An expert system using rough sets theory for aided ... - Semantic Scholar

Available online at www.sciencedirect.com

Expert Systems with Applications Expert Systems with Applications 36 (2009) 3223–3233 www.elsevier.com/locate/eswa

An expert system using rough sets theory for aided conceptual design of ship’s engine room automation Xin-Yu Shao a, Xue-Zheng Chu a, Hao-Bo Qiu a, Liang Gao a,*, Jun Yan b a

State Key Laboratory of Digital Manufacturing Equipment and Technology, Huazhong University of Science and Technology, 1037, Luoyu Road, Wuhan, Hubei 430074, PR China b China Shipbuilding Industry Corporation Wuchang Shipyard, 2, Ziyang Road, Wuhan, Hubei 430060, PR China

Abstract More and more complicated conceptual design of ship’s engine room (CDSER) heavily depends on designers’ engineering knowledge and existing ship data. To achieve intelligent design at the initial ship design stage, many researchers have made much significant progress in this field, however, most of them only focused on how to find the similar constructed ships. At present, how to utilize these existing data remains an untouched topic. In order to make good use of the existing data and reduce the dependence on designers’ experience, a novel system named Expert System for Aided Conceptual Design of Ship’s Engine Room Automation (ESACD), is elaborated in this study. With the support of the constructed Ship Data Warehouse System, two core subsystems Configuration Selection Assistant (CSA) and Design Scheme Decision Assistant (DSDA) are included in ESACD. A promising approach integrating Fuzzy c-means algorithm (FCM) and Rough Sets Theory (RST) to extract configuration rules from the stored data is adopted in CSA. According to engineers’ proposals, RST is utilized to reason knowledge in incomplete scheme information systems for getting design scheme rules in DSDA, which are useful suggestions for engineers to get better schemes at this stage. Finally, the validity and necessity of this interactive expert system are demonstrated through the CDSER of a new 50,000 DWT Handymax bulk carrier. It is proved that ESACD can efficiently facilitate rapid and intelligent design in CDSER, and reduce the cost of a new ship design. Ó 2008 Elsevier Ltd. All rights reserved. Keywords: Conceptual design of ship’s engine room (CDSER); Rough sets theory (RST); Discretization; Configuration of engine room; Design schemes

1. Introduction Conceptual design of ship’s engine room (CDSER) is one of the most important tasks in ship preliminary design, in which many perplexing subsystems should be carefully considered, such as main propulsion system, electric power system, fuel system and piping-system. The early stage of ship development includes a lot of fuzzy problems and tolerates a high degree of uncertainty. At this stage, it is inevitable that it heavily depends on the experts’ experiences and design cases of existing ships because available infor*

Corresponding author. Tel.: +86 27 87543871 85; fax: +86 27 87543074. E-mail address: [email protected] (L. Gao). 0957-4174/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.eswa.2008.01.011

mation is limited and the degree of design freedom is very high (Lee & Lee, 1999). Ship design researchers have indulged themselves in the field of computer system for aided ship design automation. The method of using computer system for simulation and detailed design of the ship subsystem has been broadly investigated (Arendt, 2000, 2004; Arendt & Kowalski, 1999; Kowalski, Arendt, Meler-Kapcia, & Zielin´ski, 2001). On the basis of these systems developed for aided ship design, the integrated system for ship design has also received much attention (Kim, Lee, Park, Park, & Jang, 2002; Han & Lee, 1996; Parsons, Singer, & Sauter, 1999). These integrated systems provide the precondition of collecting and sharing of various data on built ships in the whole ship development process.

3224

X.-Y. Shao et al. / Expert Systems with Applications 36 (2009) 3223–3233

Generally, the computer systems for aided conceptual ship design are relatively few, because knowledge of the design requirements and constraints in this early phase of a product’s life cycle is usually imprecise and incomplete, making it difficult to utilize computer-based systems or prototypes (Hsu & Liu, 2000). However, with the rapid development of artificial intelligence, utilizing expert system technology in conceptual ship design has been accepted gradually, and researchers have made much progress in this field. For instance, the application of Memory-based learning (MBL) was proposed to access reference cases for a new ship design (Lee, Kang, Ryu, & Lee, 1997). Furthermore, an intelligent assistant system was developed based on the MBL method for ship design. (Lee & Lee, 1999). For the purpose of improving the result of using MBL and meeting designers’ requirements better, fuzzy method with trapezoidal, triangular and Gaussian function was carefully studied (Kowalski, Meler-Kapcia, Zielin´ski, & Drewka, 2005). Whereas, all of these literatures are only concentrated on the methods of selecting similar constructed ships, none of them investigated the methods of handling the existing design information for getting a better design scheme under uncertain circumstances. Since conceptual ship design is a burdensome design process and lots of information should be carefully considered, it is unacceptable that expert system only provide similar built ships for the ship designers. Thus, constructing an expert system to deal with the multifarious data about the similar built ships, from which useful knowledge can be extracted for the engineer, is a new field that is worth to be deeply studied. In order to achieve the goal of mining useful knowledge for assisting design engineers in CDSER, an expert system based on Rough Sets Theory (RST) (Pawlak, 1982) is elaborated in this paper. Pawlak (1982) first proposed RST, which has been extended and applied in many research fields, such as chemic analysis (Liu, Xiang, & Qu, 2007), customer relationship management (CRM) (Tseng & Huang, 2007), accident analysis (Wong & Chung, 2007), travel demand analysis (Goh & Law, 2003) and other uses (Pawlak & Skowron, 2007). The merit of RST is that, it neither needs additional information about the data nor is necessary to correct the missing or incomplete data of attributes. Instead, rules generated are categorized into certain rules and generalized rules. In this paper, an elaborated expert system, Expert System for Aided Conceptual Design of Ship’s Engine Room Automation (ESACD) is proposed. It is composed of Configuration Selection Assistant (CSA) and Design Scheme Decision Assistant (DSDA). In CSA, following the traits of CDSER, a promising approach that combines Fuzzy c-means clustering algorithm (FCM) and RST is utilized to mine configuration knowledge from similar previous ships’ solutions. In DSDA, the method of using RST to compute rules in incomplete information systems, which can acquire optimal certain or optimal generalized rules, is adopted to get conceptual design scheme rules. The rules

