Pergamon

PII:S0967-0661 (96)00011-1
Control Eng. Practice, Vol. 4, No. 3, pp. 343-350, 1996
Copyri ght 1996 Elsevier Science Ltd
Printed in Great Britain. All ri ght s reserved
0967-0661/ 96 $15.00 + 0. 00
OPERATI NG EXPERIENCE WITH A HIGH- PRECISION TRACK
CONTROLLER FOR COMMERCIAL SHIPS
T. Hol zhi i ter* and R. Schul tze**
*Hamburg Polytechnic, Sandkamp 13a, D-24259 Westensee, Germany ([email protected])
**Anschiitz & Co., GmbH, Kiel, Germany
(Received August 1995; in final f orm December 1995)
Abs t r a c t . This article surmnarizes operating experience with a commercially produced track
controller for ships, which has been installed on a considerable number of different ships, using
a variety of position measurement systems. The controller is supplied with waypoint and radius
information. Based on this information, it constructs a full manoeuvre trajectory for track changes.
The ship is kept to this reference pat h using feedforward control from the trajectory construction in
combination with LQG feedback control. The Kalman filter can be adapted to the characteristics
of the particular position measurement system. Detailed performance results are presented for both
track-keeping and track-changing phases, recorded during a recent voyage of a Norwegian ferry.
Key Words. Ship control, track control, feedforward control, adaptive autopilot, ship model.
1. I NT RODUCT I ON
Over recent year s t he aut opi l ot on commer ci al
shi ps has evol ved f r om t he classical course con-
t rol l er as descri bed e.g. in ( Amer ongen, 1984;
Saelid, et . al . , 1984; Hol zhfi t er and St r auch, 1987)
i nt o a t r ue t r ack cont rol l er. Thi s has onl y be-
come possi bl e because of t he i ncreased accur acy
and rel i abi l i t y of posi t i on meas ur ement s due t o
(di fferent i al ) GPS. However, t he l i mi t ed posi t i on
accur acy is still t he mos t i mpor t a nt r est r i ct i on on
t r ack- keepi ng accuracy.
Thi s paper pr esent s oper at i ng exper i ence wi t h a
commer ci al l y pr oduced t r ack cont rol l er. The con-
t rol l er has been devel oped and i mpr oved over t he
l ast 6 years. The first appl i cat i on was for t he
hi gh-preci si on t r ack cont rol of mi ne hunt ers, see
(Hol zhi i t er, 1990). Fr om t he begi nni ng, however,
it was i nt ended t o appl y t he cont rol l er t o com-
merci al shi ps as soon as a de ma nd for t r ack con-
t rol arose in t hi s field. The cont r ol l er was sub-
sequent l y a da pt e d t o be used as a piece of st an-
dar d equi pment on commer ci al ships. The con-
t rol l er has cur r ent l y been i nst al l ed on a consi der-
abl e numbe r of di fferent shi ps, for i nst ance l arge
ferries, different t ypes of car go shi ps and speci al
vessels wi t h Voi t h Schnei der pr opul si on. Several
different posi t i on me a s ur e me nt s ys t ems have been
used, i ncl udi ng GPS as well as di fferent i al GPS
(whi ch was, e.g., used when r ecor di ng t he da t a
pr esent ed l at er in t hi s paper ) , Decca, Syledis com-
bi ned wi t h Doppl er log, and speci al l y i nst al l ed ra-
dio navi gat i on syst ems.
Ther e is a si gni fi cant difference bet ween a con-
t rol l er speci al l y t ai l or ed for a cer t ai n shi p (as in
t he case of t he ori gi nal i nst al l at i on on a series of
mi ne hunt ers) and a s t andar d cont rol l er t ha t has
t o be i nst al l ed on different t ypes of shi ps wi t h a
very l i mi t ed a mount of i nst al l at i on work. A t i me-
consumi ng series of t est manoeuvr es will be unac-
cept abl e t o t he cust omer in t he l at t er case. Ther e-
fore, t he r obust ness and t r ans par ency of t he con-
t rol l er is of gr eat i mpor t ance. The cont rol l er also
has t o rely on pa r a me t e r s t ha t can be identified
f r om on- boar d meas ur ement s .
