G ∈ R3×3 are the non-negative controller gain
matrices found by appropriate control synthesis methods. If we only consider the proportional
and derivative terms for the horizontal-plane modes of motion in the control objective for (8.64)
and assume that the pair (A3  B3 ) is reachable, we can formulate a linear quadratic performance
index according to ½ Z ¾
1   
 =  lim e Qe + τ pd Pτ pd   (8.75)
 →∞  0

254
 £ ¤
where error vector e ∈R6 is defined as e = e1  e2 and Q P ∈ R3×3 The following properties
hold for the error weighting matrix Q = Q ≥ 0 and the control weighting matrix P = P  0.
By minimizing the control index , the Linear Quadratic Gaussian - LQG method returns the
control gain matrix G ∈ R3×6  For LTI systems, stationary solutions of the Ricatti equation can
be found. Hence, the Ricatti equation computations reduce to solving

Ṙ = −A3 R − A3 R + RB3 P−1 B3 R − Q (8.76)

⇓ Ṙ → 0 and R → R∞ (8.77)
0 = −A3 R∞ − A3 R∞ + R∞ B3 P −1
B3 R∞ − Q (8.78)

where R∞ is a 6 × 6 dimensional non-negative symmetric matrix. The stationary LQG gain

matrix is then £ ¤
G = P−1 B3 R∞ = G G  (8.79)

Remark 8.2 We have here utilized the so-called Separation theorem expressing that control of
a linear system is conducted with stochastic excitation and the measurement vector derived by
the estimated state vector provided by an observer. Thus, the controller consists of two separate
functions: state estimation and feedback control from the estimated state vector.

Integral action
In order to meet the special command-following and disturbance-rejection performance specifi-
cations, it is necessary to append three free integrators to the control model. It is then desirable
to define a property space z which expresses the variables to be controlled towards certain set-
points (Balchen [20]). The dimension of the property space is the same as the control vector
space, that is τ  ∈ R3 implies that z ∈ R3 . The variables to be controlled to the setpoints are
fewer than those constituting the full state space e ∈ R6 . The relation between these variables
and the state space is given through the transformation

z = g(e) = e2  (8.80)

A multivariable PI algorithm is then achieved by means of an integral loop, controlling the

property z in parallel with the proportional LQ control loop. Inspired by classical SISO PI
controller tuning, one can specify the eigenvalues of the integral loop in a diagonal matrix Λ .
Since the integral loop will be slower than the proportional LQ loop, the following approximation
for the integral loop can be found

ż = G G (A3 − B3 G)−1 B3 z = Λ z (8.81)

where G ∈ R3×6 is a property matrix given by

g(e) £ ¤
G = = 03×3  I3×3  (8.82)
e
This leads to the integral loop-gain matrix
³ ´−1
G = Λ G (A3 − B3 G)−1 B3  (8.83)

where the matrix G (A3 − B3 G)−1 B3 must be non-singular. Open integrators, as they appear
both in the control loop and the estimator, should be equipped with anti-windup precautions in

255
case their outputs reach amplitude restrictions such as saturation in the physical control devices.
The resulting integral control law is written

τ̇  = Awi τ  + G z (8.84)

where Awi ∈ R3×3 is the anti-windup precaution matrix, and −G ∈ R3×3 is the non-negative
integrator gain matrix. Integrator anti-windup should be implemented in order to avoid that
the integrator integrates up beyond the saturation limits of the actuators. A simple method
to avoid this is to stop integrating when the integral term tends to a given percentage of the
actuator saturation limit. More sophisticated algorithms are found in Levine [170], Åstrøm and
Wittenmark [333] and Egeland [63].

Wind feedforward control action

Wind loads have an important impact on the vessel’s response. Especially for semisubmersibles
the wind loads are dominating. In order to obtain fast disturbance rejection with respect to
varying wind loads, it is desirable to introduce a wind feedforward controller. The wind forces
and moment in surge, sway and yaw may be estimated by a Kalman filter and a Luenberger
observer, where the wind velocity and wind direction are measured. The estimates are multiplied
by a feedforward gain matrix. The wind feedforward control law is taken to be

τ w = −G τ̂ wind  (8.85)

where τ̂ wind ∈ R3 is the vector of estimated wind forces and moment in surge, sway and yaw
respectively. It is assumed that the coupling in wind loads is covered by the wind coefficients,
so that the gain matrix G ∈ R3×3 is a non-negative diagonal matrix.

Model reference feedforward control action

In order to improve the performance of the controller during tracking operations a feedforward
control action based on input from the reference model is included. The feedforward control
action is written
τ t = M3 a + D3 v + d3 (v ) + C(v )v  (8.86)
where a and v ∈ R3 are the desired generalized reference acceleration and velocity vectors
respectively computed by appropriate reference models.

Reference model
For DP operations close to other offshore structures or ships, it is crucial to be capable to perform
controlled movements and rotations of the vessel. The automatic guidance function which takes
the vessel from the prevailing setpoint coordinates to the new setpoint is defined as the “marked
position”, and can be specified either in the reference-parallel frame or in the Earth-fixed frame.
The movement and rotation can be done for each degree of freedom (DOF) separately, or as a
fully 3 DOF coupled motion. In order to provide high-performance DP operations with bumbles
transfer between station-keeping and marked position operations, a reference model must be
introduced for the calculation of feasible trajectories of the desired vessel motion for each degree

256
of freedom. Experience achieved from full-scale experiments has demonstrated that the following
reference model in the Earth-fixed frame is appropriate

a + Ωv + Γx = Γx  (8.87)

ẋ = −A x + A η   (8.88)

where a  v and x ∈ R3 define the desired vessel acceleration, velocity and position trajectories
in the Earth-fixed frame. The vector η  ∈ R3 defines the new reference coordinates, either
relative to the previous setpoint, or as global Earth-fixed coordinates. The vector x ∈ R3
defines the filtered reference coordinates. The design parameters in the reference model consist
of a non-negative diagonal damping matrix Ω ∈ R3×3 and a diagonal stiffness matrix Γ ∈ R3×3
written as

Ω= {2    }  = 12 3 (8.89)

Γ= { 2 }  = 1  2  3 (8.90)

The first order diagonal and non-negative setpoint filter gain matrix A ∈ R3 is written

A = {1 }  = 1  2  3 (8.91)

The Earth-fixed desired vessel acceleration, velocity and position trajectories are transformed
into a moving reference-parallel frame which follows the desired Earth-fixed position and heading
trajectory. Hence,

a = R (  )a  (8.92)
v = R (  )v 

(8.93)
x = R (  )x  (8.94)

The reference model may also run in station-keeping operations, but then with the desired
velocity and acceleration equal to zero. This provides a smooth transfer between the operational
modes.

Resulting control law

The resulting horizontal-plane station keeping control law, including both feedback and feedfor-
ward terms is written
τ 3 = τ w + τ t + τ  + τ pd  (8.95)
where the different terms are defined below. A graphical illustration of the controller is
shown in Figure 8.13.

8.3.3 Horizontal-plane Controller with Roll-Pitch Damping

The horizontal-plane positioning control law (8.95) can be extended to also include roll and
pitch damping. This is motivated by the fact the hydrodynamic couplings the surge and sway
feedback loops will be extended to incorporate feedback from the low frequency estimated pitch
and roll angular velocities, denoted as ̂ and ̂ respectively.

257
Roll-pitch control law
By using a linear formulation the roll-pitch control law is formulated according to
∙ ¸
̂
τ rpd = −Grpd  (8.96)
̂

where the roll-pitch controller gain matrix Grpd ∈ R3×2 is defined as

⎡ ⎤
0 
Grpd = ⎣  0 ⎦  (8.97)
 0

and  ,  and  are the corresponding non-negative roll-pitch controller gains.

Resulting control law

The resulting positioning control law including roll-pitch damping becomes

τ 5 = τ 3 + τ rpd  (8.98)

8.3.4 Controller Analysis

The effect of the roll-pitch damping controller can be shown by analyzing the impact of the
extended control law given by (8.98) on the linearized model given in (8.64). The coupled
linearized low-frequency surge-pitch and sway-roll-yaw models will be considered. In the analysis
it is assumed that the wind feedforward controller and the integral control action will compensate
for the wind load, the current load and the mean second-order wave drift loads. Hence, the
linearized coupled surge-pitch model can be written

11 ̇ + 15 ̇ + 11  + 15  + 11  =  5surge  (8.99)

51 ̇ + 55 ̇ + 51  + 55  + 55  = 0 (8.100)

where subscript  reflects the element of the matrices (8.62) for the case  = 6 that is all six
DOFs are included. By inserting (8.99) in (8.100), the pitch dynamics can be reformulated to

51 15 51 15 51 11 51 11 51

(55 − )̇ +(55 − ) +55 +(51 − )− =−  5surge  (8.101)
11 11 11 11 11
The pure surge part of the multivariable control law in (8.98), assuming only PD and the roll-
pitch control actions is written as
11
 5surge = −  −   +   (8.102)
51
where    ≥ 0 are the PD controller gains. Substituting (8.102) in (8.101) gives the following
closed-loop pitch dynamics

(55 − 51 15 51 15

11 )̇ + (55 − 11 +  ) + 55 + (8.103)
51
(51 − 11 −  11 ) − (11 +  ) 
51 11
11  = 0
51

258
 Wind Disturbance Measurement
Measurement noise
w[k]
vw[k],w[k] v[k]

u[k] Real y[k]

Process

[k] [k]
K[k]
Wind
Observer w[k] x[k+1] x[k] y[k]
[k] C

x[k]
[k]
Observer

u[k] x[k]

Thruster
Allocation
xd[k] Reference
w[k] Signal
pd[k] (x[k]-xd[k])

i[k] ([k]-d[k])

rpd[k] (pd[k],qd[k])

t[k] Reference (d[k],ad[k])

Feed Forward

Positioning Controller

Figure 8.13: Illustration of the DP controller.

259
Assume that the horizontal-plane controller will force the surge velocity  and position deviation
 to zero. Hence, the closed-loop pitch dynamics in (8.103) can be simplified to
51 15 51 15
(55 − )̇ + (55 − +  ) + 55  = 0 (8.104)
11 11
(8.104) can be recognized as a second-order mass-damper-spring system, where it is clearly
shown that the effect of the new control law increases the damping in pitch.
A similar formulation can be found for the coupled sway-roll-yaw model. The coupled equa-
tions of motion in sway, roll and yaw are
22 ̇ + 24 ̇ + 26 ̇ + 22  + 24  + 26  + 22  =  5sway  (8.105)
42 ̇ + 44 ̇ + 46 ̇ + 42  + 44  + 46  + 44  = 0 (8.106)
62 ̇ + 64 ̇ + 66 ̇ + 62  + 64  + 66  + 66  =  5yaw  (8.107)
where  5sway and  5yaw are the control actions in sway and yaw, respectively. Thus, by including
feedback from angular roll velocity in the sway and yaw control laws

 5sway = −  −   −   −   −  (8.108)


 5yaw = −  −   −   −   −  (8.109)

a similar damping effect as in the pitch case can be obtained. Here,  ,  ,  ,  ≥ 0 are the
diagonal gains of the PD part of the control law. Moreover,      and  are the off-
diagonal controller gains with appropriate signs reflecting the hydrodynamic coupling between
sway and yaw. As for surge, assume that the sway and yaw velocities,  and , position and
angle,  and  are forced to zero by the horizontal-plane controller. Thus, the closed-loop roll
dynamics becomes
 ̇ + ( +  +  )  + 44  = 0 (8.110)
where
 26 µ ¶
42 24 46 − 42 22 62 24 64
 = 44 − + −  (8.111)
22 1−  62 26
66 22 66 22 66
 26 µ ¶
42 24 46 − 4222 62 24 64
 = 44 − + −  (8.112)
22 1− 62 26
66 22 66 22 66
42 26
42 62 22
−46
 = 22 + 22 66 (1− 62 26 )  (8.113)
  66 22

42 26 
1 46 − 22
 = 66 (1− 62 26
)
 (8.114)
 
66 22

It is clearly shown here how the damping is increased in roll by applying the extended control
law.

8.3.5 Thrust Allocation

Optimal thrust allocation
The relation between the control vector τ  ∈ R3 and the produced thrust from the  thrusters
T ∈ Rr is defined by
τ  = T3× (α)T  (8.115)

260
where T3× (α) ∈ R3×r is the thrust configuration matrix, and α ∈ Rr is the thruster orientation
vector. Let T = Ku  where K ∈ Rr×r is the diagonal matrix of thrust force coefficients written
K = { } u ∈ Rr is the control vector of either pitch-controlled, revolution-controlled or
torque- and power-controlled propeller inputs as treated in Chapter 9. The thrust provided by
the thruster unit  disregarding the thrust losses, is calculated to be

 =    (8.116)

For a fixed mounted propeller or thruster the corresponding orientation angle is set to a fixed
value reflecting the actual orientation of the device itself. In case of an azimuthing thruster,
 is an additional control input to be determined by the thrust allocation algorithm. The
commanded control action provided by the thrusters becomes

u = K−1 T+
3× (α)τ   (8.117)

where T+
3× (α) ∈ R
r×3 is the pseudo-inverse thrust configuration matrix.

Geometrical thrust induction

The effect of the commanded thruster action provided by the thrusters in (8.117) on the six
degrees-of-freedom vessel model (7.67) defined in the mathematical modelling Section 7.3 can
be calculated to be
τ thr = T6× (α)Ku  (8.118)
where τ thr ∈ R6 is the corresponding actual control vector acting the vessel as shown in Sec-
tion 7.3, and T6× (α) ∈ R6×r is the thrust configuration matrix accounting for the six DOF
contribution of the produced thruster actions. The reader should notice that (8.118) will intro-
duce roll and pitch moments, that may be important to consider for rigs as they may introduce
unintentional roll and pitch motions.

Example 8.1 Consider the thrust configuration as shown in Figure 8.14. The ship is in the
bow equipped with one azimuthing thruster and one tunnel thruster located the distance / and
 from the origin, respectively. One tunnel thruster is located stern the distance  from the
origin. Finally, two podded units are located stern at the longitudinal distance  and transverse
distance  and  from the origin, respectively. The thrusters are numbered according to: 1.
Bow azimuthing thruster, 2. Bow tunnel thruster, 3. Stern tunnel thruster, 4. Starboard pod,
and 5. Port pod. The thrust configuration matrix becomes accordingly
⎡ ⎤
cos 1 0 0 cos 4 cos 5
3×5 = ⎣ sin 1 1 1 sin 4 sin 5 ⎦
 sin 1  − − sin 4 −  cos 4  sin 5 −  cos 5

where the azimuthing angle  is defined such that  = 0 gives maximum positive surge thrust
and  = 90 gives maximum positive sway thrust.

261
 1. Bow azimuthing thruster

lba
2. Bow tunnel thruster

Surge
lbt Yaw

Sway

lst
3. Stern tunnel thruster

lp

bp bs
5. Port pod 4. Starboard pod

Figure 8.14: Example of a thruster configuration.

262
 Figure 8.15: Semi-submersible.

8.3.6 Case Study

A simulation study of a dynamically positioned semi-submersible is carried out to demonstrate
the effect of the thruster induced roll and pitch motions. The performance of the new control
strategy, denoted as τ 5  and the conventional horizontal control law, denoted as τ 3  is com-
pared. A semi-submersible, see Figure 8.15, equipped with 4 azimuthing thrusters, each able to
produce a force of 1000  located at the four corners at the two pontoons, is used in the sim-
ulations. The operational draft is equal to 24 , vessel mass at operational draft is 45000 ,
the length is 110 , and the breadth is 75 . Radius of gyration in roll is 30 , in pitch equal
to 33  and in yaw 38 . The undamped resonance periods in roll and pitch are found to be
55  and 60 , respectively.

Frequency domain analysis

The linearized system given in (8.64) is analyzed in the frequency domain. The transfer functions
of the coupled surge-pitch model subjected to disturbances are shown for the open-loop system,
closed loop system with the conventional horizontal-plane (3 DOF) controller and the new
extended roll-pitch (5 DOF) controller. In Figure 8.16 it can be seen that a significant notch-
effect is achieved about the pitch resonance frequency with the new control law. Figures 8.17
and 8.18 show how the damping in pitch is amplified about the resonance frequency applying
the new control law. Similar progress for the coupled sway-roll-yaw model can also be shown.

263
 Transfer function from disturbance in surge to surge position
0

-10

Magnitude [dB] -20

-30

-40

-50 Open Loop

Closed Loop 3 DOF
Closed Loop 5 DOF
-60
0 0.005 0.01 0.015 0.02 0.025 0.03
Frequency [Hz]

Figure 8.16: Normalized transfer function from disturbance in surge to surge position for the
open-loop system (solid), applying conventional controller τ 3 (dotted) and new controller τ 5
(dashed).

Transfer function from disturbance in surge to pitch angle

0

-10

-20
Magnitude [dB]

-30

-40

-50

-60

Open Loop
-70 Closed Loop 3 DOF
Closed Loop 5 DOF
-80
0 0.005 0.01 0.015 0.02 0.025 0.03
Frequency [Hz]

Figure 8.17: Normalized transfer function from disturbance in surge to pitch angle for the
open-loop system (solid), applying conventional controller τ 3 (dotted) and new controller τ 5
(dashed).

264
 Transfer function from disturbance in pitch to pitch angle
0

-5

-10

-15

Magnitude [dB]
-20

-25

-30

-35
Open Loop
-40 Closed Loop 3 DOF
Closed Loop 5 DOF
-45
0 0.005 0.01 0.015 0.02 0.025 0.03
Frequency [Hz]

Figure 8.18: Normalized transfer function from disturbance in pitch to pitch angle for the
open-loop system (solid), applying conventional controller τ 3 (dotted) and new controller τ 5
(dashed).

Time domain numerical simulations

The time domain numerical simulations are performed with a significant wave height of 6 , wave
peak period 10 , current velocity 05  and wind velocity 10 . The Earth-fixed directions
of waves, wind and current are colinear and equal to 225o in this simulation. The induced thrust
components in roll and pitch are clearly shown in Figures 8.21 and 8.22. Accordingly, it is the
important to control these components by the roll-pitch control law to ensure a safe and optimal
operation.
By applying the proposed new control strategy the roll and pitch amplitudes are significantly
reduced as shown in the time series of Figures 8.19 and 8.20. It is also evident that the new
controller does not reduce the horizontal positioning accuracy (Figures 8.19 and 8.20).
The effect of the suppressed roll and pitch motions is, as expected, most effective about the
resonance frequencies, as shown in the power spectra of Figure 8.23. The simulations further in-
dicate that by maintaining the same level of positioning accuracy, the total energy consumption
when applying the new controller will be lower than using the conventional design philosophy.
Oscillations in roll and pitch will, through the hydrodynamic coupling, induce motions in surge,
sway and yaw, which have to be compensated for by the controller anyhow. In the roll-pitch con-
trol strategy these coupling effects are exploited in an optimal manner to damp the oscillations,
and thereby also avoiding unintentional roll and pitch induced surge, sway and yaw motions.

265
 0.3

0.2

0.1

Surge [m]
0

-0.1

-0.2

-0.3

-0.4
0 100 200 300 400 500 600 700 800 900 1000

0.4
With roll-pitch damping
0.2 No roll-pitch damping

0
Pitch [deg]

-0.2

-0.4

-0.6

-0.8

-1
0 100 200 300 400 500 600 700 800 900 1000
time [sec]

Figure 8.19: Surge position and pitch angle applying conventional controller τ 3 (dotted) and
new controller τ 5 (solid).

0.5
Sway [m]

-0.5

-1
0 100 200 300 400 500 600 700 800 900 1000

1
Roll [deg]

0.5

-0.5

-1
0 100 200 300 400 500 600 700 800 900 1000

0.15
Heading [deg]

0.1 With roll-pitch damping

No roll-pitch damping
0.05

-0.05

-0.1
0 100 200 300 400 500 600 700 800 900 1000

time [sec]

Figure 8.20: Sway position, roll and yaw angles applying conventional controller τ 3 (dotted)
and new controller τ 5 (solid).

266
 700

600

Tau surge [kN]

500

400

300

200

100

0
0 100 200 300 400 500 600 700 800 900 1000

4
x 10
2.5
With roll-pitch damping
No roll-pitch damping
2
Tau pitch [kNm]

1.5

0.5

0
0 100 200 300 400 500 600 700 800 900 1000
time [sec]

Figure 8.21: Produced thrust components in surge and pitch applying conventional controller
τ 3 (dotted) and new controller τ 5 (solid).

2000
Tau sway [kN]

1500

1000

500

0
0 100 200 300 400 500 600 700 800 900 1000
4
x 10
0
Tau roll [kNm]

-1

-2

-3

-4

-5
0 100 200 300 400 500 600 700 800 900 1000

8000
Tau yaw [kNm]

6000

4000

2000
With roll-pitch damping
0 No roll-pitch damping

-2000
0 100 200 300 400 500 600 700 800 900 1000

time [sec]

Figure 8.22: Produced thrust components in sway, roll and yaw applying conventional controller
τ 3 (dotted) and new controller τ 5 (solid).

267
 -3
x 10
8
7

Spectrum, roll
5
4
3

2
1
0
0 0.01 0.02 0.03 0.04 0.05

-3
x 10
8
With roll-pitch damping
7 No roll-pitch damping
Spectrum, pitch

6
5
4
3

2
1
0
0 0.01 0.02 0.03 0.04 0.05

Frequency [Hz]

Figure 8.23: Power spectra of roll and pitch applying conventional controller τ 3 (dotted) and
new controller τ 5 (solid).

8.4 Hybrid Control

Hybrid control has earlier been realized in the control of airplanes and other land-based vehicles.
Although switching has been used in other applications, stability is a concern when switching
among nonlinear controllers. Extensive work on the theory of supervisory switching control has
been presented by Hespanha [114], Hespanha and Morse [115], Hespanha et al. [116]. The work
on hybrid DP control in Nguyen[201], Nguyen et al. [202], [203] and Nguyen and Sørensen [206]
are based on these results. Hybrid control enables switching either among linear or nonlinear
controllers according to the prevailing operational regimes. While dwell-time switching logic is
employed in the linear supervisory control, scale-independent hysteresis switching logic is used
in nonlinear supervisory control ensuring stability of the whole system and to prevent chattering.
Supervisory control used for DP system is more advantageous than gain scheduling control in
terms of flexibility and modularity, and decoupling between supervision and control (Hespanha,
[114]). In the supervisory control, the design of the candidate controllers is done separately with
the adaptive mechanism (the supervisor) whereas in the gain scheduling control, the candidate
controllers usually have to be particularly designed to satisfy the tuning mechanism. In the
design of hybrid-controller DP system, the supervisory control allows the switching among the
bank of controllers which are structurally different such as observer with WF filtering and PID
controller in normal sea and observer without WF filtering, AFB controller in harsh environ-
mental conditions, and inclusion of mooring models in the observer for thruster assisted position
mooring. This is not allowed in the gain scheduling control due to the similarity requirement
of the controller structure except the gains. This is important for the DP vessel that requires
the design of structurally different controllers satisfying the structural changes in the hydrody-
namics, and the performance requirements subject to varying environmental and operational
conditions.

268
8.4.1 Control Plant Model
The control plant model for the WF model is here written as a second-order linear model driven
by white noise, wpw ∈ R3 , according to

ṗw = Apw pw + Epw wpw  (8.119)

where pw ∈ R6 is the state of the WF model, Apw ∈ R6×6 is assumed Hurwitz, and describes
the first-order WF-induced motion as a mass-damper-spring system, according to
∙ ¸
03×3 I3×3
Apw =  (8.120)
−Ω2 −2ΛΩ

where Ω = diag( 1   2  3 ) is a diagonal matrix containing the dominating wave response fre-
quencies, and Λ = diag(1  2  3 ) is a diagonal matrix of damping ratios ( is often set between
0.05 and 0.2).
By assuming fixed anchor line length, the generalized mooring forces in a working point may
be approximated by a first-order Taylor expansion of the static restoring and damping forces
about a working point η = η o and ν r = ν = 0, according to

τ moor = −R> ()Gmo (η − η o ) − Dmo ν (8.121)

¯ ¯
gmo ¯ dmo ¯
where Gmo =  ¯= , Dmo =  =0  Let us assume that CRB (ν)ν, CA (ν r )ν r and the
o
nonlinear damping are small since the vessel’s velocity is small in station keeping. Based on these
assumptions and considering only surge, sway and yaw motions, v = [  ]  η = [  ] the
LF model in can be simplified such that

η̇ = R()ν (8.122)
ḃ = −T−1
b b + Eb wb 
Mν̇ = −Dν + R ()b − Gmo R> ()η + τ c 
>

where b ∈ R3 is the bias vector accounting for both slowly varying disturbances and unmodeled
dynamics, and Eb is a diagonal scaling matrix. The damping effect of the mooring system is
included in the damping matrix, such that D = DL + Dmo . τ c = [        ] is the control
input vector. The output of the control plant model, y ∈ R3 , is the position and heading of the
vessel, assumed to be a superposition of the WF and LF motion, according to

y = η + Cpw pw + v (8.123)

where v ∈ R3 is the zero-mean Gaussian white noise vector, and Cpw = [03×3 I3×3 ].

269
8.4.2 Observer Design
The nonlinear passive observer operating in open water is found by copying the control plant
model in (8.119)—(8.123) such that
p̂˙ w = Apw p̂w + K1 ỹ (8.124)
η̂˙ = R(  )ν̂ + K2 ỹ (8.125)
˙
b̂ = −T−1
b b̂ + K3 ỹ (8.126)
˙
Mν̂ = −Dν̂ + R> (  )b̂ − Gmo R> ( )η̂ + τ c
+K4 R> (  )ỹ (8.127)
ŷ = η̂ + Cpw p̂w  (8.128)
where ỹ = y − ŷ is the estimation error (in the literature also denoted as the innovation or
injection term); K1 ∈ R6×3 , K2 ∈ R3×3 , K3 ∈ R3×3 , and K4 ∈ R3×3 are the observer gain
matrices.

8.4.3 Vessel Operational Condition

As proposed in Perez et al. [221] the different control plant models subject to the various control
objectives, constraints and dynamic response of the controlled system can be formulated. Based
on this, the vessel operational condition (VOC) is defined as a space consisting of three main
dimensions (Fig. 8.24), according to:
• Mode dependence (x axis),
• Speed dependence (y axis),
• Environmental dependence (z axis).
One may consider an additional dimension which is fault-tolerant control dependence. Changes
in these dimensions result in changes in the fundamental components of the control problem (ob-
jectives, constraints and dynamic response of the controlled system).
Changes in operation mode: DP vessels operate in variety of modes, such as station keep-
ing, setpoint changes, thruster assisted position mooring (PM), low speed maneuvering, and
moderate to high speed transit operations.
Changes in speed: The changes in speed result in changes in the dynamic response of the
vessel (high-level) and thrusters (low-level). The latter will be accounted in the control allocation
scheme and the thruster and rudder controllers. In the high-level controller this can be captured
as changes in the parameters of a model, and even in the structure of the model and the controller
itself. For example, while the effects of nonlinear damping can be neglected in the zero speed
regimes, e.g. DP applications, those should be included in the controller design for higher speed
regimes.
Changes in environment: Changes in the environment result in changes of the disturbance
characteristics (frequency and intensity). The control objective of a DP vessel from calm to
moderate sea is to keep its position and heading by compensating for only LF motion. As the
sea state increases, WF motion is induced by waves with lower dominant frequencies, especially
swell in the North Sea and Barents Sea. For this reason, the control objective is to compensate
both WF and LF motions instead of only LF in moderate sea. For the low-level thruster
controllers, anti-spin control will be activated in high seas states.

270
 z E nvi ron me nt
 

e
m
t re
Ex

h
ig
H
e
at
er
od
M
x
lm
Ca
Mo de
2 g
in
ep

4
ke

g
ri n
n
io

ve
at

6
eu
St

an
m

10
d

Op era ti on al dir ectio n

ee
sp

y 20
w
Lo

Sp ee d
(m/s)
sit
an
Tr

Figure 8.24: Vessel operational conditions.

8.4.4 Concept of Hybrid Control

Structure
The idea of hybrid-controller system is the ability to automatically switch among controllers.
There are two main blocks in the hybrid-controller system: the supervisor and the controller set
(Fig. 8.25).
In the estimator-based supervision, the supervisor compares the behaviors of some admissible
models and the actual process, and decides which model is best to describe the ongoing process.
The bank of models is given by
[
 : =  = (8.129)
∈
{ẋ = A (x  u y) y = C (x  u y) :  ∈  }

There will be at least one controller designed for each model. The set of controllers is denoted
as
[
 : =  = (8.130)
∈
{ẋ = F (x  y) u = G (x  y) :  ∈ }

where  and  are the set of estimators and controllers, respectively;  and  denote the th
model and  th controller, respectively; x is the state vector of model set; y is the estimation

271
 y
yp ep
Model
set
w
Environment
Supervisor

Controller u Marine y
set Switch vessel

Figure 8.25: Hybrid DP controller.

vector; x is the state vector of the controller; y is the output of the process plant; and u is the
control force vector. When the switching is made, a switching signal,  ∈ Z+ , determines the
selected model and the controller in the loop at each instant of time.

Hysteresis switching logic

The principle of switching is intuitive in the sense that the chosen controller in the closed loop
is the one corresponding to the model behaving closest to the process plant. For this purpose,
an estimation error is defined as

e = y − y  ∈  (8.131)

Although switching based on the error vector is intuitive, this could cause the chattering
(switching back and forth) of the controllers, hence destabilizing the whole system. One of the
reasons is that the estimation error in (8.131) may contain noise. The effect of the noise can be
reduced by introducing a monitoring signal (Hespanha et al. [116]),  ∈ R, according to

̇ = − + (||e ||) (8.132)

where  ∈ R denotes a constant non-negative forgetting factor;  is the class  function;

 (0)  0; and || · || denotes any norm.
According to Hespanha et al. [116] in order to avoid chattering even more effectively, the
switching is done on a slow-down basis of observing the growth of the estimation errors. This
means that the switching is not toward  = min( ), but toward  = (1 + ) min( ), where
 ∈ R is a positive hysteresis constant. Such switching logic is called hysteresis switching logic.
In practice, the hysteresis parameter  is tuned such that it is positive. However, if  is too
large, the switching will be frozen at previous operating regimes; and if  is too small, chattering

272
may occasionally occur. The forgetting factor  is chosen such that the monitoring signal,  ,
is neither too sluggish nor aggressive.

Properties
In the formal stability analysis of the whole system, it is convenient to have the following
definition adopted from Hespanha [114]. The switched system includes the process, the controller
set, and the model set

ẋ = A (x w) (8.133)

e = C (z w)  ∈  (8.134)

where x denotes the state vector of the process, multi-controller, and multi-model; the nonlinear
function A is found by collecting the state space equations of the process, the multi-controller,
and the multi-model; and w is the environmental disturbance vector. The input to the switched
system is the disturbances caused by the wind, wave and current loads, and the output is the
model error vector, e .
According to Hespanha et al. [116], the switched system must satisfy the matching and
detectability properties.
Matching property: the multi-model should be designed such that each particular y provides
a “good” approximation of the output y. This means e is small whenever the process is inside
the corresponding  .
Detectability property: For every fixed model, the switched system must be detectable with
respect to the estimation error, e , when the switching signal is frozen at . The detectability of
the system guarantees that if the output of the system is small, then the state must be eventually
small, no matter its initial state.

8.4.5 Example: Switching from DP to PM

This example of supervisory-switched control (SSC) is from Nguyen et al.[203]. In the example
we consider moderate sea and low speed such that VOC = Mode= DP and PM, Speed,
Environment = Moderate, (see Fig. 8.24).
We will here show how station keeping operations of the shuttle tanker from DP to thruster
assisted position mooring (PM), that is switching between mode 3, 4 and 5 in Figure 8.26. The
shuttle tanker transits from port to/from offshore field using the autopilot controller (controller
1). The DP controller is controller 3. The smooth transformation from autopilot to/from DP is
the controller 2. The single point mooring (SPM) or submerged turret loading (STL) mode to
connect to the loading buoy or tower is the controller 5. The smooth transition between DP and
SPM/STL is the controller 4. In the example shown here we will look into the station keeping
operation only, that is modes 3-5. By collecting the states of the WF model in (8.119) and the
LF model in (8.122), the state vector of the observer is reformulated:

p>
x = [b b >  bˆ>  ν
w η b > ]>  x ∈ R15  (8.135)

then the model can be written compactly as,

ẋ = T>
 ()A T ()x + B τ c + K (y − y )
y = C x  (8.136)

273
 y p1 Setpoint
e p1 Operator
Model 1: Transit

Model 2: Transformation transit 
‐ station keeping
y
Station keeping
Switching
u Model 3: DP
logic 
Model 4: Transformation  and
DP ‐ PM signal 
y p5 e p5 
Model 5: PM

Model set Supervisor

Environment
GNC 1 Observer 1               Controller 1

GNC 2 Observer 2              Controller 2

y u y
DP Vessel

u
GNC 5  Observer 5             Controller 5

Controller set Switch

Figure 8.26: Supervisory-switched control.

274
where K = [K> > > > >
1  K2  K3  K4 ] ; K1 ∈ R
6×3 is the observer matrix for WF motion estimation;

T () = diag(R ()  R () I3×3 ) ∈ R15×15 is the transformation matrix, B ∈ R15×3 
> >

C ∈ R3×15 and the system matrix A (0 ) ∈ R15×15

∙ ¸
Apw (0 ) 06×9
A (0 ) =  (8.137)
09×6 ALF

are derived from (8.119) and the LF model in (8.122).

DP Observer (Mode 3): In DP the low frequency system matrix ALF ∈ R9×6 becomes
⎡ ⎤
03×3 03×3 I3×3
ALF = ⎣ 03×3 −T−1 b 03×3 ⎦ (8.138)
03×3 M −1 −1
−M D

DP Controller (Mode 3): The nonlinear output-feedback PID controller is given by

τ cDP = τ   + τ   + τ   (8.139)

One should notice that roll-pitch damping (relevant for semisubmersibles) and wind feedfor-
ward control will be enabled/disabled by the operator. Hence, also these control terms needs to
be phased in and out properly.
PM Observer (Mode 5): In PM the effect of mooring must be included such that the system
matrix such that ⎡ ⎤
03×3 03×3 I3×3
ALF = ⎣ 03×3 −T−1b 03×3 ⎦  (8.140)
−1
−M Gmo M −1 −M−1 D
PM Controller (Mode 5): The control objectives are (1) to keep the vessel in a desired
heading angle, and (2) to add damping in surge and/or sway, when the resonant oscillatory
motions happen due to environmental excitations. In Nguyen and Sørensen [206] the various
PM control modes are given. Here, for simplicity we assume full station keeping control using
PID.
τ cPM = τ   + τ  (8.141)
Transition from DP to PM (Mode 4): The switching from DP to PM can be done by proper
weighting function  as follows

τ cPM2DP =  ( − 0 )τ cDP + (8.142)

(1 −  ( − 0 ))τ cPM 

where  is the time variable and 0 is the instant time when the switching begins.
Design of Supervisory Switching Control: The set of models and the set of controllers for
the modes correspond to  in (8.129) and in (8.130), respectively. As shown in Fig. (8.26),
the error vector, e , can be formed from the estimation vector, y , and process output, y.
From this error vector, the scale-independent hysteresis switching logic will decide the existing
VOC so as to switch to the appropriate model and thereby the appropriate controller. Details
of design and stability analysis for this supervisory switching control can be found in Nguyen et
al. [202].
Experimental Results: The experiments for the switching from DP to single point mooring
(here denoted as SPM) and vice versa are presented. The experiments were carried out using

275
 4 markers

Onboard
PC

WLAN

Accelerometer

Figure 8.27: Cybership III.

the model vessel, Cybership III (Fig. 8.27), which is a 1:90 scaled model of a shuttle tanker
having a mass of m = 75 kg, length of L = 2.27 m and breadth of B = 0.4 m (in full scale:
m = 54675 ton, L = 204.3 m, B = 36 m). The vessel is equipped with two main azimuthing
podded propellers, one tunnel thruster and one front azimuth thruster. The internal hardware
architecture is controlled by an onboard computer which can communicate with onshore PC
through a WLAN. An onshore 4-camera measurement system provides Earth-fixed position and
heading. The experiment was performed in the Marine Cybernetics Laboratory (MCLab) at
NTNU. A wave maker system was used to simulate the different sea conditions.
The PM system is modelled by a mooring line. One end of the mooring line is connected to
a fixed bridge; the other end to the bow of the Cybership III. This mooring line acts as a linear
spring since the restoring forces acting on the ship due to the linearity behavior of the whole SPM
system (hawser and buoy). There are six tests carried out for three mooring line configurations:
low, medium and high stiff mooring line, under moderate and high seas. The stiffness of the
low, medium and high stiff mooring line are  = 405 567, and 81, respectively. The
vessel is subjected to JONSWAP irregular head seas with the significant wave height (H ) and
the peak period of wave (T ). Since the switching is from DP to SPM, controllers 3, 4, and 5
will be used in the SSC.
Here, we only show the experimental result for Test 6 with high mooring stiffness and sea
state: significant wave height  = 396 and wave peak period  = 104. The other
experiments showed similar results. Fig. 8.28 shows the position and heading of the vessel and
the I controller load from control system during the switching procedure. Fig. 8.29 shows the
switching signal. The sequence of the switching from DP to SPM and vice versa is as follows.

276
 50 1000

Control force (kN)

North (m)
0 0

-50 -1000
0 1000 2000 3000 4000 5000 6000

1 2 3 4 5 6
20 100

Control force (kN)

10 50
0 0
East (m)

-10 -50
-20 -100
-30 -150
-40 -200
0 1000 2000 3000 4000 5000 6000
4
x 10
5 1

Control moment (kNm)

Heading (deg)

0 0

-5 -1
0 1000 2000 3000 4000 5000 6000
Time (s)

Figure 8.28: Test 6: performance of switching from DP to SPM mode and vice versa of the shuttle
tanker: measured position and heading (grey) and I controller forces and moment (solid).

Stage 1. Controller 3: DP. The vessel is kept in fixed position and heading by the DP
system.
Stage 2. Controller 4a: the smooth transformation in (8.142) is used.
Stage 3. Controller 5a: SPM with heading control. The vessel is connected to the SPM
system.
Stage 4. Controller 5b: SPM with heading control and surge damping.
Stage 5. Controller 4b: the smooth switching from the SPM with heading control and
surge damping to DP. The reference model provides the path from the existing position and
heading of the vessel in SPM to the desired position and heading of the vessel in DP. The I
controller is reset for the purpose of anti-windup. The smooth transformation (8.142) is used
same as in Stage 2 but the reset of I controller has to be taken into account.
Stage 6. Controller 3: DP. The vessel returns to DP mode, same as Stage 1.
The six stages are shown with the numbers in Fig. 8.28. The experiments with the switching
from PM to DP mode and vice versa in different sea conditions and different mooring system
showed good performance. When the vessel was kept in a fixed heading, there existed oscillations
in surge due to resonance (Stage 2). These oscillations were as expected reduced considerably
by adding surge damping (Stage 3).

277
 Figure 8.29: Test 6: Switching signal

8.5 Weather Optimal Positioning

Thor I. Fossen and Jann Peter Strand

Conventional DP systems for ships and free-floating rigs are usually designed for station-
keeping by specifying a desired constant position (   ) and a desired constant heading angle
  . In order to minimize the ship fuel consumption, the desired heading  should in many op-
erations be chosen such that the yaw moment is zero. For vessels with port/starboard symmetry
this means that the mean environmental force due to wind, waves and currents attack through
the center line of the vessel. Then the ship must be rotated until the yaw moment is zero.
Unfortunately, it is impossible to measure or compute the direction of the mean environ-
mental force with sufficient accuracy. Hence, the desired heading   is usually taken to be
the measurement of the mean wind direction which can be easily measured. In the North Sea,
however, this can result in large off-sets from the true mean direction of the total environmental
force. The main reason for this is the unmeasured current force component and waves that
do not coincide with the wind direction. Hence, the DP system can be operated under highly
non-optimal conditions if fuel saving is the issue. A small off-set in the optimal heading angle
will result in a large use of thrust.
One attractive method for computing the weather optimal heading   is to monitor the
resulting thruster forces in the - and -directions. Hence, the bow of the ship can be turned
in one direction until the thruster force in the -direction approaches zero. This method is
appealing but the main catch in doing this is that the total resulting thruster forces in the -
and -directions have to be computed since there are no sensors doing this job directly. The
sensors only measure the angular speed and pitch angle of the propellers. Hence, the thrust for
each propeller must be computed by using a model of the thruster characteristic resulting in a
pretty rough estimate of the total thruster force in each direction.
Another principle, proposed by Pinkster and Nienhuis [225], is to control the - and  position
using a PID feedback controller, in addition to feedback from the yaw velocity, such that the
vessel tends towards the optimal heading. This principle, however, requires that the rotation
point of the vessel is located in a certain distance fore of the centre of gravity, or even fore of
the bow, and it also puts restrictions on the thruster configuration and the number of thrusters
installed.
This section describes the design of a new concept for weather optimal positioning control
(WOPC) of ships and free-floating rigs, see Fossen and Strand [84] and Strand [276]. The control

278
 Gravity
Field

mass

Force Field Due to

Wind, Waves and Current (unknown)

Figure 8.30: Weather optimal positioning principle: Equivalent to a pendulum in the gravity
field, where gravity is the unmeasured quantity.

objective is that the vessel heading should adjust automatically to the mean environmental
disturbances (wind, waves and currents) such that a minimum amount of energy is used in order
to save fuel and reduce NO /CO -emissions without using any environmental sensors. This
is particular useful for shuttle tankers and FPSOs which can be located at the same position
for long time. Also DP operated supply vessels which must keep their position for days in
loading/off-loading operations have a great WOPC fuel saving potential.
Nonlinear and adaptive backstepping designs are used to derive the WOPC system. The
concept of WOPC can also be implemented by using other control methods e.g. feedback
linearization.
The key feature of the WOPC is that no information about the environmental disturbances
is required. This is important since the mean environmental disturbances acting on a floating
vessel cannot be accurately measured or computed. We show that the ship can be exponentially
stabilized on a circle arc with constant radius by letting the bow of the ship point towards the
origin of the circle. In order to maintain a fixed position at the same time, a translatory circle
center control law is designed. Moreover, the circle center is translated on-line such that the
Cartesian position is constant while the bow of the ship is automatically turned up against the
mean environmental force (weather-vaning). This approach is motivated by a pendulum in the
gravity field where gravity is the unmeasured quantity. The circular motion of the controlled
ship, where the mean environmental force can be interpreted as an unknown force field, copies
the dynamics of a pendulum in the gravity field, see Figure 8.30.

279
8.5.1 Mathematical Modeling
Ship model
In this section we consider a Low-frequency low-speed ship model in 3 DOF

η̇ = R()ν (8.143)
Mν̇ + C(ν)ν + D(ν)ν = τ + w (8.144)

where the Earth-fixed position ( ) and heading  is represented by η = [  ] and the
vessel-fixed vessel velocities are represented by ν = [  ] . From the low-speed assumption,
which is applied in the Lyapunov stability analysis, M = M  0, Ṁ = 0 and D(ν)  0,
∀ν ∈ R3 . τ ∈ R3 is the control vector of forces in surge and sway and moment in yaw provided
by the propulsion system. Unmodelled external forces and moment due to wind, currents and
waves are lumped together into an vessel-fixed disturbance vector w ∈ R3 . The rotation matrix
R() in yaw is defined in Section 7.2

Polar coordinates
The Cartesian coordinates ( ) is related to the polar coordinates by

 = 0 +  cos  (8.145)
 = 0 +  sin  (8.146)

where (0  0 ) is the origin of a circle with radius  and polar angle 
p
 = ( − 0 )2 + ( − 0 ) (8.147)
 = atan2 (( − 0 ) ( − 0 ))  (8.148)

Time differentiation of (8.145) and (8.146), yields

̇ = ̇0 + ̇ cos  −  sin  ̇ (8.149)

̇ = ̇0 + ̇ sin  +  cos  ̇ (8.150)

Define the state vectors

p0 , [0  0 ]  x , [  ]  (8.151)
From (8.149) and (8.150) a new kinematic relationship can be written in terms of the vectors
p0 and x as
η̇ = R()H()ẋ + Lṗ0  (8.152)
where ⎡ ⎤ ⎡ ⎤
1 0 0 1 0
H() = ⎣ 0  0 ⎦  L = ⎣ 0 1 ⎦ (8.153)
0 0 1 0 0
From (8.152) the Cartesian kinematics (8.143) can be replaced by a differential equation for the
polar coordinates
ẋ = T(x)ν − T(x)R ()Lṗ0  (8.154)

280
where

T(x) , H−1 ()R ()R()

= H−1 ()R ( − ) (8.155)

Note that the conversion between Cartesian in polar coordinates is only a local diffeomorphism,
since the radius must be kept larger than a minimum value, i.e.   min  0 in order to avoid
the singular point  = 0

Ship model transformation

The ship model (8.144) can be represented by polar coordinates by using (8.154) and substituting

ν = T−1 (x)ẋ + R Lṗ0  (8.156)

−1 −1  
ν̇ = T (x)ẍ + Ṫ (x)ẋ+R Lp̈0 + Ṙ Lṗ0  (8.157)

such that

Mν̇ + C(ν)ν + D(ν)ν = τ + w

m 0
M (x)ẍ + C (ν x)ẋ + D (ν x)ẋ = T− q(ν x ṗ0  p̈0 )
+T− τ + T− w (8.158)

where

M (x) = T− (x)MT−1 (x) (8.159)

³ ´
−1
C (ν x) = T (x) C(ν) − MT (x)Ṫ(x) T−1 (x)
−
(8.160)
D (ν x) = T− (x)D(ν)T−1 (x) (8.161)

q(ν x ṗ0  p̈0 ) = MR ()Lp̈0 + MṘ ()Lṗ0
+[C(ν) + D(ν)]R ()Lṗ0  (8.162)

Here M (x), C (ν x) and D (ν x) can be shown to satisfy

M (x) = M (x)  0 , D (ν x)  0 ∀x

The ship dynamics does also satisfy the skew-symmetric property

³ ´
z Ṁ − 2C z = 0 ∀z x (8.163)

Disturbance modelling
The steady-state LF motion of the ship and also the ship’s equilibrium position depend on the
unknown environmental loads acting on the vessel. Let the environmental loads due to wind,
waves and currents be represented by:

• a slowly-varying mean force  which attacks the ship in a point (   ) in body-fixed
coordinates.

281
 X
w1
North Xw
Fe

 e

Yw
( x, y )
w2

cos  

Y

East
( x0 , y0 )
sin 

Figure 8.31: Environmental force  decomposed into the components 1 and 2 

• a slowly-varying mean direction   , relative to the Earth-fixed frame, see Figure 8.31.

The term slowly-varying simply states that the mean environmental forces are slow compared
to the vessel kinematics and dynamics. The slowly-varying terms include 2nd-order wave-induced
disturbances (wave drift), currents and mean wind forces. The WF motion is assumed to be
filtered out of the measurements by using a wave filter, see Section 8.2.
Since there are no sensors which can be used to measure (    ) and (   ) with sufficient
accuracy it is impossible to use feedforward from the environmental disturbances. This motivates
the following assumptions:

A1: The unknown mean environmental force  and it’s direction   are assumed to be constant
or at least slowly-varying.

A2: The unknown attack point (   ) is constant for each constant  

Discussion: These are good assumptions since the ship control system is only supposed to coun-
teract the slowly-varying motion components of the environmental disturbances.

From Figure 8.31 the body-fixed environmental load vector w ∈ R3 can be expressed as
⎡ ⎤ ⎡ ⎤
1 ()  cos(  − )
w = ⎣ 2 () ⎦ = ⎣  sin(  − ) ⎦ (8.164)
3 ()   sin(  − ) −   cos(  − )

282
Notice that the environmental loads vary with the heading angle  of the ship. Moreover,
q
 = 12 + 22  (8.165)
  =  + tan−1 (2 1 ) (8.166)
The environmental forces  and  with attack point (   ) are shown in Figure 8.31. It
should be noted that the attach point  =  () and  =  () will also change with the yaw
angle  This relationship will be complicated function of the hull and superstructure geometries.

8.5.2 Weather Optimal Control Objectives

The weather optimal control objectives make use of the following definitions (Fossen and Strand
[84]):
Definition 8.2 (Weather optimal heading) The weather optimal heading angle  opt is given
by the equilibrium state where the yaw moment 3 (opt ) = 0 at the same time as the bow of the
ship is turned up against weather (mean environmental disturbances), that is 2 ( opt ) = 0 This
implies that the moment arms  ( opt ) = constant and  (opt ) = 0 and
⎡ ⎤ ⎡ ⎤
1 ( opt ) −
w( opt ) = ⎣ 2 ( opt ) ⎦ = ⎣ 0 ⎦ 
3 ( opt ) 0
Hence, the mean environmental force attacks the ship in the bow (minimum drag coefficient for
water and wind loads)
Definition 8.3 (Weather optimal positioning) Weather optimal positioning (station keep-
ing) is defined as the equilibrium state where
1 ( opt ) = −  2 ( opt ) = 3 ( opt ) =  (  opt ) = 0 (8.167)
(weather optimal heading) and the position ( ) = (   ) is kept constant.
These definitions motivates the following two control objectives:
O1: Weather optimal heading control (WOHC): This is obtained by restricting the ship
to move on a circle with constant radius  =  and at the same time force the ship’s bow
to point towards the center of the circle until the weather optimal heading angle  =  opt
is reached, see Figure 8.32. An analogy to this is a pendulum in gravity field, see Figure
8.30. The position ( ) = (0 +  cos  0 +  sin ) will vary until the weather optimal
heading angle is reached. This is obtained by specifying the control objective in polar
coordinates according to
 = constant, (8.168)
̇  = 0 (8.169)
  =  +  (8.170)
Discussion: The requirement  =  implies that the ship moves on a circle
with constant radius. The second requirement ̇  = 0 implies that the tangential speed
̇ is kept small while the last requirement   =  +  ensures that the ship’s bow points
towards the center of the circle.

283
 Xw Resulting environmental
 force in the x- and y- directions
(Earth-fixed)
Fe

Yw
w3>0
w1
(Unstable equilibrium point)

w1=Fe Fe
w2=w3=0
X

w2

Y (Stable equilibrium point)

Fe w2 w1=-Fe
w1 w2=w3=0

w3<0

Figure 8.32: Principle for weather optimal heading control (WOHC).

O2: Weather Optimal Positioning Control (WOPC): In order to maintain a fixed Earth-
fixed position ( ) = (   ), the circle center p0 = [0  0 ] must be moved simultane-
ously as Control Objective O1 is satisfied. The is referred to as translatory circle center
control.

We will derive a nonlinear and adaptive backstepping controller which satisfies O1 and
O2.

8.5.3 Nonlinear and Adaptive Control Design

In this section a positioning controller is presented by using the polar coordinate representation
of the ship dynamics. Backstepping is used to derive the feedback controller (Krstić et al.
[154]). The control law will be derived in 3 successive steps:

1. Nonlinear backstepping (PD-control): the ship is forced to move on a circle arc with desired
radius   with minimum tangential velocity ̇ and desired heading   

2. Adaptive backstepping (PID-control): this is necessary to compensate for the unknown

environmental force  .

3. Translatory circle center control: the circle center (0  0 ) is translated such that the ship
maintains a constant position (   ) even though it is moving on a virtual circle arc.
Hence, the Captain of the ship will only notice that the ship is rotating a yaw angle 
about a constant position (   ) until the weather optimal heading  opt is reached

284
Nonlinear backstepping (PD-control)
A general positioning controller is derived by using vectorial backstepping. The tracking objective
is specified in polar coordinates by the smooth reference trajectory x = [       ] ∈  3 where

x  ẋ  ẍ ∈ L∞ 

Since the transformed system (8.158) is of order 2, backstepping is performed in two vectorial
steps resulting in a nonlinear PD-control law. First, we define a virtual reference trajectory as

ẋ , ẋ − Λz1  (8.171)

where z1 = x − x is the Earth-fixed tracking error and Λ  0 is a diagonal design matrix.

Furthermore, let z2 denote a measure of tracking defined according to

z2 , ẋ − ẋ = ż1 + Λz1  (8.172)

From (8.172), the following expressions are obtained

ẋ = z2 +ẋ  (8.173)
ẍ = ż2 +ẍ  (8.174)

This implies that the vessel model (8.158) can be expressed in terms of z2 , ẋ and ẍ as:

M ż2 + C z2 + D z2 = T− τ + T− q(·) − M ẍ − C ẋ − D ẋ + T− w (8.175)

Step 1: Let z1 be the first error variable, which from (8.172) has the dynamics

ż1 = −Λz1 +z2  (8.176)

A Lyapunov function candidate (LFC) for the first step is

1
1 = z1 K z1  (8.177)
2
̇1 = −z1 K Λz1 + z1 K z2  (8.178)
where K = K  0 is a constant design matrix.

Step 2: In the second step we choose a LFC motivated by the ”pseudo” kinetic energy, that is
1
2 = 1 + z2 M z2  M = M  0 (8.179)
2
Time differentiation of 2 along the trajectories of z1 and z2 yields
1
̇2 = ̇1 + z2 M ż2 + z2 Ṁ z2  (8.180)
2
which by substitution of (8.178) and (8.175) yields
1 ³ ´
̇2 = −z1 K Λz1 + z2 Ṁ − 2C z2 − z2 D z2 +z2 T− w
¡ 2 ¢
+z2 K z1 +T− τ + T− q(·) − M ẍ − C ẋ − D ẋ  (8.181)

285
By using the property (8.163) and choosing the nonlinear PD-control law as
T− τ = M ẍ + C ẋ + D ẋ − K z1 − K z2 −T− q(·) (8.182)
where K  0 is a strictly positive design matrix, we finally get
̇2 = −z1 K Λz1 − z2 (K + D )z2 +z T− w (8.183)
Notice that the dissipative term z2 D z2  0 ∀z2 6= 0 is exploited in the design as it appears in
the expression for ̇2  With the control law (8.182) the closed-loop dynamics becomes
M ż2 + (C + D + K )z2 + K z1 = T− w (8.184)
The error dynamics of the resulting system becomes nonautonomous since
∙ ¸∙ ¸ ∙ ¸∙ ¸
K 03×3 ż1 K Λ 03×3 z1
= −
03×3 M ż2 03×3 C + D + K z2
∙ ¸∙ ¸ ∙ ¸
03×3 K z1 03×1
+ + w, (8.185)
−K 03×3 z2 T−
m
M(x)ż = −K(x ν)z+Sz+B̄(x)w (8.186)
where the different matrices are defined as
∙ ¸
 K 03×3
M(x) = M (x) =  (8.187)
03×3 M (x)
∙ ¸
K Λ 03×3
K(x ν) =  0 (8.188)
03×3 C (x ν) + D (x ν) + K
∙ ¸
 03×3 K
S = −S =  (8.189)
−K 03×3
∙ ¸
03×1
B̄(x) =  (8.190)
T− (x)
In the absence of disturbances, w ≡ 0, the origin z = 0 is uniformly locally exponentially
stable (ULES) according to Lyapunov. Global results cannot be achieved due to the local
diffeomorphism between the Cartesian and polar coordinates, that is the transformation matrix
T(x) is singular for  = 0.
With disturbances w 6= 0, the closed-loop system is input-to-state-stable (ISS). In the next
section, we will show how adaptive backstepping (backstepping with integral action) can be used
to obtain ULES for the case of a non-zero disturbance vector w 6= 0

Adaptive backstepping (PID-control)

If the disturbance vector w have a non-zero mean, this will result in a steady-state offset when
using the nonlinear PD-controller. Since the ship is restricted to move on a circle arc where w
can be viewed as a force field there will be a stable and unstable equilibrium point on the circle
arc (similar to a pendulum in the gravity field). The stable equilibrium point is, see Figure 8.32
⎡ ⎤
−1
w = φ = ⎣ 0 ⎦   (8.191)
0

286
Since, the disturbance  is assumed to be slowly-varying, we can apply adaptive backstepping
to obtain integral effect in the system. Thus, in the analysis it will be assumed that
̇ = 0 (8.192)
Let the estimate of  be denoted as ̂ , and ̃ = ̂ −   An additional step in the derivation
of the backstepping control must be performed in order to obtain an adaptive update law for
̂  Moreover:

Step 3: The adaptive update law is found by adding the squared parameter estimation error
to 2  Moreover,
1 2
3 = 2 + ̃    0 (8.193)
2 
where
1
̇3 = ̇2 + ̃˙ ̃  (8.194)

The nonlinear control law (8.182) is modified to
τ = T (M ẍ + C ẋ + D ẋ − K z1 − K z2 )−q(·) − φ̂  (8.195)
where the last term φ̂ provides integral action. Hence, the z2 -dynamics becomes
M ż2 + (C + D + K )z2 + K z1 = −T− φ̃  (8.196)
This implies that
1
̇3 = −z1 K Λz1 − z2 (K + D )z2 − z2 T− φ̃ + ̃˙ ̃ 

1
= −z1 K Λz1 − z2 (K + D )z2 +̃ (−φ T z2 + ̃˙ )
   −1
(8.197)

˙
The adaptive law ̃˙ = ̂ is chosen as
˙
̂ = φ T−1 z2    0 (8.198)
such that
̇3 = −z1 K Λz1 − z2 (K + D )z2 ≤ 0 (8.199)
The non-autonomous error dynamics for the adaptive backstepping controller can be written
M(x)ż = [−K(x ν) + S]z+B(x)̃  (8.200)
̃˙ = −B (x)z (8.201)
where ∙ ¸
03×1
B(x) =  (8.202)
−T− (x)φ
In order to satisfy Control Objective O1 we must choose the controller gains according to
⎡ ⎤ ⎡ ⎤ ⎡ ⎤
1 0 0 1 0 0 1 0 0
K = ⎣ 0 0 0 ⎦  K = ⎣ 0 2 0 ⎦  Λ = ⎣ 0 0 0 ⎦  (8.203)
0 0 3 0 0 3 0 0 3
Notice that 2 = 2 = 0 This implies that the ship is free to move on the circle arc with
tangential velocity ̇ The gain 2  0 is used to increase the tangential damping (D-control)
while the radius  and heading  are stabilized by using PID-control.

