International journal of computer integrated manufacturing , tập 24, số 4, 2011

An integrated decision support system for global manufacturing co-ordination in the automotiveKeywords: global manufacturing context; dependency and co-ordination; integrated decision su

An integrated decision support system for global manufacturing co-ordination in the automotive

Keywords: global manufacturing context; dependency and co-ordination; integrated decision support; multi-criteriadecision making

1 Introduction

Over the last three decades, along with the

phenom-enon of globalisation, manufacturing management has

been experiencing a paradigm shift from the local

through the international to the global level (Meixell

and Gargeya 2005) This paradigm shift has triggered

many industries to innovate the ways they deliver their

products through globally networked production

systems A direct consequence to the automotive

industry is a fundamental change to their

organisa-tional structure Specifically, it caused the recent

emergence of a new structure and configuration of

manufacturing networks (Trappey et al 2007)

Tradi-tionally, manufacturing networks were organised in

tiers (Veloso and Kumar 2002, Mondragon and Lynos

2008) For example, original equipment manufacturers

(OEMs) would design and assemble the cars First tiers

in the manufacturing network would manufacture and

supply components directly to the automaker (e.g the

fuel pump) Second tiers would produce some of the

simpler individual parts that would be included in a

component manufactured by a first tier (e.g the

housing of the fuel pump), and third and fourth tiers

would mostly supply raw materials This relatively

simple configuration required less co-ordinationeffort across the manufacturing network, becausethe majority of the interactions and communicationsonly happened between the two consecutive tiers.However, this simple configuration no longer fits theactual structure of the industry in today’s globalisa-tion environment (Doran et al 2007) The new directsuppliers are becoming large global firms, which areeither specialised in complex systems or integrators

of a series of subsystems Studies within theInternational Motor Vehicle Program and otheroutside analysts suggest that the new configurationinvolves a division (based on roles and responsibil-ities) along the following four lines (Veloso andKumar 2002):

Systems integrator: a company capable ofdesigning and integrating systems, subassembliesand components into modules that are shipped

or placed directly by the suppliers in the makers’ assembly plants

auto- Global standardiser – systems manufacturer: acompany that sets the standard on a global basisfor a system and components These firms are

*Corresponding author Email: shaofeng.liu@plymouth.ac.uk

Vol 24, No 4, April 2011, 285–301

ISSN 0951-192X print/ISSN 1362-3052 online

Ó 2011 Taylor & Francis

DOI: 10.1080/0951192X.2011.554869

capable of design, development and

manufactur-ing of complex systems Systems manufacturers

may supply motor vehicle manufacturers directly

or indirectly through Systems integrators

Component specialist: a company that designs

and manufactures a specific component or

subsystem for a given car or platform These

firms will increasingly work as suppliers to

systems integrators and global standardisers

Raw material supplier: a company that supplies

raw materials to the OEMs or their suppliers

Some of the raw material suppliers are also

moving into component specialists to add value

to their products

With the new configuration of global

manufactur-ing networks, global firms are forced to take a

substantial responsibility in the design and engineering

of the systems, and more importantly in co-ordinating

the networks for their manufacturing, assembly and

services (Nunes et al 2005) Figure 1 illustrates the

increasing complexity of interacting relationships that

can be identified in the new flattened structure ofglobal manufacturing networks Therefore, in the newflat structure, the co-ordination requirements havebeen raised to a higher level (EIMaraghy andMahmoudi 2009) It has been acknowledged that theultimate success of operations in global manufacturingenterprises depends on the companies’ capability of co-ordination, synchronisation and integration ofbusiness activities (Weston and Cui 2008)

Global manufacturing co-ordination has beenproved to be challenging because of the overarchingissues confronting global manufacturing, namely itsdynamics, complexity, uncertainty and high risk(Pontrandolfo and Okogbaa 1999, Rudberg andWest 2008)

The dynamics of global manufacturing exists inmany respects These include the unbundling ofdifferent stages of the production process across theglobe, the growing capacity for firms to outsourceinternationally, greater product differentiation and thegrowth of the phenomenon of ‘global value chain’,whereby different businesses add value by different

Figure 1 Interactions requiring co-ordination for the new flattened structure

processes or activities at each stage of production

(Nagurney and Matsypura 2005, Needle 2005, Slack

et al 2010) Accordingly, the traditional production

model where firms were responsible for all stages of the

production process of a particular product has

changed Many manufacturers now choose to

specia-lise on particular steps in the production process, such

as design, research and development, or sales and

marketing, either within individual geographic

loca-tions or through participation in the global value

chain, or through utilising outsourcing possibilities

(Dreyer et al 2009)

The complexity of global manufacturing can be

understood from two dimensions First, there is a

complex network of inter-relationships between

differ-ent activities Second, these activities take place in a set

of contexts including the strategic (e.g management

and leadership style, business ethics), organisational

(e.g structure, ownership and size) and environmental

(e.g economy, the state, culture difference) contexts

(Needle 2005) There are complex interactions between

the activities and the context where they take place

(Kazmer and Roser 2008) Furthermore, it is also

believed that the relationship between the global

manufacturing activities and the contexts is not static

but dynamic (Liu and Young 2004, Meixell and

Gargeya 2005)

Uncertainty of global manufacturing has been well

acknowledged from the supply chain perspective, i.e

uncertainty from both the demand and the supply side

(Kazmer and Roser 2008) For example, Verdouw

et al (2010) explored how to master demand and

supply uncertainty with combined product and process

configuration Exchange-rate uncertainty and its

im-pact on price setting are discussed in Kazaz et al

(2005) In Acar et al (2010) the relative impact of three

sources of uncertainties (supply, demand and

lead-time) on cost and service performance is studied using

mathematical models Furthermore, factors such as

regional, national and international economic (e.g

inflation and recession) and political instability, as well

as the regulatory environment can raise extra

chal-lenges to the global manufacturing co-ordination

Depending on the modes of entry, there are various

degrees of risk in relation to global manufacturing

Among the six common modes of entry, exporting,

licensing and franchising are considered as relatively

low risk, while wholly owned subsidiary (also known

as FDI), international joint venture and off-shore

outsourcing are considered as high risk (Lowe et al

2009) There are many causes for the high risks, which

are usually summarised as the ‘4Cs’ – capability,

compatibility, commitment and control Capability

risk is the main cause for delays of end product and

service delivery due to the inability of suppliers to

produce on time and to the required quality (Canbolat

et al 2007) Compatibility risks arise in workingtogether and often do not emerge until the implemen-tation phase Such risks can arise as a result ofdifferences in culture, management style, personalityand administrative and accounting procedures (Rud-berg and West 2008) Many alliances fail through alack of staying power because partners are not willing

to continue the commitment in resources and effort.Control risk is normally high for weaker partner(s) in ajoint venture or strategic alliance When one partner isdominant, then the weaker partner(s) may risk havingits (or their) core competencies reduced or eliminated(Nagurney and Matsypura 2005)

In today’s highly competitive, fast-paced globalbusiness environment, there is no room for error inmaking global co-ordination decisions Companies’success (or survival) depends on the manufacturingmanagers’ capability in making consistent, rationaland optimal decisions In order to succeed in such anunforgiving environment, manufacturing managersneed efficient and effective support that can provide

an appropriate level of decision analysis and ment through using a wide range of models, along withdata and information sources available to them.This paper is concerned with integrated decisionsupport for global manufacturing co-ordination acrossmultiple functions and multiple (international) loca-tions A global context modeller is defined to addressthe dynamics, complexity, uncertainty and risks of thebusiness environment Global manufacturing perfor-mance measurements are captured through a multi-criteria scoring modeller The purpose is to integratethe global context modeller and the multi-criteriascoring modeller within an integrated decision supportsystem, which has the ability to align the manufactur-ing management decisions with the firm’s globalbusiness environment and its performance objectives

assess-A case study has been undertaken to evaluate thedecision system in the automotive industry The maincontribution of this paper (to the body of knowledge ingeneral and to global manufacturing co-ordinationsystems specifically) is that it advances the state of theart in model-driven decision support systems, byaddressing the most commonly stated shortcomings

of the traditional methodologies including lack ofmodel integration and lack of model usability/accessibility

The paper is organised as follows: Section 2 reviewswork in relation to decision making and support inglobal manufacturing Then an integrated decisionsupport system is proposed to support the decisionmaking in Section 3 Section 4 discusses the issuesrelated to integration and the system implementation.The evaluation of the decision support system is

discussed in Section 5 before conclusions are drawn in

Section 6

2 Literature review

Decisions in global manufacturing can be classified

into two types of structures: centralised and

decen-tralised (Canbolat et al 2007) Within a decendecen-tralised

decision structure, local decision makers can make

decisions based on their own goals and preferences,

without constraints from their suppliers, consumers or

partners In fact, in this case the co-ordination effect

along the manufacturing network at the global level is

minimal One severe consequence of decentralised

decision making is that it can lead to a loss of control

for the upper-level managers in the OEMs, systems

integrators and global standardisers As a result, the

OEMs will not be able to deliver the products and

services to customers to meet the specified performance

criteria Therefore, many argue that co-ordination

decisions need to be centralised so that decisions across

different functions and locations in the whole

manu-facturing network are well co-ordinated (Acar et al

2010) Research has shown that centrally co-ordinated

decisions are more advantageous Within the

centra-lised decision structure, decision makers at different

organisational levels aim to resolve conflicting interests

and work towards one common goal, i.e to meet the

global manufacturing network overall performance

objective Upper-level managers at OEMs, systems

integrators and global standardisers can interfere with

lower level decisions when needed (usually only in

‘exceptional’ circumstances) (Kouvelis and Gutierrez

1997) There are, however, implementing and control

difficulties associated with central co-ordination which

needs more investigation For example, the decision

dependencies within the whole decision network can

become really complex Therefore, decision

manage-ment such as the decision propagation path and

decision change has to be well addressed This paper

attempts to address the issues concerning the

centra-lised decision structure and explores how this type of

decision can be supported through advanced ICT

technologies and systems

Decision support system (DSS) is a well-established

research and development area, originating from

computer science and organisation management

re-presented by the work undertaken by Simon et al at

the Carnegie Institute of Technology and by Gerrity

et al at MIT, during the late 1950s and early 1960s

(Keen and Morton 1978) A DSS is defined as an

interactive computer-based system that is designed to

support solutions to decision problems (Bhatt and

Zaveri 2002, Shim et al 2002) DSS research and its

applications evolved significantly over time DSS’s

power in handling large amount of information withspeed and accuracy together with its capability ofcomputing for complex analysis has made it an idealaid for decision makers In global manufacturing,diverse DSS have been developed to support varioustypes of decisions, including systems that couldsupport facility location (Canbolat et al 2007), supplynetwork planning and control (Leu et al 2008, Dreyer

et al 2009), multi-site capacity planning and control,demand management, outsourcing decisions (Loeb-becke and Huyskens 2009), simulation and optimisa-tion (Tyagi et al 2004)

A closer look into the literature on DSS for globalmanufacturing reveals that most DSS can be classified

as data based Data-based DSS argue for the tion of ICT as enablers for immediate access toinformation/knowledge and thus reduce responsetime and increase flexibility (Guerra-Zubiaga andYoung 2006, Young et al 2007, Dreyer et al 2009).For example, a DSS utilising distributed artificialintelligence techniques (mobile agents in particular) isdeveloped for the transfer of product design andmanufacturing information throughout the globalmanufacturing network (Nassehi et al 2006, Newman

utilisa-et al.2008) With the support from the intelligent DSS,distributed decision makers can make the rightdecisions on the manufacturing resources and processplans to achieve interoperability between disparatemanufacturing venues

Provision of the right information and knowledge

is important to decision makers However, based DSS has gone one step further in supportingdecision making Along with the access to data andinformation resources at various internal and externalrepositories, model-based DSS can also provide thecapability of decision analysis and evaluation based on

model-a wide rmodel-ange of qumodel-alitmodel-ative model-and qumodel-antitmodel-ative models(Narasihan and Mahapatra 2004, Phillips-Wren et al.2009) Therefore, model-based DSS are advantageousover data-based DSS in terms of informing decisionmakers about the consequences of each decisionalternative There have been vast amount of interestsand development recently in model-based DSS forglobal manufacturing Leu et al (2008) presented aDSS for global supply network configuration based onlinear programming optimisation models In Canbolat

et al (2007), an integrated modelling approachbrought together a decision tree and multi-attributeutility theory for global manufacturing facility locationdecisions A DSS using mathematical programmingmodels for global network optimisation is discussed byTyagi et al (2004)

Despite its wide application, existing model-basedDSS have been heavily criticised Some most com-monly pronounced shortcomings include lack of model

reusability (for single purpose, throwaway efforts),

lack of integration of models to the real world

(isolation from the environment that they represent)

and lack of model utility/accessibility (not available to

non-modelling specialists and therefore with limited

usage and value) (Delen and Pratt 2006) To address

the issues related to model integration and model

utility/accessibility, first, this paper has developed the

concepts of a global context modeller and a

multi-criteria scoring modeller to adequately reflect the

complexity, uncertainty and risks of a real-world

global manufacturing environment Second, the paper

implements the models within an integrated DSS

(IDSS) based on a standard integration platform,

where non-modelling specialists can conveniently

access the models through the platform’s professional,

user-friendly interface The IDSS is designed and

developed to support the decision making in global

manufacturing co-ordination (i.e management of the

dependencies) across multiple business functions

(manufacturing, transportation and distribution) and

multiple geographical locations (different countries,

continents and free trade zones)

3 Key components of the integrated decision support

system

Figure 2 shows the architecture of the integrated

decision support system (IDSS) The architecture

comprises three basic components inherited from

traditional DSS and four new key components defined

in this paper especially for global manufacturing ordination DBMS (database management subsystem),MBMS (model base management subsystem) and UI(user interaction management subsystem) are consid-ered as the three basic components for a traditionalDSS (Hopple 1988) The IDSS takes the concept ofthese three components and instantiates them in thescenario of global manufacturing The main functions

co-of the three basic components remain the same as intraditional DSS, i.e to manage data, models andinteraction with users, which have been well discussed

in the literature (Carlsson and Turban 2002) Thissection focuses on the four key components (proposed

in this paper), i.e a global context modeller (GCM), amulti-criteria scoring modeller, a configurator and aco-ordinator

