You are on page 1of 17

M.-C. Fang et al.

/ Ocean Engineering 54 (2012) 1725


Ocean Engineering 54 (2012) 1325

13

Contents lists available


SciVerse
at
ScienceDirec
t

Ocean Engineering
journal homepage:
www.elsevier.com/locate/oceanen
g

Applying the PD controller on the roll reduction and track keeping for the ship
advancing in waves
Ming-Chung Fang a,n, Yu-Hsien Lin b, Bo-Jhe Wang a
a

Department of Systems & Naval Mechatronic Engineering, National Cheng-Kung University, Tainan City 70101, Taiwan b Fishing Boat

& Marine Engineering Research Center, National Cheng-Kung University, Tainan City 70101, Taiwan

articleinfo

abstract

Article history:
Received 1 January 2012
Accepted 1 July 2012
Available online 26 July 2012

This study aims to develop a PD control system based on the neural network algorithm to reduce the ship rolling motion in the
desired track through rudder and fin actions. The mathematical model including sea-keeping and maneuvering characteristics
is developed in the present paper and the

Keywords:
PD control
Neural network
Rudder
Stabilizer fin

nonlinear time history of ship motions is solved by the fourth order RungeKutta method. In order to
achieve the purpose of roll reduction and track keeping, the rudder and fin stabilizer are used as the control tools for the ship
advancing in the seaway. In addition, the PD controller based on the selftuning neural network algorithm is applied to achieve
the goals of multi-input and multi-output in the control system. Four different types of control modes on a container ship
model are studied and the performances are investigated in different sea states. The results indicate that the PD controller
based on the self-tuning neural network algorithm applying to the stabilizer fin control for roll reduction and the rudder control
for track keeping in the seaway would be suggested.
& 2012 Elsevier Ltd. All rights reserved.

1. Introduction
While a ship sails in the seaway, the external forces, i.e. winds, waves or
ocean currents, usually cause the variations on the motions and track.
Therefore, some devices need to be applied to the ship to maintain the
stability and orientation.
Usually the roll motion is considered to be the most severe problem in
safety while the ship sails in the seaway. Simply speaking, the serious roll
motion would affect the ship stability, comfort and efficiency of crews,
accuracy of electrical mechanism, and ship course. Therefore, roll-reduction
is especially advantageous for ships advancing in waves. Some devices have
been well employed to accomplish the ship roll reduction, e.g. bilge keels,
anti-roll tanks, gyroscopic stabilizers, moving weights and stabilizer fins
(Treakle et al., 2000). Additionally it is also well known that the rudder can
be applied to control the ship heading, trackkeeping and even the rollreduction. In this study, the rudder and the stabilizer fin will be adopted.
Although the rudder is mainly applied to alter the ship course, it also can
produce an additional roll moment affecting the roll motion and
consequently may be used as the roll-reduction device (Fang, 1991; Fossen,
1994). In addition to the rudder, the stabilizer fin might be the most effective
and widely adopted rollreduction device for ships in waves. However, in
order to apply

Corresponding author. Tel.: 886 6 2747018x211; fax: 886 6 2080592. E-mail address:
fangmc@mail.ncku.edu.tw (M.-C. Fang).

0029-8018/$ - see front matter & 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.oceaneng.2012.07.006

the stabilizer fin to reduce the ship roll motion, the ship speed must reach the
adequate value to make the stabilizer fin function more efficient (Jerrold and

Michael, 1999). Generally, both the rudder and the stabilizer fin need to
incorporate the automatic controller to achieve the optimal roll-reduction for
the ship traveling in rough seas. The previous literature (Jerrold and Michael,
1999) presented a mathematical model consisting of roll-sway-yaw motions to
discuss the roll-reduction efficiency for the stabilizer fin and the rudder by
adopting the Separate Control and Compact Control respectively.
How to select a suitable controller for the stabilizer fin and the rudder to
reduce the ship rolling motion in waves is an important technology. Several
kinds of controllers were already developed in the past, e.g. PD and PID
controls (Dove and Wright, 1991), adaptive control (Tzeng and Lin, 2000),
fuzzy control (Cao and Lee, 2003) and sliding mode control (Fang and Luo,
2007) etc. Among them, either PD or PID controller with fixed-design
parameters is a conventional autopilot system for a ship steering in waves,
which is the most widely used among many autopilot systems because of its
simplicity. In the past, the control gain parameters adopted in the PD or PID
controller were usually determined based on the experience of the operator,
trial-anderror method or experiments, which might not be optimal. Therefore,
some approaches would be needed to promote the efficiency of the controller to
reduce the ship roll motion, for example, the neural network algorithm can
handle such kind of
problem very well. The previous study (Hemerly and Nascimento, 1999) used
the neural network adaptive control algorithm to train the controller and
simulate control parameters for PID controller. The authors (Li et al., 2005)
applied the PID controller to the ship stabilizer fin based on the wavelet neural
network identification and tuning, which can improve the effectiveness and the
performance of the controller. Our former research (Fang et al., 2010) also
applied the PID controller to handle the stabilizer fin to reduce the ship rolling
motion by incorporating the self-tuning neural network and good performances
were obtained.
According to the previous researches (Fang and Luo, 2006, 2007; Fang et
al., 2010), the 6-DOF mathematical model including sea-keeping and

