Kalman Filtering

for Positioning and

Heading Control of
Ships and Offshore Rigs
THOR I. FOSSEN and TRISTAN PEREZ

ESTIMATING THE EFFECTS

OF WAVES, WIND,
AND CURRENT

PHOTO BY D.S. BERNSTEIN

32 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2009 1066-033X/09/$26.00©2009IEEE

S
hips and offshore rigs perform missions that require navigation system also performs signal quality checking and
tight motion control. During the past 30 years, there transforms the measurements to a common reference frame
has been an increasing demand for higher accuracy used by the control and guidance systems [1]–[3].
and reliability of ship motion-control systems to such Environmental forces due to waves, wind, and current are
an extent that it is difficult to conceive of operations considered disturbances to the motion-control system. These
in which motion-control systems are not an essential part, or forces, which can be described in stochastic terms, are concep-
even an enabling factor. Modern marine vessels are equipped tually separated into wave- and low-frequency components.
with sophisticated motion-control systems that have differ- Waves produce a pressure change on the hull surface,
ent objectives depending on the particular operations being which in turn induces forces. These pressure-induced forces
performed. Some of these control objectives include position have an oscillatory component that depends linearly on the
and heading regulation, path following, trajectory tracking, wave elevation. Hence, these forces have the same fre-
and wave-induced motion reduction. quency as that of the waves and are therefore referred to as
The stochastic nature of environmentally induced forces wave-frequency forces. Wave forces also have a component
has led to extensive use of the Kalman filter for estimating that depends nonlinearly on the wave elevation [4], [5].
ship-motion-related quantities and for filtering disturbances, Nonlinear wave forces are due to the quadratic dependence
both of which are of paramount importance for implement- of the pressure on the fluid-particle velocity induced by the
ing marine motion-control systems. In this article, we intro- passing of the waves. If, for example, two sinusoidal inci-
duce the elements of a ship motion-control system, describe dent waves have different frequencies, then their quadratic
the models used for position-regulation and course-keeping relationship gives pressure forces with frequencies at both
control, and discuss how the Kalman filter is used for wave the sum and difference of the incident wave frequencies as
filtering, that is, removing the oscillatory wave-induced well as zero-frequency or mean forces. Hence, the nonlinear
motion from velocity and positioning measurements. wave forces present both lower and higher frequencies than
the wave frequencies. The mean wave forces cause the
SHIP MOTION CONTROL AND WAVE FILTERING vessel to drift. The forces with a frequency content at the
Ships and offshore rigs are exposed to wind, waves, and cur- difference of the wave frequencies can have low-frequency
rent forces. The requirements of the various operations per- content, which can lead to resonance in the horizontal
formed by the vessel, together with the characteristics of the motion of vessels with mooring lines or under positioning
environmental forces, define the modes of operation and the control [5]. The high-frequency, wave-pressure-induced
objectives of the motion-control system. Figure 1 illustrates
the components of a ship motion-control system [1], [2]. The
guidance system generates a smooth and feasible desired ref- Motion-Control System
erence trajectory described in terms of position, velocity, and Desired Forces
acceleration. The trajectory is generated by algorithms that Observer
Controller Allocation
use the ship’s actual and desired position and a mathematical (Wave Filter) −
model of the ship together with information regarding mis- Actuator
sions, operator decisions, weather, and fleet operations. The Commands
control system processes motion-related signals to infer the Desired Position Actuators
state of the ship, to filter disturbances, and to generate an and Velocity
Control
appropriate command for the actuators so as to reduce the Information Forces
difference between the actual and desired ship trajectories. • Task
• Operator Guidance
The controller may have multiple modes of operation System
• Weather
depending on the type of mission performed, for example, • Fleet
course keeping, positioning, and roll- and pitch-motion
Measured Position Motion
damping. For some ships and particular operations, the
and Velocity Navigation
desired control action can be delivered in several ways due to
System
overactuation, which provides increased reliability to actua-
tor faults. Thus, multiple combinations of actuator demands
FIGURE 1 The basic components of a modern ship motion-control
can yield the same control action. In these cases, the control system can be grouped into three subsystems. The navigation
system must also solve a control allocation problem based on system processes information from the motion sensors to provide
an optimization criteria [2]. The navigation system, which reliable signals related to the actual trajectory of the vessel. The
provides reliable measurements of position and heading, col- guidance system provides desired feasible trajectories based on
lects information from the various sensors, such as GPS, information related to the vessel state and mission in progress.
The control system processes the information coming from the
speed log, compass, gyros, radar, and accelerometers. The navigation and guidance system to produce forces that correct
deviations of the vessel trajectory from the desired vessel trajec-
Digital Object Identifier 10.1109/MCS.2009.934408 tory. (Picture of courtesy of Austal Ships, Australia.)

DECEMBER 2009 « IEEE CONTROL SYSTEMS MAGAZINE 33

 The requirements of the various operations performed by the vessel define
the modes of operation and the objectives of the motion-control system.

forces, which have frequency content at the sum of the wave In low-to-medium sea states, the frequency of oscillations
frequencies, are normally too high to be considered in ship of the linear wave forces do not normally affect the operational
motion control, but these forces can contribute to structural performance of the vessel. Hence, controlling only low-
vibration in the hull. For further details about wave loads frequency motion avoids correcting the motion for every single
and their effects on ship motion, see [4] and [5]. wave, which can result in unacceptable operational conditions
Like waves, wind and current induce forces due to pres- for the propulsion system due to power consumption and
sure variation on the vessel structure. Wind forces have a potential wear of the actuators. Operations that require the
mean component and a random fluctuating component due control of only the low-frequency motion include dynamic
to gusts. In ship motion control, only the mean wind forces positioning, heading autopilots, and thruster-assisted position
are compensated since the frequency of gusts is often outside mooring. Dynamic positioning refers to the use of the propul-
the bandwidth of the vessel response [2]. Current-induced sion system to regulate the horizontal position and heading of
forces affect vessels requiring positioning control and vessels the vessel. In thruster-assisted position mooring, the propul-
at mooring. These forces have low-frequency content. The sion system is used to reduce the mean loading on the mooring
variations of these forces can be induced by changes in the lines. Additional operations that require the control of only the
speed and direction of the current relative to the vessel. low-frequency motion include slow maneuvers that arise, for
Wind, current, and nonlinear wave forces that are of example, from following underwater remotely operated vehi-
interest to ship motion control are referred to as low-frequency cles. Operations that require the control of only the wave-
forces. Figure 2 shows a typical example of observed motion frequency motions include heave compensation for deploying
in marine surface vessels. The motion components derived loads at the sea floor [6] as well as ride control of passenger
from the inducing forces are thus referred to as the low- vessels, where reducing roll and pitch motion helps avoid
frequency motion and wave-frequency motion. Due to the motion sickness [3].
characteristics of ship motion depicted in Figure 2, the The control of only low-frequency motion is achieved by
objective of a ship motion-control system may be to control appropriate filtering of the wave-frequency components
only the low-frequency motion, only the wave-frequency from the position and heading measurements and estimated
motion, or the total motion, that is, the sum of the low- and velocities. This filtering is performed before the signals are
wave-frequency motions [1]–[3]. passed on to the controller as indicated in Figure 1.
Early course-keeping autopilots used a proportional con-
troller with a deadband nonlinearity. The deadband provided
14
an effect similar to wave filtering since it delivered a null con-
trol action until the control signals were large enough to be
12 outside the deadband. The amount of deadband in the auto-
10
pilot could be changed, and this setting was called “weather”
since the size of the deadband was selected by the operator
Heading Angle (°)

