行政院國家科學委員會專題研究計畫 成果報告

行政院國家科學委員會專題研究計畫 成果報告

TFT-LCD 供應鏈之新產品開發及管理(第 3 年) 研究成果報告(完整版)

計 畫 類 別 ： 個別型

計 畫 編 號 ： NSC 97-2410-H-216-010-MY3

執 行 期 間 ： 99 年 08 月 01 日至 100 年 07 月 31 日 執 行 單 位 ： 中華大學工業管理學系

計 畫 主 持 人 ： 李欣怡

計畫參與人員： 碩士班研究生-兼任助理人員：陳建舜 碩士班研究生-兼任助理人員：吳欣蔚 碩士班研究生-兼任助理人員：黃振哲 碩士班研究生-兼任助理人員：劉純宜 博士班研究生-兼任助理人員：林俊宇

報 告 附 件 ： 出席國際會議研究心得報告及發表論文

處 理 方 式 ： 本計畫涉及專利或其他智慧財產權，2 年後可公開查詢

中 華 民 國 100 年 10 月 29 日

I

TFT-LCD 供應鏈之新產品開發及管理

中文摘要

在全球市場激烈競爭下，企業成功的關鍵因素在於如何滿足顧客需求，故一企業如何永 續經營則考驗這企業內部之核心價值。然而，對於企業而言，核心價值不再僅僅是企業管理 能力，更加考驗企業如何滿足市場所期待之產品研發能力。TFT-LCD 產業為目前台灣最亮麗 的產業之一。隨著全球 TFT-LCD 產業邁入成熟階段，未來將面臨劇烈之市場競爭及價格割喉 戰。台灣 TFT-LCD 製造之競爭優勢源於低成本、高品質、具彈性、相關產業之專業技術與完 整的群聚供應鏈等。然而在全球競爭環境下，企業生存與成功之關鍵在於如何因應市場快速 變化與顧客需求的不確定性，許多企業也都紛紛了解到，新產品開發(NPD)對企業生存極為 重要。因此，新產品開發將是保持競爭優勢並維持企業長期利潤之首要關鍵，故在產品設計 與製造上必須滿足顧客需求之產品品質及功能。因此，本研究結合品質機能展開(QFD)與模 糊網路分析法(FANP)以解決 TFT-LCD 製造商於新產品開發階段所面臨之問題。專家的主觀 判斷中，往往無法處理過多因子之比較，故本研究先運用模糊德爾菲(FDM)篩選出關鍵之因 子。為處理模糊語意及不確定因素，以及因子間之相互依存關係，本研究提出之架構結合了 QFD 及 FANP 二種方法以協助設計人員利用系統化之模式於新產品開發上。然而，一項產品 之開發需要有良好的前置規劃外，更需考量原物料品質之重要性，故供應商合作與選擇即為 相當重要之決策。本研究延續新產品開發架構，運用模糊分析網絡程序(FANP)與利益、機會、

成本與風險(BOCR)建立可靠度較佳之供應商選擇，以利決策者做為參考之依據。

關鍵詞：新產品開發(NPD)，品質機能展開(QFD)，模糊網路分析法(FANP)， TFT-LCD，

利益、機會、成本與風險(BOCR)，模糊德爾菲(FDM)。

II

New product development and management in TFT-LCD supply chain

ABSTRACT

Global competitiveness has become the biggest concern of manufacturing companies, especially in TFT-LCD industries. However, as the global TFT-LCD industry enters the mature stage, an extremely competitive and cost-cutting war is foreseeable. While providing the products with a lower cost, better quality at the right time and place is important for Taiwan’s TFT-LCD manufacturers, new product development (NPD) is essential to maintain a competitive edge and to make a decent profit in a longer term. Thus, the introduction of successful new products is a source of new sales and profits and is a necessity in the intense competitive international market.

After a product is developed, a firm needs the cooperation of upstream suppliers to provide satisfactory components and parts for manufacturing final products. Therefore, the selection of suitable suppliers has also become a very important decision. In this research, a model that incorporates quality function deployment (QFD) and fuzzy analytic network process (FANP) is built to solve the NPD problem in TFT-LCD manufacturing. Since people are not willing and capable to handle comparisons properly when there are too many factors, fuzzy Delphi method (FDM) is used first to limit the number of factors included in the model. In considering the impreciseness and vagueness in human judgments and information, and the interrelationship among factors, a QFD model incorporated with FANP is constructed to facilitate the NPD process. In addition, an analytical approach is proposed to select the most appropriate critical-part suppliers in order to maintain a high reliability of the supply chain. A fuzzy analytic network process (FANP) model, which incorporates the benefits, opportunities, costs and risks (BOCR) concept, is constructed to evaluate various aspects of suppliers. The proposed model is adopted in a TFT-LCD manufacturer in Taiwan in evaluating the expected performance of suppliers with respect to each important factor, and an overall ranking of the suppliers can be generated as a result.

Keywords: New product development (NPD); Quality function deployment (QFD); Fuzzy

analytic network process (FANP); TFT-LCD; Fuzzy Delphi method (FDM); Supplier

selection; Benefits, opportunities, costs and risks (BOCR)

III

目 錄

中文摘要...I ABSTRACT... II 目 錄...III 圖 目 錄...IV 表 目 錄... V

1. Introduction... 1

2. TFT-LCD manufacturing... 2

3. Methods... 3

3.1. Fuzzy Delphi method (FDM)... 3

3.2. Quality function deployment (QFD)... 4

3.3. Fuzzy analytic network process (FANP) ... 6

4. The proposed model for NPD and supplier selection ... 6

5. Case study ... 13

6. Conclusions... 21

References... 22

計畫成果自評...25

IV

圖 目 錄

Figure 1. TFT-LCD manufacturing process ...3

Figure 2. Gray zone of lⁱ and uⁱ. ... 4

Figure 3. Four phases of QFD... 5

Figure 4. The components of HOQ... 5

Figure 5. House of Quality (Lee et al., 2010) ... 8

Figure 6. The control hierarchy (Lee, 2009b)... 10

Figure 7. Inner dependence among CAs... 16

Figure 8. House of Quality... 16

Figure 9. The four HOQs for the case study ... 19

V

表 目 錄

Table 1. Transformation of linguistic variables ... 13

Table 2. Fuzzy Delphi method results ... 14

Table 3. Customer attributes and engineering characteristics for TFT-LCD... 15

Table 4. Relation matrix among CAs... 15

1

1. Introduction

Under a globally competitive business environment, technological innovation and satisfaction of customer needs are the keys to survival and success for firms, especially for TFT-LCD firms.

