109

Optimization System of Block Division

Using Genetic Algorithm and Product Model

* *
by  Muhammad Arif Wibisono, Student Member   Kunihiro Hamada, Member
* *
   Mitsuru Kitamura, Member  Varikkattu Karottu Kesavadev , Student Member

Summary

Block construction method is generally used in modern shipyards. In this method, interim products are built by assembling
parts, and the ship is built from these interim products. This process can be referred as “Assembling Hierarchy” in this paper. In
order to construct the ship effectively, an appropriate Assembling Hierarchy plan which considering the configurations of interim
products is required. So in this paper, an optimization system of ship Assembling Hierarchy using Genetic Algorithms (GA) and
Product Model is discussed.
In order to realize the optimization using GA, it is necessary to define the chromosomes and to examine the GA operations.
Therefore, Joint Hierarchy Chromosome is newly introduced. Joint hierarchy chromosome is genotype or encoded solution of
Assembling Hierarchy. In this chromosome, the level of Assembling Hierarchy on each joint is set as design variable. By using this
chromosome, various plans of ship Assembling Hierarchy can be generated. Effective GA operations such as selection, crossover
and mutation are also shown. Moreover, proposed system is integrated with Product Model. Therefore, fitness with penalty of the
plan is calculated by the use of Product Model data.
Finally, an example of the optimization is shown in the paper to confirm the validity of the developed method.

advanced because of the improvement of the computer ability. In

1. INTRODUCTION
 The block construction system is established as an efficient
production system in modern shipyards. In this system,
productivity and efficiency of construction are depending on
block division plans, so it is necessary to make proper block
division plan. Generally, decision of proper block division plan
still depends on expert designer who already has a long-term
experience and knowledge in shipbuilding. However, the number
of such skilled engineers decreases. Therefore, it is required to
support the block division by the use of computer.
 Furthermore, the Product Model in the shipbuilding industry
has already reached an appreciable level of practical use. It has
been proved that information such as the weight, the joint length,
and the size, etc. can be acquired from Product Model, which
become reference in doing the block division. However, because
block division is a huge scale discrete optimization problem, there
are few researches which support plan generation or decision
1)
making of block division except the research by Aoyama et.al.
 Now, the practical use of the optimization technique is also
shipbuilding, a lot of researches have already been done about
2,3) 4,5)
structural design and the work scheduling . The
effectiveness of the Genetic Algorithm (GA) in the optimization
problem concerning the design and the production of shipbuilding
has been shown by these researches. However, the research to
optimize block division has not been executed ever before,
because of the difficulty to formulate the block division problem.
 So, optimization system of the block division is discussed in
this paper. Followings are the characteristics of this study;
z Proposing the new optimization method using joint
information, this can be realized by the use of GA and
Product Model.
z Making clear the GA operations that are appropriate to
optimize block division.
Therefore, basic concepts of proposed method and details of
information processing are shown in this paper. Moreover, the
effectiveness of proposed method is discussed based on some
examples of optimization.
2. THE OPTIMIZATION OF THE BLOCK DIVISION


㧖
Graduate School of Engineering, Hiroshima University  2. 1 Characteristics of the Block Division

  The features and the demands of the block division can be

ේⓂฃℂᐔᚑ  ᐕ  ᦬  ᣣ summarized as followings according to the Ohno 6).