287
Semi-definite matrices
Since the controller gains 2 and 2 are chosen to be zero, the matrices

K ≥ 0 Λ ≥ 0 (8.204)

are only positive semi-definite resulting in a positive semi-definite 3 . Uniform local asymptotic
stability (ULAS) of the equilibrium (z̃ )= (00) can, however, be proven since the system is
ISS. We therefore consider the error dynamics of the reduced order system (z1  z2 ) given by
∙ ¸
1 0 0
z1 = Ez1  E=  (8.205)
0 0 1

This implies that

ż1 = −EΛz1 + Ez2 

= −(EΛE )z1 + Ez2  (8.206)

Notice that the last step is possible since the diagonal matrices Λ = diag{1  0 3 } satisfies

ΛE z1 = Λz1  (8.207)

Hence, the error dynamics (8.200)-(8.201) can be transformed to

M (x)ż = [−K (x ν) + S ]z +B (x)̃  (8.208)

̃˙ = −B (x)z  (8.209)

where z = [z1  z2 ] and

∙ ¸
(EK E ) 02×3
M (x) = M (x) = (8.210)
03×2 M (x)
∙ ¸
(EK E )(EΛE ) 02×3
K (x ν) = 0, (8.211)
03×2 C (x ν) + D (x ν) + K
∙ ¸
 02×2 EK
S = −S =  (8.212)
−K E 03×3
∙ ¸
02×1
B (x) =  (8.213)
T− (x)φ

We have here used the fact that K E z1 = K z1 for K = diag{1  0 3 }

Nonautonomous Lyapunov analysis

Recall that the Lyapunov function 3 is only semi-definite since K is positive semi-definite.
Since the system is ISS asymptotic tracking is guaranteed by
∙ ¸
1  1
3 = z1 (EK E )z1 + z2 M z2 + ̃2  0 (8.214)
2 
̇3 = −z1 (EK E )(EΛE )z1 − z2 (K + D )z2 ≤ 0 (8.215)

288
where EK E  0 and EΛE  0 Hence, z1  z2  ̃ ∈ L∞  ULES of the equilibrium point
(z1  z2  ̃ )=(0 00) follows by using the stability theorem of Fossen et al. [85] (see also Loria
et al. [175]) for nonlinear nonautonomous systems where 3  0 (positive definite) and ̇3 ≤ 0
(negative semi-definite). The reason that ̇3 is only negative semi-definite is that a negative
term proportional to −̃2 is missing in the expression for ̇3 

We are now ready to state the main theorem of this section.

Theorem 8.2 (Main result: ULES weather optimal position control) The equilibrium point
(z1  z2  ̃ )= (0 00) of the nonlinear system (8.208) and (8.209) with control law (8.195) and
parameter adaptation law (8.198) is ULES.
Proof. see Fossen and Strand [84].¤

Translatory circle center controller

The adaptive backstepping controller of the previous section satisfies control objective O1, that
is weather optimal heading control. Weather optimal position control, control objective O2, can
be satisfied by moving the circle center p0 = [0  0 ] such that the ship maintains a constant
position p = [ ] 
In order to meet the fixed position control objective, an update law for the circle center p0
must be derived. Recall that the Cartesian Earth-fixed position of the ship is given by
p = L η (8.216)
where L is defined in (8.153). Let p̃ , p − p denote the corresponding deviation from the
desired position vector p , [   ] . The desired position can either be constant (regulation)
or a smooth time-varying reference trajectory. The control law for translation of the circle center
is derived by considering the following LFC
1
 = p̃ p̃ (8.217)
2
where
̇ = p̃ (ṗ − ṗ ) = p̃ (L η̇ − ṗ ) (8.218)
By using (8.152), L L = I2×2 and ẋ = z2 +ẋ we get
̇ = p̃ [L (R()H()ẋ + Lṗ0 )−ṗ ]
= p̃ (ṗ0 − ṗ + L R()H()ẋ ) + p̃ L R()H()z2  (8.219)
Now, by choosing the circle center update law as
ṗ0 = ṗ − L R()H()ẋ −0 p̃ (8.220)
where   0 we get
̇ = − p̃ p̃ + p̃ L R()H()z2  (8.221)
Unfortunately a cross term in p̃ and z2 will appear in the expression for ̇3 if (8.220) is applied.
In order to guarantee that ̇3 ≤ 0 we must modify the weather optimal controller (8.195) and
add a negative term −p̃ p̃ to the expression
̇3 = −z1 (EK E )(EΛE )z1 − z2 (K + D )z2  (8.222)
to remove the cross term.

289
 GNC Laboratory
Basin wind generator
(ducted fan)

Y wave generator

Figure 8.33: Experimental set-up showing the directions of the wind and wave generators.

Weather optimal position control (WOPC)

The cross-terms involving p̃ and z2 in ̇ can be removed by modifying the nonlinear controller
(8.195) to

τ = T (M ẍ + C ẋ + D ẋ − K z1 − K z2 )−q(·) − φ̂

−T E ()R ()Lp̃ (8.223)

The last term in τ implies that

̇3 = −z1 (EK E )(EΛE )z1 − z2 (K + D )z2 − p̃ L R()H()z2  (8.224)

Consider the LFC

wopc = 3 +   (8.225)
̇wopc = −z1 (EK E )(EΛE )z1 − z2 (K + D )z2 −  p̃ p̃ (8.226)
and therefore the equilibrium point (z1  z2  ̃  p̃)=(0 00 0) is ULES.
The term p̈0 is needed in the expression for q(·). This term is computed from (8.220) as

p̈0 = p̈ −  (ṗ − ṗ ) − L R()H()ẍ

−L Ṙ()H()ẋ − L R()Ḣ()ẋ  (8.227)

8.5.4 Experimental Results

The experimental set-up is shown in Figure 8.33.
The experimental results are transformed to full scale according to Table 8.1 in Section 8.2.
In the scaling we used  = 70 meters and  = 4500 tones. A ducted fan is used to generate
a slowly-varying or constant wind disturbance.

290
Experiment 1: weather optimal heading control (WOHC)
In the first experiment the ship was allowed to move on the circle arc (circle center controller
was turned off). This is referred to as WOHC. The fixed origin and circle arc are shown in
Figure 8.34. Notice that the initial heading is approximately 30 degrees, see Figure 8.36, while
the position ( ) ≈ (13 −43), see Figure 8.35. These values are the one obtained when the fan
was initially directed in 210 degrees (opposite direction of the ship heading).
After 3000 seconds the fan was slowly rotated to 165 degrees corresponding to a weather
optimal heading of −15 degrees, see Figure 8.36. During this process, the ship starts to move on
the circle arc (with heading towards the circle center) until it is stabilized to its new heading, that
is −15 degrees. The new position on the circle arc is ( ) ≈ (3 20). This clearly demonstrates
that the ship heading converges to the optimal value (copies the dynamics of a pendulum in the
gravity field). This is done without using any external wind sensor.
In the next experiment, we will show how the circle center can be translated on-line in order
to obtain a constant position ( )

Experiment 2: Weather optimal position control (WOPC)

In the second experiment the ship should maintain its position (circle center controller is turned
on). This is referred to as WOPC. The performance during station-keeping (dynamic position-
ing) is shown in Figure 8.37 while the translation of the circle is shown in Figure 8.38. The
position controller works within an accuracy of ±1 meters which is the accuracy of the position
reference systems.
Again the weather optimal heading is changed from approximately 23 degrees to 2 degrees
but this time without changing the position ( ) of the ship. The position deviations and the
weather optimal heading are shown in Figure 8.39. These values are obtained by moving the
fan from an initial angle of 203 degrees to 182 degrees.
The last plots (Figure 8.40) shows the deviation for the radius regulator (upper plot) and the
how the circle center (0  0 ) is changed on-line by the circle center control law (lower plots) in
order to obtain a fixed position ( ) The experiment show that the ship will turn up against
an unknown disturbance (wind) at the same time as the ship maintains its position.

291
 Position in the horizontal plane

80

70

60

50
North [m]

40

30

20

10

-80 -60 -40 -20 0 20 40 60 80

East [m]

Figure 8.34: WOHC experiment showing the circular motion of the ship when the circle center
controller is turned off (WOHC).

Position in the horizontal plane

14

12

10

8
North [m]

-2
-50 -40 -30 -20 -10 0 10 20

East [m]

Figure 8.35: WOHC experiment showing the ( ) position of the model ship (m) when the fan
is rotated.

292
 Deviation in radius [m]

Heading [deg]
40

30

20

10

-10

-20
0 1000 2000 3000 4000 5000 6000 7000 8000 9000

time [sec]

Figure 8.36: WOHC experiment showing the performance of the radius regulator (upper plot)
and weather optimal heading (lower plot) versus time (s).

Position deviation in the horizontal plane

2

1.5

0.5
North [m]

-0.5

-1

-1.5

-2
-2 -1.5 -1 -0.5 0 0.5 1 1.5 2

East [m]

Figure 8.37: WOPC experiment showing station-keeping to (   ) = (0 0)

293
 500
Circle center

400 Zoom-In
2
300
North [m]

200 0

100
-2
0 -2 0 2

-100

-300 -200 -100 0 100 200 300 400 500

East [m]

Figure 8.38: WOPC experiment showing how the circle center is moving during station-keeping.

294
 Position deviation, North [m]
0.5
0.4
0.3
0.2
0.1
0
-0.1
-0.2
-0.3
-0.4
-0.5
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000

Position deviation, East [m]

2

1.5

0.5

-0.5

-1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000

Position deviation, East [m]

25

20

15

10

0
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
time [sec]

Figure 8.39: WOPC experiment showing the North and East position accuracies (upper plots)
and weather optimal heading (lower plot) versus time (seconds). The position accuracy is within
±1 m while the heading changes from 23 degrees to 2 degrees as the fan is rotated.

295
 Deviation radius [m]
1.5

0.5

-0.5

-1

-1.5
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000

Circle center, North [m]

30

25

20

15

10

-5
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000

Circle center, East [m]

20

-20

-40

-60

-80

-100

-120

-140
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
time [sec]

Figure 8.40: WOPC experiment showing the deviation for the radius regulator (upper plot)
and the translation of the circle center (0  0 ) (lower plots) versus time in seconds. The radius
deviation is within ±1 m during the rotation of the fan.

296
Chapter 9

Propulsion Control

The motivation to develop improved methods for propulsion and local thruster control was
initiated by former colleagues in ABB in the middle of 1990s. The idea was to better utilize
the nice properties of electrical drives and motors to the benefit of improved control and power
plant performance as well as reduced mechanical wear and tear of the propulsion components.
The first initial publication on this topic was Sørensen et al. [287]. Later on this work has
been addressed more extensively by several MSc theses and PhD scholarships at NTNU in
joint cooperation with industry partners. In particular, thruster control in extreme conditions
has been studied. Here, robust thruster control in extreme seas accounting for severe thrust
losses due to ventilation and in-and-out-of-water effects is developed, and an anti-spin thruster
controller, motivated by similar effects in wheel slip control on cars, is developed for thrusters
on marine vessels.
Several sections of this chapter are based on research results containing both theoretical and
experimental contributions published in papers by Smogeli and Ruth as a part of their PhD
study. In particular, material from the following publications are used: Aarseth [1], Overå [217],
Smogeli et al. [259], [260], [261], [262], [263], [266], Smogeli [264], Ruth et al. [232], [235], Ruth
and Sørensen [233], [234], Ruth [231], Sørensen and Smogeli [303] and Sørensen et al. [304].
Installed power capacity on marine vessels and offshore installations is normally limited.
Besides there is an increased focus on environmental aspects motivating technical solutions
that reduce the total energy consumption and emission of exhaust gases. The different sub-
systems/equipment installed on a vessel or offshore installation can be categorized into two parts,
producers and consumers of energy. For dynamically positioned vessels the thruster system
normally represents one of the main consumers of energy, and is regarded as a critical system
with respect to safety. On the contrary, the dynamic positioning system is only an auxiliary
system for the vessel to do a profitable operation of one kind or another, such as drilling, oil
production, loading, and so on. Hence, the thruster usage should not cause a load shedding of
those productive consumers, or in worst case cause a total power black-out because of unintended
power consumption. The strong requirements to vessel performance, operational availability
and overall safety have therefore resulted in increased focus on the total vessel concept and
the interactions between the different equipment and systems installed. Flexibility in operation
has enabled electrical power generation and distribution systems for propulsion, positioning,
oil production, drilling, and loading, where all equipment and control systems are integrated
into a common power plant network and automation network. In fully integrated systems,
functional in addition to physical integration of power and automation systems combined with

297
thorough marine process knowledge introduces new and far better opportunities to optimize
the overall vessel mission objective at lower life cycle costs. In order to accomplish this it is
essential to properly address the energy control of the consumers and producers of electrical
power onboard the vessel. If the various consumers (thrusters, pumps, compressors etc.) of
power act separately and uncoordinated from each other, the power generation system must be
dimensioned and operated with larger safety margins to account for the corresponding larger
mean power demands and unintentional power peaks.
This chapter will illustrate the importance of focusing on low level actuator control exem-
plified on thruster control in order to achieve a thorough successful control result, which does
not have negative impact on the other systems on the ship.
Other important results in the fields of thrust allocation, propulsion control and low-level
control of the thruster devices can be found in Blanke [28], Grimble and Johnson [100], Healey
et al. [113], Sørdalen [281], Sørensen et al. [287], Whitcomb and Yoerger [322] and [323],
Bachmayer et al. [12], Fossen and Blanke [86], Blanke et al. [30], Johansen et al. [134] and
Lindegaard [171].

9.1 Propellers and Thrusters

9.1.1 Shaft Propulsion
In a diesel-electric power and propulsion system, the shaft propellers are normally driven by
variable-speed electric motors. The horizontal motors may be directly connected to the shaft,
which results in a simple and mechanically robust solution, or via a gear coupling, which allows
for increased rotational speed of the motor and results in a more compact unit.
Shaft propellers are used mainly in shuttle tankers and other vessels in which the propulsion
power required is too high for conventional azimuth thrusters. By use of high-lift rudders, shaft
propellers may also be used to provide a certain degree of transverse thrust.

9.1.2 Thrusters
Tunnel thrusters produce fixed-direction transverse thrust and are often used in vessels in which
shaft propellers are amply dimensioned for the longitudinal thrust needed in DP operation. The
motors are normally vertically mounted with an L-shaped gear or horizontally mounted with
a Z-shaped gear where the geared transmission allows for higher rotational speed and smaller
motor constructions.
Tunnel thrusters may be of the variable-speed fixed-pitch (FPP) type or equipped with
constant-speed controllable-pitch propellers (CPP). The FPP solution has a simpler mechanical
construction since the pitch transmission can be omitted. Furthermore, low-thrust losses will be
reduced from levels typically 15% of rated power to essentially zero. This is the main reason for
why FPP has become so popular for DP operated vessels. The FPP thruster must be driven by
a variable-speed thruster drive.
Azimuth thrusters are rotatable devices for production of thrust in any direction. Although
it is optimized for positive-thrust direction, it should have a certain degree of negative thrust
capability as well, to maintain dynamic thrust capacity without performing continuous azimuth
rotation. The electric motor is normally vertically mounted and drives an L-shaped gear trans-
mission. With regard to azimuth thrusters, a variable-speed thruster motor drive and FPP
simplify the underwater mechanical construction and reduce low-thrust losses significantly.

298
 Slipring Unit (Power/Data Transmission)
Hydraulic Steering Unit

Ventilation Unit

Air Cooling

Installation Block

Bearing, Shaft Seals Bearing

FP Propeller Shaft Line Electric Motor

Figure 9.1: Podded propulsion.

Conventional azimuth thrusters are used with power ratings up to 6 − 7  .

9.1.3 Podded Propulsion

In podded propulsion units (Figure 9.1), a variable-speed electric motor is located in a compact
pod. The fixed-pitch propeller is mounted directly on the motor shaft. Since a mechanical
gear is avoided, the transmission efficiency is higher than in an azimuth thruster. Like the
conventional azimuth thruster, the podded propulsor is freely rotatable and may produce thrust
in both forward and aft directions if the thrust bearings are designed for it. The electrical power
is transferred to the motor via flexible cabling or slip rings for 360-degree operation. Since the
propeller pitch is fixed and there is no gear transmission, the mechanical construction is simple
and robust.
The pod can be designed for pushing or pulling operation. Especially the pulling type pod
gives the propeller a near optimum and uniform wake field, which increases the hydrodynamic
efficiency of the propeller and reduces the risk for cavitation, and hence give reduced noise and
vibrations.
Podded propulsion units have been in operation for more than eight years in cruise vessels,
ice breakers, service vessels and tankers. Recent new-built semi-submersible drilling units are
now also utilizing podded propulsion as station keeping/transit propulsion thrusters.
Podded propulsion units are today available in power ranges up to at least 25  . The
larger units provide access into the pod for visual inspection.

9.1.4 Mechanical Pod

The conventional azimuth thruster is normally used for station keeping and maneuvering. In
order to improve the hydrodynamics and steering capability that is required for propulsion, the
shape of the thruster is adapted.

299
 A mechanical pod is a mechanical drive designed for propulsion at higher ship velocities. It is
powered from an in-board motor, and the mechanical power is transferred to the propeller with
an L-shaped gear. The shape is optimized for low hydrodynamic resistance and high propulsion
efficiency.

9.1.5 Water Jets

Water jets are used mainly in high vessel speed operations. There has been presented concepts
where the water jet is driven by electrical motors, but the number of applications are limited.

9.2 Control Problem Formulation

As seen in the previous sections the positioning systems include different control functions for au-
tomatic positioning and guidance. The high-level positioning controller produces a commanded
thrust vector τ  ∈ R3 in surge, sway and yaw. The problem of finding the corresponding force
and direction of the thrusters that meets the high-level thrust commands is called thrust alloca-
tion. By using the singular value decomposition and geometrical filtering techniques the optimal
force and direction for each thrust device can be found avoiding singular thrust behavior with
reduced wear and tear and energy consumption for any thrust configuration and type of thruster,
see Sørdalen [281].
The propeller and thruster devices can be controllable pitch propeller (CPP) with fixed
speed, controllable speed with fixed pitch propeller (FPP), or controllable pitch and speed in
combination. The propeller pitch is measured at a radius 0.7 of the propeller blade, and is
defined as the travelled distance per revolution when advancing the propeller as a screw with
screw angle equal to the blade angle at 0.7. The pitch ratio  is defined as the ratio of
pitch at 0.7 is the propeller diameter. Traditionally, the propeller pitch ratio or the shaft
speed is used to indirectly control the propeller force towards the reference setpoint  specified
by the thrust allocation. In the thrust devices, a local pitch or speed controller is present. A
static mapping, see Figure 9.2, from the reference force to the actual control input (speed), d ,
which can be either pitch or propeller speed is used according to

d = (d ) (9.1)

Since normally no sensors are available for measuring the actual force developed by the
propeller, there is no guarantee for fulfilling the high-level thrust commands, and the mapping
from commanded thruster force to actual propeller force can be viewed as an open-loop system.
Thus, sophisticated control designs will be significantly degraded with respect to performance
and stability margins if the high-level control inputs are not produced by the local controllers.
Conventionally, the resulting pitch or speed set-point signals are determined from station-
ary propeller force-to-speed/pitch relations based on information about thruster characteristics
found from model tests and bollard pull tests provided by the thruster manufacturer. These
relations may later be modified during sea trials. However, as shown later they are strongly in-
fluenced by the local water flow around the propeller blades, hull design, operational philosophy,
vessel motion, waves and water current. Hence, the developed thrust  may differ substan-
tially from the commanded thrust setpoint  . In conventional positioning systems, variations
in these relations are not accounted for in the control system, resulting in reduced positioning
performance with respect to accuracy and response time. In addition, the variations may lead

300
 Disturbance

Desired Desired  Actual 

Thrust: Td speed: nd thrust: Ta
g(Td)

Mapping

Figure 9.2: Relationship between commanded thruster force, d , and actual developed propeller
force,  .

to deterioration of performance and stability in the electrical power plant network due to unin-
tentional power peaks or drops caused by load fluctuations on the propeller shafts, as shown in
Figure 9.3. The unpredictable load variations force the operator to have more available power
than necessary. This implies that the diesel generators will get more running hours at lower
loads in average, which in terms gives more wear, tear and increased maintenance costs. This
motivates finding improved methods for local thruster control.
In this section, the method of torque and power control is described using forces-to-torque/power
mappings, Sørensen et al. [287]. Hence, the force-to-pitch/speed () mappings, see (9.1), are
replaced by force-to-torque/power mappings. This approach will give more stable power loads,
and the method is less sensitive to disturbances. rpm denotes revolutions-per-minute.
Modelling of propellers is complicated by the fact that it is impossible to develop a finite-
dimensional plant model from first principles. A combination of simplified analytical and empir-
ical models is therefore the commonly used solution. This section treats modelling of propellers
from a control point of view. This means that the models are required to be accurate enough
to capture the main physical effects, and such facilitate control system design and testing.
The actual propeller thrust  and torque  are influenced by many parameters, the most
important being propeller geometry, submergence, and propeller loading — which depends on the
propeller pitch ratio and shaft speed.  and  can in general be formulated as functions of
fixed thruster parameters θ (i.e. propeller diameter, geometry, position, etc.), the shaft speed
, and variables x (i.e. pitch ratio, advance velocity, submergence, etc.):

 =  ( x  θ ) (9.2a)
 =  ( x  θ ) (9.2b)

In the following, only speed controlled FPP at low advance speed will be considered. The models
are, however, easily extendable to both controllable pitch propellers (CPP) and higher inflow
velocities, as experienced on vessels in transit, see Smogeli [264], and Ruth [231]. The pitch ratio

301
 RPM INPUT TORQUE INPUT

POWER POWER

RPM RPM

VARIABLE CONSTANT
DISTURBANCE PROPELLER PROPELLER
TORQUE TORQUE TORQUE

Figure 9.3: Torque and power control reduce the dynamic load variations and allows for lower
spinning reserves.

will in this case be a fixed parameter. The functions  (·) and  (·) may include loss effects
due to e.g. in-line and transverse velocity fluctuations, ventilation, in-and-out-of water effects,
and thruster-thruster interaction, as well as dynamic flow effects. In addition, the dynamics of
the motor and shaft must be considered. In the following sections basic propeller characteristics,
some quasi-static loss effects, and dynamic effects due to the water inflow, motor, and shaft are
considered.

9.3 Propeller Characteristics

The relationships between the propeller thrust  , torque  , shaft speed  (in revolutions-per-
second — rps), diameter , and density of water  are commonly given by Carlton [42]:
 =  (·) = sign() 4 2  (9.3)
5 2
 =  (·) = sign()    (9.4)
 and  are strictly positive thrust and torque coefficients, where the effects of thrust and
torque losses have been accounted for. The inclusion of loss effects in  and  is not
conventional, but is convenient for consistency in the description of the propeller characteristics.
The propeller power consumption  is written as:
 = 2 = sign()2 5 3  (9.5)
The expressions for  and  for deeply submerged propellers are found by so-called open
water tests, usually performed in a cavitation tunnel or a towing tank. This relationship is

302
 1st 2nd 3rd 4th
 ≥0 0 0 ≥0
 ≥0 ≥0 0 0

Table 9.1: The 4 quadrants of operation for a propeller, with advance velocity  and shaft
speed .

commonly referred to as an open-water propeller characteristics. The so-called advance ratio is

defined as

 =  (9.6)

where  is the inflow velocity to the propeller. The corresponding open-water efficiency  0 is
defined as the ratio of produced to consumed power for the propeller:
     
0 = = =  (9.7)
2 2  2

The expressions for  and  for deeply submerged thrusters are described by the following
parameters (Oosterveld and van Oossanen [216])
µ ¶
 
 = 1      (9.8)
 
µ ¶
  
 = 2         (9.9)
  

where  is the pitch ratio,   is the expanded-area ratio,  is the number of blades,
 is the Reynolds number,  is the maximum thickness of the blade section, and  is the chord
length of the blade section. The curves for  ,  , and   are shown in Figure 9.5 for different
pitch ratios for a Wageningen B-screw series based on Table 5 in Oosterveld and van Oossanen
[216], with  = 2 · 106 ,  = 4,  = 31 , and   = 052
Propellers are, with the exception of tunnel thrusters, usually asymmetric and optimized
for producing thrust in one direction. The propeller characteristics will therefore depend on
both the rotational direction of the propeller and the inflow direction. The 4 quadrants of
operation of a propeller are defined in Table 9.1. The typical open-water characteristics is a 1-
quadrant model, covering positive shaft speeds and positive advance velocities. The open-water
characteristics is extended into the 2nd quadrant, i.e. negative  . If reversing the propeller
operating direction, this can be modelled by a similar open-water characteristics for negative
shaft speeds and negative advance velocities. In a complete 4-quadrant model both positive and
negative shaft speeds and positive and negative advance velocities are covered. This is typically
done by modelling the propeller lift and drag coefficients as a function of propeller blade angle of
attack. In van Lammeren et al. [159] the 4-quadrant characteristics of some of the Wageningen
B-series propellers are given by a Fourier series representation. It is also possible to tabulate
the lift and drag coefficients as functions of propeller blade angle of attack (Bachmayer et al.
[12]). A simplification of the 1-quadrant propeller characteristics often seen in literature is to
approximate the  and  curves to be linear in  , see Blanke [28], Blanke et al. [30] and

303
Fossen and Blanke [86]:

 =  0 −  1   (9.10a)
 = 0 − 1   (9.10b)
⇓
 =   || −  ||   (9.10c)
 =   || −  ||   (9.10d)

where  0 , 0 ,  1 , and 1 are constant coefficients for  = 0 and:

 = 4  0   = 3  1 

 = 5 0   = 4 1 

This approximation is in reality only valid in the 1st quadrant. (9.10c) and (9.10d) clearly show
the dependence of the thrust and torque on  .  0 and 0 are the nominal thrust and
torque coefficients for a deeply submerged propeller with  = 0 and no thrust losses. The
corresponding nominal thrust  , torque  , and power  are expressed by:

 = sign() 0 4 2  (9.11a)

5 2
 = sign()0    (9.11b)
5 3
 = 2 = sign()20    (9.11c)

The nominal thrust, torque, and power of a reversed propeller are expressed as above , but with
 0 and 0 replaced with their reverse counterparts,  0 and 0 . Note that  0 and
0 are positive, since the thrust direction is accounted for in the expression.

Remark 9.1 Notice that with reference to the thrust allocation we here have assumed a speed
controlled propeller with  =  4  0   = | |  ,  =   In this case  =    =  ,
and subsequently  = 

Remark 9.2 As we will see later in this section the actual propeller thrust a and moment a
are the real produced thrust and moment, while d and d are the corresponding desired values
that are input to the local thruster controllers. Ideally, they should be equal. However, this is
hard or almost impossible to achieve.

In Figure 9.4 the relation between the output τ  ∈ R3 from the DP controller and the
propeller speed setpoint  is shown.

9.3.1 Example
Consider the following speed controlled thruster given by  = 089  = 4,  = 31 ,
 = 1025 3 and   = 052 The maximum and minimum  for the thruster are set
to be 300  and −100 , respectively. This is transferred to revolutions-per-second ()
according to
1
max = 300  = 300  = 5 
60
1
min = −100  = −100  = −167 
60

304
 Thrust 
Td Desired 
speed: nd
x Td1
Td2
DP y Thrust
Speed n [RPS]
Controller Allocation
 Tdn Thrust characteristics

Figure 9.4: Relation between the DP controller and the thrust setpoint for a speed controller
thruster.

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1 1.2
Advance ratio J

Figure 9.5: Open water  (solid), 10 ·  (dash) and   (dash-dot) as a function of advance
ratio  for  = 07, 089 and 11.

305
 1000
T [kN]
900

800

700

600

500

400

300

200

100

-1 1 2 3 4 5
-100 n [rps]
-200

-300

Figure 9.6: Thruster characteristics for  0 = 036 (solid)  1 = 03 (circle) and  2 =
04 (dotted)

First, assume bollard pull condition, that is  = 0. From Figure 9.5 we find that  0 = 036.
The thrust characteristics is then in [ ] given by (9.11a) according to

d_0 =  4  0 | |  
= 1025(31)4 036 | |  
= 34078 | |  

Next, assume that incoming current or vessel speed increase  such that  1 = 030

d_1 =  4  1 | |  
= 1025(31)4 030 | |  
= 28398 | |  

Finally we assume an increase in  due to negative advance speed such that  2 = 04 and

d_2 =  4  2 | | /

= 1025(31)4 04 | |  
= 37864 | |  

In Figure 9.6 the different thrust characteristics (d_0  d_1  d_2 ) are plotted for the defined
speed range [min  max ]. We observe from Figure 9.6 that a change in the propeller thrust
coefficient  will have large impact on the thrust characteristics. This may cause problems
in the dynamic positioning system and in the electrical propulsion system as they usually only

306
operate with one set of thrust characteristics, e.g. d_0 =  4  0 | |   In the real situation
the thruster characteristics will vary as shown in Figure 9.6, resulting in large moment and
power fluctuations as shown later in the chapter.

9.4 Propulsion Efficiency

The difference between the ship speed  and the inflow velocity to the propeller  is called
the wake. For a ship moving forward,  is less than  since the after-body flow changes its
magnitude between ship speed near the vessel and zero far from the vessel. The stationary
relationship for axial water inflow can be modelled as (Carlton [42])

 =  (1 − ( +  +  )) =  (1 − ) (9.12)

where  is the wake fraction number, typically in the range of 0    04.  is the wake
fraction caused by the wave motion of the water particles.  is the wake fraction caused by so-
called potential effects for a hull advancing forward in an ideal fluid, and  is the wake fraction
caused by viscous effects due to the effect of boundary layers. One should notice that (9.12)
only must serve as a rough estimate, since it does not account for the interactions between the
vessel motion and the actual wave particle flow. By applying a moment equation to a control
volume surrounding the inlet flow, the thrust can be related to the rate of change of moment of
the control volume. This will give a dynamic equation for the advance speed, see Healey et al.
[113] and Blanke et al. [30] and the references therein.
Consider a towed ship without any propeller installed, then a pressure field is set up given the
hull shape and the towing speed. If the same ship is self-propelled with a propeller at the same
speed, the pressure field experienced for the towed ship will be changed due to the propeller
action (thrust). At the aft part of the hull the propeller will increase the flow velocity, and
thereby according to Bernoulli reduce the local pressure. Hence, the suction of the propeller
generally reduces the pressure at the stern resulting in increased resistance or drag  according
to
 = a   (9.13)
where  is the thrust-deduction coefficient, typically in the range of 0    02 caused by
pressure reduction due to potential effects, viscous effects, waves, and appendices. In some
extreme cases  may become negative. In steady state the effective thrust is equal to the total
resistance  according to
 = a (1 −  ) (9.14)
The thrust coefficient behind the hull is normally assumed to be unchanged compared to
open water, while the torque coefficient will be affected by the change in inflow at the stern,
 . This is accounted for by the relative rotative efficiency
 
 2  
 =  = =  (9.15)
  
The overall propulsion efficiency can now be found as the ratio between the useful work done by
the product of drag and ship speed divided by the work required to overcome the shaft torque

 = =    (9.16)
2a

307
where  = 1−
1− is defined as the hull efficiency and is according to Newman [199] in the range


of 10—12.   is the mechanical efficiency typically in the range of 08—09.

9.5 Propeller and Thruster Losses

Modelling of thrust losses, which are constituted of complex physical phenomena, is most com-
monly solved by empirical methods in conjunction with analytical models. The following loss
effects may be considered:

• In-line velocity fluctuations (only for open and ducted propellers).

• Water inflow perpendicular to the propeller axis caused by current, vessel speed or jets
from other thrusters will introduce a force in the direction of the inflow due to deflection
of the propeller race. This is often referred to as cross-coupling drag.

• For heavily loaded propellers ventilation (air suction) caused by decreasing pressure on
the propeller blades may occur, especially when the submergence of the propeller becomes
small due to vessel’s wave frequency motion.

• For extreme conditions with large vessel motions the in-and-out-of-water effects will result
in sudden drop of thrust and torque following a hysteresis pattern.

• Both thrust reduction and change of thrust direction may occur due to thruster-hull inter-
action caused by frictional losses and pressure effects when the thruster race sweeps along
the hull. The last is referred to as the Coanda effect.

• Thruster-thruster interaction caused by influence from the propeller race from one thruster
on neighboring thrusters may lead to significant thrust reduction, if not appropriate pre-
cautions are taken in the thrust allocation algorithm.

The sensitivity to the different types of losses depends on the type of propeller and thruster
used, application of skegs and fins, hull design and operational philosophy. Main propellers
are subject to large thrust losses due to air ventilation and in-and-out-of-water effects. Rotat-
able azimuth thrusters are subject to losses caused by hull friction and interaction with other
thrusters. Tunnel thrusters are subject to losses caused by non-axial inflow due to current and
vessel speed and ventilation phenomenon in heavy seas.
The effects of the above mentioned losses are modelled by the thrust and the torque reduction
functions  and  based on a combination of analytical and empirical data. Hence, the actual
thrust  and torque  experienced by the propeller can be expressed as

 =  ( x  θ )(1 −  ) ,  ( x  θ ) (9.17)

 =  ( x  θ ) ,  ( x  θ ) (9.18)

where x represents dynamic states, such as vessel motion, propeller submergence and environ-
mental conditions. Propeller dependent parameters are represented by the vector θ . In the
following, only speed controlled fixed pitch propellers will be considered. The pitch ratio will in
this case be a fixed parameter. Research results on thrust losses may be found in for example
Shiba [253], Minsaas et al. [187], [188] and [189], and Smogeli et al. [259], [262] and [263]. Un-
fortunately, no general mathematical model is outlined for describing the thruster performance

308
in time-varying conditions, including transient effects. Especially for a dynamically positioned
vessel, with advance numbers close to zero, few results are available.
Thrust losses as shown in (9.17) and (9.18) may be expressed by the thrust and torque
reduction coefficients   and   , which express the ratio of actual to nominal thrust and torque
(Minsaas et al. [189])

 
=  ( x  θ )(1 −  ) =   =  (9.19)
  0
 
=  ( x  θ ) =   =  (9.20)
 0

In Figure 9.21 thrust reduction functions from Minsaas et al. [188] accounting for in-and-
out-of-water effects are illustrated. To investigate the effect of ventilation and loss of effective
disc area in a DP situation, two sets of model tests as presented in Smogeli et al. [259] will be
shown: 1) Stationary tests of a ducted propeller in a cavitation tunnel. 2) Dynamic tests of the
same ducted propeller subject to waves and varying submergence in open water.

9.5.1 In-line Velocity Fluctuations

Waves, currents and vessel motion induce a varying velocity field around the propeller. This
may be composed in an in-line component due to waves, current and vessel surge, sway and
yaw motion, and a transverse component due to waves and vessel heave, roll and pitch motion.
The in-line, or axial, component gives rise to changes in the advance velocity  , and hence
advance number  as shown in (9.12). As the thruster operating point is moving on the thrust
and torque curves, fluctuations in thrust, torque, power, and shaft speed are induced. Let the
advance number including only the hull wake be denoted  , then the thrust and torque loss
factors   and   due to axial flow are written

 ( )
  =  (9.21)
 ( )
 ( )
  =  (9.22)
 ( )

Figures 9.7 and 9.6 show thrust curves for varying advance speeds  , and clearly illustrates
that the resulting thrust loss is significant. The thrust and torque loss factors for in-line velocity
fluctuations will take on values both larger than and smaller than unity, since the variations
in advance speed may lead to both higher and lower propeller load than the nominal value for
 = 0. Note that this loss effect is not applicable to a tunnel thruster, since the thrust then is
independent of the in-line velocity.

9.5.2 Transverse Velocity Fluctuations

The transverse velocity fluctuations have different effects on tunnel thrusters compared to con-
ventional open and ducted propellers.

309
 5
x 10

10

2
Ta [N]

−2

−4 Va = −4 m/s
Va = 0 m/s
−6
V = 4 m/s
a
−8

−3 −2 −1 0 1 2 3
n [rps]

Figure 9.7: Example thrust curves for varying advance speed  

Open and ducted propellers

The transverse velocity component will give rise to a bending of the propeller race, and thus a
force perpendicular to the propeller axis. This force is given by
p
 =       (9.23)
with  a cross-coupling drag factor and  the transverse velocity component.  consists of
components from the vessel motion, current and wave induced velocities.  is commonly set
to 125 for ducted propellers and 031 for open propellers. From momentum considerations, the
thrust loss factor   is given by
µ ¶2 Ã p !2
     
  = 1 − =1−
 
 2
= 1 −   2 2  (9.24)

Notice that this loss effect does not affect the thruster torque directly, so that the torque loss
function   will be unity. Figure 9.8 shows an example of   as function of transverse
velocity for varying thrusts.

Tunnel thrusters
A tunnel thruster will experience large thrust losses caused by relative motion between the vessel
and water. Using the transverse velocity component  , the relative velocity  is expressed
as

 =  (9.25)


310
 1

0.8
βT,tr

0.6

T /T = 0.2
a a,max
0.4 Ta/Ta,max = 0.5
T /T
a
= 0.8
a,max

0.2
0 0.5 1 1.5 2 2.5 3 3.5 4
Vtr

Figure 9.8: Example thrust loss factor due to cross-coupling drag   versus transverse velocity
 for varying nondimensional thrust  max .
1

0.8
βT,tr

0.6

0.4

0.2
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
V [−]
rel

Figure 9.9: Example tunnel thruster speed loss factor as function of relative velocity  .

where  is the jet velocity in the wake of the thruster

s
2
 =  (9.26)
 

 is the propeller disc area. Figure 9.9 shows a typical plot of the thrust loss factor   as
function of  . Again, this effect does not affect the thruster torque, so that   = 1.

9.5.3 Ventilation and In-and-out-of Water Effects

Ventilation, cavitation, loss of effective disc area, and the Wagner effect are all closely related
physical phenomena. A generic thrust loss model due to ventilation and loss of effective disc
area has been implemented based on the experimental results presented later in this section.
An extension capturing the hysteresis introduced by the Wagner effect is under development.

311
 1

0.8

0.6
βT,vent

0.4

0.2

0
0
0.2
0.4
3
0.6 2
n/n 1
max
0.8 0
−1 h/R
1 −2

Figure 9.10: Ventilation thrust loss model as function of relative shaft speed max and relative
submergence .

The loss factor   is given as a two-dimensional function of the relative immersion  and
relative shaft speed max .  is the propeller shaft submergence,  is the propeller radius and
max the maximum shaft speed of the propeller. The loss function captures the main character-
istics of the ventilation and the loss of effective disc area, see Figure 9.10. For large submergence
  = 1, which means zero loss and full thrust. For low propeller loading, decreasing sub-
mergence leads to gradually decreasing   — and hence thrust. This corresponds to a loss of
effective disc area. For high propeller loading, decreasing submergence leads to an abrupt loss
of thrust due to ventilation. Note that the loss model will be propeller specific, and the critical
values of  and max where ventilation starts will vary.
The torque loss factor for ventilation will usually be larger than the thrust loss factor, and
may be related to the thrust reduction factor by:

  = (  )  (9.27)

where  = 065 for a ducted propeller,  = 0843 for an open propeller, and  = 0575 for a
tunnel thruster (Minsaas et al. [189]).

Experimental results from the cavitation tunnel tests

Systematic tests with a ducted propeller were carried out at the cavitation tunnel at NTNU,
Aarseth [1], Overå [217] and Smogeli et al. [259]. The advance number was kept at approximately
0.2, and the thruster shaft speed and submergence varied while measuring propeller and duct
thrust and propeller torque. Figure 9.11 shows  as function of relative submergence  and
propeller shaft speed.  is the propeller shaft submergence, and  the propeller radius. The

312
 0.25

0.2

0.15
KT

0.1

0.05

0
2
1.5
1 16
14
12
h/R 0.5 10
8
0 6
4 n

Figure 9.11: Thrust coefficient  vs. relative submergence  and propeller shaft speed 
for  = 02.

torque coefficient has a similar form. Nondimensionalized total thrust is presented in Figure 9.12.
These results agree with previous research on the subject, but contain additional information
for the DP case; at low speeds, with  close to zero, the advance number is no longer a sufficient
parameter for describing the losses. For large submergence, the thruster exhibits the expected
behavior, yielding the normal thrust characteristics with  constant and thrust proportional
to 2 . For low shaft speeds, and hence low propeller load, the thrust is constant for  = 15 to
 = 1, and then decays almost linearly from  = 1 to  = 0. This can be seen in Figure
9.12, and corresponds to lost thrust due to loss of effective disc area, as given by Minsaas et
al. [189]. For higher propeller loads, the thrust (and load torque) drops rapidly with decreasing
submergence. It is evident that for a heavily loaded propeller, proximity to the surface may lead
to an abrupt loss of thrust. As can be seen in Figure 9.12, the results indicate that a reduction
of shaft speed in such a case will increase the thrust. This was the key observation from the
experiments, motivating the design of an anti-spin thruster controller, shown later in the section.

Open water experimental results from MCLab

The results from the cavitation tunnel were all obtained at steady state, and are such stationary
results. A thruster will in reality be operating in more dynamic conditions where transient
effects are important. To investigate dynamic effects, tests in the Marine Cybernetics Laboratory
(MCLab) at NTNU were carried out with the same propeller unit (Aarseth [1], Overå [217] and
Smogeli et al. [259]). The results here were less clear than the stationary results, but confirmed
that it is advantageous to reduce the shaft speed when the thruster is subject to ventilation.

313
 1

0.8

0.6
T/Tmax

0.4

0.2

0
2
1.5 20
1 15
h/R
0.5 10 n
0 5

Figure 9.12: Nondimensionalized thrust vs. relative submergence  and propeller shaft speed
 for  = 02.

Figure 9.13 shows a time series of the propeller thrust and shaft speed for  = 15 in moderate
waves. The thrust increases for increasing , but suddenly drops sharply when the propeller
load becomes too high. The time series from these tests showed that ventilation caused the
thrust to drop sharply by as much as 85 % in a time span of only 1.5 seconds. This illustrates
the fast dynamics that the thruster control system has to deal with.

9.6 Propeller Shaft Model

Let  denote the torque generated by the propeller motor. A torque balance for the propeller
shaft is written
 ̇ =  −  −  () (9.28)
where  is the moment of inertia for the shaft, propeller, and motor,  = 2 is the angular
shaft speed, and  () is the shaft friction. The friction may for most applications be viewed
as a sum of a static friction (or starting torque)  and a linear component:

 () = tanh() +   (9.29)

where   0 is an appropriately chosen small constant, and  is a linear friction coefficient.

If desired, more sophisticated friction models may be used, including e.g. nonlinear elements,
Stribeck friction, and hysteresis effects. This has not been considered in the current work; even
though the experimental facility exhibited some nonlinear friction characteristics, the model

314
 220 15
n [rev/s] Propeller thrust [N]
200
14

180
13
160
Propeller thrust [N]

140 12

n [rev/s]
120 11

100
10
80

9
60

40 8
0 50 100 150
Time [s]

Figure 9.13: Time series of propeller thrust  and shaft speed  in moderate waves, with
ventilation at high propeller loading (high ).

315
 qP , xP

Qa
fQ (.)
qP , xP

Qm 1 n Ta
2 pIs fT (.)

Kw

Figure 9.14: Propeller shaft model with linear friction.

in (9.29) proved sufficient for control and control design purposes. If we disregard the static
friction, the torque balance (Figure 9.14) for the propeller shaft is written

 ̇ =  −  −   (9.30)

The power delivered by the motor and the actual propeller shaft power accounting for the
effect of thrust losses are given by

 =  = 2  (9.31)

 =  = 2  (9.32)

By combining (9.11a), (9.11b) and (9.32) and let  =  and  =  , it is possible to express
the actual thrust produced as function of propeller power and torque according to

 =  ( x  θ )(1 −  ) 

=  (·)(1 −  ) 4  | |  

=  (·)(1 −  )  5  | |  
  
 (·)(1 −  ) 
= ·   (9.33)
 (·)  

316
 q

Let  = sgn( )  4 , then the actual consumed power may be rewritten as
 

 =  = 2 

= 2   (·) 

s
 
= 2  (·) sgn( ) 
  4 
 32
= sgn( )2  (·) 32 √  
( ) 2 
1 1
= sgn( )2  (·) √  32  (9.34)
  (  (·)(1 −  ))32 

The actual thrust is

13
 23   (1 −  )
 = sgn( ) 23
· 23  (9.35)
(2 ·  ·  )

9.6.1 Torque Loop in an Electrical Motor Drive

The torque control is inherent in the design of most applied control schemes for variable speed
drive systems. The torque is controlled by means of motor currents and motor fluxes with high
accuracy and bandwidth, see Figure 9.15. Theoretically the rise time of the torque in PWM
(Pulse Width Modulated) drives is limited by the motor’s inductance (in load commutated
inverters, LCI’s, also by the DC choke). However, in practice the controller limits the rate
of change of torque in order to prevent damages on the mechanics. The closed loop of motor
and torque controller may for practical reasons be assumed to be equivalent with the 1st order
model, where  is the time constant in the range of 20 − 200  () and  the
commanded torque
1
̇ = ( −  ) (9.36)

or written by Laplace transform as
1
 () =  () (9.37)
1 +  

9.6.2 Resulting Thruster Dynamics

Combining the hydrodynamic thruster model, the shaft model, the motor model and the local
thruster controller yields the total thruster dynamics for one thruster unit. This is shown in
Figure 9.16, where the position of the thruster dynamics in the total vessel dynamics also is
shown.

317
 Tc Torque
controller
Tcalc -
PWM ~ PWM
c Flux
~ Converter

- controller Isa
calc Motor
Model
Isb

Induction
Motor
nm
Flux weakening
Tacho Tm

Figure 9.15: Torque loop in electrical motor drive.

Waves
Wind
Current Motion Measurements
Environment Vessel Vessel
dynamics observer

Thrust Filtered &

Motion reconstructed
Waves forces
signals
Current
Desired
Thrust
thrust
setpoints
Thruster Thrust vector Vessel
dynamics allocation control

Thrust
force

Thruster unit
Inline velocity Load torque
Vessel fluctuations
motion Ta
Transverse Qa
velocity Loss
fluctuations effects Propeller n Shaft Qm Motor Qc Local Tr
hydro- dynamics dynamics thruster
Waves Ventilation Thrust
Current
dynamics control setpoint
Water exit Motor torque
Other losses Shaft speed

Figure 9.16: Total thruster dynamics and its position in the total vessel dynamics.

318
9.6.3 Thrust, Torque, Power, and Shaft Speed Relations
The shaft speed  can be expressed in terms of the thrust  or the torque  by inverting (9.3)
and (9.4):
s
| |
 = sign( )  (9.38)
4 
s
| |
 = sign( )  (9.39)
5 

From (9.5), the magnitude of the shaft speed is expressed in terms of  by:
13

|| =  (9.40)
(2 )13 53

The thrust  can be expressed in terms of the torque  by eliminating sign()2 4 between
(9.3) and (9.4):

 =   (9.41)
 
Similarly,  can be expressed in terms of  by eliminating sign()2 4 between (9.3) and
(9.5) and inserting (9.38):
13 23  23
| | =  
23 
(9.42)
(2)23 
Finally,  can be expressed in terms of  by inserting (9.39) in (9.5):
13
13 53 
| | = 23  (9.43)
(2)23

9.7 Low-Level Thruster Controllers

The purpose of the low-level thruster controller is to relate the desired thrust  , given by the
thrust allocation routine, to the commanded motor torque  . The thruster controllers may be
divided in two control regimes, depending on the operational conditions:

(a) Thruster control in normal conditions, when experiencing low to moderate thrust losses.

(b) Thruster control in extreme conditions, when experiencing large and abrupt thrust losses
due to ventilation and in-and-out-of-water effects.

For surface vessels with FPP, shaft speed control is the industry standard, whereas torque
and power control was introduced by Sørensen et al. [287], and a combined torque and power
controller was presented in Smogeli et al. [260]. For underwater vehicles, both torque and
various shaft speed control schemes have been proposed, see Withcomb and Yoerger [323], and
the references therein. For control of diesel engines, shaft speed control is the most commonly
used solution. However, both torque and power control have been proposed and implemented,
see Blanke and Busk Nielsen [29].

319
9.7.1 Control Objectives
Shaft speed feedback control is the conventional way to control a FPP. However, as will be shown
later, this may not be the best design choice. The ultimate goal of any thruster controller is
to make the actual thrust track the desired thrust. Since no feedback from the actual thrust is
available, different ways of trying to achieve this for a FPP are:

• Shaft speed feedback control.

• Torque feedforward control.

• Power feedback control.

• Combinations of the above.

In dynamic operating conditions, other goals may be more important than tracking the
desired thrust, e.g. reducing mechanical wear and tear, and limiting power oscillations and peak
values. Predictable power consumption is of major concern to the power management system
in order to avoid blackouts and improving the performance and the stability of the electrical
power plant network. Other performance criteria than thrust production should therefore also
be taken into account. In this work, the following performance criteria are considered:

1. Thrust production in presence of disturbances, i.e. ability to locally counteract thrust

losses.

2. Mechanical wear and tear caused by transients and oscillations in motor and propeller
torque.

3. Predictable power consumption.

9.8 Thruster Control in Normal Conditions

9.8.1 Controller Structure
The low-level thruster controller is proposed to be composed of five main building blocks: refer-
ence generator, core controller, inertia compensation, friction compensation, and torque limiting.
This is shown in Figure 9.17, where a block diagram of a thruster with low-level thruster con-
troller is presented. The commanded torque from the core controller is termed  , where  = 
for shaft speed control,  =  for torque control,  =  for power control, and  =  for com-
bined torque/power control. The commanded torque from the friction and inertia compensation
schemes are termed   and  , respectively, and the total commanded torque before torque
and power limiting 0 is given by:

0 =  +   +   (9.44)

0 is then limited by a torque and power limiting function to yield the commanded torque to
the motor  .

320
 Control scheme Propulsion unit

Friction Q ff
comp. Qmax Pmax Qa
nr
Td Qm Propeller Ta
Reference Tr Core Q ci Q c0 Torque Qc Motor Shaft n hydro-
generator controller limit dynamics dynamics
dynamics
.
nr Inertia Q if
comp. n

Figure 9.17: Block diagram of a thruster with reference generator, core controller, inertia and
friction compensation, and torque limiting.

9.8.2 Control Plant Model Parameters

Since the propeller diameter  and density of water  are known and constant, the only re-
maining parameters are the thrust and torque coefficients  and  . The thrust and torque
coefficients used in the controllers are termed   and  , and denoted control coefficients.

Choosing   and 

In thruster control for station-keeping operations, estimates of the nominal thrust and torque
coefficients  0 and 0 are usually chosen as control coefficients, because the actual advance
velocity is unknown to the controller. Doppler logs, GPS, or similar may be used to give
estimates of the advance velocity, but these measurements are usually not of sufficient accuracy
for inclusion in the low-level thruster controllers.
For thruster control in transit, only the main propellers of the vessel are used. If the propeller
characteristics are known, improved controller performance may be achieved by estimating the
propeller advance velocity  using the known vessel surge speed  and an estimated hull wake
factor, or a low-pass filtered measurement from a Doppler log or GPS.   and  could
then be estimated from the propeller characteristics. Similar reasoning may be applied to an
underwater vehicle, where the propeller normally is deeply submerged and not subject to wave
effects, ventilation and water exits. Such an output feedback shaft speed controller with advance
speed compensation for AUVs can be found in Fosen and Blanke [86]. In the following, station-
keeping and low-speed manoeuvring operations of surface vessels are of main concern, such that
the control coefficients are taken as  0 and 0 .

Reverse thrust
With the exception of tunnel thrusters, most propellers are asymmetric. To achieve good per-
formance for both positive and negative thrust references, it may therefore be necessary to use
two sets of control parameters. For low-speed operations, this means that  0 and 0 should
be used for positive  , and the reverse thrust coefficients  0 and 0 for negative  . The

321
following control coefficients are therefore proposed:

  =  0  + (1 −  ) 0  (9.45a)
 = 0  + (1 −  )0  (9.45b)

where  is a smooth switching function given by:

1 1 
 =  ( ) = + tanh( ) (9.46)
2 2 
where   0 is a constant. The shaft speed reference  is given from  by:
s
| |
 =  ( ) = sign( )  (9.47)
4  

which is the inverse function of the nominal shaft speed-to-force characteristics given in (9.11a)
with  0 replaced by   . The switch between the coefficients occurs smoothly in the interval
 ∈ [−   ], such that  ≈ 0 for   − and  ≈ 1 for    . The switch “width” 
may be chosen freely. Equation (9.45) does not affect the stability properties of the controllers
when compared to using constant control coefficients, since  0 , 0 ,  0 , and 0 are
strictly positive, and the values of   and  only affect the mappings from thrust reference
to shaft speed, torque, and power reference. Using  as the switching variable instead of e.g.
the measured shaft speed  means that no measurement noise is entered into the control law
through (9.45).

9.8.3 Reference Generator

An appropriate reference generator for this system is a second-order filter with rate limiting.
The reference generator is imposed on the desired shaft speed  , where  is given from  by
 in (9.47), and  is given from  by the inverse −1 ( ), which is equal to (9.11a) with  0
replaced by   . The second-order filter has damping ratio   and natural frequency  0 , and
the shaft speed rate limit is set to ̇ , such that the reference generator is given by:

 =  ( )
̇2 = min(max(̇  −̇ ) ̇ )
 20
 () = 2 () 
2 + 2  0  +  20
 = −1 ( ) (9.48)

Notice that the reference generator gives a continuous acceleration reference ̇ , which is available
directly from the implementation of the second-order filter.

9.8.4 Inertia Compensation

For a large thruster, the rotational inertia may be a dominating dynamic term. In order to
achieve the desired closed loop properties with respect to response and tracking, it may therefore
be beneficial to include an inertia compensation term. The inertial term in the rotational
dynamics (9.28) is  ̇. Since differentiation of the measured shaft speed is undesirable, and

322
the inertia compensation is wanted only when the thrust reference is changed, the following
feedforward compensation is proposed:

 ( ) =  2 ̇  (9.49)

where  is an estimate of the rotational inertia  , and ̇ is given by the reference generator in
(9.48).