14
M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725
maneuvering characteristics for a ship steering in waves can be treated
water. The roll damping coefficient K f_ can be computed from the empirical
separately in order to simplify the complicated coupling problems. Based on
formula derived by the previous study (Hamamoto et al., 1994; Takahashi,
this model, a PD control system incorporating the self-tuning neural network
1969). The related nonlinear terms for maneuvering derivatives can be found
algorithm is applied to the stabilizer fin and the rudder for roll reduction and
from the references (Hirano and Takashina, 1980; Inoue et al., 1981b).
track keeping. In the following section, the control system incorporated in the
The FroudeKrylov forces are calculated by integrating the incident
mathematical model for the ship maneuvering in waves is described. Different
wave pressure acted on ship hull to the wave surface, and the diffraction
wave conditions and the dynamical environments are adopted to investigate the
forces are calculated with the linear strip theory and source distribution
efficiency of this control system.
method. These forces depend on the ship configuration below the
instantaneous wave surface that must be calculated at any instant of time
(Fang et al., 1993; Fang and Luo, 2008). The terms Ipp, Jpp, QF, QP, n in Eq.
2. Mathematical model
(7) represent the moment of inertia of propeller-shafting system, the
propeller added moment of inertia, the propeller torque, the main engine
The nonlinear 6-DOF equations of motions (Fang and Luo, 2006; Fang et
torque and the rpm of propeller, respectively. The rudder forces can be
al., 2010) including the control forces of the stabilizer fin and the rudder are
referred to Hirano (1980).
developed and shown in Eqs. (1)(6). In addition, the engine torque equation is
The stabilizer fin-induced force and moment can be derived and listed as
also shown simultaneously in Eq. (7).
follows,
c_
c_
y_
XFF FRDFLD
8
mu_n myoeXnc_ n mxoeu_mzoew

Surge :
1 XFKo0XRF XFF T1tpR
9
YFF FRLcosfFFLLcosfF

10
c_
c_
c
c_
ZFF FRLsinfF FLLsinfF
mn_ u mxoeu myoen_YnnYc Yc_ Yn9n9n9n9

11
KFF zFFRLcosfFFLLcosfFyFFRLsinfFFLLsinfF
Sway :
Y _ n9c_ 9Yc_ 9c_ 9c_ 9c_ 9YFKo0YDFoeYRFYFF

n9c9
12
MFF zFXFF

2
13
Heave : y oeyZy_ oey_Zyoey mw_ mzoew_ ZwoewZ
NFF xFYFF

ZFKo0ZDFoeZFF mg
where the subscripts RL, RD, LL and LD represent the fin-induced lift force
(L) and drag force (D) on the starboard side (R) and port side (L),
3
respectively. xF, yF and zF are the coordinates of the action center of the fin
force. fF is the angle between the fin and the plumb line. The related
Ixxf Ixxy_c_ Jxxoey_c_ Jxxoef Kf_ f_ YnnYc_ c_ zH Roll
equations of the lift force and the draft force can be also found with respect
:
to the reference (Lee, 1977).
KFKo0KDFoeKRF KFF
4
3. Control system
Iyyy Ixxc_ f_ J

