A Novel Methodology for Robust Dynamic Positioning

of Marine Vessels: Theory and Experiments*

Vahid Hassani1 , Asgeir J. Sørensen2 and António M. Pascoal3

Abstract— The paper describes a novel robust adaptive con- crane and heavy lift vessels, geological survey vessels, and
troller for Dynamic Positioning (DP) of marine vessels. The multi-purpose supply and intervention vessels.
proposed Robust Multiple Model Adaptive Dynamic Positioning Early dynamic positioning systems were implemented
(RMMADP) structure consists of a bank of robust controllers
designed using the Mixed-µ methodology and an identification using PID controllers. In order to restrain thruster trembling
unit. The latter is composed by a bank of (steady-state) Kalman caused by the wave-induced motion components, notch fil-
filters (KFs) that generate online the output estimation errors ters in cascade with low pass filters were used with the
(residuals) that are used to generate appropriate monitoring controllers. An improvement in performance was achieved
signals. At each sampling time, the monitoring signals are by exploiting more advanced control techniques based on
assessed to decide which controller should be selected from
the bank of the controllers. The proposed adaptive structure of optimal control and Kalman filter theory, see [2]–[4]. A
the RMMADP enables the DP system to operate in different simpler set-up based on nonlinear passive observers (to
operational conditions and hence, it is a step forward to a replace Kalman filters) was introduced in [5]–[7]. Further
so-called all-year marine DP system. Numerical simulations, developments in recent years have led to the use of nonlinear
carried out with a high fidelity nonlinear DP simulator, illus- control [8], robust control [9], [10], adaptive control [11] and
trate the efficacy of the RMMADP techniques proposed. To
bridge the gap between theory and practice, the results are hybrid control [12] theories in the design of DP systems. The
experimentally verified by model testing a DP operated ship, literature on ship DP is vast and defies a simple summary.
the Cybership III, under different sea conditions in a towing See for example [13], [14] and the references therein for a
tank equipped with a hydraulic wave maker. short presentation of the subject and its historical evolution.
Most of the current DP systems are designed to operate up
I. INTRODUCTION to certain limits of weather conditions. However, in practice
The first generation of DP systems came to existence in the the sea state may undergo large variations and therefore the
1960s for offshore drilling applications, due to the need to controller should adapt to the sea state itself. To meet this
drill in deep waters and the realization that Jack-up barges challenge, different techniques have been proposed. Among
and anchoring systems could not be used economically at them, supervisory control techniques were exploited in [12]
such depths. The first vessel equipped with a Dynamic Po- to design a hybrid DP controller. A distinctive feature of the
sitioning (DP) system was launched in 1961 [1]. The vessel, controller developed was the use of spectral techniques to
named Eureka, was property of the Shell Oil Company. estimate the wave spectrum in surge, sway, and yaw from
Nowadays, DP systems are used with a wide range of vessel position and heading measurements. The results were used
types and in different marine operations such as hydrographic to identify the sea state, based on which the appropriate con-
surveying, marine construction, wreck investigation, under- troller was selected from a pre-defined bank of controllers.
water recovery, site surveying, underwater cable and pipe However, this approach is sensitive to measurement noise
laying, and inspection and maintenance. In particular, in the and may have latency problems because it requires that the
offshore, oil, and gas industries many applications are only samples acquired be buffered to estimate the Power Density
possible with the use of DP systems for service vessels, Spectrum of the measurement time series.
drilling rigs and ships, shuttle tankers, cable and pipe layers, In this paper, availing ourselves of previous results ob-
floating production off-loading and storage units (FPSOs), tained by the authors in [9], [10], [15], [16], we propose
a new type of DP control law that we will henceforth be
*This work was supported in part by projects MORPH (EU FP7 under referred to as a Robust Multiple Model Adaptive Dynamic
grant agreement No. 288704) and PEst-OE/EEI/LA0009/2011, and was
carried out in cooperation with the Norwegian Marine Technology Research
Positioning (RMMADP) system. A bank of individual robust
Institute (MARINTEK) and the Centre for Autonomous Marine Operations DP controllers for different sea conditions (calm, moderate,
and Systems (AMOS); the Norwegian research council is acknowledged as high and extreme) are designed using mixed-µ techniques
sponsor of MARINTEK and AMOS.
1 Vahid Hassani was formerly with LARSyS, Portugal, and is now [17]. In the new structure proposed, a bank of Kalman filters
with the Norwegian Marine Technology Research Institute (MARINTEK), are designed based on a finite number of models of the vessel
Trondheim, Norway. vahid.hassani@marintek.sintef.no when it undergoes operations under different sea conditions.
2 A. J. Sørensen is with Centre for Autonomous Marine Op-
Multiple model identification tools are used to identify
erations and Systems (AMOS) and Dept. of Marine Technology,
Norwegian Univ. of Science and Technology, Trondheim, Norway. the sea state and to select the appropriate (locally) robust
asgeir.sorensen@ntnu.no DP controller from a pre-defined bank of the controllers.
3 António M. Pascoal is with the Laboratory of Robotics and
The main objective of this paper is to integrate a bank
Systems in Engineering and Science (LARSyS), Instituto Superior
Técnico (IST), Univ. Técnica de Lisboa (UTL), Lisbon, Portugal. of appropriate robust DP controllers into a state-of-the-art
antonio@isr.ist.utl.pt robust adaptive DP architecture yielding good performance
in varying operational conditions, from calm to extreme seas. with
The proposed structure extends the so-called weather window 2
Ω = diag{ω01 2
, ω02 2
, ω03 },
operational availability of a DP system without the need for
spectral identification techniques (which exhibit latency and Λ = diag{2ζ1 ω01 , 2ζ2 ω02 , 2ζ3 ω03 },
are very sensitive to measurement noise). where ω0 = [ω01 ω02 ω03 ]T and ζi are the Dominant Wave
Numerical simulations, carried out in a high fidelity non- Frequency (DWF) and relative damping ratio, respectively.
linear DP simulator, illustrate the efficacy of the RMMADP Matrix T = diag(Tx , Ty , Tψ ) is a diagonal matrix of positive
techniques proposed. To bridge the gap between theory and bias time constants and Eb ∈ R3×3 is a diagonal scaling
practice, the results are experimentally verified by model matrix. Vector ηL ∈ R3 consists of low frequency (LF),
testing of a DP operated ship, the Cybership III, under earth-fixed position (xL , yL ) and LF heading ψL of the
different simulated sea conditions in a towing tank equipped vessel relative to an earth-fixed frame, ν ∈ R3 represents
with a hydraulic wave maker. the velocities decomposed in a vessel-fixed reference, and
The structure of the paper is as follows. In section II R(ψL ) is the standard orthogonal yaw angle rotation matrix
we present the main issues that arise in the design of the (see [21] for complete details). Equation (5) describes the
RMMADP. In Section III we present the results of numerical vessels’s LF motion at low speed (see [21]), where M ∈
Monte-Carlo simulations with stochastic signals, carried out R3×3 is the generalized system inertia matrix including
in the Marine Cybernetics Simulator, that illustrate the per- zero frequency added mass components, D ∈ R3×3 is the
formance of developed DP controller in calm to extreme sea linear damping matrix, and τ ∈ R3 is a control vector
conditions. In section IV, a short description of the model- of generalized forces generated by the propulsion system,
test vessel, Cybership III, and experimental results of model- that is, the main propellers aft of the ship and thrusters
tests are presented. Conclusions and suggestions for future which can produce surge and sway forces as well as a
research are summarized in Section V. yaw moment. Vector ηtot ∈ R3 describes the vessel’s total
II. THE ROBUST MULTIPLE-MODEL ADAPTIVE motion, consisting of total position (xtot , ytot ) and total
DYNAMIC POSITIONING heading ψtot of the vessel. Finally, (7) represents the position
In what follows, the vessel model that is by now standard1 and heading measurement equation, with v ∈ R3 a zero-
is presented. See for example [6], [8], [18], [19]. The model mean Gaussian white measurement noise.
From (1)-(6), using practical assumptions, a linear model
admits the realization
with parametric uncertainty was obtained in [9] as follows:

ξ˙W = AW (ω0 )ξW + EW wW (1) ξ˙W = AW (ω0 )ξW + EW wW (8)

b
ηW = R(ψL )CW ξW (2) ηW = CW ξW (9)
−1 p
ḃ = −T −1 b + Eb wb (3) ḃ = −T b + θ1 Sb +
p p
wbf (10)
p p
η̇L = R(ψL )ν (4) η̇L = θ1 SηL + ν + θ2 Sν (11)
p
M ν̇ + Dν = τ + RT (ψtot )b (5) M ν̇ + Dν = τ + b (12)
p
ηtot = ηL + ηW (6) ηyf = ηL b
+ ηW +n (13)
b
ηy = ηtot + v, (7) where ηW are WF components of motion on the body-
coordinate axis, wbf and ηyf are a new modified disturbance
where (1) and (2) capture the 1st-order wave induced motions
in surge, sway, and yaw; equation (3) represents the 1st-order and a modified measurement defined by wbf = RT (ψy )Eb wb
Markov process approximating the unmodelled dynamics and ηyf = RT (ψy )ηtot + n, respectively, n ∈ R3 is the mea-
and the slowly varying environmental forces (in surge and surement noise, ω0 , θ1 , and θ2 are parametric uncertainties
sway) and torques (in yaw) due to waves (2nd order wave given in Table II, and the matrix S is given by
 
