Professional Documents
Culture Documents
2014 SNH Tahara Etal
2014 SNH Tahara Etal
2014 SNH Tahara Etal
net/publication/269632499
CITATIONS READS
15 386
6 authors, including:
Silvia Volpi
University of Iowa
13 PUBLICATIONS 221 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Matteo Diez on 03 February 2015.
ABSTRACT INTRODUCTION
The paper describes the development and The long-term collaboration among IIHR (IIHR-
demonstration of Simulation Based Design (SBD) Hydroscience& Engineering, the University of Iowa),
methods, which have been developed through the CNR-INSEAN (Italian Ship Model Basin) and NMRI
long-term international collaboration among three research groups focuses on the development of a
institutes. The methods are capable to solve stochastic Simulation Based Design (SBD) toolbox, which
optimization problems, and results for Robust Design provides high performance computing CFD solvers,
Optimization (RDO) and Reliability-Based Robust optimization algorithms, automatic methods for
Design Optimization (RBRDO) are presented. Non- geometry and grid manipulation, with application to
gradient type, global optimization algorithms are naval hydrodynamics. The SBD was successfully used
applied, based on the success demonstrated in authors’ in the optimization of high speed monohull, multihull,
recent studies. In addition, the concepts of Variable displacement ships, and foil-assisted semi-planing
Fidelity/Physics approach are adopted, that is, both catamaran ferries (Campana, 2006, 2009; Tahara et al.,
Unsteady RANS and Potential Flow based CFD 2008, 2011; Kandasamy, 2011, 2013).
methods are used. Recent development for geometry A new challenge of ship design optimization is to
modelling/modification methods are also shown and introduce stochastic optimization theory, which is able
found promising. In the following, an overview of the to consider not only deterministic function but more
present method is given, and results are presented and complex and realistic design requirements in a form of
discussed for the initial design of a waterjet propelled Stochastic Objective Function (SOF), accounting for
high-speed ship, namely the Delft Catamaran, where real sea stochastic environmental operating conditions
objective functions are minimized, in stochastic form (Tahara et al., 2012; He et al., 2013; Diez et al., 2013).
and associated to real-ocean design conditions. Some related works have been reported, but most of
them are based on classical deterministic optimization
scheme to minimize non-stochastic objective function.
Limitations of pure deterministic optimized
designs are of three types: i) possible degradation of
the performances in off-design conditions due to
stochastic inputs; ii) a deterministic approach to
formulate a multiobjective problem with individual
objective functions for every operational condition of
interest, easily leads to a stiff and almost intractable
optimization problem; and iii) in general, objective
function is defined in a form of non-probability
Figure 1: Computed wave field around Delft Catamaran function, which precludes consideration of SOF. Our
(URANS-CFD results). For Fr=0.5 and model scale Re. approach is to introduce stochastic optimization
methods, which redefines a deterministic optimal speed is given in a form of PDF. Finally, future work
design problem by introducing the probability of our joint project will be given.
distributions of the stochastic events that one is
interested in taking into consideration. The problem is Table 1: Main particulars for Delft catamaran 372 model.
solved in a form of multiobjective optimization Main particulars Symbol Model
problem, by using a suitable numerical scheme Length overall (m) L OA 3.8220
investigated by the authors. Consequently, the present Length between parpendiculars (m) L PP 3.6274
scheme is capable to consider the multi-objective Length on waterline (m) L WL 3.6274
nature in ship design optimization, where the Beam (m) B 1.1570
improvement of a specific aspect of the global design Clearlance n/n hull CPs (m) - 0.8470
usually causes the worsening for some others. Draft at FP (m) TF 0.1815
The present paper describes the development of a Draft at AP (m) TA 0.1815
Dsiplacement volume (m3) Δ, δ 0.0770
CFD-based multiobjective stochastic optimization
Prismatic coefficient CP 0.6160
method, where in particular, Robust Design
Block coefficient CB 0.4027
Optimization (RDO), and Reliability-Based Robust
Longitudinal C.B. L CB -0.0970
Design Optimization (RBRDO) are investigated. Two Wetted surface area (bare hull) (m2) S 1.4220
SBD approaches are developed and applied with focus
on different aspects. Specifically, NMRI-SBD and
IIHR-INSEAN SBD (hereafter referred to as SBD-A
and SBD-B, respectively) are used in a
complementary manner. SBD-A focuses on
multiobjective deterministic and stochastic RDO
problems for commercial ships, by considering calm-
water resistance (RT) and added resistance in waves
(RAW); SBD-B addresses multiobjective deterministic
and more advanced stochastic RBRDO problems for
high-speed sea lift military ships, by considering
expected value of mean RT in waves and ship
operability in real ocean environment. Figure 2: Main dimensions for Delft catamaran 372 model
Advanced geometry modelling/modification (Kandasamy et al., 2013).