8 based on weather conditions [7], [8]. Other systems used

6
lowpass and notch filters, which introduced significant phase
lag, and thus performance degradation when a high gain
4 control is required. An alternative to traditional filtering con-
2
sists of using a wave-induced motion model and an observer
to separate the wave motion from the low-frequency motion.
LF Heading (Filtered)
0 For this approach, which requires a spectral factorization
Total Heading (LF+WF)
−2 model of the disturbances, the Kalman filter is the preferred
0 20 40 60 80 100 120 140 160 180 200 observer [9]. The Kalman filter can use measurements from
Time (s)
multiple sensors with different levels of accuracy to produce
estimates of the ship velocities, which are not measured in
FIGURE 2 Example of motion of a marine vessel. The total motion most ship-positioning applications.
can be thought of as the superposition of a low-frequency (LF) and
a zero-mean oscillatory wave-frequency (WF) component. This
In the remainder of this article, we focus on two opera-
plot shows the total motion and low-frequency components of the tions where only low-frequency motion control is required,
heading angle of a vessel during a course change. dynamic positioning and course keeping, and we present

34 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2009

 Historical Aspects of Ship Motion-Control Systems
Related to Autopilots and Dynamic Positioning
M odern ship motion-control systems are the result of sev- tion feedback control system formulated as a three-term con-
eral historical milestone developments. Among these de- trol law that is known today as proportional-integral-derivative
velopments are the invention of the gyroscompass, local and (PID) control [18]. This controller was motivated by observing
global positioning systems, control and estimation theory, and the way in which a helmsman steers a ship.
computer-based control systems. The early autopilot systems of Sperry and Minorsky suf-
A gyroscope consists of a spinning mass, typically a disk or fered from performance degradation due to the oscillatory
wheel, mounted on a base so that its axis can turn freely in one component of wave-induced loads and motion. Many tech-
or more directions, and thereby the spinning mass maintains its niques for wave filtering were investigated, including lowpass
orientation due to conservation of angular momentum regard- filtering, notch filtering, and deadband techniques. These
less of the rotational motion of the base [39]. The first recorded filters had limited success since a significant phase lag was
construction of a gyroscope is credited to C.A. Bohnenberger introduced, which affected the performance of the required
in 1810, while the first electrically driven gyroscope was dem- high-gain control.
onstrated in 1890 by G.M. Hopkins [36]. The electrically driven The wave filtering problem was addressed in the 1970s us-
gyroscope led to the invention of the gyrocompass, which uses ing a Kalman filter in the context of ship positioning [9]. In par-
a gyroscope to find the true or geometric North. If the spin axis ticular, local position measurements based on electromagnetic
of the gyroscope spinning mass is horizontal, the rotation of radio waves were used to develop the first ship-positioning
the Earth will produce the gyroscope to move in precession systems. This filter is based on a model of the oscillatory wave-
until spin axis is aligned with a meridian, and thus point to the induced motion of the vessel, which is considered as colored
true North. Gyrocompasses are used on ships instead of mag- noise and is combined with a model describing the motion due
netic compasses since they are not sensitive to disturbances to the force actuators. The states of the model describing the
in the magnetic field due to the steel hulls, onboard electri- oscillating motion were referred to as high-frequency states,
cal instrumentation, and local changes of the Earth magnetic whereas the states corresponding to the motion due to the ac-
field, known as magnetic variation. In 1908, H. Anschutz pat- tion of the actuators were referred to as low-frequency states.
ented the first North-seeking gyrocompass. Three years later Today, this terminology has changed to wave-frequency and
E. Sperry was granted a patent for his ballistic compass includ- low-frequency states. The Kalman filter is used to estimate
ing vertical damping, which reduces gyrocompass measure- both states, and wave-filtering is achieved by using only the
ment errors due to the acceleration of the ship. low-frequency states as feedback signals in the motion con-
Sperry constructed the first automatic ship steering mecha- troller. This filtering approach has the advantage of reducing
nism in 1911. This device was referred to as Metal Mike since measurement noise while providing estimates of the velocities
it captured much of the behavior of a skilled helmsman. Metal and rates when these are not measured.
Mike compensated for varying sea states using automatic Kalman filtering, together with the early work of Sperry and
gain adjustments and feedback of the vessel heading angle. Minorsky, laid the foundation of modern commercial dynamic
In 1922, N. Minorsky presented a detailed analysis of a posi- positioning and autopilot systems for marine vessels.

the models commonly used to design wave filtering based sway (transverse motion), and yaw (rotation about the verti-
on Kalman filters. For an account of the development of cal axis, also called heading).
these control applications, see “Historical Aspects of Ship The position of the vessel is normally measured by a dif-
Motion-Control Systems Related to Autopilots and ferential GPS, while the heading is measured by a gyro-
Dynamic Positioning.” compass. Additional types of sensors are usually available
to ensure reliability of the positioning system, namely,
MODELS FOR DYNAMIC POSITIONING inertial measurement units, hydro-acoustic position sen-
AND KALMAN FILTER DESIGNS sors, taut wires, and laser sensors. Examples of these sen-
The advent of the global positioning system (GPS) in the sors are illustrated in Figure 3.
1980s opened the possibility for the design of accurate auto-
matic systems for station-keeping (position regulation) and Mathematical Modeling of Vessel Dynamics
low-speed maneuvering of ships and offshore rigs. Vessels for Positioning Control Systems
operating in this mode rely on feedback information from To describe the motion of a ship, two reference frames are
position measurements and heading. These measurements used, a local geographical Earth-fixed frame and a body-
are used in a motion-control system that operates in the fixed frame, which is attached to the vessel. The compo-
three planar degrees of freedom, surge (forward motion), nents of the position-orientation vector h 5 3 n, e, c 4 ^ are

DECEMBER 2009 « IEEE CONTROL SYSTEMS MAGAZINE 35

 Satellite Navigation
System v (Sway) q (Pitch)
(DGPS/GLONASS) yb

Surface
Reference
System

p (Roll) r (Yaw)
Hydroacoustic
Positioning u (Surge)
System
Taut xb w (Heave)
Wire
zb
FIGURE 3 Examples of sensors for vessel positioning. For reliability,
some vessels have multiple sensors. The surface position reference FIGURE 4 Motion variables for a marine vessel based on the
systems include laser-doppler sensors and radio navigation systems. SNAME convention [32]. The linear velocities in surge, sway, and
As a satellite system, GPS and differential GPS are the most common. heave are expressed in a body-fixed frame. Similarly, the three
For vessels operating at fixed locations, such as oil rigs, it is common components of the angular velocity vector are expressed in the
to use hydroacoustic sensors and taut wires. The latter consist of a same body-fixed frame. These angular velocity components can
weight tied to a wire, where deviations of the desired position of the be related to the time-derivatives of the roll, pitch, and yaw Euler
ship are determined from the length of the wire deployed. angles using a kinematic transformation.