induced loads), wind, and currents. The latter are given in 0 1 0
earth fixed coordinates but expressed in body-axis. In the S =  −1 0 0  .
above, ηW ∈ R3 is the vessel’s WF motion due to 1st- 0 0 0
order wave-induced disturbances, consisting of WF position
(xW , yW ) and WF heading ψW of the vessel; wW ∈ R3 and The equations describing the kinematics and the dynamics
wb ∈ R3 are zero mean Gaussian white noise vectors, and of the vessel can be represented in the following standard
[ ] [ ] form for multiple-input-multiple-output (MIMO) linear plant
03×3 I3×3 0
AW = , EW = 3×1 , models:
−Ω3×3 −Λ3×3 I3×1
[ ]
CW = 03×3 I3×3 , ẋ(t) = A(ω0 , θ1 , θ2 )x(t) + Bu(t) + Lw(t), (14a)
y(t) = Cx(t) + v(t), (14b)
1 The model described by (1)-(6) has minor differences with respect to
p T
the ones normally available in the literature. While in most of the literature where x(t) = [ξW T bp T ηL ν T ]T ∈ R15 denotes the
the WF components of motion are modeled in a fixed-earth frame, in this
paper the WF motion is modeled in body-frame. The reader is referred to state of the system, u(t) = M −1 τ ∈ R3 its control input,
[16], [20] for details and improvements of the present model. y(t) = ηyf ∈ R3 its measured noisy output, w(t) =
 T
[wW T wbf ]T ∈ R6 an input plant disturbance that cannot sea conditions; in particular, superior performance of robust
be measured, and v(t) = n ∈ R3 is the measurement DP controllers in extreme sea condition is shown in [24].
noise. The equations in (14) are simply a compact way In a RMMADP system a bank of KFs are used in
of presenting equations in (8)-(13); A(ω0 , θ1 , θ2 ), B, L order to select the correct controllers from the bank of
and C are defined in the obvious manner. Vectors w(t) the controllers. Each KF is designed based on a selected
and v(t) are zero-mean white Gaussian signals, mutually value of the unknown parameter, ω0 . The residuals of all
independent with intensities E{w(t)wT (τ )} = Qδ(t − τ ) the KFs are analyzed in a block called Monitoring Signal
and E{v(t)v T (τ )} = Rδ(t − τ ). The initial condition x(0) Evaluator (MSE). The MSE assigns a performance index
of (14) is a Gaussian random vector with mean and covari- (monitoring signal) to each KF (and the corresponding
ance given by E{x(0)} = 0 and E{x(0)xT (0)} = Σ(0), controller). Then, the monitoring signals µi (to be defined
respectively. Matrix A(ω0 , θ1 , θ2 ) contains unknown constant shortly) are associated with the ith KF (and ith controller,
parameters indexed by ω0 , θ1 , and θ2 . The parametric i.e. Ki , in the bank of the controllers). These monitoring
uncertainty interval of θ1 and θ2 is very small1 and the main signals are used to select the best local controller from the
parametric uncertainty in the the model given by (14) is the bank of robust DP controllers. Four selected values for ω0
DWF, ω0 . Table I shows the definition of the sea conditions are chosen as {0.48, 0.63, 0.92, 1.18} (rad/s). Each selected
characterized by the DWF. The sea conditions are associated value represents one of the sea states, calm, moderate, high
with the particular model of offshore supply vessel that is and extreme. The bank of the Kalman filters is designed
used in our study. We assume that DWF lies in the interval based on the selected nominal values for the dominant wave
frequency (DWF); see [24] for details on the design of a
TABLE I
Kalman filter for DP systems and [25] for details on the
D EFINITION OF S EA S TATES FROM [22]
selection of the nominal design models. Each steady state
KF has the following realization [26]:
Sea States DWF Significant Wave Height ( )
ω0 (rad/s) Hs (m) x̂i (t + 1) = A(ω0 , θ1 , θ2 )x̂i (t) + Bu(t) + Hθi y(t) − ŷi (t) ,
Calm Seas > 1.11 < 0.1 ŷi (t) = C x̂i (t),
Moderate Seas [0.74 1.11] [0.1 1.69]
High Seas [0.53 0.74] [1.69 6.0] Hθi = A(ω0 , θ1 , θ2 )Pi C T [CPi C T + R]−1 ,
Extreme Seas < 0.53 > 6.0
