You are on page 1of 11

Ocean Engineering 38 (2011) 1054–1064

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Optimization of fishing vessels using a Multi-Objective Genetic Algorithm


Mark A. Gammon 1
Defence R&D Canada—Atlantic, Dartmouth, Nova Scotia, Canada

a r t i c l e i n f o a b s t r a c t

Article history: A fishing boat hull is used as an example of how hull form optimization can be accomplished using a
Received 17 August 2010 Multi-Objective Genetic Algorithm (MOGA). The particular MOGA developed during this study allows
Accepted 5 March 2011 automatic selection of a few Pareto Optimal results for examination by the designers while searching
Editor-in-Chief: A.I. Incecik
the complete Pareto Front. The optimization uses three performance indices for resistance, seakeeping
Available online 23 May 2011
and stability to modify the hull shape to obtain optimal hull offsets as well as optimal values for the
Keywords: principal parameters of length, beam and draft. The modification of the 148/1-B fishing boat hull, the
Optimization parent hull form of the _Istanbul Technical University (_ITÜ) series of fishing boats, is presented by first
Multi-objective fixing the principal parameters and allowing the hull offsets to change, and secondly by simultaneously
Fishing vessel
allowing variation of both the principal parameters and the hull offsets. Improvements in all three
Resistance
objectives were found. For further research the methodology can be modified to allow for the addition
Seakeeping
Stability of other performance objectives, such as cost or specific mission objectives, as well as the use of
enhanced performance prediction solvers. In addition, one or more hulls could be evaluated by
experiment to validate the results of using this particular optimization approach.
& 2011 Elsevier Ltd. All rights reserved.

1. Introduction percent can lead to substantial savings, especially in large ships.


For example, using a bunker fuel charge of approximately $450/
Hull form optimization is a process that involves changing a metric ton,2 given a ship that burns 150 ton/day over a 14 day trip
ship or boat hull in order to improve performance such as across the Pacific,3 the overall cost for fuel alone would be
resistance, seakeeping, stability and so forth. The hull is a funda- $945,000. An improvement of even 2% would represent a savings
mental component of a vessel and has a significant influence on the of $18,900 per trip. Minimizing resistance by creating a slender
performance and consequently on overall success of the design. hull, for example, conflicts with stability performance, which is
A single design optimization problem optimizes a single objective increased by the greater beam. Greater beam in turn increases
function while satisfying some design requirements, while multi- viscous resistance. An optimized design requires that these
ple objective design optimization examines trade-offs between conflicting performance criteria can reach compromise.
often conflicting aspects of the design problem. The individual Evolutionary Algorithms (EA) and Artificial Neural Networks
objective or ‘cost’ functions, such as minimization of resistance by (ANN or NN) offer effective methods for conducting optimization
Day and Doctors (1997) or minimization of the total catamaran and data analysis. EA techniques may be separated into Genetic
resistance by Danisman et al. (2001) or minimization of the wash Algorithms (GAs), Evolution Strategies (ESs) and Evolutionary
from high speed vessels by Zaraphonitis et al. (2003), represent programming (EP). In this study, the term GA is predominantly
different aspects of design optimization. used to reflect the encoding and characteristics of the algorithm,
The shape of the hull impacts every aspect of a design. Three unless reference is made to a specific technique. For example, Day
performance objectives are considered as examples from among and Doctors (1997) studied hull form optimization using a GA
the numerous issues facing a design team, namely stability, technique in which the objective was to minimize resistance.
resistance and seakeeping. Stability must satisfy or exceed certain Their study varied a wide range of hull displacements and
constraints and is often modeled as a constraint rather than an examined the optimization trends that occurred on the basis of
objective function. Seakeeping obviously impacts human safety variation of the principal parameters. Yasukawa (2000) and
and comfort. Resistance is one of the chief costs in the operation Dejhalla et al. (2002) have both conducted a resistance optimiza-
of the vessel, such that minimizing resistance by even a few tion analysis of a hull form using GA methods where the objective
was also to minimize wave resistance. Those studies focused on

E-mail addresses: mark.gammon@drdc-rddc.gc.ca, mark.gammon@forces.gc.ca


1 2
Research conducted while on leave at Yildiz Technical University, Istanbul, http://www.bunkerworld.com/markets/surcharges/tsa
3
Turkey. http://www.tsacarriers.org/fs_bunker.html

0029-8018/$ - see front matter & 2011 Elsevier Ltd. All rights reserved.
doi:10.1016/j.oceaneng.2011.03.001
M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064 1055

