Journal of

Marine Science
and Engineering

Article
Mixed-Fidelity Design Optimization of Hull Form Using CFD
and Potential Flow Solvers
Gregory J. Grigoropoulos * , Christos Bakirtzoglou, George Papadakis and Dimitrios Ntouras

Laboratory of Ship and Marine Hydrodynamics, School of Naval Architecture and Marine Engineering,
National Technical University of Athens (N.T.U.A.), 15780 Athens, Greece; [email protected] (C.B.);
[email protected] (G.P.); [email protected] (D.N.)
* Correspondence: [email protected]

Abstract: The present paper proposes a new mixed-fidelity method to optimize the shape of ships
using genetic algorithms (GA) and potential flow codes to evaluate the hydrodynamics of variant
hull forms, enhanced by a surrogate model based on an Artificial Neural Network (ANN) to account
for viscous effects. The performance of the variant hull forms generated by the GA is evaluated for
calm water resistance using potential flow methods which are quite fast when they run on modern
computers. However, these methods do not take into account the viscous effects which are dominant
in the stern region of the ship. Solvers of the Reynolds-Averaged Navier-Stokes Equations (RANS)
should be used in this respect, which, however, are too time-consuming to be used for the evaluation
of some hundreds of variants within the GA search. In this study, a RANS solver is used prior to the
execution of the GA to train an ANN in modeling the effect of stern design geometrical parameters



only. Potential flow results, accounting for the geometrical design parameters of the rest of the hull,

 are combined with the aforementioned trained meta-model for the final hull form evaluation. This
Citation: Grigoropoulos, G.J.; work concentrates on the provision of a more reliable framework for the evaluation of hull form
Bakirtzoglou, C.; Papadakis, G.; performance in calm water without a significant increase of the computing time.
Ntouras, D. Mixed-Fidelity Design
Optimization of Hull Form Using Keywords: optimization; genetic algorithms; artificial neural networks; meta-models; multilevel
CFD and Potential Flow Solvers. J. optimization; potential flow; viscous flow
Mar. Sci. Eng. 2021, 9, 1234. https://
doi.org/10.3390/jmse9111234

Academic Editor:
1. Introduction
Abbas Dashtimanesh
Hydrodynamic hull form optimization is a very demanding task in terms of computer
Received: 10 September 2021 and time resources. In general, it is a multi-disciplinary process to take into account re-
Accepted: 28 October 2021 sistance, propulsion, seakeeping, and maneuvering characteristics of a vessel related to
Published: 8 November 2021 different sea states and wind directions as stated in Grigoropoulos et al. [1,2]. However,
even in the single objective case, where only calm water resistance is handled, the use of
Publisher’s Note: MDPI stays neutral fine grids for CFD evaluation is too time-consuming to be used in all steps of optimization
with regard to jurisdictional claims in via a Genetic Algorithm (GA). In this respect, metamodels or surrogate models have been
published maps and institutional affil- widely used in several engineering contexts, such as structural optimization, aeronau-
iations. tics, aerospace and ground or waterborne vehicles, including stochastic applications and
uncertainty quantification.
Thus, the necessity to reduce the computational effort in the optimization process
without sacrificing the accuracy of the outcome has led to the extensive use of metamodels.
Copyright: © 2021 by the authors. The latter are based on the number of high-fidelity evaluations required, since the compu-
Licensee MDPI, Basel, Switzerland. tational cost of the resulting algorithms is highly reduced [3]. Typical surrogate models are
This article is an open access article polynomial regression, kriging method, artificial neural networks (ANN), support vector
distributed under the terms and machines (SVM), or radial basis functions (RBF) [4–6].
conditions of the Creative Commons Once built, they are very fast (split seconds vs. hours of simulation). Of course,
Attribution (CC BY) license (https:// it is imperative to be ensured that a chosen surrogate model approximates the simula-
creativecommons.org/licenses/by/ tion sufficiently well to replace them for the design task at hand, at least at the level
4.0/).

J. Mar. Sci. Eng. 2021, 9, 1234. https://doi.org/10.3390/jmse9111234 https://www.mdpi.com/journal/jmse

J. Mar. Sci. Eng. 2021, 9, 1234 2 of 15

needed for engineering purposes at early stages when trends and dependencies need to be
understood [7].
There are various ways to incorporate metamodels within an EA, so there are various
metamodel-assisted evolutionary algorithms (MAEAs). Many relevant papers are based on
the use of offline trained metamodels, i.e., metamodels which are trained separately from
the evolution. On the other hand, in the variant of metamodel-assisted EAs (MAEAs) with
online trained metamodels, these are trained on the fly separately for each new population
member [8].
This paper proposes a mixed methodology to optimize the hull form for resistance
in calm water, using both potential and viscous flow codes, both with grids of suitable
density utilizing a limited number of viscous evaluations carried a priori (offline) to formal
optimization. It is well known that potential flow codes implementing Boundary Element
Method (BEM) are quite efficient and reliable in modeling the effect of various geometrical
design parameters on the hydrodynamic performance of ships in calm water for the
major part of the hull form except for the stern region. In the latter area, the viscous
phenomena dominate and the potential flow modeling is poor. The 3D, time domain, and
Rankine source potential flow code SWAN2 2002 is used for the potential flow calculation
of the wave making resistance Rw . The potential flow results for the bow and the middle
part of the hull form are combined with a surrogate model based on an ANN trained
by viscous flow results to account for the effect of the stern local design parameters on
its hydrodynamic performance. The in-house (U)RANS solver MaPFlow, described in
detail in Papadakis et al. [9,10], a cell centered CFD Solver that uses both structured and
unstructured grids, is suitable for the viscous flow calculations.
On the basis of the aforementioned discussion, an Artificial Neural Network (ANN) is
established to account for the effect of the stern only geometrical parameters, assuming the
parent form for the rest of the hull. This methodology assumes that the effect of the stern
design parameters on the performance of the whole hull is not altered significantly when
the bow and the middle design variables are modified. In other words, this assumption
is expressed in terms of the (at least partial) independence of the effects of the stern
design variables from the rest ones. This assumption is verified by the comparison of
the performance of the parent and the optimized hull using viscous flow calculations. A
further check is also included in the paper to verify the reliability of the ANN.
It is expected that a small number of variants of the stern hull form geometry as
evaluated by MaPFlow provide sufficient and reliable training to the ANN. The number of
the required evaluations depends, of course, on the hull form, magnitude of variances and
the complexity of vessel shape in the stern region. The latter is responsible for the number
of the geometrical parameters that should contribute to the hull form optimization.
To be more specific, the mixed fidelity optimization procedure, presented in this paper,
is performed in two separate steps, firstly, for the stern region and following for the middle
and bow region. Once, the ANN is trained sufficiently to constitute a reliable metamodel,
the first optimization step consists of ANN function’s minimization, in order to derive the
combination of stern parameters that minimize hull viscous resistance RTOTAL , as predicted
by the ANN. On the second step, the bow and middle ship geometrical variables are
optimized while maintaining the parent form for the stern of the hull in order to minimize
wave resistance Rw . Finally, all the geometrical variables, those in the bow and the middle
section of the hull, as well as those in the stern region, as tuned by the two optimization
cycles are combined in one fair hull. This final hull is evaluated by direct comparison with
CFD modeling employing a grid of the same density. The KRISO Container Ship (KCS) has
been used as a test case.
The major contribution of the work is that limited viscous flow calculations are carried
out a priori (offline) to train the ANN in order to model efficiently stern viscous effects
which dominate in relatively low Froude numbers, such as in the case examined. In this
way, a reliable optimization framework is achieved with minimum resources to estimate
J. Mar. Sci. Eng. 2021, 9, 1234 3 of 15

stern viscous pressure effects, encompassing at the same time the advantages of potential
flow codes to guide the optimization of the rest of the hull form for wave resistance.

