You are on page 1of 8

of the ASME201130thInternational

Proceedings on Ocean,OffshoreandArcticEngineering
Gonference
oMAE2011
June19-24,2011,
Rotterdam,TheNetherlands

16
oMAE2011-491

ALGORITHM
A GENERICOPTIMIZATION
FORTHE ALLOCATIONOF DP ACTUATORS

Dr. ir. E.F.G.van Daalen J.L. Cozijn


Research
Maritime (MARIN)
Netherlands
Institute Maritime
Research (MARIN)
InstituteNetherlands
Wageningen, The Netherlands Wageningen,The Netherlands

C. Loussouarn Prof. Dr. P.W.Hemker


EcolePolytechnique CentrumWiskunde& lnformatica (CWl)
France
Palaiseau. Amsterdam. The Netherlands

Finally, some options for improvementsand future


ABSTRACT I enhancements - suchas includingthruster-thruster andthruster-
hull interactionsand the effectsof current- are discussed.
In this paper we presenta generic optimizationalgorithm
for the allocation of dynamic positioning actuators,such as INTRODUCTION
azimuthingthrustersand fixed thrusters.The algorithmis based
on the well-known Lagrangemultipliers method. In the present With the offshore industry moving to ever deeperwaters,
approachthe Lagrangianfunctionalrepresentsnot only the cost more and more vesselsare equippedwith dynamicpositioning
function(the total power deliveredby all actuators),but also all (DP) systems.On a DP vessela feedbacksystemcontrolsthe
constraintsrelatedto thrustersaturationand forbiddenzonesfor thrustersto keepthe vesselin a fixed position,thus eliminating
azimuthingthrusters. the needfor mooring lines.The componentsin a DP systemare
describedin more detailbelow
In the presentedapproachthe applicationof the Lagrange
multipliers method leads to a nonlinear set of equations, Components in a DP system
becausean exact expressionfor the total power is applied and The DP systemon board a vesselcontainsseveraldifferent
the actuatorlimitationsare accountedfor in an implicit manner, hardware and software components.These componentsare
by meansof nonlinearconstraints.It is solved iteratively with shown in the schematic overview in Figure 1. The main
the Newton-Raphsonmethodand a stepby stepimplementation componentsof the DP system are briefly describedbelow.
of the constraintsrelatedto the actuatorlimitations. canbe foundin [1], [2] and [3].
More detailedexplanations

In addition.the resultsfrom the non-linearsolutionmethod o Position Measurement:The position of the control point
were comparedwith the resultsfrom a simplified set of linear (CP) on the vessel is measuredusing e.g. GPS or an
equations,basedon an approximate(quadratic)expressionfor acousticpositionreferencesystem.
the thrusterpower. The non-linearsolution was more accurate, o Extended Kalman Filter (EKF): The EKF determinesthe
while requiringonly a slightly higher computationaleffort. low frequencymotions and velocities of the vessel.The
purposeof the filter is to avoid thrusterresponseto wave
An example is shown for a thruster configurationwith 8 frequencyvesselmotions.
azimuthingthrusters,typical for a DP semi-submersible. The o Position Error: The estimatedlow frequencyposition and
resultsshow that the optimizationalgorithm is very stableand velocity are comparedto the position and velocity of the
efficient. reference point (RP). The resulting position error is
forwardedto the Controller.

CopyrightO 2011by ASME


o Controller: Basedon the horizontaloffset from the RP and
the velocity of the vessel, the Controller determinesthe OPTIMIZATIONOF THRUSTER ALLOCATION
requiredtotal surgeand sway forces and yaw moment.
o Allocation Algorithm: The Allocation Algorithm distributes Intro duction and dejinitio ns
the required total forces and moment over the available We considera vesselwith N azimuthingthrusters(other
actuatorssuch that the allocatedpower is minimized. The typesof actuatorswill be consideredlater on). Eachazimuthing
allocation algorithm (marked red in Figure 1) is the thruster i is characterrzedby the following thrusterattributes:
componentof the DP systemdiscussedin this paper.
o Thrusters:The azimuth anglesand RPMs are set, basedon o The position (x,,li) with respectto the referencepoint G .
the output of the Allocation Algorithm. The generatedtotal o The maximum thrust (u*.,
forcesand momentwill move the vesselCP towardsthe RP
position. o The maximum power P,n"*.-

