See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/252031927

Real-time Estimation of a Ship's Attitude

Article  in  Proceedings of the American Control Conference · June 2011

DOI: 10.1109/ACC.2011.5990612

CITATIONS READS
15 290

5 authors, including:

Johannes K. Eberharter
University of Applied Siences / Fachhochschule Vorarlberg, Dornbirn, Austria
24 PUBLICATIONS   106 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Computational and Line Geometric Methods View project

Applied Crane Controls View project

All content following this page was uploaded by Johannes K. Eberharter on 05 July 2015.

The user has requested enhancement of the downloaded file.

2011 American Control Conference
on O'Farrell Street, San Francisco, CA, USA
June 29 - July 01, 2011

Real-Time Estimation of a Ship’s Attitude

Sebastian Küchler, Christoph Pregizer, Johannes Karl Eberharter, Klaus Schneider, and Oliver Sawodny

Abstract— During subsea lifting operations in harsh sea motion sensor has three accelerometers for measuring surge,
conditions, the involved crane system is subjected to extensive sway, and heave and three rotation rate sensors for roll,
dynamic forces due to vertical vessel motion. Thus, active heave pitch, and yaw. Obviously, the accelerometer signals have
compensation systems can be used to compensate for vertical
vessel motion and to reduce forces acting on the crane structure. to be integrated twice and the rotation rate signals once to
Furthermore, such systems allow an exact positioning of the obtain the relative position and attitude of a vessel. To reduce
load on the seabed. However, active heave compensation systems typical errors like sensor noise, bias, and misalignment of the
always require knowledge about the vertical position of the rotation rate sensors and accelerometers, different approaches
crane depending on the ship’s heave, roll, and pitch motion. for signal conditioning can be found in the literature. For
Hence, an attitude estimation method for ships during subsea
lifting operations is proposed. To estimate the roll and pitch example, Godhavn [4] proposes integrating filters to obtain
motion of a vessel with high accuracy, rotation rate sensors are the relative position of a vessel. The filter coefficients are
fused with accelerometers using an Extended Kalman Filter. chosen in accordance to the actual sea spectrum. Other
Since an exact knowledge of the yaw motion is not required approaches for attitude estimation of a body fuse different
to determine the crane’s vertical motion, the yaw angle is sensor signals to compensate different errors of each sensor.
stabilized around zero with an additional virtual sensor signal.
The attitude estimation algorithm is evaluated with simulation A widely-used approach is to aid the IMU signals with GPS
and measurement results from an experimental setup. measurements to eliminate the resulting drift terms of the
estimated position and attitude as described by Godhavn [5]
I. INTRODUCTION or Fossen and Perez [6]. However, the proposed method
requires additional GPS signals resulting in higher costs.
There is an increasing demand for offshore installations,
Another possibility for attitude estimation is to fuse the sig-
such as underwater conveying systems for oil and gas fields
nals of the gyro rate sensors with the measured accelerations
or wind parks in the near future. Offshore oil and gas
of the IMU to estimate the errors of the rotation rate signals.
fields will be developed to a large extent with all processing
Kim and Golnaraghi [7] propose such a method. They model
equipment on the seabed. Thus, high operability on the
the rotation rates as first order systems and formulate an
underwater construction is required.
Extended Kalman Filter (EKF) estimating the attitude of a
On the other hand, dealing with such installations puts new
rigid body. Metni et al. [8] fuse signals of three gyroscopes,
challenges on the equipment for the offshore industry, be-
three accelerometers, and three magnetometers in two com-
cause waves, wind, and ocean currents easily cause the vessel
plementary filters. The derived filters are used to estimate
to move away both horizontally and vertically. Especially,
the attitude of an unmanned aerial vehicle. The first filter
the vessel’s vertical motion has a significant effect on the
fuses the gyroscopes with the accelerometers to estimate
involved crane system shown in Fig. 1 during subsea lifting
the roll and pitch angles, while the second filter is used to
operations. Thus, it is important to keep the load motion
aid the gyroscopes with the magnetometers to estimate the
unaffected by the wave induced vessel motion even under
yaw angle. The additional usage of magnetometers results in
harsh sea conditions. For decoupling the vertical load motion
a better estimation of the yaw angle. Data fusion of three
from the vertical vessel motion, passive or active heave
gyroscopes, two inclinometers, and a compass via an EKF
compensation systems can be used. All active systems have
is presented by Setoodeh et al. [9]. The paper derives an
in common that they require exact knowledge of the vertical
indirect error state model basing on attitude errors and bias
crane tip’s position where the load is attached to, as described
of the gyroscopes. The resulting EKF is realized in a way
in the literature [1], [2], [3]. However, the vertical motion
that it separates the model of the bias and the error dynamics
of the crane tip is not only affected by the vessel’s heave
resulting in two subsystems. However, all these methods
motion, but also by the vessel’s roll and pitch motion. Hence,
have problems estimating the correct roll and pitch angles
active heave compensation systems require the actual heave,
in situations when the IMU is accelerated, since they use a
roll, and pitch motion of a vessel.
model of the earth’s gravitational vector, which is only valid
All six degrees of freedom of a vessel or ship can be obtained
for an IMU that is not accelerated, to stabilize the roll and
from an inertial measurement unit (IMU). Such a standalone
pitch motion.
S. Küchler, C. Pregizer, and O. Sawodny are with This paper presents a method for estimating the attitude of a
the Institute for System Dynamics, University of ship or vessel during subsea lifting operations. The scheme
Stuttgart, P.O. Box 801140, D-70511 Stuttgart, Germany utilizes a low-cost IMU (ADIS 16365) from Analog Devices
sebastian.kuechler@isys.uni-stuttgart.de
J. K. Eberharter and K. Schneider are with Liebherr Werk Nenzing as a standalone motion sensor without aiding external sensor
GmbH, P.O. Box 10, A-6710 Nenzing, Austria sources like a GPS receiver or a compass. The IMU has

