182
Gene a i e Algo i hms in Ea ly Ship Design:
An Explo a ion o Hull Subdi ision Gene a ion
Diego De León, NHL S enden, Leeuwa den/Ne he lands, diego.de.leon.wug@nhls enden.com
He be Koelman, NHL S enden, Leeuwa den/Ne he lands, he be .koelman@nhls enden.com
Abs ac
This pape explo es he po en ial o a da a-d i en ool o aid in he ea ly ship design p ocess, h ough
he gene a ion o subdi isions o he gene al layou ia a p oo -o -concep p o o ype which le e ages
a GAN o c ea e plausible layou al e na i es. The so wa e implemen a ion in eg a es a BSP ee
s uc u e o pa ame isa ion, and a CAD geome y implemen a ion. To wo k wi hin he in insic
limi a ions o gene a i e algo i hms, he decision-making is made by a na al a chi ec , a ge ing acili-
a ing he e alua ion o mul iple concep s and b oadening he design possibili ies. The pape desc ibes
he unc ioning o he p oo -o -concep p o o ype, conside a ions on i s c ea ion and applicabili y.
1. In oduc ion
The cu en s a e o ship design is caugh in be ween he apid de elopmen o new compu a ional
echnologies, and he challenges o a undamen al change in he indus y p opelled by en i onmen al
and legisla i e pushes owa ds deca bonisa ion and sus ainabili y.
The di e si y o solu ions needed o sus ainable p opulsion and he implemen a ion o ene gy sa ing
echnologies means ha he new design p ocesses should allow o a as e e alua ion o mul iple
solu ions, which is a change in pa adigm om p e ious me hodologies ha had an immu able cons an
in hei sou ce o ene gy.
This mul iple solu ion pa adigm leads o he ques ion o how o use hese new compu a ional
echnologies o enhance he ship design p ocess. Wi hin his p ojec he p oposed answe is he as
idea ion a he beginning o he ship design p ocess using gene a i e algo i hms, o he gene a ion o
mul iple ini ial ship layou s as a base o na al a chi ec s and enginee s o e alua e and wo k on,
accele a ing he ini ial p ocess and allowing o he conside a ion o mo e possibili ies.
2. Ship subdi ision and layou a ionale
The gene al a angemen o a ship plays a c i ical ole in de e mining i s unc ional and ope a ional
pe o mance. In he ea ly design s ages, layou decisions es ablish he ounda ion o how spaces
in e ac , how sys ems a e in eg a ed, and how u u e echnologies, such as al e na i e p opulsion can
be accommoda ed. Despi e his cen al ole, he gene al layou emains one o he leas digi ally
suppo ed a eas in he ship design p ocess.
This gap is especially e iden as he ma i ime indus y shi s owa ds g eene p opulsion sys ems and
mo e modula , adap i e essels. Al e na i e p opulsion solu ions, o en come wi h unique spa ial and
enginee ing equi emen s. T adi ional design p ocesses, elying hea ily on expe in ui ion and manual
i e a ion, s uggle o e icien ly explo e he new design spaces ha hese echnologies in oduce. A ool
ha can apidly gene a e and e alua e a wide a ie y o layou con igu a ions becomes inc easingly
aluable in his con ex .
2.1. Ene gy ansi ion challenges
The ma i ime indus y has seen a se ies o changes in i s long his o y as new echnologies become
a ailable and o e mo e p ac ical means o mo ing a ship. Mul iple ansi ions mean mul iple s udy
cases ha show how he indus y and echnological landscape has aken e e y change, He dzik (2023)
bu ce ain pa allels can be obse ed.
183
E e y ansi ion has been ini ia ed by a change in echnology, he a ailabili y o a new solu ion ha
supe sedes he p e iously dominan echnology due o p ac ical, economical o echnological easons,
c ea ing a solu ion-d i en change DNV (2019). These his o ical p eceden s o changes in he indus y
di e om he one p esen ed by he cu en deca bonisa ion challenge, whe e he u ge o eplace he
dependency on ossil uels lies no in he echnical limi a ions o uel oil i sel , bu in he ex e nal
impac s i c ea es.
