Professional Documents
Culture Documents
Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng
art ic l e i nf o a b s t r a c t
Article history: This paper presents an application of knowledge-based engineering in containership cargo tank
Received 12 May 2012 structural design and optimization. Ship design is such a complicated multi-discipline task that design
Accepted 16 June 2013 experiences and examples are indispensable. In this proposed knowledge-based system, experiences of
Available online 17 July 2013
design experts, design rules and successful previous designs are stored in the knowledge base. In the
Keywords: design process of new ship structures, the relevant knowledge are automatically distracted from
Knowledge-based engineering knowledge base and executed together with the knowledge reasoning technique. Both design rules
Rules method method and Interpolation method are introduced into this system. The design results fulfill the design
Interpolation method constraints and requirements, therefore design errors and mistakes can be avoided. What's more, design
Reliability
cycle can be reduced significantly. So it can be an appropriate way applying knowledge-based
engineering in ship structural design. What's more, Finite Element Method is employed to carry out
the strength analysis, and reliability analysis is performed with Monte Carlo Simulation by introducing
uncertainty of design variables under seawater corrosion.
& 2013 Elsevier Ltd. All rights reserved.
0029-8018/$ - see front matter & 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.oceaneng.2013.06.013
J. Cui, D. Wang / Ocean Engineering 72 (2013) 124–139 125
An extensive survey of challenges in current ship and floating representation, knowledge reasoning, with its emphasis on reus-
structure design and analysis is made by Sharma et al. (2012), and ing product design knowledge, design experience and other kinds
the following challenges remain active: to develop novel of knowledge in design, developing new and optimal products at
approaches to extract design and production information from the top speed.
previous designs and use them intelligently in the new design and Knowledge acquisition refers to acquiring knowledge from
production schemes, to develop comprehensive functional design knowledge sources, such as design expertise, design standards,
approaches, etc.; to apply the modular approach in the ship/ product specifications, design experience and successful prece-
floating structure design with real concept utilization, industrial dents, etc. Fig. 1 shows a general approach of knowledge
participation and applications; integration of disciplinary simula- acquisition.
tion tools into complex engineering environment of CAD/CAM/ Knowledge representation refers to expressing problems and
CAE/FEM/CFD, integration of different geometric models; to storing knowledge in a computer-interpretable representation,
develop integrated computational approaches in CFD, FEA and which can facilitate computers to take advantage of the knowledge
manufacturing simulation based upon input geometric definition base and solve complicated and difficult problems. Simply speak-
and so on. ing, knowledge representation translates human language into
This paper discusses how to apply methods of knowledge- computer language (Zhou et al., 2007).
based engineering in containership's cargo block structures design, Knowledge-based reasoning refers to the deductive thinking
and integration of design and analysis. Thus some of the chal- process which is deducing another judgment from judgment
lenges mentioned above are addressed in this paper, e.g. the known according to a strategy. Designers express the product
methods to extract information from previous designs and use knowledge base including expert knowledge, experience knowl-
them in the target ship design, modular approach (standard edge, specifications and precedent through knowledge-based
sections or bulkheads) in ship design, and integration with FEM engineering to guide designer in the design process. The reasoning
tools based on geometric model information. The performance of methods include rule-based reasoning (RBR) and case-based
hull structural design is improved with the help of KBE. Addition- reasoning (CBR). The method of RBR is mainly used for specific
ally, reliability analysis is carried out with FEM by introducing parameters based on knowledge advisor and knowledge expert
uncertainty of design parameters, and conclusions are made by work system. The method of CBR is used for product and part
comparing results from both design rules method and Interpola- design similar with previous design cases based on product
tion method. The aims of this paper are to develop a knowledge- knowledge template.
based engineering system by the acquisition of design knowledge In this paper, CBR and RBR are both employed in ship structural
and pervious designs, to improve design efficiency and quality, and design.
to discuss the influence of design parameters uncertainty on the The user-friendly interface provided by the software system
structural performance. helps users export and import design information and manufac-
ture information to achieve a joint of deliberation of human–
computer to address problems, and the design process of
2. Knowledge-based engineering knowledge-based engineering is illustrated in Fig. 2.
The ship structural design system based on KBE includes user
Knowledge-based engineering is a kind of artificial intelligence interface, knowledge base, and inference engine. The user inter-
technology, and it uses the principles and methods of artificial face provides user a quick search and check of knowledge
intelligence to address difficult problems. The core of knowledge-
based engineering is to integrate professional knowledge, domain
knowledge, users' maturity design experience, the choice of design
parameters based on experimental data, material data, users' Knowledge base
feedback, and relevant design standards and norms into the design
of software through the logical judgments and deduction, achiev-
ing product intelligent design (Helvacioglu and Insel, 2003).