The t r ack cont rol l er was desi gned usi ng mod-
ern cont rol desi gn t echni ques such as LQG con-
trol, s uppl ement ed by model - f ol l owi ng feedfor-
war d cont rol for changi ng manoeuvr es . Thus t he
desi gn was l argel y under t aken accor di ng t o " t ext -
book st r at egi es", avoi di ng bad t ri cks t ha t are of-
t en used for maki ng t hi ngs work.
Thi s paper onl y briefly addresses t he shi p model
and t he cont rol s t r at egy used in t he aut opi l ot .
These have al r eady been descri bed elsewhere, see
e.g. (Hol zhi i t er, 1990; Hol zhfi t er and St rauch,
1987). Here t he mai n at t ent i on is pai d t o det ai l ed
per f or mance resul t s whi ch were recorded on a voy-
age of t he Nor wegi an f er r y shi p Kong Har al d f r om
Mol de t o Kr i st i ansund. The t r ack cont rol s ys t em
on t hi s shi p is t he resul t of a cooper at i on bet ween
Nor cont r ol ( posi t i on handl i ng and man- machi ne
i nt erface for t r ack pl anni ng) and Anschi i t z & Co.
( t r ack cont rol l er). Fi gur e 1 pr ovi des an overvi ew
of t he component s in t he t r ack cont rol loop. A
343
344 T. Holzhiiter and R. Schultze
DGPS I I ECDIS ~ NIA'W ~
track control and
track constructior
gyro . , , . ~ = , f = , = , e ~ c k
Steeringl
Gear /
Fig. 1. Flow of information in track control system
very significant det ai l is t he i nt erface bet ween t he
t rack pl anni ng syst em - here t he Nor cont r ol Dat a
Bri dge 2000 - and t he t r ack cont rol l er. The con-
t rol l er is pr ovi ded wi t h t he waypoi nt i nf or mat i on,
and not wi t h course-t o-st eer or cross-t rack-error
l O (
K
8C
7 (
~ 5c
' i i
5 0 0 1 1500 2 0 0 0 2 5 0 0 3 0 0 0 4(X)O 4 5 0 0 5000
T i m e [ S ]
as in mor e ol d-fashi oned t r ack- cont r ol archi t ec-
tures. The waypoi nt i nt erface makes i t possible
for all t he comput at i ons r egar di ng t r ack const ruc-
t i on and cross-t rack-error cal cul at i on t o be done
by t he t rack cont rol l er itself. Thi s is a prerequi -
site for t he feedforward cont rol design descri bed
below. Thi s paper focuses on t he t r ack cont rol l er
itself, which is mar ket ed as Naut opi l ot A - Wat ch
One ( NPA- W1) .
A general i mpressi on of t he t rack-keepi ng qual i t y
of t he syst em can be gai ned f r om Fig. 2. The
t rack er r or duri ng t rack-keepi ng phases is below
5 m. Duri ng t rack changes t he t r ack error is mea-
sured agai nst t he i nt er nal l y pr ecomput ed maneu-
vering t r aj ect or y. The error dur i ng t hese phases
is t ypi cal l y below 10 m. More det ai l ed resul t s will
be present ed in Sections 4 and 6. Accordi ng t o
t he navi gat i on officers, 90 % of t he voyages on t he
Kong Har al d are now per f or med in t rack-keepi ng
mode, see (Skips Revyen, 1994). The r emai ni ng
10 % use course cont rol and hand st eeri ng mode.
10
- 1 0
- 1 5
- 2 0
_ 2 5 0 . . . . 5 0 0 ' 3 5 0 0 = i
5 0 0 1000 1500 2 0 0 0 2 3 0 0 0 4 0 0 0 4 5 0 0 , 5 0 0 0
Time [ s ]
Fig. 2. Heading and track error during typical voyage.
