Computers in Industry 59 (2008) 905–922

Knowledge-intensive collaborative decision support for design processes:A hybrid decision support model and agent§

Xuan F. Zha a,*, Ram D. Sriram a, Marco G. Fernandez b, Farrokh Mistree b

a Design and Process Group, National Institute of Standards and Technology, Gaithersburg, MD 20899, USAb System Realization Laboratory, Georgia Institute of Technology, Atlanta, USA

A R T I C L E I N F O

Article history:

Available online 7 October 2008

Keywords:

Decision-based design

Hybrid

Decision model

Fuzzy negotiation

Autonomous decision agent

Multi-agent framework

A B S T R A C T

This paper proposes a hybrid decision support model within a multi-agent framework to facilitate

integration and collaboration for design decisions. The hybrid decision model integrates the compromise

decision support problem technique (cDSP) and the fuzzy synthetic decision model (FSD) to

quantitatively incorporate qualitative design knowledge and preferences of designers for multiple,

conflicting attributes stored in a knowledge repository so that a better understanding of the

consequences of design decisions can be achieved from a more comprehensive perspective. The focus

of this work is on the provision of a hybrid decision model and framework for improved cooperative

design decision support. The novelty of the work is in combination of different models or methods,

algorithms for making collaborative design decisions in both objective and subjective natures, in

particular the fuzzy negotiation mechanism using the hybrid decision model. The developed hybrid

decision model and framework are generic and flexible enough to be used in a variety of decision

problems. Finally, the proposed hybrid decision model and framework are illustrated with applications in

collaborative concept evaluation and selection in product family design for mass customization.

Published by Elsevier B.V.

Contents lists available at ScienceDirect

Computers in Industry

journal homepage: www.e lsev ier .com/ locate /compind

1. Introduction

Design of complex engineering systems is increasingly becominga collaborative task among designers or design teams that arephysically, geographically, and temporally distributed. Collabora-tive design task essentially involves decision-making processes thatrequire evaluation, comparison and selection of design alternativesas well as eventual optimization from a systematic perspective. Forexample, a product development team is generally composed ofrepresentatives from marketing, business development, engineer-ing, and production. Those team members utilize various decision-making techniques and design criteria to develop and evaluatevarious design alternatives. Increasing design knowledge andsupporting collaboration among designers or other decision-makers(e.g. customers) to make intelligent decisions can increase designefficiency and result in higher quality designs. Achieving a knowl-

§ Commercial equipment and software, many of which are either registered or

trademarked, are identified in order to adequately specify certain procedures. In no

case does such identification imply recommendation or endorsement by the US

National Institute of Standards and Technology, nor does it imply that the materials

or equipment identified are necessarily the best available for the purpose.* Corresponding author.

E-mail address: xfzha@ieee.org (X.F. Zha).

0166-3615/$ – see front matter . Published by Elsevier B.V.

doi:10.1016/j.compind.2008.07.009

edge-intensive design support system is hampered by at least thefollowing factors:

� E

xisting product design and decision knowledge modelling andsupport schemes have deficiencies in their representationalcapabilities. Very few provide explicit representations of thedesign decision and rationale as part of a comprehensive productdesign knowledge representation, which could not be directlyused for design decision support. These issues reinforce the needfor consensus among product development software developersand users regarding the content and form for more compre-hensive representation of design knowledge including designdecision knowledge. � M odels or methods, algorithms for making design decisions in

both objective and subjective natures are not unified andseamlessly integrated. Currently available algorithms for opti-mization and constraint satisfaction have weaknesses; morerigorous algorithms tend to be too slow, heuristics, and toounreliable. It is therefore needed to develop a hybrid decisionsupport model under an integrated framework to facilitate theseamless integration of objective and subjective aspects incollaborative design decisions.

� C urrent design decision support systems suffer one or more of the

following drawbacks: a heavy dependency on the preferences,

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922906

experience and knowledge of designers; no built-in ways fordecision-making knowledge representation; no efficient mechan-isms to share, exchange, and reuse the decision knowledge; noefficient mechanisms to coordinate decision activities, provideadvisory services, and explain the results and what-ifs. Thus, anintelligent decision tool that can provide efficient coordinationand decision-making mechanisms for designers is still needed.

Designers distributed geographically and organizationally inglobal economies worsen the above problems. Multiple firmscoordinate to develop and exchange product specifications anddesign decisions, and to reach compromised decisions bynegotiations. In these situations, designers cannot rely on informaldiscussion to make joint decisions and resolve differences orconflicts in product design, or to combine separate contributions tothe same product definition and in-process decision. The followingchallenges for product specifications and design decisions arisefrom the above problems:

� S

pecify more accurately which as-generated alternatives andevaluation and selection of them satisfy the design requirements,and which do not. This enables tools to incorporate the sameinterpretation of product specifications and design decisionsmade more reliably and accurately. � P rovide a structured and formalized comprehensive decision

representation and framework for modeling the characteristicsof multidisciplinary and multi-objective decision problems bothin subjective and objective natures, which can handle qualitativeand quantitative knowledge.

� D evelop a sound, practical trade-off based decision model that

can quantitatively incorporate qualitative knowledge andpreferences for multiple, conflicting attributes stored in a designknowledge repository so that a better understanding of theconsequences of design decisions can be achieved from a morecomprehensive perspective.

� E stablish a hybrid decision model and a comprehensive

collaboration framework which may integrate more individualdecision techniques such as the compromise decision supportmodel, decision involving uncertainty (e.g., fuzzy systems), andintelligent agents to solve decision problems and help designerscoordinate in making rapid and more intelligent decisions duringthe collaborative design process.

There have been many research efforts on the development ofintelligent design decision support systems including collaborativedecision support systems [1–8], but the challenges above are stillnot well addressed. Earlier studies focused on integrating reason-ing mechanisms in design support systems [2,9]. Recent studiesfocus on decision system or task structures, unified data andknowledge modeling and representation, concentrating on howthese are integrated [8,10–13]. Latest research indicates a trendtowards semantic data and knowledge modeling for betterinteroperability of integrated decision support systems [14,15].Rather than attempting to design a new algorithm withoutweaknesses – a task that is difficult if not impossible – someresearchers have been working on ways to organize algorithms sothat they can suppress their respective weaknesses throughcooperation, and together achieve what separately they mightnot [12,16]. Section 2 gives a comprehensive review of relatedwork on the design decision support.

This paper aims to address the challenges above with aknowledge-intensive collaboration paradigm including a unified,hybrid decision support model for decision-based design and aknowledge-supported decision support framework. The developedmodel and framework aims to serve as a basis for the seamless

integration of stakeholders and decision-makers involved in thecollaborative decision-based design process. Design decisionmodeling in this paper refers to analyzing subjective as well asobjective design conditions, generating design alternatives, andchoosing the most appropriate one among them (see Section 3). Ahigh level of interoperability for integrated design decision-making reduces the cost of decision, increases reliability, opensopportunities for innovation, and provides stronger competition inthe international marketplace. This paper covers the aspects ofdecision modeling within the above scope, such as decisionsupport process, decision-based design, decision taxonomies,interconnections of decision models (a hybrid decision model),performance evaluation, and a multi-agent collaboration frame-work. It does not address more detailed topics such as decisionprocess models, alternative generation models, and other agents inthe framework. The approaches of this paper will be applied tothese topics and the wide product life-cycle decision support infuture work, see Section 7.

The organization of this paper is as follows. Section 2 reviewsthe previous research related to design decision support andcurrent status. Section 3 discusses the design decision supportprocess and decision-based design. Knowledge-intensive decisionsupport for design processes is highlighted. Section 4 proposes ahybrid decision mode that integrates the compromise decisionsupport problem model (cDSP) and the fuzzy synthetic decisionmodel (FSD). Section 5 presents a knowledge-supported decisionagent within a multi-agent collaborative decision support frame-work. Section 6 provides an application of the proposed hybriddecision model in the concept evaluation and selection stage andtwo case studies. Section 7 summarizes and concludes the paperand gives future work.

2. Literature review

Design decision support problems necessitate the search forsuperior or satisfying [17,18] design solutions, especially in theearly stages of design, when all of the information needed tomodel a system comprehensively may not be available. Currentresearch in design decision support (particularly pertaining todecision-based design) is focused on enabling decision technol-ogies to assist product designers to make decisions during thedesign process [5,19–21], where the primary emphasis is onsupport for information management related to decision-making.The principles and methods developed from game theory,decision and risk analysis, and utility models are incorporatedinto the design process for complex systems, where consistency isnecessary at all levels of decision making in order to resolveconflict and intransivity, and to facilitate negotiation and opti-mization. Generally, the literature on design decision supportincluding integrated, collaborative decision support can beclassified into the following categories [22]: multi-criteria utilityanalysis [23,24], fuzzy set analysis, probability analysis, inte-grated design analytic methodology, and the information contentapproach [1]. The following review is focused on the first fourapproaches, due mostly to their current popularity or potential forcollaborative decision support.

Multi-criteria utility analysis, originally developed by vonNeumann and Morgenstern [25], is an analytical method forevaluating a set of alternatives, given a set of criteria. It has beenapplied widely in the areas of engineering and business for decision-making [26]. Thurston [27] has applied this technique to thematerial selection problem, where alternatives are evaluated basedon utility functions that reflect the designer’s preferences formultiple criteria. Similar work by Fernandez et al. [24] addressed thedifficulties associated with resource selection for rapid prototyping

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922 907

by synthesizing the selection Decision Support Problem with utilitytheory. Mistree et al. [19,28] modeled design evaluation as acompromise Decision Support Problem (cDSP) and employed goalprogramming techniques to make superior compromise decisions.While mathematical programming and utility analysis enhancealgorithm-rigorous optimization modeling, such methods requirethe expected performance with respect to each criterion to berepresented in a quantitative form. They are not appropriate for usein the early design stages, where some qualitative (i.e., intangible)design criteria, are involved and difficult to quantify [24,29]. TheAHP mechanism proposed by Saaty [30] is widely recognized as auseful tool to support multi-criteria decision-making (MCDM)problem.