978-1-4577-0079-8/11/$26.00 ©2011 AACC 2411

 yn n-frame n-frame is considered as the inertial frame. The second frame
is the body frame, denoted by b, and attached to the IMU.
xn
offshore crane crane tip The orthogonal axes are aligned with the sensor axes.
n
The rotation model of the sensor is derived using quater-
z
vessel/ship IMU nions rather than direction cosines and Euler angles, as the
quaternion representation does not have a problem with sin-
gularities, handles normalization better, and does not require
bulky trigonometric calculations of Euler angles. Following
T
y b , sway pitch b-frame [10], the propagation of the quaternion q bn = [q0 q1 q2 q3 ] ,
indicating the transformation from the n- to the b-frame, can
T
roll be expressed using the rotation rates ω bnb = [ωx ωy ωz ]
sea waves (rotation rate of the b-frame relative to the n-frame expressed
yaw xb , surge in the b-frame) and written as
q̇0 = − 12 (q1 ωx + q2 ωy + q3 ωz ),
z b , heave payload q̇1 = 21 (q0 ωx − q3 ωy + q2 ωz ),
(1)
q̇2 = 12 (q3 ωx + q0 ωy − q1 ωz ),
Fig. 1. Vessel/Ship with offshore crane.
q̇3 = − 12 (q2 ωx − q1 ωy − q0 ωz ).
Because every gyroscope measures an additive offset and
three accelerometers and three rotation rate sensors. To avoid sensor noise to the rotation rate itself, the measured rates of
dependency on the knowledge about a certain magnetic field, the IMU ω bimu are not the same as the actual rates of the
the algorithm does not use additional magnetometers. Special body ω bnb . Thus, each measured rate of the IMU is modeled
attention is given to the roll and pitch motion, as these as the sum of the actual body rate, an offset term, and random
influence the vertical motion of a crane during subsea lifting sensor noise. The resulting rotation rates measured through
operations as depicted in Fig. 1. Exact knowledge of the the IMU are given by
yaw motion is not required for active heave compensation
ω bimu = ω bnb + ρ + ξω , (2)
systems as it does not result in a vertical motion of the crane
tip that must be compensated. To estimate the roll and pitch T  T
where ρ = [ρx ρy ρz ] and ξω = ξωx ξωy ξωz are the
motion, the rotation rate signals of the gyroscopes are fused offset term and sensor noise of the gyroscopes.
with the accelerometers using an EKF. To guarantee good The dynamic behavior of the actual body rates ω bnb itself is
performance also in situations when the IMU is accelerated, assumed as a first order system with additive white noise as
the covariance matrices of the EKF are adapted online unknown input as it is proposed by Kim and Golnaraghi [7].
using the absolute values of the measured accelerations and Thus the ODE for each body rate may be expressed as
rotation rates. Furthermore, it is shown that a virtual sensor 1
for the yaw angle can be used to aid the estimated yaw angle ω̇x = τω ωx + ζωx ,
and to stabilize it around zero to avoid an unbounded drift 1
ω̇y = τω ωy + ζωy , (3)
of this angle resulting in numerical problems. ω̇z = 1
τω ωz + ζωz
II. SENSOR MODELING with the time constant τω > 0 and the additive white noise
In the following a dynamic model for an orientation sensor ζω . The time constant τ ω reflects the expected dynamics
is derived. The sensor used in this work is a standard low-cost of the rotation rates of the body. In the following, a very
strapdown IMU (ADIS 16365) from Analog Devices. The slow dynamic is assumed for the offset term in (2). Hence,
IMU consists of three rate gyros and three accelerometers each gyro offset is modeled as a random walk process with
in orthogonal triads. The objective of the model is to design additive white noise, yielding the following expression:
an observer estimating the pitch and roll angles of a ship or ρ̇ = ζρ , (4)
vessel during subsea lifting operations without requiring any  T
ship specific parameters. Thus it is desired that the model where ζρ = ζρx ζρy ζρz denotes the offsets’ white noise.
only consists of sensor specific parameters or dynamics. It is clear that a direct integration of the measured rates
As depicted in Fig. 1, there are two main frames used in the ω bimu using (1) would result in a drift of the quaternion
following. The navigation frame, denoted by n, is defined as qbn , due to the gyro offsets. Therefore, the rate sensors of
a North-East-Down frame (NED-frame) with the vertical axis the IMU have to be aided by an additional sensor for a
directing toward the center of the earth (down). Because the drift-free integration of the quaternion. Since it is desired
ship stays at a fixed position during subsea lifting operations, that the IMU acts as a standalone sensor for the vessel’s
the n-frame is fixed to the earth. Furthermore, the effects due attitude estimation, the accelerometers of the IMU are used
to the rotation of the earth are considered as negligible, since to compensate the gyro offsets. The basic principle of aiding
these effects are in the range of the sensor noise; hence, the rotation rate sensors with accelerometers is the fact that the

