Tesi Raspa

Universit Politecnica delle Marche
Scuola di Dottorato di Ricerca in Scienze dellIngegneria

Curriculum in Ingegneria Informatica, Gestionale e dellAutomazione
Advanced strategies for control and
fault diagnosis of marine surface
vessels
Ph.D. Dissertation of:
Paolo Raspa
Advisor:
Prof. Sauro Longhi
Curriculum Supervisor:
Prof. Sauro Longhi
XI edition - new series
Curriculum in Ingegneria Informatica, Gestionale e dellAutomazione
Advanced strategies for control and
fault diagnosis of marine surface
vessels
Ph.D. Dissertation of:
Paolo Raspa
Advisor:
Prof. Sauro Longhi
Curriculum Supervisor:
Prof. Sauro Longhi
XI edition - new series
Facolt di Ingegneria
Via Brecce Bianche 60131 Ancona (AN), Italy
To my family, friends and beloved one
Acknowledgments
This project was developed with the Department of Information Engineer-
ing of the Polytechnic University of Marche, Ancona, Italy. I wish to thank
Prof. Sauro Longhi for the opportunity to participate in such an important
experience. I am grateful to other members of my research group, Prof. Gi-
anluca Ippoliti and Dr.Ing. Flavia Benetazzo, for the wonderful work we did
together and other colleagues I worked with, who shared their experiences and
good mood and made time at work fun and less wearing. I wish to thank Prof.
Mogens Blanke and Prof. Roberto Galeazzi from the Technical University of
Denmark for introducing me to this exciting research area. I want to express
my sincere gratitude to Prof. Asgeir J. Sorensen and other colleagues and
technicians at the Norwegian University of Science of Technology for giving me
the great opportunity to test my work at the Marine Cybernetics Laboratory
for the invaluable support during my stay. I would nally say thank you to
my family, my friends and my beloved one, since without their support and
patience it would have been terribly harder to reach the end of this journey
Ancona, Gennaio 2013
Paolo Raspa
ix
Abstract
The motion of surface vessels due to waves is signicantly aected by envi-
ronmental disturbances, like waves, currents and wind. The roll-induced high
vertical accelerations increase the possibility of cargo damage and the incidence
of seasickness development, while horizontal motion due to waves, currents,
waves and wind aects the ship ooading and dynamic positioning operation.
Thus the design of control systems to damp the induced roll motion and to
guarantee a reliable stabilization to xed or dynamic position, also in case of
thruster failure, as prescribed from certication agencies. The main issue is
reaching a proper trade-o between the achievable performances and the com-
plexity of the control system in relation to the operating condition and the
coupling between the ship degree of freedom. This thesis presents some novel
contributions to the quantication of fundamental limitation in roll stabilisers
and a novel variable structure control system focused on the robustness re-
spect to model uncertainties, environmental disturbances and thruster failures
specic for rudder-n-roll-stabilers and dynamic positioning. The proposed so-
lutions are validated using simulation tools and software and experimental tests
conducted at the Marine Cybernetics Laboratory of the Norwegian University
of Science and Technologies with a scaled model vessel.
xi
Sommario
Il moto di vascelli superciali pesantemente inuenzato dalla presenza di
onde e altri disturbi ambientali. Il moto di rollio indotto dalle onde provoca
accellerazioni verticali che possono incrementare le possibilit di danni al carico
e lincidenza di malori per passeggeri ed equipaggio, mentre i moti orizzontali
indotti da correnti, onde e vento inuiscono sul preciso posizionamento della
nave nelle operazioni di ancoraggio o di accostamento della nave. E di grande
importanza per la sicurezza e il comfort di equipaggio ed eventuali passeggeri,
nonch sicurezza del carico, adabilit e operabilit della nave che il vascello
sia predisposto con opportuni sistemi di controllo in grado di smorzare il moto
di rollio o garantire un corretto posizionamento della nave in maniera adabile,
quindi anche in caso di guasti alle appendici di controllo, come prescritto dalla
normativa. Per questi aspetti il problema principale risiede nel trovare il giusto
trade-o tra le prestazioni e la complessit del sistema di controllo in rapporto
alle condizioni operative, dovute agli accoppiamenti tra I gradi di libert del
sistema. Questa tesi presenta dei contributi innovativi per la quanticazione
delle limitazioni insite nei sistemi di controllo per lo smorzamento del rollio
attraverso lo studio delle caratteristiche proprie del modello idrodinamico e
propone un innovativo sistema controllo a struttura variabile focalizzato sulla
robustezza in risposta a incertezze sul modello, ai disturbi esterni e a possibili
guasti nel sistema di attuazione della nave nei sistemi di controllo per la sta-
bilizzazione del moto di rollio e per il posizionamento dinamico della nave. Le
soluzioni proposte sono state validati tramite tool e software per la simulazione,
nonch tramite test condotti in vasca presso il Marine Cybernetics Laboratory
della Norwegian university of Science and Technologies con un modello in scala
di un vascello di supercie.
xiii
Contents
List of Figures xvii
List of Tables xix
1 Introduction 1
2 Ship Motion Models 3
2.1 Reference Frames . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Kinematics and Kinetics equations . . . . . . . . . . . . . . . . 6
2.3 Environmental Forces and moments . . . . . . . . . . . . . . . 6
2.3.1 Description of Ocean Waves . . . . . . . . . . . . . . . . 7
2.3.2 Wave forces and moments . . . . . . . . . . . . . . . . . 14
2.3.3 Ship motion in Seaway . . . . . . . . . . . . . . . . . . . 14
2.3.4 Wind forces and moments . . . . . . . . . . . . . . . . . 18
2.3.5 Ocean current forces and moments . . . . . . . . . . . . 18
3 Ship Roll Stabilizing Control 21
3.1 Rudder Roll Damping . . . . . . . . . . . . . . . . . . . . . . . 21
3.1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1.2 Roll Model for Seakeeping . . . . . . . . . . . . . . . . . 24
3.1.3 Frequency Dependent Linear Model for Rudder Roll Damp-
ing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.1.4 Fundamental Limitations in Rudder Roll Stabilisers . . 31
3.1.5 Controller Design and Optimisation . . . . . . . . . . . 34
3.1.6 Control Design . . . . . . . . . . . . . . . . . . . . . . . 36
3.1.7 Control Optimisation . . . . . . . . . . . . . . . . . . . 37
3.1.8 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.1.9 Conclusions RRD . . . . . . . . . . . . . . . . . . . . . . 40
3.2 Fin/Rudder Roll Damping . . . . . . . . . . . . . . . . . . . . . 40
3.2.1 Nonlinear Ship Model for Fin/Rudder Roll Damping . . 42
3.2.2 Control Modules . . . . . . . . . . . . . . . . . . . . . . 45
3.2.3 State Space Representation . . . . . . . . . . . . . . . . 46
3.2.4 Zero Dynamics Analysis . . . . . . . . . . . . . . . . . . 47
3.2.5 Normal Form . . . . . . . . . . . . . . . . . . . . . . . . 47
xv
Contents
3.2.6 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . 49
3.2.7 Controller Design . . . . . . . . . . . . . . . . . . . . . . 50
3.2.8 Case Study . . . . . . . . . . . . . . . . . . . . . . . . . 52
4 Dynamic Positioning 55
4.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Nonlinear Ship Model . . . . . . . . . . . . . . . . . . . . . . . 56
4.2.1 Manoueuvering Model . . . . . . . . . . . . . . . . . . . 57
4.2.2 Environmental Disturbances . . . . . . . . . . . . . . . . 58
4.2.3 Thruster Allocation . . . . . . . . . . . . . . . . . . . . 59
4.2.4 Plant model . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.3 Wave Filtering . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3.1 Multi-rate Kalman Filter . . . . . . . . . . . . . . . . . 62
4.4 Discrete Time Variable Structure Control . . . . . . . . . . . . 63
4.4.1 Control Model . . . . . . . . . . . . . . . . . . . . . . . 63
4.4.2 Computation of the sliding surfaces . . . . . . . . . . . 64
4.4.3 Computation of the control law . . . . . . . . . . . . . . 65
4.5 Fault Tolerant Control . . . . . . . . . . . . . . . . . . . . . . . 69
4.5.1 Thruster Failures . . . . . . . . . . . . . . . . . . . . . . 69
4.5.2 Fault Diagnosis . . . . . . . . . . . . . . . . . . . . . . . 69
4.5.3 Change detection . . . . . . . . . . . . . . . . . . . . . . 72
4.5.4 Fault isolation . . . . . . . . . . . . . . . . . . . . . . . 73
4.5.5 System reconguration . . . . . . . . . . . . . . . . . . . 74
4.6 Control Allocation . . . . . . . . . . . . . . . . . . . . . . . . . 75
4.6.1 Static Allocation . . . . . . . . . . . . . . . . . . . . . . 76
4.6.2 Dynamic Allocation . . . . . . . . . . . . . . . . . . . . 76
4.6.3 DLS Algorithm . . . . . . . . . . . . . . . . . . . . . . . 77
5 Experimental Results 79
6 Conclusions 85
xvi
List of Figures
2.1 Motion superposition. Courtesy of Perez [1] . . . . . . . . . . . 4
2.2 Reference frames . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Reference scheme for the Seakeeping Model . . . . . . . . . . . 7
2.4 Sea state realization for long-crested irregular sea . . . . . . . . 10
2.5 ITTC Modied Pierson-Moskowitz wave and wave slope spectra 12
2.6 Amplitude of regular wave components

i
calculated using MPM
with h
1/3
= 2.5m and T
1
= 8s and a particular realization for
both wave height and wave slope . . . . . . . . . . . . . . . . . 13
2.7 Reference frames and standard notation for the description of
ship dynamics. Courtesy of Perez [1] . . . . . . . . . . . . . . . 15
2.8 Denition of encounter angle and sailing condition. Courtesy
of Perez[1] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.9 Transformation from wave to encounter frequency for a ship
moving with forward speed 15 knots and dierent wave direc-
tions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.1 Roll moment induced by the rudder . . . . . . . . . . . . . . . 21
3.2 Motion RAO amplitude and phase for dierent encounter angles
for the example naval vessel sailing at 15kts . . . . . . . . . . . 28
3.3 Roll motion spectra

i
calculated using the motion RAO and
MPM wave slope spectrum with h
1/3
= 1m for the example
vessel at 15kts for dierent encounter angles and average wave
period T
1
. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4 Amplitude of regular roll component

i
wave slope spectrum with h
1/3
= 1m and T
1
= 7.5s and partic-
ular realizations of roll motion for the example vessel at 15kts
for dierent encounter angles . . . . . . . . . . . . . . . . . . . 30
3.5 Comparison of the Poisson Integral using the approach in [2]
(black dashed line) and the one proposed in this work, which
add the area under the black solid line triangle centred in
s
. 34
3.6 Comparison between (3.27) (red dashed line) and (3.30) (blue
solid line) for the lower bound estimation of S
+
from Poisson
integral. S
is function of the lower bound of the attenuation

bandwidth centred in
s
=
n
. M
1
= 0.5 and M
3
= 0.3 . . . . 35
xvii
List of Figures
3.7 Bode Diagram of the sensitivity function during the optimisation
process. The minimum of the optimisation is given for the red
dashed line. The desired sensitivity is shown as a black dashed
line. The circle line shows the waterbed eects when the con-
troller (3.40) is tuned up with an attenuation bandwidth centred
around 0.7 rad/s . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.8 Roll Angle, Steering Angle and Rate simulations in Sea State
3. Encounter angle = 45
. M
3
= 0.3 at
s
= 1.1135 [rad/s],
M
1
= 0.5 [0.75
s
1.25
s
] . . . . . . . . . . . . . . . . . . . 41
3.9 Reference frames and notation for ship motion description . . . 43
3.10 Main ship values: body-xed frame O
b
{x
b
, z
b
},Center of Grav-
ity (CG), Lateral Center of Gravity (LCG), Vertical Center of
Gravity (VCG), Waterline (WL), Draught (T), Aft Perpendic-
ular (AP), Front Perpendicular (FP), Length between Perpen-
dicular (Lpp). . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.11 Roll and Yaw angles obtained during a 1000 seconds simulation
in sea state 5 - encounter angle 135 deg - encounter frequency
0.86rad/s conditions. The controller action starts at t = 300s.
The reference yaw angle is changed from 0 deg to 10 deg at t =
600s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.12 Rudder and ns eorts during the simulations in sea state 5 -
e
= 45 deg conditions. The controller action starts at t = 300 s.
600 s. The values for the rudder angle and the ns force N
f
do not ever achieve their saturation value , i.e. 40 deg and 10
7
N
respectively, and a good gap from those is kept. . . . . . . . . . 53
3.13 Zero Dynamics behavior. . . . . . . . . . . . . . . . . . . . . . . 53
4.1 Vessel Reference frames . . . . . . . . . . . . . . . . . . . . . . 57
4.2 Thrusters conguration for the oshore supply vessel . . . . . . 59
4.3 Convergence to zero of the sliding surfaces in nominal condi-
tion applying the DTVSC control law. Figure shows that the
DSMEC is satised . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.4 Fault-tolerant system architecture . . . . . . . . . . . . . . . . 70
4.5 Isolation logic . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
4.6 Bench of controllers . . . . . . . . . . . . . . . . . . . . . . . . 74
4.7 Thrusters conguration for the vessel . . . . . . . . . . . . . . . 75
5.1 CyberShip 3 at the MCLab . . . . . . . . . . . . . . . . . . . . 80
5.2 CASE 1 - References (red line) and simulated n, e and us-
ing DTVS controller (blue and dotted line) and PID (black and
dash-dotted line) . . . . . . . . . . . . . . . . . . . . . . . . . . 82
xviii
List of Figures
5.3 CASE 3 - References (red line) and simulated n, e and us-
ing DTVS controller (blue and dotted line) and PID (black and
dash-dotted line) . . . . . . . . . . . . . . . . . . . . . . . . . . 82
5.4 CASE 3 - First (blue line), second (red line) and third (black
line) azimuth thrusters angle and thrust force f . . . . . . . 83
xix
List of Tables
2.1 Nomenclature to describe the ship motion with the reference
frame associated . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 World meteorological sea state denitions . . . . . . . . . . . . 11
3.1 Non-dimensional damping coecient as function of the ship
speed for the naval vessel in ?? . . . . . . . . . . . . . . . . . . 26
3.2 Control system performances . . . . . . . . . . . . . . . . . . . 40
3.3 Control system performances for the vessel sailing with U =
7m/s in slight waves for dierent encounter angles . . . . . . . 51
3.4 Control system performances for the vessel sailing with U =
7m/s in rough waves for dierent encounter angles . . . . . . . 52
4.1 Thruster conguration matrix . . . . . . . . . . . . . . . . . . . 59
4.2 Dependability matrix . . . . . . . . . . . . . . . . . . . . . . . . 70
4.3 Dependency matrix . . . . . . . . . . . . . . . . . . . . . . . . . 72
4.4 Isolation Logic . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.1 Desired Positions . . . . . . . . . . . . . . . . . . . . . . . . . . 81
5.2 Environmental conditions . . . . . . . . . . . . . . . . . . . . . 81
5.3 ISE performance index . . . . . . . . . . . . . . . . . . . . . . . 81
xxi
Chapter 1
Introduction
One of the most largely investigated problem in marine engineering are the roll
damping and the dynamic positioning. The motion of surface vessels due to
waves is signicantly aected by environmental disturbances, like waves, cur-
rents and wind. The roll-induced high vertical accelerations increase the pos-
sibility of cargo damage and the incidence of seasickness development, while
horizontal motion due to waves, currents, waves and wind aects the ship of-
oading and dynamic positioning operation. Thus the design of control systems
to damp the induced roll motion and to guarantee a reliable stabilization to
xed or dynamic position, also in case of thruster failure, as prescribed from
certication agencies. A marine craft is a very complex system to model and
control due to the variety and complexity of the phenomena aecting its dynam-
ics. As a consequence, the denition of an eective mathematical description
which is at the same time not too complicated and suciently realistic for
the design and test of control systems is still an open problem. Roll motion
prevents the whole execution of some tasks. For example, pipe-laying ves-
sels, drilling vessels, air craft carriers and navy ships handling weapons require
small displacement to be able to perform their mission. Recent developments
in ship-building technology, e.g., new light construction materials, innovative
machinery and propulsion systems, along with enhanced control equipment,
introduce new issues and suggest new control challenges. In this scenario, the
role of the ship motion control system gains a crucial role.
On the other hand exploration and exploitation of hydrocarbons have opened
up an era of dynamically positioned (DP) vessels. The number of vessels
equipped with dynamic positioning (DP) systems rose in the recent years due
to increasing oil and gas exploration at sea. The DP is an autonomous control
system that acts to maintain the vessel position and the angle of direction at a
reference point by means of the vessel propulsion and maneuvering thrusters.
Knowledge of thruster allocation, combined with information from Global Nav-
igation Satellite Systems (GNSS) and compass, is used to calculate the steering
angle and the thrust for each thruster. The control action maintains the de-
sired position and orientation according to a navigation path or a specic task
1
Chapter 1 Introduction
(absolute or relative DP). The dynamic positioning system is decisive in those
situations in which the position of the unit is bound to a specic point on the
seabed (absolute DP), or it is related to a moving unit, like when the ship is
operating with other vessels or for remotely operated underwater vehicles. A
dynamically positioned (DP) vessel is by the International Maritime Organiza-
tion (IMO) and the certifying class societies (DNV, ABS, LR, etc.) dened as a
vessel that maintains its position and heading (xed location or pre-determined
track) exclusively by means of active thrusters, [43]. Other solutions like posi-
tion mooring consider the aid of mooring lines, see [44].
The design of control systems to damp the induced roll motion and to guaran-
tee a reliable stabilization to xed or dynamic position, also in case of thruster
failure, as prescribed from certication agencies. This thesis presents a re-
view of Ship Motion Models in Chapter 2, proposes some novel contributions
to the quantication of fundamental limitation in roll stabilisers in 3 and a
novel variable structure control system focused on the robustness respect to
model uncertainties, environmental disturbances and thruster failures specic
for rudder-n-roll-stabiler in Section 3.2 of Chapter 3 A discrete-time version
control systems is presented in 4 for the the dynamic positioning of a surface
vessel. The proposed solutions are validated using simulation tools and soft-
ware and experimental tests conducted at the Marine Cybernetics Laboratory
of the Norwegian University of Science and Technologies with a scaled model
vessel.
2
Chapter 2
Ship Motion Models
In this part the dynamical model of the ship is presented in linear and non
linear form and the wave induced motion is introduced. The system structure
is introduced and simulation tests are carried out This chapter gives a general
description of the ship dynamics . A manoeuvring non linear model and linear
model are introduced. All the material presented in the chapter is derived from
Perez [1]
The seakeeping model is used, because the sea forces and moments cannot
be inputs for the maneuvering model shown in Chapter ??.
As stated in [1] the seakeeping and maneuvering models are combined for
control system testing: the seakeeping model gives the eects of a sea scenario
on the ship motion added to the outputs of the maneuvering model, which
describe the interaction between the control inputs and the induced motion,
see Fig. 2.1
2.1 Reference Frames
It is possible to consider a ship as a rigid body moving in the space. Thus its
dynamic can be studied considering a 6DOF model. There are three coordinates
to dene the orientation and three coordinates to dene the translations. The
coordinates are dened using two types of reference frames: inertial frame and
body xed frame. There are four reference frame usually considered for marine
vessels:
North-east-down frame (n-frame) (o
n
, x
n
, y
n
, z
n
). It is considered
as an inertial frame xed to the earth. The x
n
-axis points towards the
North, the y
n
-axis points towards the East and the z
n
-axis points to-
wards the center of the Earth. The origin is located in the mean water
free-surface-see Fig. 2.2.
Geometric frame (g-frame) (o
g
, x
g
, y
g
, z
g
) . It is a non-inertial frame
xed to the ship hull. The x
g
-axis points towards the bow, the y
g
-axis
3
Chapter 2 Ship Motion Models
Figure 2.1: Motion superposition. Courtesy of Perez [1]
points towards starboard and z
g
points upwards. The origin o
g
is located
along the longitudinal center line at the intersection between the base-
line (BL) and the aft perpendicular (AP), which is taken at the rudder
stock-see Fig. 2.2.
Body-xed frame (b-frame) (o
b
, x
b
, y
b
, z
b
) . It is an non-inertial frame
xed to the ship hull. The x
b
-axis points towards the bow, the y
b
-axis
points towards starboard and the positive z
b
-axis points downwards. The
axis are chosen to coincide with the principal axis of inertia. Their inter-
section determines o
b
[1]-see Fig. 2.2.
Hydrodynamic frame (h-frame) (o
h
, x
h
, y
h
, z
h
) . It is no xed to the
hull and moves at the average speed of the vessel following its path. The
x
h
-y
h
plane coincide with the mean water free surface. The positive x
h
axis points forward and it is aligned with the low frequency yaw angle