Fuzzy analysis, based on fuzzy set theory [31], is capable ofdealing with qualitative or imprecise inputs from designers. It doesso by describing the performance of each criterion with linguisticterms, such as ‘‘good,’’ ‘‘poor,’’ and ‘‘medium,’’ and has proven to bequite useful in decision-making problems with multiple goals orcriteria [32]. Wood and Antonsson [33] have demonstrated itsviability in performing computations with imprecise designparameters in mechanical design. Wood et al. [79] compared fuzzysets with probability methods and concluded that the fuzzy analysisapproach is most appropriate when imprecise design descriptionsabound, while a probability analysis approach is best suited fordealing with stochastic uncertainty. Thurston and Carnahan [29]revealed that the fuzzy analysis approach is useful and appropriateat very early stages of the preliminary design process. Knosala andPedrycz [34] utilized the analytic hierarchy process (AHP) method[30] to construct membership functions for the performance andweight of each criterion, and then applied the fuzzy weighted meanof the overall evaluation to ranking alternatives. Carnahan et al. [35]represented evaluation results and weights regarding each criterionwith linguistic terms and ranked alternatives based on the fuzzyweighted mean of distance from a fuzzy goal. While the fuzzyanalysis approach excels in capturing semantic uncertainty vialinguistic terms, it requires discreet deliberation in dealing withcrisp information specifically. A domain-specific method is neededto ‘‘fuzzify’’ each tangible criterion whose evaluation is naturallyestimated as an ordinary real variable. Another challenge is theincomparability/compatibility among various criteria [36,37],which necessitates mechanisms to be capable of converting varioustypes of performance evaluation with respect to different criteria toa common metric so as to allow for the specification of suitablemembership functions for these.

Design evaluation usually involves both tangible and intangiblecriteria, along with quantitative and qualitative performancemeasures. This necessitates a hybrid approach for combining thequantitative, normative problem structuring capabilities of opera-tions research techniques with the qualitative, descriptive problemsolving approaches used in artificial intelligence research. Forexample, Maimon and Fisher [38] presented a robot selection modelusing integer programming and a rule-based expert system.Considerable research efforts have been devoted to fuzzy goalprogramming for modeling mathematically the imprecise relation-ships implicit in fuzzy goals and soft constraints [10,39]. However,these efforts mostly model a particular aspect of uncertainties indesign evaluation, such as inexact relationships, imprecise informa-tion, and uncertain information [34]. It is difficult for a fuzzy goalprogramming model to consider all sources of uncertaintycoherently during the preliminary stages of design [35]. In addition,computational complexity is a key issue, especially if a large numberof design alternatives and criteria are involved [36,40].

To reflect customer preferences in multi-criteria design evalua-tion, the relative importance or weighting for each criterion has beenconsidered by numerous evaluation procedures [22]. Frazell [41]

assigned weights to criteria on a 0–100 scale. Sullivan [82] presenteda similar method called the linear additive model, in which rankingis included. Huang and Ghandforoush [83] offered anotherprocedure for quantifying subjective criteria. They computedintangible criteria measures as the multiplication of the intangiblecriterion weights by the subjective customer rating. Dixon et al. [84]measured the performance by degree of satisfaction, ranging fromexcellent to unacceptable, and combined this measure with prioritycategories of high, moderate, or low to evaluate the design. Nielsenet al. [42] used factor-criteria to establish the level of importance ofattributes. A priority level (absolutely necessary, important, ordesirable) is indicated for each factor-criterion and is used to guidedecision-making. The main drawback of these evaluation methods isthat they ignore inconsistency on the part of the decision maker. Theanalytic hierarchy process (AHP) was developed to deal with thedecision-maker’s inconsistency and to more accurately mimic thehuman decision-making process [30]. In AHP weights are deter-mined by means of pairwise comparisons between hierarchicaldecision levels. It has been proven to be a more rigorous procedurefor determining customer preferences and has been approachedfrom the fuzzy point of view by Boender et al. [40]. Carnahan et al.[35] also proposed to ‘‘fuzzify’’ the weights subsequent to theirhaving been obtained via AHP. However, they are not used forcollaborative evaluation and selection in design decision.

There are also many other product feasibility and qualityassessment tools that are useful for planning and evaluating thedesign concepts of products. Examples include quality functiondeployment (QFD) [43], concurrent function deployment (CFD) [44],the conceptual selection matrix [45], Taguchi’s robust designmethod [46], etc. Quality function deployment (QFD) provides aset of matrix-based techniques to quantify the organizationalcharacteristics and identify qualities that would meet customerexpectations and needs. While QFD addresses only qualitativeaspects, CFD deals with total life-cycle concerns from a concurrentengineering perspective. The concept selection matrix initiallyproposed by Pugh [45] is another matrix-based approach employedto quantify and measure product quality. It is based on a list ofproduct and customer requirements. The purpose of Taguchi’srobust design method is to reduce or control variations in a productor process [46]. Depending upon the complexity and stage of adesign, there could be a large number of iterations required. Whilethese methodologies provide high-level guidelines for designevaluation, detailed supporting techniques are essential. Prasad[44] also noted that 4Ms (models, methods, metrics and measures)are the core in integrated product development.

With the emergence of collaborative design, researchers areaddressing enabling technologies or infrastructure to assistproduct designers to make decisions in the computer or net-work-centric design environment. Some techniques are intendedto help designers collaborate or coordinate by sharing product anddesign information through formal as well as informal interac-tions, while others are geared towards conflict management [4–8,47,48]. There are other approaches for solving the resolution ofconflicts, i.e., Altshuller [49] concept of conflict resolution bychange of used components or the algorithm for conflictelimination using TRIZ (Theory of Inventive Problem Solving)and DOE (design of experiment) [50], etc. Most decision supportsystems can only calculate satisfaction levels and are based uponthe mathematical programming, utility analysis and algorithm-rigorous optimization modeling principles and approaches (e.g.,cDSP and game theory). They are data and information based, andthus cannot handle knowledge explicitly. They are more appro-priate for quantitative (tangible) criteria but not for qualitative(intangible) criteria (difficult to quantify). There is a need foradding unique analysis and reporting features, including: the

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922908

probability that a particular alternative is the best choice;assessment of the level of consensus for each alternative; guidanceon what should be done next; and documentation of the entiredecision-making process. In early stages of design decisions are ill

structured and often supported with scarce information. Multiplepotential solutions and limited predictability all contribute todesign complexity [51]. Moreover, significant functional andtechnical barriers often prevent the free flow or sharing andexchange of the necessary information and knowledge, especiallyfor design decision-making [52].

3. Knowledge-intensive design decision support

Decision-based design addresses the cognitive ‘‘structuring’’ ofa design problem; the drive for innovation where the existingdesign ‘‘structure’’ or design solution space is ill-defined orinsufficient; the need to reduce complexity by mapping to whatwe know; and the consistent use of decision technologies tooptimize the decision-making capabilities within the design spacethat has been created [48]. Decision-making generally involvesrealizing a goal by analyzing subjective as well as objectiveconditions, generating alternatives, and choosing the mostappropriate one among them. A generic decision support processcan be described as having the following interactive aspects:intelligence, design, choice and implementation [18]. It involvesseveral stages ranging from problem identification and classifica-tion, simplification of assumptions, data collection, modelformulation, solution alternative generation, evaluation, selection,model verification and validation, and testing of the proposedsolution to final implementation of the devised plan. Currentresearch is focused predominantly on how knowledge support canaid the decision-maker during the design process.

The main role of a designer is to apply scientific and engineeringknowledge to find (generate, evaluate and select) the solutions todesign problems, and then optimize those solutions within theframework composed of requirements and constraints set byphysical, environmental and human-related considerations.Design can be viewed as the process of converting informationthat characterizes the needs and requirements for a product intoknowledge about it. Based on the principle of decision-baseddesign, the design equation can be expressed as follows [19]:{K} = T{I}, where K is a knowledge output, I is an information input,and T is a transformation relationship, respectively. Thus, knowl-edge-intensive support becomes more critical in the designprocess and has been recognized as a key enabling technologyfor retaining competitive advantages in product development.

Generally, the problem of design decision (evaluation andselection) can be defined as follows: given a set of designalternatives, the designer or agent/system evaluates them andselects one among them that can best satisfy customer needs,meet design requirements and constraints, and fit the technicalcapabilities of a company. This paper proposes a knowledge-intensive design decision support scheme (KIDDSS), as depicted inFig. 1, in which the design decision support is exploited from theperspective of synthesis of design process modeling (DPM),knowledge management (KM), and decision support (DS). Fig. 1also shows a scenario of implementing the KIDDS from theperspectives of decision knowledge management (DKM), includ-ing knowledge generation and acquisition, codification, proces-sing and utilization (reasoning), etc. The principles and methodsdeveloped from decision theories including game theory, utilitytheory, probability theory, fuzzy set theory, among others, playa key role in implementing KIDDS framework as they areincorporated into the design process (see [21,53] for discussionof some of these techniques).

The importance of developing fully automatic algorithmsfor choosing design options in the face of objective and subjectivedesign criteria which often conflict are explained in the literature.The knowledge-intensive design decision support processesshown in Fig. 1 underscore that it is also difficult to developautomatic algorithms for such situations. In next sections, instead,a suite of algorithms which work together in an integrated fashionare proposed to select design choices where there are poten-tial conflicts in the design specifications. These algorithms arefeeding into and supporting the larger vision of a design supportenvironment.