3.1 Global context modellerThe purpose of defining the global context modeller(GCM) is to provide the decision makers with anappreciation for the complexity, uncertainty and risks

of the global business environment at which ordination decisions are situated The global manu-facturing context can be identified from differentperspectives, for example, from strategic, organisa-tional and environmental perspectives To manage thecharacteristics of a global manufacturing context, theGCM captures the information of the identified factors

co-Figure 2 Architecture of the IDSS

and classifies them in three main categories Figure 3 is

a class diagram of the global manufacturing context

represented with SysML (Weilkiens 2008) For the

environmental context class, five sub-classes have been

further defined: economy, social and cultural

differ-ences, technology, state and politics, and labour

market For the organisational context, four

sub-classes are defined: structure, ownership, size and

goals Two sub-classes for the strategic context are

management and leadership style, and business ethics

Attributes have been specified for all classes to capture

further details of the factors For example, important

attributes of the state and politics class include

membership of a free trade agreement (such as

NATO, EU, or NAFTA), investment incentives

(regarding taxes, energy, etc.), demand (sales market)

and infrastructure The stability of the state and

politics can be considered as either stable, disturbance

likely (e.g occasional violence) or not stable (e.g

regular war zone) The whole point of capturing global

manufacturing context information through the classes

and attributes is to allow the decision makers to use the

right information to gauge the likely level of

un-certainty and risks of the business, to appreciate the

complexity and dynamics of environment and make

informed decisions

To assess specific characteristics of global

manu-facturing, managers need to find all the necessary

information by searching through a series of classes

modelled in the GCM Table 1 gives examples of

information captured in relevant classes and attributesthat can be used to assess the characteristics ofuncertainty and risk (the definitions of the character-istics have been discussed in Section 1) As Table 1shows, to assess uncertainty, information from thefollowing classes can be used: state and politics(stability and infrastructure attributes), economy (ex-change rate attribute), supply networks (both supplyside and demand side) and technology (affecting lead-time) To estimate different aspects of the risks,information from the following classes and attributescan be used For capability assessment, users can usetechnology, size, infrastructure and labour market.Similarly, for compatibility assessment, classes oftechnology, social and cultural differences, manage-ment and leadership style, and business ethics can beused Ownership can be used to assess control factor,and the goals class can help assess commitment aspect.The impact of uncertainty to decision making is thatdecisions will be made on inaccurate information ifuncertainty is not anticipated, for example if thefluctuation of demand is not considered, then manu-facturers may either have insufficient capacity to dealwith extra demand or have excess capacity and wasteresources when the demand is actually lower Owing tolack of information about risks in the ‘4Cs’ (not able tofulfil the capability, compatibility, commitment andcontrol as defined in Section 1), decision makers couldmake wrong decisions For example, when decisionmakers are not informed of manufacturing networked

Figure 3 Class structure of the global manufacturing context

resources and their capabilities, it is impossible for

them to formulate potential alternatives and make

rational choices

Based on the information captured and organised

in the global manufacturing context, GCM can then

provide a qualitative assessment of the factors for each

facility involved in the global manufacturing network,

quantify the attributes through weighting according to

the manufacturing manager’s domain knowledge and

the decision maker’s preferences, calculate the

aggre-gated value of the factors, and estimate the potential

uncertainty and risk level for the partnerships

3.2 Multi-criteria scoring modeller

Decision criteria for global manufacturing depend on

the metrics adopted for the measurement of

manufac-turing network performance The definition of

manu-facturing network performance has been broad

because a company’s mission, strategy and objectives

can vary considerably based on the value of the

products offered to the customers (Meixell and

Gargeya 2005) Although real world manufacturing

networks emphasise a variety of performance measures

in practice, many argue that commonality does exist

and fundamental measures can be identified For

example, the five performance objectives proposed in

Slack et al (2010) are widely accepted They are cost,

quality, speed, dependability and flexibility Earlier,

the Supply Chain Council (2005) identified five

performance metrics as cost, assets, reliability, ibility and responsiveness Under globalisation, someresearchers also recognise access to new technologiesand broadened supply base as benefits (Needle 2005).The sharp economic downturn in recent years has led

flex-to an increased emphasis on cost reduction Thispaper takes the view that no single performancemetric can sufficiently represent the complexity ofglobal manufacturing and, therefore, treats the globalmanufacturing co-ordination as a multi-criteria deci-sion problem Subsequently a multi-criteria scoringmodeller (MCSM) is proposed to address the decisionproblem

For decision makers, a multi-criteria decisionproblem that requires a trade-off among the severalcriteria is difficult to solve (Nagurney and Matsypura2005) In this section, an MCSM is defined to assist inanalysing the global manufacturing co-ordinationproblem and help identify the preferred decisionalternative The MCSM has the following fivefunctions:

Function 1: Develop a list of the criteria to beconsidered For the global manufacturing co-ordination decision problem, five criteria havebeen considered based on the recommendationsfrom Slack et al (2010) and the Supply ChainCouncil (2005): cost, quality, reliability, flexibil-ity and responsiveness (speed)

Function 2: Assign a weight to each criterionthat describes the criterion’s relative importance

In the IDSS, wi represents the weight forcriterion i

Function 3: Assign a rating for each criterionthat shows how well each decision alternativesatisfies the criterion In the IDSS, rij is used torepresent the rating for criterion i and decisionalternative j

Function 4: Calculate the score for each decisionalternative In the IDSS, Sjrepresents the scorefor alternative j The equation used to compute

Sjfor each alternative is Sj¼ w1r1jþ w2r2jþ w3r3jþ w4r4jþ w5r5j Function 5: Order the decision alternativesfrom the highest score to the lowest score toprovide the MCSM’s ranking of the decisionalternatives

To realise Function 2, i.e assign a weight to eachcriterion to indicate the criterion’s relative importanceperceived by decision makers in a specific decisionmaking process, a five-point scale is specified in which

5 means very important and 1 unimportant Byrepeating this question for each of the five criteria,the MCSM can capture the weightings assigned by

Table 1 Relevant information in GCM that can be used to

assess uncertainty and risks

uncertainty

State and politics,technology,infrastructureSupply

uncertainty

Structure/supplynetworks/supply sideDemand

uncertainty

Structure/supplynetworks/demand sideExchange rate

uncertainty

Economy

infrastructure, labourmarket

Compatibility Social and cultural

differences,management andleadership style,business ethics,technology

decision makers and record them in the database for

later calculation of Wi

To realise Function 3, i.e rate each decision

alternative in terms of how well it satisfies each

criterion, a nine-point scale system specified by Saaty

(2005) is employed The scoring process must be

completed for each combination of decision

alter-natives and decision criterion Assuming the number

of decision alternatives is N, and because five decision

criteria must be considered, then a total of 5 6 N

ratings must be provided and captured in the MCSM

When N is big such as over a hundred, without

support from computer systems it is impossible for

human decision makers to comprehend the

appro-priateness of all the decision alternatives against

decision criteria, in which case the benefit of having

the MCSM is considerable The results of Function 2

(weighting the decision criteria) and 3 (rating decision

alternative against each decision criterion) will enable

Function 4 to calculate the overall satisfaction of

decision alternatives based on the aggregated weight

of all decision criteria

It should be noted that quantitative measures have

been used for the cost criterion, in which aggregated

cost has been considered (Newnes et al 2008) The

mathematical model for the aggregated cost

calcula-tion (so far informacalcula-tion about four types of cost

elements is collected and captured in the IDSS) is

where Cp is the production cost incurred for a

particular component, Ci the inventory cost incurred

for a particular storage location or warehouse, Cethe

currency exchange cost incurred for a particular

transaction, and Ct the transportation cost incurred

for a particular movement of products To sum up, the

MCSM utilises a combination of quantitative (for cost

criterion) and qualitative (for other criteria) assessment

to provide analysis of the decision alternatives

3.3 Configurator

The configurator provides the IDSS with the capability

of organising the facilities into a manufacturing

network The key for the configurator to generate a

manufacturing network is to understand the

organisa-tional structures such as the flat structure shown in the

Figure 1 Each facility’s function and characterisation

(as OEM, system integrator, global standardiser,

system manufacturer, component specialist or raw

material supplier) should be identified and the

infor-mation needs to be stored in the system database in

advance, and ready for the configurator to query

3.4 Co-ordinatorThe co-ordinator is designed to manage decisionhierarchies and dependencies among the OEM, sys-tems integrators, global standardisers, systems manu-facturers, component specialists and raw materialsuppliers in a manufacturing network if a flat structure

is configured by the configurator Alternatives of ordination strategy and mechanism are also provided

co-4 Integration and system implementation4.1 Relationships between the four key componentsWhile the four components have their distinguishingroles and functions, the specification of the relation-ships between the components holds the key forintegration Integration was and remains to be one ofthe most often used words, yet poorly defined notions(Ding et al 2009, Liu et al 2010) However, it is widelyaccepted that integration is a property of component(in the form of models, services, tools, methods,systems or subsystems) interrelations Therefore, it isbelieved that the key notion is the relationships and thenature of these relationships In the context of IDSS,integration means sharing of consistent and currentinformation, sharing of model analysis functions(through remote service calls), and sharing a commondecision making process through co-ordinated activ-ities (triggered at the right time for the right decisionmakers in the right order)

This section discusses the modelling of the ships with SysML (systems modelling language).SysML is a visual modelling language and an evolution

relation-of UML (unified modelling language) SysML aims tosupport the audience in systems engineering, particu-larly to allow them to address the integration ofsystems (Neaga and Harding 2005, Weilkiens 2008).The main reasons to choose SysML for modellingIDSS is that, with SysML, the complex relationshipsbetween the four key components, i.e the GCM, theMCSM, the Configurator and the Co-ordinator can bebetter represented, communicated and understood.Furthermore, SysML tools provide the mechanismfor the models to be transformed into programminglanguages such as Java, which could save the systemdevelopers considerable time and effort in codegeneration

Key relationships between the four key nents in the IDSS have been defined and representedusing SysML component models, as shown in Figure 4

In SysML, component diagrams define how nents/subsystems are collected into a high level systemand interfaced (through ports) with the connectionsbetween them As shown in Figure 4, between the fourkey components, communication of the messages are

compo-directed through 12 dedicated pairs of ports The

nature of the relationships is attached as labels on each

connection For example, the key information

pro-vided to other modules by the global context modeller

is global manufacturing context (discussed in Section

3.1) Information provided by the configurator

in-cludes all configuration types for the manufacturing

network The multi-criteria scoring modeller provides

decision evaluation results against decision

perfor-mance metrics Finally, the co-ordinator provides the

information about decision dependency and

propaga-tion path The ports will be mapped to the computer

network within the IDSS By understanding the

relationships between the four key components and

how the information and functions can be efficiently

and effectively communicated through dedicated

inter-acting points (i.e the port-pairs), it ensures that the

right information and functions are available at the

right time in the right place for the right decision

makers

4.2 The integration platformThe IDSS is implemented by adopting a professionalintegration platform, namely the SAP ERP NetWea-ver, which is provided by SAP (one of the world’sleading companies in professional software) SAP ERPNetWeaver supports enterprise management using webservices technology Since the 1970s there have beenmajor technology waves in software solutions: fromthe mainframe computing to client and server archi-tecture, and now to service-oriented networks (Ng and

Ip 2000) The services provided by the SAP ERPNetWeaver platform utilise the portal’s capabilities,making use of the SAP Business Information Ware-house and the Strategic Enterprise Managementfunctions such as balanced scorecard and managementcockpit (Malik 2005) SAP ERP Netweaver is an openplatform The four key components discussed inSection 3 are first developed as independent modulesusing Java programming (facilitated by an automatic

Figure 4 Relationships between the four key components in IDSS

code generation function provided by the SysML

software Enterprise Architect1) The individual

mod-ules then undergo Unit Test When the functions of

each module are texted to be valid, all four modules

are then integrated using the SAP ERP Netweaver

adaptors, which allows the users’ own modules and

tools to be plugged in and played Altogether, the

newly developed modules and the ERP Netweaver

platform form the IDSS Technical details of the

system implementation based on SysML modelling,

Java programming and the integration platform have

been discussed in authors’ previous publication (Liu

et al.2009) Equipped with the IDSS, decision makers

can use the various functions from the four key

components to produce multiple dimensional analysis

Decision makers can also use the functions embedded

within the Netweaver, such as the visualised decision

dashboards (a screenshot is shown in Figure 5), to

inform their decision making process The evaluation

of the IDSS with a case study in the automotive sector

is discussed in the next section

5 Evaluation of the integrated decision support system

through a decision case

This case study is based on the information collected

from Aeolus Automotive Corporation (AAC),

cur-rently ranked the second largest in the automotive

industry in China Its main foreign joint investors

include French Citroen, Japanese Honda and Korean

Kia Back in 1999, AAC was able to produce a total

number of 257,000 vehicles, with production mainly

focusing on heavy-duty, medium-sized and light-duty

trucks The joint investment with Citroen enabledthem to produce Fukang sedans As a benefit of thejoint ventures from Japan and Korea, AAC now alsoproduces Forte (Kia), Bluebird, Nissan GT-R andHonda Teana sedans Today, their products coverdifferent ranges of final products, systems, subsystemsand components for trucks, cars and coaches, whichhave a substantial market share in China and Asia andare accessing South American and African markets As

of 2007, AAC produced an output of over 1.1 millionvehicles The whole manufacturing network of AAC iscomplex and truly global

The key co-ordination issues encountered by AAC/Honda Teana is the management of a mixture of twotypes of dependencies, i.e type 1 dependency –between different activities (production, inventoryand distribution), and type 2 dependency - betweencomparable activities in different geographical loca-tions (across nations, regions, continents and free tradezones) The illustration of the application of the IDSS

to the AAC case uses three key assumptions The firstassumption is to take one product type, i.e the HondaTeana sedan, out of all product ranges as an example

