2016 International Conference on Industrial Informatics - Computing Technology, Intelligent Technology, Industrial Information

Integration

Ship Structural Members Design and Optimization Based on Knowledge-based

Engineering

Guan Guan Qu Yang

Department of Naval Architecture School of Distance and Continuing Education
Dalian University of Technology Dalian University of Technology
Dalian, China Dalian, China
[email protected] [email protected]

Abstract—It is always pursued that the excellent ship identified that the expert system can significantly improve
structural members are rapidly designed and modified on the the efficiency and quality of the ship design and integrated
premise of ensuring security in ship engineering. In this paper, design knowledge into design system and achieved a highly
design and optimization method for ship structure based on flexible design environment [6]. Lee applied knowledge-
knowledge-based engineering is put forward. Combining based engineering to the general arrangement of the engine
knowledge-based engineering with parametric technology, the room, making full use of engineering knowledge data [7].
knowledge base of ship structure design is established and 3D Wu proposed knowledge-based modeling for ship overall
rapid optimal design is achieved. During the design, positions design to achieve information storage and retrieval [8]. Chen
of ship structure components are driven by positional
applied knowledge-based engineering on ship deck design
parameters. Scantlings of ship structural components are
obtained by mother ships knowledge base and NURBS
[9]. Cui used KBE on midship section structural design [10].
interpolation, according to rules requirements. Main Cai developed a midship section design system based on
structures are optimized by quantum-behaved genetic AutoCAD, subdivided transverse section into several
algorithm. Examples show that this method achieves modules and obtained the min-area midship section [11].
knowledge accumulation and reuse, provides design results This paper introduces the concept of knowledge ontology
inspection, and realizes the obtainment of reliable ship for ship structure, combines knowledge-based engineering
structure. with parametric technology and establishes knowledge base
for ship structure design. During the design, positions of
Keywords- ship structure; knowledge-based engineering; structural components are driven by positional parameters.
quantum-behaved genetic algorithm; design and optimization Scantlings of structural components are obtained by parent
ship knowledge base and NURBS interpolation, according to
I. INTRODUCTION specification requirements. Main structures are optimized by
quantum-behaved genetic algorithm.
The ship design quality and efficiency affect ship
development cycle. As ship design is a complicated process II. THE BASIC PRINCIPLES
with many labor-intensive activities, how to develop and
modify the ship structure rapidly is a major problem for ship A. Knowledge-based engineering (KBE)
designers [1]. KBE is a new branch of science researching knowledge
Traditional ship structural design focuses on dealing with information, which abstracts the common problems during
data. It is a kind of interactive graphics operation, and the intelligent system research and development [12]. The
results that designers work out describe only visual shape essence of KBE is to excavate and accumulate knowledge in
and size of ship structure, while related knowledge including product development, then express, apply and maintain
design idea, expertise, parent ships information and rules are knowledge to achieve automatic solution for related
not reflected. Obviously, it’s not convenient for designers to problems.
modify and check the design model [2]. To solve this
problem, knowledge-based engineering (KBE) method is B. Knowledge ontology
applied on ship design to integrate design methods, rules and As there are so many ship forms, it’s impossible to
expertise experience into design process. Based on the develop the specific design module for each ship. However,
related knowledge, system can automatically offers the high according to their forms, ships can be induced into several
qualified design model rapidly after designers input groups. Ships in the same group have similar structural
parameters and requirements. forms. Generally, a new ship is designed according to the
Currently, KBE is well developed in vehicle and parent ship and its structural design is based on the parent
machinery [3]. However, because of the variety and ship structure. In view of above facts, the concept of
complication of ship design, application of KBE on ship knowledge ontology for ship structure is proposed.
industry is still at the beginning. University of Rostock in Knowledge ontology refers to a 3D parametric hull
Germany planned the knowledge based system for ship structural model for one specified ship group. For this model,
modeling in research programs, which could obtain design common structures of ships in group are abstracted and
criteria and provide tools from the knowledge base [4]. Park stored while special structures of ships are wiped out. In that

978-1-5090-3575-5/16 $31.00 © 2016 IEEE 268

DOI 10.1109/ICIICII.2016.20
way, a new ship structure belonging to the same ship group C. Rule base
can be quickly developed based on the knowledge ontology. Case base is a collection of parent ships.
III. KNOWLEDGE BASE CONSTRUCTION
The knowledge base is defined as a collection of
experience, rules, cases and other knowledge. Knowledge
base includes theoretical knowledge, technical rules, expert
experience, successful cases and scantling standards, etc. All
those knowledge are collected, organized and summarized
into a number of rules, analysis and problem-solving
strategies, and are placed in a particular form of the
document or database to constitute a knowledge base, which
can achieve storage and classification management of
Figure 2. Plate Thickness Figure 3. Flat-bulb steel
product design knowledge and provide best guidance and
recommendations for the designer during the design process
[13]. Fig. 1 shows the knowledge base structure.