A concise definition of KBE is given by Kim et al. (2012) as follows:
KBE has roots in computer aided design (CAD) and knowledge- Extract and
based systems and from early success as a support system for a refine
design engineer within the context of product design to the basis
for generative design, it can have a wide scope that covers the full
range of activities related to Product Lifecycle Management (PLM)
and Multidisciplinary Design Optimization (MDO). The scope of
Documents Experts Cases,etc
KBE includes design, analysis, manufacturing, and support.
As demonstrated by Zhou et al. (2007), knowledge-based engi-
neering connotation includes: knowledge acquisition, knowledge Fig. 1. Knowledge acquisition.
126 J. Cui, D. Wang / Ocean Engineering 72 (2013) 124–139
Expert Design
Design cases ……
knowledge experience
Knowledge
Inference
Machine
Knowledge Knowledge
Workspace
Explanation Acquisition
User
Interface
coefficient, if 90 ≤ L ≤ 300 m, target ship's scantlings and design rules formulations, expertise
and so on. In this paper, the standard parts library for stiffeners'
300L 3=2
C ¼ 10:75 ð2Þ cross-section is stored in access sheets as shown in Fig. 4.
100
The minimum modulus W 0 of mid-ship section should be not
smaller than: 5. Design example
2 3
W 0 ¼ CL BðC b þ 0:7Þ cm ð3Þ 5.1. Design process explanations
Additionally, considering the effects of high-strength steel's
application in structural components, we introduced material In our previous paper, the design of containership midship
coefficient K L for longitudinal strength in Table 1, and K for local section is investigated by knowledge-based engineering (Cui et al.,
strength. Detailed specifications are shown as follows. 2012), and this paper is an extended study of the previous one.
Material coefficient of local strength K is defined as follows. The structural design based on knowledge-based engineering can
be divided into 3 steps. The first step is to establish a design
K ¼ maxf235=ReH ; 0:66g ð4Þ
knowledge base including design expertise, design rules, design
where ReH denotes the yielding stress for the material (MPa). examples and other useful knowledge. The second step is to choose
the optimum design model from design examples, and the final
step is to achieve automated design through a variety of knowledge
4. Standard parts library reasoning methods such as CBR and RBR. The design interface for
longitudinal structures and design parameters of target ship are
Although ship structures are very complicated and vary greatly shown in Fig. 5.
from each other, they are constructed by standard parts from the In the above interface, users input the target ship scantlings
same standard parts library. That is because the techniques and such as length, breadth, depth etc., and then the design system
manufacturing facilities restrains its products so that the parts will automatically choose an appropriate mother ship for reference
used in building ships are structures with discrete properties. in target ship's design. Thus the longitudinal structures' locations
Model components are stored as standard parametric model and boundaries will be set by appropriate topology adjustment
components in standards parts library. Once new structural parts according to target ship's scantlings. The remaining work is to
are introduced, the library will be updated as necessary. In the determine the final structural properties such as the thickness of
design process with CBR in this paper, the target ship's structural deck, the cross-section scantling of stiffeners and the application
component type will inherit the similar design case's (mother scope for each structural property. Generally speaking, there are
ship) while the component parameters will be adjusted due to two ways in dealing with this scantling design, i.e., the design
rules method and the interpolation method. Take the ship bottom
Table 1 plate as an example, the two methods are of different emphasis in
Material coefficient K L of longitudinal strength.
the implementation of design rules.
ReH ðN=mm2 Þ KL
(1) Design rules method:
235 1 In this method, we calculate the values of t 1 and t 2 according to
315 0.78
formula (1). Taking the material coefficient into accounts, both
355 0.72
390 0.68
the values of t 1 and t 2 will be adjusted and the maximum of
these two values will be chosen as the thickness of bottom plate.