In assumption 2 the discussion limits the customers toAsian and South American markets only The thirdassumption is that three main functions along themanufacturing network are considered – production,inventory and transportation

5.1 Experimental design and data collectionBased on the above three assumptions, asimplified mini manufacturing network for the Honda

Figure 5 A screenshot of the decision dashboard

Teana – the CRV model - sedan (as shown in Figure 6)

is formulated The network includes two markets –

Asian market (India, Pakistan and Thailand) and

South American market (Brazil and Argentina); the

production network is represented by an OEM, one

systems integrator, one global standardiser, two

systems manufacturers, two component specialists,

and one raw material supplier For most of them,

there are choices of several potential suppliers

Specifically, the OEM – AAC HQ (China, Wuhan,

short for CHN-WH); systems integrator – chassis

integrator (CHN-WH or CHN-XF); global

standardi-ser – Honda HQ (Japan-H); systems manufacturers –

lighting system (CHN-GZ or Malaysia-L) and cooling

system (CHN-XF or Singapore-L); component

specia-lists – engines (Japan-H) and gear boxes (CHN-SY or

CHN-XF or CHN-LZ); raw material suppliers – steel

(CHN-DB or CHN-HB or CHN-HN) Data collected

from the company’s manufacturing specialists include

product order history spanning 12 months in 2007 for

the Honda CRV model sedan, production costs (of

raw materials, components, systems and assembly),

transportation costs (domestic and international),

inventory costs and import tariffs for the countries

involved Context data of potential facilities, given by

the manufacturing specialists in AAC on selected

environmental and organisational factors, are

sum-marised in Table 2 These data are captured in the

GCM, populated and stored in the IDSS database in

advance of the experiment

5.2 The decision procedureFigure 7 outlines the decision procedure for thedecision case by linking the listed activities in theboxes The figure also illustrates the order in whichthe four key components participate in the decisionprocedure The direction of arrows shows theinformation flow between the activities The nature

of the information flowing through the components

is labelled on the arrows As can be seen from thefigure, GCM retrieves the context information forfacilities, quantifies the data and generates anaggregated value for all the context factors Theoutputs from the GCM should enable manufacturingmanagers to have an appreciation of the uncertaintyand risks that might exist The uncertainty and risksare calculated based on the value of the attributes ofrelevant classes Based on the calculation, facilitieswith high uncertainty and risks will be distinguishedfrom those with low uncertainty and risks The oneswith high uncertainty and risks will be eliminatedand the ones with low uncertainty and risks willremain as potential candidates, which will be savedinto the IDSS database to be used at a later stagefor network configuration The MCSM then takes allthe configuration alternatives formulated by theconfigurator for multi-criteria analysis The config-urations with the lowest overall performance assessed

by MCSM will be eliminated at this stage uration alternatives with higher overall performance

Config-Figure 6 Simplified mini manufacturing network for the Honda CRV case

assessed by the MCSM are kept as feasible

config-urations and are saved into the database for later use

by the co-ordinator The co-ordinator then defines

dependencies between the facilities in the

configura-tion, quantifies the dependency complexity, and

identifies co-ordination strategies and mechanisms

In the end, the IDSS produces the combined results

of a holistic analysis, which takes into account all

three dimensions: the global manufacturing context

dimension supported by the GCM, the multi-criteria

dimension supported by the MCSM and the

co-ordination complexity dimension supported by the

co-ordinator The outputs of the holistic analysis will

be the optimal configuration and co-ordination

choice

5.3 Experimental resultsBased on the context information shown in the Table

2, IDSS assesses the uncertainty and risks of eachpotential facility in the Honda CRV case Figure 8shows the analysed results of the global context of thefacilities, in a bar chart format The results fromFigure 8 show that all facilities have a relatively highvalue of the aggregated context stability, which meansthat the uncertainty and risks can be considered asrelatively low Therefore, all the facilities can beentered into the configurator as potential candidatesfor consideration in the next stage

Using the known information from Section 5.1, theconfigurator can quickly identify the roles for all

Table 2 Illustrative data collected and captured in GCM about global context of facilities

Facility names

Demandfluctuation

Supplyuncertainty Exchange rate

Technologycompatibility

State/region

Figure 7 Activity and information flow within the decision case

candidate facilities as OEM, systems integrators,

global standardisers, system manufacturers,

compo-nent specialists, or raw material suppliers The

config-urator then searches the IDSS database for the product

structure, in this case the BOM (bill of material) for the

Honda CRV sedan, which has been stored in the

system beforehand Because product distribution will

closely depend on where customers are located, it is

essential that markets are included in the

configura-tion The outputs of the configurator execution will be

all configuration alternatives for the manufacturing

network under consideration In this case, there are

360 in total (calculated based on the combination of

the numbers of potential candidate facilities, i.e

3 6 3 6 4 6 1 6 2 6 1 6 5) Ten out of the 360

configurations have been extracted and shown in Table

3, just to illustrate what the mini-manufacturing

networks look like The large number of configuration

alternatives implies the complexity of the

co-ordina-tion Obviously it is not practical for the company to

explore all of the alternatives (otherwise, the resources

will be stretched very far and wide) How can the

manufacturing managers identify good configurations

that can best achieve the performance objectives? The

MCSM will be able to better support the decision

The above configuration alternatives are then taken

as inputs to the MCSM for multi-criteria decisionanalysis Subsequently, the MCSM assigns a weightingand rating for each criterion and configurationalternative and computes the aggregated value ofeach configuration’s overall performance against themultiple criteria Figure 9 shows the analysis results ofthe 10 configurations Based on the analysis from theMCSM, those configurations with the overall lowestscores will be eliminated at this stage For example,configurations 3 and 7 both have very low overallscores and, therefore, will not be taken to the nextstage for further consideration The other eight withhigher overall scores are considered as feasible config-urations and will enter to the next stage

Next, the co-ordinator defines the dependenciesbetween the facilities for the remaining eight configura-tion alternatives, quantifies the dependencies, andidentifies the appropriate co-ordination strategies andmechanisms for the configurations Table 4 illustratesthe dependency and co-ordination requirements for theeight configurations The complexity and difficulty ofco-ordination is quantified based on the two types of thedependencies shown in columns 2 and 3 A similar scoringsystem as used for MCSM is employed in the co-ordinator The aggregated results of the co-ordinationcomplexity and difficulty for the eight remaining config-urations are shown in the last column in the Table 4.Finally, the IDSS holistically examines the results

of the eight remaining configurations with respect to allthree key dimensions (i.e the environmental context,the overall performance against the multi-criteria andthe co-ordination complexity), then makes finalrecommendations based on the combined results.This is done through data visualisation, by generatingbubble charts and displaying them on the system userinterface, so that decision makers can have a quickglance to get an overall picture of the configurations

In the bubble chart, as shown in Figure 10, the contextdimension is illustrated on the horizontal axis, which

Table 3 Examples of the configuration alternatives for the Honda CRV mini-manufacturing network

Raw material

supplier (3)

Component specialist(1 6 3)

Systems manufacturer(2 6 2)

Globalstandardiser (1)

Systemsintegrator (2) OEM (1) Market (5)

Figure 8 Analysis results of operations environment from

GCM

Table 4 Dependency and co-ordination requirements for the eight configurations.

manufacturing and distribution; notificationneeds to be issued by IDSS when an activity

is completed in upstream of themanufacturing network Manufacturingstrategies are dominantly represented by theJapanese Pull systems and Chinese Pushsystems

All in Asia, and in the north hemisphere; nobig time difference, no seasonal difference;

transportation system relatively reliable

2

manufacturing and distribution; notificationneeds to be issued by IDSS when an activity

is completed in upstream of themanufacturing network Mostmanufacturing activities are positionedaround central China

All in Asia, small time difference;

transportation links between China andPakistan and between China and Malaysia

is convenient and reliable

1.8

manufacturing and distribution; notificationneeds to be issued by IDSS when an activity

is completed in upstream of themanufacturing network Mostmanufacturing activities spread across fourcountries (China, Japan, Malaysia andSingapore) as well as across central andnorthern China

Across Asia and South America (Brazil); timezone difference, seasonal difference; possibledelay with sea shipping

3.6

manufacturing and distribution; notificationneeds to be issued by IDSS when an activity

is completed in upstream of themanufacturing network Manufacturingactivities spread across China (central andnorth) and Japan

Across Asia and South America (Argentina);

big time zone difference, big seasondifference; possible delay with sea shipping

3.8

manufacturing and distribution;

Notification needs to be issued by IDSSwhen an activity is completed in upstream ofthe manufacturing network Manufacturingactivities spread across China (central only),Japan and Malaysia

All in Asia, small time difference;

transportation links between China andIndia and between China and Malaysia isconvenient and reliable

2.4

(continued)Figure 9 Analysis results of the overall performance for the 10 configurations

means that the configurations further away to the right

have lower uncertainty and risks The multi-criteria

dimension is expressed on the vertical axis, which

means the configurations further away on the top have

higher overall performance The co-ordination

com-plexity is represented by the size of the bubbles, i.e the

configurations with smaller sizes need less

co-ordina-tion efforts and therefore should be preferred Based

on this final holistic assessment, it is clear that

configuration 10 has high overall performance, with

relatively low uncertainty and risk from the business

environment, but with considerable complexity of the

dependency Comparatively, configuration 4 has lar context uncertainty and risks, and co-ordinationcomplexity, but with lower overall performance;Configuration 5 has the same level of overall perfor-mance as alternative 10, but with much higherinstability from the environment Therefore, config-uration 10 would be the favourable choice to decisionmakers However, configuration 10 has quite a bigbubble size which indicates that a high level ofcomplexity exists between business activities andgeographical locations along the manufacturing net-work, and therefore a tight co-ordination strategy will

simi-be needed to manage the dependencies

6 Conclusion and future workThis paper proposed an IDSS for global manufacturingco-ordination The integrated decision support system(IDSS) integrates four key components to captureinformation and to provide decision analysis that arecrucial to global manufacturing decision making Acombination of qualitative and quantitative methods hasbeen explored for the decision evaluation and analysis.Decision makers can use the evaluation and analysisresults to improve their judgement on the decisionproblems and reach more rational and consistentdecisions

One of the key contributions from this paper is thedefinition of a global context modeller (GCM) and a

manufacturing and distribution; notificationneeds to be issued by IDSS when an activity

is completed in upstream of themanufacturing network Mostmanufacturing activities spread across fourcountries (China, Japan, Malaysia andSingapore) as well as across central andnorthern China

All in Asia, but the four countries involved areacross both north hemisphere and tropicalzone; time difference is small, but withseasonal difference; transportation systemrelatively reliable

2.2

manufacturing and distribution; notificationneeds to be issued by IDSS when an activity

is completed in upstream of themanufacturing network Manufacturingactivities spread across China (central andsouth), Japan and Singapore

Across Asia and South America (Brazil); timezone difference, season difference, possibledelay with sea shipping

3.8

manufacturing and distribution; notificationneeds to be issued by IDSS when an activity

is completed in upstream of themanufacturing network Mostmanufacturing activities spread across fourcountries (China, Japan, Malaysia andSingapore) as well as across central andnorthern China

Across Asia and South America Argentina);

big time zone difference, season difference;

possible delay with sea shipping

4.0

Figure 10 Combined value of the context, multi-criteria

decision analysis and dependencies

multi-criteria scoring modeller (MCSM) The GCM

helps manufacturing managers to better understand

the key characteristics of global manufacturing The

context information captured in and provided by the

GCM can help decision makers to assess the

un-certainty and risks in a global context The MCSM

addresses the multiple criteria typical for global

manufacturing The five criteria cost, quality,

relia-bility, flexibility and responsiveness (speed) are

con-currently considered in the MCSM The integration

approach demonstrated by the IDSS facilitates current

and consistent information sharing, function sharing

and process synchronisation across activities and

geographical areas throughout the global

manufactur-ing network The information and evaluation

cap-ability provided by the four key modules in the IDSS

(i.e the GCM, the configurator, MCSM and the

co-ordinator) can support global manufacturing

man-agers to make more rationalised and informed

co-ordination decisions The decisions are aligned with

the firm’s business environments (through appreciation

of global context) and its business performance

(through aggregation of multi-criteria)

The IDSS was evaluated with a case study from the

automotive industry Both GCM and MCSM are generic

to the global manufacturing context and multi-criteria

decision making The configuration rules specified within

the configurator and the dependencies specified within

the co-ordinator can be modified and re-populated for

other applications where the manufacturing networks

have similar organisational structure

Further research has been identified as exploring

the IDSS application to other industries, such as

electronics A focus for future work will be on the

study of the flexibility of the four modules, i.e how

much changes (if any) to the modules are needed for a

new application and how much effort will be required

for the changes Future work will also seek to

consolidate the scoring system for uncertainty and

risk in the GCM Quantitative methods will be

explored for other decision criteria apart from the

cost criterion in the MCSM It is also the authors’

intention to explore the multi-criteria decision making

module with more advanced analytic network

pro-cesses instead of the currently used analytic hierarchy

process, in order to accommodate the

interdependen-cies between the decision factors

References

Acar, Y., Kadipasaoglu, S., and Schipperijin, P., 2010 A

decision framework for global supply chain management

modelling: An assessment of the impact of demand,

supply and lead-time uncertainties on performance

International Journal of Production Research, 48 (11),

3245–3268

Bhatt, G.D and Zaveri, J., 2002 The enabling role ofdecision support systems in organisational learning.Decision Support Systems, 32, 297–309

Canbolat, Y.B., Chelst, K., and Garg, N., 2007 Combiningdecision tree and MAUT for selecting a country for aglobal manufacturing facility OMEGA: The Interna-tional Journal of Management Science, 35, 312–325.Carlsson, C and Turban, E., 2002 DSS: Directions for thenext decade Decision Support Systems, 33, 105–110.Delen, D and Pratt, D.B., 2006 An integrated andintelligent DSS for manufacturing systems ExpertSystems with Applications, 30, 325–336