methods, i.e., geometric variability exploration based
on free-form deformation (FFD) and Karhunen–Loève DELFT CATAMARAN GEOMETRY AND
expansion (KLE), are used in SBD-B. KLE is capable CONDITIONS
of producing an efficient representation of design
possibilities, with deeper improvements (Diez et al., The Delft Catamaran (DC) parent design and the main
2013; Chen et al., 2014) and is coupled herein with particulars are shown in Figure 2 (see Kandasamy et
metamodels. SBD-A uses a simpler geometry al., 2013 for details). The model is free to sink and
modeling method, based on a promising design from trim. In NMRI SBD optimization, the reference
SBD-B, and investigates a larger number of objective dimensional speed is equal to 2.98m/s for a 3.627m
functions at the time, involving RT and RAW in towing tank model, corresponding to Froude number
stochastic form, addressing also waterjet effects at (Fr) equal to 0.5 (Table 1). In IIHR-INSEAN SBD, a
different speeds. full ship scale with LPP equal to 100m is considered.
Both SBD make use of the same high fidelity The design speed is taken equal to 35 kt
URANS CFD, along with non-gradient type, global (corresponding to Fr = 0.575 for a 100 m vessel).
optimization algorithms, i.e., Multiobjective Particle Table 2 shows main particulars and conditions,
Swarm Optimization (MOPSO) and Evolutionary compared to the JHSV.
Algorithm (EA), which has been shown capable for
the present problem in the authors’ recent studies. In Table 2: Delft Catamaran full scale particulars, with
comparison to JHSV.
the following, an overview of the present method is Parameter Unit Delft Catamaran JHSV
given, and results are presented and discussed for the Propulsion - water jet (2x) water jet (4x)
initial design of a waterjet propelled high-speed ship, Length m 100 103
Beam m 31.3 28.5
namely Delft Catamaran (Van't Veer, 1988, see Figure Draft m 5.0 3.83
1), where in order to consider real-ocean design Displacement t 3,225 2,397
Design speed kt 35 35÷43
conditions, a stochastic variation of ship operation
NMRI-SBD OPTIMIZATION (SBD-A) somehow. In the present study, RDO and RBRDO are
mainly focused in SBD-A and SBD-B, respectively.
General Definition of Multiobjective Deterministic
and Stochastic Optimization Problem URANS-CFD Method
RBDO
Problem
Minimize f (x ; y) PF-CFD Method
{
}
Subject toP g j (x ; y) ≤ 0 ≥ P0 , j = 1,...,q
The PF-CFD methods used to evaluate seakeeping
RBRDO problem performance is FreDOM code, which is a frequency
Minimize µ{f (x ; y)}andσ 2 {f (x ; y)} domain, Rankine-source type panel method developed
{
}
Subject toP g j (x ; y) ≤ 0 ≥ P0 , j = 1,...,q in INSEAN (Lugni et al., 2004). In the theory, the
free-surface flow generated by a ship advancing at
In extension for stochastic optimization problem, a constant forward speed in regular incoming waves is
focus can be on the objectives or on the constraints: considered. The problem is solved by using the
dependently, one can have Robust Design potential flow theory and neglecting the nonlinearities
Optimization (RDO) and Reliability-Based Design connected with the wave-body interactions. The total
Optimization (RBDO). Reliability-Based Robust velocity potential of the fluid is decomposed as the
Design Optimization (RBRDO) refers finally to a case sum of the steady and unsteady wave fields, the latter
in which objectives or constraints are considered in a consisting of the incident, the diffraction and the
radiation waves. Then the problem is split into eight
probabilistic manner. See Table 3 for the three forms
sub-problems: one steady and seven unsteady. Small
of the problem (for simplicity a single objective
amplitude of the incident waves and ship motions are
optimization case is considered), where y is the design
assumed, then the free surface conditions is linearized
parameter collecting the quantity that is independent for the unsteady problems around the steady free
of the designer choice (e.g., environmental conditions), surface. Once velocity potential is determined,
ŷ is a user assigned threshold, and q is number of pressure integration over the hull surface yields the
inequality constraints. To formulate a stochastic added mass and damping coefficients, the restoring
optimization problem, the probability density function forces and the wave exciting forces. The restoring
(PDF) of y, Pr(y), has to be evaluated or given terms are due to both the hydrostatic pressure and the
steady hydrodynamic pressure. Finally, the Response
Amplitude Operator (RAO) is obtained by coupling where S(ω) is Bretschneider spectrum. In the present
the fluid dynamic problem with the body motion study, Sea state 3 is assumed for the most common
through the hydrodynamic loads. For more details, see state in Pacific Ocean route. More details of this
Lugni et al. (2004). scenario are described in He et al. (2013).