existing hulls that were modified by varying the hull offsets database of known designs. The focus of this study is to compute the
slightly while maintaining the same principal characteristics of performance factors directly for each candidate hull. In addition to the
the length, beam, draft and displacement. While other optimiza- principal parameters, the optimal hull offsets for the hull shape or
tion methods such as simulated annealing, Lipshitz Global Opti- hull form should be determined simultaneously. That is to say, the
mization and other methods are routinely used, as stated by Gen near-optimal hull form should also include the near-optimal length,
and Cheng (2000), the inherent characteristics of genetic algo- beam and draft, as well as satisfy a displacement requirement, in
rithms, i.e. multiple directional and global search, lack of math- order to create a near-optimal design.
ematical requirements, ability to handle all types of objective
functions and ability to be combined with conventional methods, 2.1. General multi-objective problem definition
make MOGA techniques well suited to multiple objective optimi-
zation problems. In generic terms, the functional form of the problem is given as
The MOGA approach presented in this study can be extended follows. We need to determine the vector of decision variables as
to any hull form optimization problem in which the design described in Coello Coello (1996):
requirements are known and can be formulated as a multiple !
objective problem. As an example, the MOGA methodology was x ¼ ½x1 ,x2 ,x3 ,. . .,xn T
applied to fishing boats in order to determine whether a more where xj , j ¼ 1,2,. . .,n are the decision variables. As an example,
optimal boat hull could be derived. The current study represents for this study, the decision variables include the principal para-
one example of how this specific MOGA can be applied during the meters of the vessel and the hull offsets, i.e.
initial and concept stages of vessel design. The significant pro-
x1 ¼ L; x2 ¼ B; x3 ¼ T; x4 ¼ W
blem faced by the designer, whether the vessel is small or large, is
the choice of the optimal principal parameters that will lead to a where L, B and T are the length, beam and draft, respectively. The final
successful design. Most, if not all, MOGA methodologies conduct decision variable W is the hull offsets represented as a matrix.
design optimization by searching out the entire Pareto Front, The solution must satisfy the m number of inequality con-
which will be described later. While effective in determining the straints:
optimal candidates, the plethora of possible solutions leads to a gi Z 0, i ¼ 1,. . .,m
solution space nearly equal to the population sample size, which
can number in the hundreds. It is usually left to the designer to and p the number of equality constraints:
choose the more favoured design, which can be a daunting task. hi ¼ 0, i ¼ 1,. . .,p
The current methodology allows an automatic selection of the
number of optimal compromise solutions to give to the designer. where p as the number of equality constraints should be less than
In addition, the calculation of the performance objectives in this the number of decision variables n in order to avoid being over-
study is deliberately not computer intensive, enabling cost- constrained. Most design factors can be captured as constraints,
effective initial boat design studies to be conducted as in as well as limits of the solution domain. The constraints are
Gammon (2004). Future research would focus on the use of more discussed further under the design requirements.
advanced functional representations of the performance objec- The solutions must optimize the vector function:
tives as in Maisonneuve et al. (2003) using this MOGA approach, ! ! ! !
f ¼ ½f1 ð x Þ,f2 ð x Þ,. . .,fk ð x ÞT
as additional resources become available.
! ! !
This paper is structured as follows. Section 2 is concerned with The objective functions f1 ð x Þ, f2 ð x Þ and f3 ð x Þ represent
the problem formulation and in particular the definition of the resistance, seakeeping and stability indices, respectively. In shor-
multi-objective problem along with the development of the tened form:
relevant indices that represent the individual cost or performance ! !!
f ¼ opt f ð x Þ
functions for each of the objectives. Then, a particular form of x¼O
MOGA is presented in Section 3 with some methods for encoding !
f : O-Rk
the problem. Section 4 presents results of application of this
! !! !
methodology using two different examples of the Istanbul Tech- O ¼ f x A Rn 9 g ð x Þ Z 0, hð x Þ ¼ 0g
nical University (ITU) fishing hull, the first with fixed principal
where k is the number of objectives.
parameters of length, beam and draft, and the second allowing
The multi-objective definition of optimality, known as Pareto
these parameters to change simultaneously with the hull offsets.
Optimality, is defined as a point in n-dimensional space repre-
The fishing boat series as described by Kafalı et al. (1979) was
sented by
developed by ITU for Turkish fishermen in order to have a
!
standard series with known and measured characteristics in x AO
terms of seakeeping and resistance, for which experimental data !
is well known. Finally, Section 5 gives some conclusions regarding such that for every, x  A O and I¼{1,2,y,k}, I¼ either
! !
this particular approach along with the scope for future work. 8i A I ðfi ð x Þ ¼ fi ð x ÞÞor there is at least one iAI such that
! ! ! !
fi ð x Þ 4 fi ð x Þ (for maximization problems) or fi ð x Þ o fi ð x Þ (for
minimization problems).
2. Optimization problem formulation In this study, the term near-optimal is used to describe a
design choice that achieves some compromise in the performance
Determining the optimal principal parameters for length (L), beam objectives while satisfying constraints for the design features.
(B) and draft (T), as well as volumetric displacement (r), is most often
accomplished by parametric variation of a parent hull. Usually hull 2.2. Formulation of performance indices
form optimization consists of only changing offsets of an already
suitable hull in order to optimize a particular performance objective. 2.2.1. Objective 1—resistance performance index
However, at the preliminary design stage, the principal parameters of The non-dimensional total resistance coefficient CT)ship is
the vessel must be determined. These are often determined through
regression based analyses predicting performance attributes from a CT Þship ¼ Cv þ CW Þship þ ca ¼ ð1 þ kÞCF þ CW þ ca
1056 M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064

