Академический Документы
Профессиональный Документы
Культура Документы
* *
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.
㧖
Graduate School of Engineering, Hiroshima University 2. 1 Characteristics of the Block Division
(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 3. 2 Outlines of Part Level Method
level is put to each gene (Fig.5 (3)). By using this information, the Interim products are composed of various parts, therefore, an
levels of all joints are expressed, and various Assembling Assembling Hierarchy can be generated by setting the
Hierarchies are generated (Fig.5 (4)). combinations of parts. Parts Level Chromosome is defined based
Generation process of an Assembling Hierarchy based on the on this concept.
Joint Hierarchy Chromosome is shown in Fig.6. As shown in Examples of Part Level Chromosomes are shown in Fig.8. Part
Fig.6, an Assembling Hierarchy is generated step by step, from Level Chromosome contains two series of information, part series
Level 1 until highest level. and level series. Size of part series is same as a number of parts
3. 1. 2 Joint Contradictions and size of level series is same as a number of parts minus one.
In the case of using above method, there is a risk to generate Part series shows the composition of parts, and level series shows
wrong Assembling Hierarchy in which a joint is not connected connection level between two parts in part series.
though interim products are connected. An example of such a Generation process of Assembling Hierarchy is shown in Fig.9.
case is shown in Fig.7. Such a problem is called “Joint Each gene in level series corresponds to two adjoined parts in part
Contradiction” in this paper. Joint Contradiction is automatically series. By using this relation, generation starts from level 1 by
9)
checked and corrected by executing following process ; connecting one part and next part where level data stay. After that,
1) The Assembling Hierarchy is generated based on the a part is recognized as an interim product that contains the part
chromosome (Fig.7 (2)). (e.g. part D is recognized as an interim product composed of part
2) Checking the presence of products and interim products D and part C). Therefore, by repeating above process, an
which are composed of same parts though levels in the Assembling Hierarchy is generated.
Assembling Hierarchy are different (Fig.7 (3)). 3. 3 Evaluations of two Methods
3) In the case of presence, it is judged that the joint with Above two methods are evaluated by executing optimization.
the problem (Fig.7 (4)) is connected in the same level in In this evaluation, GA operations are set as follows;
a hierarchy, and a correct Assembling Hierarchy is z population size :100
generated (Fig.7 (5)). z crossover: two-point crossover 50%
z mutation: swap mutation 30 %
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
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
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
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
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
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
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
Assembling Hierarchy
Weight
57127
57040
58886
ACKNOWLEDGEMENTS
Plan-B
57262
60936
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