Consequently, Tm=2π/ωm=7.5(s) and H1/3=0.9(m) are
Evolutionary Algorithm - Real-Coded used together with Bretschneider spectrum.
Multiobjective Genetic Algorithm
Stochastic Modeling of Speed Variation in Real
EA adopted in SBD-A is the Real-coded Seas
MultiObjective Genetic Algorithm (RC-MOGA). The
authors competitively evaluated both RC-MOGA and In SBD-A, probability density function (PDF) to
more classical Binary-coded MOGA (BC-MOGA) in express ship operation speed variation in real sea is
the earlier work (Tahara et al., 2008). It is found that based on Johnson SU distribution (Johnson, 1949),
each approach offers the advantage over the other which is the four parameter function and capable to
depending on the problem setup, e.g., if the design model a random variable that can take on any value
variables are given as continuous real number, RC- between minus infinity and plus infinity. It was shown
MOGA is more suitable. Otherwise, in more general in our precursory work that on-board measurements
engineering applications details of both algorithms are by Tsujimoto el al. (1996,1998) are well represented
discussed by Deb (2001). In the present basic by using this function. In the present application, the
algorithm of RC-MOGA higher fitness f is given to function is used in the following form:
individuals of higher Pareto ranking RP, i.e., f =1/RP.
A drawback of evolutionary family algorithms, i.e., an f Fr (ΔFr) =
increase of the computational load, is overcome by
introducing parallel computing technique, i.e., ⎧⎪ 1 ⎛ ⎛ ΔFr − ξ ⎞⎞
2⎫
⎪
Message Passing Interface (MPI) protocol with AE δ exp⎨− ⎜⎜ γ + δ sinh −1 ⎜ ⎟ ⎟⎟ ⎬
2
⎩⎪ ⎝ ⎝ λ ⎠⎠ ⎪⎭ (3)
model. See Tahara et al. (2012) for more details.
⎧⎪⎛ ΔFr − ξ ⎞ 2 ⎫⎪
Evaluation of Added Resistance in Waves 2πλ2 ⎨⎜ ⎟ + 1⎬
⎪⎩⎝ λ ⎠ ⎪⎭
Added resistance in waves (RAW) are evaluated by
method of Tsujimoto et al. (2008), which utilizes an where γ and δ (δ > 0) are shape parameters; ξ is
extensive set of NMRI tank test data. In the theory, location parameter; and λ (λ > 0) is scale parameter.
RAW is divided into two parts: RAW (ω, Fr) Also, statistical parameters are analytically given, e.g.,
= RAWm (ω, Fr) + RAWr (ω, Fr) , where, RAWm and RAWr cumulative distribution function and mean value are
are decomposed components due to ship motion and given by F ( x) = Φ(γ + δ sinh−1 z) and ξ − λ ω sinh Ω ,
wave reflection, respectively. The former is given by where z = (ΔFr − ξ ) / λ , ω = exp(δ −2 ) , Ω = γ / δ , and
Maruo’s theorem (1963), and the latter is computed by Φ is cumulative distribution function of the normal
method of Tsujimoto el al. (2008). The theory is distribution. This PDF is used with assumption that
implemented in NMRI RIS code, and used in the mean speed loss ( ΔFr ) is specified as part of design
present study together with FreDOM code described conditions. The present work assumes that ΔFr = -
earlier. On the other hand, wave height is given by 0.03, -0.06, and -0.09; for which corresponding PDF
the Rayleigh distribution, i.e., distribution and function parameters are shown and
given in Figure 3 and Table 4, respectively. For
H ⎛ H2 ⎞ (FL,FU)=(0.1,0.7), where FL and FU are the lower and
f H (H ) = exp⎜⎜ − ⎟
⎟
(1) upper bounds of Fr for evaluation (i.e., ΔFr=-0.4 and
4m0 ⎝ 8m0 ⎠ 0.2, respectively), resultant cumulative probability
distribution are more than 99.5% for all cases.
with
Table 4: Parameters for assumed mean ΔFr ( ΔFr )
⎧ 5 ωm4 2 ⎛ 5 ωm4 ⎞
⎪S (ω ) = H exp⎜− ⎟ Mean ΔFr γ δ ξ λ
⎪ 16 ω 5
1 / 3 ⎜ 4 ω4 ⎟
⎨ ⎝ ⎠ (2) -0.03 2.50 2.00 0.01653 0.00
⎪ ∞ -0.06 2.50 2.00 0.03305 0.00
∫
k
⎪⎩mk = 0 ω S (ω )dω -0.09 2.50 2.00 0.04958 0.00
30 angle of the new design, while 0 and 1 correspond to 0
25 Pr.
JSU
(Average
ΔFr=-‐0.3)
and -2 degs., where the angle is defined as negative
Pr.
JSU
(Average
ΔFr=-‐0.6)
bow downward (see Figure 4), and rotational
20
movements associated with trim are given by matrix
Pr.