4. The hybrid design decision model: technical description

One of the important features of design is that the process ofdesign is a decision-making process. Designers will be forced tochoose actions in each refinement step of the design object. Thefinal design is certainly definite and unambiguous. However, whendesigners choose an initial design solution or formulate the initialmodel, it always involves considerable uncertainties. By uncer-tainty, we mean that imprecision of the design problem has thenature of possibility. Formulated design goals and constraints areoften subjectively estimated by designers. Conventional mathe-matical programming uses a certain objective function (or multi-goal function) and a set of certain constraints. Then designers mustgive precise values to each constraint or goal. Research intouncertainty in early design has been reported from the followingtwo aspects [39]: (1) using case-based approximate reasoning togenerate design solutions and (2) formulating the design problemin a fuzzy environment. We are interested in decision-makingmodel with the aid of computing support because a successfulcomputer-based model for a design process depends largely onhow many methods/models could be used to assist designers indesign decision-making. In this section, a hybrid decision model isproposed, which can integrate one or more decision-makingtechniques such as the cDSP, fuzzy systems, and fuzzy cDSP tosolve complex design decision problems.

4.1. The compromise decision support problem model (cDSP)

Decision support problems (DSPs) are generally formulated usinga combination of analysis-based information and engineeringjudgment in the form of viewpoints, post-solution sensitivityanalysis, bounds, and context for decisions to be made. Two primarytypes of decisions are supported within the DSP technique: selection

and compromise. Complex decisions are supported through theircombination. The ‘‘selection’’ type decision actually includesevaluation and indication of preference based on multiple attributesfor one among several feasible alternatives, while the ‘‘compromise’’type decision involves the improvement of a given alternativethrough modification. Another aspect of the DSP technique that isparticularly relevant to distributed collaborative design is the abilityto express decisions that are linked together such as coupled andhierarchical decisions through combinations of selection andcompromise DSPs (i.e., selection–selection, compromise–compro-mise, and selection–compromise, see Fig. 2) [21,54,55]. Thesederived decision constructs are ideally suited for modeling networksof concurrent and sequential decisions that share information andknowledge.

In the compromise decision support problem model (cDSP), asshown in Fig. 3, a hybrid of goal programming and mathematicalprogramming is used to determine the values of design variablesthat satisfy a set of constraints and achieve as closely as possible a setof conflicting goals [19,28]. We will not embark on a throughevaluation of the acceptability of conventional rigorous mathema-

Fig. 1. Knowledge-intensive design decision support scheme (KIDDSS).

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922 909

tical programming; nevertheless, it has at least two limitations: (1) aselected initial solution has much influence on the effectiveness andproductivity of the program; (2) the true constraints or goals areoften imprecise and subjective in nature rather than exact. Forexample, the boundary or performance constraints are usually not‘‘hard’’ but very ‘‘soft’’. During the design process, the designer needsto start with approximate assessment of input variables/parameters.The value of mathematical precision is often reduced with increasing

Fig. 2. Primary and d

complexity of the system. Fuzziness could be introduced into theformulation of the cDSP by way of fuzzy numbers which are used todescribe the design environment.

Fuzzy set theory has been widely used in conventional opti-mization methods so that the normal objective function and/or aset of constraints are a fuzzy field, which are assigned bymembership functions. Thus, the model of fuzzy decision-makingunder constraints is formulated, and further, the so-called fuzzy

erived decisions.

Fig. 3. Compromise decision support model (cDSP).

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922910

compromise DSP (fuzzy cDSP) is attained [10], see Appendix A.Noted that using this approach, although the system or designproblems modeled may be fuzzy, the solution is crisp. Compared tothe conventional value-based decision-making, the significance ofthis approach is that designers are no longer forced to give a precisedescription for relevant design variables and goals.

4.2. Fuzzy synthetic decision model (FSD)

As stated above, the design problem is to generate and select afamily of design solutions (DS) that must satisfy a certain set offunctional requirements (FR) and design constraints (DC). Therelationships between function requirements, design constraints,and final design solutions can thus be represented as three fuzzymatrices M (FR � DS), M (DC � DS), and M (DS � DS). A critical stepin the design process is to map a fuzzy set of requirements to a crispset of design solution alternatives. To combine expert judgmentand process-useful knowledge for fuzzy decision-making, wepropose a fuzzy synthetic decision model (FSD) for fuzzy designcomprehensive evaluation and selection in this section based onthe improved fuzzy AHP, ranking algorithms and inferencemechanisms for engineering design evaluation and selection.

4.2.1. Fuzzy analytic hierarchy process

The AHP mechanism [30] provides a compositional approachto the multi-criteria decision-making (MCDM) problem that isfirst structured into a hierarchy of interrelated elements and thena pairwise comparison of elements in terms of their dominance iselicited. The weights are given by the eigenvector associated withthe highest eigenvalue of the reciprocal ratio matrix of pairwisecomparisons. The pairwise comparison ratio used to comparethe importance of criterion i and criterion j, that is wi and wj, isdefined as:

ai j ¼wi

w j(1)

Considering a pairwise comparison matrix A = [aij] and animportance index (weight) vector W = [wi], their relationship canbe described according to:

AW ¼ nW (2)

when A is given, W and n are calculated as an eigenvector and aneigenvalue of A, respectively. The pairwise comparison matrix A

must be examined for reliability of consistency, and the consis-tency index (CI) is calculated as:

CI ¼ lmax � n

n� 1(3)

where, lmax is the maximum value of 0. If the value of CI is higherthan 0.1, then the matrix should be reset by comparing importanceagain.

Most of current approaches compose the comparison matrix A

in the AHP according to designer’s or user’s individual preferencesonly. In a collaborative design environment, negotiation should beflexible; design agents may change their offers according to theircounter offers. Each agent has its own comparison matrix A andexchanges its matrix with other agents to cooperatively adapt tochanges during the design process. Thus, the comparison matrix A

should be built dynamically. To this end, this paper combines fuzzymembership functions with the AHP to pursue the preference ofdesigners or agents dynamically, and as a result, the fuzzycomparison matrix A is generated. This will also be used in fuzzynegotiation for design collaboration.

4.2.2. Fuzzy ranking, evaluation and selection for design alternatives

We treat design evaluation as a fuzzy multi-criteria decision-making problem, in which the fuzzy ranking of design alternativesis carried out by evaluating a set of design alternatives against a setof design criteria. For example, the linguistic terms ‘‘very low,’’‘‘low,’’ ‘‘fairly low,’’ ‘‘medium,’’ ‘‘fairly high,’’ ‘‘high,’’ and ‘‘veryhigh’’ themselves constitute a fuzzy evaluation set. A set with m

design alternatives A = {a1, a2,...,am} is to be evaluated by the fuzzyevaluation set in terms of n criteria C = {C1,C2,...,Cn}. The fuzzyrating ri j to criterion Cj for alternative ai is characterized by themembership function mRi j

ðri jÞ, ri j 2R, and a set of weights W ¼fw1; w2; :::; wng is fuzzy linguistic variables characterized by themembership function mW j

ðw jÞ, w j 2Rþ. Consider the mappingfunction giðziÞ : R2n!R defined by

giðziÞ ¼Pn

j¼1ðw jri jÞPnj¼1 w j

(4)

where, zi ¼ ðw1w2:::wn; ri1 ri2:::rinÞ: Define the membership func-tion mZi

ðziÞ by

mZiðziÞ ¼ ^ j¼1;:::;n

o mW jðw jÞ ^ k¼1;:::;n

o mRikðrikÞ (5)

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922 911

where, ^ � is the minimum calculation operator. Through themapping giðziÞ : R2n!R, the fuzzy set Zi induces a fuzzy rating setRi with the membership function as:

mRiðriÞ ¼ supZigðziÞ¼ri

mZiðziÞ; ri 2R (6)

This membership function mRiðriÞ characterizes the final fuzzy

rating of design alternative ai. It does not mean the alternative withthe maximum mRi

ðriÞ is the best. The following procedure must beemployed to further characterize mRi

ðriÞ with the following twofuzzy sets [39]:

(i) a

conditional fuzzy set is defined with the membershipfunction:

mI=Rðijr1; :::rmÞ ¼ 1 if ri > rk; 8 k2 ð1;2; :::;mÞ0 otherwise

�(7)

a fuzzy set is constructed with the membership function:i¼1;:::;m

(ii)

mRðr1; :::rmÞ ¼ ^ o mRi

ðriÞ (8)

A combination of these two fuzzy sets induces a fuzzy rating setin which the best design alternative can be determined with thehighest final rating, i.e.,

mIðiÞ ¼ supr1 ;:::rmmI=Rðijr1; :::rmÞ ^ omRðr1;:::rmÞ (9)

Compared with Eq. (4), the fuzzy ranking for design alternatives ismore flexible and presents uncertainty better. Based on thismethod, designers can effectively and consistently incorporatefuzzy linguistic weights and ratings such as ‘‘good,’’ ‘‘fair,’’‘‘important,’’ and ‘‘rather important,’’ in design alternativeevaluation. The fuzzy comparison matrix A in the fuzzy AHPabove (Section 4.2.1) can be used to determine the fuzzy linguisticweights for complex design evaluation. In some cases, a simplifiedmodel is employed to integrate linguistic terms and fuzzy numbersinto the fuzzy preference model. The universe of discourse is afinite set of fuzzy numbers used to express an imprecise level ofperformance rating and weight of each criterion [56].

For a complex design problem, designers are often required toconsider not only product functionality, but also other criteriaincluding compactness and life-cycle issues, such as manufactur-ability, maintainability, reliability, and efficiency. Some of thesecriteria may contradict each other. Designers should analyze thetrade-offs among various criteria and make the ‘‘best’’ selectionfrom various criteria combinations and available alternatives forevaluation. Therefore, it is important to have a powerful searchstrategy that will lead to a near optimum solution in a reasonableamount of time. Here, the A* search algorithm [57] is employed forthis purpose. The system calculates the weighted performancerating aggregation of each retrieved alternative by analyzing thetrade-off among various criteria and their combinations as anevaluation index [81]. Thus, the evaluation index is used as aheuristic evaluation function fh by considering all the weightedperformance ratings ri ði ¼ 1;2; . . . ;mÞ of its constituent membersand the number k of its unsatisfied customer requirements, asfollows:

f h ¼Xm

i¼1

ð1=riÞ þ k (10)

where, ri 2 ½0;1� is the numerical weighted performance rating ofthe design alternative ai; 1=ri ¼ ð1;þ1Þ is defined as theperformance cost of design alternative ai. A higher weightedperformance rating of a design alternative corresponds to a lowerperformance cost.