Many companies realize that the emphasis on new products as a source of new sales and profits is a necessity in the intense competitive international market. Since poor product definition commonly leads to product failure in the marketplace or extended product development time, companies need to consider issues such as performance, aesthetics, delivery, quality and cost in developing their products. They must know the wants (like-to-have), needs (must-have), and desires (wish-to-have) of their customers as completely as possible (Ho et al., 1999), and design and manufacture products efficiently at a competitive cost within a short period of time over those offered by competitors (Chen et al., 2004). In addition, the selection of a supplier for partnership is one of the most important steps in creating a successful supply chain and in attaining reasonable profits for a firm.

A firm, in order to maintain its competitive edge, must protect its core businesses; however, it must be and usually is willing to enter buyer-supplier relationships in peripheral activities (Todeva and Knoke, 2005). To achieve the benefits of buyer-supplier integration, in terms of increased internal efficiency and profitability of both parties, identifying and selecting viable suppliers is a preliminary step that needs to be properly managed (Bottani and Rizzi, 2007). In addition to develop an understanding of suppliers’ expectations and objectives, the firm must carry out a careful screening of potential suppliers, which is a time-consuming process (Dacin and Hitt, 1997).

Nevertheless, if the process is done correctly, a higher quality, longer lasting relationship is more attainable, and a win-win solution can be achieved.

Successful introduction and acceleration of new product development (NPD) is an important source of competitive advantage, survival and renewal for many organizations (Howell et al., 2006).

Companies have to develop successful new products continuously because of fast changing technologies, shortening product lifecycles and increased globalize competition. The advantages of NPD include fast and economic (Wheelwright and Clark, 1992), increased product reliability (Sanderson and Uzumeri, 1995), increased variety, simplified managerial complexity and increased flexibility of strategic targets (Meyer and Lehnerd, 1997). In NPD, product conceptualization is the first step and is critical to the final success of the product, and quality function deployment (QFD) is a well-known comprehensive quality management system to consider customer requirements carefully starting from product conceptualization. However, conventional QFD has its shortcomings. Even though many modified QFD models have been proposed, a comprehensive model is necessary.

The introduction of successful new products is important to survive in today’s fierce competitive international market. Suppliers’ early involvement in the NPD process and the intense patterns of communication flows are driving forces for faster releases of new products, lower costs, and prompt responses to competitors’ moves (Sobrero & Roberts, 2002). Even though the research on supplier selection is abundant, the works usually only consider the critical success

2

factors in the buyer-supplier relationship and do not emphasize the NPD capabilities of the suppliers.

The negative aspects of the buyer-supplier relationship and suppliers’ NPD capabilities must be considered simultaneously in today’s competitive TFT-LCD industries.

In this project, a model that incorporates quality function deployment (QFD) and fuzzy analytic network process (FANP) is built to solve the NPD problem in TFT-LCD manufacturing.

Through literature review and interview with domain experts, a list of factors, including customer attributes (CAs) and engineering characteristics (ECs) for TFT-LCD, is prepared first. Since people are not willing and capable to handle comparisons properly when there are too many factors, fuzzy Delphi method (FDM) is used next to limit the number of factors included in the model. In considering the impreciseness and vagueness in human judgments and information, and the interrelationship among factors, a QFD model incorporated with fuzzy analytic network process (FANP) is constructed to facilitate the NPD process. The model can provide a general framework capable of helping designers to systematically consider relevant NPD information and effectively determine the key success factors for customer-driven design and manufacturing of new products.

Another objective of this project is to propose an analytical approach to select critical-part suppliers under a fuzzy environment. A fuzzy analytic network process (FANP) model, which incorporates the benefits, opportunities, costs and risks (BOCR) concept, is constructed to evaluate critical-part suppliers. Multiple factors that are positively or negatively affecting the success of the relationship are analyzed by taking into account experts’ opinions on their importance, and a performance ranking of the suppliers is obtained.

2. TFT-LCD manufacturing

TFT-LCD has a sandwich-like structure consisting of two glass substrates with a layer of liquid crystal inside. The top substrate is fitted with a color filter that contains the black matrix and resin film containing three primary-color (red, green and blue) dyes or pigments. The bottom substrate is TFT array that contains the TFTs, storage capacitors, pixel electrodes and interconnect wiring. The two glass substrates are assembled with a sealant, and spacers are used to maintain the gap between the substrates (AU Optronics, 2010). Liquid crystal material is injected between two substrates. The outer face of each glass substrate has a sheet of polarizer film. Each end of the gate has a set of bonding pads and data-signal bus-lines to attach LCD Driver IC (LDI) chips (AU Optronics, 2010).

The manufacturing of TFT-LCD, as depicted in Figure 1, can be categorized into five major processes: TFT array fabrication, color filter (BM) fabrication, color filter (RGB) fabrication, cell assembly and module assembly. A TFT-LCD manufacturer usually has different plants for TFT array fabrication, cell assembly and module assembly. On the other hand, color filters are usually purchased from color filter manufacturers, even though there is a trend for vertical integration

3

between color filter manufacturers and TFT-LCD manufacturers or a certain degree of alliance between the two.

Figure 1. TFT-LCD manufacturing process

As global information industry increases, the demand of TFT-LCD panels with low weight, slender profile, low power consumption, high resolution, high brightness and low radiance, increases tremendously. As a result, product innovation of TFT-LCD has become an important focus for TFT-LCD manufacturers for gaining a good share of the profitability in this flourishing market.

