497
1 INTRODUCTION.
Tugboatspla y asignificantrolewhenshipsincapable
ofslowmanoeuvresarehandledinrestrictedwaters.
Shipsandtheirattendingtugsareexposedtodangers
such as collision, grounding, girting, and run‐overs
when operating in close proximity in restricted
waterways. Furthermore, the hydrodynamic
interaction forces and
moments can adversely affect
thehandlingandsafetyoftheattendingtugs. Hensen
(2012)showedthattheinteractioneffectschangewith
shiptype,widthoffairway,andthedriftanglesofthe
vessels; which can cause even experienced tug
masters difficulties in identifying safe operating
envelopesfortheirtugsduring
suchmanoeuvres.In
addition, Hensen (2012) stated that these effects
become prominent when the vessels were
significantly dissimilarin size and operatedin close
proximity during tight manoeuvres. Hensen,
Merkelbach, and Wijnen (2013) questioned 160 tug
masters with regard to their awareness of the
interaction effects during such manoeuvres. Around
30%
of the tug masters had faced critical situations
due to unexpected ship‐to‐ship interaction effects in
actualship‐assistmanoeuvres.
Shiphandlingsimulatorsuseempiricalandsemi‐
empirical methods, theoretical and numerical
methods, or potential flow methods to predict
interaction effects:(Sutulo & Soares, 2009).Withthe
exception of
the latter, the others require an
interaction effect coefficients database to solve
mathematicalmodelsfedintothesimulators,withthe
database developed and validated by empirical and
numerical techniques. For example Vantorre,
Verzhbitskaya,andLaforce(2002)conductedphysical
Accuracy of Potential Flow Methods to Solve Real-
time Ship-Tug Interaction Effects within Ship
Handling Simulators
B.N.Jayarathne,D.Ranmuthugala,S.Chai&J.Fei
A
ustralianMaritimeCollege,UniversityofTasmania,Newnham,Tasmania,Australia
ABSTRACT:Thehydrodynamicinteractioneffectsbetweentwovesselsthataresignificantlydifferentinsize
operatingincloseproximitycanadverselyaffectthesafetyandhandlingofthesevessels.Manyshiphandling
simulatordesignersimplementPotentialFlow(PF)solverstocalculatereal‐time
interactioneffects.However,
thesePFsolversstruggletoaccuratelypredictthecomplicatedflowregimesthatcanoccur,forexampleasthe
flowpassesawettransomhulloronewithadriftangle.Whenitcomestopredictingtheinteractioneffectsona
tugduringashipassist,itis
essentialtoconsidertherapidchangesofthetug’sdriftangle, asthehullacts
againsttheinflowcreatingacomplicatedflowregime.ThispaperinvestigatestheabilityofthecommercialPF
solver,Futureship®,topredicttheaccurateinteractioneffectsactingontugsoperatingatadriftangleduring
ship handling
operations through a case study. This includes a comparison against Computation Fluid
Dynamics (CFD) simulations and captive model tests to examine the suitability of the PF method for such
duties.AlthoughthePFsolvercanbetunedtosolvestreamlinebodies,itneeds furtherimprovementtodeal
withhullsat
driftangles.
http://www.transnav.eu
the International Journal
on Marine Navigation
and Safety of Sea Transportation
Volume 8
Number 4
December 2014
DOI:10.12716/1001.08.04.03
498
modelteststodeterminetheshipinteractioneffectsin
head‐on and overtaking encounters of similar and
dissimilarships.Thetestresultswereusedtocreatea
new mathematical model to improve the quality of
the interaction effects within ship manoeuvring
simulators.
Researchers such as Sutulo and Soares (2009),
Sutulo,
Soares, and Otzen (2012) and Pinkster and
Bhawsinka (2013) employing Potential Flow (PF)
solvers to predict the interaction effects as an
alternative to the excessive work and high cost
involved in developing a coefficient based model.
Currentlyonlytherelativelysimple PFdouble‐body
panel method is utilised to provide
estimates of the
interaction forces and moments in real time within
simulators (Sutulo et al., 2012). Pinkster and
Bhawsinka(2013)developed acomputerprogramto
estimateandvalidatetheinteractioneffectsusingthe
simulatoroperatedbytheMaritimeResearchInstitute
Netherland (MARIN). The PF double‐body method
was employed within their
computer program for
multi‐bodycasesinvolvingshipsandportstructures.