Pmi¼1ð1=riÞ represents the accumulated perfor-

mance cost of the design alternative along the search path thus far.k is a heuristic estimate of the minimum remaining performance

cost of the design alternative along all the possible succeedingsearch paths. fh is the estimate of the total performance costs of thedesign alternative. In Eq. (10), a higher ri translates to a better-aggregated performance of each retrieved design alternative ai, anda lower m or k, i.e., a higher compactness of a design alternative,will result in a lower evaluation index fh [81]. Thus, at each step ofthe A* search process, the best design alternative, i.e., the one withthe lowest value of the heuristic evaluation function is selected, bytaking into account multiple factors, including design compactnessand other life-cycle issues such as manufacturability, assembl-ability, maintainability, reliability, and efficiency.

4.2.3. Fuzzy knowledge base and inference mechanism

When evaluating and selecting design solution alternatives,fuzzy rules regarding the characteristics of the design solutions canbe identified and used after analyzing various possible designsolutions and discussing with marketing groups. The fuzzyknowledge base is composed by two components: the linguisticterms base and the fuzzy rule base. The former can be divided intotwo parts: the fuzzy premises (or inputs) and the fuzzy conclusions(or outputs). Both parts may contain more than one premise andone conclusion. The fuzzy synthetic design decision model wouldsupport fuzzy rules with n multiple inputs and m multiple outputs(MIMO). The fuzzy knowledge base expressed by k (k = 1, 2, . . .K)MIMO-type heuristic fuzzy rules may be written in the form:

Rk : ifx1 is Xk1 is and . . . and xn is Xk

n: Then y1 is Yk1 and

. . . ym is Ykm (11)

where, fXki g

ni¼1 denotes values of linguistic variables; fxk

i gni¼1 is

conditions defined in the universe of discourse fXki g

ni¼1; fYk

j gmj¼1

stands for the value of the dependent linguistic variable; andfyk

jgmj¼1 is conclusions (outputs) defined in the universe of

discourse fYkj g

mj¼1. The global relation aggregating all K rules is

given as R ¼PK

k¼1ðRkÞ, where, S denotes any t- or s- norm or

average. For a given set k = l of fuzzy inputs or observations fXlig

ni¼1,

the fuzzy outputs (or conclusions) fYljg

mj¼1 may be expressed as

[70]: fYljg

mj¼1 ¼ ðXl

1; :::;XlnÞ�R ¼ Xl

n�:::�ðXl2�ðXl

1�RÞÞ, where, � denotesa compositional rule of inference which combines fuzzification andinference stages. Finally, determining the centre of gravity is usedas the defuzzification method for inference mechanism:

fyljg

mj¼1 ¼

PfYl

jgmj¼1ðy jÞy jP

fYljg

mj¼1ðy jÞ

(12)

4.3. Integration and cooperation of cDSP and FSD decision models

The cDSP model is basically data and information centric andmore appropriate for implementation in conjunction with tangible(quantitative) criteria rather than with intangible (qualitative)criteria. The FSD model is knowledge based and able to handle bothintangible and tangible criteria (e.g., from fuzzy requirements tocrisp design parameters). The synthesis of the cDSP and FSDmodels can generate a more powerful decision model. Fig. 4provides a schematic view of the hybrid decision model integratingthe cDSP and FSD models and their cooperation.

The integration and cooperation could be either ‘‘loose’’ or‘‘tight.’’ In the ‘‘loose’’ mode, two or more models are combinedwith parallel or serial modeling and they work together butcomplement each other. Depending on the nature of the decisionproblem, a rule-based adaptor is employed in the model. Thisadaptor serves as a regulatory switch to adapt the decisionproblems by shifting the paradigms from one decision model (e.g.,the cDSP) to another (e.g., the FSD). In the ‘‘tight’’ mode, two or

Fig. 4. Hybrid decision support model.

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922912

more models co-exist and are integrated into a single unifiedhybrid model, for example, fuzzy cDSP mentioned above.

The decision of which models are to be used is not straight-forward, it depends on many factors: degree of understanding ofthe design process, design process decomposition (stages), designcomplexity, quantitative (tangible) and qualitative (intangible)criteria, availability of expert knowledge about the design processunder consideration, level of uncertainty, etc. Taking into account

Fig. 5. Structures of the hy

these factors, several structures of the hybrid decision model areillustrated in Fig. 5. Fig. 5(a) shows a rule-based selectorincorporated in the scheme to switch on the most appropriateseparate model for the current state of the process. Each of theseparate models possesses its own input subspace and is tuned tobe optimal for corresponding design specifications. For instance,two demonstrative heuristic rules for choosing the cDSP or the FSDmodels are given as: (1) if the problem is data or information

brid decision models.

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922 913

centric and the criteria are tangible or quantitative then the cDSPmodel is used; (2) if the problem is knowledge-centric andthe criteria are intangible or qualitative then the FSD model isused. Fig. 5(b) gives a serial structure of the hybrid model, inwhich two or more separate models (e.g., cDSP and FSD) are used toformulate hybrid models and model process behaviors anddecisions at different design stages. Fig. 5(c) is a parallel structureof the hybrid model, in which two separate models (e.g., cDSPand FSD) are used to formulate hybrid models in parallel.Fig. 5(d) is a unified structure of the hybrid model, i.e., the fuzzycDSP, in which two or more models (e.g., cDSP and FSD) co-exist and are integrated into a single unified model. Fig. 5(e) and (f)are gain scheduling like structures, in which the above fuzzyranking algorithm for some design parameters or variables isadopted, or the compromised results obtained by the cDSP modelare used as references, respectively. Providing enough data fortuning multiple fuzzy models is a challenging problem when thequantity of special experiments must be restricted. A scenario ofapplications in conceptual design decision support is provided inSection 6.

The hybrid decision model is a kind of knowledge-supportedmodel that can manage both qualitative and quantitative designdecision knowledge and provide real-time or on-line support todesigners during the decision-making process. More specifically,the benefits using the hybrid decision model are: (1) the lack of aformal means of incorporating qualitative information in the cDSPis addressed; (2) design solutions are suggested and explanationsprovided; (3) use in the early design stages becomes feasible; and(4) designers are stimulated in generating new design ideas (withlearning, continuously taking place, being captured).

4.4. Used as collaborative decision-making mechanisms

Coordination has been identified as the major problem withmulti-agent systems as it is essential to any kinds of collaborations.In its definition, loosely integrated collaborative activities arecoordinated by the protocol; tightly integrated collaborativeactivities need coordination mechanisms. The coordinationproblems discussed in this paper include cooperation and conflictresolution mechanisms for multi-agent collaborative decisionsupport systems. The common ways to resolve the conflicts arearbitration and negotiation. Arbitration is based on the classicmathematical theory and reasoning rules according to the conceptof the characteristic function ‘‘to be or not to be, {yes, no} or {0,1}.’’The negotiation is a form of decision-making with two or moreactively involved agents who cannot make decisions indepen-dently but must make concessions to achieve a compromise.

The cDSP and FSD hybrid decision model proposed in Section 4can be used for resolving decision conflicts in and between agents.As mentioned in Section 4.1, the cDSP model uses a hybrid of goalprogramming and mathematical programming to determine thevalues of design variables that satisfy a set of constraints andachieve as closely as possible a set of conflicting goals. Thus, thecDSP model is a well known approach for efficient conflictresolution and negotiation; it is already widely applied in decisionconstructs [21,54,58]. These derived decision constructs are ideallysuited for modeling networks of concurrent and sequentialdecisions that share information and knowledge in collaborativedesign. Due to the limitation of space, this section only focuses onthe hybrid decision model for design decision conflict resolutionand negotiation. The negotiation is based on fuzzy set theory andreasoning rules according to the concepts of membership functionsand fuzzy comparison matrix [63]. The fuzzy negotiationmechanism using the hybrid decision model is composed of thefollowing six phases:

1. In

itial offering option settings for a design. The negotiationmechanism is started with the ‘initial option settings for adesign’ of the agent or designer. In this phase, the negotiationagents offer their negotiation conditions reflecting their relativepreferences for the design which is composed by quantitativeconditions such as parameter values and cost. However, thefuzzy values for these conditions are changed by fuzzymembership functions reflecting qualitative conditions suchas relative preferences.

2. F

uzzy membership functions. After obtaining the initial designoptions of each agent or designer, ‘fuzzy membership functions’are used to support the construction of the fuzzy pairwisecomparison matrix A. With these fuzzy membership functions,designer’s relative preferences are transformed into fuzzymembership values. During the transformation process, bell-shaped (or Z, l, p, S-type) fuzzy membership functions can beadopted.

3. P

airwise comparison matrix A. The ‘pairwise comparison matrixA’ is constructed. In this phase, the AHP comparison matrix isused to compute the relative importance of each alternative.Thus, design options of each agent are exclusively compared.

4. S

election of preferred design options. Based on the result ofcomparison in phase (3), the preferred design options areselected by one or several agents or designers. However, this isonly the first step of dynamic negotiation process.

5. R

evision of design options and negotiation. In this phase, eachagent or designer revises its ‘initial design options for a design’and continues to negotiate with its counterpart. For thispurpose, the ‘goal-seeking’ methodology (with cDSP or fuzzycDSP) is used to revise the initial design options.

6. O

ptimal design. The fuzzified pairwise comparison matrix A andthe AHP inference mechanism are used to suggest the optimaldesign, and then go to the Phase (5) to lead towards a consensuswith their counterparts. As a result, each designer or agent couldbe satisfied with the final design.