where Pi is the solution of the discrete Riccati equation
Pi = A(ω0 , θ1 , θ2 )Pi A(ω0 , θ1 , θ2 )T + LQLT
TABLE II − A(ω0 , θ1 , θ2 )Pi C T [CPi C T + R]−1 CPi A(ω0 , θ1 , θ2 )T .
I NTERVAL OF PARAMETRIC U NCERTAINTIES (16)
It is assumed that [A(ω0 , θ1 , θ2 ), L] and [A(ω0 , θ1 , θ2 ), C]
Sea Status ω0 θ1 θ2
rad/s rad/s2 rad/s are controllable and observable, respectively for all admis-
Calm Seas [1.11 1.8] Int∗ [−0.038 0.038] sible values of of ω0 , θ1 , and θ2 . The symmetric positive
Moderate Seas [0.74 1.11] Int [−0.04 0.04] definite matrices Q and R were defined before as covariance
High Seas [0.53 0.74] Int [−0.042 0.042] matrices of the plant disturbance and measurement noise,
Extreme Seas [0.39 0.53] Int [−0.04 0.04]
∗ Int=[−5 × 10−4 5 × 10−4 ]
respectively.
The output estimation errors (ỹi (t) = y(t) − ŷi (t)) and error
covariances of all the Kalman filters (Si = CPi C T + R) are
[0.39 1.8]2 that covers calm, moderate, high and extreme sea used to compute a performance signal that can be viewed as
conditions. a gaussian maximum likelihood ratio. This signal is called a
In the RMMADP deign methodology we divide the “monitoring signal” µi (t), and is defined as
parametric uncertainty region into smaller regions and we
1∑1 ( 3√ )
t
design robust controllers Ki for each subregion, using the µi (t) := ỹi (k)T Si−1 ỹi (k) + ln (2π) 2 |Si | . (17)
µ-synthesis method [17], [23]. Here, we borrow from the t 2
k=1
work in [9], [10] where four (locally) robust DP controllers,
For more details on definition and concept of the monitoring
for calm, moderate, high and extreme sea conditions, were
signals see [15]. The monitoring signals are then used to
designed and their performance evaluated both through nu-
select one of the local controllers (the one associated with
merical simulations (in the Marine Cybernetics Simulator -
Kalman filter with the smallest monitoring signal).
MCSim), and experimentally, performing model test exper-
Fig. 1 shows the architecture of the RMMADP system.
iments. Moreover, the performance of the designed robust
The RMMADP, in Fig. 1, consists of: i) the monitoring signal
DP controllers for different sea conditions is compared with
evaluator, ii) a bank of N KFs, where each local estimator is
the ones of LQG and PID controllers in [24] showing the
designed based on one of the representative parameters, and
satisfactory performance of robust DP controllers in different
iii) a bank of N robust DP controllers which are switched in
1 See [5]–[7] where θ1 and θ2 are ignored in the design process. the feedback loop based on the values of monitoring signals.
2 We use the same interval for DWF in surge, sway and yaw. A dwell-time switching policy is used to prevent chattering
 w(t ) v(t )
shows the definition of the sea condition associated with
u (t ) x(t ) " A( ,!1 ,! 2 ) x(t ) # Bu (t ) # Lw(t )
a particular model of supply vessel that is used in the
0 y (t )
y (t ) " Cx(t ) # v(t )
MCSim. The sampling time of Ts = .25 (sec) is used to
implement the RMMADP system throughout the simulation
y1 (t ) u1 and experimental tests.
1st K.F. ( 1 ) K1
S1 1
Fig. 2 shows the results of a simple simulation4 . In this
(computed off-line)
u2
y2 (t ) Monitoring K2 Switch
u (t )
experiment we examine the performance of the RMMADP
nd 2 arg min µi
2 K.F. ( 2 ) S2 Signal
.. system in different sea conditions. The first (upper) sub-
...
i
(computed off-line) Generator ... .
y N (t ) uN figure shows the switching signal in a calm sea condition.
th N KN
N K.F. ( N ) S N The second sub-figure shows the switching signal in a
(computed off-line)
Dwell-Time
(t )
moderate sea condition. The switching signal in high sea
condition is shown in the third sub-figure and finally the
Fig. 1. The RMMADP architecture. switching signal in extreme sea state is depicted in the last
(lowest) sub-figure of Fig. 2.
among the controllers, see [15] for details on the structure of 4