So in this method, we apply the rules directly to determine the into 3D model in Tribon and Finite Element model in MSC/Patran.
structural properties. Then we resort to Nastran for direct strength analysis according to
(2) Interpolation method: CCS rules and Isight for reliability analysis. Software used in this
In this method, first, we calculate the values of t 1 and t 2 paper mainly includes: Tribon (by Kockums ComputerSystem AB),
according to formula (1). Taking the material coefficient into Patran/Nastran (by MSC Software), Isight (by Dassault Systemes),
accounts, both the values of t 1 and t 2 will be adjusted and we SJTU/Tribon (by Shanghai Jiao Tong University) etc.
can get the maximum of these two values. We call this the key
design factor. Then we select two ships from database for 5.2.1. Tribon modeling
reference, and the two ships are called mother ships. After After finishing the KBE design for container ship's cargo tank,
calculating the key design factors of the mother ships, we can we exported the design data from KBE system as text files. With
resort to interpolation of key design factors for the design of original programs created in Python, we can automatically import
target ship's bottom plate, as shown below. So in this method, the design result into Tribon. The developed model of container
we apply the rules indirectly to determine the structural ship cargo tank and its shaded viewport are shown in Fig. 9.
properties.
When the design process is completed, we can get the design (1) The model's longitudinal range is “1/2 cargo hold+1 cargo hold
results in Access tables. Then we convert design results information +1/2 cargo hold”, and horizontal range the breadth of ship, and
J. Cui, D. Wang / Ocean Engineering 72 (2013) 124–139 129
Assign
dimensions for
target ship
Mothership Select
database mothership
Abstracting
Mothership
mothership
database Topological
information
transformation
Inherit
structural
styles and
materials
Determining
structural
arrangements
for target ship
Design
method
selection
Determining Determining
Mothership scantlings for scantlings for
CCS Rules
database plates and plates and
stiffeners stiffeners
Select design
variables and
allowed values
for optimization
Optimization
with Genetic
Algorithm
vertical range the depth. Additionally, after FE analysis, only 5.2.4. Loads and boundary conditions in load cases
structures in evaluation area are considered, as shown in
Fig. 12. a. Longitudinal bending moment includes the static moment and
(2) The load cases that should be considered in FE analysis are wave induced moment. Static moment corresponds to the
listed in Table 2. maximum hogging moment in 0:4 L mid-ship range. In this
Where ds denotes the structural draught, P w denotes wave case, we can get an estimation of the static moment for target
crest caused pressure. ship by referring to the DNV's regulations on static moment
and mothership's static moment, as shown in formula (6).
M SðtarÞ
M SðtarÞ ¼ M SðmotherÞ ð6Þ
In LC4 and LC5, the rolling angle takes the minimum of 30 deg M SðmotherÞ
and tan 1 ð2ðDds Þ=BÞ; As to the container loads (20 ftor 40 ft) where M S denotes the corresponding static moment of mother-
in tank and on hatch cover, we take the maximum allowed weight ship or target ship and M S denotes the static hogging moment
of containers; LC1G, LC2G and LC3G correspond to LC1, LC2 and defined by Det Norske Veritas (DNV) (2013) as follows.
LC3, respectively, together with consideration of longitudinal
moment. M S ¼ ð0:12250:015C b ÞCL2 B kN=m ð7Þ
130 J. Cui, D. Wang / Ocean Engineering 72 (2013) 124–139
Fig. 9. Designed ship cargo tank and its shaded viewport in Tribon.
t
s′ ¼ s ð14Þ
tt c
Table 2
Load cases with corresponding loads.
Load cases description Load cases ID External loads Container loads Load condition type
1 void space of 40 ft container LC1 ds (1) P w (2) Void location – Void location – Symmetric
LC1G Other locations 40 ft Other locations 40 ft
LC2 ds (1) P w (2) Void location – Void location 40 ft Symmetric
LC2G Other locations 40 ft Other locations 40 ft
LC3 ds (1) P w (2) Void location – Void location 20 ft Symmetric
LC3G Other locations 20 ft Other locations 20 ft
Rolling with 1 void space of 40 ft container LC4 ds (4) – Void location – Void location – Asymmetric
Other locations 40 ft Other locations 40 ft
Rolling LC5 ds (4) – All locations 20 ft All locations 20 ft Asymmetric
Surging LC6 – – Loads caused by Symmetric
accelerations of
40 ft containers
Boundary constraint coefficients C 1 (in accordance with com- 5.2.7. Results of strength analysis
pression on short edge and shearing), C 2 (in accordance with Fig. 15 shows the shaded drawing of stress distribution inside
compression on long edge) are determined as Table 5. the model.