Figure 1. Structure of knowledge base

Figure 4. Parameter Spreadsheet of Flat-bulb steel
A. Ship structural component library
Ships belong to complicated structures with different IV. KBE STRUCTURAL DESIGN
kinds of construction components and large scantling The first step of ship structural design is to establish a
information. However, there are many similar structural ship parametric modeling component library, extracting the
parts. Therefore, according to the structural component characteristic parameters for parametric model components.
features, the ship structural part library is established to store Second, sum up experts knowledge, design methods and
characteristic structural part information. The knowledge professional knowledge as knowledge rules, formulas,
base is embedded in the component model constructing checks and rules, behavior or other knowledge in a machine-
standard parametric model component library to store interpretable representation in order to assist designers to
standard, scanting constraints, and size information for the complete and check the design.
ship design process. New structural components can be Knowledge-based reasoning methods consist of rule-
adjusted in parameters and designers can update the standard based reasoning (RBR) and case-based reasoning (CBR).
parts library whenever necessary. The method of RBR is mainly used for specific parameters
In this paper, different structural component libraries, based on a knowledge adviser and knowledge expert work
including plate library, profile library, thru-hole library, system, such as rule reasoning. The method of CBR is used
bracket library and profile end-cuts library, are established. for component design similar with original case based
Spreadsheet is used for plate parameter design. Plate component knowledge templates, such as reasoning from
thickness are stored in spreadsheet and connected with plate parent ship information.
shapes. Designers can obtain different plates by modifying During the design, structural component positions are
the parameters in spreadsheet, as shown in Fig.2. driven by positional parameters. Scantlings are obtained by
Similarly, spreadsheet is used for profile parametric RBR and CBR method. Main structures are optimized by
design. Profile library consists of profile T, angle steel, flat- quantum-behaved genetic algorithm. Flow chart for
bulb steel and flat steel, etc. The main parameters include knowledge-based hull structural design is shown in Fig.5.
profile number, web height, web thickness, panel thickness
and panel width, etc. The sketch of flat-bulb steel is shown in V. DESIGN CASE
Fig.3. Its parameter spreadsheet is shown in Fig.4.
A. Structure component position
B. Rule base
The main structure component form is stiffened plate,
For ship structure design, rule base is a collection of
that is, a plate with several stiffeners welded on it. The plate
design codes and standards, expertise experience, etc. The
position should be confirmed at first. Then the accessory
knowledge are sorted and summarized to assist designers to
structural components (including stiffeners, openings, etc)
complete and check the design. Detailed process of building
are confirmed based on the plate position.
rule base was introduced in design case.

269
 The plate position is described by theoretical plate and 1) Parameter construction: According to Rules for
plate boundary. The theoretical plate is obtained by inputting Classification of Sea-going Steel Ships by China
positional parameters or transforming operations of other Classification Society (CCS) [14], the thickness of
plates. For example, theoretical plate for inner bottom plating longitudinally-framed sheer strake within 0.4L amidships
is described by 1800-height; 1700-mid-away side girder
should not less than:
theoretical plate is described by 1700-offset from middle
girder; Plate boundaries can be confirmed by plane equation,  t1 = 0.06s( L1 + 110 ) Fd (1)
ship shell or other plate group.  t2 = 0.9s L +75 (2)
The accessory structural components (including
stiffeners, openings, etc) are confirmed by parametric Where, s is side longitudinal spacing; L is ship length;
method. Three kinds of constraints are added as follows: L1=L, not larger than 200m; Fd is reduction coefficient The
• Subordinate constraints, that is, the accessory parameter values are shown in Fig.6.
structural components move along with the plate.
• Distance constraints, that is, the distance of the
accessory structural component from specific
position of plate is fixed.
• Boundary constraints, that is, the boundary of plate
is looked as the boundary of accessory structural
components.
The positions of the accessory structural components are
confirmed by solving above geometrical constraint system.

Figure 6. Sheer strake parameters

2) Parameter equation: According to the formulas in
rule, parameter equations are confirmed, as shown in Fig.7.