3. Methods

3.1. Fuzzy Delphi method (FDM)

Since its development by Dalkey and Helmer in 1963, the Delphi method, which facilitates consensus by converging a value through the feedback of experts after several rounds, has been widely applied in many management areas, such as forecasting, project planning and public policy analysis. However, the method does have its shortfalls: repetitive questionnaires and evaluations, declining response rate of experts, inappropriate convergence, ambiguity and uncertainty in survey questions and in response, lengthy time and high cost (Chang et al., 2000; Chang and Wang, 2006).

Therefore, today the Delphi method has been expanded and modified into numerous techniques, and the incorporation of fuzzy set theory is one of the approaches.

From a collection of numerous factors, the fuzzy Delphi method can be applied to downsize th factors into a limited number of more important factors. The procedures are as follows (Ishikawa

et al., 1993; Chang et al., 1995; Chang and Wang, 2006; Hsiao, 2006):

1. Conduct a questionnaire and ask experts for their most pessimistic (minimum) value and the most optimistic (maximum) value of the importance of each factor in the possible sub-criteria set S in a range from 1 to 10. A score is denoted as:

(

^,

)

^,

i ik ik

c

=

l u i S

∈ ( 1 ) 2. Select the minimum and maximum values and calculate geometric mean of the group’s most

Array Fabrication Process

Cell Assembly Process

Module Assembly Process

Color Filter (BM) Fabrication

Color Filter (RGB) Fabrication

4

pessimistic (minimum) index and the values of the most optimistic (maximum) index for each factor. Determine the triangular fuzzy numbers for the most pessimistic index and the most optimistic index for each factor. The triangular fuzzy number for the most pessimistic index is

( , , )

i i i i

l m u

l = l l l and for the most optimistic index is uⁱ =( ,u u u_lⁱ _mⁱ, _uⁱ).

3. Inspect the consensus of experts’ opinions and calculate the significance value for each factor.

As shown in Figure 2, the gray zone, the overlap section of lⁱ and uⁱ, is used to inspect the consensus of experts in each factor and to calculate the consensus significance value of the factor, sⁱ.

i

ll lⁱ_m uⁱ_l l_uⁱ uⁱ_m uⁱ_u

Figure 2. Gray zone of lⁱ and uⁱ.

4. Extract factors from the candidate list. Compare consensus significance value with a threshold value, T, which is determined by experts subjectively based on the geometric mean of all sⁱ. If sⁱ ≥T, select factor i for further analysis.

3.2. Quality function deployment (QFD)

A typical QFD system consists of four phases, product planning, part deployment, process planning and production planning, and each phase contains a matrix called house of quality (HOQ) (Zhang et al., 1999; Chen et al., 2004). In the product planning phase, product planning matrix contains information about what customers want, how technically customer requirements can be achieved, and the relationships between each of these aspects (Ho et al., 1999). The four phases are depicted in Figure 3 (Ho et al., 1999; Sohn and Choi, 2000; Kahraman et al., 2006). Through the above four phases, the voice of the customer is systematically cascaded into the design, process, and production of the product (Zhang et al., 1999).

The systematic procedure for the first HOQ contains seven steps, and is depicted in Figure 4 (Chan et al., 1999; Wang, 1999; Ramasamy and Selladurai, 2004):

1. Obtaining customer attributes (CAs). In addition to questionnaire, interviewing, claim and complaint information, customer needs can also be collected by focus groups or individual interviews. From the collected information, the required CAs are established.

5

Figure 3. Four phases of QFD

Figure 4. The components of HOQ

2. Developing engineering characteristics (ECs). ECs are also known as design requirements, product features, product technical requirements, engineering attributes, engineering characteristics or substitute quality characteristics (Karsak et al., 2002).

3. Building relationship between customer attributes (CAs) and engineering characteristics (ECs).

By correlating CAs and ECs, a relationship matrix is prepared indicating how much each EC affects each CA, and such a relation can either be presented by a number or a symbol.

4. Completing competitive survey and calculating relative importance of CAs. The product performance of the company and its main competitors is rated so that the competitive positions of the company’s product in terms of the CAs can be assessed (Chan et al., 1999).

5. Performing the competitive technical benchmarking. The performance of the company and its main competitors is rated with respect to each EC.

6. Determining the relationships among ECs. A correlation matrix, or “roof”, is used to show the positive and negative relationship and the degree of relationship among the ECs.

6

7. Calculating the importance of ECs and additional goals. The importance and ranking of ECs are established from the results in step 5 and step 6.

In the QFD implementation, the determination of the correct importance weights for the CAs and ECs is essential since it affects the final outcomes of the whole process significantly. The simplest method to prioritize the CAs is based on a point scoring scale, such as 1 to 5 or 1 to 10 (Griffin and Hauser, 1993; Kwong and Bai, 2003; Buyukozkan et al., 2007). However, this method cannot effectively capture human perception, and a substantial degree of subjective judgment has to be involved in the scoring process (Kwong and Bai, 2003; Buyukozkan et al., 2007). Gustafsson and Gustafsson (1994) used a conjoint analysis method to determine the relative importance of the customer requirements by employing a pairwise comparison of customer requirements.

Because of the interrelationships among CAs and among ECs, ANP is used in some recent works (Partovi, 2001; Karsak et al., 2002; Partovi and Corredoira, 2002). In all these methods, the input variables are assumed to be precise and are treated as numerical data. In addition, human decision making often contains ambiguity and uncertainty. Hence, conventional ANP are inadequate to explicitly capture the importance assessment of CAs and ECs. To confront this problem, many researchers incorporate the fuzzy set theory into QFD.

3.3. Fuzzy analytic network process (FANP)

Saaty (1996) proposes the analytic network process (ANP) approach, which is a generalization of the AHP. The ANP approach replaces hierarchies with networks, in which the relationships between levels are not easily represented as higher or lower, dominated or being dominated, directly or indirectly (Meade and Sarkis, 1999). After evaluating the importance of all factors, including goal, cluster, criteria and alternatives through pairwise comparisons, a “supermatrix” is formed, following by a weighted supermatrix that ensures column stochastic. Finally, a limit supermatrix is calculated to obtain final solutions.