5. Overview of knowledge-intensive decision support agent forcollaborative design decision

As design becomes increasingly knowledge-intensive andcollaborative, the need for computational frameworks to supporta collaborative product development environment becomes morecritical [4,12,67]. The overall knowledge-intensive collaborativedesign framework proposed in this paper is a multi-agent systemframework that consists of a design process modeling andmanagement agent, a knowledge capture agent, a knowledgerepository, co-designers, a decision support agent, meta-systemagent, etc. Designers are also agents. The core of the scheme is thedecision support agent and it is the focus of this paper. Theknowledge repository is used to store, share, and reuse corporatedesign knowledge, including decision agent. The role of knowledgefor next generation CAD systems is discussed [65]. The coordina-tion, communication and execution mechanisms between theseagents are modeled as FIPA contract net [59] based extension withknowledge-based Petri nets [13] combined with other mechan-isms supported by the hybrid decision model in Section 4.4.

The decision support agent is a container specialized inproviding decision-making services. The comparative ranking ofalternatives and selection/compromise decision-making is afundamental component of the design decision agent. As reviewedin Section 2, several formal decision models exist. The proposedhybrid decision model of cDSP and FSD in Section 4 is used as thekernel of the decision support agent. The decision support agentcontains criteria which pair design attributes (variable modules)

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922914

with preference modules (a type of variable module used to definepreference functions) and provides an overall multiple attributeevaluation and selection service while each criterion evaluates asingle attribute. The criterion relations calculate the worth of thedesign attribute based upon the preference model, while thedecision support agents automatically generate relations toaggregate single attribute evaluations for multiple attributedecisions. Thus, there are different types of decision agents basedon the hybrid decision models and structures in Section 4.3. In theproof-of-concept implementation the decision agents have beendeveloped by integrating the cDSP technique with the knowledge-based FSD model into the hybrid decision support model forcriterion or argument analysis and fusion.

Fig. 6 gives an architecture framework and scenario for aknowledge-intensive decision support agent based on the proposedcDSP and FSD hybrid decision model. It shows that how theknowledge-intensive decision support agent can be applied fordesign decision, which instantiates and complies with the knowl-edge-intensive design decision support scheme shown in Fig. 1.More specifically, the cDSP model is used to develop conceptualdesign alternatives or variants and determine similarity andcommonality between modules and variants; while the FSD modelis used to evaluate and select a design alternative that satisfiescustomer needs, meets design requirements and complies with thetechnical capabilities of the company. The final decision can bereached based on the knowledge resources retrieved through fromthe decision knowledge repository, including differentiating fea-tures and their membership functions, fuzzy rules, fuzzy rankings,etc. A web-based design decision support agent system prototype forproof-of-concept implementation has been developed to verify the

Fig. 6. Knowledge-intensive

proposed hybrid decision model and framework. Two case studiesbelow are provided to illustrate the application of the knowledgesupported decision for concept evaluation and selection in productfamily design for mass customization.

6. Application and case studies

Mass customization has been identified as a competitivestrategy by an increasing number of companies. Family basedproduct design has been recognized as an efficient and effectivemeans to realize sufficient product variety to satisfy a range ofcustomer demands in support for mass customization. Variousapproaches and strategies for designing families of products andmass customized goods are reported in the literature. A productplatform concept exploration method (PPCEM) was proposed todesign a family of products [60]. The five steps of PPCEM are: (1)create market segmentation grid; (2) classify factors and ranges;(3) build and validate metamodels; (4) aggregate product platformspecifications; and (5) develop product platform and family, inwhich formulating and exercising appropriate cDSP models are keyprocedures. To be in line with the general conceptual design stage,a family of product concepts (variants) can be explored andgenerated using the PPCEM and vary widely by selection andassembly of modules or pre-defined building blocks at differentlevels of abstraction so as to satisfy diverse customer require-ments. A wrong or even a poor selection of either a building blockor module can rarely be compensated for at a later design stage andcan result in costly redesign. Concept evaluation and selection arecrucial for product family design for mass customization [56]. Toillustrate and validate the hybrid decision model for product family

decision support agent.

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922 915

design support, two examples are provided in this section: (1) auniversal motor platform and family design selection, and (2) apower supply family design evaluation and selection.

6.1. Universal motor platform and product family design decision

Universal motors are the most common components in powertools. Black & Decker developed a family of universal motors for itspower tools [61]. They used different motors in each of their 122basic tools with hundreds of variations. The challenging issue wasto redesign the universal motor to fit into each of these 122 basictools with hundreds of variations. Through redesign and standar-dization of the product line, they were able to produce all of theirpower tools using a line of motors that varied only in the stacklength and the amount of copper wrapped within each motor. As aresult, all the motors could be produced on a single productplatform with stack lengths varying from 0.8 in. to 1.75 in. andpower outputs ranging from 60 to 650 W. Furthermore, newdesigns were developed using standardized components such asthe redesigned motor, which allowed products to be introduced,exploited and retired with minimal expense related to productdevelopment.

We demonstrate in this example that the cDSP and FSD hybriddecision model (see Fig. 5c) is used cooperatively in the decisionsupport agent for universal motor platform and family designdecision. Specifically, the decision support agent uses the hybriddecision model for universal motor platform and family designdecision:

� T

he cDSP-based PPCEM is used to design a platform for a familyof 10 universal motors, as given in Appendix B [60]. Fig. 7illustrates the cDSP model for the universal motor platform and

Fig. 7. Benchmark universal motor family spec

shows the benchmark universal motor product family specifica-tions and performance responses.

� T he knowledge-supported decision model, i.e., the FSD model, is

used to evaluate and select an appropriate motor variant fromthe motor family obtained by the cDSP model.

The decision support agent is required to select the best motorfor a customer out of 10 motor variants in the family obtainedabove. If 12 performance features, i.e., Nc, Ns, Awf (mm2), Awa(mm2), I (A), r (cm), t (mm), L (cm), T (Nm), P (W), h (%), and M (kg)are considered as criteria, Motor 10 in the family will be selected.Table 1 gives weights and partial performance ratings for eachcriterion (for motor No.1) alongside evaluation results. If the last 5performance features above, i.e., L (cm), T (Nm), P (W), h (%), and M

(kg) are considered as evaluation criteria, Motor 8 will be selected.Figs. 8 and 9 show a comparison of the obtained results using thecDSP and FSD models: (a) Comparison of the benchmark group andcDSP for mass-efficiency relations [60], and (b) Evaluation with theFSD model for 5 and 12 criteria cases.

6.2. Collaborative design evaluation and selection for power supply

family

Power supplies are necessary components of all electronicproducts. Power supply products are often customized because theirrequirements are diverse. The proposed hybrid decision model isused for decision-making in power supply family designs for masscustomization. The design of power supply products is based on themodular design approach [56,62]. In this example, the fuzzy cDSPmodel is used to generate power supply design alternatives orvariants and determine similarity and commonality betweenmodules and product variants, while the FSD model is used for

ifications and performance responses.

Table 1Weights and partial performance ratings, evaluation results (12 criteria)

Criterion no. Criterion item Criterion weight Partial performance rating

Linguistic term Fuzzy number Weight value Linguistic term Fuzzy number Rating crisp value

1 Nc High (0.7, 0.8, 0.8, 0.9) W1 = 0.80 Very Low (0.0, 0.0,0.1,0.2) r11 = 0.500

2 Ns High (0.7, 0.8, 0.8, 0.9) W2 = 0.80 Very Low (0.0, 0.0,0.1,0.2) r12 = 0.800

3 Awf (mm2) Fairly low (0.2, 0.3, 0.4, 0.5) W3 = 0.35 Very Low (0.0, 0.0,0.1,0.2) r13 = 0.075

4 Awa (mm2) Fairly low (0.2, 0.3, 0.4, 0.5) W4 = 0.35 Very Low (0.0, 0.0,0.1,0.2) r14 = 0.500

5 I (A) Fairly low (0.2, 0.3, 0.4, 0.5) W5 = 0.35 Very Low (0.0, 0.0,0.1,0.2) r15 = 0.500

6 r (cm) Medium (0.4, 0.5, 0.5, 0.6) W6 = 0.50 Very High (0.8,0.9,1.0,1.0) r16 = 0.950

7 t (mm) Fairly high (0.5, 0.6, 0.7, 0.8) W7 = 0.65 High (0.7,0.8,0.8,0.9) r17 = 0.800

8 T (Nm) High (0.7, 0.8, 0.8, 0.9) W8 = 0.80 Very Low (0.0, 0.0,0.1,0.2) r18 = 0.075

9 P (W) High (0.7, 0.8, 0.8, 0.9) W9 = 0.00 Very Low (0.0, 0.0,0.1,0.2) r19 = 0.075

10 M (kg) Very high (0.8, 0.9, 1.0, 1.0) W10 = 0.95 Very Low (0.0, 0.0,0.1,0.2) r110 = 0.075

11 L (cm) High (0.7, 0.8, 0.8, 0.9) W11 = 0.80 Very Low (0.0, 0.0,0.1,0.2) r111 = 0.075

12 h (%) Very high (0.8, 0.9, 1.0, 1.0) W12 = 0.95 Very High (0.8,0.9,1.0,1.0) r112 = 0.950

Evaluation results

Family (variants) Evaluation index Rankings

Motor 1 2.108 5

Motor 2 2.920 10

Motor 3 2.478 8

Motor 4 2.594 9

Motor 5 2.175 6

Motor 6 2.319 7

Motor 7 1.825 2

Motor 8 1.928 3

Motor 9 2.049 4

Motor 10 1.655 1

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922916

evaluating and selecting a power supply design alternative. Inaddition, the FSD model is used for conflict resolution andnegotiation in collaborative family evaluation and selection.