 110 　日本船舶海洋工学会論文集　第 6 号　 2007 年 12 月　

 (1) Consideration of hierarchy of blocks change of combinations of the parts, Aoyama et. al. proposed
 The main concern of the block division is the decision of the the enumeration method using the Cut-Set. However it is pointed
erection block. In order to decide the erection blocks, it is out that a great number of division plans are generated in this
required to evaluate the assembling process of the erection block method, and some new concepts are necessary 1).
in the foregoing stages. Therefore, it is necessary to examine the  2. 3 Requirements to Optimize the Block Division
whole assembling process from parts to ship.  Based on the above discussions, the problems for the
 (2) Evaluation items of the block division optimization of the block division can be summarized as follows;
 In order to decide the block division, following evaluations are  (1) Requirements in generation of block division plan
necessary.  In order to realize the optimization of the block division, it is
z Feasibility: It is necessary to check the period of works necessary to optimize both of the seam position and the
in consideration of the availability and specifications of combination of the parts. The concept of dimension optimization
equipments. can be applied into the optimization of seam position which has
z Productivity: It is necessary to check the productivity already researched in the field of structural optimization.
of the shipbuilding by calculating the welding time etc. Therefore, the main problem to realize the optimization of block
quantitatively. division is to develop the method for the optimization of the
z Workability: It is important that block division agrees combination of the parts. In this method, it is required to optimize
with the construction method and equipment ability of not only erection block but also all the hierarchy from parts to
the factory under consideration. For this, evaluation is ship.
done based on the features of each shipyard and  (2) Requirements in evaluation of the division plan
manager’s construction policy.  The effectiveness to use the Product Model information in the
 (3) Concurrency of block division. evaluation has been proved in the previous researches. Therefore,
 All parts information has not been fixed in early stage of it is desirable to achieve the evaluation using the Product Model
planning the block division. Parts information is decided by integrating optimization engine and Product Model.
concurrently with block division plan. Therefore, block division  2. 4 Definitions of Optimization Problem in this Paper
plan is done in multi steps.  The effectiveness of the GA in shipbuilding field has been
 2. 2 Review of Related Papers shown by the previous researches. Therefore, optimization system
 The current state of the research to support the block division using GA and Product Model is discussed in this paper (Fig.2).
by using Product Model can be summarized as follows:  Combination of parts from parts to ship is paid attention in this
 (1) Effectiveness of evaluation using Product Model
 Information on joints and the interim products in addition to
parts is managed in the Product Model. It has been shown that
effective information to evaluate the block division can be A C A C A C
7)
extracted from Product Model .
B D D
 (2) Correspondence to the concurrency B D B
 Considering the concurrency of block division, it is important Change of combination of
Change of seam position parts
to estimate the unfixed information such as small parts and so on. (a) (b)
Aoyama et.al., deal with this problem by defining the estimating Fig.1 Modifications of Block Division Plan
function for the small parts based on the previous experiences 1).
 (3) Generation and modification of block division plan Coding space encoding Solution space
 Modification of block division plan is divided roughly into two Genetic Algorithm Product Model
Module Module
types, that is, the change of seam position (Fig.1(a)) and the Chromosome generation Pre-Condition
Fitness calculation
change of the combination of the parts (Fig.1(b)). It is relatively Genetic Operations
(Crossover and mutation) Evolution operation
easy to correspond to the change of seam position because it is (Selection)
possible to apply the design modification method using the decoding
8)
dimension information and the geometrical constrains . For the Fig.2 Basic Configuration of the Optimization System
 Optimization System of Block Division Using Genetic Algorithm and Product Model 111

paper. Therefore, pre-condition and design variables are and such differences of techniques produce a big influence on the
considered as follows; solution and the calculation time. Therefore, it is necessary to
z Pre-condition: Positions of the seams between parts examine suitable GA operations to optimize the block division.
have been already fixed. In other word, joint
information and parts information has been generated in  In this paper, chromosome is examined in section 3 and GA
the Product Model. operation is examined in section 4.
z Design variables: Interim products in each stage that is
defined as combinations of parts are set as design 3. DEFINITIONS OF CHROMOSOMES
variables. In this paper, the composition of parts and the
interim products is called Assembling Hierarchy (Fig.3).  Not only parts information but also joint information is
On the other hand, “Feasibility” and “Workability” in section 2.1 described in the Product Model. Considering this characteristic of
corresponds to constraints, and “Productivity” in same section Product Model, following two methods can be considered to
corresponds to objective function in optimization problem. generate various Assembling Hierarchies;
Therefore, constraints and objective function are set as follows; z Method1: Generate interim products and their hierarchy
z Constraints: Typical constraints of equipments in each by paying attention to joints. This method is called
stage. “Joint Hierarchy Method” in this paper.
z Objective function: total welding time. z Method2: Generate interim products and their hierarchy
In order to realize above system, it is necessary to examine by paying attention to combinations of parts. This
following points; method is called “Parts Level Method” in this paper.
 (1) Definitions of the Chromosome Outlines and evaluation results of above two methods are
 In GA, various design plans on the block division are expressed discussed in this section.
as chromosomes. Therefore, it is necessary to examine the  3. 1 Outlines of Joint Hierarchy Method
chromosome for expressing various Assembling Hierarchies.  3. 1. 1 Definitions of Joint Hierarchy Chromosome
Moreover, it is required to make clear the generation method of  All joints of the ship are connected in a construction process.
Product Model from the chromosome, in consideration of the Therefore, Assembling Hierarchy can be decided by setting the
integration of Product Model and GA. assembling stage to each joint. Based on this concept, Joint
 (2) Examinations on GA operation Hierarchy Chromosome is newly introduced.
 Basic flow of GA is shown in Fig 4. Various techniques of  Joint Hierarchy Chromosome is genotype or encoded solution
crossover, mutation and selection etc. have been proposed in GA, of an Assembling Hierarchy. The examples of Joint Hierarchy
Chromosomes are shown in Fig.5. At first, a target structure is
Various A B modeled as graph model (Fig.5 (2)), then, each joint in Graph
A product
Assembling
composed of 5 parts C D E
Model is numbered. Therefore, the size of the chromosome is
Hierarchy
A D E
A B C B
A B
C D E (4) Assembling
C D E C D Hierarchy
(3) Joint Hierarchy
(1) Parts Chromosome A B
A B C D E A C B D E C D E
A B #1#2#3 #4#5#6#7
C D C D E 2 3 3 3 3 1 2 A B C D E
1 2 3 2 2 3 1 C D
A B C D E
3 2 3 2 2 3 1 A B C D E
Fig.3 Examples of Assembling Hierarchy (2) Graph Model 1 1 1 3 2 1 3
A B
#1 2 3 3 3 3 1 1 C D E
encoding evaluation A B
crossover #2 #5 3 3 1 3 2 1 3
#3 #4 A
solutions chromosomes offspring C D E B
#6 #7 1 3 3 3 3 1 1 C D E
mutation deco ding
new D E
population
solutions
selection A C B D E
fitness calculation