Although the conventional ANP has overcome some of the shortcomings of the AHP, it still cannot effectively handle problems with imprecise information. To resolve this difficulty, fuzzy set theory can be introduced to the conventional ANP, and this new type of method is called the fuzzy ANP (FANP).

4. The proposed model for NPD and supplier selection

Even though there have been many studies on the incorporation of fuzzy AHP to QFD, the applications of fuzzy ANP to QFD are rather limited. In order to consider the interrelationship among CAs and ECs more and the inner dependence among CAs and among ECs accurately, ANP, instead of AHP, should be adopted. In order to take into account the impreciseness and vagueness in human judgments and information, fuzzy set theory should be applied. Therefore, in this study,

7

we propose to use fuzzy ANP with QFD. However, people are not willing and capable to handle comparisons properly when there are too many CAs and ECs. Therefore, fuzzy Delphi method (FDM) will be used in advance to limit the number of CAs and ECs included in the model. In addition, a fuzzy analytic network process (FANP) model, which incorporates the benefits, opportunities, costs and risks (BOCR) concept, is constructed to evaluate critical-part suppliers.

Multiple factors that are positively or negatively affecting the success of the relationship are analyzed by taking into account experts’ opinions on their importance, and a performance ranking of the suppliers is obtained.

An integrated model for NPD and supplier selection is constructed. The procedures are as follows:

Step 1. Form a committee of decision makers to define the NPD problem in a TFT-LCD manufacturer. The environmental issues of the product life cycle will be considered in the NPD process. List all possible CAs and ECs in the product planning phase through methods, such as interview, questionnaire and brainstorming.

Step 2. Apply FDM to extract CAs and ECs from the candidate lists. Questionnaire is prepared to evaluate the importance of CAs (ECs), and customers, designers and related personnel are invited to fill out the questionnaire. A group average is calculated for each of

l and

_ik

u

_ik first, and the abnormal value which is outside of two standard deviations is eliminated.

The geometric mean of the pessimistic (l_mⁱ ) and the optimistic (u_mⁱ ) importance of each CA (EC), gray zone interval value

g and consensus significance value (

ⁱ sⁱ) are calculated.

Threshold value for CA (EC) is determined subjectively, and the CA (EC) with a consensus significance value greater than or equal to the threshold value is selected.

Step 3. Use ISM to determine the inner dependence among CAs and among ECs. Note that only the adjacency matrix and reachability matrix are used to construct the relationships of CAs and of ECs. Network structures for CAs and for ECs are plotted.

Step 4. Construct a HOQ. A HOQ is constructed first, as shown in Figure 5 (Karsak et al., 2002;

Lee et al., 2010). Unlike the conventional HOQ, both the inner dependence among CAs and the inner dependence among ECs are considered here. A check is entered if there is an influence of one factor to another factor.

Step 5. Prepare a questionnaire and receive feedback from experts. A questionnaire based on the structure of the HOQ is prepared using Satty’s nine-point scale of pairwise comparison.

Experts are asked to fill out the questionnaire.

Step 6. Perform consistency test. The consistency of each pairwise comparison matrix obtained from the questionnaire is examined first by calculating the consistency index (CI) and consistency ratio (CR).

8

max

1

= −

− CI n

n

λ (2)

RI

CR=CI (3)

where n is the number of items being compared in the matrix, and RI is random index (Saaty, 1980). If an inconsistency is present, the expert is asked to revise the part of the

questionnaire.

Figure 5. House of Quality (Lee et al., 2010)

Step 7. Construct fuzzy pairwise comparison matrices. The pairwise comparison matrix of each part of the questionnaire from each expert is transformed into a fuzzy pairwise comparison matrix.

Step 8. Construct fuzzy aggregated pairwise comparison matrices. Combine fuzzy pairwise comparison matrices from all experts by a geometric mean approach.

Step 9. Construct defuzzified aggregated pairwise comparison matrices. The fuzzy aggregated pairwise comparison matrices are transformed into defuzzified aggregated pairwise comparison matrices using the center of gravity (COG) method.

9

Step 10. Calculate priority vectors of the defuzzified aggregated pairwise comparison matrices.

A local priority vector is derived for each defuzzified aggregated comparison matrix as an estimate of the relative importance of the elements (Saaty, 1980; Saaty, 1996):

w w

= _max ⋅

⋅

λ

A

(4)

where A is the defuzzified aggregated pairwise comparison matrix, w is the eigenvector, and

λmax is the largest eigenvalue of A.

Step 11. Form an unweighted supermatrix. Priority vectors are entered in the appropriate columns of a matrix, known as an unweighted supermatrix, to represent the relationships in the HOQ.

1

unweighted 1 1

CG CC

EC EE

G CA EC G

CA EC

w

⎡ ⎤

= ⎢ ⎥

⎢ ⎥

⎢ ⎥

⎣ ⎦

M I

W W W

(5)

where wCG is a vector that represents the impact of the goal on CAs, WEC is a matrix that represents the impact of CAs on ECs, ^WCC indicates the interdependency of CAs, WEE

indicates the interdependency of ECs, I is the identity matrix, and entries of zeros correspond to those elements that have no influence (Saaty, 1996; Lee et al., 2010).

Step 12. Calculate a weighted supermatrix.

Step 13. Calculate the limit supermatrix and obtain the final priorities of ECs.

Step 14. Determine the goals for the NPD and the priority level of the goals. Experts are invited to determine the additional goals in the development of the product. The priority level of the goals must be determined too. Under each level, there might be more than one goal.

Step 15. Determine the relative importance of the goals under the same priority level and the relative performance of ECs with respect to each additional goal. Methods such as the Delphi method, AHP or FAHP can be applied to obtain a consensus of experts’ opinions.

Step 16. Set the preemptive GP model which considers the relative importance of the goals under the same priority level for NPD. The objective is to maximize the satisfaction in developing the product. Goals under a higher priority level must be met before the goals under a lower priority level can be met. The goals are G1,G2,…, GN, Pl is the priority level l and P1f P2f…f PL.