where Cv is the viscous resistance coefficient, CW)ship is the wave equally. The RCI is calculated as follows:
making resistance coefficient and ca is a correlation allowance. Cv
X
N 1
1
can be represented in terms of the frictional resistance coefficient RCI ¼ ðCT þ CTi þ 1 ÞðFni þ 1 Fni Þ
CF and form factor k, i.e. Cv ¼(1þk)CF. The form factor k is i¼1
2 i
determined from the model test and is assumed independent of where N is the number of Froude numbers, CTi the resistance
speed and scale. For example, in the ITU fishing hull forms the coefficient at speed i and Fni the Froude number at speed i.
tow tank test for the parent form of the ITU series, ITU 148/1-B, !
The resulting objective for f1 ð x Þ is then represented as
showed a form factor of 0.25 whose tests were done with follows:
5 models with different scales as described by Kafalı et al. !
opt f1 ð x Þ ¼ min½RCIðL,B,T,WÞ
(1979). The same form factor is assumed for the full-scale ship.
The non-dimensional form of the total resistance RT)ship is a
2.2.2. Objective 2—seakeeping performance index
function of the ship speed V and wetted surface area S of the ship:
Seakeeping performance is a complex area of analysis and needed
RT Þship resolution into a single seakeeping performance index similar to the
CT Þship ¼
0:5rV 2 S resistance coefficient index in order to be useful in the current multi-
optimization problem. In ship motion, numerous seakeeping factors
where r is the density of water.
are relevant including acceleration at various points on the vessel,
The International Towing Tank Conference (ITTC) proposed a
slamming effects, crew response and motion sickness index. Since
frictional resistance formula based on the Reynolds number. The
there are numerous seakeeping factors, and these represent aspects of
ITTC 1957 frictional line is calculated as follows:
this particular performance attribute of the hull form, it was prudent
0:075 to resolve these into a single performance index.
CF ¼
ðlogðRnÞ2Þ2 The hull form optimization hypothesis is that the best hull
where Rn is the Reynold’s number given by form is the one that minimizes all of the motions. While there
may be conflicting influences in the motions between heave, pitch
rVL and rolling, the latter was considered to be characterized by the
Rn ¼
m beam and may also be regarded as part of the stability criteria.
where m is the viscosity, V is the ship speed and the length of the The focus for the seakeeping performance is the heave and pitch
vessel is L. motions as shown in Fig. 2.
In order to determine CT)ship, we need to determine CW)ship, the For the purpose of comparing candidate hull forms, the problem is
wave making resistance coefficient. CW)model for a model is then simplified by considering only the vertical motion from the pitch
assumed to be equal to the full scale ship Cw)ship, i.e. and heave. The total vertical motion, measured at the bow, is
combined from the maximum heave and pitch motions. This is
CW Þship ¼ CW Þmodel
multiplied by the heave acceleration to give a pseudo-equivalent
Hence momentum. The result is averaged over each ship speed to determine
pseudo energy density. The mass of the vessel is deliberately left out
CT Þship ¼ CW Þmodel þ ð1 þ kÞCF þ ca
as the combined pitch and heave are figurative and the maximum
For the example in this study, wave resistance CW is calculated values of each motion do not frequently occur simultaneously.
using a transom modified Michell integral using potential flow Vertical motion could be combined into one vertical seakeeping
theory, as described in Gammon (1990). A comparison of wave motion index by integrating the values obtained at each heading
resistance with experimental, Michell integral and the modified but just head seas were utilized over each ship speed. The equation
transom integration is shown in Fig. 1. It would appear that the for the seakeeping index is given as
transom effect is considerable for vessels with a low L/B ratio as in X
N 1
1
Gammon and Alkan (2001). At the higher Froude number (Fn) of SKI ¼ ðVerti þVerti þ 1 ÞðVi þ 1 Vi Þ
2
0.5 the effect is over pronounced using the transom theory, but as i¼1

the normal vessel speed is approximately 10 knots, the prediction where Vert represents the vertical calculation at each ship speed (V)
up to Fn 0.4 is in good agreement, and considerably closer to the € rms ):
using the heave (Hrms), pitch (frms) and heave acceleration (H
experimental curve as compared to the unmodified Michell integral.   
A Resistance Coefficient Index (RCI) is formulated from each CT € rms Hrms þ L sinðf Þ
Vert ¼ H rms
value at each speed or Froude number as a representation of the 2
area under the resistance curve to measure the overall resistance As for the first objective, the optimization for the second
performance. In the current approach, the speeds are treated !
objective f2 ð x Þ is represented as follows:

!
opt f2 ð x Þ ¼ min½SKIðL,B,T,WÞ
Michell
0.030 Gammon
Experimental
2.2.3. Objective 3—stability performance index
Cw,Cr

0.020 Stability is an area of ship research that is by itself too large to


treat in detail. It is a fundamental performance criterion that is
regulated by various ship classification societies such as the American
0.010 Bureau of Shipping (ABS) and Lloyd’s and must be part of the
evaluation of any concept design. A stability performance index is
0.000 used in the hull form optimization program that was developed by
0 0.1 0.2 0.3 0.4 0.5 0.6 Gammon and Yilmaz (2003) using design model parameters for
Fn stability that had previously been modeled using regression-based
Fig. 1. Comparison of Michell and Transom-modified wave resistance with
formulas from Grubisic (2001) and Yilmaz (1999). These parameters
experimental result for ITU Fishing Boat ITU 148/1-B at a particular load case take the form of stability constraints defined for the particular ship
(L.C.1). design problem, given requirements from the International Maritime
M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064 1057

HEAVE
Wave

PITCH
DWL

PITCH
Total Vertical Motion
HEAVE

Fig. 2. Pitch and heave motion combined into total vertical motion.