Real time interaction forces and moments were fed
into the simulator using high speed computers to
solve the flow equations. However, the final results
were found to be highly sensitive to the initial
conditions,whichweretedioustosetup.
Sutulo et al. (2012) developed a PF double‐body
panelcodeonthebasisoftheclassicHessandSmith
methodtoestimateinteractioneffectsinrealtimeon
commonly used computer hardware. The results
obtained with the code were validated against
experimental data obtained in deep and shallow
watertowing
tanksforatugoperatingnearalarger
vessel.Theresults illustratedthepotential ofthe PF
double‐bodypanelmethodforpredictinginteraction
effects, while highlighting the lack of accuracy in
predicting the sway forces at small horizontal
clearances, which were expected to be more
pronounced in non‐parallel
operations, similar to
those encountered during tugs assisting ships.
Fonfach, Sutulo, and Soares (2011)didexperimental
and numerical investigations to explore the
contribution of various factors to interaction effects,
whichwerenotaccountedforbythePFmethod.They
revealed substantial influence of free‐surface effects
ontheaccuracyofpredicted
interactioneffects.
Manyresearchers(Doctors,2006;Doctors&Beck,
2005; Eliasson & Olsson, 2011; Mantzaris, 1998;
Mierlo, 2006; Pranzitelli, Nicola, & Miranda, 2011)
have investigated the capabilities of PF methods to
study various hull shapes, especially transom stern
hullswithfreesurface.Pranzitellietal.(2011)studied
the free‐surface
flow around a semi‐displacing
transomsternmotor‐yachtadvancingsteadilyincalm
water using both PF method and Computational
Fluid Dynamics (CFD), and comparing them to
Experimental Fluid Dynamics (EFD) results. It was
foundthattheresultsgeneratedfromthePFmethod
weresubstantiallydifferentbecauseoftheinabilityof
itspanelsto‘roll‐down’andintersectwitheachother
duringiterations.Theresearchersconcludedthatthe
presence of the free‐surface can make more
complicated discretisation, resulting in numerical
problemsforcomplexgeometries,suchasfortransom
sternhulls.
Consideringtheinteractioneffectsonatugduring
shipassist,
therapidchangesoftugdriftanglecauses
a large portion of the downstream wake due to the
hull to be characterised by a bluff body flow in a
similar manner to a wet transom flow, as shown in
Figure1.Thus,itisessentialtoselectaflowsolverthat
can
accuratelysolvesuchconditionsduringreal‐time
predictions.Therefore,thisstudyaimstoexaminethe
accuracy of the drag force prediction of the
commercialPF packageFutureship
®
,in wettransom
conditionsasacasestudytoinvestigateitssuitability
to use in complicated real‐time interaction effects
analysisoftugsoperatingatadriftangle.
Figure1. Tug operating parallel to the flow (top) and
operatingatadriftangle(bottom)
FS‐Flow
®
is the module used within Futureship
®
for Rankine‐Source panel code analysis (DNV GL
Maritime, 2014) and it solves the boundary value
problemofpotentialtheoryincludingnonlinearfree‐
surface.Thepotentialflowapproachassumesthatthe
fluidisinviscidandtheflowisirrotationalaroundthe
bodies.Hence,FS‐Flow
®
isequippedwithaseparate
module capable of calculating the viscousresistance
in terms of a friction line in combination with the
wavy wetted hull surface. Therefore, the dynamic
forces,static forces,andviscousforcesactingon the
bodiesareincludedinthefinalresults,althoughthe
fluidis considered
asinviscidwithinpotentialflow.
Thetotalresistanceanditscomponentsobtainedfrom
the PF solver was then compared against captive
modelexperimentsandCFDresultsgeneratedbythe
commercialCFDcode Star‐CCM+
®
toinvestigatethe
possibilityofusingthePFsoftwareforfutureanalysis
ofinteractioneffects.
2 NUMERICALANALYSIS
The setup and relevant features of the two
commercial software packages, FS‐Flow
®
and Star‐
CCM+
®
,areprovidedbelow.
499
2.1 HullFormandCoordinateSystem
A1:20scaled hullmodeloftheAustralianMaritime
College’s (AMC)35m trainingvesselTV Bluefin was
utilisedinthisstudy. Theparticularsofthefull and
model scalehulls aregiven in Table1. The twotest
conditions analysed to investigate the effects
of
transomgeneratedcomplexflowregimeswere:
drytransomwithamodeldraftof0.17m;and
wettransomwithamodeldraftof0.18m.