Fig.4 Basic Flow of Genetic Algorithm Fig.5 Examples of Joint Hierarchy Chromosomes
 112 　日本船舶海洋工学会論文集　第 6 号　 2007 年 12 月　

same as the total number of the joints. Moreover, a hierarchical
level is put to each gene (Fig.5 (3)). By using this information, the
levels of all joints are expressed, and various Assembling
Hierarchies are generated (Fig.5 (4)).
 Generation process of an Assembling Hierarchy based on the
Joint Hierarchy Chromosome is shown in Fig.6. As shown in
Fig.6, an Assembling Hierarchy is generated step by step, from
Level 1 until highest level.
 3. 1. 2 Joint Contradictions
 In the case of using above method, there is a risk to generate
wrong Assembling Hierarchy in which a joint is not connected
though interim products are connected. An example of such a
case is shown in Fig.7. Such a problem is called “Joint
Contradiction” in this paper. Joint Contradiction is automatically
9)
checked and corrected by executing following process ;
1) The Assembling Hierarchy is generated based on the
chromosome (Fig.7 (2)).
2) Checking the presence of products and interim products
which are composed of same parts though levels in the
Assembling Hierarchy are different (Fig.7 (3)).
3) In the case of presence, it is judged that the joint with
the problem (Fig.7 (4)) is connected in the same level in
a hierarchy, and a correct Assembling Hierarchy is
generated (Fig.7 (5)).
 3. 2 Outlines of Part Level Method
 Interim products are composed of various parts, therefore, an
Assembling Hierarchy can be generated by setting the
combinations of parts. Parts Level Chromosome is defined based
on this concept.
 Examples of Part Level Chromosomes are shown in Fig.8. Part
Level Chromosome contains two series of information, part series
and level series. Size of part series is same as a number of parts
and size of level series is same as a number of parts minus one.
Part series shows the composition of parts, and level series shows
connection level between two parts in part series.
 Generation process of Assembling Hierarchy is shown in Fig.9.
Each gene in level series corresponds to two adjoined parts in part
series. By using this relation, generation starts from level 1 by
connecting one part and next part where level data stay. After that,
a part is recognized as an interim product that contains the part
(e.g. part D is recognized as an interim product composed of part
D and part C). Therefore, by repeating above process, an
Assembling Hierarchy is generated.
 3. 3 Evaluations of two Methods
 Above two methods are evaluated by executing optimization.
In this evaluation, GA operations are set as follows;
z population size :100
z crossover: two-point crossover 50%
z mutation: swap mutation 30 %