2. SHI P MODEL
For t he design of t he t rack cont rol l er, a model is
necessary t hat describes all t he rel evant aspect s
of t he shi p' s mot i on. On t he ot her hand, i t has
t o be simple enough t hat its par amet er s may be
identified, in a rel i abl e way, f r om t he s t andar d on-
boar d measur ement s. Li near i t y of t he cont rol l er
will be achieved by l i neari zat i on of t he shi p' s mo-
t i on al ong a reference t r aj ect or y. Fi gure 3 shows
t he different coor di nat e syst ems i nvol ved in de-
scribing t he shi p' s mot i on.
High-Precision Track Controller for Commercial Ships 345
Fig. 3. Coordinate system for the description of ship
motion The angle is the difference between
heading and track course, u is the forward ve-
locity measured by the log, v is the velocity in
starboard direction, x is the position along the
track, y is the cross-track-error.
al ready be deduced appr oxi mat el y f r om general
dat a on t he ship.
The sway vel oci t y v is assumed t o be pr opor t i onal
t o t he yaw r at e r.
v = ~ r ( 2)
Thus t he whole model of t he ship is descri bed by
t he 3 par amet er s K, T and a. Whi l e t he on- boar d
est i mat i on of K, T does not pose probl ems, t he es-
t i mat i on of a oft en is unrel i abl e due t o t he l i mi t ed
accuracy of t he posi t i on measur ement . Ther ef or e
t he correct value of a has t o be deduced f r om
t rack-changi ng per f or mance results, see Sect i on 6
below The r emai ni ng st at es of t he ship model can
be deri ved f r om t he ki nemat i c rel at i ons accordi ng
t o Fig. 3.
- u c o s + v s i n + d ~ (4)
= u s i n + v c o s + d r (5)
For si mpl i ci t y of t he equat i ons, it is appr opr i at e
t o use a coor di nat e syst em t ha t is t angent t o t he
reference t r aj ect or y at t he poi nt t hat is closest
t o t he act ual posi t i on. The local coor di nat e sys-
t em t hen has t o be shi ft ed and r ot at ed f r om t i me
step t o t i me st ep, in such a way t hat it r emai ns
t angent t o t he reference t r aj ect or y and t hat t he
x- coor di nat e represent s t he arc l engt h al ong t he
t raj ect ory Thus, on a st r ai ght course t he local
coor di nat e syst em r emai ns fixed, while dur i ng ma-
noeuvres its ori gi n is movi ng on t he i nvol ut e curve
of t he reference t raj ect ory
The l i neari zat i on of t he hydr odynami c equat i ons
of mot i on of a ship leads t o a second-order model
in t he st at e vari abl es r at e of t ur n r and sway veloc-
i t y v (K/illstrSm, 1982). Thi s model is decri bed by
6 par amet er s and is of t en i l l -condi t i oned wi t h re-
spect t o i dent i fi cat i on when usi ng on- boar d mea-
surement s. Ther ef or e t he well-known simplified
model named af t er Nomot o ( Nomot o, et . al . , 1957)
was used i nst ead, which has onl y t wo par amet er s.
l r +K
- T T ~ + r b + w l ( 1)
The const ant i nput rb is i nt r oduced t o model con-
st ant t orques, resul t i ng f r om s t at i onar y wi nd dis-
t urbances. Because rb is slowly t i me- var yi ng, it
has t o be est i mat ed by t he Kal man filter The pro-
cess noise wl model s i r r egul ar di st urbances result-
ing f r om wi nd t ur bul ences and waves. Thi s sim-
ple model ignores t he somet i mes significant non-
l i neari t y of t he shi p' s t ur ni ng charact eri st i c, see
e.g. (Schultze, 1992). Wher eas t he par amet er s of
t he nonl i near model are difficult t o quant i f y using
t he nor mal amount of i nst al l at i on work, t he pa-
r amet er s K, T of t he si mpl e l i near model (1) can
Here is t he headi ng angl e and (x, y) are t he t wo
component s of t he posi t i on of t he ship in t he local
coor di nat e syst em. The presence of a s t at i onar y
cur r ent (dx, dr ) , which has t o be est i mat ed by t he
Kal man filter, is assumed. The f or war d vel oci t y u
of t he ship is measur ed by t he speed log, and can
t hus be viewed as a known par amet er .