GZ [m] 3. Genetic algorithm approach

GM [m] Multi-objective problems exist in a wide range of practical


GZmax applications. Though other methods such as weighted averaging
techniques can be used, multiple objective problems can be
treated by the determination of the Pareto Front, in which no
solution is dominated by another in one or more performance
criteria as described by Gen and Cheng, (2000). Genetic Algo-
rithms are stochastic search and optimization techniques that
have the following five basic components:
Φm 1 rad Φv
Φ  A genetic representation of solutions to the problem;
 A way to create an initial population of solutions;
Fig. 3. GZ curve with stability index elements.  An evaluation function rating solutions in terms of their
fitness;
 Genetic operators that alter genetic composition of offspring
Organization (IMO) codes on intact stability or other regulation during reproduction; and
sources.  Values for parameters of Genetic Algorithms.
A single measure for the stability index in order to minimize
the number of design objectives was required in order to develop For the ship designer, finding solutions all along the Pareto
a similar metric as used in other hull design problems such as for Front using these techniques raises the problem of choosing the
sailing yachts given in Larrson and Eliasson (2000), but which near-optimal compromise solutions, from which some will have
could still provide a general indication of stability performance. to be considered as more favorable as a compromise than others.
The use of the stability index based on a single stability char- Most often, the choice of a compromise solution is left to the
acteristic such as either the area under the GZ curve or the designer. The need to automate this approach or at least provide
vanishing angle can be erroneous as these single measurements some assistance in achieving a compromise or near-optimal
may be the same for different hulls. As the objective is to compare solution has been a significant driver in the development of the
different hulls then the stability index should have the ability to current optimization methodology.
differentiate between different hulls. A stability index that uses For this study investigation into a particular MOGA was used
multiple stability characteristics provides a better assessment for to address the multiple objective issues by automatically deter-
the purpose of evaluating which hulls should be considered as mining a compromise solution. The purpose of generating this
more optimal, when considering multiple objectives. MOGA solver was alluded to earlier, that is, the problem facing
As a result the use of both the GM as a constraint according to the designer is not, conversely, the search for all solutions from
regulations and also a GZ curve is utilized as depicted in Fig. 3. the entire Pareto Front, which is usually the only optimal strategy
The positive area under the GZ curve up to the vanishing angle fn available in multi-objective problems. While aggregation or other
gives a good measure of the kinetic energy that can be absorbed preferential and interactive techniques are used, the goal of this
by the hull. The angle at which maximum GZ occurs, fm, is an study was to achieve some level of automation, i.e. to be able to
indication of the angle after which the hull may have a tendency come up with a few compromise solutions, which could then be
to capsize as a result of a diminishing GZ. examined by the design team.
The hydrostatics are calculated for each hull form and the GZ
curve is used to generate a stability index as follows: 3.1. Multi-Objective Genetic Algorithm (MOGA)
Rf
STIX ¼ fm 0 v GZðfÞdf
This is taken as the area under the GZ curve up to the The following was developed in order to address some of these
vanishing angle multiplied by the angle at which maximum GZ specific issues. The canonical Genetic Algorithm by Goldberg
occurs. The resulting stability index is indicative of the overall (1989) is modified as shown in Fig. 4 by treating each objective
stability of the hull form, but does not preclude other stability sequentially. For each objective the population is evaluated
requirements. For example, additional constraints to limit stabi- separately, and the genetic operations applied after each evalua-
lity in order to avoid unwanted stiffness in the optimal results tion to generate the next population. The current optimum, if
could be used, though the current study only used a constraint there is one, is returned at each evaluation. In some ways this
based on minimum GM requirements. For the third objective approach is similar to the VEGA approach presented by Schaffer
!
f3 ð x Þ, the optimization is then represented as (1985) where different subpopulations are kept separate for a
! number of generations, and then allowed to mix. That method
opt f3 ð x Þ ¼ max½STIXðL,B,T,WÞ allowed the population to approach a compromise solution that
1058 M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064

Fig. 4. Multi-Objective Genetic Algorithm approach.

would perform well in more than one objectives. Unlike VEGA, algorithm is quite subject to the degree of processing required to
the proposed methodology selects parents that are based only one evaluate each objective. Furthermore, the accuracy of the results
objective. The next objective in the sequence is evaluated with is also subject to the individual solutions provided by the
the population generated from parents that performed well from functional evaluations of each objective.
the previous objective. The treatment of the multi-objective
problem is reduced to single objectives in sequence, similar to a 3.2. Hull modeling using chromosomes
gradient method in which each function’s evaluation leads to a
determination of the direction of search. The relative importance In order to be able to use evolutionary algorithms for hull form
of which objective is used was found to be immaterial, as long as optimization, it is necessary to develop a scheme to map the
the requisite number of iterations, of a minimum of approxi- problem into a format that can be utilized by the algorithm. The
mately 10 generations, was used. However, this could vary with parameters of the problem need to be defined. In every applica-
the design problem and only the specific design problem here is tion of an GA, the problem of mapping the parameters for
discussed. Further research in the generic application of this candidate solutions follows from the development of the Genetic
method would be required. It should be noted that the require- Algorithm. As stated by Gen and Cheng (2000), encoding the
ment to have a sufficient population size, as well as a minimum solutions may require further development of heuristics to
number of generations, means that the efficiency of the genetic manage the solution properties. For the first part of the hull form
M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064 1059