.
The positive y
h
-axis points towards starboard, and the positive z
h
-axis
passes through the time-average position of the center of gravity. The
frame is considered inertial. -see Fig. 2.2.
All the measurements taken on board are referred to the b-frame that is also
used to express the equations of motion for the problem addressed in this work.
The linear velocities respect to the b reference frame (o
b
, x
b
, y
b
, z
b
) are re-
spectively called Surge velocity, Sway velocity and Heave velocity. While the
angular velocities are respectively called Roll rate, Pitch rate, Yaw rate. They
are the time derivatives of the angles Roll ( rotation around the x
b
axis), Pitch
4
2.1 Reference Frames
Figure 2.2: Reference frames considered for the description of ship dynamics.
Courtesy of Perez [1]
Symbol Name Description Ref. frame
n North Position x coordinates of o
b
respect to the n-frame n
e East Position y coordinates of o
b
d Down Position z coordinates of o
b
u Surge velocity Linear velocity in the x direction b
v Sway velocity Linear velocity in the y direction b
w Heave velocity Linear velocity in the z direction b
p Roll rate Angular velocity referred to the x axis b
q Pitch rate Angular velocity referred to the y axis b
r Yaw rate Angular velocity referred to the z axis b
Roll angle Angular position referred to the x axis b
Pitch angle Angular position referred to the y axis b
Yaw angle or heading angle Angular position referred to the z axis b
Table 2.1: Nomenclature to describe the ship motion with the reference frame
associated
(rotation around the y
b
axis ) and Yaw (rotation around the z
b
axis). A sum-
mary of the nomenclature to describe the motion coordinates is reported in
Table 2.1.
Referring to Table 2.1 it is possible to dene

nb
[, , ]
T
: vector of the Euler angles. The angles represent the
rotations necessary to make the n-frame coincide with the b-frame. The
rotation order is , and
[n, e, d, , , ]
T
: position orientation vector.
[u, v, w, p, q, r]
T
: linear angular velocity vector
5
2.2 Kinematics and Kinetics equations
The dynamics of marine vessels can be divided in two parts, the kinematics
and the dynamics. The latter analyses the forces causing the ship motion,
while the former accounts only for the geometric aspects of the motion. The
kinematics equations, which express the relationship between the generalized
displacements in the {n} frame and the velocities in the {b} frame, are:
= J() (2.1)
From [14], in case of irrotational and constant ocean currents, the 6 degrees of
freedom (DOF) maneuvering equations of motions can be expressed in the {b}
frame by
M
r
+C (
r
)
r
+D(
r
)
r
+G =
c
+
env
(2.2)
where
r
= [u u
c
, v v
c
, w, p, q, r]
T
is the relative velocity vector between
the vessel and current, as explained in Section 2.3.5; M = (M
RB
+M
A
) is the
system inertia matrix, including the rigid-body and the added mass matrices;
C = C
RB
+C
A
() is the Coriolis-centripetal matrix, including both rigid-body
and added mass; D(
r
) = D
L
+ D
NL
(
r
,
r
) is the damping matrix, which
may be divided into a linear component, D
L
, accounting for linear wave drift
damping and laminar skin frictions and a non linear component D
NL
(
r
,
r
)
accounting for the eects of ocean currents. It is important to notice that, for
velocities of vessel close to zero, the linear damping becomes more signicant
than the nonlinear damping. The restoring term is assumed to be G() = G
under the assumption of small roll and pitch angles. The
c
are the control
forces and moments to be produced by the actuators, while
env
are the envi-
ronmental loads.
2.3 Environmental Forces and moments
This section gives a description of the mathematical models for the ocean waves,
currents and wind disturbances to be included in the ship motion model. The
theory presented in this chapter comes from [3] amd [14] A ship in a seaway is
mainly aected by the following types of environmental disturbances: waves,
currents, wind. The environmental disturbances contain both slowly varying
and high-frequency forces. Control forces and moments due to environmental
disturbances are coursed by wind and waves. Using the principle of superposi-
tion, they are added to the right side of (4.1) by dening:
env
=
wind
+
wave
.
A ship in a seaway is mainly aected by the following types of environmental
disturbances:
Wind generated waves
6
Figure 2.3: Reference scheme for the Seakeeping Model
Wind
Ocean currents
The seakeeping model uses linear equations to describe the ship response to
sea forces and moments acting on the hull, see [4]. The analysis techniques
proposed in the literature are developed in the frequency domain, aiming to
describe the ship behavior in a statistics framework. The key element for this
methods are the Response Amplitude Operators (RAO), described in Section
2.3.3, which can be seen as transfer function from an external excitation to the
ship motion as function of the excitation frequency. The RAO can combined
with the wave spectrum (described in Section 2.3.1 to obtain the power density
spectrum of the ship motion.
The scheme of the seakeeping model is given in Fig. 2.3
The assumptions to use the seakeeping model are:
The motion equations are described in an inertial reference frame (h-
frame, see Section 2.1) traveling at the mean forward speed and with the
mean direction of the ship
The added mass and damping coecient in the motion equations are fre-
quency dependent. The solution, for a set of frequencies, give amplitude
and phase for the motion components as function of the frequency (RAO)
The model is not accurate at low frequencies
Remark The motion superposition with the wave disturbance acting as an
output disturbance can be used only in the case of a single rigid body, be-
cause the seakeeping model is not comprehensive of the interactions between
dynamical systems in term of energy
2.3.1 Description of Ocean Waves
Waves have the main eects on the ship roll motion thus only this kind of
disturbance is considered in this work. Ocean waves are random in terms of
7
both time and space. Therefore, a stochastic modelling description is the most
appropriate approach to describe them.
Before characterizing them the following assumptions are adopted as stated
in [3]:
The variations of the stochastic characteristics of the sea are much slower
than the variations of the sea surface itself.
The observed sea surface, at a certain location and for short periods of
time, is considered a realization of a stationary and homogeneous, zero
mean Gaussian stochastic process.
Standard formulae for the spectral density function S() are adopted
The validity of the hypothesis of stationarity and Gaussianity, have been
investigated via extensive analysis of time series recorded from wave riding
buoys in the North Atlantic Ocean [5]. In this study, it has been reported that:
For low and moderate sea states (Signicant wave height h
1/3
< 4m), the
sea can be considered stationary for periods over 20 min. For more severe
sea states, stationarity can be questioned even for periods of 20 min.
For low to medium states (h
1/3
< 8m), Gaussian models are still accu-
rate, but deviations from Gaussianity slightly increase with the increasing
severity of the sea state.
Under these assumptions wind-generated waves can be represented as a sum
of a large number N of sinusoidal components with random phases. The sea
elevation at a location (x, y) with respect to an inertial reference frame is given
by:
(x, y, t)
N
i=1
i
(x, y, t)
=
N
i=1
i
cos(
i
t +k
i
(xcos +y sin ) +
i
)
(2.3)
Where for every regular travelling wave component i the parameters are:
i
: wave frequency [rad/s]
i
: wave amplitude [m]
k
i
: wave number [rad]
i
: random phase [rad]
: angle between the ship heading and the direction of the wave [rad]
8
Remarks For each realization, the phase
i
of each component is chosen to
be a random variable with uniform distribution on the interval [, ]. This
choice ensures the stationarity of the approximation [6]. Once the phases
i
have been chosen, an approximation of a realization of the stochastic process
(x, y, t) is obtained.
However, it should be borne in mind that since this approximation is a sum
of deterministic sinusoids, the sea elevation pattern will repeat itself after 2
/ max
i
. Where
i
is the band of frequencies centered at
i
and dened
as (
i+1
i1
)/2.
For each regular wave component i, the phase velocity, c
i
, is the velocity
with which the wave crest moves relative to ground. Assuming innite depth
of water, the following relations hold:
c
i
=
_
g
i
2

i
=
g
c
i
k
i
=
2
i
=

2
i
g
(2.4)
where
i
is the wavelength. The second expression in Eq. (2.4) is known
as the dispersion of gravity waves, and establishes that the phase velocity is
inversely proportional to its frequency. This means that long waves propagate
faster than short ones. This phenomenon is important for it means that a ship
advancing in following seas will overtake some short waves, while it will be
overtaken by some long ones.
Considering r = xcos +y sin and Eq. (2.3) the wave slope is dened as:
(x, y, t) =
d
dr
(x, y, t) =
N
i=1
k
i
i
sin(
i
t +k
i
(xcos +y sin ) +
i
) (2.5)
Remark The wave slope is important, because in the case of roll and pitch
motion of ships, it is the slope of the waves rather than the height what excites
the motion; and due to this the roll and pitch frequency responses are often
given relative to the wave slope rather than the amplitude, see [1].
Sea spectrum As stated before, the observed sea surface can be seen as a
stochastic process. Therefore the expression given in Eq. (2.3) is considered
a realization of a stationary and homogeneous, zero mean Gaussian stochastic
process. An useful method to deal with a stochastic process is to consider its
Power Density Spectrum (PSD) which gives the power distribution as function
of the frequency.
9
0
200
400
600
0
100
200
300
400
500
1
0.5
0
0.5
1
1.5
y [m]
x [m]
Sea state realization, 10 wave components
z

[
m
]
Figure 2.4: Sea state realization for long-crested irregular sea
S
() = c
1
_

()e
j
d
R
() = c
2
_

()e
j
d
(2.6)
where R
() is the autocorrelation of the stochastic process.

As stated in [7], in any particular sea state, the elevation of the ocean surface
presents irregular characteristics. If the irregularity of the observed waves are
only in the dominant wind direction so that there are mainly uni-directional
wave crests with varying separation and remaining parallel to each other the
sea is referred to as a long-crested irregular sea, see Fig. 2.4.
Based on extensive data collection, mostly in the North Atlantic ocean, a
series of idealized single side local spectra have been obtained to describe long-
crested sea.
The power density spectrum for fully-developed long-crested-sea used is rec-
ommended by the 2nd international Ship and Oshore Structures Congress
(ISSC, 1964) and 12th ant 15th International Towing Tank Conference (ITTC,
1969, 1978), under this assumptions:
Fully-developed long-crested seas condition
Innite water depth, no swell and unlimited fetch
10
The spectrum is referred as the Modied Pierson-Moskovitz (MPM) spectrum:
S
MPM
() =
A
5
exp
B
4
A =
172.75H
2
1/3
T
4
1
B =
691
T
4
1
(2.7)
H
1/3
is the signicant wave height (average of the heights of largest 1/3rd of
the waves), T
1
is the average wave period (1 / average wave frequency of the
spectrum).
The signicant wave height is used to dene the sea state. Table 2.3.1 shows
the sea state code commonly used to describe the seaway in marine applications.
Sea state code H
1/3
Seaway description
0 0 Calm (glassy)
1 [0 0.1] Calm (rippled)
2 [0.1 0.5] Smooth (wavelets)
3 [0.5 1.25] Slight
4 [1.25 2.5] Moderate
5 [2.5 4] Rough
6 [4 6] Very rough
7 [6 9] High
8 [9 14] Very high
9 > 14 Phenomenal
Table 2.2: World meteorological sea state denitions
From Eq. (2.5), it follows that the wave slope spectrum is given by (see [1]):
S
() = k
2
S
() =

2
g
S
() (2.8)
In Fig. 2.5 ITTC spectra are shown for dierent values of the signicant
wave height H
1/3
and average wave period T
1
. From the gure it is clear that
the higher the period the lower is the peak frequency.
Under Gaussianity assumption the sea elevation can be completely charac-
terized by its PSD. Considering the Eq. (2.3) it is clear that the only unknown
parameters are the amplitudes of the wave components

i
. They can be con-
veniently dened in terms of the adopted local PSD which has to satisfy:
var [(t)] =
_
+
0
S
()d (2.9)
Following [3], let us consider for simplicity the case in which the observations
11
0 0.5 1 1.5 2 2.5 3
0
1
2
3
ITTC MPM Spectra
[rad]
S
)

[
m
2
/
s
]

H
1/3
=2.5m T
1
=6s
H
1/3
=2.5m T
1
=8s
H
1/3
=2.5m T
1
=10s
H
1/3
=4m T
1
=6s
H
1/3
=4m T
1
=8s
H
1/3
=4m T
1
=10s
0 0.5 1 1.5 2 2.5 3
0
0.005
0.01
0.015
[rad]
S
)

[
m
2
/
s
]
Figure 2.5: ITTC Modied Pierson-Moskowitz wave and wave slope spectra
are made at the origin of the inertial reference frame (x=0 and y=0) with an
angle of incidence of the waves = 0 the model of the sea elevation in Eq.
(2.3) becomes:
(t)
N
i=1
i
(0, 0, t) =
N
i=1
i
cos(
i
t +
i
) (2.10)
Where, at any particular wave frequency
i
, the variance of that component
is given by:
var [(t)] =
1
2

2
i
=
_

i
+
i
2
i
2
S
()d (2.11)
The amplitudes of the wave components are then determined as:
i
=
_
2
_

i
+
i
2
i
2
S
()d
=
_
2S
i
)
i

i
2
,
i
+

i
2
_
(2.12)
The approximation given in Eq. (2.3) with amplitudes of the sinusoidal
components calculated using (2.12) can be interpreted as the PSD S() of the
sea elevation being approximated by:
12
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0
0.05
0.1
0.15
0.2
[rad]
i

[
m
]
Amplitude of regular components

Wave Height
Wave Slope
0 20 40 60 80 100 120 140 160 180 200
2
1
0
1
2
t [s]
(
t
)
,

(
t
)

[
m
]
Sea Realization

Wave Height
Wave Slope
Figure 2.6: Amplitude of regular wave components

i
with h
1/3
= 2.5m and T
1
= 8s and a particular realization for both
wave height and wave slope
S() =
N
i=1
2
i
2
i
(
i
t) (2.13)
so that the variance satises:
var [(t)]
_
+
0
S()d =
_
+
0
S() (2.14)
Given a wave spectrum S
(), a realization of the sea elevation can be

obtained combining Eqs. (2.3) and (2.12):
(x, y, t) =
N
i=1
_
2S
i
)
i
cos(
i
t +k
i
(xcos +y sin ) +
i
)

i
2
,
i
+

i
2
_
(2.15)
A realization of the sea elevation and the wave slope along with the amplitude
of the regular wave components is given in Fig. 2.6.
13
2.3.2 Wave forces and moments
When designing a ship control system it is important to evaluate the robustness
and performance in the presence of waves. A motion control system can be
simulated under inuence of wave-induced forces by separating two eects:
the 1-st order wave-induced forces,
wave1
, wave loads observed as zero-mean
oscillatory components and the 2-nd order wave-induced forces,
wave2
, wave
drifts forces observed as non-zero slowly varying components. According to [14]
wave forces can be modelled by a linear superposition of the two components,
therefore
wave
=
wave1
+
wave2
.
Wave force models depends on response amplitude operators (RAOs) com-
puted for the particular craft using a hydrodynamic program from hull geome-
try. For each degree of freedom (dof {1...6}) in (??), the wave loads and drifts
can be represented as a sum of a large number N of sinusoidal components:
{dof}
wave1
=
N
k=1
w
g
F
{dof}
wave1
A
k
cos
_
e
k
t +F
{dof}
wave1
+
k
_
{dof}
wave2
=
N
k=1
w
g
F
{dof}
wave2
A
2
k
cos (
e
k
t +
k
)
(2.16)
Where the normalised force RAO F
{dof}
wavei
= F
{dof}
wavei
(
k
,
wave
) are complex
variables depending on the wave frequency component
k
and the relative
wave direction
wave
,
w
is the water density, A
k
is the wave amplitude for the
wave component k, depending on the wave spectrum, signicant wave height
H
s
and modal frequency w
0
. The wave encounter frequency is
e
k
=
k

_
2
k
U cos (
wave
)
_
/ (2g), U is the ship forward velocity and
k
is a random
phase component drawn from an uniform distribution in [ ], to ensure
stationarity of the time series.
2.3.3 Ship motion in Seaway
The motion response of a ship due to the waves is complex. Having a velocity
of advance, a ship experiences a wave excitation at frequencies that dier from
the wave frequencies seen from a xed point. The frequency excitation ob-
served from the ship is referred to as the encounter frequency. This frequency
varies with ship speed and predominant direction between waves in a nonlinear
fashion, see [3].
Encounter frequency and encounter wave spectra The sea state observed
from the ship is aected signicantly by a Doppler shift in the frequency com-
ponents of the wave pattern. If we assume a body-xed b-frame oriented in
14
Figure 2.7: Reference frames and standard notation for the description of ship
dynamics. Courtesy of Perez [1]
Figure 2.8: Denition of encounter angle and sailing condition. Courtesy of
Perez[1]
such a way that the x
b
axis coincides with the xed reference n-frame x-axis
and the ship moving with a forward speed U, see Fig. 2.7 and Section 2.1, then
the relation between the two reference frames is given by:
n = x
b
+Ut
e = y
b
(2.17)
Substituting these equations into the Eq. (2.3), and considering a xed
direction , see Fig. 2.8, the following is obtained:
(x
b
, y
b
, t) =
N
i=1
i
cos((
i
k
i
U cos )t +k
i
(x
b
cos +y
b
sin ) +
i
) (2.18)
15
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
[rad]
e

[
r
a
d
]
Transformation to encounter frequency. U = 15 knots

45 deg
90 deg
135 deg
Figure 2.9: Transformation from wave to encounter frequency for a ship moving
with forward speed 15 knots and dierent wave directions
where the coecient multiplying t is dened as the frequency of encounter:
e
i
=
i

2
i
g
U cos (2.19)
The inverse of the non-linear transformation given in Eq. (2.19) is multi-
valued in some cases, which means that the same encounter frequency can be
obtained from dierent wave frequency, as shown in Fig. 2.9.
Following [3], the relationship between the spectrum in the wave frequency
domain and the encounter wave spectra is based on the point that the energy
does not change if the point of observation changes, therefore
E =
_