2412
accelerometers measure only the earth’s gravitation in the b- great yaw motion during subsea lifting operations, it can be
frame during phases when the IMU is not accelerated by assumed that the relative yaw angle oscillates around zero.
other forces. Furthermore, the earth’s gravitational vector Hence, a virtual sensor signal is used in the following to
measured by the IMU always directs toward the negative reflect the property that the yaw angle oscillates around zero.
z-axis in the n-frame. Thus the measured accelerations in The output equation of the virtual sensor is given by (8) with
the b-frame during phases when the gravitational vector only Ψ = 0.
acts on the IMU may be written as To design an observer, the derived sensor model and the
T relationship for the virtual sensor have to be transformed
abimu = Cbn [0 0 − g̃] + ξa , (5)
to
 ba state space
T model. Defining the state vector as x =
where g̃ = g denotes the absolute value of the earth’s qn ω bnb ρ and neglecting the IMU’s sensor noise in (2)
 T
gravitation and ξ a = ξax ξay ξaz is the additive sensor and (7) as well as the process noise in (3) and (4), equations
noise of the accelerometers. The transformation matrix C bn (1) - (4), (7), and (10) can be transformed to state space form
from the n- to the b-frame using the quaternion representa- given by
tion is given by (see [10])  1
− 2 (x2 x5 + x3 x6 + x4 x7 )

 2 1
 2 (x1 x5 − x4 x6 + x3 x7 ) 
q0 + q12 − q22 − q32 2q1 q2 + 2q0 q3  1
 2 (x4 x5 + x1 x6 − x2 x7 ) 

b
Cn =  2q1 q2 − 2q0 q3 q02 − q12 + q22 − q32  1 
2q1 q3 + 2q0 q2 2q2 q3 + 2q0 q1 − 2 (x3 x5 − x2 x6 − x1 x7 )
(6)  1 
τω x5