This p oblem-d i en change p esen s a highe deg ee o unce ain y, as he p oposed solu ions a e e y
applica ion dependen , c ea ing a new design pa adigm whe e one o he cons an s assumed du ing he
concep ion o he essel becomes a a iable o be e alua ed.
3. Technological conside a ions and algo i hm a ionale
The in eg a ion o digi al ools such as Compu e -Aided Design (CAD), Compu e -Aided Enginee ing
(CAE), and simula ion pla o ms has long been a co ne s one o mode n ship design and enginee ing.
These echnologies enable de ailed modelling, pe o mance analysis, and i e a i e e inemen ac oss
mul iple domains: om hyd odynamics and s uc u al in eg i y o machine y layou and s abili y, Roh
(2018). Howe e , despi e hei es ablished ole, hese ools a e ypically de e minis ic in na u e: hey
ope a e unde de ined ules wi h limi ed capaci y o adap o gene alize beyond hei p og ammed
scope.
This de e minis ic cha ac e , while essen ial o alida ion and ce i ica ion, can impose p ac ical
limi a ions when explo ing as and complex design spaces o esponding o eme ging pe o mance
c i e ia. The p ocess is o en cons ained by he need o explici speci ica ion, sequen ial wo k lows,
and expe in e p e a ion, lea ing li le oom o da a-d i en in ui ion, uzzy logic, o eme gen design
disco e y, Gaspa (2018). Op imisa ion amewo ks ha e add essed some o hese challenges by
in oducing i e a i e e inemen and goal-o ien ed p ocesses, ye hey s ill ely on p ede ined
pa ame e s and objec i e unc ions.
In con as , da a-d i en app oaches, pa icula ly hose enabled by machine lea ning, o e a
complemen a y pe spec i e. These me hods shi he ocus om calcula ing exac ou pu s o lea ning
pa e ns, gene alising beha iou , and unco e ing s uc u e om da a. Whe e de e minis ic ools aim o
p ecision and ep oducibili y, da a-d i en sys ems can suppo design explo a ion and design a ia ion,
enabling new o ms o suppo in ea ly-s age design and concep ual phase. C ucially, hese app oaches
do no eplace adi ional enginee ing ools bu a he ex end hei capabili ies, b idging he gap be ween
physical modelling and compu a ional in ui ion.
3.1 Da a as a esou ce
A c i ical issue is ha mos machine lea ning models, pa icula ly supe ised lea ning app oaches,
equi e labeled da a o unc ion e ec i ely, Huang (2024). This means ha aw da a mus o en be
accompanied by anno a ions ha indica e wha i . Wi hou such con ex ual labeling, da a lacks he
s uc u e necessa y o models o lea n meaning ul associa ions o make accu a e p edic ions, Ma ko a
(2022). The p ocess o labeling da a is o en ime-consuming and esou ce-in ensi e, pa icula ly in
domains ha equi e domain expe ise as ma ine enginee ing. As a esul , he a ailabili y o labeled
da a can become a bo leneck in he de elopmen and deploymen o machine lea ning sys ems.
In enginee ing and design con ex s, including he ma i ime domain, da a o machine lea ning
applica ions o igina es om a ange o sou ces. The wo mos common sou ces in p ac ice a e
ope a ional da a and his o ical design da a.
Ope a ional da a e e s o in o ma ion gene a ed du ing he ac ual use o a sys em o p oduc , such as
senso eadings, pe o mance logs, main enance eco ds, and en i onmen al condi ions. In ma i ime
applica ions, his could include engine pe o mance me ics, uel consump ion a es, ou e acking, o
s uc u al esponses unde a ious sea s a es deGeus-Moussaul (2024). This da a p o ides aluable
184
insigh s in o how designs pe o m in eal-wo ld condi ions, suppo ing asks like p edic i e
main enance, pe o mance op imiza ion, and adap i e con ol.