can substitute engineering knowledge to a certain extent, thus, ESACD effectively speeds up the CDSER process. The remainder of this paper is organized as follows. The traits of CDSER are discussed in Section 2. The innovative system is established in Section 3. In Section 4, the detailed steps and application of integrating FCM and RST to capture configuration rules in CSA are discussed, then, applying RST approach to reasoning rules in incomplete information systems in DSDA is explained in Section 5. In Section 6, the CDSER of a new 50,000 DWT Handymax bulk carrier is taken as a case study to prove the validity and necessity of the expert system. Conclusions are given in Section 7. 2. Conceptual design of ship’s engine room The quality of conceptual design of ship’s engine room (CDSER) has significant effects on the cost and performance of new ships, and plays the most crucial role in the whole ship development process. Fig. 1 illustrates the importance and complicacy of every design stage in the whole life cycle of ship design. It can be seen that the impact of decision and design freedom at conceptual ship design stage are the most important factors in the whole ship development process, but the development tools and design knowledge at this stage are much fewer. As mentioned earlier, knowledge of the design requirements and constraints in this early phase of a ship’s life cycle is usually imprecise, incomplete and uncertain. Obviously, CDSER has all the traits of conceptual ship design, because it is one of the most significant tasks in conceptual ship design process. Not only the traits discussed above, it also has its own ones as below: High complexity: Several interrelated subsystems, dozens of devices and over ten thousand meters piping-system will be preliminarily designed in engine room.

Fig. 1. Importance and complicacy of every design stage in the whole life cycle of ship design.


No additional data: No additional data can be referred in CDSER, except for costumers’ requirements and regulations of classification societies. The need for more expertise: Making all of the decisions, including selection of equipments, location of equipments, arranging the transfer of energy, routing piping-system and so on, commonly relies on engineers’ engineering knowledge. The need for high reliability and safety: Since most of mechanical, power and energy transfer equipments are in engine room, the reliability and safety of CDSER directly impact on the performance of new ships. Thus, an expert system that could mine design knowledge from raw data for improving design quality and reducing development time in CDSER is necessary. RST,

3225

which is famous for its significant ability of exacting rules from data and classifying them, has absorbed more and more attention in expert system field. Accordingly, a new application of RST in CDSER is introduced in the following section. 3. Expert System for Aided Conceptual Design of Ship’s Engine Room Automation In accordance with the traits of CDSER discussed earlier, for achieving rapid and intelligent design in the initial ship design stage, the innovative system, Expert System for Aided Conceptual Design of Ship’s Engine Room Automation (ESACD) is elaborated in this section. ESACD includes two central subsystems (CSA and DSDA), two interfaces (CBR Interface and Output Interface), Case Base

Fig. 2. Architecture and function configuration of Expert System for Aided Conceptual Design of Ship’s Engine Room Automation (ESACD).

3226


Reasoning Analyzer, Case Base, Knowledge Base and Ship Data Warehouse System. The architecture and function configuration of ESACD are illustrated in Fig. 2. As shown in Fig. 2, for the purpose of mining deep-hidden design knowledge to assist designers in quickly and correctly making decisions, Ship Data Warehouse System that provides a general data warehouse environment for ship design must be established at first. It is composed of Data Warehouse Manager System and On-Line Analytical Processing (OLAP), also contains source data (all information of constructed ships that will be analyzed). It is the foundation of all the other components and subsystems in this expert system. Users manage the data of constructed ships within it. Parts of databases in Ship Data Warehouse System are organized as follows:

selection of correlative attributes of the specific device or subsystem in engine room, and discretizing them in accordance with requirements, combination of diverse configurations in engine room for getting preliminary design schemes, drawing sketches and estimating the design schemes from all aspects, calculation of the configuration rules and the conceptual design scheme rules, modification of the conceptual design schemes with the obtained rules, and providing information of final conceptual design scheme of engine room through the graphical interface and the simulation interface to graphical tools and simulation tools, respectively.

