International Journal of Computer Integrated Manufacturing Vol 23, No 7, July 2010, 585–602 The barriers to realising sustainable process improvement: A root cause analysis of paradigms for manufacturing systems improvement B.J Hicks* and J Matthews Innovative Design and Manufacturing Research Centre, Department of Mechanical Engineering, University of Bath, UK (Received February 2010; final version received April 2010) To become world-class, manufacturing organisations employ an array of tools and methods to realise process improvement However, many of these fail to meet expectations and/or bring about new less well understood problems Hence, prior to developing further tools and methods it is first necessary to understand the reasons why such initiatives fail This paper seeks to elicit the root causes of failed implementations and consider how these may be overcome The paper begins by reviewing various paradigms for manufacturing systems improvement including design/redesign-, maintenance-, operator-, process-, product- and quality-led initiatives In addition to examining the knowledge requirements of these approaches, the barriers to realising improvement are examined through consideration and review of literature from the fields of manufacturing, management and information systems These fields are selected because of the considerable work that deals with process improvement, change management, information systems implementation and production systems The review reveals the importance of fundamental understanding and highlights the lack of current methods for generating such understanding To address this issue, the concept of machine-material interaction is introduced and a set of requirements for a supportive methodology to generate the fundamental understanding necessary to realise sustainable process improvement is developed Keywords: manufacturing improvement; tools and methods; knowledge requirements; generating understanding Introduction In today’s highly competitive global markets product quality and cost, and manufacturing efficiency and flexibility are critical factors in an organisation’s commercial success (Roth and Miller 1992, ManarroViseras et al 2005, Swink et al 2005) The dimensions associated with production and in particular quality, efficiency and flexibility ultimately define the unit cost of the finished product, and are therefore a central focus of any organisation’s business plan and performance monitoring However, the three factors of quality, efficiency and flexibility are heavily interrelated and attempts to optimise one factor can have a potentially detrimental effect on the other It is therefore important to consider the collective effect of these dimensions on the organisation’s manufacturing capability (cf Figure 1(a)) Within a manufacturing context, quality refers to the perception of the degree to which the product or service meets the customer’s expectations For any manufacturing process to be capable it must be able to produce a quality product As the customer requirements for quality increase the manufacturing capability must also evolve Manufacturing efficiency is *Corresponding author Email: b.j.hicks@bath.ac.uk ISSN 0951-192X print/ISSN 1362-3052 online Ó 2010 Taylor & Francis DOI: 10.1080/0951192X.2010.485754 http://www.informaworld.com effectively a measure of the profit or return realised from the manufacturing system or process (Hansen 2005) At the manufacturing system level this can equate to the time it takes to complete a given task or the number of staff members needed to facilitate the production of a particular item The aim of flexibility in a manufacturing system is to change the mix, volume and timing of its output and essentially describes the ability to process variant products (Matthews et al 2006) When considering the overall manufacturing capability, flexibility has the two dimensions, range and response The range flexibility states what a manufacturing system can adopt in terms of number of different products and output levels – termed product flexibility and volume flexibility; the response flexibility describes the ease with which a system can be adapted from one state to another – termed delivery and mix in Slack (2005) This response flexibility must be considered in terms of time, cost and organisational disruption In general flexibility offers the manufacturer some degrees of freedom to take advantage of demand opportunities and simultaneously provide an ability to reduce losses (Bengtsson 2001) 586 Figure B.J Hicks and J Matthews Manufacturing capability, the organisation and the business environment While attempts to improve particular aspects of, for example, the product design or the manufacturing process can lead to improvements in the areas of either quality, efficiency or flexibility, it is ultimately the sum of all systems, actors and inputs associated with the realisation of the product that determine levels of quality, efficiency and flexibility Hence, manufacturing capability is dependent upon an organisation’s people, its processes, its products and its practices (cf Figure 1(b)) Achieving a high level of manufacturing capability and the attainment of high levels of performance within each of the these areas is frequently associated with the notion of ‘World-Class Manufacturing’ (Maskell 1991) While at a given point in time an organisation may be performing at a high capability level it is the ability to sustain an optimal or near optimal level that is the characteristic of a truly world-class organisation Hence, the notion of worldclass manufacturing and ‘world-class’ organisations is more about the ability of an organisation; its people, processes, products and practices (cf Figure 1(b)), to adapt, improve and evolve within the context of the changing business environment (cf Figure 1(c)) (Riek et al 2006) This ability to respond and adapt is becoming of increasing importance as product complexity increases (Sommer 2003); customer demand for product variety increases (Jiao and Tang 1999); product lifecycles shorten (Christopher and Peck 2003); legislation concerning areas such as materials (European packaging and packaging waste directive 2004/12/EC), emissions (Ambient air quality assessment EC Directive 96/62/EC) and Health and Safety (European Machine Safety 98/37/EC) tighten; supply chains and customers become global (Gelderman and Semeijn 2006) As a consequence of the influence of people, products, processes and practices on an organisation’s manufacturing capability there exists a wide variety of tools, methods and approaches to deliver targeted improvements in a particular area However, in many cases the improvement projects fail to meet expectations and in extreme cases can fail to deliver any improvement or bring about new less well understood problems (Hicks et al 2002) Furthermore, of those International Journal of Computer Integrated Manufacturing that deliver improvements many are short-term (Keating et al 1999) and the improvements are lost when there is, for example, a change of staff, variation in materials or process inputs, altered practices, the introduction of new equipment or yet another initiative From an organisation’s perspective these programmes not only require an investment of many tens or hundreds of thousands of pounds (Chapman et al., 1997, Sterman et al 1997, Keating et al 1999) but in the case of failed initiatives incur an indirect cost which can represent a magnitude of cost and lost opportunity which far exceeds the cost of the original improvement programme For example, optimising set-up and process parameters could make the manufacturing system sensitive to variation in inputs, e.g materials, and result in significant downtime For these reasons and to ensure long-term success, manufacturing organisations need to possess a functional and holistic understanding of the production systems and the variety of tools, methods and approaches for improvement (cf Figure 1(d)) in order that they may be successfully applied and reapplied within the context of the changing business environment Furthermore, as previously stated, it is the ability of an organisation; its people, processes, products and practices to adapt, improve and evolve within the context of the changing business that enables it to be ‘World Class’ A prerequisite for achieving this is the means or capability to generate the fundamental understanding necessary to respond appropriately It is the critical dimension of understanding and the creation of methods for generating the necessary understanding that is addressed in this paper This paper first explores the motivations for manufacturing improvement and examines in detail the principles and underlying knowledge requirements of a range of common improvement paradigms The barriers to realising sustainable improvement are then discussed and the importance of generating and communicating a fundamental understanding is highlighted The need to support organisations in reinforcing and extending their fundamental understanding is further argued and the deficiencies in existing supportive techniques are described In order to overcome these deficiencies the concept of machine-material interaction is introduced and its relationship to ‘function’ and fundamental understanding is discussed The paper concludes with the development of a set of requirements for a new supportive methodology which enables machine–material interactions to be investigated, and the necessary fundamental understanding to be developed and contextualised with respect to the knowledge requirements of a range of common improvement paradigms 587 Improvement paradigms There are a wide variety of approaches and philosophies associated with the improvement of manufacturing and production systems These higher level paradigms generally involve a range of tools and methods to target, plan and implement an improvement programme For the purpose of considering these various philosophies and their corresponding tools and methods (Brassard and Ritter 1994), the approaches and the methods can be grouped under the seven areas of: equipment design/redesign, maintenance, operator-led, process-control, product modification and new product introduction, quality, and tooling design and changeover The various manufacturing paradigms and the corresponding tools and methods that can be associated with each of these seven areas are illustrated in Figure and described in detail in Tables and Of particular interest in this work are the underlying knowledge requirements necessary to successfully apply the various tools and methods These requirements are developed in Tables and from an analysis of the aims and underlying principles of the various tools and methods, which are now summarised (1) Process control As levels of automation increase and in particular, the automation of changeover and machine set-up, so does the need to possess the understanding necessary to explicitly define set-up rules and parameters Intelligent monitoring and control has been successfully applied in Component manufacture (Uraikul et al 2000, Murdock and HayesRoth 1991) and Machining processes (Hou et al 2003, Liang et al 2004) but requires indepth knowledge of the relationship between product variation and process variation - both upstream and downstream Central to the success of these methods is the need to understand and describe the acceptable variation in product attributes during all stages of production (2) Operator-led One of the key elements to the effective operation and improvement of a production system is the successful training of the operating staff (Woodcock 1972) Training is imperative to ensure changes to working practices and operating procedures are effectively taken-up For effective training to be delivered the trainer needs to possess an indepth understanding of the content (Davis and Davis 1998), which in the case of manufacturing improvement concerns both the tools and methods for improvement and the production 588 B.J Hicks and J Matthews Figure Manufacturing improvements paradigms and their corresponding tools and methods system(s) Further, the content and learning outcomes of the training have to reflect goodpractice or at least improved practices, which must be determined in advance Central to the success of the training is the need to develop a common and shared understanding across all the trainees in order to generate the same intended learning outcome(s) This is necessary to ensure consistent practices and in particular, consistent operation of equipment, control of materials and the adoption of appropriate machine settings to maintain quality and avoid excessive wear (Adebanjo and Kehoe 2001) (3) Maintenance The ability to keep a manufacturing process efficient depends heavily upon good work practices and effective maintenance This is particularly important in today’s