Duri ng t rack cont rol , t he rel at i ve headi ng angl e
is small, and (5) can t hus be l i neari zed for t he cal-
cul at i on of t he Kal man filter and cont rol l er gains.
= u + v + d r (6)
The const ant t or que rb in (1) and t he dri ft compo-
nent s d~, d r in (4), (5) are assumed t o be slowly
t i me- var yi ng, and are model l ed as r andom walk
processes,
r b ----
=
dy =
In rougher
for si mpl i ci t y:
, ~ ( 7)
, ~ ( 8)
w~ . ( 9)
sea condi t i ons coupl ed roll and pi t ch
mot i on of t he ship i nduces a peri odi c yaw mo-
t i on. The r udder cannot count er act this mot i on.
For cont rol purposes t hi s hi gh-frequency mot i on
shoul d be separ at ed f r om t he l ow-frequency mo-
t i on of t he ship in order t o avoi d unnecessary rud-
der act i on. Thi s can be done by assumi ng t he
measur ement of t he headi ng t o be di st ur bed by
colored noise CHv. The noise can be model l ed as
t he out put of a weakl y damped osci l l at or driven
by whi t e noise
~ . F = . F + ~ . ~ w5 ( 10)
CHF 2 ~ F 2DwMFCHF. (11)
- - 0J HF - -
346 T. Holzhiiter and R. Schultze
W a y p o i n t s
I
~ 1 t r a j e c t o r y
~ [ construct.
_ , _ _ _ d s t a t e _ s p a c e l
r_ cntrl'erl
r ~ = 1
o _ I
-I
ship
v V
K a l m a n [
f i l t e r I
c o o r d i n a t e ] ~
I t r a n s f . I T M
Fig. 4. Model-following control scheme.
Thus a headi ng angl e measur ement is assumed ac-
cordi ng t o
= + . F . ( 12)
The osci l l at i ng f r equency of t he di st ur bance model
(10), (11) depends on t he seaway condi t i ons and
on t he rel at i ve di rect i on of t he waves. Ther ef or e it
is est i mat ed on-l i ne usi ng recursi ve l east -squares
i dent i fi cat i on. For a mor e det ai l ed discussion see
(Hol zhfi t er and St r auch, 1987; Hol zhi i t er, 1990).
3. MODEL- FOLLOWI NG CONTROL
Since not all st at es of t he ship model descri bed in
Sect i on 2 are measur abl e, a st at e est i mat e ~ has
t o be r econst r uct ed, usi ng a Kal man filter. The
i mpl ement at i on of t he Kal man filter is in di scret e
t i me. The syst em of equat i ons (1), (3), (4), (5) and
(7)-(11) yields t he st at e vect or
~c = ( r , , x , y , rb, d, , d~, ~nF, HF) T, (13)
t he measur ement vect or z = (era xm ym) T,
and t he cont rol vari abl e u = 5, t he r udder angle.
Thi s model corresponds t o a nonl i near discrete-
t i me st at e-space model :
vct+l = f ( z t ) + Gut +wt (14)
zt = Hz t A - v t . (15)
Here vt and wt are t he measur ement and t he pro-
cess noise, respectively, bot h assumed t o be white.
Dependi ng on whet her GPS or DGPS qual i t y is
available for posi t i on measur ement , t he covari ance
mat r i ces of t he meas ur ement noise R are set t o
different values. Thus t he Kal man filter can be
adapt ed t o different posi t i on receivers. The whole
LQG- cont r ol l er is descri bed by
= + G u t ( 16)
t+1 = + K ( z , - u t) ( 17)
f i t = L( ~t - ~ t ) . ( 1 8 )
The geomet ri c nonl i near i t y of t he syst em accord-
ing t o (4), (5) is preserved in t he ext r apol at i on step
(16) of t he Kal man filter, 5~* denot i ng t he ext r a-
pol at ed st at e est i mat e.