Ship database: type, size, speed, displacement, auxiliary equipments, materials, length between perpendiculars, breath, depth, draft and etc. Engine database: the information of main engines (ME) (type, number, power, speed, weight, price, volume, oil consumption, reliability, manufacturer and so on), owner’s comments of ME and etc. Electric power plant database: the information of main power generators (PG1) (number, type, power, speed, manufacturer and so on), the information of auxiliary power generators (PG2), type of shaft generator and etc. Fuel system database: the information (type, name, manufacturer and so on) of fuel valves, fuel pumps, centrifuges, filters and etc. Piping-system database: all the information of pipingsystem (name, size, type, materials, length, max-pressure value and so on) and etc. Report and document database: the information about work report, result document, technical document, specification document and etc. Classification society database: country, names of classification societies, documents and regulations of classification societies and etc.

4. Relevant methodologies and detailed application steps in Configuration Selection Assistant 4.1. Configuration of conceptual design of ship’s engine room In general, the choice of appropriate equipments is essential for putting forward a good initial design, which affects the quality of CDSER. When customers order new ships, they usually only offer some kinds of general description, such as type, displacement, cost and speed. However, in CDSER, to satisfy ship owners’ needs and regulations of classification societies, the ship designers have to choose and locate many complicated equipments or subsystems, also confirm the configuration relations among them. Making all the decisions of configurations depends on engineers’ experience. Thus, it is great practical that an intelligent knowledge expert system is developed to assist with configuration selection. According to the traits of CDSER and the ability of Rough Sets Theory (RST), a promising approach that integrates Fuzzy c-means clustering algorithm (FCM) and RST is applied in Configuration Selection Assistant (CSA) for mining configuration rules. 4.2. Discretization of attributes in CSA

With the support of Ship Data Warehouse System, the fuzzy theory (Kowalski et al., 2005) is utilized in Case Base Reasoning Analyzer to locate existing similar ships (similar cases) for design engineers. Then relevant information is transmitted to CSA and DSDA through CBR Interface, so that configuration rules of constructed engine rooms and conceptual design scheme rules are captured in CSA and in DSDA, respectively. Main functions of this system are summarized as following: collecting and sharing all types of data about already built ships, and classifying them into different databases, choice and management of the similar cases, management of the technical and specification documents, work reports, regulations of classification societies and result documents during the entire CDSER process,

In CDSER, various attributes of equipments or subsystems have to be thought over by ship designers, some of them are non-numerical, while others may be numerical. RST cannot deal with the continuous attributes, which is the disadvantage of this theory, so one of the crucial problems is finding a suitable method for discretizing these continuous attributes. There are three different axes by which discretization methods can be classified: global vs. local, supervised vs. unsupervised and static vs. dynamic (Dougherty, Kohavi, & Sahami, 1995). When the design engineers refer to the similar cases, the values of features are usually within a part of their global value space. Applying new technology and new requirements might make data have a wider value domain, what’s more, useful knowledge about discretization of these features has never been stored in computer


3227

systems, and thus, the discretization method in ESACD belongs to both unsupervised and local method. As Dougherty summarized (Dougherty et al., 1995), k-means clustering is the algorithm that falls into these two categories. Fuzzy c-means clustering algorithm (FCM) is an improved algorithm of k-means clustering, first proposed by Bezdek (Bezdek, 1973). Compared with other algorithms, it does not need lots of the user’s experience except for the number of clusters assigned by the user at the beginning. Consequently, applying FCM to discretization of features is the appropriate approach in CDSER. The definitions and application steps of FCM in CSA are summarized as below:

Step 3 : Assign parameters c,q, e for the given attribute. Step 4 : Initialize the degree of membership matrix u0ij , which denote the relation between each attribute point and the centroid of its cluster at the beginning. Step 5 : Update each centroid of the attribute clusters with u0ij and formula (2). ðLþ1Þ Step 6 : Calculate uij , which denote the relation between each attribute point and its centroid in the update process. ðLÞ ðLþ1Þ Step 7 : If max½kuij uij k 6 e stop, all attribute points of the given objects in engine room are separated into c attribute clusters. Otherwise go to step 5.

X = {x1, x2, . . . , xn}: the sample set of the specific attribute of the objects (the given equipments or subsystems in engine room), xj = (xj1, xj2, . . . , xjk): the jth k-dimensional vector of each attribute point, c: the number of clusters given by ship designers, vi: the centroid of the ith attribute cluster composing of V = (v1, v2, . . . , vc), q: any real number greater than 1, weighting exponent for uij and controlling the ‘‘fuzziness” of the resulting attribute clusters, e: the termination criterion between 0 and 1 given by ship designers. kxj vik2: the Euclidean distance between the jth attribute point and the centroid of its cluster.

With this algorithm, numeric attributes can be effectively discretized in conceptual design environment. In reality, lots of continuous features have to be discretized to make preparation for utilizing RST.

Define the degree of membership of all attribute vectors in each attribute cluster as: h i1=ðq1Þ uij ¼

1 kxj vi k2

Pc h

1 k¼1 kxj vk k2

ð1Þ

i1=ðq1Þ

also define vi, which denote the controid of the ith attribute cluster as: Pn q j¼1 ðuij Þ xj ð2Þ vi ¼ Pn q : j¼1 ðuij Þ In CSA, the minimization of the following objective function is pursued: J q ðuij ; vk Þ ¼

n X c X q 2 ðuij Þ kxj vi k ; j¼1

c 6 n:

ð3Þ

i¼1

The detailed steps of employing FCM in CSA are summarized as following: Step 1 : Determine the objects (the given equipments or subsystems in engine room) and their correlative attributes that will be discretized. Step 2 : Confirm the sample set X of the specific attribute and the jth k-dimensional vector of each attribute point xj.