just-intime production environment, where as a consequence of reduced stock level minor breakdowns are even more likely to stop or inhibit production (Eti et al 2006a) and reduce overall equipment effectiveness (efficiency) There are two approaches for achieving this Training to ensure changes to working practices and operating procedures are effectively taken-up The importance of training in motivating the operators and promoting ‘team work’ has been widely acknowledged (Reik et al., 2006; Scholtes et al., 2003) Improve the speed and ease of exchange of subassemblies (Boothroyd et al., 2001) Focuses on the machine operator as the key component of maintenance Operator is tasked with performing the routine maintenance tasks (Wilmott, 1997) The motto of TPM is ‘‘zero error, zero work-related accident, and zero loss’’ Hence it can be thought of as ‘deterioration prevention’ and ‘maintenance reduction’, not purely the ‘fixing’ of machines Training of the operating staff Design out Maintenance or Design for Service Total Productive Maintenance (TPM) Operator-led Maintenance Five goals: to maximize equipment effectiveness; to develop a system of productive maintenance for the life of the equipment; to involve all departments that plan, design, use, or maintain equipment in implementing TPM; to involve all employees and to promote TPM through motivational management (Redman and Grieves, 2005) (continued) Machines and products are monitored/ measured by virtue of appropriate sensors – vision, proximity etc, and where undesirable measures are recorded the production system is altered automatically – both upstream to correct and downstream to compensate Intelligent monitoring and control of production system to provide near real-time correction/adjustment (Limanond et al.,1998) In-process monitoring and control Central to the success of training is the need to develop a common and shared understanding across all the trainees in order to generate the same intended learning outcome(s) This is necessary to ensure consistent practices and in particular, consistent operation of equipment, control of materials and the adoption of appropriate machine settings to maintain quality and avoid excessive wear (Adebanjo and Kehoe, 2001) Depending on the relative likelihood of failure of a particular component or subassembly more effort into improving maintainability (disassemblability and assemblability) of this component is justified The understanding necessary to explicitly define set-up rules and parameters – right first time/best compromise These rules need to be programmed into the machine controller and their adjustment may require a skilled operator and/or prior knowledge of, in many cases, sophisticated logic and machine sequencing Requires in-depth knowledge of the relationship between product variation and process variation - both upstream and downstream - in order to alter machine parameters and settings during run-up and operation Central to this is the need to describe the acceptable variation in product attributes during all stages of production For effective training to be delivered the trainer needs to possess an in-depth understanding of the content (Davis and Davis 1998), which in the case of manufacturing improvement concerns both the tools and methods and the production system(s) Further, the content and learning outcomes of the training have to reflect good-practice or at least improved practices, which must be determined in advance While this design-led approach does not directly impact upon the nature of the production process the influence of the procedures associated with disassembly and assembly on process set-up must be understood in order to reduce both exchange time and minimise run-up Relies heavily on both the management and the operators possessing an understanding of: the function of the process, suitable machine settings, the impact of wear on the process, and the effect of operating conditions (production rate and environmental conditions) Machine settings are pre-programmed into a controller and associated with a particular product Automation of the physical changes to the manufacturing system necessary to process a product variant Automation of changeover and machine set-up Process control Knowledge requirements Principles Description The principles and underlying knowledge requirements of tools and methods for manufacturing systems improvement – Part a Approach Table International Journal of Computer Integrated Manufacturing 589 Quality Table A planning and communication method (Cohen, 1993) that is widely used in the development phase of equipment and machinery for identifying the customer requirements and translating them into technical characteristics Aimed at embedding awareness of quality in all organisational processes and ultimately strives to create customer satisfaction at continually lower real costs (Oakland, 2003) A business management strategy, originally developed by Motorola, to identify and remove the causes of defects and errors in manufacturing and business processes (Adam et al., 2003) Quality Function Deployment (QFD) Total Quality Management (TQM) Six Sigma Relies on an understanding of the function of the machine /production system and the use of predictive techniques such as Failure Mode Effects and Criticality Analysis (FMECA) or Fault Tree Analysis (FTA) implementation (Hague and Johnston, 2001) An engineering framework that enables the definition of a complete maintenance regime (Moubray, 1997) It regards maintenance as the means to maintain the functions a user may require of machinery in a defined operating context As a discipline it allows manufacturers to monitor, assess, predict and generally understand the workings of the equipment (Mitchell, 2002) Developed as a ‘‘method to transform user demands into design quality, to deploy the functions forming quality, and to deploy methods for achieving the design quality into subsystems and component parts, and ultimately to specific elements of the manufacturing process’’ (Akao, 1996) Include three activities: quality of return for shareholders, quality of products/services to endusers and quality of life at work and home Uses a set of quality management methods, including statistical methods, and requires an infrastructure of people within the organization who are experts in these methods Each Six Sigma project carried out within an organization follows a defined sequence of steps and has quantified financial targets (cost reduction or profit increase) One commonly used statistic method is Control charts (Wheeler, 2000) to assess the nature of variation in a process and to facilitate forecasting and quality management Focuses on identifying and establishing the operational, maintenance and capital improvement policies that will manage the risks of equipment failure most effectively (Smith, 2005) ReliabilityCentred Maintenance (RCM) Core to this activity is the development of the knowledge and understanding of the process and product, and specifically the areas where the product quality is influenced by interaction with the process Thomas and Webb (2003) and Antony (2007a; 2007b), shows that knowledge and understanding are key factors for successful Six Sigma implementation This understanding centres on the interaction between the process and the product, and is essential for directing the measurement, analysis, improvement and control of process and process inputs (materials and staff) A design focused activity (process and products) that can be applied to a new or existing product or service and requires an understanding of function and its relationship to quality (Govers, 2001) Knowledge requirements Principles Description Approach (Continued) 590 B.J Hicks and J Matthews Approach Description Principles Tooling design Product modification and new product introduction Tooling design and changeover Goods manufacturers are often faced with the task of processing new or altered products – such as new sizes, new materials and modified configurations (Matthews et al., 2009) Central to achieving this, is the need to determine an appropriate set of machine settings This involves production trials and potentially time-consuming trial and error testing and demands that a modified product or prototype be obtained An alternative approach is to perform a comparative assessment of new products with existing products and their associated machine settings to derive an initial set of new machine settings (Giess and Culley, 2003) In some cases a suitable set-up may not be possible and the product cannot be processed Here the production team must determine how to modify the system and/or provide recommendations for altering the properties or characteristics of the product Improve production performance and in particular flexibility - without compromising efficiency – through improved design of tooling (continued) To determine the functional requirements for redesign it is necessary to understand the limitations of the existing equipment The factors that limit the capability can be inverted in order to define the rules which are necessary for successful processing Further, the rules provide a series of objective measures for the evaluation and assessment of new equipment No matter whether it is the determination of settings for a new product or the improvement in process capability through product modification, it is necessary to understand the capability of the production process and its relationship with the properties and characteristics of the product (Frey et al., 2000) The intrinsic relationship between product design and process capability is widely acknowledged (Deleryd, 1998) and there are a variety of methods for improving process capability through product modification These include Design for Assembly (DfA), Design for Manufacturability (DFM) and Design For Manufacture Assembly (DFMA) (Dewhurst and Abbatiello, 1996) Central to the success of the SMED or DFC activities is the need to be able to understand and specify in advance the machine settings (set-up point) and range of variation (run-up adjustment) necessary for the successful processing of each product variant (Mileham et al., 2003) Central to the ability to improve tooling design is the need to generate functional design rules (design requirements) (Pahl and Beitz, 1996) i.e what needs to be achieved by the process in terms of the final product Knowledge requirements The principles and underlying knowledge requirements of tools and methods for manufacturing systems improvement – Part b Key to achieving this is to determine the most appropriate design or configuration of tooling and, if appropriate, the most efficient methods for changeover between tooling configurations (i.e minimising changeovers and/or changeover time) This includes both the physical geometry (size, profile and number of) and control of the tooling (kinematics - motion, velocity and acceleration, timing and clearances) (Hicks et al., 2001) The methodologies guide the designer Changeover Improve changeover performance through a step-by-step process from through automation and/or analysing changeover capabilities techniques such as Single-Minute through to the identification of Exchange of Die (SMED) (Shingo, improvement opportunities The 1985) or Design for Changeover approach builds understanding of the (DFC) (McIntosh et al., 2001) basic concepts and methods for the identification of improvement ideas and their potential benefits Equipment redesign, modification and Where existing equipment cannot meet increases in manufacturing capability it is replacement necessary to either modify or replace the equipment In cases where the process and the design principles which underlie the equipment are close to their limits then a process and equipment redesign is necessary (Hicks et al., 2004) In either case – modification, replacement or redesign – it is a prerequisite that both capability and functional requirements are determined Table International Journal of Computer Integrated Manufacturing 591 592 Focuses on improving the efficiency and effectiveness of the overall business processes that exist within and across an organization Involves the fundamental assessment of mission and goals and the reengineering of the organization’s business processes - the steps and procedures that govern how resources are used to create products and services that meet the needs of particular customers or markets Business Process Reengineering/ Redesign (BPR) For the successful adoption of lean a functional perspective of the production system is required in order for value streams to be identified and mapped, and to ensure that value streams flow In a manufacturing context, function is the only means to add value to the product Although not all functions may add value Achieved by establishing the processes and assigning responsibility for those