Ding, L., et al., 2009 Annotation of lightweight formats forlong-term product representations International Journal ofComputer Integrated Manufacturing, 22 (11), 1037–1053.Doran, D., et al., 2007 Supply chain modularisation: Casesfrom the French automobile industry InternationalJournal of Production Economics, 106, 3–11

Dreyer, H.C., et al., 2009 Global supply chain control – Aconceptual framework for the global control centre.Production Planning and Control, 20 (2), 147–157.EIMaraghy, H.A and Mahmoudi, N., 2009 Current design

of product modules structure and global supply chainconfigurations International Journal of Computer Inte-grated Manufacturing, 22, 483–493

Guerra-Zubiaga, D.A and Young, R.I.M., 2006 A facturing model to enable knowledge maintenance indecision support systems Journal of ManufacturingSystems, 25, 122–136

manu-Hopple, G.W., 1988 The state of the art in decision supportsystems Wellesley, MA: QED Information Sciences Inc.Kazaz, B., Dada, M., and Moskowitz, H., 2005 Globalproduction planning under exchange-rate uncertainty.Management Science, 51 (7), 1101–1119

Kazmer, D and Roser, C., 2008 Analysis of design forglobal manufacturing guidelines In: DETC2007: Pro-ceedings of the ASME international design engineeringtechnologies conference and computers and information inengineering conference,901–911

Keen, P and Morton, S.M., 1978 Decision support systems:

An organisational perspective New York: Wesley

Addison-Kouvelis, P and Gutierrez, G.J., 1997 The newsvendorproblem in global market: Optimal centralised anddecentralised control policies for a two-market stochasticinventory system Management Science, 43 (5), 571–585.Leu, J.D., et al., 2008 Advantage analysis of the globalsupply network configuration using air-cargo logisticscentre in the free trade zone In: Proceedings of the 38thinternational conference on computers and industrialengineering, Vols 1–3, pp 1289–1301

Liu, S., et al., 2009 Towards the realisation of an integrateddecision support environment for organisational decisionmaking International Journal of Decision Support Sys-tems Technology, 1 (4), 38–58

Liu, S., et al., 2010 Integration of decision support systems toimprove decision support performance Knowledge andInformation Systems – An International Journal, 22, 261–286.Liu, S and Young, R.I.M., 2004 Utilizing information andknowledge models to support global manufacturing co-ordination decisions International Journal of ComputerIntegrated Manufacturing, 17, 479–492

Loebbecke, C and Huyskens, C., 2009 Development of amodel-based net sourcing decision support system using

a five-stage methodology European Journal of tional Research, 195, 653–661

Opera-Lowe, T.J., Wendell, R.E., and Hu, G., 2009 Screening

location strategies to reduce exchange rate risk European

Journal of Operations Research, 136, 573–590

Malik, S., 2005 Enterprise dashboards: Design and best

practices for IT Hoboken, NJ: John Wiley and Sons

Meixell, M.J and Gargeya, V.B., 2005 Global supply chain

design: A literature review and critique Transportation

Research Part E, 41, 531–550

Mondragon, A.E.C and Lynos, A.C., 2008 Investigating the

implications of extending synchronised sequencing in

automotive supply chain: The case of suppliers in the

European automotive International Journal of

Produc-tion Research, 46, 2867–2888

Nagurney, A and Matsypura, D., 2005 Global supply chain

network dynamics with multi-criteria decision making

under risk and uncertainty Transportation Research Part

E – Logistics and Transportation Review, 41, 585–612

Narasihan, R and Mahapatra, S., 2004 Decision models in

global supply chain management Journal of Industrial

Marketing Management, 33, 21–27

Nassehi, A., Allen, R.D., and Newman, S.T., 2006

Applica-tion of mobile agents in interoperable STEP-NC

compliant manufacturing International Journal of

Pro-duction Research, 44 (18–19), 4159–4174

Neaga, E.I and Harding, J.A., 2005 An enterprise modelling

and integration framework based on knowledge

discov-ery and data mining International Journal of Production

Research, 436, 1089–1108

Needle, D., 2005 Business in context: An introduction to

business and its environment 4th ed Andover,

Hamp-shire, UK: Thomson

Newman, S.T., et al., 2008 Strategic advantages of

interoperability for global manufacturing using CNC

technology Robotics and Computer-Integrated

Manufac-turing, 24 (6), 699–708

Newnes, L.B., et al., 2008 Predicting the whole-life cost of a

product at the conceptual design stage Journal of

Engineering Design, 19 (2), 99–112

Ng, J.K.C and Ip, W.H., 2000 The strategic design and

development of ERP and RTMS International Journal of

Computer Integrated Manufacturing, 13, 138–150

Nunes, A., Ferreira, J.J.P., and Mendonca, J.M., 2005

Distributed business process co-ordination: A

function-ally oriented infrastructure International Journal of

Computer Integrated Manufacturing, 18, 418–426

Phillips-Wren, G., et al., 2009 An integrative evaluation

framework for intelligent decision support systems

European Journal of Operational Research, 195, 642–652

Pontrandolfo, P and Okogbaa, O.G., 1999 Global facturing: A review for planning in a global corporation.International Journal of Production Research37, 1–7.Rudberg, M and West, M.B., 2008 Global operationsstrategy: Co-ordinating manufacturing networks Ome-

manu-ga, 36, 91–106

Saaty, T.L., 2005 Theory and applications of the analyticnetwork process: Decision making with benefits, opportu-nities, costs, and risks, 3rd ed Rowans

Slack, N., Chambers, S., and Johnston, R., 2010 Operationsmanagement 6th ed Harlow, UK: FT–Prentice Hall.Shim, J.P., et al., 2002 Past, present and future of decisionsupport technology Decision Support Systems, 33, 111–126

Supply Chain Council, 2005 Performance metrics for supplychain visibility [online] Available from: http://www.pa-norama.com/documents/panorama-supply-chain-perfor-mance-management.pdf [Last accessed 10 February2011]

Trappey, C.V., et al., 2007 Business and logistics hubintegration to facilitate global supply chain linkage.Proceedings of the institute of mechanical Engineers Part

B – Journal of Engineering Manufacture, 221, 1221–1233.Tyagi, R., et al., 2004 GE plastics optimises the two-echelonglobal fulfilment network at its high performancepolymers division Interfaces, 34 (5), 359–366

Veloso, F and Kumar, R., 2002 The automotive supplychain: Global rends and Asian perspectives Manila,Philippines: Asian Development Bank

Verdouw, C.N., et al., 2010 Mastering demand and supplyuncertainty with combined product and process config-uration International Journal of Computer IntegratedManufacturing, 23 (6), 515–528

Weilkiens, T., 2008 Systems engineering with SysML/UML:Modelling, analysis and design Boston: The MK/OMGPress

Weston, R.H and Cui, Z., 2008 Next generation ofmanufacturing systems In: X.T Yan, W.I Ion, and B.Eynard, eds Global design to gain a competitive edge: Anholistic and collaborative design approach based oncomputational tools London: Springer-Verlag, 701–710.Young, R.I.M., Gunendran, A.G., and Cutting-Decelle,A.F., 2007 Manufacturing knowledge sharing in PLM:

A progression towards the use of heavy weight gies International Journal of Production Research, 45,1505–1519

ontolo-Three-dimensional automatic routing for the design of moulded interconnect devices

Yong Zhuo*, Xiaolei Du and Jianqiang Zhu

Department of Mechanical and Electrical Engineering, Xiamen University, Xiamen, China

(Received 19 May 2010; final version received 2 January 2011)One of the fundamental innovations in the field of mechatronics is the direct material integration of mechanical andelectronic functions through using moulded interconnect devices (MID) technology Unlike conventional circuit boards,they are not limited to two dimensions but offer the possibility to arbitrarily lay printed circuit traces on the surfaces ofthe three-dimensional carrier Traditional 2D layout function in electronic computer aided design (ECAD), especiallyautomatic routing algorithm cannot be directly applied in MID In this article, two 3D routing algorithms with theirown advantages and disadvantages are presented The related 3D routing functions, which are not supported byconventional mechanic CAD (MCAD) and ECAD systems, have been integrated in the design system MIDCAD Withthese 3D routing functions, MIDCAD enables a more effective product design based on the MID technology

Keywords: moulded interconnect devices (MID); automatic routing; algorithm; three dimensional

1 Introduction

The use of high temperature thermoplastics and their

selective metal plating opens a new dimension of

circuit carrier design to the electronic industry:

Three-dimensional moulded interconnect devices (3D-MID)

MIDs are injection moulded thermoplastic parts with

integrated circuit traces Unlike conventional circuit

boards, they are not limited to two dimensions but

offer the possibility to arbitrarily lay printed circuit

traces on the surfaces of the 3D carrier, see Figure 1

They provide enormous technical and economic

potential and offer a remarkably improved ecological

behaviour in comparison to conventional circuit

boards (3-D MID 2004, Feldmann et al 2006)

The common mechatronic design process has been

described in VDI 2206 (VDI 2206 2004) and discussed

by Gausemeier (2005) The difference between MIDs

and conventional mechatronic products is the mutual

dependency between the geometry and the electronics

that makes it impossible to independently design one

aspect without designing the other in the same design

stage (Gausemeier and Feldmann 2006) The most

important MID design task lies in the placement of the

electronic components and the routing of circuits on

and/or within the 3D carrier The realisation of 3D

automatic routing can greatly facilitate the design

process of MID-products The concept of automatic

routing in 3D-MID can be described as follows In a

MID-product, there are sets of pins that have to be

electrically connected The routing problem is to

automatically connect all the pins in each set on the

surfaces within the given 3D space It has to be ensuredthat the interconnection paths of the different sets donot intersect, the rules for the conductor and insulatorwidth are not violated, and the total interconnectionlength is as short as possible

Automatic routing on the PCB has been welldeveloped since 1970s (Ohtsuki 1986, Lienig 2006).Because the applications of many new technologies formechatronic products have complicated the require-ments on automatic routing, during the last twodecades, a great deal of research has been carried out

on global routing and some so-called sional’ routing algorithms are developed (Cheng et al

‘three-dimen-2004, Ababei et al 2005, Wang et al 2006) However,

it is still a new task to perform automatic routing in areal 3D space Traditional 2D routing algorithmcannot be directly applied in a real 3D space T Krebs(2006) developed a 3D routing method in his designsystem, which unfolds the 3D model into a 2D plane,

so the traditional 2D routing methods can be easilyapplied But after unfolding, the relationships ofgeometric connectivity of surface cannot be completelyrepresented; some surfaces of 3D model cannot beunfolded at the same time, i.e in many situations the3D shape cannot be unfolded to a given 2D artwork.Since this method cannot express the 3D connectivityrelationship exactly, it is not a real 3D routing solutionand its application area is very limited

This article is organised as follows Section 2describes the new 3D automatic routing method Theapplications of the related routing functions in

*Corresponding author Email: zhuoyong@xmu.edu.cn

Vol 24, No 4, April 2011, 302–311

ISSN 0951-192X print/ISSN 1362-3052 online

Ó 2011 Taylor & Francis

DOI: 10.1080/0951192X.2011.554871

MIDCAD are described in Section 3 Section 4 draws a

conclusion

2 Three-dimensional automatic routing algorithm

2.1 Comparison between the routing in 3D MID and

on PCB

The routing in 3D MID design is different from the

traditional multilayer routing in circuit design The

difference between them is shown in Figure 2 and

Table 1 It can be found that the routing in 3D-MID is

more complicated than the multilayer routing The

algorithm of the latter cannot be used in the former

directly

2.2 Methodology of the 3D automatic routing

According to whether the routing is based on a grid,

for the traditional 2D routing problem in PCB design,

there are two main types of algorithms:

Maze routing The entire routing area is

repre-sented as a grid, i.e a rectangular matrix of cells,

and the maze router find a path between source

and target by finding a sequence of adjacent cells

between them Lee’s algorithm (Lee 1961) is the

first and the best-known technique in this field,

which serves as the basic of all the other maze

routers, for example Hadlock’s minimum detour

algorithm (Hadlock 1977)

Line probe routing The routing is gridless, which

is proposed through using line segments as the

representation instead of a large grid of points

The line-probe router project horizontal/vertical

line-probes from source and destination until an

intersection between two partial connections is

discovered Hightower’s algorithm (Hightower

1969) is the well-known method in this field

The new 3D routing algorithm is motivated by the

global/detail routing and the complicated 3D geometry

of MIDs The 3D routing algorithm consists of two

main steps, the first step is converting the complicated3D routing problem to a 2D routing problem, and thesecond step is using modified or extended 2D routingmethods to solve the problem Therefore, convertingrouting problems from 3D to 2D and modifying orextending traditional 2D routing algorithm are twokey techniques to solve 3D routing problems

In the following sections, two 3D routing methodsare presented, One method is based on a grid graphand extends Hadlock’s minimum detour algorithm; theother is gridless and combines the A*-algorithm (Clow

1984, Russell and Norvig 2003) and an extension ofHightower’s algorithm The contrast between twomethods will be also presented Both new 3D routingmethods offer two new features:

Table 1 Comparison between the routing in 3D MID and

on PCB

Routing in 3D MID Multilayer routing

Geometry Complicated.

Composed of various types of surfaces and curves

Simple Composed

of only rectangular planes in general.

Location of surfaces

Intersect to each other at any angle.

Parallel to each other Change of

circuit from one surface

to another

Only permitted at the common edge

of two surfaces.

Can change at any location on the surface in most cases.