the north-east positions ( n, e ) of the vessel relative to the namely, r, which the rotation rate about the axis perpen-
local geographical frame and the yaw or heading angle c dicular to the horizontal plane.
relative to the north. The components of the velocity vector A model for vessel dynamics can be expressed as
n 5 3 u, v, r 4 ^ are the surge and sway velocities (u, v) and
the yaw rate r. These variables are depicted in Figure 4. The h? 5 R ( c ) n, (1)
#
body-fixed velocities u and v are the time derivatives of ( MRB 1MA ) n 1CRB ( n ) n1d ( Vrc, gc ) 5 tcontrol 1 twind 1 twaves.
the position of the origin of the body-fixed frame relative to (2)
the origin of the local geographical frame expressed in the
body-fixed frame. Similarly, the yaw rate r is a component The kinematic transformation (1) relates the body-fixed
of the angular velocity of the body-fixed frame with respect velocities to the time derivative of the positions in the local
to the local geographical frame expressed in the body-fixed geographical frame. This transformation is given by the
frame. For ship-positioning and heading control, the trans- rotation matrix
lational motion is assumed to be confined to the horizontal
plane, and thus the angular velocity has only one component,
cos c 2sin c 0
R ( c ) 5 £ sin c cos c 0§, R 21 ( c ) 5 R^ ( c ) . (3)
0 0 1
TABLE 1 Summary of ship motion variables used for
dynamic positioning and maneuvering applications.
For further details about kinematic models used in marine
Variable Name Frame Units control systems and transformations, see [10].
n North position Earth fixed m The terms on the right-hand side of (2) represent the vec-
e East position Earth fixed m tors of forces due to control, wind, and waves, respectively.
c Heading or yaw angle rad These forces are directed along the body-fixed directions
u Surge speed Body fixed m/s
v Sway speed Body fixed m/s
xb and yb, and the moment is taken about the vertical axis
r Yaw rate Body fixed rad/s zb, so that t 5 3 X, Y, N 4 T, where X is the surge force, Y is
X Surge force Body fixed N the sway force, and N is the yaw moment. Table 1 summa-
Y Sway force Body fixed N rizes this notation and indicates the reference frames in
N Yaw moment Body fixed N-m which the variables are expressed.
h 5 3 n, e, c 4 T Generalized position
n 5 3 u, v, r 4 T Generalized velocity
The positive-definite, rigid-body mass matrix MRB and
t 5 3 X, Y, N 4 T Generalized force the skew-symmetric Coriolis-centripetal matrix CRB ( n )
are given by

36 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2009

 The objective of a ship motion-control system may be to control only the
low-frequency motion, only the wave-frequency motion, or the total motion.

m 0 0 current coefficients CXc ( grc ) , CYc ( grc ) , CNc ( grc ) , [1], [2],
MRB 5 £ 0 m mxg § , that is,
0 mxg Iz
AFc CXc ( grc )
0 0 2m ( xgr 1 v ) 1
d ( Vrc, grc ) 5 rV 2rc £ ALc CYc ( grc ) § , (8)
CRB ( n ) 5 £ 0 0 mu §, (4) 2
ALc L oa CNC ( grc )
m x gr 1 v )
( 2mu 0
where r is the water density, AFc and ALc are frontal and
where xg denotes the longitudinal position of the center of lateral projected areas of the submerged part of the hull,
gravity of the vessel relative to the body-fixed frame. The and Loa is the length of the ship. Typical current coeffi-
Coriolis-centripetal terms appear as a consequence of express- cients for a dynamically positioned vessel are shown in
ing the equations of motion in body-fixed coordinates. This Figure 5. These coefficients are determined experimen-
formulation simplifies the model since, when expressed in tally based on scale models or using computational fluid
the body-fixed frame, the moments of inertia become time dynamic models [2].
invariant and the direction of the actuator forces are simpler Unless the vessel under consideration is the subject of
to describe. When a vessel operates under positioning con- an extensive hydrodynamic analysis and scale-model test-
trol, the velocities are small. Hence, the Coriolis-centripetal ing, the current coefficients CXc ( grc ) , CYc ( grc ) , CNc ( grc ) in
terms CRB ( n ) n in (2) can be ignored for control design. (8) are difficult to estimate with accuracy. In such cases, it is
When a vessel moves in the water, the changes in pres- common practice to simplify the model (8) in terms of a
sure on the hull are proportional to the velocities and accel- linear damping term and a bias term [33] of the form
erations of the vessel relative to the fluid. The coefficients
used to express the pressure-induced forces proportional
to the accelerations are called added-mass coefficients. d ( Vrc, grc ) < Dn 2 R^ ( c ) b, (9)
These forces reflect the change in momentum in the fluid
due to the vessel accelerations. The positive-definite hydro-
dynamic added mass matrix MA is represented by
0.2
2Xu# 0 0
CX

0
MA 5 £ 0 2Yv# 2Yr# § , (5)
−0.2
0 2Yr# 2Nr# 0 50 100 150 200 250 300 350 400
Relative Current Angle (°)
where the added-mass coefficients Xu# , Yv# , Yr# , Yr# , and Nr# 1
depend on the hull shape. Despite the terminology, note
CY

0
that only Xu# and Yv# have mass units.
−1
The term d ( Vrc, gc ) on the left-hand side of (2) repre- 0 50 100 150 200 250 300 350 400
sents the current and damping forces. These nonconser- Relative Current Angle (°)
vative forces, which reflect the transfer of energy from 0.5
the vessel to the fluid, depend on the speed and direc-
CN

0
tion of the current relative to the vessel is given by
−0.5
0 50 100 150 200 250 300 350 400
Vrc 5 "u2rc 1 v2rc 5 " ( u 2 uc ) 2 1 ( v 2 vc ) 2, (6) Relative Current Angle (°)
grc 5 2atan2 ( vrc, urc ) , (7)
FIGURE 5 Current coefficients of an offshore supply vessel deter-
where uc and vc are the components of the current velocity mined using computational fluid dynamics. The current coeffi-
in the vessel body-fixed frame. Thus, the angle of the cur- cients are normalized nondimensional forces in surge CX, sway
CY, and yaw CN. The normalization is based on current speed,
rent grc is defined relative to the bow of the vessel. water density, and the projected area of the vessel below the water
It is common practice to write the current forces in line. These coefficients can also be determined experimentally
surge, sway, and yaw as functions of nondimensional from scale model tests.

DECEMBER 2009 « IEEE CONTROL SYSTEMS MAGAZINE 37

 Most of the vessels operating in the offshore industry worldwide use
Kalman filters for velocity estimation and wave filtering.

where h? p 5 n, (15)
Mn? 1 Dn 5 bp 1 tcontrol 1 twind 1 twaves, (16)
?
bp 5 0, (17)
D11 0 0 b1
D 5 D^ 5 £ 0 D22 D23 § , b 5 £ b2 § . (10)
where bp J R^ ( c ) b. Since (15)–(17) is linear, a linear
0 D32 D33 b3
observer can be used to estimate the velocity n and bias
bp using the force vectors tcontrol, twind, and twaves along
The linear damping term also models the transfer of energy with hp, which is known from the measurement of h
from the vessel to the fluid due to the waves that are gener- through (14).
ated as a consequence of the vessel motion.
Using (9), the simplified vessel model (1)–(2) becomes Wind Force Models
The wind forces and moments can be represented in a sim-
h? 5 R ( c ) n, (11) ilar way to the current forces and moments by means of
Mn? 1 Dn 5 R^ ( c ) b 1 tcontrol 1 twind 1 twaves, (12) nondimensional force coefficients [2], namely,
?
b 5 0, (13)

where M 5 MRB 1 MA in (12). The bias term is constant in AFwCXw ( grw )

1
Earth-fixed coordinates, under the assumption of constant twind 5 raV rw £ ALwCYw ( grw ) § ,
2
(18)
2
or slowly varying currents. This assumption is appropriate ALwL oaCNw ( grw )
considering the tide periods relative to the ship dynamic
response characteristics. The bias term must therefore be where ra is the air density, AFw and ALw are frontal and lat-
rotated to be included into the equation of motion (12). This eral projected wind areas, and Loa is the vessel’s overall
rotation captures the effect that the current forces change length. The wind speed Vrw and direction grw relative to
with the heading of the vessel. the vessel are given by
The bias is estimated by the observer shown in Figure 1,
which allows the resulting force term in (12) to be compen- Vrw 5 "u 2rw 1 v 2rw, (19)
sated by the motion controller. This approach thus achieves grw 5 2atan2 ( vrw, urw ) , (20)
integral action for positioning control since the bias repre-
sents the low-frequency disturbances [2]. with

urw 5 u 2 Vw cos bw, (21)