The PD controller is usually expressed in the following form containing two


control gains, i.e. proportional (KP) and derivation

o f_ c_ JyyoeyMy_ oey_

Pitch :

xx e

MyyMw_ oew_ MwoewMFKo0MDFoeMFF

(KD)
5

dDut
rt KpDutKD

Izzc Ixxy_f_ Jxxoey_f_ Jzzoec Nn_n_Nnn Nc_ c_ Nc_ 9c_ 9c_


9c_ 9Nnnc_ n2c_ Nnc_ c_ nc_ 2NffNn9f9n9f9 Yaw :
Nc_

9f9

c_ 9f9YnnYc_ c_ Yn9n9n9n9Yn9c_ 9n9c_ 9Yc9c_ 9c_ 9c_

9xH NFKo0NDFoeNRFNFF

6
Engine : 2pIpp Jppn_ QFQP

where surge, sway and heave velocities are represented by u, v and w,


respectively, whereas roll, pitch and yaw displacements are represented by f, y
and c, respectively. R is the ship resistance. The subscripts FK, RF, FF and DF
represent Froude Krylov force, rudder force, stabilizer fin force and
diffraction force, respectively. Since the strip theory is applied, the diffraction
force for surge, XDF, cannot be calculated, however, it is generally small due to
the slenderness of the ship. In Eqs. (3) and (5), the corresponding
hydrodynamic coefficients with respect to heave and pitch can be found in
reference (Kim et al., 1980), and they are calculated by Frank close-fit method.
mx, my and mz represent the added masses with respect to the x, y and z axes,
respectively, whereas Jxx, Jyy and Jzz represent the added moments of inertia with
respect to the x, y and z axes, respectively. The maneuvering derivatives of
sway and yaw motions, i.e. Yn, Yn9n9, Nc and Nn9f9, etc., are estimated by
empirical formulas (Inoue et al., 1981a) based on the model experiment in calm

14 dt

where r(t) is the control output variable and Du(t) is the system error.
According to the different objective missions with respect to different
control tools, four types of control commands applied to the stabilizer fin
and rudder are introduced in the following,
A-type: the command to control the rudder for the track keeping.
B-type: the command to control the rudder for the track keeping and roll
reduction.
C-type: the command to control the rudder for the track keeping and the
stabilizer fin for roll reduction.

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

15
D-type: the command to control the rudder for the track keeping with roll
reduction, and the stabilizer fin for roll reduction.
In Fig. 1, the illustration shows the desired wave heading and the track along
with the instantaneous ship position. The deviation between the instantaneous
ship position and the desired path can be calculated by Eq. (15).
X0Utan cY0
E
0

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

15

tan c1
2

where (X0,Y0) represents the instantaneous position of a ship in the earth-fixed


coordinate, and c indicates the desired ship heading. Generally, the line-of-sight
(LOS) guidance technique (McGookin et al., 2000) is introduced to guide the
ship to sail to the commanded waypoint. The relationship between the actual
heading (cr) and desired heading (c) is expressed as follows.
c

ce r c
Fig. 1. The illustration of the ships track in the earth-fixed coordinate system.

heading error.

Fig. 2. The flow charts of the rudder control system for (a) A-type; (b) B- type; and (c) the stabilizer fin control for C- and D-modes.

16
where ce is the

16
3.1.

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725


Ship

The A-type and B-type (i.e. control commands on rudder)

In this section, A-type and B-type control systems would be investigated.


The PD controller with A-type command is applied to deflect the rudder angle
to maintain the track keeping only whereas the B-type command is to control
the track keeping and roll reduction simultaneously. The flow charts based on
the neural network algorithm for A-type and B-type are shown in Fig. 2(a) and
(b), respectively.
Based on the Eq. (14), the PD controller commands for the A-type can be
c
c_
written as dt KspE0t1Ksdvt1Kyp et1Kyd t1 17 where d(t)
is the rudder angle. Ksp, Ksd, Kyp and Kyd are track deviation gain, sway velocity
gain, heading error gain and yaw rate gain, respectively. Correspondingly,

model with one hidden layer is adopted in NN1. E 0(t1), v(t1), ce(t1) and c
^

Ek gktg^kt

Breadth (m)

30.2

Depth (m)

16.6

Draft (m)

FP 11.0

Metacentric height (m)


Height of C. G. (m)
2

1734.43

Waterline coefficient

0.8561

Rolling period (s)

18.2

Pitch ratio

8.4

Rudder area (m2)

40.42

Longitudinal distance from rudder to C. G. (m)

90.0

Vertical distance from rudder to C. G. (m)

8.0

Maximum deflection angle (deg.)

Because roll, yaw and sway motions can be positive or negative, the
hyperbolic tangent function is therefore adopted as the activation function, i.e.

Maximum deflection rate (deg./s)

f 0x 0:51fx2

1.0

Rudder height (m)

where gk (k1, 2, 3 and 4) implies E0, v, ce and c , respectively.