9.8.5 Friction Compensation

In the torque and power controllers, the friction term  () will be compensated by a feed-
forward controller. If this is not the case, and the friction is significant, the controllers will
be inaccurate, since at steady state  ≈  no longer holds, see (9.28). If the shaft friction
becomes significant, a feedforward friction term   should therefore be included in the con-
trol law. Note that the inclusion of   in the shaft speed controller is unnecessary, since this
controller automatically compensates for the friction.
  should be designed to compensate for the friction term  (), i.e.   ≈  (),
without destabilizing the system. Therefore a feedforward compensation based on the reference
shaft speed  from (9.47) is chosen instead of a destabilizing feedback compensation based on
. Motivated by (9.29) the friction is assumed to consist of a static and a linear term, and the
friction compensation scheme is proposed as:

  ( ) =   0 ( ) +   1 ( ) (9.50)

  0 ( ) is proposed as:


  0 ( ) =  0 tanh( ) (9.51)

where  0 ,   0, and  are constants. The switch between − 0 and  0 occurs smoothly
in the interval  ∈ [−   ] such that | |   ⇒   0 ( ) = sign( ) 0 , and  = 0 ⇒
  0 ( ) = 0. Note that the switching function is of similar form as the one used in (9.46).
  1 is proposed as:
  1 ( ) = 2 1   (9.52)
where the linear friction coefficient  1 should be chosen such that  1 ≈  . In nominal
conditions,  =  ⇒   1 ( ) ≈  . For  6=  ,   1 ( ) 6=  , but this only affects the
performance about the equilibrium point.

9.8.6 Torque and Power Limiting

The rated (nominal) torque and power for continuous operation of the motor are denoted 
and  . The maximum torque and power for the motor are usually set to

max =   max =   (9.53)

where  typically is in the range of 11 − 12. The input to the motor should be limited by a
torque limiting function given by:

| | = min{|0 | max  max |2|}

323
where 0 is the commanded torque from the controller (9.44). The maximum power limit
yields hyperbolic limit curves for the torque as a function of speed. Since the maximum power
is not limited by the converter and motor ratings only, but also by the available power in the
generators, this limit should vary accordingly. By this method the power limitation will become
fast and accurate, allowing to utilize the power system power capability with a built-in blackout
prevention.

9.8.7 Shaft Speed Feedback Control

For conventional FPP a speed controller is used to achieve the commanded propeller force. Given
a specified force command  the corresponding reference (commanded) speed of the propeller
 is found by the shaft speed mapping in (9.47). The shaft speed PI controller utilizes shaft
speed feedback from the thruster, and sets the commanded motor torque  according to
Z 
 =   +  ( )  (9.54)
0

where the shaft speed error is  =  − , and  and  are the nonnegative PI gains. The
motor torque is controlled by the inner torque control loop.  is often given as  =   ,
where  is the integral time constant. The integral term in the PI controller should be limited
according to max to avoid integral windup.

9.8.8 Torque Feedforward Control

In the torque control strategy the outer speed control loop is removed, and the thruster is
controlled by its inner torque control loop with a commanded torque  as set-point. A
mapping from the reference thrust  to the reference torque  can be found from (9.11a)
and (9.11b) with  0 and 0 replaced by   and  . The commanded motor torque 
is set equal to the reference torque:

 =  =   (9.55)
 
If the response of the thruster is slow, e.g. due to high rotational inertia, the inertia compensation
scheme in (9.49) may be needed to speed up the response. This replaces the “inner torque
algorithm” (Figure 9.15).

9.8.9 Power Feedback Control

Power control is based on controlling the power consumption of the thruster motor. The inner
torque control loop is maintained, and the reference power  is found by inserting  from
(9.55) and  from (9.47) in (9.11c) such that
2
 =  2 = | |32 √ 32
 (9.56)
 

To determine the torque direction, the signed reference power  is defined as
2
 = sign( ) = sign( )| |32 √ 32
 (9.57)
 

324
The commanded motor torque  is calculated from  using feedback from the measured
shaft speed  6= 0 according to
  sign( )| |32
 = =√ 32
 (9.58)
2||   ||

Note that the power controller is singular for zero shaft speed. As in torque control the inertia
compensation scheme in (9.49) may be added if the response of the thruster is slow, replacing
the “inner power algorithm”.

9.8.10 Combined Torque and Power Control

A significant shortcoming of the power control scheme in (9.58) is the fact that it is singular for
zero shaft speed. This means that power control should not be used close to the singular point,
for example when commanding low thrust or changing the thrust direction. For low thrust
commands, torque control shows better performance in terms of constant thrust production,
since the mapping from thrust to torque is more directly related to the propeller loading than
the mapping from thrust to power. For high thrust commands, it is essential to avoid power
transients, as these lead to higher fuel consumption and possible danger of power blackout and
harmonic distortion of the power plant network. Power control is hence a natural choice for high
thrust commands. This motivates the construction of a combined torque/power control scheme,
utilizing the best properties of both controllers.
Combining the reference torque from the torque controller  given by (9.55) and the
reference torque from the power controller  given by (9.58), the commanded motor torque
 from the combined torque/power controller is defined as:
 =  () + (1 −  ())  (9.59)
 () is a weighting function:

 () = −||  for  ∈ Z (9.60)
where ,  and  are positive constants. It satisfies:
lim  () = 1 lim  () = 0
→0 →∞

and shows smooth behavior for all . Particularly,   = 0 for  = 0 and  → ±∞. The
shape of  () defines the dominant regimes of the two control schemes, and can be used to
tune the controller according to user specifications. The parameter  will act as a scaling factor
for . A small  will widen the weighting function, giving a wider transition between 0 and 1.
Increasing the parameter  sharpens the peak about  = 0, whereas  widens it and makes the
transition from 0 to 1 more steep. The shaft speed is physically limited to some max value max ,
such that  (max ) should be close to zero. It is easily shown that  is non-singular if   1.
Hence, the power control singularity for  = 0 is removed.

Remark 9.3 The torque and power controllers aim at controlling the motor torque or power,
respectively. There is no need for integral action in these controllers, since the motor torque is
controlled by its inner torque control loop. Consequently, the torque and power controllers allow
the shaft speed of the propeller to vary with the propeller loading, in order to keep the torque or
power constant.

325
 1
r = 2, p = 1, k = 1
0.9 r = 2, p = 0.5, k = 1
r = 8, p = 1, k = 1
r = 2, p = 1, k = 4
0.8

0.7

0.6
α(x)

0.5

0.4

0.3

0.2

0.1

0
−5 −4 −3 −2 −1 0 1 2 3 4 5
x

Figure 9.18: Weight function  () for varying ,  and 

Figure 9.18 shows  () for varying ,  and . The parameter  will act as a scaling
factor for the weighting variable . A small  will widen the weighting function, giving a wider
transition between 0 and 1. Increasing the parameter  sharpens the peak about  = 0, whereas
 widens it and makes the transition from 0 to 1 more steep.
The weight function must show smooth behavior for all . The derivative of  with respect
to  is
 () 
= − || −|| (− ||−1 )


=  2  ||2−1 −||
=  2 2  ||2−1  () (9.61)

which is smooth in . Particularly,   = 0 for  = 0 and  → ±∞.

9.9 Thruster Control in Extreme Conditions

9.9.1 Loss Estimation and Ventilation Detection
In the development of the anti-spin thruster controller, a scheme for thruster performance mon-
itoring and ventilation detection will be essential. This section presents a load torque observer
based on the thruster rotational dynamics in (9.28), as presented in Smogeli [264]. The load
torque estimate can be used to calculate the estimated torque loss factor ̂  . Finally, a ventila-
tion detection scheme based on ̂  and the measured motor torque  will be developed.

326
Load Torque Observer
From (9.28), the following control plant model of the thruster dynamics is proposed:
 ̇ =  −  −   0 ( ) −  1  +   
̇ =   (9.62)
where the measured  has been used instead of  . Since  ( x  θ ) in (9.4) is unknown
and may exhibit highly nonlinear behavior,  is here assumed to be a bias term driven by an
exogenous bounded disturbance  . This is necessary in order to capture the fast transients
during ventilation. The static friction compensation term   0 ( ) from (9.50) is used instead of
a static friction model based on  in order to avoid oscillations in this term about  ≈ 0. Errors
in the friction model are accounted for by   . With the measurement  =  +  contaminated
with a bounded disturbance , and ̂ = ̂, a propeller load torque observer copying the control
plant model (9.62) is:
̂˙ = 1 [ − ̂ −   0 ( ) −  1 ̂] +  ( − ̂)
˙
̂ =  ( − ̂) (9.63)
where  and  are the observer gains. In Smogeli and Sørensen [266] it is shown that with a
constant load torque  implied by  = 0, zero measurement disturbance  = 0, and perfect
friction knowledge such that   = 0, the equilibrium point of the observer estimation error is
globally exponentially stable (GES) if the observer gains are chosen as   − 1  and   0

Torque loss estimation

For DP operation the expected nominal propeller load torque ̂ may be calculated from (9.11b)
by feedback from the propeller shaft speed . Based on (9.20), the estimated torque loss factor
̂  is calculated from ̂ in (9.63) and ̂ as:

̂ ̂
̂  = =   6= 0 (9.64)
̂  5  ||

where the control coefficient  (usually equal to 0 ) is used to calculate ̂ .

Remark 9.4 There are no thrust losses for  = 0. Hence, the singularity for  = 0 is easily
avoided by redefining (9.64) as ̂  =  () + (1 −  ())̂ ̂ , where  () is a function of
the type (9.60).

Ventilation detection
In this work, the ventilation detection algorithm is implemented by defining limits for the onset
and termination of ventilation,   and    , and generating a detection signal by monitoring
the thrust loss factor ̂  . An additional criterion for detection is that the magnitude of the motor
torque  is non-increasing. A ventilation incident will then give the following evolution of the
detection signal , with time instants 1  2  3 :
1 : ̂     ⇒  = 0
2 : ̂  ≤   ∩ sign( )̇ ≤ 0 ⇒  = 1 (9.65)
3 : ̂  ≥    ⇒  = 0

327
To avoid switching and chattering of the detection signal due to measurement noise and tran-
sients, an algorithm implementing a detection delay is added, such that once ventilation has been
detected, it cannot be reset until after a given time interval  . The detection delay can be
seen as an implementation of the switching dwell-time proposed in Hespanha [114]. This ventila-
tion detection scheme has shown good performance in both simulations and experiments. Note
that   should be chosen significantly lower than 1 in order to allow for natural oscillations
about the nominal propeller loading due to time-varying inflow to the propeller.

9.9.2 Anti-spin Thruster Control

The torque and power controllers, although advantageous for normal operating conditions, will
show unacceptable behavior when subject to large thrust losses. The nature of this problem
is in many ways similar to that of a car wheel losing traction on a slippery surface during
acceleration or braking. The work on anti-spin thruster control has therefore been motivated
by similar control strategies in car anti-spin and ABS braking systems. With any of the core
controllers from Section 9.8 in use for normal operating conditions, the proposed anti-spin control
scheme is divided in two:

1. Primary anti-spin action: override the core controller and take control of the shaft speed.

2. Secondary anti-spin action: lower the setpoint of the controller to reduce the shaft speed.

Similarities to car anti-spin/ABS

Since both the structures of the physical systems and the control objectives are similar, the
development of the thruster anti-spin control system has been motivated by similar hybrid
control strategies in car anti-spin and ABS systems, see for example Haskara et al. [110] or
Johansen et al. [133]. In car anti-spin/ABS, a commonly used model is the "quarter car", see
Canudas de Wit and Tsiotras [41]

 ̇  = − +  

= ()

̇ =   (9.66)

where  is the wheel rotational inertia,  is the wheel angular velocity,  is the wheel radius,
 is the tire friction force,  is the breaking or accelerating input torque to the wheel from
brakes and motor,  is the vertical force between tire and road,  is the tire friction coefficient,
 is the wheel slip,  is the mass of a quarter car, and  is the car speed. The actuator dynamics
have been neglected. The wheel slip  is defined as Canudas de Wit and Tsiotras [41]
½
1 −  if      6= 0 (braking),
= (9.67)
1 −  if       6= 0 (accelerating).

When  = 1 there is no sliding (pure rolling), and when  = 0 there is full sliding. The shape
of the friction coefficient () depends on many parameters, which can be summed up in the
tire characteristics, the road conditions, and the vehicle operational conditions. The aim of the
anti-spin/ABS control systems is commonly to keep the wheel slip at a specified set-point or to
maintain a specified wheel acceleration Johansen et al. [133].

328
 When comparing the two models for propulsion and car anti-spin, many similarities may
be noticed. The rotational dynamics of the propeller and the wheel are identical, with inertial
terms  ̇ and  ̇  , motor input terms  and  , and load torque terms  =  ( ) and
 =  () respectively. In both cases, the aim of the anti-spin control is to optimize the
load torque through control of the motor input. For car anti-spin, a too high input torque will
lead to loss of friction through spin (acceleration) or wheel lock (braking), and hence reduced
acceleration or stopping force. For thruster anti-spin, a too high load on a propeller operating
close to the free surface will lead to loss of torque through ventilation, and hence loss of thrust.
In addition, the propeller load torque and tire friction force depend on many other parameters,
most of which are unknown to the controller.
However, some differences should also be noticed. The tire friction force can be directly
related to the vehicle dynamics, and the wheel slip can be estimated from the vehicle speed and
wheel angular velocity. This leaves only the friction coefficient () unknown. However, for most
conditions, the shape of the friction coefficient will be known, making it possible to develop a
general anti-spin scheme. The propeller load torque  is composed of several complex physical
phenomena, and must be considered as an exogenous disturbance from a control point of view.
In addition, no general mapping between the load torque and the propeller thrust  exists. And
even if this mapping could be established, the relationship between the propeller thrust and the
vessel dynamics is more complex than the car dynamics, especially for a vessel equipped with
multiple propellers. Finally, the operating conditions of the propeller will probably be changing
faster than for the wheel; the propeller could be constantly moving in and out of water, whereas
the wheel may experience less rapidly changing road conditions.

Primary anti-spin action

If a ventilation incident is detected by the detection algorithm in (9.65), the desired primary
anti-spin control action is to take control of the shaft speed until the ventilation has terminated.
This is done by modifying the core controller output  in (9.44) with a torque scaling factor
:
 =   (9.68)
where  is the anti-spin commanded torque, and  is proposed as:
½
1 for  = 0 (not ventilated),
= (9.69)
̂  for  = 1 (ventilated).

In Smogeli and Sørensen [266] it is shown that the anti-spin control law will give a bounded
shaft speed during ventilation.

Secondary anti-spin action: setpoint mapping

For optimization of thrust production and reduction of wear and tear due to dynamic propeller
loading during ventilation, it may desirable to reduce the propeller shaft speed. This can be
done by modifying the thrust reference during ventilation, since the primary anti-spin control
action assures that the shaft speed is kept close to its reference. The desired shaft speed during
ventilation,  , will be a thruster specific parameter, and must be chosen as a trade-off between
thrust production, wear and tear, and response time. The nominal shaft speed reference  from
(9.47) should not be changed to  instantaneously, as this will lead to undesired transients. It is

329
therefore proposed to add a low-pass filter with time constant   and a rate limiting algorithm to
the change from  to  at ventilation detection, and from  to  at ventilation termination.
The filtered and rate limited shaft speed reference is termed  , i.e. ̇fall  ̇  ̇rise , where
̇fall and ̇rise are the rate limits. The thrust reference corresponding to  is termed  ,
and is given by (9.11a) with  0 replaced by   :

 = sign( )  4 2  (9.70)

Since (9.70) is the inverse of (9.47),  ≡  when ventilation is not detected.

9.9.3 Implementation aspects

Perfect shaft speed control during ventilation would be possible if the ventilation incident could
be foreseen, and   ( ) was known. However, since the propeller torque is not available as
a measurement, and the ventilation incidents are random processes caused by wave elevation
and vessel motion, this is not possible. An implementable solution is to use the torque loss
estimation and ventilation detection schemes. The deviation of the shaft speed from the steady-
state solution will then depend on the deviation of the torque modification factor  from the
actual torque loss factor   ( ) during ventilation.
To achieve the best possible control over the shaft speed during ventilation,  should be
switched from 1 to ̂  immediately when ventilation is detected. However, in order to avoid
transients in the control input, the transition of  from 1 to ̂  at ventilation detection, and
from ̂  to 1 at ventilation termination, should be rate limited, i.e. ̇    ̇  ̇  . The
rate limits ̇  and ̇   would typically be of magnitude 1. Furthermore, to avoid using the
potentially noisy estimate ̂  directly in the control law, it may be beneficial to filter  by a
properly chosen low-pass filter with time constant   .
A block diagram of the resulting thruster controller with anti-spin is shown in Figure 9.19.
Note that the proposed anti-spin controller may be used with all the core controllers from Section
9.8.

Remark 9.5 The deviation of  from   ( ) during ventilation will be affected by the ven-
tilation detection scheme, the accuracy of the estimated torque loss factor ̂  , and the chosen
rate limits and low-pass filter for . The proposed rate limits and filter on  will lead to a larger
perturbation bound, and hence also a larger shaft speed bound. However, with reference to the
control objectives, this may be a necessary trade-off to avoid unnecessary wear and tear and
power transients.

Remark 9.6 A well-tuned shaft speed controller will produce a controller output similar to the
proposed anti-spin controller in (9.68) during ventilation, since the objective of the shaft speed
controller is to keep the shaft speed constant, regardless of the load disturbance. However, since
it is desired to use torque and power control in normal conditions, using a shaft speed controller
during ventilation would require switching between the two controllers. This approach is further
discussed in Smogeli [264]. If shaft speed control is chosen for normal conditions, anti-spin
control may not be necessary. However, performance improvements may still be achievable.

Remark 9.7 From a practical point of view, the proposed torque/power controller with anti-spin
requires more system knowledge than a conventional shaft speed controller. However, the most

330
 Inertia Q if Anti-spin Propulsion
compensation control unit Ta f T (.)
scheme
Qff +
Friction
Q max Pmax f Q (.)
compensation + Qa
Core Qci Qcas + Qc0 Torque Qc Motor Qm Shaft n
thruster X + limit dynamics dynamics
controller
n
Torque
observer
Tr
Qa
g z
Tras Anti-spin Ventilation Loss
control detection calculation
bQ

0
1 LP filter +
bQ rate limit g
1
z {0,1} . .
grise gfall tg
Td Tr nr
Reference g n (.) 0
LP filter + n ras
generator
rate limit g -1
n
(.) Tras
nas 1
. .
nrise nfall tn

Figure 9.19: The anti-spin control scheme, including core controller, torque observer, loss cal-
culation, ventilation detection, and anti-spin control actions.

331
important parameters are all readily available from the propeller and motor characteristics. In
addition, all industrial installations of thruster controllers require some manual tuning. Hence,
the remaining parameters, like the various rate limits and filter time constants, may be tuned on
site. The experience so far has been that the performance is acceptable within a large parameter
range.

9.10 Sensitivity to Thrust Losses

In order to analyze the performance and the robustness of the control laws given above we will
demonstrate the sensitivity subject to thrust losses. In order to compare the different control
schemes a simplified sensitivity analysis is carried out to address the most significant properties
of speed control, torque control and power control. Pitch control is not studied, but is expected
in general to have the same behavior as speed control.

9.10.1 Shaft Speed Feedback Control

Assume that theqshaft speed equals the speed reference at steady state. Using (9.11a), (9.17)
and  = sgn()  4  , the relationship between actual propeller thrust and the reference

thrust becomes

 =  (·)(1 −  ) n ,  (·)n  (9.71)
 0
Using (9.19) we can show that  (·) =   

9.10.2 Torque Feedforward Control

From a similar consideration of the torque control scheme, it is assumed that the motor torque
and the propeller torque equal the torque reference  = sign()0 5 2 at steady state.
Hence, combining (9.11a), (9.4), (9.17), and (9.18) yields

 (·)(1 −  ) 0 
 = n ,  (·)n  (9.72)
 (·)   0

Using (9.20) we can show that  (·) =  

9.10.3 Power Feedback Control

For the power control scheme the propeller and motor power tend to the power reference d at
steady state. Hence, combining (9.11a), (9.4), (9.18), and (9.56) yields
µ ¶23
 (·)(1 −  ) 0 
 = 23
n ,  (·)n  (9.73)
 (·)   0


We can show that  (·) = 23 


332
9.10.4 Combined Power and Torque Control
The analysis of the sensitivity of the combined power/torque controller (Smogeli et al. [262]) is
not as straightforward as in the preceding cases. However, since the combined controller is a
linear combination of the power and torque controllers, it is a reasonable assumption that the
sensitivity function for combined power/torque control  (·) may be expressed as

 (·) =  for  ∈ [ 23  1] (9.74)

such that for high ,  (·) ≈  (·) and for low ,  (·) ≈  (·). It can be shown that  then
may be approximated by the linear function  = 23 + 13 (), where () is the weighting
function used in the combined controller (9.59).

9.10.5 Positioning Performance

The effect of positioning performance for the different control schemes is clearly seen from the
two examples illustrated in Figures 9.20 and 9.22.

Example 9.1 Let  = 0 and  =  = 1. By using (9.8), (9.9),  0 =  ( = 0) and

0 =  ( = 0) with the same parameters as used in Figure 9.5 for  = 089, the sensi-
tivity functions  ,  , and  (9.71)—(9.73) can be computed as a function of advance ratio
 , see Figure 9.20. As expected all three sensitivity functions decrease, resulting in increased
thrust losses, for increasing . The speed control scheme is less robust for variations in  , while
the torque control scheme shows the best robustness.

Example 9.2 Thrust losses (Figure 9.21) caused by in-and-out-of-water effects can be modelled
by the thrust and the torque reduction functions  and  (9.17)- (9.18) as a function of the
submergence of the propeller shaft  divided by the propeller radius . The reduced disc area due
to the reduced submergence is accounted for in the computation of (9.8) and (9.9) according to
Lehn [163]. Let  = 0,  0 =  ( = 0) and 0 =  ( = 0) with the same parameters
as used in Figure 9.5 for  = 089, then the sensitivity functions  ,  , and  (9.71)—
(9.73) can be computed as a function of , see Figure 9.22. As expected all three sensitivity
functions decrease, resulting in increased thrust losses, for decreasing . The speed control
scheme appears to be less robust for reduced submergence, while the torque control scheme appears
to show the best robustness. However, care should be taken concerning these observations. In
order to achieve the reduced sensitivity, the torque controller will increase the thruster speed
significantly, resulting in propeller racing. This is an undesired effect. As shown here, this will
be solved by the anti-spin controller.

9.10.6 Sensitivity Functions Summary

Based on the controller formulations, the relations in the previous section, and assuming that
all controllers perform ideally, the steady-state sensitivity functions for speed control, torque
control and power control are shown in Table 9.2. For details in the derivation of the sensitivity
functions, see Sørensen et al. [287] and Smogeli [264].

333
 1

0.8

0.6

0.4

0.2

0
0 0.2 0.4 0.6 0.8 1
Advance ratio J

Figure 9.20: Sensitivity functions for the different control schemes:  (solid),  (dash-dot)
and  (dash) as function of advance ratio  for  = 089.

0.8

0.6

0.4

0.2

-0.5 0 0.5 1 1.5 2

h/R

Figure 9.21: Mean thrust losses  (dash-dot),  (solid) and reduced disk area (dash) due
to reduced submergence in calm water of a heavily loaded propeller as a function  for
 = 089, based on Fig. 4.1 in Lehn.

334
 1

0.8

0.6

0.4

0.2

-0.5 0 0.5 1 1.5 2

h/R

Figure 9.22: Sensitivity functions for the different control schemes:  (solid),  (dash-dot)
and  (dash) as function of ,  and  .

Sensitivity Speed control Torque control Power control

 
Thrust:  (·) , 
  (·) =   (·) =  

 (·) =  ( 

)23
q 13


Speed:  (·) , 
  (·) = 1  (·) =   (·) = 13

13
 
Torque:  (·) , 
  (·) =   (·) = 1  (·) = 13

12
 
Power:  (·) , 
  (·) =   (·) = 12  (·) = 1


Table 9.2: Sensitivity functions, where i=n for shaft speed control, i=q for torque control, i=p
for power control, and i=c for combined torque/power control

335
 The analysis of the sensitivity of the combined torque/power controller is not as straightfor-
ward as in the preceding cases. In Smogeli [264] it is shown that for the combined controller the
sensitivity function  (·) may be expressed as
  23+13
 (·) ≈   (  )  (9.75)

such that for high ,  (·) ≈  (·) and for low ,  (·) ≈  (·). For    002 this gives
an error of less than 5%, and has the correct limiting values for  = 0 and  = 1. Through
 the combined controller thrust sensitivity is dependent on the shaft speed , which makes it
less convenient for analysis. However, it is clear that  (·) will lie between  (·) and  (·) for
all . It is interesting to note that if the controllers have knowledge of the instantaneous values
of  and  , such that   =  and  =  at all times, all sensitivity functions
reduce to unity, since  from (9.47),  from (9.55), and  from (9.57) will be equal to the
true values ,  , and  . This means that all controllers perform identically. Also notice that
 (·) (·) ≡  (·) for all controllers, as should be expected.

9.11 Experiments
This section presents experimental results with the controllers designed above. The impor-
tance of the friction compensation scheme is investigated, and the dynamic performance of the
controllers in various operating conditions tested.
The majority of the experiments were conducted for shaft speed control, torque control,
and power control, since the main purpose of this work was to test the properties of the basic
thruster control methods. The combined torque/power controller was also tested, and behaved
exactly as expected. Its performance can be inferred from the performance of the torque and
power controllers.

9.11.1 Experimental Setup

The experiments were conducted in the Marine Cybernetics Laboratory (MCLab) at NTNU.
The MCLab basin is 40m long, 6.45m wide, and 1.5m deep, and is equipped with a towing
carriage and a wave-maker system. Two types of setups were used during the various tests. The
tested propeller was of conventional design with 4 blades, pitch ratio at 70% of the propeller
radius  = 10, and expanded blade area ratio  = 055. The propeller was tested both
with and without duct. For the results presented here a duct was used for the anti-spin tests
only. The propeller was attached to a shaft equipped with thrust and torque sensors inside an
underwater housing, and driven by an electric motor via shafts and gears with gear ratio 1:1.
The rig with motor, gears, underwater housing, shaft and propeller was fixed to the towing
carriage on a vertical slide, which was used to control the submergence of the propeller relative
to the free surface. Ventilation incidents could hence be generated by moving the propeller
vertically with a calm free surface. The motor torque was controlled from a PC onboard the
carriage, using feedback from the propeller shaft speed and the motor torque. The control code
was generated by rapid prototyping using Opal RT-Lab and source code in Matlab/Simulink.
A sketch of the experimental setup is given in Figure 9.23.
The main characteristics of the ducted propeller and drive system are summarized in Table
9.3. The nominal thrust and torque coefficients for the propeller without duct were found to be

336
Source CPU Control CPU Q
c
Simulink Executable
Matlab Opal RT-Lab Qm Motor
Opal RT-Lab
Qm ,n

Waves
Current Strut

Va Qa
Shaft
Ta
Underwater housing

Figure 9.23: Sketch of the experimental setup.

 [m]  0 0  [kgm2 ]  [Nm]  [Nms]

0.25 0.513 0.0444 0.05 1.0 0.01

Table 9.3: Main propeller data.

337
 0 = 0570 and 0 = 00750. The nominal thrust and torque coefficients for reversed thrust
were found to be  0 = 0393 and 0 = 00655.

Friction
In the experiments presented here, the shaft friction turned out to affect the performance of
the torque and power controllers. A feedforward friction compensation scheme as proposed
in Section 9.8.5 was therefore implemented. Over the course of the experiments, the friction
compensation coefficients were found to be in the range  0 ∈ [08 10] and  1 ∈ [0009 0011].

Control parameters
The control coefficients   and  used in the experiments were as given by (9.45), with
the half “width” of the transition region  = 3. The rotational inertia of the model-scale
propeller was not found to be a dominating term. Hence, the inertia compensation scheme in
(9.49) was not needed. Therefore, and since only feasible thrust reference trajectories were used,
no reference generator as in (9.48) was used. The constant plus linear friction compensation
scheme (9.50) was used, with  0 and  1 appropriately chosen in the range specified above.
The switching “width” of the smooth switching function in (9.51) was chosen as  = 05. The
shaft speed PI controller parameters (9.54) were chosen as  = 02 and  = 005 s, which
gave adequate tracking properties. For the combined controller (9.59), the weighting function
parameters (9.60) were chosen as [  ] = [4 05 1]. This gave a pure torque controller for
  05 and a pure power controller for   3.
The friction compensation was chosen with  0 = 10Nm and  1 = 001Nms. The inertia
compensation was not needed for the small model-scale propeller. The control coefficients were
chosen as   =  0 = 0513 and  = 0 = 00444. For the combined controller (9.59),
the weighting function parameters (9.60) were chosen as [  ] = [4 05 1]. This gave a pure
torque controller for   05rps and a pure power controller for   3rps. The shaft speed PI
controller parameters (9.54) were chosen as  = 02 and  = 005s.
In the load torque observer, the gains were chosen as  = 15 and  = −25. In the ventilation
detection scheme, the parameters were chosen as   = 06 and    = 09, and the detection
delay was set to  = 1s. For the proposed anti-spin controller, the settings for the primary
anti-spin action were ̇  = 1s−1 , ̇   = −1s−1 , and unless otherwise stated   = 03s. For the
secondary anti-spin action, the filter time constant was chosen as   = 005s, and the rate limits
were chosen as ̇ = 3s−2 and ̇  = −3s−2 . The desired shaft speed during ventilation was
set to  = 9rps. The latter choice was motivated from observations during the tests, where
it seemed that the most violent dynamic loading disappeared below 10rps. A maximum shaft
speed of max = 25rps was enforced.
The anti-spin control scheme appears to be robust to parameter tuning. In order to avoid
detection chattering, however, it is important not to choose   too large. In addition, if the
rate limits are chosen too low or the filter time constants too large, the anti-spin controller re-
sponse will become slow, and the shaft speed will be allowed to increase more during ventilation.
A further discussion and guidelines on tuning are given in [264].
Figure 9.24 shows time series for the proposed anti-spin controller with varying filter time
constants   at a thrust reference of 200N. Only the primary anti-spin action was used, such
that the goal of the anti-spin controller here was to keep the shaft speed constant.   = 03s

338
 14
No filter
13 τ = 0.3 s
γ
12 τ = 0.5 s
γ
τ = 1.0 s
n [1/s]
γ
11

10

7
Q [Nm]

6
m

8 9 10 11 12 13 14
Time [s]

Figure 9.24: Comparison of shaft speed and motor torque during a ventilation incident for
the combined controller with primary antispin action, using varying  filter time constants at
 = 200 N.

was found to be a good trade-off between response time and noise in the control law, and used
in the remainder of the tests presented below.

9.11.2 Nominal Tests

A total of 36 tests were run with zero advance velocity and varying motor setpoints. Each
setpoint was run for approximately 60 seconds in order to get good statistical values. The
nominal thrust and torque coefficients  0 and 0 for varying propeller shaft speed  and
 = 0 are shown in Figure 9.25, where also the mean values used in the rest of the work are
shown. The resulting mean values were  0 = 05359 and 0 = 00832. The shaft friction for
the tested propeller was found as the steady-state difference between the motor torque  and
the propeller torque  , and is plotted versus  in Figure 9.26. A second-order model was fitted
to the data and used in the rest of the work. The model is only valid in the tested shaft speed
range, i.e.  ∈ [0 10] .

339
 KT0 = 0.5359, KQ0 = 0.0832
0.6
KT0

0.5
Measured
Mean value
0.4
0.09

0.085
KQ0

0.08

0.075
2 3 4 5 6 7 8 9 10 11
n [rps]

Figure 9.25: Nominal thrust coefficient  0 and nominal torque coefficient 0 measurements
for varying shaft speed , and the mean values used in the rest of the work.

Friction model
1.4
Measured
1.2 Model
Qf [Nm]

0.8

0.6

0.4
0 2 4 6 8 10 12
n [rps]

Figure 9.26: Shaft friction  measurements for varying shaft speed  and the second order
model used in the rest of the work.

340
 Thrust characteristics

0.8

0.6

0.4

KT
0.2
10KQ
η0

0
−0.4 −0.2 0 0.2 0.4 0.6 0.8
Ja

Figure 9.27: Thrust characteristics from experiments. The continuous lines are found from 3rd
order polynominal fits to the data points.

9.11.3 Sensitivity Tests

In the sensitivity tests the thrust reference was kept constant at  = 100 , the advance
velocities were  = {−1 −05 0 05 1 15}  and the speed, torque and power controllers
were used. Each combination was run 3 times, giving a total of 54 runs. The length of the test
series were limited by the length of the basin, but gave adequate statistical values. The thrust
characteristics of the tested propeller is shown in Figure 9.27, where data points from all the
sensitivity tests with positive advance velocities are compiled together with 3 order polynomial
curve fits. The thrust and torque coefficients  and  and the open-water efficiency 0 were
calculated using the mean values of  ,  and  from each run.
The performance of the three controllers are summed up in Figure 9.28, where the propeller
thrust  , propeller torque  , shaft speed , motor torque  and motor power  are shown
for varying advance speeds. The friction compensation is not included in the plots of motor
torque and power. It is clear that the three controllers obtain their objectives: The shaft speed
controller keeps the shaft speed constant, the torque controller keeps the motor torque constant,
and the power controller keeps the motor power constant. As the advance speed increases, the
effective angle of attack of the propeller blades is decreased, and the propeller loading decreases
for a constant shaft speed. This can be seen in terms of reduced propeller thrust and torque at
increasing advance velocities for the shaft speed controller: At  = 15, the propeller thrust
is reduced from 100 to 18 . The torque and power controllers have much better performance,
since they will increase the shaft speed as the propeller loading decreases: At  = 15,
the propeller thrust is reduced from 100 to 65 and 50 respectively. The reason for the
propeller torque not remaining constant for increasing advance velocities are inaccuracies in the
friction model.
The sensitivity to variations in advance velocity  can be studied by representing  and

341
 Torque control
Power control
Shaft speed control
100
T [N]

50
a

0
4
Q [Nm]

2
a

9
n [1/s]

5
4
Q [Nm]

3
m

2
1

250
200
P [W]

150
m

100
50
−1 −0.5 0 0.5 1 1.5
V [m/s]
a

Figure 9.28: Comparison of propeller thrust  , torque  , shaft speed , motor torque  and
power  for the shaft speed, torque and power controllers in varying advance velocities  .

342
 Advance velocity sensitivity functions
1.5

0.5 sq(⋅)
sp(⋅)
sn(⋅)
0
−1 −0.5 0 0.5 1 1.5
Va [m/s]

Figure 9.29: Advance velocity sensitivity functions  (·) (shaft speed control),  (·) (torque
control) and  (·) (power control) for the tested propeller.

 as functions of  or  and ignoring all other loss effects, such that  =  () and
 =  (). The experimentally determined sensitivity functions for varying advance velocity
are shown in Figure 9.29. This confirms that the shaft speed controller is the most sensitive and
the torque controller the least sensitive to variations in advance velocity.
In all the results presented in this section, the data points for negative advance speeds are
in general less accurate than for positive advance speeds due to inhomogeneous inflow to the
propeller.

9.11.4 Dynamic Tests in Waves

To validate the dynamic performance of the controllers when subject to rapidly changing inflow
to the propeller a total of 34 tests in regular and irregular waves were performed. The thrust
reference was  = 100 and the carriage was kept stationary. A comparison of the controller
performance in regular waves with wave height 8 and period 1 is shown in Figure 9.30.
DP functionality has been simulated, such that all controllers gave the same mean thrust. The
results are summarized in the following:

• The shaft speed controller keeps the shaft speed constant, and has to vary the motor
torque and power in order to achieve this. The resulting propeller thrust and torque have
the largest variance.

• The torque controller keeps the motor torque constant, and as a result the shaft speed varies
with the loading. The resulting propeller thrust and torque have the smallest variance.

• The power controller keeps the motor power constant, and as a result both the shaft speed
and motor torque varies with the loading. The resulting propeller thrust and torque lie
between the shaft speed and torque controller values.

343
 Torque control
Power control
Shaft speed control

7
n [rps]

6.9
6.8
6.7

95
Ta [N]

90

85

3.8
Q [Nm]

3.6
a

3.4

5
Q [Nm]

4.8
m

4.6

215
P [W]

210
m

205

0 1 2 3 4 5 6 7
Time [s]

Figure 9.30: Controller comparison in regular waves. DP functionality has been simulated, such
that all controllers give the same mean thrust.

344
9.11.5 Anti-spin Tests
The anti-spin control scheme was tested in a variety of operating conditions: for constant and
time-varying thrust references, in waves, with a calm free surface and forced vertical motion
of varying amplitude  , period  , and mean submergence 0 , and in combinations of waves
and vertical motion. In the results presented here, the thrust reference was kept constant
during forced vertical motion with a calm free surface. This improved repeatability and enabled
comparison of the various controllers, since the time series could be synchronized by comparing
the relative vertical motion of the propeller. Because of the chaotic nature of ventilation, the
actual operational conditions for the propeller vary from one test run to another. However, the
repeatability is in general good. The results presented below are for 0 = 15cm,  = 15cm, and
 = 5s. Performance comparisons in other conditions, further validating the robust performance
of the anti-spin controller, are presented in Smogeli [264].
Figure 9.31 shows comparisons of the thrust  , propeller torque  , shaft speed , motor
torque  , and motor power  for four different thruster controllers during a ventilation
incident with  = 200N. The ventilation incident starts at  ≈ 112s, and terminates at  ≈ 145s.
The compared controllers are shaft speed PI control (9.54), torque control (9.55), power control
(9.58), and the proposed anti-spin controller based on combined torque/power control (9.59).
With anti-spin activated, both the primary (9.68) and the secondary (9.70) anti-spin actions
were used. The time series show that the torque controller and the power controller both lead
to propeller racing. The shaft speed controller and the anti-spin controller limit the shaft speed
as intended, with the secondary anti-spin action giving a slightly reduced shaft speed during
ventilation. The resulting thrust during ventilation is about the same for all controllers. That
is, the anti-spin controller, which reduces the shaft speed to 9rps during ventilation, produces
the same thrust as the torque controller, which races to the imposed limit of 25rps. This
corresponds well to the experimental results shown in Figure 9.12. The power consumption
of the torque controller is unacceptably high, whereas the power controller keeps the power
consumption limited. The shaft speed and anti-spin controllers give a lower power consumption
during ventilation, but this is not considered to be a problem.
Figure 9.32 shows comparisons of  ,  , ,  , and  for four controllers during a
ventilation incident with  = 300N: power control, combined control with primary anti-spin
action only (marked P), combined control with primary and secondary anti-spin action (marked
P+S), and shaft speed control. The torque controller could not be tested for this thrust reference,
since the propeller racing was too severe to give sensible results. The ventilation incident starts
at  ≈ 162s, and terminates at  ≈ 195s. Again, the thrust levels during ventilation are
almost identical for all the controllers. It can be noted that the primary anti-spin controller
and the shaft speed controller behave almost identically, confirming that the proposed anti-spin
controller has a similar performance as a well-tuned shaft speed PI controller during ventilation.
The secondary anti-spin action lowers the shaft speed as intended during ventilation. Figure 9.33
shows details from the anti-spin controller during a ventilation incident: propeller torque 
versus estimated propeller torque ̂ , estimated torque loss factor ̂  and ventilation detection
signal , torque modification factor , deviation of  from the measured   , and the desired
shaft speed  . The time series show that the anti-spin controller, including torque observer
and ventilation detection, performs as intended.

345
 Torque control
Power control
Combined control with antispin P+S
Shaft speed control

25
20
n [1/s]

15
10

200
150
Ta [N]

100
50

5
4
Qa [Nm]

3
2
1

6
Qm [Nm]

5
4
3

800
Pm [W]

600
400
200
11 12 13 14 15
Time [s]

Figure 9.31: Comparison of four controllers during a ventilation incident at  = 200N.

346
 Power control
Combined control with antispin P
Combined control with antispin P+S
Shaft speed control
25
20
n [1/s]

15
10

300
Ta [N]

200

100

6
Qa [Nm]

8
Qm [Nm]

2
800

600
Pm [W]

400

200
15 16 17 18 19 20
Time [s]

Figure 9.32: Comparison of four controllers during a ventilation incident at  = 300N.

347
 6 Qa
Q [Nm]

4 Qa
2
0

1
βQ, ζ [−]

βQ
0.5
ζ
0
1
γ [−]

0.5

0.4
γ − βQ [−]

0.2
0
−0.2

12
nras [1/s]

10

0 1 2 3 4 5
Time [s]

Figure 9.33: Time series of the main parameters in the anti-spin control law with primary and
secondary control action during a ventilation incident at  = 300N.

348
9.11.6 Discussion
The power plant on ships consists of several gas turbines or diesel engines driving electrical
generators. For safe operation there must in all load conditions be enough spinning reserves
providing sufficient available power for unpredictable variations in load in order to prevent black-
out. The black-out prevention by means of reducing load on heavy consumers must typically
respond faster than 500  to be effective. With torque and power control, the propeller load
is less sensitive to variations in the surroundings, giving less power disturbances on the network
and improved voltage and frequency quality. Additionally, the maximum power consumption
may easily be limited to the available power in both schemes, since the power limitation is
explicit in the torque and power control algorithm. This in contrast to speed controlled and
pitch controlled propellers, where the actual power load must be measured as a feedback signal
with an inherent time lag which deteriorates the black-out prevention response time. The
accurate and fast control of power and power limitation in torque and power control gives less
unpredictable load changes, and less need for available power. Hence, there will be a reduced
probability of blackout due to overload, since unintentional power peaks will be suppressed. In
traditional speed controlled drives there can be an overshoot in the commanded torque/power
of up to 5%. If this occurs for several thrusters simultaneously, significantly more power than
requested by the positioning system is consumed. With torque and power control this transient
overshoot problem is removed. The number of running generators can be reduced, such that the
average loading can be higher. This reduces wear and tear and gives less maintenance of the
prime movers. In addition torque and power control will lead to reduced mechanical wear and
tear of the propulsion components.
Low-level controllers based on shaft speed, torque, and power control for electrically driven
thrusters on vessels in normal operating conditions have been investigated, and evaluated ac-
cording to three main performance criteria. The conventional shaft speed controller gave the
thrust, torque, and power with the largest variance, and it was the least robust to disturbances in
the in-line flow velocity. The torque controller produced the thrust and torque with the smallest
variance, and was superior with respect to compensating for thrust losses due to disturbances in
the inflow. The power controller gave the least oscillations in the power, with the resulting pro-
peller thrust and torque in-between the shaft speed and torque controller values. The combined
torque and power controller gave the overall best improvement in the performance from low
to high loadings. Steady-state sensitivity functions describing the performance of the various
controllers in terms of resulting thrust, shaft speed, torque, and power when subject to thrust
losses were presented. Experiments illustrated the difference in performance and robustness of
the different controllers.
It has been shown that torque and power control is a feasible solution for high-performance
thruster control only if special precautions are taken during extreme environmental conditions,
when the propeller may be subject to ventilation and in-and-out-of water effects. To solve this
problem, an anti-spin thruster controller has been designed. The thruster performance was
monitored by a load torque observer, and the anti-spin controller was triggered by a ventilation
detection scheme. The anti-spin controller took control of and lowered the shaft speed. Exper-
iments with a model scale propeller were presented. The results showed that the torque and
power controllers with anti-spin had comparable performance to that of a well-tuned shaft speed
PI controller during ventilation, without compromising the superior performance of torque and
power control in normal conditions.

349
Chapter 10

Modeling and Control of Ocean

Structures

A flexible system is here defined as a dynamic system, which includes discernible elastic or flexible
motions in addition to the rigid body motion dynamics. When designing a control system for a
flexible system it is essential to account for the flexibility in the development of the controller
algorithm and in the location of sensors and actuators. In this chapter two different examples
of control of flexible ocean structures are shown.
The first example is about modelling and control of tensioned risers installed on so-called
tensioned leg platforms (TLPs). We will in this section investigate the possibility to reduce the
riser interference by use of automatic control of the riser top tension. In addition, the aim of this
section is to give an introduction to the concept of tension legged platforms, the finite element
method (FEM) and usage, the differences between a control plant model (CPM) and process
plant model (PPM), how to verify a mathematical model, control of slender ocean structures
and how to combine advanced modeling and control for real applications.
The second example considers dynamic positioning of a surface vessel connected to a riser
conducting a drilling or work-over operation. As the surface vessel is one of the boundary
conditions to the riser, coupling phenomena between the surface vessel motion and the riser
may cause troubles, unless precautions are taken in the dynamic positioning system. By proper
control of the vessel position, an d thereby the boundary condition, the riser is controlled.
Conventional controller designs for dynamic positioning of ships and floating marine structures
have so far been based on the principle on automatic positioning in the horizontal-plane about
desired position and heading coordinates defined by the operator. A three degrees of freedom
multivariable controller either of linear or nonlinear type, normally with feedback signals from
surge, sway and yaw position and velocities, has been regarded as adequate for the control
objective. For floating structures with small waterplane area such as semi-submersibles, feedback
from roll and pitch angular rotation velocity may also be included to avoid thrust induced roll
and pitch motions that are caused by the hydrodynamic and the geometrical couplings between
the horizontal and vertical planes. However, for certain marine operations this control philosophy
may not be the most appropriate approach ensuring safety and cost-effectiveness. For drilling
and work-over operations the main positioning objective is to minimize the bending stresses
along the riser and the riser angle magnitudes at the well head on the subsea structure, and
at the top joint as well. A positioning control strategy solely based on manual setting of the
desired position coordinates may not be the most optimal solution for these applications. Here

350
a hybrid dynamic positioning controller, that also accounts for riser angle offsets and bending
stresses is proposed. It is shown that a significant reduction in riser angle magnitude can be
achieved.

10.1 Mathematical Background

Flexible systems are (Edwards and Penney, [62]) described by partial differential equations
(PDEs) involving two and several independent variables (often time and spatial coordinates)
in contrast to ordinary differential equations (ODEs) that are dependent on a single indepen-
dent variable only (often the time). If we have spatial independent variables, then, by specifying
the values of the dependent variable on some boundary denoted as boundary conditions, arbitrary
functions in the general solution can be evaluated and a particular solution can be obtained. If
one of the independent variables is the time, then we must also specify the initial conditions in
order to obtain a particular solution. Problems involving boundary and initial conditions are
called boundary-value problems. ODE together with an initial condition is called an initial-value
problem.
In the present study, a focus will be set on implementation of the solutions. The software
package MATLAB will be used, and predefined functions within this software will be accessed
in the solutions.

10.1.1 Ordinary Differential Equations (ODE)

An ordinary differential equation (ODE) is an equation that contains one or several derivatives,
in one variable, of an unknown function f ( r). In general form, ODEs are often written as
r
M( r) = f ( r) (10.1)

where M is a matrix, r is state vector and f is a vector of functions. If M is invertible, it may
be incorporated in f  Otherwise, the system is said to be a differential-algebraic problem (DAE),
but this class of systems will not be a topic in cable mechanics. Notice that the equation is
written in vector notation. If the equation contains higher order derivatives, reordering may
bring it over to the general form.
Example 10.1 Conversion of higher order ODEs.
Given the following system
2 
+ 4 = 0 (10.2)
2
We want the system to fit into the general form, and want to have only first derivatives. The
following substitution may solve this:
1 =  (10.3)

2 =  (10.4)

Rewriting the equation we find the following system of ODEs
1
= 2  (10.5)

2
= −41  (10.6)


351
 ∙ ¸
2
this fits into the general form where M = I2×2 and f = 
−41

The solution to an ODE is a specific function r() which satisfies (10.1) in the domain of
interest, Ω() where { ∈ [0  ] → R Ω →C |  = dim(r)}  The derivative of ODEs is often
time, but we will see that also spatial variables may be taken as arguments. ODEs may be further
classified on linearity and the order of derivatives. Dependent on this classification analytical
solutions may be found, and various techniques are available, see for instance Kreyszig [153].

Initial value problems (IVP) and boundary value problems (BVP)

A problem described by ODEs may be classified as an Initial Value Problem (IVP) or a Boundary
Value Problem (BVP). If an analytical solution to a problem can be found, both IVPs and BVPs
may be solved directly by introducing the known parameters in the expression. If no analytical
solution can be found, several numerical methods are available. However, these do often demand
that a classification of the problem as IVP or BVP is done in advance.
IVPs requires that the initial values of the states are known. The solution is the states for
different values of the independent variable, e.g. time. The probably most common use of IVPs
described by ODEs is time simulation of a given system. In control problems, the systems are
often described by ODEs, and the effect of the controller may be verified by a simulation in
time. The general features of IVPs/ODEs are described in detail in Hoffman [119]. Numerical
solution of ODE/IVPs are almost solely based on finite difference methods Hoffman [119]. The
Runge-Kutta methods are widely used for ODE/IVPs, but other methods may be better for
specific systems Egeland and Gravdahl [64]. Some recent advances in the methods for solving
IVPs/ODEs may be found in Jackson and Nedialkov [129]. These topics will not be further
discussed in the present text.

Example 10.2 Initial value problem: Mass-spring-damper system.

Figure 10.1: Mass-spring-damper system.

A mass-spring-damper system may be described by the following equation

̈ + ̇ +  = 0 (10.7)
where  is the mass,  is the damping,  is the spring stiffness and  is the mass’ position. We
substitute 1 =  and 2 = ̇ This gives the following set of ODEs:
∙ ¸ ∙ ¸∙ ¸
̇1 0 1 1
=    (10.8)
̇2 − − 2
Given a set of initial values for 1 and 2  the system’s states (position and velocity of the
mass) as a function of time may be solved directly from the ODEs. This is a typical initial value
problem.

352
 If the present problem described by ODEs has specified boundary values, it is classified as
a BVP. ODE/BVPs are often steady-state problems where the boundary values are given in
space. Typical examples are heat-transfer problems and beams with lateral loading.

Example 10.3 Boundary value problem: Deflection of a beam.

z
F(x)

Figure 10.2: Uniform beam with lateral load,  ()

The deflection of a homogenous beam may be described by the following equation:

4 
 =  () (10.9)
4
where  is the modulus of elasticity,  is the moment of inertia of the beam cross section and
 () is a lateral distributed force acting on the beam. This equation is sometimes written as

2  ()
=  () (10.10)
2
 2
where  () =  2  The system may be rewritten in the following form
⎡ 1 ⎤ ⎡ ⎤

2
⎢ 2 ⎥ ⎢ 3 ⎥
⎢  ⎥ = ⎢ ⎢
⎥
⎥ (10.11)
⎣ 3 ⎦ ⎣ 4

⎦
  ()
4
 ()

where the first substitution is 1 =  This is a problem where the boundary values are known,
and the dependent variable,  is spatial. If we assume that  () is constant (or symmetric
around the middle of the beam), we realize that the deflection will be symmetric. It is now
2 (0) 2 ()
possible to apply boundary conditions to the ODEs: 1 (0) = 1 () = 0 and   2 =   2 = 0
or 3 (0) = 3 () = 0 gives four boundary conditions for the system of four equations. Notice that
the two last boundary conditions gives a pinned beam. If these were replaced by 2 (0) = 2 () = 0
it would represent a clamped beam. The problem of finding a () which fulfills the boundary
conditions is a boundary value problem.

For ODE/BVPs and PDEs (see Section 10.1.2), three types of boundary conditions may be
specified. Assume that the equation is written on standard form and that  is the solution of
the equation (for PDEs,  is often the solution). Then the following boundary condition may
be specified:

1. Dirichlet boundary condition. The value of the function is specified.

 is specified on the boundary.

353
 2. Neumann boundary condition. The value of the derivative normal to the boundary is
specified.

 is specified on the boundary.

3. Mixed boundary condition. A combination of the function and its normal derivative is
specified on the boundary, i.e.

 +   is specified on the boundary.

In Example 10.3, Dirichlet boundary conditions were applied for the rewritten problem.
Hoffman [119] shows two main classes of methods for BVPs: Finite difference methods and
methods based on trial functions which have to satisfy conditions on the boundaries. The finite
difference approach is divided into the shooting methods and the equilibrium methods.
The shooting method takes the boundary values at one end as initial values and solves the
problem as an IVP. If the solution does not satisfy the boundary values at the other end, a
tuning of the initial values or some other parameters of the system must be done before another
”shot” from the first end is performed. In this way an iterative approach may solve the problem.
The equilibrium method constructs a finite difference approximation of the exact ODE at
every point of the finite difference grid, including the boundaries. A system of coupled finite dif-
ference equations results, which must be solved simultaneously, thus relaxing the entire solution,
including the boundary conditions, simultaneously.
The second class of solving ODE/BVPs is based on approximating the solution by a linear
combination of trial functions, and determining the coefficients in the trial functions so as to
satisfy the boundary-value ODE in some optimum manner. The most common examples of this
type of methods are:

1. The Rayleigh-Ritz method.

2. The Galerkin method.

3. The finite element method.

An overview of these methods are important in cable mechanics, and will be described further
at the end of this section.

Implementation of ODE problems in MATLAB

The MATLAB computer package provides several routines which are suited to solve ODE/IVP.
The routines are based on the system’s characteristics, e.g. stiffness. The most general routine
is ode45, and should be used as a first attempt. For a tutorial and detailed reference, please see
Matlab [179].
Implementation of Example 10.2 requires two functions stored at the same location on the
computer. One main program and one function returning the derivatives of the states. Each
fixed parameter is given a value, and a time dependant solution to the system is found. Before
we proceed to the code, please recognize the system in the example to be of the form ẋ = Ax
This expression may be found in the code below:

354
function msd
m=2; % Mass
d=1.5; % Damping
k=6; % Spring constant
x0=[3 5]’; %Initial values

[t,y]=ode45(’xdot’,[0 20], x0,[],m,d,k);

figure(1)
clf
subplot(211)
plot(t,y(:,1))
grid
xlabel(’Time’)
ylabel(’Position’)

subplot(212)
plot(t,y(:,2))
grid
xlabel(’time’)
ylabel(’Velocity’)

function xd=xdot(t,x,flag,m,d,k)
A=[ 0 1;
-k/m -d/m];
xd=A*x;

The code above results in the plot found in Figure 10.3.

Implementation of ODE/BVPs are slightly more difficult than IVPs. MATLAB provides a
routine for two-point boundary-value problems. The routine is based on a collocation method
which is described in Shampine et al. [250]. Some information may also be found in Matlab
[179]. A good solution is dependent on the system’s properties and a good initial guess. If
analytical expressions for the partial derivatives can be given, the solution will be more accurate
and speed up.
Let’s solve the beam deflection problem in Example 10.3. Some reasonable parameters are
given, and the resulting code is given below:

function beam
w=0.05; %Width of beam
h=0.1; %Height of beam
L=5; %Length of beam
I=w*h^3/12; %Moment of inertia
E=90e9; %Modulus of elasticity
F=-2000; %Constant distributed force
x0=0:0.1:L; %Initial mesh

355
 4

Position
1

-1

-2
0 2 4 6 8 10 12 14 16 18 20
Time

0
Velocity

-5

-10
0 2 4 6 8 10 12 14 16 18 20
time

Figure 10.3: Time simulation of mass-spring-damper system.

sol = bvpinit(x0,[0 0 0 0]); %Creates initial guess for all states

sol = bvp4c(@beamODE,@beamBC,sol,[],E,I,F,L); %Solves the BVP
%Plots the result
figure(1)
clf
plot(sol.x,sol.y(1,:))
xlabel(’x-axis’)
ylabel(’Deflection (z-value)’)
title(’Geometry of the beam’)
%System’s ODEs
function dzdx = beamODE(x,z,E,I,F,L)
dzdx = [ z(2)
z(3)
z(4)
F/(E*I) ];
%System’s boundary values
function res = beamBC(ya,yb,E,I,F,L)
res = [ ya(1) %Pos. 1st end=0 :z(0)=0
yb(1) %Pos. 2nd end=0 :z(L)=0
ya(3) %2. derivative 1st end=0 :z’’(0)=0
yb(3)]; %2. derivative 2nd end=0 :z’’(L)=0

The code above will produce the plot shown in Figure 10.4.