2q1 q3 − 2q0 q2
 
ẋ = f (x) =  1
, (11)
2q2 q3 − 2q0 q1  . τω x6
 
 
1
q02 − q12 − q22 + q32 τω x7
 
 
 0 
Hence, using (5) and (6), the measured accelerations of the 
 0


IMU due to the gravitational vector may be written as 0
abx,imu = (2q0 q2 − 2q1 q3 )g + ξax , 
(2x1 x3 − 2x2 x4 )g

aby,imu = −(2q0 q1 + 2q2 q3 )g + ξay , (7)  −(2x1 x2 + 2x3 x4 )g 
(−x21 + x22 + x23 − x24 )g
 
abz,imu = (−q02 + q12 + q22 − q32 )g + ξaz .
x5 + x8
 
y = h(x) =  . (12)
Due to a higher clearness of Euler angles compared to x6 + x9
 
 
a quaternion representation, Euler angles are used in the x7 + x1 0
 
 
following to depict the results and explain certain effects 2(x2 x3 +x1 x4 )
x21 +x22 −x23 −x24
of the proposed algorithm. In this research, the order of
successive rotations expressing the transformation from the Note that the last row of the system’s output vector y 7
n- to the b-frame with Euler angles is defined as yaw-pitch- stands for the virtual sensor signal with the corresponding
roll [10] yielding the following relationship: measurement for the observer’s correction always set to
tan(Ψ = 0) = 0.
2(q1 q2 + q0 q3 )
tan(Ψ) = , (8) Global observability of the proposed state space model can
q02
+ q12 − q22 − q32 be proven with some simple algebraic calculations. However,
sin(Θ) = −2(q1 q3 − q0 q2 ), (9) it can be shown that the observability of the yaw angle gets
2(q2 q3 + q0 q1 ) lost, if the virtual sensor signal y 7 is not used in the system’s
tan(Φ) = 2 . (10)
q0 − q12 − q22 + q32 output vector. Thus it is expected that the yaw angle gets
estimated with an error, despite the usage of a virtual sensor
Here, Ψ, Θ, and Φ denote the yaw, pitch, and roll angles
signal.
around the z, y, and x axis of the respective coordinate
For real-time implementation of the observer, the described
systems.
state space model in continuous time has to be discretized in
III. OBSERVER DESIGN time. The discretization is performed using the Euler-forward
The derived dynamical model of the gyroscope sensors method. The observer itself is realized as a standard EKF.
can be used to design an observer estimating the attitude In the following R and Q denote the covariance matrices of
of a ship. Obviously, the objective of designing an observer the sensor noise and the process noise, respectively. Theses
that estimates the ship’s roll and pitch angle without any matrices are chosen as diagonal matrices and given by
drift can be achieved using the properties of the earth’s R = diag (rω rω rω ra ra ra rΨ ) , (13)
gravitational vector (7). However, a drift-less estimation
Q = diag (qq qq qq qq qω qω qω qρ qρ qρ ) . (14)
of the yaw angle is not possible using (7), since a yaw
motion does almost not affect the direction of the earth’s In (13) rω and ra denote the sensor noise of each rotation
gravitational vector in the b-frame. To avoid an unbounded rate sensor and each accelerometer of the IMU, while r Ψ
drift of the yaw angle that results in numerical problems, an is the sensor noise of the virtual sensor signal for the yaw
additional relation is required. Since a ship does not perform angle. In (14) q q , qω , and qρ are the process noise of the

2413
four elements of the quaternion, each rotation rate, and each 0.6

corresponding offset term. 0.4

The elements of R are used in the following to reflect
0.2
uncertainties of different sensor signals. The elements r ω