Start Level 1 Level 2 Level 3 z selection: rejecting strategy

23 33 312 2 33 33 12 23 33 3 12 Details of above GA operations and fitness calculation are shown
A #1 B
A B A B A B
#2 #3 #4 #5 in section 4. Optimization results with target structure are shown
C D E C D E
#6 #7 C D E C D E
in Fig.10. In the case of using Part Level Method, many infeasible
A B
1 2 34 5 67 CDE solutions are generated in which interim products are composed
2 333 3 12
A B CDE A B CD E of parts away, and it brings bad influences on optimization.
A B CD CD CD
CDE Therefore, joint hierarchy chromosome is selected as decoding in
C D A B C D E A B C D E
this paper.
Fig.6 Generations of Assembling Hierarchy based on the Joint (4) Assembling
Hierarchy Chromosome Hierarchy
(3) Part Level A B
(1) Parts Chromosome
C D E
A B Part Level
series series C D E
(1) Product (2) Wrong (3) Same (4) Joint with (5) Correct C D E
Model & Assembling Assembling the Problem Assembling
ABCDE 1 3 1 2 A B C D
Chromosome Hierarchy Parts Hierarchy DACBE 1 3 1 2
A B C D E
A B A B A B Joint with CADBE 1 2 3 3
A 1 B (2)
C D C D C D the problem Simplification AEDBC1 2 3 2 A B
2 3 C D E
A B A B A B A B
Model AEDBC1 3 3 2
C 1 D B
C D C D C D C D A B C D E BDEAC2 3 1 1
1231 #1 #2 #3 #4 C
A E
E A BCD 2 2 3 1
A B A B C D A B C D A B C D A B C D
A E D B C
C D A B C D A B C D A B C D A B C D

Fig.7 Corrections of Joint Contradictions Fig.8 Examples of Part Level Chromosomes

 Optimization System of Block Division Using Genetic Algorithm and Product Model 113

Start Level 1 Level 2 Level 3 Rejecting strategy and penalty strategy are enumerated as typical
ABCDE ABCDE ABCDE ABCDE strategies to handling the constraints. Therefore, these two
1 3 1 2 1 3 1 2 1 3 1 2 1 3 1 2
A B C D E A B C D E A B C D E A B C D E methods are compared in this paper.
#1 #2 #3 #4 A B  4.2.2 Rejecting Strategy
CDE
A B CDE CDE  Rejecting strategy discards all infeasible chromosomes and
CDE
A B CD A B CD A B CD leaves only feasible chromosomes. In this paper, mismatch of
A B C D A B C D E A B C D E levels in Assembling Hierarchy (above 1)) and mismatch to
Fig.9 Generations of Assembling Hierarchy Based on the Part constrains of production units (above 2)) are considered in
Level Chromosome
rejecting strategy. The flow of the optimization using rejecting
4000
Product Data strategy is shown in Fig.11. Details of information processing in
Block Size
L: 50.00 m the optimization flow are explained here.
B: 20.00 m
3600 D: 10.00 m
Weight: 24,520 kg
 (1)Pre-Conditions
Number of parts 84
Number of joints 204  Product Model data and production unit data are input by
3200
designer. This process uses the previous research result 11).
Fitness