JSU
(Average
ΔFr=-‐0.9)
15
AT. Consequently, initial sinkage σ I must be given to
10 yield the same displacement, which is done by vertical
5
translation and σi is numerically obtained as the two
ΔFr design variables are given. The computational volume
0
-‐0.200 -‐0.150 -‐0.100 -‐0.050 0.000 0.050 0.100 0.150 0.200
grid is re-generated similarly, by defining P1 and P2
Figure 3: Probability density function (PDF) of ship to be grid points in solution domain for the baseline
operation speed variation as stochastic input (SBD-A). designs for zero trim condition.
⎧ FU
⎪⎪ EVRT =
FL ∫
RT ( Fr) f Fr ( Fr)dFr
(4)
⎨ FU ∞ ∞
⎪ EV
⎪⎩ RAW
=
FL 0 0 ∫ ∫ ∫
I RAW ( Fr, T , H )dTdHdFr
Figure 4: Consideration of initial trim (SBD-A).
with I RAW = 2m0 RAW (Fr;2π / T , H ) fT (T ) f Fr (Fr) Overview of SBD-A and Setup of Numerical
S (2π / T ) Optimization
and fT (T ) = 2π
m0T 2
NMRI-SBD is composed of three basic components:
The integration in the above equation is done by i.e., OPT (optimizer) module, GM (Geometry
using 3rd order spline function. RAW is directly Modeling) module, and finally AE (Asynchronous
evaluated by using PF-CFD and NMRI-RIS code, Evaluator) module to make flexible interface with
while RT is given by URANS CFD as described later. both parallel and/or serial mode PF and/or more
The present stochastic optimization test cases includes expensive URANS-CFD. AE and CFD methods
two objective function, where EVRT and EVRAW are utilize MPI parallel computing architecture. See
related to two objective functions to be minimized. Figure 4 for an overview. NMRI-SBD was
demonstrated at NMRI by using a compact LINUX-
Shape Parameterization and Grid Re-Generation based PC-Cluster equipped by 10 gigabit network
connecting HP ProLiant BL460 nodes of Intel Xeon
Since SBD-A assumes late stage of initial design by E5-2667 (64bit, 2.9 GHz). In the present optimizations,
using limited number of design variables, a relatively the population size of genetic algorithm coincides
simple approach is used to yield a new design, i.e., with number of MPI groups, each of which utilizes 6
CPUs to fully accelerate the computation. Hence,
(
P = {A T ( β )} a1 P1 + a 2 P2 + σ I ) total number of CPUs used is n(m+1) + 1 = 61 (i.e., n
⎧a1 = (1 − α ) (5) = 6, m+1=10, where m+1 is population size). The GA
where⎨ system parameter, crossover rate, was 0.75, which is
⎩a 2 = α
the same as the one used in Tahara et al. (2012). For
So that a1 + a 2 = 1
the present URANS-CFD, three-block grid system is
used, i.e., two blocks are used for ship domain and one
where P1 and P2 are surface points for corresponding block is used for background domain. The
two baseline designs; and 0≤α≤2 and 0≤β ≤1 are computational domain is only port half side with about
design variables. β is linearly related to initial trim
4.7 million grid points. For this grid arrangement, an
extensive verification and validation studies were
performed by Chen et al. (2014), and detailed
uncertainty quantification (UQ) was presented by Diez
et al. (2014). As mentioned earlier, the present
URANS CFD utilizes CFD WJ model, which model
parameters are given for Fr=0.3, 0.5, and 0.7 for
respective Fr flow simulation. Redesign of WJ is not
considered here, i.e., the same model parameters are
used for the original and optimal designs.
Figure 6: Comparison of geometry for base designs B1 and
Present System B2. Note that B1 is the original design (red and blue outlines,
respectively).
GUI Extended CAD/OPT GUI
Figure 5: SBD-A system overview. Basic components and B.1 [Original] 0.00 100.0%O 100.0%O 2.13
module interfaces. B.1 [Original] -1.00 98.6%O 140.4%O 1.09
B.2 [2nd Baseline] 0.00 89.3%O 97.0%O 1.65
⎩ 3 T
0 0
⎧ LOA ( x )
⎪ g1 ( x ) = L −1 ≤ 0
-0.01 -0.01
X X
-0.02 -0.02
0 0.5 1 1.5 0 0.5 1 1.5
⎪ OA,ori
(Fr=0.3) (Fr=0.5) ⎪ B( x )
Z g
⎪ 2 ( x ) = −1 ≤ 0
0.02
B1
⎪ Bori
0.01 B2 Sub. ⎨ (6b)
0
⎪g3 ( x) = T ( x) −1 ≤ 0
-0.01 ⎪ Tori
-0.02 ⎪
⎪ g ( x ) = δ ( x ) − δ ori = 10 −2 ≤ 0
X
0 0.5 1 1.5
(Fr=0.7)
Figure 7: Comparison of centerline wave profile between ⎪ 4 δ ori
⎩
the baseline designs. The original and optimal designs are B1
design and B2 design (minimum-RT design for Fr=0.5, Chen
et al., 2014), respectively.
where the three functions F1 through F3, are calm-
water model scale RT for Fr=0.3, 0.5, and 0.7,
respectively. Design constraints follow those of
single-objective RT minimization performed by Chen
et al. (2014) and summarized in Table 6.