356
 Geometry of the beam
0

-0.005

-0.01

-0.015
Defelction (z-value)

-0.02

-0.025

-0.03

-0.035

-0.04

-0.045
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
x-axis

Figure 10.4: Deflection of a beam with transversal load.

10.1.2 Partial Differential Equations (PDE)

A partial differential equation (PDE) is an equation which contains one or more partial deriva-
tives of an (unknown) function of two or more independent variables. The order of the highest
derivative is called the order of the equation. PDEs may be linear, quasi-linear or nonlinear.
A further classification in elliptic, parabolic and hyperbolic equations may be carried out, but
this will not be discussed in the present text. Taylor [311] is a three-volume collection of theory
on PDEs, and this is recommended for detailed reading. A few classes of PDEs have analytical
solutions. The others have to rely on numerical solutions. Computer packages like MATLAB
may solve some easy and common classes of PDEs, but the user will some times have to write
his own algorithms to solve a specific problem.
For the current topics the PDEs considered is of the class nonlinear hyperbolic equations.
Equations in this class is among others suited to solve continuum mechanics as cable mechanics
problems, vibrating strings, membranes and the motion of compressible fluids.
In Section 10.1.1 we divided problems described by ODEs into initial value problems and
boundary value problems. For the problems involving PDEs, we will in the present text only
consider initial-boundary value problems, IBVPs. The current problems will have both spatial
and time dependent terms. This may be considered as two incorporated ODEs of time and
spatial problems. In the previous section it was mentioned that time dependent problems were
associated with IVPs and spatial equilibrium problems usually were associated with BVPs.
When it comes to PDE/IBVPs, a direct relation to ODEs can be found: The time dependence
is solved as an initial value problem and the spatial dependence is solved as a boundary value
problem.

Example 10.4 PDE for a vibrating string - the wave equation.

357
 It can be shown that a vibrating string (with small amplitudes) may be modelled as

 2 ( ) 2
2  ( )
=   (10.12)
2 2
q
where  is deflection,  is longitudinal axis and  = 0   is axial tension in the string and
0 is mass per unit length. Notice that the solution  depends on both time and space, and
that the partial derivatives of time and space are constituted in the equation. With that, this
equation qualifies to the definition of a PDE. This particular equation is called the wave equation.

Example 10.5 Bernoulli beam.

We will now extend the static formulation for a homogenous beam with transversal load in
Example 10.3. It can be shown that time dependent response from a specified load may be written

2 4
0 +  =  ( ) (10.13)
2 4
The boundary conditions are similar to the static case. Because the equation contains derivatives
of both space and time, it fulfills the criterion to be a PDE.

MATLAB basic version has one built-in routine for solving PDEs. It may only solve a limited
class of equations, and only in one spatial variable. To apply this routine, the equation should
be written in the following form
µ ¶
    
c(   ) = −  f (   ) +g(   ) (10.14)
    

where  is the spatial vector,  is time, c is a diagonal matrix and g and f are vectors.  is
the solution. It turns out that this is not suitable for our hyperbolic systems, and an additional
toolbox is needed. MATLAB’s PDE toolbox can only solve problems in a 2-dimensional region,
have to be hacked in order to solve this 1-dimensional problem. The 3rd-party software FEMLAB
is able to solve problems in 1D, 2D and 3D and time, and would be an appropriate supplement
to the MATLAB family if a commercial solver is chosen.
A more direct option is to make an implementation on your own. This will be discussed in
the next section.

10.1.3 Methods for Numerical Solution of the Relevant Systems

In this section we will focus on the numerical solution of selected ODEs and PDEs. We will
formulate the spatial terms as a boundary value problem. Insertion of the solution will lead to
a time dependent ODE which may be solved with common methods described in e.g. Egeland
and Gravdahl [64]. The basis for the subsequent discussion will be the equation for an Euler
Bernoulli beam
2 4
0 2 +  4 =  ( ) (10.15)
 

358
The Rayleigh-Ritz method
This section is mainly based on Hoffman [119], Larsen [160] and a variety of other sources which
are not referred here.
The Rayleigh-Ritz method belongs to a class called variational methods. The objective of
these is to determine a minimum or maximum of a functional which depends on other functions
such as Z 2 Z 2
 [] = (   0  00  ̇)   (10.16)
1 1
where  is the functional and  is the fundamental function. The goal is now to determine the
particular function () which extremalizes []We then define
Z 2 Z 2 µ ¶
  0  00 
 = + 0  + 00  +  ̇  
1 1    
Z 2 Z 2 µ ¶
   ()  2 ()  ()
= + 0 + 00 +   (10.17)
1 1     2  
where  is the variational operator. In order to find an extremum,  is set equal to zero. We
apply integration by parts on the last three terms inside the integral, and because we demand
 = 0 at the boundaries (according to fixed end points) we obtain
Z 2 Z 2 ∙ µ ¶ µ ¶ µ ¶¸
   2   
 = − + 2 −   = 0 (10.18)
1 1   0  00   ̇
This equation must be satisfied for arbitrary distributions of  and this requires that the
expression inside the brackets to be equal to zero. We now receives the general Euler equation
for a time dependent system with one-dimensional deflection
µ ¶ µ ¶ µ ¶
   2   
− + 2 − = 0 (10.19)
  0  00   ̇
Now, it may not be obvious how this relates to the solution of an ODE/BVP or PDE. Let
us consider the static beam problem in Example 10.3, written on the following form
2  ()
=  00 () =  () (10.20)
2
This equation contains no terms in time or  or 0 ( is the solution, as  was the solution in
the beginning of the section). The method requires that a fundamental function,  is found,
and this should yield the system equation when the Euler equation is applied (10.19). In this
particular case, the functional will be given by
¡ ¢2
 =  0 + 2  (10.21)

Applying the Euler equation (10.19) gives

µ ¶
  
=  (10.22a)
   0
 ¡ ¢
2 = 2 0  (10.22b)

 00 =  (10.22c)

359
which (fortunately) resulted in the original equation.
We assume that the solution may be written in the form
=
X
( ) =  ()  () (10.23)
=1

This equation is often written as the product of vectors

( ) = φ () q() (10.24)

Because our system is time independent,  () =   where  is a constant. The Rayleigh-Ritz
method now asks for suggestions for what the solution looks like. These suggestions should be
placed in the  which are called trial functions. It is very important that these functions are
linearly independent, i.e.  =  ⇒  = 0∀ 6= and that they satisfy the boundary conditions.
For the current problem 1 () =  , 2 () = ( − 1) and 3 () = 2 ( − 1) should be suitable.
The next step is to substitute the trial functions into the functional to obtain [ ] Our
suggestion is from (10.23) that the beam’s bending moment can be found from

() =  () = 1  + 2 ( − 1) + 3 2 ( − 1) (10.25)

Use of (10.21) then gives a function on the form [( 1  2  3 )] .Taking the variation of this,
  
we find that [( 1  2  3 )] = 1
1 + 2
2 + 2
3  which is only satisfied when


= 0 (10.26)

When a solution to (10.26) is found with respect to   the values should be inserted into (10.23).
This will now be the final solution to the problem according to the Rayleigh-Ritz method.
In our case this gives the following equations when the beam has length 1
Z 1
¡ 0 ¢2
 =  + 2   (10.27)
0
Z 1
  ¡ 0 ¢2 
=  + 2   (10.28)
2 0  2  2
Z 1
  ¡ 0 ¢2 
=  + 2   (10.29)
3 0 3 3

1 = 0 is determined from the boundary conditions. Assuming that  () =constant, some
algebra leads to the following relation:
1 1 
2 + 3 =  (10.30)
3 6 6
1 2 
2 + 3 =  (10.31)
6 15 12
which has solution 2 = 2  3 = 0 Inserting into (10.25) gives the solution indicated in Figure
2 
10.5. () may be found from integration of the relation  () =   2
Until this stage we have only considered time independent systems. When time dependence
must be considered, the  () in (10.23) will no longer be constant. The first extended challenge

360
 Momentum diagram
250

200

150
M(x)

100

50

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x-axis

Figure 10.5: Bending moment diagram for the transversally loaded beam. ’o’ is the solution
found from the Rayleigh-Ritz method, and the solid line is an exact solution for some special
parameters.

will now be to find a suitable fundamental function which fits into the Euler equation (10.19).
Hamilton’s principle states that the integral of the difference between kinetic and potential
energy is stationary (attains a minimum value) for the real path of motion when the initial and
final values are given Larsen [160]. When we are working on energy, it is natural to consider the
Lagrange functional which is defined by:

 =  −  −  (10.32)

where  is the kinetic energy,  is the potential (strain) energy and  is the potential energy
for the external load. The Lagrangian for the beam example will then be given by
Z  Z ∙ ¸ Z 
1 1
= ( −  −  )  = 0 ̇ 2 − ( 00 )2 −  ( )  =   (10.33)
0 0 2 2 0

Hamilton’s principle states that

Z 2 Z 2 Z 
  =     (10.34)
1 1 0

Insertion into the Euler equation (10.19) gives

2 4
0 +  =  ( ) (10.35)
2 4
which is the equation which we want to solve. This energy-based approach is a common way of
finding the functional which has to be utilized in the Rayleigh-Ritz method.

361
 The next step is to find appropriate functions for  () It is usual to assume that the system
exhibits harmonic vibrations., i.e. q() = q0 sin  It can be shown that the fundamental
frequency  may be found as the Rayleigh quotient, which is expressed by
q0 Kq0
2 =  (10.36)
q0 Mq0
where the stiffness matrix is given from
Z  ¡ ¢
K= φ00 φ00  (10.37)
0
and the mass matrix is given from
Z 
M= 0 φφ  (10.38)
0
This eigenvalue problem in  degrees of freedom will give  eigenfrequencies. Notice that
the eigenfrequencies depends on the shape functions (via K and M). The shape functions
should therefore be chosen from physical insight of the particular problem. With this method
restrictions on the mode of vibration are introduced. The system may be too stiff, and the
method gives an upper bound solution to the eigenfrequency  If the external load force,
 ( ) cannot be described by a linear combination of the elements in q() the solution is not
likely to reflect this force’s influence on the vibration in a very good manner.
This closes the discussion on the Rayleigh-Ritz method. Although this method is widely
used within structural mechanics, it will not be applied to any of the problems related to
cable mechanics which is discussed in this report. The Rayleigh-Ritz approach requires that a
fundamental function like (10.21) is found for the system. In the current issues of cable mechanics
only the governing differential equations are known, and this favours another solution technique
- The Galerkin method.

The Galerkin method

The Galerkin method belongs to a class of methods called residual methods. The Rayleigh-
Ritz method were based on trial functions, and the minimization of the residual (error) by the
calculus of variations. As opposed to this approach, the Galerkin method is based on minimizing
the residual over the spatial variable. This is done by the use of weighting functions.
Consider the ODE/BVP problem in Example 10.3. The residual is given by
2 ()
= −  () (10.39)
2
where  is another notation for the bending moment,  () We assume that the solution is
given by
=
X
( ) =  ()  () (10.40)
=1
 () must be linearly independent and satisfy the boundary conditions. Because the system is
not time dependent,  () =  where  is constant. Next step is to choose weighting functions,
 () for the integral of the residual. The unknown parameters  may now be solved from
Z 2
 ()()  = 0 (10.41)
1

362
Insertion of the parameters into (10.40) will lead to a solution of the problem.
Let’s see how this relates to the beam problem. The residual is given from (10.39). We
choose 1 () =  , 2 () = ( − 1) and 3 () = 2 ( − 1)These functions are linearly
independent and satisfy the boundary conditions for beam length  = 1. This leads to the
following approximation of the solution for the beam’s bending moment (10.40)
 () = () = 1  + 2 ( − 1) + 3 2 ( − 1) (10.42)
1 is chosen to be 0 (see page 360)The next step is to choose the weighting functions,  ()
These are often chosen to be equal to the shape functions:  () =  () For the beam, this
leads to the following integrated residual (10.41)
Z 1
( − 1) [22 + 3 (6 − 2) −  )]  = 0 (10.43)
0
Z 1
2 ( − 1) [22 + 3 (6 − 2) −  )]  = 0 (10.44)
0

For  =  the final result is 2 = 2 , 3 = 0 which is the same as for the similar case with
the Rayleigh-Ritz method. This will always occur when the trial functions in the Rayleigh-Ritz
method is chosen to be the same as the weighting functions in the Galerkin method (Hoffman
[119]).
Now it is time to look at time dependent systems. Much of this material is taken from
Egeland and Gravdahl [64]. Again, a beam with time dependence will be considered. The
system is given by the equation
 2 ( )  4 ( )
0 +  = ()  () (10.45)
2 4
Notice that the traditional forcing term is replaced by  ( ) = ()  () in this particular
case. We define the residual
2 4
 = 0 +  − ()  () (10.46)
2 4
Use of (10.40) and (10.41) and a choice of  () =  () leads to
Z  
X Z 
£ 0000
¤
 ()   ()  () + 0 ̈ ()  ()  =  () ()  ()  (10.47)
0 =1 0

In the Galerkin method it is standard procedure to apply partial integration to reduce the order
of derivatives in this expression. Egeland and Gravdahl [64] show that
Z  Z 
 ()0000
 ()  = 00 ()00 ()  (10.48)
0 0
when the condition ¯ ¯
000 ¯ 0 000 ¯
 () () 0 =  () () 0 = 0 (10.49)
is fulfilled. Insertion into (10.47) and reordering of terms gives
X  Z  Z  Z £
£ ¤ £ 00 00
¤ ¤
0  () ()  ̈ () +   () ()   () =  () ()   ()
=1 0 0 0
(10.50)

363
This may be written in matrix form as:

Mq̈ + Kq = b () (10.51)

where
Z £ ¤
 = 0  () ()  (10.52)
0
Z £ ¤
 =  00 ()00 ()  (10.53)
0
Z £ ¤
 =  () ()  (10.54)
0

Eq. (10.51) may be recognized as an ODE/IVP which may be solved by standard methods (see
Section 10.1.1). Usually the shape functions are chosen to be Hermitian polynomials of third
order, and this leads to a consistent mass- and stiffness matrix.
In the present example the external force were separated into a product of spatial and time
dependent functions. This may not always be possible. Instead we may use the fact that the
4
operator  4 on a finite interval  ∈ [0 ] has eigenvalues   and corresponding eigenfunctions
   ∈ {1 2 3 }  so that

 4 ()
=  4  ()  ∈ {1 2 3 }  (10.55)
4
The solutions for   and  () depend on the boundary conditions Egeland and Gravdahl [64].
Sometimes the weighting functions,  () are chosen to be Dirac’s delta function. The
method is then called the collocation method. This reduced complexity also leads to reduced
accuracy of the results, and the method is therefore seldom used.
The Galerkin method will be applied to some of the problems in cable mechanics, and a
deeper insight will hopefully be achieved with further examples.

The Finite Element Method

Up to this stage we have only described the problems with solutions based on shape functions
and one single element. In order to increase the accuracy, let’s divide the spatial domain, (),
into several smaller domains,  ()Now we can solve the spatial problem for each of the smaller
subdomains and collect the results in the end, see Figure 10.6.

I(x)
I1(x) I2(x) I3 (x) In-1(x)

x1 x2
Element no. (1) (2) (3) (...) (n-1)
Node no. 1
x
2 3 n

Figure 10.6: Division of the spatial domain, and discretization of the weighted integral ()

The elements may be solved by the Rayleigh-Ritz method, the Galerkin method or other
methods which are found suitable. In the following we will only focus on the Galerkin method.

364
For a time independent system the weighted residual integral may be written as
Z 
[()] =  ()()  (10.56)
0
When the solution domain is divided into several subdomains, the global integral may be written
as a sum of all subintegrals. Let superscript () denote element number and subscript  denote
the node number.
−1
X
[()] =  () [() ()] (10.57)
=1
where Z +1
() () ()
 [ ()] =  ()()  (10.58)

is the integral for subdomain  () i.e. interval number  Here we assume that the global exact
solution () may be approximated by the series
() = (1) () + (2) () +  + (−1) () (10.59)
where () () is a local solution for each element,  is the number of nodes and  − 1 is the
number of elements. In the Galerkin weighted residual approach, the weighting factors  ()
are chosen to be the shape functions,  for the solution. When it comes to the Finite Element
Method, we will limit the discussion of static boundary value problems to linear interpolation
functions. Two approaches are dominating: The element formulation and the nodal formulation.
In the element formulation the local solution is given by two shape functions (Figure 10.7)
() ()
() () =   () + +1 +1 ()  ≤  ≤ +1  (10.60)
where
⎧
0 ⎨ 0   
() −+1 −+1
 () = − = − ∆()  ≤   +1  (10.61)
⎩ +1 −
0 0 +1 ≤ 
⎧
⎨ 0 0   
() − −
+1 () = =  ≤   +1  (10.62)
⎩ +1 − ∆()
0 0 +1 ≤ 
The nodal formulation is given by
() = 1 1 () + 2 2 () +  +   () (10.63)
where
(−1) ()
 () =  () +  () (10.64)
and
⎧
⎨ 0 0   −1
(−1) −−1 −−1
 () =  −−1 = ∆(−1)
−1 ≤     (10.65)
⎩
0 0  ≤ 
⎧
⎨ 0 0   
()
 () = − − +1
+1 −
= − −+1
∆()
 ≤   +1  (10.66)
⎩
0 0 +1 ≤ 

365
 1.0
1.0 (i-1)
i(x)
(i)
(i)
i(x)
(i)

x
i i+1 x i-1 i i+1

Figure 10.7: Shape functions for element formulation (left) and nodal formulation (right).

The element formulation is often considered to be easier to handle in higher dimensions.

Let’s return to the static beam problem from Example 10.3. The residual may be expressed
as
 =  00 () −  () (10.67)
We apply the weighting functions and integrate over the whole solution domain
Z 
£ ¤
=  ()  00 () −  ()  = 0 (10.68)
0

A standard procedure is to reduce the order of the derivative in (10.68). Integration of the first
term by parts yields
Z  Z 
¯ 0
00 0
   =   0 − ¯   00 
0 0
Z 
0
= −   0  +  0 () () −  0 (0) (0) (10.69)
0

The last two terms in this equation involve derivative boundary conditions. For Dirichlet bound-
ary conditions these terms are not needed, but for Neumann boundary conditions they introduce
the derivative condition on the global solution domain Combination of (10.68) and (10.69) gives
Z ³ ´
0
= −  0 −    +  0 () () −  0 (0) (0) (10.70)
0

Our goal is to find a procedure or algorithm for the solution at each nodal point. Eq. (10.66)
and discretization yields
Z  Ã (−1)
! Z +1 Ã ()
!
 () 0 (−1)  () ()
= −  −  ()  + −   0 −  ()  = 0
−1   
(10.71)
Because
(−1)
 () 1
=  (10.72)
 ∆(−1)
()
 () 1
= −  (10.73)
 ∆()
we find the following integral for the system
Z  µ ¶ Z +1 µ ¶
1 0 (−1) 1 0 ()
= −  −  ()  +  −  ()  = 0 (10.74)
−1 ∆(−1)  ∆()

366
The integral of the shape functions alone is simply the area of a triangle. We may for simplicity
approximate the external force to be constant over an element. This constant force can be
written
+1 + 
̄ () =  (10.75)
2
This leads to the following integral
1
R  0
R
(−1)  (−1) ()+
 = − ∆(−1) −1   − ̄ 
1
R +1 0 ()
R +1 ()
−1
(10.76)
∆ ()    − ̄    () = 0

We know that
Z Z Ã (−1)
!

0
 −1 () (−1)

  = −1 +  
−1 −1  
Z 
 −1
= (−1)
−  =  − −1  (10.77)
−1 ∆ ∆(−1)

and that
Z 
(−1) ∆(−1)
  =  (10.78)
−1 2
Z +1
() ∆()
  =  (10.79)
 2

which yields

1 (−1) ()
(−1) ∆ 1 () ∆
=− ( − −1 ) − ̄ + (+1 −  ) − ̄ = 0 (10.80)
∆(−1) 2 ∆() 2
Eq. (10.80) is valid for nonuniform grid. If we assume that the grid is uniform, and (10.80) is
multiplied by ∆, the nodal equation for node  will be
à !
̄ (−1) + ̄ ()
+1 − 2 + −1 = ∆2  (10.81)
2

This equation holds for all nodes in the grid except for the boundary nodes if Neumann boundary
conditions are provided. Eq. (10.70) shows that the derivative conditions have to be included in
(1) (−1)
the integral at the boundaries. Notice that 1 (0) = 1 (0) = 1 and  () =  () = 1
The solution for end nodes may then be calculated from

̄ (1)
−1 + 2 = ∆2 + ∆ 10  (10.82)
2
̄ (−1)
−1 −  = ∆2 − ∆ 0  (10.83)
2
Example 10.6 Calculation for static beam problem based on a finite element model.
The static beam problem is given by

 00 =  (10.84)

367
A nodal solution involving linear shape functions is given in (10.81). Assume that this is dis-
cretized into five nodes and four elements. The boundary conditions are given from 1 = 1
and 5 = 5  where  is a constant. The unknown moments, 2 3 and 4 are given from
(10.81), and may be calculated from the solution of the following equation:
⎡ ⎤
⎡ ⎤⎡ ⎤ ⎡ ⎤ ̄ (1) ⎡ ⎤
−2 1 0 2 2 1 1 0 0 ⎢ (2) ⎥ −1
⎣ 1 −2 1 ⎦ ⎣ 3 ⎦ = ∆ ⎣ 0 1 1 0 ⎦ ⎢ ̄ ⎥ ⎣ 0 ⎦
2 ⎣ ̄ (3) ⎦ + (10.85)
0 1 −2 4 0 0 1 1 −5
̄ (4)
Notice that solution of this equation may involve calculation of the inverse of the first left-hand
matrix. This matrix grows in size when the number of elements are increased, and calculation
of inverse may require high computation time. Other solutions are possible, but those will not be
discussed in this text.

A simple implementation of the static beam problem is shown in the code below.

function fem_beam
n=30; %Number of nodes
L=1; %Length of beam
Dx=L/(n-1); %Deltax
M1=0; %Boundary condition
Mn=0; %Boundary condition
Fc =-2000; %Constant distributed force.

%Boundary condition matrix

Mb=zeros(n-2,1);
Mb(1)=M1;
Mb(length(Mb))=Mn;

%Left hand matrix

A=diag(-2*ones(n-2,1),0)+diag(ones(n-3,1),1)+diag(ones(n-3,1),-1);
Fselect=diag(ones(n-2,1),0)+diag(ones(n-3,1),1);

%Force matrix
F=ones(n-2,1)*Fc;

%Calculation of solution
M=inv(A)*Dx^2/2*Fselect*F+Mb;

%Adds the boundary solutions to the solution vector

Msol=[M1; M; Mn];
xsol=0:Dx:L;

%Plots the results

figure(1)
plot(xsol,Msol)

368
 The code produces the plot shown in Figure 10.8. Notice how the FEM solution converges
to the exact solution as the number of elements is increased.
Momentum diagram
250

200

150
M(x)

100

50

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
x-axis

Figure 10.8: Solution of the static beam problem with the Finite element method and linear
interpolation functions. The exact solution is shown with a thick solid line. FEM solutions is
shown with 3, 7 and 10 nodes (thin solid lines), and for 30 nodes (’o’).

The last step of our discussion of the Finite Element Method will include equations with
time dependence - PDEs. This material is mostly found in Egeland and Gravdahl [64]. Again,
our system is
2 4
0 2 +  4 =  ( ) (10.86)
 
where  ( ) is assumed to be written  ( ) = ()  () The procedure described for ODE/BVP
will not apply. Instead, we write the solution as

X £ ¤
( ) = 1 () 1 () + 2 () 2 ()  (10.87)
=1

where 1 and 2 are linearly independent and zero in every node apart from node  The
condition 1 () = 2 () = 0  ∈
 h−1  +1 i is also fulfilled. The physical interpretation of
the physical coordinates is
1 () = (  ) (10.88)
(  )
2() =  (10.89)

We may now define
£ ¤
ϕ = 11  21  21  22   1  2  (10.90)

q = [11  21  21  22   1  2 ]  (10.91)

369
The system may now be written as

Mq̈ + Kq = b (10.92)

where
Z 
M = 0 ϕ ϕ  (10.93)
0
Z  ¡ ¢
K = 0 ϕ00 ϕ00  (10.94)
0
Z 
b = ϕ()  (10.95)
0

Eq. (10.92) may now be solved by traditional methods for ODE/IVPs.

An example of typical shape functions which holds for beams is
⎧
⎪ (−−1 )2 (−−1 )3
⎨ 3 ∆2−1 − 2 ∆3−1
⎪ −1 ≤  ≤ 
2 3
1 =
⎪ 1 − 3 (−
∆2
)
+ 2 (−
∆3
)
  ≤  ≤ +1  (10.96)
⎪
⎩
0 otherwise
⎧ 2 3
⎪ (−−1 ) (− −1 )
⎨ − ∆−1 + ∆2−1
⎪ −1 ≤  ≤ 
(− )2 (− )3
2 =    (10.97)
⎪  − 2 ∆ + ∆2   ≤  ≤ +1
⎪
⎩
0 otherwise

These are the Hermite polynomials, and they satisfy the conditions which are set to shape
functions in the Finite Element Method.
Notice that the solution of (10.92) is a solution of the spatial variable as a boundary value
problem, and a solution in time as an initial value problem. Problems like this are often called
initial-boundary value problems, IBVPs as seen Section 10.1.2.

370
10.2 Modelling and Control Top Tensioned Risers
Anne Marthine Rustad
Department of Marine Technology, NTNU

We will in this section investigate the possibility to reduce the riser interference by use of
automatic control of the riser top tension. A control strategy based on equal payout for all
risers in an array was proposed by Huse and Kleiven [124]. Results from the model tests showed
that the equal payout by connecting all the risers to one common frame at the top end gave a
significant reduction of the probability of collision in steady current. This section is motivated
by their work, 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. A
comprehensive model of the dual riser system with TLP and current forces is presented. The
riser model is verified by use of the computer program RIFLEX [230]. The top tensioned riser
control system includes both the physical structure with identification of the limitations and the
software implementation with controller objective and algorithm. This section is based on work
published in Rustad [237] and Rustad et al. [238]. For details in references, please see Rustad
[237].
This section is organized as follows: Section 10.2.1 describes the problem statement in more
detail, while Section 10.2.2 gives and introduction to the TLP concept. Section 10.2.3 deduces
the modelling of the system, including the current forces acting as a disturbance. In Section
10.2.4 this model is verified by use of the existing software RIFLEX. Section 10.2.5 compares
the different control plant models. The controller design is presented in Section 10.2.6, while
simulations and results are found in Section 10.2.7. Section 10.2.8 summarises the section.

10.2.1 Motivation and Problem Description

Offshore petroleum production started in the most shallow waters. As these reservoirs are
exploited, combined with an increasing demand for oil and gas, the petroleum industry has been
moving to ever deeper waters. Riser technology is an 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. 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. 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 interference between risers:

1. Increasing the riser spacing.

2. Increasing the top tension.

371
 Figure 10.9: The wellbay on the Snorre TLP (www.statoil.com).

Both may result in significant cost penalties. Increasing the riser spacing means increasing
the size of the wellbay (Fig. 10.9). 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 aim of this section is to show how the number of collisions can be reduced by
applying dynamically varying feedback 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 semi-submersible
or a production ship (floating production storage offloading - FPSO), and (2) tensioned risers
with wellhead on a compliant platform, like tension leg platform (TLP), spar1 or deep draft
floater (DDF). These are the most common 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.
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, and may consist of
different risers applied for production, drilling, workover, export, etc.
A TLP is chosen as the platform concept in the example presented here. 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 due to the geometric restrictions for heave motions
caused by the axially rigid tendons. For other floaters like spars, FPSOs and semi-submersibles,
the demand for stroke capacity will be much higher, which means that other riser solutions like
steel catenary risers (SCR) or other types of flexible risers are preferred. Note however, that
most of the mathematical modelling of the TLP in Sec. 10.2.3 is independent of vessel type,
1
A spar is a vertical, cylindrical buoyant platform, usually manned.

372
and is also applicable for a semi-submersible or similar.

10.2.2 Tension Leg Platform

The main purpose of the TLP model in this example is to represent the platform motions at
the wellhead area where the risers are connected 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 subjected to active control. Heave motions of a TLP have
two components. One is caused by dynamic elastic strain and deflections of the tendons, which
will influence 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 neglected 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 section is mainly based on
Demirbilek [52], Faltinsen [66] and Larsen [40].

Background for the TLP solution

Offshore platform concepts are usually classified into two major categories; fixed and compliant.
Fixed platforms stand at the sea bed 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 water, 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 that are caused by extreme
waves will therefore be quasi-static since such waves typically will have periods above 10s. As
offshore development moved towards deeper waters, the application of conventional fixed jackets
approached its limits principally imposed by the dynamic behavior of the structures. 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 (  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 response. However, dynamics
of rigid body motions for floaters are of concern. Eigenperiods are controlled by the geometry
of the water-plane (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, semi-submersible, spar and FPSO belong to this category.
Towers are fixed to the sea floor 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.

373
 A TLP acts like a floater with regard to in-plane motion components (surge, sway and yaw),
but like a fixed platform for the out-of-plane components (heave, roll and pitch). The governing
parameters for the in-plane stiffness are the tension and length of the tendons, while the cross
section area, modulus of elasticity and length of the tendons decides out-of-plane 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 environmental forces, the choice of platform system also 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
water 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 system, 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. Hence, economic factors make the compliant platforms and the TLPs in particular
one of the leading candidates for major deep water developments, especially when a dry tree
solution is preferred.

The TLP concept

The main 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 sea floor and the pro-
duction equipment onboard the platform, heave, roll and pitch motions had to be minimized
and considered as 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 in-plane motions compliant, 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.
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 connecting the upper structure to a foundation at the sea bed.
These taut mooring are called tension legs, tethers or tendons. 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. The different parts of the TLP are illustrated in Fig. 10.10, and defined
by the API [10] as follows:

• The hull consists of the buoyant columns, pontoons and the intermediate structure brac-
ings.

• 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 sea bed and consists of templates and piles, or even a
gravity system.

• The tendons connects the platform to the foundation at the sea bed.

• The mooring system consists of the tendons and the foundation.

374
Figure 10.10: The layout of the Brutus TLP in the GoM (www.offshore-technology.com).

375
 Figure 10.11: Development of TLPs and riser depths.

• The risers include drilling, production and export risers.

• The well system includes flowlines, risers, riser tensioners, wellhead and the subsea well
templates.

• The tension leg platform includes all the above, in addition to all deck equipment and
the hull system.

Interest in TLPs dates back to 1960. During the following two decades, and especially after
the installation of the Hutton TLP in 1984, the TLP concept began attracting more attention
from the offshore industry 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. Deepwater TLPs are mainly found in the Gulf of Mexico, but there are also
some West of Africa and off the coast of Brazil. The increase in water depth for TLP is shown
in Figure 10.11

10.2.3 Mathematical Modelling

Mathematical models may be formulated in various levels of complexity. We usually distinguish
between process plant models and control plant models. 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 dynamics, including environmental disturbances, control inputs and

376
 R1 R2

I U D

Figure 10.12: The risers in a tandem arrangement. I: Incoming flow, U: Upstream riser, D:
Downstream riser.

sensor outputs. A successful numerical simulation for design and verification of our control
system 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 model-based controller. The CPM is often formulated such
that the analytical stability analysis becomes feasible.
This section will focus on mathematical model of the process plant. 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. The current is modelled as an in-plane profile with varying velocity through
the water column. The TLP is exposed to surface current, wind and waves.

Hydrodynamic interaction
Interaction between two cylinders are often classified into two categories according 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, see Zdravkovich
[331]. Top tensioned marine riser systems normally falls into the latter category, although
proximity interference also can take place. It is also found that the in-line motion are much
larger than the transverse motion. Hence, in this paper we will focus on two risers in a tandem
arrangement, were the centers of the two cylinders are aligned parallel to the free stream (Fig.
10.12). A two dimensional model will capture the most important dynamics of the riser array
system.
Kalleklev et al. [143] stated that interaction between two neighbouring risers will not have
any hydrodynamic influence on the upstream riser (R1) beyond a certain point. Hence, R1
can be treated as an isolated riser. The attention will therefore be given to the hydrodynamic
influence on the downstream riser (R2).
In the literature it is distinguished between three different kinds of excitation forces and
physical mechanisms on a pair of adjacent risers. The most important effects for assessment of
riser interaction are found in DNV-RP-F203 [57]:

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.

377
 y

Vc x

Vr(y)

Figure 10.13: The decrease in the in-line water particle velocity in the wake region, from Huse,
[123].

Here, focus will be 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 treated any further. So far we do not have a reliable model for prediction of the
WIO type of response. Inclusion of these effects may have an influence on the presented results.
R2 experiences reduced mean drag force due to shielding effects, depending on the location in
the wake. A semi-empirical static wake formulation to account for the hydrodynamic interaction
between individual risers in steady current was proposed by Huse [123]. The reduced velocity
field in the wake of the upstream cylinder, see Fig. 10.13, is given by
r
  −0693(  )2
 ( ) = 2     (10.98)

where  is the incoming current velocity on R1,  is the drag coefficient,  is the diameter,
and  is the distance away from the centerline of incoming velocity profile.  and  are defined
as
4 p
 =  +   = 1     (10.99)

where 1 = 025 and 2 = 10 for a smooth cylinder.  is the distance behind the upstream riser
R1, and  is the distance between the downstream riser R2 and a virtual wake source upstream
of R1. The mean inflow on the downstream riser is hence given by

 =  −   (10.100)

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 behaviour 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. The center to center distance should therefore preferably be
kept larger than 2D.

378
 g1 b1
g2 b2
g3
b3

i3
i2

i1

f3
f2

f1

Figure 10.14: The coordinate systems:  - fixed frame for first riser,  - global frame at the
surface,  - body frame of TLP,  - element frames.

Kinematics and the coordinate systems

Four orthogonal coordinate systems are used to describe the riser and TLP motions, see Fig.
10.14. A bold letter with subscript  = {1 2 3} denotes the unit vector along the   and 
axes in the frame respectively, i.e. f1 denotes the unit vector along the -axis for the sea floor
fixed frame.

• The  -frame ( f1 f2 f3 ) is considered inertial and is fixed to the sea floor. The positions
of all the riser nodes in the global system are described relative to this frame.

• The -frame ( 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.

• The global -frame ( g1 g2 g3 ) is fixed to the sea surface right above the  -frame with
a distance in heave direction equal to the water depth and is also considered inertial.

• The -frames ( i1 i2 i3 ) are located in the -th node of the riser and is a local frame for
each element . 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.

Riser modelling with the Finite Element Method

The partial differential equation (PDE) governing the static and dynamic behaviour 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 FE

379
Figure 10.15: 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.

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
are of major importance this will only introduce a small error. Hence, a model consisting of bar
elements is sufficient. Each bar element can be described with four degrees of freedom (dofs),
that is two translational dofs in both ends of the element.  is transverse of the element, positive
to the right along i1 , and  is along the element, positive upwards along i3 , see Fig. 10.15.
The riser is fixed to the sea bed, and the top node displacement is prescribed in the horizontal
direction, whereas it is free to move vertically. Note, however, that the top tension,   is
introduced at the upper end, which prevents the riser from collapsing. The tension vector, f ,
is an external force and will be found on the right hand side of the equation of motion according
to £ ¤
f = 01×(2+1)   (10.101)

Transformation of the riser elements There are  elements and  + 1 nodes, with node
number 1 at the sea bed and node  + 1 at the sea surface.  is the inclination of element 
relative to the global coordinate system  -frame, which is situated at the sea bed in the first node
of the upstream riser. The numbering of elements, nodes and the angle of each local element
are illustrated in Fig. 10.16. v is the current in node  expressed in the inertial  -frame. Note
that  is used as the numbering of the elements, nodes and the corresponding inclination.
The positions  and  of the nodes along the riser are found through equilibrium iterations,
and are used to calculate the sine and cosine of the inclination  of each element, needed for
use in the transformations between the global and local coordinate systems. The length of each
element  is found by use of the Pythagorean theorem.

380
Figure 10.16: Numbering of the elements encircled, nodes in boxes, current  in each node 
and the attitude  of each local element , relative to the global frame.

381
 q
∆ = +1 −   ∆ = +1 −    = ∆2 + ∆2  (10.102)
∆ ∆
cos  =  sin  =  (10.103)
 
Since we are only considering a two dimensional system with two dofs in each node, the trans-
formation matrix from  to  is written as
∙ ¸
 cos  sin 
T0 (r) =  (10.104)
− sin  cos 

where r is the displacement vector, as the inclination in each element is given as a function of
the positions, given in the sea floor fixed  -frame. For notational simplicity the superscript  is
omitted as r is always given in the  -frame. The full displacement vector is given as
£ ¤
r = 1 1 2 2 · · ·   · · ·     +1  (10.105)

For all four dofs related to an element we have

" #
 T0 (r) 02×2
T (r) =  (10.106)
02×2 T0 (r)

as the inclination of each element is a function of the positions of its end nodes.

Riser system stiffness matrix For each element a stiffness matrix is defined based on the
local coordinate system. The local stiffness matrix consists of two terms; the elastic stiffness,
k , and the geometric stiffness, k . The geometric stiffness matrix includes the axial tension
 acting on the particular element . The elastic stiffness matrix works in the axial direction,
whereas the geometric stiffness in the lateral direction. The resulting stiffness matrix k for the
element  in its own frame becomes
⎛ ⎡ ⎤ ⎡ ⎤⎞
0 0 0 0 1 0 −1 0
1⎜ ⎢ 0 1 0 −1 ⎥ ⎢
⎥ +  ⎢ 0 0 0 0 ⎥⎟ 
⎥⎟
k = k +k = ⎜  ⎢ (10.107)
 ⎝ ⎣ 0 0 0 0 ⎦ ⎣ −1 0 1 0 ⎦⎠
0 −1 0 1 0 0 0 0

where  is Young’s modulus of elasticity,  is the cross sectional area of of the riser material,
here steel, and  is the length of element  found in (10.102). The axial tension  in each
element is found as a function of the elongation of the element ;

 = ∆  ∆ =  − 0  (10.108)
0
where 0 is the initial length of an element in an untensioned riser.
Each local matrix k is transformed to the global  -frame fixed to the seabed, and all the
local stiffness matrices are assembled to form the global stiffness matrix K for the riser. The
local stiffness matrix is transformed to  -frame by
∙  ¸
  k̄11 k̄12
k̄ = T k T =  (10.109)
k̄21 k̄22

382
where k̄ is the stiffness matrix for element  in the global  -frame. For k̄11 the superscript is
used to recognize the element number, and the subscript is used to identify the different parts of
the stiffness matrix, needed for the concatenation to the global matrix. T is the transformation
matrix from the  -frame to the -frame. The full, global stiffness matrix in the  -frame for 
elements is then
⎡ 1 ⎤
k̄11 k̄112
⎢ k̄121 k̄122 + k̄211 k̄212 ⎥
⎢ ⎥
⎢ 2
k̄21 2 3
k̄22 + k̄11 ⎥
⎢ ⎥
K=⎢ .. ⎥
⎢ . ⎥
⎢ ⎥
⎣ k̄22 + k̄11 k̄12 ⎦
−1  

k̄21 k̄22
The global stiffness matrix K is always given in the  -frame, and the superscript  is hence
omitted.

Riser system mass matrix The local mass matrix consists of three terms; the structural
mass of the riser m , the internal fluid m , and the hydrodynamic added mass m . The
structural mass is given by ⎡ ⎤
2 0 1 0
  ⎢
⎢ 0 2 0 1 ⎥
⎥
m = (10.110)
6 ⎣ 1 0 2 0 ⎦
0 1 0 2
where  is the density of the riser material steel. The internal fluid is included as well
⎡ ⎤
2 0 1 0
   ⎢ 0 2 0 1 ⎥
m  = ⎢ ⎥ (10.111)
6 ⎣ 1 0 2 0 ⎦
0 1 0 2
where  is the internal area of the riser, and  is the density of the internal fluid.
The local added mass in the axial -direction is assumed to be zero, while the added mass
for a circular cylinder in the lateral -direction is equal to the displaced water volume. The
matrix for the added mass is then given as
⎡ ⎤
2 0 1 0
    ⎢
⎢ 0 0 0 0 ⎥
⎥
m =  ⎣ (10.112)
6 1 0 2 0 ⎦
0 0 0 0
where  is the area found from the exterior diameter  , and  is the density of water. The
coefficient  is called the hydrodynamic added mass coefficient. For a circular cylinder  = 1.
The total local mass matrix m for each element  in its own frame is then the sum of the three
terms written as
m = m + m  + m  (10.113)
To calculate the global mass matrix for the riser, we transform each local mass matrix from its
local -frame to the global  -frame, similar to what was done with the stiffness matrix. These
are assembled to the global mass matrix M in the same way as the stiffness matrix.

383
Structural damping The damping experienced by a riser is a combination of the structural
damping and hydrodynamic damping, resulting from both radiation and viscous dissipation of
energy, see Faltinsen [66]. 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

C = 2 K (10.114)

The coefficient 2 can be found from simple equations if the total damping level is known. The
term proportional to mass is neglected, see Berge et al. [23]. Here we have assumed that the
structural damping is 1.5% of the critical damping at the eigenperiod of  = 10s. Frequencies
lower than this will have smaller damping, and the damping is assumed linearly increasing with
logarithmic increasing , such that  = 015% for  = 100s. Thus, the material damping is
small relative to the hydrodynamic damping.

Hydrodynamic forces
The hydrodynamic forces on the risers are calculated using Morison’s equation modified for a
moving circular cylinder. The horizontal hydrodynamic force on a strip of the cylinder can be
written
2 2 1
 =    −   ̈ +    ( − ̇) | − ̇|  (10.115)
4 4 2
where  and  are the undisturbed water velocity and acceleration, ̇ is the response velocity, and
̈ is the structure’s acceleration.  =  +1 is the inertia force coefficient. The acceleration of
the water has a material derivative,  =   , 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 (10.115) 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  , which is the last term of (10.115). This drag force is calculated for each
element and summarized as concentrated forces in each node. The force vector f is found at
the right hand side in the equation of motion given in (10.120) later in the text.
The added mass  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  . 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  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  = 10 for both risers is applied. It is also important to note that variations
of  will not alter any conclusions regarding the control system design for this study since the
estimates of drag forces are not needed for control purposes.

Equilibrium iteration and numerical integration method

The modelled system contains nonlinearities in the mass, stiffness and damping matrices, in
addition to the nonlinearities due to drag and hydrodynamic damping. In addition, the top

384
 Pi

Fint
Fcur
Fext
Weff

Pi-1

Figure 10.17: Balanced internal and external force.

tension is varying in time as it is the control input. The tension variation is linearly decreasing
along the riser. The forces acting on each element, such as tension, effective weight and current
due to drag are computed in the -frame for each element and thereafter transformed to the
 -frame.
As in traditional FEM, the nonlinearities are solved numerically by incremental formulation
with the Newmark- time integration method and Newton-Raphson equilibrium iteration. The
force equilibrium between the internal force F  due to tension, and the external force F 
due to drag and effective weight (illustrated in Fig. 10.17) is found for each time step. Iteration
with the Newton-Raphson method continues until an equilibrium solution is found with desired
accuracy.
The Newmark- method for numeric integration uses the assumption of constant acceleration
in each time step. This integration method is unconditionally stable for linear systems. However,
for nonlinear systems the step length the simulations most 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.

TLP modelling
Recall that the TLP is a partly rigid and partly compliant structure. The eigenfrequencies for
in-plane motions must be below frequencies for wave energy and the frequencies for the out-of-
plane 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 1-2 minutes, well above the
range of first order waves with periods of 5-20s. 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 2-4s, and well below the period of the

385
first order waves, see Larsen [40].
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, see Faltinsen [66].
• Higher order components and sum frequencies (2   2     +   ) in waves may give sig-
nificant 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 steady-state oscillations.
• Loads on difference frequencies ( −  ) will give slowly varying wave loads that may give
rise to resonant in-plane motions. Excitation from wind gusts may appear in the same
frequency range and contribute significantly to such motions.
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 30-40s 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.

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 tensioned under all realistic conditions, irrespective of the 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.
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 stress in the riser,
excessive rotation of the ball joint, or end loads may cause damage to the well template or the
riser tensioning system. Other unwanted effects might be mechanical contact between adjacent
risers or between the riser and the platform. Consequences of such incidents must be taken into
considerations, when designing the riser tensioner system, see Larsen [39].
The platform vertical motion is the most important parameter that defines the 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 setdown of the platform is controlled by the tendons. The offset in surge,   , is
equal for the risers and the tendons, but as the risers are longer and have larger radius, their
setdowns are smaller. The relative setdown, ∆, is found as
µ q ¶ µ q ¶
2 2
∆ = ∆ − ∆ =  −  −   −  −  −    2 2 (10.116)

386
 xTLP
DsR

d DsT
lR
lT

Figure 10.18: Setdown due to tendon/riser geometry.

where ∆ and ∆ are the tendon and riser setdown respectively, and  and  are the tendon
and riser lengths. This is illustrated in Fig. 10.18. For  being the water depth, we can calculate
the setdown with known offset, riser and tendon lengths. The relative setdown is small, even
for large water depths and offsets, due to the tendon riser geometry. Hence, the requirement for
heave compensation due to offset is also small. This makes TLP favorable to other floaters which
do not have surge induced setdown, such that their relative difference between the platforms
and the risers is larger.
Other effects having an influence on the stroke are the effect of temperature, tension, pressure,
production tolerances and riser deflections, see Larsen [39]. Note that this calculations on set-
down is specific for the TLP solution and not applicable for other type of vessels.

Transformation of the TLP The motions of the TLP is given as the motion of the -

frame relative to the -frame in the -frame with the position vector position vector r  =
   
[     ]  As the riser motions are expressed in the  -frame, the rotation matrix between
the  - and the -frames is needed. The transformation from - to  -frame is given by the
diagonal matrix
R =  (1 −1 −1)  (10.117)
Since both frames are considered inertial and fixed with a translation in heave direction only, all
axis are parallel, and the -axis are always pointing in the same direction. In a two dimensional
system, only the surge motion of the TLP is of interest as the prescribed motion of the riser.
Hence, the motion in -direction for the TLP described in the -frame is equal to the -position
given in the  -frame. Thus the frames are omitted for the surge position of the TLP,   .

TLP surge modelling Analysis of floater models are usually divided into two groups; sepa-
rated 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 positions dependent

387
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. 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.
In this study we assume that the TLP motions influence the riser behavior, but that the
risers do not affect the TLP motion. Hence, a de-coupled analysis model is applied. The main
purpose of the modelled TLP is to represent the prescribed motion in surge for the top node
of the risers. Since the system model is two-dimensional, only the surge motion for the TLP is
needed. It is modeled as a low frequency harmonic motions
µ ¶
2
  =   sin  +    (10.118)
 

where   is the static TLP offset,   is the amplitude, and   is the period of the TLP
motion.

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 specified
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 displacement at the bottom (both =1
and =1 ), whereas +1 =   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. 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  = m ̈  + c ̇  + k   (10.119)

where f  is the force vector that originates from the TLP motions. m , c and k are
columns in the original riser system matrices that correspond to   , and hence are removed
from the riser system matrices used in the dynamic equation of motion.

Dynamic equation of motion

In the dynamic equation of motion, the mass-damper-spring 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) r̈ + C (r) ṙ + K (r) r = f + f − f   (10.120)

where r is the riser position vector, f is the top tension, and f is the drag forces from
current and riser motions. The superscript  is omitted as the equation of motion is given in the
 -frame. In this equation the fixed and prescribed dofs are removed from the equation, and the

388
 Follows TLP motion

Cylinder

Minimum xmin
Initial payout
Initial payout
zinit Payout
position
variation
+

Piston
Static
position
Maximum
Dynamic payout xmax
stroke
Follows riser motion

Figure 10.19: Stroke parameters of one individual cylinder.

influence from specified dofs is included on the right hand side as a force acting on the system.
The riser position vector is then
£ ¤
r= 2 2 · · ·   · · ·   +1  (10.121)

Remark that this vector correspond to the entire position vector found in Eq. (10.105), but the
prescribed dofs are left out.

Actuator and constraints

The riser tensioner system or heave compensator can be implemented as a hydraulic cylinder
with a piston. Today this setup strive to keep the tension close to constant, compensating for
the stroke parameters listed in Sec. 10.2.3. 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 physical 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 the definitions by Larsen in [39], slightly
modified, and illustrated in Fig. 10.19. The initial position refers to a riser and platform
condition without offset or environmental 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 in a particular condition. 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
forced on the riser is physically constrained with upper and lower boundaries. If this tension is

389
 Cylinder bottom and riser top Payout vs time Payout vs TLP
1214 0.5 0.5
Vertical top position [m]

1213 1 1

Payout [m]
1212 1.5 1.5

1211 2 2

1210 2.5 2.5

0 200 400 0 200 400 0 20 40 60
Time [s] Time [s] TLP position [m]

Figure 10.20: From left: a) The cylinder bottom (—) and the riser top (- -), b) the distance
between them; payout for RIFLEX(—) and Simulink (- -), and c) payout vs TLP position,
RIFLEX(—) and Simulink(- -).

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 by the yield stress for the riser material and chosen such that the 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.

Measurements
The non-compensated initial position, 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, ∆ is found from
 () = 0 − ∆ () (10.122)
Recall that this compensation for set-down is typical for a TLP and noe applicable for vessels in
general. The measured parameter used in the control loop is the payout for each piston, denoted
  . Payout is defined positive when it adds elongation to the riser and negative in the opposite
direction. Minimum payout,  min , means that the piston is as far into the cylinder as possible,
for simplicity assumed zero here. Maximum payout,  max , refers to the position with maximum
free piston length. The initial payout,  0 , is the distance between the lower end of the cylinder
and the initial position. The total dynamic payout is given as
  () =  0 +  () − +1 ()  = 1 2 (10.123)
where +1 is the vertical top position of riser . An example of dynamic payout with TLP
motions is illustrated in Fig. 10.20.

10.2.4 Model Verification

RIFLEX [230] is a commercial FEM program for static and dynamic analysis of slender marine
structures. For the purpose of control system design, a more convenient numerical procedure

390
and software code has been implemented in Matlab/Simulink. The algorithm implemented is
equivalent to what is found in RIFLEX. The motivation behind this section is to validate the riser
model from Section 10.2.3 implemented in Matlab/Simulink and verify that it is appropriate as
a model of the real world used in the simulations (i.e. prosess plant model) and for the purpose
of control system design (i.e. control plant model), despite the simplifications made in the
modelling. In this case, the model should give a good picture of the global geometry of the riser,
with focus on the maximum horizontal displacement and the vertical position of the riser top,
exposed to varying TLP positions and tensions. Features often included in riser analysis, which
are of less importance in this case are:

• Bending stiffness  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, 2-20 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 main importance. Note that
for structural analysis where stress and fatigue is of importance the simplifications listed above
can not be made and more refined models are needed. Note that it is crucial to have a clear
understanding of the purpose of the analysis before simplifications in the model is made.
The 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 ( = 20) are
chosen to increase the resolution and accuracy. Four tests are run:

1. Quasi-static verification with increasing TLP offset.

2. Quasi-static verification with increasing top tension.

3. Dynamic verification with harmonic TLP motions.

4. Dynamic verification with harmonic top tension variations.

Set-up and current profiles

The TLP/riser system is simulated 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 sea floor. 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 Section 10.2.5, this model with the

391
 Current profiles
0

200

400
Water depth [m]

600

800

1000

1200
−1 −0.5 0 0.5 1
Current velocity [m/s]

Figure 10.21: Simulated current profiles from the Ormen Lange field (—), Golf of Mexico (- -)
and a bidirectional shear current (-.).

number of elements varying between 2 and 20 are analyzed and compared to investigate the size
of the error introduced.
Three different geographic areas are chosen to compare the behaviour of the risers with
various profiles. Amongst these are design profiles representing the Ormen Lange field in the
North Sea, Aker [6], loop eddies in the Golf of Mexico (GoM), Nowlin Jr. et al. [212] and
a bidirectional shear current, see Fig. 10.21. The Ormen Lange current profile is close to a
shelf edge. This gives often strong and variable current, and might potentially cause operational
challenges. The velocities are strong all the way down to the sea bed, but even stronger close to
the sea surface due to wind generated current. The data from the Ormen Lange field is extended
from 850m to 1200m water depth and the current between these depths are considered constant
at 0.5m/s. In the current from the GoM a loop eddy is seen to reduce the current at 200-400m
water depth, whereas the velocity increase again for even larger water depths. The bidirectional
shear current is due to typically 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
current, that do not vary with depth.
A selection of the verification tests are given in the following sections. The physical riser
data are found in Table 10.1 below. The TLP data is from the ASM 1200 Larsen et al. [38].
The estimated surge eigenperiod is 136s.

392
Parameter Description Value Dimension

2 Damping coefficient 0.0477 [-]

 Diameter 0.3 [m]
 Water depth 1200 [m]
 Wall thickness 0.015 [m]
 Drag coefficient 1.0 -
 Mass coefficient 2.0 -
 Modulus of elasticity 206 [GPa]
 Yield stress steel 500 [MPa]
 Riser length 1200 [m]
 Tendon length 1166 [m]
 Upper tension limit 2700 [kN]
 Lower tension limit 1200 [kN]
0 Initial payout 0.5 [m]
 Specific weight for steel 7850 [kg/m3 ]
 Specific weight for filling 800 [kg/m3 ]
 Specific weight for sea water 1026 [kg/m3 ]
∆ Initial riser distance 15 [D]

Table 10.1: Riser data for 1200m water depth.

393
 Snapshots Setdown
1200 1200.5

1000 1200
Vertical position [m]

Top position [m]

800 1199.5

600 1199

400 1198.5

200 1198
RIFLEX N = 400
Simulink N = 20
0 1197.5
0 20 40 60 0 20 40 60
Horizontal position [m] TLP position [m]

Figure 10.22: Deflections (a) and setdown (b) for the quasi-static riser model with TLP offset
0-70m and Ormen Lange design current. RIFLEX(— x) and Simulink (- -o).

Quasi-Static verification with increasing TLP offset

The quasi-static 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. 10.22a).
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. 10.22, gave as expected a better correspondence between the
RIFLEX and Simulink calculations than fewer elements, like 5 or 10, see Section 10.2.5.
The setdown is given in Fig. 10.22b). 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.

Quasi-Static verification with increasing top tension

In the second verification of the quasi-static riser model, the tension was increased from 1200kN
to 2700kN in steps of 50kN. All current profiles were tested with TLP positioned in zero and
30m offset,   = {0 30}m. The lower tension limit was the effective weight plus a safety
margin for the structure connection at the sea bed. The upper tension limit was given by a
percentage of the yield stress for steel.
Fig. 10.23 shows the relations between top tension, vertical top position and maximum

394
 Displacement vs top position Displacement vs tension Top position vs tension
25 25 1201
Max displacement [m]

Top position [m]

20 20 1200.5

15 15 1200

10 10 1199.5

5 5 1199
1199 1200 1201 1500 2000 2500 1500 2000 2500
Top position [m] Top tension [kN] Top tension [kN]

Figure 10.23: From left: a) Maximum horizontal displacement vs vertical top position,
 ( ), b) Maximum horizontal displacement vs top tension,  ( ), c) Vertical top
position vs top tension,  ( ). All plots are for the quasi-static riser model exposed to the
Ormen Lange profile. RIFLEX(— x) and Simulink (- -).

horizontal displacement for the Ormen Lange design current. 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.

Verification with dynamically moving TLP

The dynamics of the system and the influence from the current and drag forces were tested
by dynamic surge motions of the TLP with amplitude   = 20m and periods in the range
  = 60 − 300s. The static offset was   = 30m, and the riser tension was kept constant.
Fig. 10.24 shows snapshots of the dynamic riser motion at the Ormen Lange field simulated in
Simulink with   = 120s. The static configuration and the dynamic enveloping curves are
found from RIFLEX.
Top position versus time is plotted in Fig. 10.25a), while Fig. 10.25b) shows the correspon-
dence between the top position and the TLP surge position. The deviation 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,   = 60s, not shown here.