2. Hydrodynamic Hull Form Optimization

The problem of hydrodynamic hull form optimization, focusing on its performance in
calm and rough water, has been studied since the early 1980s by many authors. The use of
parametric models with genetic algorithms to carry out a dual objective optimization for
calm water resistance and seakeeping is described in detail by Grigoropoulos et al. [11].
The methodology is ruled out by the reliability of the computer codes used for the eval-
uation of the hydrodynamic performance of the variant hull forms and resources needed.
They actually drive the optimization algorithm to a realistic optimum solution. Although
viscous flow calculations provide more reliable estimates of calm water performance, they
are too time-consuming to be used in each of the hundreds of steps of genetic algorithms.
Thus, a way to tackle this difficulty is to use potential flow solvers in the optimization
procedure and to verify the result by a viscous flow code. However, potential flow codes
are not reliable in the stern region of the ship, where viscous flow phenomena prevail.
In order to remedy this situation, it is proposed to use an Artificial Neural Network as
a meta-model to account the effect of variations of the stern shape. This ANN is trained
using a few runs of the viscous flow code.

3. The Optimization Procedure

3.1. Optimization Method
The problem we have to solve is an optimization problem where the displacement is
allowed to vary up to 1% of its initial value. The space of the design variables’ variance has
been specified by trials during a preparatory stage, to ensure realistic hull form variants.
The mathematical formulation of the optimization problem is as follows:

min R( Hi , Fn, SS)

w.r.t.l ≤ i ≤ u and D∆ ≤ const

where
R : Hull’s resistance as evaluated by potential solver or ANN.
i: Design variable vector; l and u are the lower and upper bounds, respectively.
Hi : Hull shape as affected by the changes in i parameters.
Fn : Froude number of the vessel.
SS : Sea state, calm water for cases evaluated.
D∆ : Variation in the volume of displacement.
For the optimization process, the NSGA II (Non-Dominated Sorting Genetic Algo-
rithm II), Figure 1, as described by Deb et al. [12], and which is provided inside CAESES
software [13], is selected. The architecture of the artificial neural network used for the
training is discussed in the next paragraph. Details of the genetic algorithm used in the
optimization process are briefly described below:
• A number of variant geometries is generated.
• An equal number of off-springs is formed.
• The total number of parents and offspring is then sorted to levels according to non-
domination.
• The geometries of each level are ranked with respect to their crowded distance of each
solution in the population.
• A new generation is being produced with a population number equal to the initial one.
• Steps 2 to 5 are repeated.
 step, comprised of 32 generations with 14 offspring each. In the second step, where the
five bow and middle-part hull design variables were investigated, 37 generations with 24
offspring each produced in total 888 alternative hulls. The mutation and crossover prob-
ability rates were set to 0.04 and 0.92, respectively, in both cases. In both steps, the modi-
J. Mar. Sci. Eng. 2021, 9, x FOR PEERfication
REVIEW of the design variables is controlled by the GA, while CAESES automatically cre-
J. Mar. Sci. Eng. 2021, 9, 1234 4 of 15
4 of 15
ates fair ship lines for each combination of them. The theoretical background of CEASES
is described by Harries and Abt [14].

For the two optimization cycles performed, an adequate number of generations and
off-spring was selected in order to ensure convergence for all parameters. A total of 448
variants were evaluated concerning the three stern design variables parameters in the first
step, comprised of 32 generations with 14 offspring each. In the second step, where the
five bow and middle-part hull design variables were investigated, 37 generations with 24
offspring each produced in total 888 alternative hulls. The mutation and crossover prob-
ability rates were set to 0.04 and 0.92, respectively, in both cases. In both steps, the modi-
fication of the design variables is controlled by the GA, while CAESES automatically cre-
ates fair ship lines for each combination of them. The theoretical background of CEASES
is described by Harries and Abt [14].
Figure 1.
Figure 1. NSGA-II
NSGA-II procedure.
procedure.

3.2. Design
For theModification
two optimization Procedure cycles performed, an adequate number of generations and
off-spring was selected in
CAESES software has been selected order to ensure forconvergence
the parametric for design
all parameters.
and variationA total of
of the
448 variants
parent hull. were evaluated concerning the three stern design variables parameters in the
first step, comprised
The initial geometry of 32isgenerations
representedwith by a14 setoffspring each. Inproviding
of basic curves the second step, where
topological in-
the five bow
formation inand
the middle-part
longitudinalhull design(design
direction variables were investigated,
waterline, centerline, 37 generations
deck-line) and with
a set
24 offspring
of 32 sectioneach curves. produced in total
All of them are888 alternative
either F-splineshulls. The mutation
or B-splines. F-splinesandarecrossover
used to
probability rates were set to 0.04 and 0.92, respectively,
describe areas or characteristic lines subjected to variation, which directly affect in both cases. In both steps, the
the geo-
modification of the design variables
metrical hull form parameters to be optimized. is controlled by the GA, while CAESES automatically
createsThefair ship linesisfor
geometry each
split intocombination
three regions: of them.
the mainThe hull,
theoretical background
the stern region, and of CEASES
the bow
Figure
is 1. NSGA-II
described by procedure.
Harries and Abt [14].
bulb, assigning specific design variables for each of them in order to ease the optimization
process. Hull form is described by different kinds of surfaces which reflect the changes at
3.2.Design
3.2. DesignModification
ModificationProcedure
Procedure
the parameters under investigation. Surfaces are generated either by interpolating the par-
CAESES software
CAESES
ametric-modeled software
section has
has beenor
been
curves selected
selected
by using forthe
for theso-called
the parametric
parametric design
design
engine andvariation
and
curves. variation
The ofofthe
approxima- the
parent
parent hull.
hull.
tion of the initial surfaces is very satisfactory and allows for the establishment of the eight
The
The initial
initial geometry represented by
is represented byaaset set ofbasic
basiccurves
curves providing topological
design variables in total. Five of them refer to theofbow bulb andproviding
the main hull,topological
while the in-
information
formation ininthethelongitudinal
longitudinaldirection
direction (design
(design waterline,
waterline, centerline,
centerline, deck-line)
deck-line) and anda a
set
remaining three at the stem region. Figure 2 depicts the geometry delivered by CAESES
set of 32 section
of 32 section curves.
curves. All of them are either F-splines or B-splines. F-splines are used
software, while FigureAll of themthe
3 presents arecontrol
either F-splines
lines and the or B-splines.
surface ofF-splines are used to
the bow bulb.
to describe
describe areas
areas or or characteristic
characteristic lineslines subjected
subjected to variation,
to variation, which
which directly
directly affect
affect the
the geo-
geometrical
metrical hull hull
form form parameters
parameters to be
to be optimized.
optimized.
The
Thegeometry
geometryisissplit splitinto
intothree
threeregions:
regions:the themain
mainhull,
hull,the
thestern
sternregion,
region,andandthethebow
bow
bulb,
bulb,assigning
assigningspecific
specificdesign
designvariables
variablesfor foreach
eachofofthem
themininorder
ordertotoease
easethe
theoptimization
optimization
process.
process. Hull
Hull form
form is described by
is described by different
different kindskindsof ofsurfaces
surfaceswhich
whichreflect
reflectthe
thechanges
changesat
at the parameters under investigation. Surfaces are generated
the parameters under investigation. Surfaces are generated either by interpolating the par- either by interpolating
the parametric-modeled
ametric-modeled section section
curves or curves or by
by using theusing the so-called
so-called engine curves.engine Thecurves. The
approxima-
approximation
tion of the initial of surfaces
the initial is surfaces is very satisfactory
very satisfactory and allowsand allows
for the for the establishment
establishment of the eight
of
design variables in total. Five of them refer to the bow bulb and the main hull,main
the eight design variables in total. Five of them refer to the bow bulb and the while hull,
the
while the remaining three at the stem region. Figure 2 depicts
remaining three at the stem region. Figure 2 depicts the geometry delivered by CAESES the geometry delivered by
CAESES
software,software,
while Figure while3Figure
presents 3 presents
the controlthe control
lines and lines
theand the surface
surface of thebulb.
of the bow bow bulb.