Figure 1 Current applications of the MID technology

Figure 2 Routing in 3D-MID and multilayer PCB

New 3D routing methods are based on the exact

3D connectivity relationship in the geometry

model, instead of unfolding the 3D shape into a

2D plane Therefore, the 3D electrical

connec-tivity is described explicitly

By maintaining the three-dimensional

character-istics of the original problem, our routing

approaches can effectively utilise the

three-dimensional routing space while satisfying the

design rules about the conductor and insulator

width in 3D space

2.3 The 3D grid-based routing algorithm

According to the methodology of the 3D automatic

routing, the 3D grid-based routing algorithm consists

of two main steps: (1) to deal with 3D geometry data

and to create a 3D grid based on 3D shape; (2) to route

on this 3D grid, which is similar to the traditional 2D

maze routing

The creating of the 3D grid is separated from the

routing process In other words, the complicacy of 3D

geometry data has been pre-processed before the

routing process starts According to the 3D geometry

and rules of the conductor and insulator width, the 3D

grid is created by using standard finite element method

There are still some special points to be considered for

routing problem which are shown as below:

First, since the grid is created respectively on

each surface, special attention has to be paid to

ensure that the grids on different surfaces are in

accordance with each other on the common

edges, and at the same time the rules of

conductor and insulator width are not violated

Moreover, the duplicate grid nodes on edgeshave to be identified

Second, the connectivity relationship in 3Drouting is more complicated than that in 2Drouting One cell in a 3D grid can has much moreadjacent cells than which in 2D routing, asshown in Figure 3 Moreover, if a cell is at theedge of the surface, some of its adjacent cells will

be on another surface So, the 3D relationship ofsurfaces is the key point in that case

According to the characteristics of 3D MID design,Hadlock’s minimum detour algorithm, a well-known2D automatic routing algorithm, is chosen to beextended into the routing based on 3D grid cells.Hadlock’s algorithm is based on a labelling measurenamed Detour Length, which represents the totalnumber of grid cells in an interconnection pathdirected away from the target, as shown in Figure 4.The modified Hadlock’s algorithm for the 3Dproblem is described as below

Given a vertex with coordinates (x, y, z) on a 3Dgrid graph G Assume that any horizontal, vertical ordiagonal edge connecting a pair of adjacent vertices,e.g (x, y, z) and (xþ 1, y, z), (x, y, z) and (x, y þ 1, z),

or (x, y, z) and (xþ 1, y þ 1, z), be of the same unitlength, and hence any path composed of k edges be oflength k Given a pair of vertices v¼ (xv, yv, zv) and

w¼ (xw, yw, zw), let

dminðn; wÞ ¼ max xn½j  xwj; ynj ywj; znj zwj ð1Þ

dmin (v, w) gives the lower bound to the length ofany path between v and w

Figure 3 More complicated connectivity relationship in 3D routing than that in 2D routing (a) In single-layer 2D routing, onecell has at most eight adjacent cells (b) In 3D routing, one cell can have more than eight adjacent cells

Given two distinct vertices v and q, we have one of

the following equations for any adjacent vertex w of v:

dminðw; qÞ ¼ dminðn; qÞ  1 ð2Þ

dminðw; qÞ ¼ dminðn; qÞ ð3Þdminðw; qÞ ¼ dminðn; qÞ þ 1 ð4Þ

If (2), (3) or (4) holds for these v, w and q, then let w

be called an adjacent vertex of Type 1, Type 2 or Type

3, respectively, of v with respect to q

Let p(s, t)¼ [s ¼ v0, v1, v2, , vk¼ t] be a path

which starts at a vertex s¼ v0, passes intermediate

vertices v1, v2, , vk–1in this order, and terminates at

vertex of type 3 of vh71with respect to t}

With the use of these sets, the detour length of

This equation shows that a shortest path between

any two vertices v and w is to be chosen among those

of the least detour lengths During the 3D routing

process, detour numbers respect to a specified target,rather than the distances from the source, they areentered into searched empty grid cells, and those cellswith less detour numbers are expanded with higherpriority

The modified Hadlock’s algorithm (see alsoFigure 5) has some advantages: it can ensure to findthe shortest path if there is one, and it can find theshortest path by searching a much smaller area thanthe traditional Lee algorithm does Its disadvantages:(i) it is very time consuming and requires more memoryspace This problem is especially severe when a finegrid has to be used in one relative large andcomplicated 3D circuit carrier; (ii) it can only dealwith rectangular plane, because all grids are rectangle

Figure 4 Hadlock’s minimum detour algorithm

algorithm

and the grids on different surfaces are in accordance

with each other on the common edges

2.4 The 3D gridless routing algorithm

The new 3D gridless routing method is also motivated

by the principle of 3D autorouting A*-algorithm, a

well-known search algorithm, is chosen to find routing

connection surfaces for each interconnection in order

to convert 3D routing problem to approximate 2D

routing problem Hightower’s algorithm, a basic

line-search routing algorithm, is chosen to be extended into

the routing problem on the surfaces of 3D circuit

carrier

Therefore, this method also has two major stages

First, a least cost path search algorithm using A*

search method is applied to find out a sequence of

surfaces based on the surface graph deriving from 3D

shape This is a stage when nets are routed coarsely by

a sequence of connected surfaces Second, like detailed

routing, modified version of Hightower’s algorithm

could be used in order to get detailed circuit trace on

the surfaces

2.4.1 Application of the A*-algorithm

A*-algorithm has been applied to many state-space

search problems in the field of artificial intelligence

The state-space representation could be used to solve

many problems in manufacturing field (Borenstein and

Becker 2004, Holm 2006) In routing, the space is the

routing surface; the state represents the state of the

routing search as it progresses Let us restrict our

attention to routing two-point nets for the moment

We have a start point on the surface s, to be connected

to a destination point on the surface d, and wish to find

a sequence of surfaces that represent the least cost

path

In A*-algorithm there is an evaluation function f*,

at any node n, f*(n) is an estimate of the cost of aminimal cost path constrained to go through n Let

fð Þ ¼ gn ð Þ þ hn ð Þn ð7Þ

where g*(n) is the minimal cost of a path from s to anode n For g*(n) the cost of the path which has beenfound by the search process in getting to node n isused Because the path from n to the goal node d is notsearched, h*(n) will be relied on information from theproblem domain An obvious choice for h* here, is theEuclidean distancefrom node n to the goal node d.Each node in graph for routing is correspond toone surface and has one cost point, which is used as apoint on one surface associating with a node used inthe computation of the cost It is usually the closestpoint on the current surface to the cost point on theprevious surface Therefore, cost point excludes startand end point is usually on the common edge of twoadjacent surfaces A simple choice for cost point is themiddle point of each edge Therefore, the cost g*(n) is afunction of the distance traversed on the surfacesbetween the start point and the current cost point ofnode n, and the cost h*(n) is the Euclidean distancefrom current cost point to the destination point

A simple example is shown in Figure 6, the startpoint is on the surface 1, and the end point is onthe surface 5 Using A*-algorithm the connectionsurfaces with the least cost path has been found[1 à 3 à 4 à 5] In the search, a surface graph isbuilt dynamically Each surface is represented by anode and there is an edge between two surfaces if theyare neighbours During the search, an estimate cost isgiven to the edges The advantage of this approach isthat the search is quick with a small computationaleffort The drawback is that the costs given to the edges

Figure 6 An simple example of connection surfaces search using A* algorithm

are not accurate There are also some special aspects to

be considered in generating the successors in practice,

for example: keep-out surfaces, edges of a surface

being too short, irregular geometric forms of surfaces

etc In that way, searching will be quicker and the

search result will be more reasonable

2.4.2 Extension of Hightower’s algorithm

Hightower proposed an algorithm by using line

segments instead of a large grid of points to solve

routing problems more efficiently (Hightower 1969)

Given two pins that must be interconnected arbitrarily,

the Hightower-type rectangle probe router (Cage and

Smith 1977), which is shown in Figure 7, radiates

non-grided rectangular horizontal/vertical segments with

variable widths out from each pin towards the other

pin Each new probe is orthogonal to the preceding

one and constructed in a manner that allows the path

being defined to manoeuvre around local obstacles

Continuing the process, probes are generated for the

source pin and the target pin in parallel until an

intersection between the two partial connections is

discovered

According to the characteristics of 3D

MID-product design, this well-known line search algorithm

meets the following basic requirements: high execution

speed; individual circuit widths for different nets or

subnets; surfaces can be of arbitrary shape and can

have complex internal or external boundaries; surfaces

do not need to be plane, as it is often the case with

MID-products

The original Hightower-algorithm only deals with

2D plane routing problems Therefore, the algorithm

has to be extended in order to cope with

three-dimensional routing problems related to

MID-products The extension of Hightower’s algorithm forthe 3D routing problem is described below

2.4.2.1 Data structure The algorithm works on thecontinuous plane and executes fast if appropriate datastructure are used A data structure presented byLauther (1980) has been used to develop a Hightower-type rectangle-probe router (Cage and Smith1977) with the requirement to deal with differentinterconnection widths

The most important technique for an extension ofthe algorithm from 2D routing to 3D routing is thetransformation of the point coordinates (x, y, z) to a2D format, which can be used in the 2D routingalgorithm According to the geometry representation,the surface is parameterised by two variables (u and v).The coordinate of the point (x, y, z) on the surfaces can

be described by u, v and the surface ID, e.g as shown

in Figure 8 for a point on a plane

Figure 7 Hightower-type rectangle probe router Figure 8. Plane and cylinder data structure.

The routing on different surfaces is carried out along

surface-owned U- and V-directions with different

surface ids

In our routing algorithm, the main data structures

are described by classes using an object oriented

technique In 3D routing, all obstacles (parts, wires,

vias, etc.) are distributed over different surfaces;

therefore, in order to have a class for the uniform

representation of all obstacles, a variable for the

surface ID has been added That way, one

four-dimensional binary search tree per surface to store the

all obstacles information can be used

2.4.2.2 Extension to cylinders.The surface of a

3D-MID circuit carrier is not always a plane Therefore,

the routing algorithm must also work on other types of

surfaces In the following, the extension which enables

the use of the algorithm on cylindrical surfaces is

described This type of surface can often be found

when it comes to MIDs

Using the modified data structure, we can easily

extend the algorithm to be applied on cylinders (see

also Figure 8) For a point on a cylinder has the

equation (9), the value of e1, e2, e3, radius and origin

can be got from surface ID Routing is carried out

along U- and V-direction Here V-direction is axle,

search-ray in this direction is one line on the cylinder,

routing is the same as on the plane But U-direction is

the peripheral direction of the cylinder, search-ray in

this direction is actually an arc on the cylinder, the

value of u in above equation is angle of rotation,

therefore in algorithm one parameter is added for this

direction, for cylinder the value of parameter is the

radius, for plane the value is 1.0, then routings on the

plane and cylinder are identical The extended

algo-rithm can be used on the plane and cylinder Based on

the above method, the algorithm can also extend to

other types of surfaces; the 3D routing on the surfaces

of carrier could be realised

2.4.2.3 Internal or external boundaries.The original

Hightower-algorithm only deals with simple 2D planes

with orthogonal geometry In the escape algorithm, the

finding of the escape point is only based on the

information about covers; the factor of boundaries of a

plane is not taken into account, because boundaries are

not treated as obstacles In contrast to conventional

PCBs, the surfaces of MID-products are of arbitrary

shape and have complicated internal or external

boundaries; sometimes, there are no obstacles on the

surfaces, and an escape point cannot be found based

on the escape algorithm

In the extended algorithm presented here, all

internal boundaries and those external boundaries

with concave vertices are treated as special obstacles

The standard obstacle owns one rectangle area, whilethe special obstacle consists of a line and which istreated as a special rectangle with the width null Thereare three situations to be distinguished:

If the line is parallel to the U-direction (K2 inFigure 9), then the width of the special obstacle

in V-direction is null, i.e the values v of theobstacle are equal

If the line is parallel to the V-direction (K1 inFigure 9), then the width of the special obstacle

in U-direction is null, i.e the values u of theobstacle are equal

If the boundary is an arc or a line that is notparallel to the U- and V-direction (K3 inFigure 9), then two or more special obstacleswill be created along the U- and V-direction.The function of the special obstacle is not the same

as that of a general obstacle In the procedurefindcover, the length of the search-ray is not determined

by the special obstacle, but got directly from theintersection of the real boundary, therefore the extrafunction to cut search rays using boundaries is added,when the covers are special obstacles or no cover isfound In the escape algorithm, the special obstaclesare taken into account, and have the same impact asgeneral obstacles when searching for an escape point.Through treating boundaries as special obstacles, theextended version of Hightower’s algorithm now candeal with planes having complicated internal orexternal boundaries

2.4.2.4 Depth-first search Hightower’s algorithm isactually based on a simplified depth-first heuristicexploration that offers an attractive speed/memoryalternative to breadth-first wavefront propagationtechniques used in grid-based algorithms However,

in the original version of Hightower’s algorithm eachtime only one escape point can be generated by theescape algorithm, there is no backtracking function,

Figure 9 The special obstacles in 3D-MID-routing

hence finding a solution cannot be guaranteed if the

layout and boundary of the plane are complicated

In our extended version, the escape algorithm has

been improved so that more than one escape point can

be generated each time An additional data structure to

deal with escape points is developed, where the values

of the u and v parameters are saved The escape

direction and the pointer which point to the parent

escape point are also recorded in the data structure

Based on the depth-first algorithm, a recursive function

and a depth-limit are added to the main procedure

That way, the performance of Hightower’s line search

algorithm could be enhanced The chance of finding a

path is improved even for planes with complicated

boundaries and layout situation

In contrast to previous grid-based routing methods,

the 3D gridless routing method has more advantages:

(i) routing is grid independent with much less searching

efforts; (ii) routing is applicable not only to rectangular

planes but also to cylinders and more types of surfaces

with complicated holes and boundaries; (iii) routing

can simultaneously apply to circuits with different

widths The routing on the surfaces of the circuit

carriers is completed automatically while at the same

time the design rules for the conductor and insulator

width are satisfied The drawback of this 3D gridless

routing algorithm is that it cannot always guarantee to

find the existing shortest path; in very few particularcases, no route can be found even if a solution exists