Verification with dynamically varying tension

The fourth and last of the riser model verification tests included harmonic tension variations.
The initial tension was 1950kN, with an amplitude of 750kN 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. 10.26
shows the dynamic variation of the top tension (a) and the top position (b) with period of 120s.
The motion trajectory for vertical position versus tension is seen in Fig. 10.26 c). A transient
period from the static initial condition is seen. The results from simulations with 60s are similar.

395
 Snapshots of TLP moving from right til left Snapshots of TLP moving from left til right
1200 1200

1000 1000
Vertical position [m]

Vertical position [m]

800 800

600 600

400 400

200 200

0 0
0 10 20 30 40 50 0 10 20 30 40 50
Horizontal position [m] Horizontal position [m]

Figure 10.24: 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.

Top node position Top node xz−trajectory

1200 1200

1199.8 1199.8

1199.6 1199.6
Top position [m]

Top position [m]

1199.4 1199.4

1199.2 1199.2

1199 1199

1198.8 1198.8

1198.6 1198.6

1198.4 1198.4

1198.2 1198.2
0 100 200 300 10 20 30 40 50
Time [s] TLP position [m]

Figure 10.25: 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.

396
 Top tension vs time Top position vs time Top position vs top tension
1201 1201
2500
Top tension [kN]

Top position [m]

1200.5 1200.5
2000

1500
1200 1200

0 100 200 300 0 100 200 300 1000 2000 3000

Time [s] Time [s] Top tension [kN]

Figure 10.26: 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.

There were also run tests where the motion of the top node in RIFLEX was 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 [237].

10.2.5 Control plant model analysis

In Section 10.2.3, 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 [230] in Section
10.2.4. 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 an adequate model of the riser process. The model used
in the verifications had 20 elements, and satisfied the desired level of accuracy. As defined in
Section 10.2.3, 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 section we will 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?

397
Analysis input data and set-up
Recall that the riser top is assumed to be at the level of free sea surface, i.e. 1200m above the
sea bed. 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.  =
{2 4 5 8 10 16 20}. This is done to get the nodes at the same height above the sea bed, and
hence more easily compare the horizontal displacements, especially of interest in the quasi-
static analyses. With a water depth of 1200m, the element lengths were respectively 0 =
{600 400 240 150 120 75 60}m for the different models and increasing number of elements.
The Simulink models are hereafter called CPMs. These 7 versions of the model are identical
except for the number of elements, and are compared 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 sea bed 05ms, illustrated in Fig. 10.21.

The test set-up and the simulation results for the RIFLEX model are the same as in Section
10.2.4 and are used in the analysis to illustrate the anticipated correct solution.
Fig. 10.27 shows the static riser configuration with the Ormen Lange current, but 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. 10.27
is used throughout this section, unless otherwise specified.
Only the dynamic results are included in this section. The quasi-static analysis can be found
in Rustad [237].

Dynamically moving TLP

The CPMs were exposed to harmonic TLP motions using the same setup as in Section 10.2.4.
Figs. 10.28 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.10.24a)) 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 ( = {4 5}) capture the main deflection. The
medium order CPMs ( = {8 10}) follow the solution nicely, while the highest order CPMs
( = {16 20}) are hardly seen as they are on the curve for the RIFLEX solution.
The payout in Fig. 10.28 a) shows that the first mode corresponding to the TLP motion is
dominating. The TLP motion and setdown is seen in Fig. 10.29 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. 10.24b) where the TLP moves from left to right. The
payout error
| | = |  −  = | (10.124)
is plotted in Fig. 10.30.

398
 1200
N=2
N=4
N=5
N=8
1000 N=10
N=16
N=20
N=400

800
Vertical position [m]

600

400

200

0
0 2 4 6 8 10 12
Horizontal position [m]

Figure 10.27: Riser configuration without TLP offset for an increasing number of elements
exposed to the Ormen Lange design current.

Piston payout Payout vs TLP

0.5 0.5

1 1
Payout [m]

1.5 1.5
N=2
N=4
2 2 N=5
N=8
N = 10
2.5 2.5 N = 16
N = 20
N = 400
3 3
0 100 200 300 10 20 30 40 50
Time [s] TLP position [m]

Figure 10.28: Payout as a function of time (a) and TLP offset(b) exposed to the Ormen Lange
current.

399
 Cylinder bottom and riser tops
1201.5

1201

1200.5
Vertical top position [m]

1200

1199.5

Cylinder
1199 N=2
N=4
N=5
N=8
1198.5 N=10
N=16
N=20
N=400
1198
0 50 100 150 200 250 300
Time [s]

Figure 10.29: Cylinder bottom (blue line) and riser top positions for TLP motions and Ormen
Lange current.

400
 Payout Error
1.4
N=2
N=4
N=5
1.2
N=8
N = 10
N = 16
1 N = 20

0.8
[m]

0.6

0.4

0.2

−0.2
0 50 100 150 200 250 300
Time [s]

Figure 10.30: Relative payout error for harmonic TLP motions and Ormen Lange current.

For the low order CPMs, the first mode dominates. The error for  = 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. 10.29. The error from first mode shape is also
dominating for  = {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
quasi-static and only the first mode shape was seen. Smaller current velocities gave also smaller
errors.

Analysis of dynamic variation in top tension

Finally, the dynamical CPMs were analyzed with harmonically varying tension. The set-up was
the same as for similar analysis in Section 10.2.4. The simulations shown here are with a tension
period of 120s and a riser exposed to the Ormen Lange current. Fig. 10.31 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 the
recent paragraphs. Fig. 10.31 b) illustrates how an increasing number of elements result in a

401
 Top position vs time Top position vs top tension
1201 1201

1200.8 1200.8

1200.6 1200.6
Top position [m]

1200.4 1200.4
N=2
1200.2 1200.2 N=4
N=5
N=8
1200 1200
N = 10
N = 16
1199.8 1199.8 N = 20
N = 400
1199.6 1199.6
0 100 200 300 1000 1500 2000 2500 3000
Time [s] Top tension [kN]

Figure 10.31: Riser tops as a function of time (a) and tension (b) exposed to the Ormen Lange
current.

better match between the CPM model and RIFLEX.

Fig. 10.32 shows the relative payout error. Maximum errors are seen for the low order
CPMs, with approximately 0.22m for  = 2, and 0.09m for  = 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 due to the nonlinear viscous forces, i.e. the drag forces and difference in relative
velocity when the tension is decreased and increased.

Discussion of the control plant models

In Section 10.2.4, we verified the mathematical model. In this section, we investigated 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 section 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( + 1) × 2( + 1) giving 4( + 1)2 matrix
elements, where  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  = {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
 = {8 10}, have small quasi-static deviations, and follow the dynamics with errors less than
8%. The high order CPMs with  = {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 10.2.

402
 Top Position Error
0.3
N=2
N=4
N=5
0.25 N=8
N = 10
N = 16
N = 20
0.2

0.15
[m]

0.1

0.05

−0.05
0 50 100 150 200 250 300 350 400
Time [s]

Figure 10.32: Relative payout error for a riser exposed to harmonic tension variations with
periods 120s and the Ormen Lange current.

Type  QS error Dyn error Description

FCPM 4-6 5-10% 15-20% Main physics used in control

analysis
ACPM 8-12 2-5%  8% Good model for monitoring of
the risers
PPM ≥ 15  1%  5% High accuracy for the riser
process model

Table 10.2: The categorized results of the CPM analysis.

403
 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  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.

10.2.6 Controller System Design

In Chapter 1, the typical various levels of control used in marine control systems was addressed
. The real time control structure is suggested being divided into low level actuator control, high
level plant control and local optimization. The first of these is the actual physical hardware
system with local controllers, here represented by the heave compensator or riser tensioner
system. The latter two are parts of the control system architecture or software solution. The
local optimizer consists of a guidance block which decides the reference coordinates for the
vertical top position for all risers. For each riser there is a reference block, calculating the
feasible reference trajectory. The plant controller for each riser then calculates the necessary
tension that the corresponding actuator will force on its riser system, according to the reference
trajectory and the actual riser payout.

Controller objective
Today the top tension is kept close to constant by a passive heave compensation system for each
riser
1 = 2  (10.125)
where  is the tension of riser  = 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 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. 10.33 a).
Another strategy is equal payout by connecting all risers to a common frame, proposed
by Huse and Kleiven [124], see Fig. 10.33b). 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

1 = 2 (10.126)

where   is the payout of riser . Further work proposed in Rustad [237] has shown that due
to this tension variation, two equal risers will experience different length due to axial strain
according to
∆
∆ = 0  ∆ = 1 − 2  (10.127)

where ∆ is the length variation due to the difference in top tension ∆ of the two risers.
0 is the untensioned initial riser length. Applying equal payout collision can still occur, but

404
Figure 10.33: Effect of equal tension (a), equal payout (b), and equal effective length (c).

405
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

 1 + 1 =  2 + 2  (10.128)

where  is the length of riser . This means that in contrast to the strategy of equal payout
we also compensate for the axial elasticity due to the tension variation, see Fig. 10.33c). By
introducing this way of controlling the top tension, the risers may be placed with closer spacing
without increasing the risk of collision.
Equation (10.127) is a simplification valid for equal risers. For two or more risers with differ-
ent characteristics with respect to diameter, riser material or filling, a more general expression
is needed. The riser length can be formulated as
 − 0
 =  (0 ) + ∆ ∆ = 0  (10.129)

where  (0 ) is the initial length of riser  with the initial tension 0 . ∆ is the elongation
of riser  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 ∆ in top tension will give the same
increase for all elements along the riser, and (10.129) is a good estimate on the riser elongation
due to tension. Equation (10.128) could be then be rewritten as

 1 + 1 (0 ) + ∆1 =  2 + 2 (0 ) + ∆2  (10.130)

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, three different control objectives based on top measurements are proposed:

1. Equal tension: 1 = 2 .

2. Equal payout:  1 =  2 .

3. Equal effective length:  1 + 1 =  2 + 2 .

Guidance and reference model

The guidance block is used to calculate the reference trajectory or set point for payout for each
riser tensioner system. If one risers position is used as a reference trajectory for the other in the
two-riser-system, the length variation as given in (10.129) is compensated for with

 2 =  1 + ∆  (10.131)

where   is the payout reference of riser . The elongation term could be subtracted on both
sides of (10.131) if  2 is the reference. By this, the variation in riser length elongation is
compensated for.

406
 To provide high performance in tension control between various control modes and/or set-
points, a reference model is introduced to calculate a feasible trajectory for the vertical position
of the top node decided in the guidance block. The following third order filter is demonstrated
appropriate

̈  + 2    ̇  +  2   =  2   (10.132)
1 1
̇  = −   +    (10.133)
 
where   and its derivatives are the desired payout position, velocity and acceleration trajectories.
  is the new reference coordinates in the same frame, and   is the low pass filtered coordinate.
  is the relative damping ratio,   is natural frequency, and  is the cut-off period period of
the low pass filter in (10.133). This provides a smooth transfer between different setpoints.

Plant controller
A PI-controller for each riser is introduced to adjust the top tension according to the reference
trajectory and the actual position of the individual riser
Z
  = −  −    (10.134)
 =   −    = 1 2 (10.135)

where  and  are the positive control gains for riser . The top tension in each riser is
then the pretension, 0 , plus the contribution from the controller  

 = 0 +    = 1 2 (10.136)

10.2.7 Simulations
The riser model was verified with a variety of current profiles. However, in the simulations
included in this section, a current profile with one year return period from the Ormen Lange
field is used as a basis Aker [6]. 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 center-to-center 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 three different control objectives from Section 10.2.6 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 and equal effective length 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.

407
 Snapshots Top tension
1200 2000

1000
1800
800
T1

[kN]
[m]

600 1600
T2
400
1400
200

0 1200
0 2 4 6 8 10 0 200 400 600 800
[m] Time [s]

Relative horizontal distance Payout

15 0.45
Node 3
Node 6 0.5
10
Node 9 0.55
5 0.6 ξ1
[m]
[D]

0.65 ξ
0 2

0.7
−5
0.75

−10 0.8
0 200 400 600 800 0 200 400 600 800
Time [s] Time [s]

Figure 10.34: 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).

Constant equal tension

In the first case, the top tension of the risers are constant and equal to a pretension. From
(10.125) we have
1 = 2 = 0  (10.137)
A medium top tension of 0 = 1800kN is applied, and there is no TLP offset. The incoming
current profile is increased as a second order low-pass 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. 10.34 (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. 10.34,
bottom right).
The TLP is then put in an offset position of 30m, with increasing current, seen top left in
Fig. 10.35. The TLP is not very likely to have an offset in the opposite direction of the surface
current, so offset is 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. 10.35, 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. 10.35. R1 has the largest deflection and hence setdown, which is clearly seen in
the payout plot.

408
 Snapshots Top tension
1200 2000

1000
1800
800
T1
[kN]
[m]

600 1600
T2
400
1400
200

0 1200
0 10 20 30 0 200 400 600 800
[m] Time [s]

Relative horizontal distance Payout

20 0.4
Node 3
15 Node 6 0.5
Node 9
10 0.6
ξ1
[m]
[D]

5 0.7
ξ
2
0 0.8

−5 0.9

−10 1
0 200 400 600 800 0 200 400 600 800
Time [s] Time [s]

Figure 10.35: Incrementing current and 30m TLP offset. Snapshots of the riser configuration
(top left), top tension (top right), relative horizontal distance (bottom left) and payout (bottom
right).

409
 Snapshots Top tension
1200 3000

1000 2800

800
2600 T1
[kN]
[m]

600 T2
2400
400
2200
200

0 2000
0 2 4 6 8 0 200 400 600 800
[m] Time [s]

Relative horizotnal distance Payout

16 0.1
Node 3
14 Node 6 0.12
12 Node 9
0.14
10 ξ1
[m]
[D]

0.16
8 ξ
2
0.18
6

4 0.2

2 0.22
0 200 400 600 800 0 200 400 600 800
Time [s] Time [s]

Figure 10.36: 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).

410
 Increasing the top tension to the upper limit of 0 = 2700 and keeping the TLP in a
no-offset position, no collision occurs at any nodes, seen in the snapshots (top left) and relative
horizontal distance (bottom left) in Fig. 10.36. Keeping the tension at the upper limit, collision
could be avoided for these design 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 0 = 1800 for both risers. A
fully developed current profile for the Ormen Lange field is used.

Equal payout
For the equal payout controller objective, the payout of all risers should be equal, see (10.126).
Using the measured payout of R1 as a reference for R2, we have

 2 () =  1 () (10.138)

Furthermore, a PI-controller for the top tension could be formulated from (10.134)-(10.136)

1 = 01  (10.139)
2 = 02 +  2  (10.140)
Z
 2 = −2 2 − 2 2  (10.141)
2 =  2 −  2  (10.142)

where  2 is the guidance trajectory passed through the third order filter in (10.132)-(10.133).
The results are found in Fig. 10.37, with the top tension shown in the top right figure.
This control algorithm gives equal payout for the risers, seen bottom right in Fig. 10.37.
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 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.

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 (10.126), using (10.127) and (10.129) is simplified to

 2 =  1 + ∆  (10.143)

The controller is the same as for equal payout. The results are seen in Fig. 10.38.
The snapshots found top left show how the mid-position 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. 10.38 we

411
 Snapshots Top tension
1200 2000

1000
1800
800
T1
[kN]
[m]

600 1600
T2
400
1400
200

0 1200
0 5 10 0 200 400 600 800
[m] Time [s]

Relative horizontal distance Payout

10 0.5
ξ
1
ξ2
5 0.6
[m]
[D]

0 0.7

−5 Node 3 0.8
Node 6
Node 9
−10 0.9
0 200 400 600 800 0 200 400 600 800
Time [s] Time [s]

Figure 10.37: Equal payout algorithm and control of R2. Snapshots of the riser configuration
(top left), top tension (top right), relative horizontal distance (bottom left) and payout (bottom
right).

412
 Snapshots Top tension
1200 2000

1000
1800
800
T1
[kN]
[m]

600 1600
T2
400
1400
200

0 1200
0 5 10 15 0 200 400 600 800
[m] Time [s]

Relative horizontal distance Payout

20 0.5
ξ
0.6 1
15
ξ2
0.7
10
0.8
[m]
[D]

5
0.9
0
Node 3 1
−5 Node 6 1.1
Node 9
−10 1.2
0 200 400 600 800 0 200 400 600 800
Time [s] Time [s]

Figure 10.38: 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).

413
 Snapshots Top tension
1200 2000

1000
1800
800
T1
[kN]
[m]

600 1600
T2
400
1400
200

0 1200
0 10 20 30 0 200 400 600 800
[m] Time [s]

Relative horizontal distance Payout

25 0.4
ξ
20 1
ξ2
0.6
15

10
[m]
[D]

0.8
5

0 Node 3 1
−5 Node 6
Node 9
−10 1.2
0 200 400 600 800 0 200 400 600 800
Time [s] Time [s]

Figure 10.39: 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).

414
clearly see that R2 has a larger payout to compensate for riser smaller elongation and achieves
equal effective length.
The same algorithm is applied for 30m TLP offset (Fig. 10.39). As before R2 slides out to
the right with decreasing tension, increasing the horizontal distance between the risers to avoid
collision, see Fig. 10.39 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 2 is seen to stabilize
about 1400kN for the offset case and 1300kN for the no-offset case, to the upper left in Figs.
10.38 and 10.39, 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,

 1 =  2 − ∆  (10.144)

1 = 01 +  1  (10.145)
2 = 02  (10.146)
1 =  1 −  1  (10.147)

The controller is the same as for Section 10.2.7.

Fig. 10.40 shows the snapshots of 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. 10.41. Top tension was seen to decrease compared to the
no-offset case and stabilize about 2650kN and 2400kN with tension differences of 850kN and
600kN, respectively shown in Figs. 10.40 and 10.41. 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 collisions, due to the nonlinear relationship (cf. ∆ ) between the tension and the
deflection.

TLP dynamics
In this case the TLP moves with harmonic motions in surge direction about an offset of 30m.
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 peak-to-peak amplitude 40m.
The controllers are enabled after 400s. Fig. 10.42 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

415
 Snapshots Top tension
1200
2600
1000
2400
800
2200 T
[kN]
1
[m]

600 T
2000 2
400
1800
200
1600
0
0 5 10 0 200 400 600 800
[m] Time [s]

Relative horizontal distance Payout

20 0.2

15 0.3

0.4
10
0.5
[m]
[D]

5
0.6
0
Node 3 0.7
ξ
1
−5 Node 6 0.8
Node 9 ξ
2
−10 0.9
0 200 400 600 800 0 200 400 600 800
Time [s] Time [s]

Figure 10.40: 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).

416
 Snapshots Top tension
1200
2600
1000
2400
800
2200 T1
[kN]
[m]

600 T2
2000
400
1800
200
1600
0
0 10 20 30 0 200 400 600 800
[m] Time [s]

Relative horizontal distance Payout

20

15 0.4

0.5
10
0.6
[m]
[D]

5
0.7
0
Node 3 0.8
ξ
1
−5 Node 6 0.9
Node 9 ξ2
−10 1
0 200 400 600 800 0 200 400 600 800
Time [s] Time [s]

Figure 10.41: 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).

417
 Node 9

50
40

[m]
30
20
10

Node 6

50
40
[m]

30
20
10

Node 3

50 R1
40 R2
[m]

30
20
10
0 100 200 300 400 500 600 700 800 900 1000
Time [s]

Figure 10.42: Dynamic TLP motions and control of R2 enabled after 400s. Horizontal positions
for node 3, 6 and 9 for R1 and R2.

R1 decreased, and the horizontal positions for R2 increased at the same time due to reduced
shielding, not shown here.
The riser slides out to the right when the payout  2 increases, see Fig. 10.43. The relative
horizontal position between the same three nodes is seen in Fig. 10.43 (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. 10.43 and 10.44, 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 between the risers, see Fig.
10.43. 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 peak-to-peak amplitude
about 120-140kN.
When R2 is the reference for R1, its dynamic stroke decreases, giving both R1 and R2 a
dynamic stroke about 1m, see Fig. 10.44. The mean tension level is increased with 400kN for
1 , seen in Fig. 10.44. The dynamic tension variation was about 120-130kN. Increasing the
TLP period to 300s, a more quasi-static 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

418
 Snapshots Top tension
1200 2000
T1
1000 T2
1800
800

[kN]
[m]

600 1600

400
1400
200

0 1200
0 10 20 30 40 50 0 200 400 600 800 1000
[m] Time [s]

Relative horizontal distance Payout

30 0

20 0.5

10 1
[m]
[D]

0 1.5

Node 3 ξ1
−10 Node 6 2
Node 9 ξ2
−20 2.5
0 200 400 600 800 1000 0 200 400 600 800 1000
Time [s] Time [s]

Figure 10.43: 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).

419
 Snapshots Top tension
1200
2600 T1
1000 T2
2400
800

[kN] 2200
[m]

600
2000
400

200 1800

0 1600
0 10 20 30 40 50 0 200 400 600 800 1000
[m] Time [s]

Relative horizontal distance Payout

20 0

0.5
10

1
[m]
[D]

0
1.5

−10 Node 3 ξ1
Node 6 2
Node 9 ξ2
−20 2.5
0 200 400 600 800 1000 0 200 400 600 800 1000
Time [s] Time [s]

Figure 10.44: 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).

420
closer to the static case.

10.2.8 Summary
In this section, controlling the top tension of risers to avoid collision is considered. The concept
of tension legged platforms was introduced. A mathematical model of the system was made was
proposed using the finite element method. Verification of the model was thereafter done with
comparison to RIFLEX. The difference between a process plant model and the various control
plant model was demonstrated. Three different control objectives was proposed and thereafter
investigated in simulations. Automatic control of top tension to achieve equal effective length
for two risers decreased the number of collision most efficiently, both in the static cases with
and without offset, and in the cases with dynamic TLP motions.

421
10.3 Control of Drilling Riser Angles by Dynamic Positioning
of Surface Vessels
10.3.1 Introduction
A dynamically positioned (DP) vessel maintains its position (fixed location or pre-determined
track) exclusively by means of active thrusters. Position keeping means maintaining a desired
position in the horizontal-plane within the normal excursions from the desired position and
heading during the different environmental conditions. For references see Balchen et al. [19],
Grimble and Johnson [100], Fossen [78] and Sørensen et al. [286]. Extensions to include roll and
pitch damping for semi-submersibles are proposed by Sørensen and Strand [288] and [290].
The advantages of fully DP operated vessels are the ability to operate in deep-water, the
flexibility to quickly establish position and leave location, and to start up in higher sea states than
if a mooring system should be connected. In addition, there may be restrictions to deploy anchors
due to already installed subsea structures on the seabed. Appropriate designed DP vessels, with
high quality and capability on thruster/propulsion, power, control and instrumentation systems,
are able to keep position and heading precisely and with satisfactory safety margin, even up to
extreme weather conditions. For deep-water exploration and exploitation of hydrocarbons DP
operated vessels may be the only feasibly solution.
However, drilling operations at water depths smaller than 200 − 300 m have normally been
done with moored semi-submersibles or fixed structures. Motivation for this has been the antici-
pation of lower investment costs and reduced operational risk compared to dynamic positioning.
One reason for the scepticism to apply DP operated vessels for drilling at shallow water depth,
especially performing in hot oil and gas wells, has been the risk for drive-off initiated damages of
the well head, in worst case resulting in a open well. Drive-off is an uncontrolled excursion of the
vessel, which may happen because of a failure in the DP system. The most limiting operational
factor in drilling is the tolerance for riser angle deviation relatively to the well head and at the
top joint. Ideally, the angle should be within ± 2 ◦ . Deviations larger than 5 ◦ − 8 ◦ may be
fatal. Hence, in case of a drive-off the available time windows to decide and execute a controlled
disconnect becomes more limited at shallow water compared to deep-water. On the other hand,
more complex riser dynamics is appearing for increasing water depth. A further step in ensuring
a safe and high performance DP operation is to actively take the riser angle offsets and bending
stresses into account in the DP control strategy.
Here the model-based DP controller presented in Sørensen et al. [291] and [295] accounting
for riser angle offsets in 3 dimensions is shown. Numerical simulations with a drilling semi-
submersible are carried out to demonstrate the performance of the proposed controller. In
Imakita et al. [127] a similar approach was proposed applying neural network.

10.3.2 Mathematical Modelling

Vessel motions
Kinematics The different reference frames used are as presented in Section 7 and illustrated
in Figure 10.45.

• The Earth-fixed reference frame is denoted as the    -frame. Measurements of the

vessel’s position and orientation coordinates are done in this frame relatively to a defined
origin.

422
 Vc
c  X
w V
w

d

(x , y )
XE

( xd , yd )
YR

YE

Figure 10.45: Vessel reference frames.

• The reference-parallel    -frame is also Earth-fixed but rotated to the desired heading
angle   . The origin is translated to the desired  and  position coordinates.

• The body-fixed  -frame is fixed to the vessel body with the -axis positive forwards,
-axis positive to the starboard and -axis positive downwards.

The rotation matrix J gives the linear and angular velocity of the vessel in the body-fixed
frame relative to the Earth-fixed frame
∙ ¸ ∙ ¸∙ ¸
η̇ 1 J1 (η 2 ) 03×3 ν1
η̇ = = = J (η 2 ) ν (10.148)
η̇ 2 03×3 J2 (η 2 ) ν2

The vectors defining the Earth-fixed vessel position and orientation, and the body-fixed trans-
lation and rotation velocities are given by

η 1 = [  ]  (10.149a)

η 2 = [  ]  (10.149b)

ν 1 = [  ]  (10.149c)

ν 2 = [  ]  (10.149d)

Nonlinear low-frequency vessel model The nonlinear six degrees of freedom (DOFs)
body-fixed coupled equations of the low-frequency (LF) motions in surge, sway, heave, roll,
pitch and yaw are written (Section 7.3)

Mν̇ + C (ν)ν + C (ν  )ν  +D (ν    )ν  +Dν + G(η)η = τ  +τ 2 +τ   (10.150)

The effect of current is included in the relative velocity vector according to

ν  = [ −   −     ]  (10.151)

423
The current components are defined as

 =  cos (  − )  (10.152)
 =  sin(  − ) (10.153)

where  and   are the surface current velocity and direction respectively, see Figure 10.45.
C (ν) and C (ν  ) are the skew-symmetric Coriolis and centripetal matrices of the rigid body
and potential induced added mass part of the current load. D  (ν     ) is the 6-dimensional
nonlinear damping vector. The relative drag angle is found from the following relation   =
 tan(   ) M is the 6×6 dimensional system inertial matrix including asymptotic added mass
values for  → 0, where  is the wave frequency. D is the 6×6 dimensional strictly positive linear
damping matrix caused by linear wave drift damping and laminar skin friction. G is the 6 × 6
dimensional generalized restoring coefficient matrix caused by buoyancy and gravitation. τ 
is the 6-dimensional control vector consisting of forces and moments produced by the thruster
system. τ  and τ 2 are the 6-dimensional wind and second order wave load vectors.

Linear wave-frequency model The coupled equations of wave-frequency (WF) motions in

surge, sway, heave, roll, pitch and yaw are assumed to be linear, and can in the reference-parallel
frame be formulated as (Section 7.3)

M()ν̇  +D ()ν  +Gη  = τ 1  (10.154)

Assuming small roll and pitch motions the Earth-fixed motion vector can be found by

η̇  = J(  )ν   (10.155)

The 6-dimenaional WF motion vector is defined as:

£ ¤
η  = 1  2  3  4  5  6  (10.156)

τ 1 is the 6-dimensional excitation vector, which will be modified for varying vessel heading
relative to the incident wave direction. M() is the 6 × 6 wave-frequency system inertia matrix
containing frequency dependent added mass coefficients in addition to the vessel mass and
moment of inertia. D () is the 6 × 6 wave radiation (potential) damping matrix. G is the 6 × 6
linearized restoring coefficient matrix.

Drilling riser mechanics

Basic concepts A drilling riser behaves like a tensioned beam. The steel pipe provides some
bending stiffness, but it is the geometric stiffness due to the effective tension distribution along
the riser that gives the main contribution to lateral resistance against static and dynamic forces
acting perpendicular to the beam axis. A key feature of a drilling riser is that the top end
tension is kept close to constant under influence from floater motions by a heave compensating
system at the upper end. The top tension must be sufficiently high to prevent the riser from
global buckling, meaning that the effective tension must have a positive value at the lower end
of the riser.
Some basic features of riser mechanics can for the 2-dimensional static case be illustrated
with reference to Figure 10.46. The riser is seen to have an offset position of the upper end

424
 Top
tension Tt

Vessel
motion

Vessel Top
offset angle
t

Current
forces

Bottom
angle
b

Bottom
tension
Tb

Figure 10.46: Riser mechanics.

relative to the lower end. The angles between the riser and a vertical line at the top and bottom
are denoted  and   respectively. These angles are influenced by the tension at the riser ends,
the offset and the current forces. Since both ends of a drilling riser are moment free, one can
easily establish an equation that links most of these parameters
Z 
 sin  +  sin  =  () cos ()  (10.157)


where  () is the current force per unit length for a given position along the riser, and () is
the angle of the riser at the same position.
A successful drilling operation requires that both upper and lower riser angles are within
certain limits. This can in principle be achieved by proper selection of the top tension and the
vessel offset, where the latter is controlled by the dynamic positioning system. However, the
top tension can for a given installation not exceed a predefined maximum level, meaning that
successful angle control may become impossible under some current conditions. The objective
of the dynamic position system is hence to keep the riser angles within the required limits for
drilling operations for the longest possible period of time by proper selection of the desired vessel
position coordinates.
Compared to the 2-dimensional case as presented in Sørensen et al. [291], the 3-dimensional
case is somewhat more complicated, and is illustrated in Figure 10.47. The current direction may
vary through the water column, meaning that riser deflections will not take place in one well-
defined plane. Rotation at riser ends must hence be defined by the direction of the inclination
(  and   on Figure 10.47) and its magnitude ( and  on Figure 10.47).

425
 X0 X
Y0
X’’
t
Y
Y’’ t

b
b
X’

Y’
Z

Figure 10.47: Riser kinematics.

Finite element modelling For the purpose of computing stresses in the riser, a quite
refined Finite Element Model (FEM) is generally required, see Engseth et al. [65]. However,
for the purpose of calculating top and bottom riser angles a less detailed model is necessary. A
convergence study has indicated that 10 elements predicts the angles within a few percent of
those obtained by application of several hundreds of elements, even for risers at water depths of
1000 m.
Accordingly, we presently apply a model as illustrated by Figure 10.48 with 10 elements.
In reality, the model is 3-dimensionional while only a 2-dimensional projection is shown in the
figure. The local -axis for each element passes through the two nodal points by which the
element is defined. For each element a stiffness matrix is defined based on the local coordinate
system. This stiffness matrix consists of two parts. The first part represents the elastic stiffness,
k , and the second represents the so-called geometric stiffness, k . The latter is subsequently
multiplied by the effective tension in the riser which acts at each particular element,  , where
the subscript  denotes element number . The resulting total stiffness matrix becomes for
element 
k = [k + k ]  (10.158)
The 6 kinematic components (i.e. 3 displacements and 3 rotations) at each end of the local
elements (i.e. a total of 12 components for each element) are referred to the local coordinate
system. All the 12 kinematic components for each element (number ) are collected in a vector

426
v . The elastic stiffness matrix is here expressed on the form
⎡ 12 6 ⎤
3
0 2
0 − 12 3
0 6
2
0
⎢ 0 12
0 − 6
0 − 12
0 − 62 ⎥
⎢ 6 3 2 3 ⎥
⎢ 2 0 4
0 6
0 2
0 ⎥
⎢    2  ⎥
⎢ 0 − 62 0 4
0 6
0 2 ⎥
⎢
k =  ⎢ 12  2  ⎥ (10.159)
6 12
− 62 ⎥
⎢ − 3 0 2
0 3
0 0 ⎥
⎢ 0 12
− 3 0 6
0 12
0 6 ⎥
⎢ 2 2 2 ⎥
⎣ 62 0 2
0 − 6
0 4
0 ⎦
  2 
0 − 62 0 2
 0 6
2 0 4


where  is the modulus of elasticity (Young’s modulus),  is the moment of inertia ( will then
be the bending stiffness), and  is the length of each element. One should notice that (10.159)
is only of dimension 8 × 8, and not of dimension 12 × 12 as explained above. This is due to the
disregard of vertical displacements at the local nodes 1 and 2 of each element, and the same
applies to the rotation about the beam axis (torsion). The reason for this simplification is that
these DOFs are not required for the studied case. Accordingly, a total of 4 DOFs are removed
for each element, which reduces the total number from 12 to 8. The 8 × 8 geometric stiffness
matrix is correspondingly expressed as
⎡ ⎤
36 0 3 0 −36 0 3 0
⎢ 0 36 0 −3 0 −36 0 −3 ⎥
⎢ ⎥
⎢ 3 0 4 2 0 −3 0 −2 0 ⎥
⎢ ⎥
 ⎢⎢ 0 −3 0 42 0 3 0 −2 ⎥ ⎥
 k = (10.160)
30 ⎢
⎢ −36 0 −3 0 36 0 −3 0 ⎥
⎥
⎢ 0 −36 0 3 0 36 0 3 ⎥
⎢ ⎥
⎣ 3 0 − 2 0 −3 0 4 2 0 ⎦
0 −3 0 −2 0 3 0 42

The kinematic components entering the assembled dynamic equilibrium equation for the
whole riser are referred to the global coordinate system (designated by   in Figure 10.48).
A transformation matrix T is accordingly introduced for each element. The corresponding
kinematic components in the global system are for element number  expressed as

r = T v  (10.161)

These sub-vectors are finally assembled into a total riser displacement vector (containing both
displacements and rotations), designated by r. Note that the dimension of this vector is 4 times
the number of nodes, rather than the number of elements. This is due to neighboring elements
sharing common nodes.
Subsequently, a transformation of each element stiffness matrix from the local to the global
 -frame is performed by application of the transformation matrix T in (10.161) above.
The corresponding element stiffness matrix for element number  in the global system is then
expressed by a congruence transformation as

K = T k T  (10.162)
∙  ¸
K11 K12
=  (10.163)
K21 K22

427
All the element matrices referred to the global coordinate frame are subsequently assembled to
form the system stiffness matrix, K. As explained below, this matrix depends on the positions
of all the nodes. A more explicit expression for the system matrix is obtained by introducing the
following notation, K , which designates sub-matrix  for element number . For each element,
 and  run from 1 to 2 and refer to the local nodes of the elements. Hence, there are 4 such
sub-matrices: K11  K12  K21 and K22 which are all of dimension 6 × 6 or more correctly here
for the present modelling, 4 × 4. The first of these sub-matrices represents the stiffness terms
for local node number 1 containing coupling terms between the 4 DOFs. The last sub-matrix
represents the stiffness terms for the local node number 2. The two remaining sub-matrices K12
and K21 correspond to the stiffness coupling terms between the two local nodes. The global
system stiffness matrix will then take the form
⎡ 1 ⎤
K11 K112
⎢ K121 K122 + K211 K212 ⎥
⎢ ⎥
⎢ K 2 K 2 + K 3  ⎥
K=⎢ ⎢ 21 22 11 ⎥ (10.164)
: ⎥
⎢ ⎥
⎣ K −1 + K K ⎦
22 11 12
K
21 K
22

If we assume that node number 1 is located at the top of the riser, the two first DOFs correspond
to the horizontal displacements of the surface floater (− and − coordinates in the global
system). Furthermore, these correspond to the first two columns and rows of the system matrix
K. As observed from (10.164) there are at maximum 12 non-zero (here 8) elements in each of
these rows or columns.
The boundary conditions at the upper and lower ends of the riser are subsequently intro-
duced. The number of restrained kinematic components eliminates the equations corresponding
to fixed and prescribed DOFs, which reduces the dimension of the global displacement vector.
Taking the upper end as an example, the prescribed − and − displacements are taken into
account by removing the first columns and rows of the system stiffness matrix (10.164). The
first two columns of this matrix is stored in a separate matrix for later applications (see below)
in connection with the prescribed vessel motions. The upper 2 × 2 sub-matrix is then removed,
and the same applies to all the terms which correspond to the fixed DOFs at the last node of
the model.
The resulting matrix will be of dimension  × 2 (surge and sway DOFs are prescribed),
where  is the number of DOFs which are not fixed or prescribed. As shown later in the text
(10.183b) this matrix is referred to as K  Correspondingly, the main part of the remaining
stiffness matrix is now of dimension  ×  . This is referred as K in (10.183b).
The load vector for each element is also referred to the local coordinate frame initially.
The force acting at each node of the element is obtained by integration of the force intensity
multiplied by a proper “shape function”. The static force intensity is the quadratic drag force
based on the current velocity component, which is normal to the local element axis. The local
force vector at each node is subsequently transformed to the global system by an expression of
the same type as in (10.161). The local force vector now replaces the local displacement vector
in this equation, and the force vector referred to the global system replaces the displacement
vector in the global system.
The transformation matrix and accordingly the stiffness matrix and load vector as referred to
the global coordinate frame are functions of the nodal coordinates. These are in turn defined by

428
 Local coordinate
systems

Figure 10.48: Two-dimensional projection of three-dimensional FEM riser model.

the initial position and the subsequent total displacement of each node. It has been found that
a two-step iteration gives accurate results: First the static riser displacements due to a given
current profile and a given top end displacement are computed based on stiffness matrices and
load vectors referring to a vertical riser. The new static configuration is subsequently applied for
calculation of the updated transformation matrices. Based on these the system stiffness matrix
and load vector is also updated. Finally, a more accurate static configuration is computed by
means of these new matrices.

Example 10.7 We consider a simplified two-dimensional example in order to illustrate the

structure of the element and system matrices. Obviously, the resulting model is a gross oversim-
plification and does not reflect the physics of the riser behavior properly. Still, it may hopefully
give useful insight to the various matrices and their properties. Specifically, the riser proper-
ties are given by the following data: Top tension is 1 = 2500  , tension at top of second
element is 2 = 1200  , diameter is 05 , wall thickness  = 25, modulus of elasticity
is 21 × 108 2 , water depth is 1000  (which implies an element length equal to 500 ).
The resulting moment of inertia  = 3  = 122656250 × 10−3 4  where  designates the radius
of the riser pipe.
The elastic stiffness matrix for each of the two elements are now only of dimension 4 × 4
since we retain only the in-plane degrees of freedom (i.e. one rotation and one translation at
each node). For both elements, it is accordingly expressed as
⎡ 12 ⎤
3
− 62 − 12
3
− 62
⎢ − 62 4 6 2 ⎥
k =  ⎢⎣ − 3

12

6
2
12

6 ⎦
⎥ (10.165)
 2 3 2
− 62 2

6
2
4


429
and the resulting numerical values become
⎡ ⎤
247 − 1 −6182 −247 − 1 −6182
⎢ −6182 2060625 6182 1030313 ⎥
k = ⎢
⎣ −247 − 1
⎥
6182 2473 − 1 6182 ⎦
−6182 1030313 6182 2060625

The unit of element k (1 1) is 

 . The units of the other elements follow by scaling with respect
to  as seen in (10.165). The geometric stiffness matrices are different for the two elements since
the tension in the riser varies from top to bottom. However, the form of this matrix is the same
for both elements, i.e. ⎡ ⎤
36 −3 −36 −3
 ⎢⎢ −3 4
2 3 −2 ⎥ ⎥
 k = (10.166)
30 ⎣ −36 3 36 3 ⎦
−3 −2 3 42
The numerical values of this matrix for element 1 (which corresponds to the upper half-length of
the riser) becomes equal to
⎡ ⎤
600 −25000 −600 −25000
⎢ −25000 16666667 25000 −4166667 ⎥
1 k = ⎢
⎣ −600
⎥
25000 600 25000 ⎦
−25000 −4166667 25000 16666667

and for the second element all the terms in the matrix are reduced in proportion to the tension
in the element, which implies that
⎡ ⎤
288 −12000 −288 −12000
⎢ −12000 8000000 12000 −2000000 ⎥
2 k = ⎢⎣ −288
⎥
12000 288 12000 ⎦
−12000 −2000000 12000 8000000

The sum of the elastic and geometric stiffness matrix for the first and second elements hence
become

k1 = k + 1 k  (10.167)
k2 = k + 2 k  (10.168)

As expected the geometric stiffness is much larger than the elastic stiffness. This is dependent of
the length of each element. These two element matrices are subsequently added into the system
matrix. For a deformed riser, non-trivial transformation matrices must first be applied. To
establish the stiffness matrix, we consider initially the riser to have a straight and inclined shape
between the seabed and with a relative lateral offset of 80m at the surface. Since the riser is
undeformed (i.e. forms a straight line), the transformation matrices are identical for the top
and bottom elements. The 4 × 4 transformation matrix T for  = 1 to 2 is written
∙ ¸
T 02×2
T =  (10.169)
02×2 T

430
where ∙ ¸
987 0
T10 = T20 =  (10.170)
0 10
Using (10.162) the system matrix is of dimension 6 × 6 and becomes
⎡ ⎤
6022 −256134 −6022 −256134 0 0
⎢ −256134 168727292 256134 −40636354 0 0 ⎥
⎢ ⎥
⎢ −6022 256134 8926 129976 −2904 −126158 ⎥
K =⎢ ⎢ −256134 −40636354
⎥ (10.171)
⎢ 129976 250787917 126158 −18969688 ⎥
⎥
⎣ 0 0 −2904 126158 2904 126158 ⎦
0 0 −126158 −18969688 126158 82060625

Subsequently, the columns and rows corresponding to the prescribed degrees of freedom (i.e. one
translation degree of freedom at the top corresponding to the surge motion of the vessel, and one
fixed translation at the seabed) are removed. These correspond to the first row/column and the
next to the last row/column. The final stiffness matrix which corresponds to the non-prescribed
degrees-of-freedom is now of dimension 4 × 4 and is accordingly expressed as
⎡ ⎤
168727292 256134 −40636354 0
⎢ 256134 8926 129976 −126158 ⎥
K =⎢ ⎣ −40636354 129976 250787917 −18969688 ⎦ 
⎥

0 −126158 −18969688 82060625

The vector which contains the coupling stiffness terms for the top end prescribed (non-zero)
motion becomes £ ¤
K = −256134 −6022 −256134 0 
This vector corresponds to part of the first column of (10.171), which was removed from the
original 6 × 6 stiffness matrix.

Low-frequency model Considering LF riser motion, the driving excitation mechanism is

due to forced motions caused by the surface vessel at the upper end of the riser. The direct
hydrodynamic loading from waves on the riser can then be neglected, but forces from current
and riser motions must be considered. These forces are proportional to the relative velocity
squared and can hence not be separated into damping and driving terms. Since LF motions can
become large, it is advantageous to take into account that geometrical changes will influence the
system matrices. The dynamic equilibrium equation can then be written as

M(r)r̈ + C (r)ṙ + K(r)r = τ  (r) + τ  (u  r ṙ) (10.172)

where the nodal displacement vector is designated by r. The mass matrix M, structural damping
matrix C and stiffness matrix K are all considered functions of the deformed configuration
found from the displacements. Note that the mass matrix must include anisotropic added mass
terms. Two loading terms τ  and   are present, one containing contributions from
LF vessel motions and the other with loads from the current velocity vector u and the riser
velocity vector. The surface current velocity  will enter the first element of u . Both terms will
become functions of the displacements if LF motions are large. A generally applied simplification
of (10.172) is to neglect the two dynamic terms related to inertia and damping forces. The model

431
is then said to be quasi-static, meaning that the response model is static, but external loads are
calculated by taking the riser motions into account. Such models will yield acceptable accuracy
for shallow water applications, but dynamic effects will become more important in deep-water
since the lowest eigenfrequency of the riser may approach the LF regime. A further simplification
of (10.172) is to neglect the influence from riser velocity in the load due to the current. This
type of model has been applied in the present study.

Wave-frequency model In the WF regime, the direct wave forces on the riser must be
included. This loading vector   contains a quadratic drag-term containing the relative
velocity between the riser pipe and the surrounding fluid. For this case, the dynamic equilibrium
equation accordingly is expressed as

M(r)r̈ + C (r)ṙ + K(r)r = τ   (r) + τ  (u  u̇ r ṙ) (10.173)

Again it is seen that all system matrices in principle should be functions of the displacements.
The load terms are as for the LF model separated into one term from vessel motions (WF
motions only in this case) and a hydrodynamic load term. These forces must be found from a
flow velocity profile u with contributions from current and waves, ant the wave induced flow
accelerations u̇.
Two simplifications are frequently seen: The system matrices may be kept constant and
represent the condition at static equilibrium. The model is then linear, but will still have
nonlinear drag forces. The next simplification is to linearize the drag forces and split them
into a damping and an exciting term. The linearized drag damping effect can subsequently be
transferred to the left-hand side of the equilibrium equation and added to the pure structural
damping term. This simplifies the analysis, and allows e.g. frequency domain analysis methods
to be applied. However, linearizations also implies loss of accuracy in the numerical model. Note
that for the WF case one should always apply a dynamic model since the eigenfrequencies of
the riser in almost all cases will be found within the WF regime.

10.3.3 High Level Positioning and Riser Angle Controller

The horizontal-plane positioning controller consists of feedback and feedforward controller terms.
The linear positioning control law including roll and pitch damping proposed in Section 8.3 is
extended to also consider riser angle offsets. Motivated by the fact that the positioning controller
has to reflect the nonlinear kinematics given in (10.148), the hybrid positioning and riser angle
controller will be nonlinear. Applying a nonlinear formulation does also open up for further
refinement of the control law.

Nonlinear horizontal-plane PID control law

The nonlinear horizontal-plane positioning feedback controller of PID type is formulated as

τ   = −R K e − R K3 f (e) − K e−R K z (10.174)

432
where e is the 3-dimensional position and heading deviation vector, ν e is the 3-dimensional
velocity deviation vector, and z is the 3-dimensional integrator states defined as

e = [1 2 3 ] = R (  )(η̂ − η  ) (10.175)

e = ν̂ − R (  )η̇ 
ν (10.176)
ż = η̂ − η   (10.177)

R = R( −   ) , R (  )R() (10.178)

A third order stiffness term is also included for increasing the performance and robustness of
the control system
£ ¤
f (e) = 31 32 33  (10.179)
An advantage of this is the possibility to reduce the first order proportional gain matrix, resulting
in reduced dynamic thruster action for smaller position and heading deviations. Moreover, the
third order term will make the thrusters work more aggressive for larger deviations. η  is the
3-dimensional vector defining the desired Earth-fixed position and heading coordinates. η̂ and
ν̂ are the estimated position and velocity vectors computed by a nonlinear observer formulated
as in Fossen and Strand [83]. K  K3  K and K are the 3 × 3 non-negative controller gain
matrices found by appropriate control synthesis methods.
One should notice that the integral action may be represented in Earth-fixed axes, in order
to increase performance of the positioning controller for large rotating manoeuvres.

Resulting control law

The resulting positioning control law is written

τ  = τ   +τ   +τ   (10.180)

where the wind feedforward control law and the roll-pitch control law are as formulated in
Section 8.3. The thruster allocation is as presented in Section 8.3.5.

10.3.4 Local Optimization: Optimal Setpoint Chasing

With reference to Figure 1.1 in Chapter 1 the optimal setpoint chasing will be on the local
optimization level defining setpoints to the high level controller.

Reference model
A reference model is used for obtaining smooth transitions in the chasing of the optimal position
setpoint minimizing the upper and lower riser angles,  and  , respectively. Let

η  = [    ]  (10.181)

define the final Earth-fixed vector position and heading setpoint vector. This is input to the
nonlinear third order reference model as presented in Section 8.3 and is given as

(xref  xed  vde  aed ) = f (η   xref  xed  vde  aed ;) (10.182)

This model produces a smooth desired acceleration, velocity and position reference that are
inputs to the positioning control law (10.180).

433
Riser control plant model
For simplicity it is assumed that all the natural periods for the riser are located away from the
LF range. Thus the dynamic damping and inertia terms can be neglected. This leaves only
the stiffness term in (10.172). Appropriate criteria that include dynamics and tilt and bearing
are currently subject for further research. Hence, it is assumed that the increment in riser
displacement vector ∆r is related to the in-plane increment in surface vessel position vector
∆r designated by

K ∆r = −K ∆r  (10.183a)

m
∆r = −K−1
 K ∆r  (10.183b)

K is the  ×  incremental riser stiffness matrix at the present instantaneous vessel
position corresponding to the given water current profile. K is the  × 2 incremental
stiffness matrix coupling the in-plane vessel motion (i.e. surge) to the remaining DOFs. Both
matrices are obtained as sub-matrices of the system stiffness matrix given in (10.164).
The increments of top and bottom riser angles are expressed in terms of their - and -
components. Typically, for the top angle these correspond to components 1 and 2 of the vector
in (10.164). For the bottom angle, components number  − 1 and  in the response
vector are relevant. These are accordingly expressed as
¡ ¢
∆ = − K−1  K 1 ∆ = − ∆  (10.184a)
¡ −1 ¢
∆ = − K K 2 ∆ = − ∆  (10.184b)
¡ −1 ¢
∆ = − K K  −1 ∆ = − ∆  (10.184c)
¡ −1 ¢
∆ = − K K  ∆ = − ∆  (10.184d)

where the scalar constants  and  are computed on-line.

Optimal criteria
A quadratic loss function  with the scalar weighting factors  and  is introduced
h i h i
 =  (∆ +  )2 + (∆ +  )2 +  (∆ +  )2 + (∆ +  )2  (10.185)

By solving the equation where the partial derivatives of (10.185) with respect to the - and -
components of the vessel increment are set equal to zero, the optimal direction for the increment
of the vessel position is derived as
µ ¶
∆
 = tan  (10.186)
∆

where ¡ ¢
∆ =  2 +  2 ∗ (   +    )  (10.187)
and ¡ ¢
∆ =  2 +  2 ∗ (   +    )  (10.188)

434
For the optimal direction, the corresponding optimal vessel incremental position is computed as

∗    cos( ) +    sin( ) +    cos( ) +    sin( )
∆ = 
 2 cos2 ( ) +  2 sin2 ( ) +  2 cos2 ( ) +  2 sin2 ( )
(10.189)
Then the - and -components of the optimal incremental vessel position vector are expressed
as
∗
∆ = ∆ cos( ) (10.190a)
∗
∆ = ∆ sin( ) (10.190b)

Finally, in the general 3-dimensional case the updated Earth-fixed vector position and heading
setpoint vector becomes

 ∗ =   + ∆
∗
[cos( ) sin( ) 0]  (10.191)

It is worthwhile to consider the special case that the current, wave and vessel motion takes place
in a single vertical plane which passes through the x-axis. The optimal angle then obviously
becomes zero, and the optimal vessel incremental position in the direction of the -axis simplifies
to:
∗    +   
∆ =  (10.192)
 2 +  2
Here, it is implicitly understood that the angles and coefficients refer to the in-plane quantities.

Weighting criteria
A key issue in the present algorithm is the relative weighting of the top and bottom riser angles.
This weighting must reflect the maximum permissible limits for these angles. The permissible
bottom angle is typically much smaller than the permissible top angle. This implies that the
weighting factor for the lower angle should be much larger than for the upper angle. However,
dynamic effects are different for the two angles. In particular, the dynamic response of the top
angle due to wave-induced motion of the vessel and direct hydrodynamic loading on the riser is
generally much larger than for the bottom angle. This acts in the opposite direction, i.e. the
weighting factor for the top angle needs to be increased at the cost of the weighting factor for
the bottom angle. A trade-off between the various considerations is hence required.
The present algorithm must also be supplemented with a maximum permissible radius of
vessel motion for given values of the current magnitude and sea state parameters. If the ves-
sel approaches the boundary of the permissible region, further increase of the offset must be
restricted.

10.3.5 Numerical Simulations

In this Section a simulation study of a dynamic positioned semi-submersible conduction offshore
drilling operations is carried out to demonstrate the effect of the setpoint chasing strategy. The
semi-submersible, see Figure 10.49, is equipped with 4 azimuthing thrusters each able to produce
1000 kN located at the four corners at the two pontoons.
The operational draught is equal to 24 m, the vessel mass at operational draught is 45000
ton, the length is 110 m, and the breadth is 75 m. Radius of gyration in roll is 30 m, in pitch

435
Figure 10.49: Semi-submersible for oil and gas drilling.

436
 Water Depth: Surface (0 m) Middle (500 m) Bottom (1000 m)
Simulation 1   30 ◦ 075  30 ◦ 015  30 ◦
Simulation 2   30 ◦ 075  −150 ◦ 015  −150 ◦

Table 10.3: Current profile.

Current velocity [m/sec]
1.22

1.2

1.18

1.16

1.14

1.12

1.1

1.08

1.06

1.04
0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [sec]

Figure 10.50: Surface current velocity.

equal to 33 m and in yaw 38 m. The undamped resonance periods in roll and pitch are found
to be equal to 55 sec and 60 sec, respectively.The riser length used in the simulations is equal
to 1000 m. The riser radius is 025 m, the riser wall thickness is 0025 m and the modulus of
elasticity (Young’s modulus) is  = 21 × 108 Pa. The top tension is 2500 Pa, and the tension
of the lower part is 1200 Pa. In the simulation the riser is divided into only ten elements.
The positioning performance subject to two different current profiles will be considered, see
Table 10.3.
A random walk is used to generate the surface current velocity, see Figure 10.50. It is
furthermore assumed that the sea current velocity on the middle of the water column is 75%
and at the sea bottom 15% of the surface current velocity.
The significant wave height is set equal to 7 m with peak period equal to 14 sec and direction
equal to 20 ◦  The mean wind velocity is set equal 15 m s with mean direction equal to 20 ◦  The
chosen weighting factors for the two simulations are given in Table 10.4.
Initially the semi-submersible is dynamically positioned over the well head with zero (field
zero point) as the constant setpoint. After 1000 sec the optimal setpoint chasing is activated.
Figure 10.51 shows the LF position, reference and setpoint coordinates in North and East
axes as function of the time for the two simulation cases. In simulation 1 with uniform current
profile and with highest weight on the bottom riser angle, the optimal setpoint chasing forces

Top Angle,  Bottom Angle, 

Simulation 1 1 5
Simulation 2 1 1

Table 10.4: Weighting factors.

437
 North: Setpoint (dotted), Reference (dashdot) and LF position (solid) [m] North: Setpoint (dotted), Reference (dashdot) and LF position (solid) [m]
2 12

0
10

−2

8
−4

−6
6

−8

4
−10

−12
2

−14

0
−16

−18 −2
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [sec] time [sec]

East: Setpoint (dotted), Reference (dashdot) and LF position (solid) [m] East: Setpoint (dotted), Reference (dashdot) and LF position (solid) [m]
1 6

0
5

−1
4
−2

3
−3

−4 2

−5
1

−6
0
−7

−1
−8

−9 −2
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [sec] time [sec]

(a) Simulation 1 (b) Simulation 2

Figure 10.51: North and East positions for a dynamically positioned semi-submersible, where
the optimal setpoint chasing is activated at 1000 sec. Simulation 1 is shown in left column (a)
and Simulation 2 in right (b).