Part Level Chromosome  (2)Generations of initial populations

2800
Part Level Chromosome  The size of chromosome is fixed by the use of Product Model
Joint Hierarchy Chromosome
Joint Hierarchy Chromosome data. Moreover, total number of stages of Assembling Hierarchy
2400
0 200 400 600 800 1000 could be extracted from the production unit data. Based on these,
Generation
the level of each gene is assigned by generating random numbers.
Fig.10 Comparisons of Two Methods
Then, infeasible chromosomes are rejected by executing the
rejecting strategy, and the new chromosomes are generated until
4. EXAMINATIONS ON GA OPERATIONS
population size is fulfilled. As a result, feasible chromosomes are
generated as initial population.
 4. 1 Overview of the Genetic Algorithm Operations
 (3)Fitness Calculation
 GA is a stochastic search technique based on the mechanism of
 Each shipyard has several production stages and each stage has
natural selection and natural genetics. There are two operations in
different processing speed. These processing speeds are assumed
GA, genetic operation (crossover and mutation) and evolution
10)
as welding speed in this paper. Welding speed of each stage is
operation (selection) . GA is widely used in optimization
input by designer in “Pre-Conditions” process, and total welding
problems in various fields, and various GA operations have been
time (fitness) is calculated using equation (1).
proposed. Moreover, specific optimization problem needs specific
operations to solve correctly. Therefore, GA operations to solve Genetic Algorithms Product Model Module
Module
the optimization of Assembling Hierarchy are examined in this Generations of initial Pre-Conditions
populations
paper in order to get good optimization results. Product Model Data
Random Generation (Joints and Parts)
 4. 2 Examinations to Handling Constraints Rejecting Strategy Production Unit Data
Capacities of Production Unit
 4.2.1 Strategies to Handling the Constraints First Generation Welding Speed of each Stage
 The optimization of a ship Assembling Hierarchy is a huge Fitness Calculation
Rejecting Strategy
Selection Generations of
scale discrete optimization problem and there are many
Crossover and Mutation Assembling Hierarchy
constraints in the ship construction process. Many infeasible plans
Definition of next Check of
will be generated because of the following constrains; populations Assembling Hierarchy Level &
Rejecting Strategy Constrains of Production Unit
1) The total level of generated Assembling Hierarchy does Next Generation
Fitness Calculation
not match with the number of production stages because Fitness Calculation
Generations of
of the Joint Contradiction. no Assembling Hierarchy
Stop?
yes
2) Interim products in Assembling Hierarchy cannot fulfill END
Calculations of Welding Time

the restrictions of equipments such as capacity of cranes

and so on. Fig. 11 Flow of Optimization using Rejecting Strategy
 114 　日本船舶海洋工学会論文集　第 6 号　 2007 年 12 月　

n m
(1) (2) Fitness Calculation
f ( x) ¦¦ JLij u WTi
i 1 j 1  In order to penalize infeasible solutions, equation (2) is used to
 Where f(x) : total welding time (objective function) calculate fitness;
WTi : welding time in each stage (minutes/m) F ( x) f ( x)  LP  MP (2)
§ § WT  WTH 1 · ·
JLij : Joint length in each module (m) LP JLT u ¨¨WTH  ¨¨ d u H ¸¸ ¸
WTH ¸
© © ¹¹
m : number of modules in each assembling stage
WP
MP u JLm u WTL
n : number of assembling stages WC
 (4)Selection  Where F(x) : fitness
 Here, the tournament selection (tournament size: 2) is used as f(x) : total welding time (equation(1))
selection method and details are discussed in section 4.3. LP : Level Penalty (Mismatch of levels in Assembling
 (5)Crossover and Mutation Hierarchy)
 Two-point crossover and swap mutation is used. Details of MP: Module Penalty (Interim products cannot fulfill
crossover and mutation are discussed in section 4.4. the constraints of facilities)
 (6)Population for next generation JLT = Total Joint length in all module (m)
 There is a possibility to generate many infeasible solutions by WTH : welding time in highest level (minutes/m)
crossover and mutation. Therefore, the rejecting strategy is used WTH-1 : welding time in one level below highest level
again, and only feasible chromosomes are selected. After that, d : deviation of mismatch level with target level
selected feasible chromosomes and chromosomes in the previous WP : Weights of interim product
generation are gathered, and chromosomes of population size are WC : Crane capacity
selected as individuals of the next generation by tournament JLm = Joint length of target module (m)
10)
selection. This method is based on the enlarged sampling space . WTL : welding time in target level (minutes/m)
 4.2.3 Penalty Strategy  (3)Definitions of the populations for next generation
 Penalty strategy transforms the constrained problem into an  Chromosomes selected by the tournament selection are directly
unconstrained problem by penalizing infeasible solutions, in become the populations of the next generation.
which a penalty term is added to the objective function. Flow of  4.2.4 Evaluations of the Strategies
the optimization using the penalty strategy is shown in Fig.12.  Optimization results of rejecting strategy and penalty strategy
 The main differences of this method compared with rejecting are shown in Fig.13 in which target structure is same as Fig.10.
strategy are following three points; Fitness of both strategies is almost same and there is no big
 (1)Generations of initial populations difference between optimum solutions. On the other hand, there is
 The chromosomes generated at random are adopted as the a big difference on the calculation time. The weakness of
individuals of the first generation. Therefore, infeasible rejecting strategy is taking so long time to generate initial
chromosomes are included in the first population. populations. Because of this reason, penalty strategy is selected as
a method for handling constraints.
Genetic Algorithms Product Model Module
Module
Generations of initial Pre-Conditions
4000
populations Product Model Data
(Joints and Parts) 6000
Random Generation Generating
Production Unit Data Time
Fitness Calculation Capacities of Production Unit 3600 3000
Welding Speed of each Stage
First Generation
Crossover and Mutation Fitness Calculation 0
3200
Generations of Rejecting Penalty
Fitness