4.3. Definitions of using rough sets theory in SCA RST is a relative new tool that deals with vagueness and uncertainty inherent in decision making. Due to its advantage, which includes the elimination of the need for additional information about data and the ability to extract rules directly from data itself, this theory has been used in more and more research domains. 1. Information system and decision table in CSA. Pawlak has defined the notion of an information system in his literature (Pawlak, 1982). In this paper, in CSA, an information system of objects (the given equipments or subsystems in engine room) is denoted as S = (U, A, V, f), where U is a finite non-empty set of the selected objects; A is a finite non-empty set of attributes of these given objects; V is the domain of a, in which a 2 A; f is an information function f:U ? Va for any a 2 A, where Va is called domain of an attribute a. In CSA, the decision table of objects is defined as: S = (U, A [ {d}, V, f), where U, A, V, f are the same as the ones defined in the information system; {d} is a finite set of decision attributes of the given objects. 2. Lower and upper approximation in CSA. In CSA, in an information system, to every subset of attributes M # A, a binary relation, denoted by IND(M), called the M-indiscernibility relation of these selected objects, is defined as follows: INDðMÞ ¼ fðx; yÞ 2 U U j8a 2 M; aðxÞ ¼ aðyÞg:

ð4Þ

M # A and X # U, then M-lower and M-upper approximation of X is defined respectively as follows: MX ¼ [fY 2 U =INDðMÞjY # X g;

ð5Þ

MX ¼ [fY 2 U =INDðMÞjY \ X –£g:

ð6Þ

3228


3. The discernibility matrix and discernibility function in CSA. The discernibility matrix of M is defined as follows: ðC ij Þ ¼ fa 2 Mjaðxi Þ–aðxj Þg for i; j ¼ 1; 2; . . . ; n

ð7Þ

and the discernibility functionf(M)is defined as: Y X aðx; yÞ: f ðMÞ ¼

ð8Þ

ðx;yÞ2U U

In CSA, the discernibility matrix and the discernibility function are used to extract the minimal reducts from the decision tables of the given objects.

4.4. Detailed steps and flow chart of SCA In order to help design engineers with configuration selection, the suitable integration approach is adopted in CDSER for discovering design knowledge. Fig. 3 illustrates the flow chart of Configuration Selection Assistant (CSA). Through CBR Interface, similar cases data is transmitted into CSA. Then, the user chooses the objects (the given equipments or subsystems in engine room) and the relevant attributes of them are located at the same time. If the attributes are numerical, FCM is utilized to discretized them. After that, the decision tables of these objects are completed, then RST is used to extract configuration rules, by which useful suggestions are given to ship designers for completing excellent configurations of engine room.

5. Relevant methodology and detailed application steps in design scheme decision assistant Conceptual design of ship’s engine room (CDSER) is a complex, highly dynamic and interdependent process. Generally, engineers consider all kinds of information to give two or three design schemes. Every scheme might not be perfect, but has its own advantages. Judging which scheme affects the downstream design stage better, is an intricate problem to be solved. Because multifarious structures and dimensions have not been finally determined in engine room at this stage, also some new technology and new equipments are adopted for improving the performance of new ships, it is difficult for engineers to assess their performance. As a result, incomplete data is included in the information systems. Accordingly, the RST method (Kryszkiewicz, 1999) is employed to reason in incomplete information systems for getting design scheme rules in CDSER. In Design Scheme Decision Assistant (DSDA), expert knowledge that is often impossible to be formalized with traditional methods plays a decisive role at the conceptual ship design stage. 5.1. Definitions of RST in incomplete decision table in DSDA In the incomplete decision table of conceptual design schemes, any attribute domain Va may contain special symbol ‘‘” to indicate that the value of an attribute is unknown or missing. 1. Similarity of conceptual design schemes in DSDA. In the incomplete decision table of conceptual design schemes, SIM(M) is defined as: SIMðMÞ ¼ fðx; yÞ 2 U U j8a 2 M; fa ðxÞ ¼ fa ðyÞ or f a ðxÞ ¼ or f a ðyÞ ¼ g

ð9Þ

The set of possibly indiscernible conceptual design schemes is denoted by SM(x) = {y 2 Uj(x,y) 2 SIM(M)}. Define the function oA(x) as follows: @ A ðxÞ ¼ ffd ðyÞjy 2 S M ðxÞg

ð10Þ

oA(x) is the generalized decision in the incomplete decision table of conceptual design schemes. 2. Computation of optimal certain conceptual design scheme rules.Any conceptual design scheme decision rule t ? d is called certain in S iff t ? d is definite and ktk # kdk in S. Any conceptual design scheme decision rule t ? d is optimal certain in S iff it is certain in S and no other rule constructed from a proper subset of attributes of design schemes occurring in t is certain in S. Let x 2 U and IA(x) # I{d}(x) (i.e. card(oA(x)) = 1), DU(x) is a certain x-discernibility function iff: YX aðx; yÞ; where Y U ¼ U n I fdg ðxÞ: DU ðxÞ ¼ ð11Þ Fig. 3. Flow chart of configuration selection assistant (CSA).

y2Y


3. Computation of optimal generalized conceptual design scheme rules. Let x 2 U, Dg(x) is a generalized x-discernibility function in S iff YX aðx;yÞ; where Y g ¼ U n fy 2 U jdðyÞ 2 @ AT ðxÞg: Dg ðxÞ ¼ y2Y g

ð12Þ

The prime implications of Dg(x) determine generalized xreducts of the incomplete decision table of conceptual design schemes. Any conceptual design scheme decision rule t ? d is optimal generalized in S iff it is generalized in S and no other rule constructed from a proper subset of attributes of design schemes occurring in t or d is generalized in S. In DSDA, certain x-discernibility function and generalized x-discernibility function are used to obtain the optimal certain rules and optimal generalized rules of conceptual design schemes, respectively.