model the principal parameters are considered. Given the length, automatically generated, given the fact that a B-spline surface
beam and draft the basic dimensions of the hull are defined. Gen was in fact used for modeling of the hull, a more direct and
and Cheng (2000) show how the accuracy and the upper and the accurate interpretation of the hull surface could have been
lower limits are defined for a single chromosome. Using the achieved using the control points of the B-spline surface directly
following representation for the domain [aj,bj] for each variable xj: in the optimization, rather than the use of varying hull offsets, as
one of the input parameters sets that were varied during the
2mj 1 o ðbj aj Þ  103 r 2mj 1 1
optimization.
where the accuracy of 0.001 represented by the range from aj to bj
is multiplied by 103 to move the decimal point by the required
number of digits. The power mj then represents the number of 4. Fishing vessel optimization
bits required in the chromosome. The mapping of each variable is
obtained by Fishing vessels have typically developed from what were
historically small inshore fishing vessels that gradually evolved
ðbj aj Þ
xj ¼ aj þ dchr into larger vessels, as depicted in Fig. 5. Possibly as a result of this
2mj 1 1 historical evolution, Turkish fishing vessels often have a wide
where dchr is the decimal value between 0.0 and 1.0 determined high beam and low depth, as well as low draft as described by
by the chromosome. Alkan (2004), and for the fishing vessels that were the focus of the
The parameters for the principal parameters of the hull are put study, this could result in the possibility of reduced large angle
into a format for the GA. The length, beam and draft can be stability and higher block coefficients, that as one detrimental
described by a binary representation where the limits above are factor can lead to higher resistance. It is the author’s observation
used to determine binary values. Using the equation above to that their evolution from small craft that was built and hauled up
determine the number of bits required for the ranges assumed for on shore has resulted in shallow draft and broad beam. Hull forms
the length, beam and draft and a decimal accuracy of 0.001 gives for small boats originally built up to 10 m in length have been
the following, for an example of length limitation between 10 and scaled up to vessels as large as 60 m in length for commercial
30 m, a breadth limitation between 3 and 5 m, and a draft fishing. Scaling the hull form has provided large working areas for
limitation between 1 and 3 m: the decks and shallow draft, but this has not always proven
advantageous in terms of the resistance, stability and seakeeping
Length : 2mj 1 oð3010Þ  103 r 2mj 1, m1 ¼ 15
characteristics.
Beam : 2mj 1 oð53Þ  103 r 2mj 1, m2 ¼ 11 The problem formulation begins with the requirement to
Draft : 2mj 1 o ð31Þ  103 r 2mj 1, m3 ¼ 11 satisfy some particular design characteristics, in this case the
owner’s requirements for a fishing boat hull. The fishing boat can
To represent the hull offsets, the matrix W is formed from m
be considered a difficult design problem for optimization because
stations and n waterlines such that
it is a small craft relative to ordinary cargo vessels, and require-
0 1
station 1, waterline 1 . . . station m, waterline 1 ments such as the working conditions on the deck are critical for
B ^ ... ^ C safety concerns. The fishing boat example uses a number of
W¼@ A
station 1, waterline n . . . station m, waterline n factors representing the design criteria and the owner’s require-
ments that are given next.
Recombination is approached by choosing a random point in
the matrix and in the string representing only the hull offset at
4.1. Fishing boat design characteristics
that position. The recombination can be done in several ways,
however in keeping with the GA methodology; the matrices are
The resistance, seakeeping and stability evaluations all have a
recombined following the point in the offset, swapping the
large impact on the design. For fishing boats, stability is a primary
remaining row after the offset point and the remaining column
concern and regulations concerning stability are dictated by
below the offset point.
While it is simple to use the offsets directly in the chromo-
some, in order to create hulls that are at least somewhat fair in
shape, without compromising on the use of offsets versus math-
ematical hull shapes, a method was adopted to use both iterative
B-Spline surfaces as described in Gerald and Wheatley (1999),
and a representation of the offsets using offset intervals. The
method transforms the offsets into an array of offset intervals,
with the premise being that a station can be drafted using each
neighbor. The matrix W is then transformed from the offsets at m
stations and n waterlines into differences between offsets. Using
yij as an individual element in the array of offsets
yij þ 1 ¼ yij þ Dy w ij

where Dy is the difference between adjacent offsets; wij is the


chromosome representation of next change in offset and
wij ¼ wl þ ðwu wl Þ dchr
where wl is the lower limit for difference; wu the upper limit for
difference; dchr the decimal value between 0.0 and 1.0 determined
by chromosome.
It should be noted that while for convenience the hull was Fig. 5. Typical Turkish fishing trawler 49 m in length (http://www.maritimesales.
modeled in terms of offsets, so that a table of offsets could be com/EU10.htm).
1060 M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064