The feedback cont rol f in (18) is based on t he dif-
ferences bet ween t he est i mat ed st at es ~b and t he
reference st at e vect or as obt ai ned f r om t he t ra-
j ect or y const ruct i on. The t ot al cont rol signal u
is t he sum of t he feedforward cont rol f and t he
correct i on f resul t i ng f r om t he feedback loop:
ut = fit + f i t . (19)
The i nf or mat i on flow for t hi s cont rol st r uct ur e is
summar i zed in Fig. 4. It can be viewed as a gener-
al i zat i on of t he well-known me t hod of prefiltering
t he reference signal in l i near cont rol t heor y and is
called "model -fol l owi ng" cont rol .
For t he cal cul at i on of t he Kal man gains K
f r om t he noise assumpt i ons E( v t v T) = R and
E( t o t w T) = Q, t he linear appr oxi mat i on accord-
ing t o (1), (3), (6) is used.
4. TRACK- KEEPI NG PERFORMANCE
The t rack-keepi ng per f or mance can be j udged
f r om Fig. 5. A sect i on of about 10 mi n was t aken
f r om t he voyage al ready present ed in Fig. 2, st art -
ing at t = 450 s. It is obvi ous t ha t t he t rack error
is always below 10 m. The range of one st an-
dar d devi at i on ~ = 2.8 m is i ndi cat ed. It can be
observed t hat t he cont rol l er changes t he headi ng
angl e in such a way t hat t he l ow-frequency t rack
devi at i ons are count er act ed.
Fr om t he t i me series in Fig. 5 it can be seen t hat
t her e is a significant peri odi c component in t he
headi ng signal. Thi s is due t o t he wave-i nduced
roll mot i on of t he ship, which in t ur n induces a
peri odi c r ot at i on of t he ship ar ound t he yaw axis.
For good st eeri ng qual i t y it is not necessary
t o count er act t he peri odi c headi ng devi at i ons.
High-Precision Track Controller for Commercial Ships 347
D
i - -
- 8
100 200 3 4 500 000 700
T i m e [ s ]
1 0 =
t 0 1
' 10 o
0
" 0
~ I 0 "I
10";
10 <
. . . . . . . . I
10 -= 10 -I
. . . . . . . . i
1 o
F r e q u e n c y [ r a d / s ]
t i
I I
100 200
t
.
i i i i
300 400 500 6 0 0
T i m e I s ]
1
700
10 =
101
(D
~ 10"
~ 1 0 ~
104
, . . . . . . . i . . . . . . . . i
t l O- ~ lO-1 l O 0
F r e q u e n c y [ r a d / s ]
1
- (
- !
_ 1 0 : i i i i i i
1 O0 200 300 400 500 cO0 700
T i m e [ s ]
101
$ , 0 0
1 0 . . . . . . . . i . . . . . . . . i
t O -2 10 -1 10 o
F r e q u e n c y [ t a d / s ]
Fig. 5. Track error, Heading and rudder angle during
track-keeping.
Therefore for t he t rack controller described here,
a periodic di st t i rbance was i nt roduced i nt o t he
model of t he ship via (10)-(12). The peak fre-
quency of t he di st urbance model is est i mat ed via
recursive least-squares i dent i fi cat i on.
Thus t he observed spect r um of t he headi ng sig-
nal as shown in Fig. 6 (top) is t hought of as a
superposi t i on of t he pure headi ng signal and a
periodic di st urbance , ~ due t o t he roll mot i on.
Thi s is shown schemat i cal l y in Fig. 6 (centre).
Fig. 6. Demonstration of the band stop characteristic
of the Kalman filter.
Top: Autospectra of heading Sv~ [deg2s/rad]
(thick line) and rudder angle S~ [deg2s/rad]
(thin line).
Centre: Scheme for the decomposition of mea-
sured heading spectrum Sere (thick line) into
undisturbed heading signal S (thin line) and
periodic disturbance ScHp (thin dashed line).