3229

engine room by drawing sketches of schemes, in which the main task includes initial locating of devices, primary routing of piping-system and etc. After that, besides all the influencing elements of engine room are determined, the evaluations of them are accomplished. This information is taken as basic data to constitute corresponding conceptual design scheme decision table. In DSDA, optimal certain rules of conceptual design schemes that assist ship designers in modifying the schemes are obtained. With these rules, design engineers will judge whether the new solution satisfies technological specifications, ship owners’ needs and regulations of classification societies. If the answer is positive, the final design scheme should be conformed, whereas, optimal generalized rules of design schemes for designers with relatively relaxed restrictions of design have to be extracted. This process will not be ended until the final design scheme is good enough. 6. Case study

5.2. Detailed steps and flow chart of DSDA In DSDA, RST method is adopted for reasoning in incomplete design scheme systems to deduce design scheme rules. Fig. 4 shows the flow chart of Design Scheme Decision Assistant (DSDA). Both data about primary configurations of engine room in CSA and regulations of classification societies in Ship Data Warehouse System are transmitted into it. Then, users preliminarily design new

In this section, the CDSER of a new 50,000 DWT Handymax bulk carrier is taken for example to prove the validity and necessity of the novel expert system (ESACD). The data about engine rooms of similar built ships is stored in Ship Data Warehouse System, which contains all the information of 123 constructed ships. General data about the similar cases gained by using CBR approach (Kowalski et al., 2005) are listed in Table 1. As main engines (ME) are the heart equipment in ship’s engine room, ship designers attach importance to selection of ME in CDSER. Thus, selection of ME is mainly discussed in this study case. It can be seen from Table 1 that many different types of ME have been quipped on the similarly built bulk carriers. To satisfy ship owners’ needs, the appropriate configuration must be confirmed from these ME for the new bulk carrier. Firstly, the continuous attributes must be discretized. Considering power and speed of ME, the three semantic words ‘‘High (H), Normal (N) and Low (L)” are ordinarily used to describe them. Hence, the data of Max Power and Speed in Table 1 are redefined with the three words by FCM algorithm. Ship designers assign the number of clusters 3. As a result, these points are separated into 3 clusters as follows: Max power: {7900, 7980, 8160, 8580}; {8930, 9300, 9360, 9480, 9600}; {10,225, 10,620, 10,920}. Speed: {100, 105, 110}; {123, 124, 127}; {135, 148}.

Fig. 4. Flow chart of design scheme decision assistant (DSDA).

The continuous features are mapped to discrete ones in this manner. Table 2 shows the results. Then, Man B&W 5L60MC, Man B&W 6L50MC, Man B&W 6S50ME-C, Sulzer 6RTA48T, Sulzer 7RTA52U, MITSUBISHI-6UEC60LS, MITSUBISHI-5UEC60LSII are selected from Table 1. Table 3 presents the decision table that consists of these engines with concerned discretized features and consumers’ comments.

3230


Table 1 The general data about the existing ships similar to 50,000 DWT Handymax bulk carrier Ship No.

Length (o.a.) (m)

Length (p.p) (m)

Breadth (m)

Depth (m)

Draft (t)

Dead weight (t)

Main engine

4

190.00

182.00

32.26

17.0

11.0

52,200

10 23 25

189.90 189.99 189.80

182.00 182.00 181.00

32.26 32.26 32.26

17.1 16.67 16.9

10.75 10.75 10.75

52,000 50,900 50,000

38

189.96

181.00

32.20

16.5

10.7

48,530

52

186.00

177.00

30.40

16.50

10.95

45,800

67

188.00

180.00

32.00

17.7

10.7

48,199

68

187.00

180.00

32.26

16.8

11.75

50,500

70 82

182.00 190.00

173.90 181.90

32.26 32.26

17.1 17.0

10.75 11.6

52,000 54,000

87 98

189.80 182.00

182.10 174.90

32.26 32.26

17.5 16.67

10.9 11.9

51,000 52,000

101 104

190.00 189.99

181.50 182.00

31.00 32.26

17.7 17.90

10.7 12.55

46,000 54,000

111

188.10

180.00

32.26

16.8

10.75

52,000

118

179.60

172.00

32.20

19.60

12.0

47,300

Man B&W 6S50MC Sulzer 6RTA48T Sulzer 6RTA48TB Man B&W 6L50MC Man B&W 6S50ME-C Man B&W5S50MC-C Man B&W 6S50MC-C MITSUBISHI 5UEC60LSII Sulzer 7RTA52U MITSUBISHI 6UEC60LS Sulzer 6RTA48T Man B&W 5L60MC Sulzer 6RTA52U MAN B&W 6S50MC-C MITSUBISHI 5UEC60LSII MITSUBISHI 5UEC60LA

Table 2 The discretized data of max power and speed in Table 1 with FCM Ship No.