0
S
()d =
_

0
S
(
e
)d
e
(2.20)
where
S
()d = S
(
e
)d
e
(2.21)
Therefore, using Eq. (2.19), it follows that the spectrum observed by the
ship is given by:
S
(
e
) =
S
()
d
e
d
=
S
()
1
2
g
U cos
(2.22)
16
This transformation is singular in some cases (e.g. = arccos
g
2U
). Once
the encounter wave spectrum has been obtained, it is necessary to consider
the the ship motion. Thus two approaches are generally adopted:The rst
approach consists of approximating the forces and moments generated by each
regular component using formulas for a rectangular parallelepiped and include
these terms in the equations of motion as proposed by Kllstrm, see [8] and
[9]. The second approach uses the so-called Amplitude Response Operators
(RAO) and relates the sea elevation to the particular motion component under
consideration. In this method, the forces and moments are integrated over
the wetted surface taking into account the load conditions, the geometry of
the hull and the sailing conditions. The former method neglects the moments
acting on the roll motion, therefore in the sequel the latter approach will be
followed. Both methods are based on the simplifying assumption that the
motion response of the ship is linear with respect to the wave amplitude in
regular waves, i.e., superposition holds.
Motion RAO The Motion Response Amplitude Operators represent a linear
approximation of the frequency responses of the ship motion components in
regular waves. Under the linearity assumption, superposition can be applied
to determine the motion of the ship; and therefore, a connection to the model
for describing irregular waves given in the previous section can be established.
The RAO depend on the geometry of the hull and load conditions of the
vessel as well as its speed and direction with respect of the waves. The motion
of a ship in six degrees of freedom is considered as a position and motion in
three directions: surge, sway, and heave; and as an orientation and rotation
about three axis: roll, pitch and yaw, see Fig 2.7.
The magnitudes describing the position and orientation of the ship are usu-
ally expressed in the inertial n-frame and the coordinates are noted: [ne d]
T
and [ ]
T
respectively, see Fig. 2.7 and Section 2.1. The position-orientation
vector expressed in the n-frame is = [n, e, d, , , ]
T
.
The motion RAO from the sea elevation to the motion component
i
of
will be denoted:
R
(
e
, , U) = |R
(
e
, , U)| arg (R
(
e
, , U)) (2.23)
The encounter angle is dened as zero astern and positive towards the port
side of the vessel, see Fig. 2.8. For a particular load condition, encounter angle
and speed U , the RAO are determined using strip theory computations in
which the forces and moments are integrated over the wetted surface of the
hull.
For convenience, the RAO are usually given as a function of the wave fre-
quency instead of encounter frequency. To transform the RAO to encounter
17
frequency, the transformation given in Eq. (2.19) has been used.
The motion RAO can be combined with the sea elevation or sea slope spec-
trum to obtain the motion spectra:
S
i
(
e
, , U) = |R
(
e
, , U)|
2
S
(
e
)for i=1,2...6 (2.24)
where S
(
e
) is given in Eq. (2.22).
If the motion RAO of the rotational components are normalized by the wave
slope, a similar expression holds for the wave slope spectrum:
S
i
(
e
, , U) = |R
(
e
, , U)|
2
S
(
e
)for i=1,2,3 (2.25)
where S
(
e
) is Eq. (2.8) transformed according to Eq. (2.22).
2.3.4 Wind forces and moments
The wind force on a marine vessel is proportional to the projected area above
the waterline and to the square of the wind speed relative to the vessel. The
total relative wind speed can be dened as:
U
rw
=
_
u
2
rw
+v
2
rw
(2.26)
where u
w
= u V
w
cos(
rw
) and v
w
= v V
w
sin(
rw
) are the wind speeds
components, V
w
is the wind velocity,
rw
=
w
is the relative wind angle
and
w
is the wind direction. The parameters above are expressed in the {n}
frame (see Fig. 4.1). Then the wind loads are given by [14]
wind
=
1
2
a
U
rw
|U
rw
|C
w
(
w
)A
xyz
L
xy
(2.27)
where
a
is the air density; A
xyz
is a matrix whose coecients are the lateral
(A
x
) and longitudinal (A
y
) areas of the non-submerged part of the ship pro-
jected on the xz- and yz-plane; L
xy
is a matrix whose coecients are overall
length of vessel (L
oa
) and vertical distances between transverse and longitu-
dinal origin and the wind load point of attack, (L
xy
and L
yz
); C
w
(
w
) are
the non-dimensional wind coecients. These coecients are often derived by
model testing or by semi-empirical formulas, as in [15].
2.3.5 Ocean current forces and moments
The eect of sea current can be divided into 2 components: potential and vis-
cous. The potential component of current includes the so called Munk moment
and is formulated according to [16] as C
A
(
r
) in (4.1).
18
The horizontal current components in surge and sway are dened as:
u
c
= V
c
cos(
c
) v
c
= V
c
sin(
c
), (2.28)
where V
c
and
c
are the current velocity and direction respectively, see Fig.
4.1. The total relative current vector is then dened as:
U
cr
=
_
u
2
r
+v
2
r
, (2.29)
where u
r
= u u
c
, and v
r
= v v
c
. The eect of the current are normally
included in the nonlinear damping term D
NL
(
r
,
r
) of (4.1) as a function of
the relative velocity vector
r
and the relative drag angle,
r
= arctan(v
r
, u
r
).
19
Chapter 3
Ship Roll Stabilizing Control
3.1 Rudder Roll Damping
Roll is the largest and most undesirable ship motion component, [1]. It causes
vertical and transversal accelerations which can interrupt the tasks performed
by the crew or even keep the crew from performing the task at all. It also
contributes to the development of sea sickness in both, crew and passengers with
a consequent comfort deterioration. Furthermore ships roll motion can produce
cargos damage and limit the possibilities to handle on board equipment.
The main purpose of the rudder is to control the heading direction of the
ship in course-keeping manoeuvres, but it can provide roll motion too, [9]. The
latter assertion is based on the structural fact that the rudder is located aft
and also below the centre of gravity of the vessel, as shown in Figure 3.1 thus
it can impart both yaw and roll moment. Therefore, it is possible to use the
ruder to achieve two dierent purposes: heading control and roll damping.
A RRD system is relatively inexpensive since the device is already present on
the ship, even though an up-grade of the rudder machinery could be necessary,
and does not require extra space or extra weight on the boat. In addition, the
RRD can be combined with other roll damping techniques to achieve higher
performances. In any case, a sophisticated control system is necessary to as-
Figure 3.1: Roll moment induced by the rudder
21
Chapter 3 Ship Roll Stabilizing Control
sure good performances under dierent sailing conditions. An overview of the
most important contribution in rudder roll stabiliser is reported in [29], [30]
and [1]. Ships roll stabilising controllers aim at reducing the eect on the roll
component of the waves impacting on the ships hull. From the point of view
of control design, this is seen as the eect of an output disturbance, spread
over a certain bandwidth, on the ships roll response. As a consequence, it is
reasonable to design the controller to directly shape the output sensitivity func-
tion under the Internal Model Control (IMC) framework, characterised by a
model-based approach that provides explicit relationship between the structure
of the controller and that of the plant model, [31]. Unfortunately limitations
apply on the achievable performances of the closed loop system, due to the
Non-Minimum Phase (NMP) of the rudder to roll response. Indeed, under
linearity assumptions, the transfer function from rudder angle (or rate) to roll
angle (or rate) has a Right-Half Plane (RHP) zero, hence an attempt to damp
the roll motion over a certain bandwidth amplies it over other frequencies,
see [32]. [27] and later [2] (see also references therein) proposed methods for
evaluating the relationship between the desired damping bandwidth, the unde-
sired disturbance amplication and the position of the RHP zero. The result is
the estimation of a lower bound for the disturbance amplication outside the
range of reduction. Their studies are signicant because the output disturbance
induced by the waves has energy distributed over a wide range of frequencies.
In this work we present the design phases of a controller for the RRD problem.
The controller is derived using the IMC principle, directly shaping the output
sensitivity function because it relates the waves disturbance to the ships roll
motion, following the work in [33], [30], [34]. The controller is derived solving a
non-linear constrained optimisation problem, where the optimisation function
quanties the trade-o between the damping action and the undesired ampli-
cation of the output disturbance, while problem constraints are given from
specications for the close loop controlled system. The bound on the allowable
disturbances amplication is derived from the study of the Poisson integral to
which a better approximation is provided with respect to previous results. The
controllers performances, after the optimisation stage, are tested on simula-
tion comparing them with the performances obtained with a Minimum Variance
Cheap Limiting Optimal Control approach used as a benchmark, [1].
The thesis is organised as follows. Section ?? resumes the dynamic of ship
motion in seaway. Section 3.1.4 extends the analysis of the achievable perfor-
mance for RRD systems. Section 3.1.5 shows the controller design following a
procedure that combines the sensitivity function loop shaping in the optimi-
sation theory. Section 3.1.8 presents the results of the controller testing and
draws conclusions of the work.
22
3.1.1 Background
An overview of the history of rudder roll stabilizer is adapted from [1]. The
idea of using the rudder as stabilization device emerged from observation of
unusual ship roll behaviour under autopilot operations. Taggart [17], in the
1970s, tested and autopilot under dierent sea conditions and reported an ob-
served signicant roll motion induced by the rudder. Thus the idea to use the
rudder not to increase roll motion but to damp it. The rst full scale trial of a
rudder roll stabilizer was performed in 1972 by van Gunsteren aboard the mo-
tor yacht M.S. Peggy in inner waters of The Netherlands. Although the idea of
using the rudder as roll stabilising mechanism ignited in the 1970s the results
were in general poor due to the simple control strategies allowed by analogue
computers. The beginning of 1980s brought digital computers and allowed to
implement sophisticated algorithms. Thus the experiments of roll stabilizer
achieved more successful results. The most of the citations made in contem-
porary literature regarding RRS still refers to the work done in the 1980s. A
signicant contribution in the eld of rudder roll stabilization is due to Dutch
researchers. They implemented an LQG controller changing its parameters to
avoid the saturation of the steering machinery. The controllers have been tested
with full scale trials but they did not give the expected results. Most of the
work has been reported in the doctoral thesis of van der Klugt [18] and by van
Amerongen et al. [19].Another method to include the saturation in the LQG
controller developed in The Netherlands was proposed by Laudval and Fossen
[20]. In the meanwhile the Royal Danish Navy introduced RRS on the SF300
vessels. The vessels have three rudders and three propellers and the two wing
rudders are used to damp the roll.The results of some test were presented by
Blanke et al in 1989 [21].The controllers were designed using an LQG approach
modied to take in account the actuator saturations when sailing at low speed.
The SF300 is a multi-role naval vessel and dierent variation of the loading con-
ditions can be expected according to the mission performed. This, in addition
to a motion spectra dierent than those taken in account during the controller
design produced performances lower than the expected. The controller design
was then reassessed and an H
approach was followed. Further details are

given in [22]. In Sweden, Kallstrom et all [23] reported the implementation of
RRS on ships of the Royal Swedish Navy that evolved in a commercial product
called ROLL-NIX. The control algorithm was based on LQG and incorporated
adaptation mechanism to cope with dierent weather conditions. The work on
the RRS in the USA started in the 1970s with the rst generation of analogue
controllers and RRS trials were conducted on the coast guard WECH Hamil-
ton and continued in the 1980s with the digital controllers. Some results on
RRS employed by the U.S. Navy on a DD-963 (Spruance) class destroyer were
reported in Baitis and Schmidt 1989 [24]. The use of RRS within the U.S.
23
Navy has grown and nowadays several classes are equipped with these systems.
During the 1990s there was a signicant research on the theoretical aspect of
RRS. In particular the robustness properties of the controller and adaptive
techniques gained much attention. Blanke and Christensen [25] studied the
sensitivity of the performance of LQ control to variations in the coupling co-
ecients of the equations of motion. They used a linear model based on the
hydrodynamic data estimated during the design stage of the SF300 vessels.
Kallstrom and Schultz [26] described further the merits of ROLL-NIX and its
adaptive properties. Hearns and Blanke [27] proposed the use of Quantitative
Feedback Theory to design cascade of SISO controllers for roll and yaw which
targeted the problem of uncertainty i n the model. Laudval and Fossen [28]
proposed the use of the sliding mode controller. This is, perhaps, the only
reference in the literature that uses a non linear approach.
3.1.2 Roll Model for Seakeeping
For the purpose of this thesis the only interesting ship motion is the roll motion,
which is usually computed using the motion RAO normalized by the wave slop.
Applying Eq. (2.25),stating =
1
, the roll motion spectra is:
S
(
e
, , U) = |R
(
e
, , U)|
2
S
(
e
) (3.1)
In this section the method proposed Perez in [10] has been followed to sim-
ulate roll motion for roll damping applications.
The roll motion is coupled with the other horizontal motion components of
sway and yaw. However, for seakeeping analysis it can be treated in isolation,
see [11]. Therefore, in this work, the roll motion is assumed decoupled from
the other lateral motion components.
Non-Linear Roll Motion Model
The roll motion can be expressed as:
a
+g
d
(

) +g
r
() = f(t) (3.2)
where f(t) represents the moment of the excitation forces acting on the on
the hull.
The non-linear total damping function g
d
(

) is usually given by:
g
d
(

) = K
p

+K
p|p|

2
(3.3)
The non-linear restoring function depends on the righting arm and it is usu-
ally expressed in a Taylor series expansion as:
24
g
r
() = K
+K
3
(3.4)
The hydrodynamic derivatives values K
p
, K
pp
, K
, K
are given in Section

??.
A non-linear expression for g
r
() is given by Eq.(??) and a linear approxi-
mation is given by Eq. (??), reported here:
g
r
() GMtg = K
(3.5)
where GMt is the transverse metacentric height.
The constant a = I + I
add
is the total moment of inertia consisting of the
moment of inertia I and added moment of inertia I
add
.
The roll moment of inertia I is determined in terms of the roll radius of
gyration k
roll
= 0.35BOA and displacement as I = k
2
roll
. Where
is the water density.
The roll added moment of inertia is estimated as I
add
= 0.3I.
The values of the parameters GMt, BOA and are reported in Section ??.
In [10] Perez shows the result of a comparison between the non-linear roll
motion model given in Eq. (3.2) and a tank testing data to validate the model.
Linear Roll Motion Model
A linear approximation for the roll motion given in Eq. (3.2) is obtained using
the response of the following second order linear system:
a
+b

+c = f
w
(t) (3.6)
The Laplace transform of (3.6) gives the transfer function from the wave
excitation moment f
to the roll motion:

H(s) =
(s)
F
(s)
=
K
2
0
s
2
+ 2
0
s +
2
0
(3.7)
where:

b
2
ca

0

_
c
a
K
1
c
(3.8)
and
a = I +I
add
c = K
= GMtg (3.9)
The non-dimensional damping coecient values has been calculated in [10].
Table 3.1.2 shows zeta values as function of the ship speed. As expected, the
value of the damping coecient increases with the ship speed.
25
Knowing , the value of b can be obtained from the rst term in Eq. (3.8):
b = 2
ca = 2
_
(I +I
add
)GMtg (3.10)
The RAO relates the wave amplitude
0
of a regular or sinusoidal wave to the
amplitude of the motion component at each frequency. As written in Section
2.3.3, motion components like pitch and roll are more sensitive to the wave
slope than to the wave amplitude. Therefore, the RAO for angular motions is
usually dened in terms of the wave slope, see also Eq. (3.1):
|R
(
e
, , U)|

0
k
0
(3.11)
The motion RAO is obtained as a combination of the wave excitation function
W(
e
) that relates the wave amplitude to the roll moment and the transfer
function given in Eq. (3.7) that relates the roll moment to the roll motion.
This is the reason for using linear models. Therefore an expression for the
Laplace transform of the wave-excitation moment is needed.
The function W(
e
) is usually computed using hydrodynamic and wave
diraction theory. In this work, a simple approximation based on the so-called
Froude-Krilo theory has been adopted. The simplifying assumption of the
Froude-Krilo theory is that the wave system is not disturbed by the pres-
ence of the hull; and therefore, the exiting moment is calculated by integrating
the pressure over the hull instead of solving a hydrodynamic (boundary value)
problem.
At zero speed and in beam seas, such approximation leads to the following
expression for the wave excitation moment [12]:
F() = W()
0
(I +I
add
)
2
0
e
2
T
2g
k
0
(3.12)
where T is the mean draft of the ship (value given in ??).From Eq. 3.8
c =
1
K
=
1
(I+I
add
)
2
0
, therefore Eq. (3.12) becomes:
F()
e
2
T
2g
K
k
0
(3.13)
The roll motion is usually a narrow band process and its power spectrum is
only signicant around the natural frequency of the ship. Therefore, a further
Speed [kts] 0 5 10 15 20 25
0.106 0.112 0.117 0.122 0.127 0.132
Table 3.1: Non-dimensional damping coecient as function of the ship speed
for the naval vessel in ??
26
simplication is obtained by evaluating the exponential term in Eq. (3.13) at
the roll natural frequency . Furthermore, in almost all cases the argument of
the exponential evaluated at the natural frequency is much less that 1, thus
the exponential term is close to 1. By simply neglecting the exponential term
in Eq. (3.13) no signicant errors are committed:
F()
1
K
k
0
= F
()k
0
(3.14)
In this case, the approximation of the response operator as a function of the
wave slope is obtained combining Eqs. (3.7) and (3.14) as indicated in the
following:
(, /2, 0)

0
k
0
= |H()F
()| =
H()
K
=

2
0
_
(
2
0

2
)
2
+ (2
0
)
2
arg
_
R
(, /2, 0)
_
(H()F
()) =
_
H()
K
_
= (

2
0
(
2
0

2
) + i(2
0
)
)
(3.15)
Due to the simplifying assumptions the model given in Eq. (3.15) tends to
over-estimate the roll motion since the roll response to the wave excitation will
be less than the predicted by Eq. (3.15).
The RAO is not only function of the frequency but also of the forward speed
and angle of heading with respect to the waves .
The forward speed of the ship U produces a twofold eect on the RAO.
The rst eect is the increase of the damping as we have seen in the previous
sections. The second eect is taken into account in the transformation from the
wave frequency to the encounter frequency experienced on the ship according
to Eq. (2.19). It is convenient to express the RAO rst in the wave frequency
domain and then transform it to the encounter frequency domain. Thus, the
RAO for roll expressed in the wave frequency domain at any speed and heading
can be approximated as follows:
R
(, , U)
1
K
H
_

2
g
U cos
_
sin (3.16)
in which H already incorporates the modied values of according to the
forward speed, an example is given in Fig.3.1.2.
The transformation to the encounter frequency domain is then done pair-wise
according to the transformation given in Eq. (2.19).
Finally the expression given in Eq. (3.16) can be inserted in (3.1) to obtain
the roll motion spectrum.
An example of roll motion spectra is given by Fig. 3.3 for dierent average
wave period. It can be noticed that the peaks frequencies do not change due
to the wave period, but the peak amplitude decreases with the increase of the
27
0 0.5 1 1.5 2 2.5 3 3.5
0
2
4
6
x 10
4
RAO amplitude
[rad]
|
R
|
2

[
d
e
g
/
m
]
2

0 0.5 1 1.5 2 2.5 3 3.5
200
100
0
100
200
RAO phase
[rad]
a
r
g
(
R
)

[
d
e
g
]

=45 deg
=90 deg
=135 deg
=45 deg
=90 deg
=135 deg
Figure 3.2: Motion RAO amplitude and phase for dierent encounter angles for
the example naval vessel sailing at 15kts
wave period. As expected the spectrum shift due to the encounter frequency
correction: the lower the encounter angle, the lower the spectrum peak fre-
quency. That can be a problem, because the damping of the roll motion is
harder at low frequency as it will be shown in Chapter ??.
Time-series of Roll Motion using Seakeeping Models
To simulate the motion of the ship given the motion spectrum time-series can
be, as proposed by Perez and Blanke in [3]. The procedure is similar to the
one used for the simulation of the sea elevation given in Section ??.
The roll motion can be seen as a nite sum of sinusoidal components with a
random initial phase:
(t)
N
i=1
i
cos(
e
i
t +
i
) (3.17)
Given a wave spectrum S
() and the motion RAO the amplitudes of the

28
(a) T
1
= 7.5s
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
0
0.5
1
1.5
2
2.5
3
x 10
3 Roll Motion Spectra
e
[rad]
S

[
d
e
g
/
m
]

=45 deg
=90 deg
=135 deg
(b) T
1
= 10s
Figure 3.3: Roll motion spectra

i
calculated using the motion RAO and MPM
wave slope spectrum with h
1/3
= 1m for the example vessel at 15kts
for dierent encounter angles and average wave period T
1
regular components

i
are given, similarly to Eq. (2.12), by:
i
=
_
2
_

e
i
+
e
i
2
e
i
e
i
2
S
(
e
)d
e
=
_
2
_

i
+
i
2
i
2
|R
(, , U)|
2
S
()d
=
_
2|R
i
, , U)|
2
S
i
)
i

i
2
,
i
+

i
2
_
(3.18)
Using this method the components are calculated in the wave frequency do-
main instead of the observed frequency domain. This procedure avoids problem
that arises when there are singularities, as already described in Section 2.3.3
A realization of the roll motion due to waves can be obtained combinining
Eqs. (3.17) and (3.18):
(t)
N
i=1
_
2|R
i
, , U)|
2
S
i
)
i
cos(
e
i
t +arg(R
i
, , U)) +
i
)

i
2
,
i
+

i
2
_
(3.19)
In Fig. 3.4 it possible to see a realization of the wave induced roll motion in
slight sea conditions with the vessel travelling at 15kts, It is not realistic for a
vessel to reach this speed with worse sea conditions.
29
1 2 3
0
1
2
3
x 10
3 =45 deg
e
[rad]
i

[
d
e
g
]
1 2 3
0
0.005
0.01
=90 deg
e
[rad]
5 10
0
2
4
6
x 10
3 =135 deg
e
[rad]
0 100 200
5
0
5
=45 deg
t [s]
(
t
)

[
d
e
g
]

RMS=1.4475
0 100 200
10
5
0
5
10
=90 deg
t [s]

RMS=2.7346
0 100 200
5
0
5
=135 deg
t [s]