10

Priority level l : Pl ⁼

{

G nng g^∈l

}

,

{^{1, 2 , ...} }

U_g ^ng = ^N , l=1,2,…L (6)

Min

)

1

( ^{+ −}

= ∈

⎡ ⎤

= ⎢ + ⎥

⎢ ⎥

⎣ ⎦

∑ ∑ g g g g g

L

l n n n n

l n l

Z P w d w d (7) s.t.

( )− ⁺ + ⁻ =

g g g g

n i n n n

f x d d G ,f^{or all} ng ^and i (8)

x F (F is a feasible set) (9)

∈

where l is the priority level ;

w

n_g represents the weight attached to the deviation; Gngis the targeted values; dn⁺g and dn⁻g are, respectively, over- and under-achievements of the ngth goal.

Step 17. Form a committee of decision makers to define the supplier selection problem.

Step 18. Decompose the problem into a control hierarchy. The goal of the control hierarchy, as shown in Figure 6, is to calculate the relative importance of the four merits, benefits (B), opportunities (O), costs (C) and risks (R), based on the control criteria that the firm would like to achieve in evaluating suppliers. Pairwise comparison of the importance of control criteria towards the goal and the importance of the merits towards each control criterion are calculated.

Figure 6. The control hierarchy (Lee, 2009b).

Step 19. Decompose the problem into a BOCR network. A network with four sub-networks, B, O, C and R, is constructed. Four merits, which reflect both positive and negative impacts of selecting a particular supplier, must be considered in achieving the overall goal. A

11

sub-network is formed for each of the merits. For instance, for the sub-network for benefits (B) merit, there are criteria and/or detailed criteria that are related to the achievement of the benefits of the ultimate goal. The lowest level contains the alternatives (suppliers) that are under evaluation.

Step 20. Prepare a questionnaire based on the control hierarchy and the BOCR network. Experts in the field are invited to contribute their expertise and to fill out the questionnaire.

Step 21. Determine the priorities of the control criteria. Pairwise comparison results of the importance of control criteria toward achieving the overall objective are transformed into triangular fuzzy numbers using Table 1. A fuzzy positive reciprocal matrix is formed for each expert. The geometric mean method is applied next to form an aggregate fuzzy pairwise comparison matrix for all experts, and then the centroid method is adopted to defuzzify the fuzzy numbers in the aggregate fuzzy pairwise comparison matrix. The synthesized priorities of the control criteria can be calculated after a consistency test of the matrix is passed.

Step 22. Determine the importance of benefits, opportunities, costs and risks to each control criterion. The linguistic term and the triangular fuzzy number of each scale for evaluating the importance of benefits, opportunities, costs and risks to each control criterion is assigned to be very high (7,9,9), high (5,7,9), medium (3,5,7), low (1,3,5), and very low (1,1,3). As in Step 21, the opinions of the experts are aggregated by the geometric mean method, and the centroid method is used to defuzzify the fuzzy numbers. The crisp weights of the strategic criteria are normalized.

Step 23. Calculate the priorities of the merits, b, o, c and r. By multiplying the priority of a merit on each control criterion from Step 21 with the priority of the respective control criterion from Step 22 and summing up the calculated values for the merit, the priority of a merit can be obtained. Normalize the priorities of benefits, opportunities, costs and risks, and they are b, o, c and r, respectively.

Step 24. Calculate relative importance weights (priority vector) for criteria with respect to the same merit, relative importance weights (priority vector) for detailed criteria with respect to the same upper-level sub-criterion, relative priorities for the alternatives (suppliers) with respect to each criterion (detailed criterion) using a similar procedure in the inner dependence among criteria (detailed criteria) are calculated in a similar way.

Step 25. Form an unweighted supermatrix for each sub-network. The priority vectors obtained from Step 24 are entered in the appropriate columns in the unweighted supermatrix for each merit sub-network. An unweighted supermatrix for the benefits sub-network is:

12 Benefits Criteria Alternatives

0 0 0

Benefits

0 Criteria

0 Alternatives

wCB

⎡ ⎤

= ⎢ ⎥

⎢ ⎥

⎢ ⎥

⎣ ⎦

CC AC

B W

W I

(10)

where ^WCB is a vector that represents the impact of the benefits on the criteria, WCC indicates the interdependency of the criteria, WAC is a matrix that represents the impact of criteria on each of the alternatives, I is the identity matrix, and entries of zeros correspond to those elements that have no influence.

Step 26. Calculate the weighted supermatrix for each merit sub-network. Transform the unweighted supermatrix into a weighted supermatrix to make the supermatrix stochastic.

Step 27. Calculate the limit supermatrix and obtain the priorities of the alternatives for each merit sub-network. By raising the weighted supermatrix to powers, a limit supermatrix can be obtained when a convergence is met. The priorities of the alternatives (suppliers) under a merit are calculated by normalizing the alternative-to-merit column of the limit supermatrix of the merit.

Step 28. Calculate the overall priorities of alternatives (suppliers). By synthesizing priorities of each alternative under each merit from Step 27 with the corresponding normalized weights b, o, c and r from Step 23, the overall priorities of alternatives (suppliers) can be generated.

There are five ways to aggregate the priorities of each alternative (supplier) under B, O, C and R.

1. Additive

P

i=bBi+oOi+c[(1/Ci)Normalized]+r[(1/Ri)Normalized] (11) where Bi, Oi, Ci and Ri represent respectively the synthesized results of alternative i under merit B, O, C and R, and b, o, c and r are respectively normalized weights of merit B, O, C and R.

