Download as pdf or txt
Download as pdf or txt
You are on page 1of 12

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/233498605

Optimal Compartment Layout Design for a Naval Ship Using an Improved


Genetic Algorithm

Article  in  Marine technology · July 2002

CITATIONS READS

19 401

3 authors, including:

Myung-Il Roh
Seoul National University
188 PUBLICATIONS   932 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Myung-Il Roh on 25 February 2015.

The user has requested enhancement of the downloaded file.


© Marine Technology, Vol. 39, No. 3, July 2002, pp. 159–169

Optimal Compartment Layout Design for a Naval Ship Using an


Improved Genetic Algorithm
Kyu-Yeul Lee,1 Seong-Nam Han,2 and Myung-II Roh3

With the trend in modern naval ships towards less dense payloads, space layout design has become more
important. Recent advances in computing science and increased understanding of methods for developing
mathematical models, which form the basis of the space layout design, have helped with the development
of a powerful design procedure. In this study, the compartment layout problem, which can be regarded as
the space layout design of a naval ship, is represented as a mathematical model, and a compartment
layout algorithm based on the genetic algorithm (GA) in order to solve the problem is proposed. Com-
parative testing shows that the proposed algorithm performs better than other existing algorithms for the
optimal compartment layout design. Finally, the proposed algorithm is applied to the compartment layout
problem of a naval ship and the computational results are compared with the actual compartment layout
of the naval ship.

Introduction wholly on the ability of a designer to generate alternative


layouts. Carlson & Fireman (1987) used a computer-based
THE payload density of naval ships has recently been de- design system called a general arrangement design system
creasing while the payload specific volume has been increas- (GADS) to perform the arrangement design of the naval ship.
ing. This trend is caused by the drive for increased perfor- Since the optimization module has not been included in the
mance of the ship by using electronic equipment, which has system, the layout generated from the system was feasible,
lower density, when compared with guns from the past. This but not a best solution.
change of focus has also been driven by associated demands Alternatively, facility layout problems (FLP) concerning
for maintenance and operational spaces. This trend has lim- space layout optimization have been investigated in depth by
ited the volume available on a naval ship, and has meant that researchers in various fields, such as industrial engineering,
the volume required for a naval ship’s electronic equipment management science, and architecture. Layout design inves-
has become larger than the minimum volume necessary to tigations are assisted with recent advances in computing sci-
float the ship. However, the required volume for the elec- ence and also with increased understanding of methods used
tronic equipment was limited since the size and building cost for developing mathematical models. The FLP, which has
of a naval ship increases as more volume is added. This been investigated in various fields, can be regarded as a com-
means that space layout design has become a much more partment layout problem of a naval ship.
important consideration for efficient allocation of all com- The FLP is a common industrial problem of allocating fa-
partments, including the electronic equipment, in the limited cilities to either maximize the adjacency requirement (Sep-
volume. panen 1970) or minimize the cost of transporting materials
Several approaches have been suggested by many re- between them (Koopmans 1957). These facilities can be re-
searchers in studies of optimal space layout design for a na- garded as compartments in a naval ship. Several facility
val ship. Hope (1981) and Andrews (1981,1985) suggested (“compartment”) layout algorithms to solve the FLP have
approaches aimed at improving design synthesis of arrange- been developed by various researchers (Tam 1992, Islier
ment design of the naval ship; however, they did not receive 1998). However, in these algorithms, structure, number,
much attention. Brown (1986) insisted that an attempt, us- width, and location of passages were not an important con-
ing several techniques proposed in his study, should be made sideration in solving the FLP. Also the inner structure walls,
to evaluate each layout, using a quantifiable approach, while which can be regarded as watertight transverse bulkheads of
searching for the best arrangement. Cort & Hills (1987) pro- the naval ship, were not completely considered. These algo-
posed an optimal design procedure which utilizes fuzzy rithms were developed to handle the FLP in which the pas-
theory. In this approach a number of alternative layouts that sage and inner structure walls were not explicitly considered
satisfy design requirements are generated manually or by by the designer for an optimal layout of facilities in a factory
using an automated design procedure, and then an optimal or a shop. Moreover, these algorithms could handle the FLP
layout among those alternative layouts is found by using in which a boundary shape of an available area is rectangular
fuzzy theory. However, all of the possible alternative layouts so these algorithms could not be directly applied to the com-
cannot be considered because this design methodology relies partment layout problem of the naval ship, in which the in-
ner structure walls (“watertight transverse bulkheads”) and
1 the passage should be considered and the boundary shape of
Professor, Department of Naval Architecture & Ocean Engineer- the available area is curved. A new algorithm that can solve
ing and Research Institute of Marine System Engineering, Seoul
National University, Korea.
this problem is required. In this study, a compartment layout
2
Visiting research officer, Korean Navy and Seoul National Uni- algorithm is proposed based on the genetic algorithm (GA) to
versity, Korea. solve the compartment layout problem of a naval ship. Fi-
3
Ph.D. candidate, Department of Naval Architecture & Ocean En- nally, the proposed algorithm is applied to the compartment
gineering, Seoul National University, Korea. layout problem of a naval ship and the computational results