Exis ing design da a, de i ed om pas p ojec s and legacy sys ems, whe he s o ed as CAD models,
simula ion esul s, o design ables, ep esen a eposi o y o enginee ing knowledge and design in en .
Machine lea ning models can use his in o ma ion as a basis o pa e n ecogni ion, benchma king, o
gene a ing new concep s inspi ed by p o en solu ions. This o m o da a euse suppo s he no ion o
"lea ning om expe ience," enabling algo i hms o build upon decades o accumula ed enginee ing
p ac ice.
As an al e na i e o da a om exis ing ships, syn he ic da a is a i icially gene a ed a he han collec ed
om eal-wo ld obse a ions. In enginee ing and design domains, his da a is o en p oduced using
simula ion en i onmen s o ma hema ical models ha eplica e he beha iou o physical sys ems unde
con olled condi ions. Syn he ic da a se es as a aluable complemen o eal-wo ld da ase s,
pa icula ly in scena ios whe e empi ical da a is sca ce, incomple e, sensi i e, o expensi e o ob ain.
Howe e , syn he ic da a also comes wi h limi a ions, Pica d (2023). A key conce n is ideli y, whe he
he syn he ic da a accu a ely e lec s he complexi y and a iabili y o eal-wo ld phenomena. I he
da a does no cap u e impo an nuances, models ained on i may ail o gene alize o may o e i o
a i icial pa e ns. Ano he issue is he po en ial o bias in oduced by he assump ions o
simpli ica ions embedded in he gene a ion p ocess.
4. P o o ype de elopmen
The i s p o o ype de eloped in his p ojec demons a es a ull pipeline o p ocedu al ship layou
gene a ion based on a Bina y Space Pa i ioning (BSP) ee s uc u e. A i s co e, his p o o ype
illus a es how ea ly-s age layou decisions can be algo i hmically gene a ed, geome ically modeled,
and e alua ed, all wi hin an open, modula amewo k designed o u u e ex ensibili y.
4.1. CAD geome y backg ound and BSP
To suppo he de elopmen o a gene a i e layou p o o ype, he choice o a sui able compu e -aided
design (CAD) backend is a c i ical decision. In his con ex , OpenCASCADE o e s a powe ul oolki
o 3D modeling and compu a ional geome y.
This solu ion is also pa icula ly well-sui ed o implemen ing a Bina y Space Pa i ioning (BSP) ee
app oach, which is cen al o his p ojec ’s layou gene a ion logic. BSP ees p o ide a s uc u ed way
o ecu si ely subdi ide a design space in o unc ional compa men s, making hem ideal o
ep esen ing compa men alized layou s such as ship in e io s, De Koning e all (2011).
OpenCASCADE’s geome ic and opological modeling capabili ies allow o p ecise and p agma ic
c ea ion, manipula ion, and isualiza ion o he pa i ions de ined by he BSP s uc u e. This
compa ibili y enables seamless in eg a ion be ween abs ac spa ial logic and conc e e geome ic
ep esen a ion. Each node in a BSP ee can be di ec ly mapped o a olume ic shape o compa men
in OpenCASCADE, ensu ing ha layou gene a ion emains bo h compu a ionally e icien and
geome ically meaning ul. This syne gy allows he p o o ype o ansi ion smoo hly be ween da a-
d i en logic and enginee ing wi h a alid geome y, an essen ial ea u e o ea ly design wo k lows ha
combine algo i hmic gene a ion wi h CAD in eg a ion.
4.2. Gene a i e Ad e sa ial Ne wo k
To add ess he challenge o gene a ing plausible and di e se gene al layou s o ships, his p ojec
implemen s a Gene a i e Ad e sa ial Ne wo k (GAN) a chi ec u e. GANs a e a class o machine
lea ning models pa icula ly well-sui ed o gene a i e asks, wi h some exis ing applica ions wi hin
ship design, Khan (2023).