RMS=1.1721
Figure 3.4: Amplitude of regular roll component

i
calculated using MPM wave
slope spectrum with h
1/3
= 1m and T
1
= 7.5s and particular real-
izations of roll motion for the example vessel at 15kts for dierent
encounter angles
Remarks on the implementation The technique described in this section
has been implemented to generate time series using a toolbox provided by
Perez. The great disadvantage is that the parameters for the generation of the
time series are generated oine. This means that the wave encounter angle
is xed to a constant value. This is a non realistic assumption because
the rudder always produces yaw motion, thus the heading of the ship changes
during the sea trial, while sea waves have a xed direction, thus the encounter
angle may change. Therefore it is not possible to dynamically correct the
encounter frequency, if the heading changes.
If the Motion RAO data are tabulated as a function of the wave frequency
, speed U and wave encounter angle , it possible to use another method
proposed by Perez and Blanke in [3] and in [1] to interpolate the data, using
the MSS toolbox [13] and dynamically correct the encounter frequency and
angle.
3.1.3 Frequency Dependent Linear Model for Rudder Roll
Damping
In this section, the linear plant model including the kinematics and dynamics
will be discussed
30
From (4.2) and (4.1) the roll motion dynamics have been studied considering
a 3-DOF model expressed in a b body-xed frame, thus taking into accounts
couplings in sway-roll-yaw and a constant surge velocity, as in [25]. The model
in (??) can be then rewritten as a state-space system as in (??), where the
state is given by x = [v, p, r, , ] denoting sway velocity, roll rate, yaw rate,
roll and yaw angle respectively, M is a matrix comprising mass, inertia and
added mass and added inertia, f(x) express hydrodynamics and hydrostatic
forces. I is a unit matrix. Details are found in [25].
The
c
forces are linearly dependent on the position and the angle of attack
of the rudder, which is controlled by the hydraulic machinery, using a modied
model of [35], which presents two saturations on the rudder angle and on the
steering rate. In the linearity range the machinery is represented by a rst order
linear system with a delay time t
r
, which can be neglected if 5 times larger
the dominant poles of the system. For RRD system, (??) can be linearised,
assuming a constant forward velocity U, obtaining:
G
=
K
roll
(q
1
s)(q
2
+s)
(p
1
+s)(p
2
+s)(s
2
+ 2
s +
2
)
(3.20)
where q
1
R
+
is a RHP zero.
3.1.4 Fundamental Limitations in Rudder Roll Stabilisers
Under linear superposition assumption, the performances of a RRD can be
measured by means of the output sensitivity function, the ship response to
wave-induced disturbances in roll is dened as:
S(s)
cl
(s)/
ol
(s) = (1 +C(s) G(s))
1
(3.21)
where
cl
and
ol
are respectively the closed-loop and open-loop roll angle.
In particular
ol
is the ship response to wave disturbances. The desired per-
formance for the RRD is to reduce S(j) in the frequency region where the
wave-induced motion has the most energy. The rudder to roll transfer func-
tion presented in Eq. (3.20) presents a q
1
zero in the Right-Half Plane. Thus
there is a Non-Minimum Phase (NMP) dynamics in ship response to rudder
command, which means the systems has an inverse response, [1].
Following [27] and [2] the limitations imposed by the presence of a RHP zero
in (3.20) can be studied using the Poisson integral:
_
+
log |S(j)|
q
1
q
2
1
+
2
d = log
N
p
i=1
p
i
+q
1
p
i
q
1
(3.22)
31
where q
1
is the RHP zero, N
p
is the number of unstable poles and p
i
is the i
th
unstable pole. The term w(q
1
, ) =
q
1
q
2
1
+
2
is called the Poisson kernel, which
represents a weight relative to the position of the RHP zero. Assuming a lin-
ear controller without unstable pole-zero cancellation, the open-loop system is
stable, therefore (3.22) is equal to zero. The integral expresses a weighted area
balance of the integrand function. This means that if the output disturbance
is to be attenuated in a range of frequencies (
1
2
), then there must be an
amplication for frequencies outside that range. This trade-o is concentrated
in a limited bandwidth depending on the Poisson kernel and the RHP zero. In
order to nd an estimate of M
2
S(j)
, which represents the maximum

amplication outside the damping bandwidth, the right hand side of (3.22) can
be overestimated. Therefore a general approximation ([27]) of the area shaped
from the sensitivity function can be considered. The approximation is com-
posed by three rectangles reported in Fig. 3.5. The controller is to be designed
to achieve the following performances:
|S(j)| M
1
< 1 (
1
2
)
|S(j)| = 1 (0,
l
) (
h
, +)
(3.23)
The second constraint is derived from the trade-o linking the output and
complementary sensitivity S(s) + T(s) = 1. A low value of S(j) at high
frequency generates a high value of T(j) at high frequencies with a conse-
quent measurement noise amplication. From (3.23) follows that the terms
_
l
0
log |S(j)| w(q
1
, ) d and
_
+
h
log |S(j)| w(q
1
, ) d are null, because
calculated over a zero-amplication region. Therefore we have: The integral of
the Poisson Kernel is dened as:
q
1
(
a
,
b
)
_

b
a
w(q
1
, )d =
_
tan
1

b
q
1
tan
1

a
q
1
_
(3.24)
Therefore considering the approximation of the sensitivity function in Fig. 3.5
, and the integral of the Poisson Kernel it is possible to overestimate the right
hand side of Eq. (??) as:
q
1
(
l
,
1
) log M
2
+
q
1
(
1
,
2
) log M
1
+
q
2
(
l
,
h
) log M
2
0 (3.25)
Obtaining:
c
1
log M
1
+c
2
log M
2
0 . (3.26)
Following [2],
l
= and
h
= + led to c
2
= c
1
and c
1
=
q
1
(
1
,
2
)
M
2
M
q
1
(
1
,
2
)
q
1
(
1
,
2
)
1
(3.27)
32
The result of [2], shown in (3.27), gives a lower bound to the amplication peak
S(j)
outside the damping region, but any information on the location

of S(j)
cannot be provided. An improvement to the S(j)
can be
obtained assuming a control feedback designed to achieve a maximum damping
M
3
M
1
< 1 at a frequency
s
. The specications in (3.23) hold. The
damping region is centred to
s
. That means the area overestimation embraces
the dashed triangular in Fig. 3.5. The result is given in (3.28):
_
+
log |S(j)|w(q
1
, ) d
_

1
l
log M
2
w(q
1
, ) d
+
_

s
1
_
log M
1
log M
3
1

s
+

1
log M
3

s
log M
1
1

s
_
w(q
1
, ) d+
+
_

2
s
_
log M
3
log M
1
s

2
+

s
log M
1

2
log M
3
s

2
_
w(q
1
, ) d+
+
_

h
2
(log M
2
) w(q
1
, ) d
(3.28)
The solution of inequality (3.28) is given by:
0 (
q
1
(
l
,
1
) +
q
1
(
2
,
h
)) log M
2
+
+q
1
_
log
_
2
s
+q
2
1
_
log
_
2
1
+q
2
1
__
log M
1
log M
3
1

s
+
+q
1
_
log
_
2
2
+q
2
1
_
log
_
2
s
+q
2
1
__
log M
3
log M
1
s

2
+
+

1
log M
3

s
log M
1
1

s
q
1
(
s
,
1
)+
+

s
log M
1

2
log M
3
s

2
q
1
(
2
,
s
).
(3.29)
One groups the terms dependent on the minimum damping (M
1
) and on the
maximum damping (M
3
) in (3.29). Then, choosing
l
= and
h
= +,
the lower bound M
2
on the output sensitivity positive amplication is given
by:
M
2
M
d
1
(q
1
,
s
,
1
,
2
)
q
1
(
1
,
2
)
1
M
d
2
(q
1
,
s
,
1
,
2
)
q
1
(
1
,
2
)
3
(3.30)
Equations (3.27) and (3.30) can be used to evaluate whether a controller
can be designed to full specied constraints on the location of the maximum
damping frequency
s
, the attenuation bandwidth window [
1
,
2
], the mini-
mum damping required in the damping window and the value of the maximum
damping. The comparison between the overestimation of M
2
= S
in (3.27)
and (3.30) is shown in Fig. 3.6. As function of the lower limit of the attenu-
ation bandwidth
1
, with
s
equal to the natural roll roll frequency
n
. The
location of the RHP zero is at = 0.1794[rad/s]. The over estimation of the
approach proposed in this thesis gives higher values than the approach of [2].
33
Figure 3.5: Comparison of the Poisson Integral using the approach in [2] (black
dashed line) and the one proposed in this work, which add the area
under the black solid line triangle centred in
s
The gap between the two values grows with the spreading of the damping re-
gion, meaning the importance of the information on the maximum damping
to overestimate the Poisson Integral. The contribution of this study gives the
designer a tool to establish whether the damping region can be extended to
lower limits. The results presented in Fig. 3.6 shows that any linear controller
cannot assure a damping region extended to low frequencies, close to the RHP
zero. As an example with a lower frequency bound of
1
= 0.5rad/s, a required
damping M
1
= 0.5 for [
1
2
[ and M
3
= 0.3 at the natural roll frequency, a
disturbance amplication of at least 110% is expected, if waves induced motion
has a signicant energy for (
1

2
).
3.1.5 Controller Design and Optimisation
Results of Section 3.1.4 clearly show how the closed loop performances are
aected from trade-o, when the disturbance is not known exactly in terms of
energy content. Therefore specications for a RRD should require sensitivity
reduction over a range of frequency possibly where the disturbance is likely
to be. At the same time it is required to avoid excessive amplications, in
case of changes in the sea conditions or vessel speed or heading, the energy
content can be concentrated outside the frequency range of reduction. This is
likely to happen when sailing in quartering seas (0
< < 90
), because the
transformation from wave to encounter frequencies can map signicant energy
34
Figure 3.6: Comparison between (3.27) (red dashed line) and (3.30) (blue solid
line) for the lower bound estimation of S
+
from Poisson in-
tegral. S
is function of the lower bound of the attenuation

bandwidth centred in
s
=
n
. M
1
= 0.5 and M
3
= 0.3
35
below
1
, dened in (3.23).
3.1.6 Control Design
For RRD, the control problem to be solved can be formulated as :
O : the objective is to decrease the roll motion induced on the ship hull by the
waves using the rudder as actuator.
SP : the set of specications for the closed-loop system to be satised
CL : the control law belongs to the class of closed loop control systems. It is
a function of the reference roll angle and the measured roll angle.
As reported in (3.21) the actual achievable damping is determined by the
magnitude of the sensitivity function S(s), which depends on the RHP zero
and the ship dynamics. A way to set specications on the closed loop system
is to shape a desired sensitivity function, that represents an ideal objective
dened by means of SP. The latter, similarly to (3.23), are:
1)|S
des
(j)| < M
1
< 1 [
1
,
2
]
2)|S
des
(j)| = 0 [
1
,
2
]
3) min |S
des
(j)| = M
3
< M
1
< 1 =
s
(3.31)
Using a xed controller for RRD, the attenuation bandwidth has been chosen
according to the sea and travelling conditions (e.g. sailing speed).The most
of the energy content of the roll motion is located around the roll natural
frequency, therefore an usual choice for
s
is the natural roll frequency. Thus
this condition is not always veried, therefore specications are given for a large
damping bandwidth. From (3.31), a possible form for the desired sensitivity
function is:
S
des
=
s
2
2
s
+
2
s
s
s + 1
(1 +

s
s
s
)(1 +
s
s
)
(3.32)
The parameters
s
,
s
and
s
must be chosen. If
s
=
, the attenuation
bandwidth has the negative peak in
and the damping coecient

s
can be
computed imposing the second constraint in (3.31). Considering a decoupled
feedback loop the Internal Model Principle can be used to obtain the desired
specications in (3.31), as in [31]. Using the rudder as command input the
controller C
u
, required to achieve the desired sensitivity function in (3.32),
is:
C
u
=
_
S
1
des
1
_
G
1
(3.33)
Unfortunately there are unavoidable consequences due to the presence of a
RHP zero in (3.20). Indeed, it is not possible to achieve the perfect control if
36
the internal stability of the system is required, because the process inversion
makes the controller C
u
unstable. A stable controller can be obtained from
(3.33) replacing the RHP zero in G
with its Left-Half Plane mirror, using an

all-pass term, in order to maintain the DC gain of the controller in (3.33). An
high frequency pole
h
has been added to obtain a proper transfer function:
C
(s) =
(s)
(s)
=
S
1
des
1
G
s q
1
s +q
1
1
1 +
h
s
(3.34)
Equation (3.34) is not realisable because of the zero on the imaginary axis.
Considering the roll rate as control the controller becomes strictly proper and
realisable:
C
(s) =
(s)
(s)
=
(s)
s(s)
=
C
(s)
s
(3.35)
3.1.7 Control Optimisation
The controller form in (3.35), similar to the one introduced by [34], is useful
because changing
s
,
s
and
s
in (3.32) allows the controller in (3.35) to adapt
to dierent sea conditions. Moreover a new constraint for limiting the ampli-
cation peaks should be dened and parameters chosen accordingly. Therefore
an optimisation problem should be set up, which choose controller parameters
minimising a cost function. Cost function must penalise all the controllers
leading to a sensitivity function further from the ideal characteristics dened
in (3.32), hence an expression for the index to be minimised is given by:
J = S
des
(j) S(j)
2
(3.36)
where the sensitivity function is calculated applying the controller expression
to G
dened in Eq. (3.20). Being the specications and the cost function
dened relative to the response in roll angle, (3.34) has been considered for
optimisation and then (3.35) has been used to obtain a roll rate feedback con-
troller. Then specications imposed by (3.31) are included in the optimisation
problem along with a constraint on the maximum amplication peak outside
the damping region:
c
1
: min |S(j)| = M
3
< M
1
< 1 =
s
c
2
: arg min |S(j)| [
s

s
+ ]
c
3
:|S(j)| < M
1
< 1 [
1
,
2
]
c
4
:S(j)
M
2
sup
[
1
,
2
]
(3.37)
Referring to (3.34) and (3.35), the controller expression has been changed,
considering:
37
1. Change in the Controller Gain
2. Change of the Controller Dominant Pole
3. Addition of a Lead Net
The controller gain is given by:
K
g
=
_
s
+
1
s
2
s
s
_
/(K
roll
) (3.38)
Therefore lowering K
g
contributes to lower both positive and negative peaks
of the sensitivity function S(s). The controller dominant pole is q = q
1
in
(3.34), namely the Left-Half Plane mirror of the RHP zero. Bringing q close
to imaginary axis. contributes to reduce both positive and negative peaks of
the sensitivity function. High values of the couple (K
g
, q) can lead to unstable
control systems, because the RHP zero dynamics at the denominator of S(s)
becomes dominant. Still the cost function of the optimisation process tends
to avoid this case, because of the much higher values of the cost function.
The addition of a lead net to the controller is justied because an increase of
the magnitude in the open-loop control system, brings directly a reduction of
S(s). The RRD problem combining the three actions together can be solved.
Dening the vector of controller parameters x, the controller and the sensitivity
function expressions are:
C
x
(s) = C
(s) x
1
1 +
s
q
1
1 +
s
x
2
1 +x
3
s
1 +
x
3
x
4
s
C
(s) = C
x
(s)/s
S
x
(s) =
1
1 +C
x
(s)G
(s)
(3.39)
then, from (3.36) and (3.39), the non-linear programming problem statement
is:
min
x
J = min
x
(S
des
(j) S
x
(j, x))
2
subject to constraints in (3.37)
(3.40)
The solution of (3.40) is called the Optimised-RRD (OPT-RRD) and that
is the optimal solution in terms of minimising the cost function according to
the constraints in (3.37). Figure 3.7 shows the optimisation process of the
sensitivity function S(s). The minima gives the results for the optimised RRD.
The optimisation algorithm uses a Sequential Quadratic Programming (SQP)
38
Figure 3.7: Bode Diagram of the sensitivity function during the optimisation
process. The minimum of the optimisation is given for the red
dashed line. The desired sensitivity is shown as a black dashed
line. The circle line shows the waterbed eects when the controller
(3.40) is tuned up with an attenuation bandwidth centred around
0.7 rad/s
method. In this method, a Quadratic Programming (QP) sub-problem was
solved at each iteration. The optimisation for the non linear programming
problem guarantees that the result is a local minimum and the solution depends
on the basin of attraction that contains the initial point. An investigation on
the existence of a global minima for (3.40) has been tested against a large set
of initial values in the parameter space. In order to avoid saturations on the
hydraulic machinery, a modied cost function was chosen to weight the control
eort in the optimisation problem:
min
x
J = min
x
(S
des
(j) S
x
(j, x))
2
+
+ (1 ) (C
x
(j) S
x
(j, x))
2
(3.41)
where [0 1] is a parameter to be chosen. The controller resulting from the
optimisation problem is called the Optimised-RRD with Control Eort Weight
(OPT-RRD-CEW).
3.1.8 Results
The controllers designed in (3.40) and (3.41) have been applied to the non-linear
3DOF ships model used as benchmark in [36], which data are reported in ??.
39
Controllers performances can be derived for dierent encounter angles and for-
ward speed considering the Reduction of Roll, RR()=100(1 rms(
cl
)/rms(
ol
)),
which quanties the controller damping performances for a given wave sce-
nario. As a comparison the performances of a MV-CLOC derived in [1] were
also computed. The performances of the controllers are summarised in Table
3.3 for the following conditions: M
3
= 0.3 at
s
= 1.1135 [rad/s], M
1
= 0.5 for
[0.75
s
1.25
s
]. The vessel is sailing with U = 15 [kts] in slight waves (sea
state 3). A wave disturbance for a sea state 3 coming from dierent encounter
angles has been considered.
Table 3.2: Control system performances
RR() [rms]
e
= 45

e
= 90

e
= 135
MV-CLOC 18.50% 20.79% 22.96%

Opt-RRD 20.4% 51.4% 57.23%
Opt-RRD-CEW 40.34% 56.73% 60.98%
Results show a signicant roll reduction for the OPT-RRD and OPT-RRD-
CEW with respect to the MV-CLOC. The introduction of the weighting term
substantially modies the closed loop performances at quartering seas, see Fig.
3.8. The control scheme is valid for dierent ship loads and speed, but the
controller, based on ship motion model linearisation should be recomputed.
3.1.9 Conclusions RRD
This thesis outlined the main issues related to the linear design of a rudder-roll
damping controller, presenting a novel approach for quantifying the limita-
tions imposed by the non minimum phase dynamics. An rudder-roll damping
controller was derived shaping the output sensitivity function. The controller
expression was included in a constraint optimisation problem that was used
to limit the disturbance amplication outside the attenuation bandwidth. The
optimisation problem also allows to weight the control eort to prevent satu-
ration on the rudder machinery. The closed loop performances were evaluated
and simulation results showed a signicant enhancement in roll reduction ca-
pability when compared with a Minimum Variance Cheap Limiting Optimal
Controller approach.
3.2 Fin/Rudder Roll Damping
The use of rudder for roll and steering control has two critical aspects: the
nonlinear coupling between sway-roll-yaw [[25]] and the instable zero dynamics
in the rudder-roll system [[28]]. Therefore order to counteract the roll motion
40
(a) Opt-RRD
(b) Opt-RRD-CEW
Figure 3.8: Roll Angle, Steering Angle and Rate simulations in Sea State 3.
Encounter angle = 45
. M
3
= 0.3 at
s
= 1.1135 [rad/s], M
1
=
0.5 [0.75
s
1.25
s
]
41
and overtake the limitation of rudder roll stabilizing control the joint action of
active ns and rudder are considered very attractive [[37], [28]]. In particular,
active ns have shown the best performances in roll reduction at high speed
but they result ineective at low speed. On the other hand, the rudder roll
damping (RRD) provides good performances and is more eective than ns roll
damping (FRD) at low speed due to the fact that the rudder is located in the
race of the propeller and thus operates in higher speed ows than ns. However
Several works which address the roll motion reduction by exploiting both ns
and rudder have been proposed in the literature [[38], [39]]. However, these
works are based on linear models to describe the ship dynamics, and therefore
signicant nonlinear eects such as the instability of the zero dynamics are
neglected [[40]]. Furthermore, the proposed solutions are based on independent
FRD and RRD controllers, which do not consider their mutual eect on the
roll angle, thus compromising the overall performances.
In this section, an integrated control system for heading and roll damping
of a multipurpose naval vessel by means of active ns and rudder is proposed.
The ship dynamics is described through a nonlinear multivariable model which
takes into account the hydrodynamic couplings among sway, roll and yaw, due
to both the wave eect and the control devices. A stability analysis of the zero
dynamics of the adopted model along with the design of an integrated MIMO
variable structure control (VSC) law is proposed.
The rest of the thesis is organized as follows. In Section 3.2.1 the nonlin-
ear ship model for the considered multipurpose naval vessel is provided. In
Section 3.2.4 the properties of the zero dynamics for the considered nonlin-
ear ship model are investigated. In Section 3.2.7 a Variable Structure Control
(VSC) for simultaneous course keeping and roll damping is proposed. Finally,
in Section ?? conclusions are drawn and future work is discussed.
3.2.1 Nonlinear Ship Model for Fin/Rudder Roll Damping
The control design deals with a square nonlinear ship model ane in the inputs
of the form:
x = f(x) +g (u +w
H
)
y = h(x)
(3.42)
where x R
5
is the state vector, y R
2
is the output and u R
2
is the
input, while w
H
represents the external disturbances.Its derivation from the
ship manoeuvring model combined with the rst order wave disturbance will
be shown in this section.
42
Figure 3.9: Reference frames and notation for ship motion description
Rigid Body Dynamics
The ship dynamic nonlinear model can be obtained considering the vehicle as a
rigid body and applying the Newton Mechanics to describe its 6DOF motions
[[14]]. The equations of motion in matrix form are :
M
RB
+C
RB
() =
RB
(3.43)
where = [u, v, w, p, q, r]
T
is the generalized velocity vector in body-xed
frame (b-frame) according to the SNAME (1950) notation, M
RB
is the gen-
eralized mass-inertia matrix, C
RB
() is the Coriolis-centripetal matrix due to
the handing over from an inertial reference system to a moving one. Finally
,
RB
= [
1