Fitness Calculation Assembling Hierarchy

Selection Rejecting Strategy
Check of 2800
Next Generation Assembling Hierarchy Level & Penalty Strategy
Constrains of Production Unit
no
Stop? Calculations of Welding Time 2400
yes
END Calculations of Penalty 0 200 400 600 800 1000
Generation

Fig. 12 Flow of Optimization using Penalty Strategy Fig.13 Comparisons of two Strategies
 Optimization System of Block Division Using Genetic Algorithm and Product Model 115

 4. 3 Examinations on Selection Methods 4500

Selection provides the driving force in GA, and the selection
enlarged
roulette
process is critical in it. There are two basic issues involved in 4000 enlarged
selection phase, sampling space and sampling mechanism 10). tournament
size:3
 4.3.1 Overview of Sampling Spaces regular
3500 roulette

Fitness
 In GA, sampling space is created for the selection. There are enlarged
tournament
size:2
two main types of sampling spaces, regular sampling space and 3000 regular
tournament
enlarged sampling space. In the case of regular sampling space, size:3
regular
the size of sampling space is same as population size and which is 2500 tournament
0 100 200 300 400
__size:2
500
composed of offspring chromosomes after crossover and mutation. Generation

On the other hand, enlarged sampling space is defined as sum of Fig.14 Influences of Sampling Spaces and Mechanisms
parents and offspring, therefore, size of sampling space is larger
than the population size. Table 1 Influence of Supplementary Sampling Mechanisms
 4.3.2 Overview of Sampling Mechanisms Sampling
Case Supplementary Size Fitness
 Sampling mechanism concerns the problem of how to select Mechanism
chromosomes from sampling space. This paper examines two 1 - - 2721
2 Elitist 2 2704
types of sampling mechanisms, roulette selection and tournament
3 Tournament 2% 2693
selection. In the case of tournament selection, tournament size 4 size 2 5% 2583
Truncation
also influences the optimization. 5 10% 2614
 In addition to above sampling mechanisms, supplementary 6 20% 2746
sampling mechanisms might be used together. Followings are
typical supplementary sampling mechanisms;  4. 4 Examinations on Genetic Operations
z Elitist selection: The best chromosomes in generation  4.4.1 Crossover Methods
are directly inherited to the next generation.  Crossover is the main genetic operation in GA. One-point
z Truncation selection: All chromosomes are ranked crossover and two-point crossover are typical methods for
according to their fitness and chromosomes of certain crossover. In one-point crossover, one position at the
percent in high rank are inherited to the next generation. chromosome is selected at random, and the behind parts of both
 4.3.3 Evaluations of Selection Methods individuals are exchanged each other. In two-point crossover, two
Adequate selection methods are evaluated by executing positions at the chromosome are selected at random, and the
optimizations with various evolution operations. Hence, GA middle parts of both individuals are exchanged.
operations except for evolution operation are decided based on  4.4.2 Mutation Methods
the discussions in the previous section. Fig.14 shows the Mutation is a background operation to produce spontaneous
influences of sampling spaces and sampling mechanisms. From random changes in various chromosomes. Change mutation and
this, it is clear that regular sampling space and tournament swap mutation are typical mutation methods. Change mutation
selection (size: 2) is adequate for this problem. selects a bit randomly and changes it to an allowable number.
Table 1 shows the influences of supplementary sampling Swap mutation selects two bits randomly, and exchanges bits
mechanisms in which average values of fitness of three times each other. Mutations which mutate certain length of bits is called
optimizations are shown. Truncation selection (5%) with mutation with length.
tournament selection gets a good result although its influences are  4.4.3 Evaluations on Genetic Operations
smaller than sampling space and sampling mechanism.  Crossover methods and mutation methods are evaluated by
Therefore, selection methods in this study are decided as executing optimizations in which target structure is same as
regular sampling space, tournament selection (size: 2) and Fig.10 and the other GA operations are decided based on the
truncation selection (5%). discussions in section 4.3. Average fitness of four times
optimization is shown in Table 2. Two-point crossover with swap
mutation (length: 10) gets a good result.
 116 　日本船舶海洋工学会論文集　第 6 号　 2007 年 12 月　