185
A he co e o a GAN a e wo neu al ne wo ks wi h opposing goals: he gene a o and he disc imina o .
The gene a o a emp s o c ea e candida e layou s ha mimic eal designs, s a ing om andom inpu
(noise) o guided design pa ame e s. The disc imina o , in con as , e alua es he alidi y o he
gene a ed layou s by ei he compa ing hem o a da ase o eal examples o using o he ypes o il e s.
The wo ne wo ks engage in a ze o-sum game: as he gene a o imp o es i s abili y o ool he
disc imina o , he disc imina o simul aneously becomes be e a de ec ing syn he ic designs. This
ad e sa ial aining dynamic d i es bo h ne wo ks o imp o e con inuously, esul ing in p og essi ely
highe -quali y gene a ed ou pu s, C esswell (2018).
A key ad an age o his app oach lies in i s modula i y and adap abili y. The gene a o and
disc imina o can be de eloped, ained, and adjus ed independen ly. This lexibili y allows o explo e
a a ie y o a chi ec u es, lea ning s a egies, and inpu ea u es wi hou ebuilding he en i e sys em.
The disc imina o is no limi ed o being a single neu al ne wo k. The modula i y o he model allows
he explo a ion o he use o non-neu al, compu a ional disc imina o s. Exis ing enginee ing ools can
ac as e alua o s ha p o ide a judgmen o gene a ed layou s. This app oach aligns well wi h he hyb id
na u e o ea ly-s age design, whe e domain knowledge and enginee ing logic s ill play a c ucial ole.
By allowing he disc imina o o include enginee ing ma hema ical models, simula ion ools o ule-
based alida o s, he sys em gains a signi ican ad an age in p oducing ealis ic, and also unc ional
and cons ain -complian designs. This ype o e alua ion also acili a es he use o en e speci ic ypes
o cons ain s o ob ain he desi ed ou pu .
Ano he impo an bene i is he e sa ili y o he model amewo k. Bo h ne wo ks can be eplaced o
enhanced wi h o he machine lea ning me hods, such as au oencode s, ein o cemen lea ning agen s,
o decision ees, depending on he goals o a speci ic design ask. The GAN amewo k becomes a
lexible expe imen al sandbox in which new layou s a egies can be es ed and imp o ed i e a i ely.
The use o a GAN allows his p o o ype o go beyond ule-based gene a ion by lea ning design pa e ns
di ec ly om da a. This makes i possible o suppo ea ly-s age designe s no jus wi h s a ic empla es,
bu wi h dynamically gene a ed layou s ha espond o lea ned design p e e ences and can e ol e
h ough aining. Combined wi h a human-in- he-loop wo k low, his app oach opens he doo o a
powe ul new class o design ools ha suppo bo h au oma ion and expe o e sigh .
5. Implemen a ion and wo k low
A design inpu is expec ed om he na al a chi ec u ilising he ool and om he enginee ing iles, a
his s age he .iges ile. Fo his p oo o concep p o o ype he inpu expec ed includes he ship
dimensions, basic ope a ional equi emen s such as expec ed minimal ange and desi ed ca go olume,
and basic expec ed design decisions o es , such as he uel ype o e alua e. These equi ed designs
will be subjec o change depending on he implemen a ion o da a exchange wi h o he so wa e ools
wi hin he pla o m and u u e capabili ies o he equi emen and enginee ing check modules o he
disc imina o .
The p ocess begins wi h he andom gene a ion o a BSP ee, which de ines a hie a chical spa ial
subdi ision o a cubic design olume. This olume ep esen s he in e nal space o he ship, abs ac ed
o allow lexible pa i ioning wi hou ye being cons ained by he hull shape. The gene a o is cu en ly
d i en by andom numbe gene a o s bu i is designed o be eplaced by a neu al ne wo k in u u e
i e a ions, enabling be e da a-d i en pa i ioning s a egies.