the RMMADP. The stability of the overall system is studied 3

σ(t)
2
in [15]. It is shown that in steady state the correct controller 1
is selected and during the transient all the internal signals 4

remain bounded. The control action is uσ (t), where σ = 3

σ(t)
2
argmin{µi (t)}.3 1
In DP systems, it is important that the controller adapt 4

to the sea state that may change during operations. In the 3

σ(t) 2
RMMADP methodology, in order to distinguish the sea 1

state, a multiple model structure is exploited to identify 4

3
the sub-interval that the peak frequency of the assumed
σ(t)

2
wave spectrum model lies in. Based on the results of the 1
0 50 100 150
identification, which are carried out at each sampling time time (sec)

by assessing the monitoring signals, an appropriate controller

(from the bank) is switched in the feedback loop. Fig. 2. Simulation result: Switching signals in different sea conditions;
from top to bottom: calm, moderate, high, and extreme sea condition.
III. NUMERICAL SIMULATIONS
In what follows we test the performance of our controller We stress that the performance of any adaptive system
using the Marine Cybernetics Simulator (MCSim), later must be evaluated not only for constant unknown param-
on upgraded to the Marine System Simulator (MSS). The eters but also, for time-varying parameters which undergo
MCSim is a modular multi-disciplinary simulator based on slow or rapid time-variations. In practice, the sea state
Matlab/Simulink. It was developed at the department of may experience large variations and therefore adaptation is
marine technology of the Norwegian University of Science necessary in the dynamic positioning. Fig. 3 shows the results
and Technology (NTNU). The MCSim incorporates high of the simulation where the sea state changes in time. In
fidelity models, denoted as process plant model or simulation this experiment, during the first 300 seconds the moderate
model in [19], at all levels (plants and actuators). It captures sea state is simulated and for the next 300 seconds the
hydrodynamic effects, generalized coriolis and centripetal sea state changes to high condition. Fig. 3 represents the
forces, nonlinear damping and current forces, and general- time evolution of the switching signal and the position of
ized restoring forces. It is composed of different modules the vessel. The first sub-figure shows the switching signal;
such as environmental module, vessel dynamics module, clearly RMMADP system follows the sea changes.5 The
thruster and shaft module, and Vessel control module. For remaining sub-figures in Fig. 3 depicts the time evolution
more details on the MCSim see [27], [28]. of the position of the vessel.
The results of Monte-Carlo simulations aimed at assessing IV. EXPERIMENTAL RESULTS
the performance of the RMMADP controller are presented
below. In these simulations, the different environment con- The controller designed was tested using the model vessel
ditions from calm to extreme seas are simulated using the Cybership III, at the Marine Cybernetic Laboratory (MCLab)
spectrum of the Joint North Sea Wave Project (JONSWAP) 4 All the results in the paper are presented in full scale.
[29]. Throughout the paper, four different environment con- 5 For the time varying case, 2 monitoring signals are computed for
ditions from calm to extreme seas are considered. Table I each observer. The monitoring signals reset every 40 seconds (with a 20
second time difference between the resetting time of the monitoring signals
3 In this architecture, the identification module (that relies on monitor- associated with each observer) and at each time only one of these monitoring
ing signal evaluator) computes the smallest monitoring signal µi ; i ∈ signals (for each observer) goes to the supervisor. This means that when
{1, 2, . . . N } and switches the controller Ki associated with that monitoring a monitoring signal, which is fed to supervisor, resets its value, the other
signal, after which it dwells on that selection for a predefined time, called monitoring signal whose value was reset 20 seconds ago and has by now
dwell-time. See [15] for details on dwell-time and how to compute it. reached its steady state value, will be fed to the supervisor.
 4
3
bow.6 It has a mass of m = 75 (kg), length of L = 2.27 (m)
σ (t)
2 and breadth of B = 0.4 (m). Main parameters of the model is
1
presented in Table III. The internal hardware architecture is
5
controlled by an onboard computer which can communicate
North (m)