Multiobjective Deterministic Optimization Table 7: Objective function and design parameters for
sharing points (SBD-A deterministic optimization).
The initial case for NMRI-SBD is multiobjective Initial Trim
F1 * F2 * F3 *
deterministic optimization. Focus here is to obtain Model (deg.)
Pareto optimal set as solutions, explore design space
F1 Minimum (B1 design) 0.00 100.0%O 100.0%O 100.0%O
and capture important trends of new design and flow,
F2 Minimum -0.25 105.0%O 88.2%O 94.6%O
and finally create new design dataset (NDD) which F3 Minimum 0.00 101.2%O 91.6%O 91.1%O
includes a diverse set of new designs and RT data.
NDD is used in stochastic optimization discussed later. * %Original
To represent practical operation range and limit Design variables Blending parameters* Initial Trim
Model x 1 (α) x 2 (β) a1 a2 (deg.)
computational load, two additional speeds are F1 Minimum (B1 design) 0.000 0.000 1.000 0.000 0.000
considered besides a primary target Fr is 0.5, i.e., F2 Minimum 0.840 0.126 0.160 0.840 -0.251
Fr=0.3 and 0.7. As expected from Table 5 (also our F3 Minimum 0.565 0.000 0.435 0.565 0.000
precursory work, Tahara et al., 2012,2013), the
* New design is defined by a 1B.1+a2B.2
present optimization will demonstrate a case dealing
Table 8: Design variables for selected Pareto optimal
designs (SBD-A stochastic optimization).
1.05
Design variables Blending parameters* Initial Trim
Model x 1 (α) x 2 (β) a1 a2 (deg.) Original Design
B.1 [Original] 0.000 0.000 1.000 0.000 0.000 B1
1
B.2 [2nd Baseline] 1.000 0.000 0.000 1.000 0.000
ID.034 0.781 0.165 0.182 0.781 -0.331
F*2
ID.077 0.839 0.178 0.132 0.839 -0.357
* New design is defined by a 1B.1+a2B.2
0.95
Table 9: Objective functions for baseline and selected B2
0.9
Pareto optimal designs (SBD-A stochastic optimization). 0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 1.3
Mean ΔFr=-0.03 F * Average F*1
Initial Trim
Model (deg.)
F1 * F2 * (a)
B.1 [Original] 0.00 100.0% 100.0% 100.0%
1.1
B.2 [2nd Baseline] 0.00 93.2% 144.9% 119.1%
ID.034 -0.33 92.8% 85.5% 89.2%
ID.077 -0.36 92.8% 91.4% 92.1%
1.05
Model Δ F1 * Δ F2 * Δ F * Average
F*3
Original Design
B1
B.1 [Original] 0.0% 0.0% 0.0%
1
B.2 [2nd Baseline] -6.8% 44.9% 19.1%
ID.034 -7.2% -14.5% -10.8%
B2
0.95
ID.077 -7.2% -8.6% -7.9%
* %Original
0.9 0.95 1 1.05 1.1 1.15 1.2 1.25 1.3
Mean ΔFr=-0.06 F * Average F*1
Initial Trim
Model (deg.)