438
 Heading: Setpoint (dotted), Reference (dashdot) and LF position (solid) [deg] Heading: Setpoint (dotted), Reference (dashdot) and LF position (solid) [deg]
0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0

−0.1 −0.1

−0.2 −0.2

−0.3 −0.3

−0.4 −0.4

−0.5 −0.5
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [sec] time [sec]

(a) Simulation 1 (b) Simulation 2

Figure 10.52: Heading angle for a dynamically positioned semi-submersible, where the optimal
setpoint chasing is activated at 1000 sec. Simulation 1 is marked with (a) and simulation 2 is
marked with (b).
Upper (dotted) and lower (solid) riser angle [deg] Upper (dotted) and lower (solid) riser angle [deg]
4 1.2

3 1

0.8
2

0.6
1

0.4

0
0.2

−1
0

−2 −0.2

−3 −0.4
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [sec] time [sec]

(a) Simulation 1 (b) Simulation 2

Figure 10.53: Lower  (solid) and upper  (dashed) riser angles, where the optimal setpoint
chasing is activated at 1000 sec. Simulation 1 is marked with (a) and simulation 2 is marked
with (b).

the vessel about 16 − 18 m to the South and 8 − 9 m to the West. In simulation 2, with 180 ◦
change in the current direction from the middle to the bottom of the water column, and with
equal weight on the bottom and top riser angles, the optimal setpoint chasing forces the vessel
about 9 − 10 m to the North and 4 − 5 m to the East.
In Figure 10.52 the corresponding heading is shown for both simulations. The desired head-
ing is set to 0 ◦ in both simulations. It can be seen that the heading control performance is
satisfactory.
The effect of the optimal setpoint chasing on the riser angle offsets is clearly shown in Figure
10.53. In simulation 1, the lower riser angle decreases from 16 ◦ down to 04 ◦ , while the upper
riser angle increases from 25 ◦ up to 35 ◦ . Since the lower riser angle offset is given higher
weight compared to the upper riser angle, the opposite effect happens here. In simulation 2, the
lower riser angle decreases from 10 ◦ down to 02 ◦ , while the upper riser angle changes from
015 ◦ down to −04 ◦ . Here, the weight on the lower riser angle is equal to the upper riser angle.

439
 0

−200
Profile [m]

−400

−600 −15
−10

−800 −5
0

−1000 5

−8 10
−6 −4 −2 0 15
2 4 North [m]
6 8
East [m]

Figure 10.54: Snapshots of riser profile as function of time for uniform current profile, simulation
1, where the optimal setpoint chasing is activated at 1000 sec.

Hence, both angles are given equal priority. In an industrial system dynamic weights based on
type of operation and magnitudes of the riser angle offsets have to be implemented.
In Figures 10.54 and 10.55 snapshots of the riser profile in 3-dimensions as function of time
for simulation 1 and simulation 2 are shown. In the beginning the accumulation of riser profiles
are found to be about the first desired setpoint defined to be over the field-zero-point equal
to (0 0) in North-East coordinates. Then after 1000 sec, we observe that new accumulation
areas appears corresponding to the new desired setpoint found by the optimal setpoint chasing
algorithm. In Figure 10.56 the PID controller input in surge, sway and yaw are shown.

440
 0

−200
Profile [m]

−400

−10
−600

−5
−800
0

−1000
5
−5 −4 −3 −2 −1 0 1 2 10 North [m]
3 4 5
East [m]

Figure 10.55: Snapshots of riser profile as function of time for change in current direction,
simulation 2, where the optimal setpoint chasing is activated at 1000 sec.

441
 TauCX (solid), TauPX (dotted) TauDX (dashed) and TauIX (dashdot) TauCX (solid), TauPX (dotted) TauDX (dashed) and TauIX (dashdot)
1000 1000

500
500

0
0

−500

−500

−1000

−1000
−1500

−1500
−2000

−2500 −2000
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [sec] time [sec]

TauCY (solid), TauPY (dotted) TauDY (dashed) and TauIY (dashdot) TauCY (solid), TauPY (dotted) TauDY (dashed) and TauIY (dashdot)
1000 800

600

500
400

200
0

−500 −200

−400

−1000
−600

−800
−1500

−1000

−2000 −1200
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [sec] time [sec]

TauCPSI (solid), TauPPSI (dotted) TauDPSI (dashed) and TauIPSI (dashdot) TauCPSI (solid), TauPPSI (dotted) TauDPSI (dashed) and TauIPSI (dashdot)
4000 4000

2000
2000

0
0

−2000

−2000

−4000

−4000
−6000

−6000
−8000

−10000 −8000
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000
time [sec] time [sec]

(a) Simulation 1 (b) Simulation 2

Figure 10.56: PID controller input in surge, sway and yaw, where the optimal setpoint chasing
is activated at 1000 sec. Simulation 1 is shown in left column and Simulation 2 in right.

442
Chapter 11

Modelling and Control of High

Speed Craft

This chapter is about ride control of a high speed craft concept denoted as Surface Effect Ships
(SES). As described in Chapter 10, SES is also characterized with elastic dynamics described
by partial differential equations due to its spatial varying air pressure dynamics. SES has a
catamaran type hull form which contains the air cushion with flexible structures called seals or
skirts at the fore and aft ends of the air cushion. Pressurized air is supplied into the cushion
by a lift fan system and is retained by rigid side-hulls and flexible skirt systems at the bow
and the stern. The excess pressure lifts the craft and thereby reduces its calm water resistance.
The major part of the craft weight (about 80%) is supported by the excess air cushion pressure,
while the rest of the weight is supported by the buoyancy of the side-hulls. The most common
stern seal system is the flexible rear bag system consisting of a loop of flexible material, open
at both sides with one or two internal webs restraining the aft face of the loop into a two or
three loop configuration. Pressurized air from the aft of the air cushion is supplied into the bag
system. The bag pressure is about 10 − 15% higher than the air cushion pressure. SES consist of
catamaran hull equipped with skirt in the bow and a pressurized air bag in the stern. SES have
good seakeeping capabilities and offer a comfortable ride in moderate and heavy seas compared
to conventional catamarans. However, in low sea states comfort problems due to high frequency
vertical accelerations have been experienced. These high frequency vertical accelerations are
denoted as the cobblestone effect. Full scale measurements of a 35  SES have indicated the
existence of dominating vertical accelerations around three frequencies in the frequency range
below 15 . Later in the text we will show that the vibration about the first frequency is
due to the resonance of the dynamic uniform pressure (rigid body) in the air cushion. For a
35  SES this frequency is typically around 2 . This resonance may cause excessive vertical
accelerations if a ride control system is not used. The resonances around the second and third
frequency appear to be about 5  and 10  for a 35  SES having a flexible rear bag system.
These resonances are related to the spatially varying pressure (elastic modes) in the air cushion.

443
11.1 Ride Control of Surface Effect Ships
Surface Effect Ships (SES), See Figure 11.1, are known for offering a high quality ride in heavy
sea states compared to conventional catamarans. However, in low and moderate sea states
there are comfort problems due to high frequency vertical accelerations induced by resonances
in the pressurized air cushion and a high performance ride control system is required to achieve
satisfactory passenger comfort and crew workability. To develop such a ride control system it is
essential to use a sufficiently detailed dynamic model. Previous ride control systems have been
based on the coupled equations of motion in heave and pitch as derived by Kaplan and Davis
[144], [145] and Kaplan et al. [146]. Their work were based on the assumption that the major part
of the wave induced loads from the sea was imparted to the craft as dynamic uniform air pressure
acting on the wetdeck, while a minor part of the wave induced loads from the sea was imparted to
the craft as dynamic water pressure acting on the side-hulls. This work was extended by Sørensen
et al. [282], [283] and Sørensen and Egeland [284], who included the effect of spatial pressure
variations in the air cushion. It was found that acoustic resonances in the air cushion excited
by incident sea waves can result in significant vertical vibrations. To investigate the acoustic
resonances a distributed model was derived from a boundary value problem formulation where
the air flow was represented by a velocity potential subject to appropriate boundary conditions
on the surfaces enclosing the air cushion volume. A solution was found using the Helmholtz
equation in the air cushion region. In Sørensen et al. [282] the first two acoustic modes were
included in the mathematical model, while in Sørensen et al. [283] and Sørensen and Egeland
[284] a more general model with an infinite number of acoustic modes was derived. Here, the
mathematical model presented in Sørensen and Egeland [284] is presented. This mathematical
model is used to derive a ride control system, which provides active damping of both the dynamic
uniform pressure and the acoustic resonances in the air cushion. Special attention is given to
sensor and actuator placement to achieve robust stability and high performance. The stability
of the control system is analyzed using the theory of passive systems as presented in Desoer and
Vidyasagar [53] and in Vidyasagar [318]. It is shown that under appropriate assumptions the
dynamic system to be controlled is passive, and n2 stability can be achieved using a strictly
passive controller with finite gain.
The chapter is organized as follows: In Section 11.1.1 the mathematical model is derived.
Section 11.1.2 includes the controller design and the stability analysis. Finally, in Section 11.1.3
simulation results and power spectra of pressure variations and vertical accelerations obtained
from full scale trials with a 35  SES are presented.

11.1.1 Mathematical Modelling

Kinematics
In this chapter a moving coordinate frame is defined so that the origin is located in the mean
water plane below the centre of gravity with the -, - and -axes oriented positive forwards, to
the port, and upwards respectively, see Figure 11.2 . This type of coordinate frame is commonly
used in marine hydrodynamics to analyze vertical motions and accelerations Faltinsen and Zhao
[68]. The equations of motion are formulated in this moving frame. Translation along the -axis
is called heave and is denoted 3 (). The rotation angle about the -axis is called pitch and is
denoted  5 (). Heave is defined positive upwards, and pitch is defined positive with the bow
down. We are mainly concerned about the high frequency vertical vibrations. In this frequency

444
 Figure 11.1: Surface Effect Ship.

range the hydrodynamic loads on the slender side-hulls are of minor importance. Strip theory is
used and hydrodynamic memory effects are assumed to be negligible due to the high frequency
of oscillation. Furthermore, infinite water depth is assumed.

The global continuity equation

The craft is assumed to be advancing forward in regular head sea waves. The waves are assumed
to have a small wave slope with circular frequency  0 . The circular frequency of encounter   is

  =  0 +  (11.1)

where  = 2 is the wave number,  is the sea wave length and  is the craft speed. The
circular frequency of encounter   is the apparent wave frequency as experienced on the craft
advancing forward at the speed  in head sea. The incident surface wave elevation ( ) for
regular head sea is defined as
( ) =   sin(   + ) (11.2)
where   is the wave elevation amplitude. In the case of calm water the wave elevation amplitude
is equal to 0. The water waves are assumed to pass through the air cushion undisturbed. For
simplicity a rectangular cushion is considered at the equilibrium condition with height  , beam
 and length , reaching from  = −2 at the stern (AP) to  = 2 at the bow (FP). The
beam and the height of the air cushion are assumed to be much less than the length. Hence,
a one-dimensional ideal and compressible air flow in the -direction is assumed. This means
that the longitudinal position of the centre of air cushion pressure is assumed to coincide with
the origin of the coordinate frame. The air cushion area is then given by  = . The total
pressure  ( ) in the air cushion is represented by

 ( ) =  +  () +  ( ) (11.3)

where  is the atmospheric pressure,  () is the uniform excess pressure and  ( ) is the
spatially varying excess pressure.

445
 Figure 11.2: Surface Effect Ship (SES)–coordinate frame.

The basic thermodynamic variations in the air cushion are assumed to be adiabatic. When
neglecting seal dynamics, aerodynamics and viscous effects, the external forces are given by
the water pressure acting on the side-hulls and by the dynamic air cushion pressure acting on
the wetdeck. It is assumed that the dynamic cushion pressure is excited by incoming sea wave
disturbances. In the absence of waves, the stationary excess pressure in the air cushion is equal
to the equilibrium excess pressure 0 . This means that at equilibrium  ( ) = 0 such that
(11.3) becomes
 ( ) ≈  + 0 
The rate of change of the mass of air inside the cushion is equal the net mass flux into the
cushion. The global continuity equation for mass flow into and out of the cushion can then be
written as ⎛ ⎞
Z2
 ⎜ ⎟1
 () −  () = ⎝  ( ) ()⎠  (11.4)
 
−2

where  () is the air flow into the air cushion from the lift fan system,  () is the air flow
out of the air cushion due to leakage,  () is the cushion volume, and  ( ) is the density
of the air at the pressure  ( ). The basic thermodynamic variations in the air cushion are
assumed to be adiabatic. From the adiabatic pressure-density relation it follows that
∙ ¸1
 +  () +  ( ) 
 ( ) =   (11.5)
 + 0

where  is the ratio of specific heat for air, and  is the density of the air at the equilibrium
pressure  + 0 . The nondimensional uniform pressure variations  () and the nondimensional

446
 Figure 11.3: The fan characteristic curve.

spatial pressure variations  ( ) are defined according to

 () − 0
 ( ) =  (11.6)
0
 ( )
 ( ) =  (11.7)
0
If we linearize (11.5) about the craft equilibrium operating point  () =  ( ) = 0 and
differentiate with respect to the time , we find an expression for the time derivative of  ( )
as ¡ ¢

̇ ( ) = ³ 0 ´ ̇ () + ̇ ( )  (11.8)
 1 + 0
The rate of change of the air cushion volume can be written as

̇ () =  (̇ 3 () −  ̇ 5 ()) − ̇0 () (11.9)

The last term in (11.9) represents the wave volume pumping of the dynamic uniform pressure
and can be written as
Z2
sin 
2
̇0 () =  ̇( ) =      
cos    (11.10)
2
−2

The air flow into the air cushion is given by a linearization of the fan characteristic curve
about the craft equilibrium operating point, see Fig. 11.3. It is assumed that  fans with
constant   are feeding the cushion, where fan  is located at the longitudinal position   .
Assuming no dynamic fan response, the fan characteristic curve can then be represented by
 µ
X ¶
 ¡ ¢
 () = 0 0 + 0 |0  () +  (   )  (11.11)

=1

447
where 0 is the equilibrium air flow rate of fan  when  () =   and 
 |0 is the corresponding
linear fan slope about the craft equilibrium operating point 0 and 0 . The total equilibrium
air flow rate into the cushion is then
X 
0 = 0 (11.12)
=1
The volumetric air flow out of the air cushion is proportional to the leakage area  (),
which is defined as
 () = 0 +  () (11.13)
 () represents the total leakage area and is expressed as the sum of an equilibrium leakage
area  and a controlled variable leakage area  (). The equilibrium leakage area

0 =  
0 + 0  (11.14)

will be divided into leakage areas under the bow and stern region or more precisely under the
stern and bow seals. 
0 is the stern equilibrium leakage area at  = −2 and 0
  is the bow

equilibrium leakage area at  = 2. The controlled leakage area   () of the ride control
system is written
X
¡  ¢
 () = 0 + ∆  (  )  (11.15)
=1
where  is the number of louvers. The louvers are variable vent valves located at the longitu-
dinal position  =  , which change the area of openings in the wetdeck for the purpose of
leakage control.   is defined as the mean operating value or bias of the leakage area and
∆
 (  ) is defined as the commanded variable leakage area of louver . Pressure sensors are
used to measure the pressure variations in the air cushion. Sensor  is placed at the longitudinal
position  =  . Dynamic leakage areas under the side-hulls and the seals due to craft motion
are assumed to be negligible in this analysis. This type of leakage is a hard nonlinearity and
can be analyzed using describing functions Gelb and Vander Velde [92]. Computer simulations
done in Sørensen et al. [282] indicate that the dynamic leakage terms due to craft motion can
be neglected for small amplitudes of sea wave disturbances and associated small amplitudes of
heave and pitch motions as long as the sealing ability is good.
The air flow out of the air cushion is then represented by

 () =  () +   () +  () (11.16)

where s
2 ( () +  (  ))
 () =  
0  (  ) (11.17)
0
s
2 ( () +  (   ))
  () =  0   (   ) (11.18)
0
às !

X ¡ ¢ 2 ( () +   (  ))
 () =  
0 + ∆
 (  )  (  )  (11.19)
0
=1
 is the air density at the atmospheric pressure  , and  is the orifice coefficient varying
between 061 and 1 depending of the local shape on the edges of the leakage area, see Sullivan
et al. [279]. In the numerical simulations  = 061 is used.

448
 Using Taylor expansion about 0 and 0 , the air flow out of the cushion  (), can be
linearized and written
à 
1 X  1 1
 () = 0  (  ) + 0   (   ) +   (  )+ (11.20)
2 2 2 0 
=1
à 
!  
! s
1 X X X 20
  
0 + 0  () + 0 + 0 + ∆ (  ) 0  
2 0
=1 =1 =1

At stationary equilibrium the air flow into and out of the cushion is equal to
à 
!s
X 20

0 =  0 + 0  (11.21)
0
=1

The global continuity equation given in (11.4) can be linearized and written as

Z2
0
0 ( () −  ()) = ̇ ( ) + 0 ̇ () (11.22)

−2

where  () =   () and  () =   ().  () is the linear volumetric air flow
into the air cushion from the lift fan system.  () is the linear volumetric air flow out of the
air cushion due to leakage. 0 = 0 is the equilibrium air cushion volume, where  is the
cushion length,  is the cushion beam, and 0 is the cushion height at equilibrium. Using the
results from above, we find the uniform pressure equation given in equation (11.40) later in the
text.

Boundary value problem for spatial pressure variations

Wave equation The effect of spatial pressure variations in the air cushion was investigated
by Sørensen et al. [282] and [283] in the frequency domain using Helmholtz equation. Here, the
wave equation is considered leading to a more general solution formulated in the time domain.
A distributed model is derived from a boundary value problem formulation. A one-dimensional
ideal and compressible air flow in the -direction is assumed. The motion of the craft in the
vertical plane is assumed to be small so the body boundary conditions and the free surface
condition can be linearized. The air flow dynamics is formulated in terms of potential theory.
The velocity potential  (  ) for the spatially varying pressure is assumed to satisfy the
wave equation in the cushion region and the boundary conditions enclosing the cushion volume,
see Figure 11.4.
The wave equation is given by
à !
 2  (  ) 2  2  (  )  2  (  )
− + = 0 (11.23)
2 2  2

where  is the speed of sound in air. The one-dimensional approximation of the velocity potential
is introduced as Z 0
1
  ( ) =  (  ) (11.24)
0 0

449
 Figure 11.4: The boundary value problem.

By integrating the wave equation in the -direction, the following partial differential equation
of second order in the two variables  and  is obtained
Z 2 Ã ¯ ¯ !
 (  ) ¯  (  ) ¯ 2 2
2 ¯ ¯ 2    ( )    ( )
 ¯ − ¯ + − = 0 (11.25)
0  −2  =0  =0 2 2
It is then possible to derive an analytical solution of the boundary value problem. Because the
louver area and the outflow area of the lift fan system do not cover the whole beam, we have
to integrate these boundary conditions in the -direction. The nondimensional spatial pressure
variations  ( ) can be found as
0   ( )
 ( ) = −  (11.26)
0 
 2
The unbounded differential operator  2 with boundary conditions on the finite interval  ∈
[−2 2] appearing in equation (11.25), has a set of discrete eigenvalues (Keener, [149]),
2
called the discrete spectrum of the differential operator.  2 is a time-invariant smooth, self-
adjoint differential operator which is dense in the infinite dimensional inner product space 2 (Ω),
where Ω denotes the air cushion.2 (Ω) is complete and hence a Hilbert space. The eigenvalue
equation is given by
2
−2 ( ()) = 2  ()  = 1 2 3   (11.27)
2
where  () is the eigenfunction or the mode shape function of mode , and  is the corre-
sponding eigenfrequency. The eigenfunctions for  ∈ {1 2 3  } are orthonormal, that is
Z 2
( ()  ()) =  () () =    (11.28)
−2

where (· ·) denotes the scalar product. The Kronecker delta is defined by   = 1 when  = 
and   = 0 when  6= . Letting  → ∞, the eigenfunctions  () for  ∈ {1 2 3 } form an
orthonormal basis of the Hilbert space 2 (Ω). Thus,  ( ) has a unique modal representation
for  → ∞ given by
∞ ∞
0 X   ()  X

 ( ) = −   () = ̇ () () (11.29)
0 
=1 =1

450
where ̇ () is the modal amplitude function for mode  and  () =  (). Due to linearity and
orthogonality between the modes, each mode can be considered separately and the contributions
from each of them are superposed.

Boundary conditions We can set up the following boundary conditions on the surfaces
enclosing the cushion air volume.
1. On the rigid part of the wetdeck ( = 0 )
 (  )
= ̇ 3 () − ̇ 5 () (11.30)

2. At the rigid bow and rear seal systems ( = ± 2)
 (  )
= 0 (11.31)

3. At the fan outlet ( = 0   =   )
⎛ ⎞

X ¯ ∞
 (  ) 1  ¯¯ ⎝X
= −0 ̇ () (  ) +  ()⎠  (11.32)
    ¯0
=1 =1

where   is the outlet area of fan .

4. At the controlled leakage area ( = 0   =  )
s ⎛ ⎞
X X∞ 
 (  ) 20
=  ⎝1 1
̇ () ( ) +  () −
∆ (  ) ⎠
 (11.33)
 0
=1
2
=1
2 
0

5. At the bow leakage area ( = 0  =   = 2)

Assuming no motion induced leakage, we have
s ⎛ ⎞
X∞
 (  )  20 ⎝
=− ̇ () (  ) +  ()⎠  (11.34)
 2 0
=1

6. At the stern leakage area ( = 0  =  = −2)

Assuming no motion induced leakage, we have
s ⎛ ⎞
∞
 (  )  20 ⎝X
=− ̇ () ( ) +  ()⎠  (11.35)
 2 0
=1

7. At the mean free surface ( = 0)

 (  )
=    cos ( +  )  (11.36)

The mode shape functions will be chosen so the boundary conditions on the seals are satisfied.
The following mode shape functions, in general infinite many, will satisfy the boundary condition
on the seals
µ ¶ ∙ ¸
   
 () = cos +   = 1 2 3    ∈ −   (11.37)
 2 2 2

451
From (11.27) we find that the corresponding eigenfrequency   for mode 

 =    = 1 2 3  (11.38)

The modal amplitude functions will be determined according to (11.25) and the remaining
boundary conditions. By taking the inner product of (11.25) with the mode shape function of
mode  for  ∈ {1 2 3 } and using (11.29), we find the spatially varying pressure equation for
mode  for  ∈ {1 2 3 }, see (11.43) and (11.46) later in the text. The heave force and the
pitch moment due to the spatially varying pressure are found in the following way
Z ∞
2 X
3 () = 0  ̇ () () = 0  = 1 2 3  (11.39a)
−2 =1
⎧ ³ ´2
Z ∞
2 X ⎨ 2  P 
̇ ()  = 1 3 5 
0 
5 () = −0  ̇ () () = =135 (11.39b)
−2 =1 ⎩
0  = 2 4 6 

Due to symmetry about the origin, there is no contribution from the spatially varying pressure
on the heave motion. However, the pitch moment 5 () is important to include in the pitch
equation.

Equations of motion and dynamic cushion pressure

The coupling between the dynamic uniform pressure and the spatially varying pressure and
hydrodynamic and hydrostatic coupling terms like {̇ , ̇ }, { ,  }, and {   }-terms
are assumed to be negligible regarding control system design. The equations of motions and the
dynamic air cushion pressure are:
(1) Uniform pressure equation

X
1 ̇ () + 3  () + 0  ̇ 3 () = 2 ∆
 (  ) + 0 ̇0 () (11.40)
=1

where ̇0 () is the wave volume pumping given by (11.10),and

0 0 
1 = ³ ´ (11.41a)
 1 + 0
s
20
2 =   (11.41b)
0
X µ ¯ ¶
0  ¯¯
3 = 0 − 0  (11.41c)
2  ¯0
=1

(2) Spatially varying pressure equation

∞
X µ ¶ ∙ ¸
   
 ( ) = ̇ () cos +  ∈ −   (11.42)
 2 2 2
=1

452
Odd modes about the centre of pressure,  = 1 3 5 

̈ () + 2    ̇ () + 2  () = −2 ̇ 5 () + (11.43)


X µ ¶
 
1 cos  + ∆
 (  ) + 0 ̇ ()
 2
=1

where
22 2
1 =  (11.44a)
0 0
40 2
2 =  (11.44b)
0 0 ()2
The wave volume pumping for regular head sea for  = 1 3 5  is

42  cos 
̇ () = − ³ 2 ´2     sin   (11.45)
0 0  2
 −  

Even modes about the centre of pressure,  = 2 4 6 


X µ ¶
 
̈ () + 2    ̇ () + 2  () = 1 cos  + ∆
 (  ) + 0 ̇ () (11.46)
 2
=1

where the wave volume pumping for regular head sea for  = 2 4 6  is

42  sin 
̇ () = ³ 2 ´2     cos    (11.47)
0 0 
 − 
2


The relative damping ratio for  = 1 2 3 4  is

µ
2 P 
 ¡ ¢

0 
2
20 0 + 20 0 cos2  
  + 2 −
 = ¯ =1 ¶  (11.48)
P  ¡ ¢

 ¯ 2 
0  ¯ cos    + 2 0
=1

(3) Heave equation

( + ̇ ) ̈ 3 () +  ̇ 3 () +   3 () −  0  () = 3 () (11.49)

where  is the vessel mass.

(4) Pitch equation

X µ ¶2

(33 + ̇ ) ̈ 5 () +  ̇ 5 () +  5 () − 20  ̇ () = 5 () (11.50)

=135

where 55 is the moment of inertia about the -axis.

453
 The hydrostatic {   }-terms are found in the standard way by integration over the water
plane area of the side-hulls. The hydrodynamic added-mass coefficients {̇  ̇ }, the water
wave radiation damping coefficients {   }, and the hydrodynamic excitation force in heave
3 () and moment in pitch 5 (), are derived from hydrodynamic loads on the side-hulls. The
hydrodynamic loads on the side-hulls may be calculated as presented in Faltinsen and Zhao [68],
[69], Faltinsen et al. [70], [71] and Nestegård [196]. However, since the main focus is on the
high frequency range, we have used a simplified strip theory based on Salvesen et al. [248] for
calculation of the hydrodynamic loads. Neglecting the effects of transom stern and radiation
damping, the hydrodynamic excitation forces on the side-hulls in heave and pitch are given by
sin 
3 () = 2  − 
2
( −  0  ̇ ) sin    (11.51a)
2
£¡ ¢
2  − 1 cos 
2 − sin¸ 
2
2  2 ( −  0   ̇ ) −
5 () = sin  (11.51b)
  2 ̇ cos   
2

where  is the draft of the side-hulls. In the case studied here, the submerged part of side-hulls are
assumed to have constant cross-section area. Examples of two-dimensional frequency depending
added-mass and wave radiation damping coefficients are found in Faltinsen [66]. In the control
system analysis constant two-dimensional added-mass and wave radiation damping values are
assumed. The high frequency limit of the two-dimensional added-mass coefficient found in
Faltinsen [66] is used. The selected wave radiation damping coefficient in pitch corresponds to
the value at the pitch resonance frequency determined from structural mass forces acting on the
craft and hydrodynamic forces on the side-hulls. For heave we have chosen the wave radiation
damping coefficient at the resonance frequency that will exist without the presence of the excess
air cushion pressure. These simplifications are motivated by fact that the effect of damping is
most pronounced about the corresponding resonance frequency.

Helmholtz equation
Above the boundary value problem was solved in the time domain, allowing us to study the
transient behavior of the craft. If we assume that steady state conditions have been obtained,
and that the time dependence is harmonic, the effect of spatial pressure variations in the air
cushion may be investigated in the frequency domain using the Helmholtz equation (the Wave
equation (11.23) transformed into frequency domain). We must then assume that the initial
conditions of the dynamic variables are equal to zero. Furthermore, the craft is assumed to be
exposed to regular sea waves where the transients are assumed to vanish quickly. This motivates
the assumption of harmonic time dependence,   , where  is the complex unit. Hence, we write
 3 () =  3     5 () =  5     () =      () =     (11.52)
where  3 and 5 are the heave and pitch motion amplitudes relatively to the coordinate frame
located in the water plane below the centre of gravity, and  and  are the nondimensional
uniform and spatially varying pressure amplitudes respectively. These amplitudes are generally
complex and parametric functions of the frequency of encounter   . Like in the ship motion
problem, Salvesen et al. [248], it is convenient to split the velocity potential  (  ) in the
air cushion into
(0 ( ) P
+ 3 ( )3 + 5 ( )
¢ 5 + 7 ( )  +
 (  ) =   (11.53)
=1 7+ ( )∆  
 

454
where 0 ( ) 3 ( ) 5 ( ) 7 ( ) 8 ( )  , 7+ are the spatially varying pressure
potentials due to the effect of the dynamic uniform pressure, heave and pitch motion, the
incident waves and the  control actions respectively. The subscript  must not be confused with
the complex unit . The frequency dependent modal amplitude functions will be determined
according to the Helmholtz equation for  = 0 3 5 7 8  7 +  according to
³  ´2  2  ( )  2  ( )

 ( ) + + = 0 (11.54)
 2  2
Implicitly  ( ) is a function of the circular frequency of encounter   , but this is considered
as a parametric dependence. Similarly, eq. (11.25) transformed into frequency domain becomes
Z 2 Ã ¯ ¯ !
1  ( ) ¯¯  ( ) ¯¯  2   () ³  ´2
 ¯ − ¯ + +   () = 0 (11.55)
0  −2  =0  =0 2 

  (;   ) is the one-dimensional spatially varying pressure for  = 0 3 5 7 8  7 +  with

modal representation given by
∞
X
  (;  ) =  (  ) () (11.56)
=1

where  (  ) is the frequency dependent modal amplitude function for mode  due to action
. Due to linearity and orthogonality between the modes, we can consider each mode separately
and superpose the contribution from each of them. As in the time domain, the frequency
dependent modal amplitude functions have to be determined according to equation (11.55) and
the remaining boundary conditions formulated in the frequency domain. By taking the inner
product of equation (11.55) with the mode shape function of mode  for  ∈ {1 2 3 } and
using equation (11.56), we find the frequency dependent modal amplitude functions. However
before doing so, it is convenient to divide the mode shape functions into even and odd modes
around the origin of the coordinate frame. Thus,  = 1 3 5  are representing the odd modes,
while  = 2 4 6  are representing the even modes. Denoting  =   to be the differential
operator, equation (11.56) is written
∞
X X X
 () () =  () () +  () () (11.57)
=1 =135 =246

where  () and  () are the frequency dependent modal amplitude functions for the odd
and the even modes respectively due to action . Expressions for  (),  () and the cor-
responding equations of motion and dynamic cushion pressure formulated in frequency domain
are found in Sørensen [285].

Discussion of the mathematical model

It is seen from (11.49) and (11.50) that the heave and pitch motions are coupled to the dynamic
excess pressure in the air cushion region. This is to be expected since the major part of the
SES mass is supported by the air cushion excess pressure. The dynamic air cushion pressure
is expressed as the sum of the dynamic uniform pressure and the spatially varying pressure.
An important question is how many acoustic modes should be included in the mathematical

455
model. Even if the solution is formally presented by an infinite number of acoustic modes,
the modelling assumptions will not be valid in the high frequency range when two and three
dimensional effects become significant. Then a more detailed numerical analysis is required,
like for instance a boundary element or a finite element method. In the following we will use a
finite number  acoustic modes in the mathematical model. The effect of higher order modes is
assumed to be negligible. It is important to note that the air cushion dimensions and the forward
speed affect the energy level of the vertical accelerations caused by the acoustic resonances. The
acoustic resonance frequencies are inversely proportional to the air cushion length as seen from
(11.38). The wave excitation frequency which is given by the circular frequency of encounter
  =  0 + , increase with the forward speed  . Thus waves of relatively low circular frequency
 0 may excite the craft in the frequency range of the acoustic resonances when the speed  is
high. This may result in more energy in the sea wave excitation about the resonance frequencies
since the maximum sea wave height will tend to increase when the period of the sea waves
increases. The relative damping ratio   given by (11.48) is an important parameter. As expected
the leakage terms and the fan inflow term contribute to increased damping. One should notice
that the fan slope ()| , is negative. We also observe that the longitudinal placement of
the fan and the louver systems strongly affects the relative damping ratio. In the case of a single
fan system and a single louver system, it may seem natural to place the fan and the louver in
the middle of the air cushion, that is  =  = 0. However, from (11.48) we observe that the
relative damping ratio for the odd modes will be reduced significantly if  and  are equal
to 0. Maximum damping of both the odd and even acoustic resonance modes in the case of a
single lift fan system and a single louver system is obtained for  and  equal to −2 or2.
The relative damping ratio of the first odd acoustic mode on a 35 m SES will increase from
about 0.05 to 0.2 by placing the lift fan system at one of the ends of the air cushion instead
in the middle. This gives a significant improvement in ride quality even when the ride control
system is turned off. In the same manner the active damping due to the ride control system is
maximized by placing the louver system at one of the ends of the air cushion.

Remark 11.1 Here, the Faltinsen notation is used to define the coordinate frame. The hydro-
dynamic coefficients are written according to the SNAME [268] notation.

11.1.2 Robust Dissipative Controller Design

In this section a ride control system based on the mathematical model derived in the previous
section is developed. The objective of the controller is to damp out pressure fluctuations about
the equilibrium pressure 0 in the presence of sea wave disturbances. This can be formulated
in terms of the desired value of the nondimensional dynamic uniform pressure  () = 0 and
the nondimensional spatially varying pressure  ( ) = 0, where the superscript  denotes the
desired value.
The number of modes to be damped depends on the requirements related to established cri-
teria for crew workability and passenger comfort. The mathematical model of the craft dynamics
is of high order as it contains a high number of acoustic modes. A practical implementable con-
troller has to be of reduced order. When designing a controller based on a reduced order model,
it may happen that the truncated or residual modes give a degradation of the performance, and
even instability of the closed loop system. This is analogous with the so-called spillover effect in
active damping of vibrations in mechanical structures (Balas [14]). The inadvertent excitation
of the residual modes has been termed control spillover, while the unwanted contribution of the

456
residual modes to the sensed outputs has been termed observation-spillover, see Figure 11.5.
This problem was also discussed by Gevarter [93] in connection to control of flexible vehicles.
Mode 0 in Figure 11.5 is related to the uniform pressure, while the higher order modes are
related to the spatially varying pressure.
The controller must be robust with respect to modelling errors and parametric and non-
parametric uncertainties, nonlinearities in sensors and actuators and component failure. The
use of collocated compatible actuators and sensors pairs and strictly passive controllers provides a
design technique for circumventing these problems. Then louver and sensor pairs are distributed
along the air cushion, preferentially in the longitudinal direction. The problem described here is
closely related to the problem of vibration damping in large flexible space structures. Inspired by
the work of Joshi [139] we propose to use dissipative control for vibration damping of SES. Note
that the problem of vibration damping of SES has significant differences as the dynamic system
given by (11.40)-(11.51) is of third order as opposed to similar vibration damping problems of
large flexible space structures that can be written as an equivalent second order mass, damper
and spring system. Also here we use passivity theory in the following, whereas Joshi [139] used
Lyapunov theory. In dissipative controller synthesis using a linear design model it is possible to
use standard linear synthesis methods, however, the design is complicated by the constraint that
the resulting controller must satisfy appropriate passivity properties. Here, a static dissipative
solution is presented where a passive controller with proportional feedback is used. It would
also be possible to use dynamic dissipative control based on LQG synthesis (Lozano-Leal and
Joshi [177], Joshi and Maghami [140], Joshi et al. [141]) or  ∞ control synthesis (Haddad et al.
[103]). It is straightforward to extend the stability results presented in the following to the case
of dynamic dissipative controllers.

Preliminaries
The linear vector space 2 of real square-integrable  vector functions of time  is defined by
½ ¯Z ∞ ¾
 ¯
¯
2 = z : + →  ¯ z() z()  ∞  (11.58)
0

Define the truncation operator  such that

½
z() 0≤≤
z =  z =  (11.59)
0 0
The extended space2 is then defined as
2 = {z : + →  |∀ ≥ 0 kzk 2  ∞}  (11.60)
The respective norms k·k2 and k·k 2 are defined by
kzk2 = lim kzk 2  (11.61)
 →∞

where
µZ  ¶12
12 
kzk 2 = (z z) = z() z()  (11.62)
0
The inner product (·,·) of x z ∈ 2 is
Z 
(x z) = x() z() (11.63)
0

457
Figure 11.5: Observation and control spillover, where    = cos ( + 2) and    =
cos ( + 2).

458
so that kzk2 2 = (z z) . If x z ∈ 2 , then

(x z) = lim (x z)  (11.64)

→∞

and hence kk22 = ( ). A function  : + →  belongs to ∞ if

kzk∞ =  sup |z()|  ∞ (11.65)

≥0

By essential supremum we mean

 sup |z()| = inf {| |z()|   almost everywhere}  (11.66)

≥0 ≥0

that is, |z()| ≤  except for a set of measure zero, and the  sup is the smallest number
which has that property. Notice that the ∞ space is a normed linear space but not an inner
product space. Disturbances induced by irregular sea, and regular sea as well, are persistent for
a long period of time and can be represented by sinusoids. This is the case in (11.40), (11.47),
(11.45),and (11.51 ). Such disturbances belong to the ∞ space but not to the 2 space. On
the other hand, in regular sea over a finite time interval when for example running the craft into
and out of a wave field set up by the craft itself or by other ships, the disturbances are transient
and may belong to the 2 space. This may be modelled using the truncation operator given in
(11.59) on the (11.40), (11.47), (11.45),and (11.51 ).
Passivity and strict passivity on 2 are defined as:

Definition 11.1 (Desoer and Vidyasagar [53]). Let  : 2 → 2 .  is passive if there exist
some constant  such that

(u u) ≥  ∀u ∈ 2  ∀ ≥ 0 (11.67)

 is strictly u-passive if there exist some  2  0 and some constant  such that

(u u) ≥  2 kuk2 2 +  ∀u ∈ 2  ∀ ≥ 0 (11.68)

Definition 11.2 (Desoer and Vidyasagar [53]). Let  :  →  . The mapping  is 
stable with finite gain if u ∈  whenever u ∈  and there exist finite constants  and 
such that
kuk ≤  kuk +  ∀u ∈    ∈ [1 ∞]  (11.69)
If  = 0, the mapping  is  stable with finite gain and zero bias.

State space model

The dynamic system given by (11.40)-(11.51b) is written in standard state space form according
to

ẋ() = Ax() + Bu() + Eν() (11.70a)

y() = Cx() (11.70b)

where the -dimensional state vector x() is

x() = [ 3   5  ̇ 3  ̇ 5    1  2     ̇1  ̇2   ̇ ]  (11.71)

459
() is the (3 + )-dimensional disturbance vector defined as
h i
ν() = 3  5  ̇0  ̇1  ̇2  · · ·  ̇  (11.72)

where 3 (), 5 () and the time derivative ̇ () for  = 0 1 2   are found in Section 11.1.1.
 is the number of acoustic modes. u() is the -dimensional control input vector, and  is the
number of louvers. The elements of u() = [1  2    ] are for  = 1 2  

 () = ∆
 (  ) (11.73)

where ∆
 (  ) is defined in (11.15). y() is the -dimensional measurement vector, and
 is the number of pressure sensors. The symbolic expressions for the  ×  system matrix
A, the  ×  control input matrix B, the  × (3 + ) disturbance matrix E and the  × 
measurement matrix C are defined below.
The  ×  system matrix is given by
⎡ ⎤
A15×5 05× A25×
A = ⎣ 0×5 0× I× ⎦  (11.74)
A3×5 A44×4 A5×

where 05× , 0×5 and 0× are the 5 × ,  × 5 and  ×  zero matrices respectively. I× is the
 ×  identity matrix. A15×5 is defined as
⎡ ⎤
0 0 1 0 0
⎢ 0 0 0 1 0 ⎥
⎢ ⎥
⎢    0 ⎥
A15×5 = ⎢ − +̇ 0 − + 0
 
̇ + ̇ ⎥  (11.75)
⎢   ⎥
⎣ 0 − 55 +̇ 0 − 55 +̇ 0 ⎦
0 0 − 0
1

0 −3
1

A25× is defined as ⎡ ⎤
0 0 0 0 0 ··· 0
⎢ 0 0 0 0 0 ··· 0 ⎥
⎢ ⎥
A25× =⎢
⎢ 0 0 0 0 0 ··· 0 ⎥
⎥ (11.76)
⎣ 1 0 3 0 5 ··· 0 ⎦
0 0 0 0 0 ··· 0
where µ ¶2
20  
 =  (11.77)
55 + ̇ 
A3×5 is defined as ⎡ ⎤
0 0 0 1 0
⎢ 0 0 0 0 0 ⎥
⎢ ⎥
⎢ 0 0 0 3 0 ⎥
A3×5 =⎢ ⎥ (11.78)
⎢ .. .. .. .. .. ⎥
⎣ . . . . . ⎦
0 0 0 0 0
where
40 2
 = −  (11.79)
0 0 ()2

460
A4× and A5× are defined as
© ª
A4× = diag − 2  (11.80)
© ª
A5× = diag −2    (11.81)

The  × (3 + ) disturbances matrix is defined as

⎡ ⎤
0 0 0 0 0 ··· 0
⎢ 0 0 0 0 0 ··· 0 ⎥
⎢ ⎥
⎢ 1
0 0 0 0 ··· 0 ⎥
⎢ +̇ ⎥
⎢ 0 1
0 0 0 ··· 0 ⎥
E=⎢ 55 +̇ ⎥ (11.82)
⎢ 0 ⎥
⎢ 0 0 0 0 ··· 0 ⎥
⎢ 1 ⎥
⎣ 0×(3+) ⎦
0×3 0 I×
The  ×  control input matrix is given by
⎡ ⎤
0 0 ··· 0
⎢ 0 0 ··· 0 ⎥
⎢ ⎥
⎢ 0 0 ··· 0 ⎥
⎢ ⎥
⎢ 0 0 ··· 0 ⎥
⎢ ⎥
⎢ 2 2 2 ⎥
⎢ 1 1 ··· 1 ⎥
B=⎢ ⎢ 0
⎥
⎥ (11.83)
⎢ ¡ ¢ ¡× ¢ ¡ ¢ ⎥
⎢ 1 cos  1 + 2 1 cos  
 ¡2 + 2 ¢ ···  
1 cos  ¡ + 2 ¢ ⎥
⎢ ¡ ¢ ⎥
⎢ 1 cos 2 1 +  1 cos 2 
1 cos 2  ⎥
⎢  2  2 + 2 ···   + 2 ⎥
⎢ .. ⎥
⎣ . ⎦
¡ ¢ ¡ ¢ ¡ ¢
1 cos  1 + 2

1 cos 
 2 + 2

··· 
1 cos   + 2

where
22 2
1 =  (11.84)
0 0
The  ×  measurement matrix is given by
⎡ ⎤
0 0 ··· 0
⎢ 0 0 ··· 0 ⎥
⎢ ⎥
⎢ 0 0 ··· 0 ⎥
⎢ ⎥
⎢ 0 0 ··· 0 ⎥
⎢ ⎥
⎢ 1 1 ··· 1 ⎥
⎢ ⎥
C = ⎢ 0× ⎥ (11.85)
⎢ ¢ ⎥
⎢ cos  ¡1 +  ¢ cos  ¡2 +  ¢ · · · ¡
cos  ¡ + 2 ¢ ⎥
 
⎢ ⎥
⎢ cos 2 ¡ +  ¢ cos 2 ¡ +  ¢ · · ·
 2  2
 ⎥
⎢  1 2  2 2 cos 2
  + 2 ⎥
⎢ .. ⎥
⎣ . ⎦
¡ ¢ ¡ ¢ ¡ ¢
cos 
 1 + 2

cos 
 2 + 2

··· 
cos   + 2 

Consider the case where sensors and actuators are ideal, that is linear and instantaneous
with no noise. It is assumed that the control input matrix B can be related to the measurement
matrix C so that
C = B P (11.86)

461
 Figure 11.6: Feedback system.

where P is a  ×  symmetric positive definite matrix providing correct scaling of the B matrix
to the C matrix. This is the case when there is perfect collocation between the sensors and the
louvers, i.e.  =  for all  and  = . We can derive the linear time-invariant operators
between the outputs and the inputs of the dynamic system given by (11.70). Denote  to be the
differential operator. Since the pair (A B) is controllable, and the pair (C A) is observable,
the dynamic system can be represented by

y() =  ()u() +  ()ν() = y () + y () (11.87)

where

 () = C(sI −A)−1 B (11.88)

−1
 () = C(sI −A) E (11.89)

and I is the  ×  identity matrix.

Stability properties of the control system

In this section a strictly passive controller with finite gain is proposed. Employing the definitions
of passivity as presented in Desoer and Vidyasagar [53] and in Vidyasagar [318] on an inter-
connected system consisting of a two subsystems in a standard feedback configuration (Figure
11.6), robust stability of the feedback system can be shown for certain input-output properties
of the subsystems. The following lemma showing that the  ×  system matrix A is Hurwitz
will then be utilized.

Lemma 11.1 The eigenvalues of the  ×  system matrix A given in (11.70) have negative real
parts.

Proof. Consider the autonomous system of (11.70) with u() = ν() = 0. Define the
Lyapunov function candidate
1
 (x) = x Px  0 (11.90)
2
where the  ×  diagonal positive definite matrix P is given below.  (x) is positive definite.
The time derivative of  (x) along system trajectories is
1 ¡ ¢ 1
̇ (x) = x A P + PA x = − x Qx ≤ 0 (11.91)
2 2

462
The  ×  diagonal positive semidefinite matrix Q is
© ª
Q =  02×2  Q13×3  0×  Q2×  (11.92)

The 3 × 3 diagonal positive definite Q13×3 matrix is defined

Q13×3 =  { }   = 3 4 5

where
2
33 = 33  (11.93a)
 + ̇
2
44 = 44  (11.93b)
 + ̇
23
55 = 55  (11.93c)
1
The  ×  diagonal positive definite Q2× matrix is

Q2× =  { }   = 5 +  + 1 5 +  + 2  5 + 2 (11.94)

where
(5++)(5++) = (5++)(5++) 4      = 1 2 3   (11.95)
The  ×  diagonal positive definite matrix P is found to be

P =  { }   = 1 2  5 + 2 (11.96)

where

11 = 33  (11.97a)
 + ̇

22 = 44  (11.97b)
55 + ̇
 ( + ̇ )
33 = 55 0  (11.97c)
1 0
1
44 = −  (11.97d)
1 1
1
55 = 2  (11.97e)
0
 2
 =   = 6 7  5 +  (11.97f)
1
1
 =   = 6 +  7 +   2 (11.97g)
1
and where 1 is defined in (11.44a). From (11.70) it is seen that

̇ (x) = 0 (11.98)
⇓
x = x0 = [ 3   5  0 0 0 1  2     0 0  0]  (11.99)

463
Consequently
̈3 = ̈5 = ̇ = ̈1 = ̈2 = · · · = ̈ = 0 (11.100)
which implies
3 = 5 = 1 = 2 = · · · =  = 0 (11.101)
Hence, by the invariant set theorem (Vidyasagar, [318]) the equilibrium point x0 = 0 is asymp-
totically stable and the result of Lemma follows.
3+
Define the linear time-invariant operators  :  
2 → 2 ( = ) and  : 2 →  2 ,
 3+
such that y ∈ 2 whenever v ∈ 2 . In the following lemma it is shown that the process
operator  is passive. This allows for the design of robust, stable output feedback controllers
for ride control of SES.

Lemma 11.2 The process operator  is passive.

Proof. Set ν() = 0 in (11.70) and use the Lyapunov function candidate as given in (11.90).
 (x) is positive definite. The time derivative ̇ (x) along the system trajectories is
1 ¡ ¢
̇ (x) = x A P + PA x + x PBu (11.102)
2
The × diagonal positive semidefinite matrix Q is given in (11.92). Thus, if we assume perfect
collocation between the sensor and actuator pairs, that is C = B P, (11.102) becomes
1
̇ (x) = x C u − x Qx (11.103)
2
1
= y u− x Qx (11.104)
2
Integrating (11.104) from  = 0 to  =  we obtain
Z 
1
(y  u) =  ( =  ) −  ( = 0) + x Qx  (11.105)
2 0

SinceQ ≥ 0 and  ( =  )  0, (11.105) can be written

(y  u) ≥ − ( = 0) ,  (11.106)

and the result of Lemma 2 follows.

Remark 11.2 It is seen from (11.71) that if the initial conditions are equal to zero that is
x( = 0) = 0, then  = −x ( = 0)Px( = 0) = 0.

Remark 11.3 The transfer matrix  () of the linear time-invariant operator  as defined
by (11.89) is strictly proper and all the poles have negative real parts according to Lemma 1.
Hence, if ν ∈ 3+ 
2 , then y =  v ∈ 2 ∩ ∞ .


Let the controller be defined as the linear time-invariant operator  between the input
y = y + y and the output u . Connecting the  operator together with the  and 
operators, we obtain the feedback system illustrated in Figure 11.6. The transfer matrix of 
is denoted  ().

464
Proportional control law
A strictly u-passive proportional pressure feedback controller of dimension  ×  with finite gain
is proposed according to

u () =  ()y() (11.107)

 () = G 

where G =  { }  0 is a constant diagonal feedback gain matrix of dimension  × .

This control law provides enhanced damping of the pressure variations about the resonance
frequencies.
The main result of this section is given by the following theorem.

Theorem 11.1 Consider the following feedback system (see Figure 11.6)

y =  u (11.108)
y =  ν (11.109)
y = y + y  (11.110)
u = −u = − y (11.111)
3+ 3+
where    :  
2 → 2 . Assume that  : 2 →  
2 , so that y ∈ 2 whenever ν ∈ 2 .
 is strictly u-passive with finite gain, and  is passive. Hence, the feedback system defined
above is  
2 stable, and since the feedback system given by  ,  and  is linear,2 stability
is equivalent to 
∞ (BIBO) stability.

Proof. Set v() = 0 in (11.70) and use the Lyapunov function candidate as given in (11.90).
 (x) is positive definite. If we assume perfect collocation between the sensor and actuator pairs,
that is C = B P, the time derivative ̇ (x) along the closed loop system trajectory becomes
1
̇ (x) = −y G y − x Qx
µ 2 ¶
  1
= −x C G C+ Q x (11.112)
2

where the  ×  diagonal positive semidefinite matrix Q is given in (11.92). It can be shown by
inspection that
1
C G C + Q ≥ 0 (11.113)
2
since the first term in (11.112) is on quadratic form for diagonal G  0 and hence positive
semidefinite. The time derivative ̇ (x) is negative semidefinite. Using the invariant set theorem
(Vidyasagar [318]) the equilibrium point of the closed-loop system is asymptotically stable and
the result follows.
 
2 and ∞ stability of the closed-loop system using collocated sensor and actuator pairs is
maintained regardless of the number of modes, and regardless of the inaccuracy in the knowledge
of the parameters. Thus the spillover problem is eliminated and the parameters do not have to
be known in advance to obtain stability. Notice that there are no restrictions to the location
of the collocated sensor and actuator pairs with respect to stability. However, optimizing the
performance, the longitudinal location of the sensor actuator pairs is crucial as seen in (11.48).

465
 Length overall 35 m
Equilibrium fan flow rate 150 m3  s
Linear fan slope -140 m2  s
Cushion length 28 m
Nom. cushion pressure 500 mm 
Cushion beam 8m
Cushion height 2m
Weight 150 ton
Speed 50 knots

Table 11.1: SES main dimensions.

Robustness with respect to unmodelled dynamics and sector nonlinearities in the actuators are
shown in Sørensen [283]. It is further shown by Sørensen [283] that some imprecision in the
collocated sensor and actuator pairs can be accepted without violating the stability properties
of the closed-loop system

11.1.3 Simulation and Full Scale Results

In this section numerical simulations and results from full scale trials with a 35 m SES advancing
forward at high speed in head sea waves are presented. The effect of collocation and noncolloca-
tion of the sensor and actuator pairs for the 35 m SES is investigated. The SES is equipped with
one single fan and louver system. The number of acoustic modes considered in the simulation
model is four, i.e.  = 4. Main dimensions and data of the SES craft are given in Table 11.1.

Remark 11.4 One should notice that often mm  is used as unit for the pressure. The
1 kg m kg
relation between mm  and Pa (Pascal) is 1mm  = 1000 m · 1000 m3 · 981 s2 = 981 ms2 =
981Pa

Numerical simulations
Figure 11.7 shows the Bode plot of  ( ) between the pressure sensor  () and the louver
() when the pressure sensor and actuator pair is fully collocated. The sensor and louver pair
is located at the fore end of the air cushion. When the frequency of encounter goes to zero,
the dynamic pressure tends to a static value proportional to 1 2 . This indicates that the
equilibrium pressure 0 will decrease when the equilibrium leakage area increases and vice versa.
about 01 Hz the response is close to zero. This is related to the structural mass forces acting on
the SES and the hydrodynamic forces acting on the side-hulls. The high value about 2 Hz is due
to the resonance of the dynamic uniform pressure. The high values about 6 Hz, 12 Hz, 18 Hz and
24 Hz are related to the four acoustic resonance modes. From the phase plot we observe that
the phase is varying between 90 ◦ to −90 ◦ in the whole frequency range. This is to be expected
when using collocated sensor and actuator pairs.
Figure 11.8 shows the Bode plots of  (  ) for the case the pressure sensor is located at
the fore end of the air cushion while the louver is located at the aft end of the air cushion.
From the phase plot we observe that the sensed pressure signal at the fore end is 180 ◦ out of
phase compared to the pressure signal at the aft end where the louver is located. This is to be

466
Figure 11.7: Collocated sensor and actuator pair; numerically calculated Bode plot of  (  );
 =  = 12 m  = 6 m  = 50 knots, 0 = 500 mm .

467
Figure 11.8: Noncollocated sensor and actuator pair; numerically calculated Bode plot of
 (  );  = 12 m  = −12 m  = 6 m  = 50 knots 0 = 500 mm 

468
 Figure 11.9: Experimental arrangement.

expected with noncollocated sensor and louver pairs. Noncollocated sensor and actuator pairs
introduce negative phase and may lead to instability.

Full scale results

The prototype ride control system used in the full scale experiments was based on the passive
controller as presented in Section 11.1.2. The control algorithms in the ride control system
were partly implemented on a personal computer (PC) and partly analog hardware devices were
used. An outer feedback loop was implemented on the PC, while a faster inner feedback loop
about the electro-hydraulic louver system was implemented by analog hardware devices. The
lover system consisted of two vent valves located at the same longitudinal position  = 8 m
side by side. The two vent valves were operated in parallel in the outer feedback loop. Two
pressure sensors located at 1 = 10 m and 2 = −10 m were used to measure the excess pressure
variations in the air cushion. One accelerometer located about 5 m aft of the centre of gravity was
used to measure the vertical accelerations. The inner analog controller loop around the louver
system provided the necessary opening and closing actions of the vent valves. The experiment
arrangement is illustrated in Figure 11.9. The full scale measurements were carried out in sea
states with significant wave heights estimated to vary between 03 − 06 m. The power spectra
of the vertical accelerations with and without the ride control system are presented.Figure 11.10
shows the full scale power spectra of the vertical accelerations about 5 m aft of the centre of
gravity with and without the ride control system activated. With the ride control system turned
off, we observed significant response about 2 Hz, 5 Hz and 8 Hz. The response about 2 Hz is
related to the resonance of the dynamic uniform pressure, while the response about 5 Hz and
8 Hz are related to the first odd and even resonance modes. Activating the ride control system
the response about 5 Hz was significantly amplified, while the response about 2 Hz was only
slightly reduced. In this case the pressure signal at 2 = −10 m was used in the feedback loop.
Hence, the actuator and sensor pair were completely noncollocated since the louver was located
at  = 8 m. This means that for the first odd mode, the noncollocation resulted in positive
feedback for this particular mode because the pressure at the sensor location was 180 ◦ out of
phase compared to the pressure at the actuator location in the frequency range dominated by
the first odd acoustic resonance mode. The response about 8 Hz was more or less unchanged.
Both time series were taken when the craft was advancing forward with the speed  = 45 
in head sea waves with significant wave height estimated to be  = 03 m.
Figure 11.11 shows the full scale power spectra of the excess pressure variations at 1 = 10 m

469
Figure 11.10: Noncollocated sensor and actuator pair; full-scale power spectra of the vertical
accelerations at  = −5 m of a 35 m SES with ride control system on and off; 0 = 450 mm .

Figure 11.11: Noncollocated sensor and actuator pair; full-scale power spectra of the excess
pressure at  = 10 m of a 35 m SES with ride control system on and off; 0 = 450 mm .