2. Probabilistic additive

P

i=bBi+oOi+c(1-Ci)+r(1-Ri) (12) 3. Subtractive

P

i=bBi+oOi-cCi-rRi (13) 4. Multiplicative priority powers

P

i=Bib

O

io

[(1/Ci)_Normalized]^c [(1/Ri)_Normalized]^r (14) 5. Multiplicative

P

i=Bi

O

i/Ci

R

i (15)

13

Table 1. Transformation of linguistic variables Linguistic variables Positive triangular

fuzzy numbers

Positive reciprocal triangular fuzzy numbers Extremely strong (9,9,9) (1/9,1/9,1/9)

Intermediate (7,8,9) (1/9,1/8,1/7)

Very strong (6,7,8) (1/8,1/7,1/6)

Intermediate (5,6,7) (1/7,1/6,1/5)

Strong (4,5,6) (1/6,1/5,1/4)

Intermediate (3,4,5) (1/5,1/4,1/3)

Moderately strong (2,3,4) (1/4,1/3,1/2)

Intermediate (1,2,3) (1/3,1/2,1)

Equally strong (1,1,1) (1,1,1)

5. Case study

This research focuses on both the perspectives of consumers and manufacturers, and collects CAs and ECs for TFT-LCD new product development through literature review and interview with experts. There are many CAs and ECs in TFT-LCD NPD, and it is not worthwhile and possible to include all the factors in the NPD process. Therefore, the FDM is used to collect the opinions of the experts and to select the most important factors for further FANP-QFD analysis. The results of the FDM are as shown in Table 2. To limit the number of CAs and ECS, only 6 CAs and 7 ECs are selected as shown in Table 3. These selected factors will be used in the construction of the HOQ as in Figure 7.

ISM is applied to determine the inner dependence among CAs and among ECs. Using the CAs (ECs) selected from FDM, relation matrix which shows the contextual relationship among the CAs (ECs) is established for each expert. A questionnaire is prepared to ask the contextual relationship between any two CAs (ECs), and the associated direction of the relation. For example, a relation matrix for CAs formed based on an expert’s opinions is as follow:

0 0 1 0 1 0

1 0 0 0 0 0

0 0 0 0 1 0

0 0 0 0 0 1

1 0 1 1 0 0

0 0 0 1 1 0

6 5 4 3 2

1 CA CA CA CA CA

CA

CA6 CA5 CA4 CA3 CA2 CA1

⎥⎥

⎥⎥

⎥⎥

⎥⎥

⎦

⎤

⎢⎢

⎢⎢

⎢⎢

⎢⎢

⎣

⎡

1 =

D

14

Table 2. Fuzzy Delphi method results

The geometric mean of experts’ opinions on the relationship between a pair of CAs (ECs) is calculated. A threshold value of 0.5 is used to determine whether the two CAs (ECs) are dependent or not (Yang et al., 2008). That is, a relation matrix is prepared for each expert first, and a mean relation matrix is calculated using the geometric mean method to combine relation matrices from all experts. If the geometric mean value between two CAs (ECs), i. e. π_ij, in the mean relation matrix is higher than the threshold value, x_j is deemed reachable from

x , and we

_i let π_ij =1 (Yang et al., 2008). The integrated relation matrix between CAs is calculated and is as shown in Table 4.

0 0 0 0 0 0

0 0 0 0 1 0

0 0 0 0 0 0

0 0 1 0 1 1

1 0 1 0 0 0

0 0 0 1 1 0

6 5 4 3 2

1 CA CA CA CA CA

CA

CA6 CA5 CA4 CA3 CA2 CA1

⎥⎥

⎥⎥

⎥⎥

⎥⎥

⎦

⎤

⎢⎢

⎢⎢

⎢⎢

⎢⎢

⎣

⎡

= D

15

Table 3. Customer attributes and engineering characteristics for TFT-LCD Customer attributes (CAs) Engineering characteristics (ECs) CA₁ Low power consumption EC₁ Glass cutting technology CA₂ Product quality and stability EC₂ Backlight module integrated design CA3 High-quality display EC3 Quality control of raw materials CA4 Small variations structure EC4 Quality control process CA₅ Rapid delivery EC₅ IC power-saving design CA6 Reasonable prices EC6 Power consumption control

EC7 Information Management System

Table 4. Relation matrix among CAs

CA1 CA2 CA3 CA4 CA5 CA6

CA1 0 0.5 0.6 0 0.2 0 CA2 0.3 0 0.2 0.6 0.3 0.8 CA3 0.6 0.6 0 0.7 0.2 0.2 CA4 0.2 0.3 0.2 0 0.3 0.3 CA5 0.4 0.7 0 0.3 0 0 CA6 0.2 0.4 0 0.2 0 0

The initial reachability matrix M for CAs is:

⎥⎥

⎥⎥

⎥⎥

⎥⎥

⎦

⎤

⎢⎢

⎢⎢

⎢⎢

⎢⎢

⎣

⎡

=

⎥⎥

⎥⎥

⎥⎥

⎥⎥

⎦

⎤

⎢⎢

⎢⎢

⎢⎢

⎢⎢

⎣

⎡ +

⎥⎥

⎥⎥

⎥⎥

⎥⎥

⎦

⎤

⎢⎢

⎢⎢

⎢⎢

⎢⎢

⎣

⎡

= +

=

1 0 0 0 0 0

0 1 0 0 1 0

0 0 1 0 0 0

0 0 1 1 1 1

1 0 1 0 1 0

0 0 0 1 1 1

1 0 0 0 0 0

0 1 0 0 0 0

0 0 1 0 0 0

0 0 0 1 0 0

0 0 0 0 1 0

0 0 0 0 0 1

0 0 0 0 0 0

0 0 0 0 1 0

0 0 0 0 0 0

0 0 1 0 1 1

1 0 1 0 0 0

0 0 0 1 1 0

I D

M

The final reachability matrix M* for CAs is:

⎥⎥

⎥⎥

⎥⎥

⎥⎥

⎦

⎤

⎢⎢

⎢⎢

⎢⎢

⎢⎢

⎣

⎡

=

=

=

1 0 0 0 0 0

1 1 1 0 1 0

0 0 1 0 0 0

1 0 1 1 1 1

1 0 1 0 1 0

1 0 1 1 1 1

4 3

*

M M

M

Based on M*, the inner dependence among the six CAs can be depicted as in Figure 7. The same procedure can be carried out for determining the inner dependence among ECs.