2

3

4

5

6
]
T
is the total vector of forces and moments acting on
the ship.
The 6DOF ship model can be reduced to a 3DOF model by means of some
simplifying assumptions. First, being the considered vessel characterized by a
slender hull with port-starboard symmetry, it is possible to decouple the sway-
roll-yaw dynamics from the surge-heave-pitch one [[41]]. As a consequence, the
surge velocity u can be considered constant and equal to the cruise forward
speed U and a 3DOF model in the variables v, p and r is obtained.
Moreover, let us consider the b-frame such that the reference axis coincide with
the ship principal axes of inertia. This implies the inertia matrix to be diagonal.
In addition, by assuming a port-starboard symmetry of the hull, the ship center
of gravity CG = [x
g
y
g
z
g
] can be considered located on the xz-plane of the b-
frame, so that y
b
y
g
and thus y
g
= 0.
As a result, the reduced 3DOF ship model (

M
RB
,

C
RB
,
RB
) obtained from the
6DOF model (M
RB
, C
RB
,
RB
,) (eq. 3.43) can be described by:
_
_
m mz
g
mx
g
mz
g
I
x
0
mx
g
0 I
z
_
_
_
_
v
p
r
_
_
+
_
_
0 0 m|U|
0 0 mz
g
|U|
0 0 mx
g
|U|
_
_
_
_
v
p
r
_
_
=
_
_
6
_
_
(3.44)
43
where I
x
and I
z
are the moments of inertia with respect to the origin of the
b-frame, and with the obvious denition of the matrices

M
RB
,

C
RB
and
RB
.
The total vector of forces and moments
RB
is generated by dierent phenomena
and it can be expressed by the superposition of terms arising from these:
RB
=
hyd
+
prop
+
cs
+
env
(3.45)
where
hyd
describes the hydrodynamic forces and moments;
prop
is the propul-
sion forces and moments;
cs
represents forces and moments due to the control
surfaces (active ns and rudder);
env
models forces and moments due to waves,
wind and currents.
Note that, the propulsion component
prop
will be neglected in the rest of the
thesis, since the propeller eect can be assumed completely compensated by
the ship calm water resistance, under the assumption of constant ship cruise
speed.
Moreover, the environmental term
env
will not appear in the following being
modeled as a disturbance aecting the control input channels (Sec. 3.2.7).
Hydrodynamic Forces and Moments
The hydrodynamic forces and moments in 3DOF are modeled as nonlinear
functions of the accelerations v, p, r, the velocities v, p, r and the roll angle
as shown in the following equations [[25], [1]]:
2hyd
= Y
v
v +Y
r
r +Y
p
p+
+ Y
|u|v
|u|v +Y
ur
ur +Y
v|v|
v|v| +Y
v|r|
v|r|+
+ Y
r|v|
r|v| +Y
|uv|
|uv| +Y
|ur|
|ur| +Y
uu
u
2
(3.46)
4hyd
= K
v
v +K
p
p+
+ K
|u|v
|u|v +K
ur
ur +K
v|v|
v|v| +K
v|r|
v|r|+
+ K
r|v|
r|v| +K
|uv|
|uv| + K
|ur|
|ur| +K
uu
u
2
+
+ K
|u|p
u|p| +K
p|p|
p|p| +K
3
gGZ()
(3.47)
6hyd
= N
v
v +N
r
r+
+ N
|u|v
|u|v +N
|u|r
|u|r +N
r|r|
r|r| +N
r|v|
r|v|+
+ N
|uv|
|uv| +N
u|r|
u|r| +N
p
p +N
|p|p
|p|p+
+ N
|u|p
|u|p +N
u|u|
u|u|
(3.48)
The coecients proportional to the derivative v, p and r, are collected in the
3 3 Added Mass matrix (M
A
):
M
A
=
_
_
Y
v
Y
p
Y
r
K
v
K
p
0
N
v
0 N
r
_
_
, (3.49)
44
O
b
x
b
L
pp
LCG
WL
FP AP
CG
T
z
b
VCG
Figure 3.10: Main ship values: body-xed frame O
b
{x
b
, z
b
},Center of Gravity
(CG), Lateral Center of Gravity (LCG), Vertical Center of Gravity
(VCG), Waterline (WL), Draught (T), Aft Perpendicular (AP),
Front Perpendicular (FP), Length between Perpendicular (Lpp).
while the residual coecients representing the hydrodynamic nonlinear cou-
pled terms are collected in the 3DOF vector
hyd
= [
2hyd

4hyd

6hyd
]
T
. The
numerical values of the coecients are taken from [[1]]
3.2.2 Control Modules
The control module
cs
takes into account the eects on the ship motions due to
the control surfaces, i.e. rudder (
rud
) and active ns (
ns
). It can be detailed
as follows:
cs
=
rud
+
ns
. (3.50)
The eectiveness of rudder and ns on roll and yaw motion is dierent due to
their dierent geometrical location on the hull with respect to the CG.
Rudder
The 3DOF forces and moments generated by the rudder in the b-frame can be
approximated by [[1]]:

rud

_
_
L
rud
R
rud
L
rud
LCG L
rud
_
_
_
_
rud
R
rud

rud
LCG
rud
_
_

rud
(3.51)
where L
rud

rud

rud
is the lift force acting on the rudder, while R
rud
and
LCG are the distance between the rudder center of pressure CP and the ship
CG (Fig.3.10) along the z- and x-axis of the b-frame respectively. To calculate
the lift of the rudder, the eective angle of attack
erud
has been approximated
by the mechanical angle of the rudder
rud
[[1]].
45
Active Fins
The action of the active ns on the ship motions requires a more detailed
analysis compared to rudder due to the location of the foils. In this work the
following approximate model for (3DOF) forces and moments induced by the
ns will be considered (further details can be found in [[1]]):

ns

_
_
sin(
tilt
)
2 R
ns
FCG sin(
tilt
)
_
_
N
ns
(3.52)
where R
ns
is the ns roll arm (the distance of the n CP from the ship CG)
and FCG is the longitudinal distance between the n CP and the ship CG
while
tilt
represents the ns tilt angle. N
ns
is the resulting hydrodynamic
force component normal to the n, dened as a function of the lift L
ns
and
drag force D
ns
arising on them as L
ns
cos(
e
) +D
ns
sin(_e) with e the ns
eective angle of attack between the foil and the ow.
3.2.3 State Space Representation
The adopted state space model considers also the roll angle and the yaw
angle . Therefore, the following augmented state vector is considered:
x =
_
v p r

T
R
5
. (3.53)
where

=
d
with
d
the (constant) desired yaw angle. Thus, the
augmented state-space representation of the ship nonlinear model given in
eq. (3.44) is:
M x +C(x) x = T
hyd
(x) +T
cs
[
rud
N
ns
]
T
(3.54)
where the matrices M, C, T
cs
and the vector T
hyd
are dened as follows:
M =
_

M
RB
M
A
0
0 I
2
_
, (3.55)
C(x) =
_

C
RB
0
22
E(x) 0
22
_
E(x) =
_
0 1 0
0 0 cos
_
T
hyd
(x) =
_
_

hyd
0
0
_
_
, T
cs
=
_
_

rud

ns
0 0
0 0
_
_
with 0 opportune completion matrices, I
2
a 2 2 identity matrix,
hyd
dened
in sec. 3.2.1,
rud
and
ns
the coecients vector multiplying
rud
in eq. (3.51)
and N
ns
in eq. (3.52) respectively.
Thus, the ship dynamic model of eq. (3.54) can be written as in 3.42 that is
46
repeated here for sake of clarity:
x = f(x) +g (u +w
H
)
y = h(x)
(3.56)
where x R
5
is the state vector, u = [
rud
N
ns
]
T
R
2
, w
H
= [w
H1
w
H2
]
T
R
2
, h(x) = [

]
T
R
2
and:
f(x) = M
1
C(x) x +M
1
T
hyd
(x)
g =
_
g1 g2
= M
1
T
cs
.
(3.57)
3.2.4 Zero Dynamics Analysis
The zero dynamics describes the internal behavior of a system when input
and initial conditions are chosen to force the output to zero [[42]]. The analy-
sis of the zero dynamics has an important role in the control design. Indeed,
the presence of an unstable zero dynamics, i.e., non-minimum phase system,
prevents the application of important control techniques such as feedback lin-
earization. Furthermore, it signicantly aects the achievable performances,
for instance by imposing limitation on the maximum control gain.
3.2.5 Normal Form
In order to analyze the multivariable nonlinear ship model, the eq. (3.54) are
transformed to normal form. Let {r
1
, . . . , r
m
} be the vector of relative degree
at a point x
, i.e., the number of time the i-th output y

i
must be dierentiated
to have at least one component of the input vector u explicitly appearing [[42]].
Let us now briey introduce the Lie derivative, i.e., the derivative of () along
f(), as follows:
L
f
(x) =
n
i=1
x
i
f
i
. (3.58)
It can be shown that such a multivariable nonlinear ship model has relative
degree {r
1
, r
2
} = {2, 2} at x
0
= [0, 0, 0, 0, 0]. In fact, the Lie derivatives are:
L
gj
h
i
= 0
L
gj
L
f
h
i
(x) = 0
i {1, 2}, j {1, 2} (3.59)
and the matrix A(x) dened as:
A(x) =
_
L
g1
L
f
h
1
(x) L
g2
L
f
h
1
(x)
L
g1
L
f
h
2
(x) L
g2
L
f
h
2
(x)
_
=
_
a
11
a
12
a
21
cos(x
4
) a
22
cos(x
4
)
_
(3.60)
47
is nonsingular if the following holds:
a
11
a
22
= a
12
a
21
, (3.61)
cos(x
4
) = 0. (3.62)
It should be noticed that while the condition given by eq. (3.61) is related to the
vessel parameters, the second condition given by eq. (3.62) is always satised in
practice as cos(x
4
) = 0 implies that x
4
=

2
+k , k N, which is an inadmissible
operational condition, being x
4
the roll angle. Furthermore, the system has
degree {2, 2} for any state x
0
= [x
1
, x
2
, x
3
, x
4
, x
5
] with x
4
=

2
+k , k N.
Therefore, according to [[42]], the nonlinear system describing the vessel can
be transformed to normal form by applying the following local coordinates
transformation z(x) =
_
z
1
(x), z
2
(x), z
3
(x), z
4
(x), z
5
(x)
T
at x
0
.
In particular, since the system has relative degree {2, 2}, the following choice
is possible for the rst r
1
+r
2
= 4 variables:
1
=
_

1
1
1
2
_
=
_
z
1
(x)
z
2
(x)
_
=
_
h
1
(x)
L
f
h
1
(x)
_
(3.63)
2
=
_

2
1
2
2
_
=
_
z
3
(x)
z
4
(x)
_
=
_
h
2
(x)
L
f
h
2
(x)
_
(3.64)
while the remaining (n r
1
r
2
= 1) variable transformation can be chosen
arbitrarily:
= z
5
(x) (3.65)
so that the jacobian matrix describing the local coordinates transformation
z(x) =
_
z
1
(x), z
2
(x), z
3
(x), z
4
(x), z
5
(x)
T
, is nonsingular at x
0
.
Furthermore, since G = span{g} is an involutive set, z
5
(x) can be chosen so
that:
L
gi
z
5
(x) = 0 i {1, 2}. (3.66)
with g =
_
g1 g2
.
Note that, the constraint given by eq. (3.66) can be satised by assuming:
= z
5
(x) = v + p + r (3.67)
with:
_
, , , 0, 0
ker
_
g
T
_
. (3.68)
In particular, since g
T
has the following structure:
g
T
=
_
g1
1
g1
2
g1
3
0 0
g2
1
g2
2
g2
3
0 0
_
(3.69)
48
a base for the the null space is:
ker
_
g
T
_
=
_
_
_
_
0
0
0
0
1
_
_
,
_
_
0
0
0
1
0
_
_
,
_
_
g1
2
g2
3
g1
3
g2
2
g1
1
g2
2
g1
2
g2
1
g1
3
g2
1
g1
1
g2
3
g1
1
g2
2
g1
2
g2
1
1
0
0
_
_
_
_
(3.70)
Therefore, by choosing
=
g1
2
g2
3
g1
3
g2
2
g1
1
g2
2
g1
2
g2
1
(3.71)
=
g1
3
g2
1
g1
1
g2
3
g1
1
g2
2
g1
2
g2
1
(3.72)
= 1 (3.73)
the Lie Derivative turns out to be:
L
g1
z
5
(x) = g1
1
+ g1
2
+ g1
3
, (3.74)
L
g2
z
5
(x) = g2
1
+ g2
2
+ g2
3
. (3.75)
which can be re-written in a matrix form as follows:
_
g1
1
g1
2
g1
3
0 0
g2
1
g2
2
g2
3
0 0
_
0
0
_
_
=
_
0
0
_
(3.76)
being the vector [ 0 0]
T
an element of the null space base for g
T
.
As a result, the system equations (3.54) can be written in normal form as
follows:
1
1
=
1
2
1
2
= L
2
f
h
1
(x
1
(, )) +
2
j=1
L
gj
L
f
h
1
(x
1
(, )) (u
j
+w
Hj
)
2
1
=
2
2
2
2
= L
2
f
h
2
(x
1
(, )) +
2
j=1
L
gj
L
f
h
2
(x
1
(, )) (u
j
+w
Hj
)
= q(, )
(3.77)
3.2.6 Stability Analysis
Let us now investigate the properties of the zero dynamics = q(0, ), which
can be detailed as follows:
= a|| b (3.78)
49
with a, b R and a > 0, b > 0.
In order to investigate the stability of the zero dynamics, it should be pointed
out that eq. (3.78) has an unique equilibrium point:
0
= 0 (3.79)
At this point, the stability of the system can be simply checked by exploiting
the following quadratic Lyapunov candidate:
V () =
1
2
2
. (3.80)
In fact, the negative deniteness of the derivative of the Lyapunov candidate:
V () =
= (a|| b)
2
(3.81)
is guaranteed by the fact that a|| b < 0, .
3.2.7 Controller Design
The control problem is to drive the roll angle to zero and the yaw angle
to a specic value
d
or, equivalently, to get the variable

=
d
to zero.
The control inputs are the rudder angle
rud
and the normal force N
ns
arising
on the ns. The choice of N
ns
instead of the ns angle
ns
is due to the
constrain of ane input required by the variable structure control technique.
Note that, the ns angle can be always computed from N
ns
, e.g., through the
approximation N
ns
= L
ns
given in [[1]]. According to the ship dynamics given
in eq. (3.42) the input u = [
rud
N
ns
]
T
is assumed to be aected by a 1st-order
wave disturbance w
H
= [w
H1
w
H2
]
T
. Thus, the multivariable nonlinear ship
model of eq. (3.77) can be written in a compact form as follows:
0 100 200 300 400 500 600 700 800 900 1000
20
15
10
5
0
5
10
15
20
Roll angle [deg]
t [s]
[d
e
g
]
0 100 200 300 400 500 600 700 800 900 1000
5
0
5
10
15
Yaw angle [deg]
t [s]
[d
e
g
]
Figure 3.11: Roll and Yaw angles obtained during a 1000 seconds simulation
in sea state 5 - encounter angle 135 deg - encounter frequency
0.86rad/s conditions. The controller action starts at t = 300s.
600s.
50
1
1
=
1
2
1
2
= F
1
(, ) +
2
j=1
G
1j
(u
j
+w
Hj
)
2
1
=
2
2
(3.82)
2
2
= F
2
(, ) +
2
j=1
G
2j
(u
j
+w
Hj
)
= q(, )
where F R
21
with F
i
= L
2
f
h
i
and G R
22
with G
ij
= L
gj
L
f
h
i
. In addition let
us assume the disturbance w
H
is bounded by a known bound W = [W
1
W
2
]
i.e., |w
H1
| < W
1
, |w
H2
| < W
2
. Furthermore, let us consider F and G to
be modeling uncertainties of F and G respectively with known bounds, as
described in the following equations:
F = F

F
G = G

G
(3.83)
with

F and

G the nominal models.
Then, the following two sliding surfaces are dened:
s
1
=
1
2
+
1
1
1
= 0,
1
> 0
s
2
=
2
2
+
2
2
1
= 0,
2
> 0
(3.84)
and the Lyapunov candidate V (s) =
1
2
s
T
s with s = [s
1
s
2
]
T
is considered.
At this point in order to stabilize the system the following control law can be
considered:
_
u
1
u
2
_
=

G
1
F
1
(, )
1
1
2
k
1
(, ) sgn(s
1
)
F
2
(, )
2
2
2
k
2
(, ) sgn(s
2
)
_
(3.85)
where the invertibility of

G is a consequence of the ship dynamics model, and
k = [k
1
k
2
]
T
is a gain vector required to guarantee the negative deniteness of
the

V (s)
1
.
Sea state 3
e
= 45

e
= 90

e
= 135
RSR() 89.0501 95.2255 93.9063

RSR(

) 90.1423 93.8579 89.7812
Table 3.3: Control system performances for the vessel sailing with U = 7m/s
in slight waves for dierent encounter angles
1
As far as the implementantion of the control law is concerned, the function sgn(s) is re-
placed by the continous function tanh(s). It allows to reduce chattering without aecting
the stability analysis
51
Sea state 5
e
= 45

e
= 90

e
= 135
RSR() 86.8729 90.8779 93.3571

RSR(

) 88.8968 90.5029 91.6357
Table 3.4: Control system performances for the vessel sailing with U = 7m/s
in rough waves for dierent encounter angles
3.2.8 Case Study
In the following, a case study for the proposed control technique is described.
The parameters which can be found in [[1]] are exploited for the non-linear ship
model described by eq. (3.42).
The controller is designed at cruise speed 7 m/s. The rudder saturation is set
to 40 deg while the rudder speed is limited to 20 deg /s . The ns saturation,
expressed in terms of force, is computed by considering a ns maximum angle
of 28.8 deg (ns stall angle) and a ns maximum angle rate equal to 23 deg /s,
that is respectively maximum ns force 10
7
N and maximum ns force rate
1.5 10
5
N/s.
Furthermore, the input disturbances due to the waves are obtained ltering
a zero-mean Gaussian white noise process with unity power spectrum by a
linear 2nd-order wave response model [[14]]. The lter gain is properly scaled to
reproduce the roll and yaw motion similar to those obtained by using the motion
frequency response functions (motionRAO) of the vessel in combination with
the wave spectrum [[1]]. In particular, the ITTC (Modied Pierson-Moskowitz)
wave spectrum is considered for the following sailing conditions: sea state 3 with
H
s
= 0.3 m and T = 6 s and sea state 5 with H
s
= 2.5 m and T = 7.5 s, where
H
s
is the signicant wave height and T is the average wave period. In addition,
for each sea state, quartering, beam and bow seas are considered.
As far as the eectiveness of the integrated control system for roll damping is
concerned, the percentage Reduction Statistic of Roll (RSR) dened as follows:
RSR = 100
_
1
S
c
S
u
_
(3.86)
where the subscripts c and u stand for controlled and uncontrolled respectively,
and S is the Root Mean Square (RMS) of the roll signal.
Fig. 3.11 describes the roll and yaw angles for a simulation with the following
operative condition: sea state 5 with an encounter angle of 135 deg. In particu-
lar, the VSC is not activated the rst 300 s and is turned on after 300 s for the
rest of the simulation. The reference angle in yaw changes from 0 deg to 10 deg
at t = 600 s. It should be noticed that the roll motion is signicantly reduced
while the yaw angle is promptly regulated to the desired reference after the
52
0 100 200 300 400 500 600 700 800 900 1000
5
4
3
2
1
0
1
2
3
4
5
Rudder Effort [deg]
t [s]
[d
e
g
]
0 100 200 300 400 500 600 700 800 900 1000
6
4
2
0
2
4
6
x 10
4 Fins Effort [N]
t [s]
N
f [d
e
g
]
Figure 3.12: Rudder and ns eorts during the simulations in sea state 5 -
e
= 45 deg conditions. The controller action starts at t = 300 s.
600 s. The values for the rudder angle and the ns force N
f
do not ever achieve their saturation value , i.e. 40 deg and 10
7
N
respectively, and a good gap from those is kept.
VSC is activated.
Fig. 3.12 describes the rudder and ns eorts. In particular, it can be noticed
that the rudder angle is bounded within 5 deg and the ns force is bounded
within 6 10
4
N. Indeed, these results are interesting as they are always
signicantly below the saturation thresholds imposed by their corresponding
mechanical systems.
Table 3.3 and Table 3.4 give a synoptic overview of the results obtained for
the two sea states with respect to dierent encounter angles, namely 45 deg, 90 deg and135 deg.
In particular, it can be noticed that a substantial reduction of the roll damp-
ing is achieved for both slight and rough sea conditions. Again, it is worthy
to notice that the control inputs required to obtain these performances never
reach the saturation limits, both in amplitude and rate.
0 100 200 300 400 500 600 700 800 900 1000
0.06
0.04
0.02
0
0.02
0.04
0.06
Zero Dynamics
t [s]
Figure 3.13: Zero Dynamics behavior.