processes to dedicated teams and, where appropriate, systems (Hammer and Champy, 1993).In order to maintain and improve processes an understanding of the functions and processes and the value of each function must be elicited The core principles of lean thinking are: Specify value Identify value streams Make value flow Let the customer pull value Pursue perfection The term ‘lean’ was coined by Womack et al (1990) to describe the philosophy – of reducing waste throughout a company’s value stream It is not just a set of tools for the reduction of waste (Bicheno, 2003), but a way of thinking which puts the customer first Approach Lean thinking Other manufacturing philosophises Table (Continued) Description Principles Knowledge requirements B.J Hicks and J Matthews The first is preventive maintenance which aims to reduce the probability of failure in the time period after maintenance has been applied The second is corrective maintenance, which strives to reduce the severity of equipment failures once they occur (Loftsen 2000) As noted by Waeyenbergh and Pintelon (2004) industrial systems evolve rapidly so maintenance initiatives will also have to be reviewed periodically in order to take into account the changing systems and the changing environment This calls not only for a structured maintenance concept, but also one that is flexible There are a variety of maintenance improvement methods including Design for Service (DfS) (Dewhurst and Abbatiello 1996), Total Productive Maintenance (TPM) (Willmott 1997) and Reliability Centred Maintenance (RCM) (Smith 2005) which arguably focus on the design, the operator and the engineering function respectively These various approaches depend on both the management and the operators possessing an understanding of: the function of the process, the influence of machine settings on process performance, the impact of wear on the process, and the effect of operating conditions (production rate and environmental conditions) (4) Quality In a similar manner to maintenance there are a variety of methods and initiatives that support quality control, improvement and assurance These include Quality Function Deployment (QFD) (Govers 2001), Total Quality Management (TQM) (Oakland 2003) and aspects of Six Sigma (Adams et al 2003) These various approaches require an understanding of function and its relationship to quality, and an understanding of the interaction between the process and product, which are essential for directing the measurement, analysis, improvement and control of process and process inputs (materials and staff) (Thomas and Webb ( 2003) and Antony (2007a, 2007b)) (5) Tooling design and changeover The ultimate aim of improving tooling design is to improve production performance and in particular flexibility, without compromising efficiency Key to achieving this is to determine the most appropriate design or configuration of tooling and, if appropriate, the most efficient methods for changeover between tooling configurations (i.e minimising changeovers and/or changeover time) This includes both the physical geometry (size, profile and number of) and control of the International Journal of Computer Integrated Manufacturing tooling (kinematics – motion, velocity and acceleration, timing and clearances) (Hicks et al 2001) Central to the success of the Single-Minute Exchange of Die (SMED) (Shingo 1985) or Design for Changeover (DFC) (McIntosh et al 2001) activities is the need to be able to understand and specify in advance the machine settings (set-up point) and range of variation (run-up adjustment) necessary for the successful processing of each product variant (6) Equipment redesign, modification and replacement Where an increase in manufacturing capability is sought that exceeds the existing equipment or process capability, it is necessary to either modify or replace the equipment In cases where the process and the design principles which underlie the equipment are identified to be close to their limits then a process and equipment redesign may be necessary (Hicks et al 2002) In either case – modification, replacement or redesign – it is a prerequisite that both capability and functional requirements are determined Central to determining these requirements is the need to understand the limitations of the existing equipment (Matthews et al 2007, Ding et al 2009) The factors that limit the capability can be inverted in order to define the rules which are necessary for successful processing This understanding is central to realising redesigned or new equipment that overcomes the limitations of existing equipment and ultimately improves performance (quality, efficiency and/or flexibility and capability) The rules also provide a series of objective measures for the evaluation and assessment of new equipment (Matthews et al 2008) (7) Product modification and new product introduction In today’s dynamic global markets, goods manufacturers are frequently faced with the task of processing new or altered products – such as new sizes, new materials and modified configurations (Matthews et al 2009) Central to achieving this, is the need to determine an appropriate set of machine settings that enable the product to be successfully processed No matter whether it is the determination of settings for a new product or the improvement in process capability through product modification, it is necessary to understand the capability of the production process and its relationship with the properties and characteristics of the product (Frey et al 2000) (8) Other manufacturing philosophises In addition to these seven areas of manufacturing improvement there exist a number of 593 philosophies to support improvements in manufacturing and management These include lean thinking and Business Process Reengineering The term ‘lean’ was coined by Womack et al (1990) to describe the main aim of the philosophy – the reduction of waste throughout a company’s value stream However, for some lean promoters it is not just a set of tools for the reduction of waste (Bicheno 2003), but a way of thinking which puts the customer first Once this way of thinking is adopted, lean tools are available to reduce waste and improve benefits for the customer For the successful adoption of a lean approach a functional perspective of the production system is required in order for value streams to be identified and mapped, and to ensure that value streams flow In a manufacturing context, function is the only means to add value to the product Although not all functions may add value In contrast to lean, business process reengineering or business process redesign (BPR) focuses on improving the efficiency and effectiveness of the overall business processes that exist within and across an organisation This is achieved by establishing the processes and assigning responsibility for those processes to dedicated teams and, where appropriate, systems (Hammer and Champy 1993) In order to maintain and improve processes an understanding of the functions and processes and the value of each function must be elicited The previous sections have discussed the various manufacturing improvement paradigms and corresponding tools and methods with respect to their underlying principles and the knowledge and understanding that underpin their use Further examination of the knowledge requirements reveals six fundamental knowledge concepts relating to the improvement of manufacturing systems These include: (1) An understanding of the relationship between the properties and characteristics of the product, and the machine and process settings (2) An understanding of the relationship between product variation and process variation, and their influence on quality and efficiency (3) An understanding of the influence of operator procedures on quality, efficiency, and flexibility (4) An understanding of the impact of wear and operating conditions (production rate and environmental conditions) on quality and efficiency 672 R Grangel et al Doumeingts, G., Vallespir, B., and Chen, D., 1998 Decisional modelling using the GRAI Grid In: International handbook on information systems, 313–337 Springer-Verlag Eriksson, H and Penker, M., 2000 Business modeling with UML: Business patterns at work John Wiley Grangel, R., et al., 2007 Transforming GRAI extended actigrams into UML activity diagrams: A first step to model driven interoperability In: R.J Gonc¸alves, J Muller, K Mertins, and M Zelm, eds 3rd International Conference on Interoperability for Enterprise Software and Applications, Enterprise Interoperability II, New Challenges and Approaches, March, 447–458 Springer Grangel, R., Bourey, J.P., and Berre, A., 2006 Solving problems in the parametrisation of ERPs using a modeldriven approach In: G Doumeingts, J Muller, G Morel, and B Vallespir, eds Enterprise Interoperability New Challenges and Approaches, Interoperability for Enterprise Software and Applications Conference (IESA’06) March Springer, 91–101 Grangel, R., et al., 2005 Enterprise modelling, an overview focused on software generation In: H Panetto, ed Interoperability of Enterprise Software and Applications Workshops of the INTEROP-ESA International Conference EI2N, WSI, ISIDI and IEHENA 2005 Hermes Science Publishing, 65–76 Grangel, R., Cutting-Decelle, A.F., and Bourey, J.P., 2008 A UML profile for transforming GRAI Extended Actigrams into UML In: 17th IFAC World Congress July, Seoul (Korea) IBM, 2009 IBM Rational Software Modeler Development Platform 7.5 [online] Available from: http://www-01.ibm com/software/awdtools/modeler/swmodeler/ [Accessed 07 September 2009] IDEF, 2007 Integrated DEFinition Methods [online] Available from: http://www.idef.com/ [Accessed 07 September 2009] INTEROP, 2007 Interoperability Research for Networked Enterprises Applicationsand Software NoE (IST-2003508011) [online] Available from: http://interop-vlab.eu/ [Accessed 07 September 2009] Jardim-Gonc¸alves, R., Grilo, A., and Steiger-Garcao, A., 2006 Challenging the interoperability between computers in industry with MDA and SOA Computers in Industry, 57 (8), 679–689 Jouault, F., et al., 2006 ATL: a QVT-like transformation language In: P.L Tarr and W.R Cook, eds OOPSLA Companion ACM, 719–720 LINA, 2009 research laboratory [online] Available from: http://www.lina.univnantes.fr/ [Accessed 07 September 2009] MagicDraw, 2009 Magic Draw [online] Available from: http://www.magicdraw.com/ [Accessed 07 September 2009] Marshall, C., 2000 Enterprise modeling with UML Designing successful software through business analysis Addison Wesley Milewski, M and Roberts, G., 2005 The model weaving description language (MWDL)-Towards a formal aspect oriented language for MDA model transformations In: Proceedings of the 1st Workshop on Models and Aspects Handling Crosscutting Concerns in MDSD, in conjunction with the 19th European Conference on Object-Oriented Programming OMG, 2003 MDA Guide Version 1.0.1 [online] Available from: http://www.omg.org/docs/omg/03-06-01.pdf [Accessed 07 September 2009] OMG, 2006 Object Constraint Language 2.0 [online] Available from: http://www.omg.org/spec/OCL/2.0/ [Accessed 07 September 2009] OMG, 2009 Unified Modeling Language: Superstructure, version 2.2 [online] Available from: http://www.omg.org/ spec/UML/2.2/ [Accessed 07 September 2009] Tae-Young, K., et al., 2006 A modeling framework for agile and interoperable virtual enterprises Computers in Industry, 57 (3), 204–217 Vernadat, F.B., 1996 Enterprise modeling and integration: principles and applications Chapman and Hall Williams, T.J., 1993 The Purdue enterprise reference architecture In: Proceedings of the Workshop on Design of Information Infrastructure Systems for Manufacturing, November Elsevier International Journal of Computer Integrated Manufacturing Vol 23, No 7, July 2010, 673–686 Aided analysis for quality function deployment with an Apriori-based data mining approach Zaifang Zhang, Hui Cheng and Xuening Chu* School of Mechanical Engineering, Shanghai Jiao Tong University, Shanghai, 200240 PR, China (Received December 2009; final version received May 2010) Quality function deployment (QFD) is a proven useful methodology in new product development to satisfy customer requirements (CRs) House of quality (HoQ), the general implementing mode of QFD, is aimed to identify the variables of engineering characteristics (ECs) based on the relationships between CRs and ECs Traditionally, the establishment of these relationships is mainly dependent on the designers’ experience and then the HoQ included many items difficult to handle For aiding the designers on the HoQ analysis, the paper proposes an Apriori-based data mining approach to extract knowledge from historical data The approach is mainly focused on mining potential useful association rules (including positive and negative rules) that reflect the relationships according to three objectives: support, confidence, and interestingness