16

Figure 7. Inner dependence among CAs

Te examine the practicality of the proposed model, a case study is carried out in an anonymous TFT-LCD manufacturer in Taiwan. Seven experts from the firm are asked to contribute their expertise in the study. The HOQ is shown in Figure 8. Based on the relationship among factors shown in the HOQ in Figure 2, a pairwise comparison questionnaire is prepared, and the seven experts are asked to do the questionnaire. The consistency test is performed to check all the pairwise comparison matrices from the experts. If an inconsistency is found, a revision of the inputs to the questionnaire is requested. The opinions are aggregated, and aggregated pairwise comparison matrices are prepared. The center of gravity (COG) method is applied next to prepare defuzzified comparison matrices. The priority vectors of the defuzzified aggregated pairwise comparison matrices are calculated.

High static image quality

High motion image quality

Product stability

Rapid delivery

Reasonable price Engineering characteristics

(EC)

FQFD & FANP 0.228 0.213 0.182 0.164 0.211

Customer attributes

(CA)

CA1affects CA2

CA1

CA2

EC2 affects EC1

Figure 8. House of Quality

17

Based on the relationship among factors shown in the HOQ in Fig. 9, a pairwise comparison questionnaire is prepared, and the seven experts are asked to do the questionnaire. The consistency test is performed to check all the pairwise comparison matrices from the experts, and a revision of the inputs to the questionnaire is requested if necessary. The opinions are aggregated, and aggregated pairwise comparison matrices are prepared.Use the comparison of the importance of high static image quality (CA1) and high motion image quality (CA2) as an example. The experts’

opinions are transformed into triangular fuzzy numbers, i.e., (1/6,1/5,1/4), (3,4,5), (1,1,1), (1,1,1), (1/4,1/3,1/2), (1/6,1/5,1/4) and (1/6,1/5,1/4). Geometric mean approach is employed to aggregate experts’ responses, and the synthetic triangular fuzzy number for the comparison between CA1 and CA2 is (0.4553,0.5227,0.6292). The same procedure is carried out for all pairwise comparisons of other CAs. The fuzzy aggregated pairwise comparison matrix for the CAs is:

1 2 21

3 4 5

CA1 CA2 CA3 CA4 CA5

CA 1 (0.4453, 0.5227, 0.6292) (0.2225, 0.2624, 0.3320) (0.4202, 0.5742, 0.7430) (0.3048, 0.3712, 0.4640) CA 1 (0.2225, 0.2669, 0.3441) (0.4283, 0.5870, 0.7626) (

CA CA CA

=

W% 0.2876, 0.3451, 0.4202)

1 (0.6454, 0.8824, 1.1266) (0.3598, 0.4602, 0.5656) 1 (0.2714, 0.3173, 0.3743)

1

⎡ ⎤

⎢ ⎥

⎢ ⎥

⎢ ⎥

⎢ ⎥

⎢ ⎥

⎢ ⎥

⎣ ⎦

To prepare a defuzzified comparison matrix, the center of gravity (COG) method is applied next. For example, with the synthetic triangular fuzzy number for the comparison between CA1

and CA2, the defuzzified comparison between CA1 and CA2 is 0.5324. The defuzzified aggregated pairwise comparison matrix is:

1 2 3 4 5

1

2

3

4

5

CA CA CA CA CA

CA CA 21 CA

CA CA

1 0.5324 0.2723 0.5791 0.3800 1 0.2778 0.5926 0.3510 1 0.8848 0.4619 1 0.3210

1

⎡ ⎤

⎢ ⎥

⎢ ⎥

⎢ ⎥

= ⎢ ⎥

⎢ ⎥

⎢ ⎥

⎣ ⎦

W

The priority vector of the defuzzified aggregated pairwise comparison matrix for CAs is calculated.

18

1 2

21 3

4 5

CA 0.0883 CA 0.1124 CA 0.2473 CA 0.1738 CA 0.3783 w

⎡ ⎤

⎢ ⎥

⎢ ⎥

⎢ ⎥

= ⎢ ⎥

⎢ ⎥

⎢ ⎥

⎣ ⎦

The consistency test is performed by calculating the consistency index (CI) and consistency ratio (CR):

max 5.3355 5

0.08388

1 5 1

CI n n

λ − −

= = =

− − , and

0.08388

0.07489

RI 1.12

CR = CI = = .

Since CR is less than 0.1, the experts’ judgment is consistent. If the consistency test fails, the experts are required to fill out the specific part of the questionnaire again until a consensus is met.

The obtained priorities are entered into the designated places in the supermatrix, which is the unweighted supermatrix. The unweighted supermatrix is transformed into a weighted supermatrix first, and the weighted supermatrix is raised to powers to capture all the interactions and to obtain a steady-state outcome. The resulting supermatrix is the limit supermatrix, which shows the priority weights of the ECs:

1

1 1 2 2 3 3 4 4 5 5

G EC 0.228 EC 0.213 EC 0.182 EC 0.164 EC 0.211

f ANP f

f f f

w w w

w w w

⎡ ⎤ ⎡ ⎤

⎢ ⎥ ⎢ ⎥

⎢ ⎥ ⎢ ⎥

= ⎢ ⎥ = ⎢ ⎥

⎢ ⎥ ⎢ ⎥

⎢ ⎥ ⎢ ⎥

⎢ ⎥ ⎢⎣ ⎥⎦

⎣ ⎦

High color contrast (EC1) is the most important EC with priority of 0.228, followed by low display blur (EC2) and low contamination in TFT-LCD module (EC5) with priorities of 0.213 and 0.211, respectively.

19

Figure 9. The four HOQs for the case study

Next, multiple goals with different priority levels are considered in the NPD. While increasing customer satisfaction may be the main purpose in the QFD process, other issues such as cost expenditure and technical difficulty may also need to be taken into account in the design stage.

Let G1, G2 and G3 be goals of maximizing customer satisfaction, minimizing technical difficulty, and minimizing cost expenditure, respectively. Suppose that G1 is considered to be more important than G2 and G3; therefore, two priority levels are recommended in the QFD process.