Vessel Parallel Model Formulation 3
The model (11)–(13) is nonlinear due to the kinematic uw
transformations. This model can be linearized dynami- vrw 5 v 2 V
3 w sin bw. (22)
cally by introducing the vessel parallel coordinates [2], vw
which are defined in a reference frame fixed to the vessel
with axes parallel to the Earth-fixed reference frame. The Here Vw and bw are the wind speed and direction in Earth-
vessel parallel coordinates hp are thus defined by using the fixed coordinates. Typical wind coefficients CXw ( grw ) ,
transformation CYw ( grw ) , and CNw ( grw ) for an offshore vessel are shown
in Figure 6.
hp J R^ ( c ) h, (14) The wind coefficients can be obtained either from com-
putational fluid dynamics, model tests, or by scaling coef-
where hp is the position-attitude vector expressed in body ficients of similar vessels. For control design purposes,
coordinates. For dynamic positioning applications, rotation however, wind speed and direction measurements are
#
about the z-axis is often slow. Therefore, r < 0 and often used for approximate feedforward compensation,
#
R ( c ) < 0 are good approximations. Consequently, (11)– and the errors associated with this compensation are mod-
(13) can be written in terms of hp, and the resulting model, eled as a bias in (13) and (17). That is, the bias accounts for
which is referred to as the vessel parallel reference model, is the simplified current forces as well as the wind forces. If
given by feedforward compensation is not used, then a sluggish

38 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2009

 To describe the motion of a ship, two reference frames are used,
a local geographical Earth-fixed frame and a body-fixed frame,
which is attached to the vessel.

response to changes in the direction of the wind relative to the hull and H ( jv ) represents the frequency response from
the vessel may result [2]. force to motion [4], [5]. The frequency response function
F ( jv ) , which is known as the force response amplitude
Models for Wave Loads and Wave-Induced Motion operator (RAO) in naval architecture, depends on the shape
As discussed above, wave forces are usually modeled as of the hull and the angle from which the waves approach
the sum of a linear and a nonlinear component, namely, the vessel. The frequency response H ( jv ) , known as the
force-to-motion RAO, depends on the shape of the hull and
twaves 5 tlin
waves 1 twaves.
nlin
(23) the mass distribution. These frequency response functions,
which can be computed using hydrodynamic computer
The linear oscillatory component has the same frequency codes, are the basis of ship motion evaluation [3].
as the wave elevation. The nonlinear component has both The spectrum of the wave-induced motion can be
lower and higher frequency than the wave elevation. The approximated using spectral factorization. That is, (25) can
linear and low-frequency nonlinear wave forces are rele- be approximated by the spectrum of the signal obtained
vant to ship motion control. using a linear filter driven by Gaussian noise w ( t ) with a
The low-frequency nonlinear wave forces are treated as flat spectrum Sww over the frequencies of interest. Then,
an input disturbance and modeled by a bias term (17). That
is, the bias term represents a combination of nonlinear Shwhw ( v ) < |G ( jv ) |2Sww . (26)
wave and current. The linear wave forces, however, are
usually transformed into an equivalent output disturbance. A second-order, noise-driven filter is usually appropriate
With this point of view, the measured position vector can for modeling wave-induced motion [1]–[3], namely,
be represented as

htot 5 h 1 hw, (24)

1
where htot is the total position, hw is wave-induced posi- 0.5
CX

tion due to linear wave forces, and h represents the low- 0

frequency position due to control, wind, current, and −0.5
0 50 100 150 200 250 300 350 400
nonlinear wave forces. Relative Wind Angle (°)
The sea surface elevation is typically modeled as the 1
realization of a stationary Gaussian random process [34].
CY

0
Stationarity can be considered for short periods between 20
min and 3 h depending on weather conditions, while −1
0 50 100 150 200 250 300 350 400
Gaussianity depends on the wave height and depth [35]. Relative Wind Angle (°)
The deeper the ocean at the location considered, the higher 0.2
the waves for which Gaussianity can be assumed [34], [35].
CN

0
Under these conditions, the wave elevation is completely
characterized by the wave spectrum Szz ( v ) , which changes −0.2
0 50 100 150 200 250 300 350 400
with the sea state, namely, the dominant amplitudes and Relative Wind Angle (°)
wave periods.
Under a linear response assumption, the wave-induced FIGURE 6 Wind coefficients of an offshore supply vessel deter-
motion spectrum can be represented as mined using computational fluid dynamics. The wind coefficients
are normalized forces in surge CX, sway CY, and yaw CN. The
Shwhw ( v ) 5 |H ( jv ) F ( jv )|2 Szz ( v ) , (25) normalization is done in terms of wind speed, air density, and the
projected area of the vessel above the waterline. These coeffi-
cients are determined experimentally from scale model tests for
where F ( jv ) represents the frequency response function multiple wind directions measured by the wind angle grw relative
from the wave elevation to the pressure-induced forces on to the bow of the ship.

DECEMBER 2009 « IEEE CONTROL SYSTEMS MAGAZINE 39

 v 20 s the actuator. For example, if the command to a propeller is
G(s) 5 2
. (27) the rotation speed, then the corresponding coefficient in Bu
s 1 2zv 0 s 1 v 20
maps the speed to the generated thrust.
The resulting model for a dynamic positioning observer
The parameters of the shaping filter (27) as well as the design is the 15th-order state-model
intensity Sww of the noise depend on the sea state, the
vessel shape, and the angle from which the waves approach x? 5 Ax 1 Bu 1 Ew, (36)
the vessel. y 5 Hx 1 v, (37)
A filter transfer function of the form (27) is considered
for each degree of freedom. The combination of these where x 5 3 j^, h^ p , bp , n 4 [ R is the state vector, u [ R
^ ^ ^ 15 p

transfer functions can be transformed into a single state- ( p $ 3 ) is the control vector, w 5 3 w^ ^
1 , w2 , w3 4
^ ^
[ R rep-
9

space model resents the process noise vector, and

?
j 5 Awj 1 Eww, (28) Aw 0633 0633 0633 0633
hw 5 Cwj, (29) 0336 I333 0333 0333 0
A5 ≥ ¥ , B 5 ≥ 333 ¥ , (38)
0336 0333 0333 0333 0333
where the matrices Aw, Bw, and Cw depend on the chosen 0336 2M21D M21 0333 M21Bu
state-space realization adopted. Since a model for dynamic
positioning has three degrees of freedom (surge, sway, and Ew
yaw), a second-order noise filter approximation yields a 0333
H 5 3 Cw I333 0333 0333 4 , E5 ≥ ¥. (39)
state vector j with six components. The state vector j may I333
or may not have a physical interpretation depending on the M21
particular state-space realization used.
KALMAN FILTER DESIGN FOR
Linear Model for Dynamic Positioning Observer Design SHIP DYNAMIC POSITIONING
By combining the models of the vessel response together The model given in (36) forms the basis of a Kalman filter
with the disturbance models discussed above, we obtain a design. To implement the observer on a computer, the model
model structure of the form needs to be discretized. The discretization has the form
#
j 5 Awj 1 Eww1, (30) x ( k 1 1 ) 5 Fx ( k ) 1 Du ( k ) 1 Gw ( k ) , (40)
?
hp 5 n, (31) y ( k ) 5 Hx ( k ) 1 v ( k ) , (41)
Mn? 1 Dn 5 bp 1 tcontrol 1 twind 1 w2, (32)
?
bp 5 w3, (33) where
y 5 hp 1 Cwj 1 v. (34)
F 5 exp ( Ah ) , (42)
The variables wi, i 5 1, 2, 3, are Gaussian noise vectors, D 5 A21 ( F 2 I ) B, (43)
which represent model uncertainty. Note that no uncertainty G 5 A21 ( F 2 I ) E, (44)
is associated with the kinematic equation (31) since this rela-
tion describes known geometrical aspects of motion. The h is the sampling time, and the equivalent discrete-time
measurement positions (34) are the sum of the wave motion noises w ( k ) and v ( k ) are Gaussian and white with zero
Cwj and the motion hp due to wind forces, current, and con- mean. Appropriate sampling times can be determined from
trol. The measurement vector also contain some noise v. The step responses in the various degrees of freedom of the vessel.
control forces usually have two components As a rule of thumb, the sampling period is chosen to be in the
range of 20–40 samples within the rise time of the force to
t 5 2 t^ wind 1 Buu, (35) velocity response of the fastest degree of freedom. For large
offshore vessels and rigs, the sampling time is normally in
where t^ wind is an estimate of the wind forces implemented the range of 100–500 ms. Small vessels with faster dynamics
by using feedforward compensation and Buu represents may require sampling times as low as 50 ms.
actuator forces. The wind feedforward term, which is pro- To implement a Kalman filter, the parameters of the
portional to the square of the measured wind velocity, model (40), (41) as well as the covariance of the state mea-
depends on the vessel’s projected area in the direction of surement noises in the model are necessary. The mass and
the wind [2]. The vector u is the command to the actuators, damping parameters of the model can be initially estimated
which are assumed to have much faster dynamic response from hydrodynamic computations. Then, an update of the
than the vessel; thus the coefficient Bu represents the map- parameter estimates can be based on data obtained from
ping from the actuator command to the force generated by tests performed in calm water [27]. The structure of the