Figure 2. The geometry of KRISO containership as derived by CAESES software.

Figure2.2.The
Figure Thegeometry
geometryof
ofKRISO
KRISOcontainership
containershipas
asderived
derivedby
byCAESES
CAESESsoftware.
software.
J. Mar. Sci. Eng. 2021, 9, 1234 5 of 15
J. Mar.
J. Mar. Sci.
Sci. Eng.
Eng. 2021,
2021, 9,
9, xx FOR
FOR PEER
PEER REVIEW
REVIEW 55 of
of 15
15

Figure 3.
Figure 3. The
The control
control lines
lines and
and the
the surface
surface of
of the
the bow
bow bulb.
bulb.
Figure 3. The control lines and the surface of the bow bulb.

Table1.1.
Table
Table 1.The
Theinitial
The initialvalues
initial valuesand
values andlimits
and limitsofof
limits ofvariation
variationofof
variation ofdesign
designvariables.
design variables.
variables.
Design
Design
Design Variable
Variable
Variable Lower
Lower
Lower Limit
Limit
Limit Initial
Initial
Initial Value
Value
Value UpperLimit
Upper
Upper Limit
Limit
dX8%_from
dX8%_from
dX8%_from
FP
FPFP
(m)
(m)(m) −1.6 −1.6
−1.6 0.0
0.00.0
1.5
1.5
1.5
FOS_dZ
FOS_dZ (m)
FOS_dZ (m) (m) − 1.2−1.2
−1.2 0.00.0
0.0 1.0
1.0
1.0
Bulb_dL
Bulb_dL (m)(m) −0.8−0.8 0.00.0 1.3
1.3
Bulb_dL (m) −0.8 0.0 1.3
Angle_WL (◦ ) 150 170 180
Angle_WL
Angle_WL (°) 150 170 180
Angle_Prof (◦ )(°) 65 150 81170 180
90
Angle_Prof
TransomLow_zPos
Angle_Prof (°) (°)
(m) 10.756565 11.00381
81 11.390
90
TransomLow_zPos
Curve_xPos (m) (m)
TransomLow_zPos (m) 10.75
7.610.75 11.003
8.9
11.003 11.3
10.2
11.3
TubeEnd_xPos
Curve_xPos (m) (m)
(m) 4.65 7.6 5.113
8.9 5.65
10.2
Curve_xPos 7.6 8.9 10.2
TubeEnd_xPos
TubeEnd_xPos (m) (m) 4.65
4.65 5.113
5.113 5.65
5.65
The design variables of Table 1 are described in the following:
The design
The design variables
variables of of Table
Table 11 are
are described
described in in the
the following:
following:
The longitudinal shift of the frame located 8%
dX8% aft
dX8% aft
dX8% FP:
aft FP:
FP: The longitudinal
The longitudinal shift
shift ofof the
the frame
frame located
located 8% 8% of
of ship
ship length
length LL aft
aft of
of FP.
FP.
of ship length L aft of FP.
FOS_dZ:
FOS_dZ: Vertical variation of the
Vertical variation of the lower lower point of the
pointvariation
Vertical Flat-Of-Side
of the Flat-Of-Side (FOS) area.
(FOS)ofarea.
of the lower point the
FOS_dZ:
Bulb_dL: Change of of bulbous
bulbous bowbow total
totalFlat-Of-Side
length. (FOS) area.
Bulb_dL: Change length.
Angle_WL:
Bulb_dL:
Angle_WL: Angle of
Angle of waterline
waterline atat design
designChange
draft T.
draft T.of bulbous bow total length.
Angle_WL:
Angle_Prof: Angle of rise of bulbous bow Angle of waterline
profile curve. at design draft T.
Angle_Prof: Angle of rise of bulbous bow profile curve.
Angle_Prof:
Transom- Angle of rise of bulbous bow profile curve.
Transom- Vertical position
position of
of transom
transom Vertical
lowest position
point (Figure
(Figure 4). lowest point
of transom
Low_zPos:
TransomLow_zPos: Vertical lowest point 4).
Low_zPos: (Figure 4).
Curve_xPos:
Curve_xPos:
Curve_xPos: Profile stern
Profile stern curve
curve (Figure
(Figure 4).4).
Profile stern curve (Figure 4).
TubeEnd_xPos: Variation of stern tube
TubeEnd_xPos: Variation of stern tube axis length
TubeEnd_xPos: axis length (Figure
Variation of stern
(Figure 4).tube axis length (Figure 4).
4).

Figure 4.
Figure 4. The
The three
three parameters
parameters set
set at
at the
the stern
stern region.
region.
Figure 4. The three parameters set at the stern region.

Therange
The
The rangeof
range ofthe
of thedesign
the design variables
design variables and
variables and their
and their values
their values for
values for the
for the parent
the parent hull
parent hull form
hull form are
form are pre-
are pre-
pre-
sented
sented in
sented in Table
in Table 1. The
Table1.1. The constraints
The constraints
constraints ofof the
of the geometrical
thegeometrical variables
geometrical variables have
variables have been
have been specified
been specified byaaa
by
specified by
trialand
trial
trial anderror
and errormethod
error method to
method toreduce
to reducethe
reduce thenumber
the numberof
number ofnon-realistic
of non-realistic or,
non-realistic or,more
or, moregenerally,
more generally,invalid
generally, invalid
invalid
variant
variant hull forms. However, this does not mean that all the variants generated
variant hull forms. However, this does not mean that all the variants generated duringthe
hull forms. However, this does not mean that all the variants generated during
during the
the
optimization
optimization process
process are
are realistic
realistic hulls.
hulls.
optimization process are realistic hulls.
J. Mar. Sci. Eng. 2021, 9, 1234 6 of 15

3.3. The KRISO Containership

The characteristics and the floating conditions of the containership under investigation
have been proposed by the Korean Research Organization KRISO (former MOERI), and
they are given in Table 2. The service speed of the vessel has been specified to be 24 kn
corresponding to Froude number 0.26.

Table 2. Main particulars and floating conditions of the KRISO containership.

The KRISO Container Ship KCS and Its Tested Model *

Ship Model
Scale - 1:1 1:37.89
LPP (m) 230.0 6.0702
LWL (m) 232.5 6.1357
BWL (m) 32.2 0.8498
TM (m) 10.8 0.2850
WS (m2 ) 9501 6.618
5 (m3 ) 52,062 0.957
CB - 0.651 0.651
LCG from AP (m) 111.6 2.945
VCG from BL (m) 14.324 0.378
* The detailed geometry database of the ship is provided at the site of Tokyo 2015 CFD workshop website [15].

However, this vessel has not been constructed in physical scale. Only models of this
hull form have been tested in towing tank facilities. In the current study, the full-scale
vessel has been modeled.

3.4. Potential Flow Calculation

The 3D panel, Rankine source, time domain, linear code SWAN2 2002 [16,17] has been
used to carry out the potential calculations. The mesh generation of the free-surface and
the body surface of the hull is an internal routine of SWAN2 2002. The spline sheet of the
body surface is defined by 45 nodes in a direction parallel to the x-axis, corresponding to
a number of 44 panels and by 13 nodes on the y-axis’ perpendicular. The domain of the
free-surface accounted in the calculations extends 0.5 LBP upstream, 1.5 LBP downstream,
and 1.0 LBP in the transverse distance (athwartships). Figure 5 presents the spline sheet of
the free-surface (a) and the body (b).