And the derivative of the activation function in NN1 is thus evaluated as


below

7.45

Rudder

^_

19

12.08

Hull side area (m )

Propeller diameter (m)

18

fx tanh 0:5x

Propeller

t1 are selected as the input nodes, whereas E 0t, v^t, c et and c


t are set to be the output nodes of the output layer. The error function for
NN1 is described as below
2

185.5

AP 11.0

E0(t1), v(t1), ce(t1) and c t1, are track deviation, sway velocity, heading
error and yawing angular velocity, respectively.
In order to obtain the optimal PD controller, the self-tuning PD controller
based on the neural network theory is presented here, i.e. Fig. 2. The scheme
consists of two neural networks: (i) system identification neural network (NN1)
and (ii) parameter selftuning neural network (NN2).
The NN1 is used to identify and predict the nonlinear dynamic relationship
between input signals and output signals. In the present case, the series-parallel

Length (m)

735
3.5

Main engine
MCR (PS)

33,760

100% RPM

91

20

Based on the NN1 activation function in Eq. (19) and its derivative in Eq.
(20), the following sensitivity information, @gk/@d(t1), can be obtained:
@ g^k

d t

1 4 X1g^k2Vj1Hj2Wj, k 1,2...4 21

@j
where Wj is the weight between the input node and the jth hidden node, and V j
is the weight between the jth hidden node and the output node. H j is the output
vector of the hidden layer.
Subsequently, the neural network NN2 is applied to tune the
PD control gains. The error function for NN2 is defined as
Ek 12gkt1gkt2
2

k 1,2...4

22

Dgkt1
Because the control gains are all positive, the sigmoid function with domain
between 0 and 1 is selected as the activation function, i.e. Eq. (23), and its
derivative is shown in (24).
1
fx

Table 2
The dimension of the stabilizer fin.

23 e

f 0x fx1fx
Table 1
Principal dimensions of the container ship.

24

Span (m)

5.77

Fin area (m2)

15

Aspect ratio

2.22

Maximum rate (deg./s)


Maximum sway angle (deg.)

4.8
724

Longitudinal distance from midship (m)

21.4

Transverse position from midship (m)

17.7

Vertical position from midship (m)

10.17

Angle with respect to plumb line (deg.)

72

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

Fig. 3. (a) The traverse profile of stabilizer fin; and (b) the longitudinal profiles of the rudder and
stabilizer fin.

Consequently, the output values of the NN2 are all between 0 and 1.
Therefore, some proportional factors must be multiplied to meet the
practical control requirement. The proportional factor depends on the system
characteristics and the sampling time. Since the objective values of the
output layer are unknown, the signal value of network error needs to be
further derived and stated in the following. First, the output node, Z k, is
defined as
8 Ksp, k 1
>>>>< Ksd, k 2
Zk

Kyp, k 3

25

17

Table 3
The cost function values of ship rolling angle at different wave heading.
Wave type

Controller

301

751

901

Long-crested waves

A-type

44,045.15
51,079.36
17,613.73
M.-C. Fang et al. / Ocean
Engineering 54 (2012)
1725

451

601

23,722.38

1903.59

B-type

28,930.60

34,274.73

15,708.16

23,820.51

1883.25

C-type

3513.49

9249.82

5630.16

8423.61

1164.88

D-type

3530.43

10,354.45

5339.55

8686.74

1184.15

COST

18

500000

400000

300000

200000

10000
0

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

>>>>: Kyd, k 4

19

20

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

dt^_ 1 14 X
1f

The error signal values for the output nodes is

dk Dgkt1U

@gkt U@dt1UZk1Zk,
k 1,2...426 @dt1
@Zk

where f

gkt1,

k 1,2...4

27

and f

^_

29

are set to be the output nodes of the output layer.

Consequently, two more error functions for NN2 should be included in Eq.
(22), i.e.

From Eqs. (17) and (26), the item @dk(t1)/@Zk in Eq. (26) can be obtained
as below
@dkt1

^_ 2Vj1H2j Wj

Ek gkt1gkt2
, k 5,6

@Zk
The approximate value of @gk(t)/@d(t1) can be obtained from
@g^kt=@dt1, which are offered from NN1.
Similar to A-type, two more control gains on the rolling need to be added to
the PD controller for B-type and can be written as dt
c
KspE0t1Ksdvt1Kyp et1
Kyd

c_

f_
t1Krrpft1Krrd t1

28

Dgkt12
_

where g5 and g6 implies f and f , respectively.