Max power

Speed

4 10 23 25 38 52 67 68 70 82 87 98 101 104 111 118

L L N L N L N H H H L N N N H N

N N N H N N N L H L N N H N L L

Max power (kW)

Speed (rpm)

Oil consumption (t/d)

Service speed (kn)

8580

127

30.6

14.5

8160 8930 7980

124 127 148

27.9 31.9 26.8

14.4 14.5 14.2

9480

127

33.1

14.9

7900

127

26.5

14.2

9480

127

33.1

15.0

10,225

105

36.8

15.0

10920 10620

135 100

38.2 36.0

15.2 15.1

8160 9600

124 123

27.8 34.6

14.5 14.5

9360 9480

135 127

32.0 33.5

14.9 14.3

10225

105

36.9

14.7

9300

110

33.1

14.3

This ME decision table is expressed as: S = (U, A [ {d}, V, f), where U = {E1, E2, . . . , E7} is the set of the selected ME; A is the set of conditional attributes which includes R (reliability), O (oil consumption), P (price), V (volume), H (heavy repair period) and W (weight) of ME, whereas the Comment, namely consumers’ comments of these ME, is the decision attribute {d}. The attribute domains are as follows: VR = {H (High), N (Normal), L (Low)}; VO = {H (High), N (Normal), L (Low)}; VP = {H (High), N (Normal), L (Low)}; VV = {L (Large), S (Small)}; VH = {L (Long), N (Normal), S (Short)}; VW = {H (Heavy), N (Normal), L (Light)}; VComment = {S (Satisfied), U (Unsatisfied)}. The result of calculating the discernibility function in Table 3 is obtained with formula (7) and (8) as: f ðAÞ ¼ ðR _ P Þ ^ O: Thus, the reducts of the ME decision table can be described as: {R, O} or {P, O}, moreover, {O} is the core of the ME

Table 3 The decision table of selected ME in similar ships Engine

R

O

P

V

H

W

Comment

E1 E2 E3 E4 E5 E6 E7

H N H H L L N

N H L H L N N

H L H N N N L

L L L L S S L

N N S S L L N

H H H H N L H

S U S U S U S

Table 4 The minimal simplified table of Table 3 Engine

O

Comment

E1 E2 E3 E4

L N N H

S S U U


decision table. The minimal simplified table of Table 3 is concluded in Table 4. Three decision rules are acquired from Table 4 as follows: Rule 1 : If oil consumption is low, then Comment is satisfied. Rule 2 : If oil consumption is high, then Comment is unsatisfied. Rule 3 : If oil consumption is normal, then Comment is unsatisfied or satisfied. These rules indicate which features of ME mostly affect the whole ship, and which one is able to meet ship owners’ requirements effectively. As a result, oil consumption is the core of the ME decision table, that means it is the most significant element that impacts on customers’ comments. In effect, ship owners are concerned about the profit brought by ships. Once ships have been constructed, undoubtedly, the oil consumption of ME is one of the important factors that affect on the cost of transport. Lower oil consumption ME can save more energy, sequentially, strengthen enterprises’ competitive power. Hence, in Configuration Selection Assistant (CSA), these rules are consistent with the practical situation. Actually, different ship owners may have different requirements. Some of them care about the past customers’ comments, while others may concern themselves with the performance or prices of the devices. Consequently, if regard both prices and consumers’ comments as the decision goals for configuration selection, a new ME decision table is constructed based on Table 3. It is defined as: S = (U, A {P} [ {P,d}, V, f). The set of conditional attributes ({A {P}}) and the set of decision attributes ({P, d}) are included in this multiple objective decision table. Besides considering the rules for customers’ comments, the result of calculating the discernibility function for price in Table 3 is obtained with formula (7) and (8) as: f ðAÞ ¼ R ^ O Hence, {R, O} is not only the reduct, but also the core of this decision table for price. The minimal simplified table of the new decision table for price is concluded in Table 5. Four rules are extracted from Table 5: Rule 1 : If reliability is high and oil consumption is low or normal, then price is high.

Table 5 The minimal simplified table of the new ME decision table for price Engine