First, different clusters can be obtained for families by using thefuzzy cDSP model and the fuzzy clustering model [56]. Fig. 10(a)–(c) illustrates the process of clustering design variations andinstances. From a customer’s point of view, a power supply productis defined based on the following required features (RFs): power,output voltage (OutV), output current (OutC), size, regulation,mean time between failure (MTBF). From an engineer’s point ofview, the power supply product is designed by determining valuesfor these variables (parameters) (DPs): core of transformer (Core),coil of transformer (Coil), switch frequency (SwitchF), rectifier,heat sink type (TypeHS), heat sink size (SizeHS), control loop(Control). Fig. 10(d) shows the relationships between RFs, DPs,

Fig. 8. Comparison of the benchmark group and cDSP for mass-efficiency relations

(based on [60]).

configurations (a hierarchy of building block) and clusters. Threepower supply product families I, II and III are generated based onthree different clusters, which have 4, 5 and 3 base products (BPs),respectively. Each cluster has its own ranges or limitations withregard to the particular product features and/or design parameters.When the product is configured the design requirements andconstraints are satisfied in terms of product functions or functionalfeatures. From an assembly or disassembly or maintenance pointof view, it is advantageous for those parts with low exchange ratesto be placed inside of the product. The locations of some parts,however, are fixed in advance due to design constraints.

The customers’ requirements for Family I power suppliesinclude AC/DC, 45 W, 5 V and �15 V, 150 kh, $20–50, with orwithout auto-start function or feature. The knowledge decisionsupport system (design advisor) first eliminates unacceptablealternatives and determines four acceptable alternatives: NLP40-7610, NFS40-7610, NFS40-7910, and NFS 42-7610. The final designdecision is reached based on the knowledge resources (1)–(4) inFig. 11, including differentiating features (MTBF, price, and specialoffer) and their utility/membership functions, fuzzy rules, fuzzy

Fig. 9. Evaluation with the FSD model for 5 and 12 criteria cases.

Fig. 10. Clustering design variations and instances, power supply configurations.

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922 917

rankings, etc. The final design decision made by the design decisionagent in the system is NFS42-7610 as it has maximum MTBF, mediumprice and special offer of auto-start function furthermore it isacceptable based on the rules. Table 2 gives weights and partialperformance ratings for each criterion (for NLP40-7610) andevaluation results.

Second, the proposed FSD-based negotiation mechanism isused for design conflict resolution in collaborative power supplyfamily selection for customization. Suppose that three designers

Table 2Weights and partial performance ratings

Criterion no. Criterion item Criterion weight

Linguistic term Fuzzy number W

1 MTBF High (0.7,0.8,0.8,0.9) w

2 Price Fairly high (0.5,0.6,0.7,0.8) w

3 Special offer Medium (0.4,0.5,0.5,0.6) w

Evaluation results

Family I Evaluation

NLP40-7610 2.128

NFS40-7610 2.041

NFS40-7910 2.222

NFS42-7610 1.449

and one customer may exchange and revise their suggestions oroption settings by using the FSD model during the collaboration.Detailed conditions for the negotiation in collaborative selectionare composed by MTBF, price, and special offer. In the first phase,three designers suggest their initial options (represented byoptions 1, 2, and 3, respectively) and the customer changes option(option 4) settings with his/her own conditions. The initial optionsshown in Table 3(a) are transformed into fuzzy membership valuesby using the fuzzy membership functions (Table 3(b)). In this

Partial performance rating

eight value Linguistic term Fuzzy number Rating crisp value

1 = 0.80 Medium (0.4,0.5,0.5,0.6) r11 = 0.500

2 = 0.65 High (0.7,0.8,0.8,0.9) r12 = 0.800

3 = 0.50 Very low (0.0, 0.0,0.1,0.2) r13 = 0.075

index (h) Rankings

3

2

4

1

Fig. 11. Knowledge instances used in power supply evaluation and selection.

Table 3Initial option settings of designers and fuzzy membership values for each option setting

(a) Initial option settings s of designers and customer

Designer 1 (Option 1) Designer 2 (Option 2) Designer 3 (Option 3) Customer (Option 4)

MTBF 170 170 190 100–50

Price 32 40 26 15–0

Special offer 0 0 1 0,1

(b) Fuzzy membership values for each selection setting

Designer 1 Designer 2 Designer 3

MTBF 0.44 0.44 0.90

Price 0.85 0.79 0.90

Special offer 0.00 0.00 1.00

Averaged satisfaction 0.43 0.41 0.93

Table 4Pairwise comparison matrix for negotiation

MTBF Price Special Offer

Initial comparison matrix

MTBF 1 3 5

Price 1/3 1 1/2

Special offer 1/5 2 1

Normalized comparison matrix

MTBF 0.65 0.50 0.77

Price 0.22 0.17 0.08

Special offer 0.13 0.33 0.15

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922918

phase, the bell-shaped fuzzy membership functions are adopted totransform the value of each condition. However, the customer ordesigner could set the limits of each factor used in fuzzymembership functions, which are the range of data, center value,and others. The fuzzy AHP-based ‘goal-seeking’ methodology arethen used to compute the relative importance among variables(negotiation conditions: MTBF, price and special offer) and obtainthe pairwise comparison matrix for negotiation. Table 4(a) and (b)show the paired comparisons among variables as negotiationconditions and the normalized comparison matrix. The consis-tency value of AHP is 0.01 (CR = 0.01), which means that thecomparison matrix shown in Table 4 could be used to make ameaningful decision or serve as the basis for next roundnegotiation.

To get a meaningful relative comparison matrix for allconditions, the computation example is shown for all optionalsettings using ‘MTBF’ condition only. First, the third option (option

3, Table 3(b)) is selected as the most preferred option (averagesatisfaction score = 0.93). In contrast with this result, a change ismade and the relative importance of all options is computed withall negotiation conditions, and thus could decide that the secondoption (option 2) as the most preferred option (the total

Table 6Designer and customer’s final design option setting

Final design option

MTBF 196

Price 36

Special offer Auto-start

Fig. 12. Screen snapshot for collaborative power supply evaluation and selection.

Table 5Customer’s new design option setting after the first round negotiation

Before revision (designer) After revision (customer)

Design option Fuzzy value Revised design option Fuzzy value

MTBF 170 0.86 190 0.86

Price 32 0.04 30 0.74

Special offer 0 0.84 1 1.00

Averaged satisfaction – 0.58 – 0.87

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922 919

score = 0.44), whereas the third option (option 3) is now ranked thesecond. Consequently, the customer may select the second option(option 2) as the best option. The negotiation process will becontinued between the customer and designer 2, and the customermay want to improve the satisfaction level from 0.51 to 0.7 � 0.9.To get an optimal solution, the fuzzy cDSP is used, and the goal-seeking algorithm based on the random number generation and aniterative simulation (a hybrid of goal programming and mathe-matical programming) is used. The objective function and subjectconstraints used during this process are summarized as below:

Objective: Satisfaction level = 0.9Subject to:

Fuzzy value of each condition >=current fuzzy value of eachconditionFuzzy value of each condition <=1.0

In addition, the fuzzy value of each condition is related to thefuzzy membership functions and the real value directly. Therefore,the fuzzy values are transformed into real values such as quantity,price, quality, etc. Table 5 shows the results of computations (realvalue transformed) that improve customer’s satisfaction level from0.51 to 0.9. The revision and negotiation processes will continue tofind an optimal solution or consensus among designers and thecustomer. Eventually, each negotiation condition of designers andcustomer will be agreed upon. Table 6 shows the final consensus

among designers and customer and the summarized final designoption (MTBF = 196, price = 36, special offer = auto-start).

Fig. 12 gives a screen snapshot of the proof-of-concept multi-agent design decision system for collaborative power supplyproduct evaluation and selection for customization.

7. Summary, conclusions and future work

This paper presented a hybrid decision model and a multi-agentframework for collaborative design decision support. Decisiontheories and technologies, such as design process and knowledgemodeling, optimization, distributed agents and web-based colla-boration support, are exploited to explore structured support forboth single and distributed design teams. The contribution of thework is the description of a knowledge-intensive decision supportapproach, which rests on an interrelated set of algorithms andmethods: cDSP, fuzzy cDSP, and FSD. Compared with existingmodels and methods, the proposed hybrid decision model can

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922920

provide an effective means of integrating both subjective andobjective elements in the design process, making it particularlysuitable for supporting collaborative design decisions in knowl-edge-intensive intelligent and distributed environments. As such,the design decision knowledge can be managed and real-time oron-line knowledge support can be provided for designers. Typicaltechnical barriers including incomplete and evolving information,uncertain evaluations, inconsistency of team members’ inputs, etc.could be overcome for the decision-making process. Designers,especially novices, can benefit from retrieval of knowledgeevolving from previous design decisions by gaining insight intohow an earlier decision was made or by abstracting informationand applying it to a new design. By making use of the designknowledge, the design decision support system (agent) can helpdesign teams make better confident decisions so that companiesare expected to improve the design process for more innovativeproducts and reducing the product development cycle time. Thedeveloped model and framework are flexible enough to be used ina variety of decision problems. The applications in conceptevaluation and selection in design for mass customization validatethe feasibility of the developed decision support methodology andframework.

Although substantial efforts have been made in this work todevelop a knowledge-intensive support model and framework forcollaborative design decision-making, there are still some practicaland theoretical issues to be addressed in some detailed topics andmodels and applications such as structured models and designknowledge acquisition and repositories, the expert judgment andprocess-useful knowledge for decision-making, and the decision ofwhich models to be used. There are several ways in which themodel and framework could be expanded and refined, includingdecision process models, alternative generation models, and otheragents in the framework. The system itself is a prototype for proof-of-concept; it therefore requires further development. Future workis expected on the development of the decision support agent andintegrating it with a web-based product design and realizationsystem and framework. It may also include applying theapproaches of this paper to the wide product life-cycle decisionsupport.

Acknowledgements

The authors thank Janet Allen for her inputs to the earlierversion of this paper. This work was partly supported by the U.S.Department of Commerce System Integration for ManufacturingAutomation (SIMA) program, National Institute of Standards andTechnology.

Appendix A

A.1. Mathematical form of a fuzzy cDSP [10]

A.1.1. Given

An alternative to be improved, domain dependent assumptions

The system parameters:

n: n

umber of system variables, q: in equality constraints,

p + q: n

umber of system constraints, m: n umber of system goals,

giðXÞ: f

uzzy system constraint functions, fkðdiÞ: f uzzy function of deviation variables to be minimized at

priority level k for the preemptive case.