470
Figure 11.12: Collocated sensor and actuator pair; full-scale power spectra of the vertical accel-
erations at  = −5 m of a 35 m SES with ride control system on and off; 0 = 430 mm .

in the air cushion with and without the ride control system activated. The time series were taken
from the same run as above. The pressure signal at 3 = −10 m was used in the feedback loop.
Hence, the louvers and sensors were completely noncollocated. With the ride control system
turned off, we observed response about 2 Hz and 5 Hz. Activating the ride control system the
response about 5 Hz was significantly amplified. At the resonance of the dynamic pressure about
2 Hz the response level was reduced by the ride control system.
Figure 11.12 shows the full scale power spectra of the vertical accelerations about 5 m aft
of the centre of gravity with and without the ride control system activated. In this case the
pressure signal at 1 = 10 m was used in the feedback loop. Hence, the louvers and sensors were
”almost” collocated since the louver were located at  = 8 m. With the ride control system
turned off, we observed response about 2 Hz, 5 Hz and 8 Hz. Activating the ride control system
the response about all three resonance frequencies were significantly reduced. These time series
were taken when the craft was advancing forward with the speed  = 44 knots in head sea
waves with significant wave height estimated to be  = 06 m.
Figure 11.13 shows the full scale power spectra of the excess pressure variations at 1 = 10 m
in the air cushion with and without the ride control system activated. The time series were
taken from the same run as in Figure 11.12. The pressure signal at 1 = 10 m was used in
the feedback loop. Hence, the louvers and sensors were ”almost” collocated. With the ride
control system turned off, we observed response about 2 Hz and 5 Hz at the fore end of the air
cushion. Activating the ride control system the response about all three resonance frequencies
were significantly reduced.

471
Figure 11.13: Collocated sensor and actuator pair; full-scale power spectra of the excess pressure
at  = 10 m of a 35 m SES with ride control system on and off; 0 = 430 mm .

472
Bibliography

[1] Aarseth, L. Anti-spin Control System for Marine Thrusters. Master Thesis. Department
of Marine Technology. NTNU, 2003.

[2] Aarset, M. F., J. P. Strand and T. I. Fossen. Nonlinear Vectorial Observer Backstepping
With Integral Action and Wave Filtering for Ships, Proc. of the IFAC Conf. on Control
Appl. in Marine Systems (CAMS’98), Fukuoka, Japan, October 27-30, 1998, pp. 83-89,
1998.

[3] American Bureau of Shipping (ABS). Guide for Thrusters and Dynamic Positioning Sys-
tems. New York, US, 1994.

[4] Agostinho, A. C., L. Moratelli Jr., E. A. Tannuri and H. M. Morishita. Sliding Mode Con-
trol Applied to Offshore Dynamic Positioning Systems. In Proc. of the IFAC International
Conference on Manoeuvring, Guarujá, Brazil, 2009.

[5] Allensworth, T. A Short History of Sperry Marine. http:\\www.sperry-

marine.com/pages/history.html, Litton Marine Systems, 1999.

[6] Aker Maritime. Ormen Lange TLP with full processing facilities. 37-00-NN-X15-00117,
2002.

[7] Anderson, B. D. O. and J. B. Moore. Optimal Control: linear quadratic methods. Prentice-
Hall, Englewood Cliffs, NJ, 1989.

[8] Asada H. and J. J. E. Slotine. Eds. Robot Analysis and Control. John Wiley and Sons,
NY, NY, 1986.

[9] Athans, M. and P. L. Falb. Eds. Optimal Control. McGraw-Hill Book Company, New York,
1966.

[10] American Petroleum Institute. Recommended Practice for Planning, Designing and Con-
structing Tension Leg Platforms. API RP 2T, Washinton D.C., USA, 1987.

[11] American Petroleum Institute (API). Recommended Practice for Design and Analysis of
Station keeping Systems for Floating Structures. Exploration and Production Department,
API Recommended Practice 2SK, First Edition, US, 1995.

[12] Bachmayer, R., L. Whitcomb and M. A. Grosenbaugh. An Accurate Four-Quadrant Non-

linear Dynamical Model for Marine Thrusters: Theory and Experimental Vaildation. IEEE
Journal of Oceanic Engineering, Vol. (25), No. (1), pp. 146-159, 2000.

473
[13] Bagchi, A. Optimal Control of Stochastic Systems. Prentice Hall International, UK, 1993.

[14] Balas, M. Feedback Control of Flexible Systems. IEEE Transactions on Automatic Control,
Vol. (23), No. 4, pp 673-679, 1978.

[15] Balchen, J. G., N. A. Jenssen and S. Sælid. Dynamic Positioning using Kalman Filtering
and Optimal Control Theory, IFAC/IFIP Symposium on Automation in Offshore Oil Field
Operation, Amsterdam, The Netherlands, pp. 183—186, 1976.

[16] Balchen, J. G., M. Fjeld, and O. A. Solheim. Reguleringsteknikk Bind 3. Tapir a.s., Trond-
heim, Norway, 1978, In Norwegian.

[17] Balasubramanyam, R., V.I. John and J.P. Tamby. Harmonic performance of cycloconverter
synchronous-motor marine propulsions systems. In proc. of the 1993 IEEE Industry Ap-
plications Conference Twenty-Eighth IAS Annual Meeting (IAS’93), Vol.(1), pp. 496-506,
1993.

[18] Balchen, J. G., N. A. Jenssen and S. Sælid. Dynamic Positioning of Floating Vessels based
on Kalman Filtering and Optimal Control. Proc. of the 19th IEEE Conf. on Decision and
Control, New York, US, pp. 852—864, 1980.

[19] Balchen, J. G., N. A. Jenssen, E. Mathisen and S. Sælid. A Dynamic Positioning System
Based on Kalman Filtering and Optimal Control. Modeling, Identification and Control.
Vol. (1), No. (3), pp. 135-163, 1980.

[20] Balchen, J. G. A Modified LQG Algorithm (MLQG) for Robust Control of Nonlinear
Multivariable Systems. Modeling, Identification and Control, Vol. (14), No. (3), pp. 175-
180, 1993.

[21] Balchen, J. G., T. Andresen and B. Foss. Reguleringsteknikk. 1. utgave - in Norwegian,

Tapir, Trondheim, Norge, 1999.

[22] Bennet, S. A History of Control Engineering 1800-1930. Peter Peregrinus, London, 1979.

[23] Berge, S. and A. Engseth and I. Fylling and C. M. Larsen and B. J. Leira and A. Olufsen.
Handbook on Design and Operation of Flexible Pipes. Structural Engineering, SINTEF,
STF70 A92006, Trondheim, Norway, 1992.

[24] Berntsen, P. I. B. Structural Reliability Based Position Mooring. PhD Thesis 2008:62,
Department of Marine Technology, Faculty of Engineering Science & Technology, NTNU,
Norway, 2008.

[25] Berntsen, P. I. B., O. M. Aamo, B. J. Leira and A. J. Sørensen. Structural Reliability-

Based Control of Moored Interconnected Structures. Control Engineering Practice, CEP
16-4, pp. 495-504, 2008.

[26] Bertin, D., S. Bittanti, S. Meroni and S. M. Savaresi. Dynamic Positioning of a "Single-
Thruster" Vessel by Feedback Linearization. In Proc. of the IFAC Conference on Manoeu-
vring and Control of Marine Craft, Aalborg, Denmark, 2000.

474
[27] Bishop, G.N., N.A Shelley and M.J.S. Edmonds. Electric propulsion of surface fighting
ships. In proc. of Eleventh Ship Control Systems Symposium, Vol.(2), pp. 157-171, April
1997.

[28] Blanke, M. Ship Propulsion Losses Related to Automatic Steering and Prime Mover Con-
trol. Ph. D. Thesis, Servolaboratory, Technical University of Denmark, Lyngby, 1981.

[29] Blanke, M. and Busk Nielsen, P. The Marine Engine Governor. In Proc. 2nd Int. Conf. on
Maritime Communications and Control, pp 11—20, London, UK, 1990.

[30] Blanke, M., K. P. Lindegaard and T. I. Fossen. Dynamic Model for Thrust Generation
of Marine Propellers. In proc. of the IFAC Conf. on Manoeuvring and Control of Marine
Craft (MCMC’2000), Aalborg, Denmark, 2000.

[31] Blanke, M., M. Kinnaert, J. Lunze and M. Staroswiecki. Diagnostics and Fault-Tolerant
Control. Springer-Verlag, Berlin, Germany, 2003.

[32] Blendermann, W. Die Windkräfte am Shiff. Technical Report Bericht Nr. 467. Institut für
Schiffbau der Universität Hamburg, Germany, 1986.

[33] Blokland, A.J. and B. van der Ploeg. Electric ship propulsion. In Proc. of EPE ’95. 6th
European Conference on Power Electronics and Applications, Vol. (3), pp. 29-32, 1995.

[34] Brevik, M.. Topics in Guided Motion Control. PhD Thesis 2010:63, Department of En-
gineering Cybernetics, Faculty of Information Technology, Mathematics and Electrical
Engineering, NTNU, Norway, 2010.

[35] Britting, K. R. Inertial Navigation System Analysis. Wiley Interscience, 1971.

[36] Bose, B.K. Power Electronics and AVC drives. Prentice-Hall, IEEE press, 1997.

[37] Bryson, A. E. and Y. C. Applied Optimal Control. Ginn and Co., Boston, US, 1969.

[38] Larsen, C. M., C. White, I. Fylling and N. Nordsve. Nonlinear static and dynamic behavior
of tension leg platform. In Proc. of the Conf. of Offshore Mech. and Arctic Eng., 1984.

[39] Larsen, C. M. Analysis of Tensioner System Stroke for Marine Risers. SINTEF Structures
and Concrete, STF70 F93011, Norway, 1993.

[40] Larsen, C. M. Tension Leg Platforms. Lecture notes NA550/665, Dept. of Naval Architec-
ture and Marine Engineering, University of Michigan, 1995.

[41] Canudas de Wit, C. and P. Tsiotras. Dynamic Tire Friction Models for Vehicle Traction
Control. In Proc. of the 38th Conference on Decision and Control (CDC’99), Phoenix,
Arizona, USA, 1999.

[42] Carlton, J. S. Marine Propellers and Propulsion. Oxford: Butterworth-Heinemann, UK,

1994.

[43] Caricchi, F., F. Crescimbini and O. Honorati. Modular, axial-flux, permanent-magnet

motor for ship propulsion drives. In proc. of IEEE International Electric Machines and
Drives Conference Record, pp. WB2/6.1- WB2/6.3, 1997.

475
[44] Cegelec Projects. Power and propulsion system for RRS James Clark Ross. Cegelec Con-
trols, Publication No. 499, 1994.

[45] Cegelec Projects. The electrical propulsion system of the QE2. Cegelec Controls, Publica-
tion No. 498, 1994.

[46] Chang, S.C. and R. Yacamini. The effect of cycloconverter drives on noise and vibration
in electrical machines. In proc. of EPE ’95. 6th European Conference on Power Electronics
and Applications, Vol.(3), pp. 205-210 , 1995.

[47] Chen, Chi-Tsong. Linear System Theory and Design. Third Edition, Oxford University
Press, Oxford, New York, US, 1999.

[48] Chen, Chi-Tsong. Signals and Systems. Third Edition, Oxford University Press, Oxford,
New York, US, 2004

[49] Cheney, W. and D. Kincaid. Numerical Mathematics and Computing. Second Edition,
Brooks/cole Publishing Company, Moterey, California, US, 1985.

[50] Cosulich, G., M. Fracchia, A. Mariscotti and S. Savio. Compararive dependability analysis
of electric ship propulsion systems. In proc. of EPE ’95. 6th European Conference on Power
Electronics and Applications, Vol.(3), pp. 328-33, 1995.

[51] Craig, J. J. Introduction to Robotics. Addison-Wesley, Reading, Massachusetts, 1989.

[52] Demirbilek, Z. Tension Leg Platform - A state of the art review. American Society of Civil
Engineers, 1989.

[53] Desoer, C. A., and M. Vidyasager. Feedback Systems: Input-Output Properties. Academic
Press, New York, US, 1975.

[54] Det Norske Veritas (DNV). Rules for Classification of SHIPS Newbuildings, Special Equip-
ment and Systems. Additional Class, Part 6, Chapter 7: Dynamic Positioning Systems
(DP). Norway, July 2010.

[55] Det Norske Veritas (DNV). Rules and Regulations of Mobile Offshore Units, Special Equip-
ment and Systems. Additional Class, Part 6, Chapter 2: Position Mooring (POSMOOR).
Norway, 1996.

[56] Det Norske Veritas (DNV). Rules for Planning and Execution of Marine Operations, Ho-
evik, Norway, january 2000 edition, January 2000.

[57] DNV. Recommended Practice for Riser Interference. Det Norske Veritas, DNV-RP-F203,
2005.

[58] Doerry, N.H., and J.C. Davies. Integrated power system for marine applications. Naval
Engineers Journal, pp 77-90, May 1994.

[59] Dorf, R. C. and R. H. Bishop. Modern Control Systems. Eight Edition, Addision-Wesley,
California, US, 1998.

476
[60] Durham, W. C. Constrained Control Allocation. Journal of Guidance, Control and Dy-
namics, Vol(16), No.(4), 1993.

[61] Christiansen, D. Electronics Engineer’s Handbook. Fourth Edition, IEEE Press, McGraw-
Hill, New York, US, 1996.

[62] Edwards Jr., C. H. and D. E. Penney. Differential Equations and Boundary Value Problems
- Computing and Modeling. Prentice-Hall, Inc., New Jersey, US, 1996.

[63] Egeland, O. Servoteknikk. (In Norwegian) Tapir Forlag, Trondheim, Norway, 1993.

[64] Egeland, O. and J. T. Gravdahl. Modeling and Simulation for Automatic Control. Marine
Cybernetics, Trondheim, 2002.

[65] Engseth, A., Bech, A. and C. Larsen. Efficient Method for Analysis of Flexible Risers. In
Proc. of BOSS’88, Behaviour of Offshore Structures, Trondheim, Norway, 1988

[66] Faltinsen, O. M. Sea Loads on Ships and Offshore Structures. Cambridge University Press,
Cambridge, UK, 1990.

[67] Faltinsen, O. M. and B. Sortland. Slow Drift Eddy Making Damping of a Ship. Applied
Ocean Reserach, Vol.( 9), No. (1), pp37-46, 1987.

[68] Faltinsen, O. M. and R. Zhao. Numerical predictions of of ship motions at high forward
speed. Phil. Trans. R. Soc. Series A, pp 241-252, 1991a.

[69] Faltinsen, O. M. and R. Zhao. Flow predictions around high speed ships in waves. Math-
ematical Approaches in Hydrodynamics, SIAM. 1991b.

[70] Faltinsen, O. M., J. B. Helmers, K. J. Minsaas and R. Zhao. Speed Loss and Operability
of Catamarans and SES in Seaway. In Proc. First International Conference on Fast Sea
Transportattion-FAST’91, Trondheim, Norway, 1991.

[71] Faltinsen, O. M., J. Hoff, J. Kvålsvold and R. Zhao. Global loads on high-speed catama-
rans. PRADS’92, Newcastle, England, 1992.

[72] Farmer, R. Twin 40-MW gensets give Terra Nova FPSO production/propulsion options.
Gas Turbine World, pp. 16-20, September-October 1998.

[73] Farrell, J. A. and M. Barth. The Global Positioning System & Inertial Navigation.
McGraw-Hill, 1998.

[74] Fay, H. Dynamic Positioning Systems, Principles, Design and Applications. Editions Tech-
nip, Paris, France, 1989, ISBN : 2-7108-0580-4

[75] Finne, S. and J. Grue. On the Complete Radiation-diffraction Problem and Wave-drift
Damping of Marine Bodies in the Yaw Mode of Motion. J. Fluid Mechanics, Vol. (357),
pp289-320, 1998.

[76] Fisher, K.P. and L. Moore. EMC requirements for ships. Ninth International Conference
on Electromagnetic Compatibility, pp. 161-167, 1994.

477
[77] Fitzgerald, A. E., C. H. Kingsley and S. D. Umans. Electric Machinery. Fifth edition,
McGraw-Hill, 1992.

[78] Fossen, T. I. Guidance and Control of Ocean Vehicles. John Wiley & Sons Ltd., Chichester,
England, 1994.

[79] Fossen, T. I., S. I. Sagatun and A. J. Sørensen. Identification of Dynamically Positioned

Ships. Control Engineering in Practice, CEP 4-3, pp. 369-376, 1996.

[80] Fossen, T. I. Marine Control Systems, Guidance, Navigation, and Control of Ships, Rigs
and Underwater Vehicles. Marine Cybernetics, Trondheim, Norway, 2002.

[81] Fossen, T. I. Handbook of Marine Craft Hydrodynamics and Motion Control. ISBN
9781119991496. John Wiley & Sons Ltd, UK, 2010.

[82] Fossen, T. I. and Å. Grøvlen. Nonlinear Output Feedback Control of Dynamically Posi-
tioned Ships Using Vectorial Observer Backstepping. IEEE Transactions on Control Sys-
tems Technology, TCST-6(1) :pp. 121-128, 1998.

[83] Fossen, T. I. and J. P. Strand. Passive Nonlinear Observer Design for Ships Using Lyapunov
Methods: Experimental Results with a Supply Vessel. Automatica, Vol. (35), No. (1), pp.
3-16, 1999.

[84] Fossen, T. I. and J. P. Strand. Nonlinear Passive Weather Optimal Positioning Control
(WOPC) System for Ships and Rigs: Experimental Results, Automatica, Vol. (37), No.
(5), pp. 701-715, 2001.

[85] Fossen, T. I., A. Loria and A. Teel. A Theorem for UGAS and ULES of Nonautonomous
Systems: Robust Control of Mechanical Systems and Ships. International Journal of Ro-
bust and Nonlinear Control, Vol. (11). pp. 95-108, 2001.

[86] Fossen, T. I. and M. Blanke. Nonlinear Output Feedback Control of Underwater Vehicle
Propellers Using Feedback from Estimated Axial Flow Velocity. IEEE Journal of Oceanic
Engineering, Vol. (25). No. (2), April, 2000.

[87] Fossen, T. I. and J. P. Strand. Position and Velocity Observer Design. In The Ocean
Engineering Handbook. (F. El-Hawary, Ed.). CRC Press, Florida, US, pp. 189-206, 2000.

[88] Fossen, T. I. and J. P. Strand. Weather Optimal Positioning Control. In The Ocean Engi-
neering Handbook. (F. El-Hawary, Ed.). CRC Press, Florida, US, pp. 218-234, 2000.

[89] Fossen, T. I. and T. Perez. Kalman Filtering for Positioning and Heading Control of Ships
and Offshore Rigs. IEEE Control Systems Magazine. Special Issue for the 50th Anniversary
of the Kalman-Bucy Filter. CST-29(6):32-46, 2009.

[90] Fracchia, M. and Sciutto, G. Cycloconverter drives for ship propulsion. In proc. of Power
Electronics, Electrical Drives, Advanced Electrical Motors, Symposium, Vol.(1), pp. 255-
260, 1994.

[91] Fung, P. T-K. and M. Grimble. Dynamic Ship Positioning Using Self-Tuning Kalman
Filter. IEEE Transaction on Automatic Control, Vol.(28), No. (3), pp. 339-349, 1983.

478
 [92] Gelb, A. and W. E. Vander Velde. Multiple-Input Describing Functions and Nonlinear
System Design. McGraw-Hill, New York, US, 1968.

[93] Gevarter, W. B. Basic relations for control of flexible vehicles. AIAA J. Vol.(8), pp. 666-
672. 1970.

[94] Gelb, A. (Ed.). Applied Optimal Estimation. MIT Press, Boston, MA, 1988.

[95] Goodwin, G. C., S. F. Graebe and M. E. Salgado. Control System Design. Prentice Hall,
New Jersey, US, 2001.

[96] Greenhow, M. and L. Yanbao. Added masses for circular cylinders near or penetrating
fluid boundaries - review, extension and application to water-entry, -exit and slamming.
J. of. Ocean Engng., Vol. 14, No. 4, pp. 325—348, 1987.

[97] Grimble, M. J., R. J. Patton and D. A. Wise. The Design of Dynamic Ship Positioning
Control Systems Using Stochastic Optimal Control Theory. Optimal Control Applications
& Methods, Vol. (1), pp. 167-202, 1980a.

[98] Grimble, M. J., R. J. Patton and D. A. Wise. Use of Kalman Filtering Techniques in
Dynamic Ship Positioning Systems. IEE Proc., Vol. (127), No. (3), Pt. D, pp. 93-102,
1980b.

[99] Grimble, M. J. and M. A. Johnson. Optimal Control and Stochastic Estimation. Vol (1).
John Wiley and Sons Ltd., 1988.

[100] Grimble, M. J. and M. A. Johnson. Optimal Control and Stochastic Estimation: Theory
and Applications. Vol (2). John Wiley and Sons Ltd., 1988.

[101] Grøvlen, Å. , A. J. Sørensen and M. F. Aarset. Aspects of Dynamic Positioning: Impacts

of Signal Processing and Local Thrust Control in an Integrated Power, Automation and
Positioning System. IMCA 98-Station Keeping Seminar & Workshop, Rio de Janeiro,
Brazil, 1998.

[102] Hackman, T. Electric propulsion systems for ships. ABB Review, No.3 pp. 3-12, 1992.

[103] Haddad,W. M., D. S. Bernstein and W. Wang. Dissapitive H2 /H∞ controller synthesis. In
Proc. Am. Contr. Conf. San Fransisco. CA. 1993.

[104] Hansa-Sciffahrt. Shuttle-tankers for Norwegian oilfield Heidrun. Hansa — Schiffart —

Schiffbau- Hafen, Vol.(131), No. 8, pp. 42-43, 1994.

[105] Hansen, J. F. Optimal Power Control of Marine Systems. Ph.D. thesis, Dept. of Engineer-
ing Cybernetics, Norwegian University of Science and Technology, Trondheim, Norway,
2000.

[106] Hansen, J. F., A. W. Ordys and M. J. Grimble. A toolbox for simulation of multilayer op-
timisation system with static and dynamic load distribution. In proc. of IFAC Symposium
on Large Scale Systems (LSS’98), Vol.(2), pp. 857-862, 1998.

479
[107] Hansen, J. F., A. K. Ådnanes and T. I. Fossen. Modelleing, simulation and multivari-
able model-based predictive control of marine power generation system. In proc. of IFAC
Conference: Control Applications in Marine Systems (CAMS’98), pp. 45-50, 1998.

[108] Hansen, J. F., and T. I. Fossen. Nonlinear control of marine power generation systems. In
proc. of 13th Power System Computation Conference (PSCC’99), pp. 539-544, 1999.

[109] Hansen, J. F., and T. Lauvdal and A. K. Ådnanes. Modelling and simulation of variable
speed thruster drives with full-scale verification. In: Proc. of 5th IFAC Conf. on Manoeu-
vring and Control of Marine Craft — MCMC’2000, Aalborg, Denmark, 2000.

[110] Haskara, İ., Ü. Özgüner, J. Winkelman. Wheel slip control for antispin acceleration via
dynamic spark advance. Control Engineering Practice, 8, 2000.

[111] Hazelton, D.W. HTS coils for the Navy’s superconducting homopolar motor/generator.
IEEE Transactions on Applied Superconductivity, Vol.(7) No. 2, pt.1, pp. 664-667, 1997.

[112] Henriksen, R. Stokastiske systemer - Analyse, estimering og regulering. Institutt for teknisk
kybernetikk, NTNU, 1998 (In Norwegian).

[113] Healey, A. J., S. M. Rock, S. Cody, D. Miles and J. P. Brown. Toward an Improved
Understanding of Thrust Dynamics for Underwater Vehicles. IEEE Journal of Oceanic
Engineering, Vol. (20). No. (4), pp. 354-360, 1995.

[114] Hespanha J. P. Tutorial on Supervisory Control, Lecture Notes for the Workshop Control
using Logic and Switching for the Conference on Decision and Control, Orlando, Florida,
US, 2001.

[115] Hespanha J. P. and A. S. Morse. Switching Between Stabilizing Controllers. Automatica,

38(11), pp. 1905—1917, 2002.

[116] Hespanha J. P., Liberzon D. and A.S. Morse. Hysteresis-based Switching Algorithms for
Supervisory Control of Uncertain Systems. Automatica, 39-2, pp. 263—272, 2003.

[117] Hill, W.A., G. Creelman and L. Mischke. Control strategy for an icebreaker propul-
sion system. IEEE Transactions on Industry Applications, Vol.(28), No 4, pp. 887-892,
July/August 1992.

[118] Hiraide, M., E. Kaneda, T. Sato and S. Tsuchiyama. Dynamic simulator for PEFC propul-
sion plant. FUEL CELL. 1996 Fuel Cell Seminar. Program and Abstracts, pp. 327-330,
November 1996.

[119] Hoffman, J. D. Numerical Methods for Engineers and Scientists. Second edition, Marcel
Dekker, Inc.,New York, NY 10016, USA, 2001.

[120] Hopkins, J. W. Fantasy and reality. Trans IMarE, Vol.(103), pp 193-220, 1991.

[121] Horne, C. D., A. J. Whitehead and D. Webster. Naval electric power control and moni-
toring systems — a view of the future. In proc. of Eleventh ship control symposium, UK,
pp. 339-454, April 1997.

480
[122] Huse, E. and K. Matsumoto. Mooring Line Damping Due to First- and Second-order Vessel
Motion. In Proc. 21 Annual Offshore Technology Conference, Houston, Texas, Vol. (4),
pp. 135-148, 1989.
[123] Huse, E. Interaction in Deep-Sea Riser Arrays. In Proc. of the 25th Annual Offshore
Technology Conference, OTC, Houston, TX, 1993.
[124] Huse, E. and G. Kleiven. Impulse and Energy in Deepsea Riser Collisions Owing to Wake
Interference. In Proc. of the 32nd Annual Offshore Technology Conference (OTC), Hous-
ton, TX, 2000.
[125] IEEE. Recommended practice for electric installations on shipboard. IEEE Industry Ap-
plications Society, std 45, 1998.
[126] Ihle, I.-A. F. Nonlinear Cooperated Control of Marine Craft. PhD Thesis, Department of
Engineering Cybernetics, Faculty of Information Technology, Mathematics and Electrical
Engineering, NTNU, Norway, 2006.
[127] Imakita, A., S. Tanaka, Y. Amitani and S. Takagawa . Intelligent Riser Angle Control
DPS. In: Proc. of ETCE/OMAE2000. Joint Conf., OMAE2000/OSU OFT-30001, New
Orleans, US, 2000.
[128] Isherwood, M. A. Wind resistance of Merchant Ships. Trans. Inst. Naval Arch., RINA Vol.
(115), pp. 327-338, 1972.
[129] Jackson, K. R. and N. S. Nedialkov. Some Recent Advances in Validated Methods for
IVPs for ODEs. To appear in Journal of Applied Numerical Mathematics, 2002.
[130] Jensen, G. A. Offshore Pipelaying Dynamics. PhD Thesis, Department of Engineering
Cybernetics, Faculty of Information Technology, Mathematics and Electrical Engineering,
NTNU, Norway, 2010.
[131] Jenssen, N. A. Estimation and Control in Dynamic Positioning of Vessels. NTH (NTNU),
Norway, 1981.
[132] Johannessen, K., T. S. Meling, S. Haver. Joint Distribution for Wind and Waves in the
Northern North Sea. IJOPE Vol. 12, No. 1, March 2002.
[133] Johansen, T. A., J. Kalkkuhl and I. Petersen. Hybrid Control Strategies in ABS. In. Proc.
of American Control Conference (ACC’01), Arlington, VA, USA, 2001.
[134] Johansen, T. A., T. P. Fugleseth, P. Tøndel, and T. I. Fossen. Optimal Constrained Control
Allocation in Marine Surface Vessels with Rudders. In Proceedings of 6th Conference on
Manoeuvring and Control of Marine Crafts (MCMC2003), September 16-19, 2003, Girona,
Spain.
[135] Johansen, T. A., T. I. Fossen, B. Vik. Hardware-In-the-Loop Testing of DP systems. DP
Conference, Houston, US, 2005.
[136] Johansen, T. A., A. J. Sørensen, O. J. Nordahl, O. Mo and T. I. Fossen. Experiences
from Hardware-In-the-Loop (HIL) Testing of Dynamic Positioning and Power Management
Systems. OSV Singapore, 2007.

481
[137] Johansen, T. J. and A. J. Sørensen. Experiences with HIL Simulator Testing of Power Man-
agement Systems. Dynamic Positioning Conference, Marine Technology Society, Houston,
US, 13-14 October, 2009.

[138] Johansson, R. System Modeling & Identification. Prentice Hall, Englewood Cliffs, New
Jersey, US, 1993.

[139] Joshi, S. M. Control of Large Flexible Space Structures. Springer, Berlin, Germany, 1989.

[140] Joshi, S. M. and P. G. Maghami. Dissipative compensators for flexible spacecraft control.
In Proc. Am. Contr. Conf. San Diago, CA. 1990.

[141] Joshi, S. M., P. G. Maghami and A. G. Kelkar. Dynamic dissipative compensator design
for large space structures. AIAA Guidance, Navigation, Contr. Conf. New Orleans, LA,
US, 1991.

[142] Juby, L., R. Bucknall and N. A. Haines. A harmonic reduction technique for cycloconverter
propulsion drives. In proc. of Seventh International Conference on Electrical Machines and
Drives (Conf. Publ. No.412), pp. 333-337, 1995.

[143] Kalleklev, A. J. and K. J. Mørk and N. Sødahl and M. K. Nygård and A. M. Horn.
Design Guidelines for Riser Collision. In Proc. of the 35th Annual Offshore Technology
Conference, OTC, Houston, TX, 2003.

[144] Kaplan, P and S. Davis. A simplified representation of the vertical plane dynamics of SES
craft. AIAA Paper No. 74-314, AIAA/SNAME Adv. Marine Vehciles Conf. San Diego,
CA, US, 1974.

[145] Kaplan, P. and S. Davis. System analysis techniques for designing ride control system for
SES craft in waves. 5th Ship Contr. Syst. Symp., Annapolis, MD, US, 1978.

[146] Kaplan, P., J. Bentson and S. Davis. Dynamics and hydrodynamics of surface effect ships.
Trans. SNAME. Vol.(89), pp. 211-247. 1981.

[147] Katebi, M. R., M. J. Grimble and Y. Zhang. H∞ Robust Control Design for Dynamic Ship
Positioning. In IEE Proc. on Control Theory and Appl., 144, 110-120, 1997a.

[148] Katebi, M. R., Y. Zhang and M. J. Grimble. Nonlinear Dynamic Ship Positioning. In Proc.
of the 13th IFAC World Congress. Q, 303-308, 1997b.

[149] Keener, J. P. Pronciples of Applied Mathmathics. Addison-Wesly, Redwood City, CA, US,
1988.

[150] Khalil, H. K. Nonlinear Systems. Prentice Hall, New Jersey, US, 1996.

[151] Kijima, K., K. Murata and Y. Furukawa. Design of a Dynamic Positioning System for
Self-Propulsive Barge under External Disturbances. In Proc. of the IFAC Conference on
Control Applications in Marine Systems, Fukuoka, Japan, 1998.

[152] Kirk, D. E. Optimal Control Theory. Prentice Hall, Englewood Cliffs, New Jersey 07632,
1970.

482
[153] Kreyszig, E. Advanced Engineering Mathematics. 7th edition, John Wiley and Sons, Inc.,
1993.

[154] Krstic, M., I. Kanellakopoulos and P. V. Kokotovic. Nonlinear and Adaptive Control De-
sign. John Wiley & Sons Ltd., New York, US, 1995.

[155] Kristiansen, E. and O. Egeland. Frequency-dependent Added Mass in Models for Con-
troller Design for Wave motion Damping. In Proc.of 6th Conference on Manoeuvring and
Control of Marine Crafts, Girona, Spain, 2003.

[156] Kumm, W.H. Diesel fueled ship propulsion fuel cell demonstration project. FUEL CELL.
1996 Fuel Cell Seminar. Program and Abstracts, pp. 342, November 1996.

[157] Kurimo, R. Sea trial experience of the first passenger cruiser with podded propulsors. In
proc. of 7th International Symposium on Practical Design of Ships and Mobile Units, pp
743-748, September 1998.

[158] Lamb, H. Hydrodynamics. Cambridge University Press, London, UK, 1932.

[159] van Lammeren, W. P. A., van Manen, J. D., & Oosterveld, M. W. C. The Wageningen
B-screw series. Transactions of SNAME, 77, 269—317, 1969.

[160] Larsen, C. M. Dynamic Analysis of Structures. Dept. of Marine Structures, NTNU, 2000.

[161] Laukia, K. Service proves electric propulsion design. The Motor Ship, pp. 23-26, February
1993.

[162] Lauvdal, T., A. J. Sørensen, A. K. Ådnanes, J. P. Strand,J. F. Hansen and O. J. Sørdalen.

Marintronics: Optimizing Marine Power and Automation Systems Through Industrial IT.
ABB Review, No. 1/2000.

[163] Lehn, E. Practical Methods for Estimation of Thrust Losses. Report MT51A92-003,
513003.00.06, Marintek, Trondheim, Norway, 1992.

[164] Leidi, G., and D. Zaninelli. Quality of power in ships with electric propulsion. In proc. of
ICHQP. 7th International Conference on Harmonics and Quality of Power, pp. 140-146,
Oct. 1996.

[165] Leira, B. J., A. J. Sørensen and C. M. Larsen. A Reliability-based Control Algorithm for
Dynamic Positioning of Floating Vessels. Journal of Structural Safety, Elsevier Ltd., Vol.
26, pp. 1-28, 2004.

[166] Leira, B. J., A. J. Sørensen, P. I. Berntsen and O. M. Aamo. Structural Reliability Cri-
teria and Dynamic Positioning of Marine Vessels. International Journal of Materials &
Structural Reliability, 4-2, pp. 161-174, 2006.

[167] Leite, M. S., M.A. Waythe and N.A. Haines. Propulsion design and predicted efficiency
package. In proc. of 28th Universities Power Engineering Conference, Vol.(1), pp. 413-416,
1993.

[168] Leonard, J.W. Tension Structures - Behaviour and Analysis. McGraw-Hill Inc., 1988.

483
[169] Leonhard, W. Control of Electrical Drives, 2nd edn, Springer, Berlin, 1996.

[170] Levine, W. S. The Control Handbook. CRC Press in cooperation with IEEE Press, CRC
Press, Florida, US, 1996.

[171] Lindegaard, K.-P. W. Acceleration Feedback in Dynamic Positioning Systems. PhD Thesis,
Report 2003:4-W, Department of Engineering Cybernetics, NTNU, 2003.

[172] Little , G. T., K. R. Cadd and M. Edmonds. AC AC tandem generators — the flexible,
cost effective power generation solution. Trans. IMarE, Vol.(107), Part 1, pp.
13-26, 1995.

[173] Lloyd’s Register of Shipping (LRS). Rules and Regulations for the Classification of Ships.
Part 7, Chapter 8: Positional Mooring and Thruster Assisted Positional Mooring Systems.
UK, 1997.

[174] Lloyd’s Register of Shipping (LRS). Rules and Regulations for the Classification of Ships.
Part 7, Chapter 4: Rules for the Construction and Classification of Dynamic Positioning
Systems Installed in Ships. UK, 1997.

[175] Loria, A., T. I. Fossen and A. Teel. UGAS and ULES of Non-Autonomous Systems:
Applications to Integral Action Control of Ships and Manipulators. Proceedings of the 5th
European Control Conference (ECC’99), Karlsruhe, Germany, September 1999.

[176] Loria, A., T. I. Fossen and E. Panteley. A Cascaded Approach to a Separation Principle for
Dynamic Ship Positioning, IEEE Transactions of Control Systems Technology, to appear
in 2000.

[177] Lozano-Leal, R and S. M. Joshi. On the design of dissipative LQG-type controllers. In

Proc. 27th IEEE Conf. Decision Contr., TX. 1988.

[178] Mahon, L. L. J. Diesel Generator Handbook. Butterworth and Heinemann, 1992/1996.

[179] Using MATLAB. Version 6, The MathWorks Inc., 2001.

[180] Mandzuka, S. and Z. Vukic. Use of Optimal Constrained Covariance Control (OC3 ) in
Dynamic Positioning. In Proc. of the IFAC Conference on Control Applications in Marine
Systems, Trondheim, Norway, 1995.

[181] MacDonald, J. New propulsion control system for the C.S.S. Hudson. Canadian Maritime
Industries association, 45th annual technical conference, section B3.

[182] McCoy, T. J. and J.L. Kirtley Jr. Thermosyphon-cooled axial gap electric motors for
ship propulsion applications. In proc. of ICEM96 International Conference on Electrical
Machines, Vol.(2), pp. 381-386, 1996.

[183] Mead, D. J. Passive Vibration Control. John Wiley and Sons Ltd, West Sussex UK, 1999.

[184] Meirovitch, L. Principles and Techniques of Vibrations. Prentice Hall, NJ, USA, 1997,
ISBN 0-13-270430-7.

[185] Merritt, H. Hydraulic Control Systems. John Wiley and Sons Ltd, NY, NY., 1967.

484
[186] Milne-Thomson, L. M. Theoretical Hydrodynamics. MacMillan Education Ltd., London,
UK, 1968.

[187] Minsaas, K. J., O. M. Faltinsen and B. Persson. On the Importance of Added Resistance,
Propeller Immersion and Propeller Ventilation for Large Ships in Seaway. In Proc. of the
3rd International Symposium on Practical Design in Shipbuilding (PRADS’83), Tokyo &
Seoul, pp. 149-159, 1983.

[188] Minsaas, K. J., H. J. Thon, and W. Kauczynski. Influence of Ocean Environment on

Thruster Performance. In proc.: Int. Symp. Propeller and Cavitation, supplementary vol-
ume, pp. 124-142. Shanghai: The Editorial Office of Shipbuilding of China, 1986.

[189] Minsaas, K. J., H. J. Thon and W. Kauczynski. Estimation of Required Thruster Capacity
for Operation of Offshore Vessels under Severe Weather Conditions. PRADS 1987.

[190] Minorsky, N. Directional Stability of Automatic Steered Bodies. J. Amer. Soc. of Naval
engineers, 34(2):280-309, 1922.

[191] Mohan, N. and T. M. Underland and W. P. Robbins. Power Electronics: Converters,

Applications and Design. John Wiley& Sons, 1989.

[192] Morishita, H. M. and B. J. J. Cornet. Dynamics of a Turret-FPSO and Shuttle Vessel due
to Current. In Proc. of the IFAC Conference on Control Applications in Marine Systems,
Fukuoka, Japan, 1998.

[193] Morishita, H. M., E. A. Tannuri and T. T. Bravin. Methodology for Dynamic Analsyis of
Offloading Operations. In Proc. of the IFAC Conference on Control Applications in Marine
Systems, Ancona, Italy, 2004.

[194] Myrhaug, D. Irregular seas. Lecture notes, SIN1015 Marine Dynamics, NTNU, 2000.

[195] Marine Systems Simulator (MSS), http://www.marinecontrol.org, Norwegian University

of Science and Technology, Trondheim, 2011

[196] Nestegård, A. Motions of surface effect ships. Vertias Research Report No.:90-2011. 1990.

[197] Newland, D.E. An Introduction to Random Vibrations and Spectral Analysis. Longman
Group Ltd, Essex UK., 2nd. edition, 1975, ISBN 0-582-30530-6.

[198] Newman, J. N. Second-Order, Slowly-varying Forces on Vessels in Irregular Waves. Proc.

Int. Symp. Dynamics of Marine Vehicles and Structures in Waves, Mechanical Engineering
Publication Ltd., pp. 182-186, London, 1974.

[199] Newman, J. N. Marine Hydrodynamics. MIT Press, Cambridge, MA, US, 1977.

[200] Nielsen, F. G. SIN1546 Marine Operasjoner. Faculty of Marine Technology - The Norwe-
gian University of Science and Technology, Trondheim, Norway, 2000, report UK-01-86,
in Norwegian.

[201] Nguyen, T. D. Design of Hybrid Marine Control Systems for Dynamic Positioning. De-
partment of Civil Engineering, National University of Singapore (NUS), Singapore, 2006.

485
[202] Nguyen, T. D., A. J. Sørensen, and S. T. Quek. Design of Hybrid Controller for Dynamic
Positioning from Calm to Extreme Sea Conditions. Automatica, 43-5. pp. 768-785, 2007.

[203] Nguyen T. D., A. J. Sørensen and S. T. Quek. Multi-Operational Hybrid Controller Struc-
ture for Station Keeping and Transit Operations of Marine Vessels. IEEE Transaction on
Control System Technology, Vol. 16 No. (3), pp. 491-498, 2008.

[204] Nguyen, D. T., A. H. Sørbø and A. J. Sørensen (2009). Modelling and Control for Dynamic
Positioned Vessels in Level Ice. In Proc. of Conference on Manoeuvring and Control of
Marine Craft, pp. 229-236, September 16-18. 2009, Guarujá, Brazil.

[205] Nguyen, T. D. and A. J. Sørensen. Setpoint Chasing for Thruster-Assisted Position Moor-
ing. IEEE Journal of Oceanic Engineering, Vol. 34, No. 4, pp.548-558, 2009.

[206] Nguyen, D. T. and A. J. Sørensen. Switching Control for Thruster-Assisted Position Moor-
ing. Control Engineering Practice, CEP 17-9, pp. 985-994, 2009.

[207] NORSOK. NORSOK STANDARDS, http://www.nts.no/norsok.

[208] NORSOK. Norsok Standard - Marine Operations, J-003. NORSOK - Norwegian Technol-
ogy Standard Institution, rev.2. edition, 1997, http://www.nts.no/norsok.

[209] NORSOK. Norsok Standard - Subsea Intervention, U-007. NORSOK - Norwegian Tech-
nology Standard Institution, rev.2. edition, 1998, http://www.nts.no/norsok.

[210] NORSOK. Norsok Standard - Subsea Production Systems, U-003. NORSOK - Norwegian
Technology Standard Institution, rev.2. edition, 1998, http://www.nts.no/norsok.

[211] NORSOK. Norsok Standard - Actions and Actions Effect, N-003. NORSOK - Norwegian
Technology Standard Institution, rev.1. edition, 1999, http://www.nts.no/norsok.

[212] Nowlin, W. D., A. E. Jochens, S. T. DiMarco, R. O. Reid and M. K. Howard. Deepwater

physical oceanography reanalysis and synthesis of historical data. Dept. of Oceanography,
Texas A & M Univeristy, MMS 2001-064, 2001.

[213] OCIMF. Prediction of Wind and Current Loads on VLCCs. Oil Companies International
Marine Forum, Witherby & Co. Ltd., London, UK, Second Edition, 1994.

[214] Ono, R. and S. Tsuchiyama. Study on a PEFC propulsion system for surface ships. FUEL
CELL, 1996 Fuel Cell Seminar. Program and Abstracts, pp. 517-520, November 1996.

[215] Oppenheim, A. V., A. S. Willsky with S. H. Nawab. Signals and Systems. Second Edition,
Prentice Hall, New Jersey, US, 1997.

[216] Oosterveld, M. W. C. and P. van Oossanen. Further computer-analyzed Data of the Wa-
geningen B-screw Series. Int. Shipbuilding Progress, Vol. (22), pp. 251-262, 1975.

[217] Overå, E. S. A theoretical and experimental study of a ventilating ducted propeller. Master
thesis, Department of Marine Hydrodynamics, NTNU, June 2003.

[218] Paice, D.A. Power Electronic Converter Harmonics. IEEE press, 1995.

486
[219] Parker, D. S. and C. G. Hodge. The electric warship. Power Engineering Journal, pp. 5-13,
February 1998.

[220] Parkinson, B. W. and J. J. Spiker (eds.). The Global Positioning System: Theory and
Applications. American Institute of Aeronautics and Astronautics, Reston, VA, 1996.

[221] Perez, T., A. J. Sørensen and M. Blanke. Marine Vessel Models in Changing Operational
Conditions –A Tutorial. 14th IFAC (International Federation of Automatic Control) Sym-
posium on System Identification, SYSID, Newcastle, Australia, 2006.

[222] Perez, T. and A. Donaire. Constrained Control Design for Dynamic Positioning of Marine
Vehicles with Control Allocation Modelling Identification and Control, MIC 30- 2, pp.
57-70, 2009.

[223] Pettersen, K. Y. and T. I. Fossen. Underactuated Dynamic Positioning of a Ship - Experi-

mental Results, IEEE Transactions on Control Systems Technology, TCST-8(5): 856-863,
2000

[224] Pettersen, K. Y., F. Mazenc, and H. Nijmeijer. Global Uniform Asymptotic Stabilization
of an Underactuated Surface Vessel: Experimental Results. IEEE Trans. Control Syst.
Technol., 12-6, pp. 891—903, 2004.

[225] Pinkster, J. A. and U. Nienhuis. Dynamic Positioning of Large Tankers at Sea. Proc. of
the Offshore Technology Conference (OTC’96), Houston, TX, 1996.

[226] Powell, L. Platform management systems in the Royal Navy today and tomorrow. In proc.
of Eleventh Ship Control Systems Symposium, Vol.(1) pp. 1-10, April 1997.

[227] Price, W. G. and R. E. D. Bishop. Probabilistic Theory of Ship Dynamics. Chapman and
Hall, London, UK, 1974.

[228] Rensvik, E., A. J. Sørensen and M. Rasmussen. Maritime Industrial IT. 9 International
Conference on Marine Engineering Systems (ICMES), the Helsinki University of Technol-
ogy (HUT) Ship Laboratory and on board MS SILJA SERENADE, Finland, Paper B5,
19-21 May, 2003.

[229] Richardson, K. M., C. Pollock and J. O. Flower. Design and performance of a rotor position
sensing system for a switched reluctance marine propulsion unit. In proc. of IAS’96. IEEE
Industry Applications Conference, Thirty-First IAS Annual Meeting, Vol.(1), pp. 168-173,
1996.

[230] RIFLEX. User’s Manual,version3.2.3. MARINTEKReportno.519619, Trondheim,Norway,

2003.

[231] Ruth, E. Propulsion Control and Thrust Allocation on Marine Vessels. PhD Thesis
2008:203, Department of Marine Technology, Faculty of Engineering Science & Technology,
NTNU, 2008, Norway.

[232] Ruth, E., A. J. Sørensen and T. Perez. Thrust Allocation with Linear Constrained
Quadratic Cost Function. In Proc. of the IFAC Conference on Control Applications in
Marine Systems, Bol, Croatia, 2007.

487
[233] Ruth, E. and A. J. Sørensen. A solution to the Nonconvex Linearly Constrained Quadrat-
ing Thrust Allocation Problem. In Proceedings of 8th Conference on Manoeuvring and
Control of Marine Craft, Guarujá, Brazil, 2009.
[234] Ruth, E. and A. J. Sørensen. Effect of Thruster Configuration on DP Capability. 10th
International Marine Design Conference, Vol. 1, pp.-262-271, Trondheim, Norway, 2009.
[235] Ruth, E., Ø. N. Smogeli, T. Perez, and A. J. Sørensen. Antispin Thrust Allocation for
Marine Vessels. IEEE Transaction on Control System Technology, 17-6, pp. 1257-1269,
2009.
[236] Rosu, M., V. Nahkuri, A. Arkkio, T. Jokinen, J. Mantere, and J. Westerlund. Permanent
magnet synchronous motor for ship propulsion drive. In proc. of Symposium on Power
Electronics Electrical Drives Advanced Machines Power Quality, Vol.(1), pp. C3/7-12,
June 1998.

[237] Rustad, A. M. “Modeling and Control of Top Tensioned Risers”. PhD Thesis 2007:183,
Department of Marine Technology, Faculty of Engineering Science & Technology, NTNU,
2007.
[238] Rustad, A. M., C. M. Larsen and A. J. Sørensen. FEM Modelling and Control for Collision
Prevention of Top Tensioned Risers. Journal of Marine Structures, Elsevier, Vol. 20, ISSN:
0951-8339, 2007.
[239] Sagatun, S. I. The Elastic Cable Under the Action of Concentrated and Distributed Forces.
Journal of Offshore Mechanics and Arctic Engineering, Vol.(123), pp.43-45, 2001
[240] Sagatun,S. I., T. I. Fossen, and K.P. Lindegaard. Inertance control of underwater instal-
lations. in IFAC - CAMS 2001, Glasgow, UK, July 2001.
[241] Sagatun, S. I. and O. Tveit. Increased operability subsea - more accurate tools for instal-
lation and intervention. in ISOPE2001, Stavanger, Norway, June 17-22 2001.
[242] Sagatun, S.I. Active Control of Underwater Installation. IEEE Trans. Control SYstem
Tech., vol. CS-10, no. 5, pp. 743—749, 2002.
[243] Sagatun, S. I., T. A. Johansen, and T. I. Fossen. Vessel-crane moonpool control system. In
IEEE Conf. on Control Applications (CCA’02), submitted, Glasgow, UK, Aug. 2002.
[244] Sagatun, S. I. Optimal ship - crane load guidance and control: Project 7003267 Installation
and Intervention. Tech. Rep., Norsk Hydro O and E, Research and Technology, 2002.
[245] Sagatun, S. I. Ship-crane-load kinematics and dynamics: Project 7003267 Installation and
Intervention. Tech. Rep., Norsk Hydro O and E, Research and Technology, 2002.
[246] Sagli, J. R. Coordination of manipulators with redundant degrees of freedom. Ph.D. thesis,
The Norwegian Institute of Technology, Department of Engineering Cybernetics, 1991,
ITKreport 1991:34-W/ISBN 82-7119-284-1.
[247] Sallabank, P. H., and A. J. Witehead. The practical application of modern simulation
tools throughout the design and trials of a diesel electric propulsion system. Trans. IMarE,
Vol.(107), pp. 101-117, 1995.

488
[248] Salvesen, N., E. O. Tuck and O. M. Faltinsen. Ship motions and sealoads. Trans. SNAME.
Vol.(78), pp345-356. 1970.
[249] Savage, P. G. Strapdown Inertial Navigation. Lecture Notes. Strapdown Associates Inc,
Minnetonka, MN, US, 1990.
[250] Shampine, L. F., J. Kierzenka and M. W. Reichelt. Solving Boundary Value
Problems for Ordinary Differential Equations in MATLAB with Bvp4c.
ftp://ftp.mathworks.com/pub/doc/papers/bvp/tutorial.pdf, MathWorks Inc., 2000.
[251] Schriek, D. and J. W. de Nijs. Royal Netherlands Navy M class frigate: integrated moni-
toring and control system and electric installation. Trans. IMarE, Vol.(103), pp. 269-291,
1991.
[252] Sciavicco, L. and B. Siciliano. Modeling and Control of Robot Manipulators. McGraw-Hill
Book, Singapore, 1996, ISBN 0-07-114726-8.
[253] Shiba, H. Air-Drawing of Marine Propellers. Technical Report. nr. 9. Transportation Tech-
nical Research Institute, Japan, 1953.
[254] Skjetne, R. The Maneuvering Problem. PhD Thesis 2005, Department of Engineering
Cybernetics, Faculty of Information Technology, Mathematics and Electrical Engineering,
NTNU, Norway, 2005.
[255] Smith, J. R., A.F. Stronach and A. T. Mitchell. Prediction of the electrical system behav-
iour of special purpose vessels. Trans. IMarE, Vol.(97), pp. 11-22, 1985.
[256] Smith, K. S., R. Yacamini and A. C. Williamson. Cycloconverter drives for ship propulsion.
Trans. IMarE, Vol.(105), pp. 23-52, 1993.
[257] Smith, R. C.and T. O. Zavertnik. Overview of US Navy electric propulsion technology. In
proc. of ICEM 94. International Conference on Electrical Machines, Vol.(2), pp. 483-487,
1994.
[258] Smogeli, Ø. N. Thruster Control Subject to Varying Environmental Conditions. Pre-project
thesis, Department of Marine Hydrodynamics, NTNU, November 2001.
[259] Smogeli, Ø. N., L. Aarseth, E. S. Overå, A. J. Sørensen and K. J. Minsaas. Anti-Spin
Thruster Control In Extreme Seas. In Proceedings of 6th Conference on Manoeuvring and
Control of Marine Crafts (MCMC2003), Girona, Spain, September 16-19, 2003.
[260] Smogeli, Ø. N., A. J. Sørensen and T. I. Fossen. Design of a Hybrid Power/torque Thruster
Controller with Loss Estimation. In Proceedings of the IFAC Conference on Control Ap-
plications in Marine Systems (CAMS2004), Ancona, Italy, 7 — 9 July2004.
[261] Smogeli, Ø. N., J. Hansen, A. J. Sørensen and T. A. Johansen. Anti-Spin Control for
Marine Propulsion Systems. In Proceedings of CDC’04, 43st IEEE Conference on Decision
and Control, Bahamas, 2004.
[262] Smogeli, Ø. N., E. Ruth and A. J. Sørensen. Experimental Validation of Power and Torque
Thruster Control. 20th IEEE International Symposium on Intelligent Control (ISIC’05)
and the 13th Mediterranean Conference on Control and Automation (MED’05), Cyprus,
June 27-29, 2005.

489
[263] Smogeli, Ø. N., A. J. Sørensen and K. J. Minsaas. The Concept of Anti-spin Thruster
Control. Accepted for publication in IFAC Journal of Control Engineering Practice, 2005.

[264] Smogeli, Ø. N. Control of Marine Propellers — From Normal to Extreme Conditions. PhD
Thesis 2006:187, Department of Marine Technology, Faculty of Engineering Science &
Technology, NTNU, Norway, 2006.

[265] Smogeli, Ø. N., A. J. Sørensen and K. J. Minsaas. The Concept of Anti-Spin Thruster
Control. Control Engineering Practice. CEP 16-4, pp. 465-481, 2008.

[266] Smogeli, Ø. N. and A. J. Sørensen. Antispin Thruster Control for Ships. IEEE Transaction
on Control System Technology Vol. 17, No. 6, pp. 1362-1375, 2009.

[267] Smogeli, Ø. N. Experiences From Five Years of DP Software Testing. European DP Con-
ference, London, UK, 2010.

[268] SNAME. The Society of Naval Architects and Marine Engineers. Nomenclature for Treat-
ing the Motion of a Submerged Body Through a Fluid, Technical and Research Bulletin,
No. (1-5), 1950.

[269] Spong, M. W. Adaptive Control of Flexible Joint Manipulators, Syst. Contr. Lett., vol.
vol. 13, pp. 15-21, 1989.

[270] Spooner, E., N. Haines, R. Bucknall. Slotless-armature DC drives for surface warship
propulsion. IEE Proceedings-Electric Power Applications, Vol.(143), no. 6, pp. 443-448,
1996.

[271] Stephens, R. I., K. J. Burnham and P. J. Reeve. A practical Approach to the Design
of Fuzzyy Controllers with Application to Dynamic Ship Positioning. In Proc. of IFAC
Conference on Control Applications in Marine Systems, Trondheim, Norway, 1995.

[272] Strang, G. Linear Algebra and its Applications. Third Edition, Harcourt Brace Jovanovich,
Publishers, Florida, US, 1988.

[273] Strand, J. P. and T. I. Fossen. Nonlinear Output Feedback and Locally Optimal Control
of Dynamically Positioned Ships: Experimental Results. In Proc. of the IFAC Conference
on Control Applications in Marine Systems, Fukuoka, Japan, 1998.

[274] Strand, J. P., A. J. Sørensen and T. I. Fossen. Design of Automatic Thruster Assisted
Position Mooring Systems for Ships. Modeling, Identification and Control, MIC, Vol. (19),
No. (2), 1998.

[275] Strand, J. P. and T. I. Fossen. Nonlinear Output Feedback and Locally Optimal Control
of Dynamically Positioned Ships: Experimental Results. Proc. of the IFAC Conference on
Control Applications in Marine Systems (CAMS’98), Fukuoka, Japan, October 27—30, pp.
89—95, 1998.

[276] Strand, J. P. Nonlinear Position Control Systems Design for Marine Vessels. Doctoral
Dissertation, Department of Engineering Cybernetics, Norwegian University of Science
and Technology, Trondheim, Norway, 1999.

490
[277] Strand, J. P. and T. I. Fossen. Nonlinear Passive Observer for Ships with Adaptive Wave
Filtering. In ”New Directions in Nonlinear Observer Design” (H. Nijmeijer and T. I.
Fossen, Eds.), Springer-Verlag London Ltd., pp. 113-134, 1999.