Once he BSP s uc u e is gene a ed, i is mapped in o a se o 3D compa men s using
OpenCASCADE’s solid modeling ools. The esul ing subdi ided solid se es as he i s s age o he
layou ep esen a ion. To en o ce geome ic ealism, a Boolean sub ac ion ope a ion is pe o med
using a gi en hull o m. This ope a ion ims he subdi isions o i wi hin he a ailable in e nal olume
o he essel, ensu ing ha gene a ed layou s emain wi hin easible spa ial bounds.
186
Fig.1: D^3: SEA gene al unc ioning p inciple
Following his, each compa men is e alua ed in e ms o posi ion, size, and olume, p o iding a basis
o ea ly-s age layou analysis. These me ics se e as eedback o he gene a o , o ming a simpli ied
o m o disc imina o logic. The disc imina o , in his ini ial p o o ype, applies a se o ule-based and
nume ical il e s, such as minimum compa men olume o geome ic alignmen , which help iden i y
and disca d in alid o poo ly o med layou s. This i s s ep se s he s age o u u e imp o emen s
whe e disc imina o s may inco po a e his o ical layou da a, enginee ing ules, o e en exis ing digi al
ools, including hose being de eloped in he EU Ho izon SEUS pla o m.
All layou da a, including he BSP ee and esul ing geome y, a e s o ed in a .json o ma , o e ing
human eadabili y, logging, and in eg a ion wi h o he sys ems such as SARC PIAS subdi ision ools.
Fo CAD in e ope abili y, he OpenCASCADE a chi ec u e he p o o ype is buil upon also suppo s
expo in .STEP and .IGES o ma s, ensu ing compa ibili y wi h s anda d enginee ing wo k lows and
ools. This lexibili y posi ions he p o o ype as a con ibu o o he b oade SEUS digi al pla o m,
enabling downs eam in eg a ion wi h o he ools o e alua ion, simula ion, o isualiza ion.
Whe e each subdi ision is desc ibed by he plane o which i is pa allel and he ac ion o he olume
a which i happens. The base subdi ision con inues h ough a "b anch" di ided by igh and le pa hs
o he BSP ee un il i inishes on a "lea ", whe e ex a in o ma ion can be added, in his case olume
and he in ended use o he subdi ision, o enginee ing use.
6. Resul s and analysis
The i s expe imen s wi h he p o o ype we e ca ied ou in h ee s ages. In he i s s age, BSP ees
we e gene a ed wi hou applying any cons ain s, p oducing pu ely andom subdi isions wi hin he
design olume. These esul s illus a ed he baseline beha iou o he gene a o , bu also con i med ha
comple ely uncons ained subdi isions ha e limi ed alue e en o ea ly-s age idea ion. In he second
s age, a cons ain was in oduced o ensu e a p esc ibed olume ic balance be ween uel and ca go
spaces. In he hi d s age, a minimum deck heigh cons ain was added alongside he olume
equi emen ; his con igu a ion was es ed bu no e alua ed in dep h wi hin he cu en wo k.
187
Fig.2: Simple olume dis ibu ion il e based subdi isions
The layou s gene a ed a his s age emain p elimina y and abs ac . While hey comply wi h he
imposed geome ic il e s, hey do no ye exhibi he cohe ence o unc ionali y expec ed o p ac ical
ship layou s. In hei cu en o m, he ou pu s a e no di ec ly applicable o enginee ing use. This
dis inc ion unde lines he di e ence be ween algo i hmically gene a ed esul s and layou s designed by
human labou . A na al a chi ec would no only espec geome ic and olume ic cons ain s, bu
would also inco po a e a wide ange o implici , “common sense” conside a ions. Fo example, simple
concep s such as a basic le el o symme y, which is a common ea u e in ship design due o bo h
aes he ic and p ac ical conside a ions, a e no accoun ed o in his e sion o he algo i hm. These
laye s o design logic a e di icul o cap u e h ough pu ely nume ical il e ing and highligh he gap
be ween algo i hmic easibili y and p o essional design p ac ice.