Subsequently, the more output node, Zk, (k5, 6) can be rewritten as

respectively. Meanwhile, f(t1) and f t1denote the rolling angle and the
rolling angular velocity at time t1, respectively.
The two more sensitivity information with respect to the rudder control on
the rolling motion can be expressed as below

d@tf^ 1 14 X
2Vj1Hj2 Wj 1f

( Krrp,

k5

Krrd,

k6

Zk

where Krrp and Krrd represent the roll and roll rate gains by the rudder,
_

30

31

And the related additional error signal values for the output nodes should be
k

U
@dt1

@Zk

where @dk(t1)/@Zk is
g(
@dkt1
ft1, k 5
t1
@gkt
k 5,6
32

D
33

@
j

@f

@Zk
A-type
B-type
C-type
D-type

f_ t1,

k6

Time (sec)
Fig. 5. The ship responses for self-tuning neural network based PD controller in
long crested waves with H 1=3 3 m, T 10 s, initial wave heading m301 and initial ship speed

V23 knots.

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

Fig.6. The simulations of (a) trajectory and (b) tracking errors in long crested
waves with H1=3 3 m, T 10 s, initial wave heading m301 and initial ship speed V23 knots.

3.2.
The C-type and D-type control using the rudder and the
stabilizer fin
In order to understand the effect of the control for stabilizer fin on the
roll reduction, more commands than A-and B-types to adjust the stabilizer
fin angle would be added. Specifically, the C-type is used to determine if the
establishment of the stabilizer fin is able to reduce the cost values of control
system, in comparison with the A-type whereas the D-type, comparing with
the B-type, is to examine the effect of the rudder and the stabilizer fin on the
roll reduction simultaneously. The flow chart of the control for the stabilizer
fin is illustrated in Fig. 2(c).
Similar to the A-type one, the commands to control the stabilizer fin in
type-C and type-D can be presented as follows.
f_

aFt KrfpUft1KrfdU t1

34

where a(t) is the control output fin angle. Krfp and Krfd are the roll and roll
rate gains by the stabilizer fin, respectively. The output node, Z k, is then
expressed as
(
Zk

Krfp,

k1

Krfd,

k2

35

The error signal values for the output nodes of the stabilizer fin are:
8< Dft1U@a@Fftt1 U@aF@Ztk 1 UZk1Zk, k 1

dk : Df_ t1U@a@Ff_tt1 U@aF@Ztk 1 UZk1Zk, k 2 36 Finally, for


the C- and D-types, the item @aF(t1)/@Zk is

21

introduced as below
(
@
22aFt1

ft1,

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

37

Ksd

Ksp

Kyd

Kyp

@Zk f_ t1,k 2

k1

Time (sec)
Fig. 7. The control gains on the rudder operation for track keeping in long crested waves with H 1=3 3 m, T 10 s, initial wave heading m301 and initial ship speed V23 knots.

3.3.

23

Krfp

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

The cost function and roll reduction rate

In order to judge the quality of operation, the cost function (McGookin et


al., 2000) is adopted here, i.e.,

The right-hand side of Eq. (38) is related to the track keeping and roll
motion, in which n is the total number of the time interval in the simulation
process, Dci is the ith heading error between desired heading and actual
heading, E0i is the ith track deviation between the desired track and actual
trajectory, di is the ith rudder angle, aFi is the ith stabilizer fin angle and fi is the
ith roll angle. The a1, a2, a3, a4 and a5 are the weights depending on the scale of
each parameter and set to be 10, 2.0, 0.1, 0.2 and 5.0, respectively, in the
present study.

Krfd

CTotal Xn a1Dci2a2E20i a3di2a4aFi2a5f2i 38 i 0

In order to judge the efficiency of the roll reduction rate (RRR) for the
ship maneuvering in random sea using the present PD controller, the
percentage of roll reduction (Fang and Luo, 2008; Jerrold and Michael,
1999) is defined as:
APRCS
RRR %

100
39
AP where AP is the standard deviation of roll angle
calculated from the autopilot controller without roll reduction, whereas RCS
is the one including roll-reduction control.

Krrd

Krrp

4. Results and discussion


In this paper, the container ship used by (Fang and Luo, 2006; Fang et
al., 2010) is adopted for calculations. The principal particulars of the
container ship and the stabilizer fin are listed in Table 1 and Table 2,
respectively. For simplification, the water

24

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

Time (sec)
Fig. 8. The control gains on the rudder and fin operations for roll reduction in long crested waves with H 1=3 3 m, T 10 s, initial wave heading m301 and initial ship speed V23 knots.
Table 4
The total cost function values at different wave heading.
Wave type