It is noted that in practicalapplicationsthe responseof the Each thruster can rotate about its vertical axis. The azimuth
thrustersis limited in terms of rate of turn, as well as rate of (angle)of thruster i is denotedby a, , its thrustby T and its
changein RPM. This may causedifferencesbetweenthe total state is defined by the allocatedsurge force and sway force,
thrust requestedfor by the controller and the total thrust combinedin the statevector:
generatedby the actuators, especially in relatively severe
environments, close to the limitations of the vessel's
F,= (F,,,,Fr',)' (1)
stationkeeping capabilities.

Furthermore,the effective force deliveredby the thrusters The contributionto the yaw momentabout G is
may be smaller than the nominal (bollard pull) thrust value.
This difference is caused by thruster-interaction(or thrust Mr,i = xrFr,, - /iFr,, (2)
degradation)effects.The following thruster-interactioneffects
arementioned: r The thrust T, and azimuth qi are calculated from {,, and {,,

o Thruster-hullinteraction as follows:
o Thruster-thrusterinteraction
4 =ll4ll-,[F,".i+t
fl-tt , ^
o Thruster-currentinteraction y., ?, (3a)
di = iltctArr(Fr,,f F,,,) (3b)
Thruster allocation
The presentpaper focuseson the Allocation Algorithm. In
general,there will be more variables describing the thruster Note that some anglesmay be prohibited: for instance,if one
settings(azimuthangle,RPM) than equationsto solve(required thrusteris in the streamof another,the efficiency will drop. In
forces and moment). The over-determinedset of equationsis certaincases,we will thus haveto definea "forbiddenzone"for
solved in such a way to minimize the allocated power. the azimuth.
However, the resulting optimization problem is relatively
complex,for the following reasons: The power of thruster i is givenby:

o The relationsbetweenRPM, generatedthrustand consumed 1t^' = ",7,^ or 4@t = c,(F|,,*.F"',,)^'' (4)


power arenon-linear.
o The thrustgeneratedby a thrusteris limited ('saturation'). The exact formulation for the thruster power (in bollard pull
o Certain orientationsof an azimuthing thrustermay not be conditions)is obtainedfor m--312, but this leadsto a system
'forbidden zones' may be defined to avoid
allowed. These
of non-linearequations,whereasthe choica r'/t= 2 gives a set
excessivethruster-interactionlosses,or to protect sensitive
of linear equationswhich is easierto solve.
equipmentplacedunder the vesselhull (e.g.hydrophonesor
cables).
The coefficient c, is calculatedby substitutionof the maximum
valuesof the thrustandthe power for m =312:

P.rr.i
? = - (s)
r:!:,

CopyrightO 2011by ASME


We can then describe the sYstem of thrusterswith the global At an optimum,the gradientof Lagrangianmustvanish:
state vector
V-A(2)=d and VrA(2)=d 02)
(6)
Hencewe haveto find F and P , so that:
and define the total surge force, the total sway force and the
total yaw momentby summationover all thrusters: Zc,F,,,*p,-!,p"=0 (13a)
f f N 1 V 2c,Fr.i1p, + x,p, :0 (13b)
(7)
4,,o, = LF',, t Fy,at=LFr,, , M,,tot=ZM,', lV

ZF',, = F",^ (i3c)

The total power of the thruster configurationis calculatedby ? (13d)


summationof the power of the individual thrusters: ZFr,,= F,,-
T
N
- !,F",,)= M,,- (13e)
4g'=L1'^' (8) l{x,Fr.,
I ='I