For ensuring the availability and conciseness of these extracted rules, the definitions and calculations of rule conflict and redundancy are proposed and processing procedures are also developed to unite or delete unnecessary rules The reserved rules are clustered in order to facilitate rule management and reuse Furthermore, a reuse procedure is also developed for new HoQ analysis Computational experiments of an electrically powered bicycle are used to illustrate the proposed approach and its capability of extracting useful knowledge Keywords: quality function deployment; data mining; Apriori approach; association rule Introduction Quality function deployment (QFD), a well-known product planning methodology, has been widely used in new product development to generate competitive solutions for improving customer satisfaction (Hauser and Clausing 1988, Chan and Wu 2002, Bu¨yu¨ko¨zkan and Feyzio glu 2005, Chin et al 2005, Tu et al 2003, Zhang and Chu 2009a) It can facilitate the understanding of ‘voice of customer’, i.e customer requirement (CRs), and then translate CRs into engineering characteristics (ECs), parts characteristics, process plans and production requirements (Griffin and Hauser 1993, Bu¨yu¨ko¨zkan et al 2004) The first translation, known as the house of quality (HoQ), is mainly studied in most literatures since it can largely affect the target value setting of other translations in the later design stages (Kwong and Bai 2002) The main steps in HoQ are to determine the importance weights of CRs, to establish the relationships between CRs and ECs and to identify the proper ECs Various methods have been developed to tackle with the issue, and most works are classified into two categories: establishing the HoQ (includes determining the weights and the relationships) and calculating the outputs of HoQ (i.e the appropriate values of ECs) For the former, analytic hierarchy process (AHP) as a *Corresponding author Email: xnchu@sjtu.edu.cn ISSN 0951-192X print/ISSN 1362-3052 online Ó 2010 Taylor & Francis DOI: 10.1080/0951192X.2010.492840 http://www.informaworld.com general method was used for determining the importance weights of CRs (Armacost et al 1994, Lu 1994) Considering the issue as a typical group decisionmaking problem, Lai et al (1998) used a linear programming technique to aggregate individual preferences into a group consensus Ho et al (1999) also proposed an integrated group decision-making method to aggregate team members’ opinions and minimise inconsistency over group and individual preferences Concerning the market benchmarks and inner-relationships among CRs, Feyzioglu and Bu¨yu¨ko¨zkan (2008) used a 2-additive Choquet integral based procedure for determining the final importance weights of CRs Kano model was also used to classify the CRs in order to capture the relationships between CRs and ECs precisely (Matzler and Hinterhuber 1998, Chen and Ko 2008) Considering the inevitable vagueness and uncertainty in the decision-making process, fuzzy numbers, linguistic data and even incomplete data were used to express the judgments for the weights and the relationships (Shen et al 2001, Kwong and Bai 2002, Han et al 2004) Fuzzy set theory was integrated with the existing methods, e.g fuzzy AHP (Kwong and Bai 2002), fuzzy pairwise comparison ranking method (Li et al 2006), fuzzy arithmetic and entropy method (Chan et al 1999), and fuzzy mathematical programming (Lai et al 2008) Considering the personalities of 674 Z Zhang et al decision-makers (e.g individual culture, experience and knowledge), Bu¨yu¨ko¨zkan and Feyzio glu (2005) and Bu¨yu¨ko¨zkan et al (2007) proposed two similar fuzzy group decision-making approaches to fuse multiple preference styles into the uniform judgments for determining the weights and the relationships Furthermore, Zhang and Chu (2009b) proposed an optimising model to aggregate some multi-format and multi-granularity linguistic judgments into a group consensus To extract the potential inner-relationships among CRs, Temponi (1999) developed a heuristic inference scheme based on a fuzzy logic-based method Fuzzy inference model based on design experience and knowledge was also proposed to establish fuzzy rules which define the relationships between CRs and ECs (Fung et al 1998, Harding et al 2001) Based on the quantified representations of ECs and customer perceptions in several samples, regression approach (or fuzzy regression approach) was developed to identify the quantitative relationships between CRs and ECs and then the values of new product ECs are determined based on the mathematical programming model (Kim et al 2000, Chen et al 2004, Fung et al 2006) However, a precondition for regression approach is to determine which ECs have relationships with each CR by the designers For the latter (i.e calculating the outputs of HoQ), fuzzy set theory can effectively capture the uncertainty and fuzzy nature in the process (Bu¨yu¨ko¨zkan and Feyzio glu 2004) Fuzzy calculating was used for rating ECs by integrating fuzzy weighted average method and fuzzy expected value operator (Chen et al 2006) In order to generate more reasonable ECs, some factors (e.g technology and cost) are added into the planning model of HoQ Thus, many fuzzy programming models were established to fulfill the holistic planning, e.g fuzzy inference (Vanegas and Labib 2001), inexact genetic algorithm (Bai and Kwong 2003), linear programming model (Fung et al 2002, 2003) and fuzzy multiple goal programming model (Chen and Weng 2006) Generally, these models take customer satisfaction as the objective and take technical difficulty and cost as the constraints Fuzzy relationships between CRs and ECs are used to establish the constraint function of these models Other methods were also integrated with the existing models to facilitate or modify the calculation of HoQ, e.g fuzzy analytic network process (ANP) for incorporating the inner-dependences between CRs and ECs (Kahraman et al 2006) and the method of imprecision for reflecting the compensation levels among ECs (Chen and Ngai 2008) All these methods mentioned above are effective for the analysis of HoQ However, most of them are subjective because the values (e.g the judgments for the weights or relationships) are mostly determined by the designers based on their experience and knowledge Furthermore, the bigger the item number of CRs and ECs is, the more judgments the designers have to make And then the corresponding HoQ will be difficult to handle Can the designers use the data accumulated in the engineering applications to aid the analysis? Zhang et al (1996) proposed a neural network (NN) approach by which the relationships are generalised among example data instead of subjective guesses Although the method can effectively learn complex nonlinear relationships, these relationships are still in ‘black boxes’ which the designers not hold tangibly Furthermore, the NN approach takes all CRs and ECs as the inputs and outputs of network (i.e establish the relationships among all ECs and all CRs), respectively In order to ensure the outputs of the network are reliable, it is needed that historical data should cover all the variable ranges of inputs and outputs effectively and adequately However, it is more difficult for most real-world applications Data mining, which has been widely applied in engineering design, is used to extract unknown or potential hidden knowledge in a dataset (Chen et al 1996, Harding et al 2006, Choudhary et al 2009) For HoQ, Li and Wang (2010) proposed a rough set based data mining approach to elicit a minimal rule set for establishing the relationships between CRs and product categories In the method, the products must be classified into several categories based on the values of ECs and then this is seen as category rule extraction Most importantly, CRs may have relationships with any sub-sets of ECs The classification that predefines product categories limits the extracted relationship range and then results in that some potential relationships cannot be identified Furthermore, the usefulness of the extracted rules (e.g support and confidence) needs to be discussed, which has distinct effects on the reuse of the rules The data mining in HoQ is a typical multidimension problem (i.e the front-item and back-item are both multiple) Association rules are more appropriate for HoQ, which are extracted from the historical data of HoQ Association rule extraction is to elicit the potential relationships between two item-sets for benefiting product design (Shahbaz et al 2006) Apriori developed by Agrawal and Srikant (1993 and 1994) is an effective method for association rule extraction from a predefined dataset The approach is to extract the strong association rules by finding the frequent items, which was widely used in different applications (e.g mining product maps for new product development (Liao et al 2008) and product portfolio identification (Jiao and Zhang 2005)) In the International Journal of Computer Integrated Manufacturing studies of Jiao and Zhang (2005), Shao et al (2006) and Xia and Wang (2010), a simple Apriori-based algorithm was used to mine the association rules between CRs and product specifications, product specifications and configuration alternatives, and CRs and the items of product family architecture, respectively Two indices (i.e support and confidence) are used to evaluate the extracted rules in these three studies However, how to use association rules for facilitating HoQ analysis is not adequately discussed in the existing literatures In the association rule extraction, Gray and Orlowska (1998) deemed that support and confidence are not sufficient to mine interesting rules where uninteresting or misleading rules may be extracted Then other indices should be added for evaluating the rules Some possible issues in the rule mining of HoQ should be fully discussed, e.g rule conflict and redundancy Negative rules may be also useful for CRs mapping and then they should be extracted Furthermore, the extracted rules cannot completely cover all possible ranges of CRs and ECs, how to identify the rest of ECs in HoQ should be addressed The historical data of HoQ are acquired from the existing HoQ cases or derived from the existing products An Apriori-based data mining approach is proposed for extracting association rules which express the relationships between CRs and ECs of HoQ In the approach, positive and negative rules are both extracted According to the given CRs, the positive and negative rules are used to deduce the appropriate ECs and reduce the feasible range of ECs, respectively Three objectives, support, confidence, and interestingness, are used to evaluate the usefulness of the extracted rules In data mining process, the minimums of the three objectives should be assigned for ensuring the usefulness of the rules (Chen et al 1996, Shahbaz et al 2006, Shao et al 2006) Each extracted rule has the following format: IF CRs THEN ECs However, some conflicting or redundant rules may still exist in the rule set Thus, the definitions and calculations of rule conflict and redundancy are proposed and then processing procedures are developed to identify and eliminate these rules The final rules are clustered into several categories in order to facilitate their reuse Generally, the rules cannot fully cover all CRs and ECs and only the rules the designers can understand will be used in the following HoQ analysis A simple reusing procedure is developed to guide the designers in determining the corresponding values of ECs according to the extracted rule sets The paper is organised as follows The problem formulation is developed in the next section The third section proposes an Apriori-based data mining approach to elicit useful association rules In the fourth 675 section, an example is used to illustrate the proposed method Conclusions are then presented in the final section Problem formulation 2.1 QFD and HoQ QFD was originally developed in Japan in 1972 (Akao 1990) In various applications, QFD supports product/ service development or improvement, maximise customers’ satisfaction, decrease the product development time, and improve product quality (Hauser and Clausing 1988) Each translation can be represented graphically in a matrix configuration, and the first one is known as house of quality (HoQ) (Kwong and Bai 2002) Based on the relationships between CRs and ECs, HoQ aims to identify appropriate ECs (as outputs) according to a specific set of CRs (as inputs) Its structure includes CRs with their relative importance, the correlation matrix of ECs, and the relationship matrix between CRs and ECs, and the weights and values of ECs The corresponding calculating process has the following four steps (Griffin and Hauser 1993): (1) Determine the importance weights of CRs; (2) Identify ECs and their correlation matrix; (3) Elaborate the relationship matrix