Controller

301

451

601

751

901

Long-crested waves

A-type

352,608.55

513,091.81

127,195.45

478,786.00

22,754.77

B-type

410,231.86

434,519.05

134,454.35

514,632.45

28,976.97

C-type

111,555.57

282,228.80

127,516.70

379,594.98

34,667.61

D-type

100,366.86

347,300.88

149,326.15

516,415.50

38,815.07

depth is assumed to be infinitely deep. The 4th order RungeKutta method


is applied to solve the time history simulation of the ship motion response.
The time step for the calculation is set to be 0.1 s. The location profiles of
the stabilizer fin and rudder are illustrated in Fig. 3. The sea state 6 with 3 m
of significant wave
height (H1/3) and 10 s of average period T is considered for the ship
motion simulations in long crested waves.
In order to obtain the suitable convergent weight matrix in the hidden
layer for the neural network PD controller, the sufficient learning time is

needed. In the present study, the case for the ship advancing in long-crested
waves with wave heading m601 is adopted to train the ship to learn how
to handle the roll reduction and course keeping because the maximum roll
motion amplitude occurs in this case (Fang et al., 2010). The convergent
weight matrix is obtained after 10,000 s of learning time and then selected
as the input base for the following neural network based PD controller. This
technique can save lots of learning time for the PD controller based on the
self-tuning neural network algorithm. 4.1. Ship advancing in long-crested
waves with fixed heading

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725


25
long crested waves with H1=3 3 m, T 10 s and initial ship speed V23 knots. The ship heading
shifts from c301 to c01 at t300 s.

In this section, the investigation is focused on the simulation of ship


steering in random seas. The course adopted here is a straight line guiding
the ship to sail to the forward direction with respect to several wave
headings, i.e. m301,451,601,751,901. The analyses in long-crested waves
based on the ISSC spectrum are selected for discussions. The estimation
method is based on the technique used by (Fang and Luo, 2008). The initial
ship speed, the maximum rudder deflection, the rudder deflection rate limit
and the maximum fin angle are set to be 23 knots, 351, 3.5 1/s, and 241,
respectively. Besides, the other important information with respect to the
rudder and main engine are also listed in Table 1 for reference.
The cost function values of the rolling angle, heading error, the tracking
error, the rudder angle and the stabilizer fin angle for four types of controller
with respect to the different wave headings are summarized in Table 3 and
Fig. 4, respectively, which can be applied to judge the performance of each
type of control. For example, at initial heading 301, the rolling cost for Btype is lower than A-type, i.e. Table 3, but the heading error and tracking
error are about 11.5 times higher, i.e. Fig. 4(a) and (b) and the rudder
operation cost is almost the same, i.e. Fig. 4(c). It means that the rudder
commands on the tracking control in B-type are interrupted by the roll
reduction request. Including the extra fin control, i.e. C-type and D-type, the
performance of the roll reduction is improved very much and the tracking
errors are not increased, consequently, the rudder operation cost decreases
with the inclusion of the fin operation. The results in Table 3 indicate that
the roll reduction is generally obvious if the roll reduction control is
included in the rudder command, i.e. B-type. Comparing with the A-type,
the results of RRR is found to be 34%, 68% and 72% for the B-type, C-type
and D-type, respectively.
The time series of roll response, heading error, rudder angle, stabilizer
fin angle, trajectory, and tracking error with respect to the initial heading
301 are presented in Figs. 5 and 6, respectively. The results in Fig. 5 show
that the C- and D-types indeed have the

Time (sec)
Fig. 9. The ship responses for self-tuning neural network based PD controller in

good performance in roll reduction and heading control. From Fig. 6(a) and (b),
we can see that including the roll reduction control on B-type indeed makes the
ship tracking error be up to 710 m. Since roll and yaw responses are coupled
with each other, the ships heading is consequently influenced by the additional
roll motion. The corresponding control gains for tracking and roll reduction
based on the operation of the rudder and the fin are shown in Fig. 7, i.e. Kyp,
Kyd, Ksp, Ksd, and Fig. 8, i.e. Krrp, Krrd, Krfp, Krfd. The results in these figures show
that most of the gains are self-tuning in the nearly constant values except the all
control gains for A-type and the yaw gain control for B-type. This fact means
that we can find the suitable gains automatically using the present technique
instead of taking time by trial and error method.
The same conclusions as the heading of m301 can be applied to the other
initial headings, i.e. m451,601,751,901. It is also interesting to find that the
costs of the heading error, tracking error and rudder operation for the headings
601 and 901 are not of much difference.
The total cost function values with respect to the different type control are
also summarized in Table 4 for reference. From the results, we find that
incorporating the command of the stabilizer fin is generally able to reduce the
cost function values, especially for stern waves. By comparing the cost function
values in Fig. 4, Tables 3 and 4, most of PD controllers with the C- and D-types
can make the total cost function values smaller by applying the stabilizer fin
which can decrease the rudder usage. This phenomenon might be due to the
coupled effect between roll response and yaw response.