JULY 2002 0025-3316/02/3903-0159$00.47/0 MARINE TECHNOLOGY 159


are compared with the actual compartment layout of the na- Subject to
val ship after efficiency of the proposed algorithm is evalu-
ated. g1 = a m in
k - a k # 0 (3)
g2 = a k- a m ax
k # 0 (4)
Optimal compartment layout problem m in
g3 = a k - ak # 0 (5)
Compartment layout problem having watertight m ax
g 4 = ak - a # 0 (6)
transverse bulkheads and passages k

M
The compartment layout problem in this study is con-
cerned with finding the best compartment layout. The design
g5 = (
k=1
ak - Aavailable # 0 (7)
objective is to minimize the total cost of transporting mate-
rials and to maximize the adjacency requirement between g 6 = ~ xri - xw
s
.t.b
! ~ xw
s
.t.b
- xj! #
l
0 (8)
compartments while at the same time satisfying the con-
k = 1, … , M and s = 1, … , P
straints of areas, aspect ratios of the compartments, and wa-
tertight transverse bulkheads and passages. Finding the best where M is the number of compartments, fi,j is the material
compartment layout means determining sequence and areas flow between compartments i and j, di,j is the distance be-
of the compartments to be allocated, and the location of pas- tween centroids of compartments i and j, and bij is the adja-
sages. In Fig. 1 it is assumed, for convenience sake, that the cency factor which represents the adjacency ratio between
boundary shape of the available area is rectangular. Input compartments i and j. The adjacency factor bi,j is determined
data given by a designer are: from distance di,j between compartments i and j as follows.
1. Number of compartments to be allocated to the avail- bi,j = 1.0 ; if 0 < di,j # dm a x / 6
able area. bi,j = 0.8 ; if dm a x / 6 < di,j # dm ax / 3
2. Available area and its boundary shape. bi,j = 0.6 ; if dm a x / 3 < di,j # dm ax / 2
3. Upper and lower bounds of the required area for each bi,j = 0.4 ; if dm a x / 2 < di,j # 2dm a x / 3
compartment. bi,j = 0.2 ; if 2dm a x / 3 < di,j # 5dm a x / 6
4. Upper and lower bounds of the required aspect ratio for bi,j = 0.0 ; if 5dm ax / 6 < di,j # dm a x
each compartment.
5. Material flows between compartments. where dm ax is the maximum distance between the compart-
6. Adjacency values between compartments. ments.
7. Number and positions of watertight transverse bulk- ci,j is the adjacency value (0–5) between compartments i
heads. and j. The adjacency value ci,j between compartments is a
8. Number and widths of each vertical and horizontal pas- functional relationship which is not always quantifiable and
sage. is sometimes vague and difficult to define. In fact, the opti-
9. Upper and lower bounds of the position of each vertical mization result can vary depending on this value. However,
and horizontal passage. in this study, the quantifiable value for ci,j proposed by Lee
(1988) is used as shown below.
Formulation of compartment layout problem
ci,j = 0 ; it is undesirable for compartments i and j to be
The problem described in the above section is mathemati- located close together
cally formulated as follows: ci,j = 1 ; it is unimportant for compartments i and j to be
located close together
M- 1 M
ci,j = 2 ; it is ordinary for compartments i and j to be
Minimize F1 = ( (
i= 1 j=i+1
~ fi,j 2 di,j! ; total transport cost
ci,j = 3
located close together
; it is important for compartments i and j to be
(1) located close together
and ci,j = 4 ; it is especially important for compartments i
and j to be located close together
M- 1 M
ci,j = 5 ; it is absolutely necessary for compartments i
Maximize F2 = ( (
i=1 j=i+ 1
~ bi,j 2 ci,j! ; adjacency requirement and j to be located close together
(2) a k is the aspect ratio of the compartment k, a m k
in
and a m
k
ax

are the lower and upper bounds of the aspect ratio a k , ak is


the assigned area of the compartment k, am k
in
and am k
ax
are
the lower and upper bounds of the assigned area ak , Aavailable
is the available area, P is the number of the watertight trans-
verse bulkheads, xw s
.t.b
is the position (x-coordinate) of the
watertight transverse bulkhead, and xri and xlj are the x-
coordinates of the right boundary of the compartment i and
the left boundary of the compartment j.
In the above formulas, equations (3) and (4) show that the
aspect ratio a k of the compartment k should vary between the
prescribed upper (a m k
in
) and lower bounds (a m k
ax
), equations
(5) and (6) show that the area ak of the compartment k should
vary between the prescribed upper (am k
in
) and lower bounds
ma x
(ak ). Equation (7) shows that the compartments total area
should be less than the available area Aavailable . Equation (8)
Fig. 1 An example of best layout of compartments (1–8) having watertight trans- shows there should be no interference between the water-
verse bulkheads and passages. tight transverse bulkheads. These constraints are used to