3.5. Viscous Flow Calculation

Regarding the CFD solver, an in-house (U)RANS solver, MaPFlow, was employed.
MaPFlow is a cell centered CFD solver that can use both structured and unstructured
grids, capable of solving compressible flows, as well as fully incompressible flows using
the artificial compressibility method. For the reconstruction of the flow field, a 2nd order
piecewise linear interpolation scheme is used. The limiter of Venkatakrishnan [18] is
utilized when needed. The viscous fluxes are discretized using a central 2nd order scheme.
Turbulence closures implemented on MaPFlow include the one-equation turbulence
model of Spalart (SA) [19] as well as the two-equation turbulence model of Menter
(k-ω SST) [20]. Regarding laminar to turbulent transition modeling, the correlation γ-Reθ
model of Langtry and Menter [21] has been implemented.
MaPFlow can handle both steady and unsteady flows. Time integration is achieved in
an implicit manner permitting large CFL numbers. The unsteady calculations use a 2nd
order time accurate scheme combined with the dual time-stepping technique to facilitate
convergence. MaPFlow is able to handle moving/deforming geometries through the
arbitrary Eulerian Lagrangian formulation.
 been used to carry out the potential calculations. The mesh generation of the free-surface
and the body surface of the hull is an internal routine of SWAN2 2002. The spline sheet of
the body surface is defined by 45 nodes in a direction parallel to the -axis, corresponding
to a number of 44 panels and by 13 nodes on the -axis’ perpendicular. The domain of the
J. Mar. Sci. Eng. 2021, 9, 1234 free-surface accounted in the calculations extends 0.5 LBP upstream, 1.5 LBP downstream,
7 of 15
and 1.0 LBP in the transverse distance (athwartships). Figure 5 presents the spline sheet of
the free-surface (a) and the body (b).
J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 7 of 15

Figure 5. Spline sheet on the computational greed (a) free surface; (b) body surface.

3.5. Viscous Flow Calculation

Regarding the CFD solver, an in-house (U)RANS solver, MaPFlow, was employed.
MaPFlow is a cell centered CFD solver that can use both structured and unstructured
grids, capable of solving compressible flows, as well as fully incompressible flows using
the artificial compressibility method. (a) For the reconstruction of the flow field, a 2nd order
piecewise linear interpolation scheme is used. The limiter of Venkatakrishnan [18] is uti-
lized when needed. The viscous fluxes are discretized using a central 2nd order scheme.
Turbulence closures implemented on MaPFlow include the one-equation turbulence
model of Spalart (SA) [19] as well as the two-equation turbulence model of Menter (k-ω
SST) [20]. Regarding laminar to turbulent transition modeling, the correlation γ-Reθ
model of Langtry and Menter [21] has been implemented.
MaPFlow can handle both steady and unsteady flows. Time integration is achieved
in an implicit manner permitting large CFL numbers. The unsteady calculations use a 2nd
order time accurate scheme combined with the dual time-stepping technique to facilitate
convergence. MaPFlow is able to handle (b) moving/deforming geometries through the arbi-
trary Eulerian
Figure 5. Spline Lagrangian formulation. greed (a) free surface; (b) body surface.
sheet on the computational
Regarding the free surface treatment, the Volume of Fluid (VOF) method is em-
Regarding
ployed, and twothe free surface
phase flows are treatment,
describedthe
byVolume of Fluid (VOF)
two immiscible fluidsmethod is employed,
with their interface
and
beingtwo phaseimplicitly
defined flows areasdescribed by two
a discontinuity inimmiscible
the density fluids withsystem
field. The their interface
of equationsbeing
is
defined
solved in implicitly as a discontinuity
a non-segregated manner,inutilizing
the density
the field.
Kunz The system of equations
Preconditioner, is solved
as discussed by
in
Yuea non-segregated manner,density
and Wu [22], to remove utilizing the Kunz Preconditioner,
dependencies from the system’s as discussed by Yue and
eigenvalues.
Wu [22], to remove density dependencies from the system’s eigenvalues.
For the CFD simulations, in order to reduce the computational domain, half of the
For
hull is the CFD
resolved simulations,
with symmetryinconditions
order to reduce
applied theoncomputational domain,
the side. The high half ofnumber
Reynolds the hull
is
of resolved
full scalewith symmetry
simulations conditions
poses applied
a significant on the side.
challenge for CFDThesimulations
high Reynolds number
since of
fully re-
full scalesimulations
solved simulations areposes a significant challenge
computationally for CFD
prohibitive. simulations
Following since fully resolved
a grid-independence
simulations
study, a gridare computationally
consisting prohibitive.
of approximately Following
5 million cellsaisgrid-independence
employed. In the wall study, a grid
region,
consisting of approximately 5 million cells is employed. In the wall region,
wall functions are employed; nevertheless, a structured-like region composed of 25 layers wall functions
are employed;
is used aroundnevertheless, a structured-like
the solid boundary. region
Lastly, the hullcomposed
was resolvedof 25 using
layersapproximately
is used around
the solid boundary. Lastly, the hull was resolved using approximately
200,000 elements. A snapshot of the computational grid employed can be seen 200,000 elements.
in FigureA
snapshot
6. of the computational grid employed can be seen in Figure 6.

Figure
Figure6.6.Top
Topview
viewofofthe
thecomputational
computationalgrid used
grid inin
used thethe
free surface
free region.
surface TheThe
region. gridgrid
is refined
is refined to
to accurately capture the resulting wave system.
accurately capture the resulting wave system.

The span of the domain was 5 LBP BP in

inthe
thestreamwise
streamwise direction,
direction, 33 LLBP
BPin
inthe
theside
sidedirec-
direc-
BP in
tion, and 6 LBP in the
the vertical
verticaldirection.
direction.OnOnthe
theship
shiphull,
hull,aano-slip
no-slipcondition
conditionwas wasapplied,
applied,
symmetry (zero
symmetry (zerogradient
gradientininthethenormal
normal direction)
direction) conditions
conditions were
were applied
applied on symmetry
on the the sym-
metry while
plane, plane,awhile a freestream
freestream conditioncondition was imposed
was imposed on the
on the rest rest
of the of the domain.
domain. Addi-a
Additionally,
tionally, azone
damping damping zone was
was adopted toadopted to avoid reflections
avoid reflections from the wave
from the generated generated wave sys-
system.
tem.
J. Mar. Sci. Eng. 2021, 9, 1234 8 of 15
J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 8 of 15

Regarding
Regardingthe they+y+ values,
values, the
the average
average y+y+ was 150, while a maximum of 300, and due
to
tothat,
that,wall
wallfunctions
functions werewere adopted.
adopted. Unfortunately,
Unfortunately, this
this was
was aa mandatory
mandatory compromise
cost-wise in order to make full scale simulations
cost-wise in order to make full scale simulations feasible. feasible.
For
Forall
allthe
the CFD
CFD simulations,
simulations, aa time
time step
step of 0.1 s using a second-order implicit scheme
isis used
used which yields a convective CFL around
which yields a convective CFL around 3.3. Nevertheless, it was adopted to save
computational
computationaltime timesince
sincethe
theflows
flowsconsidered
consideredhere
hereconverge
convergetotoaa steady
steady state.
state. It is evident
It is evident
from
fromFigure
Figure 77that
that both
both the
the time
time step
step and
and the
the grid
grid spacing
spacing selected
selected are
are tuned
tuned inin order to
order to
properly
properlycapture
capturethetheresulting
resultingwave
wavesystem.
system.

Figure7.
Figure 7. Contour
Contour of
of the
the resulting
resulting wave
wave system
system for
for the
the original
original KCS
KCS hull
hull case—Contour
case—Contour of
of Elevation
Elevation
of the water level. Half of the model was resolved to save computational resources.
of the water level. Half of the model was resolved to save computational resources.