40 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2009

model matrices (38), (39) determines the structure of the N

matrices of the discrete-time model (40), (41). Then, for u^ 5 arg min det a ` ( k, u ) ` ( k, u ) T (48)
u k51
example, standard grey-box, state-space prediction-error
methods can be used to estimate the parameters [12]. with
The measurement noise is usually associated with sen-
sors. After correcting the sensor bias, the covariance s2vi of the e ( k, u ) 5 h^ pw ( k ) 2 Cw ( u ) j^ ( k ) , (49)
measurement noise of the sensor i can be estimated by the j^ ( k 1 1 ) 5 Aw ( u ) j^ ( k ) 1 Lw ( u ) e ( k, u ) , (50)
sample covariance from a data record taken while the vessel
is at port with no motion. Since the noise of different sensors where hpw ( k ) is replaced by the estimate h^ pw ( k ) obtained
is often uncorrelated, the covariance matrix of the vector of from detrending the measured data. Equations (48)–(50)
measurement noise is chosen to be diagonal, that is, comprise a standard prediction error estimation problem
whose solution is related to the maximum likelihood esti-
R 5 diag ( s2v1, s2v2 , c, s2vp ) . mate of the parameter vector u [11], [12].
Once the parameters of the mode are estimated, the
The estimation of the covariance Q of the state noise w in covariance Q ^ P of the innovations can also be estimated
(40) is more complex since it depends on the sea state, the from the sample covarianace of the predictions errors.
heading of the vessel relative to the environmental distur- Then, the Kalman filter can be implemented with the inno-
bances, and how uncertain the model (40), (41) is [3]. This vation wave-frequency model, and thus we can chose
covariance matrix is chosen to be block diagonal, that is, Q1 5 Q ^ P. This choice entails no loss of information.
An alternative to the procedure described above consists
Q 5 diag ( Q1, Q2, Q3 ) . (45) of fixing the damping of the filters (27) to a value in the range
0.01 to 0.1 as suggested in [13] and estimate only the natural
The matrix Q1 [ R333 is the covariance of the noise w1, frequency and noise covariance [13], [14]. This estimation
which drives the noise filter representing linear wave-in- approach is summarized in [1], where recursive least squares
duced motion, Q2 [ R333 is the covariance of the noise w2, is used for parameter estimation. A related approach, also
which represents the uncertainty in the equation of motion, based on recursive least squares, is given in [3].
and Q3 [ R333 is the covariance of the noise w3, which From the procedures discussed above, all the parame-
represents the uncertainty in the bias term that models the ters necessary to implement the Kalman filter for a dynam-
rest of the environmental forces, as indicated in (30)–(33). ically positioned vessel are obtained. The Kalman filter can
The matrices Q2 and Q3 are chosen to be diagonal. The be formulated as follows [16]. The state estimate or correc-
entries of the matrix Q2 are taken as a fraction of the vari- tion is given by
ance of the position measurement noises. The entries of Q3
x^ ( k ) 5 x ( k ) 1 K ( k ) 3 y ( k ) 2 H x ( k ) 4 , (51)
are high values. These choices provide a filter with an
appropriate balance of the uncertainty in various parts of while the state prediction is given by
the model. The covariance Q1 is estimated together with the
parameters of the wave-frequency motion model (27) from x ( k 1 1 ) 5 F x^ ( k ) 1 D u ( k ) . (52)
data measured before and during the operation of the
vessel. The parameters are re-estimated after significant The Kalman gain K ( k ) is computed as
changes in heading or at regular time intervals of 20 min,
which is the time period for which the sea state can be con- K ( k ) 5 P ( k ) HT ( HP ( k ) HT 1 R ) 21, (53)
sidered to be stationary [3], [34], [35]. Since the vessel is in a
positioning control mode, the total motion measured can be where P ( k ) satisfies the recursion
recorded and detrended to obtain an estimate of the wave-
induced motion vector h^ pw ( k ) , or, equivalently, a first-order P ( k ) 5 ( I 2 K ( k ) H ) 21P ( k ) ,
highpass filter can be used [14]. These data can then be used P k 1 1 ) 5 FP ( k ) FT 1 Q.
( (54)
to estimate the parameters of the wave-induced motion
model, for which it is convenient to consider the directly The matrices P ( k ) and P ( k ) are the conditional covari-
parameterized innovations form [12] ances of the estimation errors x ( k ) 2 x^ ( k ) and x ( k ) 2x ( k ) ,
respectively.
j^ ( k 1 1 ) 5 Aw ( u ) j^ ( k ) 1 Lw ( u ) e ( k ) , (46) The algorithm (51)–(54) is a standard Kalman filter.
hpw ( k ) 5 Cw ( u ) j^ ( k ) 1 e ( k ) , (47) The recursive form (54) may not be the best choice for
real-time implementation since numerical issues may be
where u is the vector of parameters to be estimated, and associated with matrix inversion and covariance matrix
e ( k ) is the vector of innovations. The parameter estimation propagation. These errors could result in covariance
can then be formulated as matrices that are not positive semidefinite. To address

DECEMBER 2009 « IEEE CONTROL SYSTEMS MAGAZINE 41

these issues, various implementations of the equations are ulation, the wave filter is switched on 500 s after the vessel
used to propagate and update the estimates and the cova- is under positioning control. Then, at 600 s the vessel posi-
riance. These implementations, known as square root fil- tion is changed 10 m forward. Figure 7(a) shows the mea-
tering, are algebraically equivalent to (54), but exhibit sured and wave-filtered surge position. Figure 7(b) shows
improved numerical precision and stability. For details the measured and wave-filtered surge velocity. Figure 7(c)
see [16] and [17]. shows the force generated by the controller.
The filter (51)–(54) gives the state estimate During the first 600 s, while the wave filter is switched
off, the wave-indued motion produces significant control
x^ ( k ) 5 3 j^ ( k )^, h^ p ( k )^, b^ p ( k )^, n^ ( k )^ 4 ^. (55) action. Once the wave-filter is switched on, the control
action at wave-frequencies is reduced.
The components h^ p ( k ) and n^ ( k ) , which are the low-fre-
quency motion components, are used by the controller, and Observer Based on Nonlinear Models
in this way, the wave filtering is achieved. That is, the A more advanced implementation of the observer results
Kalman filter uses a model of the wave-frequency motion by using the nonlinear vessel model (1), (2), which includes
that facilitates estimation of the low- and wave- frequency the quadratic damping forces and current loads. This model
components of the position and heading measurements. can give better accuracy, but the resulting observer is com-
The bias estimate provided by the Kalman filter can be putationally more intensive, and the parameters of the non-
used in the controller to implement integral action unless linear model are more difficult to estimate. The filtering
the controller has an integrator; see [2] for further details problem for this case can be solved using the extended
on dynamic positioning control design. Kalman filter. An alternative to the extended Kalman filter
is to use a passivation design and Lyapunov theory. This
Example of Kalman Filter Performance approach is demonstrated in [33].
for a Vessel Under Positioning Control
Figure 7 shows simulation data of a Kalman wave filter on KALMAN FILTER DESIGN FOR SHIP
a 15-m fishing vessel under positioning control. In this sim- COURSE-KEEPING AUTOPILOTS
Wave filtering for positioning systems must be imple-
mented in three degrees of freedom, surge, sway, and yaw.
When designing ship autopilots for automatic heading con-
Surge Position (m)