Bottom: Band stop characteristic of the
Kalman filter. The ratio ~ (thin line)
of the square roots of rudder and heading spec-
tral density respectively is roughly equal to
the transfer function of the controller (thick
line), clearly showing the bandstop character-
istic. The transfer function for a conventional
PID-type controller is shown for comparison
(dashed line).
348
T. Holzhiiter and R. Schultze
Thi s ship model resul t s in a bandst op charact eri s-
tic of t he Kal man filter, see (I-Iolzhiiter, 1989) for
details. In Fig. 6 ( bot t om) t he r at i o of t he t wo
spect r a of r udder and headi ng f r om Fig. 6 ( t op)
is shown, t oget her wi t h t he t r ansf er f unct i on of
t he t rack cont rol l er f r om headi ng t o r udder . The
bandst op charact eri st i c can be cl earl y seen f r om
bot h t he r at i o of t he measur ed spect r a and t he
known t r ansf er f unct i on of t he cont rol l er, which
fit t oget her qui t e well. The t r ansf er f unct i on of
a classical PI D- t ype aut opi l ot is shown for com-
parison. It woul d resul t in much mor e r udder mo-
t i on in t he f r equency r ange ar ound w ~ 0.5 t a d/ s ,
wi t hout any i mpr ovement in st eeri ng qual i t y. Thi s
seaway-filtering f eat ur e of t he aut opi l ot proved t o
be very successful.
5. MODEL- BASED TRACK CHANGES
On di spl ay devices such as ECDI S or ARPA sys-
t ems, t he pl anned t r ack changes are usual l y shown
as circles. An arc wi t h radi us prescri bed by t he
user st ar t s near t he end of t he cur r ent leg, at t he
so-called wheel -over-poi nt ( WOP*) , and ends on
t he subsequent leg, see t he dashed lines in Fig. 7.
Thi s pr esent at i on is, however, not realistic in t he
l i ght of t he rest ri ct i ons of ship dynami cs. It is
i mpossi bl e t o change t he t ur nr at e r of t he ship in-
st ant aneousl y, which can easily be deduced f r om
(1).
Fi gure 7 shows a t r ack change wi t h a t ur ni ng ra-
dius of 0.2 naut i cal miles st ar t i ng wi t h a t rack
course of 30.2 in t he lower l ef t - hand corner and
endi ng wi t h a t r ack course of 130.7 at t he lower
ri ght . The t r ack legs are shown as dashed lines
connect ed at t he waypoi nt .
G
- , ~
- l OG
- 15Q
" E - 2 0 0
-4 6 0
i _ i I
1. 5
I
- 0 5 q
- 2 0
i / ~ ~ Waypoln!
i
/
L o n g i t u d e [m]
J . . . . . . i . . . . . . . . . . . . . . . . . . . . J
Time [s]
Time [s]
I
E . . . . . . . . . . . . . . . . . . . . . . . . . . . .
c l OO
0 / r i . . . . . . . . . . . . . . . . . . . . . . . i i i i i . . . . . I -
n
- 2 0 0 20 4 0 6 0 8 0 100 120
T i m e [ s ]
I nst ead of si mpl e circles, t he t r ack cont rol l er NPA-
Wl makes use of model - based t r ack const ruct i on.
It differs f r om t he si mpl e ci rcul ar arc (dashed line)
by t wo feat ures. Fi rst , t he model - based wheel-
over-poi nt WOP lies about one ship l engt h before
of t he begi nni ng ( WOP * ) of t he si mpl e arc. Sec-
ond, t he cent er of t he circle is shi ft ed by some
met ers.