26

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

From the overall investigation on the total cost function with respect to the
different wave headings in Table 4, we can conclude

Time (sec)

Time (sec)
Fig. 10. The time variations of control gains for (a) A-type and (b) C-type in long crested waves with H 1=3 3 m, T 10 s and initial ship speed V23 knots. The ship heading shifts from c301 to
01 at t300 s.

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

27

the C-type controller, i.e. the command to control the rudder for the track
keeping and the stabilizer fin for roll reduction, might be the best choice for
the ship control, which has the better tracking keeping and roll reduction.

4.2.

Ship advancing in the dynamic heading change

This section aims to investigate the dynamic behaviors for a ship


operated in random waves by considering the variation of the ship headings.
Since the C-type controller has the better performance on the aspect of rollreduction and track-keeping, the comparisons between A-type and C-type
are made here to realize the ship dynamic behaviors in the heading change.
The ship heading is initially set to be in the stern quartering seas with 301
and changes to 01 or 601after 300 s. The sea state is assumed to be the same
as the last section.
As a result of the change of ship heading angle from 301 to 01, the ship
responses have significant well modification after t300 s as shown in Fig.
9 since the ship sailing direction coincides with the wave direction, i.e. 01.
In this case, both rolling motion and heading error become smaller after
heading change, the needs for both operations of the rudder and the fin are
vanishing finally. The results also indicate that the C-type controller works
very well on the roll reduction before the heading change and the rudder
operation is indeed less. The control gains for both types are also shown in
Fig. 10 for reference, in which Ksd has the larger variation for C-type which
may be due to sway velocity is getting less from the quartering sea to the
following sea. For A-type, Ksp becomes zero after 300 s which means there
is
10000

8000

Time (sec)

A-type
C-type

Fig. 12. The ship responses for self-tuning neural network based PD controller in
long crested waves with H1=3 3 m, T 10 s and initial ship speed V23 knots. The ship heading
shifts from c301 to c601 at t300 s.

X (m)

6000

4000

Wave direction

2000

0
-1000

1000
Y (m)

2000

3000

no gain needed for the tracking variation any more in longitudinal waves in this
case. It may also imply that the sway velocity gain needs more modification
than the yawing rate gain to adapt to the tracking error caused by the ship
heading change. The trajectory and tracking error are shown in Fig. 11 for
reference, which shows both tracking errors are also vanishing and both
trajectories are directed to the north finally. The two trajectories are different
because the positions of the heading change are different due to the different
motion response effect. Similar conclusions can be applied to the case for the
change of ship heading angle from 301 to 601, i.e. See Figs. 1214, except the
gain Ksp for A-type is not vanishing after 300 s because the tracking deviation
still exists in this case. Generally the performances of tracking keeping and roll
reduction for the ship in the dynamic heading change can be still handled well
by using the present self-tuning control technique.

5. Conclusion

Fig. 11. The simulations of (a) trajectory and (b) tracking errors for A- and C-types
in long crested waves with H1=3 3 m, T 10 s and initial ship speed V23 knots.
The ship heading shifts from c301 to c01 at t300 s.

In this study, the technique of the self-tuning neural network based PD


controller, which is used to adjust the rudder and the stabilizer fin to control the
track keeping and reduce the ship rolling motion, is successfully developed.
According to a series of numerical calculation and comparison, the self-tuning
neural network based PD controller is indeed a convenient and sufficient tool to
search for the optimal control gains. It is very suitable for a

28

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725

Time (sec)

Time (sec)
Fig. 13. The time variations of control gains for (a) A- type and (b) C-type in long crested waves with H 1=3 3 m, T 10 s and initial ship speed V23 knots. The ship
heading shifts from c301 to c601 at t300 s.
responsible tool. Finally, the results also verify that

the self-tuning neural


network based PD controller adopted here can respond to the dynamical
heading change rapidly as well, even for the different dynamic ocean
environment.

ship to automatically obtain the optimal control gains to reduce the rolling
motion and control the track keeping at different sea states. In addition,
although the stabilizer fin can be applied to reduce the rolling motion, it may
increase the resistance and heading error, which needs careful considerations.
Comparatively, the results indicate that C-mode, i.e. the compact controller
including the rudder for track keeping and stabilizer fin for roll reduction, may
have the better performance on the ship advancing at sea than other types of
controllers. It means that specified task had better be watched by the specified