ρ̂i (t) [◦ ]
and rΨ are chosen to be constant with r Ψ ≫ rω assigning 0

more uncertainties to the virtual sensor than to the rotation -0.2

rate sensors. As already mentioned, the output equations
ρ̂x
-0.4
of the accelerometers (7) are only valid during phases ρ̂y
ρ̂z
nom.
when the IMU is only affected by the earth’s gravitational -0.6
0 100 200 300 400 500
vector. On the other hand, this means that the measured t [s]
accelerations are very uncertain if the IMU is accelerated by
Fig. 2. Estimated offsets ρ̂x , ρ̂y , and ρ̂z of the rotation rate sensors together
other external forces. To take these situations into account, with their nominal values during simulation.
the accelerometers’ sensor noise r a is adapted to different
situations as proposed by Favre et. al [11]. However, to
avoid hard switches between different parameter values as given. After that measurement results from an experimental
it was done by Favre, r a is adapted linearly to situations setup are presented.
when external forces act on the IMU. Such situations are
characterized by the difference of the absolute value of the A. Simulation
earth’s gravitational vector g̃ and the absolute value of the The motion sequence used for simulation is obtained from
measured accelerations ã = abimu  as well as the absolute a six degrees of freedom simulation of a ship [12]. The model
value of the measured rates ω̃ = ωbimu , if the IMU is not used in the following corresponds to a ship that is 82.8m
located in the center of gravity of the ship. Thus, r a is given long and has 6,362 tons of weight. The model is excited
by with a JONSWAP spectrum with a significant wave height of
4m. The resulting rotation rates in the b-frame are disturbed
ra = r a + βa αεa (|ã − g̃|) + βω αεω (ω̃) , (15) with a constant offset and sensor noise; the translational
where r a is the minimum value of r a during non-accelerated accelerations are also disturbed with sensor noise.
situations, βa , βω > 0 are the linear adaption coefficients, It is clear that the performance of the attitude estimation de-
and αε (η) is defined by pends mainly on the correct offset estimation of the rotation
 rate sensors. Fig. 2 shows the estimated offsets of all three
η − ε,
 if η > ε sensors in the b-frame together with the nominal ones used
αε (η) = 0, if − η ≤ ε ≤ η (16) in the simulation. The initial values for the estimated offsets
ρˆx (0), ρˆy (0), and ρˆz (0) are obtained from an initialization

η + ε, if η < −ε

phase at the beginning. During this phase the rotation rate
to generate a dead band where r a is not adapted. The sensor’s mean values are taken over a certain time horizon
parameters βa and βω are obtained from experiments and and used to initialize each state to reduce the observer’s
chosen in a way that r a ≫ ra holds during phases when decay time. In the following ˆ• denotes the estimated values
the IMU is accelerated by forces other than the earth’s of a certain state. As can be seen, the estimated offsets ρ̂
gravitation. converge to the nominal offsets after a short decay time.
The elements qq and qρ of the diagonal matrix Q given in Furthermore, the estimated offsets are almost constant after
(14) are selected to be constant with q q ≫ qρ , as the offset they reached their nominal values. This also holds for the
terms are almost constant compared to the quaternion. The offset of the z-axis ρ̂ z , which can be attributed to the virtual
process noise of the rotation rates q ω is equal to the unknown sensor for the yaw angle. Since the performance of the offset
white noise parameters of (3). To guarantee a fast dynamic estimation is good, it is expected that the ship’s attitude is
of the states corresponding to the rotation rates during fast also estimated correctly.
rotations, the process noise q ω is adapted to the absolute Fig. 3 compares the estimated Euler angles Φ̂, Θ̂, and Ψ̂ with
value of the measured rates. It can be expressed as the simulated ones for the same simulation as in Fig. 2. For
a better depiction, the figure only shows a time slot of the
qω = q ω + κω αεω (ω̃) (17)
whole sequence after the initial errors are decayed. Figs. 3a
with the minimum value q ω and αε (η) as defined in (16). The and 3b show that the error of the estimated pitch angle Θ̂
linear adaption coefficient κ ω is obtained from experiments. is slightly higher than the error of the estimated roll angle
Φ̂. That is due to the fact that the absolute pitch motion
IV. RESULTS are bigger than the absolute roll motion. Nevertheless, both
In the following the proposed observer for ship’s attitude errors are in an acceptable range. Fig. 3c indicates that
estimation is evaluated. Special attention is given to the the yaw angle gets also estimated with an tolerable error,
performance of estimating the roll and pitch angle, as these however the performance is not as good as the performance
angles are required for active heave compensation during for the roll and pitch angles. The maximum errors and
subsea lifting operations. First, some simulation results are the Root Mean Square (RMS) errors for each axis are