3.6.
3.6. Artificial
Artificial Neural
Neural Network
Network (ANN)
(ANN)
During
Duringthe the last
last decade,
decade, the the application
application of Artificial Neural Networks has stepped up
in
inevery
everyscientific
scientificfield.
field. From
From plain
plain vanilla
vanilla networks
networks to to unsupervised deep convolutional
networks,
networks,ANN ANNisisable able to to model
model or detect complex nonlinear relationships within systems
without
withoutusing usingthe thephysics
physicsof of the
the system.
system. Furthermore, they are a valuable tool, since they
can
can bridge fragmented data to efficiently identify
bridge fragmented data to efficiently identify system
system characteristics or make up for a
lack of analytical relations within
lack of analytical relations within complex systems.complex systems.
Artificial
Artificialneural
neuralnetwork
networktheory theoryis isbased
based onon thethe
analysis
analysis of biological
of biologicalnervous systems
nervous sys-
consisting
tems consistingof neurons and their
of neurons andconnections.
their connections.A mathematical
A mathematical modelmodel
of a neural network
of a neural net-is
created,
work is based
created, onbased
this structure and signal
on this structure andtransmission. ANNs are
signal transmission. ANNscomposed of internal
are composed of
parameters to be specified
internal parameters to be through
specifiedthe processthe
through of training.
process of Such parameters
training. are the weights
Such parameters are
by which the inputs of each neuron are multiplied so that
the weights by which the inputs of each neuron are multiplied so that the corresponding the corresponding output
emerges. Explicitly,
output emerges. the output
Explicitly, the of the neuron
output is calculated
of the neuron by the sum
is calculated by theof sum
all the
of inputs,
all the
weighted
inputs, weighted by the weights of the connections from the inputs to the neuron. Addi-a
by the weights of the connections from the inputs to the neuron. Additionally,
bias term aisbias
tionally, introduced
term is to this sum. to
introduced This
thisweighted
sum. This sum is often sum
weighted called is the activation,
often called thewhich
acti-
is, then,which
vation, passedis,through
then, passeda (usually
through nonlinear)
a (usually activation
nonlinear) function
activation to produce
function the output.
to produce
Ultimately,
the output.the output ofthe
Ultimately, theoutput
last neuron, subsequently
of the last the output of
neuron, subsequently thethe model
output of in
thegeneral,
model
isincompared against actual values and the difference between
general, is compared against actual values and the difference between predicted predicted and real values andof
the same parameter is estimated through a metric function.
real values of the same parameter is estimated through a metric function. This part of theThis part of the optimization
algorithm
optimization is critical.
algorithm Theistraining
critical.aims
The at the minimization
training of the mean difference,
aims at the minimization of the mean or loss
dif-
as it is called in the ANN field, by updating the weights in each
ference, or loss as it is called in the ANN field, by updating the weights in each iteration. iteration.
In
Inthe
the field
field ofof naval
naval architecture,
architecture, artificial
artificial neural
neural networks
networks (ANN)(ANN) have have gained
gained pop-
pop-
ularity.
ularity. In In recent
recent years,
years, applications
applications of of ANNs
ANNs for for modelling
modelling and and predicting
predicting vessel
vessel hull
hull
form
form [23], calm water resistance [24,25], added resistance in waves [26], speed and
[23], calm water resistance [24,25], added resistance in waves [26], speed and fuel
fuel
consumption
consumption [27], [27], maneuverability
maneuverability qualities qualities [28],
[28], and
and seakeeping
seakeeping characteristics
characteristics [29] [29] are
are
presented in several studies. In most cases, the use of artificial
presented in several studies. In most cases, the use of artificial neural networks offers sat- neural networks offers
satisfactory results.
isfactory results.
Compared
Comparedtotothe theabove-mentioned
above-mentionedpublications,publications,this thispaper’s
paper’s distinctive
distinctive feature
featurelieslies
in
the relatively low number of input data available for the training
in the relatively low number of input data available for the training of an ANN. As dis- of an ANN. As discussed
in the Introduction,
cussed the ANNthe
in the Introduction, is trained
ANN is with only with
trained 27 examples
only 27as input, which
examples correspond
as input, which
to the 27 (=33) combinations of three values per variable, the two selected limiting ones, and
correspond to the 27 (=33) combinations of three values per variable, the two selected lim-
the value of the parent hull for the three chosen stern geometrical variables affecting the
iting ones, and the value of the parent hull for the three chosen stern geometrical variables
optimization scheme. Usually, the stern design variables are two to four and three values
affecting the optimization scheme. Usually, the stern design variables are two to four and
per variable are sufficient to train a reliable ANN, taking into account that the variation of
J. Mar. Sci. Eng. 2021, 9, 1234 9 of 15

the variables is limited. The use of limited CFD calculations is a major advantage of the
proposed methodology. The derived ANN is handled by a GA, which, after 440 evaluations,
reaches the optimum combination of the stern variables.
Selection of a suitable artificial neural network structure is probably the hardest part of
the problem and critical to obtaining accurate predictions. Since dealing with a regression
problem, the multilayer perceptron (MLP) concept was applied consisting of an input,
hidden, and output layer, as well as utilizing a backpropagation learning algorithm. The
development of the ANN was performed in Python assisted by the Tensorflow/Keras
neural network library.
The usual search process for the optimal neural network goes through the following
steps: data normalization, division of data set, selection of ANN model architecture,
and finally the assessment of ANN model results. In this study, the small number of
available training data significantly hindered this process. Input data were normalized
using a custom Min/Max scalar function centered around parent hull resistance value
with a 20% reserve. This reserve was used in order to ease the first step of optimization
process, ANN’s function minimization search, by allowing us to extrapolate values beyond
observed ranges. Moreover, during ANN model’s training, no validation set was used. It
was decided to use every data point available for the more efficient training of the network
and take the risk to validate the model’s prediction at the final stage of optimization
procedure via CFD calculations:
    
R parent_hull − Rvariant + R parent_hull − Rmin × 1.2
Norm. Function := (1)
( Rmax − Rmin ) × 1.2

In order to identify the ANN architecture that is better suited to the problem, many
trials were conducted with different configurations. The number of input neurons was
set to three, representing the three variables set at the vessel stern region, and the output
node was set to one referring to the target value of CFD calculation. The rest of the
ANN configuration as determined by the number of hidden layers, the number and type
of neurons that comprise each one of them, the training algorithm, learning rate, and
the backpropagation optimizer method went through exhaustive numerical experiments,
probably an inevitable stage when developing an ANN. An overview of the performance
of the best ANN models is presented in Table 3. The number of neurons at each layer and
their activation functions can also be seen.

Table 3. Overview of the best ANN performance.

Training Set
Code
Input Layer Hidden Layer (s) * Output Layer MSE MAE
N-1 3 6(S)–4(S) 1(S) 2.1 × 10−3 3.3 × 10−2
N-2 3 6(S) 1(S) 4.3 × 10−2 1.7 × 10−1
N-3 3 6(S)–3(R) 1(S) 4.4 × 10−2 1.7 × 10−1
N-4 3 6(S)–3(R)–3(R) 1(S) 4.0 × 10−2 1.6 × 10−1
N-5 3 12(S) 1(S) 4.6 × 10−2 1.9 × 10−1
* Activation Functions: R—ReLU (Rectified Linear Unit), S—Sigmoid.

In this work, the Mean Squared Error (MSE) function was used as the loss or cost
function under minimization during models’ training. Notice should be kept, though, on
the Mean Absolute Error (MAE) as well. The progression of MSE and MAE values during
training is presented in Figure 8.
 J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 10 of 15
J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 10 of 15

J. Mar. Sci. Eng. 2021, 9, 1234 10 of 15

J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 10 of 15

Figure 8. MSE and MAE during networks’ training.