between CRs and ECs; (4) Action plan (determine the weights of ECs and then identify their values) 2.2 Association rule extraction for HoQ Data mining for HoQ is to find useful knowledge from the accumulated cases for aiding the designers on the following HoQ analysis Specifically, the purpose is to extract the reasonable association rules which express the strong relationships between CRs and ECs All these rules will be used in new product development by the designers for identifying the appropriate ECs The problem formulation is given in Figure For a product, existing cases (i.e QFD cases or other design cases) expressed by CRs with corresponding ECs are used to form a case base Here, other design cases indicate the successful product cases which were not developed by using the QFD tool because enterprises may use various methods and tools to develop new products with CRs Corresponding CRs and ECs in these cases can be easily gathered to elicit useful rules An Apriori-based data mining approach is used to find out all useful association rules between CRs and ECs Through the analysis of rule conflict, redundancy and clustering, the useful rules are stored in a rule base Note that only these rules which can be understood by the designers are preserved for the following reuse and 676 Z Zhang et al Figure Problem formulation others will be deleted or stored in another base for future analysis For a new product development, suppose (n7j) ECs can be deduced by using corresponding rules in the rule base and then (m7i) CRs which are only related to these ECs are not considered in the new HoQ Thus, it is more convenient for the designers to analyse because the HoQ will be just established for i CRs and j ECs as shown in the shadow of Figure of association intensity A simple formula is given as follows: conðX ) YÞ ¼ ð2Þ where T(X) is the number of items in the database T satisfying all the conditions in the front-items X (3) 2.3 TðX & YÞ TðXÞ Interestingness Three objectives for association rule extraction Suppose T * {(CRk, ECk)jk ¼ 1,2, , P} is a case dataset, where P is the number of the cases A general manner of association rules is expressed as: IF X THEN Y, where X  CRS and Y  ECS (1) Support Support is a general index in rule extraction The index is a measure of the importance of association rules for expressing the universal degrees of association existing A simple formula is given as follows: T ðX & Y Þ supðX ) YÞ ¼ jTj int ðX ) YÞ ¼ ð1Þ where T(X & Y) is the number of instances in the database T satisfying all the conditions in both frontitems X and back-items Y, and jTj is the total number of instances in the database T (2) The two above-mentioned indexes are widely used in association rule mining Despite being useful measures, it is not sufficient to mine interesting rules where uninteresting or misleading rules may be extracted (Gray and Orlowska 1998) Therefore, an index of interestingness developed by Gray and Orlowska (1998) is adopted to eliminate such rules The corresponding formula is given as follows: Confidence The index is a measure of the veracity of association rules for expressing the characteristics supðX ) YÞ À1 supðXÞ Â supðYÞ ð3Þ The index is a measure of the independence between X and Y When the relationship between X and Y is dependent, independent, or weaker than independent, int(X ) Y) is greater than 0, equal to 0, or smaller than 0, respectively In the study, when int(X ) Y) is equal to or smaller than 0, the rules are taken as uninteresting rules and then deleted from the extracted rule set The three indexes are deemed as of equal importance in this study International Journal of Computer Integrated Manufacturing 3.1 An Apriori-based data mining approach for HoQ Basic definitions and concepts According to the pioneering work of Agrawal and Srikant (1994), some basic definitions and concepts are given for the Apriori-based data mining approach Definition 1: k-itemset The itemset is a set of some items If item number of an itemset is k, the itemset is called k-itemset Definition 2: Frequent itemset For an itemset, if occurrence frequencies of all items are all greater than or equal to the product of min-sup and jTj, the itemset is called frequent itemset Min-sup is the threshold value for the index of support Definition 3: Apriori-gen function For a set Lk71 of all frequent (k–1)-itemsets, the Apriori-gen function returns a superset of the set of all frequent k-itemsets The implementing process of the function is divided into two steps: join step (join (k–1)-itemsets to generate all possible k-itemsets which comprises a candidate set) and prune step (in the candidate set, delete all k-itemsets in which one or several (k–1)-subsets does not belong to Lk71) For an association rule ~ ri, suppose f1(~ ri) ¼ sup(~ ri), f2(~ ri) ¼ con(~ ri), and f3(~ ri) ¼ int(~ ri), where fj(~ ri) is the corresponding objective function of the three indexes respectively and j {1, 2, 3} Definition 4: Strong association rule For a random association rule (~ ri), when f1(~ ri) ! min-sup, f2(~ ri) ! min-con, and f3(~ ri) ! min-int, the rule is deemed as a strong association rule Min-con and min-int are the threshold values for the last two indexes (i.e confidence and interestingness), respectively Considering the sequence relations of CRs and ECs in HoQ, the generated rules have a unified form: IF CRi THEN ECj, where CRi and ECj are any sub-sets of CRs and ECs, respectively The rule is seen as positive association rule Furthermore, negative association rule is recently in high regard in rule extraction According to the study of Wu et al (2002), the definition of negative association rule is given as follows Definition 5: Negative association rule For an itemset (X, Y), the corresponding negative association rules are: (1) X ) Y, (2) X ) Y, (3) X ) Y, where X and Y represent the negative items of X and Y, respectively If X ) Y is a strong association rule, these negative rules are also strong association rules when satisfying the three conditions in Definition 4, respectively The requirement of X ) Y is a strong rule and ensures the significance of these negative rules The negative rules can not only reduce the feasible range of ECs (e.g X ) Y and X ) Y) but also can deduce the appropriate ECs (e.g X ) Y) For a 677 negative association rule, the formulae of the three indexes are given as follows (Dong et al 2004): Support: supðXÞ ¼ À supðXÞ ð4-1Þ supðX ) YÞ ¼ supðXÞ À supðX ) YÞ ð4-2Þ supðX ) YÞ ¼ supðYÞ À supðX ) YÞ ð4-3Þ supðX ) YÞ ¼ À supðXÞ À supðYÞ þ supðX ) YÞ ð4-4Þ Confidence: conðX ) YÞ ¼ À conðX ) YÞ ð5-1Þ conðX ) YÞ ¼ ðsupðYÞ À supðX ) YÞÞ ð1 À supðXÞÞ ð5-2Þ conðX ) YÞ ¼ À ðsupðYÞ À supðX ) YÞÞ=ð1 À supðXÞÞ ð5-3Þ Interestingness: intðX ) YÞ ¼ ðsupðX ) YÞÞ=ðsupðXÞ Ã ð1 À supðYÞÞÞ À ð6-1Þ intðX ) YÞ ¼ ðsupðX ) YÞÞ=ðð1 À supðXÞÞ Ã supðYÞÞ À ð6-2Þ intðX ) YÞ ¼ ðsupðX ) YÞÞ=ðð1 À supðXÞÞ Ã ð1 À supðYÞÞÞ À ð6-3Þ In order to facilitate management and reuse, positive rules and negative rules are deemed as two respective rule categories to deal with 3.2 Rule conflict, redundancy and clustering In the data mining process of HoQ, some conflicting or redundant rules may be extracted Furthermore, the extracted rules should be clustered to facilitate the analysis and reuse by the designers Related definitions are given as follows Definition 6: Rule conflict For a strong association rule ~ r1: IF X1 THEN Y1, if there exists another strong rule ~ r2: IF X2 THEN Y2, where X1 ¼ X2, Y1 and Y2 are different options for an identical EC (e.g speed is 50 km/s and 100 km/s, respectively), the two rules are conflicted, i.e rule conflict 678 Z Zhang et al In order to eliminate the conflicting rules, a measure is proposed to identify the dominated rules (1) When the following inequalities fi(~ r1) ! fi(~ r2) are all satisfied and at least one of the inequalities fi(~ r1) fi(~ r2) is satisfied, keep the rule ~ r1 as the dominated rule and then delete the rule ~ r2 which is deemed as a conflicting rule (2) When all the equalities fi(~ r1) ¼ fi(~ r2) are satisfied, the two rules are both preserved in the rule set because their priorities cannot be identified In the reuse process, the possible conflicting results deduced by these conflicting rules are needed a further analysis by the designers (3) Suppose the number of satisfying inequalities fi(~ r1) fi(~ r2) and fj(~ r1) fj(~ r2) are I and J, respectively, where I, J ¼ 0, 1, 2, or 3, and I þ J When I ¼ and J ¼ 1, it is hard to identify which is the dominated rule in a realworld application The values between each two of the three indexes have no comparability For example, suppose two rules ~ r1 (sup ¼ 0.3, ¼ 0.5 and int ¼ 0.6) and ~ r2 (sup ¼ 0.4, ¼ 0.5 and int ¼ 0.3) Obviously, f1(~ r 1) f1(~ r2), f2(~ r1) ¼ f2(~ r2) and f3(~ r1) f3(~ r2) However, the priorities of the two rules cannot be identified in the current conditions The differences of two sup or int values (i.e 0.1 and 0.3) not have a uniform dimension to compare In this study, whether a rule is a dominated one or not is determined by the number of objective priorities based on the ‘most’ principle (i.e most of the inequalities are satisfied) & The dominated rule ¼ ~ r1 ~ r2 IJ ð7Þ Definition 7: Rule redundancy For two strong association rules ~ r1: IF X1 THEN Y1 and ~ r2: IF X2 THEN Y2, where X1  X2 and Y2  Y1, the rule ~ r2 is seen as a redundant rule when at least two inequalities fi(~ r1) ! fi(~ r2) are satisfied The judgment of rule redundancy is also based on the principle of ‘most’ considering the diverse function objectives Definition 8: Rule clustering In order to facilitate the analysis and reuse of extracted rules, analogous rules can be clustered into one category by calculating their distances Suppose two strong association rules ~ r1: IFÈ X1 THEN r2: IF XÉ2 THENÈ Y2, where É Y1 and É È~ X1 ¼ x1i ji È2 m1 ; XÉ ¼ xj jj m2 ; Y1 ¼ yi ji n1 , and Y2 ¼ y2j jj n2 , where m1, m2, n1, and n2 is the number of items in X1, X2, Y1 and Y2, respectively HoQ is aimed to translate CRs into ECs and then the weight of front-items is deemed more important than that of the back-items in rule clustering The distance of the two rules is defined as follows: r2 Þ ¼ 2DðX1 ; X2 Þ þ DðY1 ; Y2 Þ Dð~ r1 ; ~ ð8Þ   corr x1i ; x2j maxðm1 ; m2 Þ;   X corr y1i ; y2j DðY1 ; Y2 Þ ¼ maxðn1 ; n2 Þ; > x1i 6¼ x2j <   > corr x1i ; x2j ¼ 0:75 x1i ; x2j CR ; and > > : x1i ¼ x2j > y1i 6¼ y2j <   > corr y1i ; y2j ¼ 0:75 y1i ; y2j EC ; > > : y1 ¼ y2 DðX1 ; X2 Þ ¼ X i j where i m1 or n1, j m2 or n2 Here, x1i ; x2j CR and y1i ; y2j EC express they belong to the same CR and EC, respectively For example, ‘Ease of use ¼ MH’ and ‘Ease of use ¼ H’ in the Section are belong to the CR2 ‘Ease of use’ Only when the distance D(~ r 1, ~ r 2) exceeds a predefined threshold value, the two rules can be clustered into a category 3.3 Apriori-based data mining procedure An Apriori-based algorithm procedure is developed to fulfill the data mining of relationships between CRs and ECs, which includes three main steps: find all possible frequent itemsets, generate strong association rules from these frequent itemsets and prune and cluster these rules Step Find all frequent itemsets (1) For a dataset T, set the threshold values of support, confidence and interestingness (i.e minsup, min-con and min-int) (2) Generate a set L1 ¼ {c1 ¼ (x1 or y1) (CRp or ECp)jc1 count ! min-sup jTj} (3) For k ¼ 2; Lk71 6¼ f k þþ) {Ck ¼ Apriori – gen(Lk71); // Ck; is the set of all candidate k-itemsets For each case t T {Ct ¼ Subset(Ck, t); // Candidates contained in t For each case (xi, yk–i) Ct // i (0,1, , k) is the item number (xi, yk–i) count þþ; } Lk ¼ {(xi, yk–i) Ck j (xi, yk–i) Á count ! minsup jTj}; Generate a È set L1k ¼ fðxi ; ykÀiÉÞ j i 6¼ or k; g (4) Return L ¼ L12 ; L13 ; ; L1k International Journal of Computer Integrated Manufacturing Step Generate the strong association rules For each frequent itemsets, if con(X, Y) ! Mincon and int(X, Y) ! Min-int, the rule X ) Y is preserved as a strong association rule The rule is generated from the frequent intemsets and then their supports are always equal to or greater than min-sup For the negative rules: X ) Y, X ) Y, and X ) Y, these rules will be preserved as strong association rules only when all their index values (i.e support, confidence and interestingness) are not less than the corresponding threshold values In order to ensure the usefulness of extracted rules, the three threshold values should not be very low Step Prune and cluster these strong association rules In order to facilitate the analysis and reuse by the designers, the positive rules and negative rules are clustered respectively (1) Check and delete the conflicting rules First, find all possible conflicting rules Second, fulfill the dominance-based calculation to compare these conflicting rules Finally, preserve the dominated rules and then delete the corresponding conflicting rules (2) Check and