160 JULY 2002 MARINE TECHNOLOGY


ensure that a compartment is not partitioned into two sub- their facility layout design. Details about the GA can be
compartments. found in many references (Goldberg 1989, Davis 1991). A
Since this problem described by equations (1) to (8) has two scheme of the proposed compartment layout algorithm based
objective functions (F1 and F2 ), it is a multi-objective optimi- on the GA in this study is shown in Fig. 2.
zation problem but can be converted to a single-objective op-
timization problem by using the weighting method (Cohon Representation of compartment layout
1978). Applying weight factors (w1 and w 2 ) to the objective
Compartment layout can be represented as a chromosome
functions (F1 and F2 ) in equations (1) and (2) results in the
by an encoding process. The chromosome can be represented
following form.
as the compartment layout be a decoding process. In this
Minimize F = w 1 ? F1 + w2 ? F2 study, a method to model the compartment layout in a four-
M- 1 M segmented chromosome, including positions of passages, is
= w1 ? ( (
i= 1 j=i+ 1
~ fi,j ´ di,j! proposed and is shown in Fig. 3 The first segment of the
chromosome represents a sequence of compartments to be
M- 1 M
allocated. The second segment represents areas of the com-
+ w2 ? ( (
i= 1 j=i+ 1
~ C - bi, j ´ ci, j! (9) partments. The areas of the compartments are given in the
same order as the first segment and are allowed to vary be-
where C is the maximum adjacency rating (say 5), and w1 tween their upper and lower bounds. The third and fourth
and w2 are the weight factors which are equivalent to the segments, respectively, represent positions of the horizontal
tradeoff between the total cost of transporting materials (F1 ) and vertical passages. The positions of the passages are al-
and the adjacency requirement (F2 ). Therefore, the ratios w1 lowed to vary between their upper and lower bounds. Figure
and w2 can vary the optimization result. In this study, the 3 shows an example of the compartment layout together with
suitable values for w1 and w2 (w1 4 1.0 and w2 4 50.0) were the corresponding representation of the four-segmented chro-
chosen so that F1 and F2 , the objective functions, have the mosome.
same effect on the optimization result.
Using a penalty function method, the constrained optimi- Distance calculation method between compartments
zation problem, as defined in equations (3) to (9), can be In the existing compartment layout algorithms a rectilin-
converted to an unconstrained optimization problem stat- ear distance method has been used to calculate the distance
ed as: between compartments di,j. This is included in equation (10)
M- 1 M M- 1 M and needed to determine the adjacency factor bi,j in equation
Minimize F8 = w1 ? ( (
i=1 j= i+1
~ fi, j ´ di, j! + w2 ? ( (
i=1 j= i+ 1
(10). In this method, the distance di,j is calculated by sum-
ming the vertical and horizontal distances between the com-
6 partments, as shown below.
~ C - bi, j ´ ci, j! + (
u= 1
$ Ru ´ max ~ gu, 0 ! %
di,j 4 |xi ­ xj| + |yi ­ yj| (11)
(10) The above method cannot be applied directly to a compart-
where Ru are penalty coefficients. ment layout problem having passages, because materials are
to be moved via the passages. The problem is the method can
miscalculate the actual distance between compartments. In
Proposed algorithm for compartment the present study a new distance calculation method was
layout problem introduced, to determine the distance between compart-
ments, by using graph theory. In this new distance calcula-
Overview of proposed algorithm tion method all relationships between compartments and

The algorithm proposed in the present study is based on


the genetic algorithm (GA), which was first proposed by Hol-
land (1975). The GA is classified as an evolutionary search
and optimization technique which considers the design pro-
cess an evolutionary one. The GA attempts to find the best
solution by generating a collection (“population”) of potential
solutions (“individuals”). Through selection, crossover, and
mutation operations, more accurate solutions are hoped to be
generated from the current set of potential solutions. This
iteration continues until the algorithm finds an acceptably
good solution. The GA has many advantages over other op-
timization techniques, including the ability to deal with
qualitatively different types of domains, such as continuous
variable domains, discrete or quantized variable domains, or
mixed-type variable domains. The classical approach con-
verts all design variables of the problem to the domain of
binary integers and encodes each individual in the popula-
tion as a binary string (e.g., 010010). The crossover and mu-
tation operations then manipulate the bits in the binary
strings (0 or 1). The GA has been successful in solving nu-
merous science and engineering problems, such as traveling
salesman problems (TSP) (Chatterjee 1996), transport prob-
lems (Wren 1995), multi-objective optimization problems
(Kim 1994), and ship structural optimization problems (Lee Fig. 2 Scheme of proposed algorithm for compartment layout problem having
2001). And, currently many researchers are using the GA in watertight transverse bulkheads and passages