F1 * F2 *
(b)
B.1 [Original] 0.00 100.0% 100.0% 100.0% 1.15
B.2 [2nd Baseline] 0.00 94.7% 186.8% 140.7%
ID.034 -0.33 94.3% 87.9% 91.1%
ID.077 -0.36 94.2% 92.1% 93.2%
1.1
Model Δ F1 * Δ F2 * Δ F * Average
1.05
0.0%
F*3
* %Original B2
0.95
2
effects of blending and consideration of initial trim;
1.8
i.e., reductions of (F1, F2, F3) as much as (+20.3%O, -
1.6
B2
1.4
F*2
value), respectively. Those for the minimum F3 design
1.2
Original Design B1
are (+5.1%O, -5.8%O, -3.4%O), i.e., gain of F2 is
1
smaller than that for the minimum F2 design, but
0.8
ID 077
ID 034
worsening of F1 is also smaller. The blending weights
0.6
(a1, a2, where new design before having initial trim is 0.8 1 1.2
F*1
1.4 1.6
B2
ID 077
⎧− 0.03
1.2
⎧ F1 = EV RT
ID 034
⎪
Original Design B1
Min. ⎨ (7)
F
⎩ 2 = EV ⎪− 0.09
0.8
RAW ID 077
⎩
0.6
0.0004
A(X) Original
0.0002
0.002
A(X) Optimal
0.0000
A(X)-A(X)_Original = 0
0.001
A(X)-A(X)_Original
-0.0002
-0.0004
LCB
Optimal Original (a)
0
0.04
0.02
(B.1 Design)
0
Z
-0.02
Fore Body
-0.04
After Body
-0.06
-0.15 -0.1 -0.05 0 0.05 0.1 0.15
Y
0.04
0.02
(ID-077 Design)
0
Z
-0.02
Fore Body
-0.04
After Body
-0.06
-0.15 -0.1 -0.05 0 0.05 0.1 0.15 (b)
Y
0.15 ID-077
0.1
0.05
-0.05
-0.1
-0.15
0 0.2 0.4 0.6 0.8 1
X
operation/environment y. The above formulation !"2 (x) = ))) ! %& h (x,S,U , # ) $ 0 '(
k
(13)
represents a RBRDO problem, since both objective S ,U ,# k=1
IIHR-INSEAN problem #1 gives the deterministic speed is assumed to range from 20% to 100% of the
baseline solution for calm water, has been given in maximum design speed (0.115 ≤ Fr ≤ 0.575). A
Chen et al. (2014), and is included here for reference. uniform distribution is assumed for the speed; the
stochastic sea state ranges from 1 to 7 and follows the
IIHR-INSEAN problem #2 gives the deterministic probability of occurrence in the North Pacific Ocean.
baseline solution for reduced total resistance in wave Further extension by including variable heading (with
and improved seakeeping performance, considering a uniform distribution from head to following waves) is
single deterministic regular head wave. Accordingly, under investigation.
ϕ1 and ϕ2 are defined respectively as the mean total
resistance in wave and a seakeeping merit factor Deterministic and Stochastic SBD Optimization
(SMF) given by Toolbox
is found along third and fourth basis. Figure 19 (d) resistance and 84% for the SMF, for stochastic regular
shows the same sensitivity analysis for the second head waves representative of sea state 5, at Fr=0.425.
objective of the problem #3. A reduction of 10% is Table 15 provides an overall summary of current SBD
found along the fourth basis for the stochastic SMF formulations and solutions, comparing SBD-A and
(through 1D UQ). Finally, Figure 19 (b) and (e) SBD-B. Opt. #3 design performances are also
present the sensitivity analysis for the stochastic included, for SBD-B problems #2 and #4.
objectives in problem #4. A 10% reduction is found
along the first basis for the expected value of the mean CONCLUSIONS
resistance in wave, whereas a 7% increase in
operability is found along the fourth basis. The high-fidelity hull optimization of a waterjet
Figure 17 shows the solution for problems #2, #3 propelled ship is successfully demonstrated by two
and #4. Optimization evolution is shown, by SBD approaches, i.e., NMRI-SBD and IIHR-
sequential DoEs and metamodel-based MODPSO. INSEAN-SBD, referred to as SBD-A and SBD-B. The
Selected designs are shown in Figures 18 and 19, two SBD methods were used in a complementary
including comparison with the original hull and the manner, providing applications to classical design
calm-water optimized design (Opt. #1) from Chen et problems for commercial ships (SBD-A), and
al. (2014). The associated performances in regular advanced applications for high-speed sea lift military
head wave representative of sea state 5 at Fr=0.425 are vessels (SBD-B).
shown in Figure 20, in terms of force, heave and pitch For SBD-A, advantage of EA based framework is
motions. Opt. #3 is selected as a candidate for model again indicated, i.e., a scalarization approach, which is
tests, providing 24% reduction for expected mean used in classical single objective optimization scheme
(a) Original (b) Opt. #1
Figure 19: Comparison of section lines for original and optimized designs (SBD-B).
and combines the several objective functions with Stochastic optimization problems are solved using a
given weights may be misleading in consideration of, multi-objective reformulation for RBRDO. Sensitivity
e.g., EVRT and EVRAW, for which magnitude of analysis and design optimization have been shown,
functional gain are considerably different. A way to based on KLE-reduced dimensionality space,
consider stochastic design conditions is also shown stochastic regular wave UQ models, sequential DoEs
promising this time, where a stochastic variation of and metamodel-based MODPSO. Significant
ship operation speed is given in a form of PDF, which improvements have been achieved for all problems.