Figure 7. Sheer strake parameters relationship schematic diagram

3) Thickness rule construction for Sheer strakes: Rule
editor is used to establish the rule shown in 1) and store into
rule base in order to recycle. Once the design parameter
values change, satisfying results will be calculated
automatically according to rule base. Thickness rule for
sheer strakes are shown in Fig.8.

Figure 8. Sheer strake thickness rule schematic diagram

4) Thickness check for sheer strake: Thickness check is
performed to make sure the design result satisfying CCS
Figure 5. Flow Chart for Knowledge-based Ship Structural Design rules. The check for sheer strake thickness is illustrated in
Fig.9. When sheer strake thickness does not satisfy the
B. Rule-based reasoning (RBR)
requirements, the system gives warnings accordingly, as are
First, the basic parameters are defined. Then thickness of illustrated in Fig.10.
the plates and profile attributes of stiffeners are obtained Other ship structure components are similar to sheer
according to the classification society rules. Second, strake design process, so that they are not covered again
relationships between parameters are confirmed according to here.
the formulas in rules. And the rule base is established. The
results for each parameter value are automatically got.
Simultaneously, the inspection is also built to check the
results. If the results are not satisfied, designers will be
prompted to learn the questions and modify them in time.
Take sheer strake design as an example.

270
 Figure 9. thickness check for Sheer strake interpolation rule for strength is built according to parent
ship parameters, as shown in Fig.14. The interpolation curve
is built by parent ship parametersNW ǃ NW ǃ
NW. The strength deck thickness W is obtained by
interpolation. The above rule is stored in rule base to be
reused. The design case is calculated automatically once the
design parameter or parent ship information changes.
Figure 10. Thickness check message for Sheer strake

C. Case-based reasoning (CBR)

CBR reasons scantlings of the structural components Figure 12. Deck parameters relationship schematic diagram
according to parent ship form information. N similar parent
ships are chosen. We take the strength deck as an example as
follows:
According to the rules, thickness of longitudinal framing
strength deck should not less than:
 t1 = 0.06s( L1 + 110 ) Fd (3)
 t2 = 0.9s L +75 (4) Figure 13. Deck key value rule schematic diagram

Where, s is side longitudinal spacing; L is ship length;

L1=L, not larger than 200m; Fd is reduction coefficient
Take the larger value between t1 and t2. If t2 is larger, take
(4) as the key formula. Take k = s L +75 as the key value of
strength deck. Then thickness of strength deck can be
calculated by NURBS interpolation. Based on the degree
elevation, triple NURBS function (shown as (5)) is used to
express both linear and quadratic interpolation. In that way,
NURBS interpolation can be generally used with even only 2
or 3 parent ships. Figure 14. Mother parameters schematic diagram
n

 ª k ( u ) º ¦ B i , 3 ( u )W iV i (5) t3
deck t hi ckness t

« t (u ) » =
i=0
n t
¬ ¼
¦i= 0
B i ,3 ( u )W i t2
t1
Where, Vi is control vertex, which is back calculated by
parent ship information, Wi is weight factor, Bi ,3 (u ) is triple
B-spline primary function, u is node, which is obtained by 0 k1 k2 k k3
accumulated chord method. key val ue k 
Detailed design procedure of CBR method is shown as Figure 15. Deck thickness rule schematic diagram
following. 4) Thickness check for sheer strake: See RBR method.
1) Parameter construction: Fig.11 shows the parameters CBR uses formulas of rules for design results checking,
of strength deck. and CBR obtains the initial scantlings by parent ship base
interpolation. Because scantlings of parent ship contain
expertise experience, CBR also effectively inherits and
reuses original knowledge.
D. Structural scantling optimization based on knowledge
Under the premise that all structure component positions
are not changed, parametric model is transferred into
knowledge-based optimization model based on the
knowledge.
1) Optimization Model: The model is built as follows:
Figure 11. Deck parameters schematic diagram
Design variable: plate thickness, profile section shape
2) Parameter equation: According to the rules, Optimization goal: structure weight minimization,
n m
parameter equations are built, as shown in Fig.12.
3) Rule base definition for strength deck: Rule editor is
 min W = ¦ pi + ¦ s j (6)
i =1 j =1
used to establish key value rule, as shown in Fig.13. Then