The t r ack- changi ng manoeuvr e, shown as t he solid
line in Fig. 7, consists of t hr ee phases correspond-
ing t o t hr ee geomet r i cal segment s. The first one is
an i nt er medi at e phase of t ur nr at e increase. Dur-
ing t hi s phase, t he cont rol l er uses r oughl y 1.5
t i mes mor e r udder t han is needed for t he subse-
quent phase of const ant t ur nr at e. In t hat second
phase, t he ship moves on a circle havi ng a user-
prescri bed radi us like t he circle di spl ayed on t he
ECDI S or ARPA screen. Before reachi ng t he new
st r ai ght t rack, t he t hi r d phase st art s, reduci ng t he
t ur nr at e unt i l fi nal l y ar r i vi ng at t he new t rack
wi t h t ur nr at e zero. Dur i ng t hi s t hi r d phase, t he
Fig. 7. Maneuvering trajectory for a turning r ~i us
of 0.2 NM with wheel-over-point WOP. T h e
track legs are shown as dashed lines intersect-
ing at the waypoint. The dashed line indicates
the approximating circle with its correspond-
ing wheel-over-point WOP*. Note t hat t h e
ship is shown in correct size.
cont rol l er uses count er - r udder . The t hr ee phases
of t he t rack change - accel erat i on, const ant t ur n-
rat e, and decel erat i on - are cl earl y visible f r om
t he t i me series of t he t ur nr at e in Fig. 7.
Fi gure 7 shows an exampl e wi t h a radi us of onl y
0.2 naut i cal miles, whi ch is qui t e smal l for a ship
like t he Kong Har al d. Thi s smal l radi us was cho-
sen for reasons of cl ari t y because wi t h a mor e t yp-
ical radi us of 0.5 naut i cal miles t he maneuver i ng
t r aj ect or y coul d har dl y be di st i ngui shed f r om t he
ci rcul ar arc on t he pl ot in Fig. 7. Al t hough t he
di st ance bet ween t he t wo arcs is onl y some met ers,
t he benefi t s of t he model - based t r ack const r uct i on
High-Precision Track Controller for Commercial Ships 349
1 4 0 0
1 2 0 ~
" o
4 ~
2 ~
o 5 o o l O O O
L o n g i t u d e [ m ]
' t . . . . . . . . . . .
Time [ s ]
, : , , ,
" 1 7 2 0 : ~ 4 6 . . . . . . . . . . . :: . . . . . . . . . . . . . . . . . . . . . . . . . : . . . . .
o , i , , ~ i
0 5 0 1 0 0 1 5 0 2 0 0 2 5 0
Ti me [ sl
~ 1 0 i
~ 5
Time I s ]
t he r udder mot i on. I t can be seen t hat t he rud-
der angl e onl y r oughl y resembles t he feedforward
pat t er n present ed in Fig. 7. Thi s is par t l y due t o
t he fact t hat on i nst al l at i on t he sway mot i on par-
amet er a was chosen t oo large. Thi s makes t he
act ual t r aj ect or y t ake a shar per t ur n t han neces-
sary, resul t i ng in a final phase of r eappr oach t o
t he t rack.
The pr obl em is t hat t he par amet er a can onl y be
t uned f r om t rack-changi ng per f or mance. It can-
not be identified, because t he st at i st i cal errors of
t he posi t i on measur ement are t oo l arge t o allow
t he comput at i on of sway velocity, and t ypi cal l y no
sway sensor is available on commerci al ships. Al-
t hough on Kong Har al d a sway sensor is installed,
however, this measur ement is not available t o t he
t rack cont rol l er. As a consequence of t he mea-
surement s present ed in Fig. 8 t he par amet er a on
Kong Har al d has been changed, but no perfor-
mance records are yet available.
In addi t i on it can be observed f r om Fig. 8 t hat
t he st at i onar y r udder angl e is significantly less
t han could be expect ed f r om t he linear model
par amet er K. Thi s is due t o a nonl i near t ur n-
ing charact eri st i c, which can be observed on many
ships (Schultze, 1992). Aft er t he i ni t i al i zat i on of
t he t ur n, a moder at e angul ar vel oci t y of t he ship
can oft en be mai nt ai ned wi t h ver y l i t t l e rudder.
Therefore, t he feedback cont rol l oop of t he t rack
cont rol l er is significantly modi f yi ng t he feedfor-
ward r udder which was shown in Fig. 7. Not e
t hat t he ship has a beam of 20 m. Thus, a t rack
error of 10 m means t hat t he edge of t he ship is
still on t he t rack.