2414
 6 0.8
4 ρ̂x
0.6 ρ̂y
2 ρ̂z

Φ [◦ ]
0
0.4
-2

ρ̂i (t) [◦ ]
-4 0.2
-6
350 400 450 500 550 0
t [s]
(a) Φ and Φ̂. -0.2

-0.4
0 200 400 600 800 1000 1200
6
t [s]
4
2
Fig. 5. Estimated offsets ρ̂x , ρ̂y , and ρ̂z of the rotation rate sensors
Θ [◦ ]

0
-2 obtained from the experimental setup.
-4
-6
350 400 450 500 550
t [s]
(b) Θ and Θ̂. TABLE II
E STIMATION ERRORS FROM EXPERIMENTAL SETUP
6
4
2 Φ [◦ ] Θ [◦ ] Ψ [◦ ]
Ψ [◦ ]

0 max. error 0.39 0.29 0.44

-2
-4 RMS error 0.14 0.10 0.15
-6
350 400 450 500 550 max. amplitude 8.36 5.13 1.68
t [s]
(c) Ψ and Ψ̂.
Fig. 3. Simulated (dashed line) and estimated (solid line) Euler angles
together with their estimation errors (dotted line).

TABLE I
paths for all six degrees of freedom of the KUKA robot are
S IMULATED ESTIMATION ERRORS
generated with the same ship model used for simulation. The
simulated motion are recorded and used afterward to move
Φ [◦ ] Θ [◦ ] Ψ [◦ ] the KUKA robot. The measured IMU signals are used for
max. error 0.16 0.42 0.37 an online estimation of the IMU’s attitude with the proposed
RMS error 0.06 0.16 0.18 method. The estimated Euler angles are compared to the
max. amplitude 3.03 5.55 1.04 angles obtained from the incremental encoders of the KUKA
robot denoted as reference signals in the following. Fig. 4
shows the complete setup.
Fig. 5 presents the estimated offsets ρ̂ of all three rotation
rate sensors. In contrast to the simulation results, the nominal
values of the offsets are unknown. Thus Fig. 5 only shows
the estimated values. However, it indicates that after a short
decay time every offset oscillates around a constant value, as
it is expected from simulation. In addition, the figure clearly
KUKA illustrates the effect of the virtual sensor for the yaw angle.
The offset of the z-axis ρ̂ z does not diverge away.
Fig. 6 illustrates the corresponding Euler angles for 900s ≤
IMU t ≤ 1100s. Figs. 6a and 6b obviously show that the estimated
roll and pitch angles are in good accordance with the
reference signals, while Fig. 6c depicts that the yaw angle
does not diverge and oscillates around zero. Again, Tab. II
Fig. 4. Experimental setup consisting of a KUKA robot and an IMU from
Analog Devices (ADIS 16365).
summarizes the maximum error values, the RMS errors and
the maximum amplitudes of the motion sequences for each
axis. From the results follows that the estimation error of
summarized in Tab. I together with the maximum amplitude the roll and pitch angles stay beyond acceptable values. The
of the corresponding motion sequence. error for the yaw axis is relatively higher; however the correct
estimation of the yaw angle is not required for active heave
B. Experimental setup compensation systems as mentioned above.
The experimental setup used to evaluate the algorithm for The adapted values for the sensor noise of the accelerometers
attitude estimation consists of an IMU from Analog Devices ra and the process noise of the rotation rates q ω as given in
(ADIS 16365) and a KUKA robot. The accelerometers of (15) and (17) are demonstrated in Fig. 7 for the test sequence.
√ of ±18g and the measurement noise
the IMU have a range The figure shows that q ω is almost constant, while ra changes
is rated at 0.5mg/ Hz. The gyros√have a range of ±150 ◦ /s due to the absolute value of the measured accelerations ã
and a noise rating of 0.044 ◦ /s/ Hz [13]. The reference which is given in Fig. 8.