271
 Where, W is the structure weight; n is plate number; m is  F (x ) = f (x ) + λ ¦ J max[0, g j (x )](8)
j =1
profile number; pi is weight of plate i ; sj is weight of profile
j; Where, x is design variable, x = ([ x c ],[ x d ])T , [ x c ] is
Constraint condition: the classification society rule, continuous design variable set, [ x d ] is discrete design
buckling strength requirements and construction technology variable set. f ( x ) is objective function; g j (x ) is constraint
requirements.
2) Knowledge engineering application for structure size condition j. λ is penalty factor, λ = 0.8 ; J is sum of
optimization: constraints.
a) Component library application: Ship structure E. Result comparison and analysis
components are mostly standard. Based on component Table 1 shows the longitudinal scantlings design result
libraries, the design variable parameter set is stored in comparison for oil tanker. The dimension information of the
spreadsheet to implement the management of the different oil tanker is: Length between perpendicular: 190 m, Breadth:
sizes of plates and profiles. 32.2m, Draft: 14.2m, Depth: 18.2m. Deck and bottom are
b) Rule base application: A large number of equations made of high-strength steel, other parts are made of ordinary
for goal functions and constraint conditions is stored in rule steel. The original parameter for quantum-behaved genetic
base using KBE. Knowledge in rule base can be reused by algorithm is: Population size: 40, genetic algebra: 200. t4~ t9
designers to avoid repeating equations. is thickness of deck, sheer strake, side plate, bilge strake,
bottom plate and keel plate. S is section area. As shown in
3) Optimization method: Table 1, the result of section area by CBR method is larger
Ship structural scantling optimization belongs to mixed than by RBR method. RBR method is better considering
integer programming. The design variables include discrete economy. However, CBR method achieves the inheritance
and continuous variables. Plates are chosen from thickness and reuse of existing knowledge thus avoids the loss of
library, stiffeners are chosen from standard profile library. knowledge. The section area is reduced after optimization.
For profile T, its thickness of web and panel, which are Fig.17 shows the evolutionary process of quantum-behaved
chosen from thickness library, are discrete variables, while genetic algorithm (QGA) and standard genetic algorithm
the width of panel and web height are continuous variables. (SGA). As shown in Fig.17, QGA is superior to SGA in
Genetic algorithm can solve the problem [15]. However, convergence speed, so combining quantum computation with
standard genetic algorithm has poor local search capacity genetic algorithm can effectively improve the search
[16, 17]. In order to improve the design variable search performance of genetic algorithm.
capacity of complex ship structure, quantum genetic
algorithm is used which is suitable for mixed variable [18].
TABLE I. 2 REASONING SOLUTIONS AND OPTIMIZATION RESULTS
Quantum-behaved genetic algorithm introduces the
universe concept into the genetic algorithm, which has better t4 t5 t6 t7 t8 t9 S
species diversity and higher search ability than the standard item
/mm /mm /mm /mm /mm /mm /m2
genetic algorithm. Quantum-behaved genetic algorithm
improves the good global search ability and convergence RBR 16 17 19 20 20 22 3.521
speed of the algorithm, which is suitable for ship structure CBR 18 18 20 22 22 24 3.684
global optimization. The calculation process is shown in Fig. optimal value 18 15 18 20 21 21 3.487
16.
Optimal fitness value of each generation

Figure 16. Flowchart of quantum-behaved genetic algorithm

As shown in Fig.16, quantum-behaved genetic algorithm
uses quantum rotating gate to update the individual of
population. Evolution generations
§ α i' · § cosθ − sin θ ·§ α i · (7) Figure 17. the QGA and the SGA evolutionary process
¨ ¸=¨
¨ β ' ¸ ¨ sin θ ¸¨ ¸
© i¹ © cosθ ¸¹¨© β i ¸¹
Where, (α i , β i ) is probability amplitude of chromosome
T F. 3D model
The ship structure component positions and sizes were
quantum bit rotation gate I before update. (α , β ) is ' ' T
i i obtained according to above process. 3D model was
probability amplitude of chromosome NO. i quantum bit established by knowledge and feature modeling technology.
rotation gate I after update. θ is rotate angle. Fig. 18 shows the structure subsection model for oil tanker.
Fitness function is built by penalty function method:

272
 [2] D. Zhao and H. M. Kong. “Knowledge-Based Ship Structure
Parametric Design,” Shipbuilding Technology, vol. 23, Mar. 2005,
pp. 9-11.
[3] C. N. Jin, B. Liu, Y. Shi, et al. “A Study of Geometric Issue in CAD
System for Car-Body Parts Design,” Journal of Jilin University, vol.
36, Mar. 2006, pp. 116-119.
[4] M. Zimmermann and R. Bronsart. “Application of Knowledge-Based
Engineering Methods for Standardization and Quality Assurance in
Ship Structural Design,” 2nd World Maritime Technology
Conference, London, Dec. 2007, pp. 57-64
[5] J. H. Park and R. L. Storch. “Overview of Ship-design Expert
Systems,” Expert Systems, vol. 19, Sep. 2002, pp. 136-141.
[6] S. Akagi and K. Fujita. “Knowledge Based Geometric Modeling
System for Preliminary Design Using Object-Oriented Approach,”
Montrea,l Que, Can: Publ by ASME, New York, NY, USA. 1989.
Figure 18. Hull block structure schematic diagram
[7] K. H. Lee and Y. S. Yeun. “Data Analysis and Utilization Method
VI. CONCLUSIONS Based on Genetic Programming in Ship Design,” Computational
Science and Its Applications-Iccsa 2006, Berlin, Dec. 2006, pp. 1199-
Knowledge-based engineering design for ship structure 1209.
has been discussed in this paper. The concept of knowledge [8] Y. H. Wu and H. J. Shaw. “Document Based Knowledge Base
ontology for ship structure was proposed. And knowledge Engineering Method for Ship Basic Design,” Ocean Engineering, vol.
base for ship structure design was established combining the 38, Sep. 2011, pp. 1508-1521.
parametric technology with knowledge-based engineering [9] J. F. Chen, H. Z. Yang, R. H. Jiang, et al. “Application of
Knowledge-Based Engineering Methods for Hull Structural Member
theory. Knowledge ontology model was built with Design,” Ship Science and Technology, vol.32, Oct. 2010, pp. 16-20.
information (including specification requirements and parent [10] J. J. Cui, D. Y. Wang, L. J. Xia, et al. “Mid-Ship Section Structural
ship, etc) embedded. During the design, structural Design and Optimization Based on Knowledge Based Engineering,”
component positions were driven by parameters. Structural Journal of Shanghai Jiaotong University, vol.46, May 2012, pp. 368-
scantlings are obtained by RBR and CBR methods. 373.
Quantum-behaved genetic algorithm was applied to optimize [11] G. Y. Cai, Y. M. Qiu and W. D. Lu. “CAD System for Mid-Ship
main structures to achieve the better results. Section Structural Design of Container Ship Based on Knowledge,”
Journal of Shanghai Jiaotong University, vol. 31, Nov. 1997, pp. 61-
This method achieved knowledge reuse and 64.
accumulation, provided design results inspection, so that
[12] C. B. Chapman and M. Pinfold. “The Application of A Knowledge
reliable ship structure is quickly obtained automatically, Based Engineering Approach to the Rapid Design and Analysis of An
reducing the probability of design errors. Automotive Structure,” Advances in Engineering Software, vol.12,
For structural component optimization, the positions are Dec. 2001, pp. 903-912.
first determined by parameter driven then scantlings are [13] H. Z. Yang and L. J. Xia. “Research Progress of Knowledge Based
optimized. However, the positions and scantlings of Engineering in Ship Design and Construction,” Ship Engineering,
vol.30, May 2008, pp. 7-12.
structural component are not involved in the optimizing
model, which should be further studied in future. [14] China Classification Society. “Rules for classification of sea-going
steel ships,” Beijing: China Communications Press, 2009.
ACKNOWLEDGMENT [15] G. Renner, A. Ekart. “Genetic Algorithm in Computer Aided
Design,” Computer Aided Design, vol.35, May 2003, pp.709-726.
The authors are grateful to Natural Science Foundation [16] C. Aladahalli, J. Cagan and K. Shimada. “Objective Function Effect
of China (51609036), China Postdoctoral Science Based Pattern Search Theoretical Framework Inspired by 3d
Foundation (2014M561234, 2015T80256) , the Doctoral Component Layout,” Journal of Mechanical Design, vol. 129, May
Starting up Foundation of Liaoning Province (201501176) 2007, pp. 243-254.
and the Fundamental Research Funds for the Central [17] K.S.N. Pawan and D. Kalyanmoy. “A Computationally Effective
Multi-Objective Search and Optimization Technique Using Coarse-
Universities (DUT16RC(4)26) for their financial supports.
to-Fine Grain Modeling,” 2002 PPSN workshop on evolutionary
multi-objective optimization, Spain: Springer, Dec. 2002, pp. 2081-
REFERENCES 2088.
[18] A. Narayanan and M. Moore. “Quantum-Inspired Genetic Algorithms
[1] H. Z. Yang, J. F. Chen, N. Ma, et al. “Implementation of Knowledge- ,” Proceedings of IEEE International Conference on Evolutionary
Based Engineering Methodology in Ship Structural Design,” Computation, Nagoya, Japan, Dec. 1996, pp. 61-66.
Computer-Aided Design, vol. 44, Sep. 2012, pp. 196-202.

273