[278] Strand, J. P. and A. J. Sørensen. Marine Positioning Systems. In The Ocean Engineering
Handbook. F. El-Hawary, Ed., CRC Press, Florida, US, Chapter 3, pp. 163-176, 2000.

[279] Sullivan, P. J., F. Gosselin and M. J. Hinchey. Dynamic Response of an Air Cushion Lift
Fan. Intersociety High Performance Marine Vehicle Conference and Exhibit - HPMV’92,
American Society of Naval Engineers, Washington D. C., US, 1992.

[280] Sælid, S., N. A. Jenssen and J. G. Balchen. Design and Analysis of a Dynamic Positioning
System Based on Kalman Filtering and Optimal Control. IEEE Transactions on Automatic
Control, Vol. (28), No. (3), pp. 331-339, 1983.

[281] Sørdalen, O. J. Optimal Thrust Allocation for Marine Vessels. IFAC Journal of Control
Engineering Practice, Vol. (5), No. (9), pp. 1223-1231, 1997.

[282] Sørensen, A. J., S. Steen and O. M. Faltinsen. Cobbelstone effect on SES. High Perfor-
mance Marine Vehicle Conf. - HPMV’92, ASNE, Washington D.C., US, 1992.

[283] Sørensen A. J., S. Steen and O. M. Faltinsen. SES Dynamics in the Vertical Plane. Journal
of Ship Technology Research. Vol(40), pp. 71-94, 1993.

[284] Sørensen, A. J. and O. Egeland. Desig of Ride Control System for Surface Effect Ships
using Dissipative Control. Automatica. Vol(31), No. (2), pp. 183-199, 1995.

[285] Sørensen, A. J. Modelling and Control of SES Dynamics in the Vertical Plane. Ph.D.
thesis, The Norwegian Institute of Technology, Trondheim, Norway, 1993.

[286] Sørensen, A. J., S. I. Sagatun and T. I. Fossen. Design of a Dynamic Positioning System
Using Model-Based Control. IFAC Journal of Control Engineering Practice, Vol. (4), No.
(3), pp. 359-368, 1996.

[287] Sørensen, A. J., A. K. Ådnanes, T. I. Fossen and J. P. Strand. A New Method of Thruster
Control in Positioning of Ships Based on Power Control. In Proc.: 4 th IFAC Conf. on
Manoeuvring and Control of Marine Craft (MCMC’97). Brijuni, Croatia, pp 172-179, 1997.

[288] Sørensen, A. J. and J. P. Strand . Positioning of Semi-submersibles with Roll and Pitch
Damping. In Proc. of 4th IFAC Conf. on Control Applications on Marine Systems —
CAMS’98, pp. 67-73, Fukuoka, Japan, 1998.

[289] Sørensen, A. J. , J. P. Strand and T. I. Fossen. Thruster Assisted Position Mooring of

Turret-Anchored FPSOs. In Proc. of the IEEE Conference on Control Applications (CCA)
and CACD, Hawaii, US, 1999.

[290] Sørensen, A. J. and J. P. Strand. Positioning of Small-Waterplane-Area Marine Construc-

tions with Roll and Pitch Damping. IFAC Journal of Control Engineering Practice,.Vol.
(8), No. (2), pp. 205-213, 2000.

491
[291] Sørensen, A. J., B. Leira, J. P. Strand and C. M. Larsen. Modelling and Control of Riser
Angles and Stresses in Dynamic Positioning. In: Proc. of 5th IFAC Conf. on Manoeuvring
and Control of Marine Craft — MCMC’2000, pp. 275-280, Aalborg, Denmark, 2000.

[292] Sørensen, A. J. and M. Ronæss. Mathematical Modeling of Dynamically Positioned and

Thruster Assisted Anchored Marine Vessels. In The Ocean Engineering Handbook. F.
El-Hawary, Ed., CRC Press, Florida, US, Chapter 3, pp. 176-189, 2000.

[293] Sørensen, A. J., T. I. Fossen and J. P. Strand. Design of Controllers for Positioning of
Marine Vessels. In The Ocean Engineering Handbook. F. El-Hawary, Ed., CRC Press,
Florida, US, Chapter 3, pp. 207-218, 2000.

[294] Sørensen, A. J. Methods for Thrust Control. In The Ocean Engineering Handbook. F.
El-Hawary, Ed., CRC Press, Florida, US, Chapter 3, pp. 235-247, 2000.

[295] Sørensen, A. J., B. Leira, J. P. Strand and C. M. Larsen. Optimal Setpoint Chasing in
Dynamic Positioning of Deep-water Drilling and Intervention Vessels. Journal of Robust
and Nonlinear Control. John Wiley & Sons Ltd., Chichester, UK. Vol. (11), pp. 1187-1205,
2001.

[296] Sørensen, A. J., J. P. Strand and H. Nyberg. Dynamic Positioning of Ships and Floaters
in Extreme Seas. In Proc. of MTS/IEEE Oceans 2002, Biloxi, Mississippi, US, 2002.

[297] Sørensen, A. J., K.-P. W. Lindegaard and E. D. D. Hansen. Locally Multiobjective H2 and
Hinf Control of Large-scale Interconnected Marine Structures. In Proceedings of CDC’02,
41st IEEE Conference on Decision and Control, Las Vegas, US, December 2002.

[298] Sørensen, A. J., E. Pedersen and Ø. Smogeli. Simulation-Based Design and Testing of Dy-
namically Positioned Marine Vessels. In Proceedings of International Conference on Ma-
rine Simulation and Ship Maneuverability, MARSIM’03, August 25 - 28, 2003, Kanazawa,
Japan.

[299] Sørensen, A. J. Structural Issues in the Design of Marine Control Systems. Key Note Lec-
ture, In Proceedings of the IFAC Conference on Control Applications in Marine Systems
(CAMS2004), Ancona, Italy, 7 — 9 July, 2004.

[300] Sørensen, A. J. Structural Issues in the Design and Operation of Marine Control Systems.
Accepted for publication in IFAC Journal of Annual Reviews in Control, 2005.

[301] Sørensen, A. J. and A. K. Ådnanes. Reconfigurable Marine Control Systems and Electri-
cal Propulsion Systems for Ships. ASNE Reconfiguration and Survivability Symposium,
February 16-18, 2005 (RS 2005), Florida, US.

[302] Sørensen, A. J., S. T. Quek and T. D. Nguyen. Improved Operability and Safety of DP Ves-
sels Using Hybrid Control Concept. International Conference on Technology & Operation
of Offshore Support Vessels (OSV 2005), Singapore, 2005.

[303] Sørensen, A. J. and Ø. N. Smogeli. Torque and Power Control of Electrically Driven Marine
Propellers. Control Engineering Practice, CEP 17-9, pp. 1053-1064, 2009.

492
[304] Sørensen, A. J., Ø. N. Smogeli and E. Ruth. Propulsion Control Strategies for Fixed
Pitch Propellers at Low Advance Speed. 1st Int. Symp.on Marine Propulsors, pp. 139-153,
Trondheim, Norway, 2009.
[305] Sørensen, A. J.. A Survey of Technology Advances and Future Perspectives of Dynamic
Positioning Systems. Plenary Lecture in Proceedings of the IFAC Conference on Con-
trol Applications in Marine Systems (CAMS2010), pp. 132-148, September 15-17, 2010,
Rostock- Warnemünde, Germany.
[306] Sørheim, H. R. Dynamic Positioning in Single-Point Mooring - A Theoretical Analysis
of Motions, and Design and Evaluation of an Optimal Control System. NTH (NTNU),
Trondheim, Norway, 1982.
[307] Tannuri, E. A. and D. C. Donha. H∞ Controller Design for Dynamic Positioning of Turret
Moored FPSO. In Proc. of the IFAC Conference on Manoeuvring and Control of Marine
Craft , Aalborg, Denmark, 2000.
[308] Tannuri, E. A. and H. M. Morishita (2006). Experimental and Numerical Evaluation of a
Typical Dynamic Positioning System. Applied Ocean Research, 28-2, pp. 133-146, 2006.
[309] Tannuri, E. A., A. C. Saad and H. M. Morishita. Offloading Operation with a DP Shuttle
Tanker: Comparison Between Full Scale Measurements and Numerical Simulatiopn Re-
sults. In Proc. of the IFAC International Conference on Manoeuvring, Guarujá, Brazil,
2009.
[310] Tannuri, E. A., A.C. Agostinho, H.M. Morishita, L. and Moratelli Jr. Dynamic Position-
ing Systems: An Experimental Analysis of Sliding Mode Control. Control Engineering
Practice, CEP 18-10, pp. 1121-1132, 2010.
[311] Taylor, M. E. Partial Differential Equations I, II and III - Nonlinear Equations. Vol. (117),
Springer-Verlag New York Inc., 1996.
[312] Tinney, M.D. and J Hensler. Efficiency and power density improvements in electric propul-
sion systems. In proc. of INEC 94, Cost effective maritime defence, pp. 175-183, 1994.
[313] Titterton, D. H. and J. L. Weston. Strapdown Inertial Navigation Technology. IEE, Lon-
don, UK, 1997.
[314] Torsethaugen, K. Model for a doubly peaked wave spectrum. Technical report STF22
A96204. SINTEF. Prepared for Norsk Hydro.
[315] Triantafyllou, M. S. Cable Mechanics With Marine Applications. Lecture Notes, Dept. of
Ocean Engineering, Massachusetts Institute of Technology (MIT), 1990.
[316] Triantafyllou, M. S. Cable Mechanics for Moored Floating Systems. BOSS’94, Boston,
MA, US, pp 67-77, 1994.
[317] Triantafyllou, M. S. and D. K. P. Yue. Damping of Moored Floating Structures, Proc. 26th
Annual Offshore Technology Conference, Houston, Texas, Vol.( 3), pp. 215-224, 1994.
[318] Vidyasager, M. Nonlinear Sytem Analysis. 2nd Edition. Prentice Hall, Englewood Cliffs,
NJ., US, 1993.

493
[319] Vik, B. and T. I. Fossen. Nonlinear Observers for Integration of Satellite and Inertial
Navigation Systems: Experimental Results. Submitted to Automatica, 2002.

[320] Volovodov, S. K., M. G. Chernjaev, A. J. Kaverinsky,S.S. Volovodov and B. P. Lampe.

Control Principle for Dynamic Positioning of Offshore Drilling Platforms and Ships In
Proc. 11. Symp. Maritime Elektrotechnik, Elektronik und Informationstechnik, Rostock,
Germany, 2004.

[321] Watanabe, K., H. Suzuki, and K. Yoshida. Active control for installation of underwater
structures. J. Mar. Sci. Tech., vol. 98-3, 1998.

[322] Whitcomb, L. L. and D. R. Yoerger. Development, Comparison, and Preliminary Ex-

perimental Validation of Nonlinear Dynamic Thruster Models. IEEE Journal of Oceanic
Engineering, Vol. (24). No. (4), pp. 481-494, 1999a.

[323] Whitcomb, L. L. and D. R. Yoerger. Preliminary Experiments in Model-Based Thruster

Control for Underwater Vehicle Positioning. IEEE Journal of Oceanic Engineering, Vol.
(24). No. (4), pp. 495-506, 1999b.

[324] White, R. D. and K. J. Bradley. The analysis of large marine drives using synchronous
machines and cyclo-converters. In proc. of International Conference on the Evolution and
Modern Aspects of Synchronous Machines, Vol.(1), pp. 393-398, 1991.

[325] White, R. D. and K. J. Bradley. The effect of system characteristic on the analysis of large
industrial and marine drive using cyclo-converter fed synchronous machines. In proc. of
Power Electronics and Variable-Speed Drives, IEE, No. 399, pp. 374-379. 1994.

[326] Wichers, J. E. W. Position Control - From Anchoring to DP System. UETP, Marine

Science and Technology, Trondheim, Norway, Section 18, pp.1-23, 1993.

[327] Wilne, T. Developments to improve the reliability of CPP thrusters since the first IMCA
publication, including cooperation of suppliers and a through life care concept. In proc. of
IMCA Station Keeping Seminar and Workshop, Houston, USA 31.oct-1.nov 1996.

[328] Yacamini, R. and S. C. Chang. Noise generation in ship propulsion motors. In proc. of
UPEC 93, pp.952-956, 1993.

[329] Yakubovich, V. A. A Frequency Theorem in Control Theory. Siberian Mathematical Jour-

nal 14(2):384—420, 1973.

[330] Yoshida, I. and M. Oka. Power feature required for PEFC powered electric propulsion
ship. 1996 Fuel Cell Seminar. Program and Abstracts, pp. 323-326, November 1996.

[331] Zdravkovich, M. M. Forces on pipe clusters. In Proc. of the Conference Separated Flow
around Marine Structures, pp. 201-226, Trondheim, Norway, 1985.

[332] Zdravkovich, M. M. Flow around circular cylinders. Oxford Science Publications, Oxford,
2003.

[333] Åström, K. J. and B. Wittenmark. Computer-Controlled Systems. Third Edition, Prentice

Hall, New Jersey, US, 1997.

494
[334] Ådnanes, A. K. High efficiency, high performance permanent magnet synchronous motor
drives. PhD dissertation, NTNU 1991:60.

[335] Ådnanes, A. K. Variable speed FP vs. fixed speed CP. In proc. of IMCA Station Keeping
Seminar and Workshop, Houston, USA, 31.oct-1.nov 1996.

[336] Ådnanes, A. K. How to Ensure System Integrity for the Power Plant in Vessels with DP
or Thruster Assisted Mooring. Offshore Electrical Power Systems International Meeting,
Rio de Janeiro, Brazil, 1998.

[337] Ådnanes, A. K., A. J. Sørensen and T. Hackman. Essential characteristics of electric

propulsion and thruster drives in DP vessels. In proc. of Dynamic Positioning Conference,
Houston, USA, 22-23 October 1997.

495
Appendix A

Linear Algebra

Some important linear algebra properties are listed below (Strang [272]).

Definition A.1 The matrix A ∈Rn×n is symmetric if

A = A  (A.1)

Definition A.2 The quadratic form is defined as

 = x Ax ∀ x ∈Rn  (A.2)

where  is a scalar.

Definition A.3 The matrix A ∈Rn×n is skew-symmetric if

A = −A  (A.3)

The quadratic form of a skew-symmetric matrix is

µ ¶
  1¡ 
¢ 1¡ 
¢
x Ax = x A+A + A−A x =0 ∀ x ∈Rn  (A.4)
2 2

Definition A.4 A necessary and sufficient condition for the symmetric matrix  ∈Rn×n to be
positive definite is
x Ax 0 ∀ x ∈Rn  (A.5)
All the eigenvalues of A will be positive, that is 1  2     0.

Definition A.5 A necessary and sufficient condition for the symmetric matrix A ∈Rn×n to be
positive semidefinite is
x Ax ≥0 ∀ x ∈Rn  (A.6)
All the eigenvalues of A will be positive or equal to zero, that is 1  2    ≥ 0.

Definition A.6 The non-symmetric matrix A 6= A  where A ∈Rn×n is positive if

x Ax ≥0 ∀ x ∈Rn  (A.7)

496
Definition A.7 The non-symmetric matrix A ∈Rn×n is strictly positive if

x Ax 0 ∀ x ∈Rn  (A.8)

¡ ¢ ¡ ¢
Let the non-symmetric matrix, A 6= A  be reformulated as A = 12 A + A + 12 A − A The
non-symmetric matrix A is strictly positive if the symmetric matrix defined as
1¡ ¢
A + A (A.9)
2
is positive definite, that is
1¡ ¢
x A + A x  0 (A.10)
2
¡ ¢
The matrix 12 A − A will be skew-symmetric matrix with the quadratic form equal to zero.
Hence (A.8) can be reformulated as
1 ¡ ¢
x Ax = x A + A x 0 ∀ x ∈Rn  (A.11)
2

Remark A.1 In textbooks for notational simplicity x Ax 0 is written A 0

497
498
Appendix B

Digital PID-Controllers

B.1 Continuous-Time PID-Controllers

It is well known from text books in control theory (Åstrøm and Wittenmark [333], Dorf and
Bishop [59], Balchen et al. [21], Egeland [63], The Control Handbook [170]) that, a single-input
single-output (SISO) continuous-time proportional-integral-derivative (PID) controller is often
formulated as
() =  () +  () +  () (B.1)
where
 () =  () (B.2)
 () =  ̇() (B.3)
Z 
 () =  () (B.4)

The deviation error between the state  and the desired setpoint  is defined as () =  () −
() and ̇() = ̇ ()− ̇(). The controller gains are defined such that  ≥ 0 is the proportional
gain,  ≥ 0 is the derivative gain, and  ≥ 0 is the integrator gain. Often the proportional and
integral gains are rewritten as
 =    (B.5)

 =  (B.6)

where  is the derivative time and  is the integration time.
Similarly, a multivariable PID controller can be written
Z 
u = K e + K ė + K e() (B.7)

where e = x −x is the deviation error vector between the state vector x and the desired setpoint
vector x , and ė = ẋ − ẋ. K ≥ 0, K ≥ 0 and K ≥ 0 are the corresponding proportional,
derivative and integral gain matrices.
In the -domain (Laplace) incorporating limitation on the derivative term and assuming that
0 ≤  ¿  , the PID formulation becomes
  
() =  () +  () + () (B.8)
(1 +  )  

499
where 0   ≤ 1

B.2 Discrete-Time PID-Controllers

A simple approximation of (B.8) can be done in order to derive a digital PID controller. The
proportional part can be written
 [] =  [] (B.9)
where
[] =  [] − [] (B.10)
From (B.8) the derivative part can be rewritten according to

 
 () =  () (B.11)
(1 +  )
m
 () +   () =   () (B.12)

Using inverse Laplace transform we find the time domain formulation of (B.12) to be

 ̇ +  =   ̇ (B.13)

If we apply Euler’s backward difference method, we achieve a discrete approximation of the

continuous-time derivative controller given in (B.13) to be

  
( [] −  [ − 1]) +  [] = ([] − [ − 1])  (B.14)
 
where  is the sampling period. Eq. (B.14) can be reformulated as

  
 [] = ¡ 
¢  [ − 1] + ([] − [ − 1])  (B.15)
 +  ( +  )

The integral term (B.4) is reformulated according to

 ̇ =  () (B.16)

and is approximated by a Euler’s forward difference method according to


( [] −  [ − 1]) =  [ − 1] (B.17)

Eq. (B.17) is reformulated as

 
 [] =  [ − 1] + [ − 1] (B.18)

Thus the resulting discrete PID becomes by summarizing (B.9), (B.15) and (B.17)

[] =  [] +  [] +  [] (B.19)

500
Remark B.1 Several other formulations of digital PID controllers exist, see Åstrøm and Wit-
tenmark [333] and Egeland [63]. These are based on other approximation methods. One example
on this is to apply Tustin’s bilinear transformation on (B.8) and then find the -transform ac-
cording to  () :  (), where
2 1 −  −1
=  (B.20)
 1 +  −1
Using shift-operator calculus the discrete PID controller is found.

Remark B.2 Integrator anti-windup should be implemented in order to avoid that the integrator
integrates up beyond the saturation limits of the actuators. A simple method to avoid this is to
stop integrating when the integral term tends to a given percentage of the actuator saturation
limit. More sophisticated algorithms are found in Levine [170], Åstrøm and Wittenmark [333]
and Egeland [63].

Remark B.3 For a constant setpoint ̇() = ̇ ()− ̇() simplifies to ̇() = −̇() as ̇ () = 0.

501
Appendix C

Definition of Electro-technical Terms

C.1 DC — Direct Current

The DC (Direct Current) voltage source is unipolar with a positive and negative voltage terminal.
Examples of DC voltage sources are:

• Batteries.

• Static converters.

• DC generators.

• Fuel cells.

Connecting an electrically conductive load to the terminal forces a current to flow, Figure
C.1. The polarity of the voltage source is annotated with + (positive) and — (negative). The
arrow shows positive direction of current flow.
If the DC voltage is  and the electrical resistance in the load is , the current  is found
by Ohm’s law

=  (C.1)

The power dissipated in the resistance equals the power delivered by the power source
2

 =   = =  2  (C.2)


Figure C.1: Electrical circuit, with voltage source  and a resistive load .

502
 Ri Ibulb
+
+
Edc Vdc Rbulb
- -

Figure C.2: Battery with no-load voltage  and inner resistance  connected to a light-bulb
with resistance  .

Example C.1 : Flash-light bulb.

A flash light is equipped with a 9 Volt (V) battery and 0.25 Watt (W) light bulb. The current
in the bulb is then

= = 278

where  = , and the resistance in the bulb is
2

 = = 324Ω

where Ω = . Here it is assumed that the battery is an ideal voltage source. Since such does
not yet exist, the battery will normally be represented by an ideal voltage source, i.e. the no-load
voltage, in series with an inner resistance, Figure C.2.

Example C.2 : Flash-light bulb with non-ideal battery.

If the battery in the previous example had an inner resistance of  = 20 Ω, and a no-load
voltage of  = 92 , the current would be

 = = 267
 + 
The voltage on the battery terminals is then
 =  −   = 867
The power dissipated in the bulb becomes
2
 =   = 023
The power loss in the battery is
2
 =   = 14
The efficiency of the battery is then

= = 0943 (943%)
 + 

503
Figure C.3: AC voltages are sinusoidal time varying, 50 Hz voltages have a cycle time of 20 ms.

Example C.3 : Short-circuited battery with inner resistance.

A short circuit (direct connection of two conductive leads with different voltage potentials)
gives a failure current which often is dimensioning for cable, terminations, etc. Upon short-
circuit, the inner resistance in the battery and in the leads and short-circuit itself will limit the
short-circuit current. With a low resistance short-circuit (0 Ω), the fault current of the battery
in the example above is

 = = 046


C.2 AC — Alternating Current

C.2.1 One phase AC sources and resistive loads
The AC (Alternating Current) voltage source is bipolar and normally sinusoidal time varying.
The frequency of the variations in European land network is 50 (Herz), i.e. 50 cycles per
second, see Figure C.3. In US networks, the frequency is 60. In ship networks, both 50
and 60 are used. For some special equipment, especially in navy applications, 400Hz can be
seen. Examples of AC voltage sources are:

• Generators (also called Alternators).

• Static converters.

• Uninterrupted Power Supplies (UPS), which are static converters supplied from batteries
if supply voltage vanish, applied for power supply to critical equipment.

Figure C.3 shows a sinusoidal time varying voltage of frequency  = 50. It can be
expressed
() = ̂ sin 2  = ̂ sin  (C.3)
Here the  = 2 is the angular frequency, expressed in . If this voltage source is connected
to a resistive load (ohmic load), as shown in Figure C.4, a current will flow through the circuit.

504
 i(t)
+
V(t) R
-

Figure C.4: AC voltage source connected to a resistive load.

And assuming that there is no voltage drop in the voltage source, the current will be given by
the resistance in the load
() ̂
() = = sin  (C.4)
 
From this it is seen, that the current is proportional to the voltage, and that the voltage and
current are in phase, i.e. phase angle equals to zero. This is always true for a pure resistive load.
Since the voltages and currents are time varying, the power consumed by the resistor is also
time varying
1
() = ()() = ̂ ˆ sin2 2 = ̂ ˆ (1 − cos 2)  (C.5)
2
Normally, the instantaneous power is less interesting than the average power. This is found
by integrating the power and averaging it over e.g. one cycle time,  according to
Z
1
 = () = ()() = ̂ ˆ sin2 2  

0
Z2 Z2
1 ˆ 1 ˆ2 ˆ2
= ̂  sin2   =  sin2   =   (C.6)
2 2 2
0 0

where we have used  = 2 and  =  +  = . Introducing the root mean square
value () for the current
v
u Z2
u
u 1 ˆ2 ˆ
 = t  sin2   = √  (C.7)
2 2
0

we can see that the power equation simplifies to a form similar as for DC systems
2
 =    (C.8)

With similar definition of  value for the voltage, it yields

 =    (C.9)

and
 =    (C.10)

505
 Vc Neutral point
+

- - Va
+ +
-

+
Vb

Figure C.5: Three phase voltage generator.

 values for voltages and currents enable us to use well known formulas from DC systems in
AC systems, when average power is concerned. This means almost always in analysis of electric
network in this context.

Example C.4 :A lamp connected to 230V AC network.

A light bulb of 25 is connected to a 230 , 50 voltage. The current in the bulb is then

 = = 011

and the resistance in the bulb is

 = = 209 Ω


C.2.2 Three phase AC sources and resistive loads

The power is in all larger installations generated as a three-phase voltage, because:

• Three-phase distribution systems give a better utilization of the material in the generator
itself, cabling, and other distribution equipment and loads.

• Smooth torque in rotating electrical machines, i.e. motors and generators.

• Satisfactory start-up torque with proper design of rotating electrical machines.

• Smoother rectified voltage in static frequency converters.

A three-phase voltage generator can be regarded as three single-phase voltage sources con-
nected in a common point called the neutral point, see Figure C.5.
The voltages are called symmetrical if they all are sinusoidal with same amplitude and 120
phase shift, see upper part of Figure C.6
√ √
 () = 2 sin   () = 2 sin ( − 120 ) 
√
 () = 2 sin ( − 240 )  (C.11)

506
 400

300

200

100
phaseA
0 phaseB
0 5 10 15 20 phaseC 25
-100

-200

-300

-400

800

600

400

200 A-B
B-C
0
C-A
0 5 10 15 20 25
-200

-400

-600

-800

Figure C.6: (Upper) Phase voltages - phase to neutral,  = 230 . (Lower) Line voltages -
line to line,  = 400 .

507
In analysis of voltages and currents, it is convenient to express the sinusoidal time varying
variables by using complex notation, which for voltage means
³√ ´ √ ¡ ¢
 () = Re 2  = 2 Re 
√
= 2 sin  (C.12)

³√ 2
´ √ ³ 2
´
 () = Re 2 − 3 = 2 Re − 3
√
= 2 sin ( − 120 )  (C.13)

³√ 2
´ √ ³ 2
´
 () = Re 2 + 3 = 2 Re + 3
√
= 2 sin ( − 240 )  (C.14)

The sinusoidal variables can be graphically presented in a complex vector diagram, see
Figure C.8. The vectors are in electro-technical terms annotated phasors, which per definition
is represented as a vector with length equal to the  value of the variable and a phase angle
which represents the angular phase to a reference value, e.g. one of the phases. For the voltages
this notation gives
2 2
 () =     () =  − 3   () =  + 3  (C.15)

The line-voltages (line to line voltages) is found as the voltage difference between two phases,
see lower part of Figure C.6

 () =  () −  ()  () =  () −  ()  () =  () −  () (C.16)

In the phasor diagram, the phasor of the line voltage is the vector difference between the phasors
of the corresponding phases

 =  −    =  −    =  −   (C.17)

Hence, the  value which is the length of the phasor, for all the line voltages is the same and
√
 = | −  | = 2 cos 30 = 3  (C.18)

If connecting a symmetrical load, i.e. equal load, to a symmetrical three-phase voltage source
as shown in Figure C.7, there will flow symmetrical currents. The currents in each of the load
branches are the same in amplitude but phase shifted. If the loads are resistive, as shown in
Figure C.6, the currents in each of the branches are
√ 
 () = 2 sin  (C.19)

√ 
 () = 2 sin ( − 120 )  (C.20)

√ 
 () = 2 sin ( − 240 )  (C.21)


508
 Ic
Vc
+ R

- - Va Ia
+ +
-
R
+ R
Vb Ib

Figure C.7: Three-phase voltage source connected to three-phase resistive load.

In phasor notation
 
 =   =   (C.22)

2  − 2
 =  − 3 =  3  (C.23)

2  + 2
 =  + 3 =  3  (C.24)

The power dissipated by the load is

2  √
 = 3   = 3  = 3   = 3   (C.25)

It is seen, that in a three-phase system, the power transfer capability of a three-phase cable is ,
i.e. 73% above a one phase transmission, with only 50% more conductor material, which means
that three-phase transmission is more cost efficient than one-phase.

C.2.3 Impedance
The electric network on a ship will typically consist of a set of alternating generators (three-
phase voltage sources), a distribution system with switchboards and transformers, and loads. For
electrical analysis of the network, the components are modeled by use of equivalent parameters,
such as voltage, or current sources and impedances.
In complex notation, the impedance of the circuit components are also expressed as complex
values:
- Resistors:  =  i.e. positive real.
- Inductors:  =  =  , i.e. positive imaginary.
1
- Capacitors:  =  = − , i.e. negative imaginary.
Ohm’s law can be applied also to phasors. The relation between the current phasor  through
an impedance, and the corresponding voltage phasor  over the impedance is

 =   (C.26)

509
 Im Im

Vc Rotating with Vca

frequency =2f
Vc
Va Va
Re Re
Vb Vab
Vbc
Vb

Figure C.8: Phasor diagram of three-phase voltages.

One can now see that a current through a resistor will be in phase with the voltage, since the
resistor is a positive real. An inductor is a positive imaginary, hence the current will lag the
voltage by 90 . For a capacitor, the current will lead the voltage by 90 .
For a combination of resistors and inductors, the current’s phase angle will be leading between
0 and 90 , dependent on the  ratio. Similarly, the current’s phase angle will be lagging
between 0 and −90 in a resistor-capacitor combination.
In phasor notation, the following definitions of power related terms apply:
- Apparent power ( ):  =   ∗ =  + 
- Power ( ):  = Re() = Re( )
- Reactive power (  ):  = Im() = Im( )
For a resistor, the apparent power equals to the power, since the voltage over the resistor
and the current through it are in phase. This is also seen by calculating the apparent power

 =   ∗ =    ∗ =  ||2  (C.27)

, which is a real value. For an inductor with only inductance, and an capacitor with only
capacitance, the power is similarly zero, and hence apparent power and reactive power are
equal.
Defining the phase angle between voltage and current as  (phi), where positive phase angle
means leading (capacitive) current and negative phase angle means lagging (inductive) current,
it is seen that

 = || cos  (C.28)

 = || sin  (C.29)

The cos  is called the power factor (PF) of the load. The power factor is 10 if the currents are
in phase with the voltages, and lower if the load is inductive or capacitive.
Since reactive power is a measure of currents which load the network without contributing
to usable power, but only gives losses in the distribution and generation system, it is usually
desired to design and operate a system with highest possible power factor. If the load had been
purely resistive, the power factor would be 10, but due to rotating machinery and transmission

510
 400

300

200
voltage
100
current
0
0 5 10 15 20 25
-100

-200

-300

-400

400

300

200
voltage
100
current
0
0 5 10 15 20 25
-100

-200

-300

-400

Figure C.9: (Upper) 90 leading current - capacitive, and (Lower) 90 lagging current - inductive.
Voltage: 230 , Current: 100 .

511
 400

300

200
voltage
100
current
0
0 5 10 15 20 25
-100

-200

-300

-400

Figure C.10: Voltage and current waveforms of Example C.5. Note: currents magnified 10x.
Current is lagging voltage by 828 .

lines with inductances and capacitances, the power factor will in all practical applications be
lower, normally lagging (inductive).

Example C.5 : Single-phase voltage source with inductive-resistive load.

A one phase, 230 50 voltage source is connected to an inductor with inductance  =
30 and resistance  = 2Ω. The impedance of the load is

 =  +  =  +  2  = (2 + 157) Ω

Selecting  = 230 real, then the current in the circuit is

 230
= = = (183 − 144)  = 145−144 
 (2 + 157) Ω

which is a current of magnitude 145, and it is lagging the voltage with a phase angle of 828 ,
which means that the power factor is cos 828 = 0125. The apparent power of the load is

 =   ∗ = 230 (183 + 144)   = (421 + 3312)  

The dissipated power is hence  = 421 and the reactive power,  = 3312  . Verifying the
results by checking the power consumption in the resistor

 =  ||2 = 2 (145)2  = 421

The inductor contributes to current and reactive power, but the power dissipation, or losses, only
occur in the resistor and results in heat generation. The power factor is
 
  = cos  = =p = 0126
||  2 + 2

512
Example C.6 : Three-phase voltage source with inductive-resistive load.
A three-phase, 400 (line to line) 50 voltage source is connected to a three-phase inductor
with inductance  = 30 and resistance  = 2 Ω per phase. Comparing with Example C.5, it
is seen that the loads will have the same voltage conditions, with a phase to neutral voltage of
230 . Hence, in the three-phase system, the following applies:
- Apparent power:  = 3  ∗ = (1263 √ + 9936)  
|| = 3 | | || = 3 | | || = 10016   = 10016  
- Power:  = Re() = 1263 
- Reactive power:  = Im() = 9936  
The power factor is then || = 0126, which is the same as for the one-phase Example
C.5, a fact that should not be too surprising, since the three-phase system actually is a triple of
the one-phase system. Therefore, analysis of symmetrical three-phase systems is usually carried
out by only regarding one of the phases, since all quantities are equivalent, and phase shifted by
120 and 240 in the remaining two phases. Such models are called per phase representation of
the three-phase system.

C.3 Modelling of Components in Electric Power Generation and

Distribution
For analysis and calculation of electric components in the electric power generation and distri-
bution system, simplified models consisting of voltage sources and impedances are used. For
most calculations, linear models are used, which means that the discrete voltage sources and
impedances are regarded to be constant, independent on voltage, current, and frequency. For
more accurate analysis, one might need to model nonlinear phenomena, such as magnetic satu-
ration of inductors and frequency dependent resistor due to current crowding. Nonlinear models
are not regarded here.

C.3.1 Generator model

The generator has a complex model. Physically, it is constructed of a rotating magnetization
winding on the rotor, which also includes damper windings for dynamic stability. Stator consists
of three-phase windings connected to a common neutral point. The induced voltage in the
windings is proportional to the rotational speed of the magnetic flux (in stationary equal to the
rotational speed of the rotor) and the magnitude of the magnetic flux. The total magnetic flux
in the stator windings consists of magnetic flux from the magnetizing current in the rotor, and
armature reaction, i.e. magnetic flux set up by the currents in the stator windings.
For accurate modeling of the generator, a time domain model with 7 degrees of freedom
model must be applied, however, for most analysis, one represents the generator as a voltage
source behind an inductor and a resistor. The inductance represents the inductive behavior
of the windings and the resistance represents the ohmic losses in the windings. Usually, the
resistance is small and can be neglected. One distinguish between stationary, transient, and
sub-transient models of the generator. The reason is, that this simplified method of modeling
neglects dynamic behavior of the generator, and tries to represent it with stationary models. For
fast transients, in the order of 10  or shorter, the sub-transient model is applied. For transients
in the order of 100  − 500 , the transient model is used, and for stationary behavior, i.e.
seconds and above, the stationary model is representative.

513
 Vc
Induced voltage,
Rs Stationary: Ea = E
Transient: Ea = E’
jX Subtransient:: Ea = E’’
Ec
Va
Eb Stator inductance, jX=jL
Ea jX Rs Stationary: L=Ls (synchronous inductance)
jX Transient: L=Ls’ (transient stator inductance)
Rs Subtransient:: L=Ls’’ (subtransient stator inductance)

Stator resistance, Rs
Vb

Figure C.11: Model of generators. Resistances are normally neglected (small, assumed zero).

Induced voltage,
Stationary: Ea = E
Transient: Ea = E’
jX Rs
Subtransient:: Ea = E’’

Stator inductance, jX=jL

Ea Va Stationary: L=Ls (synchronous inductance)
Transient: L=Ls’ (transient stator inductance)
Subtransient:: L=Ls’’ (subtransient stator inductance)
Stator resistance, Rs

Figure C.12: Per phase diagram of generator.

Typically, sub-transient models are used to find peak values of short circuit currents (in first
period after short circuit), transient models are used for calculating voltage variations during
start-up of motors and transformers, and stationary models are used to calculate stationary
load flow. The motivation is to give an engineering tool which is representative and sufficiently
simple for calculation by analytically means. Figure C.11 shows such three-phase equivalent
diagram for a generator. Physically, the induced voltages represents magnetic flux that are
characteristic and can be assumed quasi-constant for the transient period of interest and the
stator inductance is representative for the dynamically induced voltages in this time period. It
must be noted that the diagram is simplified by disregarding the effects of saliency in the rotor
due to the non-cylindrical construction. Also, the capacitive coupling between the windings and
the ground is neglected, for load flow calculations this is acceptable.
Since the voltage and currents are equal, except of being phase shifted, in the three phases,
it is normal only to draw the diagrams for the components in one of the phases, a so-called per
phase diagram, as for the generator is shown in Figure C.12.
The phasor diagram will then be as shown in Figure C.13. The stator voltage equals to the

514
 Im jLs I a
E
E’ jL's I a
E’’ jL's' I a
Va Re
Ia Ra Ia

Figure C.13: Phasor diagram for generator, phase a only. The diagram is drawn for an arbitrary
inductive current with power factor 08 (phase angle  = 368 )

induced voltage minus voltage drops over stator inductance and resistance, i.e.

 =  +   +   (C.30)

0
 =  +   + 0  (C.31)
00
 =  +   + 00  (C.32)

Transformers
The purpose of the transformer is normally to convert between different voltage levels. It also
provides possibilities for isolating different parts of the system in order to reduce the spreading
of electromagnetic and harmonic noise, and to operate with different earthing philosophy in
different parts of the power system.
The equivalent per phase diagram for the transformer is shown in Figure C.14. For an
accurate calculation of losses, the complete diagram should be used, which includes winding
losses in stator primary and secondary ( and  ), and magnetizing losses ( ).  and
 represents the so-called leakage inductance in primary and secondary windings, while 
represents the magnetizing inductance.
From a load flow calculation, the magnetizing currents and losses are negligible and one
can use the simplified diagram. Here, the primary and secondary resistances and inductances
are represented by common equivalents, with subscript . The voltage transformation ratio is
described by the  : 1 ratio, e.g. a 3300  to 690  transformer has  = 48.

C.3.2 Distribution switchboards

Distribution switchboards are normally modeled as ideal conductors, without losses, inductance
or capacitance. This is of cause a simplification, but valid for all practical considerations.

515
 Rp Lp Ls Rs Rk Lk

Vp Rm Lm Vs Vp Vs

n:1 n:1

Figure C.14: Equivalent diagrams for transformer. Left: Complete diagram for loss calculations.
Right: Simplified diagram for load flow and transient calculations.

Figure C.15: Equivalent diagram for cable, pi-diagram (left) and simplified diagram for load
flow calculations (right).

C.3.3 Cables
The modeling of a cable depends on it’s length. For shorter cables, such as up to the order
of tens of meters, one can normally neglect the influence on load flow. It the length is higher,
one must regard the losses, inductance in the conductor and some times also the capacitance to
ground in the calculations. Figure C.15 describes the pi-equivalent of a cable, and the simplified
model used for load flow calculations, neglecting capacitive coupling to ground.

C.3.4 Synchronous motors

Synchronous motors are basically a construction identically to the synchronous generator, and
modeled similarly. Synchronous motors are rarely used on vessels, unless in connection with a
variable speed drive.

C.3.5 Asynchronous motors

Asynchronous (or induction) motors are used for direct driven or variable speed drives for most of
the motor applications on a vessel. It has a rugged and cost efficient design, and low maintenance
costs. Physically it is constructed by a stator similar to the synchronous machine. The rotor is
constructed cylindrically, with short-circuited windings, normally not connected to stationary
equipment. Hence, the currents in the rotor windings are induced by the magnetic flux in the
air gap set up by the stator currents.
The equivalent diagram for the asynchronous motor is similar to a transformer with short

516
 Rs Ls Lr Rr / s Rs Ls L*r R*r / s

Vs Rm Lm Vs Rm Lm

n:1

Figure C.16: Equivalent diagrams for asynchronous motors.

circuited secondary. For most of the calculations, one is not interested in detailed behavior of
the rotor windings, so the quantities in the rotor winding are referred to equivalent quantities
at the stator side, and hence one can take the ideal transformer equivalent out of the diagram,
as shown in Figure C.16.

517
Appendix D

Standards, rules, and regulations on

electric propulsion

This section is written by Geir Ottersen with Det norske Veritas, for ICMES 2000. This is
recommended to be read by all involved in design of electric propulsion systems.

D.1 Introduction
This section presents an overview of existing standards, rules and regulations for, or related to,
electric propulsion, based on a search among European and American sources. On the standard
side, the concentration has been around IEC (International Electrotechnical Commission) or
other possible standard makers, i.e. other large electrical/maritime engineer associations.
The overview of the rules is based on some of the major ship classification societies.
The “statutory” regulations concerning this subject are taken from IMO (International Mar-
itime Organization). The different national coastal or maritime- law requirements are not re-
ferred to or looked into in great detail as they generally do not extend further than IMO regu-
lations with respect to special requirements towards/solutions for electric propulsion. National
regulations refer in some cases to class society rules on this subject; for example in the case with
US Coast Guard and ABS rules.
When a search like this is carried out, it strikes you that not much is actually written this
far on this topic with regards to standards and rules/regulations, the only special- standard in
existence was issued by the IEC in 1984, and is obviously in need of a revision.
On the other hand, a lot of articles have been published from conferences or in magazines in
the last 10 years, explaining and selling the concept of electric propulsion. The situation is in
fact that the industry (vendors) is leading the show in this arena, specified to a certain extent
by the ship owners.
With regards to international regulations (i.e. SOLAS), the requirements around conven-
tional diesel engine propulsion can to some extent also be applied to electric propulsion.
One must keep in mind: It is the function Propulsion that is important and not the choice of
technology. Electric propulsion is a choice of technology and should not be developed in a pure
electro-technical environment, without considering the overall functionality of the vessel. (Vice
versa, it is often forgotten that choosing electric technology for propulsion also should have an
impact on the manning of the vessel and the competence for operation and maintenance of the
electrical propulsion system.)

518
D.2 International regulations, applicable standards or code of
practice
Apart from the IEC standard no other similar international standard exists up to now - as far
as we know.
IMarE has publications about the subject from their conferences of 95 & 98 and a code of
practice written about maritime el. engineering in general in which el. propulsion is one chapter
(very dated however). IEEE has electric propulsion requirements included in the mentioned
books for ship and offshore and are more or less based upon the IEC 92-501. The IEEE Standard
45,”Recommended Practice for Electric Installations on Shipboard”, clause 36 is treating the el.
propulsion subject in a broad manner. The clause 36 also seems to be somewhat based on the
IEC 92-501 without, however, referring to it.
Typical requirements, i.e., to propulsion control and properties is: (SOLAS 26-31 quoted
freely):

• Propulsion power shall be controllable from both main engine room and bridge (however,
not at the same time, and governed by engine room control), and by manual means in
engine room if remote control is failing.

• Propulsion system is to have emergency stop independent of normal control (ref. to SOLAS
31).

• Means shall be provided whereby normal operation of propulsion machinery can be sus-
tained or restored within reasonable time if essential auxiliaries or main components be-
come inoperative, failure in one of the system components of the el. propulsion systems
must be taken into consideration in this context.

• A slow down in propulsion capacity is normally accepted in case of a major failure in the
propulsion system.

• In case of POD systems, the regulations affecting steering gears in 29 and 30 is valid for
the POD steering properties and in particular in the case of a total black-out/emergency
power situation.

US Coast Guard has a clause in their rules, in sec.111.35-1 about El. Propulsion of where
they refer to ABS requirements to be fulfilled for el. propulsion plants.
Also LR rules are accepted according to a new agreement (as far as we know). In that respect
most other class society rules should be accepted, as they seem to be quite similar with respect
to technical requirements to el. propulsion plants.
There seems to be a lack of a combined rule book concerning electrical propulsion containing
requirements both from the machinery and electrical side, although DnV now is making an
attempt towards this and the other societies are in all probability also doing so as well.
RINA has the most documented thus far, (also hidden in the pure machinery chapters)
with regards to what they require from an el. propulsion plant and LR has the least but
is very system- function aimed. One can of course discuss what is best serving the purpose:
overall system functional requirements or detailed specifications.(The “regulation- world” is , as
a matter of fact, moving towards a more system- functional- requirements regime. The main
important “common” requirements from the class societies is as follows (this will also reflect the

519
most important system and component requirements of IEC 92-501 and will to some degree also
be influenced by the IMO regulations related to propulsion machinery in general):

1. System requirements:

• The propulsion system is to have sufficient power (torque) for maneuvering the vessel
and for going astern.
• With the ship travelling at its maximum speed, the propulsion system is to be capable
of stopping and reversing the vessel in an agreed time.(the power system shall absorb
the regenerated power without causing tripping of generators or prime-movers).
• Adequate torque margins (run up and normal) shall be provided in propulsion systems
to guard against the motor pulling out of synchronism or stalling during start, rough
weather when turning and in icy conditions.
• Where propulsion power is derived from a central common power plant, the control
system is to ensure a safe distribution of power between propulsion and ship ser-
vices with tripping of non-essential consumers/or reduction in propulsion power if
necessary.
• The number and capacity of generators in a common power plant must be sufficient
in order to maintain power to essential and important equipment including safe and
effective propulsion of the vessel. This must be the case even with one generator
tripping out due to failure, or taken out for maintenance.
• When vessels have two or more propulsion systems on one common propeller shaft, it
shall be possible to disconnect any one of the system or system components without
affecting the operation of the other systems.
• The propulsion is to be stable throughout normal operating range and load changes
due to wave motions. No over speeding shall occur of the propulsion motors upon
sudden loss of load.
• A failure in one system component (i.e. excitation circuits) should not lead to the
complete loss of propulsion (slowing down will normally be accepted).

2. Component/equipment requirements: (incl. protection and monitoring of system)

• Semiconductor converters are to be able to run on a reduced load upon loss of forced
ventilation
• Ventilation or cooling systems of equipment in the propulsion system is to be moni-
tored for correct temperature levels.
• Provisions are to be made to protect against severe overloads and electrical faults
likely to result in damage to the electrical system
• The main propulsion circuit shall be monitored for earth faults (disconnected if nec-
essary)
• Protection against over voltages and harmonics on the system or on the general
electrical system due to starting, stopping & reversing the propulsion plant are to be
provided.

520
 • Indicating instrument in form of: speed indicators, propulsion, direction, and avail-
able power to be provided at control stations.

3. Installation/ testing requirements:

• Electric propulsion systems are to be installed and made assessable for maintenance
and repair, especially the with regards to the motors.(not modified after POD systems
have arrived)
• In general, no other specific installation requirement other than normal practice for
ships
• Work and site testing of system components for normal electrical systems.
• Commission/Sea trial of the propulsion system normally to be agreed in each case as
this is always a custom made systems.

D.3 Concluding remarks

Class society rules, IEC6092-501, IEEE Std.45, and to some extent the IEE electrical books
for ship & offshore, are the documents that contain the majority of guidelines and system
requirements for the construction installation and testing of electric propulsion system and its
components.
You will not find any guidelines when it comes to which semiconductor technique is to be
used, nor will you find suggestions for building up of system structures (except for the IEEE 45,
which shows some suggested one-line diagrams for electrical propulsion ships).
Additional information may be found in the many various articles recently written about
electric propulsion by vendors and consultants.
The IEC 92-501 standard is due for revision (last edition: 1984) and should reflect today’s
world of electric propulsion, specifically with regards to the developments taken place since 1984.
It would be advantageous to focus our efforts on one common standard we can all follow rather
than the publication of random articles about standards from the many different associations.
Standard and “rule makers” must learn from vendors who are leaders in the development
of electric propulsion. The common goal should be a cooperation between the machinery and
electrical industries in the creation of standards and rules for Electric Propulsion.
For the time being the gathering and learning of information regarding the rules and regula-
tions, which are necessary in order to design an electrical propulsion plant, is unorganized and
unsystematic. It appears that no standards or rules give guidance in the avoiding of malfunctions
within or as a result of el. propulsion systems.
EMC related faults or disturbances in the control system leading to lose of control of the
propulsion system have been a problem with some vessels. In addition, failures in the propulsion
systems have been dominated by component failures such as motor windings breaking down due
to lack of control of over voltages occurring in the power system. It has also been an all to
common problem with harmonics from the propulsion motor converters damaging equipment in
the power or control system.
One must although mention the IEEE 519 with regards to harmonics as a good standard
and covers far more detail than the same type of standard in the IEC system.

521
 Org. El. prop. El. Power Motors Semicond. Other Install &
(special) generation, & & sw. gear equipment testing
distribution Generators assemblies (i.e cables)
and Control
IEC 92-501 92-201 92-301 92-304 92-353 92-401
(1984) 92-202 92-302 92-354
61892-5 92-504 92-503
(draft 99) 92-503
IEEE Clause 36 Clause 1- Covered by Covered by Covered by Covered by 36
Std. no: 45 to 37 36 36 36 &10 & 10
(under rev.)
IMarE CoP See CoP See CoP See CoP See CoP See CoP
IEE See note * & ** See note * See note * See note * See note * See note *

IMarE = The Institute of Marine Engineers, UK , CoP = Code of Practice called” Electricity applied to marine
marine engineering” 1977,(special chapter about el. propulsion, among general el. engineering)
* “Regulation for the electrical and electronic equipment of ships”, sections 25 and 26
** “Recommendations for electrical &electronic equipment of mobile and fixed offshore installations” section
25.10. MOU electric propulsion.

Figure D.1: International standards one can use when specifying, designing and verifying electric
propulsion plants.

Finally it is the responsibility of the standard and rule creators to forge ahead and meet
the obvious challenges they face and develop a more systematic and organized method for
standardizing of the rules and regulations for electric propulsion.
Comments to Figure D.1: CENELEC standards (Europe Standards) are not mentioned as
they are not aiming at marine purpose electrical power equipment. However, they are in general
very similar to the IEC for basic electrical equipment. The relatively thin IEC 60092-501 is
intended especially for electric propulsion plants. However, applicable IEC standards of the
60092 series intended for ships are also listed, as these naturally support IEC 600092-501. The
60092- ship series refer to the basic IEC series like the 34 standard for rotating machines, when
for example treating motors and generators in, 92-301. The new offshore vessel standard series
IEC 61892 part 1-7 in which part 5 deals with MOU electric propulsion is only at a draft stadium,
but there is some considerable requirements in this documents based on the (old) 92-501 mixed
with class and engineering societies requirements. This part 5 will be issued during 2000.IEC
92- 501will be revised in the near future according to the Norwegian secretary of IEC TC —18
(Technical Committee 18),
Comments to Figure D.2: IMO is not addressing electric propulsion in a great detail but
the mentioned Amendment from 96, concerning Reg.41, gives some requirements to Main power
systems- when the system are supplying electrical propulsion systems as well as normal auxiliary
power. The new § 5 requires that: where electrical power is necessary for propulsion and steering,
loss of one main power source ( a generator) shall not render the ship safety with respect to
the power balance - further on ,- the main switchboard busbars of el. propulsion ships shall
be divided by a circuit breaker ensuring discrimination and fault clearing ( 50% of the plant
maintained one must add) in case of a major fault occurring at the switchboard. Regulations
concerning the system behavior of machinery and propulsion is of course more or less applicable

522
 Regulation Electric Main propulsion Control of Misc. req. General
propulsio In general propulsion Relating to Electrical
n Propulsion Req.
(special)
IMO 96 Part.C, Reg 26, Pt.C, Reg. Pt.E,Reg. Pt.D,reg.
(SOLAS) Amendm 27 28, *29, 30 31 & 49 46, 49, 51, 40 to 45
ent 53 & 54
Pt.D.Re
g.41
new § 5
IMO None Ch.7, Ch.7, Ch.4, Ch.5,
(MODU) Sec.1 & 9 Sec.2, 4,& 7, Sec.2 & 5 Sec.1, 2 & 5
CODE Ch.8, Ch.8,
Sec.5,& 6 Sec.7, 8, &
9
SOLAS = Safety Of Life At Sea,
MODU = Mobile Offshore Drilling Units
* ) In case of POD systems which also acts as rudder/steering gear.

Figure D.2: International regulations, which must be fulfilled when designing an electric propul-
sion plant.

to electric propulsion. The most important requirements from the mentioned regulations, in
addition to the new §5 of Reg.41 of SOLAS (numbers/paragraphs stated in table) are affecting
the control aspect of the propulsion system around availability/capability and load shedding
requirements. The control system requirements is contained in the regulation 31.1. of SOLAS,
revised in 1998, to also contain requirements aimed at computer control (DCS) systems for
control of propulsion Also to mention the validity of the normal situation today where one
common power plant is utilized for both propulsion and general electric power.
Comments to Figure D.3: Generally, the class societies have more or less similar requirements
to main class and additional “classes” therefore, not all class rulebooks are searched/reviewed.
Generally, the class rules for el. propulsion are based on the IEC 92-501 standard (no other
standard quoted at least). Some requirements are taken from their ”own chest” and is contained
in the “el. propulsion chapter” of the electrical rules by the different societies.

523
 Class El. prop (special) Propulsion/machi Additional Gen.el./aut Rules
society nery in general Class related to El. prop.
affecting
propulsion
DnV Part 4, ch.4 Sec.11 Pt.4, ch.2 Pt.6, ch.2 Pt 4, ch.5 Aut/Instr.
(under rev.) (under rev.) (RP) Pt 6, (* Pt4, ch4)
ch.3 (E0)
Pt.6, ch.7
(DP)
LR Pt.6. sec.15 Pt5, Pt.5,ch.1, 2, 6 Pt. 7 Pt.6 sec1-7
ch.20 &18
GL Pt.1, sec.12 Pt.1 Ch.2 RP Pt.6
ABS Pt.4, Ch.8, Sec.5 Pt.4, Ch.3 - Pt.4, Ch.9
(2000
rules)
RINA Sec.D, Ch.2 Sec.C - Sec.D
Ch.2,3,4,9
BV Pt.III,sec. 18-13 - - Pt.III
(IACS) none Yes - -

” –“ in the column, means that a search for relevant rule chapter was not carried out or not applicable.
E0: Unattended machinery space
DP: Dynamic positioning
RP: Redundant Propulsion

Figure D.3: Classification society’s rules with respect to electric propulsion.

524
Index

Aliasing, 138 Lowpass Filter, 159

Backward Euler’s Method, 135, 147 Marine Automation System, 10

Mooring Models, 224
Cable Models, 210
Causal, 123 Notch Filter, 160
Characteristic Equation, 132 Nyquist Frequency, 136
Class Requirements, 30
Collocated Control, 461, 462 Observable, 125, 134
Continuous-time, 121, 124, 129 Observer, 163
Control Plant Model, 176, 232, 251, 268, 396 Optimal Setpoint Chasing, 430
Control Structure, 1, 227
Parabolic Function, 127
Controllable, 125, 133
Partial Differential Equation, 350, 378, 446
Dead Reckoning, 157, 230 Partial Differential Equations, 440
Difference Equation, 130 Passive Observer, 231, 235
Discrete-time, 121, 129, 131 Passivity, 456
DP Functionality, 17 Position Mooring System, 7, 12, 174, 224
Dynamic Positioning System, 1, 7, 12, 174, 227, Power Management System, 21
419 Power System, 8
Process Plant Model, 37, 176
Euler Angle Transformation, 194 Propellers and Thrusters, 297
Euler’s Method, 131, 135, 147 Propulsion Control, 296
Propulsion System, 9
FEM, 378, 423
FIR Filter, 162 Ramp Function, 127
Fourier Transform, 128 Reference Frames, 193
Ride Control, 440
Helmholtz Equation, 451 Riser, 370, 419
Highpass Filter, 160
HIL Testing, 38, 174 Sampler , 129
Hybrid Control, 267 Sampling Period, 123, 129
Shannon’s Sampling Theorem, 138
IIR Filter, 162 Shift-operator Calculus, 143
Impulse Function, 127 Signal Processing, 2, 150
Signal Reconstruction, 123
Kalman Filter, 167, 230
Signal Testing, 150
Laplace Transform, 127 Step Function, 126
Low-frequency Model Stroboscopic, 123
LF Model, 199, 200 supervisory-switched control, 272

525
Surface Effect Ships, 440
Switching, 271

Thrust Allocation, 20, 259, 303

TLP, 372
Trapezoidal Method
Tustin’s Method, 136, 147

Vessel Operational Condition, 269

Water Current Model, 190

Wave Equation, 446
Wave Filtering, 230
Wave Spectrum, 176
Wave-frequency Model
WF Model, 199, 208
Weather Optimal Positioning, 277
Wind Model, 189
Windowing, 151

z-Transform, 140
Zero-Order-Hold, 122, 129, 131

526