Table 2 Influences of Crossover and Mutation

z Block Size
Swap Mutation Change Mutation y L: 250.00 m
x B: 57.00 m
Length 1 10 50 1 5 15
D: 20.00 m
One-point Weight: 6164.3 t
2876 2898 2961 2981 2739 2793
Crossover Number of parts 406
Two-point Number of joints 1129
1056
2608 2532 2700 2589 2579 2668
Crossover
Fig. 15 Product Data
However, the efficiency of optimization is influenced by not
only crossover method and mutation method but also crossover
Level 1 Level 2 Level 3
rate and mutation rate. Therefore, adequate rates for above
Grand Dock
Assembling Assembling
methods have been examined by executing optimizations with Stage Stage Stage
PARTS (Welding time (Welding time SHIP
(Welding time
changing crossover rate and mutation rate, and 90% crossover 1 minutes/m) 5 minutes/m) 10 minutes/m)

and 20% mutation gets a good and robust results. Crane 1 Crane 2
(Capacity: max.weight: 350 t; (Capacity: max.weight: 630 t;
 4. 5 GA Operations Proposed by this Study max.size: 35 x 35 x 35 m) max.size: 60 x 60 x 60 m)

 Based on the discussions in this section, following GA Fig. 16 Production Unit Data
operations are selected to optimize Assembling Hierarchy;
z Strategy: Penalty strategy
seam position
z Selection:
• Regular sampling space
y
• Tournament selection size 2 with Truncation 5%
z Crossover: Two-point crossover, rate: 90% x crane 2
product top view
Capacity (size)
z Mutation: Swap mutation with length 10 (5% of
chromosome length), rate: 20%
y

x
5. AN EXAMPLE OF THE OPTIMIZATION
z crane 1
Capacity (size)
 A prototype system for the optimization of Assembling x product side view

Hierarchy has been developed based on the previous research Fig. 17 Seam Positions and Crane Capacities(Size)
11)
result . Therefore, product data and production unit data are
generated by the use of the base system. Product data used in this
150000
section is shown in Fig.15. Production units are composed of 3
levels i.e. assembling stage, grand assembling stage and dock
stage. Each stage has different welding time and different 130000

constraints. Parameters set in production units are shown in

Fig.16. Fig.17 shows seam positions and size capacity of cranes. 110000
As shown in the figure, the weight constraint is more predominant
Fitness

for the erection block than the size constraints. Considering the
90000
weight constraints, 10 erection blocks should be generated. On
the other hand, the size of the interim product is restricted by the
size constraints in Level 1. 70000

Optimization result is shown in Fig.18, as well as their

attributes in Table 3. Moreover, some other plans are compared 50000
with the optimum solution in Table 4, in which only a few 0 200 400 600 800 1000
Generation
combinations of parts are different and different points are shown
by the bold lines. Fig. 18 Results of the Optimization
 Optimization System of Block Division Using Genetic Algorithm and Product Model 117

Table 3 Attributes of the Optimum Solution  From these, it is understood that the block division is a
multi-peak optimization problem, and only a few changes
Module
Level Constraints produce a quite large influence on fitness. Instead of these
Name Weight (t) Size (m)
3 III-1 6164.3 250 x 57 x 30 䋭 difficulties of the block division problems, an appropriate solution
II-1-1 625.5 25.1 x 57.0 x 30.0
is obtained by the proposed method. Therefore, it can be judged
II-1-2 625.3 25.2 x 57.0 x 30.0
II-1-3 608.4 25.2 x 57.0 x 30.0 630 (t) that the validity of the proposal method is shown.
2 60 x 60 x 60
II-1-4 608.4 25.0 x 57.0 x 30.0  As for calculation time, 125 hours is required for this
(m)
x x x
x x x optimization by using the PC (CPU: Pentium4 2.8GHZ, Memory:
II-1-10 619.8 25.1 x 57.0 x 30.0 1GB).
I-1-1-1 191 25.1 x 16.5 x 30.0
I-1-1-2 243.6 25.1 x 24.0 x 30.0
I-1-1-3 191 25.1 x 16.5 x 30.0 350 (t) 6. CONCLUSIONS
1 35 x 35 x 35
I-1-2-1 191 25.2 x 16.5 x 30.0
(m)
x x x
x x x  An optimization system for ship Assembling Hierarchy using
I-1-10-3 202.7 25.1 x 16.5 x 30.0
GA and Product Model is developed in this paper. Conclusions of
this paper are summarized as follows;
Table 4 Comparisons of Generated Plans (1) Joint Hierarchy Chromosome is introduced as a genotype of