Figure 8. MSE and MAE during networks’ training.
The network (N-1) that showed the lowest MSE and MAE error was selected in order
to The network
evaluate (N-1) that showed the lowest
of theMSE and MAE error wasas selected
inputs,intwo order
Figure 8. MSEMSEhull
andmeta-models.
and MAE during
MAE
It consists
during networks’ training. three stern parameters hid-
toden
evaluate
layershull meta-models.
comprised of six Itandconsists of the three
four nodes, stern parameters
respectively, as inputs,
and the output twoIn
layer. hid-
our
den layers
effort The comprised
to network
overcome
network (N-1) of six
the that
that and
problem
showed four
of the nodes,
vanishing
the lowest respectively,
lowest MSEgradients
MSE andand MAE and the
and saddle
MAE output
error was points,
was layer.
the in
selected In our
sigmoid
in order
The (N-1) showed error selected order
effort
to to overcome
function
evaluate was
hullused the
as problem
an
meta-models. activation
It of vanishing
function
consists of theatgradients
input,
three stern and
first and saddle
output
parameters points,
aslayer of
inputs,the
the
twosigmoid
network,
hidden
to evaluate hull meta-models. It consists of the three stern parameters as inputs, two hid-
function
while was was
tanh usedutilized
as
of an activation
atsix
the function
second at input, first and output layer with
of the network,
layers
den comprised
layers comprised sixofand four
and four layer.
nodes, nodes, Stochastic
respectively, andgradient
respectively, theand descent
output
the layer.
outputInlayer.momentum
our effort
In ourto
while
was
overcometanh was
selected
the utilized
against
problem at vanishing
ADAM
of the assecond
the layer. Stochastic
backpropagation
gradients and gradient
optimizer
saddle points,descent
method,
the with momentum
learning
sigmoid rate,
function and
was
effort to overcome the problem of vanishing gradients and saddle points, the sigmoid
was selected
momentum against
were ADAM asand
the backpropagation optimizer method, learning rate, and
used
function as an
was used set
activation to
as an 0.16
function
activationat 0.7, respectively.
input,
functionfirstatand Selected
output
input, ANN
first layer
and model’s
of the
output architecture
network,
layer of thewhile and
tanh
network,
momentum
was utilized were
atareset
the to
second0.16 and 0.7, respectively. Selected ANN model’s architecture and
performance
while tanh was presented
utilized atlayer.
the Stochastic
at second
Figures 9 and
layer. gradient
10. Thedescent
Stochastic Pearson with
gradient momentum
coefficient
descent was was
with selected
calculated
momentum at
performance
against
0.975. ADAM are as
presented
the at Figures 9 and
backpropagation 10. Themethod,
optimizer Pearsonlearning
coefficient
rate,was
andcalculated
momentum at
was selected against ADAM as the backpropagation optimizer method, learning rate, and
0.975.
were set to 0.16 and 0.7, respectively. Selected ANN model’s architecture and performance
momentum were set to 0.16 and 0.7, respectively. Selected ANN model’s architecture and
are presented at Figures 9 and 10. The Pearson coefficient was calculated at 0.975.
performance are presented at Figures 9 and 10. The Pearson coefficient was calculated at
0.975.

Figure 9. Architecture
Figure 9. Architecture of
of the
theselected
selectedANN
ANNmodel.
model.
Figure 9. Architecture of the selected ANN model.

Figure 9. Architecture of the selected ANN model.

Figure 10.
Figure 10. ANN
ANN input
input vs.
vs. predicted
predicted values.
values.
Figure 10. ANN input vs. predicted values.
The small
The small size
size of
of the
the training
training data
data led
led to
to an
an increase
increase of
of epochs
epochsperformed
performedininorder
orderto to
get The small size
satisfactory
get satisfactory of the
results.
results. training
For the data
best
For the best led to an
configuration increase of
found, epochs
the performed
number of in
epochs order
was
configuration found, the number of epochs was set at to
set at
Figure 10. ANN input vs. predicted values.
get satisfactory results. For the best configuration found, the number of epochs
the scale of 150,000. Despite the large number of epochs, the training procedure requiredwas set at
less than 5 minsize
The small to complete at an i7data
of the training 9700K.
led to an increase of epochs performed in order to
get satisfactory results. For the best configuration found, the number of epochs was set at
J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 11 of 15

J. Mar. Sci. Eng. 2021, 9, 1234 11 of 15

the scale of 150,000. Despite the large number of epochs, the training procedure required
less than 5 min to complete at an i7 9700K.

4.
4. Results
Results
The results obtained
The results obtained from
from the
the implementation
implementation of of the
the above-described
above-described methodology
methodology
are presented in this chapter.
are presented in this chapter.
In
In the
the below
below figures,
figures, the
the outcome
outcome of of the
the first
first step
step ofof optimization
optimization procedure
procedure is is pre-
pre-
sented.
sented. After
After the
the training
training ofof the ANN model,
the ANN model, an an NSGA-II
NSGA-II optimization
optimization algorithm
algorithm was was
executed
executed for forthe
theminimization
minimization of of
thethe
ANNANN function
function modeling
modeling vesselvessel total viscous
total viscous re-
resistance
sistance
R TOTAL . R
In .
Figure
TOTAL In Figure
11a–c, 11a–c,
the the convergence
convergence of the of the
three three
input input
parametersparameters represent-
representing the
ing the variables
variables at the hullat the hull
stern stern is observed.
is observed. Moreover, Moreover,
in Figure in Figure
11d, 11d, it isthat
it is evident evident that
the ANN
the ANN function was minimized. This minimization translates to a
function was minimized. This minimization translates to a 9% reduction at vessel’s RTOTAL9% reduction at ves-
sel’s RTOTALto
compared compared
the parent tohull
the parent hull asby
as evaluated evaluated
the ANNbymodel. the ANN model.

(a) (b)

(c) (d)
Figure 11.
Figure 11. (a–d)
(a–d) Convergence of the
Convergence of the three
three stern
stern parameters
parameters (a–c)
(a–c) and
and the
the minimization
minimization of
of ANN’s
ANN’s function
function (d).
(d).

To verify this result, a direct CFD evaluation

evaluation of the hull form with the optimum stern
variables was performed. Dynamic
Dynamic sinkage
sinkage and
and trim as calculated by the potential flow
solver were used in the viscous evaluation.
As presented
presentedininTable
Table4, 4, geometrical
geometrical parameters,
parameters, assessed
assessed by ANN
by the the ANN and opti-
and optimized
mized
via via NSGA-II
NSGA-II at thestern
at the vessel vessel sterncontributed
region, region, contributed to the reduction
to the reduction of RTOTAL byof 15%
RTOTAL by
after
validation by MaPFlow,
15% after validation using the same
by MaPFlow, usingdiscretization scheme and
the same discretization turbulence
scheme model for
and turbulence
the evaluation
model of the parent
for the evaluation andparent
of the the optimum
and thehull form. However,
optimum hull form.as stated in as
However, Section
stated3.5,
in
the y+ of3.5,
Section thethe
simulations was relatively
y+ of the simulations waslarge; thus, itlarge;
relatively mustthus,
be validated using
it must be successively
validated using
denser grids denser
successively in the hull region.
grids in theApart from that,
hull region. results
Apart fromobtained by theobtained
that, results trained neural
by the
networkneural
trained are quite satisfactory
network are quitedespite the small
satisfactory sizethe
despite ofsmall
the training dataset.
size of the The
training small
dataset.
discrepancy in the RTOTAL reduction between network prediction and high-fidelity viscous
evaluation verifies the reliability of the ANN meta-model.
J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 12 of 15
J. Mar. Sci. Eng. 2021, 9, x FOR PEER REVIEW 12 of 15