,nr";rr.", systemof equationscanbe written as


Optimization of allocation - rninimization of power
Our aim is to optimize the energy consumed by the
Ai=6 <+ i=AtE (14)
thrusters:we search the state vector F that minimizes the
objectfunction P,li') underthe following setof constraints:
where the system matrix, the solution vector and the right-hand
side vector are given by
o The total surgeand sway forces and the total yaw moment
haveto matchthe requiredvalues: .
I Ct t R'-l _ [Fl ' [0rr.,-]
A=l-_
R"(F) = 4,,* - 4,,o,= 0 (ea) LR 0,,,-l tB) L',"qr
R,(F) = Fr,-- 4,,o,= o (eb)
where 0,,,, it a zero sub-matrixwith m rows and n columns,
R,(F)=M,,^-M,,tot=o (9c)
C(t) is the Hessiansub-matrixof the approximatepower P(')
with size 2N x2N and R is the requirementconstraintssub-
o The thrustis limited by the maximumthrust:
matrixwith size 3x2N .

M , ( F ) = T k . , - ( F : , ,+ F : , , ) > o i = 1 , 2 , . . . , N (lo) Exact solution without constraints


We are looking for a methodto correctthe solutionfound
If the thrust of thruster i reachesthe maximum value, we from (14-15) and find a better minimization of the power
'saturated'.
will saythat the thrusteris computedin the exactway, i.e. with m=312. To this end,we
usethe Newton-Raphsonmethod[5]:
o Finally we may have constraintson the azimuth: for each
thruster, we define a forbidden zone, of width 2A,a, and We consider the problem L(x) =9, for which an
direction ao,,, wherethe azimuthis not allowed' approximate solution x0 could be determined using an
approximateexpressionof I. We will approachthe function
Approximate solution without constraints I to first order (in one dimensionwe can say that we consider
First we apply the Lagrange multipliers method to the Z almostequalto its tangentat that point, as shownin Figure
approximateproblem with m = 2 without any constraintson 2):
thrust and azimuth. The Lagrangian of our minimization
problem is composedof the total power (8) and the three L(x) x L(x)+ L'(x)(x-xo) ( 16a)
constraints(9a-c):
A betterapproximationx, is then obtainedby solving
ntz)1F, (r;- P.n<Fl
D=P!"?) (11)
0 = Z(xo)+Z'(xo)(x,-ro) (16b)

CopyrightO 2011by ASME


minimization algorithm : we needto have a first evaluationof
This leadsto the following iterativeprocess: F to be ableto implement(an approximationofl the constraint
in a linear way. To handlethe actuatorslimitationswe define a
L(x") subset J listing all indices I correspondingto saturated
i n - t = X n"- T - (16c)
L' (x,) thrusters.The equationsfor the maximum thrust constraintof
thoseactuatorswill thenbe addedin the problem.The resulting
setof equationscanbe written againas in (14), where
The monotonyof L guaruntees
the convergence.

Applying this method to | ,,', Rr M'(F)I [F.l _ [ o:",,I


n=l ^- o r , , o r , " (l r' t)=l , I ' u = l ! = |
L =VF3t2) (17)
lu@r) o"(r),, o"(rr,"(r)
) I,lt) LD(4)j
(24\
we have to solve
where M is a N("f x2N matrix and b Fu is a N(-/)xl
(F@)f(t+t) = CQt2)(F(t))F(t) -y p<!z)1p{*)7
C<3tz) (18) vector, where N(J) is the number of indices in -/ . In the
initial step,we take J empty, that is to say that we solve (14-
where 6t:/z) i, the Hessian matrix og ptlzt . l5). We obtainan initialallocation{ tnattakesno limitations
into account.This initial allocation is then used to start the
Starting from the solution Fo of the problem (14-15), the iterationprocess:we computefor eachactuator
iterative processis then:
7,,^,,(a,): f*,,(arctan(Fr,,fF,,,)) (25)
^'
[ct3r,t1Ftt';lf.,*.',.l=
L R o,,,ll B I , If the azimuth a, is in the forbiddenzone,we take the closest
(l e)
I )F,:- yp{tz)
cot,,(F,*, (F,-,
)l azimuth allowed (on the edge of the forbiddenzone), say ai ,
and set:
L & j
F,,, =4(o'o).oroi *d Fr,, =4("'u)sina, (26)
Exact sofurton with constraints
Taking into account the maximum thrust constraint,the
Indeed,even if the azimuthis forbidden,the thrust allocatedto
approximate Lagrangian becomes:
this thruster might be low. Therefore,we prefer to keep its
value,andto constructthe matrix M andcorresponding entries
p,it)=pJ?'(4- p. R(F)- p. uth
A(2)(F, (20) in the right-handsidevectorwe would ratherusethe equation

with
and(12) is supplemented = (Fi ,)' + (Fr,,)'+ (7,' )'
zF:,tF,,t+ 2Fi ,,Fr,, ( 2i)