regulation societies. As an example in this study, a typical GM Table 1


requirement is assumed of 0.40 m. For prediction of the GM Fishing boat design characteristics.
pertaining to stability an empirically derived formula for fishing
Characteristic Requirement Formulation
boat models can be used (Grubisic, 2001).
The GM formula is given as follows using breadth (B) and GM Minimum GMmax ¼ 0:163e0:742B=D
depth (D): GM¼ 0.40
Depth As relates to VFH Dmin ¼ 0:266L0:77
GMmax ¼ 0:163e0:742B=D Fish hold volume (VFH) VFH ¼95.2 m3 VFH ¼ 0:38 LFH ðB DÞ1:08
Fish hold length As related to VFH LFH ¼ 0:157 L1:26
The minimum depth (Dmin) that can be used is a function of the wl
Depth Related to draft (T) D ¼1.27T
length (L): Waterplane coefficient Cwpl ¼ 0.80 Cwpl ¼waterplane area/(L B)
(Cwpl)
Dmin ¼ 0:266L0:77
Length (L) 10.0r Lr 30.0 m Change in parameter
In addition to the stability requirement, an owner’s require- example 2
Beam (B) 3.0r Br 5.0 m Change in parameter
ment is assumed by specifying a fish-hold volume as used by
example 2
Grubisic (2001). Fish hold volume (VFH) is largely governed by the Draft (T) 1.0r Tr 3.0 m Change in parameter
size of the vessel, and is directly related to the economic value of example 2
the bull. For comparison, a design value is assumed as an example
for a fish hold volume of 95.2 m3 as in the study by Grubisic
(2001). In that model the fish hold is obtained by the following the following conditions and limits are imposed, and are sum-
relation by first obtaining a fish hold length (LFH). This is marized in Table 1 along with the other criteria.
calculated using a correlation formula with respect to the length In order to include these constraints in the optimization
of the waterline (Lwl): process, the penalty method is utilized. The penalty is found per
candidate using a method by Gen and Cheng (2000) according to
LFH ¼ 0:157L1:26
wl the two main design requirements for fish hold volume and GM:
Fish hold volume is then obtained by  
1 DVFH DGM
penalty ¼ 1 þ
VFH ¼ 0:38  LFH ðB  DÞ1:08 2 DVFH max DGM max
One parameter that is not included specifically is the depth of where DVFH is the deviation of the fish hold from the required and
the model. For our purpose, as we are mostly concerned with the DGM is the deviation of GM from the required. For example, for the
underwater portion of the hull, the depth is simply modeled as a fish hold volume:
function of draft (T) according to the following relation:
D ¼ 1:27  T if rVFH ¼ Required VFH VFH
( )
In addition to the minimum fish hold requirement and the GM rVFH , if rVFH 4 0
DVFH ¼
requirement, a desired volumetric displacement can be used, 0, if rVFH o 0
which can also be considered as an owner’s requirement. An
additional design requirement, in the form of a constraint, was The maximum and minimum are from among the population
imposed for a waterplane coefficient of 0.80. This represents a in each generation. The penalty times the performance objective
secondary hull form coefficient, rather than a principal parameter, gives the fitness function for the hull for the particular objective.
which is used as a means of maintaining a workable deck area, For example, for RCI, the fitness function is then given by
but not directly to influence the hull form. If other methods for  
ensuring deck area are used as a constraint then this waterplane RCIRCImin
Fitness ¼ 1  penalty
coefficient restriction does not have to be used. RCImax RCImin
The fishing boat example introduced some restrictions on the
As the previous outline shows, each objective can be tested
hull for the length, beam and draft. Grubisic (2001) used a length
accordingly with each of the design requirements included as a
restriction between 10 and 30 m. In all of the runs in this study, it
constraint. Alternatively if there is the possibility of maximizing
was shown that the length tends towards the maximum limit.
or minimizing a particular design requirement, then these can be
Unless the optimization penalizes length, possibly due to cost or
included as objectives. However additional objectives take addi-
another restriction, which could be based on restrictions accord-
tional computation time whereas using constraints take almost
ing to the type of fishing as imposed by quotas, or port restric-
no additional time at all, therefore, where possible, the use of
tions, as well as by the cost of the vessel, the tendency is for the
constraints should be considered rather than objectives. This does
length to move towards the maximum allowable length. Hence,
not preclude the fact that design requirements could be modeled
should a shorter vessel be required then a limit must be imposed
as objectives.
by the maximum allowable length.
In some cases the displacement was considered as a require-
It should be noted that cost has not been included in this
ment. It is modeled in a similar way as the previous constraints,
study, though this will an overriding consideration in any real
but because of the importance of displacement as a design
optimization problem. For our study, reliable costing data was
requirement, it was given a larger priority. Taking it as a separate
unobtainable for fishing boat construction, as most of the data is
term rather than averaging it together with other constraints
commercial and propriety information. Also in the current meth-
resulted in the following formulation:
odology, imposing arbitrary limits on the hull form using known
coefficients is replaced by a search of the design space driven by   
Dr 1 DVFH DGM
performance indices. Since these secondary coefficients are not penalty ¼ 1 þ
Drmax 2 DVFH max DGM max
used it is prudent to limit the main dimensions so as to
investigate a reasonable design space. In addition to restricting The deviation is calculated in a similar manner for the fish hold
the beam, limits are also imposed on the draft. Besides the design volume or GM requirement and the penalty is again used as
requirements, for the example fishing vessel used in this study multiplication factor of the fitness for each objective.
M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064 1061

4.2. Exploration of genetic algorithm functionality 0.0018


0.00175

Resistance Index
Because using a Genetic Algorithm is a stochastic and heuristic 0.0017 20 Hulls
approach, it is useful to conduct some initial investigations that 0.00165
determine to what extent the algorithm can be used to change 50 Hulls
0.0016 100 Hulls
parameters and produce useful results. ITU fishing hull series
0.00155
have been used mainly to compare the optimal hulls derived with
the well-known resistance characteristics of the original series. As 0.0015
a series, the different hull forms are similar enough in application 0.00145
to test the methodology for the development of an optimal fishing 0.0014
boat. Since none of the hulls are exactly alike, they represent a 0 20 40 60 80 100
good series of offsets to use for the optimization. Generations
Using the ITU 1B hull for comparison, the optimization was
Fig. 8. Comparison of number of hulls for each generation using ITU 1B.
first conducted at a single of Froude number (Fn¼0.28176) to
examine the feasibility of the optimization. As shown in Fig. 6,
length, beam and draft, as well as fish hold volume and displaced
volume, vary for the best or increasingly optimal hull forms. The
optimal stability characteristic is shown in the GM value given in
Fig. 7 is near the required value.
In addition to the optimization of a single hull some investiga-
tion into the GA aspect was conducted. A comparison of the
number of hulls per generation indicates that while according to
GA practice 100 hulls would yield superior results, even 20 hulls
can provide good results as shown in Fig. 8.

4.3. ITU fishing boat with fixed principal parameters

The ITU fishing boat (ITU 148/1B) is optimized with the given
constraints on the length, beam, and draft. It is convenient for
Fig. 9. ITU 148/1-B original and modified hull with fixed principal parameters.
comparison to look at the resulting hull form if the principal
parameters remain constant and the displacement is set as a target
objective in the form of a constraint, as previously described. In this Performance for ITU 1B with Fixed Dimensions
case the only change is the hull form and in the offsets. As the
displacement is set as a constraint, the fish hold constraint can
0.0520
STIX

actually be removed. Two iterations of the B-spline surface are used


to obtain a fair hull, and the maximum variation in the offsets 0.0500
is 790% of each offset interval. Fig. 9 shows the change in the hull
form in which the principal dimensions are fixed for ITU 148/1B using
an initial population of 20 hull variants that are optimized over 100 4.6000
SKI

150 4.3000
L (m) V,FHV (m3)

120 0.0072
Fish Hold Volume
90 Volume
RCI

Length
0.0070
60
30
0 10 20 30 40 50 60
0 Generation
1 3 5 6 9 10 12 15 16 21 37 38 40 48 49
Generations Fig. 10. Stability, Seakeeping and resistance index performance by generation.