JULY 2002 MARINE TECHNOLOGY 161


Fig. 3 Example of compartment layout and corresponding representation of 4-segmented chromosome

passages are first represented as an adjacency graph. Then to solve efficiently the compartment layout problems that
the shortest path between compartments i and j, and the have the watertight transverse bulkheads and passages.
distance di,j between these compartments, is determined us- These operations were incorporated into the proposed algo-
ing Dijkstra’s algorithm of the graph theory (McHugh 1990). rithm. Additionally, a modification operation was included in
Figure 4 shows all paths between compartments 1 and 10, the crossover operation and a refinement operation was
and the adjacency graph among compartment 1, 10, and the added to handle void spaces. These “void spaces” are gener-
passages to calculate the distance between compartments 1 ated while converting a chromosome into a compartment lay-
and 10. out in the decoding process.
Selection operation—The selection operation is a process
Improved genetic operations performed to select two individuals (“parents”), from the cur-
rent population, for the next genetic operation. Here, a pro-
As mentioned above, the proposed compartment layout al-
portionate selection method is employed, as it is the most
gorithm is based on the GA. In the GA, four genetic opera-
popular of the stochastic selection methods and it is some-
tions known as selection, crossover, inversion, and mutation
times called a roulette-wheel selection method. In the pro-
are typically used to generate new individuals (“children”). In
portionate selection method an individual is selected based
the present study, these operations of the GA were improved
on selection probability pselection (i). This is shown in equation
(12) where, Ft(i) is a fitness value of an ith individual.
Ft~ i !
pselection~ i ! = (12)
( i
Ft~ i!

The probability of selecting the individual from the current


population is purely a function of the individual’s relative
fitness. A fitness function (Ft) of the individual can be for-
mulated as follows:
1
Ft = - F8 or Ft = ~ if F8 > 0 ! (13)
F8
Here, the second formula of equation (13) is used. By this
selection method, individuals having higher fitness will par-
ticipate in the creation of the next generation more often
than those having lower fitness.
Crossover operation—The crossover operation is a process
performed to generate new individuals (“children”) from two
individuals (“parents”) selected by the selection operation.
Two crossover operations, a modified crossover and a one-
point crossover, are used in this study. The modification op-
eration is applied to the children generated from these cross-
over operations to satisfy integrity in the third and fourth
segments.
Modified crossover operation—The modified crossover op-
eration is based on the assumption that the individual hav-
Fig. 4 All paths between compartments 1 and 10, and adjacency graph to cal- ing higher fitness value between the parents should endow
culate distance more genes to the child. This operation is simultaneously

162 JULY 2002 MARINE TECHNOLOGY


applied to the first and second segments of each parent. Ini-
tially s1 positions in the first and second segments of the first
parent are randomly selected. This is to generate the first
and second segments of the first child. The value of s1 rep-
resents the number of genes of the first parent to be replaced
with those of the second parent. Then s1 is determined on the
basis of the first parent’s fitness when compared with the
second parent’s fitness. This is as shown in equation (14).
$ Ft~ p1 ! + Ft~ p2 ! % Ft~ p1!
-
s1 =
Ft~ p1 ! + Ft~ p2!
´ n ~ discard decimals ! , s2 = n - s1 (14) Fig. 6 An example of one-point crossover operation applied to third and fourth
segments of parents
In equation (14) Ft(p1) is the fitness of the first parent.
Ft(p2) is the fitness of the second parent, s2 is the number of
genes of the first parent to be transmitted to the first child, over operation to satisfy integrity in the third and fourth
and n is the number of the genes in the first and second segments of the children. The integrity represents that sums
segments. From equation (14) the parent having higher fit- of the genes in the third and fourth segments of each parent
ness has a smaller s1 value than the other parent due to the should consist with the height (12) and the breadth (20) of the
assumption mentioned above. The next step in this operation available area shown in Fig. 3. The modification operation
if for the genes in the s2 positions of the first parent to be satisfies integrity through modifying the last gene of the
transmitted to the corresponding positions of the first child. third and fourth segments. Figure 7 shows an example of the
Finally, the genes in the s1 positions are reordered according modification operation applied to the third and fourth seg-
to the order of the corresponding genes in the second parent ments to satisfy the integrity in the third and fourth seg-
and then they are transmitted to the corresponding positions ments of the first and second children.
of the first child. These similar steps are applied to the sec- Inversion operation—The inversion operation, which can
ond parent to also generate the first and second segments of be considered as self-crossing, is used to increase population
the second child. Figure 5 shows an example of the modified diversity like the inversion operation. The inversion opera-
crossover operation applied to the first and second segments tion is simultaneously applied to the first and second seg-
for generating the first and second children. ments of the first child generated from the crossover opera-
One-point crossover operation—The one-point crossover tion. Two genes in the first and second segments of the first
operation is simultaneously applied to the third and fourth child are randomly selected and are exchanged with each
segments of each parent. A split line is randomly determined other. Figure 8 shows an example of the inversion operation
in these segments, and then genes behind the split line are applied to the first and second segments of the first child.
exchanged between the parents. Figure 6 shows an example Mutation operation—The mutation operation is applied to
of the one-point crossover operation applied to the third and the second segment of the second child generated from the
fourth segments to generate the third and fourth segments of crossover operation. This normally causes random changes.
the first and second children. In this operation two genes in the second segment of the
Modification operation—As mentioned in the section “Rep- second child are randomly selected and a difference value is
resentation of compartment layout,” the third and fourth seg- also randomly determined. The difference value is then
ments of the children, respectively, represent positions of the added to the first gene and at the same time, subtracted from
horizontal and vertical passages. The sums of the genes in the second gene. Figure 9 shows an example of the mutation
the third and fourth segments are, respectively, equal to the operation applied to the second segment of the second child.
height and breadth of the available area. The modification Refinement operation—The refinement operation was
operation is applied to the children generated from the cross- added in the proposed algorithm to handle void spaces. Void
spaces are generated while converting a chromosome into a
compartment layout during the decoding process. When new
children generated from the operations mentioned above are
converted into the compartment layout by the decoding proc-
ess, a number of void spaces are generated as shown in Fig.
10. This is caused by avoiding interferences between the com-
partments and the watertight transverse bulkheads when
the compartments are allocated according to the sequence of
the compartments in the first segment of the chromosome.