The small discrepancy in the RTOTAL reduction between network prediction and high-fidel-
The small discrepancy in the RTOTAL reduction between network prediction and high-fidel-
J. Mar. Sci. Eng. 2021, 9, 1234 ity viscous evaluation verifies the reliability of the ANN meta-model. 12 of 15
ity viscous evaluation verifies the reliability of the ANN meta-model.
Table 4. Results of the first step of optimization.
Table 4. Results of the first step of optimization.
TransomLow_zPos
Table 4. Results Curve_xPos
of the first step of optimization. TubeEnd_xPos RTOTAL
Model TransomLow_zPos Curve_xPos TubeEnd_xPos RTOTAL
Model (m) (m) (m) (kN)
Model TransomLow_zPos (m) (m)
Curve_xPos (m) (m)
TubeEnd_xPos (m)(m) (kN)
RTOTAL (kN)
Parent Hull 11.003 8.900 5.113 1780
Parent Hull Parent
11.003 Hull 11.003
8.900 8.900 5.113 5.113 17801780
1611(ANN)/1510(
Opt_Hull_Step-1 Opt_Hull_Step-1
10.750 10.750
9.539 9.539 5.651 5.651 1611(ANN)/1510(CFD)
1611(ANN)/1510(
Opt_Hull_Step-1 10.750 9.539 5.651 CFD)
CFD)
In
In the
the 2nd
2nd step
step of
of the
the optimization
optimization procedure,
procedure, thethe five
five design
design variables
variables at
at the
the bow
bow
and In the
the 2nd region
middle step of ofthethe
optimization
vessel went procedure,
through the fiveNSGA-II
another design variables at thecycle.
optimization bow
and the middle region of the vessel went through another NSGA-II optimization cycle.
and the12middle
Figures and 13.region
This of thehull
time, vessel wentwere
variants through another
evaluated via NSGA-II
the optimization
potential solver cycle.
SWAN2
Figures 12 and 13. This time, hull variants were evaluated via the potential solver SWAN2
Figures 12 and
as 13. This time, Ithull variantsnoted
were evaluated via the potential solver SWAN2
as described
described inin Section
Section 3.4.
3.4. It should
should be
be noted that
that the
the vessel
vessel stern
stern region
region was
was kept
kept at
at
as described
parent shape in Section
while only 3.4.
the It should
five be
parametersnoted that
were the
varied. vessel stern
Reduction region
of wavewas kept at
pressure
parent shape while only the five parameters were varied. Reduction of wave pressure
parent shape been
resistance while only thesmall. five parameters were varied. ReductionTable
of wave pressure
resistance has
has been relatively
relatively small. Results
Results are
are presented
presentedat atfollowing
following Table5.5.
resistance has been relatively small. Results are presented at following Table 5.

Figure 12.
12. Convergence of
of the 2nd
2nd step of
of optimization.
Figure 12.Convergence
Figure Convergence ofthe
the 2ndstep
step ofoptimization.
optimization.

Figure 13. Comparison of wave elevation distribution by SWAN2.

Figure13.
Figure 13.Comparison
Comparisonof
ofwave
waveelevation
elevationdistribution
distributionby
bySWAN2.
SWAN2.
Table 5. Results of the second step of optimization.
Table5.5.Results
Table Resultsofofthe
thesecond
secondstep
stepofofoptimization.
optimization.
dX8%fromFP FOS_dZ Bulb_dL Angle_WL Angle_Prof RW
Model dX8%fromFP
dX8%fromFP FOS_dZ Bulb_dL Angle_WL Angle_Prof RW
Model
Model (m) FOS_dZ (m)(m) Bulb_dL (m)
(m) Angle_WL(°)
(◦ ) Angle_Prof ((°)
◦) (kN)
Rw (kN)
(m) (m) (m) (m) (°) (°) (kN)
Parent Hull 0.0 0.0 0.0 170 81 703
Parent
Parent HullHull 0.0
0.0 −0.054 0.0
0.0 0.986 0.0
0.0 −0.684 170
170171.493 81
81 89.877 703
703 676
Opt_Hull_Step-2
Opt_Hull_Step-2
Opt_Hull_Step-2 −0.054 −0.054 0.986 0.986 −0.684
−0.684 171.493
171.493 89.877 89.877 676 676

As a final step, the optimum design variables as those that emerged from both op-
timization steps were combined to derive the overall optimized hull form, presented
in Figure 14. Its performance was assessed via viscous calculation. Results were quite
encouraging and are presented in Table 6.
 As a final step, the optimum design variables as those that emerged from both opti-
J. Mar. Sci. Eng. 2021, 9, 1234
mization steps were combined to derive the overall optimized hull form, presented in
13 of 15
Figure 14. Its performance was assessed via viscous calculation. Results were quite en-
couraging and are presented in Table 6

Figure
Figure 14.14. Comparison
Comparison of of optimized
optimized (red)
(red) andand original
original hull
hull stations.
stations.

Table 6. 6.
Table Comparison of of
Comparison thethe
parent and
parent final
and optimized
final hull.
optimized hull.

Quantity
Quantity Parent Hull
Parent Hull Final
FinalHull
Hull
R
RTOTAL 1780
1780 1460
1460
TOTAL
Cb
Cb 0.6507
0.6507 0.6500
0.6500
Cp
Cp 0.9029
0.9029 0.9019
0.9019
Cm
Cm 0.7206
0.7206 0.7206
0.7206
LCB 111.92 111.84
LCB
WS (at speed)
111.92
9764
111.84
9739
WS (at speed) 9764 9739
5. Conclusions
5. Conclusions
On the basis of the results presented in the previous sections, it seems that the pro-
On the basis of the results presented in the previous sections, it seems that the pro-
posed mixed fidelity method is quite efficient in optimizing the hull form for calm water
posed mixed fidelity method is quite efficient in optimizing the hull form for calm water
resistance of ships using genetic algorithms and ANN. The method combines the capabil-
resistance of ships
ities of the using
potential flowgenetic
codes algorithms
to evaluate and ANN.
the The method
hydrodynamic combines
effects of thethe capabil-
geometrical
ities of the
design potential flow
parameters of thecodes
fore andto evaluate
middle the bodyhydrodynamic
of a ship witheffects of the of
the strength geometrical
the viscous
design
flow codes to estimate the respective effect of the design variables in the aftthe
parameters of the fore and middle body of a ship with the strength of viscous
body. ANN
flow codes to estimate the respective effect of the design variables in
was able to produce results of sufficient accuracy to be useful for the preliminary predictionthe aft body. ANN
wasof able to resistance
vessel produce resultsdespite of asufficient accuracy
small number of to be useful
training for the
inputs. It preliminary predic-
is worth underlining
tion of vessel resistance despite a small number of training inputs.
that the development of a case specific neural network model is needed. Besides tradi- It is worth underlining
that the programming
tional development of a case specific
languages, though,neural
onlinenetwork
platforms model is needed.
and libraries for Besides
developmenttradi-of
tional programming languages, though, online platforms and
artificial neural networks are widely available nowadays. Depending on the number oflibraries for development
of inputs/outputs,
artificial neural networks
the magnitude are widely available
of their variances nowadays.
and size of Depending
training data on the
an number
ANN canofbe
inputs/outputs, the magnitude of their
effortlessly integrated at the optimization process variances and size of training data an ANN can be
effortlessly integrated at the optimization process
Employing automated artificial neural network software can greatly accelerate the
Employing
process automated
of designing artificialan
and training neural
ANNnetwork
capable of software
modeling canhull
greatly accelerate
resistance the
problems.
process of designing
The agreement and training
between the optimuman ANN capable
value of theof modeling hull resistance
objective function and the problems.
results of
Thetheagreement
RANS solver between the optimum
is satisfactory. value
This of the objective
is considerably function
easier and thethan
and quicker results of the
traditional
RANS solver is satisfactory. This is considerably easier and quicker
statistical methods. Whilst it is important to choose a reasonable artificial neural networkthan traditional sta-
tistical methods. Whilst it is important to choose a reasonable artificial
architecture, the exact number of neurons in the hidden layer is not too critical. Quality neural network
architecture,
and quantity theofexact
inputnumber
data areofthe neurons in the hidden layer is not too critical. Quality
key factors.
and quantity
The finalof input
outcomedataofare thethe key factors. procedure is validated by evaluating the op-
optimization
The final
timum hull outcome
form via of thetheRANSoptimization
solver. As procedure
mentioned is validated by evaluating
above, dynamic sinkagethe andopti-
trim
mumwerehull formfor
utilized viathe
theviscous
RANSevaluation.
solver. As Formentioned above, evaluation,
a more precise dynamic sinkagethough,and the trim
actual
hullutilized
were wetted forsurface should evaluation.
the viscous have been takenFor ainto
more account.
precise Full scale simulations
evaluation, though, the and subse-
actual
quent large Reynolds number still pose a significant challenge for numerical experiments
since fully resolved simulations are computationally prohibitive, especially if combined
into an optimization scheme. Therefore, the urge to enhance the fidelity of optimization
schemes with accurate and efficient methods for metamodels to integrate viscous effects
into potential flows results with different tools such as ANNs is a challenging task.
J. Mar. Sci. Eng. 2021, 9, 1234 14 of 15

It should be noted that the potential of the present approach can be further explored
by increasing the number of training data to improve the fidelity of the ANN. Similar
ANNs could be used to account for other aspects of ship performance, such as seakeeping
and maneuverability characteristics, which, however, are currently estimated satisfactorily
by fast codes using strip theory or 3D potential flow methods. Finally, to ensure the
significance of the optimization simulations, results shall be validated by towing tank
experiments. The construction of scaled models of the parent and the optimum hull forms
and their testing in the towing tank would provide more confidence about the efficiency of
the optimization scheme.