V - A ( 2 )= d (2r)
This definesboth M and D F, and will keepthe azimuth

He n c e( 13a)and( l3b ) b e c o me : angle a, frxed.Notethat by settingthe azimuthto the closest


edgeof the forbiddenzone,we risk to havea swapfrom one
Zc,F,.,+ p, - !ip" *24F,,, :0 (22a) edgeto the otherover the time steps.To avoid this, we keepin
2c,Fr,.i+p, * xiP, *2P,Fr,, =0 (22b) memorythe previoustime stepazimuthand if it is alreadyon
the edgeof the forbiddenzone,we will choosethe sameedge.
If the constraint(10) is violated,we set {tnew)so that
The gradient with respect to p does not change, hence (l3c-e)
still hold. The vanishing of the gradient with respect to It
fr(ncw)
. = 4,.-,,(4,)
- _J ;p,01
p\ots) (28)
yields: r,, i
7(old)-'

Fl,,+ Fj,,= T3^*,, (23)


In both cases,the index i is addedto J , and we iteratethe
processwith a non-emptymatrix M this time. The iteration
Theseequationsare non-linearin F, and this preventsus from stops when J does not change any more, which at least
implementing directly the maximum thrust constraint in the

4 CopyrightO 2011bYASME
happenswhen all the actuatorsare saturated.Sometimesit will AND RECOMMENDATIONS
CONCLUSIONS
thennot be possibleto matchthe requiredforces.
Based on the results of the comparison between the
Startingfrom the solution of (14, 24), Newton'smethodis used existing Lagrange allocation method and the improved method
asin (18),andleadsto solve: described in this paper, the following conclusions were drawn.