3 Integration of 3D automatic routing in theMIDCAD system

An integrated system for the design of MID hasbeen developed (Zhuo et al 2006, 2009) This so-called MIDCAD system is developed by using Pro/TOOLKIT and based on the commercial mechanicCAD (MCAD) system Pro/ENGINEER The basicelements of the MIDCAD system are the 3D layoutfunctions for the definition of electronic componentsand the circuits on or within the 3D circuit carrier,which are not supported by conventional MCADund electronic computer aided design (ECAD)systems

Apart from the interactive and manual routingfunctions, the automatic routing of the circuit tracks

on the three-dimensional surface is now developed andintegrated in MIDCAD When designing an MIDcomponent, the following procedures including theapplication of automatic routing have to be performed(Figure 10):

P1: MCAD à MIDCAD; Create the geometry model

of the circuit carrier by using a standard

Figure 10 Application of 3D automatic routing in MIDCAD system (a) 3D circuit carrier (b) Placement of 3D electronicparts (c) 3D-autorouting/3D-circuits (d) Definition of circuit connections

MCAD-application (Figure 10a), to get the

permissive 3D space for the routing of the

interconnections

P2: ECAD à MIDCAD; the logical design, i.e the

information about the connections and electric

parts of the circuit, has to be imported from an

ECAD system; the simple circuit connection can

also be directly defined in the MIDCAD system

Here, the electronic part library and MID-feature

library can be used, which is offered by the

MIDCAD system After the definition of the

circuit connection, the 3D connectivity

relation-ship is expressed by air lines

P3: Placing; the placing function allows the placement

of the electronic components on the surface of the

circuit carrier A special control module is

responsible for collision detection and compliance

of critical distances among components Figure

10c shows the result after placing the components

P5: Routing; the user can define keep-out surfaces or

areas on which no circuit is allowed to pass

Afterwards, the 3D automatic routing can be

performed Due to the complexity of the 3D

MID-design, it can be difficult for the routing function

to reach a completely satisfying result

automati-cally, therefore manual routing, rip-up and

re-route functions can be used to improve the

interconnection paths

P6: Post-processing; after setting the cross section of

circuits, the automatically generated 3D routed

circuit tracks can be displayed (Figure 10d) If the

routing result is not satisfying enough, manual

optimisation is required

Another routing example for a 2-shot moulded

MID-component is shown in Figure 11 The routing

result is realised by a combination of autorouting and

manual optimisation Utilising the geometry model of

the carrier and the routing result of the 3D circuit, the

geometry model of the first and the second shot for themould design are automatically created in the MID-CAD system

4 Conclusion

In this article, the new 3D MID routing is introducedand two routing methods with their own advantagesand disadvantages are presented Rather than unfold-ing the 3D model into a 2D plane, we introduce twonew routing approaches which maintain the three-dimensional characteristics while decomposing thecomplex three-dimensional routing problem into a set

of two dimensional routing problems The related 3Drouting functions, which are not supported by con-ventional MCAD und ECAD systems, are integrated

in the MIDCAD system, so that MIDCAD enables amore effective product design based on the MIDtechnology

In the further development, it is intended todevelop more efficient 3D routing algorithms fordifferent routing situations on one side The twointroduced routing methods are suitable for mostMID routing problems, but routing runs only in XY-direction or UV-direction on the surface, the diagonaldirection using the above algorithms is not supported

In some situations, especially in 2-shot moulded component, the circuit tracks can be laid in randomdirection and on irregularly surfaces, the topologicalrouting method, which is not subject to geometricconstraints will be concerned and developed Becausethe most circuit nets are multi-point nets with morethan two connection pins, this 3D MID routingproblem is more complex and must be solved in thefuture On the other side, it is also very important to

MID-Figure 11 Routing example for a 2-shot moulded MID-component

create an effective communication between 3D MID

routing and later simulation and manufacturing during

MID design process These aspects also make part of

our research preoccupation and will be treated in

further works One potential area is a derivation of

geometric data of the 3D circuits for the laser

structuring process in order to generate the

corre-sponding control programs

Acknowledgements

The research presented in this article is partially supported by

National Natural Science Foundation of China (50975241)

and the Natural Science Foundation of Fujian Province

(2009J01266)

References

3-D MID e.V., 2004 Technologie 3D-MID – Ra¨umliche

elektronische Baugruppen – Herstellungsverfahren,

Gebrauchsanforderungen, Materialkennwerte Mu¨nchen:

Carl Hanser Verlag

Ababei, C., Mogal, H., and Bazargan, K., 2005

Three-dimensional place and route for FPGAs In: Proceedings

of the 2005 Asia and South Pacific Design Automation

Conference, 18–21 January, Shanghai, China, Vol 2,

New York, NY: ACM, 773–778

Borenstein, D and Becker, J.L., 2004 State space

represen-tation of manufacturing operation plans in the presence

of flexible routing capability International Journal of

Computer Integrated Manufacturing, 17 (5), 451–466

Cage, W.G and Smith II, R.J., 1977 A rectangle-probe

router for multilayer P C boards In: Proceedings of the

14th design Automation Conference New Orleans, LA

Piscataway, NJ: IEEE Press, 13–23

Cheng, L.-R., et al., 2004 Congestion estimation for 3D

routing In: Proceedings of the IEEE Computer Society

Annual Symposium on VLSI Emerging Trends in VLSI

Systems Design (ISVLSI’04), 19–20 February 2004

Lafayette, LA Washington, DC: IEEE Computer

Society

Clow, G.W., 1984 A global routing algorithm for general

cells In: Proceedings of the ACM IEEE 21st Design

Automation Conference, 25–27 June 1984 Piscataway,

NJ: IEEE Press, 45–51

Feldmann, K., Pfeffer, M., and Reinhardt, A., 2006 Creative

developments and innovative technologies for the further

success of MID In: Proceedings of the 7th International

Congress on Molded Interconnect Devices, 27–28

Septem-ber 2006 Fuerth, Germany Nu¨rnSeptem-berg: Research

Asso-ciation Molded Interconnect Devices 3-D MID e.V., 1–

15

Gausemeier, J., 2005 From mechatronics to self-optimizingconcepts and structures in mechanical engineering:new approaches to design methodology InternationalJournal of Computer Integrated Manufacturing, 18 (7),550–560

Gausemeier, J and Feldmann, K., 2006 Integrative twicklung ra¨umlicher elektronischer Baugruppen,Mu¨nchen: Carl Hanser Verlag

En-Hadlock, F.O., 1977 A shortest path algorithm for gridgraphs Networks, 7 (4), 323–334

Hightower, D.W., 1969 A solution to line-routing problem

on the continuous plane In: Proceedings of 6th DesignAutomation Workshop, 1–24

Holm, H., 2006 Information structures for design of back resource distribution control systems for discretepart manufacture International Journal of ComputerIntegrated Manufacturing, 19 (1), 24–36

feed-Lauther, U., 1980 A data structure for gridless routing In:Proceedings of the 17th design automation conference.New York, NY: ACM, 603–609

Lee, C.Y., 1961 An algorithm for path connections and itsapplication IRE Transactions on Electronic Computers,EC–10, 346–365

Lienig, J., 2006 Layoutsynthese elektronischer Schaltungen –Grundlegende Algorithmen fu¨r die Entwurfsautomatisier-ung Berlin, Heidelberg: Springer-Verlag

Krebs, T., 2006 Applications of the NEXTR – microline3D-link for the design and manufacturing of MIDproducts In: Proceedings of the 7th International Con-gress on Molded Interconnect Devices, 27–28 September

2006 Fuerth, Germany: Research Association MoldedInterconnect Devices 3-D MID e.V., 237–243

Ohtsuki, T., 1986 Layout design and verification Holland: Elsevier Science Publishers B.V

North-Russell, S.J and Norvig, P., 2003 Artificial intelligence – amodern approach 2nd ed Upper Saddle River, NJ:Pearson Education, 59–100

VDI 2206, 2004 Design methodology for mechatronicsystems Du¨sseldorf: VDI-Verlag

Wang, H.-L., et al., 2006 Three-dimensional multi-piperoute optimization based on genetic algorithms IFIPInternational Federation for Information Processing, 207,177–183

Zhuo, Y., Alvarez, C., and Feldmann, K., 2006 An grated design system for moulded interconnect devices(3D-MID) In: Proceedings of the 3rd CIRP SponsoredConference on Digital Enterprise Technology, Setubal,Portugal

inte-Zhuo, Y., Alvarez, C., and Feldmann, K., 2009 Horizontaland vertical integration of product data for the design ofmoulded interconnect devices International Journal ofComputer Integrated Manufacturing, 22 (11), 1024–1036

Power assessment as a high-level partner selection criterion for new product development projects

Marc Zolghadri*, Aicha Amrani, Salah Zouggar and Philippe Girard

IMS-Bordeaux Labs, 351, Cours de la liberation, Talence, 33405 France(Received 17 July 2010; final version received 7 January 2010)

In new product development (NPD) projects, various partners may be involved at different phases and theirparticipation can lead to either success or failure of the project Therefore, a company that launches an NPD projecthas to carefully select the most appropriate partners Currently, partner selection in supply chains is often performedaccording to well-known criteria such as cost, delay and quality However, problems could emerge in such processesdue to the unavoidable power disequilibrium of the parties involved The strongest party will sooner or later forcethe weaker ones to accept more challenging constraints The use and abuse of power will lead to doubtless mistrustand frustration Therefore, the awareness of the suppliers’ power before any collaboration is of upmost importanceand in realistic situations, it should be used as a high-level selection criterion This article argues that the power ofpartners represents a significant issue for the achievement of a coherent supplier selection strategy An innovativemethod is suggested to assess the power of each potential partner based on its performances The joint use of apower-based selection approach and a performance-based selection approach is illustrated at the end of the article.This study demonstrates how power consideration can help decision makers in selecting more relevant partners.Keywords: power; new product development project; partner selection

1 Introduction

New product development (NPD) constitutes a key

strategy to keep a competitive advantage in the current

economic context, which is characterised by increasing

customer requirements (Huang et al 2003) Possible

improvements for NPD projects were discussed in

various works (Croom 2000) Some of them underline

web applications (Huang and Mak 2001) for

support-ing product design processes or Internet-based

colla-boration (Nidamarthi et al 2001) More recently,

Aldanondo et al (2008) dealt with preliminary design

through constraint satisfaction problem solving The

authors suggest constraint filtering techniques to

provide interactive assistance to designers These

approaches focus on the technological improvement

of NPD projects, while network considerations could

guarantee the success of these projects in another way

To improve quality and cost and to reduce the NPD

project lead time, the focal company (FC) that

launches the product development project seeks

adequate partners to involve in the NPD To perform

such projects, the FC often adopts the co-development

strategy (Emden et al 2006), so the suppliers are

involved early to increase the overall performance of

the NPD project

Both the Harvard Auto Industry project and, later

on, the International Motor Vehicle Program have

mentioned the success of such supplier involvement inthe car industry, mainly in Japan (Bidault et al 1998).Bidault et al (1998) concluded that buyer–supplierrelationships have evolved in Japan from adversarial (inthe early 1960s) to cooperative management with equitylinks, technology transfer and managerial assistance.Thus, as a pioneering sector, the car industry showedthat subcontracting with suppliers should be a realbusiness strategy going far beyond, looking at suppliers

as capacity buffers This led to a new scientific branchcalled early supplier involvement (ESI)

According to Dowlatshahi (1998), ESI concerns

‘the integration of the capabilities that suppliers cancontribute to NPD projects’ van Echtelt et al (2008)perceived ESI as a more sophisticated concept under-lining the responsibility of suppliers: ‘the suppliers areexpected to carry out tasks required by the customer,and they assume even the responsibility for thedevelopment of a part, process or service’

Many authors believe that implementation of ESIhas led to better performance due to the followingfactors: successful innovation (Rothwell 1974, Imai

et al 1986, Clark and Fujimoto 1991, Womack et al.1991), better use of suppliers’ technological compe-tence (Slade 1993), reduced costs and time to market,improved quality and productivity, speed (Imai et al

1986, Clark and Fujimoto 1991, Womack et al 1991,Kamath and Liker 1994, Ragatz et al 2002, Song and

*Corresponding author Email: Marc.zolghadri@ims-bordeaux.fr

Vol 24, No 4, April 2011, 312–327

ISSN 0951-192X print/ISSN 1362-3052 online

Ó 2011 Taylor & Francis

DOI: 10.1080/0951192X.2011.554872

Beneditto 2008), improved design for manufacturing

(Wasti and Lieker 1997) and decreasing risk of

design-related delay (Hartley et al 1997)

These improvements were discussed from the

point of view of the customer The positive effects

have also been described for suppliers (Nishigushi

1994, Heide and John 1990, cited in Labahn and

Krapfel 2000) Reducing inventories along with lower

administrative, sales and overhead costs (Kalwanin

and Narayandas 1995) are other factors that lead

suppliers to ESI

Nevertheless, ESI is not exclusively advantageous

Based on asurvey, Johnsen (2009) demonstrates that

there are serious concerns about the real benefits of ESI

in NPD projects He analyses the research performed

over three decades Many researchers have considered

the positive effect of supplier involvement on product

development performance (Cusumano and Takeishi

1991; Lamming 1993; Kamath and Liker 1994), while

others observe less positive effects of ESI (Eisenhardt

and Tabrizi 1995) Wasti and Lieker (1997) identified a

positive effect of ESI when technological uncertainties

exist, while Swink (1999) showed that the product

newness could be less positive for ESI The long-term

alliances between firms were also considered as

innovation leverage, but Primo and Amundson (2002)

note that alliances could alter innovation possibilities

in the supply chain Table 1 summarises the potential

benefits and risks that a customer or a supplier may

see when participating in an ESI relationship

Subsequent to these reported works suggesting that

ESI has some critical issues (Hartley and Jones 1997,

Wasti and Lieker 1997, Bidault et al 1998, Swink 1999,

Johnsen 2009), the large push of academics towardsESI in the 1980s and 1990s has become moremoderated