Finally, Fig. 3.13 describes the behavior of the zero dynamics. According to
the theoretical analysis, the zero dynamics = q(0, ) settles around the equi-
librium point
0
= 0 after the VSC is activated at time t = 300 s. Furthermore,
the peak at time t = 600 s is due to the switch of the yaw angle reference from
0 deg to 10 deg.
53
Chapter 4
Dynamic Positioning
4.1 Background
Up to now, most dynamic positioning systems have been used for positioning
drill ships in deep water, and other oshore operations, such as diving support
and anchor handling. Furthermore, DP systems have been applied increasingly
to shuttle tankers during ooading operation with a oating production storage
and ooading (see [14] and [43]). The rst DP systems were designed using
conventional PID controllers in cascade with low pass and/or notch lters to
suppress the wave induced motion components. From 1980, a new model-
based control concept, which is based on stochastic optimal control theory and
Kalman ltering techniques, was employed to address the DP problem by [45].
Later extensions and modications of the latter work have been proposed by
numerous authors, see [43] [46], [47] and [48] and references therein. In [49],
in [50] and in [51] the sliding mode control is used with a Passive Nonlinear
Observer for the DP problem.
Like other technological systems a DP system is subject to fault (e.g. loss
of position) which may be caused either by computer, electrical, hydraulic and
thruster failures. The International Marine Contractors Association dened
guidelines for fault-tolerant system design, see the [52] report. A requirement
for DP Class 2 vessel is that loss of position is not to occur in the event of a single
failure in any active component or system (generators, thrusters, switchboards,
remote controlled valves, etc.). Moreover thrusters cause 21% of the incidents
related to DP, according to the report in [53]. Thruster reliability for DP
has been addressed by [54] as a problem to be solved. Therefore it is very
important to recover in case of any thruster failure. Moreover the [52] report
requires redundancy of all active components, thrusters included, in order to
meet the single failure criteria given above. Redundancy, as an alternative
means of providing the same function, is provided by simply having at least
two of every thruster component, but more generally what is really required
is fault tolerance which is the ability of a system to continue in operation
following a failure.
55
Chapter 4 Dynamic Positioning
Single Point Thruster failures in DP applications is extensively described in
[55]. Systematic fault tolerant control was studied for the station keeping of a
marine vessel by [56] and a structure-graph approach for fault diagnosis and
control reconguration was validated by sea tests. Fault-tolerant approaches
to handle thrust failures have been recently proposed in [57] and [58]: classical
PID control are used and reconguration after faults is implemented by means
of the virtual actuator approach in [59]. A problem is that PID controllers could
not be enough robust to handle reconguration transients during non-steady
state maneuvering.
This thesis presents an innovative solution for the DP control system of a
vessel which is based on Discrete-Time Variable Structure Controller (DTVSC)
and Wave Filtering using a (Multi-rate) Extended Kalman Filter. The intro-
duction of DTVSC allows to take into account the issue of control law digital-
ization directly. Moreover it ensures robustness with respect to model uncer-
tainties and input disturbances acting on the actuators. An Extended Kalman
lter (EKF) is designed in order to estimate the disturbances induced by the
rst order wave forces on the thruster. This is done to minimize the thruster ef-
forts. The estimation is improved by means of a Multi-Rate Extended Kalman
Filter (MREKF) which allows to take into account dierences in working fre-
quency of the sensors.
The thesis is organized as follows. The kinematic and dynamic equations,
the thruster allocation and the wave model are presented in Section 4.2. The
lter techniques are discussed in Section 4.3. The fault diagnosis system is
presented in Section 4.5. The control system, in particular the DTVS controller,
is reported in Section 4.4. Simulation results are presented in Section 5. The
thesis ends with conclusions and comments.
4.2 Nonlinear Ship Model
The models for ship motion control system design use superposition of either
motion or force [60]. The motion superposition model is the most commonly
adopted for control system design. Motion can be conceptually decomposed as
superposition of three contribution:
slowly-varying disturbance motion produced by second-order waves
eects, current and wind
control-induced motion described by a manouvering model, clarifying
the the relationship between control action and its eects on the motion.
The dynamics is very slow for the class of oshore supply vessels, thus is
referred as Low Frequency (LF) dynamics.
56
Figure 4.1: Vessel Reference frames
wave-induced motion where the wave frequency oscillatory motion in-
duced by rst-order waves is described by a seakeeping model. This
dynamics is referred as Wave Frequency (WF) dynamics.
4.2.1 Manoueuvering Model
The motion of a surface vessel for the development of a dynamic positioning
system, is described by a model which is based on the common assumption
that only horizontal forces must be counteracted, as stated in [14]. In this
hypotheses the generalized velocity vector [u, v, r]
T
in the ship body-xed
frame {b} is considered, where u is the surge velocity, v is the sway velocity and
r the yaw rate, see Fig. 4.1. Ship position is referred to the local geographical
inertial North-east-down frame {n}, xed to the Earth and described by the
generalized position [n, e, ]
T
, where n and e are the ship position in the
{n} frame and is the ship orientation referred to the n frame.
Under the assumption of low-speed manoeuvring the quadratic terms of ve-
locity, Coriolis terms and non-linear damping terms are neglected. With the
57
recalled notation, the DP model is described by the following dynamics:
M
LF
+D
LF
=
c
+
env
, (4.1)
where
LF
is the low-frequency generalised velocity vector, M is the rigid body
generalised mass, which includes inertia and added mass, D is the linear damp-
ing component,
c
and
env
are the generalized control force and environmental
disturbances, respectively. The kinematic equation has the following form:

LF
= R()
LF
, (4.2)
where
LF
is the low-frequency ship position and R =
_
_
cos sin 0
sin cos 0
0 0 1
_
_
describes rotation from {b} frame to {n} frame.
4.2.2 Environmental Disturbances
Environmental disturbances include both slowly varying and high frequency
forces. The slowly varying disturbances include second order wave drift, ocean
currents and wind forces. These eects are modeled by a bias b, described by
the following dynamic equation:
b = w
b
, (4.3)
where w
b
N(0, Q
b
) is a white noise [14]. This environmental disturbance
force, referred to ship body frame {b} has the form [14]
env
= R()
T
b.
First order wave forces eects on ship motion are modeled in the Wave Fre-
quency (WF) Dynamics as a second order frequency domain transfer function
driven by the white noise w
w
N(0, Q
w
) [14], which state space form is:
_

2
_
..
=
_
0 1
2
0
2
w
0
_
. .
A
w
_
2
_
..
+
_
0
K
w
_
. .
E
w
w
w
WF
=
_
0 1
_
. .
C
w
_
2
_
(4.4)
where
WF
represents three linear ship wave response models in surge, sway
and yaw with state vector
6
, K
w
= 2
0
w
represents a constant gain,
58
Figure 4.2: Thrusters conguration for
the oshore supply vessel
Table 4.1: Thruster conguration
matrix
Thruster Torque Angle
Port Main Propeller u
1
0 deg
Starboard Main Propeller u
2
0 deg
Bow Tunnel Thruster u
3
90 deg
Aft Tunnel Thruster u
4
90 deg
Bow Azimuth Thruster u
5

5
Aft Azimuth Thruster u
6

6
w
characterized the wave intensity,
w
is a damping coecient and
0
is
the dominating wave frequency. These parameters are estimated using the
technique described in [11] when signicant changes in heading or at regular
intervals of 20 minutes, which is the time period for which the sea state can
be considered to be stationary and can be adjusted to simulate trials in high
or moderate seas. Assuming that the vessel is in position control mode the
measured motion
m
is recorded and detrended to provide an oine estimation
of the wave induced motion
WF
. Comparing the latter with the output of
(4.4), an estimation of the of the covariance Q
w
of the state measurement noise
is provided from the sample covariance of prediction errors
=
WF

WF
,
[14].
4.2.3 Thruster Allocation
Marine vessels with n DOF are characterized by n generalized control forces
c

n
are distributed among the r thrusters in terms of control inputs u
r
:
c
= T()u (4.5)
where u is the thrust force vector. The thruster conguration matrix T()
depends on the location and orientation of the thrusters. The considered 3DOF
oshore supply vessel has two main propellers, two tunnel thrusters and two
azimuth thrusters, as shown in Figure 4.7. Azimuth thruster headings are
maintained at a xed angle . Thruster conguration matrix

T follows:
T = T( ) =
_
_
1 1 0 0 cos
5
cos
6
0 0 1 1 sin
5
sin
6
l
1
l
2
l
3
l
4
l
5
sin
5
l
6
sin
6
_
_
(4.6)
As evident from (4.6) and Figure 4.7, the allocation of the 6 thrusters is
symmetrical with respect to the longitudinal axis of the vessel. Considering the
59
two main propellers, the two tunnel thrusters and the two azimuth thrusters
as three couple of symmetrical thrusters, the same torque demand is applied to
both the thruster in each couple of symmetrical thrusters, when there are not
faults and disturbances on the actuators. Therefore the following condition is
assumed:
u
i
= u
i+1
for i = 1, 3, 5 (4.7)
The inclusion of faults and input disturbances to the system follows:
c
=

T(u +d +f), (4.8)
where thruster failures are modeled as additive faults, because a loss of torque
for a thruster is modeled using an equal torque for the corresponding element
of the f torque vector. The same applies for input disturbances d, modeled
as slowly varying bias, described by the dynamics equation

d = w
d
, where
w
d
N(0, Q
d
) is a white noise source.
4.2.4 Plant model
Under the considered hypotheses above the considered dynamic positioning
(DP) model has the following form:
= A
w
+E
w
w
w
,

LF
= R()
LF
,
b = w
b
,
d = w
d
,
M
LF
= D
LF
+

T(u +d +f) R()
T
b +w
LF
,
m
=
LF
+
WF
+w
.
(4.9)
where w
LF
N(0, Q
) is a white noise process representing the model innacu-

racies, w
LF
N(0, Q
) Dening the state vector x = [

T
,
T
LF
, b
T
, d
T
,
T
LF
]
T
21
and the measured ship position
m

3
in the {n} frame as the superpo-
sition of LF and WF dynamics, the DP model in (4.9) has the following non
60
4.3 Wave Filtering
linear state space form:
x =
_
_
A
w
0 0 0 0
0 0 0 0 R()
0 0 0 0 0
0 0 0 0 0
0 0 M
1
R()
T
M
1

T M
1
D
_
_
x
. .
f(x)
+
_
_
0
0
0
0
M
1

T
_
_
. .
B
(u +d +f) +
_
_
E
w
0 0 0
0 0 0 0
0 I 0 0
0 0 I 0
0 0 0 M
1
_
_
. .
E
w,
(4.10)
m
=
_
C
w
I 0 0 0
_
. .
H
x +w
,
(4.11)
where w = [w
T
w
, w
T
b
, w
T
d
, w
T
]
T
is the state white noise vector with covariance
Q = diag {Q
w
, Q
b
, Q
d
, Q
}, f(x) is the non linear state transition term, B is

the control input matrix and E is the state disturbance matrix, H is the output
transition matrix and w
N(0, Q
) is the white noise process modeling the

measurement error.
4.3 Wave Filtering
In order to avoid large chattering phenomena in the control system due to
uncontrollable oscillatory motion, produced by the 1-st order waves, the LF
and WF contribution to the ship motion must be separated. For this purpose
a wave lter estimates the low-frequency motion
LF
and
WF
. Since the
GNSS measurement frequency can be as low as 1-10Hz, it is advantageous [14]
to implement the discrete-time version of the Extended Kalman Filtering [61]
to estimate the motion components based on the model in (4.10)-(4.11).
Assuming the control input u(t) = u(k) for t [kT
c
, (k+1)T
s
] with sampling
time T
s
. To obtain an extended Kalman lter with an eective state prediction
equation the model in (4.10) and (4.11) has been linearized about the current
state prediction estimate x
k+1/k
. The terms B, E and H are linear, the state
transition matrix f(x) is linearized with the following form
A
c
=

x
f(x) |
x
k+1/k
(4.12)
Subsequent discretization using a Zero Order Hold (ZOH) with period T
s
of
61
the linearized model has the following form:
x(k) = A
D
(x(k 1)) +B
D
u
c
(k 1) +E
D
w(k 1)
y(k) = Hx(k) +w
(k),
(4.13)
with:
A
D
= e
A
c
T
s
(4.14)
B
D
=
_
T
s
0
e
A
c
dB = A
1
c
(e
A
c
T
s
I)B (4.15)
E
D
=
_
T
s
0
e
A
c
dE = A
1
c
(e
A
c
T
s
I)E (4.16)
Q
D
=
_
T
s
0
e
A
c
Qe
A
R
c

d = QT
s
+
_
A
c
Q+QA
T
c
_
T
2
s
2
+AQA
T
T
3
s
3
(4.17)
The model results in the following EKF with the following form [62] (where
explicit dependence on T
s
has been dropped for simplicity of notation):
P
k+1/k
= A
d
P
k/k
A
T
d
+E
D
QE
T
D
,
K
k+1
= P
k+1/k
H
T
(HP
k+1/k
H
T
+R)
1
,
x
k+1/k+1
= x
k+1/k
+K
k+1
(
m
(k) H
D
x
k+1/k
),
P
k+1/k+1
= (I K
k+1
H)P
k+1/k
,
x
k+1/k
=
k1
x
k/k
+f( x
k/k1
) A
c
x
k/k1
+BT
s
u
k
(4.18)
where R is the output covariance weight matrix. For more details see [14] and
[57].
In order to implement the Kalman lter the parameters of model (4.13) must
be estimated.
4.3.1 Multi-rate Kalman Filter
The ship motion
m
in Eq. (4.11) is measured by devices like GNSS and
compass. These sensors do not work at the same frequency and not all measures
are not available at any sampling instant. The multi-rate model based on the
delta-functions is considered. The introduction of the delta functions modies
the expression of the Kalman gain indicating the presence or the absence of
measurements at each sampling instant, as follows:
K
k+1
= P
k+1/k
H
T
(HP
k+1/k
H
T
+R)
1

k
(4.19)
and only in this formulation when there are no measurements available,
K
matrix is zero and it is only necessary to run the prediction equations.
62
4.4 Discrete Time Variable Structure Control
When new measurements are available,
K
matrix is unitary and all the
Kalman lter equations is run.
When measurements are not available, the multi-rate Kalman lter only
predicts, placing a zero in the
k
matrix.
The Kalman lter gain is composed by a prediction and a correction, made
through the sensor measurements. When measurements are not available, the
multi-rate Kalman lter only predicts, placing a zero in the
k
matrix. For
more details see [63].
The DP control system, proposed by this thesis, is based on the Variable Struc-
ture Control, a discontinuous nonlinear control, where the control law switches
between two functions, depending on the . The aim is to force the ship posi-
tion to follow a behaviour dened by sliding motion surfaces, which dynamics
must be asymptotically stable [64]. The Discrete-Time version of the Variable
Structure Control (DTVSC), [65], [66], [67], allows the implementation of the
control law on the DP computer system, while ensuring robustness with respect
to model uncertainties and input disturbances acting on the actuators. More-
over in this thesis the DTVSC is proposed for handling the thruster failures
above described.
4.4.1 Control Model
The DTVSC proposed in this thesis consider the control of ship position by
using the velocities
LF
in the {b}. The control model is derived from(4.10)
and takes the following form:

LF
= M
1
D
LF
M
1
R()b +M
1

Tu +M
1

Td +M
1

Tf. (4.20)
The DTVSC is applied to the discretization of the ship dynamics model in
(4.20) with the sampling time T
s
using a Zero Order Hold (ZOH):
LF
(k + 1) = e
M
1
DT
s
. .
F
LF
(k) +
_
T
s
0
e
M
1
D
M
1

T d
. .
G
u(k)
_
T
s
0
e
M
1
D
M
1
R()b d
. .
G
NL
+
_
T
s
0
e
M
1
D
M
1

T d
. .
G
d
d(k) +
_
T
s
0
e
M
1
D
M
1

T d
. .
G
f
f(k)
(4.21)
where it is assumed that d(k) and f(k) are constant within the sampling interval
and that the G
NL
quantity, accounting for the eects of slowly-varying distur-
63
bances is bounded. Denoting the control generalized force vector in body-xed
{b} frame as:
(k) = [
u
(k)
v
(k)
r
(k)]
T
= G u(k) (4.22)
the discrete time model used for deriving the control law takes the form:
LF
(k + 1) = F
LF
(k) +
(k) +G
NL
(k)G
d
d(k) +G
f
f(k) (4.23)
It is assumed To account for possible model uncertainties, it is assumed that
model parameters may dier from their nominal values for some unknown but
bounded quantities. G
d
and G
f
nominal values are null in nominal conditions
(G
d
= 0 and G
f
= 0), when no faults and disturbances are inuencing the
system.
F = F + F; G = G+ G; G
d
= G
d
; G
f
= G
f
(4.24)
Considering the ship velocity
LF
(k) = [u(k), v(k), r(k)]
T
and the velocity ref-
erence
LF
(k) = [u
(k), v(k)
, r
(k)]
T
, obtained using the inverse kinematic equation
from (4.2) and the imposed reference trajectory, the reference error has the
following expression:
LF
(k) =
LF
(k)
LF
(k) (4.25)
The procedure to design a Discrete Time Variable Structure Controller is
then composed by two steps
1. Computation of a sliding surface s(k) which has a stable internal dynamics
2. Establishing a control law which steers the closed-loop system towards
the sliding surface and ensures the system trajectories to stay as close as
possible to the surface.
4.4.2 Computation of the sliding surfaces
To resolve the DP tracking problem, the tracking error in (4.25) must be min-
imized. Therefore the following two-steps are proposed:
s(k) =
LF
(k) +
1
LF
(k 1) +
2
LF
(k 2) = 0 (4.26)
Stable internal dynamics of the sliding surfaces is ensured by a proper choice
of the parameter vectors
1
,
2
, which ensure the zeros of (4.26) are inside the
unit circle:
Roots(p()) =
2
+
1
+
2
< 1 (4.27)
64
4.4.3 Computation of the control law
As stated in [65] and [68] a quasi-sliding motion on the surface s(k) = 0 is
achieved if and only if the following Discrete-Time Sliding Mode Existence
Condition (DSMEC) is veried by the sliding dynamics:
|s(k + 1)| < |s(k)| k (4.28)
Theorem 1. Given the DP discrete time model in (4.23) and two-steps sliding
surfaces in (4.26) respecting the condition (4.27), the following Discrete-Time
Variable Structure Control Law
(k) =
eq
(k) +
n
(k) (4.29)
satises the Discrete-Time Sliding Mode Existence Condition (4.28), where the
equivalent control term
eq
is given by:
eq
(k) =
LF
(k) F
LF
(k)
1
LF
(k)
2
LF
(k 1) (4.30)
and the discontinuous control term
n
is given by
n
(k) =
_
(|s(k)| ) if |s(k)| >
s(k) +
n
(k 1) if |s(k)|
(4.31)
with parameters given by the following
= F
LF
max
+ G
d
d
max
+ G
f
f
max
+
(4.32)
LF
(k + 1)
LF
(k)| (4.33)
|
u
| < 1 (4.34)
Where
LF
max
is maximum generalised speed during DP operation, d
max
is the
largest disturbance that can act on the thrusters and f
max
the largest additive
fault that can aect the thrust force vector.
Proof. In order to prove that the control law in Theorem 1 veries the DSMEC,
condition (4.28) is rewritten. By short mathematical computations, condition
(4.28) is equivalent to the ([68]):
s(k)
T
s(k + 1) <
1
2
(s(k + 1))
T
(s(k + 1)) (4.35)
where it is dened the incremental dierence of the sliding surface as follows:
s(k + 1) = s(k + 1) s(k) (4.36)
65
In order to computer the incremental dierence (4.36) of the sliding surface,
the one step ahead of the sliding surface (4.26) must be calculated:
s(k + 1) =
LF
(k + 1) +
1
LF
(k) +
2
LF
(k 1) = 0, (4.37)
The expression in (4.37) is manipulated substituting the expression of the ship
dynamics in (4.23) and considering the model uncertainties in (4.24):
s(k+1) = F
LF
(k) +
1
LF
(k) +
2
LF
(k 1)
. .
known
LF
(k+1)+F
LF
(k)+
(k) = 0,
(4.38)
In (4.38) the underbraced terms do not depend from unknown parameters or
future values. The control force at the k-th sample
(k) must be able to cancel