can be modified according to desired operation Solutions to SBD problems in head waves fall in the
conditions such as course and season. same region of the design space and are similar
For SBD-B, four deterministic/stochastic (Figures 18 and 19). Opt. #3 design has been selected
optimization problems have been presented for Delft as a candidate for model tests, providing 24 and 84%
catamaran in calm water and head wave. The latter reduction for expected mean resistance and SMF, in
address different level of complexity, from single stochastic regular head waves representative of sea
deterministic regular wave to fully stochastic ocean state 5, at Fr=0.425. Opt. #3 design also achieves
environment with variable speed and sea state. significant improvements for problems #2 and #4,
with a reduction of 23% for SMF in deterministic Diez, M., Chen, X., Campana, E.F., “Reliability-based
regular wave, and 9% for expected mean resistance in robust design optimization for ships in real ocean
variable sea state/speed, with an associated increase of environment,” Proc. 12th International Conference
operability of 6% (Table 15). Calm water simulations on Fast Sea Transportation, FAST2013,
are in progress for resistance decomposition studies Amsterdam, The Netherlands, 2013.
and supporting the final decision on the optimized Eberhart R., Shi Y., “Comparing inertia weights and
design for model tests. Verification and validation of constriction factors in particle swarm
URANS simulations, along with EFD UQ for final optimization,” Proc. Congress on Evolutionary
design will be conducted accordingly. Computation, La Jolla, CA, USA, 2000.
Hasselmann, K., Barnett, TP., Bouws, E., Carlson, H.,
ACKNOWLEDGMENTS Cartwright, DE., Enke, K., Ewing, JA., Gienapp,
H., Hasselmann, DE., Kruseman, P., Meerburg, A.,
This work has been supported by the US Navy Office Mller, P., Olbers, DJ., Richter, K., Sell, W.,
of Naval Research (Grant N00014-09-1-0979, Walden, H., “Measurements of wind-wave growth
N62909-11-1-7054, N00014-11-1-0237, N62909-11- and swell decay during the Joint North Sea Wave
1-7011) under the administration of Dr. Ki-Han Kim Project (JONSWAP),” Ergnzungsheft zur
Dr. Woei-Min Lin, and Prof. Kenji Uchino; and Deutschen Hydrographischen Zeitschrift Reihe,
Grant-in-Aid for Scientific Research, Japan (Project A8, 12, 1973.
Number 21360436 and 24360363); and the Italian He W, Diez M, Zou Z, Campana EF, Stern F,
Flagship Project RITMARE, coordinated by the “URANS study of Delft catamaran total/added
Italian National Research Council and funded by the resistance, motions and slamming loads in head
Italian Ministry of Education, within the National sea including irregular wave and uncertainty
Research Program 2011–2013. quantification for variable regular wave and
geometry,” Ocean Engineering, Vol. 74, 2013, pp.
REFERENCES 189-217.
Huang, J., P. Carrica, and F. Stern, “Semi-coupled
Campana, E.F., Peri, D., Tahara, Y., Stern, F., “Shape air/water immersed boundary approach for
Optimization in Ship Hydrodynamics using curvilinear dynamic overset grids with application
Computational Fluid Dynamics,” Computer to ship hydrodynamics, Int. J. Numer. Meth. Fluids,
Methods in Applied Mechanics and Engineering, Vol. 58, 2008, pp. 591–624.
Vol. 196, 2006, pp. 634–651. Johnson, N., “Systems of frequency curves generated
Campana, E.F., Peri, D., Tahara, Y., Kandasamy, M., by methods of translation,” Biometrika, Vol. 36,
Stern, F., “Numerical Optimization Methods for 1949.
Ship Hydrodynamic Design,” SNAME Kandasamy, M., Ooi, S.K., Carrica, P. & Stern, F.,
Transactions, Annual Meeting, Providence. Rhode “Integral force/moment waterjet model for CFD
Island, USA, 2009. simulations,” ASME J. Fluids Eng., Vol. 132, No.
Carrica, P., Wilson, R.V. & Stern, F., “Unsteady 10, 2010.
RANS Simulation of the Ship Forward Speed Kandasamy, M., He, W., Tahara, Y., Peri, D.,
Diffraction Problem,” Computers and Fluids, Vol. Campana, E., Wilson, W. & Stern, F.,
35, 2006, pp. 545-570. “Optimization of waterjet-propelled high speed
Chen, Xi, Diez, M., Kandasamy, M., Zhang, Z., ships - JHSS and Delft Catamatran,” Proc. 11th
Campana, E.F., Stern, F., “High-fidelity global International Conference Fast Sea Transportation,
optimization of shape design by dimensionality FAST 2011, Honolulu, Hawaii, USA, 2011.
reduction, metamodels and deterministic particle Kandasamy, M., Peri, D., Tahara, Y., Wilson, W.,
swarm,” Engineering Optimization, DOI: Miozzi, M., Georgiev, S., Milanov, E., Campana,
10.1080/0305215X.2014.895340 EF., Stern, F., “Simulation based design
Coleman, H. W. and Stern, F., “Uncertainties and optimization of waterjet propelled Delft
CFD code validation,” ASME J. Fluids Eng., Vol. catamaran,” International Shipbuilding Progress,
119, No. 4, pp. 795–803, 1997. Vol. 60, 2013, pp. 277-308.