The expe imen s also e eal he inhe en complexi y o in e nal subdi ision in ship design. E ec i e
compa men alisa ion is a mul i- a iable p oblem, whe e while decisions can be educed o a se o
nume ical ules as by enginee ing p ac ice, a comple e sys ema ic analysis would equi e a e y high
numbe o conside a ions. Al hough ha d cons ain s p o ide a necessa y ounda ion, he le el o
complexi y may bene i om mo e abs ac easoning. Fo his eason, he use o exis ing subdi ision
da a as e e ence pa e ns eme ges as an impo an complemen a y s a egy. By lea ning om
es ablished examples, he algo i hm may app oxima e mo e complex design ules ha canno easily be
o malised, exploi ing he known ad an ages o da a-d i en app oaches.
Ano he obse a ion is ha he p o o ype is no in ended as an op imisa ion amewo k. T adi ional
op imisa ion me hods a ge pe o mance measu es and ope a e wi hin ixed se s o cons ain s and
objec i es. In con as , he p esen app oach seeks o suppo as e idea ion a he ea lies design s ages.
I s alue lies in he capaci y o gene a e mul iple plausible al e na i es quickly, enabling designe s o
e alua e a b oade ange o possibili ies han would be p ac ical h ough manual i e a ion alone. In his
sense, he p o o ype is posi ioned as an augmen a ion o human design capabili ies, no a eplacemen
o hem.
While he esul s con i m he easibili y o BSP-based gene a i e subdi ision and demons a e he
po en ial o he app oach as a ool o apid concep gene a ion, hey also highligh he necessi y o
addi ional laye s o e alua ion logic, inco po a ion o e e ence da a, and sus ained human o e sigh .
The ool’s ole is no o deli e inal o op imised layou s, bu o accele a e he idea ion p ocess and
augmen he designe ’s capaci y o explo e mul iple pa hways in he ea ly phases o ship design.
188
7. Conclusions and u u e wo k
The p o o ype demons a es he easibili y o using BSP ees o algo i hmic gene a ion o ship
subdi isions and alida es he app oach as a ounda ion o u he de elopmen . I s p ima y
con ibu ion is no he deli e y o op imised o di ec ly applicable layou s, bu he demons a ion o a
gene a i e amewo k capable o p oducing apid concep al e na i es. By accele a ing ea ly-s age
idea ion, such ools can augmen he wo k o na al a chi ec s, p o iding a b oade ange o candida e
layou s o conside a he ou se o he design p ocess.
The esul s also con i m ha human supe ision emains indispensable. While he gene a ed layou s
a e geome ically consis en , hey lack he con ex , expe ience and o he implici “common sense”
knowledge ha human designe s na u ally apply. This ein o ces he need o a human-in- he-loop
wo k low in which he algo i hm ac s as a gene a o and he designe as e alua o and decision-make .
Fu u e de elopmen will ocus on h ee complemen a y di ec ions. Fi s , he in eg a ion o ex e nal
so wa e solu ions, al eady pa o he b oade p ojec pla o m, o e s a pa h o apidly ex ending he
se o e alua i e il e s a ailable. By d awing on es ablished ools a he han de eloping e e y
componen in isola ion, i becomes possible o conside ably aise he quali y and ealism o he
gene a ed esul s. Inc easing he numbe and sophis ica ion o il e s is expec ed o di ec ly ansla e
in o mo e cohe en and p ac ically ele an layou s, wi hin eason and he capabili ies o he ools o
explo e.
Second, he inclusion o aining da a om exis ing subdi isions will enable he sys em o mo e beyond
pu ely nume ical cons ain s. Lea ning om his o ical o e e ence layou s allows he gene a o o
app oxima e he complexi y o design logics in a simila way ha human expe ience se es as a sho cu
o he complex enginee ing ma hema ics in he design p ocess. Howe e , one o he main limi a ions
in his di ec ion is he a ailabili y o subdi ision da a in o ma s ha a e sui able o di ec aining.