A.1.2. Find

Fuzzy system design variables, Xi, i = 1,. . .,n,

Fuzzy deviation variables, d�i ; d

þi ; i ¼ 1; . . . ;m,

A.1.3. Satisfy

Fuzzy system constraints (linear, nonlinear)

giðXÞ = 0; i = 1, .., p

giðXÞ 0; i = p + 1, . . ., p + q

Fuzzy system goals (linear, nonlinear)

AiðXÞ þ d�i � d

þi ¼ Gi; i ¼ 1; . . . m

Fuzzy bounds

Xmin

i Xi Xmax

i ; i ¼ 1; . . . ;n

d�i ; d

þi 0; d

�i � d

þi ¼ 0; i ¼ 1; . . . ;m

A.1.4. Minimize

Minimize the deviation function

f ¼ ½ f1ðd�; dþi Þ; :::; f kðd

�; dþi Þ�

Appendix B

B.1. cDSP for designing a group of individual universal motors [60]

B.1.1. Given

Universal motor equations

B.1.2. Find

The system variables:

Nc,j, n

umber of wire turns on the armature rj, r adius of the motor

Ns,j, n

umber of wire turns on each pole on the field tj, th ickness of the stator

Awa,j, c

ross-sectional area of the wire on the armature Ij, c urrent drawn by the motor

Awf,j, c

ross-sectional area of the wire on the field Li,j, s tack length

B.1.3. Satisfy

The system constraints (linear, nonlinear)

Magnetizing intensity: H

j 5000 A turns/m Feasible geometry: t j < rj

Torque: T

j = {0.05, 0.1, 0.125, 0.15, 0.2, 0.25, 0.3,0.35, 0.4, 0.5} Nm

Power: P

j = 300 W Efficiency: h j 0.15

Mass: M

j 2.0 kg

The system goals (linear, nonlinear):

Efficiency: ð

h j=0:70Þ þ d1; j � dþ1; j ¼ 1:0 Mass: ð M j=0:50Þ þ d2; j � dþ2; j ¼ 1:0

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922 921

The bounds on the system variables:

100 Nc,j 1500 turns, 0.5 t

j 10.0 mm

1 Ns,j 500 turns, 0.1 I

j 6.0 A

0.01 Awa,j 1.0 mm2, 1.0 r

j 10.0 cm

0.01 Awf,j 1.0 mm2, 0.0566 L

j 5.18 cm

The bounds on the derivation variables:

d�i; j; dþi; j0; d�i; j � dþi; j ¼ 0 i ¼ 1;2

B.1.4. Minimize:

The deviation function

Zi ¼ ½0:5ðd�1; jÞ þ 0:5ðdþ2; jÞ� All; j ¼ 1; :::;10

References

[1] N.P. Suh, The Principles of Design, Oxford University Press, New York, NY, 1990.[2] C. Tong, R.D. Sriram, Artificial Intelligence in Engineering Design, Academic Press,

1992.[3] W.M. Shen, D.H. Norrie, J.P. Barthes, Multi-Agent Systems for Concurrent Intel-

ligent Design and Manufacturing, Taylor & Francis, 2001.[4] R.D. Sriram, Distributed and Integrated Collaborative Engineering Design, Sarven

Publishers, Glenwood, MD, USA, 2002.[5] D.W. Rosen, Y. Chen, J. Gerhard, J.K. Allen, F. Mistree, Design decision templates

and their implementation for distributed design and solid freeform fabrication, in:Proceedings of ASME DETC00, Paper No. DETC00/DAC-14293, Baltimore, Mary-land, September 10–13, 2000.

[6] J.H. Panchal, M.K. Chamberlain, D.W. Rosen, F. Mistree, A service based archi-tecture for information and asset utilization in distributed product realization, in:9th AIAA/ISSMO Symposium on Multidisciplinary Analysis and Optimization,Atlanta, GA: AIAA-2002-5617, 2002.

[7] A. Xiao, H. Choi, R. Kulkarni, J. Allen, D. Rosen, F. Mistree, S.C. Feng, A Web-baseddistributed product realization environment, in: Proceedings of ASME DETC 2001,Paper No.: DETC2001/CIE-21766, Pittsburgh, Pennsylvania, 2001.

[8] J.F. Gerhard, J.K. Allen, D.W. Rosen, F. Mistree. A distributed product realizationenvironment for design and manufacturing, Proceedings of ASME DETC, PaperNumber: DETC2000/CIE-14624, 2000.

[9] J. Gero, Knowledge-Based Design Systems, Addison-Wesley, 1990.[10] J.K. Allen, R.S. Krishnamachari, J. Masetta, D. Pearce, D. Rigby, F. Mistree, Fuzzy

compromise: an effective way to solve hierarchical design problems, StructuralOptimization 4 (1992) 115–120.

[11] W.D. Li, S.K. Ong, A.Y.C. Nee, L. Ding, C.A. McMahon, Intelligent process planningoptimization for product cost estimation, in: Integrated Intelligent Systems forEngineering Design, vol. 149, Frontiers in Artificial Intelligence and Applications,IOS Press, 2006.

[12] X.F. Zha, R.D. Sriram, W.F. Lu, Knowledge-intensive collaborative decision supportfor design processes, in: Proceedings of ASME DETC 2003, Paper No: DETC2003/DAC-48747, Chicago, 2003.

[13] X.F. Zha, Y.E. Lim, W.F. Lu, A knowledge intensive multi-agent system forcooperative/ collaborative assembly modeling and process planning, transactionof the SDPS, Journal of Integrated Design and Process Science 7 (March (1)) (2003)99–122.

[14] M. Schoop, A. Becks, C. Quix, T. Burwick, C. Engels, M. Jarke. Enhancing decisionand negotiation support in enterprise networks through semantic web technol-ogies, 2002, www.sewasie.org/documents/xsw2002.pdf.

[15] D.K.W Chiu, S.C. Cheung, P.C.K. Hung, H.-F. Leung. Facilitating e-negotiationprocesses with semantic web technologies, Proceedings of the 38th AnnualHawaii International Conference on System Sciences (HICSS’05) Track 1, p.36a, 2005.

[16] S. Talukdar, L. Baerentzen, A. Gove, P. de Souza, Asynchronous teams: cooperationschemes for autonomous agents, Journal of Heuristics 4 (1998) 295–321.

[17] H.A. Simon, The Sciences of the Artificial, The MIT Press, Cambridge, MA, 1996.[18] H.A. Simon, Administrative Behavior, Free Press, New York, 1976.[19] F. Mistree, B. Bras, W.F. Smith, J.K. Allen, Modeling design processes: a conceptual

decision-based perspective, Engineering Design & Automation 1 (4) (1995) 209–321.

[20] M.G. Fernandez, D.W. Rosen, J.K. Allen, F. Mistree. Digital interfaces: the key toeffective decision-making in distributed collaborative design and manufacture,ASME Design Engineering Technical Conferences, Montreal, Canada, 2002, ASME.Paper Number: ASME DETC2002/CIE-34466.

[21] M.G. Fernandez, D.W. Rosen, J.K. Allen, F. Mistree. On a decision support frame-work for distributed collaborative design and manufacture, 9th AIAA/ISSMOSymposium on Multidisciplinary Analysis and Optimization, Atlanta, GA, USA,2002, AIAA. Paper Number: AIAA-2002-5496.

[22] J.X. Jiao, M.M. Tseng, Fuzzy ranking for concept evaluation in configuration designfor mass customization, Concurrent Engineering: Research and Application 6 (3)(1998) 189–206.

[23] C.C. Seepersad, F. Mistree, J.K. Allen. A Quantitative approach for designingmultiple product platforms for an evolving portfolio of products, ASME DesignEngineering Technical Conferences, Advances in Design Automation. Montreal,Canada, 2002, ASME, DETC2002/DAC-34096.

[24] M.G. Fernandez, C.C. Seepersad, D.W. Rosen, J.K. Allen, F. Mistree, Utility-baseddecision support for selection in engineering design, ASME Design EngineeringTechnical Conferences, Pittsburgh, PA, USA, 2001, ASME. Paper Number: ASMEDETC/DAC-21106.

[25] J. von Neumann, O. Morgenstern, Theory of Games and Economic Behavior,Princeton University Press, 1947.

[26] C.L. Hwang, K. Yoon, Multiple Attribute Decision Making: Methods andApplications, Springer, Berlin, 1981.

[27] D.L. Thurston, A formal method for subjective design evaluation with multipleattributes, Research in Engineering Design 3 (2) (1991) 105–122.

[28] F. Mistree, O.F. Hughes, B.A. Bras. The compromise decision support problem andthe adaptive linear programming algorithm, structural optimization: status andpromise, In: M.P. Kamatt (Ed.), AIAA, Washington D.C., Paper 11, 1993, pp. 247–286.

[29] D.L. Thurston, J.V. Carnahan, Fuzzy rating and utility analysis in preliminarydesign evaluation of multiple attributes, ASME Journal on Mechanical Design 114(4) (1992) 648–658.

[30] T.L. Saaty, The Analytic Hierarchy Process, McGraw-Hill, New York, NY, 1991.[31] L.A. Zadeh, Fuzzy sets, Information and Control 8 (1965) 338–353.[32] H.J. Zimmermann, Fuzzy Sets, Decision Making, and Expert Systems, Kluwer

Academic Publishers, Boston, 1987.[33] K.L. Wood, E.K. Antonsson, Computations with imprecise parameters in engineer-

ing design: background and theory, ASME Journal of Mechanism, Transmissions,and Automation in Design 11 (1) (1989) 616–625.

[34] R. Knosala, W. Pedrycz, Evaluation of design alternatives in mechanical engineer-ing, Fuzzy Sets and Systems 47 (3) (1992) 269–280.

[35] J.V. Carnahan, D.L. Thurston, T. Liu, Fuzzy rating for multi-attributedesign decision-making, ASME Journal of Mechanical Design 116 (2) (1994)511–521.