Then we should modify the critical buckling stresses as follows. Where, sL denotes the longitudinal stress; sW denotes the
horizontal and vertical stress; se denotes the von-mises stress for
s xcr shell elements or maximum combined stress for beam elements
8 and bar elements. τ denotes the average shearing stress along web
>
> s xcr_e when s xcr_e ≤ s2S height; λ denotes the buckling safety factor.
> ðycr_eÞ
>
>
> ðycr_eÞ
>
< 0 1 From Tables 7 and 8, we can conclude that both design
ycr ¼ methods are feasible solutions according to strength analysis by
> B C
>
> sS B sS C sS FE method, including both yielding strength and buckling analysis.
>
> @1 s xcr_e Awhen s xcr_e 4 2
>
>
: ðycr_eÞ
ðycr_eÞ
Table 4
Coefficients in corresponding loading conditions.
We employed the Multi-Island Genetic Algorithm (MIGA) to each island are migrated to another island, and then TGA are
optimize the structures in order to reduce the weight of ship cargo carried out again in each island with new individuals. As demon-
tank. MIGA is built based on traditional genetic algorithm (TGA). strated by Li et al. (2011), the iteration flow chart of MIGA is shown
TGA get inspiration from Darwin's theory of evolution. Each in Fig. 16. The basic parameter settings in this paper are shown in
candidate solution for the optimization problem is encoded like Table 10.
chromosome, then fitness for each solution is calculated which The optimization results for design rules method and Inter-
considers both the target function and violations of restraints. polation method are listed in Tables 11 and 12, respectively. After
Candidates for the next generation are generated by choosing from optimization, we get a weight reduction of 0.3% and 1.7%, respec-
current population of individuals according to fitness, crossing tively, for design rules method and Interpolation method, while
between individuals, and mutation of individuals. Finally the best the structural strength satisfies the rules requirements.
solution in the final generation is chosen as the optimum solution. We can conclude that the design solutions by design rules
The character of MIGA different from TGA is that the individuals method have economical advantage over that by design rules
are divided into several groups called “island”. All the operations of method, i.e., it provides solutions with lower cost. Furthermore,
TGA are carried out in each island, and the individuals selected in the solutions by Design Rules method are very close to the
optimum solutions in terms of design weight.
Table 5
Boundary constraint coefficients in different conditions.
Boundary conditions C1 C2
Population
End
initialization
In double bottom or side tank Elsewhere
Table 6
Buckling strength safety factor in different loading conditions.
Copy,
Stress conditions Plate length width ratio New
Mutation and
pffiffiffi pffiffiffi Population
1≤ ≤ 2l l
4 2 Crossover
s s
Table 7
Strength analysis results of solution from design rules method.
Structure name Checked load cases Allowed stress, N/mm2 Buckling safety factor
sL sW se τ λ
Table 8
Strength analysis results of solution from Interpolation method.
Structure name Checked load cases Allowed stress, N=mm2 Buckling safety factor
sL sW se τ λ
Table 9 Table 10
Design variables in structural optimization. Parameter settings in Multi-Island Genetic
Algorithm.
shell_bottom Thickness of bottom plate
shell_coam_flat Thickness of coaming flat plate Number of generations 10
shell_coam_web Thickness of coaming web plate Number of Islands 5
shell_inner_bottom Thickness of inner bottom plate Rate of crossover 1
shell_lbhd2 Thickness of longitudinal bulkhead above inner bottom Rate of migration 0.01
at y ¼ 7 11580 Rate of mutation 0.01
shell_strg_11580 Thickness of longitudinal bulkhead under inner bottom Sub-population size 10
at y ¼ 7 11580
shell_strg_4026 Thickness of longitudinal stringer at y ¼ 7 4026
shell_transtors_2 Thickness of transverse torsion box plate at ends along
breadth 5.2.9. Reliability analysis
shell_transtors_3 Thickness of transverse torsion box plate in middle of Due to the corrosion in sea water, the structural design
breadth
shell_wtbhd_4 Thickness of lower part of water tight bulkhead
scantlings are subject to reduction. To find the important design
stiff_bottom_2 Scantling of longitudinal section in bottom plate variables, we employed the sensitivity analysis to assess the
stiff_inner_2 Scantling of longitudinal section in inner bottom plate strength restraints that are most likely to fail (according to FE
stiff_lbl_3 Scantling of longitudinal section between 12196 and results) in the corrosion area. The results are shown in Fig. 17.