In this respect, some authors revealed that thesuppliers must be selected according to more realisticcriteria (Hartley et al 1997, Wasti and Lieker 1997,LaBahn and Krapfel 2000, Petersen et al 2005,Koufteros et al 2007, Song and Beneditto 2008,Johnsen 2009) Perterson et al (2005) go beyond thesupplier’s efficiency and recommend evaluation ofsuppliers by the customer in terms of complementa-rities of capabilities and culture

Somehow, suppliers’ involvement failed in someNPD projects due to dysfunctions during theircollaboration for some unanticipated reasons Mis-understanding, distrust, frustrations or even morecomplicated situations (judiciaries’ issues) emergedamong collaborators due to the use or abuse of power

by stronger parties

It is therefore necessary to improve the supplierselection process taking account of higher levelselection criteria or long-term possibilities for instance.Selecting the most relevant potential partner meansmore than selecting the highest performer The FCcould think of long time collaboration leading finally

to win–win relationships despite the immediate level performances of a potential partner (Liker 2003).This leads to supplier development Sanchez-Rodri-guez et al (1996) and Hartley and Jones (1997) focused

low-on supplier development practices and revealed howsupplier development activities could help FC

to increase its purchasing performance Recently,Abdullah et al (2008) showed that firms need to

Table 1 Benefits and risks of ESI

Successful innovation (Rothwell l974,DeBresson and Amesse 1991,Womack et al 1991, Imai et al

1986, Clark and Fujimoto 1991)

Technologically lesspredictable projects(Eisenhardt and Tabrizil995)

Reducing costs reduce

administrative,

selling and overheads

costs (Kalwanin and

Narayandas 1995)

Customer appropriating ofsupplier’s technology(LaBahn 2000)

Higher performances: reduced costsand time to market and improvedquality and productivity, speed(Womack et al 1991, Imai et al

1986, Clark and Fujimoto 1991,Kamath and Licker 1994, Ragatz

et al 2002, Song and Bernardino2008) Improved design formanufacturing (Wasti and Lieker1997)

No reduction in market (Hartley et al.1997)

time-to-Decreasing risk of design-related delay(Hartley et al 1997)

evolve from traditional to strategic purchasing through

standardisation of components, and to do so, the

firms’ attitudes should change from confrontation to

trust and partnership

In our research, we focus on the power of partners

and more precisely on the power imbalances that

influence the partnership Indeed, the power advantage

can be destructive if the more powerful party is prone

to exploit the weaker one

This article argues that the power of partners has to

be assessed and analysed as clearly as possible far in

advance in order to guarantee a win–win collaboration

or at least to offer companies a clearer view of their

respective power Our research aims to provide a

methodology for this analysis The article is structured

as follows The literature review in Section 2 analyses

two points: partner selection and power in supply

chains This section discusses previous works and

justifies our contribution Section 3 discusses necessary

concepts gathered in a power-based partner selection

approach This section shows how the performance

metrics can be transformed into power inducers These

power inducers are then aggregated by a method,

which borrows some of its components from the AHP

invented by Saaty (2005) The purpose of this method

is to illustrate the feasibility of the power-based partner

selection approach The mathematical issues are not

the main focus of this article Possible improvements of

the aggregation techniques will be discussed in the last

section of the article The proposed approach is then

applied to an illustrative case designed by our research

team After the selection of some performance metrics,

the power-based approach is applied to this case and

the results are compared to those obtained by a pure

performance-based AHP approach By analysing

these results, it is possible to highlight their

comple-mentarities Finally, a concluding discussion and some

perspectives are given at the end of the article

2 Review of partners selection criteria and power

assessment

2.1 Partners selection

The criteria for partner selection have been discussed

in many studies, the selection of partners being a key

success factor for companies in the past years (Lau

et al 2002, Benyoucef and Ding 2003) Finding the

relevant selection criteria and developing an

appro-priate partner selection model is gradually becoming

the most important issue to consider before any

alliance formation (Wu et al 2009) because the

resulting partners can profoundly impact the financial

and operational health of the company This impact is

even deeper if the partners contribute not only to the

realisation of the target product but also to its design

In one of the earliest works in this field, Dickson(1996) found that quality and delivery delay were some

of the most important selection criteria Weber et al.(1991) suggested a classification of selection criteriaand found that price, delivery, quality, productioncapacity and the geographical position were the mostused selection criteria

Some authors saw the necessity of structuringselection criteria, and they began to think of a widecriteria system to guide decision makers in choosingpartners according to their industrial, technical andenvironmental context Geringer (1991) puts forward adistinction between task-related selection criteria(associated with strategic resources and skills) andpartner-related selection criteria (associated with mea-surement of how partners can effectively worktogether) Barbarosoglu and Yazgac (1997) proposed

a hierarchical structure of criteria summarising thesupplier’s characteristics: performance assessment,business structure/manufacturing capability assess-ment and quality system assessment Huang andKeskar (2007) put the selection criteria into threecategories: product related, supplier related and societyrelated The product-related criteria are structured intoreliability, responsiveness and flexibility metrics Costand financial, and assets and infrastructure categoriesare the subclasses of the supplier-related category.Finally, safety and environmental criteria belong to thesociety-related category

Araz and Ozkarahan (2007) pointed out that thetraditional selection criteria (cost, quality and delivery)are not enough for strategic supplier selection, andthey suggested other criteria such as quality manage-ment practices, long-term management practices,financial strength, technology, innovativeness, coop-erative attitude of the supplier, and co-design and cost-reduction capabilities of the supplier However, thepartner selection theory still needs more research tomake it relevant to managers’ needs (Chung et al 2000;Hitt et al 2000), mainly in terms of environmental(power of firms) and social (reputation, position in themarket, etc.) aspects (Wu et al 2009) Recently, Feng

et al (2010) addressed this question by studying thepartner selection process through ‘individual’ or

‘collaboration’ utility The authors argue that theindividual utility considers a single candidate partner(performance criteria), while the collaboration utilitydeals with relationships among involved partners Inother words, the collaboration utility underlines thefact that a partner should have not only good intrinsicperformance (individually) but also good interactionswith others in the supply chain (Derrouiche et al.2007) This is related to the system theory in which asingle component cannot be studied without the entiresystem to which it belongs and the interactions it has

with other components, see von Bertalanffy 1976 This

systemic view is a fundamental axiom adopted in this

article

2.2 Consideration of partners’ power in management

and social sciences

In a company that looks for selecting partners for

realisation and/or development, some modules have to

be aware of the collaborative situation that could rise

from a potential imbalance of power Tenbrunsel et al

(1997) state, ‘one factor that impacts the partnership is

the power of negotiators’

The concept of power became popular in the

engineering fields when Porter (1980) presented his

Five Force model Power represents an important

factor behind the supply chain development and

deployment according to Crook et al (2007); the

power of partners is an unavoidable reality that

influences the collaborative relationships Power

en-ables stronger firms to gain favourable exchange terms

from others, or more broadly, to coerce others to do

what they would not otherwise do (cf Emerson 1962,

Pfeffer and Salanick 1978)

Gaski (1984) reports on the concept of power

defined by Cartwright (1959): ‘When an agent O

performs an act resulting in some change in another

agent P, we say that O influences P If O has the

capability to influence P, we say that O has power over

P’ The research reported here is based on the

definition of power suggested by Martin (1992) as the

‘success of one group in obtaining compliance with its

wishes regardless of the opposition of others’ His

studies of power focus mainly on the so-called

zero-sum model of power between two parties, in which an

increase of power of party A inevitably involves a

reduction of power of party B This reflects the

assumption that in a given situation ‘there is a fixed

amount of power, which is indivisible’

Emerson (1962) defines power as an inherent

property of the relation; it is not an attribute of the

actor, which underlines the systemic view of the power

as mentioned in the last section Emerson (1962) links

two concepts: power and dependency He defines the

power of A over B as a consequence of the dependency

of B to A: ‘The power of A is the amount of resistance

on the part of B which can be potentially overcome

by A The dependence of actor B upon actor A is:

(1) directly proportional to B’s motivational

invest-ment in goals mediated by A, and (2) inversely

proportional to the availability of those goals to B

outside of the A–B relation’

The fundamental axiom claimed by Emerson

(1962) is that an imbalanced relation is unstable, and

he studied processes that tend to reduce this imbalance

These processes are called cost reduction and cing operations Cost reduction refers to all activitiesthat target a minimisation of the ‘cost’ involved for oneparty in meeting the demands of the other This mainlyrefers to the consensus that the weaker party acceptsbecause it looks for attaining the goals Balancingoperations aims at acting on motivation and attain-ability of goals through four possible actions to balancethe power between A and B by: ‘(1) the reduction ofB’s motivational investments in goals mediated by A,(2) cultivating B’s alternative sources for gratification

balan-of those goals, (3) increasing A’s motivational ment in goals mediated by B, and (4) denying A’salternative sources for achieving those goals’

invest-This concept of mediation is according to Frenchand Raven (1959), largely cited in the scientificliterature, who distinguish mediated from non-mediated power (Flynn et al 2008, Zhao et al 2008).The mediated power expresses the power controlled bythe customer on the supplier (which can reward orcoerce a manufacturer) In contrast, non-mediatedpower represents the perception of the customer’spower by the supplier The supplier itself decideswhether and how much it will be influenced by acustomer (perception of expert power, referent powerand legitimate power) Often, the customer may noteven be aware that these powers exist

Much of the literature about power asserts that apower advantage is destructive, because a morepowerful party tends to exploit its advantage (Pruitt

1981, McAlister et al 1986) Lawler (1992) states that

an imbalance of power fosters the use of hostile ratherthan conciliatory tactics LaBahn and Krapfel (2000)note that the power dynamics within buyer–supplierrelationships should not be underestimated Theyaffirm that ‘powerful customers, who abuse theirpower advantage and behave opportunistically, mayruin the trust that is a critical ingredient in supplierinvolvement projects’ It can be concluded that power-ful suppliers in product development projects maythreaten the collaborative relationships with thecompany In choosing partners, it is then advised topay attention to the power of suppliers and, moregenerally, to all partners

Although power has been studied largely in theaforementioned works, a great lack of understandingstill exists in power assessment It is remarkable to seethat many theories exist about the effect of the power

on a bilateral relationship but, as far as we have found,the issue of power assessment has rarely beenaddressed in engineering fields except in Cho andChu (1994) These authors use the basic concepts ofPorter’s Five Force framework (Porter 1980) distin-guishing intrinsic bargaining power and managers’propensity to exert it They postulate that intrinsic

bargaining power comes from structural variables that

constitute the whole industry, called industry-specific

Propensity to exert power is influenced by variables

related to situations that a specific firm faces, called

firm-specific In this model, the bargaining power of

each partner is the product of the intrinsic bargaining

power and the factors influencing the party’s

propen-sity to exert this power This method was applied by

the authors to an industrial case from the shoes

industry

2.3 Positioning of our contribution

From this state of the art, it is outstanding to note that

rarely the question of power is considered in selection

processes in the engineering-related literature, and

power assessment techniques are rarely suggested The

methodology proposed in this article is an attempt to

assess the power of each potential partner to influence

the power imbalance or at least to generate an

awareness of it for the actors A supplier could better

reject an order of a stronger customer, or a customer

can look for a weaker supplier

In short, the basic axioms of our research are as

follows:

2.3.1 Relations instead of individuals: a systemic view

The main selection criteria in the existing methods,

referred here as performance-based selection

ap-proaches, are associated with partners A performance

indicator informs us about a partner, for instance,

notifying us that its delivery delay is 3 weeks

Such a performance indicator is an inherent

attribute of a partner

In this case, the partner is considered in an isolated

manner without any relation to other actors The focus

of the performance-based selection approaches is then

on individuals and their performances These proaches compare potential partners together bycomparing their performance indicators (price, delay,etc.)

ap-The research reported here switches its focus fromindividuals to relations These relations exist betweenthose potential partners and the focal company.Therefore, it transforms performance indicators intopower inducers A power inducer is an inherentproperty of the relation linking a potential partner tothe focal company

The transformation principle consists in ing the focal company will by obtaining the collabora-tion of the potential partner according to its own goalsand context (strategy, market, etc.) To do so, the focalcompany should judge whether the performance metricvalue (the 3 weeks of delivery delay for instance) of thepotential partner is interesting enough or not accord-ing to its goals This is somehow the ‘price’ that thefocal company is ready to pay to benefit from theperformance of that potential partner (the attribute ofthe partnership among two actors) One given perfor-mance value can be judged very interesting by acustomer in some situations while in another case,(s)he may not find it relevant (Figure 1)

determin-2.3.2 Relative power assessmentMeasurement is a process that defines the magnitude of

a quantity (the delay for instance) referring to a unit ofmeasurement However, as there is no way to measurethe power of a company, in the proposed approach inthis article, the target is to compare the assessedrelative power of companies The result is to obtain

an order among them expressed by stronger than orweaker thanor equals to

Figure 1 Performance-based partner selection vs power-based partner selection

2.3.3 Power is multi-dimensional

Power is made of a lot of dimensions Some are linked

to the market situation while others are related to the

company Authors restrict intentionally their research

presented here only to those dimensions connected to

the performance of the companies Other aspects are

not considered here The purpose of the approach is

to allow analysts to use this power assessment as a

diagnosis tool

3 Power-based partner selection approach

To describe the so-called power-based partner

selec-tion approach, the following convenselec-tions are adopted:

FC: Focal company willing to select partners for

an NPD project

Pk: Potential partner k of the FC

(FC, Pk): Business relationship of FC with Pk,

k2 {1, , n}

ci, I2 {1, , m}: Aggregated performance

criter-ion (i.e cost, delays, quality, fill rate, etc.)

cij, j2 {1, , s}: Detailed performance criterion j

of the aggregated criterion i

We suggest studying the partnership (FC, Pk) of

FC with a potential partner Pk related to one

performance criterion ci This will be noted as (FC,

Pk, 5ci4)