this dynamics in order to satisfy the equivalence to zero and to compensate for
the uncertainties in F
LF
(k). Thus the controller expression is divided in
two parts, and equivalent control term
eq
and a discontinuous control term
n
:
(k) =
eq
(k)+
n
(k) =
LF
(k) F
LF
(k)
1
LF
(k)
2
LF
(k 1)
. .
eq
+
n
(k)
(4.39)
Then, by (4.39), the Equation (4.38)
s(k + 1) =
n
(k) + F
LF
(k) +
LF
(k)
LF
(k + 1)
. .
h(k)
(4.40)
The right hand of (4.40) is divided in two parts, the
n
, which expression
depends only on the nominal plant and the uncertainty variable h(k), which
depends on unknown or uncertain terms. Then (4.40) is rewritten as follows:
s(k + 1) =
n
(k) +h(k) (4.41)
Combining (4.36) and (4.41) the following is obtained:
s(k + 1) = s(k + 1) s(k)) =
n
(k) +h(k) s(k) (4.42)
Using (4.38) and (4.42) the DSMEC in (4.35) becomes a systems of 3 second-
order inequalities
n
T
(k)
n
(k) + 2h
T
(k)
n
(k) +h
T
(k)h(k) s
T
(k)s(k) < 0 k (4.43)
Solving the systems of second order inequalities in (4.43) respect to
n
, the
66
following solution is obtained
h(k) |s(k)| <
n
(k) < h(k) +|s(k)| (4.44)
Eq. (4.44) shows that the value of the
n
depends on on an unknown variable
h(k). Dening the maximum variation of the reference velocity as
:
|
LF
(k)
LF
(k + 1)| <
k (4.45)
and the maximum value of the reference velocity as
LF
max
|
LF
(k)| <
LF
max
k (4.46)
the vector of maximum uncertainties of the system has the form:
|h(k)| < = F
LF
max
+ G
d
d
max
+ G
f
f
max
+
k (4.47)
Therefore the uncertainty variable h(k) is bounded by (4.47). Assuming the
sector:
|s(k)| > , (4.48)
the following is a conservative choice of
n
for the solution of (4.44) to respect
the DSMEC in the sector given by (4.48):
h(k)|s(k)| < |s(k)| <
n
(k) < +|s(k)| < h(k)+|s(k)| k (4.49)
Considering (4.48) and (4.52), implies:
n
< |(|s(k)| )|
n
= (|s(k)| ) (4.50)
given
|s(k)| >
|
u
| < 1
When |s(k)| < the DSMEC is not guaranteed. However the tracking error
can stay in the sector given by |s(k)| < , assuming that the uncertainty
variable h(k) is slowly variant and using the approach of Time Delay Control
method. Considering (4.41), the sliding surface is rewritten as
s(k) =
n
(k 1) +h(k 1) (4.51)
Imposing the condition of the Time Delay Control s(k + 1) = 0 and assuming
h(k) h(k 1), by some algebra manipulation the following control law is
67
Figure 4.3: Convergence to zero of the sliding surfaces in nominal condition
applying the DTVSC control law. Figure shows that the DSMEC
is satised
proposed:
n
(k + 1) = s(k) +
n
(k) if |s(k)| (4.52)
Integral action for drift forces compensation using drift forces b estimation
from the MREKF is used to ensure steady state zero error, as stated in [69].
I
(k) =
I
(k1)+K
I
T
s
(
LF
(k1)
LF
(k1)) or
I
(k) = R(

)
b (4.53)
In order to obtain zero steady-state errors, it is possible to sum the integral
action (4.53) to DTVSC in the control law, see [14]. The nal control law u is
given combining the DTVSC and Integral actions:
c
(k) =
(k) +
I
(k) (4.54)
Figure 4.4.3 shows the convergence of the sliding surfaces.
68
4.5 Fault Tolerant Control
4.5.1 Thruster Failures
The work [55] stated that regardless the amount of hardware redundancy in-
stalled, all control systems could fail in an instant even if they were thought to
be redundant. However, through proper fault tolerant techniques the chances
are greatly reduced. Thruster failure example scenarios are listed in [55]. DP
classication are give according to DP class requirements. In order to meet the
DP class 2 requirements the thrusters must either fail to zero or as is. How-
ever, in reality, if the pitch hydraulic sticks, the thruster can still fail to full
pitch. The only way to prevent this is to stop a thruster when a signicant
discrepancy is detected between command and feedback. This arrangement is
rare indeed, and in any case, produces other less desirable failure modes.
In the considered scenario the aim is to meet the requirements of DP Class
2 vessel: loss of position is not to occur in the event of a single failure in any
thruster. Thruster failures caused by electrical or communication systems are
addressed by this work, including complete loss of position and thruster stall
at dierent torque values. In the scenario considered faults are modelled as
additive faults, as explained in Section 4.2.3.
4.5.2 Fault Diagnosis
The fault diagnosis module proposed in this work is designed to detect failures
on thruster, which can either fail to zero or as is. The fault diagnosis module
includes fault analysis module (composed by structural properties and resid-
ual generation, that uses two model based techniques), fault detection module
(using a change detection algorithm) and logic decision module (used to isolate
faults). The fault diagnosys is used to recongure the control input accord-
ing to the fault tolerant system in Fig. 4.4 Fig. 4.4 is a fault-tolerant system
architecture.
Structural Analysis
The structural proprieties of the dynamic system are analyzed by the use of the
tool developed by Lorentzen and Blanke [70]. The dependability matrix of the
actuator subsystems in Table 4.2 is used to evaluate the structural properties.
Table 4.2 shows the relation between the residuals derived by the tool proposed
in [70].
Analyzing the depenability matrix of Table 4.2 the following statements are
derived: every violation on each subsystem constraint is detectable, because
any fault changing the relations of each subsystems results in a violation of the
parity equation. A violation of the main propeller subsystem constraint results
69
Figure 4.4: Fault-tolerant system architecture
Table 4.2: Dependability matrix
Subsystem Main Propeller Tunnel Thrusters Azimuth Thrusters
Parity equation 1 x
Parity equation 2 x x
Parity equation 3 x x
70
in a violation of the parity equation 1, therefore faults on the main propeller
susbsystems are isolable.
Parity Space Approach
Residuals for the fault diagnosis system are obtained applying the Parity Space
Approach, a model based technique for linear systems to obtain residuals which
are independent from disturbances and initial state of the system. The lineari-
sation of the ship dynamics model in (4.10) around
LF
= 0 and
LF
= 0,
follows:

LF
= M
1
D
LF
+M
1

Tu +M
1

Td +M
1

Tf
y =
LF
(4.55)
where the failure of a thruster is modeled as an additive fault and enviromental
disturbances are neglected and the output is the ship velocity vector
LF
:
Using the Parity Space Approach proposed in [59], a residual generator is
obtained ltering input u and output y data. According to the considered
approach, the lter described from two transfer functions, V
ru
(s) and V
ry
(s):
(V
ru
(s)V
ry
(s))
p(s)
= Q(s)F(s) (4.56)
where Q(s) is an arbitrary polynomial matrix, F(s) is a matrix which rows
form an irreducible polynomial basis (then any suitable matrix (V
ru
(s)V
ry
(s))
is obtained by combinations of the rows of F(s)) and p(s) is a common denom-
inator for V
ru
(s) and V
ry
(s) . Making the hypothesis of null initial state, the
residual expression in Laplace domain is given by:
r
psa
(s) = (V
ru
(s) V
ry
(s))
_
y(s)
u(s)
_
(4.57)
Given the expression of V
ru
and V
ry
the residual expression from parity space
approach follows:
r
psa
(s) =
Q(s)F(s)
p(s)
_
y(s)
u(s)
_
(4.58)
The dependency matrix for the residual obtained from parity space approach
is shown in Table 4.3. Three residuals are found using the considered Parity
Space method, namely r
psa
j
, where j 1, 2, 3. The terms f
i
refers to a fault in
the i-th thruster, with i 1, 2, .., 6 for the considered reference ship shown in
Fig. 4.7. Results in Table 4.3 match and, as stated by the structural analysis,
which as a consequence of the symmetry of the actuator system, which are
equal in pairs in nominal conditions. When a fault occurs on the actuators,
one of the subsystems does not verify the constraints, because the two actuator
71
Table 4.3: Dependency matrix
f
1
f
2
f
3
f
4
f
5
f
6
r
psa
1
x x 0 0 x x
r
psa
2
0 0 x x x x
r
psa
3
x x x x x x
torques of the pair are dierent in the faulty case.
Luenberger observer
In order to solve the fault isolation problem , the Luenberger observer is used
to nd other residuals[71]. According to the introduced notations, the DP ship
linear model, without considering enviromental disturbances and faults, has
the following form:

LF
= M
1
D
. .
A
LF
+M
1

T
. .
B
u
y =
LF
(4.59)
Then Leuenberger state estimate takes the form:
LF
= (AG)
L
+Bu
c
+Gy (4.60)
where G is a matrix chosen such as the eigenvalues of (AG) are in the left-half
plane. The residual are calculated as the dierence between the measured and
estimated velocity:
r
Lue
(t) = (4.61)
4.5.3 Change detection
To impose the robustness the cumulative sum (CUSUM) a change detector is
implemented. To identify faults causing changes in the mean of the residuals,
without changing their standard deviation, the cumulative sum algorithm in-
troduced in [72] is proposed for improving the robustness of the fault detection
system. Denoting with r
k
the residual value at the time k,
0
and
2
the mean
and variance, respectively, of the residual and the change of the mean the
decision function is:
g
+
k
=

2
max
_
0, g
+
k1
+r
k

0

2
_
(4.62)
72
Figure 4.5: Isolation logic
Table 4.4: Isolation Logic
f
1
f
2
f
3
f
4
f
5
f
6
alm
PSA
1
1 1 0 0 0 0
alm
PSA
2
0 0 1 1 0 0
alm
PSA
3
0 0 0 0 1 1
alm
Leu
1
0 1 1 0 0 0
alm
Leu
2
1 0 0 1 1 1
The decision function g
+
k
is evaluated against a threshold h to trigger an
alarm. The implemention the CUSUM algorithm requires the knowledge of
the mean value of the residue after the fault. The mean value of the resid-
ual after the fault can be overestimated, as stated in [59]. The threshold h
is a trade-o between the delay in identifying the fault and the false alarm
frequency. The CUSUM algorithm is applied to the residual obtained from
the Parity Space Approach and the Luenberger observer approach. If the
decision function is greater than the threshold h the alarm for the respec-
tive residual is triggered as described in Table 4.4, where the alarm vector
alm = [alm
PSA
1
, alm
PSA
2
, alm
PSA
3
, alm
Leu
1
, alm
Leu
2
]
T
.
4.5.4 Fault isolation
The proposed fault isolation model allows to identify the faulty thruster com-
bining the results of change detection algorithm applied on residuals processed
by Parity Space Approach and the Luenberger Observer. The proposed fault
isolation logic is shown in Fig. 4.5 and Table 4.4, where the i-th fault is iso-
lated when the alarm alm vector has the value given in the i-th column. As
an example a fault on the main propeller (thruster number 1) is isolated when
alm
PSA
1
= 1 and alm
Leu
1
= 1 and all the others alarms are not triggered.
73
Figure 4.6: Bench of controllers
It is worth to notice that faults aecting the two azimutal thrusters cannot
be isolated: when alm
PSA
3
= 1 and alm
Leu
2
= 1 it is conrmed a fault aected
either one or two azimuth truster, but it is not conrmed which one is faulty.
To overcome this problem a solution is proposed in Section ??.
4.5.5 System reconguration
Fault handling should incorporate a reconguration step to recover from the
actuator faults, though performance specication should be less restrictive.
The system reconguration algorithm includes controllers for scenarios and
employes the results of fault diagnosis to enable the appropriate controller c
i
that can handle the fault f
i
. Its necessary to build a bench of controllers, one
for each possible fault, as in Fig. 4.6.
In nominal conditions the thrust forces are chosen using (4.7), therefore in
this condition, by the thruster allocation algorithm described in Section 4.2.3,
the controller c
0
allocates the same thrust on each pair of actuator. This is
due to the symmetry of the thrusters to respect to the ship geometry, as said
in Subsection 4.2.3.
(k) +
I
(k) = Gu(k)
u
i
(k) = u
i+1
(k) for i = 1, 3, 5
(4.63)
When a fault is detected on the i th thruster the relative thrust force is set to
74
4.6 Control Allocation
Figure 4.7: Thrusters conguration for the vessel
zero (complete loss) or a constant value (stall) while the (4.63) is solved respect
to the other thrusters and the redundant and symmetric actuator of the pair
must be used. The controller with the new transfer allocation algorithm is so
called c
i
.
In this section the focus is on the thrust allocation problem in case of roteable
thrusters, like the azimuth thrusters, that is: how much thrust should be pro-
vided by each of the thrust devices. Thruster conguration is shown in Fig.
4.7
The aim of the control allocation is to compute azimuth thruster angle
c
and torque f
c
commands in order to satisfy the control law in (4.63) for all
thrust devices. The allocation problem can be formulated as the calculation of
a pair (
c
, f
c
) that satises:
c
= T(
c
)f
c
(4.64)
where
c
= is given by the control law in (4.63) and T(
c
), considering a
vessel with n azimuth rotatable thrusters, can be written as
T() = AB() (4.65)
So, equation (4.64) becomes:
= T()f = AB()
. .
t
f
(4.66)
where A is a 3 2n matrix:
75
A =
_
_
1 0 .. 1 0
0 1 .. 0 1
l
y1
l
x1
.. l
yn
l
xn
_
_
(4.67)
and B is a 2n n matrix
B =
_
_
cos(
1
) 0 .. 0
sin(
1
) 0 .. 0
.. .. .. ..
0 0 .. cos(
n
)
0 0 .. sin(
n
)
_
_
(4.68)
For azimuth i, the angle
i
is given by:
i
= arctan(t
y
i
/t
x
i
) (4.69)
The static solution is obtained using the Damped least squares (DLS) method,
that allows to obtain the thrust command also near a singular point. The
dynamic problem is solved by assuming that the thrust command can be per-
turbed about the static thrust command. More information are provided in
[73]. The allocation problem is divided into a static and dynamic allocation
problem, to take into account the thruster dynamics. The thruster is modelled
as a Low-Pass (LP) lter in static conditions. For small command deviations,
it is modelled as a High-Pass (HP) lter.
4.6.1 Static Allocation
The LP-ltered command is given by
+ =
c
(4.70)
where = diag(
1
T
1
,
1
T
2
,
1
T
3
) is a diagonal matrix containing time constant for
surge, sway and yaw.
The LP-ltered aximuth angles and thrust forces are
i
= arctan (t
y
i
/t
x
i
)
f = B
()t = B
T
()t
(4.71)
4.6.2 Dynamic Allocation
Let assume small deviations in thrust command around the LP-ltered
thrust command . The change in the azimuth angles and thrust forces
f can be computed assuming a perturbation (, f) about the point (, f):
76
= J(, f)
_
f
_
T
(4.72)
where J
_
, f
_
is the Jacobian matrix dened as
J(, f) =
_
T()f
T()
_
=f=f
(4.73)
The dynamic thrust command is computed by using a limited HP-lter
of the thrust command
c
. The lter has the form of the following transfer
function:
i
(s) =
K
i
s
s
2
+s
i
n,i
s +
2
n,i
c,i
(s)
(4.74)
where
n,i
is the cut-o frequency and
i
is the relative damping ratio in
surge, sway and yaw respectively.
4.6.3 DLS Algorithm
The control allocation problem is solved superimposing the static and the dy-
namic solutions. Thruster angle is given by
c
= + (4.75)
The DLS method provides the thruster torques
f
c
= T
DLS
(
c
)
c
(4.76)
where T
DLS
(
c
) is the pseudo-inverse matrix based on the DLS, see [73].
77
Chapter 5
Experimental Results
The 1:30 scale model of a naval surveillance vessel,the CyberShip 3 (CS3),
shown in Figure5.1 was used for testing. This ship is located at the Marine
Cybernetics Laboratory (MCLab), Norwegian University of Science and Tech-
nology. For more details about the CS3, see [74] and [75]. The ship is equipped
with three azimuthal thrusters, as shown in Fig. 4.7. Hence we have 6 con-
trol variables for 3 DOF, u = [u
1
, u
2
, u
3
]
T
and = [
1
,
2
,
3
]
T
. Here we
assume state feedback, an extension to output feedback using state observers
will be a next step. Thrusters positions are L
y1
= 0.11m, L
y2
= 0.11m,
L
x1
= L
x2
= 0.789m and L
x3
= 0.636m. Referring to Fig. 4.7, the maximum
thrust of the main thrusters is 21.9 N and the maximum thrust of the fore
thruster is 10 N.
The thruster allocation matrix T() is:
T() =
_
c(
1
) c(
2
) c(
3
)
s(
1
) s(
2
) s(
3
)
l
y1
c(
1
) l
x1
s(
1
) l
y2
c(
2
) l
x2
s(
2
) l
x3
s(
3
)
_
(5.1)
Where c(
i
) and s(
i
) stand for cos(
i
) and sin(
i
) respectively. So, the A
and B() matrix are
A =
_
1 0 1 0 1 0
0 1 0 1 0 1
l
y1
l
x1
l
y2
l
x2
l
y3
l
x3
_
(5.2)
B() =
_
_
c(
1
) 0 0
s(
1
) 0 0
0 c(
2
) 0
0 s(
2
) 0
0 0 c(
3
)
0 0 s(
3
)
_
_
(5.3)
Thruster dynamics are described by the following parameters: time constants
for the LP-lter are T
i
= 10s, the cut-o frequencies are
n,i
= 0.63, for
i = 1, .., 3 and the damping ratio is
i
= 1 for i 1, 2, 3.
The performance index is considered for a quantitative comparison between
results of various simulations. The performance index used here is the In-
79
Chapter 5 Experimental Results
Figure 5.1: CyberShip 3 at the MCLab
tegral of Squared Error ISE. For the DP system index depend on the er-
ror between position and orientation references and position and orientation
measured by sensors, namely e(t) =
des
(t) (t). The index expression is
ISE =
1
t
f
t
0
_
t
f
t
0
e(t)
2
dt.
In the following we show the comparison between the proposed control system
and a conventional PID controller, using a conguration similar to the one
reported in [73] for the Cybership I, considering the following case scenarios:
Case 1 The scale model is sailing in calm water with slight disturbances
from wind and currents in a short straight movement.
Case 2 The scale model is sailing in slight water with moderate disturbances
from wind and currents in a short straight movement.
Case 3 The scale model is sailing in calm water with slight disturbances
from wind and currents, desired position and changes 2 times.
Reference positions in the {n} frame are shown in Table 5.1. Environmental
conditions during tests, in terms of signicant wave height H
s
, modal frequency
w
0
, wind velocity V
w
and current speed V
c
, are shown in Table 5.2. Results
in term of ISE performance index are in Table 5.3. The vessel is sailing in
quartering sea. This aects the index for the n direction, which is therefore
considerably bigger compared to e for all the case scenarios. Results from CASE
1 simulations are shown in Fig. 5.2. Results from the CASE 3 simulations are
80
Table 5.1: Desired Positions
Case Time (s) x
d
(m) y
d
(m)
d
(deg)
1 0-1000 0 0 0
1 1000-2000 10 10 20
2 0-1000 0 0 0
2 1000-2000 10 10 20
3 0-800 5 3 arctan (y
d
/x
d
)
3 800-1400s 6 3 arctan (y
d
/x
d
)
3 1400-2000 6 7 arctan (y
d
/x
d
)
Table 5.2: Environmental conditions
Case H
s
(m)
0
(rad/s) V
w
(m/s) V
c
(m/s)
1 0.1 1.1109 1.5 0.05
2 0.3 1.0003 3 0.1
3 0.1 1.1109 1.5 0.05
shown in Fig. 5.3. Azimuth thruster angle and torques set by the control
allocation algorithms are also show in Fig. 5.4.
Simulation results show that the DTVS controller guarantees a satisfactory
smoother performance with respect to the PID controller in every case con-
sidered. The DTVSC can eventually require larger control eorts and could
be more complex to deploy and computationally expensive with respect to
the PID controller, it guarantees better performance. Results of the CASE 1
simulations, depicted in Fig. 5.2(b), show that using the PID controller the
ship orientation slowly converge to the reference orientation, while the DTVSC
controller rapidly brings the system state nearby the sliding surfaces, while the
integral action counteract the 2-nd order slowly varying drift forces. While
Case 3 simulation results in Fig. 5.3(b) show that DTVS control system per-
forms better during reference position changes, notably for the yaw rotation
angle. Fig. 5.4(c) and Fig. 5.4(d) shows that the price to pay for the per-
formance improvement of the DTVSC is the control eort, while Fig. 5.4(a)
and Fig. 5.4(b) show how the DLS method implemented for the allocation
algorithm helps in preventing further increases of the control eorts, when the
Table 5.3: ISE performance index
n (surge position) e (sway position) (yaw rotation)
DTVSC PID DTVSC PID DTVSC PID
Case 1 0.0023 0.03 7.47 10
6
7.99 10
6
1.51 10
5
1.61 10
5
Case 2 0.0073 0.07 2.1 10
5
2.13 10
5
3.64 10
5
3.98 10
5
Case 3 0.0057 0.024 5.2 10
6
5.88 10
6
1.51 10
5
1.62 10
5
81
0 2 4 6 8 10
0
2
4
6
8
10
e (m)
n