delete the redundant rules Only when front-items and/or back-items of two rules have an inclusion relationship, the two rules may be redundant For each rule~ r ¼ (X ) Y), search all rules f~ r g ¼ fðA ) BÞ j A  X; B  Yg And then, comparing the three objective values, the rule ~ r is a redundant rule when at least two inequalities fi ð~ r Þ fi ð~ rÞ are satisfied (3) Cluster all rules In order to facilitate the reuse, the rules that have single CR are classified into different categories according to each CR (named as single-CR category) Namely the rules of the same CR are classified into one category For the other rules (their front-items have two or more CRs), the density based spatial clustering of applications with noise (DBSCAN) is used for clustering them into different categories (Ester 1996) (named as multi-CRs category) According to the predefined radius and threshold item values, the method can discover clusters for these rules by using Equation (8) This study does not discuss the algorithm efficiency Generally, the magnitude in HoQ cases is tens or even hundreds in real-world applications (e.g automobile, engineering machinery) Furthermore, the study is to find the easily understood rules of which the item number is low, e.g 2, or Therefore, whether the proposed algorithm has high efficiency does not bring a distinct impact on the data mining in HoQ It is needed 679 to point out that the efficiency of data mining approach is still an important problem in some fields (e.g sale records of supermarket) 3.4 Reuse of the association rules The reuse procedure is given as follows Step Based on the pre-defined CR set of the ongoing product development, search the matching rules for deriving the proper ECs First, search the matching rules for individual CR in single-CR categories Second, according to the maximal frontitem number, generate all combinations of CRs If the maximum is 3, the item numbers of the combinations will be and Take the high-frequency CR as the main item in each multi-CRs category Search whether the item is included in the pre-defined CR set If yes, match all CR combinations with each rule in a multiCRs category If not, match all CR combinations with the left rules This may distinctly decrease the searching domain if the main item is not in the CR set When all matching rules are found out, an EC set which can partially or fully satisfy the CR set is identified Step Verify the EC set for repetition and conflict The repeated items will be united For the conflicting ECs, first compare their objective values and then the EC with bigger values will be held Here, an EC may occur more than once and then each objective value is calculated by summing the corresponding index values of the item For the negative results (deduced from the negative rules), their objective values are seen as negative and then should be subtracted from the total values of the item If their objective values are equal for multiple ECs, which ECs are preserved will be determined by the designers An illustrative example 4.1 Rule extraction An example is cited from the study of Shao et al (2006) to illustrate the proposed Apriori-based data mining approach on how to mine useful association rules between CRs and ECs The example is gathered from a company which is a producer of electrically powered bicycles (EPB) To satisfy various CRs, different product specifications are provided in the early design stage CRs are summarised into six items as shown in Table To capture the imprecision or uncertainty in the decision-making process of HoQ, linguistic variables are used to express these items The corresponding linguistic variables are very low (VL), low (L), medium low (ML), medium (M), medium high (MH), high (H), and very high (VH) 680 Z Zhang et al The product specifications of EPB which are seen as ECs in HoQ are represented by eleven attributes as given in Table Fifteen transaction records are gathered in Table for data mining In Shao et al (2006), fuzzy clustering method was used to cluster customers and product specifications of these records, a variable precision rough set method is used to extract useful information from them, and an Apriori approach is used to mine the configuration Table Customer requirements (Shao et al 2006) Customer requirements No CR1 CR2 CR3 Performance/cost Ease of use Little electrical consumption Aesthetics Security Little noise CR4 CR5 CR6 Table No VL, L, ML, M, MH, H, VH VL, L, ML, M, MH, H, VH VL, L, ML, M, MH, H, VH VL, L, ML, M, MH, H, VH VL, L, ML, M, MH, H, VH VL, L, ML, M, MH, H, VH Product specifications (Shao et al 2006) Product specifications { PS1 PS2 PS3 PS PS PS PS7 PS8 PS9 PS10 PS11 { Options Maximum speed Driving range Mass of EPB Maximum load Electrical consumption/ 100 miles Maximum noise Rating voltage Rating power Rating output torque Over-current protection Under-voltage protection Options 16, 18, 20 (km/h) 40, 50, 60 (km) 32, 36, 40 (kg) 70, 75, 80 (kg) 1.0, 1.2 (km.h) 45, 50, 60 (db(A)) 24, 36 (V) 180, 280, 320 (W) 6.25, 6.5, (N.M) 12–15, 12.5–15.5 (A) 29.5–31.5, 30.5–32.5 (V) PS is the abbreviation of product specification Table rules between product specifications and configuration alternatives In this study, these records are used as the historical data of HoQ to illustrate how to elicit association rules for facilitating the analysis of HoQ Here, CRs and product specifications in these records are deemed as the inputs (i.e CRs) and outputs (i.e ECs) of HoQ As mentioned above, how to mine useful association rules and use these rules in HoQ needs to be further discussed, e.g rule evaluation, rule reduction, rule clustering and rule reuse In the study of Li and Wang (2010), the rough set based data mining method can only support mining classification rule but not association rule In other words, the decision variables must be classified into several categories when using the method And the rules only express the relationships between CRs and a sub-set of ECs as a whole (i.e product category) However, in the analysis of HoQ, the potential rules are association rules which are many-to-many relationships Category rule extraction cannot randomly identify the relationships between CRs and any subsets of ECs And then some potential strong rules between CRs and these ECs may be lost For example, the example in Li and Wang (2010) used two PSs (maximum speed and driving range) for classifying the product and then only the relationships between CRs and product category (this also can be seen as the relationships between CRs and the set of the two ECs) are elicited The possible relationships between CRs and one EC are not identified, e.g CR1 is supposed to have a strong relationship with PS2 but not with the set of PSs and and then the strong relationship will be lost in the classification rule extraction Although the relationships between CRs and other ECs can be also extracted by using the method, some potential relationships may be omitted because the artificial selection of EC set limits the variable range For example, the Business lists of customer requirements and product specifications (Shao et al 2006) No CR1 CR2 CR3 CR4 CR5 CR6 PS1 PS2 PS3 PS4 PS5 PS6 PS7 PS8 PS9 PS10 PS11 10 11 12 13 14 15 M MH MH MH MH VH H H H VH MH MH MH MH MH MH MH MH MH MH H VH VH VH MH VH H H VH MH ML ML ML ML MH MH MH MH ML MH M M MH MH H H MH MH ML H MH MH L L H M M M MH MH H MH MH MH MH MH H H H M M M MH M H H MH MH H H MH MH MH MH MH MH MH MH ML MH 16 16 16 18 18 18 18 16 18 20 18 20 16 20 20 50 50 50 40 40 40 40 50 40 60 40 60 50 60 60 36 36 32 36 36 36 36 36 36 40 40 40 36 36 40 75 70 70 80 80 80 70 70 80 75 80 75 80 75 75 1.2 1 1.2 1.2 1.2 1.2 1.2 1 1.2 1.2 1.2 60 45 45 45 50 60 60 50 60 50 45 50 50 60 45 36 36 36 24 24 24 24 24 36 36 36 36 36 24 36 180 180 280 280 280 320 320 320 180 320 320 320 180 280 320 7 6.5 6.5 6.5 6.5 6.25 6.5 6.5 6.25 7 6.25 6.25 13.5 14 14 14 14 14 13.5 13.5 13.5 14 13.5 14 13.5 14 14 30.5 30.5 30.5 30.5 31.5 31.5 30.5 31.5 31.5 31.5 31.5 31.5 31.5 31.5 30.5 681 International Journal of Computer Integrated Manufacturing relationships between CRs and PS set (PSs 1and 2) or PS set (PSs and 4) can be mined by using the method twice However, the potential relationships between CRs with other combinations of PSs (e.g PSs and 3, PSs and 4, etc) still cannot be identified Obviously, it is unpractical to find the relationships between CRs and each of all possible combinations of PSs by using the method repeatedly In conclusion, the classification limits the relationship formally but ignores the randomicity of relationships between CRs and ECs in HoQ In the study, two kinds of rules are extracted: (1) high support with a low confidence and interestingness; (2) high confidence and interestingness but support is not especially cared for For the first category, some rules may be obvious and well-known, but other rules may be still not known to the designers tangibly This can enrich the common knowledge For the second Table Rules between CRs and ECs (high support with low confidence and interestingness) Type Positive association rules No IF (CRs) THEN (ECs) Sup Con Int Electrical consumption/100 miles ¼ Over-current protection ¼ 14 Under-voltage protection ¼ 30.5 Rating voltage ¼ 24 Under-voltage protection ¼ 31.5 Over-current protection ¼ 14 Over-current protection ¼ 14 Rating power ¼ 320 Over-current protection ¼ 14 0.33 0.4 0.33 0.33 0.4 0.33 0.33 0.47 0.33 0.71 0.86 0.71 0.71 0.86 0.83 0.83 0.64 0.79 0.43 0.79 0.79 0.43 0.39 0.39 0.36 0.67 Rating voltage ¼ 36 0.4 0.67 11 Ease of use ¼ MH Ease of use ¼ MH Ease of use ¼ MH Little electrical consumption ¼ MH Little electrical consumption ¼ MH Aesthetics ¼ MH Security ¼ MH Little noise ¼ MH Performance/cost ¼ MH and Ease of use ¼ MH Performance/cost ¼ MH and Little noise ¼ MH Ease of use ¼ MH 0.33 0.71 1.14 12 Little electrical consumption ¼ MH 0.33 0.71 0.79 13 Little electrical consumption ¼ MH 0.33 0.71 0.79 14 ! Performance/cost ¼ MH{ 0.33 0.83 0.39 15 16 ! Performance/cost ¼ MH ! Ease of use ¼ MH 0.4 0.47 0.88 0.5 0.46 17 18 19 20 21 0.33 0.47 0.47 0.33 0.4 0.63 0.88 0.88 0.63 0.6 0.56 0.46 0.46 0.56 0.5 ! Rating voltage ¼ 36 0.4 0.67 0.67 23 ! Ease of use ¼ MH ! Ease of use ¼ MH ! Little electrical consumption ¼ MH ! Little electrical consumption ¼ MH ! (Performance/cost ¼ MHand Ease of use ¼ MH) ! (Performance/cost ¼ MH and Little noise ¼ MH) ! Ease of use ¼ MH Electrical consumption/100 miles ¼ and Over-current protection ¼ 14 Mass of EPB ¼ 36 and Rating voltage ¼ 24 Mass of EPB ¼ 36 andUnder-voltage protection ¼ 31.5 ! Electrical consumption/100 miles ¼ ! Maximum noise ¼ 45 ! Electrical consumption/100 miles ¼ ! Over-current protection ¼ 14 ! Under-voltage protection ¼ 30.5 ! Rating voltage ¼ 24 ! Under-voltage protection ¼ 31.5 ! Over-current protection ¼ 14 0.53 0.5 24 ! Little electrical consumption ¼ MH 0.47 0.88 0.46 25 ! Little electrical consumption ¼ MH ! (Electrical consumption/100 miles ¼ and Over-current protection ¼ 14) ! (Mass of EPB ¼ 36 and Rating voltage ¼ 24) ! (Mass of EPB ¼ 36 and Undervoltage protection ¼ 31.5) 0.47 0.88 0.46 10 Negative association rules 22 { category, the rules may provide new interesting insights for the designers (Cohen et al 2001) Correspondingly, the min-sup can be set as a low value in this category For the first mining category, the minimums of high frequency sets, confidence and interestingness are assumed as 5, 0.6 and 0.35, respectively In order to facilitate the rule understanding by the designers, suppose that the number of items in a rule is not greater than The rules with high number of items are not considered and discussed in the study The extracted results are given in Table Obviously, there exist some redundant rules Rules 1, 4, 16, 17, 19–21 are identified as redundant rules which will be deleted from the rule sets For example, for three rules ~ r 1, ~ r2 and ~ r11, ~ r1 is a redundant rule of ~ r11 because all the inequalities fi ð~ r2 is r11 Þ ! fi ð~ r1 Þ are satisfied However, ~ not a redundant rule of ~ r11 because just the inequality ‘!’ is a negative expression for the item 682 Z Zhang et al f3 ð~ r11 Þ ! f3 ð~ r2 Þ is satisfied The negative rules can be used to reduce the feasible range of ECs or deduce appropriate ECs For example, the rule 15 in Table 4, if the CR ‘Performance/cost’ is not MH in a new HoQ, the EC ‘Maximum noise’ is not equal to 45 The designers will identify the EC value just from 50 and 60 Furthermore, suppose a rule: IF ‘! Ease of use ¼ MH’ THEN ‘Mass of EPB ¼ 36’ If the CR ‘Ease of use’ is not MH in a new HoQ, the EC ‘Mass of EPB’ is equal to 36 For the second, the minimums of high frequency sets, confidence and interestingness are assumed as 3, 0.9 and 1.5, respectively Generally, very low support rules are more inherently uninteresting (Cohen et al Table 2001) They may increase the number of rules and then make them difficult for reuse Thus, the minimum of high frequency sets is assigned as but not a lower value (e.g or 2) The item number in a rule is also not greater than The results are given in Table Through the checking of rule conflict and redundancy, rules 26–28 are identified as redundant rules which will be deleted from the rule sets Suppose all the extracted rules are understood and approved by the designers Assign the radius as 1.5 and the threshold value of item number in a cluster as 2, the rules are clustered into 15 categories as shown in Table Main items for each multi-CRs category are highlighted in bold fonts Rules between CRs and ECs (low support with high confidence and interestingness) Type Positive association rules No 26 IF (CRs) 1.5 ¼ H and Over-current protection ¼ 13.5 0.2 1.5 ¼ H and Little Over-current protection ¼ 13.5 0.2 1.5 ¼H Mass of EPB ¼ 36 and Over-current protection ¼ 13.5 Electrical consumption/100 miles ¼ 1.2 and Over-current protection ¼ 13.5 Maximum noise ¼ 45 0.2 0.2 0.2 Under-voltage protection ¼ 30.5 0.2 1.5 Rating voltage ¼ 24 0.2 1.5 Over-current protection ¼ 13.5 Over-current protection ¼ 13.5 0.2 0.27 1 1.5 1.5 Electrical consumption/100 miles ¼ 1.2 and Under-voltage protection ¼ 31.5 Under-voltage protection ¼ 30.5 0.2 1.5 0.27 1.5 Electrical consumption/100 miles ¼ 0.2 1.5 Maximum noise ¼ 45 0.2 Under-voltage protection ¼ 30.5 0.2 1.5 Electrical consumption/100 miles ¼ 0.27 1.5 Maximum noise ¼ 60 0.2 Rating voltage ¼ 24 0.2 1.5 Maximum load ¼ 80 0.2 1.5 Maximum noise ¼ 45 0.2 Under-voltage protection ¼ 30.5 0.2 1.5 Rating voltage ¼ 36 Mass of EPB ¼ 40 0.2 0.2 1 2.75 Performance/cost ¼ H 31 Performance/cost ¼ MH and Little electrical consumption ¼ ML Performance/cost ¼ MH and Little electrical consumption ¼ ML Ease of use ¼ VH and Little electrical consumption ¼ MH Ease of use ¼ VH and Security ¼ H Ease of use ¼ VH and Little noise ¼ MH Ease of use ¼ H 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 Int 0.2 30 32 Con Over-current protection ¼ 13.5 29 28 Sup ¼ H and Ease of Performance/cost use ¼ VH Performance/cost Security ¼ H Performance/cost noise ¼ MH Performance/cost 27 THEN (ECs) Ease of use ¼ MH and Little electrical consumption ¼ MH Ease of use ¼ MH and Aesthetics ¼ MH Ease of use ¼ MH and Aesthetics ¼ MH Ease of use ¼ MH and Aesthetics ¼ MH Ease of use ¼ MH and Little noise ¼ MH Little electrical consumption ¼ MH and Aesthetics ¼ MH Little electrical consumption ¼ MH and Aesthetics ¼ MH Little electrical consumption ¼ MH and Security ¼ MH Little electrical consumption ¼ ML andSecurity ¼ MH Little electrical consumption ¼ ML andSecurity ¼ MH Aesthetics ¼ M Security ¼ M and Little noise ¼ MH 683 International Journal of Computer Integrated Manufacturing Table Clustering results of these rules Type Positive association rules Cluster IF (CRs) Performance/cost ¼ H Performance/cost ¼ H Ease of use ¼ MH Ease of use ¼ MH Ease of use ¼ MH Ease of use ¼ H Little electrical consumption ¼ MH Little electrical consumption ¼ MH Little electrical consumption ¼ MH 10 11 Negative association rules 12 13 14 Aesthetics ¼ M Aesthetics ¼ MH Security ¼ MH Little noise ¼ MH Performance/cost ¼ MH and Ease of use ¼ MH Ease of use ¼ VH and Security ¼ H Ease of use ¼ VH and Little noise ¼ MH Ease of use ¼ MH and Aesthetics ¼ MH Ease of use ¼ MH and Aesthetics ¼ MH Ease of use ¼ MH and Aesthetics ¼ MH Ease of use ¼ MH and Little noise ¼ MH Little electrical consumption ¼ MH and Aesthetics ¼ MH Little electrical consumption ¼ MH and Aesthetics ¼ MH Performance/cost ¼ MH and Little electrical consumption ¼ ML Little electrical consumption ¼ ML and Security ¼ MH Little electrical consumption ¼ MH and Security ¼ MH Little electrical consumption ¼ ML and Security ¼ MH Performance/cost ¼ MH and Little electrical consumption ¼ ML Ease of use ¼ MH and Little electrical consumption ¼ MH Ease of use ¼ VH and Little electrical consumption ¼ MH Performance/cost ¼ MH and Little noise ¼ MH Security ¼ M and Little noise ¼ MH ! ! ! ! Performance/cost ¼ MH Performance/cost ¼ MH Ease of use ¼ MH Ease of use ¼ MH ! Little electrical consumption ¼ MH ! Little electrical consumption ¼ MH 15 ! (Performance/cost ¼ MHand Little noise ¼ MH) THEN (ECs) Sup Con Int Mass of EPB ¼ 36 and Over-current protection ¼ 13.5 Electrical consumption/100 miles ¼ 1.2 and Over-current protection ¼ 13.5 Over-current protection ¼ 14 Under-voltage protection ¼ 30.5 Electrical consumption/100 miles ¼ and Over-current protection ¼ 14 Electrical consumption/100 miles ¼ 1.2 and Under-voltage protection ¼ 31.5 Under-voltage protection ¼ 31.5 Mass of EPB ¼ 36 and Rating voltage ¼ 24 Mass of EPB ¼ 36 andUnder-voltage protection ¼ 31.5 Rating voltage ¼ 36 Over-current protection ¼ 14 Over-current protection ¼ 14 Rating power ¼ 320 Over-current protection ¼ 14 0.2 0.2 0.4 0.33 0.33 0.86 0.71 0.71 0.43 0.79 1.14 0.2 1.5 0.4 0.33 0.86 0.71 0.43 0.79 0.33 0.71 0.79 0.2 0.33 0.33 0.47 0.33 0.83 0.83 0.64 0.39 0.39 0.36 0.67 Over-current protection ¼ 13.5 Over-current protection ¼ 13.5 0.2 0.27 1 1.5 1.5 Electrical consumption/100 miles ¼ 0.2 1.5 Maximum noise ¼ 45 0.2 Under-voltage protection ¼ 30.5 0.2 1.5 Electrical consumption/100 miles ¼ 0.27 1.5 Rating voltage ¼ 24 0.2 1.5 Maximum noise ¼ 60 0.2 Maximum noise ¼ 45 0.2 Maximum noise ¼ 45 0.2 Maximum load ¼ 80 0.2 1.5 Under-voltage protection ¼ 30.5 0.2 1.5 Under-voltage protection ¼ 30.5 0.2 1.5 Under-voltage protection ¼ 30.5 0.27 1.5 Rating voltage ¼ 24 0.2 1.5 Rating voltage ¼ 36 0.4 0.33 Mass of EPB ¼ 40 0.2 2.75 ! Electrical consumption/100 miles ¼ ! Maximum noise ¼ 45 ! Under-voltage protection ¼ 30.5 ! (Electrical consumption/100 miles ¼ and Over-current protection ¼ 14) ! (Mass of EPB ¼ 36 and Rating voltage ¼ 24) ! (Mass of EPB ¼ 36 and Undervoltage protection ¼ 31.5) ! Rating voltage ¼ 36 0.33 0.4 0.47 0.53 0.83 0.88 0.39 0.5 0.46 0.5 0.47 0.88 0.46 0.47 0.88 0.46 0.4 0.67 0.67 684 Z Zhang et al Table Derived ECs for the given CRs No CRs CR1 Performance/cost ¼ H CR2 Ease of use ¼ MH CR3 Little electrical consumption ¼ MH Aesthetics ¼ MH Security ¼ H Little noise ¼ MH Ease of use ¼ MH and Aesthetics ¼ MH Ease of use ¼ MH and Little noise ¼ MH Little electrical consumption ¼ MH and Aesthetics ¼ MH Ease of use ¼ MH and Little electrical consumption ¼ MH ! Performance/cost ¼ MH CR4 CR5 CR6 CR2 and CR4 CR2 and CR6 CR3 and CR4 CR2 and CR3 ! CR1 4.2 The derived ECs Mass of EPB ¼ 36, Over-current protection ¼ 13.5, and Electrical consumption/100 miles ¼ 1.2 Over-current protection ¼ 14, Under-voltage protection ¼ 30.5, and Electrical consumption/100 miles ¼ Mass of EPB ¼ 36, Rating voltage ¼ 24, and Under-voltage protection ¼ 31.5 Over-current protection ¼ 14 NULL Rating power ¼ 320 Electrical consumption/100 miles ¼ 1, Maximum noise ¼ 45, and Under-voltage protection ¼ 30.5 Electrical consumption/100 miles ¼ Rating voltage ¼ 24 and Maximum noise ¼ 60 Under-voltage protection ¼ 30.5 ! Electrical consumption/100 miles ¼ and ! Maximum noise ¼ 45 Guide to a new product development Suppose a CR set for new product development, i.e (CR1 ¼ H, CR2 ¼ MH, CR3 ¼ MH, CR4 ¼ MH, CR5 ¼ H, CR6 ¼ MH,), and then a sub-set of ECs is derived in Table based on the corresponding rules searched from Table Obviously, there exist some repeated or conflicting ECs The repeated ECs are united For the conflicting ECs, their total objective values are calculated by summing corresponding objective values For example, ‘Electrical consumption/100 miles ¼ 1’ and ‘Electrical consumption/100 miles ¼ 1.2’ are conflicted Obviously, the EC ‘Electrical consumption/100 miles ¼ 1’ is preserved because its total values of three objectives (i.e 0.47, 1.88 and 3.75) are greater than these (i.e 0.2, 1, and 2) of ‘Electrical consumption/100 miles ¼ 1.2’ If the three objective values are equal, the results are only determined by the designers The preserved ECs are highlighted in bold fonts as shown in Table 7, i.e (Mass of EPB ¼ 36, Electrical consumption/100 miles ¼ 1, Maximum noise ¼ 60, Rating voltage ¼ 24, Rating power ¼ 320, Over-current protection 14, and Under-voltage protection ¼ 30.5) A new HoQ is still needed to identify the rest of ECs, in which the number of ECs is just (i.e Maximum speed, Driving range, Maximum load, and Rating output torque) Obviously, the HoQ is more convenient for the designers to handle Conclusions The paper proposes an Apriori-based data mining approach in order to provide the designers with sufficient design knowledge to establish the relationships between CRs and ECs in HoQ The approach is aimed to find out the useful association rules that reflect the relationships according to three objectives: support, confidence and interestingness Considering the possible unnecessary rules in the extracted rules, rule conflict and redundancy should be checked and then corresponding processing procedures are developed These rules are still affirmed by the designers with their experience The approved rules can be used for identifying appropriate ECs in the following HoQs In order to facilitate rule management and reuse, these approved rules should be clustered A reuse procedure is also developed for the new HoQ analysis The proposed approach and its availability are illustrated by an EPB case The results show that the extracted rules can effectively help the designers to fulfill the analysis of HoQ and also can distinctly reduce the complexity of HoQ In the proposed Apriori-based data mining approach, the identification of minimums of three objectives (i.e support, confidence, and interestingness) is mainly depended on the experience of designers Obviously, these values have some effects on the results of data mining How to determine these values more objectively and credibly needs to be discussed In order to ensure the reliably of results, a lot of data (i.e design cases of QFD) are required The proposed approach is more suitable for these applications with mass of data A real-world application is needed to test the proposed method Furthermore, the proposed approach for other steps (i.e parts deployment, process planning and production planning) in QFD will be considered International Journal of Computer Integrated Manufacturing Acknowledgements The project was supported by Shanghai Technology Innovative Activity Planning Program (09dz1124600, 10dz1121600), National High-Tech RandD Program for CIMS, China (No 2007AA04Z140), Research Fund for Doctoral Program of Higher Education, China (No 20070248020) and Shanghai Leading Academic Discipline, China (No Y0102) The authors express sincere appreciation to the anonymous referees for their helpful comments to improve the quality of the paper Thanks also go to Dr Li Yu for invaluable advice and encouragement References Agrawal, R., Imielinski, T., and Swami, A., 1993 Mining association rules between sets of items in large databases In: Proceedings of the ACM SIGMOD International Conference on Management of Data (ACM SIGMOD 930 ), Washington, USA, 207–216 Agrawal, R and Srikant, R., 1994 Fast algorithms for mining association rules In: Proceedings of the 20th International Conference on Very Large Databases (VLDB0 94), Santiago, Chile, 487–499 Akao, Y., 1990 Quality function deployment: integrating customer requirements into product design Cambridge, MA: The Productivity Press Armacost, R.L., Componation, P.J., Mullens, M.A., et al., 1994 An AHP framework for prioritizing customer requirements in QFD: An industrialized housing application IIE Transactions, 26 (4), 72–79 Bai, H and Kwong, C.K., 2003 Inexact genetic algorithm approach to target values setting of engineering requirements in QFD International Journal of Production Research, 41 (16), 3861–3881 Bu¨yu¨ko¨zkan, G., Derelı, T., and Baykasoglu, A., 2004 A survey on the methods and tools of concurrent new product development and agile manufacturing Journal of Intelligent Manufacturing, 15 (6), 731–751 Bu¨yu¨ko¨zkan, G and Feyzioglu, O., 2004 A fuzzy-logicbased decision-making approach for new product development International Journal of Production Economics, 90, 27–45 Bu¨yu¨ko¨zkan, G and Feyzioglu, O., 2005 Group decision making to better respond customer needs in software development Computers and Industrial Engineering, 48, 427–441 Bu¨yu¨ko¨zkan, G., Feyzioglu, O., and Ruan, D., 2007 Fuzzy group decision-making to multiple preference formats in quality function deployment Computers in Industry, 58 (5), 392–402 Chan, L.K., Kao, H.P., Ng, A., et al., 1999 Rating the importance of customer needs in quality function deployment by fuzzy and entropy methods International Journal of Production Research, 37 (11), 2499–2518 Chan, L.K and Wu, M.L., 2002 Quality function deployment: a literature review European Journal of Operational Research, 143, 463–497 Chen, L.-H and Ko, W.C., 2008 A fuzzy nonlinear model for quality function deployment considering Kano’s concept Mathematical and Computer Modelling, 48 (3–4), 581–593 Chen, L.