(FG))
CGt2) Rr M, ( Fo )-lIF'*." 1. We have developed a method to solve the thruster
allocation with minimrzation of the exact power. The use
R or,, o li ll -p of the Newton-Raphson method is to be recommended:
M(FJ o o depending on the configuration, it may lead to significant
l L p (2e) power (energy) savings and there are no drawbacks to its
- vp(ttzt(FGr)
lcorztlFt*,)F,: use (no loss of accuracy, satisfying computation time).
An iterative process has also been studied to take the
t % 2.
actuators limitations into account. This process includes
L 'tql some subtle points about the way to handle the forbidden
zone of the azimuthing thrusters, which make the
At the end of the iterative process, we obtain an allocation of algorithm time-dependent. The algorithm can handle all
forces that minimizes the power computed in the exact way, tlpes of thrusters
respects the actuators and matches the required forces in the 3. This results in a very well-balanced allocation with a sound
best achievableway. mathematical-physical basis: it both matches the
requirements and it respects the limitations of the
acfuators.
RESULTS
However, some improvements may still be made. Here we
The test configurationconsistsof 4 starboardthrustersand give some suggestions that may be interesting to explore in the
4 port sidethrusters,seeFigure 3 for an outline and Table I for future.
the main dimensionsand the thrustercharactetistics.
When the minimization algorithm fails
The required surge and sway forces and yaw moment that As we have shown previously,by adding equationsto the
we use as input for the allocation algorithm are taken from a minimization problem we may arrive to a point where the
model test and have been filtered to take out the wave systemhasno solution:
frequencyvariations.The requiredforces do not representany
specific environment,but are the output signals of the DP o Either the 3 equationsfor forcesand momentrequirements,
systemscontroller, see Figure l. The results correspondto combinedwith the maximum thrust constraintsequations,
simulationsfor a 5 minutestime interval. Note, however,that fully determine the allocation. This happenswith very
each time step is solved independentlyfrom the others. To simple configurations,but is very rare with more complex
'-l-loop',we
demonstratethe interestof the re-allocationin the configurations:in complexconfigurations,thereare a lot of
- -
first present for referencepurposes the resultsobtainedby degreesof freedom,and there will be differentpossibilities
solvingthe allocationproblem without limitationsfor the thrust to meet the requirements(when it is possible).In these
and azimuth.As can be seenfrom Figure 4, the solutionsmatch cases,we need to add equationsto chooseone possibility:
exactly the required forces (with a relative error ( lO-tt). that is the role of the energyminimization.
However, the allocation algorithm asks some thrusters to o Or the total thrust required is too high, and due to the
performbeyondtheir maximum thrust, as shownin Figure 5. If actuatorslimitations it will not be possibleto reach it: the
we truncate these thrust values, then the total forces and allocationproblemhasno solution.
momentare quite different from the requiredvalues,seeFigure
6. It is noted that this is a quite crude method to deal with To handle thesesituations(which are pointed out by the
thrustersaturationand that allocationalgorithmsin real-life DP nullity of the systemmatrix determinant),we suggest:
systemswill probablyuse a more advancedapproach.
Figures7 and 8 demonstratethe effect of the 7-loop'. The o First, try to solvethe linear systemobtainedfrom the forces
allocationalgorithmnow accountsfor the maximum thrust and and moment requirementsand actuatorslimitations' This
forbidden zone restrictions immediately. It re-distributesthe will give the only solution of the allocationproblem if it
extra forcesleft by the truncationof the thrust of PSI and PS2 exists. As it happensonly for very simple and unrealistic
to the other thrusters:we seethat now at somepoint the other configurations,this hasnot beenimplementedyet.
thrustersreachtheir maximum capacitytoo. When all thrusters o Then, if it appearsthat the requirementscannotbe met, we
are saturated,the required forces may not be exactly matched, should look for a compromise.As a matter of fact, we
but the accuracyis still satisfactory(relativeerror < l0-'). cannot just set automatically all the actuatorsto their

CopyrightO 2011byASME
maximumthrust value, becauseit may lead to a global yaw Figure l: Schematicoverview of a DP system.
moment very different from the required one. For the REQUIRED
POSITION
moment, we compare the accuracy of the different
allocationsin the 7-loop' in terms of yaw moment, and
choosethe bestone.This is a way to find a compromise,but
it might be done in a more systematicand accurateway, for
instanceby using a penaltyfunction.

Time stepdependency
The resultspresentedaboveare computedfor an input of 5
minutes,but each time step is taken independently.In reality, ALGORITH14
the actuatorscannot changetheir RPM and azimuth instantly,
and given the RPM and azimuth at time step ft, it may be
impossibleto reach the RPM and azimuth that the allocation
request at time step k+7: the time interval will not be
\b\
sufficient, and the actuatorwill only reach in-betweenvalues.
The effectivetotal forces and moment could then be different Figure 2: Schematicalrepresentation of Newton's method.Red
from the requiredforcesand moment,becausethe thrustersare curve indicatesexact function, blue and greencurvesindicate
lagging behind. This typically happensin relatively severe successive linear (tangent)approximations.
environments,close to the limitations of the vessel's
stationkeepingcapabilities.In mild environments,the thrusters
will generallybe capableof deliveringthe requestedRPMs and
azimuthangles.

TAB LE S A ND F I G U R ES

Table 1: Test configurationmain dimensions and thruster


characteristics.
Description Value
Leneth of pontoon 9 7 . 5 0m
Width of pontoon 23.20m
Distance between pontoons 32.80m
Figure 3: Test configurationwith 8 azimuthingthrusters:4 on
Max mum thrust 496 kN
the portsidefloater (PSl, PS2, PS3 and PS4) and 4 on the
Max mum power 3000kw starboardfloater (SBl, SB2, SB3 and SB4). The forbidden
zonesare indicatedas grey sectors.