Deb, K., Multi-Objective Optimization using Kennell C.G., White B.L., Comstock E.N.,
Evolutionary Algorithms, John Wiley & Sons, “Innovative Naval Designs for North Atlantic
2001. Opeartions,” SNAME Transactions, Vol. 93, 1985,
Diez M., Peri D., “Robust optimization for ship pp. 261-281.
conceptual design,” Ocean Engineering, Vol. Lugni, C., Colagrossi, A., Landrini, M. & Faltinsen,
37, 2010, pp. 966-977. O.M., “Experimental and numerical study of semi-
displacement mono-hull and catamaran in calm multiobjective optimization of a surface
water and incident waves,” Proc. 25th Symposium combatant,” J. Marine Science and Technology,
on Naval Hydrodynamics, St. John's, Canada, Vol. 13, No. 2, 2008, pp. 95-116.
2004. Tahara, Y., Hino, T., Kandasamy, M., He, W., Stern,
Maruo, H., “Resistance in waves, research on F., “CFD-based multiobjective optimization of
seakeeping qualities of ships in Japan,” J. Society waterjet propelled high speed ships,” Proc. 11th
of Naval Architects of Japan, Vol. 8, 1963, pp. 67- International Conference Fast Sea Transportation,
102. FAST 2011, Honolulu, Hawaii, USA, 2011.
Pellegrini, R., Rinaldi, F., Campana, E.F., Diez, M., Tahara, Y., Peri, D., Campana, EF., Stern, F., “Single
Fasano, G., Iemma, U., Liuzzi, G., Lucidi, S., and multiobjective design optimization of a fast
Serani, A. & Stern, F., “Application of derivative- multihull ship: numerical and experimental
free multi- objective algorithms to reliability-based results,” J. Marine Science and Technology, Vol.
robust design optimization of a high-speed 16, No. 4, 2011, pp. 412-433.
catamaran in real ocean environment,” Proc. 4th Tahara, Y., Kobayashi, H., Kandasamy, M., He, W.,
International Conference on Engineering Peri, D., Diez, M., Campana, E., Stern, F., “CFD-
Optimization, EngOpt 2014, 8–11 September 2014, based multiobjective stochastic optimization of
Lisbon, Portugal. waterjet propelled high speed ships,” Proc. 29th
Peri, D., Kandasamy, M., Tahara, Y., Wilson, W., Symposium on Naval Hydrodynamics,
Miozzi, M., Campana, EF., Stern, F., “Simulation Gothenburg, Sweden, 2012.
based design with variable physics modeling and Tsujimoto, M., Naito, S., “A New Analytic Method of
experimental verification of a water-Jet propelled Voyage data,” Proc. 2nd New Ship & Marine Tec.
catamaran,” Proc. 29th Symposium on Naval 21st Century, 1998.
Hydrodynamics, Gothenburg, Sweden, 2012. Tsujimoto, M., Naito, S., “Estimation of Speed
Stern, F., Wilson, R. V., Coleman, H. W., and Margin by Probabilistic Methods,” Proc. 3rd
Paterson, E. G., “Comprehensive approach to Korea-Japan Joint Workshop on Ship and Marine
verification and validation of CFD simulations— Hydrodynamics, 1996.
Part 1: Methodology and Procedures,” ASME J. Tsujimoto, M., Shibata, K., Kuroda, M., Takagi, K.,
Fluids Eng., Vol. 123, No. 4, 2001, pp. 793–802. “A Practical Correction Method for Added
Stern, F., Wilson, R., and Shao, J., “Quantitative Resistance in Waves,” J. Japan Society of Naval
approach to V&V of CFD simulations and Architects and Ocean Engineers, Vol. 8, 2008,
certification of CFD codes,” Int. J. Numer. Meth. pp.141-146.
Fluids, Vol. 50, 2006, pp. 1335–1355. Van't Veer R., “Experimental results of motions and
Tahara, Y., Takai, T., “High-performance multi- structural loads on the 372 catamaran model in
objective evolutionary algorithms for head and oblique waves,” Technical University of
computational fluid dynamics-based design Delft Report, 1130, 1988.
optimization,” Proc. 3rd PAAMES and Xing, T. and Stern, F., Factors of safety for
AMEC2008, Chiba, Japan, 2008, pp. 313 – 323. Richardson extrapolation, ASME J. Fluids Eng.,
Tahara, Y., Peri, D., Campana, E.F., Stern, F., Vol. 132, No. 6, 2010.
“Computational fluid dynamics-based
Table 15: SBD optimization problems summary.