-H and Weng, M.C., 2006 An evaluation approach to engineering design in QFD processed using fuzzy goal programming models European Journal of Operational Research, 172, 230–248 685 Chen, M.-S., Han, J., and Yu, P.S., 1996 Data mining: an overview from a database perspective IEEE Transactions on Knowledge and Data Engineering, (6), 866–883 Chen, Y.Z., Fung, R.Y.K., and Tang, J.F., 2006 Rating technical attributes in fuzzy QFD by integrating fuzzy weighted average method and fuzzy expected value operator European Journal of Operational Research, 174, 1553–1566 Chen, Y., Tang, J., Fung, R.Y.K., et al., 2004 Fuzzy regression-based mathematical programming model for quality function deployment International Journal of Production Research, 42 (5), 1009–1027 Chen, Y.Z and Ngai, E.W.T., 2008 A fuzzy QFD program modelling approach using the method of imprecision International Journal of Production Research, 46 (24), 6823–6840 Chin, K.S., Lam, J., Chan, J.S.F., et al., 2005 A CIMOSA presentation of an integrated product design review framework International Journal of Computer Integrated Manufacturing, 84 (4), 260–278 Choudhary, A.K., Harding, J.A., and Tiwari, M.K., 2009 Data mining in manufacturing: a review based on the kind of knowledge Journal of Intelligent Manufacturing, 20, 501–521 Cohen, E., Datar, M., Fujiwara, S., et al., 2001 Finding interesting associations without support pruning IEEE Transactions on Knowledge and Data Engineering, 13 (1), 64–78 Dong, X.J., Wang, S.J., Song, H.T., et al., 2004 Study on negative association rules Transactions of Beijing Institute of Technology, 24 (11), 978–981 Ester, M., Kriegel, H.-P., Sander, J., et al., 1996 A density-based algorithm for discovering clusters in large spatial database with noise In: Proceedings of 2nd International Conference on Knowledge Discovery and Data Mining (KDD-96), Portland, Oregon, 226–231 Feyzio glu, O and Bu¨yu¨ko¨zkan, G., 2008 An integrated group decision-making approach for new product development International Journal of Computer Integrated Manufacturing, 21 (4), 366–375 Fung, R.Y.K., Popplewell, K., and Xie, J., 1998 An intelligent hybrid system for customer requirements analysis and product attribute targets determination International Journal of Production Research, 36 (1), 13–34 Fung, R.Y.K., Tang, J.F., Tu, P.Y.L., et al., 2003 Modelling of quality function deployment planning with resource allocation Research of Engineering Design, 14, 247–255 Fung, R.Y K., Tang, J., Tu, Y., et al., 2002 Product design resources optimization using a non-linear fuzzy quality function deployment model International Journal of Production Research, 40 (3), 585–599 Fung, R.Y.K., Chen, Y.Z., Tang, J.F., et al., 2006 Estimating the functional relationships for quality function deployment under uncertainties Fuzzy Sets and Systems, 157, 98–120 Gray, B and Orlowska, M.E., 1998 CCAIIA: Clustering categorical attributes into interesting association rules In: Proceedings of the second Pacific-Asia Conference on Research and Development in Knowledge Discovery and Data Mining, 15–17 132–143 Griffin, A and Hauser, J.R., 1993 The voice of the Customer Marketing Science, 12 (1), 1–27 686 Z Zhang et al Han, C.H., Kim, J.K., and Choi, S.H., 2004 Prioritizing engineering characteristics in quality function deployment with incomplete information: a linear partial ordering approach International Journal of Production Economics, 91, 235–249 Harding, J.A., Popplewell, K., Fung, R.Y.K., et al., 2001 An intelligent information framework relating customer requirements and product characteristics Computers in Industry, 44, 51–65 Harding, J.A., Shahbaz, M., Srinivas, S., and Kusiak, A., 2006 Data mining in manufacturing: a review Journal of Manufacturing Science and Engineering, 128, 969–976 Ho, E.S.S.A., Lai, Y.J., and Chang, S.I., 1999 An integrated group decision-making approach to quality function deployment IIE Transactions, 31, 553–567 Hauser, J.R and Clausing, D., 1988 The house of quality Harvard Business Review, 66, 63–73 Jiao, J.X and Zhang, Y.Y., 2005 Product portfolio identification based on association rule mining Computer-Aided Design, 37, 149–172 Kahraman, C., Ertay, T., and Bu¨yu¨ko¨zkan, G., 2006 A fuzzy optimization model for QFD planning process using analytic network approach European Journal of Operational Research, 171, 390–411 Kim, K.J., Moskowitz, H., Dhingra, A., et al., 2000 Fuzzy multicriteria models for quality function deployment European Journal of Operational Research, 121, 504–518 Kwong, C.K and Bai, H., 2002 A fuzzy AHP approach to the determination of importance weights of customer requirements in quality function deployment Journal of Intelligent Manufacturing, 13 (5), 367–377 Lai, Y.J., Ho, E.S.S.A., and Chang, S.I., 1998 Identifying customer preferences in quality function deployment using group decision-making techniques In: J Usher, U Roy, and H Parsaei, eds Integrated Product and Process Development Wiley: New York, 1–28 Lai, X., Xie, M., Tan, K.C., and Yang, B., 2008 Ranking of customer requirements in a competitive environment Computers and Industrial Engineering, 54, 202–214 Li, J.R and Wang, Q.H., 2010 A rough set based data mining approach for house of quality analysis International Journal of Production Research, 48 (7), 2095–2107 Li, X.D., Tu, Y.L., Xue, D.Y., et al., 2006 Importance measures for customer requirements and technical attributes in fuzzy environment In: Proceedings of the 17th IASTED International Conference Modelling and Simulation, Montreal, QC, Canada, 193–200 Liao, S.H., Hsieh, C.L., and Huang, S.P., 2008 Mining product maps for new product development Expert System with Applications, 34, 50–62 Lu, M.H., Madu, C.N., Kuei, C., et al., 1994 Integrating QFD, AHP and benchmarking in strategic marketing Journal of Business and Industrial Marketing, (1), 41–50 Matzler, K and Hinterhuber, H.H., 1998 How to make product development projects more successful by integrating Kano’s model of customer satisfaction into quality function deployment Technovation, 18 (1), 25–38 Shahbaz, M., Srinivas, S., Harding, J.A., and Turner, M., 2006 Product design and manufacturing process improvement using association rules Proceedings of the Institution of Mechanical Engineers, Part B: Journal of Engineering Manufacture, 220, 243–254 Shao, X.Y., Wang, Z.H., Li, P.G., et al., 2006 Integrating data mining and rough set for customer group-based discovery of product configuration rules International Journal of Production Research, 44 (14), 2789–2811 Shen, X.X., Tan, K.C., and Xie, M., 2001 The implementation of quality function deployment based on linguistic data Journal of Intelligent Manufacturing, 12 (1), 65–75 Temponi, C., Yen, J., and Tiao, W.A., 1999 House of quality: a fuzzy logic-based requirements analysis European Journal of Operational Research, 117, 340–354 Tu, Y.L., Fung, R.Y.K., Tang, J.F., et al., 2003 Computeraided customer interface for rapid product development International Journal of Advanced Manufacturing Technology, 21, 743–753 Vanegas, L.V and Labib, A.W., 2001 A fuzzy quality function deployment (FQFD) model for deriving optimum targets International Journal of Production Research, 39 (1), 99–120 Wu, X.D., Zhang, C.Q., and Zhang, S.C., 2002 Mining both positive and negative association rules In: Proceedings of the 19th International Conference on Machine Learning (ICML-2002), San Francisco, USA, 658–665 Xia, S.S and Wang, L.Y., 2010 Customer requirements mapping method based on association rule mining for mass customization International Journal of Computer Applications in Technology, 37 (3–4), 198–203 Zhang, X.P., Bode, J., and Ren, S.J., 1996 Neural networks in quality function deployment Computers and Industrial Engineering, 31 (3/4), 669–673 Zhang, Z.F and Chu, X.N., 2009a A new integrated decision-making approach for design alternative selection for supporting complex product development International Journal of Computer Integrated Manufacturing, 22 (3), 179–198 Zhang, Z.F and Chu, X.N., 2009b Fuzzy group decisionmaking for multi-format and multi-granularity linguistic judgments in quality function deployment Expert System with Applications, 36 (5), 9150–9158 [...]... (0.26 6,0 .34 1,0 .37 1,0 .468) (0.00 4,0 .04 6,0 .06 4,0 .130) (0.24 5,0 .31 4,0 .35 3,0 .445) (0.00 1,0 .04 5,0 .04 7,0 .115) (0.00 4,0 .12 5,0 .29 7,0 .942) MS1 (0.07 8,0 .13 9,0 .16 3,0 .244) (0.23 5,0 .32 8,0 .37 6,0 .505) (0.21 6,0 .28 3,0 .30 1,0 .404) (0.07 6,0 .13 0,0 .14 5,0 .222) (0.18 1,0 .24 7,0 .28 1,0 .365) (0.04 5,0 .10 4,0 .12 0,0 .194) (0.00 4,0 .13 8,0 .32 1,0 .947) MS2 M4 M5 (0.26 5,0 .34 5,0 .37 8,0 .485) (0.10 1,0 .17 2,0 .21 0,0 .318) (0.15 1,0 .21 8,0 .24 5,0 .326) (0.30 9,0 .37 4,0 .38 5,0 .488)... (0.30 9,0 .37 4,0 .38 5,0 .488) (0.16 6,0 .23 1,0 .26 5,0 .347) (0.29 3,0 .37 3,0 .40 4,0 .509) (0.00 7,0 .20 1,0 .42 3,0 .964) MS3 0 (0.02 3,0 .06 6,0 .07 8,0 .143) 0 0 0 0 (0.02 7,0 .08 0,0 .09 7,0 .185) 0 (0.00 3,0 .06 1,0 .14 3,0 .801) (0.04 7,0 .15 1,0 .17 3,0 .277) 1 (0.17 0,0 .27 5,0 .31 3,0 .418) Hydraulic module 0 (0.06 2,0 .11 1,0 .13 3,0 .210) (0.04 3,0 .11 5,0 .13 4,0 .222) 0 (1.00 0,1 .00 0,1 .00 0,1 .000) 0 (0.20 1,0 .28 6,0 .32 8,0 .477) 0 (0.00 3,0 .08 2,0 .19 0,0 .857)... 0 0 PEC7 International Journal of Computer Integrated Manufacturing 609 610 (0.01 7,0 .05 3, 0.06 4,0 .129) (0.16 5,0 .28 5, 0.32 4,0 .446) (0.01 0,0 .05 5, 0.08 4,0 .317) MH (0.00 7,0 .03 6, 0.05 6,0 .221) H (0.00 4,0 .01 6, 0.02 4,0 .098) M 0 (0.01 7,0 .05 3, 0.06 4,0 .129) (0.27 7,0 .37 2, 0.41 7,0 .556) (0.01 7,0 .07 6, 0.11 4,0 .392) H 0 (0.01 0,0 .05 8, 0.09 1,0 .350) H CV (0.01 1,0 .04 5, 0.06 7,0 .246) M RW CR8 (0.01 2,0 .04 5, 0.05 6,0 .119) 0... 0.19 1,0 .305) (0.00 9,0 .05 6, 0.08 7,0 .340) H MEC1 (0.09 0,0 .14 1, 0.16 7,0 .251) 0 (0.03 2,0 .09 4, 0.11 2,0 .208) (0.04 2,0 .08 2, 0.10 2,0 .180) 0 0 0 (0.09 0,0 .14 1, 0.16 7,0 .251) (0.03 8,0 .10 4, 0.12 1,0 .211) 0 PEC2 0 0 (0.04 7,0 .15 1, 0.17 3,0 .277) (0.01 1,0 .11 5, 0.13 6,0 .241) PEC2 0 0 0 0 0 (0.01 8,0 .09 9, 0.14 7,0 .461) MH-H MEC2 0 (0.01 2,0 .04 5, 0.05 6,0 .119) 0 0 (0.03 0,0 .09 1, 0.11 1,0 .221) (0.17 3,0 .24 4, 0.27 0,0 .362) 0 (0.04 8,0 .08 8,. .. 0 (0.24 3,0 .34 9, 0.39 9,0 .594) 0 (0.04 3,0 .08 2, 0.10 1,0 .169) 0 PEC5 0 0 0 0 PEC5 0 0 0 0 (0.01 7,0 .06 9, 0.09 8,0 .315) H MEC5 (0.01 7,0 .05 2, 0.06 3,0 .120) 0 0 0 0 (0.39 3,0 .48 3, 0.51 2,0 .627) 0 (0.03 8,0 .10 4, 0.12 1,0 .211) 0 PEC6 0 0 0 0 PEC6 (continued) 0 0 (0.03 0,0 .09 1, 0.11 1,0 .221) 0 0 (0.03 7,0 .10 4, 0.12 3,0 .217) (0.00 7,0 .03 7, 0.05 7,0 .226) MH MEC6 0 (0.36 4,0 .47 7, 0.52 3,0 .636) (0.03 2,0 .09 4, 0.11 2,0 .208) 0 0 0... (0.04 8,0 .08 8, 0.10 7,0 .178) 0 PEC3 0 0 (0.04 7,0 .15 1, 0.17 3,0 .277) 0 PEC3 0 0 0 0 0 (0.00 7,0 .02 7, 0.04 0,0 .152) MH MEC3 0 0 (0.32 8,0 .43 9, 0.48 5,0 .659) 0 0 0 0 0 0 PEC4 0 0 0 0 PEC4 (0.02 1,0 .07 9, 0.09 9,0 .206) (0.08 1,0 .14 1, 0.16 3,0 .244) 0 (0.07 9,0 .14 9, 0.18 0,0 .290) (0.01 3,0 .06 0, 0.09 2,0 .329) Automatic control; MEC4 (0.04 8,0 .08 8, 0.10 7,0 .178) 0 (0.08 3,0 .13 5, 0.16 3,0 .258) 0 (0.36 4,0 .47 7, 0.52 3,0 .636) 0 (0.24 3,0 .34 9,. .. (0.67 1,0 .75 4,0 .77 8,0 .855) 0 (0.16 8,0 .23 3,0 .26 6,0 .368) (0.17 5,0 .24 2,0 .28 7,0 .379) 0 (0.22 6,0 .31 7,0 .36 1,0 .455) 0 0 0 0 (0.24 9,0 .33 1,0 .37 5,0 .489) (0.09 2,0 .15 8,0 .19 6,0 .279) (0.13 9,0 .21 9,0 .25 5,0 .382) (0.07 4,0 .13 6,0 .17 5,0 .272) 0 0 (0.00 9,0 .16 6,0 .34 1,0 .935) (0.00 3,0 .11 8,0 .27 5,0 .914) The results are given in Table 3 M2 M3 MEC1 MEC2 MEC3 MEC4 MEC5 MEC6 RW (0.00 5,0 .05 7,0 .07 2,0 .142) (0.23 5,0 .32 8,0 .37 6,0 .505) (0.26 6,0 .34 1,0 .37 1,0 .468)... matrix of engine module, hydraulic module and electric module Table 2 (0.32 2,0 .40 9,0 .44 3,0 .557) (0.03 6,0 .10 0,0 .12 4,0 .215) (0.05 5,0 .10 8,0 .10 9,0 .199) (0.33 1,0 .41 9,0 .43 0,0 .513) (0.09 3,0 .14 6,0 .17 3,0 .246) (0.34 9,0 .43 9,0 .47 1,0 .575) (0.00 6,0 .19 2,0 .40 6,0 .961) MS4 0 (0.05 5,0 .10 3,0 .12 5,0 .199) 0 0 0 (0.08 1,0 .14 6,0 .17 4,0 .264) (0.07 4,0 .13 6,0 .17 5,0 .272) (0.75 8,0 .84 8,0 .86 5,0 .956) (0.00 2,0 .05 4,0 .13 3,0 .807) M6 (0.04 7,0 .15 1,0 .17 3,0 .277)... higher the RPN, the more important and serious the failure could be Fuzzy theory has been incorporated with RPN calculation (Chang et al 199 9, Braglia et al 200 3, Wang et al 2009) In the study, fuzzy RPNs are still used A fuzzy ten-point scale is adopted for evaluating O, S and D, i.e ( 1, 1, 2 ), ( 1, 2, 3 ), ( 2, 3, 4 ), ( 3, 4, 5 ), ( 4, 5, 6 ), ( 5, 6, 7 ), ( 6, 7, 8 ), ( 7, 8, 9 ), ( 8, 9, 10 ), ( 9, 1 0, 10) The fuzzy... strategies (MSs) of engine module PEC1 PEC2 PEC3 PEC4 PEC5 PEC6 PEC7 PEC8 RW CV M1 (0.14 5,0 .22 2,0 .24 6,0 .329) (0.11 7,0 .17 9,0 .20 7,0 .296) (0.42 5,0 .52 4,0 .56 2,0 .686) (1.00 0,1 .00 0,1 .00 0,1 .000) 0 (0.21 6,0 .30 0,0 .33 8,0 .448) (0.01 9,0 .07 0,0 .08 6,0 .172) (0.04 4,0 .13 5,0 .15 2,0 .243) (0.00 8,0 .18 2,0 .37 2,0 .940) 1 (0.04 7,0 .15 1,0 .17 3,0 .277) (0.04 7,0 .15 1,0 .17 3,0 .277) (b) Relationship matrix between PEC and modules of product Engine