−5
with onshore PC through a WLAN. The PC onboard the
ship uses QNX real-time operating system (target PC). The
1
control system is developed on a PC in the control room (host
East (m)

−1
PC) under Simulink/Opal and downloaded to the target PC
using automatic C-code generation and wireless Ethernet.
Heading (deg)

2
0
The motion capture unit (MCU), installed in the MCLab,
−2 provides Earth-fixed position and heading of the vessel. The
0 100 200 300
Time (s)
400 500 600 MCU consists of 3 onshore cameras mounted on the towing
carriage and a marker mounted on the vessel. The cameras
Fig. 3. Simulation result: Evolution of the switching signal and the position emit infrared light and receive the light reflected from the
of the vessel in a time varying scenario marker.
To simulate the different sea conditions, a hydraulic wave
maker system was used. It consists of a single flap covering
of the Department of Marine Technology at the Norwegian
the whole breadth of the basin, and a computer controlled
University of Science and Technology (NTNU). The per-
motor moving the flap. It can produce regular and irregular
formance of the controllers was tested under different sea
waves with different spectrums. We have used the JONSWAP
conditions produced by a hydraulic wave maker.
spectrum to simulate the different sea conditions for our
A. Overview of the CybershipIII experiment.
CyberShip III is a 1:30 scaled model of an offshore vessel Fig. 5 shows the results of the experiment where the
operating in the North Sea. Fig. 4 shows the vessel at wave maker system simulates different sea state conditions.
the basin in the MCLab and table III presents the main Since the robust DP controllers are designed for a linearized
parameters of the model and full scale vessel. system, a simple PI controller is used as an initializer to
bring the vessel to the desired point (the origin) in the basin.
TABLE III The first (upper) sub-figure shows the wave elevation profile
M ODEL MAIN PARAMETERS recorded by a probe installed five meters away from the wave
maker. The second sub-figure shows the switching signal.
Model Full Scale
Overall Length 2.275 m 68.28 m
As in the case of simulations, the RMMADP system keeps
Length between track of the sea state. However, we should stress that we have
perpendiculars 1.971 m 59.13 m tuned the RMMADP during a few tests and Fig. 5 shows the
Breadth 0.437 m 13.11 m final tuned system. We also should highlight that the wave
Breadth at water line 0.437 m 13.11 m
Draught 0.153 m 4.59 m profiles generated by the wave maker for high and extreme
Draught front perpendicular 0.153 m 4.59 m seas were not very different due to the limited capacity of
Draught aft. perpendicular 0.153 m 4.59 m the wave maker. Therefore, identifying the high and extreme
Depth to main deck 0.203 m 6.10 m
Weight (hull) 17.5 kg Unknown
sea conditions proved to be more difficult than identifying the
Weight (normal load) 74,2 kg 22.62 tons other conditions. The remaining sub-figures in Fig. 5 shows
Longitudal center of gravity 100 cm 30 m the time evolution of the position and heading of the vessel.
Vertical center of gravity 19.56 cm 5.87 m
Propulsion motors max V. CONCLUSIONS AND FUTURE RESEARCH
shaft power (6% gear loss) 81 W 3200 HP
A Robust Multiple Model Adaptive Dynamic Positioning
Tunnel thruster max
shaft power (6% gear loss) 27 W 550 HP (RMMADP) system was described. The RMMADP system
Maximum Speed Unknown 11 knots can operate in time-varying operational conditions, from
calm to extreme seas, thus making it a good candidate for
all-year operations. Furthermore, the system dispenses with
the need for spectral identification techniques (which exhibit
delays and are very sensitive to measurement noise). The
proposed RMMADP design built upon recent developments
on robust adaptive techniques using a multiple model struc-
ture. The RMMADP consists of a bank of robust controllers
designed using Mixed-µ methodology and an identification
unit. The identification unit uses a bank of (steady-state)
Fig. 4. Cybership III. Kalman filters (KFs) that generates online appropriate moni-
toring signals. At each sampling time, the monitoring signals
Cybership III is equipped with two pods located at the aft. 6 For technical reasons in this experiment the tunnel thruster was deacti-
A tunnel thruster and an azimuth thruster are installed in the vated.
 4 High Sea
Extreme Sea [8] G. Torsetnes, J. Jouffroy, and T. I. Fossen, “Nonlinear dynamic