Fig. 6. Volume, length and fish hold volume for increasing optimal hulls.
generations. The last optimal hull (in solid lines) is overlaid with the
original hull (in dashed lines) in Fig. 9. The changes in the hull form
6
are not very great, as expected, though some difference in the sections
5 can be seen. The extreme ends of the hull appear to have widened
whereas the mid-ship sections, though nearly the same, have
GM,B,T (m)

4 Beam
narrowed.
Draft
3 The waterline except at the mid-section shows a tendency to
GM
2 narrow. This is probably in response to the objective to minimize
the resistance, which is subsequently made up in the rest of the
1 body by having fuller sections elsewhere. However, since the
0 optimization is not solely a function of resistance, this observa-
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 tion is made on the basis of only one performance index, and may
Generations in fact be subject to other performance factors.
The performance indices for resistance, seakeeping and stabi-
Fig. 7. Maximum GM with beam and draft for increasingly optimal hull forms. lity are plotted in Fig. 10 to show how the results evolve over the
1062 M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064

different generations. The stability index at the top graph starts ITU 1B Heave Motion for Original and Modified
low and increases rapidly, over 60 generations. The seakeeping given Fixed Dimensions
index also starts off high and is lowered over subsequent genera- 0.7
tions as seakeeping motion is minimized. Finally, the resistance Original Heave
0.6
index also starts off high and lowers, quite quickly, becoming Modified Heave
more or less the same after 10 generations and is maintained over 0.5
the course of the next 60 generations. After 60 generations, no

Heave
0.4
further Pareto Optimal hulls were found.
Comparing the actual performance from the evolved hull with 0.3
the original ITU 148/1-B hull, for resistance, the total and wave
0.2
resistance coefficients are shown in Fig. 11. The wave resistance
at higher Froude numbers is lower, from 3.85  10  2 to 0.1
3.28  10  2 at a Froude number of 0.5, which corresponds to a
reduction of 14.8% of the wave resistance. 0
The pitch is somewhat larger at lower Froude numbers but is 0 0.1 0.2 0.3 0.4 0.5 0.6
reduced at higher Froude numbers, as seen in Fig. 12. The heave Froude Number
as shown in Fig. 13 is lower at smaller Froude numbers and is
Fig. 13. ITU 148/1-B original and modified heave motion.
coincidental at larger Froude numbers. The overall result is to
lower the seakeeping index.
Though the stability index increased, these improvements in GZ Curve fro ITU 1b Original and Modified
resistance and seakeeping come with a nominal cost in the GZ with Fixed Dimensions
stability from the original hull form, as dynamic stability as given 0.3
by the area under the GZ curve, shown in Fig. 14, is reduced. The 0.25 Original
stability can vary and a different optimal form having good
Modified
resistance and seakeeping, as well as stability characteristics, 0.2
can also be chosen. For the ITU 1B example the GM of 1.111 m
GZ (m)

was used for the original hull form while a GM of 1.057 m was 0.15
obtained for the modified hull. The KM, which is independent of 0.1
KG, is 2.749 m and 2.65 m for the original and modified hulls,
respectively. 0.05

0
Resistance Coefficients for ITU 1B given Fixed Dimensions 0 10 20 30 40 50 60 70
-0.05
0.045
Original Wave resistance Heel (deg)
0.040 Modified Wave resistance
Resistance Coefficients

0.035 Original Total Fig. 14. GZ curve for original and modified ITU 148/1-B given fixed dimensions.
Modified total
0.030
0.025
0.020
0.015
0.010
0.005
0.000
0 0.1 0.2 0.3 0.4 0.5 0.6
Froude Number

Fig. 11. ITU 148/1-B original and modified total and wave resistance.

ITU 1B Original and Modified Pitch


8 Fig. 15. ITU 148/1-B original and evolved hull form with change in principal
7 parameters.

6
5 4.4. ITU fishing boat with change in principal parameters
Pitch

4
If the principal parameters are allowed to vary according to the
3
limits described previously, some quite different and unusual
2 results occur. Using a fish hold volume requirement of 95.2 m3, as
Original Pitch
1 in the example by Grubisic (2001), and re-running the ITU 148/1-B
Modified Pitch
example yields the optimal hull as shown in Fig. 15. No constraint is
0 set for the actual displaced volume in this particular run. The beam in
0 0.1 0.2 0.3 0.4 0.5 0.6
Froude Number this case is quite wide and the draft quite shallow. The limits in the
main dimensions explored a space with a minimum draft of 1.5 m, a
Fig. 12. ITU 148/1-B original and modified pitch motion. maximum beam of 8 m and a maximum length of 30 m. In trying to
M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064 1063

Fig. 16. RCI optimal performance by generation for ITU 148/1-B with change in Fig. 18. STIX optimal performance by generation for ITU 148/1-B with change in
principal parameters. principal parameters.