M.-C. Fang et al. / Ocean Engineering 54 (2012) 1725


29
Inoue, S., Hirano, M., Kijima, K., Takashima, J., 1981a. A practical calculation method of ship
maneuvering motion. Int. Shipbuild. Prog. 28, 207222.
Inoue, S., Hirano, M., Kijima, K., 1981b. Hydrodynamic derivatives on ship maneuvering. Int.
Shipbuild. Prog. 28, 112125.
Jerrold, N.S., Michael, G.P., 1999. Rudder/fin roll stabilization of the USCG WMEC 901 class
vessel. Mar. Technol. 36, 157170.
Kim, C.H., Chou, F.S., Tien, D., 1980. Motions and hydrodynamic loads of a ship advancing in
oblique waves. Trans. Soc. Nav. Archit. Mar. Eng. 88, 225256.
Lee, C.M., 1977. Prediction of motion, stability and wave load of small-waterplanearea-twin-hull
ships. Soc. Nav. Archit. Mar. Eng. Trans. 85, 94130.
Li, H., Jin, H., Guo, C., 2005. PID based on wavelet neural network identification and tuning and
its application to fin stabilizer. In: Proceedings of the IEEE, International Conference on
Mechatronics and Automation. Niagara Falls, Canada.
McGookin, E.W., Murray-Smith, D.J., Li, Y., Fossen, T.I., 2000. Ship steering control system
optimization using genetic algorithms. Control Eng. Pract. 8, 429443.
Takahashi, T., 1969. Mechanism of rolling and application. Report of Mitsubishi Heavy Industry
Nagasaki Technical Institute. p. 2842 (in Japanese).
Treakle, T.W., Mook, D.T., Liapis, S.I., Nayfeh, A.H., 2000. A timedomain method to evaluate the
use of moving weight to reduce the roll motion of a ship. Ocean Eng. 27 (12), 13211343.
Tzeng, C.Y., Lin, K.F., 2000. Adaptive ship steering autopilot design with saturating and slew rate
limiting actuator. Int. J. Adaptive Control Signal Process 14, 411426.

Time (sec)
Fig. 14. The simulations of (a) trajectory and (b) tracking errors for A- and C-types
in long crested waves with H 1=3 3 m, T 10 s and initial ship speed V23 knots. The ship
heading shifts from c301 to c601 at t300 s.

Acknowledgment
This work is partially sponsored by National Science Council for their
financial support under the Grant no. 99-2915-I-006-044.
References
Cao, Y.S., Lee, T.H., 2003. Maneuvering of surface vessels using a fuzzy logic controller. J.
Ship Res. 47 (2), 101130.
Dove, M.J., Wright, C.B., 1991. Computer methods in marine and offshore engineering.
Southampton.
Fang, M.C., 1991. Roll reduction by rudder control for two ships during underway replenishment.
J. Ship Res. 35, 141150.
Fang, M.C., Lee, M.L., Lee, C.K., 1993. Time simulation of water shipping for a ship advancing
in large longitudinal-waves. J. Ship Res. 37 (2), 126137.
Fang, M.C., Luo, J.H., 2006. The application of the sliding mode controller on the ship roll
reduction in random waves using genetic algorithm. Nav. Eng. J. 118 (4), 3747.
Fang, M.C., Luo, J.H., 2007. On the track keeping and roll reduction of the ship in random waves
using different sliding mode controllers. Ocean Eng. 34 (34), 479488.
Fang, M.C., Luo, J.H., 2008. The ship track keeping with roll reduction using a multiple-states PD
controller on the rudder operation. Mar. Technol. Sname News 45 (1), 2127.
Fang, M.C., Zhuo, Y.Z., Lee, Z.Y., 2010. The application of the self-tuning neural network PID
controller on the ship roll reduction in random waves 37 (7), 529538Ocean Eng. 37 (7),
529538.
Fossen, T., 1994. Guidance and Control of Ocean Vehicles. John Wiley and Sons, New York.
Hamamoto, M., Matsuda, A., Ise, Y., 1994. Ship motion and the dangerous zone of a ship in
severe following seas. J. Soc. Nav. Archit. Jpn. 175, 6978.
Hemerly, E.M., Nascimento, C.L., 1999. An NN-based approach for tuning servocontrollers.
Neural Networks 12 (3), 513518.
Hirano, M., Takashina, J., 1980. A calculation of ship turning motion taking coupling effect due to
heel into consideration. Trans. West-Jpn. Soc. Nav. Archit. 59, 7181.
Hirano, M., 1980. On the calculation method of ship maneuvering motion at initial design phase.
J. Soc. Nav. Archit. Jpn. 147, 144153.

You might also like