For simplifying the computational efforts, a recently proposed model is adopted in this case study (Lee et al., 2010). With G1 belonging to priority level 1 and G2 and G3 belonging to priority level 2, the preemptive fuzzy goal programming model is as follows:

G1:

1 1 1 11 11

2 12 12 3 13 13

4 4 14 14 5 15 15

( ) ( )

( ) ( )

( ) ( )

+ − + −

+ − + −

+ − + −

× + = × + +

× + + × + +

× + + × +

∑

fi fi fi f f f

f f f f f f

f f f f f f f

w d d w d d

w d d w d d

w u d d w d d

11 11 12 12

13 13 4 14 14

15 15

0.2288 ( ) 0.2137 ( )

0.1823 ( ) 0.1642 ( )

0.2110 ( )

f f f f

f f f f f

f f

d d d d

d d u d d

d d

+ − + −

+ − + −

+ −

= × + + × +

+ × + + × +

+ × +

G2:

1 1 1 11 11

2 12 12 3 13 13

4 4 14 14 5 15 15

( ) ( )

( ) ( )

( ) ( )

i i i

t t t t t t

t t t t t t

t t t t t t t

w d d w d d

w d d w d d

w u d d w d d

+ − + −

+ − + −

+ − + −

× + = × +

+ × + + × + +

× + + × +

∑

11 11 12 12

13 13 4 14 14

15 15

0.1925 ( ) 0.2564 ( )

0.3215 ( ) 0.1582 ( )

0.0714 ( )

t t t t

t t t t t

t t

d d d d

d d u d d

d d

+ − + −

+ − + −

+ −

= × + + × +

+ × + + × +

+ × +

20

G3:

1 1 1 11 11

2 12 12 3 13 13

4 4 14 14 5 15 15

( ) ( )

( ) ( )

( ) ( )

i i i

c c c c c c

c c c c c c

c c c c c c c

w d d w d d

w d d w d d

w u d d w d d

+ − + −

+ − + −

+ − + −

× + = × +

+ × + + × + +

× + + × +

∑

11 11 12 12

13 13 4 14 14

15 15

0.2512 ( ) 0.2025 ( )

0.1584 ( ) 0.1796 ( )

0.2083 ( )

c c c c

c c c c c

c c

d d d d

d d u d d

d d

+ − + −

+ − + −

+ −

= × + + × +

+ × + + × +

+ × +

s.t:f x1( )1 −d⁺_f11+d⁻_f11 =g x_f1( 1)

21 21 2

2( 2) _{f f f} ( 2)

f x −d⁺ +d⁻ =g x

31 31 3

3( )3 _{f f f} ( 3)

f x −d⁺ +d⁻ =g x

41 41 4

4( 4) _{f f f} ( 4)

f x −d⁺ +d⁻ =g x

51 51 5

5( 5) _{f f f} ( 5)

f x −d⁺ +d⁻ =g x

11 11 11

1( )1 _{t t t} ( 1)

t x −d⁺ +d⁻ =g x

21 21 2

2( 2) _{t t t} ( 2)

t x −d⁺ +d⁻ =g x

31 31 3

3( )3 _{t t t} ( 3)

t x −d⁺ +d⁻ =g x

41 41 4

4( 4) _{t t t} ( 4)

t x −d⁺ +d⁻ =g x

51 51 5

5( 5) _{t t t} ( 5)

t x −d⁺ +d⁻ =g x

11 11 1

1( )1 _{c c c} ( 1)

c x −d⁺ +d⁻ =g x

21 21 2

2( 2) _{c c c} ( 2)

c x −d⁺ +d⁻ =g x

31 31 3

3( )3 _{c c c} ( 3)

c x −d⁺ +d⁻ =g x

41 41 4

4( 4) _{c c c} ( 4)

c x −d⁺ +d⁻ =g x

51 51 5

5( 5) _{c c c} ( 5)

c x −d⁺ +d⁻ =g x

F

X ∈ (F is a feasible set)

21

where ^{f xi( )i} ,^{t xi( )i} andc xi

( )

i

are respectively the membership function for customer satisfaction, technical difficulty and cost expenditure for all ECs, _{( )}

fi i

g x , _{( )}

ti i

g x and _{( )}

ci i

g x are respectively the targeted values for customer satisfaction, technical difficulty and cost expenditure for all ECs.

Based on different goals, different types of membership function can be used. For example, ( )

2 2

f x is the membership function for customer satisfaction of EC2, low display blurriness. If a maximum satisfaction is achieved, f x2( )2 =1 ; if a minimum satisfaction is achieved, f x2( )2 =0.

Due to copyright transfer to publishers for conference and journal papers, please refer to the publication list in the last page for complete case studies.

6. Conclusions

With limited resources, including time, cost and human power, a firm can only focus on a certain parts of its research and design. Therefore, how to develop and manufacture a product that can acquire the highest expected benefits for the firm is an important task. In this research, a systematic process that incorporates FDM, ISM and FANP into QFD was proposed for new product development. Through comprehensive literature review and interview with experts, a list of CAs that customers perceive as important for a TFT-LCD panel and a list of ECs that may be necessary for TFT-LCD panel were prepared. The most important factors from the CA and EC lists were selected by the FDM. The ISM was applied to determine the inner dependence among CAs and among ECs. The results were used to construct the HOQ, and the priorities of CAs and ECs were generated through FANP so that the inner dependence among CAs and among ECs and the linguistic uncertainty of experts could be incorporated in the calculation. In addition, a fuzzy analytic network process (FANP) model with the consideration of benefits, opportunities, costs and risks (BOCR) was constructed for supplier selection. While there are numerous supplier selection models available, most models usually only stress on the criteria that are required by a buyer, but not consider the opportunities, costs and risks aspects of the buyer when selecting a supplier.

Therefore, this research provided a comprehensive model that considers the four merits simultaneously and takes into account the interrelationships among the factors. In addition, fuzzy set theory was incorporated to overcome the uncertainty and ambiguity in human decision-making process. A case study of a TFT-LCD manufacturer in selecting the most appropriate critical-part manufacturers was introduced to examine the practicality of the proposed model.

The proposed model can help designers systematically consider relevant NPD information and effectively determine key CAs and ECs for designing and manufacturing of new products, and it can facilitate the process of selecting the most appropriate critical-part manufacturers. The model not only can be applied by a TFT-LCD manufacturer, it can also be adjusted by firms in other high-tech industries to suit the particular needs. The generated results can provide valuable references in making NPD decisions and selecting suppliers for cooperation.