Fig. 19. SKI versus RCI optimal for ITU148/1-B with change in principal parameters.
Fig. 17. SKI optimal performance by generation for ITU 148/1-B with change in
principal parameters.

the last Pareto Optimal 67th generation data point plotted


indicating how the optimization is working.

achieve minimum resistance the hull is evolving towards maximum


length, while for stability the hull tends towards the maximum beam. 5. Conclusions
The shallow draft is driven by the minimization of resistance given
that there is no restriction on displacement. The displacement A method for conducting optimization of hull forms was
achieved in this case was only 110 m3. However much that this wide applied to Turkish fishing vessels with the intent of improving
flat hull is notable in Turkish fishing fleets, the results may be the resistance, seakeeping and stability performance for a given
impractical. set of constraints. A fishing boat hull is used as an example of how
As in the previous case, only 100 generations with a popula- hull form optimization can be accomplished using a Multi-
tion of 20 variants was run. Two iterations of the B-spline surface Objective Genetic Algorithm (MOGA). The particular MOGA
are used to obtain a smooth hull and 790% variation in the developed during this study allows automatic selection of a few
offsets interval is allowed. The body plan shown in Fig. 15 for the Pareto Optimal results for examination by the designers while
optimal hull shows a larger beam to achieve a larger vessel to searching the complete Pareto Front. The optimization uses the
match the requirement for the fish hold volume. This results in a three performance indices for resistance, seakeeping and stability
vessel with more displacement but, interestingly, a somewhat to modify the hull shape and obtain optimal hull offsets, as well as
shallower hull. optimal values for the principal parameters of length, beam and
Resistance index trends for optimal RCI versus generation is draft. The modification of the Istanbul Technical University (ITU)
shown in Fig. 16. Large improvements in the RCI is seen in the 148/1-B fishing boat series hull was presented by first fixing the
first 10 generations, with nominal changes after 10 generations principal parameters and allowing the hull offsets to change, and
and no change after 70 generations. Similarly, the SKI perfor- secondly by simultaneously allowing variation of both the princi-
mance by generation is shown in Fig. 17 and indicates improve- pal parameters and the hull offsets. Improvements in all three
ments in the SKI up to 65 generations. The stability index is objectives were found. For further research the methodology can
shown in Fig. 18 and also shows an increase in stability index. be modified to allow for the addition of other performance
Fig. 19 shows one view into the performance for resistance and objectives, such as cost or specific mission objectives, as well as
seakeeping as they tend towards their respective minimums, with the use of enhanced performance prediction solvers. In addition,
1064 M.A. Gammon / Ocean Engineering 38 (2011) 1054–1064

one or more hulls could be evaluated by experiment to validate Gen, M., Cheng, R., 2000. Genetic Algorithms and Engineering Optimization.
the results of using this particular optimization approach. Wiley-Interscience, New York.
Gerald, G.F., Wheatley, P.O., 1999. Applied Numerical Analysis, 6th edition
Addison-Wesley Longman, Reading, Massachusetts.
References Goldberg, D.E., 1989. Genetic Algorithms in Search, Optimization and Machine
Learning. Addison-Wesley Longman, Reading, Massachusetts.
Grubisic, I., 2001. Improvement of the fishing vessel concept design model. In:
Alkan, A.D., 2004. The Turkish fishing fleet: technical features and design Proceedings of the 7th Symposium on Techniques and Technology of Fishing
evaluations. Brodogradnja 52 (1), 21–29. Vessels, Ancona, Italy.
Coello Coello, C.A., 1996. An Empirical Study of Evolutionary Techniques for Kafalı, K., S- aylan, Ö., S- alcı, A., 1979. Development of hull forms of fishing boats
Multiobjective Optimization in Engineering Design. Ph.D. Thesis. Tulane suitable for Turkish Waters (in Turkish). TÜB_ITAK (The Scientific and Techno-
University, New Orleans, Louisiana. logical Research Council of Turkey), Project Number G-416, Istanbul, Turkey.
Danisman, D.B., Goren, O., Insel, M., Atlar, M., 2001. An optimization study for the Larrson, L., Eliasson, R., 2000. Principles of Yacht Design, London. Adlard Nautical.
bow form of high speed displacement catamarans. Marine Technology 38 (2). Maisonneuve, J.J., Harries, S., Marzi, J., Raven, H.C., Viviani, U., Piippo, H., 2003.
Day, A.H., Doctors, L.J., 1997. Resistance optimization of displacement vessels on Towards optimal design of ship hull shapes. In: Proceedings of the 8th
the basis of principal parameters. Journal of Ship Research 48 (4), 249–259. International Marine Design Conference, National Technical University of
Dejhalla, R., Mrsa, Z., Vukovic, S., 2002. A genetic algorithm approach to minimum Athens, Athens, vol. 2, pp. 31–42.
ship wave resistance problem. Marine Technology 39 (3), 187–195. Schaffer, J.D., 1985. Multiple objective optimization with vector evaluated genetic
Gammon, M.A., 1990. Modifying Thin-Ship Wave Resistance Computation for algorithms. In: Genetic Algorithms and their Applications: Proceedings of the First
Transom Stern Ships. Master of Applied Science Thesis. Dalhousie University, International Conference on Genetic algorithms, Pittsburgh, PA, pp. 93–100.
Halifax, Canada. Yasukawa, H., 2000. Ship form improvement using genetic algorithm. Ship
Gammon, M.A., Alkan, A.D., 2001. A resistance study of low L/B vessels with Technology Research 47, 35–44.
Transoms. In: Proceedings of the Techniques and Technology of Fishing Yilmaz, H., 1999. Effect of form parameters on stability of fishing vessels. In:
Vessels Conference, Ancona, Italy. Proceedings of the 7th Symposium on Techniques and Technology in Fishing
Gammon, M.A., 2004. Ship Hull Form Optimization by Evolutionary Algorithm. Vessels, Ancona, Italy.
Ph.D. Thesis. Yildiz Technical University, Istanbul, Turkey. Zaraphonitis, G., Papanikolaou, A., Mourkoyannis, D., 2003. Hull form optimization
Gammon, M.A., Yilmaz, H., 2003. Stability evaluation for fishing boat hull of high speed vessels with respect to wash and powering. In: Proceedings of
optimization. In: Proceedings of the Techniques and Technology of Fishing the 8th International Marine Design Conference, vol. 2. National Technical
Vessels Conference, Ancona, Italy. University of Athens, Athens.

You might also like