(
m
)
(a) References and simulated path (b) Reference and simulated yaw rotation
angle
Figure 5.2: CASE 1 - References (red line) and simulated n, e and using
DTVS controller (blue and dotted line) and PID (black and dash-
dotted line)
0 1 2 3 4 5 6 7
0
1
2
3
4
5
6
e (m)
n

(
m
)
(a) References and simulated path
0 500 1000 1500 2000
20
10
0
10
20
30
40
50
time (s)
p
s
i

(
d
e
g
)
(b) Reference and simulated yaw rotation
angle
Figure 5.3: CASE 3 - References (red line) and simulated n, e and using
DTVS controller (blue and dotted line) and PID (black and dash-
dotted line)
82
0 500 1000 1500 2000
100
50
0
50
100
150
200
time (s)
a
l
f
a

D
T
V
S
C

(
d
e
g
)
(a) Azimuth angle using DTVS controller
0 500 1000 1500 2000
100
50
0
50
100
150
200
time (s)
a
l
f
a

P
I
D

(
d
e
g
)
(b) Azimuth angle using PID
0 500 1000 1500 2000
2
1.5
1
0.5
0
0.5
1
1.5
2
time (s)
f

D
T
V
S
C

(
N
)
(c) Thruster torque f using DTVSC
0 500 1000 1500 2000
2
1.5
1
0.5
0
0.5
1
1.5
2
time (s)
f

P
I
D

(
N
)
(d) Thruster torque f using PID
Figure 5.4: CASE 3 - First (blue line), second (red line) and third (black line)
azimuth thrusters angle and thrust force f
83
azimuth thrusters conguration is near a singular value. Results in Table 5.3
show that the worse environmental conditions of CASE 2 respect to CASE 1
are reected by far larger ISE performance indices both for PID and DTVS
controller. Anyway the latter always show better perfomances in terms of ISE
index compared to the PID.
84
Chapter 6
Conclusions
In this work, the design of an integrated control system for heading and roll
damping of a multipurpose naval vessel by means of active ns and rudder has
been investigated. A multivariable nonlinear model which takes into account
the hydrodynamic couplings among sway, roll and yaw due to both the wave
eect and the control devices has been considered. A stability analysis of the
zero dynamics for the proposed nonlinear model along with the design of a
MIMO variable structure control law has been performed. Simulation results
have been carried out to corroborate the theoretical analysis. Moreover, a val-
idation of the proposed mathematical analysis for several parameters data-set
describing dierent operative conditions for dierent vessels will be performed.
The problem of dynamic positioning plays a key role in all those cases where
it is not possible to anchor the ship at the seabed, or where the ships position
is bound to a specic point on the bottom. In this thesis an architecture is
identied for the autonomous dynamic positioning of marine vessel equipped
with azimuth thruster using a non-linear discrete control and robust control
allocation technique for the resolution of the DP problem. Here we assume state
feedback, an extension to output feedback using state observers will be a next
step. From a control perspective, it was shown that the DTVS controller satisfy
the sliding mode existence condition in a sector depending on the maximum
uncertainties of the system. Simulations conrm the robustness of the control
scheme in presence of environmental disturbances also in moderate seas and
show that DTVSC outperformed classic PID controller.
85
Bibliography
[1] T. Perez, Ship motion control : course keeping and roll stabilisation using
rudder and ns. London: Springer, 2005.
[2] G. C. Goodwin, T. Perez, M. Seron, C. Y. Tzeng, and Ching Yaw Tzeng,
On fundamental limitations for rudder roll stabilization of ships, in
Proceedings of the 39th IEEE Conference on Decision and Control, vol. 5,
2000, pp. 4705 4710 vol.5. [Online]. Available: 10.1109/CDC.2001.914671
[3] T. Perez and M. Blanke, Simulation of ship motion in seaway, Technical
University of Denmark, Lingby, Denmark, Tech. Rep. EE02037, 2002.
[4] T. Perez and T. Fossen, Time-domain models of marine surface vessels
for simulation and control design based on seakeeping computations, in
Proc.of the 7th Conference on Manoeuvring and Control of Marine Craft
MCMC, Lisbon, Portugal, Sept 2006.
[5] S. Haverre and T. Moan, On some uncertainties related to short term
stochastic modelling of ocean waves. Probabilistic Oshore Mechanics,
Progress in Engineering Science, 1985.
[6] A. Gelb and W. E. V. Velde, Multiple-input describing functions and
nonlinear system designs, 1968.
[7] W. Price and R. Bishop, Probabilistic theory of ship dynamics, 1974.
[8] C. G. Kllstrm, Identication and adaptive control applied to ship steer-
ing, Ph.D. dissertation, Lund Institute of Technology, Lund, Sweden,
1989.
[9] T. Fossen, Guidance and control of ocean vehicles. Chichester, New York:
John Wiley and Sons, Ltd Sons, Ltd, 1994.
[10] T. Perez, Non-linear and quasi-linearized roll models for seakeeping anal-
ysis of a patrol boat, Newcastle, Australia, Tech. Rep., nov 2002.
[11] A. R. J. M. Lloyd, Seakeeping: Ship behaviour in rough weather, Ellis
Hoorwood Series in Marine Technology, 1989.
87
Bibliography
[12] C. Vassiaopoulos, Ship rolling at zero speed in beam seas with non-linear
damping restoration, J. Ship Research, 1971.
[13] MSS, Marine Systems Simulator, Norwegian University of Science
and Technology, Trondheim, Norway, 2007. [Online]. Available: http:
//www.marinecontrol.org
[14] T. I. Fossen, Handbook of Marine Craft Hydrodynamics and Motion Con-
trol. Chichester, UK: John Wiley & Sons, Ltd, Apr. 2011.
[15] R. Isherwood, Wind Resistance of Merchant Ships, Trans. Inst. Naval
Arch., RINA, vol. 115, pp. 327338, 1972.
[16] A. Srensen, S. Sagatun, and T. Fossen, Design of a dynamic positioning
system using model-based control, Control Engineering Practice, vol. 4,
no. 3, pp. 359 368, 1996.
[17] R. Taggart, Anomalous behaviour of merchant ship steering systems,
Marine Technology, pp. 205215, 1970.
[18] P. G. M. Van der Klugt, Rudder Roll Stabilization, Ph.D. dissertation,
Delft University of Technology , The Netherlands, 1987.
[19] J. Van Amerongen, P. G. M. Van der Klugt, and H. R. Van
Nauta Lemke, Rudder roll stabilization for ships, Automatica,
vol. 26, no. 4, pp. 679690, Jul. 1990. [Online]. Available:
http://doc.utwente.nl/72885/http://www.sciencedirect.com/science/
article/B6V21-47WTGG3-V4/2/caa378a523b12153719a169fc8742cb8
[20] T. Laudval and T. I. T. Fossen, Rudder Roll Stabilization of Ships
Subject to Input Rate Saturation Using a Gain Scheduled Control Law,
in In Proc. of IFAC CAMS98, no. October 1998, 1998, pp. pages 121126.
[Online]. Available: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=
10.1.1.49.1100
[21] M. Blanke, P. Haals, and K. K. Andreasen, Rudder roll damping ex-
perience in denmark, in Proc. of IFAC workshop CAMS 1989, Lyngby,
Denmark, 1989.
[22] J. Adrian, J. Bentsen, M. Blanke, and K. Larsen, Rudder roll damp-
ing in coastal region sea conditions, Proc. of 5th IFAC Conference on
Manoeuvring and Control of Marine Craft, MCMC2000, 2000.
[23] C. G. Kllstrm, P. Wessel, and S. Sjolander, Roll reduction by rudder
control, Spring meeting-STAR Symposium, 3rd IMSDC, 1998.
88
Bibliography
[24] E. Baitis and L. V. Schmidt, Ship roll stabilization in the u.s. navy,
Naval Engineering journal, pp. 1013, 1989.
[25] M. Blanke and A. Christensen, Rudder Roll Damping Autopilot Robust-
ness to Sway-Yaw-Roll Couplings, in In Proceedings of 10th Ship Control
Systems Symposium, Ottawa, Canada, 1993, pp. 93119.
[26] C. G. Kllstrm and W. L. Schultz, An integrated rudder control system
for roll damping and course manteinance, Proc. of the 9th Ship Control
System Symposium, pp. 3278, 1990.
[27] G. Hearns and M. Blanke, Quantitative analysis and design of a rud-
der roll damping controller, in Proc. of 4th IFAC Conference on Control
Applications in Marine Systems, Fukuoka, Japan, oct 1998.
[28] T. Laudval and T. Fossen, Nonlinear rudder-roll damping of non-
minimum phase ships using sliding mode control, Proc. of the European
Control Conference, Brussel, Belgium, 1997.
[29] T. Perez and M. Blanke, Ship roll damping control, Annual Reviews
in Control, vol. 36, no. 1, pp. 129 147, 2012. [Online]. Available:
http://www.sciencedirect.com/science/article/pii/S1367578812000119
[30] , Ship roll motion control, in Proc. CAMS 2010, Rostock, Germany,
2010, pp. 112.
[31] C. Y. Tzeng, C. Y. Wu, and Y. L. Chu, A sensitivity function approach
to the design of rudder roll stabilization controller, Journal of Marine
Science and Technology, no. 2, pp. 100112, 2001.
[32] J. C. Doyle, B. A. Francis, and A. R. Tannenbaum, Feedback Control
Theory. Macmillan Publishing Company, 1992.
[33] M. Blanke, Sensitivity specication based design of rudder-roll damping,
Technical University of Denmark DTU, Lyngby, Denmark, Tech. Rep.,
2008.
[34] M. Blanke, J. Adrian, K. Larsen, and J. Bentsen, Rudder roll damp-
ing in coastal region sea conditions, Proc. of 5th IFAC Conference on
Manoeuvring and Control of Marine Craft, MCMC2000, 2000.
[35] J. Van Amerongen, Adaptive steering of ships - a model reference ap-
proach to improved maneouvering and economic course keeping, Ph.D.
dissertation, Delft University of Technology, The Netherlands, 1982.
89
Bibliography
[36] T. Perez, A. Ross, and T. I. Fossen, A 4-dof simulink model of a coastal
patrol vessel for manoeuvring in waves, in Proc. of the 15th Conference
on Marine Craft Maneuvering and Control (MCMC06), Lisboa, Portugal,
2006.
[37] V. Tzeng and C. Wu, On the design and analysis of ship stabilizing n
controller, Marine Science and Technology, vol. 8, no. 2, 2000.
[38] G. N. Roberts, M. T. Sharif, R. Sutton, and A. Agarwal, Robust control
methodology applied to the design of a combined steering/stabiliser system
for warships, Control Theory and Applications, IEEE Proceedings of, vol.
144, no. 2, pp. 128136, mar. 1997.
[39] R. Katebi, A two layer controller for integrated n and rudder roll sta-
bilization, in Proc. of the Conference on Control Applications in Marine
Systems (CAMS 2004), 2004, pp. 101106.
[40] T. Fossen and T. Lauvdal, Nonlinear stability analysis of ship autopilots
in sway, roll and yaw, in Proc. of the 3rd Conf. on Manoeuvring and
Control of Marine Craft (MCMC94), 1994, pp. 113124.
[41] O. Faltinsen, Sea Loads on Ships and Oshore Structures. Trondheim,
Norway: Marine Cybernetics, 2002.
[42] A. Isidori, Nonlinear Control Systems. Secaucus, NJ, USA: Springer-
Verlag New York, Inc., 1995.
[43] A. J. Srensen, A survey of dynamic positioning control systems, Annual
Reviews in Control, vol. 35, no. 1, pp. 123136, 2011.
[44] D. Nguyen and A. Sorensen, Setpoint chasing for thruster-assisted posi-
tion mooring, Oceanic Engineering, IEEE Journal of, vol. 34, no. 4, pp.
548558, oct. 2009.
[45] J. G. Balchen, N. A. Jenssen, E. Mathisen, and S. Slid, A dynamic posi-
tioning system based on kalman ltering and optimal control, Modeling,
Identication and Control, vol. 1, no. 3, pp. 135163, 1980.
[46] T. I. Fossen, Nonlinear passive control and observer design for ships,
Modeling, Identication and Control, vol. 21, no. 3, pp. 129184, 2000.
[47] J. Strand and T. I. Fossen, Nonlinear passive observer design for ships
with adaptive wave ltering, in New Directions in nonlinear observer
design. Springer Berlin / Heidelberg, 1999, vol. 244, pp. 113134.
90
Bibliography
[48] S. Fang, B. J. Leira, and M. Blanke, Reliability-based dynamic posi-
tioning of oating vessels with riser and mooring system, in Interna-
tional Conference on Computational Methods in Marine Engineering (MA-
RINE), Lisbon, Portugal, Sept. 2011.
[49] G. Xia, X. Shi, M. Fu, H. Wang, and X. Bian, Design of dynamic po-
sitioning systems using hybrid CMAC-based PID controller for a ship,
in Mechatronics and Automation, 2005 IEEE International Conference,
vol. 2, 2005, pp. 825 830 Vol. 2.
[50] E. A. Tannuri and A. C. Agostinho, Higher order sliding mode
control applied to dynamic positioningsystems, Rostock-Warnemnde,
Germany, sep 2010, pp. 132 137. [Online]. Available: http:
//dx.doi.org/10.3182/20100915-3-DE-3008.00048
[51] E. Tannuri, A. Agostinho, H. Morishita, and L. Moratelli, Dynamic posi-
tioning systems: An experimental analysis of sliding mode control, Con-
trol Engineering Practice, vol. 18, no. 10, pp. 1121 1132, 2010.
[52] MSC/IMCA, Guidelines for Vessels with Dynamic Positioning Systems,
Internation Maritime Organization, Tech. Rep., 1994.
[53] IMCA, Dynamic Positioning Station Keeping Incidents - Incidents re-
ported for 2007, Internation Maritime Organization, Tech. Rep., 1994.
[54] D. Phillips, The dynamic positioning of ships: the problems solved? in
Control 96, UKACC International Conference on (Conf. Publ. No. 427),
vol. 2, sept. 1996, pp. 1214 1219 vol.2.
[55] D. F. Phillips, Classic single-point failures of redundant dp systems, in
Proc. of the Dynamic Positioning Conference, 1998.
[56] M. Blanke, Diagnosis and fault-tolerant control for ship station keeping,
in Intelligent Control, 2005. Proceedings of the 2005 IEEE International
Symposium on, Mediterrean Conference on Control and Automation, june
2005, pp. 1379 1384.
[57] M. Fu, F. Ding, M. Li, and P. Yi, A nonlinear estimate lter designed
for ship dynamic positioning, in 8th IEEE International Conference on
Control and Automation, Jun. 2010, pp. 914919.
[58] M. Fu, J. Ning, and Y. Wei, Fault-tolerant control of dynamic positioning
vessel by means of a virtual thruster, in Mechatronics and Automation
(ICMA), 2011 International Conference on, aug. 2011, pp. 1706 1710.
[59] M. Blanke, M. Kinnaert, J. Lunze, and M. Staroswiecki, Diagnosis and
Fault-Tolerant Control, 2nd ed. Berlin: Springer, 2006.
91
Bibliography
[60] T. Perez, T. Fossen, and A. Srensen, A discussion about seakeeping
and manoeuvring models for surface vessels, Centre for Ships and Ocean
Structures (CESOS), Norwegian University of Science and Technology
NTNU, Trondheim, Norway, MSS-TR-001, 2004. [Online]. Available:
http://www.cesos.ntnu.no/mss/
[61] F. Benetazzo, G. Ippoliti, S. Longhi, and P. Raspa, Discrete Time Vari-
able Structure Control for the Dynamic Positioning of an Oshore Supply
Vessel, in Preprints of the 2012 IFAC Workshop on Automatic Control
in Oshore Oil and Gas Production, 2012.
[62] L. Jetto, S. Longhi, and G. Venturini, Development and experimental
validation of an adaptive extended kalman lter for the localization of
mobile robots, IEEE Transactions on Robotics and Automation, vol. 15,
pp. 219229, 1999.
[63] M. Mora and J. Tornero, Path planning and trajectory generation using
multi-rate predictive articial potential elds, in IEEE/RSJ International
Conference on Intelligent Robots and Systems, sept. 2008, pp. 2990 2995.
[64] V. Utkin, Sliding modes in control and optimization. Berlin Allemagne:
Springer-Verlag, 1992.
[65] K. Furuta, Sliding mode control of a discrete system, Systems and Con-
trol Letters, vol. 14, no. 2, pp. 145 152, 1990.
[66] M. Corradini, G. Ippoliti, S. Longhi, and G. Orlando, A quasi-sliding
mode approach for robust control and speed estimation of PM synchronous
motors, IEEE Transactions on Industrial Electronics, vol. 59, no. 2, pp.
1096 1104, Feb. 2012.
[67] M. Corradini and G. Orlando, A discrete adaptive variable-structure con-
troller for mimo systems, and its application to an underwater rov, IEEE
Transactions on Control Systems Technology, vol. 5, no. 3, pp. 349 359,
may 1997.
[68] S. Sarpturk, Y. Istefanopulos, and O. Kaynak, On the stability of discrete-
time sliding mode systems, IEEE Transaction of Automatic Control, vol.
AC-32, pp. 930932, 1987.
[69] A. Loria, T. I. Fossen, and E. Panteley, A separation principle for dy-
namic positioning of ships: theoretical and experimental results, Control
Systems Technology, IEEE Transactions on, vol. 8, no. 2, pp. 332343,
Mar. 2000.
[70] T. Lorentzen and M. Blanke, Satool software reference, Tech. Rep., 2005.
92
Bibliography
[71] D. G. Luenberger, An introduction to observers, IEEE Trans. on Auto-
matic Control, vol. 16, no. 6, pp. 596602, 1971.
[72] M. Basseville and I. V. Nikiforov, Detection of Abrupt Changes - Theory
and Application. Englewood Clis N.J.: Prentice Hall, 1993.
[73] S. P. Berge and T. I. Fossen, Robust Control Allocation Of Overactuated
Ships; Experiments With A Model Ship, in Proc. of the 4th IFAC Con-
ference on Manoeuvring and Control of Marine Craft, Brijuni, Croatia,
1997, pp. 166171.
[74] T. S. Solvin, Underactuated dynamic positioning with model test, Mas-
ters thesis, Norwegian University of Science and Technology, Trondheim,
Norway, 2011.
[75] T. D. Nguyen, A. J. Srensen, and S. Tong Quek, Design of hybrid con-
troller for dynamic positioning from calm to extreme sea conditions, Au-
tomatica, vol. 43, pp. 768785, May 2007.
93

Tesi Raspa

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

Tesi Raspa

Hochgeladen von

Copyright:

Verfügbare Formate

Universit Politecnica delle Marche

Scuola di Dottorato di Ricerca in Scienze dellIngegneria

is function of the lower bound of the attenuation

() is the autocorrelation of the stochastic process.

(), a realization of the sea elevation can be

approach was followed. Further details are

are given in Section

to the roll motion:

() and the motion RAO the amplitudes of the

, which represents the maximum

outside the damping region, but any information on the location

cannot be provided. An improvement to the S(j)

is function of the lower bound of the attenuation

and the damping coecient

with its Left-Half Plane mirror, using an

MV-CLOC 18.50% 20.79% 22.96%

, i.e., the number of time the i-th output y

RSR() 89.0501 95.2255 93.9063

RSR() 86.8729 90.8779 93.3571

Figure 3.13: Zero Dynamics behavior.

) is a white noise process representing the model innacu-

) Dening the state vector x = [

}, f(x) is the non linear state transition term, B is

) is the white noise process modeling the

(k) must be able to cancel

Das könnte Ihnen auch gefallen