R

O

P

E1 E2 E3 E4 E5 E6 E7

H N H H L L N

N H L H L N N

H L H N N N L

3231

Rule 2 : If reliability is high and oil consumption is high, then price is normal. Rule 3 : If reliability is low and oil consumption is low or normal, then price is normal. Rule 4 : If reliability is normal and oil consumption is high or normal, then price is low. The four rules express that if reliability is high and oil consumption is low, price of ME should be high. Contrarily, price should be low. All of these configuration rules are consistent with the practical situation in CDSER. In CSA, the useful information gives a valid approach to help ship designers to think over the conceptual design of engine room from all aspects. The appropriate suggestions guide ship designers to get better configurations based on past solutions. For the purpose of improving the performance of the new 50,000 DWT Handymax bulk carrier, design engineers not only select the appropriate main engines (ME) from these similar cases, but also try to equip one of the four ME that have not been equipped on those similar bulk carriers. The main technical parameters of the four ME are shown in Table 6. With engineers’ experience, if new ME were equipped, some components (coupling, shaft, screw, generator and so on) of the ship subsystems should be changed accordingly. Designers select diverse correlative components for the four ME. When preliminary selection and location of them are finished, six feasible schemes have been confirmed. Then, engineers assess these six design schemes from the aspects about which they are concerned. From their view points, five features (valid utilization degree, maintenance cost, building cost, reliability in voyage and energy consumption cost) are thought important for evaluation of the schemes. Because there is no information and experience of using the new ME in the past, except for the manufacture technical parameters that are offered by the provider, some attributes of these design schemes are hard to be estimated in practice. So, the incomplete data is inevitably included in decision table. Table 7 is the decision table containing incomplete data of the six design schemes. The symbol ‘‘” is used to denote the incomplete values of the attributes.

Table 6 The main technical parameters of the ME wanted to be equipped on the new bulk carrier Main engine

Bore (mm)

Stroke (mm)

Length (mm)

Weight (kg)

Max power (kW)

Speed (rpm)

Man B&W 5S60MC MITSUBISHI 7UEC52LS Sulzer 7RTA48T Sulzer 5RTA58T

600

2292

7184

319,000

10,200

105

520

1850

8285

293,000

9310

120

480

2000

6968

220,000

9520

124

580

2416

6330

280,000

10,000

103

3232


Table 7 The incomplete decision table of the six conceptual design schemes Scheme

V

M

C

R

E

Grade

oA

S1 S2 S3 S4 S5 S6

H H N L N H

H L H L L

H H L H L H

H H L

H H H H

II I II III II II

{II} {I, II} {II} {III} {II} {I, II}

This decision table is expressed as: S = (U, A [ {d}, V, f), where U = {S1, S2, . . . , S6} is the set of these conceptual design schemes of engine room; A is the set of conditional attributes which consists of V (valid utilization degree), M (maintenance cost), C (building cost), R (reliability in voyage) and E (energy consumption cost), whereas the decision attribute {d} is Grade, which represents the evaluation grade of conceptual design schemes. The attribute domains are: VV = {H (High), N (Normal), L (Low)}; VM = {H (High), L (Low)}; VC = {H (High), L (Low)}; VR = {H (High), L (Low)}; VE = {H (High), L (Low)}; VGrade = {I, II, III}, where I, II and III stand for the good, normal and poor evaluation grade of design schemes, respectively. S A ðS 1 Þ ¼ fS 1 g;

S A ðS 2 Þ ¼ fS 2 ; S 6 g;

S A ðS 3 Þ ¼ fS 3 ; S 5 g;

S A ðS 4 Þ ¼ fS 4 g;

S A ðS 5 Þ ¼ fS 3 ; S 5 g;

S A ðS 6 Þ ¼ fS 2 ; S 6 g:

The optimal certain rules of Table 7 are concluded with formula (11) as follows: Rule 1 : If maintenance cost is low or valid utilization degree is high or reliability in voyage is high and energy consumption cost is low, then Grade is I. Rule 2 : If valid utilization degree is normal, then Grade is II. Rule 3 : If building cost is low, then Grade is II. Rule 4 : If maintenance cost is high and reliability in voyage is high or valid utilization degree is high, then Grade is II. Rule 5 : If valid utilization degree is low, then Grade is III; Rule 6 : If maintenance cost is high and reliability in voyage is low or energy consumption cost is low, then Grade is III. As the result of analyzing the design schemes in Design Scheme Decision Assistant (DSDA), Rule 1 illuminates that if maintenance cost is low or valid utilization degree is high or reliability in voyage is high and energy consumption cost is low, then the evaluation grade of the design schemes would be good. Hence, following this rule, to accomplish a good design scheme, ship designers should amend the design scheme to be easily maintained or availably utilized or very reliable and lower energy consumed. On the contrary, Rule 5 and 6 explain which features and what values of them would result in the poor grade of

design schemes. In practice, they should be avoided because of their unsatisfying performance. In DSDA, if no satisfying design scheme is acquired with these optimal certain rules, optimal generalized rules with a wider range of restrictions are advised to be adopted to help designers. Generalized decision attributes are calculated with formula (10), and shown in Table 7. Optimal generalized rules of conceptual design schemes in Table 7 are gotten with formula (12) as following: Rule 1 : If valid utilization degree is high, then Grade is I or II. Rule 2 : If maintenance cost is low, then Grade is I or II. Rule 3 : If valid utilization degree is normal, then Grade is II. Rule 4 : If building cost is low, then Grade is II. Rule 5 : If maintenance cost is high and valid utilization is high or reliability in voyage is high, then Grade is II. Rule 6 : If valid utilization degree is low, then Grade is III. Rule 7 : If maintenance cost is high and reliability in voyage is low, then Grade is III. From the above results, some optimal generalized rules can be found the same as the optimal certain ones. However, some optimal generalized rules have two decision attributes, so they can provide more suggestions with relatively relaxed restrictions of design for design engineers than optimal certain rules do. In the light of Rule 1, in order to have a good or normal grade of conceptual design scheme, ship designers should amend the design scheme of engine room to be highly validly utilized. While, with respect to Rule 7, it expresses that if poor grade is to be avoided, engineers should prevent the conceptual design schemes of engine room being hard maintained and unstable in voyage. All the above discussions show that configuration rules and conceptual design scheme rules that discover the hidden information in CSDER, are consistent with the practical situation appropriately. The results verify that the proposed expert system (ESACD) substitutes engineers’ knowledge effectively, consequently, it largely assists ship designers in configuration selection and conceptual design schemes decision in CDSER. It can be assumed ESACD is the suitable expert system that shortens development time and promotes design quality in CDSER. 7. Conclusions In this paper, the original expert system, ESACD is established for mining deep hidden design rules in CDSER. Two core subsystems: Configuration Selection Assistant (CSA) and Design Scheme Decision Assistant (DSDA) are included in it. In CSA, a suitable approach combining FCM and RST is applied to handle the information about engine rooms of similar built ships for getting configuration rules. In DSDA, by employing RST to reason