15 trol, it is only necessary to consider a model and perform

10 Measured
wave filtering for the yaw degree of freedom [1], [2], [8],
Filtered
5 [15]. In this section, we discuss the type of models that are
0
normally used to implement observers for autopilot wave
−5
300 350 400 450 500 550 600 650 700 750 800 filtering and how the Kalman filter is used. For a discus-
Time (s) sion on the development of ship autopilots, see “Historical
(a) Aspects of Ship Motion-Control Systems Related to Autopi-
Surge Control Force (N) Surge Velocity (m/s)

1 lots and Dynamic Positioning.”

0.5
0 The Nomoto Model for Ship Heading Response
−0.5 Measured Autopilots are used to correct deviations from a desired
Filtered
–1
300 350 400 450 500 550 600 650 700 750 800 equilibrium heading, and thus a linear model is sufficient
Time (s) for control design [15]. The response in yaw rate due to a
(b) small deviation in the angle of a control actuator, such as
× 104
1 a rudder or the steering nozzle of a water jet, can be
0.5 derived from (2) by isolating the yaw motion, which is
0 given by
Demand
−0.5
Actual
–1 #
300 350 400 450 500 550 600 650 700 750 800 ( Iz 2 Nr# ) r 2 Nrr 5 Ndd, (56)
Time (s)
(c) where Iz is the moment of inertia in yaw, Nr# , Nr, and Nd are
hydrodynamic coefficients, r is the yaw rate, and d is the
FIGURE 7 Simulated data of Kalman-based-wave filter perfor- actuator angle. This model, which is known as the first-order
mance for a 15-m fishing vessel under positioning control. The Nomoto model [38], can be written as the transfer function
wave filter is switched on at 500 s, and the vessel is moved for-
ward by 10 m at 600 s. (a) shows the measured and wave-filtered
surge position. (b) shows the measured and wave-filtered surge r(s) K
5 . (57)
velocity. (c) shows the surge force generated by the controller. d ( s ) 1 1 Ts

42 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2009

 When a vessel moves in the water, the changes in pressure on the
hull are proportional to the velocities and accelerations
of the vessel relative to the fluid.

The time constant and low-frequency gain are given by 1

t^ wind 5 raV 2rw ALw L oa CNw ( grw ) . (66)
2
Iz 2 Nr#
T5 , (58)
2Nr
The uncertainty in (66) has a low-frequency content, which is
Nd
K5 2 , (59) compensated by the integral action of the controller [2]. The
Nr
rudder command is computed from the control input tN as
which can be estimated from trials in calm water.
1
Using the motion superposition assumption, as in the case d5 t . (67)
Nd N
of positioning control design, we model low-frequency envi-
ronmental disturbances with a bias moment term in the equa- To implement the control law, both c and r are needed.
tion of motion. Then, the state-space model can be written as Most ships have only compass measurements c, and thus
the turning rate r must be estimated. In addition, it is neces-
#
c 5 r, (60) sary to perform wave filtering such that the oscillatory
# 1 1 wave-induced motions are avoided in the feedback loop.
r 5 2 r 1 tN 1 b, (61)
# T m
b 5 0, (62) Models for Course Autopilots Observer Design
As in the case of positioning, we consider the first-order,
where m 5 Iz 2 Nr# and wave-induced motion as an output disturbance. Hence the
K measured yaw angle can be decomposed into
tN 5 m d 5 N d d (63)
T
denotes the control yaw moment. ctot 5 c 1 cw, (68)

Autopilot Design where c is the response due to the control action and low-
For autopilot control design, it is common to design a pro- frequency disturbance and where cw represents the first-
portional-integral-derivative controller with feedforward order wave-induced motion.
from wind and a smooth time-varying reference signal To estimate of c and r, and thus wave filtering, the
cd ( t ) according to model used for a Kalman filter design is given by

# 1
#
tN ( s ) 5 2t^ wind 1 mard 2 rd b
T j w 5 c w, (69)
5 #
tFF cw 5 2v 20 jw 2 2zv 0 cw 1 w1, (70)
#
t c 5 r, (71)
|1K |
2maKpc | #
d r 1 Ki3 c ( t ) dtb, (64) 1 1
8 0 r^ 5 2 r 1 ( twind 1 tN ) 1 b 1 w2, (72)
# T m
tPID b 5 w 3, (73)
where tN is the controller yaw moment and tFF is a feedfor-
#
ward term using the reference signal rd 5 cd. The heading where v 0 and z are the parameters of the filter (27) used to
|
and yaw rate errors are denoted by c 5 c 2 cd and |r 5 r 2 rd, represent the wave-induced yaw motion, b is the bias term,
respectively. The control gains Kp, Kd, and Ki must be chosen and w1, w2, and w3 are Gaussian white noises. The mea-
such that the third-order linear error dynamics surement equation is

y 5 c 1 cw 1 v, (74)
|# 1 |1K c
t
| ( t ) dt 5 0
r 1 a 1 Kd b |
r 1 Kp c i3 (65)
T 0 where v represents zero-mean Gaussian white measure-
ment noise introduced by the sensor. Notice that neither
is asymtotically stable. Notice that the control law (64) the yaw rate r nor the wave states jw and cw are measured.
depends on wind feedforward, where The resulting state-space model is

DECEMBER 2009 « IEEE CONTROL SYSTEMS MAGAZINE 43

 x? 5 Ax 1 Bu 1 Ew, (75) To implement a Kalman filter for a course autopilot, the
y 5 Hx 1 v, (76) model (75) can be discretized as in the case of dynamic posi-
tioning, and the parameters T and m of the response from
where the control forces can be estimated from tests performed in
calm water. Then, the parameters v 0 and z of the first-order,
x 5 3 jw, cw, c, r, b 4 ^, (77) wave-induced model and the covariance of the driving
u 5 twind 1 tN, (78) noise w1 can be estimated from data collected during the
w 5 3 w1, w2, w3 4 ^, (79) operation of the vessel [3]. The procedure for the parameter
estimation is similar to that discussed for dynamic position-
and ing. The only significant difference is that the parameters of
the first-order, wave-induced model must be re-estimated
0 1 0 0 0 0 not only if significant heading changes occur but also if sig-
2v 20 22zv 0 0 0 0 0 nificant speed changes occur. Changes in the speed of the
A5E 0 0 0 1 0U , b5E 0 U , (80) vessel produce a Doppler effect, and the frequencies of the
0 0 0 2T21 1 m21 waves seen from the vessel change with both the speed and
0 0 0 0 0 0 direction of the waves relative to the vessel [3].

0 0 0 Example Kalman Filter Design

1 0 0 for Course-Keeping Autopilot
E 5 EE0
0 0 0U ,
0U h^ 5 3 0, 1, 1, 0, 0 4 . (81) To illustrate the performance of Kalman-filter-based wave fil-
0 1 0 tering, we consider the case of an autopilot application taken
0 0 1 from the Marine Systems Simulator (MSS) [40]. This simulation
package implemented in Matlab-Simulink provides models of
vessels and a library of Simulink blocks for heading autopi-
lot control system design and blocks for a Kalman-filter-based
40
Heading (°)

wave filter from heading-only measurements.