--l
-J

CopyrightO 2011by ASME


Figure 4: Resultsfor test configurationwithout limitations on Figure 6: Results for test configuration with limitations on
thrust and azimuth; total surge force (top left), sway force thrust applied a posteriori; total surge force (top left), sway
(middle left), yaw moment (bottom left), thrust (top right), force (middle left), yaw moment (bottom left), thrust (top
azimuth (middle right) and power (bottom right); green line : right), azimuth (middle right) and power (bottom right); green
approximatesolution(a), blue line : exactsolution(b), red line line : approximatesolution (a), blue line : exact solution (b),
= requiredvalues(req). red line = requiredvalues(req).

1 3 13

- 1.rj
z
5 >
---
- 1,1 11]
t:
-'rz L 1.2

,l ,l

'160

s l o
z
.2,
,.j

- 25 . .180
U

5 .{ni
} .
:; - t o o,i

Figure 5: Resultsfor test configurationwithout limitations on Figure 7: Results for test configuration with limitations on
thrust and azimuth;thrust (left) and azimuth (right) of portside thrust and azimuth accounted for directly in optimization
thrustersPSl-PS4;greenline : approximate'solution (a), blue algorithm;total surgeforce (top left), sway force (middle left),
line : exact solution (b), red line : maximum thrust,red bar: yaw moment (bottom left), thrust (top right), azimuth(middle
forbiddenzone. right) and power (bottom right); green line : approximate
solution (a), blue line : exact solution (b), red line : required
lEo
I values(req).
e0l
0l
'uoI
I
1 8 0r
0
1 8 0|
t
eol
- l
'! :l -.-
ul
-so
' t B oI . 0 t 2
3
0
'100 i
|
.f D.5 sol 7 -r.:ii.-......,---
I
.-
: n d ot I

?n, 'eo
I
180 |
i -ti

I
0 I
';5i " '
1 8 0r 0 t :
;',f0, . tO
:! , $
iol' : s o
.69:
-:
: .,--..
f ,):i -.--''---*.* -;i { .so 7 l/ r '^'-,

t r o . I t t t a o E . $
E :
0 1 2 3 4 5
t X . - , ic c i
t lrinl I iftrl
-^,
.tml
0 1 : 1 2 3 4 5
I imrl

CopyrightO 2011by ASME


Figure 8: Results for test configuration with limitations on
thrust and azimuth accounted for directly in optimization R E FE R E N C E S
algorithm;thrust (left) and azimuth (right) of portsidethrusters
"Time-domainAnalysis
PSI-PS4; green line : approximatesolution (a), blue line : [] Senaris, J.J., (MARIN), for DP
exact solution (b), red line : maximum thrust, red bar : Simulations", OMAE2009-79587, OMAE Conference,
forbiddenzone. Honolulu,2009.

.- 1S0
[2] Aalbers,A.B., Jansen,R.B.H.J.,Kuipers,R.J.P.E.and van
€ n o
Walree,R., "Developmentsin dynamicpositioning systems
for offshore stationkeeping and ffioading", OMAE
,L '90
' - - 0 t ) x 1 5
Conference,Copenhagen, 1995.
_ 180
? 05 E m

[3] Aalbers, A.B. (MARIN) and Merchant, A.A. (Keppel-


Tr1 4, : I a t
N -

Pnr. j 8t . e0t
% i : i , r 5
lgo I
0 t i 3 , l 5
FELS), "The HydrodynamicModel Testingfor ClosedLoop
DP Assisted Mooring", OTC1996-8261, Oflshore
TechnologyConference,Houston,1996.
"V[/hat
[4]Van Drjk, R.R.T. and Aalbers, A.B. (MARIN),
3 uo-
Happens in Water- The use of Hydrodynamicsto Improve
';1
01 r '-I !' ;"-:*:i
| -

f
DP", MTS Dynamic Positioning Conference,Houston,
l{t ;;, , -.,*/
! : ; ,

2001.
I lr.in:

[5] Hildebrand, F.B., Introduction to Numerical Analysis, 2nd


edition.DoverBookson AdvancedMathematics, 1987.

@2011byASME
Copyright

You might also like