Wave Profile (m)

Moderate Sea
2
positioning of ships with gain-scheduled wave filtering,” in Proc. IEEE
Calm Sea Moderate Sea
Conference on Decision and Control (CDC’04), Paradise Iceland,
0
Bahamas, 2004.
−2 Initialization [9] V. Hassani, A. J. Sørensen, and A. M. Pascoal, “Robust dynamic
−4 positioning of offshore vessels using mixed-µ synthesis, part I: Design-
4 ing process,” in Proc. ACOOG 2012 - IFAC Workshop on Automatic
Control in Offshore Oil and Gas Production, Trondheim, Norway,
σ (t)

2
2012.
0 [10] ——, “Robust dynamic positioning of offshore vessels using mixed-
10
µ synthesis, part II: Simulation and experimental results,” in Proc.
North (m)

0
ACOOG 2012 - IFAC Workshop on Automatic Control in Offshore Oil
−10
and Gas Production, Trondheim, Norway, 2012.
2 [11] E. A. Tannuri, L. K. Kubota, and C. P. Pesce, “Adaptive techniques
East (m)

0 applied to offshore dynamic positioning systems,” Journal of the

−2 Brazilian Society of Mechanical Sciences and Engineering, vol. 28,
Heading (deg)

5 no. 3, pp. 323–330, 2006.