14800 platform
From the above figures, we can find that one or two specific
stiff_lbl_5 Scantling of longitudinal section between 6988 and 9592
platform design variables play the dominating role in the corresponding
strength restraint. So we can determine the design variables.
136 J. Cui, D. Wang / Ocean Engineering 72 (2013) 124–139
From the Sensitivity Analysis results, we can see that the safety we can assume that the safety factors are also independent from
factors depends mostly on one or two design variables, i.e., λinner each other, so we can simplify the reliability problem as follows.
shell_inner_bottom; λlbhd2 shell_lbhd2;
P saf e ¼ Pðλinner ≥1 and λlbhd2 ≥1 and λlsothers ≥1 and λstrg_11580 ≥1
λlsothers shell_strg_4026, shell_bottom; λstrg_11580 shell_strg_
and sbeam ≤231Þ
11580;
≅Pðλinner ≥1Þ Pðλlbhd2 ≥1Þ Pðλlsothers ≥1Þ Pðλstrg_11580 ≥1Þ
sbeam shell_bottom: Pðsbeam ≤231Þ ð17Þ
We can see that the safety factors depend on different design By calculating the corresponding area under probability density
variables, as the design variables are independent from each other, function curve, we can get that Pðλinner ≥1Þ ¼ 99:6%; Pðλlbhd2 ≥1Þ ¼
98:0%; Pðλlsothers ≥1Þ ¼ 100%;
Pðλstrg_11580 ≥1Þ ¼ 99:5%; and Pðsbeam ≤231Þ ¼ 83:8% for design
probability density function fi(x)
lower limit ai mean value μi P saf e ¼ 1:0 0:994 1:0 1:0 0:978 ¼ 97:2%:
design variable xi
We can conclude that the design solutions by Interpolation
Fig. 18. Probability density function of variable. method have much greater reliability than that by design rules
Probability distribution of inner bottom buckling safety factor Probability distribution of longitudinal bulkhead buckling safety factor
6 10
9
Probability distribution
Probability distribution
5
8
Monte Carlo Simulation points Monte Carlo Simulation points
4 Minmum square error fitting 7
Minmum square error fitting
6
3 5
4
2
3
1 2
1
0 0
1 1.05 1.1 1.15 1.2 1.25 1.3 1.35 0.95 1 1.05 1.1 1.15 1.2
buckling safety factor λinner buckling safety factor λIbhd2
Probability distribution
8
8
7
7 Monte Carlo Simulation points
6 Minmum square error fitting
6
5
5
4 4
3 3
2 2
1 1
0 0
1.08 1.1 1.12 1.14 1.16 1.18 1.2 1.22 1.24 1.26 1.28 1 1.05 1.1 1.15 1.2 1.25
buckling safety factor λlsothers buckling safety factor λstrg-11580
0.16
0.14 Monte Carlo Simulation points
0.12 Minmum square error fitting
0.1
0.08
0.06
0.04
0.02
0
224 226 228 230 232 234 236 238
stress in longitudinals and stiffeners σbeam
Probability distribution of inner bottom buckling safety factor Probability distribution of longitudinal bulkhead buckling safety factor
9
6
Probability distribution
Probability distribution
8
5 Monte Carlo Simulation points 7 Monte Carlo Simulation points
Minmum square error fitting
6 Minmum square error fitting
4
5
3 4
2 3
2
1
1
0 0
1.05 1.1 1.15 1.2 1.25 1.3 1.35 1.4 1 1.05 1.1 1.15 1.2 1.25
buckling safety factor λinner buckling safety factor λIbhd2
Probability distribution of longitudinal stringer2 buckling safety factor Probability distribution of longitudinal stringer buckling safety factor
12 9
Probability distribution
Probability distribution
8
10
7
Monte Carlo Simulation points
8 6 Monte Carlo Simulation points
Minmum square error fitting
Minmum square error fitting
5
6
4
4 3
2
2
1
0 0
1.15 1.2 1.25 1.3 1 1.05 1.1 1.15 1.2 1.25 1.3
buckling safety factor λIsothers buckling safety factor λstrg-11580
0.14
0.12 Monte Carlo Simulation points
0.1 Minmum square error fitting
0.08
0.06
0.04
0.02
0
220 222 224 226 228 230 232 234
stress in stiffeners and longitudinals σbeam
method, i.e., it provides better robustness in terms of sea water methodology. As expertise and design considerations are dealt
caused corrosion uncertainties. The solutions by Interpolation with as implicit knowledge, this design idea may be more
method may represent the considerations of robust design practical.