3.1 Overall approachCommonly, performance-based partner selection ap-proaches rank potential partners based on theirperformance Techniques (such as AHP) allow analysts

to solve such ranking problems Figure 2 presents themain differences between performance-based andpower-based partners’ selection approaches

3.2 Concept of powerp(FC, Pk,5ci4)

A given business relationship (FC, Pk) with a partner

Pk exists if and only if FC and Pk are dependent,(Emerson 1962) This means that Pk has a resourcethat FC needs This business relationship, generateddue to that resource dependency, can be characterised

by several performance criteria ci(i.e brand image ofthe partner, cost and delay of the delivered items, after-sale services proposed by the partner, the masteredtechnology, etc.) Each criterion cican be evaluated byone or several detailed performance criteria if neces-sary In this case, it can be written that ci¼ {ci1, ,

cij, } For instance, the cost can be the aggregation

of two performance measures: the cost per unit of asupplied item and the cost of after-sale service.Let us study the partnership described only by one

of its performance criterion (FC, Pk, 5ci4) The FC’smanagers are asked to judge their will in obtaining thecollaboration of this partner knowing the value of thatperformance metric This corresponds to a supply and

Figure 2 Power-based and performance-based partner selection approaches

demand law because it defines the will of a demander

to obtain something from a supplier In other words,

a performance value, which is an attribute of the

potential supplier and the Pk, is transformed to an

attribute of the relationship that could be established

between the FC and Pk This is the core idea of the

approach introduced here

The judgment of the managers will be expressed by a

power inducer, which is an image of that performance

metric 5ci4 The power inducer can be notated as

p(FC, Pk, 5ci4) or more easily by pk,ij This idea is

illustrated through a simple example: a customer A

needs a tool that a supplier B can provide (i.e A

depends on B) The customer A will judge this possible

partnership based on several metrics, among others the

cost of the after-sale service asked by B If this price is

highly competitive, A’s desire to obtain this partnership

will be very high A is thus ready to make concessions

in order to obtain the co-operation of B, and this will

generate a power relation between them In this case,

and regarding the cost of the after-sale service,

intuitively one can deduce that A is weaker than B

After-sale service cost could be one of the

components of an aggregated criterion In this case,

an aggregation should be applied to transform pk,ij

to pk,i Obviously, the power of A and B is not only

based on the after-sale service The A–B relationship

should also be considered according to other

perfor-mance metrics This will provide various power

inducers that must be then aggregated into one final

value that represents the relative power of the two

parties, A and B

As a general remark, the transformation of

performance measures into power inducers is necessary

because the performance measures are heterogeneous

(cost, delay, willingness, etc.) and incomparable, i.e

there is no way to directly combine them Thus, they

should be transformed into homogenous metrics The

threshold technique introduced in the next section

accomplishes this transformation

Before discussing the threshold technique in detail,

let us describe the possible values of the power of a

relationship regarding a criterion and each

perfor-mance measure of a potential partner It is proposed

here that the power inducer can have three possible

values describing three different situations:

(1) Partner Pkis stronger than FC corresponds to

two joint conditions:

FC and Pkare dependent

and

The performance of this potential

relation-ship may result in a good chance of success

for the FC

or FC is more interested in this relationship thanthat potential partner is

The FC is ready to negotiate and to accept some of thepartner’s requests in order to obtain this partnership

We note this situation by ‘Pkþ’ For instance, thiscould be the case for a very low price of a suppliedcomponent or its very high quality offered by thesupplier

(2) FC is stronger than Pk corresponds to thesituation where the following conditions aresatisfied:

FC and Pkare dependent

and The FC is aware of the fact that itscollaboration with this potential suppliercould result in a better chance of successfor the partner

or FC is less interested in this relationship thanthat potential partner is

The FC is stronger than the partner, even if the FC’sneed remains We note this situation ‘FCþ’

(3) FC and partner Pk are balanced Two tions describe this situation:

condi- FC and Pkare dependent

and Regarding the considered performance, therelationship presents same level of opportu-nities and/or risks for both the FC and thepartner

Thus, the FC and the partner forces are balanced Thissituation is notated by ‘¼ ’

3.3 The threshold techniqueThe threshold technique transforms heterogeneouscriteria values into homogenous power inducers Itconsists of determining the two thresholds, T1and T2

by the FC managers for every selection criteria(Figure 2) The basic action of the threshold techniquecan be written as follows:

FC; Pk; < ci>

ð Þ !thresholds pk;iFC; Pk; < cij>

!thresholds pk;ij(

These thresholds allow three possible power

situations:

Rule 1:

‘FCþ’ (i.e FC is stronger) is declared for

values belonging to [T2, Limit Value];

(the partner’s performance value is worse

than T2)

Rule 2:

‘Pkþ’ (i.e partner is stronger) for values

belonging to [Limit Value, T1]; (the partner’s

performance value is better than T1)

Rule 3:

‘¼ ’ (i.e balanced power) for values belonging

to the interval [T1, T2]

The determination of thresholds and the necessary

analysis can be done by the FC’s experts For instance,

the return velocity and fill rate thresholds can be

provided by the quality department

By considering the examples in Figure 3, it is seen

that two situations are possible in the determination of

the power inducer In example 1, the FC is stronger

when the measured performance is as big as possible

This is the case of delivery cycle time If the partner has

a very long delivery cycle time, longer than a

determined threshold T2, then the FC can be judged

as stronger than the partner because of the bad

performance of the partner This corresponds to

FCþ In example 2, the FC is stronger when the

measured performance is as small as possible, such as

fill rate If a partner has a bad fill rate, less than

threshold T1, then the FC is stronger In short, the FC

is stronger when the performance of the partner is low

(high delivery lead time or small fill rate)

Figure 3 Threshold technique

3.4 Assessment of powerThe power inducers, pk,ijand pk,i, are then aggregatedstep by step into pk,iand pkto assess the power of the

FC and its potential partner This means the following:

pk;i;8i !agg: pk(

Various aggregation methods can be applied todetermine the power-knowing power inducers Thisaggregation will be done either in literal or innumerical ways The literal analysis technique shouldmanipulate literal values such as FCþ, Pkþ or ¼ ,while in the numerical technique, a numeric value isassociated with each of these three possible balancesituations: ‘0’ corresponds to a balanced situation, ‘þ1’

to FCþ and ‘71’ to Pkþ allowing further calculations

We will use the numerical valuation to calculate thepower of partners hereafter

In Section 3.1, it was considered that the gated criterion cimay be composed of detailed criteria

aggre-ci¼ {ci1, , cij, } Let aij be the weights of detailedcriteria cij, and bi the weights of sets of aggregatedcriteria of ci In this case, the relative power of apotential partner is calculated according to thefollowing equations:

pk;ij;8j !agg: pk;i:pðFC; Pk; < Ci>Þ ¼Xs

j¼1ðaij pk;ijÞ

pk;i;8i !agg: pk:pðFC; Pk; < C >Þ

¼Xm i¼1

bipk;i¼Xm

i¼1

bi Xs j¼1ðaij pk;ijÞ

These are the weighted sums where weights aij

and bi represent the importance of one criterion over

others The determination of these weights must be

assessed based on the application In our first

application presented here, we decided to use Saaty’s

(2005) AHP preference scale and techniques to

calculate these weights (see Table 2) because they

take account users’ preferences and can be obtained by

audits Interested readers should refer to Saaty (2005)

for a detailed description of the technique

As a reminder, hereafter the calculation of aijand bi

is detailed The AHP principle for weight calculation

compares criteria pairwise by using the preference

values defined by Saaty (2005) (see Table 4)

Let Qpbe the matrix of these pairwise preferences

In this case, the possible matrices are Q(pk,ij)¼ [qlr]s*s

and Q(pk,i)¼ [qlr]m*m, where s and m are the number of

considered criteria (detailed and aggregated) The

value of qlr represents the preference of the criterion l

over the criterion r The possible values of qlr are

{1, 2, , 9}, representing Saaty’s (2005) preference

scale (see Table 4) Thus, q32¼ 3 means that criterion

3 has a ‘moderate importance’ relative to criterion 2

The whole preference table is identified, step by step,

by comparing criteria pairwise knowing that qlr¼ qlr71

" #

Xs l¼1

Ys r¼1

qlr

" #1

s

1 s

ð2Þ

bi¼

Ym r¼1qlr

" #1

m

Xm i1

Ym r1qlr

Let:

O¼ {P1, , Pk, , Pn} be a set of potentialpartners,

C¼ {c1, ci, , cm} be a set of aggregatedcriteria,

z¼ {ci1, , cij, , cis} be a set of detailedcriteria

For each Pk2 OFor each ci2 CStep 1: Calculate power inducer pk, and pk,ijof partner

Pkregarding the criteria ciand cijFor every ci,8i

Let Lmin(i), Lmax(i)be the minimum and maximumlimit values of ci, respectively

Let Ti1and Ti2be two thresholds fixed by experts:

pk;ij¼

þ1 if Tij;2 Cij LmaxðijÞ

0 if Tij;1 Cij Tij;2

1 if LminðijÞ Cij Tij;1

Step 3: Calculate the powerUsing the Equation (1) calculate the powerp(FC, P)¼ p 2 [71,1]

Table 2 Preference scale used in the AHP method

Judgement or verbal preference

Numericalassessment

Intermediate values (to be used

if trade-offs are required)

2, 4, 6 and 8

4 Illustration and analysis

In this section, the power-based selection approach

is applied to a case study designed in our laboratory

to illustrate the capability of the proposed

approach

A possible way to extend the use of bicycles is

to transform basic bicycles into electrical

power-aided ones by assembling electrical power-assist kits

This solution (basic bicycleþ power-assist kit) is

cheaper (*400–600e) than electrical power-assist

bicycles sold by manufacturers (*1000–1500e)

Therefore, it is reasonable to assume that a large

customer can launch a call for tenders in order to

improve its existing bicycles into power-assisted

ones

A typical call for tenders for such kits could

contain the following elements:

Technical specification of existing bicycles: male

and female bicycles with seven gears

Cost constraints: the total price must be less than

100e/kit (motor and battery)

Delay: the delay for preparing 100 bicycles

should be less than 1 week

Quantity: the project could concern 4000 bicycles

(typical for a big town)

Battery autonomy: 30 km for a total weight of

100 kg

Battery weight: less than 7 kg

Charge duration: maximum 4 h

Motor module weight: less than 3 kg

Aesthetics: should be included in an

aerody-namic box that is easily assembled to the bicycle

body

A bicycle manufacturer, called Centaur Bicycle,

could answer such a call for tenders Centaur Bicycle

can either choose to design and manufacture all of the

necessary modules or in a more realistic situation, it

could require help from some suppliers These

suppli-ers should design and manufacture the modules asked

by Centaur Bicycle For traditional manufacturers

such as Centaur Bicycle, it is quite reasonable to

hypothesise that they are able to design and

manu-facture mechanical devices necessary to assemble the

motor to the bicycle while sourcing batteries and the

motor-assistance module In this case, Centaur Bicycle

should determine its best supplier(s) for providing the

motor-assistance module in an engineered-to-order

way

Let us say that Centaur Bicycle has prepared a

short list of five potential suppliers of motor-assistance

modules using its business intelligence These partners

are labelled P , P , P, P and P

4.2 Application of the power-based partner selectionapproach

A collection of selection criteria was presented to themembers of our research team in order to choose some

of the most relevant criteria that could be used in such

a situation The final list of criteria and their variationdomain (Boolean, real, a list of possibilities, etc.), theassociated thresholds and the corresponding powerinducers are presented in Table 3 Readers should keep

in mind that only the thresholds are new in thisassessment methodology while all performance metricsvalues and their related domains are known even forany performance-based selection approach The powerinducers are obtained simply once the thresholds aresuggested by users

The elementary criteria were grouped into gated criteria as follows: A¼ ða1; a3Þ; B ¼ ðb2; b3Þ;

aggre-C¼ ðc2; c9Þ; D ¼ ðd2; d4Þ; E ¼ ðe1; e2; e3Þ; F ¼ ð f1; f2Þ;

G¼ ðg1; g3Þ; H ¼ ðh5; h8Þ; I ¼ ði4; i5Þ; J ¼ ð j5; j8; j15Þ:Then, for every aggregated criterion, its detailedcriteria were judged pairwise according to the AHPpreference scale The same approach was applied to theaggregated criteria A, B, , J which were comparedpairwise This allows the calculation of the weights ofaggregated and detailed criteria, biand aij(see Table 4)

4.3 Application of AHP as the performance-basedpartner selection approach

In addition to the power-based selection approach, aperformance-based selection approach was applied tothis case for comparison The chosen performance-based approach was the pure AHP method, as shown

in Figure 4 Alternatives (the potential partners in thevocabulary of this article) are assessed against each ofthe criterion and by using the weights of criteriaobtained by using Equations (1) and (2), and the finalranking of suppliers based on their performances iscalculated

In this application of the AHP approach, we usedthe geometric mean to achieve the ranking (seeBudescu et al 1986, Barzilai et al 1987, Golany andKress 1993) The results are equal to common eigenvalue calculations often used in the literature Some ofthe intermediate results are provided in the appendix ofthe article, while the final results are shown in thebottom rows of the Figure 5

4.4 Analysis of resultsThe powers of each potential supplier P1, , P5 areobtained by the calculation of power inducers related

to each detailed criterion (22 in total) according toEquation (1)

The final results of the power-based selectionapproach are shown in the top rows of Figure 5 Itmight be concluded that the FC would be dominated

by P1(70.55) while it dominates P3 (þ0.13) In anycase, the power balance does not offer a strongposition to FC It can be concluded that the FC will

be in more or less balanced power situations towards

P2, P3, P4and P5 as indicated by power values veryclose to zero, or in the worst situation, it will bedominated by P1

It is interesting to observe that the ranking ofpartners obtained by the performance-based approachshows that P1 is more efficient than its competitors

Table 4 Weights of detailed and aggregated criteria

Figure 5 Profiles of performance values and power values

Figure 4 The performance-based partners selection approach using the AHP technique