0 [12] T. D. Nguyen, A. J. Sørensen, and S. T. Quek, “Design of hybrid con-
−5
troller for dynamic positioning from calm to extreme sea conditions,”
500 1000 1500 2000 2500 3000 Automatica, vol. 43, no. 5, pp. 768–785, 2007.
Time (s)
[13] A. J. Sørensen, “Structural issues in the design and operation of marine
control systems,” Annual Reviews in Control, vol. 29, pp. 125–149,
Fig. 5. Experimental results: evolution of the wave profile, the switching 2005.
signal and the position of the vessel in time varying scenario (data is scaled [14] ——, “A survey of dynamic positioning control systems,” Annual
from model to full scale). Reviews in Control, vol. 35, pp. 123–136, 2011.
[15] V. Hassani, J. Hespanha, M. Athans, and A. M. Pascoal, “Stability
analysis of robust multiple model adaptive control,” in Proc. of The
18th IFAC World Congress, Milan, Italy, 2011.
are processed to select which controller should be selected [16] V. Hassani, A. J. Sørensen, and A. M. Pascoal, “Multiple model
from the bank of controllers. Numerical simulations, carried adaptive wave filtering for dynamic positioning of marine vessels,”
out with a high fidelity nonlinear DP simulator illustrated in Proc. ACC’12 - American Control Conference, Montreal, Canada,
2012.
the efficacy of the RMMADP techniques proposed. The [17] G. J. Balas, “mixed-µ software (unpublished version),” 2009, private
results were experimentally verified by model testing a DP communication.
operated ship, the Cybership III, under different simulated [18] T. I. Fossen and J. P. Strand, “Passive nonlinear observer design for
ships using lyapunov methods: Full-scale experiments with a supply
sea conditions in a towing tank with a hydraulic wave maker. vessel,” Automatica, vol. 35, pp. 3–16, 1999.
The experimental data confirms that the method developed [19] A. J. Sørensen, “Lecture notes on marine control systems,” Norwegian
holds promise for practical applications. University of Science and Technology, Tech. Rep. Report UK-11-76,
2011.
[20] V. Hassani, A. J. Sørensen, A. M. Pascoal, and A. P. Aguiar, “Devel-
ACKNOWLEDGMENT oping a linear model for wave filtering and dynamic positioning,” in
Proc. CONTROLO12 - the 10th Portuguese Conference on Automatic
We thank our colleagues A. Pedro Aguiar, J. Hespanha and Control, Madeira, Portugal, 2012.
Michael Athans for many discussions on adaptive control. [21] T. I. Fossen, Handbook of Marine Craft Hydrodynamics and Motion
Control. Chichester, UK: John Wiley & Sons. Ltd, 2011.
We would also like to thank T. Wahl, Øyvind Smogeli, M. [22] W. G. Price and R. E. D. Bishop, Probabilistic Theory of Ship
Etemaddar, E. Peymani, M. Shapouri, and B. Ommani for Dynamics. London, UK: Chapman and Hall, 1974.
their great assistance during the model tests at MCLab. [23] S. Skogestad and I. Postlethwaite, Multivariable Feedback Control:
Analysis and Design (2nd Edition). Wiley, 2006.
[24] V. Hassani, A. J. Sørensen, and A. M. Pascoal, “Evaluation of
R EFERENCES three dynamic ship positioning controllers: from calm to extreme
conditions,” in Proc. NGCUV 2012 - IFAC Workshop on Navigation,
[1] D. Bray, Dynamic Positioning, 2nd ed., ser. Oilfield Seamanship Guidance and Control of Underwater Vehicles, Porto, Portugal, 2012.
Series. London, UK: Oilfield Publications Ltd (OPL), 2003, vol. 9. [25] S. Fekri, M. Athans, and A. Pascoal, “Issues, progress and new results
[2] J. Balchen, N. Jenssen, and S. Sælid, “Dynamic positioning using in robust adaptive control,” Int. J. of Adaptive Control and Signal
Kalman filtering and optimal control theory,” in the IFAC/IFIP Sympo- Processing, vol. 20, pp. 519–579, 2006.
sium On Automation in Offshore Oil Field Operation, Bergen, Norway, [26] B. D. O. Anderson and J. B. Moore, Optimal Filtering. New Jersey,
1976, pp. 183–186. USA: Prentice-Hall, 1979.
[3] M. J. Grimble, R. J. Patton, and D. A. Wise, “The design of dynamic [27] A. J. Sørensen, E. Pedersen, and O. Smogeli, “Simulation-based
ship positioning control systems using extended Kalman filtering design and testing of dynamically positioned marine vessels,” in
techniques,” in Proc. IEEE Oceans Conference (Oceans’79), San Proc. of International Conference on Marine Simulation and Ship
Diego, CA, 1979, pp. 488–497. Maneuverability (MARSIM’03), Kanazawa, Japan, 2003.
[4] S. Sælid, N. A. Jenssen, and J. Balchen, “Design and analysis of a [28] T. I. Fossen and T. Perez, “Marine systems simulator (MSS),”
dynamic positioning system based on Kalman filtering and optimal “www.marinecontrol.org”, 2009.
control,” IEEE Transactions on Automatic Control, vol. 28, no. 3, pp. [29] K. Hasselmann, T. P. Barnett, E. Bouws, H. Carlson, D. Cartwright,
331–339, 1983. K. Enke, J. A. Ewing, H. Gienapp, D. E. Hasselmann, P. Kruseman,
[5] J. P. Strand, “Nonlinear position control systems design for marine A. Meerburg, P. Müller, D. J. Olbers, K. Richter, W. Sell, and
vessels,” Ph.D. dissertation, Dept. of Eng. Cybernetics, Norwegian H. Walden, “Measurements of wind-wave growth and swell decay
University of Science and Technology, Trondheim, Norway, 1999. during the joint north sea wave project (JONSWAP),” Ergnzungsheft
[6] J. P. Strand and T. I. Fossen, “Nonlinear passive observer for ships zur Deutschen Hydrographischen Zeitschrift Reihe, vol. 8, no. 12, pp.
with adaptive wave filtering,” New Directions in Nonlinear Observer 1–95, 1973.
Design (H. Nijmeijer and T. I. Fossen, Eds.), Springer-Verlag London
Ltd., pp. 113–134, 1999.
[7] T. I. Fossen, “Nonlinear passive control and observer design for ships,”
Modeling, Identification and Control (MIC), vol. 21, no. 3, pp. 129–
184, 2000.