2415
 10
to these motion. The derived approach fuses the rotation
5
rates of an IMU with its accelerometer measurements to
Φ [◦ ]
0
-5
estimate the vessel’s attitude. To stabilize the unobservable
-10
yaw motion around zero, a virtual sensor signal for the yaw
900 950 1000 1050 1100
t [s] angle is applied. The parameters of the EKF are adapted
(a) Φ and Φ̂. online and depend on the measured absolute values of the
accelerometers and rotation rate sensors.
10
5
Through simulation and measurement results, the attitude
estimation algorithm was validated. It was shown that the
Θ [◦ ]

0
-5
estimated vessel’s roll and pitch motion are in good accor-
-10 dance with the reference signals. Furthermore, the estimation
900 950 1000 1050 1100
t [s] error of the yaw angle was also in an acceptable range.
(b) Θ and Θ̂.
R EFERENCES
10
[1] S. Küchler, T. Mahl, J. Neupert, K. Schneider, and O. Sawodny,
5 “Active Control for an Offshore Crane Using Prediction of the Vessel’s
Ψ [◦ ]

0 Motion,” IEEE/ASME Transactions on Mechatronics, vol. 16, no. 2,

-5 pp. 297–309, 2011.
[2] K. Do and J. Pan, “Nonlinear control of an active heave compensation
-10
900 950 1000 1050 1100 system,” Ocean Engineering, vol. 35, no. 5-6, pp. 558 – 571, 2008.
t [s] [3] S. Messineo and A. Serrani, “Offshore crane control based on adaptive
(c) Ψ and Ψ̂. external models,” Automatica, vol. 45, no. 11, pp. 2546 – 2556, 2009.
Fig. 6. Estimated Euler angles (solid line) and the corresponding references [4] J.-M. Godhavn, “Adaptive tuning of heave filter in motion sensor,” in
(dashed line) obtained from the experimental setup. The estimation errors Proceedings of OCEANS, Nice, 1998, pp. 174 – 178.
are also depicted (dotted line). [5] ——, “High quality heave measurements based on GPS RTK and
accelerometer technology,” in Proceedings of MTS/IEEE OCEANS,
vol. 1, Providence, RI, 2000, pp. 309 – 314.
1.5 [6] T. Fossen and T. Perez, “Kalman filtering for positioning and heading
control of ships and offshore rigs,” IEEE Control Systems Magazine,
1
qω [−]

vol. 29, no. 6, pp. 32 – 46, 2009.

0.5 [7] A. Kim and M. Golnaraghi, “A quaternion-based orientation estimation
algorithm using an inertial measurement unit,” in Proceedings of
0
900 950 1000 1050 1100 Position Location and Navigation Symposium, Monterey, CA, 2004,
t [s] pp. 268 – 272.
(a) qω . [8] N. Metni, J.-M. Pflimlin, T. Hamel, and P. Soures, “Attitude and gyro
bias estimation for a vtol uav,” Control Engineering Practice, vol. 14,
3000 no. 12, pp. 1511 – 1520, 2006.
2500 [9] P. Setoodeh, A. Khayatian, and E. Frajah, “Attitude estimation by
ra [−]

2000
separate-bias kalman filter-based data fusion,” The Journal of Navi-
gation, vol. 57, no. 2, pp. 261–273, 2004.
1500 [10] D. Titterton and J. Weston, Strapdown Inertial Navigation Technology,
1000
900 950 1000 1050 1100
2nd ed., N. Stewart and H. Griffiths, Eds. Stevenage, United
t [s] Kingdom: Institution of Electrical Engineers, 2004.
(b) ra . [11] J. Favre, B. Jolles, O. Siegrist, and K. Aminian, “Quaternion-based
fusion of gyroscopes and accelerometers to improve 3d angle mea-
Fig. 7. Adapted sensor noise of the accelerometers ra and adapted process surement,” Electronics Letters, vol. 42, no. 11, pp. 612–614, 2006.
noise of the rotation rates qω used for the EKF. [12] MSS, “Marine systems simulator,” http://www.marinecontrol.org,
2010, viewed 20.11.2010.
10.5 [13] ADIS 16365, Six Degrees of Freedom Inertial Sensor, Analog Devices,
2009.

10
 
ã sm2

9.5

9
900 950 1000 1050 1100
t [s]

Fig. 8. Absolute value of the measured accelerations ã used to adapt the

sensor noise ra .

V. CONCLUSION
This paper has presented an attitude estimation approach
for ships or vessels used during subsea lifting operations.
Since active heave compensation systems require actual roll
and pitch motion of a vessel, special attention was given

2416

View publication stats