methodology.
Acknowledgements Jin-feng Chen, He-zhen Yang, Ru-hong Jiang, De-yu Wang, 2009. Application of
knowledge-based engineering methods for hull structure member design.
In: International Conference on Computer Application in Shipbuilding. pp. 87–93.
The present paper is supported by both the project of Ministry Cui, Jin-ju, Wang, De-yu, Xia, Li-juan, Ma, Chong, 2012. Mid-ship section structural
of Education and Finance (No. 200512) and the project of the State design and optimization based on knowledge based engineering. J. Shanghai
Key Laboratory of ocean engineering (GKZD010053-10). Jiao Tong Univ. 46, 368–373.
Kulon, J., Mynors, D.J., Broomhead, P., 2006. A knowledge-based engineering design
tool for metal forging. J. Mater. Process. Technol. 177, 331–335.
References Roh, M.I., Lee, K.Y., 2007. Generation of the 3D CAD model of the hull structure at
the initial ship design stage and its application. Comput. Ind. 58, 539–557.
Cai, Qian-ya, Qiu, Yong-ming, Lu, Wei-dong, 1997. Knowledge based CAD system for
Olsson, A., Sandberg, G., Dahlblom, O., 2003. On Latin hypercube sampling for
structural reliability analysis. Struct. Saf. 25 (1), 47–68. containership midship section structural design. J. Shanghai Jiao Tong Univ. 31,
Chapman, C.B., Pinfold, M., 2001. The application of a knowledge based engineering 61–64.
approach to the rapid design and analysis of an automotive structure. Adv. Eng. Sharma, R., Kim, Tae-wan, 2010. Development of a logic-based product life-cycle
Software 32, 903–912. management (LBPLM) System for Shipbuilding Industry—Conceptual Develop-
China Classification Society (CCS), 2005. Guidelines for Direct Strength Analysis of ment. J. Ship Prod. Des. 26 (4), 231–251.
Container Ship. China Communications Press. Sharma, R., Kim, Tae-wan, Lee Storch, Richard, Hopman, Hans J.J., Ove Erikstad,
China Classification Society (CCS), 2009. Rules and Regulations for the Construction Stein, 2012. Challenges in computer applications for ship and floating structure
and Classification of Sea-going Steel Ships. China Communications Press. design and analysis. Comput. Aided Des. 44 (3), 166–185.
Det Norske Veritas (DNV), 2013. Rules for Classification of Ships. Det Norske Veritas Kim, Tae-wan, Lee Storch, Richard, Hopman, Hans J.J., Ove Erikstad, Stein, 2012.
AS, (DNV). Editorial: applications in ship and floating structure design and analysis.
Lee, Dongkon, 2006. Knowledge-based system for safety control of damaged ship. Comput. Aided Des. 44 (3), 163–165.
Knowl. Based Syst. 19, 187–191. Zhou, X.H., Qiu, Y., Hua, G., 2007. A feasible approach to the integration of CAD and
Helvacioglu, S, Insel, M, 2003. An expert system approach to container ship layout CAPP. Comput. Aided Des. 39, 324–338.
design. Int. Shipbuilding Prog. 50, 19–34. Zhao, Zhen, 2002. Research on KBE Technologies for Intelligent Blanking Process
Li, Huili, Lang, Lihui, Zhang, Jianyong, Yang, Huai, 2011. Cost optimization method of Planning facing Theory and System of Innovative Design. Shanghai Jiao Tong
large-scale prestressed wire winded framework on Multiple-Island Genetic University Doctorate Dissertation.
Algorithm. Chin. J. Aeronaut. 24 (5), 673–680.