Fig. 8. Track-changing manoeuvre.
are significant in real -worl d appl i cat i ons where a
pr edi ct abl e r udder movement wi t hout peaks at
t he begi nni ng of a t ur n (Fig. 7) is of great im-
por t ance.
6. AN EXAMPLE OF TRACK- CHANGI NG
PERFORMANCE
To demonst r at e t he t r ack- changi ng per f or mance
of t he cont rol l er NPA- W1, t he l argest t rack
change f r om Fig. 2 is shown in Fig. 8. It takes
place at about t = 300 s wi t h a change in set
course of about 45 o. The x- y- gr aph and t he t i me
series of t rack error, headi ng, and r udder angle are
present ed in Fig. 8. The desired radi us of t ur ni ng
was 0.5 NM.
The t rack er r or is below 15 m for t he whole of
t he t ur n. It is i nt er est i ng t o t ake a closer look at
7. CONCLUSI ONS
Thr ee aspect s of t he design have proved t o be very
successful. The first is t he model - based const ruc-
t i on of a reference t rack. Thi s leads t o a smoot h
manoeuvr e t r aj ect or y which can be followed by
t he ship, and resul t s in qui t e pr edi ct abl e r udder
movement s.
The second aspect is t he r educt i on of r udder mo-
t i on in rough seaway by using an appr opr i at e
adapt i ve Kal man filter. Apar t f r om less wear and
t ear of t he r udder ' s hydr aul i c mechani sm, t hi s also
decreases fuel consumpt i on.
Finally, t he use of a Kal man filter for st at e
est i mat i on allows easy adapt at i on of t he filter-
ing charact eri st i cs t o different t ypes of posi t i on-
measur ement syst ems.
The resul t s present ed i l l ust rat e t hat aft er a long
er a of course cont rol as t he s t andar d aut opi l ot
mode, t he t rack cont rol mode is now finding its
350 T. Holzhiiter and R. Schultze
way into the commercial wheelhouse. This was
only made possible after differential GPS provided
the appropriate accuracy of position measure-
ment. It was demonstrated t hat under these con-
ditions, an advanced track controller can achieve
track errors well below 10 m.
8. REFERENCES
Amerongen, J.v. (1984). Adaptive steering of
ships - A model reference approach. Automat.
ica, Vol. 20, No. l , pp. 3-14.
Holzhiiter, T. and Strauch, H. (1987). An adap-
tive autopilot for ships: design and operational
experience. Proc. lOth I FAC World Congress,
Munich, Germany, pp. 226-231.
Holzhiiter, T. (1989). Robust identification in an
adaptive track controller for ships. Proc. 3rd
IFA C Symposium on adaptive systems in con-
trol and signal processing, Glasgow, UK, pp.
275-280.
Holzhiiter, T. (1990). A high precision track
controller for ships. Proc. 11th I FAC World
Congress, Tallin, Estonia, vol.8 pp. 118-123.
K/illstrSm, C.G. (1982). Identification and Adap-
tive Control applied to Ship Steering. PhD.
Thesis Lund Institute of Technology.
Nomoto, K., Taguchi, T., Honda, K. and Hi-
rano, S. (1957). On the steering quality of ships.
Int.Shipbuilding Progr., Vol. 4 No.35, pp. 354-
370.
Saelid, S., Svanes, T., Onshus, T., Jensen, N.A.
(1984). Design considerations, analysis and
practical experience with an adaptive ship' s au-
topilot. Proc. 9th IFAC World Congress, Bu-
dapest, Hungary, pp. 35-40.
Schultze, R. (1992). Robust identification for
adaptive control: the dynamic Hinkley-
detector. Proc. 4th IFAC Symposium on Adap-
tive Systems in Control and Signal Processing,
Grenoble, France, pp. 23-28.
Skips Revyen (1994). Distansen Trondheim-
Kristiansund N - feste del av den elektroniske
sjeveien offisielt pnet. Skips Revyen, Nr. 4, pp.
32-33.