Fig. 5 An example of modified crossover operation applied to first and second Fig. 7 An example of modification operation applied to third and fourth segments
segments of parents of children

JULY 2002 MARINE TECHNOLOGY 163


Fig. 8 An example of inversion operation applied to first and second segments of
first child

Fig. 9 An example of mutation operation applied to second segment of second Fig. 11 Example of layout of compartments and corresponding three-segmented
child chromosome of Islier’s algorithm

objective function values and the ratio of computation times


are shown. From Table 1 it can be seen that the proposed
algorithm is superior to Islier’s algorithm as there are better
values of best and mean objective functions produced from
the proposed algorithm, and it required less computation
time.
Figure 12 shows final layouts obtained from the proposed
and Islier’s algorithms for the case of 20 compartments.

Example application of proposed algorithm


Development of a ship’s arrangement is one of the most
Fig. 10 Compartment layouts before and after refinement operation for first child critical aspects of a ship’s design and is an integral part of the
design development, during each phase of design, due to of
three significant factors (Carlson 1987). First, there is in-
Here the refinement operation is performed to eliminate the
creased emphasis on integrating the individual functions of a
void spaces and for efficient utilization of the available area.
ship to form an integrated ship system. Second, most naval
Figure 10 shows the results of the compartment layout before
ships are space critical, their size and resulting cost being
the refinement operation including void spaces and after the
governed by spatial requirements. Third, the ship arrange-
refinement operation for the first child.
ment is a necessary input to most other ship design tasks.
These three factors lead to close management and design
Comparative test of proposed algorithm with emphasis on the ship’s arrangement. As the design develops,
an existing algorithm space is clearly the element in the design process that can be
most effectively controlled. Space allocation is a simple con-
To evaluate the efficiency of the proposed algorithm (par- trol mechanism accepted by members of the design team.
ticularly the improved genetic operations) a comparative test With the trend in modern naval ships towards less dense
was performed between it and Islier’s algorithm (1998). For a payloads, space layout design has become more important.
more accurate comparison, the objective function, the chro- Recent advances in computing science and increased under-
mosome structure, and the representation method of the standing of methods for developing mathematical models,
compartment layout of the proposed algorithm were modified which form the basis of space layout design, have helped with
by being made equal to those of Islier’s algorithm. The com- the development of a powerful design procedure. In the
partment layout of Islier’s algorithm is represented as a present study, the proposed algorithm was applied to the
three-segmented chromosome shown in Fig. 11(c). Figure compartment layout problems, which can be regarded as
11(a) shows the placement procedure of the compartment. space layout design, of a naval ship (FF-21 multi-mission
The compartment layout of Islier’s algorithm is constructed frigate (Afanasieff 1994)). The proposed algorithm was ap-
not on the available area shown in Fig. 3 but on the grid- plied to compartment layouts of an aftbody having the avail-
based available area. This available area is divided into unit able area of a rectangular boundary shape. The forebody was
area squares or rectangles by a grid, as shown in Fig. 11(b). considered as having the available area of a curved boundary
For this case the compartments can have polygonal as well as shape on the second deck of the FF-21, as is shown in Fig. 13.
rectangular shapes, as is shown in Fig. 11(b), as each com- The results from running the proposed algorithm were com-
partment is allocated along the grid. Figure 11 shows an pared with actual compartment layout (aftbody and fore-
example of the representation method of the compartment body) on the second deck of the FF-21. The compartment
layout of Islier’s algorithm. layout on the second deck of the FF-21 and the regions of the
The testing was performed for several cases in which the aftbody and the forebody to which the proposed algorithm
compartments number varies from 8, 12, 16, and 20, and the was applied are shown in Fig. 13.
maximum number of generations being 500 with a popula- Compartment layout problem of aftbody on second
tion size of 500. Both algorithms were run ten times for each deck of FF-21
case in the Pentium III system (667MHz, 256MB RAM) and
the results obtained are summarized in Table 1 where the The actual compartment layout of the aftbody on the sec-
best among ten objective function values, the mean of ten ond deck (frame no. 68–92) of the FF-21 is shown in Fig. 14