30
20
The vessel considered is a 160-m mariner class vessel
10 with a nominal service speed of 15 kn, or 7.7 m/s. The
0 parameters of a complete and validated nonlinear model of
0 50 100 150 200 250
Time (s) the form (1)–(2) are given in [1]. From the step tests per-
(a) formed on the nonlinear model, a first-order Nomoto model
Low−Frequency Motion KF Estimate
Heading Rate (°/s)

0.6
0.4
0.2 40
Heading (°)

0 30
−0.2 20
0 50 100 150 200 250
10 Vessel
Time (s)
(b) 0 Desired
−10
Low−Frequency Motion KF Estimate 0 50 100 150 200 250
4 Time (s)
Heading (°)

2 (a)
Rudder Angle (°)

0 15
−2
−4 10
0 50 100 150 200 250 5
Time (s)
(c) 0

Wave−Frequency Motion −5
KF Estimate 0 50 100 150 200 250
Time (s)
(b)
FIGURE 8 Performance of a Kalman filter for heading autopilot for
a mariner cargo ship. (a) shows the true low-frequency heading c
and its Kalman filter estimate c^ . (b) shows the true low-frequency FIGURE 9 Performance of a heading autopilot for a mariner cargo
heading rate r and its Kalman filter estimate r^. (c) shows the ship with a wave filter. (a) shows the time series of the desired
wave-frequency component of the heading cw and its Kalman heading cd and the actual vessel heading c. (b) shows the rudder
filter estimate c^ w. The Kalman filter, which uses measurements of angle d. Because only the low-frequency estimates of the heading
the sum of low- and wave-frequency heading, estimates these angle and rate produced by the Kalman filter are passed to the
states and the rate using a model of the vessel and a model of the controller, the motion of the rudder does not respond to the wave
wave-induced motion. motion, and thus wave filtering is achieved.

44 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2009

 The objective of a positioning and heading control system is regulation while
compensating only for the low-frequency environmental forces.

(57) is identified with the parameters K 5 0.185 s21 and associated with wave-frequency motion. Once these states
T 5 107.3 s. Based on the time constant, a sampling period are estimated by the Kalman filter, only the low-frequency
of 0.5 s is chosen for the implementation of the Kalman states are used as feedback signals in the controller.
filter. The standard deviation of the noise of the compass In this article, we have described the main components
sensor is 0.5°. From a record of heading motion while the of a ship motion-control system and two particular motion-
rudder is kept constant, the parameters of the first-order, control problems that require wave filtering, namely,
wave-induced motion model are estimated, namely, z 5 dynamic positioning and heading autopilot. Then, we dis-
0.1, v 0 5 1.2 rad/s, and the standard deviation of the noise cussed the models commonly used for vessel response and
driving the filter is sw1 5"300 rad/s. showed how these models are used for Kalman filter
Using the above data, a Kalman filter is designed. Fig- design. We also briefly discussed parameter and noise
ures 8 and 9 show the performance of the Kalman filter. covariance estimation, which are used for filter tuning. To
Figure 8(a) and (b) shows the true low-frequency heading illustrate the performance, a case study based on numerical
angle and rate together with the Kalman filter estimates. simulations for a ship autopilot was considered.
Figure 8(c) shows the first-order, wave-induced heading The material discussed in this article conforms to
angle component and its estimate. Figure 9 shows the per- modern commercially available ship motion-control
formance of the control loop. Figure 9(a) shows the desired systems. Most of the vessels operating in the offshore
and the actual heading angle of the controlled vessel. Figure industry worldwide use Kalman filters for velocity esti-
9(b) depicts the rudder angle. In this figure, we can appreci- mation and wave filtering. Thus, the article provides an
ate the effect of the wave filtering since the rudder angle has up-to-date tutorial and overview of Kalman-filter-based
no motion at the wave frequency. wave filtering.

CONCLUSIONS AUTHOR INFORMATION

Ships and oil rigs perform operations that require accurate Thor I. Fossen received the M.Sc. degree in naval architecture
positioning and heading control. Vessel motion is affected in 1987 from the Norwegian University of Science and Tech-
by environmental forces, which can be separated into low- nology (NTNU) and the Ph.D. in control systems from NTNU
and wave-frequency forces. The low-frequency forces are in 1991. In 1989–1990, he was a Fulbright Scholar in aerody-
caused by wind, current, and wave forces that depend non- namics and flight control at the Department of Aeronautics
linearly on the wave elevation. The wave-frequency forces and Astronautics at the University of Washington, Seattle. In
are caused by oscillatory pressure changes that depend lin- 1993, he was appointed professor of guidance and control at
early on the wave elevation. NTNU. He is the author of Guidance and Control of Ocean Ve-
The objective of a positioning and heading control system hicles and Marine Control Systems and coauthor of New Direc-
is regulation while compensating only for the low-frequency tions in Nonlinear Observer Design. He has contributed to the
environmental forces. This mode of operation results in development of several industrial autopilot and dynamic
energy savings and prevents actuator wear. For example, in positioning systems. He is a founder of Marine Cybernetics,
a 24-h positioning offshore operation, the propulsion system which offers services for hardware-in-the-loop testing of ma-
must not compensate for every single wave but only for the rine control systems. He has also been involved in the design
low-frequency disturbances. The implementation of such of the SeaLaunch trim and heel correction systems. His work
control systems often requires estimates of velocities from on weather optimal positioning control for marine vessels re-
measurements of position and heading, while also filtering ceived the Automatica Prize Paper Award in 2002.
the oscillatory components of motion due to the waves. This Tristan Perez ([email protected]) com-
type of filtering, known as wave filtering, is a key aspect in pleted his electronics engineering degree at the National
the motion control of marine surface vessels. University of Rosario, Argentina, in 1999 and the Ph.D. in
An effective way to address both velocity estimation control engineering at the University of Newcastle, Austra-
and wave filtering is by augmenting the vessel model with lia, in 2003. In 2002, he worked as a naval architect at ADI-
a model of the oscillatory wave-induced motion and using Limited Australia, where he evaluated ship motion perfor-
a Kalman filter to estimate the states of the various parts of mance in waves. In 2003–2004 he was a research fellow at the
the model. That is, the model has state variables associated Mechatronics Research Centre, University of Wales, United
with the low-frequency motion and other state variables Kingdom, where he worked on control and fault detection