Author Contributions: Conceptualization, G.J.G.; Supervision, G.J.G.; Methodology, G.J.G., C.B.; Vis-
cous flow calculations, G.P., D.N.; Potential flow calculations, C.B.; ANN development, C.B.; Writing
G.J.G., G.P. and C.B. All authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: The data of KCS containership presented in this study are openly
available at “Tokyo 2015 A Workshop on CFD in Ship Hydrodynamics”: https://t2015.nmri.go.jp/
(accessed on 2 September 2021).
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Grigoropoulos, G.J. Hull Form optimization for hydrodynamic performance. Mar. Technol. 2004, 41, 167–182.
2. Grigoropoulos, G.J.; Chalkias, D. Hull form Optimization in Calm and Rough Water. Comput.-Aided Des. J. 2010, 42, 977–984.
[CrossRef]
3. Volpi, S.; Gaul, N.; Diez, M.; Song, H.; Iema, U.; Campana, E.; Choi, K.; Stern, F. Development and validation of a dynamic
metamodel based on stochastic radial basis functions and uncertainty quantification. Struct. Multidiscip. Optim. 2015, 51, 347–368.
[CrossRef]
4. Sclavounos, P.; Yu, M. Artificial Intelligence machine Learning in marine Hydrodynamics. In Proceedings of the International
Conference on Ocean, Offshore and Arctic Engineering, Madrid, Spain, 17–22 June 2018.
5. Chen, X.; Diez, M.; Kandashamy, M.; Zhang, Z.; Campana, E.; Stern, F. High-fielity global optimization of shape design by
dimensionality reduction, metamodels and deterministic particle swarm. In Engineering Optimization; Taylor & Francis: Abingdon,
UK, 2014; pp. 473–494.
6. Harries, S.; Abt, C. CAESES—The HOLISHIP platform for process integration and design optimization. In A Holistic Approach to
Ship Design; Springer: Berlin/Heidelberg, Germany, 2019; pp. 276–291.
7. Papanikolaou, A.; Flikkema, M.; Harries, S.; Marzi, J.; Le Nena, R.; Torben, S.; Yrjanainen, A. Tools and applications for the
holistic ship design. In Proceedings of the 8th Transport Research Arena, Helsinki, Finalnd, 27–30 April 2020.
8. Giannakoglou, K. Design of optimal aerodynamic shapes using stochastic optimization methods and computational intelligence.
Prog. Aerosp. Sci. 2002, 38, 43–76. [CrossRef]
9. Papadakis, G. Development of a Hybrid Compressible Vortex Particle Method and Application to External Problems Including
Helicopter Flows, NTUA. 2016. Available online: https://dspace.lib.ntua.gr/xmlui/handle/123456789/40024?locale-attribute=en
(accessed on 22 July 2021).
10. Papadakis, G.; Filippas, E.; Ntouras, D.; Belebassakis, K. Effects of viscocity and non-linearity on 3D flapping foil thruster for
marine applications. In Proceedings of the OCEANS 2019, Marseille, France, 17–20 June 2019.
11. Grigoropoulos, G.J.; Perdikari, T.; Asouti, V.; Giannakoglou, K. MDO of Hull Forms Using Low-Cost Evolutionary Algorithms.
In Proceedings of the NATO-RTO, Advanced Vehicle Technology Panel AVT-173, Sofia, Bulgaria, 19 May 2011.
12. Deb, K.; Pratap, A.; Agarwal, S.; Meyarivan, T. A Fast and Elitist Multiobjective Genetic Algorithm: NSGA-II. IEEE Trans. Evol.
Comput. 2002, 6, 182–198. [CrossRef]
13. CAESES. CAESES Software Manual; Friendship Systems GmBH: Potsdam, Germany, 2020.
14. Harries, S.; Abt, C. Parametric Curve Design applying fairness criteria. In Proceedings of the International Workshop on Creating
Fair and Shape-Preserving Curves and Surfaces, Potsdam/Berlin, Germany, 14–17 September 1997.
15. Tokyo. A Workshop on CFD in Ship Hydrodynamics. 2015. Available online: https://t2015.nmri.go.jp/ (accessed on
2 September 2021).
16. Sklavounos, P. Computation of Wave Ship Interactions. Adv. Mar. Hydrodyn. Comput. Mech. 1995, 2618. Available online:
http://resolver.tudelft.nl/uuid:da01c58e-1285-414b-83da-e31b812b11dc (accessed on 25 October 2021).
J. Mar. Sci. Eng. 2021, 9, 1234 15 of 15

17. SWAN2. User Manual: Ship Flow Simulation in Calm Water and in Waves; Boston Marine Consulting Inc.: Boston, MA, USA, 2002.
18. Venkatakrishnan, V. On the accuracy of limiters and convergence to steady state solutions. In Proceedings of the 31st Aerospace
Sciences Meeting, Reno, NV, USA, 14–17 June 1993.
19. Spalart, P.; Allmaras, S.; Reno, J. A One-Equatlon Turbulence Model for Aerodynamic Flows Boeing Commercial Airplane Group. In
Proceedings of the 30th Aerospace Sciences Meeting and Exhibit, Reno, NV, USA, 6–9 January 1992; AIAA: Reston, VA, USA, 1992.
20. Menter, F. Two-equation eddy-viscosity turbulence models for engineering applications. AIAA J. 1994, 32, 1598–1605. [CrossRef]
21. Langtry, R.; Menter, F. Correlation-Based Transition Modeling for Unstructured Parallelized Computational Fluid Dynamics
Codes. AIAA J. 2009, 47, 2894–2906. [CrossRef]
22. Yue, D.; Wu, S. An improvement to the Kunz preconditioner and numerical investigation of hydrofoil interactions in tandem.
Int. J. Comput. Fluid Dyn. 2018, 32, 167–185. [CrossRef]
23. Taniguchi, T.; Ichinose, Y. Hull form design support tool based on machine learning. In Proceedings of the 19th Conference on
Computer and IT Applications in the Maritime Industry (COMPIT), Pontignano, Italy, 17–19 August 2020.
24. Grabowska, K.; Szczuko, P. Ship resistance prediction with Artificial Neural Networks. In Proceedings of the Signal Processing:
Algorithms, Architectures, Arrangements, and Applications (SPA), Poznan, Poland, 23–25 September 2015 .
25. Margari, V.; Kanellopoulou, A.; Zaraphonitis, G. On the use of Artificial Neural Networks for the calm water resistance prediction
of MARAD Systematic Series’ hullforms. Ocean. Eng. 2018, 165, 528–537. [CrossRef]
26. Cepowski, T. The prediction of ship added resistance at the preliminary design stage by the use of an artificial neural network.
Ocean. Eng. 2020, 195, 106657. [CrossRef]
27. Tarelko, W.; Rudzki, K. Applying artificial neural networks for modelling ship speed and fuel consumption. Neural Comput. Appl.
2020, 32, 17379–17395. [CrossRef]
28. Abramowski, T. Application of artificial neural networks to assessment of ship manoeuvrability qualities. Pol. Marit. Res. 2008,
15, 15–21. [CrossRef]
29. Martins, P.; Lobo, V. Estimating Maneuvering and Seakeeping Characteristics with Neural Networks. In Proceedings of the
OCEANS 2007—Europe, Aberdeen, UK, 29 September–4 October 2007.