164 JULY 2002 MARINE TECHNOLOGY


Table 1 Comparison of computational results of proposed and Islier’s algorithms for various com-
partment layout cases (OFV: objective function value)

Fig. 13 Compartment layout plan of second deck of FF-21

and it has the available area of a rectangular boundary shape room”). Therefore, if only the total transport cost F1 in equa-
with 20 compartments, two watertight transverse bulkheads, tion (1), an objective function, is considered, optimization can
two horizontal passages, and two vertical passages. generate a compartment layout in which these compart-
Optimization by the proposed algorithm was performed ments having relatively low material flows are separately
100 times on a Pentium III system (667 MHz, 256 MB RAM). located. However, it is generally understood that these com-
The optimization result and the best compartment layout partments should be adjacent to each other. Thus, the adja-
obtained from 100 runs are shown in Table 2 and Fig. 14, cency requirement F2 in equation (2) should be additionally
respectively. Also the actual compartment layout of the aft- considered as the objective function in order to obtain a rea-
body and the convergence history of the objective function sonable, optimal compartment layout. For verification, very
value during iteration are also shown in Fig. 14. Table 2 small material flows fi, j between the electronic equipment
shows the best, mean, and standard deviation of the objective rooms (“compartments 9 and 15”), and among the ship office
function values found by 100 runs. (“compartment 11”) and the officer rooms (“compartments 17,
To solve this problem, assumed input data for material 18, and 19”) are assigned as input data since material move-
flow fi,j and the adjacency value ci,j were used for optimiza- ments are hardly observed between them. Of the adjacency
tion. Material movement was hardly observed between com- conditions, the conditions that the compartments 9 and 15
partments 9 (“electronic equipment room”) and 15 (“elec- should be adjacent to each other (i.e., c9 ,15 4 5), and the
tronic equipment room”), and among compartments 11 (“ship compartments 11, 17, 18, and 19 should be adjacent to one
office”), 17 (“officer room”), 18 (“officer room”), and 19 (“officer another (i.e., c11,17 4 c11,18 4 . . . 4 c18,19 4 5) were also

JULY 2002 MARINE TECHNOLOGY 165


Fig. 14 Actual and computed compartment layouts of aftbody on second deck of FF-21 and convergence history
of objective function value during iteration

Table 2 Optimization result by proposed algorithm for 100 runs (OFV: objective function value)

given. The optimized compartment layout, generated by the Compartment layout problem of forebody on second
proposed algorithm, satisfies the adjacency conditions, as deck of FF-21
shown in Fig. 14. In Fig. 14, two electronic equipment rooms
were adjacent to each other and the ship office and the three
The actual compartment layout of the forebody on the sec-
officer rooms were also adjacent to one another. The proposed
algorithm considered both the total transport cost and the ond deck (frame no. 17–44) of the FF-21 is shown in Fig. 15
adjacency requirement in its generation of the optimal facil- and it has the available area of a curved boundary shape, 20
ity layout. compartments, two watertight transverse bulkheads, two
However, the result obtained from the proposed algorithm horizontal passages, and one vertical passage.
is somewhat different from the actual compartment layout. Optimization by the proposed algorithm was performed
Of course, the actual compartment layout shown in Fig. 14 100 times on a Pentium III system (667 MHz, 256 MB RAM).
may not be an optimal compartment layout of the second The optimization result and the best compartment layout
deck of the FF-21. Here, however, the actual compartment obtained from 100 runs are shown in Table 3 and Fig. 15,
layout was used only for comparison purpose with the results respectively. The actual compartment layout of the forebody
by the proposed algorithm. This difference is considered to and the convergence history of the objective function value
arise from the input data assumptions made for material during iteration are also shown in Fig. 15. Table 3 shows the
flows and the adjacency values between the compartments in best, mean, and standard deviation of the objective function
the proposed algorithm, because no quantifiable data for values found by 100 runs. The adjacency conditions used to
them were available on the second deck of the FF-21. The solve this problem were as follows: compartments 1 (“scul-
weight factors (w1 and w2 ) may have had a role in the ap- lery”), 2 (“food service office”), and 3 (“dry provisions room”)
parent difference; they are not always quantifiable, vague should be adjacent to one another (i.e., c1,2 4 c1,3 4 . . . 4
and hard to define, were assumed to be w 1 4 1.0 and w2 4 c2,3, 4 5); compartments 9 (“crew messroom and lounge”) and
50.0, and used for optimization. In fact, the various optimal 10 (“chief petty officer messroom and lounge”) should be ad-
compartment layouts could be obtained by varying the jacent to each other (i.e., c9 ,10 4 5); and compartments 13
weight factors that reflect the influence of F1 (“total trans- (“galley”), 14 (“refrigerating machine room”), 15 (“chill store
port cost”) and F2 (“adjacency requirement”) on the optimi- room”), and 16 (“freeze store room”) should be adjacent to one
zation result. The compartment layout design of an actual another (i.e., c13,14 4 c13,15 4 . . . 4 c1 5,16 4 5). The com-
ship is developed from the designers’ experiences. If data partment layout satisfying the adjacency conditions was ob-
about material flows and adjacency values can be quantita- tained as shown in Fig. 15, where the scullery, food service
tively determined, and the weight factors more precisely de- office, and dry provisions room are adjacent to one another;
fined, then the proposed compartment layout algorithm the crew messroom and lounge, and the chief petty officer
would be able to yield a satisfactory compartment layout of messroom and lounge, are adjacent to each other; and the
an actual ship. That is, the proposed algorithm would be able galley, refrigerating machine room, and chill and freeze store
to yield better results than the actual compartment layout rooms are also adjacent to one another. The optimal facility
shown in Fig. 14. layout, generated by the proposed algorithm, incorporates