DECEMBER 2009 « IEEE CONTROL SYSTEMS MAGAZINE 45

of thrusters for underwater vehicles. From 2004 to 2007, he [18] N. Minorsky, “Directional stability of automatic steered bodies,” J.
Amer. Soc. Naval Eng., vol. 34, no. 2, pp. 280–309, 1922.
was a senior researcher at the Centre for Ships and Ocean
Structures (CeSOS), Trondheim, Norway, and a lecturer in [19] J. S. Sargent and P. N. Cowgill, “Design considerations for dynamically
positioned utility vessels,” in Proc. 8th Offshore Technology Conf., Dallas,
marine control systems at the Norwegian University of Sci- 1976, pp. 13–19.
ence and Technology (NTNU). Since October 2007, he has [20] M. J. Morgan, Ed., Dynamic Positioning of Offshore Vessels. Tulsa, OK:
been with the Centre for Complex Dynamic Systems and Petroage Pub, June 1978.
Control at the University of Newcastle, Australia, where he [21] J. G. Balchen, N. A. Jenssen, and S. Sælid, “Dynamic positioning of
leads a research program on performance optimization of floating vessels based on Kalman filtering and optimal control,” in Proc.
19th IEEE Conf. Decision and Control, New York, NY, 1980, pp. 852–864.
marine control systems. In 2008, he was appointed adjunct
[22] J. G. Balchen, N. A. Jenssen, E. Mathisen, and S. Sælid. (1980). Dynamic
associate professor of ship dynamics at NTNU, Norway.
positioning system based on Kalman filtering and optimal control. Model.
He is the author of Ship Motion Control, a member of the Identification Contr. [Online], 1(3), pp. 135–163. Available: http://www.mic-
IFAC Technical Committee on Marine Control Systems, journal.no/micarchives.asp
and a member of the Royal Institution of Naval Architects, [23] M. J. Grimble, R. J. Patton, and D. A. Wise, “The design of dynamic po-
sitioning control systems using stochastic optimal control theory,” Optimal
United Kingdom. He provides consultancy services for the
Contr. Appl. Methods, vol. 1, pp. 167–202, 1980.
marine industry and has collaborated on the development
[24] M. J. Grimble, R. J. Patton, and D. A. Wise, “Use of Kalman filtering
of commercial ship autopilots, positioning systems, and techniques in dynamic ship positioning systems,” Proc. Inst. Elec. Eng., vol.
ride control. He can be contacted at the Centre for Complex 127, no. 3, pp. 93–102, 1980.
Dynamic Systems and Control (CDSC), The University of [25] P. T. K. Fung and M. J. Grimble, “Dynamic ship positioning using a
Newcastle, Callaghan, NSW 2308, Australia. self tuning Kalman filter,” IEEE Trans. Automat. Contr., vol. 28, no. 3, pp.
339–349, 1983.
[26] S. Sælid, N. A. Jenssen, and J. G. Balchen, “Design and analysis of a dy-
REFERENCES namic positioning system based on Kalman filtering and optimal control,”
[1] T. I. Fossen, Guidance and Control of Ocean Vehicles. New York: Wiley, 1994. IEEE Trans. Automat. Contr., vol. 28, no. 3, pp. 331–339, 1983.
[2] T. I. Fossen, Marine Control Systems: Guidance, Navigation and Control of Ships, [27] T. I. Fossen, S. I. Sagatun, and A. J. Sørensen, “Identification of dynami-
Rigs and Underwater Vehicles. Trondheim, Norway: Marine Cybernetics, 2002. cally positioned ships,” Contr. Eng. Pract., vol. 4, no. 3, pp. 369–376, 1996.
[3] T. Perez, Ship Motion Control: Course Keeping and Roll Reduction Using Rudder [28] T. I. Fossen and Å. Grøvlen, “Nonlinear output feedback control of dy-
and Fins (Advances in Industrial Control Series). London: Springer-Verlag, 2005. namically positioned ships using vectorial observer backstepping,” IEEE
[4] J. N. Newman, Marine Hydrodynamics. Cambridge, MA: MIT Press, 1977. Trans. Contr. Syst. Technol., vol. 6, no. 1, pp. 121–128, 1998.
[5] O. M. Faltinsen, Sea Loads on Ships and Offshore Structures. Cambridge, [29] T. I. Fossen and J. P. Strand. (1999). A tutorial on nonlinear backstep-
U.K.: Cambridge Univ. Press, 1990. ping: Applications to ship control. Model. Identification Contr. [Online],
[6] T. Perez and P. Steinmann, “Modelling and performance of an active 20(2), pp. 83–135. Available: http://www.mic-journal.no/micarchives.asp
heave compensator for offshore operations,” in Proc. IFAC Conf. Control Ap- [30] A. J. Sørensen, S. I. Sagatun, and T. I. Fossen, “Design of a dynamic
plications in Marine Systems (CAMS), Bol, Croatia, Sept. 2007, pp. 30–36. positioning system using model based control,” J. Contr. Eng. Pract., vol. 4,
[7] R. Taggart, “Anomalous behaviour of merchant ship steering systems,” no. 3, pp. 359–368, 1996.
SNAME Trans. Marine Technol., vol. 7, no. 2, pp. 205–215, Apr. 1970. [31] A. J. Sørensen, T. I. Fossen, and J. P. Strand, “Design of controllers for
[8] J. van Amerongen. (1982). Adaptive Steering of Ships [Online]. Ph.D. Dis- positioning of marine vessels,” in The Ocean Engineering Handbook, F. El-
sertation, Delft Univ. Technology, The Netherlands. Available: http:// Hawary, Ed. Boca Raton, FL: CRC, 2000, ch. 3, pp. 207–218.
www.ce.utwente.nl/amn [32] SNAME, “Nomenclature for treating the motion of a submerged body
[9] J. G. Balchen, N. A. Jenssen, and S. Sælid, “Dynamic positioning using through a fluid,” in Technical and Research Bulletin No. 1–5. New York:
Kalman filtering and optimal control theory,” in Proc. IFAC/IFIP Symp. Au- The Society of Naval Architects and Marine Engineers (SNAME), 1950,
tomation in Offshore Oil Field Operation, Bergen, Norway, 1976, pp. 183–186. pp. 1–15.
[10] T. Perez and T. I. Fossen. (2007). Kinematic models for manoeuvring [33] T. I. Fossen and J. P. Strand, “Passive nonlinear observer design for
and seakeeping of marine vessels. Model. Identification Contr. [Online]. ships using Lyapunov methods: Experimental results with a supply ves-
(Bulletin of the Norwegian Society of Automatic Control),28(1), pp. 19–31. sel,” Automatica, vol. 35, no. 1, pp. 3–16, 1999.
Available: http://www.mic-journal.no/micarchives.asp
[34] M. Ochi, Ocean Waves: The Stochastic Approach (Ocean Technology Se-
[11] A. Harvey, Forecasting Structural Time Series Models and the Kalman Filter. ries). Cambridge, U.K.: Cambridge Univ. Press, 1998.
Cambridge, U.K.: Cambridge Univ. Press, 1989.
[35] S. Haverre and T. Moan. “On some uncertainties related to short term
[12] L. Ljung, System Identification: Theory for the User. Englewood Cliffs, NJ:
stochastic modeling of ocean waves,” in Probabilistic Offshore Mechanics,
Prentice-Hall, 1999.
Progress in Engineering Science. Southampton, U.K.: CML, 1985.
[13] T. Holzhüter, “On robustness of course keeping autopilots,” in Proc.
[36] T. Allensworth. A Short History of Sperry Marine (2009) [Online].
IFAC Workshop on Control Applications in Marine Systems, Genova, Italy, 1992,
Available: http://www.sperrymarine.northropgrumman.com/Company-
pp. 235–244.
Information/Corporate-History/Sperry-History/
[14] T. Holzhüter, “A commercial adaptive autopilot for ships: Design and
experimental experience,” in Proc. 10th IFAC World Congress, Munich, Ger- [37] S. Bennett, “Ship stabilization: history,” in Concise Encyclopedia of Traf-
many, 1987, pp. 226–230. fic and Transportation Systems, M. Papageorgiou, Ed. New York: Pergamon,
1991, pp. 454–459.
[15] D. Clarke, “The foundations of steering and manoeuvring,” in Proc. IFAC
Conf. Control Applications, Plenary Address, Girona, Spain, 2003, pp. 10–25. [38] K. Nomoto, T. Taguchi, K. Honda, and S. Hirano, “On the steering qual-
ities of ships,” Int. Shipbuilding Prog., vol. 4, no. 35, pp. 354–370, 1957.
[16] P. Maybeck, Stochastic Models, Estimation and Control, vols. I–III. New
York: Academic, 1979. [39] R. N. Arnold and L. Maunder. Gyrodynamics and Its Engineering Applica-
[17] T. Kailath, A. H. Sayed, and B. Hassibi, Linear Estimation (Prentice Hall tions. New York: Academic, 1961.
Information and System Sciences Series). Englewood Cliffs, NJ: Prentice- [40] Marine Systems Simulator (2009) [Online]. Available: www.
Hall, 2000. marinecontrol.org

46 IEEE CONTROL SYSTEMS MAGAZINE » DECEMBER 2009