Table1.MainParticularsoftheHullForm
_______________________________________________
MainParticularsUnit FullScale ModelScale
_______________________________________________
LengthWaterline,L m 32.150 1.608
WettedSurfacearea,S m
2
384.15 0.96
DryTransomDraft m 3.480.17
WetTransomDraft m 3.600.18
_______________________________________________
A three‐dimensional model scale hull form was
developed using the commercial software
Rhinoceros
®
V5.0 and imported into the two
packages. The coordinate system for the analysis is
showninFigure2.Theflowvelocityvectorwasinthe
positive X direction while the horizontal plane
throughtheoriginwasconsideredasthefreesurface.
2.2 DomainandMeshinFS‐Flow
®
Flow velocities ranged from 0.34m/s to 1.04m/s in
model scale, acting along the positive X direction,
withthevesselallowedtotrimandheaveduringthe
analysis. The free surface had a rectangular shape,
with the inlet boundary at a distance equal to the
scaled model waterline length (L
m) upstream of the
origin,the outletboundaryat 3L
mdownstream from
the origin, and a total domain width of 1.1L
m. The
dimensionswereselectedtomatchthoseoftheAMC
towingtank,exceptforthelength,whichwasshorten
toreducethecomputationaleffortwithoutadversely
affectingwakeresolution. Themeshconfiguration is
illustrated in Figure 2, which was developed in FS‐
Flow
®
.
Figure2.CoordinatesandShipModelwithFreesurface in
FS‐Flow
®
The mesh independence study was conducted
through mesh refinements without affecting the
stability of the solver. The drag coefficient at a
forwardspeedof1.04m/swastestedfordrytransom
condition for the models with different panel
numbers to obtain an appropriate mesh. This
approachprovidedsufficientlyaccurateresultswhile
maintaining
low computational effort. The finest
mesh investigated had 4220 panels; while a 3490
panel mesh was selected as a suitable mesh for
steady‐statesimulationsasitspredictionswerewithin
1.5%ofthatforthefinestmesh(seeFigure3).
Figure3.Absolute%differenceofDragCoefficientagainst
finestpanelmeshfortheFS‐Flow
®
model
2.3 SetupandMeshinStar‐CCM+
®
Star‐CCM+
®
uses a finite volume technique to solve
the Reynolds Averaged Navier‐Stokes (RANS)
equations (CD‐Adapco, 2014). In order to directly
compare the CFD and EFD results, the width and
depthoftheAMCtowingtankwerereplicatedinthe
numerical fluid domain, although the length was
reduced to 10.0m
to decrease the mesh load while
ensuring the pressure andwakefields generated by
thehullweresufficientlyresolvedwithinthedomain.
In addition, since the flow around the hull is
symmetrical aboutthe centerline, onlythe starboard
half ofthe hull wasmodeledin ordertoreduce the
computational domain
and thus the associated
computational effort. The vessel was fixed in all
degreesof freedom,using particulartrim andheave
conditions obtained from the FS‐Flow
®
simulation
results. The computations were performed using
hexahedraltrimmedmeshgeneratedbyStar‐CCM+
®
.
Following a mesh independence study (Figure 4), a
mesh with approximately 3.5 million cells was
selected for the investigation as the percentage
difference reduced to below 0.5% beyond this size
mesh.
The near wall spacing on the vessel is defined
usingthedimensionlessdistance(y
+
)measuredfrom
the wall surface to the edge of the first layer. The
resolution of the boundary layer was estimated by
prescribingthenumberofinflationprismslayers,the
growthrate,andthefirstnodedistancefromthewall
(
y
)reflectedbythenon‐dimensionaldistancevalue
(y
+
)asdefinedinequation(1).
13
14
me
yL y 80R
(1)
Theminimumtotalthicknessoftheinflationlayers
around the hull was matched to 2 times Prandtl’s
1/7th power law of theoretical estimate of turbulent
boundarylayerthicknessoveraflatplate,i.e.2×0.16
L
m/Re
1/7
(White,2003).
The y
+
study was conducted between y
+
~1 to
y
+
~100 with the k‐ SST turbulence model, which
changefromthelowReynoldswalltreatmentmodel
to the empirical‐based wall function formulation
around y
+
=10. From Figure 5 it is seen that the %
variationofthedragcoefficientisaround5%atay
+
of
30.Thus, y
+
~30 was selected as a compromise
between accuracy and the solver time and effort.
However, it should be noted that this y
+
value is