166 JULY 2002 MARINE TECHNOLOGY


Fig. 15 Actual and computed compartment layouts of forebody on second deck of FF-21 and convergence history
of objective function value during iteration
Table 3 Optimization result by proposed algorithm for 100 runs (OFV: objective function value)

both the total transport cost consideration and the adjacency using offset data in text file format to solve the FF-21 com-
requirement. partment layout problem having an available area of a
However, the layout obtained from the proposed algorithm curved boundary shape. When each compartment is allocated
is also somewhat different from the actual compartment lay- according to the representation method of the compartment
out. The reasons for this difference are similar to those of the layout mentioned in the section “Representation of compart-
compartment layout problem of the aftbody mentioned in the ment layout,” then the x-coordinate of the right boundary (xr )
foregoing section. In the next section, the representation of of the compartment is determined by using the x-coordinate
the hull form and calculation of the position of the compart- of the left boundary (xl) and the assigned area (a) of the
ment will be described. compartment. To do this an area curve with respect to the
Representation of hull form and calculation of x-coordinate of the right boundary (xr ) of the compartment
area/position of compartment should be generated first. Then the x-coordinate of the right
boundary (xr ) of the compartment, having the arbitrary as-
In this study, the hull form of the FF-21 was represented signed area, is determined using the curve-plane intersec-
with the NURBS (Non Uniform Rational B-Spline) curve by tion. Figure 16 shows the procedure followed to determine

Fig. 16 Scheme of representation of hull form and calculation of right boundary x-coordinate
(xr) for compartment with arbitrary area

JULY 2002 MARINE TECHNOLOGY 167


Fig. 17 Chromosome structure of proposed algorithm for multideck layout problems

the x-coordinate of the right boundary (x19 r ) of compartment path between compartments i and j, and the distance di,j
No. 19 when the x-coordinate of the left boundary (x19 l ) and between these compartments, is determined using the Dijk-
the assigned area (a19 ) of compartment No. 19 are given. stra’s algorithm of the graph theory. Figure 18 shows the
distance calculations in the single-deck and multideck layout
Expansion of proposed algorithm to multideck problems.
layout problems Modification of the genetic operations—Since the chromo-
some structure must be modified to treat the three-
The proposed algorithm in this study was applied only to dimensional multideck layout problems as mentioned above,
two-dimensional ship compartment layout problems (“single- the genetic operations such as crossover, mutation, and re-
deck layout problems”) to show its applicability to three- finement, which are applied to the chromosomes, must also
dimensional ship compartment layout problems. However, a be modified. This can be done by expanding the genetic op-
compartment layout problem of a naval ship is a three- erations for two-dimensional compartment layout problems
dimensional multideck layout problem. That is, ship com- of the proposed algorithm because the chromosome structure
partments have functional relations with other compart- to treat the multideck layout problems is composed of several
ments not only horizontally but also vertically. And, support four-segmented chromosomes as shown in Fig. 17. Figure 19
spaces on different deck levels such as vertical ladders must shows an example of a modified crossover operation to treat
be considered to minimize the movement of personnel as well the multideck layout problems.
as material. The proposed algorithm can be expanded to ap-
ply to more realistic multideck layout problems of three-
dimensional ship compartment layout problems with some
modifications as follows.
Modification of the four-segmented chromosome to repre-
sent the compartment layout—The chromosome structure is
the most important item to consider when applying the pro-
posed algorithm to three-dimensional multideck layout prob-
lems. The structure must be modified by combining several
four-segmented chromosomes as shown in Fig. 17, where
each four-segmented chromosome represents a compartment
layout on the corresponding deck level.
Modification of the distance calculation method between
the compartments—The distance calculation method also
must be modified to treat the multideck layout problems. In
three-dimensional multideck layout problems, the distance
calculation method between compartments can be expanded
by from that of the two-dimensional compartment layout
problem. That is, the two-dimensional distance calculation
method can be expanded into a three-dimensional case. All
relationships between compartments, passages, and vertical
ladders (referred to passages between different decks) are
first represented as an adjacency graph. Then the shortest Fig. 18 Distance calculation in single-deck and multideck layout problems