O e coming his ba ie will be essen ial o ully le e aging he po en ial o da a-d i en app oaches,
and po en ial app oaches exploi he use o syn he ic da a and machine-lea ning-based da a agging and
ex ac ion ools.
The p o o ype es ablishes a solid ounda ion o BSP-based gene a i e subdi ision and demons a es
i s po en ial as a ool o accele a ing idea ion in ea ly-s age ship design. I s u he de elopmen will
depend on en iching he disc imina o h ough in eg a ion wi h ex e nal so wa e and embedding
knowledge de i ed om exis ing subdi isions, ul ima ely enabling a hyb id app oach whe e
compu a ional gene a ion and human expe ise wo k in andem.
Acknowledgemen s
Wi h s ong g a i ude o he EU Ho izon SEUS p ojec and i s pa ne s: NTNU, NHL S enden, UTU,
SARC B.V., CADMATIC, Gondan Shipya ds & Uls ein Shipya ds. To he Eu opean Union as he main
sponso o he p ojec .
Re e ences
CRESWELL, A.; WHITE, T.; DUMOULIN, V.; ARULKUMARAN, K.; SENGUPTA, B.;
BHARATH, A.A. (2018), Gene a i e ad e sa ial ne wo ks: An o e iew, IEEE Signal P ocessing
Magazine 35/1, pp.53–65
DE GEUS-MOUSSAULT, S.; SEUBERS, H.; LINSKENS, H.; CORADDU, A.; PRUYN, J. (2024),
Ope a ional Da a o Sea Ma gin Calcula ions in Ea ly Ship Design, 15 h IMDC Con ., Ams e dam
DE KONING, D.; KOELMAN, H.J.; HOPMAN, H. (2011), A no el ship subdi ision me hod and i s
applica ion in cons ain managemen o ship layou design, 10 h COMPIT Con ., Be lin, pp.292–304
189
DNV (2019), Compa ison o al e na i e ma ine uels, h ps://sea-lng.o g/wp-con en /uploads/2020/04/
Al e na i e-Ma ine-Fuels-S udy_ inal_ epo _25.09.19.pd
GASPAR, H.M. (2018), Da a-d i en ship design, 9 h COMPIT Con ., Pa one
HERDZIK, J. (2023), Ma ine Fuel F om The Pas o The Fu u e, Scien i ic Jou nals o he Ma i ime
Uni e si y o Szczecin, h ps:// eposi o y.am.szczecin.pl/handle/123456789/2780
HUANG, Q.; ZHAO, T. (2024), Da a Collec ion and Labeling Techniques o Machine Lea ning,
h ps://a xi .o g/abs/2407.12793
KHAN, S.; GOUCHER-LAMBERT, K.; KOSTAS, K.; KAKLIS, P. (2023), Shiphullgan: A gene ic
pa ame ic modelle o ship hull design using deep con olu ional gene a i e model, Compu e
Me hods Appl. Mech. Eng. 411
MARKOVA, V. (2022), Impac o Da a P ep ocessing on Machine Lea ning Pe o mance,
Academia.edu, h ps://www.academia.edu/67070662/Impac _o _Da a_P ep ocessing_on_Machine_
Lea ning_Pe o mance
PICARD, C.; SCHIFFMANN, J.; AHMED, F. (2023), DATED: Guidelines o C ea ing Syn he ic
Da ase s o Enginee ing Design Applica ions, a Xi p ep in a Xi :2305.09018
ROH, M.-I.; LEE, K.-Y. (2018), Compu a ional Ship Design, Sp inge
VEELO, B.N. (2004), Va ia ions o Shape in Indus ial Geome ic Models, PhD Thesis, NTNU,
T ondheim