incomplete design scheme systems, optimal certain design scheme rules and optimal generalized design scheme rules are calculated to assist designers in modifying and deciding the final preliminary design scheme of engine room. In this elaborated expert system, the easily understood knowledge that is denoted by the design rules facilitates rapid and intelligent design in CDSER. CDSER of a new 50,000 DWT bulk carrier was taken as a case study in this work, where the validity and benefit of ESACD is correctly proved. The results show that this intelligent expert system can successfully extract useful suggestions at the initial ship design stage, effectively reduce the dependence on engineers’ knowledge and cut down the ship design cost and time. Acknowledgements This research was supported by the National Basic Research Program (973 Program) of China under Grant No. 2004CB719405 and Hi-Tech Research and Development Program of China under Grant No. 2007AA04Z120. References Arendt, R. (2000). An application of on expert system for aided design of power ship subsystem automation. In Proceedings of the sixth international conference on methods and models in automation and robotics. Miedzyzdroje, Poland (pp. 121–126). Arendt, R. (2004). The application of an expert system for simulation investigations in the aided design of ship power systems automation. Expert Systems with Applications, 27(3), 493–499. Arendt, R. & Kowalski, Z. (1999). An application of simulation investigation in an expert system for ship automation aided design. In Proceedings of the international conference unconventional electromechanical and electrical systems. Vol. 2, St. Petersburg. (pp. 541– 546). Bezdek, J. C. (1973). Fuzzy mathematics in pattern classification. Ph.D.dissertation, Cornell University, Ithaca. Dougherty, J., Kohavi, R., & Sahami, M. (1995). Supervised and unsupervised discretization of continuous features. In Machine learn-

3233

ing: Proceedings of the twelfth international conference (pp. 194–202). San Francisco, CA: Morgan Kaufmann Publishers. Goh, C., & Law, R. (2003). Incorporating the rough sets theory into travel demand analysis. Tourism Management, 24(5), 511–517. Han, S. H., & Lee, D. (1996). Integration of ship design software modules utilizing the superapplication concept. Journal of Marine Science and Technology, 1(5), 268–273. Hsu, W., & Liu, B. (2000). Conceptual design: issues and challenges. Computer-Aided Design, 32(14), 849–850. Kim, H., Lee, J. K., Park, J. H., Park, B. J., & Jang, D. S. (2002). Applying digital manufacturing technology to ship production and the maritime environment. Integrated Manufacturing Systems, 13(5), 295–305. Kowalski, Z., Arendt, R., Meler-Kapcia, M., & Zielin´ski, S. (2001). An expert system for aided design of ship systems automation. Expert Systems with Applications, 20(3), 261–266. Kowalski, Z., Meler-Kapcia, M., Zielin´ski, S., & Drewka, M. (2005). methodology application in an expert system for aided design ship’s engine room automation. Expert Systems with Applications, 29(2), 256–263. Kryszkiewicz, M. (1999). Rules in incomplete information systems. Information Sciences, 113(3-4), 271–292. Lee, D., Kang, J. H., Ryu, K. R., & Lee, K. H. (1997). Applying memorybased learning to indexing of reference ships for case-based conceptual ship design. In Proceedings of international conference on case-based reasoning97, Rhode Island. (pp. 74–83). Lee, D., & Lee, K. H. (1999). An approach to case-based system for conceptual ship design assistent. Expert Systems with Applications, 16(2), 97–104. Liu, H., Xiang, B., & Qu, L. (2007). The application of rough sets in SAR analysis of N1-site substituted fluoroquinolones. Chemometrics and Intelligent Laboratory Systems, 87(2), 155–160. Parsons, M. G., Singer, D. J., & Sauter, J. A. (1999). A hybrid agent approach forset-based conceptual ship design. In Proceedings of the 10th international conference on computer applications in shipbuilding. Cambridge, MA (pp. 207–221). Pawlak, Z. (1982). Rough Sets. International Journal of Computer and Information Sciences, 11(5), 341–356. Pawlak, Z., & Skowron, A. (2007). Rudiments of rough sets. Information Sciences, 177(1), 3–27. Tseng, T., & Huang, C. (2007). Rough set-based approach to feature selection in customer relationship management. Omega, 35(4), 365–383. Wong, J., & Chung, Y. (2007). Rough set approach for accident chains exploration. Accident Analysis & Prevention, 39(3), 629–637.