168 JULY 2002 MARINE TECHNOLOGY


Conclusions and discussions References
AFANASIEFF , L. AND MABRY, J. P. 1994 The design of the FF-21 multi-
In general, up to the present time, the compartment layout mission frigate. Naval Engineers Journal, 106, 3, 150–162.
design of the naval ship has been performed manually. In the ANDREWS , D. 1981 Creative ship design. Transactions RINA, 447–472.
study herein, the compartment layout algorithm by the op- ANDREWS , D. 1986 An integrated approach to ship synthesis. Transac-
timization technique was proposed for solving the compart- tions RINA, 73–104.
BROWN , D. K. 1986 Defining a warship. Naval Engineers Journal, 98, 2,
ment layout problem involving watertight transverse bulk- 31–40.
heads and passages. The compartment layout problem with CARLSON, C. M. AND FIREMAN, H. 1987 General arrangement design com-
watertight transverse bulkheads and passages was mathe- puter system and methodology. Naval Engineers Journal, 99, 3, 261–273.
matically formulated. The layout of compartments was mod- CHATTERJEE, S., CARRERA, C., AND LYNCH, L. A. 1996 Genetic algorithms
and traveling salesman problems. European Journal of Operational Re-
eled in a four-segmented chromosome; this included positions search, 93, 3, 490–510.
of passages. A new method was proposed for calculating dis- COHON, J. L. 1978 Multiobjective Programming and Planning. New
tances between the compartments using the Dijkstra’s algo- York: Academic Press Inc.
rithm of the graph theory. A comparison with an existing CORT, A. AND HILLS, W. 1987 Space layout design using computer as-
sisted methods. Naval Engineers Journal, 99, 3, 249–260.
algorithm was performed to evaluate the proposed algo- DAVIS , L. Ed. 1991 Handbook of Genetic Algorithms. New York: Van
rithm’s efficiency. The comparison results show that the pro- Nostrand-Reinhold.
posed algorithm is superior to the existing one. Finally, the GOLDBERG , D. E. 1989 Genetic Algorithms in Search, Optimization, and
proposed algorithm was applied to ship compartment layout Machine Learning. Reading, MA: Addison Wesley.
HOPE, J. P. 1981 The process of naval ship general arrangement design
problems and the computational results were compared with and analysis. Naval Engineers Journal, 93, 4, 29–38.
the actual ship compartment layout. From this, the applica- HOLLAND , J. H. 1975 Adaptation in Natural and Artificial Systems, Ann
bility of the proposed algorithm was shown and discussed. Arbor: The University of Michigan Press.
The proposed algorithm in this study was first applied to ISLIER, A. A. 1998 A genetic algorithm approach for multiple criteria
facility design. International Journal of Production Research, 36, 6,
two-dimensional ship compartment layout problems to show 1549–1569.
its applicability to three-dimensional ship compartment lay- KIM , H. 1994 Multicriteria structural optimization by genetic algo-
out problems. What remains is the modification of the pro- rithm. Ph.D. thesis, Seoul National University of Korea, written in Ko-
posed algorithm for application to more realistic multideck rean.
KOOPMANS, T. C. AND BECKMANN , M. 1957 Assignment problems and the
layout problems of the three-dimensional ship compartment. location of economic activities. Econometrica, 25, 1, 53–76.
Finally, the optimization results obtained from the proposed LEE , H. J. 1988 Heuristic graph-theoretic approach in facility layout
algorithm could be improved by developing new genetic op- problem: The development of a decision support system. Ph.D. thesis, The
erations and testing the improved algorithm. University of Texas at Arlington.
LEE , K. Y. AND ROH, M. I. 2001 An efficient genetic algorithm using
gradient information for ship structural design optimization. Journal of
Ship Technology Research, 48, 4, 161–170.
Acknowledgments MC HUGH, J. A. 1990 Algorithm Graph Theory. New Jersey: Prentice-
Hall Inc.
This work was supported partially by the Korea Science SEPPANEN , J. AND MOORE , J. M. 1970 Facilities planning and graph
and Engineering Foundation (KOSEF) as part of the re- theory. International Journal of Management Science, 17, 242–253.
TAM , K. Y. 1992 Genetic algorithms, function optimization and facility
search project “Web based Multidisciplinary Robust Ship De- layout design. European Journal of Operational Research, 63, 2, 322–346.
sign” and also sponsored partially by the Research Institute WREN , A. AND WREN , D. O. 1995 A genetic algorithm for public transport
of Marine Systems Engineering of Seoul National University. driver scheduling. Computers & Operations Research, 22, 1, 101–110.

JULY 2002 MARINE TECHNOLOGY 169

View publication stats

You might also like