[36] J. Wang, A fuzzy outranking method for conceptual design evaluation, Interna-tional Journal of Production Research 35 (4) (1997) 995–1010.

[37] J. Siskos, J. Lochard, J. Lombard. A multicriteria decision-making methodologyunder fuzziness: application to the evaluation of radiological protection nuclearpower plants, TIMS/Studies in Management Sciences, In: H.J. Zimmermann (Ed.),1984, Amsterdam, North-Holland, pp. 261–283.

[38] O. Maimon, E. Fisher, Analysis of robotic technology alternatives, in: Proceedingsof the 1985 Annual Industrial Engineering Conference, 1985, pp. 227–236.

[39] J.K. Gui, Methodology for modeling complete product assemblies, PhD Disserta-tion, Helsinki University of Technology, Finland, 1993.

[40] C.G. Boender, J.G. de Graan, F.A. Lootsma, Multicriteria decision analysis withfuzzy pairwise comparisons, Fuzzy Sets and Systems 29 (1989) 133–143.

[41] E. Frazell, Suggested techniques enable multi-criteria evaluation of materialhandling alternatives, Industrial Engineering 17 (2) (1985).

[42] E.H. Nielsen, J.R. Dixon, M.K. Simmons, GERES: a knowledge based materialselection program for injection molded resins, in: Proceedings of the ASME1986 Computers in Engineering Conference, Chicago, IL, (1986), pp. 255–262.

[43] D. Clausing, Total Quality Development: A Step-by-Step Guide to World ClassConcurrent Engineering, ASME Press, New York, 1994.

[44] B. Prasad, Concurrent Engineering Fundamentals, vol. 1–2, Prentice Hall PTR, NJ,1996.

[45] S. Pugh, Total Design: Integrating Methods for Successful Product Engineering,Addition-Wesley Publishing Co. Inc., 1991.

[46] G. Taguchi, Introduction to Quality Engineering, Asian Productivity Organization,Tokyo, Japan, 1986.

[47] J.M. Tien, Toward a decision informatics paradigm: a real-time, information-based approach to decision making, IEEE Transactions on Systems, Man andCybernetics 33 (1) (2003) 102–113.

[48] R.D. Sriram, S. Szykman, D. Durham, Special Issue on collaborative engineering,Guest Editorial, Journal of Computing and Information Science in Engineering 6(June (2)) (2006) 93–95.

[49] G. Altshuller, The Innovation Algorithm, Technical Innovation Center, Inc.,Worcester, MA, 1999.

[50] J. Hsing, et al., Conflict Resolution Using TRIZ and Design of Experiment (DOE), in:Proceedings of TRIZCON2001, The Altshuller Institute, March, 2001.

[51] J.P. Lambright, C. Ume, A flat composite panel design advisory system usingknowledge based and case based reasoning, Journal of Mechanical Design 118(1996 December) 461–475.

[52] G. Forgionne, Decision technology system to deliver effective concurrentengineering, Concurrent Engineering: Research and Applications 2 (2) (1994)67–76.

[53] G. Hazelrigg. System Engineering: An Approach to Information-basedDesign, Prentice-Hall International Series in Industrial and System Engineering,1996.

[54] A. Xiao, H. Choi, J.K. Allen, D.W. Rosen, F. Mistree. Collaborative decision makingacross digital interfaces, Proceedings of ASME DETC 2002, Paper No.: DETC2002/DAC-34073, Montreal, Canada, 2002.

[55] G.M. Mocko. A knowledge framework for integrating multiple perspectives indecision-centric design, PhD Thesis, Georgia Institute of Technology, 2006.

[56] X.F. Zha, R.D. Sriram, W.F. Lu, Evaluation and selection in product design for masscustomization: a knowledge decision approach, artificial intelligence for engi-neering design, Analysis and Manufacturing (AIEDAM) 18 (1) (2004) 87–109.

X.F. Zha et al. / Computers in Industry 59 (2008) 905–922922

[57] R.D. Sriram, Intelligent Systems for Engineering: A Knowledge-based Approach,Springer-Verlag, London, UK, 1997.

[58] G. Hernandez, C.C. Seepersad, J.K. Allen, F. Mistree, A method for interactivedecision-making in collaborative, distributed engineering design, InternationalJournal of Agile Manufacturing Systems 5 (2) (2002) 47–65.

[59] FIPA 2001, http://www.fipa.org/specs/fipa00029/.[60] T.W. Simpson, J.R. Maier, F. Mistree, Product platform design: method and

application, Research in Engineering Design 13 (1) (2001) 2–22.[61] M.H. Meyer, A.P. Lehnerd, The Power of Product Platforms, The Free Press, New

York, 1997.[62] M.M. Tseng, J.X. Jiao, Product family modeling for mass customization, Computers

in Industry 35 (3–4) (1998) 495–498.[63] J.S. Kim, Negotiation support in electronic commerce using fuzzy membership

functions and AHP, in: Proceedings of the 6th Pacific Rim International Workshopon Multi-Agents (PRIMA), Seoul, Korea, (2003), pp. 93–104.

[64] C. Mascle, Decision support system in a design for assembly and disassemblymethodology, in: Proceedings of the 5th IEEE International Symposium onAssembly and Task Planning, Besancon, France, July 10–11, 2003.

[65] S. Szykman, R.D. Sriram, W. Regli, The role of knowledge in next-generationproduct development system, ASME Journal of Computing and InformationScience in Engineering 1 (1) (2001) 3–11.

[66] K.L. Wood, E.K. Antonsson, J.L. Beck, Representing imprecision in engineering:comparing fuzzy and probability calculus, Research in Engineering Design 1 (3–4)(1989) 187–203.

[67] X.F. Zha, R.D. Sriram, Knowledge-intensive collaborative decision support fordesign process, in: J.N.D. Gupta, G.A. Forgionne, M. Mora (Eds.), IntelligentDecision-making Support Systems, Springer, 2006 (chapter 16).

[68] W.Y. Zhang, S.B. Tor, G.A. Britton, A heuristic state-space approach to the func-tional design of mechanical systems, International Journal of Advanced Manu-facturing Technology 19 (2002) 235–244.

[69] W.G. Sullivan, Models IEs Can Use to include strategic, non-monetary factors inautomation decisions, Industrial Engineering 18 (1986) 42–50.

[70] P. Huang, P. Ghandforoush, Procedures Given for Evaluating, Selecting Robots,Idustrial Engineering 16 (4) (1984) 44–48.

[71] J.R. Dixon, A. Howe, P.R. Cohen, M.K. Simmons, Dominic I: Progress TowardsDomain Independence in Design by Iterative Redesign, in: Proceedings of theASME 1986 Computers in Engineering Conference, Chicago, IL, Vol. 1, (1986), pp.199–212.

Xuan F. Zha is currently working for the Design and

Process Group of the Manufacturing System Integration

Division at the US National Institute of Standards and

Technology. He is co-founder and chief scientist of

Extension System International. He is also an associated

member of Gigascale Systems Research Center, USA and

guest full professor of Shanghai JiaoTong University

China. His current research interests include remote

monitoring and intelligent service systems, ontology-

based product and system lifecycle engineering, man–

machine–environment system engineering, collabora-

tive sustainable product development, knowledge

management, artificial intelligence and soft computing

in design and manufacturing, embedded and mechatronic systems, robotics, and so

forth. Zha has authored or coauthored over 120 papers, book chapters, and reports

in those areas. Zha serves regular reviewers for numerous research grants, book

series, international journals and conferences. Zha also serves as international

conferences’ program committee, session chair, and review coordinator, and

editorial board member and guest editor for several international journals. Zha is a

Senior Member of IEEE and SME.

Ram D. Sriram is currently leading the Design and

Process group in the Manufacturing Systems Integra-

tion Division at the National Institute of Standards and

Technology, where he conducts research on standards

for interoperability of computer-aided design systems.

Prior to that he was on the engineering faculty (1986–

1994) at the Massachusetts Institute of Technology

(MIT) and was instrumental in setting up the Intelligent

Engineering Systems Laboratory. At MIT, Sriram

initiated the MIT-DICE project, which was one of the

pioneering projects in collaborative engineering. Sriram

has co-authored or authored over 200 papers, books,

and reports in computer-aided engineering, including several books. Sriram was a

founding co-editor of the International Journal for AI in Engineering. In 1989, he was

awarded a Presidential Young Investigators Award from the National Science

Foundation, U.S.A. Sriram is a fellow of the American Society of Mechanical

Engineers and a Senior Member of the Institute of Electrical and Electronics

Engineers. Sriram has a B.Tech. from IIT, Madras, India, and an M.S. and a Ph.D. from

Carnegie Mellon University, Pittsburgh, USA.

Marco Gero Fernandez received his PhD in Mechanical

Engineering from the George W. Woodruff School of

Mechanical Engineering at the Georgia Institute of

Technology in 2005. As a graduate student, he was a

National Science Foundation Research Fellow, a

National Science Foundation Integrative Graduate

Education and Research Traineeship Program Fellow,

and a Georgia Tech Presidential Fellow. He received a

BSE from Duke University in 1999, an MS from the

Georgia Institute of Technology in 2005, and an MBA

from the College of Management at the Georgia

Institute of Technology in 2006. He was a member

of the Systems Realization Laboratory from 1999 to

2006 and currently works in Business Transformation Consulting for the Consulting

Services Practice of Capgemini.

Professor Farrokh Mistree’s design experience spans

mechanical, aeronautical, structural, and industrial

engineering. His teaching experience spans courses in

engineering design, naval architecture, solid mechanics,

operations research and computer science. His current

research focus is on learning how to manage design

freedom in multi-scale design (from molecular to

reduced order models) to facilitate the integrated design

of materials, product and design process chains.And

what does Farrokh Mistree want to do in the future? He

wants to have fun—fun in defining the emerging science-

based discipline of design. Fun in providing an oppor-

tunity for highly motivated and talented people to learn how to achieve their

dreams.http://www.srl.gatech.edu/Members/fmistree.