Con- an Assembling Hierarchy. The characteristic of this

Weld. Time)
O.F. (Total

straints chromosome is that the level of Assembling Hierarchy on

Fitness

Assembling Hierarchy
Weight

each joint is set as design variable. The effectiveness of this

Size

chromosome is shown by comparing it with the

chromosome paying attention to the combination of parts.
Optimum Solution

(2) GA operations to optimize Assembling Hierarchy are

57127

57127

٤ ٤ examined by executing various optimizations. As a result,

penalty strategy, tournament selection with regular
sampling space, swap mutation with length and two-point
crossover could be judged as appropriate operations in the
problem.
(3) Optimizations of Assembling Hierarchy by the use of the
Plan-A

57040

58886

٤ × proposed method are executed, and appropriate solution can

be obtained. Therefore, validity of the proposed method is
confirmed.

ACKNOWLEDGEMENTS
Plan-B

57262

60936

× ٤  In the development of prototype system, the system developed

by Emeritus Prof. Toshiharu Nomoto and Prof. Kazuhiro Aoyama
of the University of Tokyo was effectively used. Authors would
like to express gratitude to them. The financial support offered by
the Japan Society for the Promotion of Science (Grant-in-Aid for
Scientific Research, Scientific Research (B) 18360418), is
Plan-C

57737

57737

٤ ٤ gratefully acknowledged.
 118 　日本船舶海洋工学会論文集　第 6 号　 2007 年 12 月
REFERENCES
1) Aoyama, K., Takechi, S. and Nomoto, T.: Basic study on
the design support system for construction block on CIM in
shipbuilding, Journal of the Society of Naval Architects of
Japan, Vol.188, (1999), pp.387-398.
2) Kitamura, M., Nobukawa, H., Djenod, K.: A Study on
Optimum Structural Design of Container Ship, Journal of
the Society of Naval Architects of Japan, Vol.186, (1999),
pp.353-360.
3) Nobukawa, H., Yang, F., Kitamura, M.: Optimization of
Engine Room Structure under Static and Dynamic
Constrains Using Genetic Algorithms, Journal of the
Society of Naval Architects of Japan, Vol.183, (1998),
pp.315-322.
4) Arai, M.: A Study on the Optimization of the Hull-block
Construction Process by Multi-Objective Genetic Algorithm,
Journal of the Society of Naval Architects, Japan, Vol.190,
(2001), pp.459-467.
5) Chen, S., and Liu, Y.: The Application of Multi-level
Genetic Algorithms in Assembly Planning, Journal of
Industrial Technology, Vol.17, No.4, (2001), pp.1-9.
　
6) Ohno, I.: Systemization of Production Control in the
Shipyard, Journal of the Society of Naval Architects, Japan,
Vol.187, (1999), pp.321-328.
7) Aoyama, K., and Nomoto, T.: Information Models and
Functions for CIM in Shipbuilding, Journal of Marine
Science and Technology, Vol.2, (1997), pp.148-162.
8) Nomoto, T., Aoyama, K. and Aramaki, K.: Basic Studies on
the Design of Plate Structures by Describing of Design
Process, Journal of the Society of Naval Architects of Japan,
Vol.172, (1992), pp.675-682.
9) Wibisono, M. A., Hamada, K., and Kitamura, M.:
Optimization System for Ship Assembling Hierarchy Using
Genetic Algorithm and Product Model, TEAM 2006, Seoul,
South Korea, (2006), pp.220-227.
10) Gen, M. and Cheng, R.: Genetic Algorithms and
Engineering Design, John Wiley & Sons, New York,
(1997).
11) Hamada, K., Bentin, M., and Kitamura, M.: Master
Planning Supporting System for Cooperative Ship
Assembling, Transactions of The West-Japan Society of
Naval Architects, Vol.104, (2002), pp.229-239.